blob: e76b559179054b93191b8959ddcad9ad01535e8c [file] [log] [blame]
Thomas Gleixner5b497af2019-05-29 07:18:09 -07001// SPDX-License-Identifier: GPL-2.0-only
Alexei Starovoitov51580e72014-09-26 00:17:02 -07002/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
Alexei Starovoitov969bf052016-05-05 19:49:10 -07003 * Copyright (c) 2016 Facebook
Joe Stringerfd978bf72018-10-02 13:35:35 -07004 * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io
Alexei Starovoitov51580e72014-09-26 00:17:02 -07005 */
Yonghong Song838e9692018-11-19 15:29:11 -08006#include <uapi/linux/btf.h>
Alexei Starovoitov51580e72014-09-26 00:17:02 -07007#include <linux/kernel.h>
8#include <linux/types.h>
9#include <linux/slab.h>
10#include <linux/bpf.h>
Yonghong Song838e9692018-11-19 15:29:11 -080011#include <linux/btf.h>
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010012#include <linux/bpf_verifier.h>
Alexei Starovoitov51580e72014-09-26 00:17:02 -070013#include <linux/filter.h>
14#include <net/netlink.h>
15#include <linux/file.h>
16#include <linux/vmalloc.h>
Thomas Grafebb676d2016-10-27 11:23:51 +020017#include <linux/stringify.h>
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080018#include <linux/bsearch.h>
19#include <linux/sort.h>
Yonghong Songc195651e2018-04-28 22:28:08 -070020#include <linux/perf_event.h>
Martin KaFai Laud9762e82018-12-13 10:41:48 -080021#include <linux/ctype.h>
KP Singh6ba43b72020-03-04 20:18:50 +010022#include <linux/error-injection.h>
KP Singh9e4e01d2020-03-29 01:43:52 +010023#include <linux/bpf_lsm.h>
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070024#include <linux/btf_ids.h>
Alexei Starovoitov51580e72014-09-26 00:17:02 -070025
Jakub Kicinskif4ac7e02017-10-09 10:30:12 -070026#include "disasm.h"
27
Jakub Kicinski00176a32017-10-16 16:40:54 -070028static const struct bpf_verifier_ops * const bpf_verifier_ops[] = {
Alexei Starovoitov91cc1a92019-11-14 10:57:15 -080029#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
Jakub Kicinski00176a32017-10-16 16:40:54 -070030 [_id] = & _name ## _verifier_ops,
31#define BPF_MAP_TYPE(_id, _ops)
Andrii Nakryikof2e10bf2020-04-28 17:16:08 -070032#define BPF_LINK_TYPE(_id, _name)
Jakub Kicinski00176a32017-10-16 16:40:54 -070033#include <linux/bpf_types.h>
34#undef BPF_PROG_TYPE
35#undef BPF_MAP_TYPE
Andrii Nakryikof2e10bf2020-04-28 17:16:08 -070036#undef BPF_LINK_TYPE
Jakub Kicinski00176a32017-10-16 16:40:54 -070037};
38
Alexei Starovoitov51580e72014-09-26 00:17:02 -070039/* bpf_check() is a static code analyzer that walks eBPF program
40 * instruction by instruction and updates register/stack state.
41 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
42 *
43 * The first pass is depth-first-search to check that the program is a DAG.
44 * It rejects the following programs:
45 * - larger than BPF_MAXINSNS insns
46 * - if loop is present (detected via back-edge)
47 * - unreachable insns exist (shouldn't be a forest. program = one function)
48 * - out of bounds or malformed jumps
49 * The second pass is all possible path descent from the 1st insn.
Zhen Lei8fb33b62021-05-25 10:56:59 +080050 * Since it's analyzing all paths through the program, the length of the
Gary Lineba38a92017-03-01 16:25:51 +080051 * analysis is limited to 64k insn, which may be hit even if total number of
Alexei Starovoitov51580e72014-09-26 00:17:02 -070052 * insn is less then 4K, but there are too many branches that change stack/regs.
53 * Number of 'branches to be analyzed' is limited to 1k
54 *
55 * On entry to each instruction, each register has a type, and the instruction
56 * changes the types of the registers depending on instruction semantics.
57 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
58 * copied to R1.
59 *
60 * All registers are 64-bit.
61 * R0 - return register
62 * R1-R5 argument passing registers
63 * R6-R9 callee saved registers
64 * R10 - frame pointer read-only
65 *
66 * At the start of BPF program the register R1 contains a pointer to bpf_context
67 * and has type PTR_TO_CTX.
68 *
69 * Verifier tracks arithmetic operations on pointers in case:
70 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
71 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
72 * 1st insn copies R10 (which has FRAME_PTR) type into R1
73 * and 2nd arithmetic instruction is pattern matched to recognize
74 * that it wants to construct a pointer to some element within stack.
75 * So after 2nd insn, the register R1 has type PTR_TO_STACK
76 * (and -20 constant is saved for further stack bounds checking).
77 * Meaning that this reg is a pointer to stack plus known immediate constant.
78 *
Edward Creef1174f72017-08-07 15:26:19 +010079 * Most of the time the registers have SCALAR_VALUE type, which
Alexei Starovoitov51580e72014-09-26 00:17:02 -070080 * means the register has some value, but it's not a valid pointer.
Edward Creef1174f72017-08-07 15:26:19 +010081 * (like pointer plus pointer becomes SCALAR_VALUE type)
Alexei Starovoitov51580e72014-09-26 00:17:02 -070082 *
83 * When verifier sees load or store instructions the type of base register
Joe Stringerc64b7982018-10-02 13:35:33 -070084 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are
85 * four pointer types recognized by check_mem_access() function.
Alexei Starovoitov51580e72014-09-26 00:17:02 -070086 *
87 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
88 * and the range of [ptr, ptr + map's value_size) is accessible.
89 *
90 * registers used to pass values to function calls are checked against
91 * function argument constraints.
92 *
93 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
94 * It means that the register type passed to this function must be
95 * PTR_TO_STACK and it will be used inside the function as
96 * 'pointer to map element key'
97 *
98 * For example the argument constraints for bpf_map_lookup_elem():
99 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
100 * .arg1_type = ARG_CONST_MAP_PTR,
101 * .arg2_type = ARG_PTR_TO_MAP_KEY,
102 *
103 * ret_type says that this function returns 'pointer to map elem value or null'
104 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
105 * 2nd argument should be a pointer to stack, which will be used inside
106 * the helper function as a pointer to map element key.
107 *
108 * On the kernel side the helper function looks like:
109 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
110 * {
111 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
112 * void *key = (void *) (unsigned long) r2;
113 * void *value;
114 *
115 * here kernel can access 'key' and 'map' pointers safely, knowing that
116 * [key, key + map->key_size) bytes are valid and were initialized on
117 * the stack of eBPF program.
118 * }
119 *
120 * Corresponding eBPF program may look like:
121 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
122 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
123 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
124 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
125 * here verifier looks at prototype of map_lookup_elem() and sees:
126 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
127 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
128 *
129 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
130 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
131 * and were initialized prior to this call.
132 * If it's ok, then verifier allows this BPF_CALL insn and looks at
133 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
134 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
Zhen Lei8fb33b62021-05-25 10:56:59 +0800135 * returns either pointer to map value or NULL.
Alexei Starovoitov51580e72014-09-26 00:17:02 -0700136 *
137 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
138 * insn, the register holding that pointer in the true branch changes state to
139 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
140 * branch. See check_cond_jmp_op().
141 *
142 * After the call R0 is set to return type of the function and registers R1-R5
143 * are set to NOT_INIT to indicate that they are no longer readable.
Joe Stringerfd978bf72018-10-02 13:35:35 -0700144 *
145 * The following reference types represent a potential reference to a kernel
146 * resource which, after first being allocated, must be checked and freed by
147 * the BPF program:
148 * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET
149 *
150 * When the verifier sees a helper call return a reference type, it allocates a
151 * pointer id for the reference and stores it in the current function state.
152 * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into
153 * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type
154 * passes through a NULL-check conditional. For the branch wherein the state is
155 * changed to CONST_IMM, the verifier releases the reference.
Joe Stringer6acc9b42018-10-02 13:35:36 -0700156 *
157 * For each helper function that allocates a reference, such as
158 * bpf_sk_lookup_tcp(), there is a corresponding release function, such as
159 * bpf_sk_release(). When a reference type passes into the release function,
160 * the verifier also releases the reference. If any unchecked or unreleased
161 * reference remains at the end of the program, the verifier rejects it.
Alexei Starovoitov51580e72014-09-26 00:17:02 -0700162 */
163
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700164/* verifier_state + insn_idx are pushed to stack when branch is encountered */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +0100165struct bpf_verifier_stack_elem {
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700166 /* verifer state is 'st'
167 * before processing instruction 'insn_idx'
168 * and after processing instruction 'prev_insn_idx'
169 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +0100170 struct bpf_verifier_state st;
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700171 int insn_idx;
172 int prev_insn_idx;
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +0100173 struct bpf_verifier_stack_elem *next;
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -0700174 /* length of verifier log at the time this state was pushed on stack */
175 u32 log_pos;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700176};
177
Alexei Starovoitovb285fcb2019-05-21 20:14:19 -0700178#define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192
Alexei Starovoitovceefbc92018-12-03 22:46:06 -0800179#define BPF_COMPLEXITY_LIMIT_STATES 64
Daniel Borkmann07016152016-04-05 22:33:17 +0200180
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +0100181#define BPF_MAP_KEY_POISON (1ULL << 63)
182#define BPF_MAP_KEY_SEEN (1ULL << 62)
183
Daniel Borkmannc93552c2018-05-24 02:32:53 +0200184#define BPF_MAP_PTR_UNPRIV 1UL
185#define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \
186 POISON_POINTER_DELTA))
187#define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV))
188
189static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux)
190{
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +0100191 return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON;
Daniel Borkmannc93552c2018-05-24 02:32:53 +0200192}
193
194static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux)
195{
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +0100196 return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV;
Daniel Borkmannc93552c2018-05-24 02:32:53 +0200197}
198
199static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux,
200 const struct bpf_map *map, bool unpriv)
201{
202 BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV);
203 unpriv |= bpf_map_ptr_unpriv(aux);
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +0100204 aux->map_ptr_state = (unsigned long)map |
205 (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL);
206}
207
208static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux)
209{
210 return aux->map_key_state & BPF_MAP_KEY_POISON;
211}
212
213static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux)
214{
215 return !(aux->map_key_state & BPF_MAP_KEY_SEEN);
216}
217
218static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux)
219{
220 return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON);
221}
222
223static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state)
224{
225 bool poisoned = bpf_map_key_poisoned(aux);
226
227 aux->map_key_state = state | BPF_MAP_KEY_SEEN |
228 (poisoned ? BPF_MAP_KEY_POISON : 0ULL);
Daniel Borkmannc93552c2018-05-24 02:32:53 +0200229}
Martin KaFai Laufad73a12017-03-22 10:00:32 -0700230
Yonghong Song23a2d702021-02-04 15:48:27 -0800231static bool bpf_pseudo_call(const struct bpf_insn *insn)
232{
233 return insn->code == (BPF_JMP | BPF_CALL) &&
234 insn->src_reg == BPF_PSEUDO_CALL;
235}
236
Martin KaFai Laue6ac2452021-03-24 18:51:42 -0700237static bool bpf_pseudo_kfunc_call(const struct bpf_insn *insn)
238{
239 return insn->code == (BPF_JMP | BPF_CALL) &&
240 insn->src_reg == BPF_PSEUDO_KFUNC_CALL;
241}
242
Yonghong Song69c087b2021-02-26 12:49:25 -0800243static bool bpf_pseudo_func(const struct bpf_insn *insn)
244{
245 return insn->code == (BPF_LD | BPF_IMM | BPF_DW) &&
246 insn->src_reg == BPF_PSEUDO_FUNC;
247}
248
Daniel Borkmann33ff9822016-04-13 00:10:50 +0200249struct bpf_call_arg_meta {
250 struct bpf_map *map_ptr;
Daniel Borkmann435faee12016-04-13 00:10:51 +0200251 bool raw_mode;
Daniel Borkmann36bbef52016-09-20 00:26:13 +0200252 bool pkt_access;
Daniel Borkmann435faee12016-04-13 00:10:51 +0200253 int regno;
254 int access_size;
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700255 int mem_size;
John Fastabend10060502020-03-30 14:36:19 -0700256 u64 msize_max_value;
Martin KaFai Lau1b986582019-03-12 10:23:02 -0700257 int ref_obj_id;
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -0700258 int map_uid;
Alexei Starovoitovd83525c2019-01-31 15:40:04 -0800259 int func_id;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -0800260 struct btf *btf;
Hao Luoeaa6bcb2020-09-29 16:50:47 -0700261 u32 btf_id;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -0800262 struct btf *ret_btf;
Hao Luoeaa6bcb2020-09-29 16:50:47 -0700263 u32 ret_btf_id;
Yonghong Song69c087b2021-02-26 12:49:25 -0800264 u32 subprogno;
Daniel Borkmann33ff9822016-04-13 00:10:50 +0200265};
266
Alexei Starovoitov8580ac92019-10-15 20:24:57 -0700267struct btf *btf_vmlinux;
268
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700269static DEFINE_MUTEX(bpf_verifier_lock);
270
Martin KaFai Laud9762e82018-12-13 10:41:48 -0800271static const struct bpf_line_info *
272find_linfo(const struct bpf_verifier_env *env, u32 insn_off)
273{
274 const struct bpf_line_info *linfo;
275 const struct bpf_prog *prog;
276 u32 i, nr_linfo;
277
278 prog = env->prog;
279 nr_linfo = prog->aux->nr_linfo;
280
281 if (!nr_linfo || insn_off >= prog->len)
282 return NULL;
283
284 linfo = prog->aux->linfo;
285 for (i = 1; i < nr_linfo; i++)
286 if (insn_off < linfo[i].insn_off)
287 break;
288
289 return &linfo[i - 1];
290}
291
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700292void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt,
293 va_list args)
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700294{
Jakub Kicinskia2a7d572017-10-09 10:30:15 -0700295 unsigned int n;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700296
Jakub Kicinskia2a7d572017-10-09 10:30:15 -0700297 n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args);
Jakub Kicinskia2a7d572017-10-09 10:30:15 -0700298
299 WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1,
300 "verifier log line truncated - local buffer too short\n");
301
302 n = min(log->len_total - log->len_used - 1, n);
303 log->kbuf[n] = '\0';
304
Alexei Starovoitov8580ac92019-10-15 20:24:57 -0700305 if (log->level == BPF_LOG_KERNEL) {
306 pr_err("BPF:%s\n", log->kbuf);
307 return;
308 }
Jakub Kicinskia2a7d572017-10-09 10:30:15 -0700309 if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1))
310 log->len_used += n;
311 else
312 log->ubuf = NULL;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700313}
Jiri Olsaabe08842018-03-23 11:41:28 +0100314
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -0700315static void bpf_vlog_reset(struct bpf_verifier_log *log, u32 new_pos)
316{
317 char zero = 0;
318
319 if (!bpf_verifier_log_needed(log))
320 return;
321
322 log->len_used = new_pos;
323 if (put_user(zero, log->ubuf + new_pos))
324 log->ubuf = NULL;
325}
326
Jiri Olsaabe08842018-03-23 11:41:28 +0100327/* log_level controls verbosity level of eBPF verifier.
328 * bpf_verifier_log_write() is used to dump the verification trace to the log,
329 * so the user can figure out what's wrong with the program
Quentin Monnet430e68d2018-01-10 12:26:06 +0000330 */
Jiri Olsaabe08842018-03-23 11:41:28 +0100331__printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
332 const char *fmt, ...)
333{
334 va_list args;
335
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700336 if (!bpf_verifier_log_needed(&env->log))
337 return;
338
Jiri Olsaabe08842018-03-23 11:41:28 +0100339 va_start(args, fmt);
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700340 bpf_verifier_vlog(&env->log, fmt, args);
Jiri Olsaabe08842018-03-23 11:41:28 +0100341 va_end(args);
342}
343EXPORT_SYMBOL_GPL(bpf_verifier_log_write);
344
345__printf(2, 3) static void verbose(void *private_data, const char *fmt, ...)
346{
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700347 struct bpf_verifier_env *env = private_data;
Jiri Olsaabe08842018-03-23 11:41:28 +0100348 va_list args;
349
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700350 if (!bpf_verifier_log_needed(&env->log))
351 return;
352
Jiri Olsaabe08842018-03-23 11:41:28 +0100353 va_start(args, fmt);
Martin KaFai Lau77d2e052018-03-24 11:44:23 -0700354 bpf_verifier_vlog(&env->log, fmt, args);
Jiri Olsaabe08842018-03-23 11:41:28 +0100355 va_end(args);
356}
Alexei Starovoitovcbd35702014-09-26 00:17:03 -0700357
Alexei Starovoitov9e15db62019-10-15 20:25:00 -0700358__printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
359 const char *fmt, ...)
360{
361 va_list args;
362
363 if (!bpf_verifier_log_needed(log))
364 return;
365
366 va_start(args, fmt);
367 bpf_verifier_vlog(log, fmt, args);
368 va_end(args);
369}
370
Martin KaFai Laud9762e82018-12-13 10:41:48 -0800371static const char *ltrim(const char *s)
372{
373 while (isspace(*s))
374 s++;
375
376 return s;
377}
378
379__printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env,
380 u32 insn_off,
381 const char *prefix_fmt, ...)
382{
383 const struct bpf_line_info *linfo;
384
385 if (!bpf_verifier_log_needed(&env->log))
386 return;
387
388 linfo = find_linfo(env, insn_off);
389 if (!linfo || linfo == env->prev_linfo)
390 return;
391
392 if (prefix_fmt) {
393 va_list args;
394
395 va_start(args, prefix_fmt);
396 bpf_verifier_vlog(&env->log, prefix_fmt, args);
397 va_end(args);
398 }
399
400 verbose(env, "%s\n",
401 ltrim(btf_name_by_offset(env->prog->aux->btf,
402 linfo->line_off)));
403
404 env->prev_linfo = linfo;
405}
406
Yonghong Songbc2591d2021-02-26 12:49:22 -0800407static void verbose_invalid_scalar(struct bpf_verifier_env *env,
408 struct bpf_reg_state *reg,
409 struct tnum *range, const char *ctx,
410 const char *reg_name)
411{
412 char tn_buf[48];
413
414 verbose(env, "At %s the register %s ", ctx, reg_name);
415 if (!tnum_is_unknown(reg->var_off)) {
416 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
417 verbose(env, "has value %s", tn_buf);
418 } else {
419 verbose(env, "has unknown scalar value");
420 }
421 tnum_strn(tn_buf, sizeof(tn_buf), *range);
422 verbose(env, " should have been in %s\n", tn_buf);
423}
424
Daniel Borkmannde8f3a82017-09-25 02:25:51 +0200425static bool type_is_pkt_pointer(enum bpf_reg_type type)
426{
427 return type == PTR_TO_PACKET ||
428 type == PTR_TO_PACKET_META;
429}
430
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800431static bool type_is_sk_pointer(enum bpf_reg_type type)
432{
433 return type == PTR_TO_SOCKET ||
Martin KaFai Lau655a51e2019-02-09 23:22:24 -0800434 type == PTR_TO_SOCK_COMMON ||
Jonathan Lemonfada7fd2019-06-06 13:59:40 -0700435 type == PTR_TO_TCP_SOCK ||
436 type == PTR_TO_XDP_SOCK;
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800437}
438
John Fastabendcac616d2020-05-21 13:07:26 -0700439static bool reg_type_not_null(enum bpf_reg_type type)
440{
441 return type == PTR_TO_SOCKET ||
442 type == PTR_TO_TCP_SOCK ||
443 type == PTR_TO_MAP_VALUE ||
Yonghong Song69c087b2021-02-26 12:49:25 -0800444 type == PTR_TO_MAP_KEY ||
Yonghong Song01c66c42020-06-30 10:12:40 -0700445 type == PTR_TO_SOCK_COMMON;
John Fastabendcac616d2020-05-21 13:07:26 -0700446}
447
Joe Stringer840b9612018-10-02 13:35:32 -0700448static bool reg_type_may_be_null(enum bpf_reg_type type)
449{
Joe Stringerfd978bf72018-10-02 13:35:35 -0700450 return type == PTR_TO_MAP_VALUE_OR_NULL ||
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800451 type == PTR_TO_SOCKET_OR_NULL ||
Martin KaFai Lau655a51e2019-02-09 23:22:24 -0800452 type == PTR_TO_SOCK_COMMON_OR_NULL ||
Yonghong Songb121b342020-05-09 10:59:12 -0700453 type == PTR_TO_TCP_SOCK_OR_NULL ||
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700454 type == PTR_TO_BTF_ID_OR_NULL ||
Yonghong Songafbf21d2020-07-23 11:41:11 -0700455 type == PTR_TO_MEM_OR_NULL ||
456 type == PTR_TO_RDONLY_BUF_OR_NULL ||
457 type == PTR_TO_RDWR_BUF_OR_NULL;
Joe Stringerfd978bf72018-10-02 13:35:35 -0700458}
459
Alexei Starovoitovd83525c2019-01-31 15:40:04 -0800460static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg)
461{
462 return reg->type == PTR_TO_MAP_VALUE &&
463 map_value_has_spin_lock(reg->map_ptr);
464}
465
Martin KaFai Laucba368c2019-03-18 10:37:13 -0700466static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type)
467{
468 return type == PTR_TO_SOCKET ||
469 type == PTR_TO_SOCKET_OR_NULL ||
470 type == PTR_TO_TCP_SOCK ||
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700471 type == PTR_TO_TCP_SOCK_OR_NULL ||
472 type == PTR_TO_MEM ||
473 type == PTR_TO_MEM_OR_NULL;
Martin KaFai Laucba368c2019-03-18 10:37:13 -0700474}
475
Martin KaFai Lau1b986582019-03-12 10:23:02 -0700476static bool arg_type_may_be_refcounted(enum bpf_arg_type type)
Joe Stringerfd978bf72018-10-02 13:35:35 -0700477{
Martin KaFai Lau1b986582019-03-12 10:23:02 -0700478 return type == ARG_PTR_TO_SOCK_COMMON;
Joe Stringerfd978bf72018-10-02 13:35:35 -0700479}
480
Lorenz Bauerfd1b0d62020-09-21 13:12:26 +0100481static bool arg_type_may_be_null(enum bpf_arg_type type)
482{
483 return type == ARG_PTR_TO_MAP_VALUE_OR_NULL ||
484 type == ARG_PTR_TO_MEM_OR_NULL ||
485 type == ARG_PTR_TO_CTX_OR_NULL ||
486 type == ARG_PTR_TO_SOCKET_OR_NULL ||
Yonghong Song69c087b2021-02-26 12:49:25 -0800487 type == ARG_PTR_TO_ALLOC_MEM_OR_NULL ||
488 type == ARG_PTR_TO_STACK_OR_NULL;
Lorenz Bauerfd1b0d62020-09-21 13:12:26 +0100489}
490
Joe Stringerfd978bf72018-10-02 13:35:35 -0700491/* Determine whether the function releases some resources allocated by another
492 * function call. The first reference type argument will be assumed to be
493 * released by release_reference().
494 */
495static bool is_release_function(enum bpf_func_id func_id)
496{
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700497 return func_id == BPF_FUNC_sk_release ||
498 func_id == BPF_FUNC_ringbuf_submit ||
499 func_id == BPF_FUNC_ringbuf_discard;
Joe Stringer840b9612018-10-02 13:35:32 -0700500}
501
Jakub Sitnicki64d85292020-04-29 20:11:52 +0200502static bool may_be_acquire_function(enum bpf_func_id func_id)
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800503{
504 return func_id == BPF_FUNC_sk_lookup_tcp ||
Lorenz Baueredbf8c02019-03-22 09:54:01 +0800505 func_id == BPF_FUNC_sk_lookup_udp ||
Jakub Sitnicki64d85292020-04-29 20:11:52 +0200506 func_id == BPF_FUNC_skc_lookup_tcp ||
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700507 func_id == BPF_FUNC_map_lookup_elem ||
508 func_id == BPF_FUNC_ringbuf_reserve;
Jakub Sitnicki64d85292020-04-29 20:11:52 +0200509}
510
511static bool is_acquire_function(enum bpf_func_id func_id,
512 const struct bpf_map *map)
513{
514 enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC;
515
516 if (func_id == BPF_FUNC_sk_lookup_tcp ||
517 func_id == BPF_FUNC_sk_lookup_udp ||
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700518 func_id == BPF_FUNC_skc_lookup_tcp ||
519 func_id == BPF_FUNC_ringbuf_reserve)
Jakub Sitnicki64d85292020-04-29 20:11:52 +0200520 return true;
521
522 if (func_id == BPF_FUNC_map_lookup_elem &&
523 (map_type == BPF_MAP_TYPE_SOCKMAP ||
524 map_type == BPF_MAP_TYPE_SOCKHASH))
525 return true;
526
527 return false;
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800528}
529
Martin KaFai Lau1b986582019-03-12 10:23:02 -0700530static bool is_ptr_cast_function(enum bpf_func_id func_id)
531{
532 return func_id == BPF_FUNC_tcp_sock ||
Martin KaFai Lau1df8f552020-09-24 17:03:50 -0700533 func_id == BPF_FUNC_sk_fullsock ||
534 func_id == BPF_FUNC_skc_to_tcp_sock ||
535 func_id == BPF_FUNC_skc_to_tcp6_sock ||
536 func_id == BPF_FUNC_skc_to_udp6_sock ||
537 func_id == BPF_FUNC_skc_to_tcp_timewait_sock ||
538 func_id == BPF_FUNC_skc_to_tcp_request_sock;
Martin KaFai Lau1b986582019-03-12 10:23:02 -0700539}
540
Brendan Jackman39491862021-03-04 18:56:46 -0800541static bool is_cmpxchg_insn(const struct bpf_insn *insn)
542{
543 return BPF_CLASS(insn->code) == BPF_STX &&
544 BPF_MODE(insn->code) == BPF_ATOMIC &&
545 insn->imm == BPF_CMPXCHG;
546}
547
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700548/* string representation of 'enum bpf_reg_type' */
549static const char * const reg_type_str[] = {
550 [NOT_INIT] = "?",
Edward Creef1174f72017-08-07 15:26:19 +0100551 [SCALAR_VALUE] = "inv",
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700552 [PTR_TO_CTX] = "ctx",
553 [CONST_PTR_TO_MAP] = "map_ptr",
554 [PTR_TO_MAP_VALUE] = "map_value",
555 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700556 [PTR_TO_STACK] = "fp",
Alexei Starovoitov969bf052016-05-05 19:49:10 -0700557 [PTR_TO_PACKET] = "pkt",
Daniel Borkmannde8f3a82017-09-25 02:25:51 +0200558 [PTR_TO_PACKET_META] = "pkt_meta",
Alexei Starovoitov969bf052016-05-05 19:49:10 -0700559 [PTR_TO_PACKET_END] = "pkt_end",
Petar Penkovd58e4682018-09-14 07:46:18 -0700560 [PTR_TO_FLOW_KEYS] = "flow_keys",
Joe Stringerc64b7982018-10-02 13:35:33 -0700561 [PTR_TO_SOCKET] = "sock",
562 [PTR_TO_SOCKET_OR_NULL] = "sock_or_null",
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800563 [PTR_TO_SOCK_COMMON] = "sock_common",
564 [PTR_TO_SOCK_COMMON_OR_NULL] = "sock_common_or_null",
Martin KaFai Lau655a51e2019-02-09 23:22:24 -0800565 [PTR_TO_TCP_SOCK] = "tcp_sock",
566 [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null",
Matt Mullins9df1c282019-04-26 11:49:47 -0700567 [PTR_TO_TP_BUFFER] = "tp_buffer",
Jonathan Lemonfada7fd2019-06-06 13:59:40 -0700568 [PTR_TO_XDP_SOCK] = "xdp_sock",
Alexei Starovoitov9e15db62019-10-15 20:25:00 -0700569 [PTR_TO_BTF_ID] = "ptr_",
Yonghong Songb121b342020-05-09 10:59:12 -0700570 [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_",
Hao Luoeaa6bcb2020-09-29 16:50:47 -0700571 [PTR_TO_PERCPU_BTF_ID] = "percpu_ptr_",
Andrii Nakryiko457f4432020-05-29 00:54:20 -0700572 [PTR_TO_MEM] = "mem",
573 [PTR_TO_MEM_OR_NULL] = "mem_or_null",
Yonghong Songafbf21d2020-07-23 11:41:11 -0700574 [PTR_TO_RDONLY_BUF] = "rdonly_buf",
575 [PTR_TO_RDONLY_BUF_OR_NULL] = "rdonly_buf_or_null",
576 [PTR_TO_RDWR_BUF] = "rdwr_buf",
577 [PTR_TO_RDWR_BUF_OR_NULL] = "rdwr_buf_or_null",
Yonghong Song69c087b2021-02-26 12:49:25 -0800578 [PTR_TO_FUNC] = "func",
579 [PTR_TO_MAP_KEY] = "map_key",
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700580};
581
Edward Cree8efea212018-08-22 20:02:44 +0100582static char slot_type_char[] = {
583 [STACK_INVALID] = '?',
584 [STACK_SPILL] = 'r',
585 [STACK_MISC] = 'm',
586 [STACK_ZERO] = '0',
587};
588
Alexei Starovoitov4e920242017-11-30 21:31:36 -0800589static void print_liveness(struct bpf_verifier_env *env,
590 enum bpf_reg_liveness live)
591{
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -0800592 if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE))
Alexei Starovoitov4e920242017-11-30 21:31:36 -0800593 verbose(env, "_");
594 if (live & REG_LIVE_READ)
595 verbose(env, "r");
596 if (live & REG_LIVE_WRITTEN)
597 verbose(env, "w");
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -0800598 if (live & REG_LIVE_DONE)
599 verbose(env, "D");
Alexei Starovoitov4e920242017-11-30 21:31:36 -0800600}
601
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800602static struct bpf_func_state *func(struct bpf_verifier_env *env,
603 const struct bpf_reg_state *reg)
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700604{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800605 struct bpf_verifier_state *cur = env->cur_state;
606
607 return cur->frame[reg->frameno];
608}
609
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -0800610static const char *kernel_type_name(const struct btf* btf, u32 id)
Alexei Starovoitov9e15db62019-10-15 20:25:00 -0700611{
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -0800612 return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off);
Alexei Starovoitov9e15db62019-10-15 20:25:00 -0700613}
614
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800615static void print_verifier_state(struct bpf_verifier_env *env,
616 const struct bpf_func_state *state)
617{
618 const struct bpf_reg_state *reg;
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700619 enum bpf_reg_type t;
620 int i;
621
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800622 if (state->frameno)
623 verbose(env, " frame%d:", state->frameno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700624 for (i = 0; i < MAX_BPF_REG; i++) {
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -0700625 reg = &state->regs[i];
626 t = reg->type;
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700627 if (t == NOT_INIT)
628 continue;
Alexei Starovoitov4e920242017-11-30 21:31:36 -0800629 verbose(env, " R%d", i);
630 print_liveness(env, reg->live);
631 verbose(env, "=%s", reg_type_str[t]);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700632 if (t == SCALAR_VALUE && reg->precise)
633 verbose(env, "P");
Edward Creef1174f72017-08-07 15:26:19 +0100634 if ((t == SCALAR_VALUE || t == PTR_TO_STACK) &&
635 tnum_is_const(reg->var_off)) {
636 /* reg->off should be 0 for SCALAR_VALUE */
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700637 verbose(env, "%lld", reg->var_off.value + reg->off);
Edward Creef1174f72017-08-07 15:26:19 +0100638 } else {
Hao Luoeaa6bcb2020-09-29 16:50:47 -0700639 if (t == PTR_TO_BTF_ID ||
640 t == PTR_TO_BTF_ID_OR_NULL ||
641 t == PTR_TO_PERCPU_BTF_ID)
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -0800642 verbose(env, "%s", kernel_type_name(reg->btf, reg->btf_id));
Martin KaFai Laucba368c2019-03-18 10:37:13 -0700643 verbose(env, "(id=%d", reg->id);
644 if (reg_type_may_be_refcounted_or_null(t))
645 verbose(env, ",ref_obj_id=%d", reg->ref_obj_id);
Edward Creef1174f72017-08-07 15:26:19 +0100646 if (t != SCALAR_VALUE)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700647 verbose(env, ",off=%d", reg->off);
Daniel Borkmannde8f3a82017-09-25 02:25:51 +0200648 if (type_is_pkt_pointer(t))
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700649 verbose(env, ",r=%d", reg->range);
Edward Creef1174f72017-08-07 15:26:19 +0100650 else if (t == CONST_PTR_TO_MAP ||
Yonghong Song69c087b2021-02-26 12:49:25 -0800651 t == PTR_TO_MAP_KEY ||
Edward Creef1174f72017-08-07 15:26:19 +0100652 t == PTR_TO_MAP_VALUE ||
653 t == PTR_TO_MAP_VALUE_OR_NULL)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700654 verbose(env, ",ks=%d,vs=%d",
Edward Creef1174f72017-08-07 15:26:19 +0100655 reg->map_ptr->key_size,
656 reg->map_ptr->value_size);
Edward Cree7d1238f2017-08-07 15:26:56 +0100657 if (tnum_is_const(reg->var_off)) {
658 /* Typically an immediate SCALAR_VALUE, but
659 * could be a pointer whose offset is too big
660 * for reg->off
661 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700662 verbose(env, ",imm=%llx", reg->var_off.value);
Edward Cree7d1238f2017-08-07 15:26:56 +0100663 } else {
664 if (reg->smin_value != reg->umin_value &&
665 reg->smin_value != S64_MIN)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700666 verbose(env, ",smin_value=%lld",
Edward Cree7d1238f2017-08-07 15:26:56 +0100667 (long long)reg->smin_value);
668 if (reg->smax_value != reg->umax_value &&
669 reg->smax_value != S64_MAX)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700670 verbose(env, ",smax_value=%lld",
Edward Cree7d1238f2017-08-07 15:26:56 +0100671 (long long)reg->smax_value);
672 if (reg->umin_value != 0)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700673 verbose(env, ",umin_value=%llu",
Edward Cree7d1238f2017-08-07 15:26:56 +0100674 (unsigned long long)reg->umin_value);
675 if (reg->umax_value != U64_MAX)
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700676 verbose(env, ",umax_value=%llu",
Edward Cree7d1238f2017-08-07 15:26:56 +0100677 (unsigned long long)reg->umax_value);
678 if (!tnum_is_unknown(reg->var_off)) {
679 char tn_buf[48];
Edward Creef1174f72017-08-07 15:26:19 +0100680
Edward Cree7d1238f2017-08-07 15:26:56 +0100681 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700682 verbose(env, ",var_off=%s", tn_buf);
Edward Cree7d1238f2017-08-07 15:26:56 +0100683 }
John Fastabend3f50f132020-03-30 14:36:39 -0700684 if (reg->s32_min_value != reg->smin_value &&
685 reg->s32_min_value != S32_MIN)
686 verbose(env, ",s32_min_value=%d",
687 (int)(reg->s32_min_value));
688 if (reg->s32_max_value != reg->smax_value &&
689 reg->s32_max_value != S32_MAX)
690 verbose(env, ",s32_max_value=%d",
691 (int)(reg->s32_max_value));
692 if (reg->u32_min_value != reg->umin_value &&
693 reg->u32_min_value != U32_MIN)
694 verbose(env, ",u32_min_value=%d",
695 (int)(reg->u32_min_value));
696 if (reg->u32_max_value != reg->umax_value &&
697 reg->u32_max_value != U32_MAX)
698 verbose(env, ",u32_max_value=%d",
699 (int)(reg->u32_max_value));
Edward Creef1174f72017-08-07 15:26:19 +0100700 }
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700701 verbose(env, ")");
Edward Creef1174f72017-08-07 15:26:19 +0100702 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700703 }
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700704 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
Edward Cree8efea212018-08-22 20:02:44 +0100705 char types_buf[BPF_REG_SIZE + 1];
706 bool valid = false;
707 int j;
708
709 for (j = 0; j < BPF_REG_SIZE; j++) {
710 if (state->stack[i].slot_type[j] != STACK_INVALID)
711 valid = true;
712 types_buf[j] = slot_type_char[
713 state->stack[i].slot_type[j]];
714 }
715 types_buf[BPF_REG_SIZE] = 0;
716 if (!valid)
717 continue;
718 verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE);
719 print_liveness(env, state->stack[i].spilled_ptr.live);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700720 if (state->stack[i].slot_type[0] == STACK_SPILL) {
721 reg = &state->stack[i].spilled_ptr;
722 t = reg->type;
723 verbose(env, "=%s", reg_type_str[t]);
724 if (t == SCALAR_VALUE && reg->precise)
725 verbose(env, "P");
726 if (t == SCALAR_VALUE && tnum_is_const(reg->var_off))
727 verbose(env, "%lld", reg->var_off.value + reg->off);
728 } else {
Edward Cree8efea212018-08-22 20:02:44 +0100729 verbose(env, "=%s", types_buf);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700730 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700731 }
Joe Stringerfd978bf72018-10-02 13:35:35 -0700732 if (state->acquired_refs && state->refs[0].id) {
733 verbose(env, " refs=%d", state->refs[0].id);
734 for (i = 1; i < state->acquired_refs; i++)
735 if (state->refs[i].id)
736 verbose(env, ",%d", state->refs[i].id);
737 }
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -0700738 if (state->in_callback_fn)
739 verbose(env, " cb");
740 if (state->in_async_callback_fn)
741 verbose(env, " async_cb");
Jakub Kicinski61bd5212017-10-09 10:30:11 -0700742 verbose(env, "\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700743}
744
Lorenz Bauerc69431a2021-04-29 14:46:54 +0100745/* copy array src of length n * size bytes to dst. dst is reallocated if it's too
746 * small to hold src. This is different from krealloc since we don't want to preserve
747 * the contents of dst.
748 *
749 * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could
750 * not be allocated.
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700751 */
Lorenz Bauerc69431a2021-04-29 14:46:54 +0100752static void *copy_array(void *dst, const void *src, size_t n, size_t size, gfp_t flags)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700753{
Lorenz Bauerc69431a2021-04-29 14:46:54 +0100754 size_t bytes;
755
756 if (ZERO_OR_NULL_PTR(src))
757 goto out;
758
759 if (unlikely(check_mul_overflow(n, size, &bytes)))
760 return NULL;
761
762 if (ksize(dst) < bytes) {
763 kfree(dst);
764 dst = kmalloc_track_caller(bytes, flags);
765 if (!dst)
766 return NULL;
767 }
768
769 memcpy(dst, src, bytes);
770out:
771 return dst ? dst : ZERO_SIZE_PTR;
772}
773
774/* resize an array from old_n items to new_n items. the array is reallocated if it's too
775 * small to hold new_n items. new items are zeroed out if the array grows.
776 *
777 * Contrary to krealloc_array, does not free arr if new_n is zero.
778 */
779static void *realloc_array(void *arr, size_t old_n, size_t new_n, size_t size)
780{
781 if (!new_n || old_n == new_n)
782 goto out;
783
784 arr = krealloc_array(arr, new_n, size, GFP_KERNEL);
785 if (!arr)
786 return NULL;
787
788 if (new_n > old_n)
789 memset(arr + old_n * size, 0, (new_n - old_n) * size);
790
791out:
792 return arr ? arr : ZERO_SIZE_PTR;
793}
794
795static int copy_reference_state(struct bpf_func_state *dst, const struct bpf_func_state *src)
796{
797 dst->refs = copy_array(dst->refs, src->refs, src->acquired_refs,
798 sizeof(struct bpf_reference_state), GFP_KERNEL);
799 if (!dst->refs)
800 return -ENOMEM;
801
802 dst->acquired_refs = src->acquired_refs;
803 return 0;
804}
805
806static int copy_stack_state(struct bpf_func_state *dst, const struct bpf_func_state *src)
807{
808 size_t n = src->allocated_stack / BPF_REG_SIZE;
809
810 dst->stack = copy_array(dst->stack, src->stack, n, sizeof(struct bpf_stack_state),
811 GFP_KERNEL);
812 if (!dst->stack)
813 return -ENOMEM;
814
815 dst->allocated_stack = src->allocated_stack;
816 return 0;
817}
818
819static int resize_reference_state(struct bpf_func_state *state, size_t n)
820{
821 state->refs = realloc_array(state->refs, state->acquired_refs, n,
822 sizeof(struct bpf_reference_state));
823 if (!state->refs)
824 return -ENOMEM;
825
826 state->acquired_refs = n;
827 return 0;
828}
829
830static int grow_stack_state(struct bpf_func_state *state, int size)
831{
832 size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE;
833
834 if (old_n >= n)
835 return 0;
836
837 state->stack = realloc_array(state->stack, old_n, n, sizeof(struct bpf_stack_state));
838 if (!state->stack)
839 return -ENOMEM;
840
841 state->allocated_stack = size;
842 return 0;
Joe Stringerfd978bf72018-10-02 13:35:35 -0700843}
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700844
Joe Stringerfd978bf72018-10-02 13:35:35 -0700845/* Acquire a pointer id from the env and update the state->refs to include
846 * this new pointer reference.
847 * On success, returns a valid pointer id to associate with the register
848 * On failure, returns a negative errno.
849 */
850static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx)
851{
852 struct bpf_func_state *state = cur_func(env);
853 int new_ofs = state->acquired_refs;
854 int id, err;
855
Lorenz Bauerc69431a2021-04-29 14:46:54 +0100856 err = resize_reference_state(state, state->acquired_refs + 1);
Joe Stringerfd978bf72018-10-02 13:35:35 -0700857 if (err)
858 return err;
859 id = ++env->id_gen;
860 state->refs[new_ofs].id = id;
861 state->refs[new_ofs].insn_idx = insn_idx;
862
863 return id;
864}
865
866/* release function corresponding to acquire_reference_state(). Idempotent. */
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800867static int release_reference_state(struct bpf_func_state *state, int ptr_id)
Joe Stringerfd978bf72018-10-02 13:35:35 -0700868{
869 int i, last_idx;
870
Joe Stringerfd978bf72018-10-02 13:35:35 -0700871 last_idx = state->acquired_refs - 1;
872 for (i = 0; i < state->acquired_refs; i++) {
873 if (state->refs[i].id == ptr_id) {
874 if (last_idx && i != last_idx)
875 memcpy(&state->refs[i], &state->refs[last_idx],
876 sizeof(*state->refs));
877 memset(&state->refs[last_idx], 0, sizeof(*state->refs));
878 state->acquired_refs--;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700879 return 0;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700880 }
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700881 }
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -0800882 return -EINVAL;
Joe Stringerfd978bf72018-10-02 13:35:35 -0700883}
884
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800885static void free_func_state(struct bpf_func_state *state)
886{
Alexei Starovoitov58963512018-01-08 07:51:17 -0800887 if (!state)
888 return;
Joe Stringerfd978bf72018-10-02 13:35:35 -0700889 kfree(state->refs);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800890 kfree(state->stack);
891 kfree(state);
892}
893
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700894static void clear_jmp_history(struct bpf_verifier_state *state)
895{
896 kfree(state->jmp_history);
897 state->jmp_history = NULL;
898 state->jmp_history_cnt = 0;
899}
900
Alexei Starovoitov1969db42017-11-01 00:08:04 -0700901static void free_verifier_state(struct bpf_verifier_state *state,
902 bool free_self)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700903{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800904 int i;
905
906 for (i = 0; i <= state->curframe; i++) {
907 free_func_state(state->frame[i]);
908 state->frame[i] = NULL;
909 }
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700910 clear_jmp_history(state);
Alexei Starovoitov1969db42017-11-01 00:08:04 -0700911 if (free_self)
912 kfree(state);
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700913}
914
915/* copy verifier state from src to dst growing dst stack space
916 * when necessary to accommodate larger src stack
917 */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800918static int copy_func_state(struct bpf_func_state *dst,
919 const struct bpf_func_state *src)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700920{
921 int err;
922
Joe Stringerfd978bf72018-10-02 13:35:35 -0700923 memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs));
924 err = copy_reference_state(dst, src);
925 if (err)
926 return err;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700927 return copy_stack_state(dst, src);
928}
929
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800930static int copy_verifier_state(struct bpf_verifier_state *dst_state,
931 const struct bpf_verifier_state *src)
932{
933 struct bpf_func_state *dst;
934 int i, err;
935
Lorenz Bauer06ab6a52021-04-29 14:46:55 +0100936 dst_state->jmp_history = copy_array(dst_state->jmp_history, src->jmp_history,
937 src->jmp_history_cnt, sizeof(struct bpf_idx_pair),
938 GFP_USER);
939 if (!dst_state->jmp_history)
940 return -ENOMEM;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700941 dst_state->jmp_history_cnt = src->jmp_history_cnt;
942
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800943 /* if dst has more stack frames then src frame, free them */
944 for (i = src->curframe + 1; i <= dst_state->curframe; i++) {
945 free_func_state(dst_state->frame[i]);
946 dst_state->frame[i] = NULL;
947 }
Daniel Borkmann979d63d2019-01-03 00:58:34 +0100948 dst_state->speculative = src->speculative;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800949 dst_state->curframe = src->curframe;
Alexei Starovoitovd83525c2019-01-31 15:40:04 -0800950 dst_state->active_spin_lock = src->active_spin_lock;
Alexei Starovoitov25897262019-06-15 12:12:20 -0700951 dst_state->branches = src->branches;
952 dst_state->parent = src->parent;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -0700953 dst_state->first_insn_idx = src->first_insn_idx;
954 dst_state->last_insn_idx = src->last_insn_idx;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -0800955 for (i = 0; i <= src->curframe; i++) {
956 dst = dst_state->frame[i];
957 if (!dst) {
958 dst = kzalloc(sizeof(*dst), GFP_KERNEL);
959 if (!dst)
960 return -ENOMEM;
961 dst_state->frame[i] = dst;
962 }
963 err = copy_func_state(dst, src->frame[i]);
964 if (err)
965 return err;
966 }
967 return 0;
968}
969
Alexei Starovoitov25897262019-06-15 12:12:20 -0700970static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st)
971{
972 while (st) {
973 u32 br = --st->branches;
974
975 /* WARN_ON(br > 1) technically makes sense here,
976 * but see comment in push_stack(), hence:
977 */
978 WARN_ONCE((int)br < 0,
979 "BUG update_branch_counts:branches_to_explore=%d\n",
980 br);
981 if (br)
982 break;
983 st = st->parent;
984 }
985}
986
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700987static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx,
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -0700988 int *insn_idx, bool pop_log)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700989{
990 struct bpf_verifier_state *cur = env->cur_state;
991 struct bpf_verifier_stack_elem *elem, *head = env->head;
992 int err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700993
994 if (env->head == NULL)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700995 return -ENOENT;
Alexei Starovoitov17a52672014-09-26 00:17:06 -0700996
Alexei Starovoitov638f5b92017-10-31 18:16:05 -0700997 if (cur) {
998 err = copy_verifier_state(cur, &head->st);
999 if (err)
1000 return err;
1001 }
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -07001002 if (pop_log)
1003 bpf_vlog_reset(&env->log, head->log_pos);
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001004 if (insn_idx)
1005 *insn_idx = head->insn_idx;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001006 if (prev_insn_idx)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001007 *prev_insn_idx = head->prev_insn_idx;
1008 elem = head->next;
Alexei Starovoitov1969db42017-11-01 00:08:04 -07001009 free_verifier_state(&head->st, false);
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001010 kfree(head);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001011 env->head = elem;
1012 env->stack_size--;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001013 return 0;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001014}
1015
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01001016static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
Daniel Borkmann979d63d2019-01-03 00:58:34 +01001017 int insn_idx, int prev_insn_idx,
1018 bool speculative)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001019{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001020 struct bpf_verifier_state *cur = env->cur_state;
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01001021 struct bpf_verifier_stack_elem *elem;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001022 int err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001023
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07001024 elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001025 if (!elem)
1026 goto err;
1027
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001028 elem->insn_idx = insn_idx;
1029 elem->prev_insn_idx = prev_insn_idx;
1030 elem->next = env->head;
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -07001031 elem->log_pos = env->log.len_used;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001032 env->head = elem;
1033 env->stack_size++;
Alexei Starovoitov1969db42017-11-01 00:08:04 -07001034 err = copy_verifier_state(&elem->st, cur);
1035 if (err)
1036 goto err;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01001037 elem->st.speculative |= speculative;
Alexei Starovoitovb285fcb2019-05-21 20:14:19 -07001038 if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) {
1039 verbose(env, "The sequence of %d jumps is too complex.\n",
1040 env->stack_size);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001041 goto err;
1042 }
Alexei Starovoitov25897262019-06-15 12:12:20 -07001043 if (elem->st.parent) {
1044 ++elem->st.parent->branches;
1045 /* WARN_ON(branches > 2) technically makes sense here,
1046 * but
1047 * 1. speculative states will bump 'branches' for non-branch
1048 * instructions
1049 * 2. is_state_visited() heuristics may decide not to create
1050 * a new state for a sequence of branches and all such current
1051 * and cloned states will be pointing to a single parent state
1052 * which might have large 'branches' count.
1053 */
1054 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001055 return &elem->st;
1056err:
Alexei Starovoitov58963512018-01-08 07:51:17 -08001057 free_verifier_state(env->cur_state, true);
1058 env->cur_state = NULL;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001059 /* pop all elements and return */
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -07001060 while (!pop_stack(env, NULL, NULL, false));
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001061 return NULL;
1062}
1063
1064#define CALLER_SAVED_REGS 6
1065static const int caller_saved[CALLER_SAVED_REGS] = {
1066 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
1067};
1068
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001069static void __mark_reg_not_init(const struct bpf_verifier_env *env,
1070 struct bpf_reg_state *reg);
Edward Creef1174f72017-08-07 15:26:19 +01001071
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07001072/* This helper doesn't clear reg->id */
1073static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm)
Edward Creeb03c9f92017-08-07 15:26:36 +01001074{
Edward Creeb03c9f92017-08-07 15:26:36 +01001075 reg->var_off = tnum_const(imm);
1076 reg->smin_value = (s64)imm;
1077 reg->smax_value = (s64)imm;
1078 reg->umin_value = imm;
1079 reg->umax_value = imm;
John Fastabend3f50f132020-03-30 14:36:39 -07001080
1081 reg->s32_min_value = (s32)imm;
1082 reg->s32_max_value = (s32)imm;
1083 reg->u32_min_value = (u32)imm;
1084 reg->u32_max_value = (u32)imm;
1085}
1086
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07001087/* Mark the unknown part of a register (variable offset or scalar value) as
1088 * known to have the value @imm.
1089 */
1090static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm)
1091{
1092 /* Clear id, off, and union(map_ptr, range) */
1093 memset(((u8 *)reg) + sizeof(reg->type), 0,
1094 offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type));
1095 ___mark_reg_known(reg, imm);
1096}
1097
John Fastabend3f50f132020-03-30 14:36:39 -07001098static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm)
1099{
1100 reg->var_off = tnum_const_subreg(reg->var_off, imm);
1101 reg->s32_min_value = (s32)imm;
1102 reg->s32_max_value = (s32)imm;
1103 reg->u32_min_value = (u32)imm;
1104 reg->u32_max_value = (u32)imm;
Edward Creeb03c9f92017-08-07 15:26:36 +01001105}
1106
Edward Creef1174f72017-08-07 15:26:19 +01001107/* Mark the 'variable offset' part of a register as zero. This should be
1108 * used only on registers holding a pointer type.
1109 */
1110static void __mark_reg_known_zero(struct bpf_reg_state *reg)
1111{
Edward Creeb03c9f92017-08-07 15:26:36 +01001112 __mark_reg_known(reg, 0);
Edward Creef1174f72017-08-07 15:26:19 +01001113}
1114
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08001115static void __mark_reg_const_zero(struct bpf_reg_state *reg)
1116{
1117 __mark_reg_known(reg, 0);
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08001118 reg->type = SCALAR_VALUE;
1119}
1120
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001121static void mark_reg_known_zero(struct bpf_verifier_env *env,
1122 struct bpf_reg_state *regs, u32 regno)
Edward Creef1174f72017-08-07 15:26:19 +01001123{
1124 if (WARN_ON(regno >= MAX_BPF_REG)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001125 verbose(env, "mark_reg_known_zero(regs, %u)\n", regno);
Edward Creef1174f72017-08-07 15:26:19 +01001126 /* Something bad happened, let's kill all regs */
1127 for (regno = 0; regno < MAX_BPF_REG; regno++)
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001128 __mark_reg_not_init(env, regs + regno);
Edward Creef1174f72017-08-07 15:26:19 +01001129 return;
1130 }
1131 __mark_reg_known_zero(regs + regno);
1132}
1133
Dmitrii Banshchikov4ddb7412021-02-13 00:56:40 +04001134static void mark_ptr_not_null_reg(struct bpf_reg_state *reg)
1135{
1136 switch (reg->type) {
1137 case PTR_TO_MAP_VALUE_OR_NULL: {
1138 const struct bpf_map *map = reg->map_ptr;
1139
1140 if (map->inner_map_meta) {
1141 reg->type = CONST_PTR_TO_MAP;
1142 reg->map_ptr = map->inner_map_meta;
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -07001143 /* transfer reg's id which is unique for every map_lookup_elem
1144 * as UID of the inner map.
1145 */
1146 reg->map_uid = reg->id;
Dmitrii Banshchikov4ddb7412021-02-13 00:56:40 +04001147 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
1148 reg->type = PTR_TO_XDP_SOCK;
1149 } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP ||
1150 map->map_type == BPF_MAP_TYPE_SOCKHASH) {
1151 reg->type = PTR_TO_SOCKET;
1152 } else {
1153 reg->type = PTR_TO_MAP_VALUE;
1154 }
1155 break;
1156 }
1157 case PTR_TO_SOCKET_OR_NULL:
1158 reg->type = PTR_TO_SOCKET;
1159 break;
1160 case PTR_TO_SOCK_COMMON_OR_NULL:
1161 reg->type = PTR_TO_SOCK_COMMON;
1162 break;
1163 case PTR_TO_TCP_SOCK_OR_NULL:
1164 reg->type = PTR_TO_TCP_SOCK;
1165 break;
1166 case PTR_TO_BTF_ID_OR_NULL:
1167 reg->type = PTR_TO_BTF_ID;
1168 break;
1169 case PTR_TO_MEM_OR_NULL:
1170 reg->type = PTR_TO_MEM;
1171 break;
1172 case PTR_TO_RDONLY_BUF_OR_NULL:
1173 reg->type = PTR_TO_RDONLY_BUF;
1174 break;
1175 case PTR_TO_RDWR_BUF_OR_NULL:
1176 reg->type = PTR_TO_RDWR_BUF;
1177 break;
1178 default:
Dan Carpenter33ccec52021-02-17 10:45:25 +03001179 WARN_ONCE(1, "unknown nullable register type");
Dmitrii Banshchikov4ddb7412021-02-13 00:56:40 +04001180 }
1181}
1182
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02001183static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg)
1184{
1185 return type_is_pkt_pointer(reg->type);
1186}
1187
1188static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg)
1189{
1190 return reg_is_pkt_pointer(reg) ||
1191 reg->type == PTR_TO_PACKET_END;
1192}
1193
1194/* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
1195static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg,
1196 enum bpf_reg_type which)
1197{
1198 /* The register can already have a range from prior markings.
1199 * This is fine as long as it hasn't been advanced from its
1200 * origin.
1201 */
1202 return reg->type == which &&
1203 reg->id == 0 &&
1204 reg->off == 0 &&
1205 tnum_equals_const(reg->var_off, 0);
1206}
1207
John Fastabend3f50f132020-03-30 14:36:39 -07001208/* Reset the min/max bounds of a register */
1209static void __mark_reg_unbounded(struct bpf_reg_state *reg)
1210{
1211 reg->smin_value = S64_MIN;
1212 reg->smax_value = S64_MAX;
1213 reg->umin_value = 0;
1214 reg->umax_value = U64_MAX;
1215
1216 reg->s32_min_value = S32_MIN;
1217 reg->s32_max_value = S32_MAX;
1218 reg->u32_min_value = 0;
1219 reg->u32_max_value = U32_MAX;
1220}
1221
1222static void __mark_reg64_unbounded(struct bpf_reg_state *reg)
1223{
1224 reg->smin_value = S64_MIN;
1225 reg->smax_value = S64_MAX;
1226 reg->umin_value = 0;
1227 reg->umax_value = U64_MAX;
1228}
1229
1230static void __mark_reg32_unbounded(struct bpf_reg_state *reg)
1231{
1232 reg->s32_min_value = S32_MIN;
1233 reg->s32_max_value = S32_MAX;
1234 reg->u32_min_value = 0;
1235 reg->u32_max_value = U32_MAX;
1236}
1237
1238static void __update_reg32_bounds(struct bpf_reg_state *reg)
1239{
1240 struct tnum var32_off = tnum_subreg(reg->var_off);
1241
1242 /* min signed is max(sign bit) | min(other bits) */
1243 reg->s32_min_value = max_t(s32, reg->s32_min_value,
1244 var32_off.value | (var32_off.mask & S32_MIN));
1245 /* max signed is min(sign bit) | max(other bits) */
1246 reg->s32_max_value = min_t(s32, reg->s32_max_value,
1247 var32_off.value | (var32_off.mask & S32_MAX));
1248 reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value);
1249 reg->u32_max_value = min(reg->u32_max_value,
1250 (u32)(var32_off.value | var32_off.mask));
1251}
1252
1253static void __update_reg64_bounds(struct bpf_reg_state *reg)
Edward Creeb03c9f92017-08-07 15:26:36 +01001254{
1255 /* min signed is max(sign bit) | min(other bits) */
1256 reg->smin_value = max_t(s64, reg->smin_value,
1257 reg->var_off.value | (reg->var_off.mask & S64_MIN));
1258 /* max signed is min(sign bit) | max(other bits) */
1259 reg->smax_value = min_t(s64, reg->smax_value,
1260 reg->var_off.value | (reg->var_off.mask & S64_MAX));
1261 reg->umin_value = max(reg->umin_value, reg->var_off.value);
1262 reg->umax_value = min(reg->umax_value,
1263 reg->var_off.value | reg->var_off.mask);
1264}
1265
John Fastabend3f50f132020-03-30 14:36:39 -07001266static void __update_reg_bounds(struct bpf_reg_state *reg)
1267{
1268 __update_reg32_bounds(reg);
1269 __update_reg64_bounds(reg);
1270}
1271
Edward Creeb03c9f92017-08-07 15:26:36 +01001272/* Uses signed min/max values to inform unsigned, and vice-versa */
John Fastabend3f50f132020-03-30 14:36:39 -07001273static void __reg32_deduce_bounds(struct bpf_reg_state *reg)
1274{
1275 /* Learn sign from signed bounds.
1276 * If we cannot cross the sign boundary, then signed and unsigned bounds
1277 * are the same, so combine. This works even in the negative case, e.g.
1278 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1279 */
1280 if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) {
1281 reg->s32_min_value = reg->u32_min_value =
1282 max_t(u32, reg->s32_min_value, reg->u32_min_value);
1283 reg->s32_max_value = reg->u32_max_value =
1284 min_t(u32, reg->s32_max_value, reg->u32_max_value);
1285 return;
1286 }
1287 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1288 * boundary, so we must be careful.
1289 */
1290 if ((s32)reg->u32_max_value >= 0) {
1291 /* Positive. We can't learn anything from the smin, but smax
1292 * is positive, hence safe.
1293 */
1294 reg->s32_min_value = reg->u32_min_value;
1295 reg->s32_max_value = reg->u32_max_value =
1296 min_t(u32, reg->s32_max_value, reg->u32_max_value);
1297 } else if ((s32)reg->u32_min_value < 0) {
1298 /* Negative. We can't learn anything from the smax, but smin
1299 * is negative, hence safe.
1300 */
1301 reg->s32_min_value = reg->u32_min_value =
1302 max_t(u32, reg->s32_min_value, reg->u32_min_value);
1303 reg->s32_max_value = reg->u32_max_value;
1304 }
1305}
1306
1307static void __reg64_deduce_bounds(struct bpf_reg_state *reg)
Edward Creeb03c9f92017-08-07 15:26:36 +01001308{
1309 /* Learn sign from signed bounds.
1310 * If we cannot cross the sign boundary, then signed and unsigned bounds
1311 * are the same, so combine. This works even in the negative case, e.g.
1312 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1313 */
1314 if (reg->smin_value >= 0 || reg->smax_value < 0) {
1315 reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
1316 reg->umin_value);
1317 reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
1318 reg->umax_value);
1319 return;
1320 }
1321 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1322 * boundary, so we must be careful.
1323 */
1324 if ((s64)reg->umax_value >= 0) {
1325 /* Positive. We can't learn anything from the smin, but smax
1326 * is positive, hence safe.
1327 */
1328 reg->smin_value = reg->umin_value;
1329 reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
1330 reg->umax_value);
1331 } else if ((s64)reg->umin_value < 0) {
1332 /* Negative. We can't learn anything from the smax, but smin
1333 * is negative, hence safe.
1334 */
1335 reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
1336 reg->umin_value);
1337 reg->smax_value = reg->umax_value;
1338 }
1339}
1340
John Fastabend3f50f132020-03-30 14:36:39 -07001341static void __reg_deduce_bounds(struct bpf_reg_state *reg)
1342{
1343 __reg32_deduce_bounds(reg);
1344 __reg64_deduce_bounds(reg);
1345}
1346
Edward Creeb03c9f92017-08-07 15:26:36 +01001347/* Attempts to improve var_off based on unsigned min/max information */
1348static void __reg_bound_offset(struct bpf_reg_state *reg)
1349{
John Fastabend3f50f132020-03-30 14:36:39 -07001350 struct tnum var64_off = tnum_intersect(reg->var_off,
1351 tnum_range(reg->umin_value,
1352 reg->umax_value));
1353 struct tnum var32_off = tnum_intersect(tnum_subreg(reg->var_off),
1354 tnum_range(reg->u32_min_value,
1355 reg->u32_max_value));
1356
1357 reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off);
Edward Creeb03c9f92017-08-07 15:26:36 +01001358}
1359
John Fastabend3f50f132020-03-30 14:36:39 -07001360static void __reg_assign_32_into_64(struct bpf_reg_state *reg)
Edward Creeb03c9f92017-08-07 15:26:36 +01001361{
John Fastabend3f50f132020-03-30 14:36:39 -07001362 reg->umin_value = reg->u32_min_value;
1363 reg->umax_value = reg->u32_max_value;
1364 /* Attempt to pull 32-bit signed bounds into 64-bit bounds
1365 * but must be positive otherwise set to worse case bounds
1366 * and refine later from tnum.
1367 */
John Fastabend3a71dc32020-05-29 10:28:40 -07001368 if (reg->s32_min_value >= 0 && reg->s32_max_value >= 0)
John Fastabend3f50f132020-03-30 14:36:39 -07001369 reg->smax_value = reg->s32_max_value;
1370 else
1371 reg->smax_value = U32_MAX;
John Fastabend3a71dc32020-05-29 10:28:40 -07001372 if (reg->s32_min_value >= 0)
1373 reg->smin_value = reg->s32_min_value;
1374 else
1375 reg->smin_value = 0;
John Fastabend3f50f132020-03-30 14:36:39 -07001376}
1377
1378static void __reg_combine_32_into_64(struct bpf_reg_state *reg)
1379{
1380 /* special case when 64-bit register has upper 32-bit register
1381 * zeroed. Typically happens after zext or <<32, >>32 sequence
1382 * allowing us to use 32-bit bounds directly,
1383 */
1384 if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) {
1385 __reg_assign_32_into_64(reg);
1386 } else {
1387 /* Otherwise the best we can do is push lower 32bit known and
1388 * unknown bits into register (var_off set from jmp logic)
1389 * then learn as much as possible from the 64-bit tnum
1390 * known and unknown bits. The previous smin/smax bounds are
1391 * invalid here because of jmp32 compare so mark them unknown
1392 * so they do not impact tnum bounds calculation.
1393 */
1394 __mark_reg64_unbounded(reg);
1395 __update_reg_bounds(reg);
1396 }
1397
1398 /* Intersecting with the old var_off might have improved our bounds
1399 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1400 * then new var_off is (0; 0x7f...fc) which improves our umax.
1401 */
1402 __reg_deduce_bounds(reg);
1403 __reg_bound_offset(reg);
1404 __update_reg_bounds(reg);
1405}
1406
1407static bool __reg64_bound_s32(s64 a)
1408{
Alexei Starovoitovb02709582020-12-08 19:01:51 +01001409 return a > S32_MIN && a < S32_MAX;
John Fastabend3f50f132020-03-30 14:36:39 -07001410}
1411
1412static bool __reg64_bound_u32(u64 a)
1413{
Daniel Borkmann10bf4e82021-04-23 13:59:55 +00001414 return a > U32_MIN && a < U32_MAX;
John Fastabend3f50f132020-03-30 14:36:39 -07001415}
1416
1417static void __reg_combine_64_into_32(struct bpf_reg_state *reg)
1418{
1419 __mark_reg32_unbounded(reg);
1420
Alexei Starovoitovb02709582020-12-08 19:01:51 +01001421 if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) {
John Fastabend3f50f132020-03-30 14:36:39 -07001422 reg->s32_min_value = (s32)reg->smin_value;
John Fastabend3f50f132020-03-30 14:36:39 -07001423 reg->s32_max_value = (s32)reg->smax_value;
Alexei Starovoitovb02709582020-12-08 19:01:51 +01001424 }
Daniel Borkmann10bf4e82021-04-23 13:59:55 +00001425 if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value)) {
John Fastabend3f50f132020-03-30 14:36:39 -07001426 reg->u32_min_value = (u32)reg->umin_value;
John Fastabend3f50f132020-03-30 14:36:39 -07001427 reg->u32_max_value = (u32)reg->umax_value;
Daniel Borkmann10bf4e82021-04-23 13:59:55 +00001428 }
John Fastabend3f50f132020-03-30 14:36:39 -07001429
1430 /* Intersecting with the old var_off might have improved our bounds
1431 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1432 * then new var_off is (0; 0x7f...fc) which improves our umax.
1433 */
1434 __reg_deduce_bounds(reg);
1435 __reg_bound_offset(reg);
1436 __update_reg_bounds(reg);
Edward Creeb03c9f92017-08-07 15:26:36 +01001437}
1438
Edward Creef1174f72017-08-07 15:26:19 +01001439/* Mark a register as having a completely unknown (scalar) value. */
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001440static void __mark_reg_unknown(const struct bpf_verifier_env *env,
1441 struct bpf_reg_state *reg)
Edward Creef1174f72017-08-07 15:26:19 +01001442{
Alexei Starovoitova9c676b2018-09-04 19:13:44 -07001443 /*
1444 * Clear type, id, off, and union(map_ptr, range) and
1445 * padding between 'type' and union
1446 */
1447 memset(reg, 0, offsetof(struct bpf_reg_state, var_off));
Edward Creef1174f72017-08-07 15:26:19 +01001448 reg->type = SCALAR_VALUE;
Edward Creef1174f72017-08-07 15:26:19 +01001449 reg->var_off = tnum_unknown;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001450 reg->frameno = 0;
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07001451 reg->precise = env->subprog_cnt > 1 || !env->bpf_capable;
Edward Creeb03c9f92017-08-07 15:26:36 +01001452 __mark_reg_unbounded(reg);
Edward Creef1174f72017-08-07 15:26:19 +01001453}
1454
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001455static void mark_reg_unknown(struct bpf_verifier_env *env,
1456 struct bpf_reg_state *regs, u32 regno)
Edward Creef1174f72017-08-07 15:26:19 +01001457{
1458 if (WARN_ON(regno >= MAX_BPF_REG)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001459 verbose(env, "mark_reg_unknown(regs, %u)\n", regno);
Alexei Starovoitov19ceb412017-11-30 21:31:37 -08001460 /* Something bad happened, let's kill all regs except FP */
1461 for (regno = 0; regno < BPF_REG_FP; regno++)
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001462 __mark_reg_not_init(env, regs + regno);
Edward Creef1174f72017-08-07 15:26:19 +01001463 return;
1464 }
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001465 __mark_reg_unknown(env, regs + regno);
Edward Creef1174f72017-08-07 15:26:19 +01001466}
1467
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001468static void __mark_reg_not_init(const struct bpf_verifier_env *env,
1469 struct bpf_reg_state *reg)
Edward Creef1174f72017-08-07 15:26:19 +01001470{
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001471 __mark_reg_unknown(env, reg);
Edward Creef1174f72017-08-07 15:26:19 +01001472 reg->type = NOT_INIT;
1473}
1474
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001475static void mark_reg_not_init(struct bpf_verifier_env *env,
1476 struct bpf_reg_state *regs, u32 regno)
Daniel Borkmanna9789ef2017-05-25 01:05:06 +02001477{
Edward Creef1174f72017-08-07 15:26:19 +01001478 if (WARN_ON(regno >= MAX_BPF_REG)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001479 verbose(env, "mark_reg_not_init(regs, %u)\n", regno);
Alexei Starovoitov19ceb412017-11-30 21:31:37 -08001480 /* Something bad happened, let's kill all regs except FP */
1481 for (regno = 0; regno < BPF_REG_FP; regno++)
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001482 __mark_reg_not_init(env, regs + regno);
Edward Creef1174f72017-08-07 15:26:19 +01001483 return;
1484 }
Daniel Borkmannf54c7892019-12-22 23:37:40 +01001485 __mark_reg_not_init(env, regs + regno);
Daniel Borkmanna9789ef2017-05-25 01:05:06 +02001486}
1487
Andrey Ignatov41c48f32020-06-19 14:11:43 -07001488static void mark_btf_ld_reg(struct bpf_verifier_env *env,
1489 struct bpf_reg_state *regs, u32 regno,
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08001490 enum bpf_reg_type reg_type,
1491 struct btf *btf, u32 btf_id)
Andrey Ignatov41c48f32020-06-19 14:11:43 -07001492{
1493 if (reg_type == SCALAR_VALUE) {
1494 mark_reg_unknown(env, regs, regno);
1495 return;
1496 }
1497 mark_reg_known_zero(env, regs, regno);
1498 regs[regno].type = PTR_TO_BTF_ID;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08001499 regs[regno].btf = btf;
Andrey Ignatov41c48f32020-06-19 14:11:43 -07001500 regs[regno].btf_id = btf_id;
1501}
1502
Jiong Wang5327ed32019-05-24 23:25:12 +01001503#define DEF_NOT_SUBREG (0)
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001504static void init_reg_state(struct bpf_verifier_env *env,
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001505 struct bpf_func_state *state)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001506{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001507 struct bpf_reg_state *regs = state->regs;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001508 int i;
1509
Edward Creedc503a82017-08-15 20:34:35 +01001510 for (i = 0; i < MAX_BPF_REG; i++) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001511 mark_reg_not_init(env, regs, i);
Edward Creedc503a82017-08-15 20:34:35 +01001512 regs[i].live = REG_LIVE_NONE;
Edward Cree679c7822018-08-22 20:02:19 +01001513 regs[i].parent = NULL;
Jiong Wang5327ed32019-05-24 23:25:12 +01001514 regs[i].subreg_def = DEF_NOT_SUBREG;
Edward Creedc503a82017-08-15 20:34:35 +01001515 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001516
1517 /* frame pointer */
Edward Creef1174f72017-08-07 15:26:19 +01001518 regs[BPF_REG_FP].type = PTR_TO_STACK;
Jakub Kicinski61bd5212017-10-09 10:30:11 -07001519 mark_reg_known_zero(env, regs, BPF_REG_FP);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001520 regs[BPF_REG_FP].frameno = state->frameno;
Daniel Borkmann6760bf22016-12-18 01:52:59 +01001521}
1522
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001523#define BPF_MAIN_FUNC (-1)
1524static void init_func_state(struct bpf_verifier_env *env,
1525 struct bpf_func_state *state,
1526 int callsite, int frameno, int subprogno)
1527{
1528 state->callsite = callsite;
1529 state->frameno = frameno;
1530 state->subprogno = subprogno;
1531 init_reg_state(env, state);
1532}
1533
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07001534/* Similar to push_stack(), but for async callbacks */
1535static struct bpf_verifier_state *push_async_cb(struct bpf_verifier_env *env,
1536 int insn_idx, int prev_insn_idx,
1537 int subprog)
1538{
1539 struct bpf_verifier_stack_elem *elem;
1540 struct bpf_func_state *frame;
1541
1542 elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
1543 if (!elem)
1544 goto err;
1545
1546 elem->insn_idx = insn_idx;
1547 elem->prev_insn_idx = prev_insn_idx;
1548 elem->next = env->head;
1549 elem->log_pos = env->log.len_used;
1550 env->head = elem;
1551 env->stack_size++;
1552 if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) {
1553 verbose(env,
1554 "The sequence of %d jumps is too complex for async cb.\n",
1555 env->stack_size);
1556 goto err;
1557 }
1558 /* Unlike push_stack() do not copy_verifier_state().
1559 * The caller state doesn't matter.
1560 * This is async callback. It starts in a fresh stack.
1561 * Initialize it similar to do_check_common().
1562 */
1563 elem->st.branches = 1;
1564 frame = kzalloc(sizeof(*frame), GFP_KERNEL);
1565 if (!frame)
1566 goto err;
1567 init_func_state(env, frame,
1568 BPF_MAIN_FUNC /* callsite */,
1569 0 /* frameno within this callchain */,
1570 subprog /* subprog number within this prog */);
1571 elem->st.frame[0] = frame;
1572 return &elem->st;
1573err:
1574 free_verifier_state(env->cur_state, true);
1575 env->cur_state = NULL;
1576 /* pop all elements and return */
1577 while (!pop_stack(env, NULL, NULL, false));
1578 return NULL;
1579}
1580
1581
Alexei Starovoitov17a52672014-09-26 00:17:06 -07001582enum reg_arg_type {
1583 SRC_OP, /* register is used as source operand */
1584 DST_OP, /* register is used as destination operand */
1585 DST_OP_NO_MARK /* same as above, check only, don't mark */
1586};
1587
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001588static int cmp_subprogs(const void *a, const void *b)
1589{
Jiong Wang9c8105b2018-05-02 16:17:18 -04001590 return ((struct bpf_subprog_info *)a)->start -
1591 ((struct bpf_subprog_info *)b)->start;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001592}
1593
1594static int find_subprog(struct bpf_verifier_env *env, int off)
1595{
Jiong Wang9c8105b2018-05-02 16:17:18 -04001596 struct bpf_subprog_info *p;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001597
Jiong Wang9c8105b2018-05-02 16:17:18 -04001598 p = bsearch(&off, env->subprog_info, env->subprog_cnt,
1599 sizeof(env->subprog_info[0]), cmp_subprogs);
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001600 if (!p)
1601 return -ENOENT;
Jiong Wang9c8105b2018-05-02 16:17:18 -04001602 return p - env->subprog_info;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001603
1604}
1605
1606static int add_subprog(struct bpf_verifier_env *env, int off)
1607{
1608 int insn_cnt = env->prog->len;
1609 int ret;
1610
1611 if (off >= insn_cnt || off < 0) {
1612 verbose(env, "call to invalid destination\n");
1613 return -EINVAL;
1614 }
1615 ret = find_subprog(env, off);
1616 if (ret >= 0)
Yonghong Song282a0f42021-02-26 12:49:24 -08001617 return ret;
Jiong Wang4cb3d992018-05-02 16:17:19 -04001618 if (env->subprog_cnt >= BPF_MAX_SUBPROGS) {
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001619 verbose(env, "too many subprograms\n");
1620 return -E2BIG;
1621 }
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001622 /* determine subprog starts. The end is one before the next starts */
Jiong Wang9c8105b2018-05-02 16:17:18 -04001623 env->subprog_info[env->subprog_cnt++].start = off;
1624 sort(env->subprog_info, env->subprog_cnt,
1625 sizeof(env->subprog_info[0]), cmp_subprogs, NULL);
Yonghong Song282a0f42021-02-26 12:49:24 -08001626 return env->subprog_cnt - 1;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001627}
1628
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001629struct bpf_kfunc_desc {
1630 struct btf_func_model func_model;
1631 u32 func_id;
1632 s32 imm;
1633};
1634
1635#define MAX_KFUNC_DESCS 256
1636struct bpf_kfunc_desc_tab {
1637 struct bpf_kfunc_desc descs[MAX_KFUNC_DESCS];
1638 u32 nr_descs;
1639};
1640
1641static int kfunc_desc_cmp_by_id(const void *a, const void *b)
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001642{
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001643 const struct bpf_kfunc_desc *d0 = a;
1644 const struct bpf_kfunc_desc *d1 = b;
1645
1646 /* func_id is not greater than BTF_MAX_TYPE */
1647 return d0->func_id - d1->func_id;
1648}
1649
1650static const struct bpf_kfunc_desc *
1651find_kfunc_desc(const struct bpf_prog *prog, u32 func_id)
1652{
1653 struct bpf_kfunc_desc desc = {
1654 .func_id = func_id,
1655 };
1656 struct bpf_kfunc_desc_tab *tab;
1657
1658 tab = prog->aux->kfunc_tab;
1659 return bsearch(&desc, tab->descs, tab->nr_descs,
1660 sizeof(tab->descs[0]), kfunc_desc_cmp_by_id);
1661}
1662
1663static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id)
1664{
1665 const struct btf_type *func, *func_proto;
1666 struct bpf_kfunc_desc_tab *tab;
1667 struct bpf_prog_aux *prog_aux;
1668 struct bpf_kfunc_desc *desc;
1669 const char *func_name;
1670 unsigned long addr;
1671 int err;
1672
1673 prog_aux = env->prog->aux;
1674 tab = prog_aux->kfunc_tab;
1675 if (!tab) {
1676 if (!btf_vmlinux) {
1677 verbose(env, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n");
1678 return -ENOTSUPP;
1679 }
1680
1681 if (!env->prog->jit_requested) {
1682 verbose(env, "JIT is required for calling kernel function\n");
1683 return -ENOTSUPP;
1684 }
1685
1686 if (!bpf_jit_supports_kfunc_call()) {
1687 verbose(env, "JIT does not support calling kernel function\n");
1688 return -ENOTSUPP;
1689 }
1690
1691 if (!env->prog->gpl_compatible) {
1692 verbose(env, "cannot call kernel function from non-GPL compatible program\n");
1693 return -EINVAL;
1694 }
1695
1696 tab = kzalloc(sizeof(*tab), GFP_KERNEL);
1697 if (!tab)
1698 return -ENOMEM;
1699 prog_aux->kfunc_tab = tab;
1700 }
1701
1702 if (find_kfunc_desc(env->prog, func_id))
1703 return 0;
1704
1705 if (tab->nr_descs == MAX_KFUNC_DESCS) {
1706 verbose(env, "too many different kernel function calls\n");
1707 return -E2BIG;
1708 }
1709
1710 func = btf_type_by_id(btf_vmlinux, func_id);
1711 if (!func || !btf_type_is_func(func)) {
1712 verbose(env, "kernel btf_id %u is not a function\n",
1713 func_id);
1714 return -EINVAL;
1715 }
1716 func_proto = btf_type_by_id(btf_vmlinux, func->type);
1717 if (!func_proto || !btf_type_is_func_proto(func_proto)) {
1718 verbose(env, "kernel function btf_id %u does not have a valid func_proto\n",
1719 func_id);
1720 return -EINVAL;
1721 }
1722
1723 func_name = btf_name_by_offset(btf_vmlinux, func->name_off);
1724 addr = kallsyms_lookup_name(func_name);
1725 if (!addr) {
1726 verbose(env, "cannot find address for kernel function %s\n",
1727 func_name);
1728 return -EINVAL;
1729 }
1730
1731 desc = &tab->descs[tab->nr_descs++];
1732 desc->func_id = func_id;
1733 desc->imm = BPF_CAST_CALL(addr) - __bpf_call_base;
1734 err = btf_distill_func_proto(&env->log, btf_vmlinux,
1735 func_proto, func_name,
1736 &desc->func_model);
1737 if (!err)
1738 sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]),
1739 kfunc_desc_cmp_by_id, NULL);
1740 return err;
1741}
1742
1743static int kfunc_desc_cmp_by_imm(const void *a, const void *b)
1744{
1745 const struct bpf_kfunc_desc *d0 = a;
1746 const struct bpf_kfunc_desc *d1 = b;
1747
1748 if (d0->imm > d1->imm)
1749 return 1;
1750 else if (d0->imm < d1->imm)
1751 return -1;
1752 return 0;
1753}
1754
1755static void sort_kfunc_descs_by_imm(struct bpf_prog *prog)
1756{
1757 struct bpf_kfunc_desc_tab *tab;
1758
1759 tab = prog->aux->kfunc_tab;
1760 if (!tab)
1761 return;
1762
1763 sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]),
1764 kfunc_desc_cmp_by_imm, NULL);
1765}
1766
1767bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog)
1768{
1769 return !!prog->aux->kfunc_tab;
1770}
1771
1772const struct btf_func_model *
1773bpf_jit_find_kfunc_model(const struct bpf_prog *prog,
1774 const struct bpf_insn *insn)
1775{
1776 const struct bpf_kfunc_desc desc = {
1777 .imm = insn->imm,
1778 };
1779 const struct bpf_kfunc_desc *res;
1780 struct bpf_kfunc_desc_tab *tab;
1781
1782 tab = prog->aux->kfunc_tab;
1783 res = bsearch(&desc, tab->descs, tab->nr_descs,
1784 sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm);
1785
1786 return res ? &res->func_model : NULL;
1787}
1788
1789static int add_subprog_and_kfunc(struct bpf_verifier_env *env)
1790{
Jiong Wang9c8105b2018-05-02 16:17:18 -04001791 struct bpf_subprog_info *subprog = env->subprog_info;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001792 struct bpf_insn *insn = env->prog->insnsi;
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001793 int i, ret, insn_cnt = env->prog->len;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001794
Jiong Wangf910cef2018-05-02 16:17:17 -04001795 /* Add entry function. */
1796 ret = add_subprog(env, 0);
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001797 if (ret)
Jiong Wangf910cef2018-05-02 16:17:17 -04001798 return ret;
1799
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001800 for (i = 0; i < insn_cnt; i++, insn++) {
1801 if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn) &&
1802 !bpf_pseudo_kfunc_call(insn))
Yonghong Song69c087b2021-02-26 12:49:25 -08001803 continue;
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001804
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07001805 if (!env->bpf_capable) {
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001806 verbose(env, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n");
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001807 return -EPERM;
1808 }
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001809
1810 if (bpf_pseudo_func(insn)) {
1811 ret = add_subprog(env, i + insn->imm + 1);
1812 if (ret >= 0)
1813 /* remember subprog */
1814 insn[1].imm = ret;
1815 } else if (bpf_pseudo_call(insn)) {
1816 ret = add_subprog(env, i + insn->imm + 1);
1817 } else {
1818 ret = add_kfunc_call(env, insn->imm);
1819 }
1820
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001821 if (ret < 0)
1822 return ret;
1823 }
1824
Jiong Wang4cb3d992018-05-02 16:17:19 -04001825 /* Add a fake 'exit' subprog which could simplify subprog iteration
1826 * logic. 'subprog_cnt' should not be increased.
1827 */
1828 subprog[env->subprog_cnt].start = insn_cnt;
1829
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07001830 if (env->log.level & BPF_LOG_LEVEL2)
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001831 for (i = 0; i < env->subprog_cnt; i++)
Jiong Wang9c8105b2018-05-02 16:17:18 -04001832 verbose(env, "func#%d @%d\n", i, subprog[i].start);
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001833
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07001834 return 0;
1835}
1836
1837static int check_subprogs(struct bpf_verifier_env *env)
1838{
1839 int i, subprog_start, subprog_end, off, cur_subprog = 0;
1840 struct bpf_subprog_info *subprog = env->subprog_info;
1841 struct bpf_insn *insn = env->prog->insnsi;
1842 int insn_cnt = env->prog->len;
1843
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001844 /* now check that all jumps are within the same subprog */
Jiong Wang4cb3d992018-05-02 16:17:19 -04001845 subprog_start = subprog[cur_subprog].start;
1846 subprog_end = subprog[cur_subprog + 1].start;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001847 for (i = 0; i < insn_cnt; i++) {
1848 u8 code = insn[i].code;
1849
Maciej Fijalkowski7f6e4312020-09-16 23:10:07 +02001850 if (code == (BPF_JMP | BPF_CALL) &&
1851 insn[i].imm == BPF_FUNC_tail_call &&
1852 insn[i].src_reg != BPF_PSEUDO_CALL)
1853 subprog[cur_subprog].has_tail_call = true;
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07001854 if (BPF_CLASS(code) == BPF_LD &&
1855 (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND))
1856 subprog[cur_subprog].has_ld_abs = true;
Jiong Wang092ed092019-01-26 12:26:01 -05001857 if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32)
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001858 goto next;
1859 if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL)
1860 goto next;
1861 off = i + insn[i].off + 1;
1862 if (off < subprog_start || off >= subprog_end) {
1863 verbose(env, "jump out of range from insn %d to %d\n", i, off);
1864 return -EINVAL;
1865 }
1866next:
1867 if (i == subprog_end - 1) {
1868 /* to avoid fall-through from one subprog into another
1869 * the last insn of the subprog should be either exit
1870 * or unconditional jump back
1871 */
1872 if (code != (BPF_JMP | BPF_EXIT) &&
1873 code != (BPF_JMP | BPF_JA)) {
1874 verbose(env, "last insn is not an exit or jmp\n");
1875 return -EINVAL;
1876 }
1877 subprog_start = subprog_end;
Jiong Wang4cb3d992018-05-02 16:17:19 -04001878 cur_subprog++;
1879 if (cur_subprog < env->subprog_cnt)
Jiong Wang9c8105b2018-05-02 16:17:18 -04001880 subprog_end = subprog[cur_subprog + 1].start;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08001881 }
1882 }
1883 return 0;
1884}
1885
Edward Cree679c7822018-08-22 20:02:19 +01001886/* Parentage chain of this register (or stack slot) should take care of all
1887 * issues like callee-saved registers, stack slot allocation time, etc.
1888 */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001889static int mark_reg_read(struct bpf_verifier_env *env,
Edward Cree679c7822018-08-22 20:02:19 +01001890 const struct bpf_reg_state *state,
Jiong Wang5327ed32019-05-24 23:25:12 +01001891 struct bpf_reg_state *parent, u8 flag)
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001892{
1893 bool writes = parent == state->parent; /* Observe write marks */
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07001894 int cnt = 0;
Edward Creedc503a82017-08-15 20:34:35 +01001895
1896 while (parent) {
1897 /* if read wasn't screened by an earlier write ... */
Edward Cree679c7822018-08-22 20:02:19 +01001898 if (writes && state->live & REG_LIVE_WRITTEN)
Edward Creedc503a82017-08-15 20:34:35 +01001899 break;
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -08001900 if (parent->live & REG_LIVE_DONE) {
1901 verbose(env, "verifier BUG type %s var_off %lld off %d\n",
1902 reg_type_str[parent->type],
1903 parent->var_off.value, parent->off);
1904 return -EFAULT;
1905 }
Jiong Wang5327ed32019-05-24 23:25:12 +01001906 /* The first condition is more likely to be true than the
1907 * second, checked it first.
1908 */
1909 if ((parent->live & REG_LIVE_READ) == flag ||
1910 parent->live & REG_LIVE_READ64)
Alexei Starovoitov25af32d2019-04-01 21:27:42 -07001911 /* The parentage chain never changes and
1912 * this parent was already marked as LIVE_READ.
1913 * There is no need to keep walking the chain again and
1914 * keep re-marking all parents as LIVE_READ.
1915 * This case happens when the same register is read
1916 * multiple times without writes into it in-between.
Jiong Wang5327ed32019-05-24 23:25:12 +01001917 * Also, if parent has the stronger REG_LIVE_READ64 set,
1918 * then no need to set the weak REG_LIVE_READ32.
Alexei Starovoitov25af32d2019-04-01 21:27:42 -07001919 */
1920 break;
Edward Creedc503a82017-08-15 20:34:35 +01001921 /* ... then we depend on parent's value */
Jiong Wang5327ed32019-05-24 23:25:12 +01001922 parent->live |= flag;
1923 /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */
1924 if (flag == REG_LIVE_READ64)
1925 parent->live &= ~REG_LIVE_READ32;
Edward Creedc503a82017-08-15 20:34:35 +01001926 state = parent;
1927 parent = state->parent;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001928 writes = true;
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07001929 cnt++;
Edward Creedc503a82017-08-15 20:34:35 +01001930 }
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07001931
1932 if (env->longest_mark_read_walk < cnt)
1933 env->longest_mark_read_walk = cnt;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08001934 return 0;
Edward Creedc503a82017-08-15 20:34:35 +01001935}
1936
Jiong Wang5327ed32019-05-24 23:25:12 +01001937/* This function is supposed to be used by the following 32-bit optimization
1938 * code only. It returns TRUE if the source or destination register operates
1939 * on 64-bit, otherwise return FALSE.
1940 */
1941static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn,
1942 u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t)
1943{
1944 u8 code, class, op;
1945
1946 code = insn->code;
1947 class = BPF_CLASS(code);
1948 op = BPF_OP(code);
1949 if (class == BPF_JMP) {
1950 /* BPF_EXIT for "main" will reach here. Return TRUE
1951 * conservatively.
1952 */
1953 if (op == BPF_EXIT)
1954 return true;
1955 if (op == BPF_CALL) {
1956 /* BPF to BPF call will reach here because of marking
1957 * caller saved clobber with DST_OP_NO_MARK for which we
1958 * don't care the register def because they are anyway
1959 * marked as NOT_INIT already.
1960 */
1961 if (insn->src_reg == BPF_PSEUDO_CALL)
1962 return false;
1963 /* Helper call will reach here because of arg type
1964 * check, conservatively return TRUE.
1965 */
1966 if (t == SRC_OP)
1967 return true;
1968
1969 return false;
1970 }
1971 }
1972
1973 if (class == BPF_ALU64 || class == BPF_JMP ||
1974 /* BPF_END always use BPF_ALU class. */
1975 (class == BPF_ALU && op == BPF_END && insn->imm == 64))
1976 return true;
1977
1978 if (class == BPF_ALU || class == BPF_JMP32)
1979 return false;
1980
1981 if (class == BPF_LDX) {
1982 if (t != SRC_OP)
1983 return BPF_SIZE(code) == BPF_DW;
1984 /* LDX source must be ptr. */
1985 return true;
1986 }
1987
1988 if (class == BPF_STX) {
Ilya Leoshkevich83a28812021-03-01 16:40:19 +01001989 /* BPF_STX (including atomic variants) has multiple source
1990 * operands, one of which is a ptr. Check whether the caller is
1991 * asking about it.
1992 */
1993 if (t == SRC_OP && reg->type != SCALAR_VALUE)
Jiong Wang5327ed32019-05-24 23:25:12 +01001994 return true;
1995 return BPF_SIZE(code) == BPF_DW;
1996 }
1997
1998 if (class == BPF_LD) {
1999 u8 mode = BPF_MODE(code);
2000
2001 /* LD_IMM64 */
2002 if (mode == BPF_IMM)
2003 return true;
2004
2005 /* Both LD_IND and LD_ABS return 32-bit data. */
2006 if (t != SRC_OP)
2007 return false;
2008
2009 /* Implicit ctx ptr. */
2010 if (regno == BPF_REG_6)
2011 return true;
2012
2013 /* Explicit source could be any width. */
2014 return true;
2015 }
2016
2017 if (class == BPF_ST)
2018 /* The only source register for BPF_ST is a ptr. */
2019 return true;
2020
2021 /* Conservatively return true at default. */
2022 return true;
2023}
2024
Ilya Leoshkevich83a28812021-03-01 16:40:19 +01002025/* Return the regno defined by the insn, or -1. */
2026static int insn_def_regno(const struct bpf_insn *insn)
Jiong Wangb325fbc2019-05-24 23:25:13 +01002027{
Ilya Leoshkevich83a28812021-03-01 16:40:19 +01002028 switch (BPF_CLASS(insn->code)) {
2029 case BPF_JMP:
2030 case BPF_JMP32:
2031 case BPF_ST:
2032 return -1;
2033 case BPF_STX:
2034 if (BPF_MODE(insn->code) == BPF_ATOMIC &&
2035 (insn->imm & BPF_FETCH)) {
2036 if (insn->imm == BPF_CMPXCHG)
2037 return BPF_REG_0;
2038 else
2039 return insn->src_reg;
2040 } else {
2041 return -1;
2042 }
2043 default:
2044 return insn->dst_reg;
2045 }
Jiong Wangb325fbc2019-05-24 23:25:13 +01002046}
2047
2048/* Return TRUE if INSN has defined any 32-bit value explicitly. */
2049static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn)
2050{
Ilya Leoshkevich83a28812021-03-01 16:40:19 +01002051 int dst_reg = insn_def_regno(insn);
2052
2053 if (dst_reg == -1)
Jiong Wangb325fbc2019-05-24 23:25:13 +01002054 return false;
2055
Ilya Leoshkevich83a28812021-03-01 16:40:19 +01002056 return !is_reg64(env, insn, dst_reg, NULL, DST_OP);
Jiong Wangb325fbc2019-05-24 23:25:13 +01002057}
2058
Jiong Wang5327ed32019-05-24 23:25:12 +01002059static void mark_insn_zext(struct bpf_verifier_env *env,
2060 struct bpf_reg_state *reg)
2061{
2062 s32 def_idx = reg->subreg_def;
2063
2064 if (def_idx == DEF_NOT_SUBREG)
2065 return;
2066
2067 env->insn_aux_data[def_idx - 1].zext_dst = true;
2068 /* The dst will be zero extended, so won't be sub-register anymore. */
2069 reg->subreg_def = DEF_NOT_SUBREG;
2070}
2071
Edward Creedc503a82017-08-15 20:34:35 +01002072static int check_reg_arg(struct bpf_verifier_env *env, u32 regno,
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002073 enum reg_arg_type t)
2074{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002075 struct bpf_verifier_state *vstate = env->cur_state;
2076 struct bpf_func_state *state = vstate->frame[vstate->curframe];
Jiong Wang5327ed32019-05-24 23:25:12 +01002077 struct bpf_insn *insn = env->prog->insnsi + env->insn_idx;
Jiong Wangc342dc12019-04-12 22:59:37 +01002078 struct bpf_reg_state *reg, *regs = state->regs;
Jiong Wang5327ed32019-05-24 23:25:12 +01002079 bool rw64;
Edward Creedc503a82017-08-15 20:34:35 +01002080
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002081 if (regno >= MAX_BPF_REG) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002082 verbose(env, "R%d is invalid\n", regno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002083 return -EINVAL;
2084 }
2085
Jiong Wangc342dc12019-04-12 22:59:37 +01002086 reg = &regs[regno];
Jiong Wang5327ed32019-05-24 23:25:12 +01002087 rw64 = is_reg64(env, insn, regno, reg, t);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002088 if (t == SRC_OP) {
2089 /* check whether register used as source operand can be read */
Jiong Wangc342dc12019-04-12 22:59:37 +01002090 if (reg->type == NOT_INIT) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002091 verbose(env, "R%d !read_ok\n", regno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002092 return -EACCES;
2093 }
Edward Cree679c7822018-08-22 20:02:19 +01002094 /* We don't need to worry about FP liveness because it's read-only */
Jiong Wangc342dc12019-04-12 22:59:37 +01002095 if (regno == BPF_REG_FP)
2096 return 0;
2097
Jiong Wang5327ed32019-05-24 23:25:12 +01002098 if (rw64)
2099 mark_insn_zext(env, reg);
2100
2101 return mark_reg_read(env, reg, reg->parent,
2102 rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002103 } else {
2104 /* check whether register used as dest operand can be written to */
2105 if (regno == BPF_REG_FP) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002106 verbose(env, "frame pointer is read only\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002107 return -EACCES;
2108 }
Jiong Wangc342dc12019-04-12 22:59:37 +01002109 reg->live |= REG_LIVE_WRITTEN;
Jiong Wang5327ed32019-05-24 23:25:12 +01002110 reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002111 if (t == DST_OP)
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002112 mark_reg_unknown(env, regs, regno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002113 }
2114 return 0;
2115}
2116
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002117/* for any branch, call, exit record the history of jmps in the given state */
2118static int push_jmp_history(struct bpf_verifier_env *env,
2119 struct bpf_verifier_state *cur)
2120{
2121 u32 cnt = cur->jmp_history_cnt;
2122 struct bpf_idx_pair *p;
2123
2124 cnt++;
2125 p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER);
2126 if (!p)
2127 return -ENOMEM;
2128 p[cnt - 1].idx = env->insn_idx;
2129 p[cnt - 1].prev_idx = env->prev_insn_idx;
2130 cur->jmp_history = p;
2131 cur->jmp_history_cnt = cnt;
2132 return 0;
2133}
2134
2135/* Backtrack one insn at a time. If idx is not at the top of recorded
2136 * history then previous instruction came from straight line execution.
2137 */
2138static int get_prev_insn_idx(struct bpf_verifier_state *st, int i,
2139 u32 *history)
2140{
2141 u32 cnt = *history;
2142
2143 if (cnt && st->jmp_history[cnt - 1].idx == i) {
2144 i = st->jmp_history[cnt - 1].prev_idx;
2145 (*history)--;
2146 } else {
2147 i--;
2148 }
2149 return i;
2150}
2151
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07002152static const char *disasm_kfunc_name(void *data, const struct bpf_insn *insn)
2153{
2154 const struct btf_type *func;
2155
2156 if (insn->src_reg != BPF_PSEUDO_KFUNC_CALL)
2157 return NULL;
2158
2159 func = btf_type_by_id(btf_vmlinux, insn->imm);
2160 return btf_name_by_offset(btf_vmlinux, func->name_off);
2161}
2162
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002163/* For given verifier state backtrack_insn() is called from the last insn to
2164 * the first insn. Its purpose is to compute a bitmask of registers and
2165 * stack slots that needs precision in the parent verifier state.
2166 */
2167static int backtrack_insn(struct bpf_verifier_env *env, int idx,
2168 u32 *reg_mask, u64 *stack_mask)
2169{
2170 const struct bpf_insn_cbs cbs = {
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07002171 .cb_call = disasm_kfunc_name,
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002172 .cb_print = verbose,
2173 .private_data = env,
2174 };
2175 struct bpf_insn *insn = env->prog->insnsi + idx;
2176 u8 class = BPF_CLASS(insn->code);
2177 u8 opcode = BPF_OP(insn->code);
2178 u8 mode = BPF_MODE(insn->code);
2179 u32 dreg = 1u << insn->dst_reg;
2180 u32 sreg = 1u << insn->src_reg;
2181 u32 spi;
2182
2183 if (insn->code == 0)
2184 return 0;
2185 if (env->log.level & BPF_LOG_LEVEL) {
2186 verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask);
2187 verbose(env, "%d: ", idx);
2188 print_bpf_insn(&cbs, insn, env->allow_ptr_leaks);
2189 }
2190
2191 if (class == BPF_ALU || class == BPF_ALU64) {
2192 if (!(*reg_mask & dreg))
2193 return 0;
2194 if (opcode == BPF_MOV) {
2195 if (BPF_SRC(insn->code) == BPF_X) {
2196 /* dreg = sreg
2197 * dreg needs precision after this insn
2198 * sreg needs precision before this insn
2199 */
2200 *reg_mask &= ~dreg;
2201 *reg_mask |= sreg;
2202 } else {
2203 /* dreg = K
2204 * dreg needs precision after this insn.
2205 * Corresponding register is already marked
2206 * as precise=true in this verifier state.
2207 * No further markings in parent are necessary
2208 */
2209 *reg_mask &= ~dreg;
2210 }
2211 } else {
2212 if (BPF_SRC(insn->code) == BPF_X) {
2213 /* dreg += sreg
2214 * both dreg and sreg need precision
2215 * before this insn
2216 */
2217 *reg_mask |= sreg;
2218 } /* else dreg += K
2219 * dreg still needs precision before this insn
2220 */
2221 }
2222 } else if (class == BPF_LDX) {
2223 if (!(*reg_mask & dreg))
2224 return 0;
2225 *reg_mask &= ~dreg;
2226
2227 /* scalars can only be spilled into stack w/o losing precision.
2228 * Load from any other memory can be zero extended.
2229 * The desire to keep that precision is already indicated
2230 * by 'precise' mark in corresponding register of this state.
2231 * No further tracking necessary.
2232 */
2233 if (insn->src_reg != BPF_REG_FP)
2234 return 0;
2235 if (BPF_SIZE(insn->code) != BPF_DW)
2236 return 0;
2237
2238 /* dreg = *(u64 *)[fp - off] was a fill from the stack.
2239 * that [fp - off] slot contains scalar that needs to be
2240 * tracked with precision
2241 */
2242 spi = (-insn->off - 1) / BPF_REG_SIZE;
2243 if (spi >= 64) {
2244 verbose(env, "BUG spi %d\n", spi);
2245 WARN_ONCE(1, "verifier backtracking bug");
2246 return -EFAULT;
2247 }
2248 *stack_mask |= 1ull << spi;
Andrii Nakryikob3b50f02019-07-08 20:32:44 -07002249 } else if (class == BPF_STX || class == BPF_ST) {
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002250 if (*reg_mask & dreg)
Andrii Nakryikob3b50f02019-07-08 20:32:44 -07002251 /* stx & st shouldn't be using _scalar_ dst_reg
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002252 * to access memory. It means backtracking
2253 * encountered a case of pointer subtraction.
2254 */
2255 return -ENOTSUPP;
2256 /* scalars can only be spilled into stack */
2257 if (insn->dst_reg != BPF_REG_FP)
2258 return 0;
2259 if (BPF_SIZE(insn->code) != BPF_DW)
2260 return 0;
2261 spi = (-insn->off - 1) / BPF_REG_SIZE;
2262 if (spi >= 64) {
2263 verbose(env, "BUG spi %d\n", spi);
2264 WARN_ONCE(1, "verifier backtracking bug");
2265 return -EFAULT;
2266 }
2267 if (!(*stack_mask & (1ull << spi)))
2268 return 0;
2269 *stack_mask &= ~(1ull << spi);
Andrii Nakryikob3b50f02019-07-08 20:32:44 -07002270 if (class == BPF_STX)
2271 *reg_mask |= sreg;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002272 } else if (class == BPF_JMP || class == BPF_JMP32) {
2273 if (opcode == BPF_CALL) {
2274 if (insn->src_reg == BPF_PSEUDO_CALL)
2275 return -ENOTSUPP;
2276 /* regular helper call sets R0 */
2277 *reg_mask &= ~1;
2278 if (*reg_mask & 0x3f) {
2279 /* if backtracing was looking for registers R1-R5
2280 * they should have been found already.
2281 */
2282 verbose(env, "BUG regs %x\n", *reg_mask);
2283 WARN_ONCE(1, "verifier backtracking bug");
2284 return -EFAULT;
2285 }
2286 } else if (opcode == BPF_EXIT) {
2287 return -ENOTSUPP;
2288 }
2289 } else if (class == BPF_LD) {
2290 if (!(*reg_mask & dreg))
2291 return 0;
2292 *reg_mask &= ~dreg;
2293 /* It's ld_imm64 or ld_abs or ld_ind.
2294 * For ld_imm64 no further tracking of precision
2295 * into parent is necessary
2296 */
2297 if (mode == BPF_IND || mode == BPF_ABS)
2298 /* to be analyzed */
2299 return -ENOTSUPP;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002300 }
2301 return 0;
2302}
2303
2304/* the scalar precision tracking algorithm:
2305 * . at the start all registers have precise=false.
2306 * . scalar ranges are tracked as normal through alu and jmp insns.
2307 * . once precise value of the scalar register is used in:
2308 * . ptr + scalar alu
2309 * . if (scalar cond K|scalar)
2310 * . helper_call(.., scalar, ...) where ARG_CONST is expected
2311 * backtrack through the verifier states and mark all registers and
2312 * stack slots with spilled constants that these scalar regisers
2313 * should be precise.
2314 * . during state pruning two registers (or spilled stack slots)
2315 * are equivalent if both are not precise.
2316 *
2317 * Note the verifier cannot simply walk register parentage chain,
2318 * since many different registers and stack slots could have been
2319 * used to compute single precise scalar.
2320 *
2321 * The approach of starting with precise=true for all registers and then
2322 * backtrack to mark a register as not precise when the verifier detects
2323 * that program doesn't care about specific value (e.g., when helper
2324 * takes register as ARG_ANYTHING parameter) is not safe.
2325 *
2326 * It's ok to walk single parentage chain of the verifier states.
2327 * It's possible that this backtracking will go all the way till 1st insn.
2328 * All other branches will be explored for needing precision later.
2329 *
2330 * The backtracking needs to deal with cases like:
2331 * R8=map_value(id=0,off=0,ks=4,vs=1952,imm=0) R9_w=map_value(id=0,off=40,ks=4,vs=1952,imm=0)
2332 * r9 -= r8
2333 * r5 = r9
2334 * if r5 > 0x79f goto pc+7
2335 * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff))
2336 * r5 += 1
2337 * ...
2338 * call bpf_perf_event_output#25
2339 * where .arg5_type = ARG_CONST_SIZE_OR_ZERO
2340 *
2341 * and this case:
2342 * r6 = 1
2343 * call foo // uses callee's r6 inside to compute r0
2344 * r0 += r6
2345 * if r0 == 0 goto
2346 *
2347 * to track above reg_mask/stack_mask needs to be independent for each frame.
2348 *
2349 * Also if parent's curframe > frame where backtracking started,
2350 * the verifier need to mark registers in both frames, otherwise callees
2351 * may incorrectly prune callers. This is similar to
2352 * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences")
2353 *
2354 * For now backtracking falls back into conservative marking.
2355 */
2356static void mark_all_scalars_precise(struct bpf_verifier_env *env,
2357 struct bpf_verifier_state *st)
2358{
2359 struct bpf_func_state *func;
2360 struct bpf_reg_state *reg;
2361 int i, j;
2362
2363 /* big hammer: mark all scalars precise in this path.
2364 * pop_stack may still get !precise scalars.
2365 */
2366 for (; st; st = st->parent)
2367 for (i = 0; i <= st->curframe; i++) {
2368 func = st->frame[i];
2369 for (j = 0; j < BPF_REG_FP; j++) {
2370 reg = &func->regs[j];
2371 if (reg->type != SCALAR_VALUE)
2372 continue;
2373 reg->precise = true;
2374 }
2375 for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) {
2376 if (func->stack[j].slot_type[0] != STACK_SPILL)
2377 continue;
2378 reg = &func->stack[j].spilled_ptr;
2379 if (reg->type != SCALAR_VALUE)
2380 continue;
2381 reg->precise = true;
2382 }
2383 }
2384}
2385
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002386static int __mark_chain_precision(struct bpf_verifier_env *env, int regno,
2387 int spi)
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002388{
2389 struct bpf_verifier_state *st = env->cur_state;
2390 int first_idx = st->first_insn_idx;
2391 int last_idx = env->insn_idx;
2392 struct bpf_func_state *func;
2393 struct bpf_reg_state *reg;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002394 u32 reg_mask = regno >= 0 ? 1u << regno : 0;
2395 u64 stack_mask = spi >= 0 ? 1ull << spi : 0;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002396 bool skip_first = true;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002397 bool new_marks = false;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002398 int i, err;
2399
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07002400 if (!env->bpf_capable)
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002401 return 0;
2402
2403 func = st->frame[st->curframe];
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002404 if (regno >= 0) {
2405 reg = &func->regs[regno];
2406 if (reg->type != SCALAR_VALUE) {
2407 WARN_ONCE(1, "backtracing misuse");
2408 return -EFAULT;
2409 }
2410 if (!reg->precise)
2411 new_marks = true;
2412 else
2413 reg_mask = 0;
2414 reg->precise = true;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002415 }
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002416
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002417 while (spi >= 0) {
2418 if (func->stack[spi].slot_type[0] != STACK_SPILL) {
2419 stack_mask = 0;
2420 break;
2421 }
2422 reg = &func->stack[spi].spilled_ptr;
2423 if (reg->type != SCALAR_VALUE) {
2424 stack_mask = 0;
2425 break;
2426 }
2427 if (!reg->precise)
2428 new_marks = true;
2429 else
2430 stack_mask = 0;
2431 reg->precise = true;
2432 break;
2433 }
2434
2435 if (!new_marks)
2436 return 0;
2437 if (!reg_mask && !stack_mask)
2438 return 0;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002439 for (;;) {
2440 DECLARE_BITMAP(mask, 64);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002441 u32 history = st->jmp_history_cnt;
2442
2443 if (env->log.level & BPF_LOG_LEVEL)
2444 verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx);
2445 for (i = last_idx;;) {
2446 if (skip_first) {
2447 err = 0;
2448 skip_first = false;
2449 } else {
2450 err = backtrack_insn(env, i, &reg_mask, &stack_mask);
2451 }
2452 if (err == -ENOTSUPP) {
2453 mark_all_scalars_precise(env, st);
2454 return 0;
2455 } else if (err) {
2456 return err;
2457 }
2458 if (!reg_mask && !stack_mask)
2459 /* Found assignment(s) into tracked register in this state.
2460 * Since this state is already marked, just return.
2461 * Nothing to be tracked further in the parent state.
2462 */
2463 return 0;
2464 if (i == first_idx)
2465 break;
2466 i = get_prev_insn_idx(st, i, &history);
2467 if (i >= env->prog->len) {
2468 /* This can happen if backtracking reached insn 0
2469 * and there are still reg_mask or stack_mask
2470 * to backtrack.
2471 * It means the backtracking missed the spot where
2472 * particular register was initialized with a constant.
2473 */
2474 verbose(env, "BUG backtracking idx %d\n", i);
2475 WARN_ONCE(1, "verifier backtracking bug");
2476 return -EFAULT;
2477 }
2478 }
2479 st = st->parent;
2480 if (!st)
2481 break;
2482
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002483 new_marks = false;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002484 func = st->frame[st->curframe];
2485 bitmap_from_u64(mask, reg_mask);
2486 for_each_set_bit(i, mask, 32) {
2487 reg = &func->regs[i];
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002488 if (reg->type != SCALAR_VALUE) {
2489 reg_mask &= ~(1u << i);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002490 continue;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002491 }
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002492 if (!reg->precise)
2493 new_marks = true;
2494 reg->precise = true;
2495 }
2496
2497 bitmap_from_u64(mask, stack_mask);
2498 for_each_set_bit(i, mask, 64) {
2499 if (i >= func->allocated_stack / BPF_REG_SIZE) {
Alexei Starovoitov2339cd62019-09-03 15:16:17 -07002500 /* the sequence of instructions:
2501 * 2: (bf) r3 = r10
2502 * 3: (7b) *(u64 *)(r3 -8) = r0
2503 * 4: (79) r4 = *(u64 *)(r10 -8)
2504 * doesn't contain jmps. It's backtracked
2505 * as a single block.
2506 * During backtracking insn 3 is not recognized as
2507 * stack access, so at the end of backtracking
2508 * stack slot fp-8 is still marked in stack_mask.
2509 * However the parent state may not have accessed
2510 * fp-8 and it's "unallocated" stack space.
2511 * In such case fallback to conservative.
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002512 */
Alexei Starovoitov2339cd62019-09-03 15:16:17 -07002513 mark_all_scalars_precise(env, st);
2514 return 0;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002515 }
2516
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002517 if (func->stack[i].slot_type[0] != STACK_SPILL) {
2518 stack_mask &= ~(1ull << i);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002519 continue;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002520 }
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002521 reg = &func->stack[i].spilled_ptr;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002522 if (reg->type != SCALAR_VALUE) {
2523 stack_mask &= ~(1ull << i);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002524 continue;
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002525 }
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002526 if (!reg->precise)
2527 new_marks = true;
2528 reg->precise = true;
2529 }
2530 if (env->log.level & BPF_LOG_LEVEL) {
2531 print_verifier_state(env, func);
2532 verbose(env, "parent %s regs=%x stack=%llx marks\n",
2533 new_marks ? "didn't have" : "already had",
2534 reg_mask, stack_mask);
2535 }
2536
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002537 if (!reg_mask && !stack_mask)
2538 break;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002539 if (!new_marks)
2540 break;
2541
2542 last_idx = st->last_insn_idx;
2543 first_idx = st->first_insn_idx;
2544 }
2545 return 0;
2546}
2547
Alexei Starovoitova3ce6852019-06-28 09:24:09 -07002548static int mark_chain_precision(struct bpf_verifier_env *env, int regno)
2549{
2550 return __mark_chain_precision(env, regno, -1);
2551}
2552
2553static int mark_chain_precision_stack(struct bpf_verifier_env *env, int spi)
2554{
2555 return __mark_chain_precision(env, -1, spi);
2556}
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002557
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07002558static bool is_spillable_regtype(enum bpf_reg_type type)
2559{
2560 switch (type) {
2561 case PTR_TO_MAP_VALUE:
2562 case PTR_TO_MAP_VALUE_OR_NULL:
2563 case PTR_TO_STACK:
2564 case PTR_TO_CTX:
Alexei Starovoitov969bf052016-05-05 19:49:10 -07002565 case PTR_TO_PACKET:
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02002566 case PTR_TO_PACKET_META:
Alexei Starovoitov969bf052016-05-05 19:49:10 -07002567 case PTR_TO_PACKET_END:
Petar Penkovd58e4682018-09-14 07:46:18 -07002568 case PTR_TO_FLOW_KEYS:
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07002569 case CONST_PTR_TO_MAP:
Joe Stringerc64b7982018-10-02 13:35:33 -07002570 case PTR_TO_SOCKET:
2571 case PTR_TO_SOCKET_OR_NULL:
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08002572 case PTR_TO_SOCK_COMMON:
2573 case PTR_TO_SOCK_COMMON_OR_NULL:
Martin KaFai Lau655a51e2019-02-09 23:22:24 -08002574 case PTR_TO_TCP_SOCK:
2575 case PTR_TO_TCP_SOCK_OR_NULL:
Jonathan Lemonfada7fd2019-06-06 13:59:40 -07002576 case PTR_TO_XDP_SOCK:
Martin KaFai Lau65726b52020-01-08 16:34:54 -08002577 case PTR_TO_BTF_ID:
Yonghong Songb121b342020-05-09 10:59:12 -07002578 case PTR_TO_BTF_ID_OR_NULL:
Yonghong Songafbf21d2020-07-23 11:41:11 -07002579 case PTR_TO_RDONLY_BUF:
2580 case PTR_TO_RDONLY_BUF_OR_NULL:
2581 case PTR_TO_RDWR_BUF:
2582 case PTR_TO_RDWR_BUF_OR_NULL:
Hao Luoeaa6bcb2020-09-29 16:50:47 -07002583 case PTR_TO_PERCPU_BTF_ID:
Gilad Reti744ea4e2021-01-13 07:38:07 +02002584 case PTR_TO_MEM:
2585 case PTR_TO_MEM_OR_NULL:
Yonghong Song69c087b2021-02-26 12:49:25 -08002586 case PTR_TO_FUNC:
2587 case PTR_TO_MAP_KEY:
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07002588 return true;
2589 default:
2590 return false;
2591 }
2592}
2593
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002594/* Does this register contain a constant zero? */
2595static bool register_is_null(struct bpf_reg_state *reg)
2596{
2597 return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0);
2598}
2599
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002600static bool register_is_const(struct bpf_reg_state *reg)
2601{
2602 return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off);
2603}
2604
Yonghong Song5689d492020-10-08 18:12:38 -07002605static bool __is_scalar_unbounded(struct bpf_reg_state *reg)
2606{
2607 return tnum_is_unknown(reg->var_off) &&
2608 reg->smin_value == S64_MIN && reg->smax_value == S64_MAX &&
2609 reg->umin_value == 0 && reg->umax_value == U64_MAX &&
2610 reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX &&
2611 reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX;
2612}
2613
2614static bool register_is_bounded(struct bpf_reg_state *reg)
2615{
2616 return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg);
2617}
2618
Jann Horn6e7e63c2020-04-17 02:00:06 +02002619static bool __is_pointer_value(bool allow_ptr_leaks,
2620 const struct bpf_reg_state *reg)
2621{
2622 if (allow_ptr_leaks)
2623 return false;
2624
2625 return reg->type != SCALAR_VALUE;
2626}
2627
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002628static void save_register_state(struct bpf_func_state *state,
2629 int spi, struct bpf_reg_state *reg)
2630{
2631 int i;
2632
2633 state->stack[spi].spilled_ptr = *reg;
2634 state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
2635
2636 for (i = 0; i < BPF_REG_SIZE; i++)
2637 state->stack[spi].slot_type[i] = STACK_SPILL;
2638}
2639
Andrei Matei01f810a2021-02-06 20:10:24 -05002640/* check_stack_{read,write}_fixed_off functions track spill/fill of registers,
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002641 * stack boundary and alignment are checked in check_mem_access()
2642 */
Andrei Matei01f810a2021-02-06 20:10:24 -05002643static int check_stack_write_fixed_off(struct bpf_verifier_env *env,
2644 /* stack frame we're writing to */
2645 struct bpf_func_state *state,
2646 int off, int size, int value_regno,
2647 int insn_idx)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002648{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002649 struct bpf_func_state *cur; /* state of the current function */
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002650 int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002651 u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg;
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002652 struct bpf_reg_state *reg = NULL;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002653
Lorenz Bauerc69431a2021-04-29 14:46:54 +01002654 err = grow_stack_state(state, round_up(slot + 1, BPF_REG_SIZE));
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002655 if (err)
2656 return err;
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002657 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
2658 * so it's aligned access and [off, off + size) are within stack limits
2659 */
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002660 if (!env->allow_ptr_leaks &&
2661 state->stack[spi].slot_type[0] == STACK_SPILL &&
2662 size != BPF_REG_SIZE) {
2663 verbose(env, "attempt to corrupt spilled pointer on stack\n");
2664 return -EACCES;
2665 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002666
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002667 cur = env->cur_state->frame[env->cur_state->curframe];
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002668 if (value_regno >= 0)
2669 reg = &cur->regs[value_regno];
Daniel Borkmann2039f262021-07-13 08:18:31 +00002670 if (!env->bypass_spec_v4) {
2671 bool sanitize = reg && is_spillable_regtype(reg->type);
2672
2673 for (i = 0; i < size; i++) {
2674 if (state->stack[spi].slot_type[i] == STACK_INVALID) {
2675 sanitize = true;
2676 break;
2677 }
2678 }
2679
2680 if (sanitize)
2681 env->insn_aux_data[insn_idx].sanitize_stack_spill = true;
2682 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002683
Yonghong Song5689d492020-10-08 18:12:38 -07002684 if (reg && size == BPF_REG_SIZE && register_is_bounded(reg) &&
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07002685 !register_is_null(reg) && env->bpf_capable) {
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002686 if (dst_reg != BPF_REG_FP) {
2687 /* The backtracking logic can only recognize explicit
2688 * stack slot address like [fp - 8]. Other spill of
Zhen Lei8fb33b62021-05-25 10:56:59 +08002689 * scalar via different register has to be conservative.
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002690 * Backtrack from here and mark all registers as precise
2691 * that contributed into 'reg' being a constant.
2692 */
2693 err = mark_chain_precision(env, value_regno);
2694 if (err)
2695 return err;
2696 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002697 save_register_state(state, spi, reg);
2698 } else if (reg && is_spillable_regtype(reg->type)) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002699 /* register containing pointer is being spilled into stack */
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002700 if (size != BPF_REG_SIZE) {
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002701 verbose_linfo(env, insn_idx, "; ");
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002702 verbose(env, "invalid size of register spill\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002703 return -EACCES;
2704 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002705 if (state != cur && reg->type == PTR_TO_STACK) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002706 verbose(env, "cannot spill pointers to stack into stack frame of the caller\n");
2707 return -EINVAL;
2708 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002709 save_register_state(state, spi, reg);
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002710 } else {
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002711 u8 type = STACK_MISC;
2712
Edward Cree679c7822018-08-22 20:02:19 +01002713 /* regular write of data into stack destroys any spilled ptr */
2714 state->stack[spi].spilled_ptr.type = NOT_INIT;
Jiong Wang0bae2d42018-12-15 03:34:40 -05002715 /* Mark slots as STACK_MISC if they belonged to spilled ptr. */
2716 if (state->stack[spi].slot_type[0] == STACK_SPILL)
2717 for (i = 0; i < BPF_REG_SIZE; i++)
2718 state->stack[spi].slot_type[i] = STACK_MISC;
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002719
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002720 /* only mark the slot as written if all 8 bytes were written
2721 * otherwise read propagation may incorrectly stop too soon
2722 * when stack slots are partially written.
2723 * This heuristic means that read propagation will be
2724 * conservative, since it will add reg_live_read marks
2725 * to stack slots all the way to first state when programs
2726 * writes+reads less than 8 bytes
2727 */
2728 if (size == BPF_REG_SIZE)
2729 state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
2730
2731 /* when we zero initialize stack slots mark them as such */
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002732 if (reg && register_is_null(reg)) {
2733 /* backtracking doesn't work for STACK_ZERO yet. */
2734 err = mark_chain_precision(env, value_regno);
2735 if (err)
2736 return err;
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002737 type = STACK_ZERO;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07002738 }
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002739
Jiong Wang0bae2d42018-12-15 03:34:40 -05002740 /* Mark slots affected by this stack write. */
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002741 for (i = 0; i < size; i++)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002742 state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] =
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002743 type;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002744 }
2745 return 0;
2746}
2747
Andrei Matei01f810a2021-02-06 20:10:24 -05002748/* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is
2749 * known to contain a variable offset.
2750 * This function checks whether the write is permitted and conservatively
2751 * tracks the effects of the write, considering that each stack slot in the
2752 * dynamic range is potentially written to.
2753 *
2754 * 'off' includes 'regno->off'.
2755 * 'value_regno' can be -1, meaning that an unknown value is being written to
2756 * the stack.
2757 *
2758 * Spilled pointers in range are not marked as written because we don't know
2759 * what's going to be actually written. This means that read propagation for
2760 * future reads cannot be terminated by this write.
2761 *
2762 * For privileged programs, uninitialized stack slots are considered
2763 * initialized by this write (even though we don't know exactly what offsets
2764 * are going to be written to). The idea is that we don't want the verifier to
2765 * reject future reads that access slots written to through variable offsets.
2766 */
2767static int check_stack_write_var_off(struct bpf_verifier_env *env,
2768 /* func where register points to */
2769 struct bpf_func_state *state,
2770 int ptr_regno, int off, int size,
2771 int value_regno, int insn_idx)
2772{
2773 struct bpf_func_state *cur; /* state of the current function */
2774 int min_off, max_off;
2775 int i, err;
2776 struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL;
2777 bool writing_zero = false;
2778 /* set if the fact that we're writing a zero is used to let any
2779 * stack slots remain STACK_ZERO
2780 */
2781 bool zero_used = false;
2782
2783 cur = env->cur_state->frame[env->cur_state->curframe];
2784 ptr_reg = &cur->regs[ptr_regno];
2785 min_off = ptr_reg->smin_value + off;
2786 max_off = ptr_reg->smax_value + off + size;
2787 if (value_regno >= 0)
2788 value_reg = &cur->regs[value_regno];
2789 if (value_reg && register_is_null(value_reg))
2790 writing_zero = true;
2791
Lorenz Bauerc69431a2021-04-29 14:46:54 +01002792 err = grow_stack_state(state, round_up(-min_off, BPF_REG_SIZE));
Andrei Matei01f810a2021-02-06 20:10:24 -05002793 if (err)
2794 return err;
2795
2796
2797 /* Variable offset writes destroy any spilled pointers in range. */
2798 for (i = min_off; i < max_off; i++) {
2799 u8 new_type, *stype;
2800 int slot, spi;
2801
2802 slot = -i - 1;
2803 spi = slot / BPF_REG_SIZE;
2804 stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE];
2805
2806 if (!env->allow_ptr_leaks
2807 && *stype != NOT_INIT
2808 && *stype != SCALAR_VALUE) {
2809 /* Reject the write if there's are spilled pointers in
2810 * range. If we didn't reject here, the ptr status
2811 * would be erased below (even though not all slots are
2812 * actually overwritten), possibly opening the door to
2813 * leaks.
2814 */
2815 verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d",
2816 insn_idx, i);
2817 return -EINVAL;
2818 }
2819
2820 /* Erase all spilled pointers. */
2821 state->stack[spi].spilled_ptr.type = NOT_INIT;
2822
2823 /* Update the slot type. */
2824 new_type = STACK_MISC;
2825 if (writing_zero && *stype == STACK_ZERO) {
2826 new_type = STACK_ZERO;
2827 zero_used = true;
2828 }
2829 /* If the slot is STACK_INVALID, we check whether it's OK to
2830 * pretend that it will be initialized by this write. The slot
2831 * might not actually be written to, and so if we mark it as
2832 * initialized future reads might leak uninitialized memory.
2833 * For privileged programs, we will accept such reads to slots
2834 * that may or may not be written because, if we're reject
2835 * them, the error would be too confusing.
2836 */
2837 if (*stype == STACK_INVALID && !env->allow_uninit_stack) {
2838 verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d",
2839 insn_idx, i);
2840 return -EINVAL;
2841 }
2842 *stype = new_type;
2843 }
2844 if (zero_used) {
2845 /* backtracking doesn't work for STACK_ZERO yet. */
2846 err = mark_chain_precision(env, value_regno);
2847 if (err)
2848 return err;
2849 }
2850 return 0;
2851}
2852
2853/* When register 'dst_regno' is assigned some values from stack[min_off,
2854 * max_off), we set the register's type according to the types of the
2855 * respective stack slots. If all the stack values are known to be zeros, then
2856 * so is the destination reg. Otherwise, the register is considered to be
2857 * SCALAR. This function does not deal with register filling; the caller must
2858 * ensure that all spilled registers in the stack range have been marked as
2859 * read.
2860 */
2861static void mark_reg_stack_read(struct bpf_verifier_env *env,
2862 /* func where src register points to */
2863 struct bpf_func_state *ptr_state,
2864 int min_off, int max_off, int dst_regno)
2865{
2866 struct bpf_verifier_state *vstate = env->cur_state;
2867 struct bpf_func_state *state = vstate->frame[vstate->curframe];
2868 int i, slot, spi;
2869 u8 *stype;
2870 int zeros = 0;
2871
2872 for (i = min_off; i < max_off; i++) {
2873 slot = -i - 1;
2874 spi = slot / BPF_REG_SIZE;
2875 stype = ptr_state->stack[spi].slot_type;
2876 if (stype[slot % BPF_REG_SIZE] != STACK_ZERO)
2877 break;
2878 zeros++;
2879 }
2880 if (zeros == max_off - min_off) {
2881 /* any access_size read into register is zero extended,
2882 * so the whole register == const_zero
2883 */
2884 __mark_reg_const_zero(&state->regs[dst_regno]);
2885 /* backtracking doesn't support STACK_ZERO yet,
2886 * so mark it precise here, so that later
2887 * backtracking can stop here.
2888 * Backtracking may not need this if this register
2889 * doesn't participate in pointer adjustment.
2890 * Forward propagation of precise flag is not
2891 * necessary either. This mark is only to stop
2892 * backtracking. Any register that contributed
2893 * to const 0 was marked precise before spill.
2894 */
2895 state->regs[dst_regno].precise = true;
2896 } else {
2897 /* have read misc data from the stack */
2898 mark_reg_unknown(env, state->regs, dst_regno);
2899 }
2900 state->regs[dst_regno].live |= REG_LIVE_WRITTEN;
2901}
2902
2903/* Read the stack at 'off' and put the results into the register indicated by
2904 * 'dst_regno'. It handles reg filling if the addressed stack slot is a
2905 * spilled reg.
2906 *
2907 * 'dst_regno' can be -1, meaning that the read value is not going to a
2908 * register.
2909 *
2910 * The access is assumed to be within the current stack bounds.
2911 */
2912static int check_stack_read_fixed_off(struct bpf_verifier_env *env,
2913 /* func where src register points to */
2914 struct bpf_func_state *reg_state,
2915 int off, int size, int dst_regno)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002916{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002917 struct bpf_verifier_state *vstate = env->cur_state;
2918 struct bpf_func_state *state = vstate->frame[vstate->curframe];
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002919 int i, slot = -off - 1, spi = slot / BPF_REG_SIZE;
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002920 struct bpf_reg_state *reg;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002921 u8 *stype;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002922
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08002923 stype = reg_state->stack[spi].slot_type;
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002924 reg = &reg_state->stack[spi].spilled_ptr;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002925
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002926 if (stype[0] == STACK_SPILL) {
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002927 if (size != BPF_REG_SIZE) {
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002928 if (reg->type != SCALAR_VALUE) {
2929 verbose_linfo(env, env->insn_idx, "; ");
2930 verbose(env, "invalid size of register fill\n");
2931 return -EACCES;
2932 }
Andrei Matei01f810a2021-02-06 20:10:24 -05002933 if (dst_regno >= 0) {
2934 mark_reg_unknown(env, state->regs, dst_regno);
2935 state->regs[dst_regno].live |= REG_LIVE_WRITTEN;
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002936 }
2937 mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64);
2938 return 0;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002939 }
Alexei Starovoitov9c3997602014-10-28 15:11:41 -07002940 for (i = 1; i < BPF_REG_SIZE; i++) {
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07002941 if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07002942 verbose(env, "corrupted spill memory\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002943 return -EACCES;
2944 }
2945 }
2946
Andrei Matei01f810a2021-02-06 20:10:24 -05002947 if (dst_regno >= 0) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002948 /* restore register state from stack */
Andrei Matei01f810a2021-02-06 20:10:24 -05002949 state->regs[dst_regno] = *reg;
Alexei Starovoitov2f18f622017-11-30 21:31:38 -08002950 /* mark reg as written since spilled pointer state likely
2951 * has its liveness marks cleared by is_state_visited()
2952 * which resets stack/reg liveness for state transitions
2953 */
Andrei Matei01f810a2021-02-06 20:10:24 -05002954 state->regs[dst_regno].live |= REG_LIVE_WRITTEN;
Jann Horn6e7e63c2020-04-17 02:00:06 +02002955 } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) {
Andrei Matei01f810a2021-02-06 20:10:24 -05002956 /* If dst_regno==-1, the caller is asking us whether
Jann Horn6e7e63c2020-04-17 02:00:06 +02002957 * it is acceptable to use this value as a SCALAR_VALUE
2958 * (e.g. for XADD).
2959 * We must not allow unprivileged callers to do that
2960 * with spilled pointers.
2961 */
2962 verbose(env, "leaking pointer from stack off %d\n",
2963 off);
2964 return -EACCES;
Edward Creedc503a82017-08-15 20:34:35 +01002965 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002966 mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002967 } else {
Andrei Matei01f810a2021-02-06 20:10:24 -05002968 u8 type;
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002969
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002970 for (i = 0; i < size; i++) {
Andrei Matei01f810a2021-02-06 20:10:24 -05002971 type = stype[(slot - i) % BPF_REG_SIZE];
2972 if (type == STACK_MISC)
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002973 continue;
Andrei Matei01f810a2021-02-06 20:10:24 -05002974 if (type == STACK_ZERO)
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002975 continue;
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08002976 verbose(env, "invalid read from stack off %d+%d size %d\n",
2977 off, i, size);
2978 return -EACCES;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002979 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002980 mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64);
Andrei Matei01f810a2021-02-06 20:10:24 -05002981 if (dst_regno >= 0)
2982 mark_reg_stack_read(env, reg_state, off, off + size, dst_regno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002983 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07002984 return 0;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07002985}
2986
Andrei Matei01f810a2021-02-06 20:10:24 -05002987enum stack_access_src {
2988 ACCESS_DIRECT = 1, /* the access is performed by an instruction */
2989 ACCESS_HELPER = 2, /* the access is performed by a helper */
2990};
2991
2992static int check_stack_range_initialized(struct bpf_verifier_env *env,
2993 int regno, int off, int access_size,
2994 bool zero_size_allowed,
2995 enum stack_access_src type,
2996 struct bpf_call_arg_meta *meta);
2997
2998static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno)
Daniel Borkmanne4298d22019-01-03 00:58:31 +01002999{
Andrei Matei01f810a2021-02-06 20:10:24 -05003000 return cur_regs(env) + regno;
3001}
3002
3003/* Read the stack at 'ptr_regno + off' and put the result into the register
3004 * 'dst_regno'.
3005 * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'),
3006 * but not its variable offset.
3007 * 'size' is assumed to be <= reg size and the access is assumed to be aligned.
3008 *
3009 * As opposed to check_stack_read_fixed_off, this function doesn't deal with
3010 * filling registers (i.e. reads of spilled register cannot be detected when
3011 * the offset is not fixed). We conservatively mark 'dst_regno' as containing
3012 * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable
3013 * offset; for a fixed offset check_stack_read_fixed_off should be used
3014 * instead.
3015 */
3016static int check_stack_read_var_off(struct bpf_verifier_env *env,
3017 int ptr_regno, int off, int size, int dst_regno)
3018{
3019 /* The state of the source register. */
3020 struct bpf_reg_state *reg = reg_state(env, ptr_regno);
3021 struct bpf_func_state *ptr_state = func(env, reg);
3022 int err;
3023 int min_off, max_off;
3024
3025 /* Note that we pass a NULL meta, so raw access will not be permitted.
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003026 */
Andrei Matei01f810a2021-02-06 20:10:24 -05003027 err = check_stack_range_initialized(env, ptr_regno, off, size,
3028 false, ACCESS_DIRECT, NULL);
3029 if (err)
3030 return err;
3031
3032 min_off = reg->smin_value + off;
3033 max_off = reg->smax_value + off;
3034 mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno);
3035 return 0;
3036}
3037
3038/* check_stack_read dispatches to check_stack_read_fixed_off or
3039 * check_stack_read_var_off.
3040 *
3041 * The caller must ensure that the offset falls within the allocated stack
3042 * bounds.
3043 *
3044 * 'dst_regno' is a register which will receive the value from the stack. It
3045 * can be -1, meaning that the read value is not going to a register.
3046 */
3047static int check_stack_read(struct bpf_verifier_env *env,
3048 int ptr_regno, int off, int size,
3049 int dst_regno)
3050{
3051 struct bpf_reg_state *reg = reg_state(env, ptr_regno);
3052 struct bpf_func_state *state = func(env, reg);
3053 int err;
3054 /* Some accesses are only permitted with a static offset. */
3055 bool var_off = !tnum_is_const(reg->var_off);
3056
3057 /* The offset is required to be static when reads don't go to a
3058 * register, in order to not leak pointers (see
3059 * check_stack_read_fixed_off).
3060 */
3061 if (dst_regno < 0 && var_off) {
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003062 char tn_buf[48];
3063
3064 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Andrei Matei01f810a2021-02-06 20:10:24 -05003065 verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n",
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003066 tn_buf, off, size);
3067 return -EACCES;
3068 }
Andrei Matei01f810a2021-02-06 20:10:24 -05003069 /* Variable offset is prohibited for unprivileged mode for simplicity
3070 * since it requires corresponding support in Spectre masking for stack
3071 * ALU. See also retrieve_ptr_limit().
3072 */
3073 if (!env->bypass_spec_v1 && var_off) {
3074 char tn_buf[48];
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003075
Andrei Matei01f810a2021-02-06 20:10:24 -05003076 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3077 verbose(env, "R%d variable offset stack access prohibited for !root, var_off=%s\n",
3078 ptr_regno, tn_buf);
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003079 return -EACCES;
3080 }
3081
Andrei Matei01f810a2021-02-06 20:10:24 -05003082 if (!var_off) {
3083 off += reg->var_off.value;
3084 err = check_stack_read_fixed_off(env, state, off, size,
3085 dst_regno);
3086 } else {
3087 /* Variable offset stack reads need more conservative handling
3088 * than fixed offset ones. Note that dst_regno >= 0 on this
3089 * branch.
3090 */
3091 err = check_stack_read_var_off(env, ptr_regno, off, size,
3092 dst_regno);
3093 }
3094 return err;
3095}
3096
3097
3098/* check_stack_write dispatches to check_stack_write_fixed_off or
3099 * check_stack_write_var_off.
3100 *
3101 * 'ptr_regno' is the register used as a pointer into the stack.
3102 * 'off' includes 'ptr_regno->off', but not its variable offset (if any).
3103 * 'value_regno' is the register whose value we're writing to the stack. It can
3104 * be -1, meaning that we're not writing from a register.
3105 *
3106 * The caller must ensure that the offset falls within the maximum stack size.
3107 */
3108static int check_stack_write(struct bpf_verifier_env *env,
3109 int ptr_regno, int off, int size,
3110 int value_regno, int insn_idx)
3111{
3112 struct bpf_reg_state *reg = reg_state(env, ptr_regno);
3113 struct bpf_func_state *state = func(env, reg);
3114 int err;
3115
3116 if (tnum_is_const(reg->var_off)) {
3117 off += reg->var_off.value;
3118 err = check_stack_write_fixed_off(env, state, off, size,
3119 value_regno, insn_idx);
3120 } else {
3121 /* Variable offset stack reads need more conservative handling
3122 * than fixed offset ones.
3123 */
3124 err = check_stack_write_var_off(env, state,
3125 ptr_regno, off, size,
3126 value_regno, insn_idx);
3127 }
3128 return err;
Daniel Borkmanne4298d22019-01-03 00:58:31 +01003129}
3130
Daniel Borkmann591fe982019-04-09 23:20:05 +02003131static int check_map_access_type(struct bpf_verifier_env *env, u32 regno,
3132 int off, int size, enum bpf_access_type type)
3133{
3134 struct bpf_reg_state *regs = cur_regs(env);
3135 struct bpf_map *map = regs[regno].map_ptr;
3136 u32 cap = bpf_map_flags_to_cap(map);
3137
3138 if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) {
3139 verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n",
3140 map->value_size, off, size);
3141 return -EACCES;
3142 }
3143
3144 if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) {
3145 verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n",
3146 map->value_size, off, size);
3147 return -EACCES;
3148 }
3149
3150 return 0;
3151}
3152
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003153/* check read/write into memory region (e.g., map value, ringbuf sample, etc) */
3154static int __check_mem_access(struct bpf_verifier_env *env, int regno,
3155 int off, int size, u32 mem_size,
3156 bool zero_size_allowed)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003157{
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003158 bool size_ok = size > 0 || (size == 0 && zero_size_allowed);
3159 struct bpf_reg_state *reg;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003160
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003161 if (off >= 0 && size_ok && (u64)off + size <= mem_size)
3162 return 0;
3163
3164 reg = &cur_regs(env)[regno];
3165 switch (reg->type) {
Yonghong Song69c087b2021-02-26 12:49:25 -08003166 case PTR_TO_MAP_KEY:
3167 verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n",
3168 mem_size, off, size);
3169 break;
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003170 case PTR_TO_MAP_VALUE:
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003171 verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n",
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003172 mem_size, off, size);
3173 break;
3174 case PTR_TO_PACKET:
3175 case PTR_TO_PACKET_META:
3176 case PTR_TO_PACKET_END:
3177 verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
3178 off, size, regno, reg->id, off, mem_size);
3179 break;
3180 case PTR_TO_MEM:
3181 default:
3182 verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n",
3183 mem_size, off, size);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003184 }
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003185
3186 return -EACCES;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003187}
3188
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003189/* check read/write into a memory region with possible variable offset */
3190static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno,
3191 int off, int size, u32 mem_size,
3192 bool zero_size_allowed)
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003193{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003194 struct bpf_verifier_state *vstate = env->cur_state;
3195 struct bpf_func_state *state = vstate->frame[vstate->curframe];
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003196 struct bpf_reg_state *reg = &state->regs[regno];
3197 int err;
3198
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003199 /* We may have adjusted the register pointing to memory region, so we
Edward Creef1174f72017-08-07 15:26:19 +01003200 * need to try adding each of min_value and max_value to off
3201 * to make sure our theoretical access will be safe.
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003202 */
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07003203 if (env->log.level & BPF_LOG_LEVEL)
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003204 print_verifier_state(env, state);
Daniel Borkmannb7137c42019-01-03 00:58:33 +01003205
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003206 /* The minimum value is only important with signed
3207 * comparisons where we can't assume the floor of a
3208 * value is 0. If we are using signed variables for our
3209 * index'es we need to make sure that whatever we use
3210 * will have a set floor within our range.
3211 */
Daniel Borkmannb7137c42019-01-03 00:58:33 +01003212 if (reg->smin_value < 0 &&
3213 (reg->smin_value == S64_MIN ||
3214 (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) ||
3215 reg->smin_value + off < 0)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003216 verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003217 regno);
3218 return -EACCES;
3219 }
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003220 err = __check_mem_access(env, regno, reg->smin_value + off, size,
3221 mem_size, zero_size_allowed);
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003222 if (err) {
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003223 verbose(env, "R%d min value is outside of the allowed memory range\n",
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003224 regno);
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003225 return err;
3226 }
3227
Edward Creeb03c9f92017-08-07 15:26:36 +01003228 /* If we haven't set a max value then we need to bail since we can't be
3229 * sure we won't do bad things.
3230 * If reg->umax_value + off could overflow, treat that as unbounded too.
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003231 */
Edward Creeb03c9f92017-08-07 15:26:36 +01003232 if (reg->umax_value >= BPF_MAX_VAR_OFF) {
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003233 verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n",
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003234 regno);
3235 return -EACCES;
3236 }
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003237 err = __check_mem_access(env, regno, reg->umax_value + off, size,
3238 mem_size, zero_size_allowed);
3239 if (err) {
3240 verbose(env, "R%d max value is outside of the allowed memory range\n",
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003241 regno);
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003242 return err;
3243 }
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08003244
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003245 return 0;
3246}
3247
3248/* check read/write into a map element with possible variable offset */
3249static int check_map_access(struct bpf_verifier_env *env, u32 regno,
3250 int off, int size, bool zero_size_allowed)
3251{
3252 struct bpf_verifier_state *vstate = env->cur_state;
3253 struct bpf_func_state *state = vstate->frame[vstate->curframe];
3254 struct bpf_reg_state *reg = &state->regs[regno];
3255 struct bpf_map *map = reg->map_ptr;
3256 int err;
3257
3258 err = check_mem_region_access(env, regno, off, size, map->value_size,
3259 zero_size_allowed);
3260 if (err)
3261 return err;
3262
3263 if (map_value_has_spin_lock(map)) {
3264 u32 lock = map->spin_lock_off;
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08003265
3266 /* if any part of struct bpf_spin_lock can be touched by
3267 * load/store reject this program.
3268 * To check that [x1, x2) overlaps with [y1, y2)
3269 * it is sufficient to check x1 < y2 && y1 < x2.
3270 */
3271 if (reg->smin_value + off < lock + sizeof(struct bpf_spin_lock) &&
3272 lock < reg->umax_value + off + size) {
3273 verbose(env, "bpf_spin_lock cannot be accessed directly by load/store\n");
3274 return -EACCES;
3275 }
3276 }
Alexei Starovoitov68134662021-07-14 17:54:10 -07003277 if (map_value_has_timer(map)) {
3278 u32 t = map->timer_off;
3279
3280 if (reg->smin_value + off < t + sizeof(struct bpf_timer) &&
3281 t < reg->umax_value + off + size) {
3282 verbose(env, "bpf_timer cannot be accessed directly by load/store\n");
3283 return -EACCES;
3284 }
3285 }
Edward Creef1174f72017-08-07 15:26:19 +01003286 return err;
Gianluca Borellodbcfe5f2017-01-09 10:19:46 -08003287}
3288
Alexei Starovoitov969bf052016-05-05 19:49:10 -07003289#define MAX_PACKET_OFF 0xffff
3290
Udip Pant7e407812020-08-25 16:20:00 -07003291static enum bpf_prog_type resolve_prog_type(struct bpf_prog *prog)
3292{
Toke Høiland-Jørgensen3aac1ea2020-09-29 14:45:50 +02003293 return prog->aux->dst_prog ? prog->aux->dst_prog->type : prog->type;
Udip Pant7e407812020-08-25 16:20:00 -07003294}
3295
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01003296static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
Thomas Graf3a0af8f2016-11-30 17:10:10 +01003297 const struct bpf_call_arg_meta *meta,
3298 enum bpf_access_type t)
Brenden Blanco4acf6c02016-07-19 12:16:56 -07003299{
Udip Pant7e407812020-08-25 16:20:00 -07003300 enum bpf_prog_type prog_type = resolve_prog_type(env->prog);
3301
3302 switch (prog_type) {
Daniel Borkmann5d66fa72018-10-24 22:05:45 +02003303 /* Program types only with direct read access go here! */
Thomas Graf3a0af8f2016-11-30 17:10:10 +01003304 case BPF_PROG_TYPE_LWT_IN:
3305 case BPF_PROG_TYPE_LWT_OUT:
Mathieu Xhonneux004d4b22018-05-20 14:58:16 +01003306 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
Martin KaFai Lau2dbb9b92018-08-08 01:01:25 -07003307 case BPF_PROG_TYPE_SK_REUSEPORT:
Daniel Borkmann5d66fa72018-10-24 22:05:45 +02003308 case BPF_PROG_TYPE_FLOW_DISSECTOR:
Daniel Borkmannd5563d32018-10-24 22:05:46 +02003309 case BPF_PROG_TYPE_CGROUP_SKB:
Thomas Graf3a0af8f2016-11-30 17:10:10 +01003310 if (t == BPF_WRITE)
3311 return false;
Gustavo A. R. Silva87317452020-10-02 18:42:17 -05003312 fallthrough;
Daniel Borkmann5d66fa72018-10-24 22:05:45 +02003313
3314 /* Program types with direct read + write access go here! */
Daniel Borkmann36bbef52016-09-20 00:26:13 +02003315 case BPF_PROG_TYPE_SCHED_CLS:
3316 case BPF_PROG_TYPE_SCHED_ACT:
Brenden Blanco4acf6c02016-07-19 12:16:56 -07003317 case BPF_PROG_TYPE_XDP:
Thomas Graf3a0af8f2016-11-30 17:10:10 +01003318 case BPF_PROG_TYPE_LWT_XMIT:
John Fastabend8a31db52017-08-15 22:33:09 -07003319 case BPF_PROG_TYPE_SK_SKB:
John Fastabend4f738ad2018-03-18 12:57:10 -07003320 case BPF_PROG_TYPE_SK_MSG:
Daniel Borkmann36bbef52016-09-20 00:26:13 +02003321 if (meta)
3322 return meta->pkt_access;
3323
3324 env->seen_direct_write = true;
Brenden Blanco4acf6c02016-07-19 12:16:56 -07003325 return true;
Stanislav Fomichev0d01da62019-06-27 13:38:47 -07003326
3327 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3328 if (t == BPF_WRITE)
3329 env->seen_direct_write = true;
3330
3331 return true;
3332
Brenden Blanco4acf6c02016-07-19 12:16:56 -07003333 default:
3334 return false;
3335 }
3336}
3337
Edward Creef1174f72017-08-07 15:26:19 +01003338static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
Yonghong Song9fd29c02017-11-12 14:49:09 -08003339 int size, bool zero_size_allowed)
Edward Creef1174f72017-08-07 15:26:19 +01003340{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07003341 struct bpf_reg_state *regs = cur_regs(env);
Edward Creef1174f72017-08-07 15:26:19 +01003342 struct bpf_reg_state *reg = &regs[regno];
3343 int err;
3344
3345 /* We may have added a variable offset to the packet pointer; but any
3346 * reg->range we have comes after that. We are only checking the fixed
3347 * offset.
3348 */
3349
3350 /* We don't allow negative numbers, because we aren't tracking enough
3351 * detail to prove they're safe.
3352 */
Edward Creeb03c9f92017-08-07 15:26:36 +01003353 if (reg->smin_value < 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003354 verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
Edward Creef1174f72017-08-07 15:26:19 +01003355 regno);
3356 return -EACCES;
3357 }
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08003358
3359 err = reg->range < 0 ? -EINVAL :
3360 __check_mem_access(env, regno, off, size, reg->range,
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003361 zero_size_allowed);
Edward Creef1174f72017-08-07 15:26:19 +01003362 if (err) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003363 verbose(env, "R%d offset is outside of the packet\n", regno);
Edward Creef1174f72017-08-07 15:26:19 +01003364 return err;
3365 }
Jiong Wange6478152018-11-08 04:08:42 -05003366
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003367 /* __check_mem_access has made sure "off + size - 1" is within u16.
Jiong Wange6478152018-11-08 04:08:42 -05003368 * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff,
3369 * otherwise find_good_pkt_pointers would have refused to set range info
Andrii Nakryiko457f4432020-05-29 00:54:20 -07003370 * that __check_mem_access would have rejected this pkt access.
Jiong Wange6478152018-11-08 04:08:42 -05003371 * Therefore, "off + reg->umax_value + size - 1" won't overflow u32.
3372 */
3373 env->prog->aux->max_pkt_offset =
3374 max_t(u32, env->prog->aux->max_pkt_offset,
3375 off + reg->umax_value + size - 1);
3376
Edward Creef1174f72017-08-07 15:26:19 +01003377 return err;
3378}
3379
3380/* check access to 'struct bpf_context' fields. Supports fixed offsets only */
Yonghong Song31fd8582017-06-13 15:52:13 -07003381static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003382 enum bpf_access_type t, enum bpf_reg_type *reg_type,
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003383 struct btf **btf, u32 *btf_id)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003384{
Daniel Borkmannf96da092017-07-02 02:13:27 +02003385 struct bpf_insn_access_aux info = {
3386 .reg_type = *reg_type,
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003387 .log = &env->log,
Daniel Borkmannf96da092017-07-02 02:13:27 +02003388 };
Yonghong Song31fd8582017-06-13 15:52:13 -07003389
Jakub Kicinski4f9218a2017-10-16 16:40:55 -07003390 if (env->ops->is_valid_access &&
Andrey Ignatov5e43f892018-03-30 15:08:00 -07003391 env->ops->is_valid_access(off, size, t, env->prog, &info)) {
Daniel Borkmannf96da092017-07-02 02:13:27 +02003392 /* A non zero info.ctx_field_size indicates that this field is a
3393 * candidate for later verifier transformation to load the whole
3394 * field and then apply a mask when accessed with a narrower
3395 * access than actual ctx access size. A zero info.ctx_field_size
3396 * will only allow for whole field access and rejects any other
3397 * type of narrower access.
Yonghong Song31fd8582017-06-13 15:52:13 -07003398 */
Yonghong Song23994632017-06-22 15:07:39 -07003399 *reg_type = info.reg_type;
Yonghong Song31fd8582017-06-13 15:52:13 -07003400
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003401 if (*reg_type == PTR_TO_BTF_ID || *reg_type == PTR_TO_BTF_ID_OR_NULL) {
3402 *btf = info.btf;
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003403 *btf_id = info.btf_id;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003404 } else {
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003405 env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003406 }
Alexei Starovoitov32bbe002016-04-06 18:43:28 -07003407 /* remember the offset of last byte accessed in ctx */
3408 if (env->prog->aux->max_ctx_offset < off + size)
3409 env->prog->aux->max_ctx_offset = off + size;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003410 return 0;
Alexei Starovoitov32bbe002016-04-06 18:43:28 -07003411 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003412
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003413 verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07003414 return -EACCES;
3415}
3416
Petar Penkovd58e4682018-09-14 07:46:18 -07003417static int check_flow_keys_access(struct bpf_verifier_env *env, int off,
3418 int size)
3419{
3420 if (size < 0 || off < 0 ||
3421 (u64)off + size > sizeof(struct bpf_flow_keys)) {
3422 verbose(env, "invalid access to flow keys off=%d size=%d\n",
3423 off, size);
3424 return -EACCES;
3425 }
3426 return 0;
3427}
3428
Martin KaFai Lau5f456642019-02-08 22:25:54 -08003429static int check_sock_access(struct bpf_verifier_env *env, int insn_idx,
3430 u32 regno, int off, int size,
3431 enum bpf_access_type t)
Joe Stringerc64b7982018-10-02 13:35:33 -07003432{
3433 struct bpf_reg_state *regs = cur_regs(env);
3434 struct bpf_reg_state *reg = &regs[regno];
Martin KaFai Lau5f456642019-02-08 22:25:54 -08003435 struct bpf_insn_access_aux info = {};
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003436 bool valid;
Joe Stringerc64b7982018-10-02 13:35:33 -07003437
3438 if (reg->smin_value < 0) {
3439 verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3440 regno);
3441 return -EACCES;
3442 }
3443
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003444 switch (reg->type) {
3445 case PTR_TO_SOCK_COMMON:
3446 valid = bpf_sock_common_is_valid_access(off, size, t, &info);
3447 break;
3448 case PTR_TO_SOCKET:
3449 valid = bpf_sock_is_valid_access(off, size, t, &info);
3450 break;
Martin KaFai Lau655a51e2019-02-09 23:22:24 -08003451 case PTR_TO_TCP_SOCK:
3452 valid = bpf_tcp_sock_is_valid_access(off, size, t, &info);
3453 break;
Jonathan Lemonfada7fd2019-06-06 13:59:40 -07003454 case PTR_TO_XDP_SOCK:
3455 valid = bpf_xdp_sock_is_valid_access(off, size, t, &info);
3456 break;
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003457 default:
3458 valid = false;
Joe Stringerc64b7982018-10-02 13:35:33 -07003459 }
3460
Martin KaFai Lau5f456642019-02-08 22:25:54 -08003461
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003462 if (valid) {
3463 env->insn_aux_data[insn_idx].ctx_field_size =
3464 info.ctx_field_size;
3465 return 0;
3466 }
3467
3468 verbose(env, "R%d invalid %s access off=%d size=%d\n",
3469 regno, reg_type_str[reg->type], off, size);
3470
3471 return -EACCES;
Joe Stringerc64b7982018-10-02 13:35:33 -07003472}
3473
Daniel Borkmann4cabc5b2017-07-21 00:00:21 +02003474static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
3475{
Daniel Borkmann2a159c62018-10-21 02:09:24 +02003476 return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno));
Daniel Borkmann4cabc5b2017-07-21 00:00:21 +02003477}
3478
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +01003479static bool is_ctx_reg(struct bpf_verifier_env *env, int regno)
3480{
Daniel Borkmann2a159c62018-10-21 02:09:24 +02003481 const struct bpf_reg_state *reg = reg_state(env, regno);
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +01003482
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003483 return reg->type == PTR_TO_CTX;
3484}
3485
3486static bool is_sk_reg(struct bpf_verifier_env *env, int regno)
3487{
3488 const struct bpf_reg_state *reg = reg_state(env, regno);
3489
3490 return type_is_sk_pointer(reg->type);
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +01003491}
3492
Daniel Borkmannca369602018-02-23 22:29:05 +01003493static bool is_pkt_reg(struct bpf_verifier_env *env, int regno)
3494{
Daniel Borkmann2a159c62018-10-21 02:09:24 +02003495 const struct bpf_reg_state *reg = reg_state(env, regno);
Daniel Borkmannca369602018-02-23 22:29:05 +01003496
3497 return type_is_pkt_pointer(reg->type);
3498}
3499
Daniel Borkmann4b5defd2018-10-21 02:09:25 +02003500static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno)
3501{
3502 const struct bpf_reg_state *reg = reg_state(env, regno);
3503
3504 /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */
3505 return reg->type == PTR_TO_FLOW_KEYS;
3506}
3507
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003508static int check_pkt_ptr_alignment(struct bpf_verifier_env *env,
3509 const struct bpf_reg_state *reg,
David S. Millerd1174412017-05-10 11:22:52 -07003510 int off, int size, bool strict)
Alexei Starovoitov969bf052016-05-05 19:49:10 -07003511{
Edward Creef1174f72017-08-07 15:26:19 +01003512 struct tnum reg_off;
David S. Millere07b98d2017-05-10 11:38:07 -07003513 int ip_align;
David S. Millerd1174412017-05-10 11:22:52 -07003514
3515 /* Byte size accesses are always allowed. */
3516 if (!strict || size == 1)
3517 return 0;
3518
David S. Millere4eda882017-05-22 12:27:07 -04003519 /* For platforms that do not have a Kconfig enabling
3520 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
3521 * NET_IP_ALIGN is universally set to '2'. And on platforms
3522 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
3523 * to this code only in strict mode where we want to emulate
3524 * the NET_IP_ALIGN==2 checking. Therefore use an
3525 * unconditional IP align value of '2'.
David S. Millere07b98d2017-05-10 11:38:07 -07003526 */
David S. Millere4eda882017-05-22 12:27:07 -04003527 ip_align = 2;
Edward Creef1174f72017-08-07 15:26:19 +01003528
3529 reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off));
3530 if (!tnum_is_aligned(reg_off, size)) {
3531 char tn_buf[48];
3532
3533 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003534 verbose(env,
3535 "misaligned packet access off %d+%s+%d+%d size %d\n",
Edward Creef1174f72017-08-07 15:26:19 +01003536 ip_align, tn_buf, reg->off, off, size);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07003537 return -EACCES;
3538 }
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003539
Alexei Starovoitov969bf052016-05-05 19:49:10 -07003540 return 0;
3541}
3542
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003543static int check_generic_ptr_alignment(struct bpf_verifier_env *env,
3544 const struct bpf_reg_state *reg,
Edward Creef1174f72017-08-07 15:26:19 +01003545 const char *pointer_desc,
3546 int off, int size, bool strict)
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003547{
Edward Creef1174f72017-08-07 15:26:19 +01003548 struct tnum reg_off;
3549
3550 /* Byte size accesses are always allowed. */
3551 if (!strict || size == 1)
3552 return 0;
3553
3554 reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off));
3555 if (!tnum_is_aligned(reg_off, size)) {
3556 char tn_buf[48];
3557
3558 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003559 verbose(env, "misaligned %saccess off %s+%d+%d size %d\n",
Edward Creef1174f72017-08-07 15:26:19 +01003560 pointer_desc, tn_buf, reg->off, off, size);
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003561 return -EACCES;
3562 }
3563
3564 return 0;
3565}
3566
David S. Millere07b98d2017-05-10 11:38:07 -07003567static int check_ptr_alignment(struct bpf_verifier_env *env,
Daniel Borkmannca369602018-02-23 22:29:05 +01003568 const struct bpf_reg_state *reg, int off,
3569 int size, bool strict_alignment_once)
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003570{
Daniel Borkmannca369602018-02-23 22:29:05 +01003571 bool strict = env->strict_alignment || strict_alignment_once;
Edward Creef1174f72017-08-07 15:26:19 +01003572 const char *pointer_desc = "";
David S. Millerd1174412017-05-10 11:22:52 -07003573
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003574 switch (reg->type) {
3575 case PTR_TO_PACKET:
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02003576 case PTR_TO_PACKET_META:
3577 /* Special case, because of NET_IP_ALIGN. Given metadata sits
3578 * right in front, treat it the very same way.
3579 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003580 return check_pkt_ptr_alignment(env, reg, off, size, strict);
Petar Penkovd58e4682018-09-14 07:46:18 -07003581 case PTR_TO_FLOW_KEYS:
3582 pointer_desc = "flow keys ";
3583 break;
Yonghong Song69c087b2021-02-26 12:49:25 -08003584 case PTR_TO_MAP_KEY:
3585 pointer_desc = "key ";
3586 break;
Edward Creef1174f72017-08-07 15:26:19 +01003587 case PTR_TO_MAP_VALUE:
3588 pointer_desc = "value ";
3589 break;
3590 case PTR_TO_CTX:
3591 pointer_desc = "context ";
3592 break;
3593 case PTR_TO_STACK:
3594 pointer_desc = "stack ";
Andrei Matei01f810a2021-02-06 20:10:24 -05003595 /* The stack spill tracking logic in check_stack_write_fixed_off()
3596 * and check_stack_read_fixed_off() relies on stack accesses being
Jann Horna5ec6ae2017-12-18 20:11:58 -08003597 * aligned.
3598 */
3599 strict = true;
Edward Creef1174f72017-08-07 15:26:19 +01003600 break;
Joe Stringerc64b7982018-10-02 13:35:33 -07003601 case PTR_TO_SOCKET:
3602 pointer_desc = "sock ";
3603 break;
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08003604 case PTR_TO_SOCK_COMMON:
3605 pointer_desc = "sock_common ";
3606 break;
Martin KaFai Lau655a51e2019-02-09 23:22:24 -08003607 case PTR_TO_TCP_SOCK:
3608 pointer_desc = "tcp_sock ";
3609 break;
Jonathan Lemonfada7fd2019-06-06 13:59:40 -07003610 case PTR_TO_XDP_SOCK:
3611 pointer_desc = "xdp_sock ";
3612 break;
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003613 default:
Edward Creef1174f72017-08-07 15:26:19 +01003614 break;
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003615 }
Jakub Kicinski61bd5212017-10-09 10:30:11 -07003616 return check_generic_ptr_alignment(env, reg, pointer_desc, off, size,
3617 strict);
Daniel Borkmann79adffc2017-03-31 02:24:03 +02003618}
3619
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003620static int update_stack_depth(struct bpf_verifier_env *env,
3621 const struct bpf_func_state *func,
3622 int off)
3623{
Jiong Wang9c8105b2018-05-02 16:17:18 -04003624 u16 stack = env->subprog_info[func->subprogno].stack_depth;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003625
3626 if (stack >= -off)
3627 return 0;
3628
3629 /* update known max for given subprogram */
Jiong Wang9c8105b2018-05-02 16:17:18 -04003630 env->subprog_info[func->subprogno].stack_depth = -off;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003631 return 0;
3632}
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003633
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003634/* starting from main bpf function walk all instructions of the function
3635 * and recursively walk all callees that given function can call.
3636 * Ignore jump and exit insns.
3637 * Since recursion is prevented by check_cfg() this algorithm
3638 * only needs a local stack of MAX_CALL_FRAMES to remember callsites
3639 */
3640static int check_max_stack_depth(struct bpf_verifier_env *env)
3641{
Jiong Wang9c8105b2018-05-02 16:17:18 -04003642 int depth = 0, frame = 0, idx = 0, i = 0, subprog_end;
3643 struct bpf_subprog_info *subprog = env->subprog_info;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003644 struct bpf_insn *insn = env->prog->insnsi;
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +02003645 bool tail_call_reachable = false;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003646 int ret_insn[MAX_CALL_FRAMES];
3647 int ret_prog[MAX_CALL_FRAMES];
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +02003648 int j;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003649
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003650process_func:
Maciej Fijalkowski7f6e4312020-09-16 23:10:07 +02003651 /* protect against potential stack overflow that might happen when
3652 * bpf2bpf calls get combined with tailcalls. Limit the caller's stack
3653 * depth for such case down to 256 so that the worst case scenario
3654 * would result in 8k stack size (32 which is tailcall limit * 256 =
3655 * 8k).
3656 *
3657 * To get the idea what might happen, see an example:
3658 * func1 -> sub rsp, 128
3659 * subfunc1 -> sub rsp, 256
3660 * tailcall1 -> add rsp, 256
3661 * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320)
3662 * subfunc2 -> sub rsp, 64
3663 * subfunc22 -> sub rsp, 128
3664 * tailcall2 -> add rsp, 128
3665 * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416)
3666 *
3667 * tailcall will unwind the current stack frame but it will not get rid
3668 * of caller's stack as shown on the example above.
3669 */
3670 if (idx && subprog[idx].has_tail_call && depth >= 256) {
3671 verbose(env,
3672 "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n",
3673 depth);
3674 return -EACCES;
3675 }
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003676 /* round up to 32-bytes, since this is granularity
3677 * of interpreter stack size
3678 */
Jiong Wang9c8105b2018-05-02 16:17:18 -04003679 depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32);
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003680 if (depth > MAX_BPF_STACK) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003681 verbose(env, "combined stack size of %d calls is %d. Too large\n",
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003682 frame + 1, depth);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003683 return -EACCES;
3684 }
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003685continue_func:
Jiong Wang4cb3d992018-05-02 16:17:19 -04003686 subprog_end = subprog[idx + 1].start;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003687 for (; i < subprog_end; i++) {
Alexei Starovoitov7ddc80a2021-07-14 17:54:15 -07003688 int next_insn;
3689
Yonghong Song69c087b2021-02-26 12:49:25 -08003690 if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i))
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003691 continue;
3692 /* remember insn and function to return to */
3693 ret_insn[frame] = i + 1;
Jiong Wang9c8105b2018-05-02 16:17:18 -04003694 ret_prog[frame] = idx;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003695
3696 /* find the callee */
Alexei Starovoitov7ddc80a2021-07-14 17:54:15 -07003697 next_insn = i + insn[i].imm + 1;
3698 idx = find_subprog(env, next_insn);
Jiong Wang9c8105b2018-05-02 16:17:18 -04003699 if (idx < 0) {
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003700 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
Alexei Starovoitov7ddc80a2021-07-14 17:54:15 -07003701 next_insn);
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003702 return -EFAULT;
3703 }
Alexei Starovoitov7ddc80a2021-07-14 17:54:15 -07003704 if (subprog[idx].is_async_cb) {
3705 if (subprog[idx].has_tail_call) {
3706 verbose(env, "verifier bug. subprog has tail_call and async cb\n");
3707 return -EFAULT;
3708 }
3709 /* async callbacks don't increase bpf prog stack size */
3710 continue;
3711 }
3712 i = next_insn;
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +02003713
3714 if (subprog[idx].has_tail_call)
3715 tail_call_reachable = true;
3716
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003717 frame++;
3718 if (frame >= MAX_CALL_FRAMES) {
Paul Chaignon927cb782019-03-20 13:58:27 +01003719 verbose(env, "the call stack of %d frames is too deep !\n",
3720 frame);
3721 return -E2BIG;
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003722 }
3723 goto process_func;
3724 }
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +02003725 /* if tail call got detected across bpf2bpf calls then mark each of the
3726 * currently present subprog frames as tail call reachable subprogs;
3727 * this info will be utilized by JIT so that we will be preserving the
3728 * tail call counter throughout bpf2bpf calls combined with tailcalls
3729 */
3730 if (tail_call_reachable)
3731 for (j = 0; j < frame; j++)
3732 subprog[ret_prog[j]].tail_call_reachable = true;
Daniel Borkmann5dd0a6b2021-07-12 22:57:35 +02003733 if (subprog[0].tail_call_reachable)
3734 env->prog->aux->tail_call_reachable = true;
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +02003735
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003736 /* end of for() loop means the last insn of the 'subprog'
3737 * was reached. Doesn't matter whether it was JA or EXIT
3738 */
3739 if (frame == 0)
3740 return 0;
Jiong Wang9c8105b2018-05-02 16:17:18 -04003741 depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32);
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003742 frame--;
3743 i = ret_insn[frame];
Jiong Wang9c8105b2018-05-02 16:17:18 -04003744 idx = ret_prog[frame];
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -08003745 goto continue_func;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08003746}
3747
David S. Miller19d28fb2018-01-11 21:27:54 -05003748#ifndef CONFIG_BPF_JIT_ALWAYS_ON
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -08003749static int get_callee_stack_depth(struct bpf_verifier_env *env,
3750 const struct bpf_insn *insn, int idx)
3751{
3752 int start = idx + insn->imm + 1, subprog;
3753
3754 subprog = find_subprog(env, start);
3755 if (subprog < 0) {
3756 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
3757 start);
3758 return -EFAULT;
3759 }
Jiong Wang9c8105b2018-05-02 16:17:18 -04003760 return env->subprog_info[subprog].stack_depth;
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -08003761}
David S. Miller19d28fb2018-01-11 21:27:54 -05003762#endif
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -08003763
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08003764int check_ctx_reg(struct bpf_verifier_env *env,
3765 const struct bpf_reg_state *reg, int regno)
Daniel Borkmann58990d12018-06-07 17:40:03 +02003766{
3767 /* Access to ctx or passing it to a helper is only allowed in
3768 * its original, unmodified form.
3769 */
3770
3771 if (reg->off) {
3772 verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n",
3773 regno, reg->off);
3774 return -EACCES;
3775 }
3776
3777 if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
3778 char tn_buf[48];
3779
3780 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3781 verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf);
3782 return -EACCES;
3783 }
3784
3785 return 0;
3786}
3787
Yonghong Songafbf21d2020-07-23 11:41:11 -07003788static int __check_buffer_access(struct bpf_verifier_env *env,
3789 const char *buf_info,
3790 const struct bpf_reg_state *reg,
3791 int regno, int off, int size)
Matt Mullins9df1c282019-04-26 11:49:47 -07003792{
3793 if (off < 0) {
3794 verbose(env,
Yonghong Song4fc00b72020-07-28 15:18:01 -07003795 "R%d invalid %s buffer access: off=%d, size=%d\n",
Yonghong Songafbf21d2020-07-23 11:41:11 -07003796 regno, buf_info, off, size);
Matt Mullins9df1c282019-04-26 11:49:47 -07003797 return -EACCES;
3798 }
3799 if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
3800 char tn_buf[48];
3801
3802 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3803 verbose(env,
Yonghong Song4fc00b72020-07-28 15:18:01 -07003804 "R%d invalid variable buffer offset: off=%d, var_off=%s\n",
Matt Mullins9df1c282019-04-26 11:49:47 -07003805 regno, off, tn_buf);
3806 return -EACCES;
3807 }
Yonghong Songafbf21d2020-07-23 11:41:11 -07003808
3809 return 0;
3810}
3811
3812static int check_tp_buffer_access(struct bpf_verifier_env *env,
3813 const struct bpf_reg_state *reg,
3814 int regno, int off, int size)
3815{
3816 int err;
3817
3818 err = __check_buffer_access(env, "tracepoint", reg, regno, off, size);
3819 if (err)
3820 return err;
3821
Matt Mullins9df1c282019-04-26 11:49:47 -07003822 if (off + size > env->prog->aux->max_tp_access)
3823 env->prog->aux->max_tp_access = off + size;
3824
3825 return 0;
3826}
3827
Yonghong Songafbf21d2020-07-23 11:41:11 -07003828static int check_buffer_access(struct bpf_verifier_env *env,
3829 const struct bpf_reg_state *reg,
3830 int regno, int off, int size,
3831 bool zero_size_allowed,
3832 const char *buf_info,
3833 u32 *max_access)
3834{
3835 int err;
3836
3837 err = __check_buffer_access(env, buf_info, reg, regno, off, size);
3838 if (err)
3839 return err;
3840
3841 if (off + size > *max_access)
3842 *max_access = off + size;
3843
3844 return 0;
3845}
3846
John Fastabend3f50f132020-03-30 14:36:39 -07003847/* BPF architecture zero extends alu32 ops into 64-bit registesr */
3848static void zext_32_to_64(struct bpf_reg_state *reg)
3849{
3850 reg->var_off = tnum_subreg(reg->var_off);
3851 __reg_assign_32_into_64(reg);
3852}
Matt Mullins9df1c282019-04-26 11:49:47 -07003853
Jann Horn0c17d1d2017-12-18 20:11:55 -08003854/* truncate register to smaller size (in bytes)
3855 * must be called with size < BPF_REG_SIZE
3856 */
3857static void coerce_reg_to_size(struct bpf_reg_state *reg, int size)
3858{
3859 u64 mask;
3860
3861 /* clear high bits in bit representation */
3862 reg->var_off = tnum_cast(reg->var_off, size);
3863
3864 /* fix arithmetic bounds */
3865 mask = ((u64)1 << (size * 8)) - 1;
3866 if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) {
3867 reg->umin_value &= mask;
3868 reg->umax_value &= mask;
3869 } else {
3870 reg->umin_value = 0;
3871 reg->umax_value = mask;
3872 }
3873 reg->smin_value = reg->umin_value;
3874 reg->smax_value = reg->umax_value;
John Fastabend3f50f132020-03-30 14:36:39 -07003875
3876 /* If size is smaller than 32bit register the 32bit register
3877 * values are also truncated so we push 64-bit bounds into
3878 * 32-bit bounds. Above were truncated < 32-bits already.
3879 */
3880 if (size >= 4)
3881 return;
3882 __reg_combine_64_into_32(reg);
Jann Horn0c17d1d2017-12-18 20:11:55 -08003883}
3884
Andrii Nakryikoa23740e2019-10-09 13:14:57 -07003885static bool bpf_map_is_rdonly(const struct bpf_map *map)
3886{
3887 return (map->map_flags & BPF_F_RDONLY_PROG) && map->frozen;
3888}
3889
3890static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val)
3891{
3892 void *ptr;
3893 u64 addr;
3894 int err;
3895
3896 err = map->ops->map_direct_value_addr(map, &addr, off);
3897 if (err)
3898 return err;
Andrii Nakryiko2dedd7d2019-10-11 10:20:53 -07003899 ptr = (void *)(long)addr + off;
Andrii Nakryikoa23740e2019-10-09 13:14:57 -07003900
3901 switch (size) {
3902 case sizeof(u8):
3903 *val = (u64)*(u8 *)ptr;
3904 break;
3905 case sizeof(u16):
3906 *val = (u64)*(u16 *)ptr;
3907 break;
3908 case sizeof(u32):
3909 *val = (u64)*(u32 *)ptr;
3910 break;
3911 case sizeof(u64):
3912 *val = *(u64 *)ptr;
3913 break;
3914 default:
3915 return -EINVAL;
3916 }
3917 return 0;
3918}
3919
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003920static int check_ptr_to_btf_access(struct bpf_verifier_env *env,
3921 struct bpf_reg_state *regs,
3922 int regno, int off, int size,
3923 enum bpf_access_type atype,
3924 int value_regno)
3925{
3926 struct bpf_reg_state *reg = regs + regno;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003927 const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id);
3928 const char *tname = btf_name_by_offset(reg->btf, t->name_off);
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003929 u32 btf_id;
3930 int ret;
3931
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003932 if (off < 0) {
3933 verbose(env,
3934 "R%d is ptr_%s invalid negative access: off=%d\n",
3935 regno, tname, off);
3936 return -EACCES;
3937 }
3938 if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
3939 char tn_buf[48];
3940
3941 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3942 verbose(env,
3943 "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n",
3944 regno, tname, off, tn_buf);
3945 return -EACCES;
3946 }
3947
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08003948 if (env->ops->btf_struct_access) {
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003949 ret = env->ops->btf_struct_access(&env->log, reg->btf, t,
3950 off, size, atype, &btf_id);
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08003951 } else {
3952 if (atype != BPF_READ) {
3953 verbose(env, "only read is supported\n");
3954 return -EACCES;
3955 }
3956
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003957 ret = btf_struct_access(&env->log, reg->btf, t, off, size,
3958 atype, &btf_id);
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08003959 }
3960
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003961 if (ret < 0)
3962 return ret;
3963
Andrey Ignatov41c48f32020-06-19 14:11:43 -07003964 if (atype == BPF_READ && value_regno >= 0)
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08003965 mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id);
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08003966
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07003967 return 0;
3968}
3969
Andrey Ignatov41c48f32020-06-19 14:11:43 -07003970static int check_ptr_to_map_access(struct bpf_verifier_env *env,
3971 struct bpf_reg_state *regs,
3972 int regno, int off, int size,
3973 enum bpf_access_type atype,
3974 int value_regno)
3975{
3976 struct bpf_reg_state *reg = regs + regno;
3977 struct bpf_map *map = reg->map_ptr;
3978 const struct btf_type *t;
3979 const char *tname;
3980 u32 btf_id;
3981 int ret;
3982
3983 if (!btf_vmlinux) {
3984 verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n");
3985 return -ENOTSUPP;
3986 }
3987
3988 if (!map->ops->map_btf_id || !*map->ops->map_btf_id) {
3989 verbose(env, "map_ptr access not supported for map type %d\n",
3990 map->map_type);
3991 return -ENOTSUPP;
3992 }
3993
3994 t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id);
3995 tname = btf_name_by_offset(btf_vmlinux, t->name_off);
3996
3997 if (!env->allow_ptr_to_map_access) {
3998 verbose(env,
3999 "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n",
4000 tname);
4001 return -EPERM;
4002 }
4003
4004 if (off < 0) {
4005 verbose(env, "R%d is %s invalid negative access: off=%d\n",
4006 regno, tname, off);
4007 return -EACCES;
4008 }
4009
4010 if (atype != BPF_READ) {
4011 verbose(env, "only read from %s is supported\n", tname);
4012 return -EACCES;
4013 }
4014
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004015 ret = btf_struct_access(&env->log, btf_vmlinux, t, off, size, atype, &btf_id);
Andrey Ignatov41c48f32020-06-19 14:11:43 -07004016 if (ret < 0)
4017 return ret;
4018
4019 if (value_regno >= 0)
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004020 mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id);
Andrey Ignatov41c48f32020-06-19 14:11:43 -07004021
4022 return 0;
4023}
4024
Andrei Matei01f810a2021-02-06 20:10:24 -05004025/* Check that the stack access at the given offset is within bounds. The
4026 * maximum valid offset is -1.
4027 *
4028 * The minimum valid offset is -MAX_BPF_STACK for writes, and
4029 * -state->allocated_stack for reads.
4030 */
4031static int check_stack_slot_within_bounds(int off,
4032 struct bpf_func_state *state,
4033 enum bpf_access_type t)
4034{
4035 int min_valid_off;
4036
4037 if (t == BPF_WRITE)
4038 min_valid_off = -MAX_BPF_STACK;
4039 else
4040 min_valid_off = -state->allocated_stack;
4041
4042 if (off < min_valid_off || off > -1)
4043 return -EACCES;
4044 return 0;
4045}
4046
4047/* Check that the stack access at 'regno + off' falls within the maximum stack
4048 * bounds.
4049 *
4050 * 'off' includes `regno->offset`, but not its dynamic part (if any).
4051 */
4052static int check_stack_access_within_bounds(
4053 struct bpf_verifier_env *env,
4054 int regno, int off, int access_size,
4055 enum stack_access_src src, enum bpf_access_type type)
4056{
4057 struct bpf_reg_state *regs = cur_regs(env);
4058 struct bpf_reg_state *reg = regs + regno;
4059 struct bpf_func_state *state = func(env, reg);
4060 int min_off, max_off;
4061 int err;
4062 char *err_extra;
4063
4064 if (src == ACCESS_HELPER)
4065 /* We don't know if helpers are reading or writing (or both). */
4066 err_extra = " indirect access to";
4067 else if (type == BPF_READ)
4068 err_extra = " read from";
4069 else
4070 err_extra = " write to";
4071
4072 if (tnum_is_const(reg->var_off)) {
4073 min_off = reg->var_off.value + off;
4074 if (access_size > 0)
4075 max_off = min_off + access_size - 1;
4076 else
4077 max_off = min_off;
4078 } else {
4079 if (reg->smax_value >= BPF_MAX_VAR_OFF ||
4080 reg->smin_value <= -BPF_MAX_VAR_OFF) {
4081 verbose(env, "invalid unbounded variable-offset%s stack R%d\n",
4082 err_extra, regno);
4083 return -EACCES;
4084 }
4085 min_off = reg->smin_value + off;
4086 if (access_size > 0)
4087 max_off = reg->smax_value + off + access_size - 1;
4088 else
4089 max_off = min_off;
4090 }
4091
4092 err = check_stack_slot_within_bounds(min_off, state, type);
4093 if (!err)
4094 err = check_stack_slot_within_bounds(max_off, state, type);
4095
4096 if (err) {
4097 if (tnum_is_const(reg->var_off)) {
4098 verbose(env, "invalid%s stack R%d off=%d size=%d\n",
4099 err_extra, regno, off, access_size);
4100 } else {
4101 char tn_buf[48];
4102
4103 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
4104 verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n",
4105 err_extra, regno, tn_buf, access_size);
4106 }
4107 }
4108 return err;
4109}
Andrey Ignatov41c48f32020-06-19 14:11:43 -07004110
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004111/* check whether memory at (regno + off) is accessible for t = (read | write)
4112 * if t==write, value_regno is a register which value is stored into memory
4113 * if t==read, value_regno is a register which will receive the value from memory
4114 * if t==write && value_regno==-1, some unknown value is stored into memory
4115 * if t==read && value_regno==-1, don't care what we read from memory
4116 */
Daniel Borkmannca369602018-02-23 22:29:05 +01004117static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno,
4118 int off, int bpf_size, enum bpf_access_type t,
4119 int value_regno, bool strict_alignment_once)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004120{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004121 struct bpf_reg_state *regs = cur_regs(env);
4122 struct bpf_reg_state *reg = regs + regno;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08004123 struct bpf_func_state *state;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004124 int size, err = 0;
4125
4126 size = bpf_size_to_bytes(bpf_size);
4127 if (size < 0)
4128 return size;
4129
Edward Creef1174f72017-08-07 15:26:19 +01004130 /* alignment checks will add in reg->off themselves */
Daniel Borkmannca369602018-02-23 22:29:05 +01004131 err = check_ptr_alignment(env, reg, off, size, strict_alignment_once);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004132 if (err)
4133 return err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004134
Edward Creef1174f72017-08-07 15:26:19 +01004135 /* for access checks, reg->off is just part of off */
4136 off += reg->off;
4137
Yonghong Song69c087b2021-02-26 12:49:25 -08004138 if (reg->type == PTR_TO_MAP_KEY) {
4139 if (t == BPF_WRITE) {
4140 verbose(env, "write to change key R%d not allowed\n", regno);
4141 return -EACCES;
4142 }
4143
4144 err = check_mem_region_access(env, regno, off, size,
4145 reg->map_ptr->key_size, false);
4146 if (err)
4147 return err;
4148 if (value_regno >= 0)
4149 mark_reg_unknown(env, regs, value_regno);
4150 } else if (reg->type == PTR_TO_MAP_VALUE) {
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07004151 if (t == BPF_WRITE && value_regno >= 0 &&
4152 is_pointer_value(env, value_regno)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004153 verbose(env, "R%d leaks addr into map\n", value_regno);
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07004154 return -EACCES;
4155 }
Daniel Borkmann591fe982019-04-09 23:20:05 +02004156 err = check_map_access_type(env, regno, off, size, t);
4157 if (err)
4158 return err;
Yonghong Song9fd29c02017-11-12 14:49:09 -08004159 err = check_map_access(env, regno, off, size, false);
Andrii Nakryikoa23740e2019-10-09 13:14:57 -07004160 if (!err && t == BPF_READ && value_regno >= 0) {
4161 struct bpf_map *map = reg->map_ptr;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004162
Andrii Nakryikoa23740e2019-10-09 13:14:57 -07004163 /* if map is read-only, track its contents as scalars */
4164 if (tnum_is_const(reg->var_off) &&
4165 bpf_map_is_rdonly(map) &&
4166 map->ops->map_direct_value_addr) {
4167 int map_off = off + reg->var_off.value;
4168 u64 val = 0;
4169
4170 err = bpf_map_direct_read(map, map_off, size,
4171 &val);
4172 if (err)
4173 return err;
4174
4175 regs[value_regno].type = SCALAR_VALUE;
4176 __mark_reg_known(&regs[value_regno], val);
4177 } else {
4178 mark_reg_unknown(env, regs, value_regno);
4179 }
4180 }
Andrii Nakryiko457f4432020-05-29 00:54:20 -07004181 } else if (reg->type == PTR_TO_MEM) {
4182 if (t == BPF_WRITE && value_regno >= 0 &&
4183 is_pointer_value(env, value_regno)) {
4184 verbose(env, "R%d leaks addr into mem\n", value_regno);
4185 return -EACCES;
4186 }
4187 err = check_mem_region_access(env, regno, off, size,
4188 reg->mem_size, false);
4189 if (!err && t == BPF_READ && value_regno >= 0)
4190 mark_reg_unknown(env, regs, value_regno);
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07004191 } else if (reg->type == PTR_TO_CTX) {
Edward Creef1174f72017-08-07 15:26:19 +01004192 enum bpf_reg_type reg_type = SCALAR_VALUE;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004193 struct btf *btf = NULL;
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07004194 u32 btf_id = 0;
Alexei Starovoitov19de99f2016-06-15 18:25:38 -07004195
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07004196 if (t == BPF_WRITE && value_regno >= 0 &&
4197 is_pointer_value(env, value_regno)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004198 verbose(env, "R%d leaks addr into ctx\n", value_regno);
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07004199 return -EACCES;
4200 }
Edward Creef1174f72017-08-07 15:26:19 +01004201
Daniel Borkmann58990d12018-06-07 17:40:03 +02004202 err = check_ctx_reg(env, reg, regno);
4203 if (err < 0)
4204 return err;
4205
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004206 err = check_ctx_access(env, insn_idx, off, size, t, &reg_type, &btf, &btf_id);
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07004207 if (err)
4208 verbose_linfo(env, insn_idx, "; ");
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004209 if (!err && t == BPF_READ && value_regno >= 0) {
Edward Creef1174f72017-08-07 15:26:19 +01004210 /* ctx access returns either a scalar, or a
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02004211 * PTR_TO_PACKET[_META,_END]. In the latter
4212 * case, we know the offset is zero.
Edward Creef1174f72017-08-07 15:26:19 +01004213 */
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004214 if (reg_type == SCALAR_VALUE) {
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004215 mark_reg_unknown(env, regs, value_regno);
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004216 } else {
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004217 mark_reg_known_zero(env, regs,
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004218 value_regno);
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004219 if (reg_type_may_be_null(reg_type))
4220 regs[value_regno].id = ++env->id_gen;
Jiong Wang5327ed32019-05-24 23:25:12 +01004221 /* A load of ctx field could have different
4222 * actual load size with the one encoded in the
4223 * insn. When the dst is PTR, it is for sure not
4224 * a sub-register.
4225 */
4226 regs[value_regno].subreg_def = DEF_NOT_SUBREG;
Yonghong Songb121b342020-05-09 10:59:12 -07004227 if (reg_type == PTR_TO_BTF_ID ||
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004228 reg_type == PTR_TO_BTF_ID_OR_NULL) {
4229 regs[value_regno].btf = btf;
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07004230 regs[value_regno].btf_id = btf_id;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004231 }
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004232 }
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004233 regs[value_regno].type = reg_type;
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004234 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004235
Edward Creef1174f72017-08-07 15:26:19 +01004236 } else if (reg->type == PTR_TO_STACK) {
Andrei Matei01f810a2021-02-06 20:10:24 -05004237 /* Basic bounds checks. */
4238 err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t);
Daniel Borkmanne4298d22019-01-03 00:58:31 +01004239 if (err)
4240 return err;
Alexei Starovoitov87266792017-05-30 13:31:29 -07004241
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08004242 state = func(env, reg);
4243 err = update_stack_depth(env, state, off);
4244 if (err)
4245 return err;
Alexei Starovoitov87266792017-05-30 13:31:29 -07004246
Andrei Matei01f810a2021-02-06 20:10:24 -05004247 if (t == BPF_READ)
4248 err = check_stack_read(env, regno, off, size,
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004249 value_regno);
Andrei Matei01f810a2021-02-06 20:10:24 -05004250 else
4251 err = check_stack_write(env, regno, off, size,
4252 value_regno, insn_idx);
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02004253 } else if (reg_is_pkt_pointer(reg)) {
Thomas Graf3a0af8f2016-11-30 17:10:10 +01004254 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004255 verbose(env, "cannot write into packet\n");
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004256 return -EACCES;
4257 }
Brenden Blanco4acf6c02016-07-19 12:16:56 -07004258 if (t == BPF_WRITE && value_regno >= 0 &&
4259 is_pointer_value(env, value_regno)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004260 verbose(env, "R%d leaks addr into packet\n",
4261 value_regno);
Brenden Blanco4acf6c02016-07-19 12:16:56 -07004262 return -EACCES;
4263 }
Yonghong Song9fd29c02017-11-12 14:49:09 -08004264 err = check_packet_access(env, regno, off, size, false);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004265 if (!err && t == BPF_READ && value_regno >= 0)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004266 mark_reg_unknown(env, regs, value_regno);
Petar Penkovd58e4682018-09-14 07:46:18 -07004267 } else if (reg->type == PTR_TO_FLOW_KEYS) {
4268 if (t == BPF_WRITE && value_regno >= 0 &&
4269 is_pointer_value(env, value_regno)) {
4270 verbose(env, "R%d leaks addr into flow keys\n",
4271 value_regno);
4272 return -EACCES;
4273 }
4274
4275 err = check_flow_keys_access(env, off, size);
4276 if (!err && t == BPF_READ && value_regno >= 0)
4277 mark_reg_unknown(env, regs, value_regno);
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004278 } else if (type_is_sk_pointer(reg->type)) {
Joe Stringerc64b7982018-10-02 13:35:33 -07004279 if (t == BPF_WRITE) {
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004280 verbose(env, "R%d cannot write into %s\n",
4281 regno, reg_type_str[reg->type]);
Joe Stringerc64b7982018-10-02 13:35:33 -07004282 return -EACCES;
4283 }
Martin KaFai Lau5f456642019-02-08 22:25:54 -08004284 err = check_sock_access(env, insn_idx, regno, off, size, t);
Joe Stringerc64b7982018-10-02 13:35:33 -07004285 if (!err && value_regno >= 0)
4286 mark_reg_unknown(env, regs, value_regno);
Matt Mullins9df1c282019-04-26 11:49:47 -07004287 } else if (reg->type == PTR_TO_TP_BUFFER) {
4288 err = check_tp_buffer_access(env, reg, regno, off, size);
4289 if (!err && t == BPF_READ && value_regno >= 0)
4290 mark_reg_unknown(env, regs, value_regno);
Alexei Starovoitov9e15db62019-10-15 20:25:00 -07004291 } else if (reg->type == PTR_TO_BTF_ID) {
4292 err = check_ptr_to_btf_access(env, regs, regno, off, size, t,
4293 value_regno);
Andrey Ignatov41c48f32020-06-19 14:11:43 -07004294 } else if (reg->type == CONST_PTR_TO_MAP) {
4295 err = check_ptr_to_map_access(env, regs, regno, off, size, t,
4296 value_regno);
Yonghong Songafbf21d2020-07-23 11:41:11 -07004297 } else if (reg->type == PTR_TO_RDONLY_BUF) {
4298 if (t == BPF_WRITE) {
4299 verbose(env, "R%d cannot write into %s\n",
4300 regno, reg_type_str[reg->type]);
4301 return -EACCES;
4302 }
Colin Ian Kingf6dfbe312020-07-27 18:54:11 +01004303 err = check_buffer_access(env, reg, regno, off, size, false,
4304 "rdonly",
Yonghong Songafbf21d2020-07-23 11:41:11 -07004305 &env->prog->aux->max_rdonly_access);
4306 if (!err && value_regno >= 0)
4307 mark_reg_unknown(env, regs, value_regno);
4308 } else if (reg->type == PTR_TO_RDWR_BUF) {
Colin Ian Kingf6dfbe312020-07-27 18:54:11 +01004309 err = check_buffer_access(env, reg, regno, off, size, false,
4310 "rdwr",
Yonghong Songafbf21d2020-07-23 11:41:11 -07004311 &env->prog->aux->max_rdwr_access);
4312 if (!err && t == BPF_READ && value_regno >= 0)
4313 mark_reg_unknown(env, regs, value_regno);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004314 } else {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004315 verbose(env, "R%d invalid mem access '%s'\n", regno,
4316 reg_type_str[reg->type]);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004317 return -EACCES;
4318 }
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004319
Edward Creef1174f72017-08-07 15:26:19 +01004320 if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ &&
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004321 regs[value_regno].type == SCALAR_VALUE) {
Edward Creef1174f72017-08-07 15:26:19 +01004322 /* b/h/w load zero-extends, mark upper bits as known 0 */
Jann Horn0c17d1d2017-12-18 20:11:55 -08004323 coerce_reg_to_size(&regs[value_regno], size);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07004324 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004325 return err;
4326}
4327
Brendan Jackman91c960b2021-01-14 18:17:44 +00004328static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004329{
Brendan Jackman5ffa2552021-01-14 18:17:47 +00004330 int load_reg;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004331 int err;
4332
Brendan Jackman5ca419f2021-01-14 18:17:46 +00004333 switch (insn->imm) {
4334 case BPF_ADD:
4335 case BPF_ADD | BPF_FETCH:
Brendan Jackman981f94c2021-01-14 18:17:49 +00004336 case BPF_AND:
4337 case BPF_AND | BPF_FETCH:
4338 case BPF_OR:
4339 case BPF_OR | BPF_FETCH:
4340 case BPF_XOR:
4341 case BPF_XOR | BPF_FETCH:
Brendan Jackman5ffa2552021-01-14 18:17:47 +00004342 case BPF_XCHG:
4343 case BPF_CMPXCHG:
Brendan Jackman5ca419f2021-01-14 18:17:46 +00004344 break;
4345 default:
Brendan Jackman91c960b2021-01-14 18:17:44 +00004346 verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm);
4347 return -EINVAL;
4348 }
4349
4350 if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) {
4351 verbose(env, "invalid atomic operand size\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004352 return -EINVAL;
4353 }
4354
4355 /* check src1 operand */
Edward Creedc503a82017-08-15 20:34:35 +01004356 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004357 if (err)
4358 return err;
4359
4360 /* check src2 operand */
Edward Creedc503a82017-08-15 20:34:35 +01004361 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004362 if (err)
4363 return err;
4364
Brendan Jackman5ffa2552021-01-14 18:17:47 +00004365 if (insn->imm == BPF_CMPXCHG) {
4366 /* Check comparison of R0 with memory location */
4367 err = check_reg_arg(env, BPF_REG_0, SRC_OP);
4368 if (err)
4369 return err;
4370 }
4371
Daniel Borkmann6bdf6ab2017-06-29 03:04:59 +02004372 if (is_pointer_value(env, insn->src_reg)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07004373 verbose(env, "R%d leaks addr into mem\n", insn->src_reg);
Daniel Borkmann6bdf6ab2017-06-29 03:04:59 +02004374 return -EACCES;
4375 }
4376
Daniel Borkmannca369602018-02-23 22:29:05 +01004377 if (is_ctx_reg(env, insn->dst_reg) ||
Daniel Borkmann4b5defd2018-10-21 02:09:25 +02004378 is_pkt_reg(env, insn->dst_reg) ||
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08004379 is_flow_key_reg(env, insn->dst_reg) ||
4380 is_sk_reg(env, insn->dst_reg)) {
Brendan Jackman91c960b2021-01-14 18:17:44 +00004381 verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n",
Daniel Borkmann2a159c62018-10-21 02:09:24 +02004382 insn->dst_reg,
4383 reg_type_str[reg_state(env, insn->dst_reg)->type]);
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +01004384 return -EACCES;
4385 }
4386
Brendan Jackman37086bf2021-02-02 13:50:02 +00004387 if (insn->imm & BPF_FETCH) {
4388 if (insn->imm == BPF_CMPXCHG)
4389 load_reg = BPF_REG_0;
4390 else
4391 load_reg = insn->src_reg;
4392
4393 /* check and record load of old value */
4394 err = check_reg_arg(env, load_reg, DST_OP);
4395 if (err)
4396 return err;
4397 } else {
4398 /* This instruction accesses a memory location but doesn't
4399 * actually load it into a register.
4400 */
4401 load_reg = -1;
4402 }
4403
Brendan Jackman91c960b2021-01-14 18:17:44 +00004404 /* check whether we can read the memory */
Yonghong Song31fd8582017-06-13 15:52:13 -07004405 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
Brendan Jackman37086bf2021-02-02 13:50:02 +00004406 BPF_SIZE(insn->code), BPF_READ, load_reg, true);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004407 if (err)
4408 return err;
4409
Brendan Jackman91c960b2021-01-14 18:17:44 +00004410 /* check whether we can write into the same memory */
Brendan Jackman5ca419f2021-01-14 18:17:46 +00004411 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
4412 BPF_SIZE(insn->code), BPF_WRITE, -1, true);
4413 if (err)
4414 return err;
4415
Brendan Jackman5ca419f2021-01-14 18:17:46 +00004416 return 0;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004417}
4418
Andrei Matei01f810a2021-02-06 20:10:24 -05004419/* When register 'regno' is used to read the stack (either directly or through
4420 * a helper function) make sure that it's within stack boundary and, depending
4421 * on the access type, that all elements of the stack are initialized.
4422 *
4423 * 'off' includes 'regno->off', but not its dynamic part (if any).
4424 *
4425 * All registers that have been spilled on the stack in the slots within the
4426 * read offsets are marked as read.
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004427 */
Andrei Matei01f810a2021-02-06 20:10:24 -05004428static int check_stack_range_initialized(
4429 struct bpf_verifier_env *env, int regno, int off,
4430 int access_size, bool zero_size_allowed,
4431 enum stack_access_src type, struct bpf_call_arg_meta *meta)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004432{
Daniel Borkmann2a159c62018-10-21 02:09:24 +02004433 struct bpf_reg_state *reg = reg_state(env, regno);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08004434 struct bpf_func_state *state = func(env, reg);
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07004435 int err, min_off, max_off, i, j, slot, spi;
Andrei Matei01f810a2021-02-06 20:10:24 -05004436 char *err_extra = type == ACCESS_HELPER ? " indirect" : "";
4437 enum bpf_access_type bounds_check_type;
4438 /* Some accesses can write anything into the stack, others are
4439 * read-only.
4440 */
4441 bool clobber = false;
4442
4443 if (access_size == 0 && !zero_size_allowed) {
4444 verbose(env, "invalid zero-sized read\n");
4445 return -EACCES;
4446 }
4447
4448 if (type == ACCESS_HELPER) {
4449 /* The bounds checks for writes are more permissive than for
4450 * reads. However, if raw_mode is not set, we'll do extra
4451 * checks below.
4452 */
4453 bounds_check_type = BPF_WRITE;
4454 clobber = true;
4455 } else {
4456 bounds_check_type = BPF_READ;
4457 }
4458 err = check_stack_access_within_bounds(env, regno, off, access_size,
4459 type, bounds_check_type);
4460 if (err)
4461 return err;
4462
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004463
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004464 if (tnum_is_const(reg->var_off)) {
Andrei Matei01f810a2021-02-06 20:10:24 -05004465 min_off = max_off = reg->var_off.value + off;
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004466 } else {
Andrey Ignatov088ec262019-04-03 23:22:39 -07004467 /* Variable offset is prohibited for unprivileged mode for
4468 * simplicity since it requires corresponding support in
4469 * Spectre masking for stack ALU.
4470 * See also retrieve_ptr_limit().
4471 */
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07004472 if (!env->bypass_spec_v1) {
Andrey Ignatov088ec262019-04-03 23:22:39 -07004473 char tn_buf[48];
Edward Creef1174f72017-08-07 15:26:19 +01004474
Andrey Ignatov088ec262019-04-03 23:22:39 -07004475 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Andrei Matei01f810a2021-02-06 20:10:24 -05004476 verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n",
4477 regno, err_extra, tn_buf);
Andrey Ignatov088ec262019-04-03 23:22:39 -07004478 return -EACCES;
4479 }
Andrey Ignatovf2bcd052019-04-03 23:22:37 -07004480 /* Only initialized buffer on stack is allowed to be accessed
4481 * with variable offset. With uninitialized buffer it's hard to
4482 * guarantee that whole memory is marked as initialized on
4483 * helper return since specific bounds are unknown what may
4484 * cause uninitialized stack leaking.
4485 */
4486 if (meta && meta->raw_mode)
4487 meta = NULL;
4488
Andrei Matei01f810a2021-02-06 20:10:24 -05004489 min_off = reg->smin_value + off;
4490 max_off = reg->smax_value + off;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004491 }
4492
Daniel Borkmann435faee12016-04-13 00:10:51 +02004493 if (meta && meta->raw_mode) {
4494 meta->access_size = access_size;
4495 meta->regno = regno;
4496 return 0;
4497 }
4498
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004499 for (i = min_off; i < max_off + access_size; i++) {
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08004500 u8 *stype;
4501
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004502 slot = -i - 1;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004503 spi = slot / BPF_REG_SIZE;
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08004504 if (state->allocated_stack <= slot)
4505 goto err;
4506 stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE];
4507 if (*stype == STACK_MISC)
4508 goto mark;
4509 if (*stype == STACK_ZERO) {
Andrei Matei01f810a2021-02-06 20:10:24 -05004510 if (clobber) {
4511 /* helper can write anything into the stack */
4512 *stype = STACK_MISC;
4513 }
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08004514 goto mark;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004515 }
Yonghong Song1d68f222020-05-09 10:59:15 -07004516
4517 if (state->stack[spi].slot_type[0] == STACK_SPILL &&
4518 state->stack[spi].spilled_ptr.type == PTR_TO_BTF_ID)
4519 goto mark;
4520
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07004521 if (state->stack[spi].slot_type[0] == STACK_SPILL &&
Yonghong Songcd17d382020-12-09 17:33:49 -08004522 (state->stack[spi].spilled_ptr.type == SCALAR_VALUE ||
4523 env->allow_ptr_leaks)) {
Andrei Matei01f810a2021-02-06 20:10:24 -05004524 if (clobber) {
4525 __mark_reg_unknown(env, &state->stack[spi].spilled_ptr);
4526 for (j = 0; j < BPF_REG_SIZE; j++)
4527 state->stack[spi].slot_type[j] = STACK_MISC;
4528 }
Alexei Starovoitovf7cf25b2019-06-15 12:12:17 -07004529 goto mark;
4530 }
4531
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08004532err:
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004533 if (tnum_is_const(reg->var_off)) {
Andrei Matei01f810a2021-02-06 20:10:24 -05004534 verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n",
4535 err_extra, regno, min_off, i - min_off, access_size);
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004536 } else {
4537 char tn_buf[48];
4538
4539 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
Andrei Matei01f810a2021-02-06 20:10:24 -05004540 verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n",
4541 err_extra, regno, tn_buf, i - min_off, access_size);
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004542 }
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -08004543 return -EACCES;
4544mark:
4545 /* reading any byte out of 8-byte 'spill_slot' will cause
4546 * the whole slot to be marked as 'read'
4547 */
Edward Cree679c7822018-08-22 20:02:19 +01004548 mark_reg_read(env, &state->stack[spi].spilled_ptr,
Jiong Wang5327ed32019-05-24 23:25:12 +01004549 state->stack[spi].spilled_ptr.parent,
4550 REG_LIVE_READ64);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004551 }
Andrey Ignatov2011fcc2019-03-28 18:01:57 -07004552 return update_stack_depth(env, state, min_off);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004553}
4554
Gianluca Borello06c1c042017-01-09 10:19:49 -08004555static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
4556 int access_size, bool zero_size_allowed,
4557 struct bpf_call_arg_meta *meta)
4558{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004559 struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
Gianluca Borello06c1c042017-01-09 10:19:49 -08004560
Edward Creef1174f72017-08-07 15:26:19 +01004561 switch (reg->type) {
Gianluca Borello06c1c042017-01-09 10:19:49 -08004562 case PTR_TO_PACKET:
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02004563 case PTR_TO_PACKET_META:
Yonghong Song9fd29c02017-11-12 14:49:09 -08004564 return check_packet_access(env, regno, reg->off, access_size,
4565 zero_size_allowed);
Yonghong Song69c087b2021-02-26 12:49:25 -08004566 case PTR_TO_MAP_KEY:
4567 return check_mem_region_access(env, regno, reg->off, access_size,
4568 reg->map_ptr->key_size, false);
Gianluca Borello06c1c042017-01-09 10:19:49 -08004569 case PTR_TO_MAP_VALUE:
Daniel Borkmann591fe982019-04-09 23:20:05 +02004570 if (check_map_access_type(env, regno, reg->off, access_size,
4571 meta && meta->raw_mode ? BPF_WRITE :
4572 BPF_READ))
4573 return -EACCES;
Yonghong Song9fd29c02017-11-12 14:49:09 -08004574 return check_map_access(env, regno, reg->off, access_size,
4575 zero_size_allowed);
Andrii Nakryiko457f4432020-05-29 00:54:20 -07004576 case PTR_TO_MEM:
4577 return check_mem_region_access(env, regno, reg->off,
4578 access_size, reg->mem_size,
4579 zero_size_allowed);
Yonghong Songafbf21d2020-07-23 11:41:11 -07004580 case PTR_TO_RDONLY_BUF:
4581 if (meta && meta->raw_mode)
4582 return -EACCES;
4583 return check_buffer_access(env, reg, regno, reg->off,
4584 access_size, zero_size_allowed,
4585 "rdonly",
4586 &env->prog->aux->max_rdonly_access);
4587 case PTR_TO_RDWR_BUF:
4588 return check_buffer_access(env, reg, regno, reg->off,
4589 access_size, zero_size_allowed,
4590 "rdwr",
4591 &env->prog->aux->max_rdwr_access);
Lorenz Bauer0d004c022020-09-21 13:12:18 +01004592 case PTR_TO_STACK:
Andrei Matei01f810a2021-02-06 20:10:24 -05004593 return check_stack_range_initialized(
4594 env,
4595 regno, reg->off, access_size,
4596 zero_size_allowed, ACCESS_HELPER, meta);
Lorenz Bauer0d004c022020-09-21 13:12:18 +01004597 default: /* scalar_value or invalid ptr */
4598 /* Allow zero-byte read from NULL, regardless of pointer type */
4599 if (zero_size_allowed && access_size == 0 &&
4600 register_is_null(reg))
4601 return 0;
4602
4603 verbose(env, "R%d type=%s expected=%s\n", regno,
4604 reg_type_str[reg->type],
4605 reg_type_str[PTR_TO_STACK]);
4606 return -EACCES;
Gianluca Borello06c1c042017-01-09 10:19:49 -08004607 }
4608}
4609
Dmitrii Banshchikove5069b9c2021-02-13 00:56:41 +04004610int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg,
4611 u32 regno, u32 mem_size)
4612{
4613 if (register_is_null(reg))
4614 return 0;
4615
4616 if (reg_type_may_be_null(reg->type)) {
4617 /* Assuming that the register contains a value check if the memory
4618 * access is safe. Temporarily save and restore the register's state as
4619 * the conversion shouldn't be visible to a caller.
4620 */
4621 const struct bpf_reg_state saved_reg = *reg;
4622 int rv;
4623
4624 mark_ptr_not_null_reg(reg);
4625 rv = check_helper_mem_access(env, regno, mem_size, true, NULL);
4626 *reg = saved_reg;
4627 return rv;
4628 }
4629
4630 return check_helper_mem_access(env, regno, mem_size, true, NULL);
4631}
4632
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08004633/* Implementation details:
4634 * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL
4635 * Two bpf_map_lookups (even with the same key) will have different reg->id.
4636 * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after
4637 * value_or_null->value transition, since the verifier only cares about
4638 * the range of access to valid map value pointer and doesn't care about actual
4639 * address of the map element.
4640 * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps
4641 * reg->id > 0 after value_or_null->value transition. By doing so
4642 * two bpf_map_lookups will be considered two different pointers that
4643 * point to different bpf_spin_locks.
4644 * The verifier allows taking only one bpf_spin_lock at a time to avoid
4645 * dead-locks.
4646 * Since only one bpf_spin_lock is allowed the checks are simpler than
4647 * reg_is_refcounted() logic. The verifier needs to remember only
4648 * one spin_lock instead of array of acquired_refs.
4649 * cur_state->active_spin_lock remembers which map value element got locked
4650 * and clears it after bpf_spin_unlock.
4651 */
4652static int process_spin_lock(struct bpf_verifier_env *env, int regno,
4653 bool is_lock)
4654{
4655 struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
4656 struct bpf_verifier_state *cur = env->cur_state;
4657 bool is_const = tnum_is_const(reg->var_off);
4658 struct bpf_map *map = reg->map_ptr;
4659 u64 val = reg->var_off.value;
4660
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08004661 if (!is_const) {
4662 verbose(env,
4663 "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n",
4664 regno);
4665 return -EINVAL;
4666 }
4667 if (!map->btf) {
4668 verbose(env,
4669 "map '%s' has to have BTF in order to use bpf_spin_lock\n",
4670 map->name);
4671 return -EINVAL;
4672 }
4673 if (!map_value_has_spin_lock(map)) {
4674 if (map->spin_lock_off == -E2BIG)
4675 verbose(env,
4676 "map '%s' has more than one 'struct bpf_spin_lock'\n",
4677 map->name);
4678 else if (map->spin_lock_off == -ENOENT)
4679 verbose(env,
4680 "map '%s' doesn't have 'struct bpf_spin_lock'\n",
4681 map->name);
4682 else
4683 verbose(env,
4684 "map '%s' is not a struct type or bpf_spin_lock is mangled\n",
4685 map->name);
4686 return -EINVAL;
4687 }
4688 if (map->spin_lock_off != val + reg->off) {
4689 verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock'\n",
4690 val + reg->off);
4691 return -EINVAL;
4692 }
4693 if (is_lock) {
4694 if (cur->active_spin_lock) {
4695 verbose(env,
4696 "Locking two bpf_spin_locks are not allowed\n");
4697 return -EINVAL;
4698 }
4699 cur->active_spin_lock = reg->id;
4700 } else {
4701 if (!cur->active_spin_lock) {
4702 verbose(env, "bpf_spin_unlock without taking a lock\n");
4703 return -EINVAL;
4704 }
4705 if (cur->active_spin_lock != reg->id) {
4706 verbose(env, "bpf_spin_unlock of different lock\n");
4707 return -EINVAL;
4708 }
4709 cur->active_spin_lock = 0;
4710 }
4711 return 0;
4712}
4713
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07004714static int process_timer_func(struct bpf_verifier_env *env, int regno,
4715 struct bpf_call_arg_meta *meta)
4716{
4717 struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
4718 bool is_const = tnum_is_const(reg->var_off);
4719 struct bpf_map *map = reg->map_ptr;
4720 u64 val = reg->var_off.value;
4721
4722 if (!is_const) {
4723 verbose(env,
4724 "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n",
4725 regno);
4726 return -EINVAL;
4727 }
4728 if (!map->btf) {
4729 verbose(env, "map '%s' has to have BTF in order to use bpf_timer\n",
4730 map->name);
4731 return -EINVAL;
4732 }
Alexei Starovoitov68134662021-07-14 17:54:10 -07004733 if (!map_value_has_timer(map)) {
4734 if (map->timer_off == -E2BIG)
4735 verbose(env,
4736 "map '%s' has more than one 'struct bpf_timer'\n",
4737 map->name);
4738 else if (map->timer_off == -ENOENT)
4739 verbose(env,
4740 "map '%s' doesn't have 'struct bpf_timer'\n",
4741 map->name);
4742 else
4743 verbose(env,
4744 "map '%s' is not a struct type or bpf_timer is mangled\n",
4745 map->name);
4746 return -EINVAL;
4747 }
4748 if (map->timer_off != val + reg->off) {
4749 verbose(env, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n",
4750 val + reg->off, map->timer_off);
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07004751 return -EINVAL;
4752 }
4753 if (meta->map_ptr) {
4754 verbose(env, "verifier bug. Two map pointers in a timer helper\n");
4755 return -EFAULT;
4756 }
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -07004757 meta->map_uid = reg->map_uid;
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07004758 meta->map_ptr = map;
4759 return 0;
4760}
4761
Daniel Borkmann90133412018-01-20 01:24:29 +01004762static bool arg_type_is_mem_ptr(enum bpf_arg_type type)
4763{
4764 return type == ARG_PTR_TO_MEM ||
4765 type == ARG_PTR_TO_MEM_OR_NULL ||
4766 type == ARG_PTR_TO_UNINIT_MEM;
4767}
4768
4769static bool arg_type_is_mem_size(enum bpf_arg_type type)
4770{
4771 return type == ARG_CONST_SIZE ||
4772 type == ARG_CONST_SIZE_OR_ZERO;
4773}
4774
Andrii Nakryiko457f4432020-05-29 00:54:20 -07004775static bool arg_type_is_alloc_size(enum bpf_arg_type type)
4776{
4777 return type == ARG_CONST_ALLOC_SIZE_OR_ZERO;
4778}
4779
Andrey Ignatov57c3bb72019-03-18 16:57:10 -07004780static bool arg_type_is_int_ptr(enum bpf_arg_type type)
4781{
4782 return type == ARG_PTR_TO_INT ||
4783 type == ARG_PTR_TO_LONG;
4784}
4785
4786static int int_ptr_type_to_size(enum bpf_arg_type type)
4787{
4788 if (type == ARG_PTR_TO_INT)
4789 return sizeof(u32);
4790 else if (type == ARG_PTR_TO_LONG)
4791 return sizeof(u64);
4792
4793 return -EINVAL;
4794}
4795
Lorenz Bauer912f4422020-08-21 11:29:46 +01004796static int resolve_map_arg_type(struct bpf_verifier_env *env,
4797 const struct bpf_call_arg_meta *meta,
4798 enum bpf_arg_type *arg_type)
4799{
4800 if (!meta->map_ptr) {
4801 /* kernel subsystem misconfigured verifier */
4802 verbose(env, "invalid map_ptr to access map->type\n");
4803 return -EACCES;
4804 }
4805
4806 switch (meta->map_ptr->map_type) {
4807 case BPF_MAP_TYPE_SOCKMAP:
4808 case BPF_MAP_TYPE_SOCKHASH:
4809 if (*arg_type == ARG_PTR_TO_MAP_VALUE) {
Lorenz Bauer6550f2d2020-09-28 10:08:02 +01004810 *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON;
Lorenz Bauer912f4422020-08-21 11:29:46 +01004811 } else {
4812 verbose(env, "invalid arg_type for sockmap/sockhash\n");
4813 return -EINVAL;
4814 }
4815 break;
4816
4817 default:
4818 break;
4819 }
4820 return 0;
4821}
4822
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004823struct bpf_reg_types {
4824 const enum bpf_reg_type types[10];
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07004825 u32 *btf_id;
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004826};
4827
4828static const struct bpf_reg_types map_key_value_types = {
4829 .types = {
4830 PTR_TO_STACK,
4831 PTR_TO_PACKET,
4832 PTR_TO_PACKET_META,
Yonghong Song69c087b2021-02-26 12:49:25 -08004833 PTR_TO_MAP_KEY,
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004834 PTR_TO_MAP_VALUE,
4835 },
4836};
4837
4838static const struct bpf_reg_types sock_types = {
4839 .types = {
4840 PTR_TO_SOCK_COMMON,
4841 PTR_TO_SOCKET,
4842 PTR_TO_TCP_SOCK,
4843 PTR_TO_XDP_SOCK,
4844 },
4845};
4846
Randy Dunlap49a2a4d2020-10-06 19:16:13 -07004847#ifdef CONFIG_NET
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07004848static const struct bpf_reg_types btf_id_sock_common_types = {
4849 .types = {
4850 PTR_TO_SOCK_COMMON,
4851 PTR_TO_SOCKET,
4852 PTR_TO_TCP_SOCK,
4853 PTR_TO_XDP_SOCK,
4854 PTR_TO_BTF_ID,
4855 },
4856 .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
4857};
Randy Dunlap49a2a4d2020-10-06 19:16:13 -07004858#endif
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07004859
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004860static const struct bpf_reg_types mem_types = {
4861 .types = {
4862 PTR_TO_STACK,
4863 PTR_TO_PACKET,
4864 PTR_TO_PACKET_META,
Yonghong Song69c087b2021-02-26 12:49:25 -08004865 PTR_TO_MAP_KEY,
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004866 PTR_TO_MAP_VALUE,
4867 PTR_TO_MEM,
4868 PTR_TO_RDONLY_BUF,
4869 PTR_TO_RDWR_BUF,
4870 },
4871};
4872
4873static const struct bpf_reg_types int_ptr_types = {
4874 .types = {
4875 PTR_TO_STACK,
4876 PTR_TO_PACKET,
4877 PTR_TO_PACKET_META,
Yonghong Song69c087b2021-02-26 12:49:25 -08004878 PTR_TO_MAP_KEY,
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004879 PTR_TO_MAP_VALUE,
4880 },
4881};
4882
4883static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } };
4884static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } };
4885static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } };
4886static const struct bpf_reg_types alloc_mem_types = { .types = { PTR_TO_MEM } };
4887static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } };
4888static const struct bpf_reg_types btf_ptr_types = { .types = { PTR_TO_BTF_ID } };
4889static const struct bpf_reg_types spin_lock_types = { .types = { PTR_TO_MAP_VALUE } };
Hao Luoeaa6bcb2020-09-29 16:50:47 -07004890static const struct bpf_reg_types percpu_btf_ptr_types = { .types = { PTR_TO_PERCPU_BTF_ID } };
Yonghong Song69c087b2021-02-26 12:49:25 -08004891static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } };
4892static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } };
Florent Revestfff13c42021-04-19 17:52:39 +02004893static const struct bpf_reg_types const_str_ptr_types = { .types = { PTR_TO_MAP_VALUE } };
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07004894static const struct bpf_reg_types timer_types = { .types = { PTR_TO_MAP_VALUE } };
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004895
Lorenz Bauer0789e13b2020-09-23 17:01:55 +01004896static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = {
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004897 [ARG_PTR_TO_MAP_KEY] = &map_key_value_types,
4898 [ARG_PTR_TO_MAP_VALUE] = &map_key_value_types,
4899 [ARG_PTR_TO_UNINIT_MAP_VALUE] = &map_key_value_types,
4900 [ARG_PTR_TO_MAP_VALUE_OR_NULL] = &map_key_value_types,
4901 [ARG_CONST_SIZE] = &scalar_types,
4902 [ARG_CONST_SIZE_OR_ZERO] = &scalar_types,
4903 [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types,
4904 [ARG_CONST_MAP_PTR] = &const_map_ptr_types,
4905 [ARG_PTR_TO_CTX] = &context_types,
4906 [ARG_PTR_TO_CTX_OR_NULL] = &context_types,
4907 [ARG_PTR_TO_SOCK_COMMON] = &sock_types,
Randy Dunlap49a2a4d2020-10-06 19:16:13 -07004908#ifdef CONFIG_NET
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07004909 [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types,
Randy Dunlap49a2a4d2020-10-06 19:16:13 -07004910#endif
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004911 [ARG_PTR_TO_SOCKET] = &fullsock_types,
4912 [ARG_PTR_TO_SOCKET_OR_NULL] = &fullsock_types,
4913 [ARG_PTR_TO_BTF_ID] = &btf_ptr_types,
4914 [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types,
4915 [ARG_PTR_TO_MEM] = &mem_types,
4916 [ARG_PTR_TO_MEM_OR_NULL] = &mem_types,
4917 [ARG_PTR_TO_UNINIT_MEM] = &mem_types,
4918 [ARG_PTR_TO_ALLOC_MEM] = &alloc_mem_types,
4919 [ARG_PTR_TO_ALLOC_MEM_OR_NULL] = &alloc_mem_types,
4920 [ARG_PTR_TO_INT] = &int_ptr_types,
4921 [ARG_PTR_TO_LONG] = &int_ptr_types,
Hao Luoeaa6bcb2020-09-29 16:50:47 -07004922 [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types,
Yonghong Song69c087b2021-02-26 12:49:25 -08004923 [ARG_PTR_TO_FUNC] = &func_ptr_types,
4924 [ARG_PTR_TO_STACK_OR_NULL] = &stack_ptr_types,
Florent Revestfff13c42021-04-19 17:52:39 +02004925 [ARG_PTR_TO_CONST_STR] = &const_str_ptr_types,
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07004926 [ARG_PTR_TO_TIMER] = &timer_types,
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004927};
4928
4929static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004930 enum bpf_arg_type arg_type,
4931 const u32 *arg_btf_id)
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004932{
4933 struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
4934 enum bpf_reg_type expected, type = reg->type;
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004935 const struct bpf_reg_types *compatible;
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004936 int i, j;
4937
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004938 compatible = compatible_reg_types[arg_type];
4939 if (!compatible) {
4940 verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type);
4941 return -EFAULT;
4942 }
4943
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004944 for (i = 0; i < ARRAY_SIZE(compatible->types); i++) {
4945 expected = compatible->types[i];
4946 if (expected == NOT_INIT)
4947 break;
4948
4949 if (type == expected)
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004950 goto found;
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004951 }
4952
4953 verbose(env, "R%d type=%s expected=", regno, reg_type_str[type]);
4954 for (j = 0; j + 1 < i; j++)
4955 verbose(env, "%s, ", reg_type_str[compatible->types[j]]);
4956 verbose(env, "%s\n", reg_type_str[compatible->types[j]]);
4957 return -EACCES;
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004958
4959found:
4960 if (type == PTR_TO_BTF_ID) {
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07004961 if (!arg_btf_id) {
4962 if (!compatible->btf_id) {
4963 verbose(env, "verifier internal error: missing arg compatible BTF ID\n");
4964 return -EFAULT;
4965 }
4966 arg_btf_id = compatible->btf_id;
4967 }
4968
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004969 if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off,
4970 btf_vmlinux, *arg_btf_id)) {
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004971 verbose(env, "R%d is of type %s but %s is expected\n",
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08004972 regno, kernel_type_name(reg->btf, reg->btf_id),
4973 kernel_type_name(btf_vmlinux, *arg_btf_id));
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07004974 return -EACCES;
4975 }
4976
4977 if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
4978 verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n",
4979 regno);
4980 return -EACCES;
4981 }
4982 }
4983
4984 return 0;
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004985}
4986
Yonghong Songaf7ec132020-06-23 16:08:09 -07004987static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
4988 struct bpf_call_arg_meta *meta,
4989 const struct bpf_func_proto *fn)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004990{
Yonghong Songaf7ec132020-06-23 16:08:09 -07004991 u32 regno = BPF_REG_1 + arg;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07004992 struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
Yonghong Songaf7ec132020-06-23 16:08:09 -07004993 enum bpf_arg_type arg_type = fn->arg_type[arg];
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01004994 enum bpf_reg_type type = reg->type;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004995 int err = 0;
4996
Daniel Borkmann80f1d682015-03-12 17:21:42 +01004997 if (arg_type == ARG_DONTCARE)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07004998 return 0;
4999
Edward Creedc503a82017-08-15 20:34:35 +01005000 err = check_reg_arg(env, regno, SRC_OP);
5001 if (err)
5002 return err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005003
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07005004 if (arg_type == ARG_ANYTHING) {
5005 if (is_pointer_value(env, regno)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005006 verbose(env, "R%d leaks addr into helper function\n",
5007 regno);
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07005008 return -EACCES;
5009 }
Daniel Borkmann80f1d682015-03-12 17:21:42 +01005010 return 0;
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07005011 }
Daniel Borkmann80f1d682015-03-12 17:21:42 +01005012
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02005013 if (type_is_pkt_pointer(type) &&
Thomas Graf3a0af8f2016-11-30 17:10:10 +01005014 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005015 verbose(env, "helper access to the packet is not allowed\n");
Alexei Starovoitov6841de82016-08-11 18:17:16 -07005016 return -EACCES;
5017 }
5018
Lorenz Bauer912f4422020-08-21 11:29:46 +01005019 if (arg_type == ARG_PTR_TO_MAP_VALUE ||
5020 arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE ||
5021 arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) {
5022 err = resolve_map_arg_type(env, meta, &arg_type);
5023 if (err)
5024 return err;
5025 }
5026
Lorenz Bauerfd1b0d62020-09-21 13:12:26 +01005027 if (register_is_null(reg) && arg_type_may_be_null(arg_type))
5028 /* A NULL register has a SCALAR_VALUE type, so skip
5029 * type checking.
5030 */
5031 goto skip_type_check;
Jiri Olsafaaf4a72020-08-25 21:21:18 +02005032
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07005033 err = check_reg_type(env, regno, arg_type, fn->arg_btf_id[arg]);
Lorenz Bauerf79e7ea2020-09-21 13:12:27 +01005034 if (err)
5035 return err;
5036
Martin KaFai Laua968d5e2020-09-24 17:03:44 -07005037 if (type == PTR_TO_CTX) {
Lorenz Bauerfeec7042020-09-21 13:12:23 +01005038 err = check_ctx_reg(env, reg, regno);
5039 if (err < 0)
5040 return err;
Lorenz Bauerd7b94542020-09-21 13:12:21 +01005041 }
5042
Lorenz Bauerfd1b0d62020-09-21 13:12:26 +01005043skip_type_check:
Lorenz Bauer02f7c952020-09-21 13:12:22 +01005044 if (reg->ref_obj_id) {
Andrii Nakryiko457f4432020-05-29 00:54:20 -07005045 if (meta->ref_obj_id) {
5046 verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n",
5047 regno, reg->ref_obj_id,
5048 meta->ref_obj_id);
5049 return -EFAULT;
5050 }
5051 meta->ref_obj_id = reg->ref_obj_id;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005052 }
5053
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005054 if (arg_type == ARG_CONST_MAP_PTR) {
5055 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -07005056 if (meta->map_ptr) {
5057 /* Use map_uid (which is unique id of inner map) to reject:
5058 * inner_map1 = bpf_map_lookup_elem(outer_map, key1)
5059 * inner_map2 = bpf_map_lookup_elem(outer_map, key2)
5060 * if (inner_map1 && inner_map2) {
5061 * timer = bpf_map_lookup_elem(inner_map1);
5062 * if (timer)
5063 * // mismatch would have been allowed
5064 * bpf_timer_init(timer, inner_map2);
5065 * }
5066 *
5067 * Comparing map_ptr is enough to distinguish normal and outer maps.
5068 */
5069 if (meta->map_ptr != reg->map_ptr ||
5070 meta->map_uid != reg->map_uid) {
5071 verbose(env,
5072 "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n",
5073 meta->map_uid, reg->map_uid);
5074 return -EINVAL;
5075 }
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07005076 }
Daniel Borkmann33ff9822016-04-13 00:10:50 +02005077 meta->map_ptr = reg->map_ptr;
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -07005078 meta->map_uid = reg->map_uid;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005079 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
5080 /* bpf_map_xxx(..., map_ptr, ..., key) call:
5081 * check that [key, key + map->key_size) are within
5082 * stack limits and initialized
5083 */
Daniel Borkmann33ff9822016-04-13 00:10:50 +02005084 if (!meta->map_ptr) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005085 /* in function declaration map_ptr must come before
5086 * map_key, so that it's verified and known before
5087 * we have to check map_key here. Otherwise it means
5088 * that kernel subsystem misconfigured verifier
5089 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005090 verbose(env, "invalid map_ptr to access map->key\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005091 return -EACCES;
5092 }
Paul Chaignond71962f2018-04-24 15:07:54 +02005093 err = check_helper_mem_access(env, regno,
5094 meta->map_ptr->key_size, false,
5095 NULL);
Mauricio Vasquez B2ea864c2018-10-18 15:16:20 +02005096 } else if (arg_type == ARG_PTR_TO_MAP_VALUE ||
Martin KaFai Lau6ac99e82019-04-26 16:39:39 -07005097 (arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL &&
5098 !register_is_null(reg)) ||
Mauricio Vasquez B2ea864c2018-10-18 15:16:20 +02005099 arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005100 /* bpf_map_xxx(..., map_ptr, ..., value) call:
5101 * check [value, value + map->value_size) validity
5102 */
Daniel Borkmann33ff9822016-04-13 00:10:50 +02005103 if (!meta->map_ptr) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005104 /* kernel subsystem misconfigured verifier */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005105 verbose(env, "invalid map_ptr to access map->value\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005106 return -EACCES;
5107 }
Mauricio Vasquez B2ea864c2018-10-18 15:16:20 +02005108 meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE);
Paul Chaignond71962f2018-04-24 15:07:54 +02005109 err = check_helper_mem_access(env, regno,
5110 meta->map_ptr->value_size, false,
Mauricio Vasquez B2ea864c2018-10-18 15:16:20 +02005111 meta);
Hao Luoeaa6bcb2020-09-29 16:50:47 -07005112 } else if (arg_type == ARG_PTR_TO_PERCPU_BTF_ID) {
5113 if (!reg->btf_id) {
5114 verbose(env, "Helper has invalid btf_id in R%d\n", regno);
5115 return -EACCES;
5116 }
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08005117 meta->ret_btf = reg->btf;
Hao Luoeaa6bcb2020-09-29 16:50:47 -07005118 meta->ret_btf_id = reg->btf_id;
Lorenz Bauerc18f0b62020-09-21 13:12:25 +01005119 } else if (arg_type == ARG_PTR_TO_SPIN_LOCK) {
5120 if (meta->func_id == BPF_FUNC_spin_lock) {
5121 if (process_spin_lock(env, regno, true))
5122 return -EACCES;
5123 } else if (meta->func_id == BPF_FUNC_spin_unlock) {
5124 if (process_spin_lock(env, regno, false))
5125 return -EACCES;
5126 } else {
5127 verbose(env, "verifier internal error\n");
5128 return -EFAULT;
5129 }
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07005130 } else if (arg_type == ARG_PTR_TO_TIMER) {
5131 if (process_timer_func(env, regno, meta))
5132 return -EACCES;
Yonghong Song69c087b2021-02-26 12:49:25 -08005133 } else if (arg_type == ARG_PTR_TO_FUNC) {
5134 meta->subprogno = reg->subprogno;
Lorenz Bauera2bbe7c2020-09-21 13:12:24 +01005135 } else if (arg_type_is_mem_ptr(arg_type)) {
5136 /* The access to this pointer is only checked when we hit the
5137 * next is_mem_size argument below.
5138 */
5139 meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MEM);
Daniel Borkmann90133412018-01-20 01:24:29 +01005140 } else if (arg_type_is_mem_size(arg_type)) {
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005141 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005142
John Fastabend10060502020-03-30 14:36:19 -07005143 /* This is used to refine r0 return value bounds for helpers
5144 * that enforce this value as an upper bound on return values.
5145 * See do_refine_retval_range() for helpers that can refine
5146 * the return value. C type of helper is u32 so we pull register
5147 * bound from umax_value however, if negative verifier errors
5148 * out. Only upper bounds can be learned because retval is an
5149 * int type and negative retvals are allowed.
Yonghong Song849fa502018-04-28 22:28:09 -07005150 */
John Fastabend10060502020-03-30 14:36:19 -07005151 meta->msize_max_value = reg->umax_value;
Yonghong Song849fa502018-04-28 22:28:09 -07005152
Edward Creef1174f72017-08-07 15:26:19 +01005153 /* The register is SCALAR_VALUE; the access check
5154 * happens using its boundaries.
Gianluca Borello06c1c042017-01-09 10:19:49 -08005155 */
Edward Creef1174f72017-08-07 15:26:19 +01005156 if (!tnum_is_const(reg->var_off))
Gianluca Borello06c1c042017-01-09 10:19:49 -08005157 /* For unprivileged variable accesses, disable raw
5158 * mode so that the program is required to
5159 * initialize all the memory that the helper could
5160 * just partially fill up.
5161 */
5162 meta = NULL;
5163
Edward Creeb03c9f92017-08-07 15:26:36 +01005164 if (reg->smin_value < 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005165 verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n",
Edward Creef1174f72017-08-07 15:26:19 +01005166 regno);
5167 return -EACCES;
5168 }
Gianluca Borello06c1c042017-01-09 10:19:49 -08005169
Edward Creeb03c9f92017-08-07 15:26:36 +01005170 if (reg->umin_value == 0) {
Edward Creef1174f72017-08-07 15:26:19 +01005171 err = check_helper_mem_access(env, regno - 1, 0,
5172 zero_size_allowed,
5173 meta);
Gianluca Borello06c1c042017-01-09 10:19:49 -08005174 if (err)
5175 return err;
Gianluca Borello06c1c042017-01-09 10:19:49 -08005176 }
Edward Creef1174f72017-08-07 15:26:19 +01005177
Edward Creeb03c9f92017-08-07 15:26:36 +01005178 if (reg->umax_value >= BPF_MAX_VAR_SIZ) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005179 verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
Edward Creef1174f72017-08-07 15:26:19 +01005180 regno);
5181 return -EACCES;
5182 }
5183 err = check_helper_mem_access(env, regno - 1,
Edward Creeb03c9f92017-08-07 15:26:36 +01005184 reg->umax_value,
Edward Creef1174f72017-08-07 15:26:19 +01005185 zero_size_allowed, meta);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07005186 if (!err)
5187 err = mark_chain_precision(env, regno);
Andrii Nakryiko457f4432020-05-29 00:54:20 -07005188 } else if (arg_type_is_alloc_size(arg_type)) {
5189 if (!tnum_is_const(reg->var_off)) {
Brendan Jackman28a8add2021-01-12 12:39:13 +00005190 verbose(env, "R%d is not a known constant'\n",
Andrii Nakryiko457f4432020-05-29 00:54:20 -07005191 regno);
5192 return -EACCES;
5193 }
5194 meta->mem_size = reg->var_off.value;
Andrey Ignatov57c3bb72019-03-18 16:57:10 -07005195 } else if (arg_type_is_int_ptr(arg_type)) {
5196 int size = int_ptr_type_to_size(arg_type);
5197
5198 err = check_helper_mem_access(env, regno, size, false, meta);
5199 if (err)
5200 return err;
5201 err = check_ptr_alignment(env, reg, 0, size, true);
Florent Revestfff13c42021-04-19 17:52:39 +02005202 } else if (arg_type == ARG_PTR_TO_CONST_STR) {
5203 struct bpf_map *map = reg->map_ptr;
5204 int map_off;
5205 u64 map_addr;
5206 char *str_ptr;
5207
Florent Revesta8fad732021-04-23 01:55:43 +02005208 if (!bpf_map_is_rdonly(map)) {
Florent Revestfff13c42021-04-19 17:52:39 +02005209 verbose(env, "R%d does not point to a readonly map'\n", regno);
5210 return -EACCES;
5211 }
5212
5213 if (!tnum_is_const(reg->var_off)) {
5214 verbose(env, "R%d is not a constant address'\n", regno);
5215 return -EACCES;
5216 }
5217
5218 if (!map->ops->map_direct_value_addr) {
5219 verbose(env, "no direct value access support for this map type\n");
5220 return -EACCES;
5221 }
5222
5223 err = check_map_access(env, regno, reg->off,
5224 map->value_size - reg->off, false);
5225 if (err)
5226 return err;
5227
5228 map_off = reg->off + reg->var_off.value;
5229 err = map->ops->map_direct_value_addr(map, &map_addr, map_off);
5230 if (err) {
5231 verbose(env, "direct value access on string failed\n");
5232 return err;
5233 }
5234
5235 str_ptr = (char *)(long)(map_addr);
5236 if (!strnchr(str_ptr + map_off, map->value_size - map_off, 0)) {
5237 verbose(env, "string is not zero-terminated\n");
5238 return -EINVAL;
5239 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07005240 }
5241
5242 return err;
5243}
5244
Lorenz Bauer01262402020-08-21 11:29:47 +01005245static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id)
5246{
5247 enum bpf_attach_type eatype = env->prog->expected_attach_type;
Udip Pant7e407812020-08-25 16:20:00 -07005248 enum bpf_prog_type type = resolve_prog_type(env->prog);
Lorenz Bauer01262402020-08-21 11:29:47 +01005249
5250 if (func_id != BPF_FUNC_map_update_elem)
5251 return false;
5252
5253 /* It's not possible to get access to a locked struct sock in these
5254 * contexts, so updating is safe.
5255 */
5256 switch (type) {
5257 case BPF_PROG_TYPE_TRACING:
5258 if (eatype == BPF_TRACE_ITER)
5259 return true;
5260 break;
5261 case BPF_PROG_TYPE_SOCKET_FILTER:
5262 case BPF_PROG_TYPE_SCHED_CLS:
5263 case BPF_PROG_TYPE_SCHED_ACT:
5264 case BPF_PROG_TYPE_XDP:
5265 case BPF_PROG_TYPE_SK_REUSEPORT:
5266 case BPF_PROG_TYPE_FLOW_DISSECTOR:
5267 case BPF_PROG_TYPE_SK_LOOKUP:
5268 return true;
5269 default:
5270 break;
5271 }
5272
5273 verbose(env, "cannot update sockmap in this context\n");
5274 return false;
5275}
5276
Maciej Fijalkowskie4119012020-09-16 23:10:09 +02005277static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env)
5278{
5279 return env->prog->jit_requested && IS_ENABLED(CONFIG_X86_64);
5280}
5281
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005282static int check_map_func_compatibility(struct bpf_verifier_env *env,
5283 struct bpf_map *map, int func_id)
Kaixu Xia35578d72015-08-06 07:02:35 +00005284{
Kaixu Xia35578d72015-08-06 07:02:35 +00005285 if (!map)
5286 return 0;
5287
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005288 /* We need a two way check, first is from map perspective ... */
5289 switch (map->map_type) {
5290 case BPF_MAP_TYPE_PROG_ARRAY:
5291 if (func_id != BPF_FUNC_tail_call)
5292 goto error;
5293 break;
5294 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5295 if (func_id != BPF_FUNC_perf_event_read &&
Yonghong Song908432c2017-10-05 09:19:20 -07005296 func_id != BPF_FUNC_perf_event_output &&
Alexei Starovoitova7658e12019-10-15 20:25:04 -07005297 func_id != BPF_FUNC_skb_output &&
Eelco Chaudrond831ee82020-03-06 08:59:23 +00005298 func_id != BPF_FUNC_perf_event_read_value &&
5299 func_id != BPF_FUNC_xdp_output)
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005300 goto error;
5301 break;
Andrii Nakryiko457f4432020-05-29 00:54:20 -07005302 case BPF_MAP_TYPE_RINGBUF:
5303 if (func_id != BPF_FUNC_ringbuf_output &&
5304 func_id != BPF_FUNC_ringbuf_reserve &&
Andrii Nakryiko457f4432020-05-29 00:54:20 -07005305 func_id != BPF_FUNC_ringbuf_query)
5306 goto error;
5307 break;
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005308 case BPF_MAP_TYPE_STACK_TRACE:
5309 if (func_id != BPF_FUNC_get_stackid)
5310 goto error;
5311 break;
Martin KaFai Lau4ed8ec52016-06-30 10:28:43 -07005312 case BPF_MAP_TYPE_CGROUP_ARRAY:
David S. Miller60747ef2016-08-18 01:17:32 -04005313 if (func_id != BPF_FUNC_skb_under_cgroup &&
Sargun Dhillon60d20f92016-08-12 08:56:52 -07005314 func_id != BPF_FUNC_current_task_under_cgroup)
Martin KaFai Lau4a482f32016-06-30 10:28:44 -07005315 goto error;
5316 break;
Roman Gushchincd339432018-08-02 14:27:24 -07005317 case BPF_MAP_TYPE_CGROUP_STORAGE:
Roman Gushchinb741f162018-09-28 14:45:43 +00005318 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
Roman Gushchincd339432018-08-02 14:27:24 -07005319 if (func_id != BPF_FUNC_get_local_storage)
5320 goto error;
5321 break;
John Fastabend546ac1f2017-07-17 09:28:56 -07005322 case BPF_MAP_TYPE_DEVMAP:
Toke Høiland-Jørgensen6f9d4512019-07-26 18:06:55 +02005323 case BPF_MAP_TYPE_DEVMAP_HASH:
Toke Høiland-Jørgensen0cdbb4b2019-06-28 11:12:35 +02005324 if (func_id != BPF_FUNC_redirect_map &&
5325 func_id != BPF_FUNC_map_lookup_elem)
John Fastabend546ac1f2017-07-17 09:28:56 -07005326 goto error;
5327 break;
Björn Töpelfbfc504a2018-05-02 13:01:28 +02005328 /* Restrict bpf side of cpumap and xskmap, open when use-cases
5329 * appear.
5330 */
Jesper Dangaard Brouer6710e112017-10-16 12:19:28 +02005331 case BPF_MAP_TYPE_CPUMAP:
5332 if (func_id != BPF_FUNC_redirect_map)
5333 goto error;
5334 break;
Jonathan Lemonfada7fd2019-06-06 13:59:40 -07005335 case BPF_MAP_TYPE_XSKMAP:
5336 if (func_id != BPF_FUNC_redirect_map &&
5337 func_id != BPF_FUNC_map_lookup_elem)
5338 goto error;
5339 break;
Martin KaFai Lau56f668d2017-03-22 10:00:33 -07005340 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
Martin KaFai Laubcc6b1b2017-03-22 10:00:34 -07005341 case BPF_MAP_TYPE_HASH_OF_MAPS:
Martin KaFai Lau56f668d2017-03-22 10:00:33 -07005342 if (func_id != BPF_FUNC_map_lookup_elem)
5343 goto error;
Martin KaFai Lau16a43622017-08-17 18:14:43 -07005344 break;
John Fastabend174a79f2017-08-15 22:32:47 -07005345 case BPF_MAP_TYPE_SOCKMAP:
5346 if (func_id != BPF_FUNC_sk_redirect_map &&
5347 func_id != BPF_FUNC_sock_map_update &&
John Fastabend4f738ad2018-03-18 12:57:10 -07005348 func_id != BPF_FUNC_map_delete_elem &&
Jakub Sitnicki9fed9002020-02-18 17:10:20 +00005349 func_id != BPF_FUNC_msg_redirect_map &&
Jakub Sitnicki64d85292020-04-29 20:11:52 +02005350 func_id != BPF_FUNC_sk_select_reuseport &&
Lorenz Bauer01262402020-08-21 11:29:47 +01005351 func_id != BPF_FUNC_map_lookup_elem &&
5352 !may_update_sockmap(env, func_id))
John Fastabend174a79f2017-08-15 22:32:47 -07005353 goto error;
5354 break;
John Fastabend81110382018-05-14 10:00:17 -07005355 case BPF_MAP_TYPE_SOCKHASH:
5356 if (func_id != BPF_FUNC_sk_redirect_hash &&
5357 func_id != BPF_FUNC_sock_hash_update &&
5358 func_id != BPF_FUNC_map_delete_elem &&
Jakub Sitnicki9fed9002020-02-18 17:10:20 +00005359 func_id != BPF_FUNC_msg_redirect_hash &&
Jakub Sitnicki64d85292020-04-29 20:11:52 +02005360 func_id != BPF_FUNC_sk_select_reuseport &&
Lorenz Bauer01262402020-08-21 11:29:47 +01005361 func_id != BPF_FUNC_map_lookup_elem &&
5362 !may_update_sockmap(env, func_id))
John Fastabend81110382018-05-14 10:00:17 -07005363 goto error;
5364 break;
Martin KaFai Lau2dbb9b92018-08-08 01:01:25 -07005365 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
5366 if (func_id != BPF_FUNC_sk_select_reuseport)
5367 goto error;
5368 break;
Mauricio Vasquez Bf1a2e442018-10-18 15:16:25 +02005369 case BPF_MAP_TYPE_QUEUE:
5370 case BPF_MAP_TYPE_STACK:
5371 if (func_id != BPF_FUNC_map_peek_elem &&
5372 func_id != BPF_FUNC_map_pop_elem &&
5373 func_id != BPF_FUNC_map_push_elem)
5374 goto error;
5375 break;
Martin KaFai Lau6ac99e82019-04-26 16:39:39 -07005376 case BPF_MAP_TYPE_SK_STORAGE:
5377 if (func_id != BPF_FUNC_sk_storage_get &&
5378 func_id != BPF_FUNC_sk_storage_delete)
5379 goto error;
5380 break;
KP Singh8ea63682020-08-25 20:29:17 +02005381 case BPF_MAP_TYPE_INODE_STORAGE:
5382 if (func_id != BPF_FUNC_inode_storage_get &&
5383 func_id != BPF_FUNC_inode_storage_delete)
5384 goto error;
5385 break;
KP Singh4cf1bc12020-11-06 10:37:40 +00005386 case BPF_MAP_TYPE_TASK_STORAGE:
5387 if (func_id != BPF_FUNC_task_storage_get &&
5388 func_id != BPF_FUNC_task_storage_delete)
5389 goto error;
5390 break;
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005391 default:
5392 break;
5393 }
5394
5395 /* ... and second from the function itself. */
5396 switch (func_id) {
5397 case BPF_FUNC_tail_call:
5398 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
5399 goto error;
Maciej Fijalkowskie4119012020-09-16 23:10:09 +02005400 if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) {
5401 verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005402 return -EINVAL;
5403 }
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005404 break;
5405 case BPF_FUNC_perf_event_read:
5406 case BPF_FUNC_perf_event_output:
Yonghong Song908432c2017-10-05 09:19:20 -07005407 case BPF_FUNC_perf_event_read_value:
Alexei Starovoitova7658e12019-10-15 20:25:04 -07005408 case BPF_FUNC_skb_output:
Eelco Chaudrond831ee82020-03-06 08:59:23 +00005409 case BPF_FUNC_xdp_output:
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005410 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
5411 goto error;
5412 break;
Daniel Borkmann5b029a32021-08-23 21:02:09 +02005413 case BPF_FUNC_ringbuf_output:
5414 case BPF_FUNC_ringbuf_reserve:
5415 case BPF_FUNC_ringbuf_query:
5416 if (map->map_type != BPF_MAP_TYPE_RINGBUF)
5417 goto error;
5418 break;
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005419 case BPF_FUNC_get_stackid:
5420 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
5421 goto error;
5422 break;
Sargun Dhillon60d20f92016-08-12 08:56:52 -07005423 case BPF_FUNC_current_task_under_cgroup:
Daniel Borkmann747ea552016-08-12 22:17:17 +02005424 case BPF_FUNC_skb_under_cgroup:
Martin KaFai Lau4a482f32016-06-30 10:28:44 -07005425 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
5426 goto error;
5427 break;
John Fastabend97f91a72017-07-17 09:29:18 -07005428 case BPF_FUNC_redirect_map:
Jesper Dangaard Brouer9c270af2017-10-16 12:19:34 +02005429 if (map->map_type != BPF_MAP_TYPE_DEVMAP &&
Toke Høiland-Jørgensen6f9d4512019-07-26 18:06:55 +02005430 map->map_type != BPF_MAP_TYPE_DEVMAP_HASH &&
Björn Töpelfbfc504a2018-05-02 13:01:28 +02005431 map->map_type != BPF_MAP_TYPE_CPUMAP &&
5432 map->map_type != BPF_MAP_TYPE_XSKMAP)
John Fastabend97f91a72017-07-17 09:29:18 -07005433 goto error;
5434 break;
John Fastabend174a79f2017-08-15 22:32:47 -07005435 case BPF_FUNC_sk_redirect_map:
John Fastabend4f738ad2018-03-18 12:57:10 -07005436 case BPF_FUNC_msg_redirect_map:
John Fastabend81110382018-05-14 10:00:17 -07005437 case BPF_FUNC_sock_map_update:
John Fastabend174a79f2017-08-15 22:32:47 -07005438 if (map->map_type != BPF_MAP_TYPE_SOCKMAP)
5439 goto error;
5440 break;
John Fastabend81110382018-05-14 10:00:17 -07005441 case BPF_FUNC_sk_redirect_hash:
5442 case BPF_FUNC_msg_redirect_hash:
5443 case BPF_FUNC_sock_hash_update:
5444 if (map->map_type != BPF_MAP_TYPE_SOCKHASH)
John Fastabend174a79f2017-08-15 22:32:47 -07005445 goto error;
5446 break;
Roman Gushchincd339432018-08-02 14:27:24 -07005447 case BPF_FUNC_get_local_storage:
Roman Gushchinb741f162018-09-28 14:45:43 +00005448 if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
5449 map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
Roman Gushchincd339432018-08-02 14:27:24 -07005450 goto error;
5451 break;
Martin KaFai Lau2dbb9b92018-08-08 01:01:25 -07005452 case BPF_FUNC_sk_select_reuseport:
Jakub Sitnicki9fed9002020-02-18 17:10:20 +00005453 if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY &&
5454 map->map_type != BPF_MAP_TYPE_SOCKMAP &&
5455 map->map_type != BPF_MAP_TYPE_SOCKHASH)
Martin KaFai Lau2dbb9b92018-08-08 01:01:25 -07005456 goto error;
5457 break;
Mauricio Vasquez Bf1a2e442018-10-18 15:16:25 +02005458 case BPF_FUNC_map_peek_elem:
5459 case BPF_FUNC_map_pop_elem:
5460 case BPF_FUNC_map_push_elem:
5461 if (map->map_type != BPF_MAP_TYPE_QUEUE &&
5462 map->map_type != BPF_MAP_TYPE_STACK)
5463 goto error;
5464 break;
Martin KaFai Lau6ac99e82019-04-26 16:39:39 -07005465 case BPF_FUNC_sk_storage_get:
5466 case BPF_FUNC_sk_storage_delete:
5467 if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
5468 goto error;
5469 break;
KP Singh8ea63682020-08-25 20:29:17 +02005470 case BPF_FUNC_inode_storage_get:
5471 case BPF_FUNC_inode_storage_delete:
5472 if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE)
5473 goto error;
5474 break;
KP Singh4cf1bc12020-11-06 10:37:40 +00005475 case BPF_FUNC_task_storage_get:
5476 case BPF_FUNC_task_storage_delete:
5477 if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
5478 goto error;
5479 break;
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005480 default:
5481 break;
Kaixu Xia35578d72015-08-06 07:02:35 +00005482 }
5483
5484 return 0;
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005485error:
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005486 verbose(env, "cannot pass map_type %d into func %s#%d\n",
Thomas Grafebb676d2016-10-27 11:23:51 +02005487 map->map_type, func_id_name(func_id), func_id);
Alexei Starovoitov6aff67c2016-04-27 18:56:21 -07005488 return -EINVAL;
Kaixu Xia35578d72015-08-06 07:02:35 +00005489}
5490
Daniel Borkmann90133412018-01-20 01:24:29 +01005491static bool check_raw_mode_ok(const struct bpf_func_proto *fn)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005492{
5493 int count = 0;
5494
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005495 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005496 count++;
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005497 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005498 count++;
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005499 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005500 count++;
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005501 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005502 count++;
Alexei Starovoitov39f19ebb2017-01-09 10:19:50 -08005503 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
Daniel Borkmann435faee12016-04-13 00:10:51 +02005504 count++;
5505
Daniel Borkmann90133412018-01-20 01:24:29 +01005506 /* We only support one arg being in raw mode at the moment,
5507 * which is sufficient for the helper functions we have
5508 * right now.
5509 */
5510 return count <= 1;
5511}
5512
5513static bool check_args_pair_invalid(enum bpf_arg_type arg_curr,
5514 enum bpf_arg_type arg_next)
5515{
5516 return (arg_type_is_mem_ptr(arg_curr) &&
5517 !arg_type_is_mem_size(arg_next)) ||
5518 (!arg_type_is_mem_ptr(arg_curr) &&
5519 arg_type_is_mem_size(arg_next));
5520}
5521
5522static bool check_arg_pair_ok(const struct bpf_func_proto *fn)
5523{
5524 /* bpf_xxx(..., buf, len) call will access 'len'
5525 * bytes from memory 'buf'. Both arg types need
5526 * to be paired, so make sure there's no buggy
5527 * helper function specification.
5528 */
5529 if (arg_type_is_mem_size(fn->arg1_type) ||
5530 arg_type_is_mem_ptr(fn->arg5_type) ||
5531 check_args_pair_invalid(fn->arg1_type, fn->arg2_type) ||
5532 check_args_pair_invalid(fn->arg2_type, fn->arg3_type) ||
5533 check_args_pair_invalid(fn->arg3_type, fn->arg4_type) ||
5534 check_args_pair_invalid(fn->arg4_type, fn->arg5_type))
5535 return false;
5536
5537 return true;
5538}
5539
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005540static bool check_refcount_ok(const struct bpf_func_proto *fn, int func_id)
Joe Stringerfd978bf72018-10-02 13:35:35 -07005541{
5542 int count = 0;
5543
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005544 if (arg_type_may_be_refcounted(fn->arg1_type))
Joe Stringerfd978bf72018-10-02 13:35:35 -07005545 count++;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005546 if (arg_type_may_be_refcounted(fn->arg2_type))
Joe Stringerfd978bf72018-10-02 13:35:35 -07005547 count++;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005548 if (arg_type_may_be_refcounted(fn->arg3_type))
Joe Stringerfd978bf72018-10-02 13:35:35 -07005549 count++;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005550 if (arg_type_may_be_refcounted(fn->arg4_type))
Joe Stringerfd978bf72018-10-02 13:35:35 -07005551 count++;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005552 if (arg_type_may_be_refcounted(fn->arg5_type))
Joe Stringerfd978bf72018-10-02 13:35:35 -07005553 count++;
5554
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005555 /* A reference acquiring function cannot acquire
5556 * another refcounted ptr.
5557 */
Jakub Sitnicki64d85292020-04-29 20:11:52 +02005558 if (may_be_acquire_function(func_id) && count)
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005559 return false;
5560
Joe Stringerfd978bf72018-10-02 13:35:35 -07005561 /* We only support one arg being unreferenced at the moment,
5562 * which is sufficient for the helper functions we have right now.
5563 */
5564 return count <= 1;
5565}
5566
Lorenz Bauer9436ef62020-09-21 13:12:20 +01005567static bool check_btf_id_ok(const struct bpf_func_proto *fn)
5568{
5569 int i;
5570
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07005571 for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) {
Lorenz Bauer9436ef62020-09-21 13:12:20 +01005572 if (fn->arg_type[i] == ARG_PTR_TO_BTF_ID && !fn->arg_btf_id[i])
5573 return false;
5574
Martin KaFai Lau1df8f552020-09-24 17:03:50 -07005575 if (fn->arg_type[i] != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i])
5576 return false;
5577 }
5578
Lorenz Bauer9436ef62020-09-21 13:12:20 +01005579 return true;
5580}
5581
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005582static int check_func_proto(const struct bpf_func_proto *fn, int func_id)
Daniel Borkmann90133412018-01-20 01:24:29 +01005583{
5584 return check_raw_mode_ok(fn) &&
Joe Stringerfd978bf72018-10-02 13:35:35 -07005585 check_arg_pair_ok(fn) &&
Lorenz Bauer9436ef62020-09-21 13:12:20 +01005586 check_btf_id_ok(fn) &&
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005587 check_refcount_ok(fn, func_id) ? 0 : -EINVAL;
Daniel Borkmann435faee12016-04-13 00:10:51 +02005588}
5589
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02005590/* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
5591 * are now invalid, so turn them into unknown SCALAR_VALUE.
Edward Creef1174f72017-08-07 15:26:19 +01005592 */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005593static void __clear_all_pkt_pointers(struct bpf_verifier_env *env,
5594 struct bpf_func_state *state)
Alexei Starovoitov969bf052016-05-05 19:49:10 -07005595{
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01005596 struct bpf_reg_state *regs = state->regs, *reg;
Alexei Starovoitov969bf052016-05-05 19:49:10 -07005597 int i;
5598
5599 for (i = 0; i < MAX_BPF_REG; i++)
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02005600 if (reg_is_pkt_pointer_any(&regs[i]))
Jakub Kicinski61bd5212017-10-09 10:30:11 -07005601 mark_reg_unknown(env, regs, i);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07005602
Joe Stringerf3709f62018-10-02 13:35:29 -07005603 bpf_for_each_spilled_reg(i, state, reg) {
5604 if (!reg)
Alexei Starovoitov969bf052016-05-05 19:49:10 -07005605 continue;
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02005606 if (reg_is_pkt_pointer_any(reg))
Daniel Borkmannf54c7892019-12-22 23:37:40 +01005607 __mark_reg_unknown(env, reg);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07005608 }
5609}
5610
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005611static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
5612{
5613 struct bpf_verifier_state *vstate = env->cur_state;
5614 int i;
5615
5616 for (i = 0; i <= vstate->curframe; i++)
5617 __clear_all_pkt_pointers(env, vstate->frame[i]);
5618}
5619
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08005620enum {
5621 AT_PKT_END = -1,
5622 BEYOND_PKT_END = -2,
5623};
5624
5625static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open)
5626{
5627 struct bpf_func_state *state = vstate->frame[vstate->curframe];
5628 struct bpf_reg_state *reg = &state->regs[regn];
5629
5630 if (reg->type != PTR_TO_PACKET)
5631 /* PTR_TO_PACKET_META is not supported yet */
5632 return;
5633
5634 /* The 'reg' is pkt > pkt_end or pkt >= pkt_end.
5635 * How far beyond pkt_end it goes is unknown.
5636 * if (!range_open) it's the case of pkt >= pkt_end
5637 * if (range_open) it's the case of pkt > pkt_end
5638 * hence this pointer is at least 1 byte bigger than pkt_end
5639 */
5640 if (range_open)
5641 reg->range = BEYOND_PKT_END;
5642 else
5643 reg->range = AT_PKT_END;
5644}
5645
Joe Stringerfd978bf72018-10-02 13:35:35 -07005646static void release_reg_references(struct bpf_verifier_env *env,
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005647 struct bpf_func_state *state,
5648 int ref_obj_id)
Joe Stringerfd978bf72018-10-02 13:35:35 -07005649{
5650 struct bpf_reg_state *regs = state->regs, *reg;
5651 int i;
5652
5653 for (i = 0; i < MAX_BPF_REG; i++)
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005654 if (regs[i].ref_obj_id == ref_obj_id)
Joe Stringerfd978bf72018-10-02 13:35:35 -07005655 mark_reg_unknown(env, regs, i);
5656
5657 bpf_for_each_spilled_reg(i, state, reg) {
5658 if (!reg)
5659 continue;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005660 if (reg->ref_obj_id == ref_obj_id)
Daniel Borkmannf54c7892019-12-22 23:37:40 +01005661 __mark_reg_unknown(env, reg);
Joe Stringerfd978bf72018-10-02 13:35:35 -07005662 }
5663}
5664
5665/* The pointer with the specified id has released its reference to kernel
5666 * resources. Identify all copies of the same pointer and clear the reference.
5667 */
5668static int release_reference(struct bpf_verifier_env *env,
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005669 int ref_obj_id)
Joe Stringerfd978bf72018-10-02 13:35:35 -07005670{
5671 struct bpf_verifier_state *vstate = env->cur_state;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005672 int err;
Joe Stringerfd978bf72018-10-02 13:35:35 -07005673 int i;
5674
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005675 err = release_reference_state(cur_func(env), ref_obj_id);
5676 if (err)
5677 return err;
Joe Stringerfd978bf72018-10-02 13:35:35 -07005678
Martin KaFai Lau1b986582019-03-12 10:23:02 -07005679 for (i = 0; i <= vstate->curframe; i++)
5680 release_reg_references(env, vstate->frame[i], ref_obj_id);
5681
5682 return 0;
Joe Stringerfd978bf72018-10-02 13:35:35 -07005683}
5684
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005685static void clear_caller_saved_regs(struct bpf_verifier_env *env,
5686 struct bpf_reg_state *regs)
5687{
5688 int i;
5689
5690 /* after the call registers r0 - r5 were scratched */
5691 for (i = 0; i < CALLER_SAVED_REGS; i++) {
5692 mark_reg_not_init(env, regs, caller_saved[i]);
5693 check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
5694 }
5695}
5696
Yonghong Song14351372021-02-26 12:49:23 -08005697typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env,
5698 struct bpf_func_state *caller,
5699 struct bpf_func_state *callee,
5700 int insn_idx);
5701
5702static int __check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
5703 int *insn_idx, int subprog,
5704 set_callee_state_fn set_callee_state_cb)
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005705{
5706 struct bpf_verifier_state *state = env->cur_state;
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005707 struct bpf_func_info_aux *func_info_aux;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005708 struct bpf_func_state *caller, *callee;
Yonghong Song14351372021-02-26 12:49:23 -08005709 int err;
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005710 bool is_global = false;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005711
Alexei Starovoitovaada9ce2017-12-25 13:15:42 -08005712 if (state->curframe + 1 >= MAX_CALL_FRAMES) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005713 verbose(env, "the call stack of %d frames is too deep\n",
Alexei Starovoitovaada9ce2017-12-25 13:15:42 -08005714 state->curframe + 2);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005715 return -E2BIG;
5716 }
5717
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005718 caller = state->frame[state->curframe];
5719 if (state->frame[state->curframe + 1]) {
5720 verbose(env, "verifier bug. Frame %d already allocated\n",
5721 state->curframe + 1);
5722 return -EFAULT;
5723 }
5724
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005725 func_info_aux = env->prog->aux->func_info_aux;
5726 if (func_info_aux)
5727 is_global = func_info_aux[subprog].linkage == BTF_FUNC_GLOBAL;
Martin KaFai Lau34747c42021-03-24 18:51:36 -07005728 err = btf_check_subprog_arg_match(env, subprog, caller->regs);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005729 if (err == -EFAULT)
5730 return err;
5731 if (is_global) {
5732 if (err) {
5733 verbose(env, "Caller passes invalid args into func#%d\n",
5734 subprog);
5735 return err;
5736 } else {
5737 if (env->log.level & BPF_LOG_LEVEL)
5738 verbose(env,
5739 "Func#%d is global and valid. Skipping.\n",
5740 subprog);
5741 clear_caller_saved_regs(env, caller->regs);
5742
Ilya Leoshkevich45159b22021-02-12 05:04:08 +01005743 /* All global functions return a 64-bit SCALAR_VALUE */
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005744 mark_reg_unknown(env, caller->regs, BPF_REG_0);
Ilya Leoshkevich45159b22021-02-12 05:04:08 +01005745 caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG;
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005746
5747 /* continue with next insn after call */
5748 return 0;
5749 }
5750 }
5751
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07005752 if (insn->code == (BPF_JMP | BPF_CALL) &&
5753 insn->imm == BPF_FUNC_timer_set_callback) {
5754 struct bpf_verifier_state *async_cb;
5755
5756 /* there is no real recursion here. timer callbacks are async */
Alexei Starovoitov7ddc80a2021-07-14 17:54:15 -07005757 env->subprog_info[subprog].is_async_cb = true;
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07005758 async_cb = push_async_cb(env, env->subprog_info[subprog].start,
5759 *insn_idx, subprog);
5760 if (!async_cb)
5761 return -EFAULT;
5762 callee = async_cb->frame[0];
5763 callee->async_entry_cnt = caller->async_entry_cnt + 1;
5764
5765 /* Convert bpf_timer_set_callback() args into timer callback args */
5766 err = set_callee_state_cb(env, caller, callee, *insn_idx);
5767 if (err)
5768 return err;
5769
5770 clear_caller_saved_regs(env, caller->regs);
5771 mark_reg_unknown(env, caller->regs, BPF_REG_0);
5772 caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG;
5773 /* continue with next insn after call */
5774 return 0;
5775 }
5776
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005777 callee = kzalloc(sizeof(*callee), GFP_KERNEL);
5778 if (!callee)
5779 return -ENOMEM;
5780 state->frame[state->curframe + 1] = callee;
5781
5782 /* callee cannot access r0, r6 - r9 for reading and has to write
5783 * into its own stack before reading from it.
5784 * callee can read/write into caller's stack
5785 */
5786 init_func_state(env, callee,
5787 /* remember the callsite, it will be used by bpf_exit */
5788 *insn_idx /* callsite */,
5789 state->curframe + 1 /* frameno within this callchain */,
Jiong Wangf910cef2018-05-02 16:17:17 -04005790 subprog /* subprog number within this prog */);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005791
Joe Stringerfd978bf72018-10-02 13:35:35 -07005792 /* Transfer references to the callee */
Lorenz Bauerc69431a2021-04-29 14:46:54 +01005793 err = copy_reference_state(callee, caller);
Joe Stringerfd978bf72018-10-02 13:35:35 -07005794 if (err)
5795 return err;
5796
Yonghong Song14351372021-02-26 12:49:23 -08005797 err = set_callee_state_cb(env, caller, callee, *insn_idx);
5798 if (err)
5799 return err;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005800
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08005801 clear_caller_saved_regs(env, caller->regs);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005802
5803 /* only increment it after check_reg_arg() finished */
5804 state->curframe++;
5805
5806 /* and go analyze first insn of the callee */
Yonghong Song14351372021-02-26 12:49:23 -08005807 *insn_idx = env->subprog_info[subprog].start - 1;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005808
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07005809 if (env->log.level & BPF_LOG_LEVEL) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005810 verbose(env, "caller:\n");
5811 print_verifier_state(env, caller);
5812 verbose(env, "callee:\n");
5813 print_verifier_state(env, callee);
5814 }
5815 return 0;
5816}
5817
Yonghong Song314ee052021-02-26 12:49:27 -08005818int map_set_for_each_callback_args(struct bpf_verifier_env *env,
5819 struct bpf_func_state *caller,
5820 struct bpf_func_state *callee)
5821{
5822 /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn,
5823 * void *callback_ctx, u64 flags);
5824 * callback_fn(struct bpf_map *map, void *key, void *value,
5825 * void *callback_ctx);
5826 */
5827 callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1];
5828
5829 callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY;
5830 __mark_reg_known_zero(&callee->regs[BPF_REG_2]);
5831 callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr;
5832
5833 callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE;
5834 __mark_reg_known_zero(&callee->regs[BPF_REG_3]);
5835 callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr;
5836
5837 /* pointer to stack or null */
5838 callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3];
5839
5840 /* unused */
5841 __mark_reg_not_init(env, &callee->regs[BPF_REG_5]);
5842 return 0;
5843}
5844
Yonghong Song14351372021-02-26 12:49:23 -08005845static int set_callee_state(struct bpf_verifier_env *env,
5846 struct bpf_func_state *caller,
5847 struct bpf_func_state *callee, int insn_idx)
5848{
5849 int i;
5850
5851 /* copy r1 - r5 args that callee can access. The copy includes parent
5852 * pointers, which connects us up to the liveness chain
5853 */
5854 for (i = BPF_REG_1; i <= BPF_REG_5; i++)
5855 callee->regs[i] = caller->regs[i];
5856 return 0;
5857}
5858
5859static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
5860 int *insn_idx)
5861{
5862 int subprog, target_insn;
5863
5864 target_insn = *insn_idx + insn->imm + 1;
5865 subprog = find_subprog(env, target_insn);
5866 if (subprog < 0) {
5867 verbose(env, "verifier bug. No program starts at insn %d\n",
5868 target_insn);
5869 return -EFAULT;
5870 }
5871
5872 return __check_func_call(env, insn, insn_idx, subprog, set_callee_state);
5873}
5874
Yonghong Song69c087b2021-02-26 12:49:25 -08005875static int set_map_elem_callback_state(struct bpf_verifier_env *env,
5876 struct bpf_func_state *caller,
5877 struct bpf_func_state *callee,
5878 int insn_idx)
5879{
5880 struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx];
5881 struct bpf_map *map;
5882 int err;
5883
5884 if (bpf_map_ptr_poisoned(insn_aux)) {
5885 verbose(env, "tail_call abusing map_ptr\n");
5886 return -EINVAL;
5887 }
5888
5889 map = BPF_MAP_PTR(insn_aux->map_ptr_state);
5890 if (!map->ops->map_set_for_each_callback_args ||
5891 !map->ops->map_for_each_callback) {
5892 verbose(env, "callback function not allowed for map\n");
5893 return -ENOTSUPP;
5894 }
5895
5896 err = map->ops->map_set_for_each_callback_args(env, caller, callee);
5897 if (err)
5898 return err;
5899
5900 callee->in_callback_fn = true;
5901 return 0;
5902}
5903
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07005904static int set_timer_callback_state(struct bpf_verifier_env *env,
5905 struct bpf_func_state *caller,
5906 struct bpf_func_state *callee,
5907 int insn_idx)
5908{
5909 struct bpf_map *map_ptr = caller->regs[BPF_REG_1].map_ptr;
5910
5911 /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn);
5912 * callback_fn(struct bpf_map *map, void *key, void *value);
5913 */
5914 callee->regs[BPF_REG_1].type = CONST_PTR_TO_MAP;
5915 __mark_reg_known_zero(&callee->regs[BPF_REG_1]);
5916 callee->regs[BPF_REG_1].map_ptr = map_ptr;
5917
5918 callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY;
5919 __mark_reg_known_zero(&callee->regs[BPF_REG_2]);
5920 callee->regs[BPF_REG_2].map_ptr = map_ptr;
5921
5922 callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE;
5923 __mark_reg_known_zero(&callee->regs[BPF_REG_3]);
5924 callee->regs[BPF_REG_3].map_ptr = map_ptr;
5925
5926 /* unused */
5927 __mark_reg_not_init(env, &callee->regs[BPF_REG_4]);
5928 __mark_reg_not_init(env, &callee->regs[BPF_REG_5]);
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07005929 callee->in_async_callback_fn = true;
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07005930 return 0;
5931}
5932
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005933static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx)
5934{
5935 struct bpf_verifier_state *state = env->cur_state;
5936 struct bpf_func_state *caller, *callee;
5937 struct bpf_reg_state *r0;
Joe Stringerfd978bf72018-10-02 13:35:35 -07005938 int err;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005939
5940 callee = state->frame[state->curframe];
5941 r0 = &callee->regs[BPF_REG_0];
5942 if (r0->type == PTR_TO_STACK) {
5943 /* technically it's ok to return caller's stack pointer
5944 * (or caller's caller's pointer) back to the caller,
5945 * since these pointers are valid. Only current stack
5946 * pointer will be invalid as soon as function exits,
5947 * but let's be conservative
5948 */
5949 verbose(env, "cannot return stack pointer to the caller\n");
5950 return -EINVAL;
5951 }
5952
5953 state->curframe--;
5954 caller = state->frame[state->curframe];
Yonghong Song69c087b2021-02-26 12:49:25 -08005955 if (callee->in_callback_fn) {
5956 /* enforce R0 return value range [0, 1]. */
5957 struct tnum range = tnum_range(0, 1);
5958
5959 if (r0->type != SCALAR_VALUE) {
5960 verbose(env, "R0 not a scalar value\n");
5961 return -EACCES;
5962 }
5963 if (!tnum_in(range, r0->var_off)) {
5964 verbose_invalid_scalar(env, r0, &range, "callback return", "R0");
5965 return -EINVAL;
5966 }
5967 } else {
5968 /* return to the caller whatever r0 had in the callee */
5969 caller->regs[BPF_REG_0] = *r0;
5970 }
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005971
Joe Stringerfd978bf72018-10-02 13:35:35 -07005972 /* Transfer references to the caller */
Lorenz Bauerc69431a2021-04-29 14:46:54 +01005973 err = copy_reference_state(caller, callee);
Joe Stringerfd978bf72018-10-02 13:35:35 -07005974 if (err)
5975 return err;
5976
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005977 *insn_idx = callee->callsite + 1;
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07005978 if (env->log.level & BPF_LOG_LEVEL) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08005979 verbose(env, "returning from callee:\n");
5980 print_verifier_state(env, callee);
5981 verbose(env, "to caller at %d:\n", *insn_idx);
5982 print_verifier_state(env, caller);
5983 }
5984 /* clear everything in the callee */
5985 free_func_state(callee);
5986 state->frame[state->curframe + 1] = NULL;
5987 return 0;
5988}
5989
Yonghong Song849fa502018-04-28 22:28:09 -07005990static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type,
5991 int func_id,
5992 struct bpf_call_arg_meta *meta)
5993{
5994 struct bpf_reg_state *ret_reg = &regs[BPF_REG_0];
5995
5996 if (ret_type != RET_INTEGER ||
5997 (func_id != BPF_FUNC_get_stack &&
Dave Marchevskyfd0b88f2021-04-16 13:47:02 -07005998 func_id != BPF_FUNC_get_task_stack &&
Daniel Borkmann47cc0ed2020-05-15 12:11:17 +02005999 func_id != BPF_FUNC_probe_read_str &&
6000 func_id != BPF_FUNC_probe_read_kernel_str &&
6001 func_id != BPF_FUNC_probe_read_user_str))
Yonghong Song849fa502018-04-28 22:28:09 -07006002 return;
6003
John Fastabend10060502020-03-30 14:36:19 -07006004 ret_reg->smax_value = meta->msize_max_value;
John Fastabendfa123ac2020-03-30 14:36:59 -07006005 ret_reg->s32_max_value = meta->msize_max_value;
Alexei Starovoitovb02709582020-12-08 19:01:51 +01006006 ret_reg->smin_value = -MAX_ERRNO;
6007 ret_reg->s32_min_value = -MAX_ERRNO;
Yonghong Song849fa502018-04-28 22:28:09 -07006008 __reg_deduce_bounds(ret_reg);
6009 __reg_bound_offset(ret_reg);
John Fastabend10060502020-03-30 14:36:19 -07006010 __update_reg_bounds(ret_reg);
Yonghong Song849fa502018-04-28 22:28:09 -07006011}
6012
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006013static int
6014record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta,
6015 int func_id, int insn_idx)
6016{
6017 struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx];
Daniel Borkmann591fe982019-04-09 23:20:05 +02006018 struct bpf_map *map = meta->map_ptr;
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006019
6020 if (func_id != BPF_FUNC_tail_call &&
Daniel Borkmann09772d92018-06-02 23:06:35 +02006021 func_id != BPF_FUNC_map_lookup_elem &&
6022 func_id != BPF_FUNC_map_update_elem &&
Mauricio Vasquez Bf1a2e442018-10-18 15:16:25 +02006023 func_id != BPF_FUNC_map_delete_elem &&
6024 func_id != BPF_FUNC_map_push_elem &&
6025 func_id != BPF_FUNC_map_pop_elem &&
Yonghong Song69c087b2021-02-26 12:49:25 -08006026 func_id != BPF_FUNC_map_peek_elem &&
Björn Töpele6a47502021-03-08 12:29:06 +01006027 func_id != BPF_FUNC_for_each_map_elem &&
6028 func_id != BPF_FUNC_redirect_map)
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006029 return 0;
Daniel Borkmann09772d92018-06-02 23:06:35 +02006030
Daniel Borkmann591fe982019-04-09 23:20:05 +02006031 if (map == NULL) {
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006032 verbose(env, "kernel subsystem misconfigured verifier\n");
6033 return -EINVAL;
6034 }
6035
Daniel Borkmann591fe982019-04-09 23:20:05 +02006036 /* In case of read-only, some additional restrictions
6037 * need to be applied in order to prevent altering the
6038 * state of the map from program side.
6039 */
6040 if ((map->map_flags & BPF_F_RDONLY_PROG) &&
6041 (func_id == BPF_FUNC_map_delete_elem ||
6042 func_id == BPF_FUNC_map_update_elem ||
6043 func_id == BPF_FUNC_map_push_elem ||
6044 func_id == BPF_FUNC_map_pop_elem)) {
6045 verbose(env, "write into map forbidden\n");
6046 return -EACCES;
6047 }
6048
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006049 if (!BPF_MAP_PTR(aux->map_ptr_state))
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006050 bpf_map_ptr_store(aux, meta->map_ptr,
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07006051 !meta->map_ptr->bypass_spec_v1);
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006052 else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr)
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006053 bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON,
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07006054 !meta->map_ptr->bypass_spec_v1);
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006055 return 0;
6056}
6057
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006058static int
6059record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta,
6060 int func_id, int insn_idx)
6061{
6062 struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx];
6063 struct bpf_reg_state *regs = cur_regs(env), *reg;
6064 struct bpf_map *map = meta->map_ptr;
6065 struct tnum range;
6066 u64 val;
Daniel Borkmanncc52d912019-12-19 22:19:50 +01006067 int err;
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006068
6069 if (func_id != BPF_FUNC_tail_call)
6070 return 0;
6071 if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) {
6072 verbose(env, "kernel subsystem misconfigured verifier\n");
6073 return -EINVAL;
6074 }
6075
6076 range = tnum_range(0, map->max_entries - 1);
6077 reg = &regs[BPF_REG_3];
6078
6079 if (!register_is_const(reg) || !tnum_in(range, reg->var_off)) {
6080 bpf_map_key_store(aux, BPF_MAP_KEY_POISON);
6081 return 0;
6082 }
6083
Daniel Borkmanncc52d912019-12-19 22:19:50 +01006084 err = mark_chain_precision(env, BPF_REG_3);
6085 if (err)
6086 return err;
6087
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006088 val = reg->var_off.value;
6089 if (bpf_map_key_unseen(aux))
6090 bpf_map_key_store(aux, val);
6091 else if (!bpf_map_key_poisoned(aux) &&
6092 bpf_map_key_immediate(aux) != val)
6093 bpf_map_key_store(aux, BPF_MAP_KEY_POISON);
6094 return 0;
6095}
6096
Joe Stringerfd978bf72018-10-02 13:35:35 -07006097static int check_reference_leak(struct bpf_verifier_env *env)
6098{
6099 struct bpf_func_state *state = cur_func(env);
6100 int i;
6101
6102 for (i = 0; i < state->acquired_refs; i++) {
6103 verbose(env, "Unreleased reference id=%d alloc_insn=%d\n",
6104 state->refs[i].id, state->refs[i].insn_idx);
6105 }
6106 return state->acquired_refs ? -EINVAL : 0;
6107}
6108
Florent Revest7b155232021-04-19 17:52:40 +02006109static int check_bpf_snprintf_call(struct bpf_verifier_env *env,
6110 struct bpf_reg_state *regs)
6111{
6112 struct bpf_reg_state *fmt_reg = &regs[BPF_REG_3];
6113 struct bpf_reg_state *data_len_reg = &regs[BPF_REG_5];
6114 struct bpf_map *fmt_map = fmt_reg->map_ptr;
6115 int err, fmt_map_off, num_args;
6116 u64 fmt_addr;
6117 char *fmt;
6118
6119 /* data must be an array of u64 */
6120 if (data_len_reg->var_off.value % 8)
6121 return -EINVAL;
6122 num_args = data_len_reg->var_off.value / 8;
6123
6124 /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const
6125 * and map_direct_value_addr is set.
6126 */
6127 fmt_map_off = fmt_reg->off + fmt_reg->var_off.value;
6128 err = fmt_map->ops->map_direct_value_addr(fmt_map, &fmt_addr,
6129 fmt_map_off);
Florent Revest8e8ee102021-04-23 01:55:42 +02006130 if (err) {
6131 verbose(env, "verifier bug\n");
6132 return -EFAULT;
6133 }
Florent Revest7b155232021-04-19 17:52:40 +02006134 fmt = (char *)(long)fmt_addr + fmt_map_off;
6135
6136 /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we
6137 * can focus on validating the format specifiers.
6138 */
Florent Revest48cac3f2021-04-27 19:43:13 +02006139 err = bpf_bprintf_prepare(fmt, UINT_MAX, NULL, NULL, num_args);
Florent Revest7b155232021-04-19 17:52:40 +02006140 if (err < 0)
6141 verbose(env, "Invalid format string\n");
6142
6143 return err;
6144}
6145
Jiri Olsa9b99edc2021-07-14 11:43:55 +02006146static int check_get_func_ip(struct bpf_verifier_env *env)
6147{
6148 enum bpf_attach_type eatype = env->prog->expected_attach_type;
6149 enum bpf_prog_type type = resolve_prog_type(env->prog);
6150 int func_id = BPF_FUNC_get_func_ip;
6151
6152 if (type == BPF_PROG_TYPE_TRACING) {
6153 if (eatype != BPF_TRACE_FENTRY && eatype != BPF_TRACE_FEXIT &&
6154 eatype != BPF_MODIFY_RETURN) {
6155 verbose(env, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n",
6156 func_id_name(func_id), func_id);
6157 return -ENOTSUPP;
6158 }
6159 return 0;
Jiri Olsa9ffd9f32021-07-14 11:43:56 +02006160 } else if (type == BPF_PROG_TYPE_KPROBE) {
6161 return 0;
Jiri Olsa9b99edc2021-07-14 11:43:55 +02006162 }
6163
6164 verbose(env, "func %s#%d not supported for program type %d\n",
6165 func_id_name(func_id), func_id, type);
6166 return -ENOTSUPP;
6167}
6168
Yonghong Song69c087b2021-02-26 12:49:25 -08006169static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
6170 int *insn_idx_p)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006171{
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006172 const struct bpf_func_proto *fn = NULL;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07006173 struct bpf_reg_state *regs;
Daniel Borkmann33ff9822016-04-13 00:10:50 +02006174 struct bpf_call_arg_meta meta;
Yonghong Song69c087b2021-02-26 12:49:25 -08006175 int insn_idx = *insn_idx_p;
Alexei Starovoitov969bf052016-05-05 19:49:10 -07006176 bool changes_data;
Yonghong Song69c087b2021-02-26 12:49:25 -08006177 int i, err, func_id;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006178
6179 /* find function prototype */
Yonghong Song69c087b2021-02-26 12:49:25 -08006180 func_id = insn->imm;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006181 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006182 verbose(env, "invalid func %s#%d\n", func_id_name(func_id),
6183 func_id);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006184 return -EINVAL;
6185 }
6186
Jakub Kicinski00176a32017-10-16 16:40:54 -07006187 if (env->ops->get_func_proto)
Andrey Ignatov5e43f892018-03-30 15:08:00 -07006188 fn = env->ops->get_func_proto(func_id, env->prog);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006189 if (!fn) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006190 verbose(env, "unknown func %s#%d\n", func_id_name(func_id),
6191 func_id);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006192 return -EINVAL;
6193 }
6194
6195 /* eBPF programs must be GPL compatible to use GPL-ed functions */
Daniel Borkmann24701ec2015-03-01 12:31:47 +01006196 if (!env->prog->gpl_compatible && fn->gpl_only) {
Daniel Borkmann3fe28672018-06-02 23:06:33 +02006197 verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006198 return -EINVAL;
6199 }
6200
Jiri Olsaeae2e832020-08-25 21:21:19 +02006201 if (fn->allowed && !fn->allowed(env->prog)) {
6202 verbose(env, "helper call is not allowed in probe\n");
6203 return -EINVAL;
6204 }
6205
Daniel Borkmann04514d12017-12-14 21:07:25 +01006206 /* With LD_ABS/IND some JITs save/restore skb from r1. */
Martin KaFai Lau17bedab2016-12-07 15:53:11 -08006207 changes_data = bpf_helper_changes_pkt_data(fn->func);
Daniel Borkmann04514d12017-12-14 21:07:25 +01006208 if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) {
6209 verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n",
6210 func_id_name(func_id), func_id);
6211 return -EINVAL;
6212 }
Alexei Starovoitov969bf052016-05-05 19:49:10 -07006213
Daniel Borkmann33ff9822016-04-13 00:10:50 +02006214 memset(&meta, 0, sizeof(meta));
Daniel Borkmann36bbef52016-09-20 00:26:13 +02006215 meta.pkt_access = fn->pkt_access;
Daniel Borkmann33ff9822016-04-13 00:10:50 +02006216
Martin KaFai Lau1b986582019-03-12 10:23:02 -07006217 err = check_func_proto(fn, func_id);
Daniel Borkmann435faee12016-04-13 00:10:51 +02006218 if (err) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006219 verbose(env, "kernel subsystem misconfigured func %s#%d\n",
Thomas Grafebb676d2016-10-27 11:23:51 +02006220 func_id_name(func_id), func_id);
Daniel Borkmann435faee12016-04-13 00:10:51 +02006221 return err;
6222 }
6223
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08006224 meta.func_id = func_id;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006225 /* check args */
Dmitrii Banshchikov523a4cf2021-02-26 00:26:29 +04006226 for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) {
Yonghong Songaf7ec132020-06-23 16:08:09 -07006227 err = check_func_arg(env, i, &meta, fn);
Alexei Starovoitova7658e12019-10-15 20:25:04 -07006228 if (err)
6229 return err;
6230 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006231
Daniel Borkmannc93552c2018-05-24 02:32:53 +02006232 err = record_func_map(env, &meta, func_id, insn_idx);
6233 if (err)
6234 return err;
6235
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +01006236 err = record_func_key(env, &meta, func_id, insn_idx);
6237 if (err)
6238 return err;
6239
Daniel Borkmann435faee12016-04-13 00:10:51 +02006240 /* Mark slots with STACK_MISC in case of raw mode, stack offset
6241 * is inferred from register state.
6242 */
6243 for (i = 0; i < meta.access_size; i++) {
Daniel Borkmannca369602018-02-23 22:29:05 +01006244 err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B,
6245 BPF_WRITE, -1, false);
Daniel Borkmann435faee12016-04-13 00:10:51 +02006246 if (err)
6247 return err;
6248 }
6249
Joe Stringerfd978bf72018-10-02 13:35:35 -07006250 if (func_id == BPF_FUNC_tail_call) {
6251 err = check_reference_leak(env);
6252 if (err) {
6253 verbose(env, "tail_call would lead to reference leak\n");
6254 return err;
6255 }
6256 } else if (is_release_function(func_id)) {
Martin KaFai Lau1b986582019-03-12 10:23:02 -07006257 err = release_reference(env, meta.ref_obj_id);
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08006258 if (err) {
6259 verbose(env, "func %s#%d reference has not been acquired before\n",
6260 func_id_name(func_id), func_id);
Joe Stringerfd978bf72018-10-02 13:35:35 -07006261 return err;
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08006262 }
Joe Stringerfd978bf72018-10-02 13:35:35 -07006263 }
6264
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07006265 regs = cur_regs(env);
Roman Gushchincd339432018-08-02 14:27:24 -07006266
6267 /* check that flags argument in get_local_storage(map, flags) is 0,
6268 * this is required because get_local_storage() can't return an error.
6269 */
6270 if (func_id == BPF_FUNC_get_local_storage &&
6271 !register_is_null(&regs[BPF_REG_2])) {
6272 verbose(env, "get_local_storage() doesn't support non-zero flags\n");
6273 return -EINVAL;
6274 }
6275
Yonghong Song69c087b2021-02-26 12:49:25 -08006276 if (func_id == BPF_FUNC_for_each_map_elem) {
6277 err = __check_func_call(env, insn, insn_idx_p, meta.subprogno,
6278 set_map_elem_callback_state);
6279 if (err < 0)
6280 return -EINVAL;
6281 }
6282
Alexei Starovoitovb00628b2021-07-14 17:54:09 -07006283 if (func_id == BPF_FUNC_timer_set_callback) {
6284 err = __check_func_call(env, insn, insn_idx_p, meta.subprogno,
6285 set_timer_callback_state);
6286 if (err < 0)
6287 return -EINVAL;
6288 }
6289
Florent Revest7b155232021-04-19 17:52:40 +02006290 if (func_id == BPF_FUNC_snprintf) {
6291 err = check_bpf_snprintf_call(env, regs);
6292 if (err < 0)
6293 return err;
6294 }
6295
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006296 /* reset caller saved regs */
Edward Creedc503a82017-08-15 20:34:35 +01006297 for (i = 0; i < CALLER_SAVED_REGS; i++) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006298 mark_reg_not_init(env, regs, caller_saved[i]);
Edward Creedc503a82017-08-15 20:34:35 +01006299 check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
6300 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006301
Jiong Wang5327ed32019-05-24 23:25:12 +01006302 /* helper call returns 64-bit value. */
6303 regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG;
6304
Edward Creedc503a82017-08-15 20:34:35 +01006305 /* update return register (already marked as written above) */
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006306 if (fn->ret_type == RET_INTEGER) {
Edward Creef1174f72017-08-07 15:26:19 +01006307 /* sets type to SCALAR_VALUE */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006308 mark_reg_unknown(env, regs, BPF_REG_0);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006309 } else if (fn->ret_type == RET_VOID) {
6310 regs[BPF_REG_0].type = NOT_INIT;
Roman Gushchin3e6a4b32018-08-02 14:27:22 -07006311 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL ||
6312 fn->ret_type == RET_PTR_TO_MAP_VALUE) {
Edward Creef1174f72017-08-07 15:26:19 +01006313 /* There is no offset yet applied, variable or fixed */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006314 mark_reg_known_zero(env, regs, BPF_REG_0);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006315 /* remember map_ptr, so that check_map_access()
6316 * can check 'value_size' boundary of memory access
6317 * to map element returned from bpf_map_lookup_elem()
6318 */
Daniel Borkmann33ff9822016-04-13 00:10:50 +02006319 if (meta.map_ptr == NULL) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006320 verbose(env,
6321 "kernel subsystem misconfigured verifier\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006322 return -EINVAL;
6323 }
Daniel Borkmann33ff9822016-04-13 00:10:50 +02006324 regs[BPF_REG_0].map_ptr = meta.map_ptr;
Alexei Starovoitov3e8ce292021-07-14 17:54:11 -07006325 regs[BPF_REG_0].map_uid = meta.map_uid;
Daniel Borkmann4d31f302018-11-01 00:05:53 +01006326 if (fn->ret_type == RET_PTR_TO_MAP_VALUE) {
6327 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE;
Alexei Starovoitove16d2f12019-01-31 15:40:05 -08006328 if (map_value_has_spin_lock(meta.map_ptr))
6329 regs[BPF_REG_0].id = ++env->id_gen;
Daniel Borkmann4d31f302018-11-01 00:05:53 +01006330 } else {
6331 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
Daniel Borkmann4d31f302018-11-01 00:05:53 +01006332 }
Joe Stringerc64b7982018-10-02 13:35:33 -07006333 } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) {
6334 mark_reg_known_zero(env, regs, BPF_REG_0);
6335 regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL;
Lorenz Bauer85a51f82019-03-22 09:54:00 +08006336 } else if (fn->ret_type == RET_PTR_TO_SOCK_COMMON_OR_NULL) {
6337 mark_reg_known_zero(env, regs, BPF_REG_0);
6338 regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON_OR_NULL;
Martin KaFai Lau655a51e2019-02-09 23:22:24 -08006339 } else if (fn->ret_type == RET_PTR_TO_TCP_SOCK_OR_NULL) {
6340 mark_reg_known_zero(env, regs, BPF_REG_0);
6341 regs[BPF_REG_0].type = PTR_TO_TCP_SOCK_OR_NULL;
Andrii Nakryiko457f4432020-05-29 00:54:20 -07006342 } else if (fn->ret_type == RET_PTR_TO_ALLOC_MEM_OR_NULL) {
6343 mark_reg_known_zero(env, regs, BPF_REG_0);
6344 regs[BPF_REG_0].type = PTR_TO_MEM_OR_NULL;
Andrii Nakryiko457f4432020-05-29 00:54:20 -07006345 regs[BPF_REG_0].mem_size = meta.mem_size;
Hao Luo63d9b802020-09-29 16:50:48 -07006346 } else if (fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL ||
6347 fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID) {
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006348 const struct btf_type *t;
6349
6350 mark_reg_known_zero(env, regs, BPF_REG_0);
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08006351 t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL);
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006352 if (!btf_type_is_struct(t)) {
6353 u32 tsize;
6354 const struct btf_type *ret;
6355 const char *tname;
6356
6357 /* resolve the type size of ksym. */
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08006358 ret = btf_resolve_size(meta.ret_btf, t, &tsize);
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006359 if (IS_ERR(ret)) {
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08006360 tname = btf_name_by_offset(meta.ret_btf, t->name_off);
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006361 verbose(env, "unable to resolve the size of type '%s': %ld\n",
6362 tname, PTR_ERR(ret));
6363 return -EINVAL;
6364 }
Hao Luo63d9b802020-09-29 16:50:48 -07006365 regs[BPF_REG_0].type =
6366 fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ?
6367 PTR_TO_MEM : PTR_TO_MEM_OR_NULL;
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006368 regs[BPF_REG_0].mem_size = tsize;
6369 } else {
Hao Luo63d9b802020-09-29 16:50:48 -07006370 regs[BPF_REG_0].type =
6371 fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ?
6372 PTR_TO_BTF_ID : PTR_TO_BTF_ID_OR_NULL;
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08006373 regs[BPF_REG_0].btf = meta.ret_btf;
Hao Luoeaa6bcb2020-09-29 16:50:47 -07006374 regs[BPF_REG_0].btf_id = meta.ret_btf_id;
6375 }
KP Singh3ca10322020-11-06 10:37:43 +00006376 } else if (fn->ret_type == RET_PTR_TO_BTF_ID_OR_NULL ||
6377 fn->ret_type == RET_PTR_TO_BTF_ID) {
Yonghong Songaf7ec132020-06-23 16:08:09 -07006378 int ret_btf_id;
6379
6380 mark_reg_known_zero(env, regs, BPF_REG_0);
KP Singh3ca10322020-11-06 10:37:43 +00006381 regs[BPF_REG_0].type = fn->ret_type == RET_PTR_TO_BTF_ID ?
6382 PTR_TO_BTF_ID :
6383 PTR_TO_BTF_ID_OR_NULL;
Yonghong Songaf7ec132020-06-23 16:08:09 -07006384 ret_btf_id = *fn->ret_btf_id;
6385 if (ret_btf_id == 0) {
6386 verbose(env, "invalid return type %d of func %s#%d\n",
6387 fn->ret_type, func_id_name(func_id), func_id);
6388 return -EINVAL;
6389 }
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08006390 /* current BPF helper definitions are only coming from
6391 * built-in code with type IDs from vmlinux BTF
6392 */
6393 regs[BPF_REG_0].btf = btf_vmlinux;
Yonghong Songaf7ec132020-06-23 16:08:09 -07006394 regs[BPF_REG_0].btf_id = ret_btf_id;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006395 } else {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006396 verbose(env, "unknown return type %d of func %s#%d\n",
Thomas Grafebb676d2016-10-27 11:23:51 +02006397 fn->ret_type, func_id_name(func_id), func_id);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07006398 return -EINVAL;
6399 }
Alexei Starovoitov04fd61ab2015-05-19 16:59:03 -07006400
Martin KaFai Lau93c230e2020-10-19 12:42:12 -07006401 if (reg_type_may_be_null(regs[BPF_REG_0].type))
6402 regs[BPF_REG_0].id = ++env->id_gen;
6403
Lorenz Bauer0f3adc22019-03-22 09:53:59 +08006404 if (is_ptr_cast_function(func_id)) {
Martin KaFai Lau1b986582019-03-12 10:23:02 -07006405 /* For release_reference() */
6406 regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id;
Jakub Sitnicki64d85292020-04-29 20:11:52 +02006407 } else if (is_acquire_function(func_id, meta.map_ptr)) {
Lorenz Bauer0f3adc22019-03-22 09:53:59 +08006408 int id = acquire_reference_state(env, insn_idx);
6409
6410 if (id < 0)
6411 return id;
6412 /* For mark_ptr_or_null_reg() */
6413 regs[BPF_REG_0].id = id;
6414 /* For release_reference() */
6415 regs[BPF_REG_0].ref_obj_id = id;
6416 }
Martin KaFai Lau1b986582019-03-12 10:23:02 -07006417
Yonghong Song849fa502018-04-28 22:28:09 -07006418 do_refine_retval_range(regs, fn->ret_type, func_id, &meta);
6419
Jakub Kicinski61bd5212017-10-09 10:30:11 -07006420 err = check_map_func_compatibility(env, meta.map_ptr, func_id);
Kaixu Xia35578d72015-08-06 07:02:35 +00006421 if (err)
6422 return err;
Alexei Starovoitov04fd61ab2015-05-19 16:59:03 -07006423
Song Liufa28dcb2020-06-29 23:28:44 -07006424 if ((func_id == BPF_FUNC_get_stack ||
6425 func_id == BPF_FUNC_get_task_stack) &&
6426 !env->prog->has_callchain_buf) {
Yonghong Songc195651e2018-04-28 22:28:08 -07006427 const char *err_str;
6428
6429#ifdef CONFIG_PERF_EVENTS
6430 err = get_callchain_buffers(sysctl_perf_event_max_stack);
6431 err_str = "cannot get callchain buffer for func %s#%d\n";
6432#else
6433 err = -ENOTSUPP;
6434 err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n";
6435#endif
6436 if (err) {
6437 verbose(env, err_str, func_id_name(func_id), func_id);
6438 return err;
6439 }
6440
6441 env->prog->has_callchain_buf = true;
6442 }
6443
Song Liu5d99cb2c2020-07-23 11:06:45 -07006444 if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack)
6445 env->prog->call_get_stack = true;
6446
Jiri Olsa9b99edc2021-07-14 11:43:55 +02006447 if (func_id == BPF_FUNC_get_func_ip) {
6448 if (check_get_func_ip(env))
6449 return -ENOTSUPP;
6450 env->prog->call_get_func_ip = true;
6451 }
6452
Alexei Starovoitov969bf052016-05-05 19:49:10 -07006453 if (changes_data)
6454 clear_all_pkt_pointers(env);
6455 return 0;
6456}
6457
Martin KaFai Laue6ac2452021-03-24 18:51:42 -07006458/* mark_btf_func_reg_size() is used when the reg size is determined by
6459 * the BTF func_proto's return value size and argument.
6460 */
6461static void mark_btf_func_reg_size(struct bpf_verifier_env *env, u32 regno,
6462 size_t reg_size)
6463{
6464 struct bpf_reg_state *reg = &cur_regs(env)[regno];
6465
6466 if (regno == BPF_REG_0) {
6467 /* Function return value */
6468 reg->live |= REG_LIVE_WRITTEN;
6469 reg->subreg_def = reg_size == sizeof(u64) ?
6470 DEF_NOT_SUBREG : env->insn_idx + 1;
6471 } else {
6472 /* Function argument */
6473 if (reg_size == sizeof(u64)) {
6474 mark_insn_zext(env, reg);
6475 mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64);
6476 } else {
6477 mark_reg_read(env, reg, reg->parent, REG_LIVE_READ32);
6478 }
6479 }
6480}
6481
6482static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn)
6483{
6484 const struct btf_type *t, *func, *func_proto, *ptr_type;
6485 struct bpf_reg_state *regs = cur_regs(env);
6486 const char *func_name, *ptr_type_name;
6487 u32 i, nargs, func_id, ptr_type_id;
6488 const struct btf_param *args;
6489 int err;
6490
6491 func_id = insn->imm;
6492 func = btf_type_by_id(btf_vmlinux, func_id);
6493 func_name = btf_name_by_offset(btf_vmlinux, func->name_off);
6494 func_proto = btf_type_by_id(btf_vmlinux, func->type);
6495
6496 if (!env->ops->check_kfunc_call ||
6497 !env->ops->check_kfunc_call(func_id)) {
6498 verbose(env, "calling kernel function %s is not allowed\n",
6499 func_name);
6500 return -EACCES;
6501 }
6502
6503 /* Check the arguments */
6504 err = btf_check_kfunc_arg_match(env, btf_vmlinux, func_id, regs);
6505 if (err)
6506 return err;
6507
6508 for (i = 0; i < CALLER_SAVED_REGS; i++)
6509 mark_reg_not_init(env, regs, caller_saved[i]);
6510
6511 /* Check return type */
6512 t = btf_type_skip_modifiers(btf_vmlinux, func_proto->type, NULL);
6513 if (btf_type_is_scalar(t)) {
6514 mark_reg_unknown(env, regs, BPF_REG_0);
6515 mark_btf_func_reg_size(env, BPF_REG_0, t->size);
6516 } else if (btf_type_is_ptr(t)) {
6517 ptr_type = btf_type_skip_modifiers(btf_vmlinux, t->type,
6518 &ptr_type_id);
6519 if (!btf_type_is_struct(ptr_type)) {
6520 ptr_type_name = btf_name_by_offset(btf_vmlinux,
6521 ptr_type->name_off);
6522 verbose(env, "kernel function %s returns pointer type %s %s is not supported\n",
6523 func_name, btf_type_str(ptr_type),
6524 ptr_type_name);
6525 return -EINVAL;
6526 }
6527 mark_reg_known_zero(env, regs, BPF_REG_0);
6528 regs[BPF_REG_0].btf = btf_vmlinux;
6529 regs[BPF_REG_0].type = PTR_TO_BTF_ID;
6530 regs[BPF_REG_0].btf_id = ptr_type_id;
6531 mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *));
6532 } /* else { add_kfunc_call() ensures it is btf_type_is_void(t) } */
6533
6534 nargs = btf_type_vlen(func_proto);
6535 args = (const struct btf_param *)(func_proto + 1);
6536 for (i = 0; i < nargs; i++) {
6537 u32 regno = i + 1;
6538
6539 t = btf_type_skip_modifiers(btf_vmlinux, args[i].type, NULL);
6540 if (btf_type_is_ptr(t))
6541 mark_btf_func_reg_size(env, regno, sizeof(void *));
6542 else
6543 /* scalar. ensured by btf_check_kfunc_arg_match() */
6544 mark_btf_func_reg_size(env, regno, t->size);
6545 }
6546
6547 return 0;
6548}
6549
Edward Creeb03c9f92017-08-07 15:26:36 +01006550static bool signed_add_overflows(s64 a, s64 b)
6551{
6552 /* Do the add in u64, where overflow is well-defined */
6553 s64 res = (s64)((u64)a + (u64)b);
6554
6555 if (b < 0)
6556 return res > a;
6557 return res < a;
6558}
6559
Daniel Borkmannbc895e82021-01-20 00:24:24 +01006560static bool signed_add32_overflows(s32 a, s32 b)
John Fastabend3f50f132020-03-30 14:36:39 -07006561{
6562 /* Do the add in u32, where overflow is well-defined */
6563 s32 res = (s32)((u32)a + (u32)b);
6564
6565 if (b < 0)
6566 return res > a;
6567 return res < a;
6568}
6569
Daniel Borkmannbc895e82021-01-20 00:24:24 +01006570static bool signed_sub_overflows(s64 a, s64 b)
Edward Creeb03c9f92017-08-07 15:26:36 +01006571{
6572 /* Do the sub in u64, where overflow is well-defined */
6573 s64 res = (s64)((u64)a - (u64)b);
6574
6575 if (b < 0)
6576 return res < a;
6577 return res > a;
David S. Millerd1174412017-05-10 11:22:52 -07006578}
6579
John Fastabend3f50f132020-03-30 14:36:39 -07006580static bool signed_sub32_overflows(s32 a, s32 b)
6581{
Daniel Borkmannbc895e82021-01-20 00:24:24 +01006582 /* Do the sub in u32, where overflow is well-defined */
John Fastabend3f50f132020-03-30 14:36:39 -07006583 s32 res = (s32)((u32)a - (u32)b);
6584
6585 if (b < 0)
6586 return res < a;
6587 return res > a;
6588}
6589
Alexei Starovoitovbb7f0f92017-12-18 20:12:00 -08006590static bool check_reg_sane_offset(struct bpf_verifier_env *env,
6591 const struct bpf_reg_state *reg,
6592 enum bpf_reg_type type)
6593{
6594 bool known = tnum_is_const(reg->var_off);
6595 s64 val = reg->var_off.value;
6596 s64 smin = reg->smin_value;
6597
6598 if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) {
6599 verbose(env, "math between %s pointer and %lld is not allowed\n",
6600 reg_type_str[type], val);
6601 return false;
6602 }
6603
6604 if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) {
6605 verbose(env, "%s pointer offset %d is not allowed\n",
6606 reg_type_str[type], reg->off);
6607 return false;
6608 }
6609
6610 if (smin == S64_MIN) {
6611 verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n",
6612 reg_type_str[type]);
6613 return false;
6614 }
6615
6616 if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) {
6617 verbose(env, "value %lld makes %s pointer be out of bounds\n",
6618 smin, reg_type_str[type]);
6619 return false;
6620 }
6621
6622 return true;
6623}
6624
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006625static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env)
6626{
6627 return &env->insn_aux_data[env->insn_idx];
6628}
6629
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006630enum {
6631 REASON_BOUNDS = -1,
6632 REASON_TYPE = -2,
6633 REASON_PATHS = -3,
6634 REASON_LIMIT = -4,
6635 REASON_STACK = -5,
6636};
6637
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006638static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg,
Daniel Borkmannbb01a1b2021-05-21 10:19:22 +00006639 u32 *alu_limit, bool mask_to_left)
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006640{
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006641 u32 max = 0, ptr_limit = 0;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006642
6643 switch (ptr_reg->type) {
6644 case PTR_TO_STACK:
Piotr Krysiuk1b1597e2021-03-16 09:47:02 +01006645 /* Offset 0 is out-of-bounds, but acceptable start for the
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006646 * left direction, see BPF_REG_FP. Also, unknown scalar
6647 * offset where we would need to deal with min/max bounds is
6648 * currently prohibited for unprivileged.
Piotr Krysiuk1b1597e2021-03-16 09:47:02 +01006649 */
6650 max = MAX_BPF_STACK + mask_to_left;
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006651 ptr_limit = -(ptr_reg->var_off.value + ptr_reg->off);
Daniel Borkmannb658bbb2021-03-23 09:04:10 +01006652 break;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006653 case PTR_TO_MAP_VALUE:
Piotr Krysiuk1b1597e2021-03-16 09:47:02 +01006654 max = ptr_reg->map_ptr->value_size;
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006655 ptr_limit = (mask_to_left ?
6656 ptr_reg->smin_value :
6657 ptr_reg->umax_value) + ptr_reg->off;
Daniel Borkmannb658bbb2021-03-23 09:04:10 +01006658 break;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006659 default:
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006660 return REASON_TYPE;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006661 }
Daniel Borkmannb658bbb2021-03-23 09:04:10 +01006662
6663 if (ptr_limit >= max)
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006664 return REASON_LIMIT;
Daniel Borkmannb658bbb2021-03-23 09:04:10 +01006665 *alu_limit = ptr_limit;
6666 return 0;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006667}
6668
Daniel Borkmannd3bd7412019-01-06 00:54:37 +01006669static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env,
6670 const struct bpf_insn *insn)
6671{
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07006672 return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K;
Daniel Borkmannd3bd7412019-01-06 00:54:37 +01006673}
6674
6675static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux,
6676 u32 alu_state, u32 alu_limit)
6677{
6678 /* If we arrived here from different branches with different
6679 * state or limits to sanitize, then this won't work.
6680 */
6681 if (aux->alu_state &&
6682 (aux->alu_state != alu_state ||
6683 aux->alu_limit != alu_limit))
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006684 return REASON_PATHS;
Daniel Borkmannd3bd7412019-01-06 00:54:37 +01006685
Brendan Jackmane6ac5932021-02-17 10:45:09 +00006686 /* Corresponding fixup done in do_misc_fixups(). */
Daniel Borkmannd3bd7412019-01-06 00:54:37 +01006687 aux->alu_state = alu_state;
6688 aux->alu_limit = alu_limit;
6689 return 0;
6690}
6691
6692static int sanitize_val_alu(struct bpf_verifier_env *env,
6693 struct bpf_insn *insn)
6694{
6695 struct bpf_insn_aux_data *aux = cur_aux(env);
6696
6697 if (can_skip_alu_sanitation(env, insn))
6698 return 0;
6699
6700 return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0);
6701}
6702
Daniel Borkmannf5288192021-03-24 11:25:39 +01006703static bool sanitize_needed(u8 opcode)
6704{
6705 return opcode == BPF_ADD || opcode == BPF_SUB;
6706}
6707
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006708struct bpf_sanitize_info {
6709 struct bpf_insn_aux_data aux;
Daniel Borkmannbb01a1b2021-05-21 10:19:22 +00006710 bool mask_to_left;
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006711};
6712
Daniel Borkmann9183671a2021-05-28 15:47:32 +00006713static struct bpf_verifier_state *
6714sanitize_speculative_path(struct bpf_verifier_env *env,
6715 const struct bpf_insn *insn,
6716 u32 next_idx, u32 curr_idx)
6717{
6718 struct bpf_verifier_state *branch;
6719 struct bpf_reg_state *regs;
6720
6721 branch = push_stack(env, next_idx, curr_idx, true);
6722 if (branch && insn) {
6723 regs = branch->frame[branch->curframe]->regs;
6724 if (BPF_SRC(insn->code) == BPF_K) {
6725 mark_reg_unknown(env, regs, insn->dst_reg);
6726 } else if (BPF_SRC(insn->code) == BPF_X) {
6727 mark_reg_unknown(env, regs, insn->dst_reg);
6728 mark_reg_unknown(env, regs, insn->src_reg);
6729 }
6730 }
6731 return branch;
6732}
6733
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006734static int sanitize_ptr_alu(struct bpf_verifier_env *env,
6735 struct bpf_insn *insn,
6736 const struct bpf_reg_state *ptr_reg,
Daniel Borkmann6f55b2f2021-03-22 15:45:52 +01006737 const struct bpf_reg_state *off_reg,
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006738 struct bpf_reg_state *dst_reg,
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006739 struct bpf_sanitize_info *info,
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006740 const bool commit_window)
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006741{
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006742 struct bpf_insn_aux_data *aux = commit_window ? cur_aux(env) : &info->aux;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006743 struct bpf_verifier_state *vstate = env->cur_state;
Daniel Borkmann801c6052021-04-29 15:19:37 +00006744 bool off_is_imm = tnum_is_const(off_reg->var_off);
Daniel Borkmann6f55b2f2021-03-22 15:45:52 +01006745 bool off_is_neg = off_reg->smin_value < 0;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006746 bool ptr_is_dst_reg = ptr_reg == dst_reg;
6747 u8 opcode = BPF_OP(insn->code);
6748 u32 alu_state, alu_limit;
6749 struct bpf_reg_state tmp;
6750 bool ret;
Piotr Krysiukf2323262021-03-16 09:47:02 +01006751 int err;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006752
Daniel Borkmannd3bd7412019-01-06 00:54:37 +01006753 if (can_skip_alu_sanitation(env, insn))
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006754 return 0;
6755
6756 /* We already marked aux for masking from non-speculative
6757 * paths, thus we got here in the first place. We only care
6758 * to explore bad access from here.
6759 */
6760 if (vstate->speculative)
6761 goto do_sim;
6762
Daniel Borkmannbb01a1b2021-05-21 10:19:22 +00006763 if (!commit_window) {
6764 if (!tnum_is_const(off_reg->var_off) &&
6765 (off_reg->smin_value < 0) != (off_reg->smax_value < 0))
6766 return REASON_BOUNDS;
6767
6768 info->mask_to_left = (opcode == BPF_ADD && off_is_neg) ||
6769 (opcode == BPF_SUB && !off_is_neg);
6770 }
6771
6772 err = retrieve_ptr_limit(ptr_reg, &alu_limit, info->mask_to_left);
Piotr Krysiukf2323262021-03-16 09:47:02 +01006773 if (err < 0)
6774 return err;
6775
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006776 if (commit_window) {
6777 /* In commit phase we narrow the masking window based on
6778 * the observed pointer move after the simulated operation.
6779 */
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006780 alu_state = info->aux.alu_state;
6781 alu_limit = abs(info->aux.alu_limit - alu_limit);
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006782 } else {
6783 alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0;
Daniel Borkmann801c6052021-04-29 15:19:37 +00006784 alu_state |= off_is_imm ? BPF_ALU_IMMEDIATE : 0;
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006785 alu_state |= ptr_is_dst_reg ?
6786 BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST;
Daniel Borkmanne042aa52021-07-16 09:18:21 +00006787
6788 /* Limit pruning on unknown scalars to enable deep search for
6789 * potential masking differences from other program paths.
6790 */
6791 if (!off_is_imm)
6792 env->explore_alu_limits = true;
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006793 }
6794
Piotr Krysiukf2323262021-03-16 09:47:02 +01006795 err = update_alu_sanitation_state(aux, alu_state, alu_limit);
6796 if (err < 0)
6797 return err;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006798do_sim:
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006799 /* If we're in commit phase, we're done here given we already
6800 * pushed the truncated dst_reg into the speculative verification
6801 * stack.
Daniel Borkmanna7036192021-05-04 08:58:25 +00006802 *
6803 * Also, when register is a known constant, we rewrite register-based
6804 * operation to immediate-based, and thus do not need masking (and as
6805 * a consequence, do not need to simulate the zero-truncation either).
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006806 */
Daniel Borkmanna7036192021-05-04 08:58:25 +00006807 if (commit_window || off_is_imm)
Daniel Borkmann7fedb632021-03-24 10:38:26 +01006808 return 0;
6809
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006810 /* Simulate and find potential out-of-bounds access under
6811 * speculative execution from truncation as a result of
6812 * masking when off was not within expected range. If off
6813 * sits in dst, then we temporarily need to move ptr there
6814 * to simulate dst (== 0) +/-= ptr. Needed, for example,
6815 * for cases where we use K-based arithmetic in one direction
6816 * and truncated reg-based in the other in order to explore
6817 * bad access.
6818 */
6819 if (!ptr_is_dst_reg) {
6820 tmp = *dst_reg;
6821 *dst_reg = *ptr_reg;
6822 }
Daniel Borkmann9183671a2021-05-28 15:47:32 +00006823 ret = sanitize_speculative_path(env, NULL, env->insn_idx + 1,
6824 env->insn_idx);
Xu Yu08032782019-03-21 18:00:35 +08006825 if (!ptr_is_dst_reg && ret)
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006826 *dst_reg = tmp;
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006827 return !ret ? REASON_STACK : 0;
6828}
6829
Daniel Borkmannfe9a5ca2021-05-28 13:47:27 +00006830static void sanitize_mark_insn_seen(struct bpf_verifier_env *env)
6831{
6832 struct bpf_verifier_state *vstate = env->cur_state;
6833
6834 /* If we simulate paths under speculation, we don't update the
6835 * insn as 'seen' such that when we verify unreachable paths in
6836 * the non-speculative domain, sanitize_dead_code() can still
6837 * rewrite/sanitize them.
6838 */
6839 if (!vstate->speculative)
6840 env->insn_aux_data[env->insn_idx].seen = env->pass_cnt;
6841}
6842
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01006843static int sanitize_err(struct bpf_verifier_env *env,
6844 const struct bpf_insn *insn, int reason,
6845 const struct bpf_reg_state *off_reg,
6846 const struct bpf_reg_state *dst_reg)
6847{
6848 static const char *err = "pointer arithmetic with it prohibited for !root";
6849 const char *op = BPF_OP(insn->code) == BPF_ADD ? "add" : "sub";
6850 u32 dst = insn->dst_reg, src = insn->src_reg;
6851
6852 switch (reason) {
6853 case REASON_BOUNDS:
6854 verbose(env, "R%d has unknown scalar with mixed signed bounds, %s\n",
6855 off_reg == dst_reg ? dst : src, err);
6856 break;
6857 case REASON_TYPE:
6858 verbose(env, "R%d has pointer with unsupported alu operation, %s\n",
6859 off_reg == dst_reg ? src : dst, err);
6860 break;
6861 case REASON_PATHS:
6862 verbose(env, "R%d tried to %s from different maps, paths or scalars, %s\n",
6863 dst, op, err);
6864 break;
6865 case REASON_LIMIT:
6866 verbose(env, "R%d tried to %s beyond pointer bounds, %s\n",
6867 dst, op, err);
6868 break;
6869 case REASON_STACK:
6870 verbose(env, "R%d could not be pushed for speculative verification, %s\n",
6871 dst, err);
6872 break;
6873 default:
6874 verbose(env, "verifier internal error: unknown reason (%d)\n",
6875 reason);
6876 break;
6877 }
6878
6879 return -EACCES;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006880}
6881
Andrei Matei01f810a2021-02-06 20:10:24 -05006882/* check that stack access falls within stack limits and that 'reg' doesn't
6883 * have a variable offset.
6884 *
6885 * Variable offset is prohibited for unprivileged mode for simplicity since it
6886 * requires corresponding support in Spectre masking for stack ALU. See also
6887 * retrieve_ptr_limit().
6888 *
6889 *
6890 * 'off' includes 'reg->off'.
6891 */
6892static int check_stack_access_for_ptr_arithmetic(
6893 struct bpf_verifier_env *env,
6894 int regno,
6895 const struct bpf_reg_state *reg,
6896 int off)
6897{
6898 if (!tnum_is_const(reg->var_off)) {
6899 char tn_buf[48];
6900
6901 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
6902 verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n",
6903 regno, tn_buf, off);
6904 return -EACCES;
6905 }
6906
6907 if (off >= 0 || off < -MAX_BPF_STACK) {
6908 verbose(env, "R%d stack pointer arithmetic goes out of range, "
6909 "prohibited for !root; off=%d\n", regno, off);
6910 return -EACCES;
6911 }
6912
6913 return 0;
6914}
6915
Daniel Borkmann073815b2021-03-23 15:05:48 +01006916static int sanitize_check_bounds(struct bpf_verifier_env *env,
6917 const struct bpf_insn *insn,
6918 const struct bpf_reg_state *dst_reg)
6919{
6920 u32 dst = insn->dst_reg;
6921
6922 /* For unprivileged we require that resulting offset must be in bounds
6923 * in order to be able to sanitize access later on.
6924 */
6925 if (env->bypass_spec_v1)
6926 return 0;
6927
6928 switch (dst_reg->type) {
6929 case PTR_TO_STACK:
6930 if (check_stack_access_for_ptr_arithmetic(env, dst, dst_reg,
6931 dst_reg->off + dst_reg->var_off.value))
6932 return -EACCES;
6933 break;
6934 case PTR_TO_MAP_VALUE:
6935 if (check_map_access(env, dst, dst_reg->off, 1, false)) {
6936 verbose(env, "R%d pointer arithmetic of map value goes out of range, "
6937 "prohibited for !root\n", dst);
6938 return -EACCES;
6939 }
6940 break;
6941 default:
6942 break;
6943 }
6944
6945 return 0;
6946}
Andrei Matei01f810a2021-02-06 20:10:24 -05006947
Edward Creef1174f72017-08-07 15:26:19 +01006948/* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
Edward Creef1174f72017-08-07 15:26:19 +01006949 * Caller should also handle BPF_MOV case separately.
6950 * If we return -EACCES, caller may want to try again treating pointer as a
6951 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
6952 */
6953static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
6954 struct bpf_insn *insn,
6955 const struct bpf_reg_state *ptr_reg,
6956 const struct bpf_reg_state *off_reg)
Josef Bacik48461132016-09-28 10:54:32 -04006957{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08006958 struct bpf_verifier_state *vstate = env->cur_state;
6959 struct bpf_func_state *state = vstate->frame[vstate->curframe];
6960 struct bpf_reg_state *regs = state->regs, *dst_reg;
Edward Creef1174f72017-08-07 15:26:19 +01006961 bool known = tnum_is_const(off_reg->var_off);
Edward Creeb03c9f92017-08-07 15:26:36 +01006962 s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value,
6963 smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value;
6964 u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value,
6965 umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value;
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00006966 struct bpf_sanitize_info info = {};
Josef Bacik48461132016-09-28 10:54:32 -04006967 u8 opcode = BPF_OP(insn->code);
Daniel Borkmann24c109b2021-03-23 08:51:02 +01006968 u32 dst = insn->dst_reg;
Daniel Borkmann979d63d2019-01-03 00:58:34 +01006969 int ret;
Josef Bacik48461132016-09-28 10:54:32 -04006970
Edward Creef1174f72017-08-07 15:26:19 +01006971 dst_reg = &regs[dst];
Josef Bacik48461132016-09-28 10:54:32 -04006972
Daniel Borkmann6f161012018-01-18 01:15:21 +01006973 if ((known && (smin_val != smax_val || umin_val != umax_val)) ||
6974 smin_val > smax_val || umin_val > umax_val) {
6975 /* Taint dst register if offset had invalid bounds derived from
6976 * e.g. dead branches.
6977 */
Daniel Borkmannf54c7892019-12-22 23:37:40 +01006978 __mark_reg_unknown(env, dst_reg);
Daniel Borkmann6f161012018-01-18 01:15:21 +01006979 return 0;
Josef Bacik48461132016-09-28 10:54:32 -04006980 }
6981
Edward Creef1174f72017-08-07 15:26:19 +01006982 if (BPF_CLASS(insn->code) != BPF_ALU64) {
6983 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
Yonghong Song6c693542020-06-18 16:46:31 -07006984 if (opcode == BPF_SUB && env->allow_ptr_leaks) {
6985 __mark_reg_unknown(env, dst_reg);
6986 return 0;
6987 }
6988
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08006989 verbose(env,
6990 "R%d 32-bit pointer arithmetic prohibited\n",
6991 dst);
Edward Creef1174f72017-08-07 15:26:19 +01006992 return -EACCES;
6993 }
David S. Millerd1174412017-05-10 11:22:52 -07006994
Joe Stringeraad2eea2018-10-02 13:35:30 -07006995 switch (ptr_reg->type) {
6996 case PTR_TO_MAP_VALUE_OR_NULL:
6997 verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n",
6998 dst, reg_type_str[ptr_reg->type]);
Edward Creef1174f72017-08-07 15:26:19 +01006999 return -EACCES;
Joe Stringeraad2eea2018-10-02 13:35:30 -07007000 case CONST_PTR_TO_MAP:
Yonghong Song7c696732020-09-08 10:57:02 -07007001 /* smin_val represents the known value */
7002 if (known && smin_val == 0 && opcode == BPF_ADD)
7003 break;
Gustavo A. R. Silva87317452020-10-02 18:42:17 -05007004 fallthrough;
Joe Stringeraad2eea2018-10-02 13:35:30 -07007005 case PTR_TO_PACKET_END:
Joe Stringerc64b7982018-10-02 13:35:33 -07007006 case PTR_TO_SOCKET:
7007 case PTR_TO_SOCKET_OR_NULL:
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -08007008 case PTR_TO_SOCK_COMMON:
7009 case PTR_TO_SOCK_COMMON_OR_NULL:
Martin KaFai Lau655a51e2019-02-09 23:22:24 -08007010 case PTR_TO_TCP_SOCK:
7011 case PTR_TO_TCP_SOCK_OR_NULL:
Jonathan Lemonfada7fd2019-06-06 13:59:40 -07007012 case PTR_TO_XDP_SOCK:
Joe Stringeraad2eea2018-10-02 13:35:30 -07007013 verbose(env, "R%d pointer arithmetic on %s prohibited\n",
7014 dst, reg_type_str[ptr_reg->type]);
Edward Creef1174f72017-08-07 15:26:19 +01007015 return -EACCES;
Joe Stringeraad2eea2018-10-02 13:35:30 -07007016 default:
7017 break;
Edward Creef1174f72017-08-07 15:26:19 +01007018 }
7019
7020 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
7021 * The id may be overwritten later if we create a new variable offset.
Josef Bacik48461132016-09-28 10:54:32 -04007022 */
Edward Creef1174f72017-08-07 15:26:19 +01007023 dst_reg->type = ptr_reg->type;
7024 dst_reg->id = ptr_reg->id;
Josef Bacikf23cc642016-11-14 15:45:36 -05007025
Alexei Starovoitovbb7f0f92017-12-18 20:12:00 -08007026 if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) ||
7027 !check_reg_sane_offset(env, ptr_reg, ptr_reg->type))
7028 return -EINVAL;
7029
John Fastabend3f50f132020-03-30 14:36:39 -07007030 /* pointer types do not carry 32-bit bounds at the moment. */
7031 __mark_reg32_unbounded(dst_reg);
7032
Daniel Borkmann7fedb632021-03-24 10:38:26 +01007033 if (sanitize_needed(opcode)) {
7034 ret = sanitize_ptr_alu(env, insn, ptr_reg, off_reg, dst_reg,
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00007035 &info, false);
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01007036 if (ret < 0)
7037 return sanitize_err(env, insn, ret, off_reg, dst_reg);
Daniel Borkmann7fedb632021-03-24 10:38:26 +01007038 }
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01007039
Josef Bacik48461132016-09-28 10:54:32 -04007040 switch (opcode) {
7041 case BPF_ADD:
Edward Creef1174f72017-08-07 15:26:19 +01007042 /* We can take a fixed offset as long as it doesn't overflow
7043 * the s32 'off' field
7044 */
Edward Creeb03c9f92017-08-07 15:26:36 +01007045 if (known && (ptr_reg->off + smin_val ==
7046 (s64)(s32)(ptr_reg->off + smin_val))) {
Edward Creef1174f72017-08-07 15:26:19 +01007047 /* pointer += K. Accumulate it into fixed offset */
Edward Creeb03c9f92017-08-07 15:26:36 +01007048 dst_reg->smin_value = smin_ptr;
7049 dst_reg->smax_value = smax_ptr;
7050 dst_reg->umin_value = umin_ptr;
7051 dst_reg->umax_value = umax_ptr;
Edward Creef1174f72017-08-07 15:26:19 +01007052 dst_reg->var_off = ptr_reg->var_off;
Edward Creeb03c9f92017-08-07 15:26:36 +01007053 dst_reg->off = ptr_reg->off + smin_val;
Daniel Borkmann09625902018-11-01 00:05:52 +01007054 dst_reg->raw = ptr_reg->raw;
Edward Creef1174f72017-08-07 15:26:19 +01007055 break;
7056 }
Edward Creef1174f72017-08-07 15:26:19 +01007057 /* A new variable offset is created. Note that off_reg->off
7058 * == 0, since it's a scalar.
7059 * dst_reg gets the pointer type and since some positive
7060 * integer value was added to the pointer, give it a new 'id'
7061 * if it's a PTR_TO_PACKET.
7062 * this creates a new 'base' pointer, off_reg (variable) gets
7063 * added into the variable offset, and we copy the fixed offset
7064 * from ptr_reg.
7065 */
Edward Creeb03c9f92017-08-07 15:26:36 +01007066 if (signed_add_overflows(smin_ptr, smin_val) ||
7067 signed_add_overflows(smax_ptr, smax_val)) {
7068 dst_reg->smin_value = S64_MIN;
7069 dst_reg->smax_value = S64_MAX;
7070 } else {
7071 dst_reg->smin_value = smin_ptr + smin_val;
7072 dst_reg->smax_value = smax_ptr + smax_val;
7073 }
7074 if (umin_ptr + umin_val < umin_ptr ||
7075 umax_ptr + umax_val < umax_ptr) {
7076 dst_reg->umin_value = 0;
7077 dst_reg->umax_value = U64_MAX;
7078 } else {
7079 dst_reg->umin_value = umin_ptr + umin_val;
7080 dst_reg->umax_value = umax_ptr + umax_val;
7081 }
Edward Creef1174f72017-08-07 15:26:19 +01007082 dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off);
7083 dst_reg->off = ptr_reg->off;
Daniel Borkmann09625902018-11-01 00:05:52 +01007084 dst_reg->raw = ptr_reg->raw;
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02007085 if (reg_is_pkt_pointer(ptr_reg)) {
Edward Creef1174f72017-08-07 15:26:19 +01007086 dst_reg->id = ++env->id_gen;
7087 /* something was added to pkt_ptr, set range to zero */
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08007088 memset(&dst_reg->raw, 0, sizeof(dst_reg->raw));
Edward Creef1174f72017-08-07 15:26:19 +01007089 }
Josef Bacik48461132016-09-28 10:54:32 -04007090 break;
7091 case BPF_SUB:
Edward Creef1174f72017-08-07 15:26:19 +01007092 if (dst_reg == off_reg) {
7093 /* scalar -= pointer. Creates an unknown scalar */
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007094 verbose(env, "R%d tried to subtract pointer from scalar\n",
7095 dst);
Edward Creef1174f72017-08-07 15:26:19 +01007096 return -EACCES;
7097 }
7098 /* We don't allow subtraction from FP, because (according to
7099 * test_verifier.c test "invalid fp arithmetic", JITs might not
7100 * be able to deal with it.
Edward Cree93057062017-07-21 14:37:34 +01007101 */
Edward Creef1174f72017-08-07 15:26:19 +01007102 if (ptr_reg->type == PTR_TO_STACK) {
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007103 verbose(env, "R%d subtraction from stack pointer prohibited\n",
7104 dst);
Edward Creef1174f72017-08-07 15:26:19 +01007105 return -EACCES;
7106 }
Edward Creeb03c9f92017-08-07 15:26:36 +01007107 if (known && (ptr_reg->off - smin_val ==
7108 (s64)(s32)(ptr_reg->off - smin_val))) {
Edward Creef1174f72017-08-07 15:26:19 +01007109 /* pointer -= K. Subtract it from fixed offset */
Edward Creeb03c9f92017-08-07 15:26:36 +01007110 dst_reg->smin_value = smin_ptr;
7111 dst_reg->smax_value = smax_ptr;
7112 dst_reg->umin_value = umin_ptr;
7113 dst_reg->umax_value = umax_ptr;
Edward Creef1174f72017-08-07 15:26:19 +01007114 dst_reg->var_off = ptr_reg->var_off;
7115 dst_reg->id = ptr_reg->id;
Edward Creeb03c9f92017-08-07 15:26:36 +01007116 dst_reg->off = ptr_reg->off - smin_val;
Daniel Borkmann09625902018-11-01 00:05:52 +01007117 dst_reg->raw = ptr_reg->raw;
Edward Creef1174f72017-08-07 15:26:19 +01007118 break;
7119 }
Edward Creef1174f72017-08-07 15:26:19 +01007120 /* A new variable offset is created. If the subtrahend is known
7121 * nonnegative, then any reg->range we had before is still good.
7122 */
Edward Creeb03c9f92017-08-07 15:26:36 +01007123 if (signed_sub_overflows(smin_ptr, smax_val) ||
7124 signed_sub_overflows(smax_ptr, smin_val)) {
7125 /* Overflow possible, we know nothing */
7126 dst_reg->smin_value = S64_MIN;
7127 dst_reg->smax_value = S64_MAX;
7128 } else {
7129 dst_reg->smin_value = smin_ptr - smax_val;
7130 dst_reg->smax_value = smax_ptr - smin_val;
7131 }
7132 if (umin_ptr < umax_val) {
7133 /* Overflow possible, we know nothing */
7134 dst_reg->umin_value = 0;
7135 dst_reg->umax_value = U64_MAX;
7136 } else {
7137 /* Cannot overflow (as long as bounds are consistent) */
7138 dst_reg->umin_value = umin_ptr - umax_val;
7139 dst_reg->umax_value = umax_ptr - umin_val;
7140 }
Edward Creef1174f72017-08-07 15:26:19 +01007141 dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off);
7142 dst_reg->off = ptr_reg->off;
Daniel Borkmann09625902018-11-01 00:05:52 +01007143 dst_reg->raw = ptr_reg->raw;
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02007144 if (reg_is_pkt_pointer(ptr_reg)) {
Edward Creef1174f72017-08-07 15:26:19 +01007145 dst_reg->id = ++env->id_gen;
7146 /* something was added to pkt_ptr, set range to zero */
Edward Creeb03c9f92017-08-07 15:26:36 +01007147 if (smin_val < 0)
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08007148 memset(&dst_reg->raw, 0, sizeof(dst_reg->raw));
Edward Creef1174f72017-08-07 15:26:19 +01007149 }
7150 break;
7151 case BPF_AND:
7152 case BPF_OR:
7153 case BPF_XOR:
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007154 /* bitwise ops on pointers are troublesome, prohibit. */
7155 verbose(env, "R%d bitwise operator %s on pointer prohibited\n",
7156 dst, bpf_alu_string[opcode >> 4]);
Edward Creef1174f72017-08-07 15:26:19 +01007157 return -EACCES;
7158 default:
7159 /* other operators (e.g. MUL,LSH) produce non-pointer results */
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007160 verbose(env, "R%d pointer arithmetic with %s operator prohibited\n",
7161 dst, bpf_alu_string[opcode >> 4]);
Edward Creef1174f72017-08-07 15:26:19 +01007162 return -EACCES;
7163 }
7164
Alexei Starovoitovbb7f0f92017-12-18 20:12:00 -08007165 if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type))
7166 return -EINVAL;
7167
Edward Creeb03c9f92017-08-07 15:26:36 +01007168 __update_reg_bounds(dst_reg);
7169 __reg_deduce_bounds(dst_reg);
7170 __reg_bound_offset(dst_reg);
Daniel Borkmann0d6303d2019-01-03 00:58:30 +01007171
Daniel Borkmann073815b2021-03-23 15:05:48 +01007172 if (sanitize_check_bounds(env, insn, dst_reg) < 0)
7173 return -EACCES;
Daniel Borkmann7fedb632021-03-24 10:38:26 +01007174 if (sanitize_needed(opcode)) {
7175 ret = sanitize_ptr_alu(env, insn, dst_reg, off_reg, dst_reg,
Daniel Borkmann3d0220f2021-05-21 10:17:36 +00007176 &info, true);
Daniel Borkmann7fedb632021-03-24 10:38:26 +01007177 if (ret < 0)
7178 return sanitize_err(env, insn, ret, off_reg, dst_reg);
Daniel Borkmann0d6303d2019-01-03 00:58:30 +01007179 }
7180
Edward Creef1174f72017-08-07 15:26:19 +01007181 return 0;
7182}
7183
John Fastabend3f50f132020-03-30 14:36:39 -07007184static void scalar32_min_max_add(struct bpf_reg_state *dst_reg,
7185 struct bpf_reg_state *src_reg)
7186{
7187 s32 smin_val = src_reg->s32_min_value;
7188 s32 smax_val = src_reg->s32_max_value;
7189 u32 umin_val = src_reg->u32_min_value;
7190 u32 umax_val = src_reg->u32_max_value;
7191
7192 if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) ||
7193 signed_add32_overflows(dst_reg->s32_max_value, smax_val)) {
7194 dst_reg->s32_min_value = S32_MIN;
7195 dst_reg->s32_max_value = S32_MAX;
7196 } else {
7197 dst_reg->s32_min_value += smin_val;
7198 dst_reg->s32_max_value += smax_val;
7199 }
7200 if (dst_reg->u32_min_value + umin_val < umin_val ||
7201 dst_reg->u32_max_value + umax_val < umax_val) {
7202 dst_reg->u32_min_value = 0;
7203 dst_reg->u32_max_value = U32_MAX;
7204 } else {
7205 dst_reg->u32_min_value += umin_val;
7206 dst_reg->u32_max_value += umax_val;
7207 }
7208}
7209
John Fastabend07cd2632020-03-24 10:38:15 -07007210static void scalar_min_max_add(struct bpf_reg_state *dst_reg,
7211 struct bpf_reg_state *src_reg)
7212{
7213 s64 smin_val = src_reg->smin_value;
7214 s64 smax_val = src_reg->smax_value;
7215 u64 umin_val = src_reg->umin_value;
7216 u64 umax_val = src_reg->umax_value;
7217
7218 if (signed_add_overflows(dst_reg->smin_value, smin_val) ||
7219 signed_add_overflows(dst_reg->smax_value, smax_val)) {
7220 dst_reg->smin_value = S64_MIN;
7221 dst_reg->smax_value = S64_MAX;
7222 } else {
7223 dst_reg->smin_value += smin_val;
7224 dst_reg->smax_value += smax_val;
7225 }
7226 if (dst_reg->umin_value + umin_val < umin_val ||
7227 dst_reg->umax_value + umax_val < umax_val) {
7228 dst_reg->umin_value = 0;
7229 dst_reg->umax_value = U64_MAX;
7230 } else {
7231 dst_reg->umin_value += umin_val;
7232 dst_reg->umax_value += umax_val;
7233 }
John Fastabend3f50f132020-03-30 14:36:39 -07007234}
7235
7236static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg,
7237 struct bpf_reg_state *src_reg)
7238{
7239 s32 smin_val = src_reg->s32_min_value;
7240 s32 smax_val = src_reg->s32_max_value;
7241 u32 umin_val = src_reg->u32_min_value;
7242 u32 umax_val = src_reg->u32_max_value;
7243
7244 if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) ||
7245 signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) {
7246 /* Overflow possible, we know nothing */
7247 dst_reg->s32_min_value = S32_MIN;
7248 dst_reg->s32_max_value = S32_MAX;
7249 } else {
7250 dst_reg->s32_min_value -= smax_val;
7251 dst_reg->s32_max_value -= smin_val;
7252 }
7253 if (dst_reg->u32_min_value < umax_val) {
7254 /* Overflow possible, we know nothing */
7255 dst_reg->u32_min_value = 0;
7256 dst_reg->u32_max_value = U32_MAX;
7257 } else {
7258 /* Cannot overflow (as long as bounds are consistent) */
7259 dst_reg->u32_min_value -= umax_val;
7260 dst_reg->u32_max_value -= umin_val;
7261 }
John Fastabend07cd2632020-03-24 10:38:15 -07007262}
7263
7264static void scalar_min_max_sub(struct bpf_reg_state *dst_reg,
7265 struct bpf_reg_state *src_reg)
7266{
7267 s64 smin_val = src_reg->smin_value;
7268 s64 smax_val = src_reg->smax_value;
7269 u64 umin_val = src_reg->umin_value;
7270 u64 umax_val = src_reg->umax_value;
7271
7272 if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
7273 signed_sub_overflows(dst_reg->smax_value, smin_val)) {
7274 /* Overflow possible, we know nothing */
7275 dst_reg->smin_value = S64_MIN;
7276 dst_reg->smax_value = S64_MAX;
7277 } else {
7278 dst_reg->smin_value -= smax_val;
7279 dst_reg->smax_value -= smin_val;
7280 }
7281 if (dst_reg->umin_value < umax_val) {
7282 /* Overflow possible, we know nothing */
7283 dst_reg->umin_value = 0;
7284 dst_reg->umax_value = U64_MAX;
7285 } else {
7286 /* Cannot overflow (as long as bounds are consistent) */
7287 dst_reg->umin_value -= umax_val;
7288 dst_reg->umax_value -= umin_val;
7289 }
John Fastabend3f50f132020-03-30 14:36:39 -07007290}
7291
7292static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg,
7293 struct bpf_reg_state *src_reg)
7294{
7295 s32 smin_val = src_reg->s32_min_value;
7296 u32 umin_val = src_reg->u32_min_value;
7297 u32 umax_val = src_reg->u32_max_value;
7298
7299 if (smin_val < 0 || dst_reg->s32_min_value < 0) {
7300 /* Ain't nobody got time to multiply that sign */
7301 __mark_reg32_unbounded(dst_reg);
7302 return;
7303 }
7304 /* Both values are positive, so we can work with unsigned and
7305 * copy the result to signed (unless it exceeds S32_MAX).
7306 */
7307 if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) {
7308 /* Potential overflow, we know nothing */
7309 __mark_reg32_unbounded(dst_reg);
7310 return;
7311 }
7312 dst_reg->u32_min_value *= umin_val;
7313 dst_reg->u32_max_value *= umax_val;
7314 if (dst_reg->u32_max_value > S32_MAX) {
7315 /* Overflow possible, we know nothing */
7316 dst_reg->s32_min_value = S32_MIN;
7317 dst_reg->s32_max_value = S32_MAX;
7318 } else {
7319 dst_reg->s32_min_value = dst_reg->u32_min_value;
7320 dst_reg->s32_max_value = dst_reg->u32_max_value;
7321 }
John Fastabend07cd2632020-03-24 10:38:15 -07007322}
7323
7324static void scalar_min_max_mul(struct bpf_reg_state *dst_reg,
7325 struct bpf_reg_state *src_reg)
7326{
7327 s64 smin_val = src_reg->smin_value;
7328 u64 umin_val = src_reg->umin_value;
7329 u64 umax_val = src_reg->umax_value;
7330
John Fastabend07cd2632020-03-24 10:38:15 -07007331 if (smin_val < 0 || dst_reg->smin_value < 0) {
7332 /* Ain't nobody got time to multiply that sign */
John Fastabend3f50f132020-03-30 14:36:39 -07007333 __mark_reg64_unbounded(dst_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007334 return;
7335 }
7336 /* Both values are positive, so we can work with unsigned and
7337 * copy the result to signed (unless it exceeds S64_MAX).
7338 */
7339 if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) {
7340 /* Potential overflow, we know nothing */
John Fastabend3f50f132020-03-30 14:36:39 -07007341 __mark_reg64_unbounded(dst_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007342 return;
7343 }
7344 dst_reg->umin_value *= umin_val;
7345 dst_reg->umax_value *= umax_val;
7346 if (dst_reg->umax_value > S64_MAX) {
7347 /* Overflow possible, we know nothing */
7348 dst_reg->smin_value = S64_MIN;
7349 dst_reg->smax_value = S64_MAX;
7350 } else {
7351 dst_reg->smin_value = dst_reg->umin_value;
7352 dst_reg->smax_value = dst_reg->umax_value;
7353 }
7354}
7355
John Fastabend3f50f132020-03-30 14:36:39 -07007356static void scalar32_min_max_and(struct bpf_reg_state *dst_reg,
7357 struct bpf_reg_state *src_reg)
John Fastabend07cd2632020-03-24 10:38:15 -07007358{
John Fastabend3f50f132020-03-30 14:36:39 -07007359 bool src_known = tnum_subreg_is_const(src_reg->var_off);
7360 bool dst_known = tnum_subreg_is_const(dst_reg->var_off);
7361 struct tnum var32_off = tnum_subreg(dst_reg->var_off);
7362 s32 smin_val = src_reg->s32_min_value;
7363 u32 umax_val = src_reg->u32_max_value;
7364
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007365 if (src_known && dst_known) {
7366 __mark_reg32_known(dst_reg, var32_off.value);
John Fastabend3f50f132020-03-30 14:36:39 -07007367 return;
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007368 }
John Fastabend07cd2632020-03-24 10:38:15 -07007369
7370 /* We get our minimum from the var_off, since that's inherently
7371 * bitwise. Our maximum is the minimum of the operands' maxima.
7372 */
John Fastabend3f50f132020-03-30 14:36:39 -07007373 dst_reg->u32_min_value = var32_off.value;
7374 dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val);
7375 if (dst_reg->s32_min_value < 0 || smin_val < 0) {
7376 /* Lose signed bounds when ANDing negative numbers,
7377 * ain't nobody got time for that.
7378 */
7379 dst_reg->s32_min_value = S32_MIN;
7380 dst_reg->s32_max_value = S32_MAX;
7381 } else {
7382 /* ANDing two positives gives a positive, so safe to
7383 * cast result into s64.
7384 */
7385 dst_reg->s32_min_value = dst_reg->u32_min_value;
7386 dst_reg->s32_max_value = dst_reg->u32_max_value;
7387 }
John Fastabend3f50f132020-03-30 14:36:39 -07007388}
7389
7390static void scalar_min_max_and(struct bpf_reg_state *dst_reg,
7391 struct bpf_reg_state *src_reg)
7392{
7393 bool src_known = tnum_is_const(src_reg->var_off);
7394 bool dst_known = tnum_is_const(dst_reg->var_off);
7395 s64 smin_val = src_reg->smin_value;
7396 u64 umax_val = src_reg->umax_value;
7397
7398 if (src_known && dst_known) {
John Fastabend4fbb38a2020-09-24 11:45:06 -07007399 __mark_reg_known(dst_reg, dst_reg->var_off.value);
John Fastabend3f50f132020-03-30 14:36:39 -07007400 return;
7401 }
7402
7403 /* We get our minimum from the var_off, since that's inherently
7404 * bitwise. Our maximum is the minimum of the operands' maxima.
7405 */
John Fastabend07cd2632020-03-24 10:38:15 -07007406 dst_reg->umin_value = dst_reg->var_off.value;
7407 dst_reg->umax_value = min(dst_reg->umax_value, umax_val);
7408 if (dst_reg->smin_value < 0 || smin_val < 0) {
7409 /* Lose signed bounds when ANDing negative numbers,
7410 * ain't nobody got time for that.
7411 */
7412 dst_reg->smin_value = S64_MIN;
7413 dst_reg->smax_value = S64_MAX;
7414 } else {
7415 /* ANDing two positives gives a positive, so safe to
7416 * cast result into s64.
7417 */
7418 dst_reg->smin_value = dst_reg->umin_value;
7419 dst_reg->smax_value = dst_reg->umax_value;
7420 }
7421 /* We may learn something more from the var_off */
7422 __update_reg_bounds(dst_reg);
7423}
7424
John Fastabend3f50f132020-03-30 14:36:39 -07007425static void scalar32_min_max_or(struct bpf_reg_state *dst_reg,
7426 struct bpf_reg_state *src_reg)
John Fastabend07cd2632020-03-24 10:38:15 -07007427{
John Fastabend3f50f132020-03-30 14:36:39 -07007428 bool src_known = tnum_subreg_is_const(src_reg->var_off);
7429 bool dst_known = tnum_subreg_is_const(dst_reg->var_off);
7430 struct tnum var32_off = tnum_subreg(dst_reg->var_off);
Daniel Borkmann5b9fbeb2020-10-07 15:48:58 +02007431 s32 smin_val = src_reg->s32_min_value;
7432 u32 umin_val = src_reg->u32_min_value;
John Fastabend3f50f132020-03-30 14:36:39 -07007433
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007434 if (src_known && dst_known) {
7435 __mark_reg32_known(dst_reg, var32_off.value);
John Fastabend3f50f132020-03-30 14:36:39 -07007436 return;
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007437 }
John Fastabend07cd2632020-03-24 10:38:15 -07007438
7439 /* We get our maximum from the var_off, and our minimum is the
7440 * maximum of the operands' minima
7441 */
John Fastabend3f50f132020-03-30 14:36:39 -07007442 dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val);
7443 dst_reg->u32_max_value = var32_off.value | var32_off.mask;
7444 if (dst_reg->s32_min_value < 0 || smin_val < 0) {
7445 /* Lose signed bounds when ORing negative numbers,
7446 * ain't nobody got time for that.
7447 */
7448 dst_reg->s32_min_value = S32_MIN;
7449 dst_reg->s32_max_value = S32_MAX;
7450 } else {
7451 /* ORing two positives gives a positive, so safe to
7452 * cast result into s64.
7453 */
Daniel Borkmann5b9fbeb2020-10-07 15:48:58 +02007454 dst_reg->s32_min_value = dst_reg->u32_min_value;
7455 dst_reg->s32_max_value = dst_reg->u32_max_value;
John Fastabend3f50f132020-03-30 14:36:39 -07007456 }
7457}
7458
7459static void scalar_min_max_or(struct bpf_reg_state *dst_reg,
7460 struct bpf_reg_state *src_reg)
7461{
7462 bool src_known = tnum_is_const(src_reg->var_off);
7463 bool dst_known = tnum_is_const(dst_reg->var_off);
7464 s64 smin_val = src_reg->smin_value;
7465 u64 umin_val = src_reg->umin_value;
7466
7467 if (src_known && dst_known) {
John Fastabend4fbb38a2020-09-24 11:45:06 -07007468 __mark_reg_known(dst_reg, dst_reg->var_off.value);
John Fastabend3f50f132020-03-30 14:36:39 -07007469 return;
7470 }
7471
7472 /* We get our maximum from the var_off, and our minimum is the
7473 * maximum of the operands' minima
7474 */
John Fastabend07cd2632020-03-24 10:38:15 -07007475 dst_reg->umin_value = max(dst_reg->umin_value, umin_val);
7476 dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask;
7477 if (dst_reg->smin_value < 0 || smin_val < 0) {
7478 /* Lose signed bounds when ORing negative numbers,
7479 * ain't nobody got time for that.
7480 */
7481 dst_reg->smin_value = S64_MIN;
7482 dst_reg->smax_value = S64_MAX;
7483 } else {
7484 /* ORing two positives gives a positive, so safe to
7485 * cast result into s64.
7486 */
7487 dst_reg->smin_value = dst_reg->umin_value;
7488 dst_reg->smax_value = dst_reg->umax_value;
7489 }
7490 /* We may learn something more from the var_off */
7491 __update_reg_bounds(dst_reg);
7492}
7493
Yonghong Song2921c902020-08-24 23:46:08 -07007494static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg,
7495 struct bpf_reg_state *src_reg)
7496{
7497 bool src_known = tnum_subreg_is_const(src_reg->var_off);
7498 bool dst_known = tnum_subreg_is_const(dst_reg->var_off);
7499 struct tnum var32_off = tnum_subreg(dst_reg->var_off);
7500 s32 smin_val = src_reg->s32_min_value;
7501
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007502 if (src_known && dst_known) {
7503 __mark_reg32_known(dst_reg, var32_off.value);
Yonghong Song2921c902020-08-24 23:46:08 -07007504 return;
Daniel Borkmann049c4e12021-05-10 13:10:44 +00007505 }
Yonghong Song2921c902020-08-24 23:46:08 -07007506
7507 /* We get both minimum and maximum from the var32_off. */
7508 dst_reg->u32_min_value = var32_off.value;
7509 dst_reg->u32_max_value = var32_off.value | var32_off.mask;
7510
7511 if (dst_reg->s32_min_value >= 0 && smin_val >= 0) {
7512 /* XORing two positive sign numbers gives a positive,
7513 * so safe to cast u32 result into s32.
7514 */
7515 dst_reg->s32_min_value = dst_reg->u32_min_value;
7516 dst_reg->s32_max_value = dst_reg->u32_max_value;
7517 } else {
7518 dst_reg->s32_min_value = S32_MIN;
7519 dst_reg->s32_max_value = S32_MAX;
7520 }
7521}
7522
7523static void scalar_min_max_xor(struct bpf_reg_state *dst_reg,
7524 struct bpf_reg_state *src_reg)
7525{
7526 bool src_known = tnum_is_const(src_reg->var_off);
7527 bool dst_known = tnum_is_const(dst_reg->var_off);
7528 s64 smin_val = src_reg->smin_value;
7529
7530 if (src_known && dst_known) {
7531 /* dst_reg->var_off.value has been updated earlier */
7532 __mark_reg_known(dst_reg, dst_reg->var_off.value);
7533 return;
7534 }
7535
7536 /* We get both minimum and maximum from the var_off. */
7537 dst_reg->umin_value = dst_reg->var_off.value;
7538 dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask;
7539
7540 if (dst_reg->smin_value >= 0 && smin_val >= 0) {
7541 /* XORing two positive sign numbers gives a positive,
7542 * so safe to cast u64 result into s64.
7543 */
7544 dst_reg->smin_value = dst_reg->umin_value;
7545 dst_reg->smax_value = dst_reg->umax_value;
7546 } else {
7547 dst_reg->smin_value = S64_MIN;
7548 dst_reg->smax_value = S64_MAX;
7549 }
7550
7551 __update_reg_bounds(dst_reg);
7552}
7553
John Fastabend3f50f132020-03-30 14:36:39 -07007554static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg,
7555 u64 umin_val, u64 umax_val)
John Fastabend07cd2632020-03-24 10:38:15 -07007556{
John Fastabend07cd2632020-03-24 10:38:15 -07007557 /* We lose all sign bit information (except what we can pick
7558 * up from var_off)
7559 */
John Fastabend3f50f132020-03-30 14:36:39 -07007560 dst_reg->s32_min_value = S32_MIN;
7561 dst_reg->s32_max_value = S32_MAX;
7562 /* If we might shift our top bit out, then we know nothing */
7563 if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) {
7564 dst_reg->u32_min_value = 0;
7565 dst_reg->u32_max_value = U32_MAX;
7566 } else {
7567 dst_reg->u32_min_value <<= umin_val;
7568 dst_reg->u32_max_value <<= umax_val;
7569 }
7570}
7571
7572static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg,
7573 struct bpf_reg_state *src_reg)
7574{
7575 u32 umax_val = src_reg->u32_max_value;
7576 u32 umin_val = src_reg->u32_min_value;
7577 /* u32 alu operation will zext upper bits */
7578 struct tnum subreg = tnum_subreg(dst_reg->var_off);
7579
7580 __scalar32_min_max_lsh(dst_reg, umin_val, umax_val);
7581 dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val));
7582 /* Not required but being careful mark reg64 bounds as unknown so
7583 * that we are forced to pick them up from tnum and zext later and
7584 * if some path skips this step we are still safe.
7585 */
7586 __mark_reg64_unbounded(dst_reg);
7587 __update_reg32_bounds(dst_reg);
7588}
7589
7590static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg,
7591 u64 umin_val, u64 umax_val)
7592{
7593 /* Special case <<32 because it is a common compiler pattern to sign
7594 * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are
7595 * positive we know this shift will also be positive so we can track
7596 * bounds correctly. Otherwise we lose all sign bit information except
7597 * what we can pick up from var_off. Perhaps we can generalize this
7598 * later to shifts of any length.
7599 */
7600 if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0)
7601 dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32;
7602 else
7603 dst_reg->smax_value = S64_MAX;
7604
7605 if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0)
7606 dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32;
7607 else
7608 dst_reg->smin_value = S64_MIN;
7609
John Fastabend07cd2632020-03-24 10:38:15 -07007610 /* If we might shift our top bit out, then we know nothing */
7611 if (dst_reg->umax_value > 1ULL << (63 - umax_val)) {
7612 dst_reg->umin_value = 0;
7613 dst_reg->umax_value = U64_MAX;
7614 } else {
7615 dst_reg->umin_value <<= umin_val;
7616 dst_reg->umax_value <<= umax_val;
7617 }
John Fastabend3f50f132020-03-30 14:36:39 -07007618}
7619
7620static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg,
7621 struct bpf_reg_state *src_reg)
7622{
7623 u64 umax_val = src_reg->umax_value;
7624 u64 umin_val = src_reg->umin_value;
7625
7626 /* scalar64 calc uses 32bit unshifted bounds so must be called first */
7627 __scalar64_min_max_lsh(dst_reg, umin_val, umax_val);
7628 __scalar32_min_max_lsh(dst_reg, umin_val, umax_val);
7629
John Fastabend07cd2632020-03-24 10:38:15 -07007630 dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val);
7631 /* We may learn something more from the var_off */
7632 __update_reg_bounds(dst_reg);
7633}
7634
John Fastabend3f50f132020-03-30 14:36:39 -07007635static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg,
7636 struct bpf_reg_state *src_reg)
7637{
7638 struct tnum subreg = tnum_subreg(dst_reg->var_off);
7639 u32 umax_val = src_reg->u32_max_value;
7640 u32 umin_val = src_reg->u32_min_value;
7641
7642 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7643 * be negative, then either:
7644 * 1) src_reg might be zero, so the sign bit of the result is
7645 * unknown, so we lose our signed bounds
7646 * 2) it's known negative, thus the unsigned bounds capture the
7647 * signed bounds
7648 * 3) the signed bounds cross zero, so they tell us nothing
7649 * about the result
7650 * If the value in dst_reg is known nonnegative, then again the
Tobias Klauser18b24d72021-01-21 18:43:24 +01007651 * unsigned bounds capture the signed bounds.
John Fastabend3f50f132020-03-30 14:36:39 -07007652 * Thus, in all cases it suffices to blow away our signed bounds
7653 * and rely on inferring new ones from the unsigned bounds and
7654 * var_off of the result.
7655 */
7656 dst_reg->s32_min_value = S32_MIN;
7657 dst_reg->s32_max_value = S32_MAX;
7658
7659 dst_reg->var_off = tnum_rshift(subreg, umin_val);
7660 dst_reg->u32_min_value >>= umax_val;
7661 dst_reg->u32_max_value >>= umin_val;
7662
7663 __mark_reg64_unbounded(dst_reg);
7664 __update_reg32_bounds(dst_reg);
7665}
7666
John Fastabend07cd2632020-03-24 10:38:15 -07007667static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg,
7668 struct bpf_reg_state *src_reg)
7669{
7670 u64 umax_val = src_reg->umax_value;
7671 u64 umin_val = src_reg->umin_value;
7672
7673 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7674 * be negative, then either:
7675 * 1) src_reg might be zero, so the sign bit of the result is
7676 * unknown, so we lose our signed bounds
7677 * 2) it's known negative, thus the unsigned bounds capture the
7678 * signed bounds
7679 * 3) the signed bounds cross zero, so they tell us nothing
7680 * about the result
7681 * If the value in dst_reg is known nonnegative, then again the
Tobias Klauser18b24d72021-01-21 18:43:24 +01007682 * unsigned bounds capture the signed bounds.
John Fastabend07cd2632020-03-24 10:38:15 -07007683 * Thus, in all cases it suffices to blow away our signed bounds
7684 * and rely on inferring new ones from the unsigned bounds and
7685 * var_off of the result.
7686 */
7687 dst_reg->smin_value = S64_MIN;
7688 dst_reg->smax_value = S64_MAX;
7689 dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val);
7690 dst_reg->umin_value >>= umax_val;
7691 dst_reg->umax_value >>= umin_val;
John Fastabend3f50f132020-03-30 14:36:39 -07007692
7693 /* Its not easy to operate on alu32 bounds here because it depends
7694 * on bits being shifted in. Take easy way out and mark unbounded
7695 * so we can recalculate later from tnum.
7696 */
7697 __mark_reg32_unbounded(dst_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007698 __update_reg_bounds(dst_reg);
7699}
7700
John Fastabend3f50f132020-03-30 14:36:39 -07007701static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg,
7702 struct bpf_reg_state *src_reg)
John Fastabend07cd2632020-03-24 10:38:15 -07007703{
John Fastabend3f50f132020-03-30 14:36:39 -07007704 u64 umin_val = src_reg->u32_min_value;
John Fastabend07cd2632020-03-24 10:38:15 -07007705
7706 /* Upon reaching here, src_known is true and
7707 * umax_val is equal to umin_val.
7708 */
John Fastabend3f50f132020-03-30 14:36:39 -07007709 dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val);
7710 dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val);
John Fastabend07cd2632020-03-24 10:38:15 -07007711
John Fastabend3f50f132020-03-30 14:36:39 -07007712 dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32);
7713
7714 /* blow away the dst_reg umin_value/umax_value and rely on
7715 * dst_reg var_off to refine the result.
7716 */
7717 dst_reg->u32_min_value = 0;
7718 dst_reg->u32_max_value = U32_MAX;
7719
7720 __mark_reg64_unbounded(dst_reg);
7721 __update_reg32_bounds(dst_reg);
7722}
7723
7724static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg,
7725 struct bpf_reg_state *src_reg)
7726{
7727 u64 umin_val = src_reg->umin_value;
7728
7729 /* Upon reaching here, src_known is true and umax_val is equal
7730 * to umin_val.
7731 */
7732 dst_reg->smin_value >>= umin_val;
7733 dst_reg->smax_value >>= umin_val;
7734
7735 dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64);
John Fastabend07cd2632020-03-24 10:38:15 -07007736
7737 /* blow away the dst_reg umin_value/umax_value and rely on
7738 * dst_reg var_off to refine the result.
7739 */
7740 dst_reg->umin_value = 0;
7741 dst_reg->umax_value = U64_MAX;
John Fastabend3f50f132020-03-30 14:36:39 -07007742
7743 /* Its not easy to operate on alu32 bounds here because it depends
7744 * on bits being shifted in from upper 32-bits. Take easy way out
7745 * and mark unbounded so we can recalculate later from tnum.
7746 */
7747 __mark_reg32_unbounded(dst_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007748 __update_reg_bounds(dst_reg);
7749}
7750
Jann Horn468f6ea2017-12-18 20:11:56 -08007751/* WARNING: This function does calculations on 64-bit values, but the actual
7752 * execution may occur on 32-bit values. Therefore, things like bitshifts
7753 * need extra checks in the 32-bit case.
7754 */
Edward Creef1174f72017-08-07 15:26:19 +01007755static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
7756 struct bpf_insn *insn,
7757 struct bpf_reg_state *dst_reg,
7758 struct bpf_reg_state src_reg)
7759{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07007760 struct bpf_reg_state *regs = cur_regs(env);
Edward Creef1174f72017-08-07 15:26:19 +01007761 u8 opcode = BPF_OP(insn->code);
Mao Wenanb0b3fb62020-04-18 09:37:35 +08007762 bool src_known;
Edward Creeb03c9f92017-08-07 15:26:36 +01007763 s64 smin_val, smax_val;
7764 u64 umin_val, umax_val;
John Fastabend3f50f132020-03-30 14:36:39 -07007765 s32 s32_min_val, s32_max_val;
7766 u32 u32_min_val, u32_max_val;
Jann Horn468f6ea2017-12-18 20:11:56 -08007767 u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32;
John Fastabend3f50f132020-03-30 14:36:39 -07007768 bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64);
Daniel Borkmanna6aaece2021-03-23 09:30:01 +01007769 int ret;
Jann Hornb7992072018-10-05 18:17:59 +02007770
Edward Creeb03c9f92017-08-07 15:26:36 +01007771 smin_val = src_reg.smin_value;
7772 smax_val = src_reg.smax_value;
7773 umin_val = src_reg.umin_value;
7774 umax_val = src_reg.umax_value;
Edward Creef1174f72017-08-07 15:26:19 +01007775
John Fastabend3f50f132020-03-30 14:36:39 -07007776 s32_min_val = src_reg.s32_min_value;
7777 s32_max_val = src_reg.s32_max_value;
7778 u32_min_val = src_reg.u32_min_value;
7779 u32_max_val = src_reg.u32_max_value;
7780
7781 if (alu32) {
7782 src_known = tnum_subreg_is_const(src_reg.var_off);
John Fastabend3f50f132020-03-30 14:36:39 -07007783 if ((src_known &&
7784 (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) ||
7785 s32_min_val > s32_max_val || u32_min_val > u32_max_val) {
7786 /* Taint dst register if offset had invalid bounds
7787 * derived from e.g. dead branches.
7788 */
7789 __mark_reg_unknown(env, dst_reg);
7790 return 0;
7791 }
7792 } else {
7793 src_known = tnum_is_const(src_reg.var_off);
John Fastabend3f50f132020-03-30 14:36:39 -07007794 if ((src_known &&
7795 (smin_val != smax_val || umin_val != umax_val)) ||
7796 smin_val > smax_val || umin_val > umax_val) {
7797 /* Taint dst register if offset had invalid bounds
7798 * derived from e.g. dead branches.
7799 */
7800 __mark_reg_unknown(env, dst_reg);
7801 return 0;
7802 }
Daniel Borkmann6f161012018-01-18 01:15:21 +01007803 }
7804
Alexei Starovoitovbb7f0f92017-12-18 20:12:00 -08007805 if (!src_known &&
7806 opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) {
Daniel Borkmannf54c7892019-12-22 23:37:40 +01007807 __mark_reg_unknown(env, dst_reg);
Alexei Starovoitovbb7f0f92017-12-18 20:12:00 -08007808 return 0;
7809 }
7810
Daniel Borkmannf5288192021-03-24 11:25:39 +01007811 if (sanitize_needed(opcode)) {
7812 ret = sanitize_val_alu(env, insn);
7813 if (ret < 0)
7814 return sanitize_err(env, insn, ret, NULL, NULL);
7815 }
7816
John Fastabend3f50f132020-03-30 14:36:39 -07007817 /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops.
7818 * There are two classes of instructions: The first class we track both
7819 * alu32 and alu64 sign/unsigned bounds independently this provides the
7820 * greatest amount of precision when alu operations are mixed with jmp32
7821 * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD,
7822 * and BPF_OR. This is possible because these ops have fairly easy to
7823 * understand and calculate behavior in both 32-bit and 64-bit alu ops.
7824 * See alu32 verifier tests for examples. The second class of
7825 * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy
7826 * with regards to tracking sign/unsigned bounds because the bits may
7827 * cross subreg boundaries in the alu64 case. When this happens we mark
7828 * the reg unbounded in the subreg bound space and use the resulting
7829 * tnum to calculate an approximation of the sign/unsigned bounds.
7830 */
Edward Creef1174f72017-08-07 15:26:19 +01007831 switch (opcode) {
7832 case BPF_ADD:
John Fastabend3f50f132020-03-30 14:36:39 -07007833 scalar32_min_max_add(dst_reg, &src_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007834 scalar_min_max_add(dst_reg, &src_reg);
John Fastabend3f50f132020-03-30 14:36:39 -07007835 dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off);
Edward Creef1174f72017-08-07 15:26:19 +01007836 break;
7837 case BPF_SUB:
John Fastabend3f50f132020-03-30 14:36:39 -07007838 scalar32_min_max_sub(dst_reg, &src_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007839 scalar_min_max_sub(dst_reg, &src_reg);
John Fastabend3f50f132020-03-30 14:36:39 -07007840 dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off);
Josef Bacik48461132016-09-28 10:54:32 -04007841 break;
7842 case BPF_MUL:
John Fastabend3f50f132020-03-30 14:36:39 -07007843 dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off);
7844 scalar32_min_max_mul(dst_reg, &src_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007845 scalar_min_max_mul(dst_reg, &src_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007846 break;
7847 case BPF_AND:
John Fastabend3f50f132020-03-30 14:36:39 -07007848 dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off);
7849 scalar32_min_max_and(dst_reg, &src_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007850 scalar_min_max_and(dst_reg, &src_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007851 break;
7852 case BPF_OR:
John Fastabend3f50f132020-03-30 14:36:39 -07007853 dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off);
7854 scalar32_min_max_or(dst_reg, &src_reg);
John Fastabend07cd2632020-03-24 10:38:15 -07007855 scalar_min_max_or(dst_reg, &src_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007856 break;
Yonghong Song2921c902020-08-24 23:46:08 -07007857 case BPF_XOR:
7858 dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off);
7859 scalar32_min_max_xor(dst_reg, &src_reg);
7860 scalar_min_max_xor(dst_reg, &src_reg);
7861 break;
Josef Bacik48461132016-09-28 10:54:32 -04007862 case BPF_LSH:
Jann Horn468f6ea2017-12-18 20:11:56 -08007863 if (umax_val >= insn_bitness) {
7864 /* Shifts greater than 31 or 63 are undefined.
7865 * This includes shifts by a negative number.
Edward Creeb03c9f92017-08-07 15:26:36 +01007866 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07007867 mark_reg_unknown(env, regs, insn->dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007868 break;
7869 }
John Fastabend3f50f132020-03-30 14:36:39 -07007870 if (alu32)
7871 scalar32_min_max_lsh(dst_reg, &src_reg);
7872 else
7873 scalar_min_max_lsh(dst_reg, &src_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007874 break;
7875 case BPF_RSH:
Jann Horn468f6ea2017-12-18 20:11:56 -08007876 if (umax_val >= insn_bitness) {
7877 /* Shifts greater than 31 or 63 are undefined.
7878 * This includes shifts by a negative number.
Edward Creeb03c9f92017-08-07 15:26:36 +01007879 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07007880 mark_reg_unknown(env, regs, insn->dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007881 break;
7882 }
John Fastabend3f50f132020-03-30 14:36:39 -07007883 if (alu32)
7884 scalar32_min_max_rsh(dst_reg, &src_reg);
7885 else
7886 scalar_min_max_rsh(dst_reg, &src_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007887 break;
Yonghong Song9cbe1f5a2018-04-28 22:28:11 -07007888 case BPF_ARSH:
7889 if (umax_val >= insn_bitness) {
7890 /* Shifts greater than 31 or 63 are undefined.
7891 * This includes shifts by a negative number.
7892 */
7893 mark_reg_unknown(env, regs, insn->dst_reg);
7894 break;
7895 }
John Fastabend3f50f132020-03-30 14:36:39 -07007896 if (alu32)
7897 scalar32_min_max_arsh(dst_reg, &src_reg);
7898 else
7899 scalar_min_max_arsh(dst_reg, &src_reg);
Yonghong Song9cbe1f5a2018-04-28 22:28:11 -07007900 break;
Josef Bacik48461132016-09-28 10:54:32 -04007901 default:
Jakub Kicinski61bd5212017-10-09 10:30:11 -07007902 mark_reg_unknown(env, regs, insn->dst_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007903 break;
7904 }
7905
John Fastabend3f50f132020-03-30 14:36:39 -07007906 /* ALU32 ops are zero extended into 64bit register */
7907 if (alu32)
7908 zext_32_to_64(dst_reg);
Jann Horn468f6ea2017-12-18 20:11:56 -08007909
John Fastabend294f2fc2020-03-24 10:38:37 -07007910 __update_reg_bounds(dst_reg);
Edward Creeb03c9f92017-08-07 15:26:36 +01007911 __reg_deduce_bounds(dst_reg);
7912 __reg_bound_offset(dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007913 return 0;
7914}
7915
7916/* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
7917 * and var_off.
7918 */
7919static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
7920 struct bpf_insn *insn)
7921{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08007922 struct bpf_verifier_state *vstate = env->cur_state;
7923 struct bpf_func_state *state = vstate->frame[vstate->curframe];
7924 struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg;
Edward Creef1174f72017-08-07 15:26:19 +01007925 struct bpf_reg_state *ptr_reg = NULL, off_reg = {0};
7926 u8 opcode = BPF_OP(insn->code);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07007927 int err;
Edward Creef1174f72017-08-07 15:26:19 +01007928
7929 dst_reg = &regs[insn->dst_reg];
Edward Creef1174f72017-08-07 15:26:19 +01007930 src_reg = NULL;
7931 if (dst_reg->type != SCALAR_VALUE)
7932 ptr_reg = dst_reg;
Alexei Starovoitov75748832020-10-08 18:12:37 -07007933 else
7934 /* Make sure ID is cleared otherwise dst_reg min/max could be
7935 * incorrectly propagated into other registers by find_equal_scalars()
7936 */
7937 dst_reg->id = 0;
Edward Creef1174f72017-08-07 15:26:19 +01007938 if (BPF_SRC(insn->code) == BPF_X) {
7939 src_reg = &regs[insn->src_reg];
Edward Creef1174f72017-08-07 15:26:19 +01007940 if (src_reg->type != SCALAR_VALUE) {
7941 if (dst_reg->type != SCALAR_VALUE) {
7942 /* Combining two pointers by any ALU op yields
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007943 * an arbitrary scalar. Disallow all math except
7944 * pointer subtraction
Edward Creef1174f72017-08-07 15:26:19 +01007945 */
Alexei Starovoitovdd066822018-09-12 14:06:10 -07007946 if (opcode == BPF_SUB && env->allow_ptr_leaks) {
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007947 mark_reg_unknown(env, regs, insn->dst_reg);
7948 return 0;
Edward Creef1174f72017-08-07 15:26:19 +01007949 }
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007950 verbose(env, "R%d pointer %s pointer prohibited\n",
7951 insn->dst_reg,
7952 bpf_alu_string[opcode >> 4]);
7953 return -EACCES;
Edward Creef1174f72017-08-07 15:26:19 +01007954 } else {
7955 /* scalar += pointer
7956 * This is legal, but we have to reverse our
7957 * src/dest handling in computing the range
7958 */
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07007959 err = mark_chain_precision(env, insn->dst_reg);
7960 if (err)
7961 return err;
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007962 return adjust_ptr_min_max_vals(env, insn,
7963 src_reg, dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007964 }
7965 } else if (ptr_reg) {
7966 /* pointer += scalar */
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07007967 err = mark_chain_precision(env, insn->src_reg);
7968 if (err)
7969 return err;
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007970 return adjust_ptr_min_max_vals(env, insn,
7971 dst_reg, src_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007972 }
7973 } else {
7974 /* Pretend the src is a reg with a known value, since we only
7975 * need to be able to read from this state.
7976 */
7977 off_reg.type = SCALAR_VALUE;
Edward Creeb03c9f92017-08-07 15:26:36 +01007978 __mark_reg_known(&off_reg, insn->imm);
Edward Creef1174f72017-08-07 15:26:19 +01007979 src_reg = &off_reg;
Alexei Starovoitov82abbf82017-12-18 20:15:20 -08007980 if (ptr_reg) /* pointer += K */
7981 return adjust_ptr_min_max_vals(env, insn,
7982 ptr_reg, src_reg);
Edward Creef1174f72017-08-07 15:26:19 +01007983 }
7984
7985 /* Got here implies adding two SCALAR_VALUEs */
7986 if (WARN_ON_ONCE(ptr_reg)) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08007987 print_verifier_state(env, state);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07007988 verbose(env, "verifier internal error: unexpected ptr_reg\n");
Edward Creef1174f72017-08-07 15:26:19 +01007989 return -EINVAL;
7990 }
7991 if (WARN_ON(!src_reg)) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08007992 print_verifier_state(env, state);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07007993 verbose(env, "verifier internal error: no src_reg\n");
Edward Creef1174f72017-08-07 15:26:19 +01007994 return -EINVAL;
7995 }
7996 return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg);
Josef Bacik48461132016-09-28 10:54:32 -04007997}
7998
Alexei Starovoitov17a52672014-09-26 00:17:06 -07007999/* check validity of 32-bit and 64-bit arithmetic operations */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01008000static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008001{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07008002 struct bpf_reg_state *regs = cur_regs(env);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008003 u8 opcode = BPF_OP(insn->code);
8004 int err;
8005
8006 if (opcode == BPF_END || opcode == BPF_NEG) {
8007 if (opcode == BPF_NEG) {
8008 if (BPF_SRC(insn->code) != 0 ||
8009 insn->src_reg != BPF_REG_0 ||
8010 insn->off != 0 || insn->imm != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008011 verbose(env, "BPF_NEG uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008012 return -EINVAL;
8013 }
8014 } else {
8015 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
Edward Creee67b8a62017-09-15 14:37:38 +01008016 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) ||
8017 BPF_CLASS(insn->code) == BPF_ALU64) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008018 verbose(env, "BPF_END uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008019 return -EINVAL;
8020 }
8021 }
8022
8023 /* check src operand */
Edward Creedc503a82017-08-15 20:34:35 +01008024 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008025 if (err)
8026 return err;
8027
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008028 if (is_pointer_value(env, insn->dst_reg)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008029 verbose(env, "R%d pointer arithmetic prohibited\n",
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008030 insn->dst_reg);
8031 return -EACCES;
8032 }
8033
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008034 /* check dest operand */
Edward Creedc503a82017-08-15 20:34:35 +01008035 err = check_reg_arg(env, insn->dst_reg, DST_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008036 if (err)
8037 return err;
8038
8039 } else if (opcode == BPF_MOV) {
8040
8041 if (BPF_SRC(insn->code) == BPF_X) {
8042 if (insn->imm != 0 || insn->off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008043 verbose(env, "BPF_MOV uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008044 return -EINVAL;
8045 }
8046
8047 /* check src operand */
Edward Creedc503a82017-08-15 20:34:35 +01008048 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008049 if (err)
8050 return err;
8051 } else {
8052 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008053 verbose(env, "BPF_MOV uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008054 return -EINVAL;
8055 }
8056 }
8057
Arthur Fabrefbeb1602018-07-31 18:17:22 +01008058 /* check dest operand, mark as required later */
8059 err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008060 if (err)
8061 return err;
8062
8063 if (BPF_SRC(insn->code) == BPF_X) {
Jiong Wange434b8c2018-12-07 12:16:18 -05008064 struct bpf_reg_state *src_reg = regs + insn->src_reg;
8065 struct bpf_reg_state *dst_reg = regs + insn->dst_reg;
8066
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008067 if (BPF_CLASS(insn->code) == BPF_ALU64) {
8068 /* case: R1 = R2
8069 * copy register state to dest reg
8070 */
Alexei Starovoitov75748832020-10-08 18:12:37 -07008071 if (src_reg->type == SCALAR_VALUE && !src_reg->id)
8072 /* Assign src and dst registers the same ID
8073 * that will be used by find_equal_scalars()
8074 * to propagate min/max range.
8075 */
8076 src_reg->id = ++env->id_gen;
Jiong Wange434b8c2018-12-07 12:16:18 -05008077 *dst_reg = *src_reg;
8078 dst_reg->live |= REG_LIVE_WRITTEN;
Jiong Wang5327ed32019-05-24 23:25:12 +01008079 dst_reg->subreg_def = DEF_NOT_SUBREG;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008080 } else {
Edward Creef1174f72017-08-07 15:26:19 +01008081 /* R1 = (u32) R2 */
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008082 if (is_pointer_value(env, insn->src_reg)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008083 verbose(env,
8084 "R%d partial copy of pointer\n",
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008085 insn->src_reg);
8086 return -EACCES;
Jiong Wange434b8c2018-12-07 12:16:18 -05008087 } else if (src_reg->type == SCALAR_VALUE) {
8088 *dst_reg = *src_reg;
Alexei Starovoitov75748832020-10-08 18:12:37 -07008089 /* Make sure ID is cleared otherwise
8090 * dst_reg min/max could be incorrectly
8091 * propagated into src_reg by find_equal_scalars()
8092 */
8093 dst_reg->id = 0;
Jiong Wange434b8c2018-12-07 12:16:18 -05008094 dst_reg->live |= REG_LIVE_WRITTEN;
Jiong Wang5327ed32019-05-24 23:25:12 +01008095 dst_reg->subreg_def = env->insn_idx + 1;
Jiong Wange434b8c2018-12-07 12:16:18 -05008096 } else {
8097 mark_reg_unknown(env, regs,
8098 insn->dst_reg);
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008099 }
John Fastabend3f50f132020-03-30 14:36:39 -07008100 zext_32_to_64(dst_reg);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008101 }
8102 } else {
8103 /* case: R = imm
8104 * remember the value we stored into this reg
8105 */
Arthur Fabrefbeb1602018-07-31 18:17:22 +01008106 /* clear any state __mark_reg_known doesn't set */
8107 mark_reg_unknown(env, regs, insn->dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01008108 regs[insn->dst_reg].type = SCALAR_VALUE;
Jann Horn95a762e2017-12-18 20:11:54 -08008109 if (BPF_CLASS(insn->code) == BPF_ALU64) {
8110 __mark_reg_known(regs + insn->dst_reg,
8111 insn->imm);
8112 } else {
8113 __mark_reg_known(regs + insn->dst_reg,
8114 (u32)insn->imm);
8115 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008116 }
8117
8118 } else if (opcode > BPF_END) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008119 verbose(env, "invalid BPF_ALU opcode %x\n", opcode);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008120 return -EINVAL;
8121
8122 } else { /* all other ALU ops: and, sub, xor, add, ... */
8123
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008124 if (BPF_SRC(insn->code) == BPF_X) {
8125 if (insn->imm != 0 || insn->off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008126 verbose(env, "BPF_ALU uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008127 return -EINVAL;
8128 }
8129 /* check src1 operand */
Edward Creedc503a82017-08-15 20:34:35 +01008130 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008131 if (err)
8132 return err;
8133 } else {
8134 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008135 verbose(env, "BPF_ALU uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008136 return -EINVAL;
8137 }
8138 }
8139
8140 /* check src2 operand */
Edward Creedc503a82017-08-15 20:34:35 +01008141 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008142 if (err)
8143 return err;
8144
8145 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
8146 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008147 verbose(env, "div by zero\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008148 return -EINVAL;
8149 }
8150
Rabin Vincent229394e82016-01-12 20:17:08 +01008151 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
8152 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
8153 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
8154
8155 if (insn->imm < 0 || insn->imm >= size) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008156 verbose(env, "invalid shift %d\n", insn->imm);
Rabin Vincent229394e82016-01-12 20:17:08 +01008157 return -EINVAL;
8158 }
8159 }
8160
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07008161 /* check dest operand */
Edward Creedc503a82017-08-15 20:34:35 +01008162 err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07008163 if (err)
8164 return err;
8165
Edward Creef1174f72017-08-07 15:26:19 +01008166 return adjust_reg_min_max_vals(env, insn);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008167 }
8168
8169 return 0;
8170}
8171
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008172static void __find_good_pkt_pointers(struct bpf_func_state *state,
8173 struct bpf_reg_state *dst_reg,
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008174 enum bpf_reg_type type, int new_range)
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008175{
8176 struct bpf_reg_state *reg;
8177 int i;
8178
8179 for (i = 0; i < MAX_BPF_REG; i++) {
8180 reg = &state->regs[i];
8181 if (reg->type == type && reg->id == dst_reg->id)
8182 /* keep the maximum range already checked */
8183 reg->range = max(reg->range, new_range);
8184 }
8185
8186 bpf_for_each_spilled_reg(i, state, reg) {
8187 if (!reg)
8188 continue;
8189 if (reg->type == type && reg->id == dst_reg->id)
8190 reg->range = max(reg->range, new_range);
8191 }
8192}
8193
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08008194static void find_good_pkt_pointers(struct bpf_verifier_state *vstate,
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02008195 struct bpf_reg_state *dst_reg,
David S. Millerf8ddadc2017-10-22 13:36:53 +01008196 enum bpf_reg_type type,
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008197 bool range_right_open)
Alexei Starovoitov969bf052016-05-05 19:49:10 -07008198{
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008199 int new_range, i;
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008200
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008201 if (dst_reg->off < 0 ||
8202 (dst_reg->off == 0 && range_right_open))
Edward Creef1174f72017-08-07 15:26:19 +01008203 /* This doesn't give us any range */
8204 return;
8205
Edward Creeb03c9f92017-08-07 15:26:36 +01008206 if (dst_reg->umax_value > MAX_PACKET_OFF ||
8207 dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF)
Edward Creef1174f72017-08-07 15:26:19 +01008208 /* Risk of overflow. For instance, ptr + (1<<63) may be less
8209 * than pkt_end, but that's because it's also less than pkt.
8210 */
8211 return;
8212
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008213 new_range = dst_reg->off;
8214 if (range_right_open)
8215 new_range--;
8216
8217 /* Examples for register markings:
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008218 *
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008219 * pkt_data in dst register:
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008220 *
8221 * r2 = r3;
8222 * r2 += 8;
8223 * if (r2 > pkt_end) goto <handle exception>
8224 * <access okay>
8225 *
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008226 * r2 = r3;
8227 * r2 += 8;
8228 * if (r2 < pkt_end) goto <access okay>
8229 * <handle exception>
8230 *
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008231 * Where:
8232 * r2 == dst_reg, pkt_end == src_reg
8233 * r2=pkt(id=n,off=8,r=0)
8234 * r3=pkt(id=n,off=0,r=0)
8235 *
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008236 * pkt_data in src register:
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008237 *
8238 * r2 = r3;
8239 * r2 += 8;
8240 * if (pkt_end >= r2) goto <access okay>
8241 * <handle exception>
8242 *
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008243 * r2 = r3;
8244 * r2 += 8;
8245 * if (pkt_end <= r2) goto <handle exception>
8246 * <access okay>
8247 *
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008248 * Where:
8249 * pkt_end == dst_reg, r2 == src_reg
8250 * r2=pkt(id=n,off=8,r=0)
8251 * r3=pkt(id=n,off=0,r=0)
8252 *
8253 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
Daniel Borkmannfb2a3112017-10-21 02:34:21 +02008254 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
8255 * and [r3, r3 + 8-1) respectively is safe to access depending on
8256 * the check.
Alexei Starovoitov969bf052016-05-05 19:49:10 -07008257 */
Daniel Borkmann2d2be8c2016-09-08 01:03:42 +02008258
Edward Creef1174f72017-08-07 15:26:19 +01008259 /* If our ids match, then we must have the same max_value. And we
8260 * don't care about the other reg's fixed offset, since if it's too big
8261 * the range won't allow anything.
8262 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
8263 */
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008264 for (i = 0; i <= vstate->curframe; i++)
8265 __find_good_pkt_pointers(vstate->frame[i], dst_reg, type,
8266 new_range);
Alexei Starovoitov969bf052016-05-05 19:49:10 -07008267}
8268
John Fastabend3f50f132020-03-30 14:36:39 -07008269static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008270{
John Fastabend3f50f132020-03-30 14:36:39 -07008271 struct tnum subreg = tnum_subreg(reg->var_off);
8272 s32 sval = (s32)val;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008273
John Fastabend3f50f132020-03-30 14:36:39 -07008274 switch (opcode) {
8275 case BPF_JEQ:
8276 if (tnum_is_const(subreg))
8277 return !!tnum_equals_const(subreg, val);
8278 break;
8279 case BPF_JNE:
8280 if (tnum_is_const(subreg))
8281 return !tnum_equals_const(subreg, val);
8282 break;
8283 case BPF_JSET:
8284 if ((~subreg.mask & subreg.value) & val)
8285 return 1;
8286 if (!((subreg.mask | subreg.value) & val))
8287 return 0;
8288 break;
8289 case BPF_JGT:
8290 if (reg->u32_min_value > val)
8291 return 1;
8292 else if (reg->u32_max_value <= val)
8293 return 0;
8294 break;
8295 case BPF_JSGT:
8296 if (reg->s32_min_value > sval)
8297 return 1;
Daniel Borkmannee114dd2021-02-05 17:20:14 +01008298 else if (reg->s32_max_value <= sval)
John Fastabend3f50f132020-03-30 14:36:39 -07008299 return 0;
8300 break;
8301 case BPF_JLT:
8302 if (reg->u32_max_value < val)
8303 return 1;
8304 else if (reg->u32_min_value >= val)
8305 return 0;
8306 break;
8307 case BPF_JSLT:
8308 if (reg->s32_max_value < sval)
8309 return 1;
8310 else if (reg->s32_min_value >= sval)
8311 return 0;
8312 break;
8313 case BPF_JGE:
8314 if (reg->u32_min_value >= val)
8315 return 1;
8316 else if (reg->u32_max_value < val)
8317 return 0;
8318 break;
8319 case BPF_JSGE:
8320 if (reg->s32_min_value >= sval)
8321 return 1;
8322 else if (reg->s32_max_value < sval)
8323 return 0;
8324 break;
8325 case BPF_JLE:
8326 if (reg->u32_max_value <= val)
8327 return 1;
8328 else if (reg->u32_min_value > val)
8329 return 0;
8330 break;
8331 case BPF_JSLE:
8332 if (reg->s32_max_value <= sval)
8333 return 1;
8334 else if (reg->s32_min_value > sval)
8335 return 0;
8336 break;
Jiong Wang092ed092019-01-26 12:26:01 -05008337 }
Jiong Wanga72dafa2019-01-26 12:26:00 -05008338
John Fastabend3f50f132020-03-30 14:36:39 -07008339 return -1;
8340}
8341
8342
8343static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode)
8344{
8345 s64 sval = (s64)val;
8346
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008347 switch (opcode) {
8348 case BPF_JEQ:
8349 if (tnum_is_const(reg->var_off))
8350 return !!tnum_equals_const(reg->var_off, val);
8351 break;
8352 case BPF_JNE:
8353 if (tnum_is_const(reg->var_off))
8354 return !tnum_equals_const(reg->var_off, val);
8355 break;
Jakub Kicinski960ea052018-12-19 22:13:04 -08008356 case BPF_JSET:
8357 if ((~reg->var_off.mask & reg->var_off.value) & val)
8358 return 1;
8359 if (!((reg->var_off.mask | reg->var_off.value) & val))
8360 return 0;
8361 break;
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008362 case BPF_JGT:
8363 if (reg->umin_value > val)
8364 return 1;
8365 else if (reg->umax_value <= val)
8366 return 0;
8367 break;
8368 case BPF_JSGT:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008369 if (reg->smin_value > sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008370 return 1;
Daniel Borkmannee114dd2021-02-05 17:20:14 +01008371 else if (reg->smax_value <= sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008372 return 0;
8373 break;
8374 case BPF_JLT:
8375 if (reg->umax_value < val)
8376 return 1;
8377 else if (reg->umin_value >= val)
8378 return 0;
8379 break;
8380 case BPF_JSLT:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008381 if (reg->smax_value < sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008382 return 1;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008383 else if (reg->smin_value >= sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008384 return 0;
8385 break;
8386 case BPF_JGE:
8387 if (reg->umin_value >= val)
8388 return 1;
8389 else if (reg->umax_value < val)
8390 return 0;
8391 break;
8392 case BPF_JSGE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008393 if (reg->smin_value >= sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008394 return 1;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008395 else if (reg->smax_value < sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008396 return 0;
8397 break;
8398 case BPF_JLE:
8399 if (reg->umax_value <= val)
8400 return 1;
8401 else if (reg->umin_value > val)
8402 return 0;
8403 break;
8404 case BPF_JSLE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008405 if (reg->smax_value <= sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008406 return 1;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008407 else if (reg->smin_value > sval)
Alexei Starovoitov4f7b3e82018-12-03 22:46:05 -08008408 return 0;
8409 break;
8410 }
8411
8412 return -1;
8413}
8414
John Fastabend3f50f132020-03-30 14:36:39 -07008415/* compute branch direction of the expression "if (reg opcode val) goto target;"
8416 * and return:
8417 * 1 - branch will be taken and "goto target" will be executed
8418 * 0 - branch will not be taken and fall-through to next insn
8419 * -1 - unknown. Example: "if (reg < 5)" is unknown when register value
8420 * range [0,10]
Jiong Wang092ed092019-01-26 12:26:01 -05008421 */
John Fastabend3f50f132020-03-30 14:36:39 -07008422static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode,
8423 bool is_jmp32)
Jiong Wang092ed092019-01-26 12:26:01 -05008424{
John Fastabendcac616d2020-05-21 13:07:26 -07008425 if (__is_pointer_value(false, reg)) {
8426 if (!reg_type_not_null(reg->type))
8427 return -1;
8428
8429 /* If pointer is valid tests against zero will fail so we can
8430 * use this to direct branch taken.
8431 */
8432 if (val != 0)
8433 return -1;
8434
8435 switch (opcode) {
8436 case BPF_JEQ:
8437 return 0;
8438 case BPF_JNE:
8439 return 1;
8440 default:
8441 return -1;
8442 }
8443 }
Jiong Wang092ed092019-01-26 12:26:01 -05008444
John Fastabend3f50f132020-03-30 14:36:39 -07008445 if (is_jmp32)
8446 return is_branch32_taken(reg, val, opcode);
8447 return is_branch64_taken(reg, val, opcode);
Jann Horn604dca52020-03-30 18:03:23 +02008448}
8449
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008450static int flip_opcode(u32 opcode)
8451{
8452 /* How can we transform "a <op> b" into "b <op> a"? */
8453 static const u8 opcode_flip[16] = {
8454 /* these stay the same */
8455 [BPF_JEQ >> 4] = BPF_JEQ,
8456 [BPF_JNE >> 4] = BPF_JNE,
8457 [BPF_JSET >> 4] = BPF_JSET,
8458 /* these swap "lesser" and "greater" (L and G in the opcodes) */
8459 [BPF_JGE >> 4] = BPF_JLE,
8460 [BPF_JGT >> 4] = BPF_JLT,
8461 [BPF_JLE >> 4] = BPF_JGE,
8462 [BPF_JLT >> 4] = BPF_JGT,
8463 [BPF_JSGE >> 4] = BPF_JSLE,
8464 [BPF_JSGT >> 4] = BPF_JSLT,
8465 [BPF_JSLE >> 4] = BPF_JSGE,
8466 [BPF_JSLT >> 4] = BPF_JSGT
8467 };
8468 return opcode_flip[opcode >> 4];
8469}
8470
8471static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg,
8472 struct bpf_reg_state *src_reg,
8473 u8 opcode)
8474{
8475 struct bpf_reg_state *pkt;
8476
8477 if (src_reg->type == PTR_TO_PACKET_END) {
8478 pkt = dst_reg;
8479 } else if (dst_reg->type == PTR_TO_PACKET_END) {
8480 pkt = src_reg;
8481 opcode = flip_opcode(opcode);
8482 } else {
8483 return -1;
8484 }
8485
8486 if (pkt->range >= 0)
8487 return -1;
8488
8489 switch (opcode) {
8490 case BPF_JLE:
8491 /* pkt <= pkt_end */
8492 fallthrough;
8493 case BPF_JGT:
8494 /* pkt > pkt_end */
8495 if (pkt->range == BEYOND_PKT_END)
8496 /* pkt has at last one extra byte beyond pkt_end */
8497 return opcode == BPF_JGT;
8498 break;
8499 case BPF_JLT:
8500 /* pkt < pkt_end */
8501 fallthrough;
8502 case BPF_JGE:
8503 /* pkt >= pkt_end */
8504 if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END)
8505 return opcode == BPF_JGE;
8506 break;
8507 }
8508 return -1;
8509}
8510
Josef Bacik48461132016-09-28 10:54:32 -04008511/* Adjusts the register min/max values in the case that the dst_reg is the
8512 * variable register that we are working on, and src_reg is a constant or we're
8513 * simply doing a BPF_K check.
Edward Creef1174f72017-08-07 15:26:19 +01008514 * In JEQ/JNE cases we also adjust the var_off values.
Josef Bacik48461132016-09-28 10:54:32 -04008515 */
8516static void reg_set_min_max(struct bpf_reg_state *true_reg,
John Fastabend3f50f132020-03-30 14:36:39 -07008517 struct bpf_reg_state *false_reg,
8518 u64 val, u32 val32,
Jiong Wang092ed092019-01-26 12:26:01 -05008519 u8 opcode, bool is_jmp32)
Josef Bacik48461132016-09-28 10:54:32 -04008520{
John Fastabend3f50f132020-03-30 14:36:39 -07008521 struct tnum false_32off = tnum_subreg(false_reg->var_off);
8522 struct tnum false_64off = false_reg->var_off;
8523 struct tnum true_32off = tnum_subreg(true_reg->var_off);
8524 struct tnum true_64off = true_reg->var_off;
8525 s64 sval = (s64)val;
8526 s32 sval32 = (s32)val32;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008527
Edward Creef1174f72017-08-07 15:26:19 +01008528 /* If the dst_reg is a pointer, we can't learn anything about its
8529 * variable offset from the compare (unless src_reg were a pointer into
8530 * the same object, but we don't bother with that.
8531 * Since false_reg and true_reg have the same type by construction, we
8532 * only need to check one of them for pointerness.
8533 */
8534 if (__is_pointer_value(false, false_reg))
8535 return;
Daniel Borkmann4cabc5b2017-07-21 00:00:21 +02008536
Josef Bacik48461132016-09-28 10:54:32 -04008537 switch (opcode) {
8538 case BPF_JEQ:
Josef Bacik48461132016-09-28 10:54:32 -04008539 case BPF_JNE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008540 {
8541 struct bpf_reg_state *reg =
8542 opcode == BPF_JEQ ? true_reg : false_reg;
8543
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07008544 /* JEQ/JNE comparison doesn't change the register equivalence.
8545 * r1 = r2;
8546 * if (r1 == 42) goto label;
8547 * ...
8548 * label: // here both r1 and r2 are known to be 42.
8549 *
8550 * Hence when marking register as known preserve it's ID.
Josef Bacik48461132016-09-28 10:54:32 -04008551 */
John Fastabend3f50f132020-03-30 14:36:39 -07008552 if (is_jmp32)
8553 __mark_reg32_known(reg, val32);
8554 else
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07008555 ___mark_reg_known(reg, val);
Josef Bacik48461132016-09-28 10:54:32 -04008556 break;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008557 }
Jakub Kicinski960ea052018-12-19 22:13:04 -08008558 case BPF_JSET:
John Fastabend3f50f132020-03-30 14:36:39 -07008559 if (is_jmp32) {
8560 false_32off = tnum_and(false_32off, tnum_const(~val32));
8561 if (is_power_of_2(val32))
8562 true_32off = tnum_or(true_32off,
8563 tnum_const(val32));
8564 } else {
8565 false_64off = tnum_and(false_64off, tnum_const(~val));
8566 if (is_power_of_2(val))
8567 true_64off = tnum_or(true_64off,
8568 tnum_const(val));
8569 }
Jakub Kicinski960ea052018-12-19 22:13:04 -08008570 break;
Josef Bacik48461132016-09-28 10:54:32 -04008571 case BPF_JGE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008572 case BPF_JGT:
8573 {
John Fastabend3f50f132020-03-30 14:36:39 -07008574 if (is_jmp32) {
8575 u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1;
8576 u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32;
8577
8578 false_reg->u32_max_value = min(false_reg->u32_max_value,
8579 false_umax);
8580 true_reg->u32_min_value = max(true_reg->u32_min_value,
8581 true_umin);
8582 } else {
8583 u64 false_umax = opcode == BPF_JGT ? val : val - 1;
8584 u64 true_umin = opcode == BPF_JGT ? val + 1 : val;
8585
8586 false_reg->umax_value = min(false_reg->umax_value, false_umax);
8587 true_reg->umin_value = max(true_reg->umin_value, true_umin);
8588 }
Edward Creeb03c9f92017-08-07 15:26:36 +01008589 break;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008590 }
Josef Bacik48461132016-09-28 10:54:32 -04008591 case BPF_JSGE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008592 case BPF_JSGT:
8593 {
John Fastabend3f50f132020-03-30 14:36:39 -07008594 if (is_jmp32) {
8595 s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1;
8596 s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008597
John Fastabend3f50f132020-03-30 14:36:39 -07008598 false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax);
8599 true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin);
8600 } else {
8601 s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1;
8602 s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval;
8603
8604 false_reg->smax_value = min(false_reg->smax_value, false_smax);
8605 true_reg->smin_value = max(true_reg->smin_value, true_smin);
8606 }
Josef Bacik48461132016-09-28 10:54:32 -04008607 break;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008608 }
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008609 case BPF_JLE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008610 case BPF_JLT:
8611 {
John Fastabend3f50f132020-03-30 14:36:39 -07008612 if (is_jmp32) {
8613 u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1;
8614 u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32;
8615
8616 false_reg->u32_min_value = max(false_reg->u32_min_value,
8617 false_umin);
8618 true_reg->u32_max_value = min(true_reg->u32_max_value,
8619 true_umax);
8620 } else {
8621 u64 false_umin = opcode == BPF_JLT ? val : val + 1;
8622 u64 true_umax = opcode == BPF_JLT ? val - 1 : val;
8623
8624 false_reg->umin_value = max(false_reg->umin_value, false_umin);
8625 true_reg->umax_value = min(true_reg->umax_value, true_umax);
8626 }
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008627 break;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008628 }
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008629 case BPF_JSLE:
Jiong Wanga72dafa2019-01-26 12:26:00 -05008630 case BPF_JSLT:
8631 {
John Fastabend3f50f132020-03-30 14:36:39 -07008632 if (is_jmp32) {
8633 s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1;
8634 s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008635
John Fastabend3f50f132020-03-30 14:36:39 -07008636 false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin);
8637 true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax);
8638 } else {
8639 s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1;
8640 s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval;
8641
8642 false_reg->smin_value = max(false_reg->smin_value, false_smin);
8643 true_reg->smax_value = min(true_reg->smax_value, true_smax);
8644 }
Daniel Borkmannb4e432f2017-08-10 01:40:02 +02008645 break;
Jiong Wanga72dafa2019-01-26 12:26:00 -05008646 }
Josef Bacik48461132016-09-28 10:54:32 -04008647 default:
Jann Horn0fc31b12020-03-30 18:03:24 +02008648 return;
Josef Bacik48461132016-09-28 10:54:32 -04008649 }
8650
John Fastabend3f50f132020-03-30 14:36:39 -07008651 if (is_jmp32) {
8652 false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off),
8653 tnum_subreg(false_32off));
8654 true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off),
8655 tnum_subreg(true_32off));
8656 __reg_combine_32_into_64(false_reg);
8657 __reg_combine_32_into_64(true_reg);
8658 } else {
8659 false_reg->var_off = false_64off;
8660 true_reg->var_off = true_64off;
8661 __reg_combine_64_into_32(false_reg);
8662 __reg_combine_64_into_32(true_reg);
8663 }
Josef Bacik48461132016-09-28 10:54:32 -04008664}
8665
Edward Creef1174f72017-08-07 15:26:19 +01008666/* Same as above, but for the case that dst_reg holds a constant and src_reg is
8667 * the variable reg.
Josef Bacik48461132016-09-28 10:54:32 -04008668 */
8669static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
John Fastabend3f50f132020-03-30 14:36:39 -07008670 struct bpf_reg_state *false_reg,
8671 u64 val, u32 val32,
Jiong Wang092ed092019-01-26 12:26:01 -05008672 u8 opcode, bool is_jmp32)
Josef Bacik48461132016-09-28 10:54:32 -04008673{
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008674 opcode = flip_opcode(opcode);
Jann Horn0fc31b12020-03-30 18:03:24 +02008675 /* This uses zero as "not present in table"; luckily the zero opcode,
8676 * BPF_JA, can't get here.
Edward Creeb03c9f92017-08-07 15:26:36 +01008677 */
Jann Horn0fc31b12020-03-30 18:03:24 +02008678 if (opcode)
John Fastabend3f50f132020-03-30 14:36:39 -07008679 reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32);
Edward Creef1174f72017-08-07 15:26:19 +01008680}
8681
8682/* Regs are known to be equal, so intersect their min/max/var_off */
8683static void __reg_combine_min_max(struct bpf_reg_state *src_reg,
8684 struct bpf_reg_state *dst_reg)
8685{
Edward Creeb03c9f92017-08-07 15:26:36 +01008686 src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value,
8687 dst_reg->umin_value);
8688 src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value,
8689 dst_reg->umax_value);
8690 src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value,
8691 dst_reg->smin_value);
8692 src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value,
8693 dst_reg->smax_value);
Edward Creef1174f72017-08-07 15:26:19 +01008694 src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off,
8695 dst_reg->var_off);
Edward Creeb03c9f92017-08-07 15:26:36 +01008696 /* We might have learned new bounds from the var_off. */
8697 __update_reg_bounds(src_reg);
8698 __update_reg_bounds(dst_reg);
8699 /* We might have learned something about the sign bit. */
8700 __reg_deduce_bounds(src_reg);
8701 __reg_deduce_bounds(dst_reg);
8702 /* We might have learned some bits from the bounds. */
8703 __reg_bound_offset(src_reg);
8704 __reg_bound_offset(dst_reg);
8705 /* Intersecting with the old var_off might have improved our bounds
8706 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
8707 * then new var_off is (0; 0x7f...fc) which improves our umax.
8708 */
8709 __update_reg_bounds(src_reg);
8710 __update_reg_bounds(dst_reg);
Edward Creef1174f72017-08-07 15:26:19 +01008711}
8712
8713static void reg_combine_min_max(struct bpf_reg_state *true_src,
8714 struct bpf_reg_state *true_dst,
8715 struct bpf_reg_state *false_src,
8716 struct bpf_reg_state *false_dst,
8717 u8 opcode)
8718{
8719 switch (opcode) {
8720 case BPF_JEQ:
8721 __reg_combine_min_max(true_src, true_dst);
8722 break;
8723 case BPF_JNE:
8724 __reg_combine_min_max(false_src, false_dst);
Edward Creeb03c9f92017-08-07 15:26:36 +01008725 break;
Daniel Borkmann4cabc5b2017-07-21 00:00:21 +02008726 }
Josef Bacik48461132016-09-28 10:54:32 -04008727}
8728
Joe Stringerfd978bf72018-10-02 13:35:35 -07008729static void mark_ptr_or_null_reg(struct bpf_func_state *state,
8730 struct bpf_reg_state *reg, u32 id,
Joe Stringer840b9612018-10-02 13:35:32 -07008731 bool is_null)
Thomas Graf57a09bf2016-10-18 19:51:19 +02008732{
Martin KaFai Lau93c230e2020-10-19 12:42:12 -07008733 if (reg_type_may_be_null(reg->type) && reg->id == id &&
8734 !WARN_ON_ONCE(!reg->id)) {
Edward Creef1174f72017-08-07 15:26:19 +01008735 /* Old offset (both fixed and variable parts) should
8736 * have been known-zero, because we don't allow pointer
8737 * arithmetic on pointers that might be NULL.
8738 */
Edward Creeb03c9f92017-08-07 15:26:36 +01008739 if (WARN_ON_ONCE(reg->smin_value || reg->smax_value ||
8740 !tnum_equals_const(reg->var_off, 0) ||
Edward Creef1174f72017-08-07 15:26:19 +01008741 reg->off)) {
Edward Creeb03c9f92017-08-07 15:26:36 +01008742 __mark_reg_known_zero(reg);
8743 reg->off = 0;
Edward Creef1174f72017-08-07 15:26:19 +01008744 }
8745 if (is_null) {
8746 reg->type = SCALAR_VALUE;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07008747 /* We don't need id and ref_obj_id from this point
8748 * onwards anymore, thus we should better reset it,
8749 * so that state pruning has chances to take effect.
8750 */
8751 reg->id = 0;
8752 reg->ref_obj_id = 0;
Dmitrii Banshchikov4ddb7412021-02-13 00:56:40 +04008753
8754 return;
8755 }
8756
8757 mark_ptr_not_null_reg(reg);
8758
8759 if (!reg_may_point_to_spin_lock(reg)) {
Martin KaFai Lau1b986582019-03-12 10:23:02 -07008760 /* For not-NULL ptr, reg->ref_obj_id will be reset
8761 * in release_reg_references().
8762 *
8763 * reg->id is still used by spin_lock ptr. Other
8764 * than spin_lock ptr type, reg->id can be reset.
Joe Stringerfd978bf72018-10-02 13:35:35 -07008765 */
8766 reg->id = 0;
8767 }
Thomas Graf57a09bf2016-10-18 19:51:19 +02008768 }
8769}
8770
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008771static void __mark_ptr_or_null_regs(struct bpf_func_state *state, u32 id,
8772 bool is_null)
8773{
8774 struct bpf_reg_state *reg;
8775 int i;
8776
8777 for (i = 0; i < MAX_BPF_REG; i++)
8778 mark_ptr_or_null_reg(state, &state->regs[i], id, is_null);
8779
8780 bpf_for_each_spilled_reg(i, state, reg) {
8781 if (!reg)
8782 continue;
8783 mark_ptr_or_null_reg(state, reg, id, is_null);
8784 }
8785}
8786
Thomas Graf57a09bf2016-10-18 19:51:19 +02008787/* The logic is similar to find_good_pkt_pointers(), both could eventually
8788 * be folded together at some point.
8789 */
Joe Stringer840b9612018-10-02 13:35:32 -07008790static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno,
8791 bool is_null)
Thomas Graf57a09bf2016-10-18 19:51:19 +02008792{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08008793 struct bpf_func_state *state = vstate->frame[vstate->curframe];
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008794 struct bpf_reg_state *regs = state->regs;
Martin KaFai Lau1b986582019-03-12 10:23:02 -07008795 u32 ref_obj_id = regs[regno].ref_obj_id;
Daniel Borkmanna08dd0d2016-12-15 01:30:06 +01008796 u32 id = regs[regno].id;
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008797 int i;
Thomas Graf57a09bf2016-10-18 19:51:19 +02008798
Martin KaFai Lau1b986582019-03-12 10:23:02 -07008799 if (ref_obj_id && ref_obj_id == id && is_null)
8800 /* regs[regno] is in the " == NULL" branch.
8801 * No one could have freed the reference state before
8802 * doing the NULL check.
8803 */
8804 WARN_ON_ONCE(release_reference_state(state, id));
Joe Stringerfd978bf72018-10-02 13:35:35 -07008805
Paul Chaignonc6a9efa2019-04-24 21:50:42 +02008806 for (i = 0; i <= vstate->curframe; i++)
8807 __mark_ptr_or_null_regs(vstate->frame[i], id, is_null);
Thomas Graf57a09bf2016-10-18 19:51:19 +02008808}
8809
Daniel Borkmann5beca082017-11-01 23:58:10 +01008810static bool try_match_pkt_pointers(const struct bpf_insn *insn,
8811 struct bpf_reg_state *dst_reg,
8812 struct bpf_reg_state *src_reg,
8813 struct bpf_verifier_state *this_branch,
8814 struct bpf_verifier_state *other_branch)
8815{
8816 if (BPF_SRC(insn->code) != BPF_X)
8817 return false;
8818
Jiong Wang092ed092019-01-26 12:26:01 -05008819 /* Pointers are always 64-bit. */
8820 if (BPF_CLASS(insn->code) == BPF_JMP32)
8821 return false;
8822
Daniel Borkmann5beca082017-11-01 23:58:10 +01008823 switch (BPF_OP(insn->code)) {
8824 case BPF_JGT:
8825 if ((dst_reg->type == PTR_TO_PACKET &&
8826 src_reg->type == PTR_TO_PACKET_END) ||
8827 (dst_reg->type == PTR_TO_PACKET_META &&
8828 reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
8829 /* pkt_data' > pkt_end, pkt_meta' > pkt_data */
8830 find_good_pkt_pointers(this_branch, dst_reg,
8831 dst_reg->type, false);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008832 mark_pkt_end(other_branch, insn->dst_reg, true);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008833 } else if ((dst_reg->type == PTR_TO_PACKET_END &&
8834 src_reg->type == PTR_TO_PACKET) ||
8835 (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
8836 src_reg->type == PTR_TO_PACKET_META)) {
8837 /* pkt_end > pkt_data', pkt_data > pkt_meta' */
8838 find_good_pkt_pointers(other_branch, src_reg,
8839 src_reg->type, true);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008840 mark_pkt_end(this_branch, insn->src_reg, false);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008841 } else {
8842 return false;
8843 }
8844 break;
8845 case BPF_JLT:
8846 if ((dst_reg->type == PTR_TO_PACKET &&
8847 src_reg->type == PTR_TO_PACKET_END) ||
8848 (dst_reg->type == PTR_TO_PACKET_META &&
8849 reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
8850 /* pkt_data' < pkt_end, pkt_meta' < pkt_data */
8851 find_good_pkt_pointers(other_branch, dst_reg,
8852 dst_reg->type, true);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008853 mark_pkt_end(this_branch, insn->dst_reg, false);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008854 } else if ((dst_reg->type == PTR_TO_PACKET_END &&
8855 src_reg->type == PTR_TO_PACKET) ||
8856 (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
8857 src_reg->type == PTR_TO_PACKET_META)) {
8858 /* pkt_end < pkt_data', pkt_data > pkt_meta' */
8859 find_good_pkt_pointers(this_branch, src_reg,
8860 src_reg->type, false);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008861 mark_pkt_end(other_branch, insn->src_reg, true);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008862 } else {
8863 return false;
8864 }
8865 break;
8866 case BPF_JGE:
8867 if ((dst_reg->type == PTR_TO_PACKET &&
8868 src_reg->type == PTR_TO_PACKET_END) ||
8869 (dst_reg->type == PTR_TO_PACKET_META &&
8870 reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
8871 /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */
8872 find_good_pkt_pointers(this_branch, dst_reg,
8873 dst_reg->type, true);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008874 mark_pkt_end(other_branch, insn->dst_reg, false);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008875 } else if ((dst_reg->type == PTR_TO_PACKET_END &&
8876 src_reg->type == PTR_TO_PACKET) ||
8877 (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
8878 src_reg->type == PTR_TO_PACKET_META)) {
8879 /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */
8880 find_good_pkt_pointers(other_branch, src_reg,
8881 src_reg->type, false);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008882 mark_pkt_end(this_branch, insn->src_reg, true);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008883 } else {
8884 return false;
8885 }
8886 break;
8887 case BPF_JLE:
8888 if ((dst_reg->type == PTR_TO_PACKET &&
8889 src_reg->type == PTR_TO_PACKET_END) ||
8890 (dst_reg->type == PTR_TO_PACKET_META &&
8891 reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) {
8892 /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */
8893 find_good_pkt_pointers(other_branch, dst_reg,
8894 dst_reg->type, false);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008895 mark_pkt_end(this_branch, insn->dst_reg, true);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008896 } else if ((dst_reg->type == PTR_TO_PACKET_END &&
8897 src_reg->type == PTR_TO_PACKET) ||
8898 (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
8899 src_reg->type == PTR_TO_PACKET_META)) {
8900 /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */
8901 find_good_pkt_pointers(this_branch, src_reg,
8902 src_reg->type, true);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08008903 mark_pkt_end(other_branch, insn->src_reg, false);
Daniel Borkmann5beca082017-11-01 23:58:10 +01008904 } else {
8905 return false;
8906 }
8907 break;
8908 default:
8909 return false;
8910 }
8911
8912 return true;
8913}
8914
Alexei Starovoitov75748832020-10-08 18:12:37 -07008915static void find_equal_scalars(struct bpf_verifier_state *vstate,
8916 struct bpf_reg_state *known_reg)
8917{
8918 struct bpf_func_state *state;
8919 struct bpf_reg_state *reg;
8920 int i, j;
8921
8922 for (i = 0; i <= vstate->curframe; i++) {
8923 state = vstate->frame[i];
8924 for (j = 0; j < MAX_BPF_REG; j++) {
8925 reg = &state->regs[j];
8926 if (reg->type == SCALAR_VALUE && reg->id == known_reg->id)
8927 *reg = *known_reg;
8928 }
8929
8930 bpf_for_each_spilled_reg(j, state, reg) {
8931 if (!reg)
8932 continue;
8933 if (reg->type == SCALAR_VALUE && reg->id == known_reg->id)
8934 *reg = *known_reg;
8935 }
8936 }
8937}
8938
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01008939static int check_cond_jmp_op(struct bpf_verifier_env *env,
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008940 struct bpf_insn *insn, int *insn_idx)
8941{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08008942 struct bpf_verifier_state *this_branch = env->cur_state;
8943 struct bpf_verifier_state *other_branch;
8944 struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs;
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07008945 struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008946 u8 opcode = BPF_OP(insn->code);
Jiong Wang092ed092019-01-26 12:26:01 -05008947 bool is_jmp32;
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07008948 int pred = -1;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008949 int err;
8950
Jiong Wang092ed092019-01-26 12:26:01 -05008951 /* Only conditional jumps are expected to reach here. */
8952 if (opcode == BPF_JA || opcode > BPF_JSLE) {
8953 verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008954 return -EINVAL;
8955 }
8956
8957 if (BPF_SRC(insn->code) == BPF_X) {
8958 if (insn->imm != 0) {
Jiong Wang092ed092019-01-26 12:26:01 -05008959 verbose(env, "BPF_JMP/JMP32 uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008960 return -EINVAL;
8961 }
8962
8963 /* check src1 operand */
Edward Creedc503a82017-08-15 20:34:35 +01008964 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008965 if (err)
8966 return err;
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008967
8968 if (is_pointer_value(env, insn->src_reg)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07008969 verbose(env, "R%d pointer comparison prohibited\n",
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07008970 insn->src_reg);
8971 return -EACCES;
8972 }
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07008973 src_reg = &regs[insn->src_reg];
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008974 } else {
8975 if (insn->src_reg != BPF_REG_0) {
Jiong Wang092ed092019-01-26 12:26:01 -05008976 verbose(env, "BPF_JMP/JMP32 uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008977 return -EINVAL;
8978 }
8979 }
8980
8981 /* check src2 operand */
Edward Creedc503a82017-08-15 20:34:35 +01008982 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07008983 if (err)
8984 return err;
8985
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07008986 dst_reg = &regs[insn->dst_reg];
Jiong Wang092ed092019-01-26 12:26:01 -05008987 is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32;
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07008988
John Fastabend3f50f132020-03-30 14:36:39 -07008989 if (BPF_SRC(insn->code) == BPF_K) {
8990 pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32);
8991 } else if (src_reg->type == SCALAR_VALUE &&
8992 is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) {
8993 pred = is_branch_taken(dst_reg,
8994 tnum_subreg(src_reg->var_off).value,
8995 opcode,
8996 is_jmp32);
8997 } else if (src_reg->type == SCALAR_VALUE &&
8998 !is_jmp32 && tnum_is_const(src_reg->var_off)) {
8999 pred = is_branch_taken(dst_reg,
9000 src_reg->var_off.value,
9001 opcode,
9002 is_jmp32);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08009003 } else if (reg_is_pkt_pointer_any(dst_reg) &&
9004 reg_is_pkt_pointer_any(src_reg) &&
9005 !is_jmp32) {
9006 pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode);
John Fastabend3f50f132020-03-30 14:36:39 -07009007 }
9008
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07009009 if (pred >= 0) {
John Fastabendcac616d2020-05-21 13:07:26 -07009010 /* If we get here with a dst_reg pointer type it is because
9011 * above is_branch_taken() special cased the 0 comparison.
9012 */
9013 if (!__is_pointer_value(false, dst_reg))
9014 err = mark_chain_precision(env, insn->dst_reg);
Alexei Starovoitov6d94e742020-11-10 19:12:11 -08009015 if (BPF_SRC(insn->code) == BPF_X && !err &&
9016 !__is_pointer_value(false, src_reg))
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -07009017 err = mark_chain_precision(env, insn->src_reg);
9018 if (err)
9019 return err;
9020 }
Daniel Borkmann9183671a2021-05-28 15:47:32 +00009021
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07009022 if (pred == 1) {
Daniel Borkmann9183671a2021-05-28 15:47:32 +00009023 /* Only follow the goto, ignore fall-through. If needed, push
9024 * the fall-through branch for simulation under speculative
9025 * execution.
9026 */
9027 if (!env->bypass_spec_v1 &&
9028 !sanitize_speculative_path(env, insn, *insn_idx + 1,
9029 *insn_idx))
9030 return -EFAULT;
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07009031 *insn_idx += insn->off;
9032 return 0;
9033 } else if (pred == 0) {
Daniel Borkmann9183671a2021-05-28 15:47:32 +00009034 /* Only follow the fall-through branch, since that's where the
9035 * program will go. If needed, push the goto branch for
9036 * simulation under speculative execution.
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07009037 */
Daniel Borkmann9183671a2021-05-28 15:47:32 +00009038 if (!env->bypass_spec_v1 &&
9039 !sanitize_speculative_path(env, insn,
9040 *insn_idx + insn->off + 1,
9041 *insn_idx))
9042 return -EFAULT;
Alexei Starovoitovfb8d2512019-06-15 12:12:19 -07009043 return 0;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009044 }
9045
Daniel Borkmann979d63d2019-01-03 00:58:34 +01009046 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx,
9047 false);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009048 if (!other_branch)
9049 return -EFAULT;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009050 other_branch_regs = other_branch->frame[other_branch->curframe]->regs;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009051
Josef Bacik48461132016-09-28 10:54:32 -04009052 /* detect if we are comparing against a constant value so we can adjust
9053 * our min/max values for our dst register.
Edward Creef1174f72017-08-07 15:26:19 +01009054 * this is only legit if both are scalars (or pointers to the same
9055 * object, I suppose, but we don't support that right now), because
9056 * otherwise the different base pointers mean the offsets aren't
9057 * comparable.
Josef Bacik48461132016-09-28 10:54:32 -04009058 */
9059 if (BPF_SRC(insn->code) == BPF_X) {
Jiong Wang092ed092019-01-26 12:26:01 -05009060 struct bpf_reg_state *src_reg = &regs[insn->src_reg];
Jiong Wang092ed092019-01-26 12:26:01 -05009061
Edward Creef1174f72017-08-07 15:26:19 +01009062 if (dst_reg->type == SCALAR_VALUE &&
Jiong Wang092ed092019-01-26 12:26:01 -05009063 src_reg->type == SCALAR_VALUE) {
9064 if (tnum_is_const(src_reg->var_off) ||
John Fastabend3f50f132020-03-30 14:36:39 -07009065 (is_jmp32 &&
9066 tnum_is_const(tnum_subreg(src_reg->var_off))))
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009067 reg_set_min_max(&other_branch_regs[insn->dst_reg],
Jiong Wang092ed092019-01-26 12:26:01 -05009068 dst_reg,
John Fastabend3f50f132020-03-30 14:36:39 -07009069 src_reg->var_off.value,
9070 tnum_subreg(src_reg->var_off).value,
Jiong Wang092ed092019-01-26 12:26:01 -05009071 opcode, is_jmp32);
9072 else if (tnum_is_const(dst_reg->var_off) ||
John Fastabend3f50f132020-03-30 14:36:39 -07009073 (is_jmp32 &&
9074 tnum_is_const(tnum_subreg(dst_reg->var_off))))
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009075 reg_set_min_max_inv(&other_branch_regs[insn->src_reg],
Jiong Wang092ed092019-01-26 12:26:01 -05009076 src_reg,
John Fastabend3f50f132020-03-30 14:36:39 -07009077 dst_reg->var_off.value,
9078 tnum_subreg(dst_reg->var_off).value,
Jiong Wang092ed092019-01-26 12:26:01 -05009079 opcode, is_jmp32);
9080 else if (!is_jmp32 &&
9081 (opcode == BPF_JEQ || opcode == BPF_JNE))
Edward Creef1174f72017-08-07 15:26:19 +01009082 /* Comparing for equality, we can combine knowledge */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009083 reg_combine_min_max(&other_branch_regs[insn->src_reg],
9084 &other_branch_regs[insn->dst_reg],
Jiong Wang092ed092019-01-26 12:26:01 -05009085 src_reg, dst_reg, opcode);
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07009086 if (src_reg->id &&
9087 !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) {
Alexei Starovoitov75748832020-10-08 18:12:37 -07009088 find_equal_scalars(this_branch, src_reg);
9089 find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]);
9090 }
9091
Edward Creef1174f72017-08-07 15:26:19 +01009092 }
9093 } else if (dst_reg->type == SCALAR_VALUE) {
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009094 reg_set_min_max(&other_branch_regs[insn->dst_reg],
John Fastabend3f50f132020-03-30 14:36:39 -07009095 dst_reg, insn->imm, (u32)insn->imm,
9096 opcode, is_jmp32);
Josef Bacik48461132016-09-28 10:54:32 -04009097 }
9098
Alexei Starovoitove688c3d2020-10-14 10:56:08 -07009099 if (dst_reg->type == SCALAR_VALUE && dst_reg->id &&
9100 !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) {
Alexei Starovoitov75748832020-10-08 18:12:37 -07009101 find_equal_scalars(this_branch, dst_reg);
9102 find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]);
9103 }
9104
Jiong Wang092ed092019-01-26 12:26:01 -05009105 /* detect if R == 0 where R is returned from bpf_map_lookup_elem().
9106 * NOTE: these optimizations below are related with pointer comparison
9107 * which will never be JMP32.
9108 */
9109 if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K &&
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -07009110 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
Joe Stringer840b9612018-10-02 13:35:32 -07009111 reg_type_may_be_null(dst_reg->type)) {
9112 /* Mark all identical registers in each branch as either
Thomas Graf57a09bf2016-10-18 19:51:19 +02009113 * safe or unknown depending R == 0 or R != 0 conditional.
9114 */
Joe Stringer840b9612018-10-02 13:35:32 -07009115 mark_ptr_or_null_regs(this_branch, insn->dst_reg,
9116 opcode == BPF_JNE);
9117 mark_ptr_or_null_regs(other_branch, insn->dst_reg,
9118 opcode == BPF_JEQ);
Daniel Borkmann5beca082017-11-01 23:58:10 +01009119 } else if (!try_match_pkt_pointers(insn, dst_reg, &regs[insn->src_reg],
9120 this_branch, other_branch) &&
9121 is_pointer_value(env, insn->dst_reg)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009122 verbose(env, "R%d pointer comparison prohibited\n",
9123 insn->dst_reg);
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07009124 return -EACCES;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009125 }
Alexei Starovoitov06ee7112019-04-01 21:27:40 -07009126 if (env->log.level & BPF_LOG_LEVEL)
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -08009127 print_verifier_state(env, this_branch->frame[this_branch->curframe]);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009128 return 0;
9129}
9130
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009131/* verify BPF_LD_IMM64 instruction */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01009132static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009133{
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +02009134 struct bpf_insn_aux_data *aux = cur_aux(env);
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07009135 struct bpf_reg_state *regs = cur_regs(env);
Hao Luo4976b712020-09-29 16:50:44 -07009136 struct bpf_reg_state *dst_reg;
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +02009137 struct bpf_map *map;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009138 int err;
9139
9140 if (BPF_SIZE(insn->code) != BPF_DW) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009141 verbose(env, "invalid BPF_LD_IMM insn\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009142 return -EINVAL;
9143 }
9144 if (insn->off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009145 verbose(env, "BPF_LD_IMM64 uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009146 return -EINVAL;
9147 }
9148
Edward Creedc503a82017-08-15 20:34:35 +01009149 err = check_reg_arg(env, insn->dst_reg, DST_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009150 if (err)
9151 return err;
9152
Hao Luo4976b712020-09-29 16:50:44 -07009153 dst_reg = &regs[insn->dst_reg];
Jakub Kicinski6b173872016-09-21 11:43:59 +01009154 if (insn->src_reg == 0) {
Jakub Kicinski6b173872016-09-21 11:43:59 +01009155 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
9156
Hao Luo4976b712020-09-29 16:50:44 -07009157 dst_reg->type = SCALAR_VALUE;
Edward Creeb03c9f92017-08-07 15:26:36 +01009158 __mark_reg_known(&regs[insn->dst_reg], imm);
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009159 return 0;
Jakub Kicinski6b173872016-09-21 11:43:59 +01009160 }
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009161
Hao Luo4976b712020-09-29 16:50:44 -07009162 if (insn->src_reg == BPF_PSEUDO_BTF_ID) {
9163 mark_reg_known_zero(env, regs, insn->dst_reg);
9164
9165 dst_reg->type = aux->btf_var.reg_type;
9166 switch (dst_reg->type) {
9167 case PTR_TO_MEM:
9168 dst_reg->mem_size = aux->btf_var.mem_size;
9169 break;
9170 case PTR_TO_BTF_ID:
Hao Luoeaa6bcb2020-09-29 16:50:47 -07009171 case PTR_TO_PERCPU_BTF_ID:
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -08009172 dst_reg->btf = aux->btf_var.btf;
Hao Luo4976b712020-09-29 16:50:44 -07009173 dst_reg->btf_id = aux->btf_var.btf_id;
9174 break;
9175 default:
9176 verbose(env, "bpf verifier is misconfigured\n");
9177 return -EFAULT;
9178 }
9179 return 0;
9180 }
9181
Yonghong Song69c087b2021-02-26 12:49:25 -08009182 if (insn->src_reg == BPF_PSEUDO_FUNC) {
9183 struct bpf_prog_aux *aux = env->prog->aux;
9184 u32 subprogno = insn[1].imm;
9185
9186 if (!aux->func_info) {
9187 verbose(env, "missing btf func_info\n");
9188 return -EINVAL;
9189 }
9190 if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) {
9191 verbose(env, "callback function not static\n");
9192 return -EINVAL;
9193 }
9194
9195 dst_reg->type = PTR_TO_FUNC;
9196 dst_reg->subprogno = subprogno;
9197 return 0;
9198 }
9199
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +02009200 map = env->used_maps[aux->map_index];
9201 mark_reg_known_zero(env, regs, insn->dst_reg);
Hao Luo4976b712020-09-29 16:50:44 -07009202 dst_reg->map_ptr = map;
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009203
Alexei Starovoitov387544b2021-05-13 17:36:10 -07009204 if (insn->src_reg == BPF_PSEUDO_MAP_VALUE ||
9205 insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE) {
Hao Luo4976b712020-09-29 16:50:44 -07009206 dst_reg->type = PTR_TO_MAP_VALUE;
9207 dst_reg->off = aux->map_off;
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +02009208 if (map_value_has_spin_lock(map))
Hao Luo4976b712020-09-29 16:50:44 -07009209 dst_reg->id = ++env->id_gen;
Alexei Starovoitov387544b2021-05-13 17:36:10 -07009210 } else if (insn->src_reg == BPF_PSEUDO_MAP_FD ||
9211 insn->src_reg == BPF_PSEUDO_MAP_IDX) {
Hao Luo4976b712020-09-29 16:50:44 -07009212 dst_reg->type = CONST_PTR_TO_MAP;
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +02009213 } else {
9214 verbose(env, "bpf verifier is misconfigured\n");
9215 return -EINVAL;
9216 }
9217
Alexei Starovoitov17a52672014-09-26 00:17:06 -07009218 return 0;
9219}
9220
Daniel Borkmann96be4322015-03-01 12:31:46 +01009221static bool may_access_skb(enum bpf_prog_type type)
9222{
9223 switch (type) {
9224 case BPF_PROG_TYPE_SOCKET_FILTER:
9225 case BPF_PROG_TYPE_SCHED_CLS:
Daniel Borkmann94caee8c2015-03-20 15:11:11 +01009226 case BPF_PROG_TYPE_SCHED_ACT:
Daniel Borkmann96be4322015-03-01 12:31:46 +01009227 return true;
9228 default:
9229 return false;
9230 }
9231}
9232
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009233/* verify safety of LD_ABS|LD_IND instructions:
9234 * - they can only appear in the programs where ctx == skb
9235 * - since they are wrappers of function calls, they scratch R1-R5 registers,
9236 * preserve R6-R9, and store return value into R0
9237 *
9238 * Implicit input:
9239 * ctx == skb == R6 == CTX
9240 *
9241 * Explicit input:
9242 * SRC == any register
9243 * IMM == 32-bit immediate
9244 *
9245 * Output:
9246 * R0 - 8/16/32-bit skb data converted to cpu endianness
9247 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01009248static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009249{
Alexei Starovoitov638f5b92017-10-31 18:16:05 -07009250 struct bpf_reg_state *regs = cur_regs(env);
Daniel Borkmann6d4f1512020-01-06 22:51:57 +01009251 static const int ctx_reg = BPF_REG_6;
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009252 u8 mode = BPF_MODE(insn->code);
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009253 int i, err;
9254
Udip Pant7e407812020-08-25 16:20:00 -07009255 if (!may_access_skb(resolve_prog_type(env->prog))) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009256 verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009257 return -EINVAL;
9258 }
9259
Daniel Borkmanne0cea7c2018-05-04 01:08:14 +02009260 if (!env->ops->gen_ld_abs) {
9261 verbose(env, "bpf verifier is misconfigured\n");
9262 return -EINVAL;
9263 }
9264
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009265 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
Alexei Starovoitovd82bccc2016-04-12 10:26:19 -07009266 BPF_SIZE(insn->code) == BPF_DW ||
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009267 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009268 verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n");
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009269 return -EINVAL;
9270 }
9271
9272 /* check whether implicit source operand (register R6) is readable */
Daniel Borkmann6d4f1512020-01-06 22:51:57 +01009273 err = check_reg_arg(env, ctx_reg, SRC_OP);
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009274 if (err)
9275 return err;
9276
Joe Stringerfd978bf72018-10-02 13:35:35 -07009277 /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as
9278 * gen_ld_abs() may terminate the program at runtime, leading to
9279 * reference leak.
9280 */
9281 err = check_reference_leak(env);
9282 if (err) {
9283 verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n");
9284 return err;
9285 }
9286
Alexei Starovoitovd83525c2019-01-31 15:40:04 -08009287 if (env->cur_state->active_spin_lock) {
9288 verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n");
9289 return -EINVAL;
9290 }
9291
Daniel Borkmann6d4f1512020-01-06 22:51:57 +01009292 if (regs[ctx_reg].type != PTR_TO_CTX) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009293 verbose(env,
9294 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009295 return -EINVAL;
9296 }
9297
9298 if (mode == BPF_IND) {
9299 /* check explicit source operand */
Edward Creedc503a82017-08-15 20:34:35 +01009300 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009301 if (err)
9302 return err;
9303 }
9304
Daniel Borkmann6d4f1512020-01-06 22:51:57 +01009305 err = check_ctx_reg(env, &regs[ctx_reg], ctx_reg);
9306 if (err < 0)
9307 return err;
9308
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009309 /* reset caller saved regs to unreadable */
Edward Creedc503a82017-08-15 20:34:35 +01009310 for (i = 0; i < CALLER_SAVED_REGS; i++) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009311 mark_reg_not_init(env, regs, caller_saved[i]);
Edward Creedc503a82017-08-15 20:34:35 +01009312 check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
9313 }
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009314
9315 /* mark destination R0 register as readable, since it contains
Edward Creedc503a82017-08-15 20:34:35 +01009316 * the value fetched from the packet.
9317 * Already marked as written above.
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009318 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009319 mark_reg_unknown(env, regs, BPF_REG_0);
Jiong Wang5327ed32019-05-24 23:25:12 +01009320 /* ld_abs load up to 32-bit skb data. */
9321 regs[BPF_REG_0].subreg_def = env->insn_idx + 1;
Alexei Starovoitovddd872b2014-12-01 15:06:34 -08009322 return 0;
9323}
9324
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009325static int check_return_code(struct bpf_verifier_env *env)
9326{
brakmo5cf1e912019-05-28 16:59:36 -07009327 struct tnum enforce_attach_type_range = tnum_unknown;
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08009328 const struct bpf_prog *prog = env->prog;
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009329 struct bpf_reg_state *reg;
9330 struct tnum range = tnum_range(0, 1);
Udip Pant7e407812020-08-25 16:20:00 -07009331 enum bpf_prog_type prog_type = resolve_prog_type(env->prog);
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08009332 int err;
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07009333 struct bpf_func_state *frame = env->cur_state->frame[0];
9334 const bool is_subprog = frame->subprogno;
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08009335
KP Singh9e4e01d2020-03-29 01:43:52 +01009336 /* LSM and struct_ops func-ptr's return type could be "void" */
Dmitrii Banshchikovf782e2c2020-11-13 17:17:56 +00009337 if (!is_subprog &&
9338 (prog_type == BPF_PROG_TYPE_STRUCT_OPS ||
Udip Pant7e407812020-08-25 16:20:00 -07009339 prog_type == BPF_PROG_TYPE_LSM) &&
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08009340 !prog->aux->attach_func_proto->type)
9341 return 0;
9342
Zhen Lei8fb33b62021-05-25 10:56:59 +08009343 /* eBPF calling convention is such that R0 is used
Martin KaFai Lau27ae79972020-01-08 16:35:03 -08009344 * to return the value from eBPF program.
9345 * Make sure that it's readable at this time
9346 * of bpf_exit, which means that program wrote
9347 * something into it earlier
9348 */
9349 err = check_reg_arg(env, BPF_REG_0, SRC_OP);
9350 if (err)
9351 return err;
9352
9353 if (is_pointer_value(env, BPF_REG_0)) {
9354 verbose(env, "R0 leaks addr as return value\n");
9355 return -EACCES;
9356 }
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009357
Dmitrii Banshchikovf782e2c2020-11-13 17:17:56 +00009358 reg = cur_regs(env) + BPF_REG_0;
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07009359
9360 if (frame->in_async_callback_fn) {
9361 /* enforce return zero from async callbacks like timer */
9362 if (reg->type != SCALAR_VALUE) {
9363 verbose(env, "In async callback the register R0 is not a known value (%s)\n",
9364 reg_type_str[reg->type]);
9365 return -EINVAL;
9366 }
9367
9368 if (!tnum_in(tnum_const(0), reg->var_off)) {
9369 verbose_invalid_scalar(env, reg, &range, "async callback", "R0");
9370 return -EINVAL;
9371 }
9372 return 0;
9373 }
9374
Dmitrii Banshchikovf782e2c2020-11-13 17:17:56 +00009375 if (is_subprog) {
9376 if (reg->type != SCALAR_VALUE) {
9377 verbose(env, "At subprogram exit the register R0 is not a scalar value (%s)\n",
9378 reg_type_str[reg->type]);
9379 return -EINVAL;
9380 }
9381 return 0;
9382 }
9383
Udip Pant7e407812020-08-25 16:20:00 -07009384 switch (prog_type) {
Daniel Borkmann983695f2019-06-07 01:48:57 +02009385 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
9386 if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG ||
Daniel Borkmann1b66d252020-05-19 00:45:45 +02009387 env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG ||
9388 env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME ||
9389 env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME ||
9390 env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME ||
9391 env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME)
Daniel Borkmann983695f2019-06-07 01:48:57 +02009392 range = tnum_range(1, 1);
Stanislav Fomichev77241212021-01-27 11:31:39 -08009393 if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND ||
9394 env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND)
9395 range = tnum_range(0, 3);
Gustavo A. R. Silvaed4ed402019-07-11 11:22:33 -05009396 break;
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009397 case BPF_PROG_TYPE_CGROUP_SKB:
brakmo5cf1e912019-05-28 16:59:36 -07009398 if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) {
9399 range = tnum_range(0, 3);
9400 enforce_attach_type_range = tnum_range(2, 3);
9401 }
Gustavo A. R. Silvaed4ed402019-07-11 11:22:33 -05009402 break;
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009403 case BPF_PROG_TYPE_CGROUP_SOCK:
9404 case BPF_PROG_TYPE_SOCK_OPS:
Roman Gushchinebc614f2017-11-05 08:15:32 -05009405 case BPF_PROG_TYPE_CGROUP_DEVICE:
Andrey Ignatov7b146ce2019-02-27 12:59:24 -08009406 case BPF_PROG_TYPE_CGROUP_SYSCTL:
Stanislav Fomichev0d01da62019-06-27 13:38:47 -07009407 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009408 break;
Alexei Starovoitov15ab09b2019-10-28 20:24:26 -07009409 case BPF_PROG_TYPE_RAW_TRACEPOINT:
9410 if (!env->prog->aux->attach_btf_id)
9411 return 0;
9412 range = tnum_const(0);
9413 break;
Yonghong Song15d83c42020-05-09 10:59:00 -07009414 case BPF_PROG_TYPE_TRACING:
Yonghong Songe92888c72020-05-13 22:32:05 -07009415 switch (env->prog->expected_attach_type) {
9416 case BPF_TRACE_FENTRY:
9417 case BPF_TRACE_FEXIT:
9418 range = tnum_const(0);
9419 break;
9420 case BPF_TRACE_RAW_TP:
9421 case BPF_MODIFY_RETURN:
Yonghong Song15d83c42020-05-09 10:59:00 -07009422 return 0;
Daniel Borkmann2ec06162020-05-16 00:39:18 +02009423 case BPF_TRACE_ITER:
9424 break;
Yonghong Songe92888c72020-05-13 22:32:05 -07009425 default:
9426 return -ENOTSUPP;
9427 }
Yonghong Song15d83c42020-05-09 10:59:00 -07009428 break;
Jakub Sitnickie9ddbb72020-07-17 12:35:23 +02009429 case BPF_PROG_TYPE_SK_LOOKUP:
9430 range = tnum_range(SK_DROP, SK_PASS);
9431 break;
Yonghong Songe92888c72020-05-13 22:32:05 -07009432 case BPF_PROG_TYPE_EXT:
9433 /* freplace program can return anything as its return value
9434 * depends on the to-be-replaced kernel func or bpf program.
9435 */
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009436 default:
9437 return 0;
9438 }
9439
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009440 if (reg->type != SCALAR_VALUE) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009441 verbose(env, "At program exit the register R0 is not a known value (%s)\n",
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009442 reg_type_str[reg->type]);
9443 return -EINVAL;
9444 }
9445
9446 if (!tnum_in(range, reg->var_off)) {
Yonghong Songbc2591d2021-02-26 12:49:22 -08009447 verbose_invalid_scalar(env, reg, &range, "program exit", "R0");
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009448 return -EINVAL;
9449 }
brakmo5cf1e912019-05-28 16:59:36 -07009450
9451 if (!tnum_is_unknown(enforce_attach_type_range) &&
9452 tnum_in(enforce_attach_type_range, reg->var_off))
9453 env->prog->enforce_expected_attach_type = 1;
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -07009454 return 0;
9455}
9456
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009457/* non-recursive DFS pseudo code
9458 * 1 procedure DFS-iterative(G,v):
9459 * 2 label v as discovered
9460 * 3 let S be a stack
9461 * 4 S.push(v)
9462 * 5 while S is not empty
9463 * 6 t <- S.pop()
9464 * 7 if t is what we're looking for:
9465 * 8 return t
9466 * 9 for all edges e in G.adjacentEdges(t) do
9467 * 10 if edge e is already labelled
9468 * 11 continue with the next edge
9469 * 12 w <- G.adjacentVertex(t,e)
9470 * 13 if vertex w is not discovered and not explored
9471 * 14 label e as tree-edge
9472 * 15 label w as discovered
9473 * 16 S.push(w)
9474 * 17 continue at 5
9475 * 18 else if vertex w is discovered
9476 * 19 label e as back-edge
9477 * 20 else
9478 * 21 // vertex w is explored
9479 * 22 label e as forward- or cross-edge
9480 * 23 label t as explored
9481 * 24 S.pop()
9482 *
9483 * convention:
9484 * 0x10 - discovered
9485 * 0x11 - discovered and fall-through edge labelled
9486 * 0x12 - discovered and fall-through and branch edges labelled
9487 * 0x20 - explored
9488 */
9489
9490enum {
9491 DISCOVERED = 0x10,
9492 EXPLORED = 0x20,
9493 FALLTHROUGH = 1,
9494 BRANCH = 2,
9495};
9496
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -07009497static u32 state_htab_size(struct bpf_verifier_env *env)
9498{
9499 return env->prog->len;
9500}
9501
Alexei Starovoitov5d839022019-05-21 20:17:05 -07009502static struct bpf_verifier_state_list **explored_state(
9503 struct bpf_verifier_env *env,
9504 int idx)
9505{
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -07009506 struct bpf_verifier_state *cur = env->cur_state;
9507 struct bpf_func_state *state = cur->frame[cur->curframe];
9508
9509 return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)];
Alexei Starovoitov5d839022019-05-21 20:17:05 -07009510}
9511
9512static void init_explored_state(struct bpf_verifier_env *env, int idx)
9513{
Alexei Starovoitova8f500a2019-05-21 20:17:06 -07009514 env->insn_aux_data[idx].prune_point = true;
Alexei Starovoitov5d839022019-05-21 20:17:05 -07009515}
Alexei Starovoitovf1bca822014-09-29 18:50:01 -07009516
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009517enum {
9518 DONE_EXPLORING = 0,
9519 KEEP_EXPLORING = 1,
9520};
9521
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009522/* t, w, e - match pseudo-code above:
9523 * t - index of current instruction
9524 * w - next instruction
9525 * e - edge
9526 */
Alexei Starovoitov25897262019-06-15 12:12:20 -07009527static int push_insn(int t, int w, int e, struct bpf_verifier_env *env,
9528 bool loop_ok)
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009529{
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009530 int *insn_stack = env->cfg.insn_stack;
9531 int *insn_state = env->cfg.insn_state;
9532
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009533 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009534 return DONE_EXPLORING;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009535
9536 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009537 return DONE_EXPLORING;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009538
9539 if (w < 0 || w >= env->prog->len) {
Martin KaFai Laud9762e82018-12-13 10:41:48 -08009540 verbose_linfo(env, t, "%d: ", t);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009541 verbose(env, "jump out of range from insn %d to %d\n", t, w);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009542 return -EINVAL;
9543 }
9544
Alexei Starovoitovf1bca822014-09-29 18:50:01 -07009545 if (e == BRANCH)
9546 /* mark branch target for state pruning */
Alexei Starovoitov5d839022019-05-21 20:17:05 -07009547 init_explored_state(env, w);
Alexei Starovoitovf1bca822014-09-29 18:50:01 -07009548
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009549 if (insn_state[w] == 0) {
9550 /* tree-edge */
9551 insn_state[t] = DISCOVERED | e;
9552 insn_state[w] = DISCOVERED;
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009553 if (env->cfg.cur_stack >= env->prog->len)
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009554 return -E2BIG;
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009555 insn_stack[env->cfg.cur_stack++] = w;
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009556 return KEEP_EXPLORING;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009557 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -07009558 if (loop_ok && env->bpf_capable)
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009559 return DONE_EXPLORING;
Martin KaFai Laud9762e82018-12-13 10:41:48 -08009560 verbose_linfo(env, t, "%d: ", t);
9561 verbose_linfo(env, w, "%d: ", w);
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009562 verbose(env, "back-edge from insn %d to %d\n", t, w);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009563 return -EINVAL;
9564 } else if (insn_state[w] == EXPLORED) {
9565 /* forward- or cross-edge */
9566 insn_state[t] = DISCOVERED | e;
9567 } else {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009568 verbose(env, "insn state internal bug\n");
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009569 return -EFAULT;
9570 }
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009571 return DONE_EXPLORING;
9572}
9573
Yonghong Songefdb22d2021-02-26 12:49:20 -08009574static int visit_func_call_insn(int t, int insn_cnt,
9575 struct bpf_insn *insns,
9576 struct bpf_verifier_env *env,
9577 bool visit_callee)
9578{
9579 int ret;
9580
9581 ret = push_insn(t, t + 1, FALLTHROUGH, env, false);
9582 if (ret)
9583 return ret;
9584
9585 if (t + 1 < insn_cnt)
9586 init_explored_state(env, t + 1);
9587 if (visit_callee) {
9588 init_explored_state(env, t);
Alexei Starovoitov86fc6ee2021-07-14 17:54:13 -07009589 ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env,
9590 /* It's ok to allow recursion from CFG point of
9591 * view. __check_func_call() will do the actual
9592 * check.
9593 */
9594 bpf_pseudo_func(insns + t));
Yonghong Songefdb22d2021-02-26 12:49:20 -08009595 }
9596 return ret;
9597}
9598
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009599/* Visits the instruction at index t and returns one of the following:
9600 * < 0 - an error occurred
9601 * DONE_EXPLORING - the instruction was fully explored
9602 * KEEP_EXPLORING - there is still work to be done before it is fully explored
9603 */
9604static int visit_insn(int t, int insn_cnt, struct bpf_verifier_env *env)
9605{
9606 struct bpf_insn *insns = env->prog->insnsi;
9607 int ret;
9608
Yonghong Song69c087b2021-02-26 12:49:25 -08009609 if (bpf_pseudo_func(insns + t))
9610 return visit_func_call_insn(t, insn_cnt, insns, env, true);
9611
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009612 /* All non-branch instructions have a single fall-through edge. */
9613 if (BPF_CLASS(insns[t].code) != BPF_JMP &&
9614 BPF_CLASS(insns[t].code) != BPF_JMP32)
9615 return push_insn(t, t + 1, FALLTHROUGH, env, false);
9616
9617 switch (BPF_OP(insns[t].code)) {
9618 case BPF_EXIT:
9619 return DONE_EXPLORING;
9620
9621 case BPF_CALL:
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -07009622 if (insns[t].imm == BPF_FUNC_timer_set_callback)
9623 /* Mark this call insn to trigger is_state_visited() check
9624 * before call itself is processed by __check_func_call().
9625 * Otherwise new async state will be pushed for further
9626 * exploration.
9627 */
9628 init_explored_state(env, t);
Yonghong Songefdb22d2021-02-26 12:49:20 -08009629 return visit_func_call_insn(t, insn_cnt, insns, env,
9630 insns[t].src_reg == BPF_PSEUDO_CALL);
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009631
9632 case BPF_JA:
9633 if (BPF_SRC(insns[t].code) != BPF_K)
9634 return -EINVAL;
9635
9636 /* unconditional jump with single edge */
9637 ret = push_insn(t, t + insns[t].off + 1, FALLTHROUGH, env,
9638 true);
9639 if (ret)
9640 return ret;
9641
9642 /* unconditional jmp is not a good pruning point,
9643 * but it's marked, since backtracking needs
9644 * to record jmp history in is_state_visited().
9645 */
9646 init_explored_state(env, t + insns[t].off + 1);
9647 /* tell verifier to check for equivalent states
9648 * after every call and jump
9649 */
9650 if (t + 1 < insn_cnt)
9651 init_explored_state(env, t + 1);
9652
9653 return ret;
9654
9655 default:
9656 /* conditional jump with two edges */
9657 init_explored_state(env, t);
9658 ret = push_insn(t, t + 1, FALLTHROUGH, env, true);
9659 if (ret)
9660 return ret;
9661
9662 return push_insn(t, t + insns[t].off + 1, BRANCH, env, true);
9663 }
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009664}
9665
9666/* non-recursive depth-first-search to detect loops in BPF program
9667 * loop == back-edge in directed graph
9668 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +01009669static int check_cfg(struct bpf_verifier_env *env)
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009670{
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009671 int insn_cnt = env->prog->len;
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009672 int *insn_stack, *insn_state;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009673 int ret = 0;
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009674 int i;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009675
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009676 insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009677 if (!insn_state)
9678 return -ENOMEM;
9679
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009680 insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009681 if (!insn_stack) {
Alexei Starovoitov71dde682019-04-01 21:27:43 -07009682 kvfree(insn_state);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009683 return -ENOMEM;
9684 }
9685
9686 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
9687 insn_stack[0] = 0; /* 0 is the first instruction */
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009688 env->cfg.cur_stack = 1;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009689
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009690 while (env->cfg.cur_stack > 0) {
9691 int t = insn_stack[env->cfg.cur_stack - 1];
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009692
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009693 ret = visit_insn(t, insn_cnt, env);
9694 switch (ret) {
9695 case DONE_EXPLORING:
9696 insn_state[t] = EXPLORED;
9697 env->cfg.cur_stack--;
9698 break;
9699 case KEEP_EXPLORING:
9700 break;
9701 default:
9702 if (ret > 0) {
9703 verbose(env, "visit_insn internal bug\n");
9704 ret = -EFAULT;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -08009705 }
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009706 goto err_free;
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009707 }
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009708 }
9709
Wedson Almeida Filho59e2e272020-11-21 01:55:09 +00009710 if (env->cfg.cur_stack < 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009711 verbose(env, "pop stack internal bug\n");
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009712 ret = -EFAULT;
9713 goto err_free;
9714 }
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009715
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009716 for (i = 0; i < insn_cnt; i++) {
9717 if (insn_state[i] != EXPLORED) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -07009718 verbose(env, "unreachable insn %d\n", i);
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009719 ret = -EINVAL;
9720 goto err_free;
9721 }
9722 }
9723 ret = 0; /* cfg looks good */
9724
9725err_free:
Alexei Starovoitov71dde682019-04-01 21:27:43 -07009726 kvfree(insn_state);
9727 kvfree(insn_stack);
Alexei Starovoitov7df737e2019-04-19 07:44:54 -07009728 env->cfg.insn_state = env->cfg.insn_stack = NULL;
Alexei Starovoitov475fb782014-09-26 00:17:05 -07009729 return ret;
9730}
9731
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009732static int check_abnormal_return(struct bpf_verifier_env *env)
9733{
9734 int i;
9735
9736 for (i = 1; i < env->subprog_cnt; i++) {
9737 if (env->subprog_info[i].has_ld_abs) {
9738 verbose(env, "LD_ABS is not allowed in subprogs without BTF\n");
9739 return -EINVAL;
9740 }
9741 if (env->subprog_info[i].has_tail_call) {
9742 verbose(env, "tail_call is not allowed in subprogs without BTF\n");
9743 return -EINVAL;
9744 }
9745 }
9746 return 0;
9747}
9748
Yonghong Song838e9692018-11-19 15:29:11 -08009749/* The minimum supported BTF func info size */
9750#define MIN_BPF_FUNCINFO_SIZE 8
9751#define MAX_FUNCINFO_REC_SIZE 252
9752
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009753static int check_btf_func(struct bpf_verifier_env *env,
9754 const union bpf_attr *attr,
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009755 bpfptr_t uattr)
Yonghong Song838e9692018-11-19 15:29:11 -08009756{
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009757 const struct btf_type *type, *func_proto, *ret_type;
Peter Oskolkovd0b28182019-01-16 10:43:01 -08009758 u32 i, nfuncs, urec_size, min_size;
Yonghong Song838e9692018-11-19 15:29:11 -08009759 u32 krec_size = sizeof(struct bpf_func_info);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009760 struct bpf_func_info *krecord;
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009761 struct bpf_func_info_aux *info_aux = NULL;
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009762 struct bpf_prog *prog;
9763 const struct btf *btf;
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009764 bpfptr_t urecord;
Peter Oskolkovd0b28182019-01-16 10:43:01 -08009765 u32 prev_offset = 0;
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009766 bool scalar_return;
Dan Carpentere7ed83d2020-06-04 11:54:36 +03009767 int ret = -ENOMEM;
Yonghong Song838e9692018-11-19 15:29:11 -08009768
9769 nfuncs = attr->func_info_cnt;
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009770 if (!nfuncs) {
9771 if (check_abnormal_return(env))
9772 return -EINVAL;
Yonghong Song838e9692018-11-19 15:29:11 -08009773 return 0;
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009774 }
Yonghong Song838e9692018-11-19 15:29:11 -08009775
9776 if (nfuncs != env->subprog_cnt) {
9777 verbose(env, "number of funcs in func_info doesn't match number of subprogs\n");
9778 return -EINVAL;
9779 }
9780
9781 urec_size = attr->func_info_rec_size;
9782 if (urec_size < MIN_BPF_FUNCINFO_SIZE ||
9783 urec_size > MAX_FUNCINFO_REC_SIZE ||
9784 urec_size % sizeof(u32)) {
9785 verbose(env, "invalid func info rec size %u\n", urec_size);
9786 return -EINVAL;
9787 }
9788
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009789 prog = env->prog;
9790 btf = prog->aux->btf;
Yonghong Song838e9692018-11-19 15:29:11 -08009791
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009792 urecord = make_bpfptr(attr->func_info, uattr.is_kernel);
Yonghong Song838e9692018-11-19 15:29:11 -08009793 min_size = min_t(u32, krec_size, urec_size);
9794
Yonghong Songba64e7d2018-11-24 23:20:44 -08009795 krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009796 if (!krecord)
9797 return -ENOMEM;
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009798 info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN);
9799 if (!info_aux)
9800 goto err_free;
Yonghong Songba64e7d2018-11-24 23:20:44 -08009801
Yonghong Song838e9692018-11-19 15:29:11 -08009802 for (i = 0; i < nfuncs; i++) {
9803 ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size);
9804 if (ret) {
9805 if (ret == -E2BIG) {
9806 verbose(env, "nonzero tailing record in func info");
9807 /* set the size kernel expects so loader can zero
9808 * out the rest of the record.
9809 */
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009810 if (copy_to_bpfptr_offset(uattr,
9811 offsetof(union bpf_attr, func_info_rec_size),
9812 &min_size, sizeof(min_size)))
Yonghong Song838e9692018-11-19 15:29:11 -08009813 ret = -EFAULT;
9814 }
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009815 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009816 }
9817
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009818 if (copy_from_bpfptr(&krecord[i], urecord, min_size)) {
Yonghong Song838e9692018-11-19 15:29:11 -08009819 ret = -EFAULT;
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009820 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009821 }
9822
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009823 /* check insn_off */
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009824 ret = -EINVAL;
Yonghong Song838e9692018-11-19 15:29:11 -08009825 if (i == 0) {
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009826 if (krecord[i].insn_off) {
Yonghong Song838e9692018-11-19 15:29:11 -08009827 verbose(env,
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009828 "nonzero insn_off %u for the first func info record",
9829 krecord[i].insn_off);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009830 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009831 }
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009832 } else if (krecord[i].insn_off <= prev_offset) {
Yonghong Song838e9692018-11-19 15:29:11 -08009833 verbose(env,
9834 "same or smaller insn offset (%u) than previous func info record (%u)",
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009835 krecord[i].insn_off, prev_offset);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009836 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009837 }
9838
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009839 if (env->subprog_info[i].start != krecord[i].insn_off) {
Yonghong Song838e9692018-11-19 15:29:11 -08009840 verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n");
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009841 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009842 }
9843
9844 /* check type_id */
Yonghong Songba64e7d2018-11-24 23:20:44 -08009845 type = btf_type_by_id(btf, krecord[i].type_id);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08009846 if (!type || !btf_type_is_func(type)) {
Yonghong Song838e9692018-11-19 15:29:11 -08009847 verbose(env, "invalid type id %d in func info",
Yonghong Songba64e7d2018-11-24 23:20:44 -08009848 krecord[i].type_id);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009849 goto err_free;
Yonghong Song838e9692018-11-19 15:29:11 -08009850 }
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -08009851 info_aux[i].linkage = BTF_INFO_VLEN(type->info);
Alexei Starovoitov09b28d72020-09-17 19:09:18 -07009852
9853 func_proto = btf_type_by_id(btf, type->type);
9854 if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto)))
9855 /* btf_func_check() already verified it during BTF load */
9856 goto err_free;
9857 ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL);
9858 scalar_return =
9859 btf_type_is_small_int(ret_type) || btf_type_is_enum(ret_type);
9860 if (i && !scalar_return && env->subprog_info[i].has_ld_abs) {
9861 verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n");
9862 goto err_free;
9863 }
9864 if (i && !scalar_return && env->subprog_info[i].has_tail_call) {
9865 verbose(env, "tail_call is only allowed in functions that return 'int'.\n");
9866 goto err_free;
9867 }
9868
Martin KaFai Laud30d42e2018-12-05 17:35:44 -08009869 prev_offset = krecord[i].insn_off;
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009870 bpfptr_add(&urecord, urec_size);
Yonghong Song838e9692018-11-19 15:29:11 -08009871 }
9872
Yonghong Songba64e7d2018-11-24 23:20:44 -08009873 prog->aux->func_info = krecord;
9874 prog->aux->func_info_cnt = nfuncs;
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009875 prog->aux->func_info_aux = info_aux;
Yonghong Song838e9692018-11-19 15:29:11 -08009876 return 0;
9877
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009878err_free:
Yonghong Songba64e7d2018-11-24 23:20:44 -08009879 kvfree(krecord);
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009880 kfree(info_aux);
Yonghong Song838e9692018-11-19 15:29:11 -08009881 return ret;
9882}
9883
Yonghong Songba64e7d2018-11-24 23:20:44 -08009884static void adjust_btf_func(struct bpf_verifier_env *env)
9885{
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009886 struct bpf_prog_aux *aux = env->prog->aux;
Yonghong Songba64e7d2018-11-24 23:20:44 -08009887 int i;
9888
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009889 if (!aux->func_info)
Yonghong Songba64e7d2018-11-24 23:20:44 -08009890 return;
9891
9892 for (i = 0; i < env->subprog_cnt; i++)
Alexei Starovoitov8c1b6e62019-11-14 10:57:16 -08009893 aux->func_info[i].insn_off = env->subprog_info[i].start;
Yonghong Songba64e7d2018-11-24 23:20:44 -08009894}
9895
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009896#define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \
9897 sizeof(((struct bpf_line_info *)(0))->line_col))
9898#define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE
9899
9900static int check_btf_line(struct bpf_verifier_env *env,
9901 const union bpf_attr *attr,
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009902 bpfptr_t uattr)
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009903{
9904 u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0;
9905 struct bpf_subprog_info *sub;
9906 struct bpf_line_info *linfo;
9907 struct bpf_prog *prog;
9908 const struct btf *btf;
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009909 bpfptr_t ulinfo;
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009910 int err;
9911
9912 nr_linfo = attr->line_info_cnt;
9913 if (!nr_linfo)
9914 return 0;
Bixuan Cui0e6491b2021-09-11 08:55:57 +08009915 if (nr_linfo > INT_MAX / sizeof(struct bpf_line_info))
9916 return -EINVAL;
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009917
9918 rec_size = attr->line_info_rec_size;
9919 if (rec_size < MIN_BPF_LINEINFO_SIZE ||
9920 rec_size > MAX_LINEINFO_REC_SIZE ||
9921 rec_size & (sizeof(u32) - 1))
9922 return -EINVAL;
9923
9924 /* Need to zero it in case the userspace may
9925 * pass in a smaller bpf_line_info object.
9926 */
9927 linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info),
9928 GFP_KERNEL | __GFP_NOWARN);
9929 if (!linfo)
9930 return -ENOMEM;
9931
9932 prog = env->prog;
9933 btf = prog->aux->btf;
9934
9935 s = 0;
9936 sub = env->subprog_info;
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009937 ulinfo = make_bpfptr(attr->line_info, uattr.is_kernel);
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009938 expected_size = sizeof(struct bpf_line_info);
9939 ncopy = min_t(u32, expected_size, rec_size);
9940 for (i = 0; i < nr_linfo; i++) {
9941 err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size);
9942 if (err) {
9943 if (err == -E2BIG) {
9944 verbose(env, "nonzero tailing record in line_info");
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009945 if (copy_to_bpfptr_offset(uattr,
9946 offsetof(union bpf_attr, line_info_rec_size),
9947 &expected_size, sizeof(expected_size)))
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009948 err = -EFAULT;
9949 }
9950 goto err_free;
9951 }
9952
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -07009953 if (copy_from_bpfptr(&linfo[i], ulinfo, ncopy)) {
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009954 err = -EFAULT;
9955 goto err_free;
9956 }
9957
9958 /*
9959 * Check insn_off to ensure
9960 * 1) strictly increasing AND
9961 * 2) bounded by prog->len
9962 *
9963 * The linfo[0].insn_off == 0 check logically falls into
9964 * the later "missing bpf_line_info for func..." case
9965 * because the first linfo[0].insn_off must be the
9966 * first sub also and the first sub must have
9967 * subprog_info[0].start == 0.
9968 */
9969 if ((i && linfo[i].insn_off <= prev_offset) ||
9970 linfo[i].insn_off >= prog->len) {
9971 verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n",
9972 i, linfo[i].insn_off, prev_offset,
9973 prog->len);
9974 err = -EINVAL;
9975 goto err_free;
9976 }
9977
Martin KaFai Laufdbaa0b2018-12-19 13:01:01 -08009978 if (!prog->insnsi[linfo[i].insn_off].code) {
9979 verbose(env,
9980 "Invalid insn code at line_info[%u].insn_off\n",
9981 i);
9982 err = -EINVAL;
9983 goto err_free;
9984 }
9985
Martin KaFai Lau23127b32018-12-13 10:41:46 -08009986 if (!btf_name_by_offset(btf, linfo[i].line_off) ||
9987 !btf_name_by_offset(btf, linfo[i].file_name_off)) {
Martin KaFai Lauc454a462018-12-07 16:42:25 -08009988 verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i);
9989 err = -EINVAL;
9990 goto err_free;
9991 }
9992
9993 if (s != env->subprog_cnt) {
9994 if (linfo[i].insn_off == sub[s].start) {
9995 sub[s].linfo_idx = i;
9996 s++;
9997 } else if (sub[s].start < linfo[i].insn_off) {
9998 verbose(env, "missing bpf_line_info for func#%u\n", s);
9999 err = -EINVAL;
10000 goto err_free;
10001 }
10002 }
10003
10004 prev_offset = linfo[i].insn_off;
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -070010005 bpfptr_add(&ulinfo, rec_size);
Martin KaFai Lauc454a462018-12-07 16:42:25 -080010006 }
10007
10008 if (s != env->subprog_cnt) {
10009 verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n",
10010 env->subprog_cnt - s, s);
10011 err = -EINVAL;
10012 goto err_free;
10013 }
10014
10015 prog->aux->linfo = linfo;
10016 prog->aux->nr_linfo = nr_linfo;
10017
10018 return 0;
10019
10020err_free:
10021 kvfree(linfo);
10022 return err;
10023}
10024
10025static int check_btf_info(struct bpf_verifier_env *env,
10026 const union bpf_attr *attr,
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -070010027 bpfptr_t uattr)
Martin KaFai Lauc454a462018-12-07 16:42:25 -080010028{
10029 struct btf *btf;
10030 int err;
10031
Alexei Starovoitov09b28d72020-09-17 19:09:18 -070010032 if (!attr->func_info_cnt && !attr->line_info_cnt) {
10033 if (check_abnormal_return(env))
10034 return -EINVAL;
Martin KaFai Lauc454a462018-12-07 16:42:25 -080010035 return 0;
Alexei Starovoitov09b28d72020-09-17 19:09:18 -070010036 }
Martin KaFai Lauc454a462018-12-07 16:42:25 -080010037
10038 btf = btf_get_by_fd(attr->prog_btf_fd);
10039 if (IS_ERR(btf))
10040 return PTR_ERR(btf);
Alexei Starovoitov350a5c42021-03-07 14:52:48 -080010041 if (btf_is_kernel(btf)) {
10042 btf_put(btf);
10043 return -EACCES;
10044 }
Martin KaFai Lauc454a462018-12-07 16:42:25 -080010045 env->prog->aux->btf = btf;
10046
10047 err = check_btf_func(env, attr, uattr);
10048 if (err)
10049 return err;
10050
10051 err = check_btf_line(env, attr, uattr);
10052 if (err)
10053 return err;
10054
10055 return 0;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070010056}
10057
Edward Creef1174f72017-08-07 15:26:19 +010010058/* check %cur's range satisfies %old's */
10059static bool range_within(struct bpf_reg_state *old,
10060 struct bpf_reg_state *cur)
Alexei Starovoitov969bf052016-05-05 19:49:10 -070010061{
Edward Creeb03c9f92017-08-07 15:26:36 +010010062 return old->umin_value <= cur->umin_value &&
10063 old->umax_value >= cur->umax_value &&
10064 old->smin_value <= cur->smin_value &&
Daniel Borkmannfd675182021-02-05 20:48:21 +010010065 old->smax_value >= cur->smax_value &&
10066 old->u32_min_value <= cur->u32_min_value &&
10067 old->u32_max_value >= cur->u32_max_value &&
10068 old->s32_min_value <= cur->s32_min_value &&
10069 old->s32_max_value >= cur->s32_max_value;
Edward Creef1174f72017-08-07 15:26:19 +010010070}
10071
Edward Creef1174f72017-08-07 15:26:19 +010010072/* If in the old state two registers had the same id, then they need to have
10073 * the same id in the new state as well. But that id could be different from
10074 * the old state, so we need to track the mapping from old to new ids.
10075 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
10076 * regs with old id 5 must also have new id 9 for the new state to be safe. But
10077 * regs with a different old id could still have new id 9, we don't care about
10078 * that.
10079 * So we look through our idmap to see if this old id has been seen before. If
10080 * so, we require the new id to match; otherwise, we add the id pair to the map.
10081 */
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010082static bool check_ids(u32 old_id, u32 cur_id, struct bpf_id_pair *idmap)
Edward Creef1174f72017-08-07 15:26:19 +010010083{
10084 unsigned int i;
10085
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010086 for (i = 0; i < BPF_ID_MAP_SIZE; i++) {
Edward Creef1174f72017-08-07 15:26:19 +010010087 if (!idmap[i].old) {
10088 /* Reached an empty slot; haven't seen this id before */
10089 idmap[i].old = old_id;
10090 idmap[i].cur = cur_id;
10091 return true;
10092 }
10093 if (idmap[i].old == old_id)
10094 return idmap[i].cur == cur_id;
10095 }
10096 /* We ran out of idmap slots, which should be impossible */
10097 WARN_ON_ONCE(1);
10098 return false;
10099}
10100
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010101static void clean_func_state(struct bpf_verifier_env *env,
10102 struct bpf_func_state *st)
10103{
10104 enum bpf_reg_liveness live;
10105 int i, j;
10106
10107 for (i = 0; i < BPF_REG_FP; i++) {
10108 live = st->regs[i].live;
10109 /* liveness must not touch this register anymore */
10110 st->regs[i].live |= REG_LIVE_DONE;
10111 if (!(live & REG_LIVE_READ))
10112 /* since the register is unused, clear its state
10113 * to make further comparison simpler
10114 */
Daniel Borkmannf54c7892019-12-22 23:37:40 +010010115 __mark_reg_not_init(env, &st->regs[i]);
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010116 }
10117
10118 for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) {
10119 live = st->stack[i].spilled_ptr.live;
10120 /* liveness must not touch this stack slot anymore */
10121 st->stack[i].spilled_ptr.live |= REG_LIVE_DONE;
10122 if (!(live & REG_LIVE_READ)) {
Daniel Borkmannf54c7892019-12-22 23:37:40 +010010123 __mark_reg_not_init(env, &st->stack[i].spilled_ptr);
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010124 for (j = 0; j < BPF_REG_SIZE; j++)
10125 st->stack[i].slot_type[j] = STACK_INVALID;
10126 }
10127 }
10128}
10129
10130static void clean_verifier_state(struct bpf_verifier_env *env,
10131 struct bpf_verifier_state *st)
10132{
10133 int i;
10134
10135 if (st->frame[0]->regs[0].live & REG_LIVE_DONE)
10136 /* all regs in this state in all frames were already marked */
10137 return;
10138
10139 for (i = 0; i <= st->curframe; i++)
10140 clean_func_state(env, st->frame[i]);
10141}
10142
10143/* the parentage chains form a tree.
10144 * the verifier states are added to state lists at given insn and
10145 * pushed into state stack for future exploration.
10146 * when the verifier reaches bpf_exit insn some of the verifer states
10147 * stored in the state lists have their final liveness state already,
10148 * but a lot of states will get revised from liveness point of view when
10149 * the verifier explores other branches.
10150 * Example:
10151 * 1: r0 = 1
10152 * 2: if r1 == 100 goto pc+1
10153 * 3: r0 = 2
10154 * 4: exit
10155 * when the verifier reaches exit insn the register r0 in the state list of
10156 * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch
10157 * of insn 2 and goes exploring further. At the insn 4 it will walk the
10158 * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ.
10159 *
10160 * Since the verifier pushes the branch states as it sees them while exploring
10161 * the program the condition of walking the branch instruction for the second
10162 * time means that all states below this branch were already explored and
Zhen Lei8fb33b62021-05-25 10:56:59 +080010163 * their final liveness marks are already propagated.
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010164 * Hence when the verifier completes the search of state list in is_state_visited()
10165 * we can call this clean_live_states() function to mark all liveness states
10166 * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state'
10167 * will not be used.
10168 * This function also clears the registers and stack for states that !READ
10169 * to simplify state merging.
10170 *
10171 * Important note here that walking the same branch instruction in the callee
10172 * doesn't meant that the states are DONE. The verifier has to compare
10173 * the callsites
10174 */
10175static void clean_live_states(struct bpf_verifier_env *env, int insn,
10176 struct bpf_verifier_state *cur)
10177{
10178 struct bpf_verifier_state_list *sl;
10179 int i;
10180
Alexei Starovoitov5d839022019-05-21 20:17:05 -070010181 sl = *explored_state(env, insn);
Alexei Starovoitova8f500a2019-05-21 20:17:06 -070010182 while (sl) {
Alexei Starovoitov25897262019-06-15 12:12:20 -070010183 if (sl->state.branches)
10184 goto next;
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070010185 if (sl->state.insn_idx != insn ||
10186 sl->state.curframe != cur->curframe)
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010187 goto next;
10188 for (i = 0; i <= cur->curframe; i++)
10189 if (sl->state.frame[i]->callsite != cur->frame[i]->callsite)
10190 goto next;
10191 clean_verifier_state(env, &sl->state);
10192next:
10193 sl = sl->next;
10194 }
10195}
10196
Edward Creef1174f72017-08-07 15:26:19 +010010197/* Returns true if (rold safe implies rcur safe) */
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010198static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold,
10199 struct bpf_reg_state *rcur, struct bpf_id_pair *idmap)
Edward Creef1174f72017-08-07 15:26:19 +010010200{
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010201 bool equal;
10202
Edward Creedc503a82017-08-15 20:34:35 +010010203 if (!(rold->live & REG_LIVE_READ))
10204 /* explored state didn't use this */
10205 return true;
10206
Edward Cree679c7822018-08-22 20:02:19 +010010207 equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010208
10209 if (rold->type == PTR_TO_STACK)
10210 /* two stack pointers are equal only if they're pointing to
10211 * the same stack frame, since fp-8 in foo != fp-8 in bar
10212 */
10213 return equal && rold->frameno == rcur->frameno;
10214
10215 if (equal)
Edward Creef1174f72017-08-07 15:26:19 +010010216 return true;
10217
10218 if (rold->type == NOT_INIT)
10219 /* explored state can't have used this */
10220 return true;
10221 if (rcur->type == NOT_INIT)
Alexei Starovoitov969bf052016-05-05 19:49:10 -070010222 return false;
Edward Creef1174f72017-08-07 15:26:19 +010010223 switch (rold->type) {
10224 case SCALAR_VALUE:
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010225 if (env->explore_alu_limits)
10226 return false;
Edward Creef1174f72017-08-07 15:26:19 +010010227 if (rcur->type == SCALAR_VALUE) {
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010228 if (!rold->precise && !rcur->precise)
10229 return true;
Edward Creef1174f72017-08-07 15:26:19 +010010230 /* new val must satisfy old val knowledge */
10231 return range_within(rold, rcur) &&
10232 tnum_in(rold->var_off, rcur->var_off);
10233 } else {
Jann Horn179d1c52017-12-18 20:11:59 -080010234 /* We're trying to use a pointer in place of a scalar.
10235 * Even if the scalar was unbounded, this could lead to
10236 * pointer leaks because scalars are allowed to leak
10237 * while pointers are not. We could make this safe in
10238 * special cases if root is calling us, but it's
10239 * probably not worth the hassle.
Edward Creef1174f72017-08-07 15:26:19 +010010240 */
Jann Horn179d1c52017-12-18 20:11:59 -080010241 return false;
Edward Creef1174f72017-08-07 15:26:19 +010010242 }
Yonghong Song69c087b2021-02-26 12:49:25 -080010243 case PTR_TO_MAP_KEY:
Edward Creef1174f72017-08-07 15:26:19 +010010244 case PTR_TO_MAP_VALUE:
Edward Cree1b688a12017-08-23 15:10:50 +010010245 /* If the new min/max/var_off satisfy the old ones and
10246 * everything else matches, we are OK.
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080010247 * 'id' is not compared, since it's only used for maps with
10248 * bpf_spin_lock inside map element and in such cases if
10249 * the rest of the prog is valid for one map element then
10250 * it's valid for all map elements regardless of the key
10251 * used in bpf_map_lookup()
Edward Cree1b688a12017-08-23 15:10:50 +010010252 */
10253 return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 &&
10254 range_within(rold, rcur) &&
10255 tnum_in(rold->var_off, rcur->var_off);
Edward Creef1174f72017-08-07 15:26:19 +010010256 case PTR_TO_MAP_VALUE_OR_NULL:
10257 /* a PTR_TO_MAP_VALUE could be safe to use as a
10258 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
10259 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
10260 * checked, doing so could have affected others with the same
10261 * id, and we can't check for that because we lost the id when
10262 * we converted to a PTR_TO_MAP_VALUE.
10263 */
10264 if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL)
10265 return false;
10266 if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)))
10267 return false;
10268 /* Check our ids match any regs they're supposed to */
10269 return check_ids(rold->id, rcur->id, idmap);
Daniel Borkmannde8f3a82017-09-25 02:25:51 +020010270 case PTR_TO_PACKET_META:
Edward Creef1174f72017-08-07 15:26:19 +010010271 case PTR_TO_PACKET:
Daniel Borkmannde8f3a82017-09-25 02:25:51 +020010272 if (rcur->type != rold->type)
Edward Creef1174f72017-08-07 15:26:19 +010010273 return false;
10274 /* We must have at least as much range as the old ptr
10275 * did, so that any accesses which were safe before are
10276 * still safe. This is true even if old range < old off,
10277 * since someone could have accessed through (ptr - k), or
10278 * even done ptr -= k in a register, to get a safe access.
10279 */
10280 if (rold->range > rcur->range)
10281 return false;
10282 /* If the offsets don't match, we can't trust our alignment;
10283 * nor can we be sure that we won't fall out of range.
10284 */
10285 if (rold->off != rcur->off)
10286 return false;
10287 /* id relations must be preserved */
10288 if (rold->id && !check_ids(rold->id, rcur->id, idmap))
10289 return false;
10290 /* new val must satisfy old val knowledge */
10291 return range_within(rold, rcur) &&
10292 tnum_in(rold->var_off, rcur->var_off);
10293 case PTR_TO_CTX:
10294 case CONST_PTR_TO_MAP:
Edward Creef1174f72017-08-07 15:26:19 +010010295 case PTR_TO_PACKET_END:
Petar Penkovd58e4682018-09-14 07:46:18 -070010296 case PTR_TO_FLOW_KEYS:
Joe Stringerc64b7982018-10-02 13:35:33 -070010297 case PTR_TO_SOCKET:
10298 case PTR_TO_SOCKET_OR_NULL:
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -080010299 case PTR_TO_SOCK_COMMON:
10300 case PTR_TO_SOCK_COMMON_OR_NULL:
Martin KaFai Lau655a51e2019-02-09 23:22:24 -080010301 case PTR_TO_TCP_SOCK:
10302 case PTR_TO_TCP_SOCK_OR_NULL:
Jonathan Lemonfada7fd2019-06-06 13:59:40 -070010303 case PTR_TO_XDP_SOCK:
Edward Creef1174f72017-08-07 15:26:19 +010010304 /* Only valid matches are exact, which memcmp() above
10305 * would have accepted
10306 */
10307 default:
10308 /* Don't know what's going on, just say it's not safe */
10309 return false;
10310 }
Alexei Starovoitov969bf052016-05-05 19:49:10 -070010311
Edward Creef1174f72017-08-07 15:26:19 +010010312 /* Shouldn't get here; if we do, say it's not safe */
10313 WARN_ON_ONCE(1);
Alexei Starovoitov969bf052016-05-05 19:49:10 -070010314 return false;
10315}
10316
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010317static bool stacksafe(struct bpf_verifier_env *env, struct bpf_func_state *old,
10318 struct bpf_func_state *cur, struct bpf_id_pair *idmap)
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010319{
10320 int i, spi;
10321
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010322 /* walk slots of the explored stack and ignore any additional
10323 * slots in the current stack, since explored(safe) state
10324 * didn't use them
10325 */
10326 for (i = 0; i < old->allocated_stack; i++) {
10327 spi = i / BPF_REG_SIZE;
10328
Alexei Starovoitovb2339202018-12-13 11:42:31 -080010329 if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) {
10330 i += BPF_REG_SIZE - 1;
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -080010331 /* explored state didn't use this */
Gianluca Borellofd05e572017-12-23 10:09:55 +000010332 continue;
Alexei Starovoitovb2339202018-12-13 11:42:31 -080010333 }
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -080010334
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010335 if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID)
10336 continue;
Alexei Starovoitov19e2dbb2018-12-13 11:42:33 -080010337
10338 /* explored stack has more populated slots than current stack
10339 * and these slots were used
10340 */
10341 if (i >= cur->allocated_stack)
10342 return false;
10343
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -080010344 /* if old state was safe with misc data in the stack
10345 * it will be safe with zero-initialized stack.
10346 * The opposite is not true
10347 */
10348 if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC &&
10349 cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO)
10350 continue;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010351 if (old->stack[spi].slot_type[i % BPF_REG_SIZE] !=
10352 cur->stack[spi].slot_type[i % BPF_REG_SIZE])
10353 /* Ex: old explored (safe) state has STACK_SPILL in
Randy Dunlapb8c1a302020-08-06 20:31:41 -070010354 * this stack slot, but current has STACK_MISC ->
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010355 * this verifier states are not equivalent,
10356 * return false to continue verification of this path
10357 */
10358 return false;
10359 if (i % BPF_REG_SIZE)
10360 continue;
10361 if (old->stack[spi].slot_type[0] != STACK_SPILL)
10362 continue;
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010363 if (!regsafe(env, &old->stack[spi].spilled_ptr,
10364 &cur->stack[spi].spilled_ptr, idmap))
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010365 /* when explored and current stack slot are both storing
10366 * spilled registers, check that stored pointers types
10367 * are the same as well.
10368 * Ex: explored safe path could have stored
10369 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
10370 * but current path has stored:
10371 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
10372 * such verifier states are not equivalent.
10373 * return false to continue verification of this path
10374 */
10375 return false;
10376 }
10377 return true;
10378}
10379
Joe Stringerfd978bf72018-10-02 13:35:35 -070010380static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur)
10381{
10382 if (old->acquired_refs != cur->acquired_refs)
10383 return false;
10384 return !memcmp(old->refs, cur->refs,
10385 sizeof(*old->refs) * old->acquired_refs);
10386}
10387
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010388/* compare two verifier states
10389 *
10390 * all states stored in state_list are known to be valid, since
10391 * verifier reached 'bpf_exit' instruction through them
10392 *
10393 * this function is called when verifier exploring different branches of
10394 * execution popped from the state stack. If it sees an old state that has
10395 * more strict register state and more strict stack state then this execution
10396 * branch doesn't need to be explored further, since verifier already
10397 * concluded that more strict state leads to valid finish.
10398 *
10399 * Therefore two states are equivalent if register state is more conservative
10400 * and explored stack state is more conservative than the current one.
10401 * Example:
10402 * explored current
10403 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
10404 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
10405 *
10406 * In other words if current stack state (one being explored) has more
10407 * valid slots than old one that already passed validation, it means
10408 * the verifier can stop exploring and conclude that current state is valid too
10409 *
10410 * Similarly with registers. If explored state has register type as invalid
10411 * whereas register type in current state is meaningful, it means that
10412 * the current state will reach 'bpf_exit' instruction safely
10413 */
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010414static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_state *old,
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010415 struct bpf_func_state *cur)
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010416{
10417 int i;
10418
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010419 memset(env->idmap_scratch, 0, sizeof(env->idmap_scratch));
10420 for (i = 0; i < MAX_BPF_REG; i++)
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010421 if (!regsafe(env, &old->regs[i], &cur->regs[i],
10422 env->idmap_scratch))
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010423 return false;
10424
Daniel Borkmanne042aa52021-07-16 09:18:21 +000010425 if (!stacksafe(env, old, cur, env->idmap_scratch))
Alexei Starovoitov1a0dc1a2016-05-05 19:49:09 -070010426 return false;
Edward Creef1174f72017-08-07 15:26:19 +010010427
Joe Stringerfd978bf72018-10-02 13:35:35 -070010428 if (!refsafe(old, cur))
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010429 return false;
10430
10431 return true;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010432}
10433
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010434static bool states_equal(struct bpf_verifier_env *env,
10435 struct bpf_verifier_state *old,
10436 struct bpf_verifier_state *cur)
Edward Creedc503a82017-08-15 20:34:35 +010010437{
Edward Creedc503a82017-08-15 20:34:35 +010010438 int i;
10439
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010440 if (old->curframe != cur->curframe)
10441 return false;
10442
Daniel Borkmann979d63d2019-01-03 00:58:34 +010010443 /* Verification state from speculative execution simulation
10444 * must never prune a non-speculative execution one.
10445 */
10446 if (old->speculative && !cur->speculative)
10447 return false;
10448
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080010449 if (old->active_spin_lock != cur->active_spin_lock)
10450 return false;
10451
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010452 /* for states to be equal callsites have to be the same
10453 * and all frame states need to be equivalent
10454 */
10455 for (i = 0; i <= old->curframe; i++) {
10456 if (old->frame[i]->callsite != cur->frame[i]->callsite)
10457 return false;
Lorenz Bauerc9e73e32021-04-29 14:46:56 +010010458 if (!func_states_equal(env, old->frame[i], cur->frame[i]))
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010459 return false;
10460 }
10461 return true;
10462}
10463
Jiong Wang5327ed32019-05-24 23:25:12 +010010464/* Return 0 if no propagation happened. Return negative error code if error
10465 * happened. Otherwise, return the propagated bit.
10466 */
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010467static int propagate_liveness_reg(struct bpf_verifier_env *env,
10468 struct bpf_reg_state *reg,
10469 struct bpf_reg_state *parent_reg)
10470{
Jiong Wang5327ed32019-05-24 23:25:12 +010010471 u8 parent_flag = parent_reg->live & REG_LIVE_READ;
10472 u8 flag = reg->live & REG_LIVE_READ;
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010473 int err;
10474
Jiong Wang5327ed32019-05-24 23:25:12 +010010475 /* When comes here, read flags of PARENT_REG or REG could be any of
10476 * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need
10477 * of propagation if PARENT_REG has strongest REG_LIVE_READ64.
10478 */
10479 if (parent_flag == REG_LIVE_READ64 ||
10480 /* Or if there is no read flag from REG. */
10481 !flag ||
10482 /* Or if the read flag from REG is the same as PARENT_REG. */
10483 parent_flag == flag)
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010484 return 0;
10485
Jiong Wang5327ed32019-05-24 23:25:12 +010010486 err = mark_reg_read(env, reg, parent_reg, flag);
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010487 if (err)
10488 return err;
10489
Jiong Wang5327ed32019-05-24 23:25:12 +010010490 return flag;
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010491}
10492
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010493/* A write screens off any subsequent reads; but write marks come from the
10494 * straight-line code between a state and its parent. When we arrive at an
10495 * equivalent state (jump target or such) we didn't arrive by the straight-line
10496 * code, so read marks in the state must propagate to the parent regardless
10497 * of the state's write marks. That's what 'parent == state->parent' comparison
Edward Cree679c7822018-08-22 20:02:19 +010010498 * in mark_reg_read() is for.
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010499 */
10500static int propagate_liveness(struct bpf_verifier_env *env,
10501 const struct bpf_verifier_state *vstate,
10502 struct bpf_verifier_state *vparent)
10503{
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010504 struct bpf_reg_state *state_reg, *parent_reg;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010505 struct bpf_func_state *state, *parent;
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010506 int i, frame, err = 0;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010507
10508 if (vparent->curframe != vstate->curframe) {
10509 WARN(1, "propagate_live: parent frame %d current frame %d\n",
10510 vparent->curframe, vstate->curframe);
10511 return -EFAULT;
10512 }
Edward Creedc503a82017-08-15 20:34:35 +010010513 /* Propagate read liveness of registers... */
10514 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
Jakub Kicinski83d16312019-03-21 14:34:36 -070010515 for (frame = 0; frame <= vstate->curframe; frame++) {
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010516 parent = vparent->frame[frame];
10517 state = vstate->frame[frame];
10518 parent_reg = parent->regs;
10519 state_reg = state->regs;
Jakub Kicinski83d16312019-03-21 14:34:36 -070010520 /* We don't need to worry about FP liveness, it's read-only */
10521 for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) {
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010522 err = propagate_liveness_reg(env, &state_reg[i],
10523 &parent_reg[i]);
Jiong Wang5327ed32019-05-24 23:25:12 +010010524 if (err < 0)
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010525 return err;
Jiong Wang5327ed32019-05-24 23:25:12 +010010526 if (err == REG_LIVE_READ64)
10527 mark_insn_zext(env, &parent_reg[i]);
Edward Creedc503a82017-08-15 20:34:35 +010010528 }
Edward Creedc503a82017-08-15 20:34:35 +010010529
Jiong Wang1b04aee2019-04-12 22:59:34 +010010530 /* Propagate stack slots. */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010531 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE &&
10532 i < parent->allocated_stack / BPF_REG_SIZE; i++) {
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010533 parent_reg = &parent->stack[i].spilled_ptr;
10534 state_reg = &state->stack[i].spilled_ptr;
Jiong Wang55e7f3b2019-04-12 22:59:36 +010010535 err = propagate_liveness_reg(env, state_reg,
10536 parent_reg);
Jiong Wang5327ed32019-05-24 23:25:12 +010010537 if (err < 0)
Jiong Wang3f8cafa2019-04-12 22:59:35 +010010538 return err;
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010539 }
Edward Creedc503a82017-08-15 20:34:35 +010010540 }
Jiong Wang5327ed32019-05-24 23:25:12 +010010541 return 0;
Edward Creedc503a82017-08-15 20:34:35 +010010542}
10543
Alexei Starovoitova3ce6852019-06-28 09:24:09 -070010544/* find precise scalars in the previous equivalent state and
10545 * propagate them into the current state
10546 */
10547static int propagate_precision(struct bpf_verifier_env *env,
10548 const struct bpf_verifier_state *old)
10549{
10550 struct bpf_reg_state *state_reg;
10551 struct bpf_func_state *state;
10552 int i, err = 0;
10553
10554 state = old->frame[old->curframe];
10555 state_reg = state->regs;
10556 for (i = 0; i < BPF_REG_FP; i++, state_reg++) {
10557 if (state_reg->type != SCALAR_VALUE ||
10558 !state_reg->precise)
10559 continue;
10560 if (env->log.level & BPF_LOG_LEVEL2)
10561 verbose(env, "propagating r%d\n", i);
10562 err = mark_chain_precision(env, i);
10563 if (err < 0)
10564 return err;
10565 }
10566
10567 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
10568 if (state->stack[i].slot_type[0] != STACK_SPILL)
10569 continue;
10570 state_reg = &state->stack[i].spilled_ptr;
10571 if (state_reg->type != SCALAR_VALUE ||
10572 !state_reg->precise)
10573 continue;
10574 if (env->log.level & BPF_LOG_LEVEL2)
10575 verbose(env, "propagating fp%d\n",
10576 (-i - 1) * BPF_REG_SIZE);
10577 err = mark_chain_precision_stack(env, i);
10578 if (err < 0)
10579 return err;
10580 }
10581 return 0;
10582}
10583
Alexei Starovoitov25897262019-06-15 12:12:20 -070010584static bool states_maybe_looping(struct bpf_verifier_state *old,
10585 struct bpf_verifier_state *cur)
10586{
10587 struct bpf_func_state *fold, *fcur;
10588 int i, fr = cur->curframe;
10589
10590 if (old->curframe != fr)
10591 return false;
10592
10593 fold = old->frame[fr];
10594 fcur = cur->frame[fr];
10595 for (i = 0; i < MAX_BPF_REG; i++)
10596 if (memcmp(&fold->regs[i], &fcur->regs[i],
10597 offsetof(struct bpf_reg_state, parent)))
10598 return false;
10599 return true;
10600}
10601
10602
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010010603static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010604{
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010010605 struct bpf_verifier_state_list *new_sl;
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070010606 struct bpf_verifier_state_list *sl, **pprev;
Edward Cree679c7822018-08-22 20:02:19 +010010607 struct bpf_verifier_state *cur = env->cur_state, *new;
Alexei Starovoitovceefbc92018-12-03 22:46:06 -080010608 int i, j, err, states_cnt = 0;
Alexei Starovoitov10d274e2019-08-22 22:52:12 -070010609 bool add_new_state = env->test_state_freq ? true : false;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010610
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010611 cur->last_insn_idx = env->prev_insn_idx;
Alexei Starovoitova8f500a2019-05-21 20:17:06 -070010612 if (!env->insn_aux_data[insn_idx].prune_point)
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010613 /* this 'insn_idx' instruction wasn't marked, so we will not
10614 * be doing state search here
10615 */
10616 return 0;
10617
Alexei Starovoitov25897262019-06-15 12:12:20 -070010618 /* bpf progs typically have pruning point every 4 instructions
10619 * http://vger.kernel.org/bpfconf2019.html#session-1
10620 * Do not add new state for future pruning if the verifier hasn't seen
10621 * at least 2 jumps and at least 8 instructions.
10622 * This heuristics helps decrease 'total_states' and 'peak_states' metric.
10623 * In tests that amounts to up to 50% reduction into total verifier
10624 * memory consumption and 20% verifier time speedup.
10625 */
10626 if (env->jmps_processed - env->prev_jmps_processed >= 2 &&
10627 env->insn_processed - env->prev_insn_processed >= 8)
10628 add_new_state = true;
10629
Alexei Starovoitova8f500a2019-05-21 20:17:06 -070010630 pprev = explored_state(env, insn_idx);
10631 sl = *pprev;
10632
Alexei Starovoitov9242b5f2018-12-13 11:42:34 -080010633 clean_live_states(env, insn_idx, cur);
10634
Alexei Starovoitova8f500a2019-05-21 20:17:06 -070010635 while (sl) {
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070010636 states_cnt++;
10637 if (sl->state.insn_idx != insn_idx)
10638 goto next;
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -070010639
Alexei Starovoitov25897262019-06-15 12:12:20 -070010640 if (sl->state.branches) {
Alexei Starovoitovbfc6bb72021-07-14 17:54:14 -070010641 struct bpf_func_state *frame = sl->state.frame[sl->state.curframe];
10642
10643 if (frame->in_async_callback_fn &&
10644 frame->async_entry_cnt != cur->frame[cur->curframe]->async_entry_cnt) {
10645 /* Different async_entry_cnt means that the verifier is
10646 * processing another entry into async callback.
10647 * Seeing the same state is not an indication of infinite
10648 * loop or infinite recursion.
10649 * But finding the same state doesn't mean that it's safe
10650 * to stop processing the current state. The previous state
10651 * hasn't yet reached bpf_exit, since state.branches > 0.
10652 * Checking in_async_callback_fn alone is not enough either.
10653 * Since the verifier still needs to catch infinite loops
10654 * inside async callbacks.
10655 */
10656 } else if (states_maybe_looping(&sl->state, cur) &&
10657 states_equal(env, &sl->state, cur)) {
Alexei Starovoitov25897262019-06-15 12:12:20 -070010658 verbose_linfo(env, insn_idx, "; ");
10659 verbose(env, "infinite loop detected at insn %d\n", insn_idx);
10660 return -EINVAL;
10661 }
10662 /* if the verifier is processing a loop, avoid adding new state
10663 * too often, since different loop iterations have distinct
10664 * states and may not help future pruning.
10665 * This threshold shouldn't be too low to make sure that
10666 * a loop with large bound will be rejected quickly.
10667 * The most abusive loop will be:
10668 * r1 += 1
10669 * if r1 < 1000000 goto pc-2
10670 * 1M insn_procssed limit / 100 == 10k peak states.
10671 * This threshold shouldn't be too high either, since states
10672 * at the end of the loop are likely to be useful in pruning.
10673 */
10674 if (env->jmps_processed - env->prev_jmps_processed < 20 &&
10675 env->insn_processed - env->prev_insn_processed < 100)
10676 add_new_state = false;
10677 goto miss;
10678 }
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010679 if (states_equal(env, &sl->state, cur)) {
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070010680 sl->hit_cnt++;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010681 /* reached equivalent register/stack state,
Edward Creedc503a82017-08-15 20:34:35 +010010682 * prune the search.
10683 * Registers read by the continuation are read by us.
Edward Cree8e9cd9c2017-08-23 15:11:21 +010010684 * If we have any write marks in env->cur_state, they
10685 * will prevent corresponding reads in the continuation
10686 * from reaching our parent (an explored_state). Our
10687 * own state will get the read marks recorded, but
10688 * they'll be immediately forgotten as we're pruning
10689 * this state and will pop a new one.
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010690 */
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010691 err = propagate_liveness(env, &sl->state, cur);
Alexei Starovoitova3ce6852019-06-28 09:24:09 -070010692
10693 /* if previous state reached the exit with precision and
10694 * current state is equivalent to it (except precsion marks)
10695 * the precision needs to be propagated back in
10696 * the current state.
10697 */
10698 err = err ? : push_jmp_history(env, cur);
10699 err = err ? : propagate_precision(env, &sl->state);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010700 if (err)
10701 return err;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010702 return 1;
Edward Creedc503a82017-08-15 20:34:35 +010010703 }
Alexei Starovoitov25897262019-06-15 12:12:20 -070010704miss:
10705 /* when new state is not going to be added do not increase miss count.
10706 * Otherwise several loop iterations will remove the state
10707 * recorded earlier. The goal of these heuristics is to have
10708 * states from some iterations of the loop (some in the beginning
10709 * and some at the end) to help pruning.
10710 */
10711 if (add_new_state)
10712 sl->miss_cnt++;
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070010713 /* heuristic to determine whether this state is beneficial
10714 * to keep checking from state equivalence point of view.
10715 * Higher numbers increase max_states_per_insn and verification time,
10716 * but do not meaningfully decrease insn_processed.
10717 */
10718 if (sl->miss_cnt > sl->hit_cnt * 3 + 3) {
10719 /* the state is unlikely to be useful. Remove it to
10720 * speed up verification
10721 */
10722 *pprev = sl->next;
10723 if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) {
Alexei Starovoitov25897262019-06-15 12:12:20 -070010724 u32 br = sl->state.branches;
10725
10726 WARN_ONCE(br,
10727 "BUG live_done but branches_to_explore %d\n",
10728 br);
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070010729 free_verifier_state(&sl->state, false);
10730 kfree(sl);
10731 env->peak_states--;
10732 } else {
10733 /* cannot free this state, since parentage chain may
10734 * walk it later. Add it for free_list instead to
10735 * be freed at the end of verification
10736 */
10737 sl->next = env->free_list;
10738 env->free_list = sl;
10739 }
10740 sl = *pprev;
10741 continue;
10742 }
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070010743next:
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070010744 pprev = &sl->next;
10745 sl = *pprev;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010746 }
10747
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010748 if (env->max_states_per_insn < states_cnt)
10749 env->max_states_per_insn = states_cnt;
10750
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -070010751 if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES)
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010752 return push_jmp_history(env, cur);
Alexei Starovoitovceefbc92018-12-03 22:46:06 -080010753
Alexei Starovoitov25897262019-06-15 12:12:20 -070010754 if (!add_new_state)
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010755 return push_jmp_history(env, cur);
Alexei Starovoitov25897262019-06-15 12:12:20 -070010756
10757 /* There were no equivalent states, remember the current one.
10758 * Technically the current state is not proven to be safe yet,
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010759 * but it will either reach outer most bpf_exit (which means it's safe)
Alexei Starovoitov25897262019-06-15 12:12:20 -070010760 * or it will be rejected. When there are no loops the verifier won't be
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010761 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
Alexei Starovoitov25897262019-06-15 12:12:20 -070010762 * again on the way to bpf_exit.
10763 * When looping the sl->state.branches will be > 0 and this state
10764 * will not be considered for equivalence until branches == 0.
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010765 */
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010766 new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL);
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010767 if (!new_sl)
10768 return -ENOMEM;
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010769 env->total_states++;
10770 env->peak_states++;
Alexei Starovoitov25897262019-06-15 12:12:20 -070010771 env->prev_jmps_processed = env->jmps_processed;
10772 env->prev_insn_processed = env->insn_processed;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010773
10774 /* add new state to the head of linked list */
Edward Cree679c7822018-08-22 20:02:19 +010010775 new = &new_sl->state;
10776 err = copy_verifier_state(new, cur);
Alexei Starovoitov1969db42017-11-01 00:08:04 -070010777 if (err) {
Edward Cree679c7822018-08-22 20:02:19 +010010778 free_verifier_state(new, false);
Alexei Starovoitov1969db42017-11-01 00:08:04 -070010779 kfree(new_sl);
10780 return err;
10781 }
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070010782 new->insn_idx = insn_idx;
Alexei Starovoitov25897262019-06-15 12:12:20 -070010783 WARN_ONCE(new->branches != 1,
10784 "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010785
Alexei Starovoitov25897262019-06-15 12:12:20 -070010786 cur->parent = new;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010787 cur->first_insn_idx = insn_idx;
10788 clear_jmp_history(cur);
Alexei Starovoitov5d839022019-05-21 20:17:05 -070010789 new_sl->next = *explored_state(env, insn_idx);
10790 *explored_state(env, insn_idx) = new_sl;
Jakub Kicinski7640ead2018-12-12 16:29:07 -080010791 /* connect new state to parentage chain. Current frame needs all
10792 * registers connected. Only r6 - r9 of the callers are alive (pushed
10793 * to the stack implicitly by JITs) so in callers' frames connect just
10794 * r6 - r9 as an optimization. Callers will have r1 - r5 connected to
10795 * the state of the call instruction (with WRITTEN set), and r0 comes
10796 * from callee with its full parentage chain, anyway.
10797 */
Edward Cree8e9cd9c2017-08-23 15:11:21 +010010798 /* clear write marks in current state: the writes we did are not writes
10799 * our child did, so they don't screen off its reads from us.
10800 * (There are no read marks in current state, because reads always mark
10801 * their parent and current state never has children yet. Only
10802 * explored_states can get read marks.)
10803 */
Alexei Starovoitoveea1c222019-06-15 12:12:21 -070010804 for (j = 0; j <= cur->curframe; j++) {
10805 for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++)
10806 cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i];
10807 for (i = 0; i < BPF_REG_FP; i++)
10808 cur->frame[j]->regs[i].live = REG_LIVE_NONE;
10809 }
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010810
10811 /* all stack frames are accessible from callee, clear them all */
10812 for (j = 0; j <= cur->curframe; j++) {
10813 struct bpf_func_state *frame = cur->frame[j];
Edward Cree679c7822018-08-22 20:02:19 +010010814 struct bpf_func_state *newframe = new->frame[j];
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010815
Edward Cree679c7822018-08-22 20:02:19 +010010816 for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) {
Alexei Starovoitovcc2b14d2017-12-14 17:55:08 -080010817 frame->stack[i].spilled_ptr.live = REG_LIVE_NONE;
Edward Cree679c7822018-08-22 20:02:19 +010010818 frame->stack[i].spilled_ptr.parent =
10819 &newframe->stack[i].spilled_ptr;
10820 }
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010821 }
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010822 return 0;
10823}
10824
Joe Stringerc64b7982018-10-02 13:35:33 -070010825/* Return true if it's OK to have the same insn return a different type. */
10826static bool reg_type_mismatch_ok(enum bpf_reg_type type)
10827{
10828 switch (type) {
10829 case PTR_TO_CTX:
10830 case PTR_TO_SOCKET:
10831 case PTR_TO_SOCKET_OR_NULL:
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -080010832 case PTR_TO_SOCK_COMMON:
10833 case PTR_TO_SOCK_COMMON_OR_NULL:
Martin KaFai Lau655a51e2019-02-09 23:22:24 -080010834 case PTR_TO_TCP_SOCK:
10835 case PTR_TO_TCP_SOCK_OR_NULL:
Jonathan Lemonfada7fd2019-06-06 13:59:40 -070010836 case PTR_TO_XDP_SOCK:
Alexei Starovoitov2a027592019-10-15 20:25:02 -070010837 case PTR_TO_BTF_ID:
Yonghong Songb121b342020-05-09 10:59:12 -070010838 case PTR_TO_BTF_ID_OR_NULL:
Joe Stringerc64b7982018-10-02 13:35:33 -070010839 return false;
10840 default:
10841 return true;
10842 }
10843}
10844
10845/* If an instruction was previously used with particular pointer types, then we
10846 * need to be careful to avoid cases such as the below, where it may be ok
10847 * for one branch accessing the pointer, but not ok for the other branch:
10848 *
10849 * R1 = sock_ptr
10850 * goto X;
10851 * ...
10852 * R1 = some_other_valid_ptr;
10853 * goto X;
10854 * ...
10855 * R2 = *(u32 *)(R1 + 0);
10856 */
10857static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev)
10858{
10859 return src != prev && (!reg_type_mismatch_ok(src) ||
10860 !reg_type_mismatch_ok(prev));
10861}
10862
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010010863static int do_check(struct bpf_verifier_env *env)
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010864{
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -070010865 bool pop_log = !(env->log.level & BPF_LOG_LEVEL2);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080010866 struct bpf_verifier_state *state = env->cur_state;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010867 struct bpf_insn *insns = env->prog->insnsi;
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010868 struct bpf_reg_state *regs;
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010869 int insn_cnt = env->prog->len;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010870 bool do_print_state = false;
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010871 int prev_insn_idx = -1;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010872
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010873 for (;;) {
10874 struct bpf_insn *insn;
10875 u8 class;
10876 int err;
10877
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010878 env->prev_insn_idx = prev_insn_idx;
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010879 if (env->insn_idx >= insn_cnt) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070010880 verbose(env, "invalid insn idx %d insn_cnt %d\n",
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010881 env->insn_idx, insn_cnt);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010882 return -EFAULT;
10883 }
10884
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010885 insn = &insns[env->insn_idx];
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010886 class = BPF_CLASS(insn->code);
10887
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010888 if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070010889 verbose(env,
10890 "BPF program is too large. Processed %d insn\n",
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010891 env->insn_processed);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010892 return -E2BIG;
10893 }
10894
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010895 err = is_state_visited(env, env->insn_idx);
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010896 if (err < 0)
10897 return err;
10898 if (err == 1) {
10899 /* found equivalent state, can prune the search */
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010900 if (env->log.level & BPF_LOG_LEVEL) {
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010901 if (do_print_state)
Daniel Borkmann979d63d2019-01-03 00:58:34 +010010902 verbose(env, "\nfrom %d to %d%s: safe\n",
10903 env->prev_insn_idx, env->insn_idx,
10904 env->cur_state->speculative ?
10905 " (speculative execution)" : "");
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010906 else
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010907 verbose(env, "%d: safe\n", env->insn_idx);
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070010908 }
10909 goto process_bpf_exit;
10910 }
10911
Alexei Starovoitovc3494802018-12-03 22:46:04 -080010912 if (signal_pending(current))
10913 return -EAGAIN;
10914
Daniel Borkmann3c2ce602017-05-18 03:00:06 +020010915 if (need_resched())
10916 cond_resched();
10917
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010918 if (env->log.level & BPF_LOG_LEVEL2 ||
10919 (env->log.level & BPF_LOG_LEVEL && do_print_state)) {
10920 if (env->log.level & BPF_LOG_LEVEL2)
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010921 verbose(env, "%d:", env->insn_idx);
David S. Millerc5fc9692017-05-10 11:25:17 -070010922 else
Daniel Borkmann979d63d2019-01-03 00:58:34 +010010923 verbose(env, "\nfrom %d to %d%s:",
10924 env->prev_insn_idx, env->insn_idx,
10925 env->cur_state->speculative ?
10926 " (speculative execution)" : "");
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080010927 print_verifier_state(env, state->frame[state->curframe]);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010928 do_print_state = false;
10929 }
10930
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070010931 if (env->log.level & BPF_LOG_LEVEL) {
Daniel Borkmann7105e822017-12-20 13:42:57 +010010932 const struct bpf_insn_cbs cbs = {
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070010933 .cb_call = disasm_kfunc_name,
Daniel Borkmann7105e822017-12-20 13:42:57 +010010934 .cb_print = verbose,
Jiri Olsaabe08842018-03-23 11:41:28 +010010935 .private_data = env,
Daniel Borkmann7105e822017-12-20 13:42:57 +010010936 };
10937
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010938 verbose_linfo(env, env->insn_idx, "; ");
10939 verbose(env, "%d: ", env->insn_idx);
Jiri Olsaabe08842018-03-23 11:41:28 +010010940 print_bpf_insn(&cbs, insn, env->allow_ptr_leaks);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010941 }
10942
Jakub Kicinskicae19272017-12-27 18:39:05 -080010943 if (bpf_prog_is_dev_bound(env->prog->aux)) {
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010944 err = bpf_prog_offload_verify_insn(env, env->insn_idx,
10945 env->prev_insn_idx);
Jakub Kicinskicae19272017-12-27 18:39:05 -080010946 if (err)
10947 return err;
10948 }
Jakub Kicinski13a27df2016-09-21 11:43:58 +010010949
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070010950 regs = cur_regs(env);
Daniel Borkmannfe9a5ca2021-05-28 13:47:27 +000010951 sanitize_mark_insn_seen(env);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070010952 prev_insn_idx = env->insn_idx;
Joe Stringerfd978bf72018-10-02 13:35:35 -070010953
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010954 if (class == BPF_ALU || class == BPF_ALU64) {
Alexei Starovoitov1be7f752015-10-07 22:23:21 -070010955 err = check_alu_op(env, insn);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010956 if (err)
10957 return err;
10958
10959 } else if (class == BPF_LDX) {
Jakub Kicinski3df126f2016-09-21 11:43:56 +010010960 enum bpf_reg_type *prev_src_type, src_reg_type;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070010961
10962 /* check for reserved fields is already done */
10963
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010964 /* check src operand */
Edward Creedc503a82017-08-15 20:34:35 +010010965 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010966 if (err)
10967 return err;
10968
Edward Creedc503a82017-08-15 20:34:35 +010010969 err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010970 if (err)
10971 return err;
10972
Alexei Starovoitov725f9dc2015-04-15 16:19:33 -070010973 src_reg_type = regs[insn->src_reg].type;
10974
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010975 /* check that memory (src_reg + off) is readable,
10976 * the state of dst_reg will be updated by this func
10977 */
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010978 err = check_mem_access(env, env->insn_idx, insn->src_reg,
10979 insn->off, BPF_SIZE(insn->code),
10980 BPF_READ, insn->dst_reg, false);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070010981 if (err)
10982 return err;
10983
Daniel Borkmannc08435e2019-01-03 00:58:27 +010010984 prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010010985
10986 if (*prev_src_type == NOT_INIT) {
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070010987 /* saw a valid insn
10988 * dst_reg = *(u32 *)(src_reg + off)
Jakub Kicinski3df126f2016-09-21 11:43:56 +010010989 * save type to validate intersecting paths
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070010990 */
Jakub Kicinski3df126f2016-09-21 11:43:56 +010010991 *prev_src_type = src_reg_type;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070010992
Joe Stringerc64b7982018-10-02 13:35:33 -070010993 } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) {
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070010994 /* ABuser program is trying to use the same insn
10995 * dst_reg = *(u32*) (src_reg + off)
10996 * with different pointer types:
10997 * src_reg == ctx in one branch and
10998 * src_reg == stack|map in some other branch.
10999 * Reject it.
11000 */
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011001 verbose(env, "same insn cannot be used with different pointers\n");
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070011002 return -EINVAL;
11003 }
11004
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011005 } else if (class == BPF_STX) {
Jakub Kicinski3df126f2016-09-21 11:43:56 +010011006 enum bpf_reg_type *prev_dst_type, dst_reg_type;
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070011007
Brendan Jackman91c960b2021-01-14 18:17:44 +000011008 if (BPF_MODE(insn->code) == BPF_ATOMIC) {
11009 err = check_atomic(env, env->insn_idx, insn);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011010 if (err)
11011 return err;
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011012 env->insn_idx++;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011013 continue;
11014 }
11015
Brendan Jackman5ca419f2021-01-14 18:17:46 +000011016 if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) {
11017 verbose(env, "BPF_STX uses reserved fields\n");
11018 return -EINVAL;
11019 }
11020
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011021 /* check src1 operand */
Edward Creedc503a82017-08-15 20:34:35 +010011022 err = check_reg_arg(env, insn->src_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011023 if (err)
11024 return err;
11025 /* check src2 operand */
Edward Creedc503a82017-08-15 20:34:35 +010011026 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011027 if (err)
11028 return err;
11029
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070011030 dst_reg_type = regs[insn->dst_reg].type;
11031
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011032 /* check that memory (dst_reg + off) is writeable */
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011033 err = check_mem_access(env, env->insn_idx, insn->dst_reg,
11034 insn->off, BPF_SIZE(insn->code),
11035 BPF_WRITE, insn->src_reg, false);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011036 if (err)
11037 return err;
11038
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011039 prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010011040
11041 if (*prev_dst_type == NOT_INIT) {
11042 *prev_dst_type = dst_reg_type;
Joe Stringerc64b7982018-10-02 13:35:33 -070011043 } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011044 verbose(env, "same insn cannot be used with different pointers\n");
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070011045 return -EINVAL;
11046 }
11047
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011048 } else if (class == BPF_ST) {
11049 if (BPF_MODE(insn->code) != BPF_MEM ||
11050 insn->src_reg != BPF_REG_0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011051 verbose(env, "BPF_ST uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011052 return -EINVAL;
11053 }
11054 /* check src operand */
Edward Creedc503a82017-08-15 20:34:35 +010011055 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011056 if (err)
11057 return err;
11058
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +010011059 if (is_ctx_reg(env, insn->dst_reg)) {
Joe Stringer9d2be442018-10-02 13:35:31 -070011060 verbose(env, "BPF_ST stores into R%d %s is not allowed\n",
Daniel Borkmann2a159c62018-10-21 02:09:24 +020011061 insn->dst_reg,
11062 reg_type_str[reg_state(env, insn->dst_reg)->type]);
Daniel Borkmannf37a8cb2018-01-16 23:30:10 +010011063 return -EACCES;
11064 }
11065
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011066 /* check that memory (dst_reg + off) is writeable */
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011067 err = check_mem_access(env, env->insn_idx, insn->dst_reg,
11068 insn->off, BPF_SIZE(insn->code),
11069 BPF_WRITE, -1, false);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011070 if (err)
11071 return err;
11072
Jiong Wang092ed092019-01-26 12:26:01 -050011073 } else if (class == BPF_JMP || class == BPF_JMP32) {
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011074 u8 opcode = BPF_OP(insn->code);
11075
Alexei Starovoitov25897262019-06-15 12:12:20 -070011076 env->jmps_processed++;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011077 if (opcode == BPF_CALL) {
11078 if (BPF_SRC(insn->code) != BPF_K ||
11079 insn->off != 0 ||
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080011080 (insn->src_reg != BPF_REG_0 &&
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070011081 insn->src_reg != BPF_PSEUDO_CALL &&
11082 insn->src_reg != BPF_PSEUDO_KFUNC_CALL) ||
Jiong Wang092ed092019-01-26 12:26:01 -050011083 insn->dst_reg != BPF_REG_0 ||
11084 class == BPF_JMP32) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011085 verbose(env, "BPF_CALL uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011086 return -EINVAL;
11087 }
11088
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080011089 if (env->cur_state->active_spin_lock &&
11090 (insn->src_reg == BPF_PSEUDO_CALL ||
11091 insn->imm != BPF_FUNC_spin_unlock)) {
11092 verbose(env, "function calls are not allowed while holding a lock\n");
11093 return -EINVAL;
11094 }
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080011095 if (insn->src_reg == BPF_PSEUDO_CALL)
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011096 err = check_func_call(env, insn, &env->insn_idx);
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070011097 else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL)
11098 err = check_kfunc_call(env, insn);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080011099 else
Yonghong Song69c087b2021-02-26 12:49:25 -080011100 err = check_helper_call(env, insn, &env->insn_idx);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011101 if (err)
11102 return err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011103 } else if (opcode == BPF_JA) {
11104 if (BPF_SRC(insn->code) != BPF_K ||
11105 insn->imm != 0 ||
11106 insn->src_reg != BPF_REG_0 ||
Jiong Wang092ed092019-01-26 12:26:01 -050011107 insn->dst_reg != BPF_REG_0 ||
11108 class == BPF_JMP32) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011109 verbose(env, "BPF_JA uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011110 return -EINVAL;
11111 }
11112
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011113 env->insn_idx += insn->off + 1;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011114 continue;
11115
11116 } else if (opcode == BPF_EXIT) {
11117 if (BPF_SRC(insn->code) != BPF_K ||
11118 insn->imm != 0 ||
11119 insn->src_reg != BPF_REG_0 ||
Jiong Wang092ed092019-01-26 12:26:01 -050011120 insn->dst_reg != BPF_REG_0 ||
11121 class == BPF_JMP32) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011122 verbose(env, "BPF_EXIT uses reserved fields\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011123 return -EINVAL;
11124 }
11125
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080011126 if (env->cur_state->active_spin_lock) {
11127 verbose(env, "bpf_spin_unlock is missing\n");
11128 return -EINVAL;
11129 }
11130
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080011131 if (state->curframe) {
11132 /* exit from nested function */
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011133 err = prepare_func_exit(env, &env->insn_idx);
Alexei Starovoitovf4d7e402017-12-14 17:55:06 -080011134 if (err)
11135 return err;
11136 do_print_state = true;
11137 continue;
11138 }
11139
Joe Stringerfd978bf72018-10-02 13:35:35 -070011140 err = check_reference_leak(env);
11141 if (err)
11142 return err;
11143
Alexei Starovoitov390ee7e2017-10-02 22:50:23 -070011144 err = check_return_code(env);
11145 if (err)
11146 return err;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070011147process_bpf_exit:
Alexei Starovoitov25897262019-06-15 12:12:20 -070011148 update_branch_counts(env, env->cur_state);
Alexei Starovoitovb5dc0162019-06-15 12:12:25 -070011149 err = pop_stack(env, &prev_insn_idx,
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -070011150 &env->insn_idx, pop_log);
Alexei Starovoitov638f5b92017-10-31 18:16:05 -070011151 if (err < 0) {
11152 if (err != -ENOENT)
11153 return err;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011154 break;
11155 } else {
11156 do_print_state = true;
11157 continue;
11158 }
11159 } else {
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011160 err = check_cond_jmp_op(env, insn, &env->insn_idx);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011161 if (err)
11162 return err;
11163 }
11164 } else if (class == BPF_LD) {
11165 u8 mode = BPF_MODE(insn->code);
11166
11167 if (mode == BPF_ABS || mode == BPF_IND) {
Alexei Starovoitovddd872b2014-12-01 15:06:34 -080011168 err = check_ld_abs(env, insn);
11169 if (err)
11170 return err;
11171
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011172 } else if (mode == BPF_IMM) {
11173 err = check_ld_imm(env, insn);
11174 if (err)
11175 return err;
11176
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011177 env->insn_idx++;
Daniel Borkmannfe9a5ca2021-05-28 13:47:27 +000011178 sanitize_mark_insn_seen(env);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011179 } else {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011180 verbose(env, "invalid BPF_LD mode\n");
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011181 return -EINVAL;
11182 }
11183 } else {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011184 verbose(env, "unknown insn class %d\n", class);
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011185 return -EINVAL;
11186 }
11187
Daniel Borkmannc08435e2019-01-03 00:58:27 +010011188 env->insn_idx++;
Alexei Starovoitov17a52672014-09-26 00:17:06 -070011189 }
11190
11191 return 0;
11192}
11193
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011194static int find_btf_percpu_datasec(struct btf *btf)
11195{
11196 const struct btf_type *t;
11197 const char *tname;
11198 int i, n;
11199
11200 /*
11201 * Both vmlinux and module each have their own ".data..percpu"
11202 * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF
11203 * types to look at only module's own BTF types.
11204 */
11205 n = btf_nr_types(btf);
11206 if (btf_is_module(btf))
11207 i = btf_nr_types(btf_vmlinux);
11208 else
11209 i = 1;
11210
11211 for(; i < n; i++) {
11212 t = btf_type_by_id(btf, i);
11213 if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC)
11214 continue;
11215
11216 tname = btf_name_by_offset(btf, t->name_off);
11217 if (!strcmp(tname, ".data..percpu"))
11218 return i;
11219 }
11220
11221 return -ENOENT;
11222}
11223
Hao Luo4976b712020-09-29 16:50:44 -070011224/* replace pseudo btf_id with kernel symbol address */
11225static int check_pseudo_btf_id(struct bpf_verifier_env *env,
11226 struct bpf_insn *insn,
11227 struct bpf_insn_aux_data *aux)
11228{
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011229 const struct btf_var_secinfo *vsi;
11230 const struct btf_type *datasec;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011231 struct btf_mod_pair *btf_mod;
Hao Luo4976b712020-09-29 16:50:44 -070011232 const struct btf_type *t;
11233 const char *sym_name;
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011234 bool percpu = false;
Kaixu Xiaf16e6312020-11-11 13:03:46 +080011235 u32 type, id = insn->imm;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011236 struct btf *btf;
Kaixu Xiaf16e6312020-11-11 13:03:46 +080011237 s32 datasec_id;
Hao Luo4976b712020-09-29 16:50:44 -070011238 u64 addr;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011239 int i, btf_fd, err;
Hao Luo4976b712020-09-29 16:50:44 -070011240
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011241 btf_fd = insn[1].imm;
11242 if (btf_fd) {
11243 btf = btf_get_by_fd(btf_fd);
11244 if (IS_ERR(btf)) {
11245 verbose(env, "invalid module BTF object FD specified.\n");
11246 return -EINVAL;
11247 }
11248 } else {
11249 if (!btf_vmlinux) {
11250 verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
11251 return -EINVAL;
11252 }
11253 btf = btf_vmlinux;
11254 btf_get(btf);
Hao Luo4976b712020-09-29 16:50:44 -070011255 }
11256
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011257 t = btf_type_by_id(btf, id);
Hao Luo4976b712020-09-29 16:50:44 -070011258 if (!t) {
11259 verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id);
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011260 err = -ENOENT;
11261 goto err_put;
Hao Luo4976b712020-09-29 16:50:44 -070011262 }
11263
11264 if (!btf_type_is_var(t)) {
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011265 verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id);
11266 err = -EINVAL;
11267 goto err_put;
Hao Luo4976b712020-09-29 16:50:44 -070011268 }
11269
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011270 sym_name = btf_name_by_offset(btf, t->name_off);
Hao Luo4976b712020-09-29 16:50:44 -070011271 addr = kallsyms_lookup_name(sym_name);
11272 if (!addr) {
11273 verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n",
11274 sym_name);
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011275 err = -ENOENT;
11276 goto err_put;
Hao Luo4976b712020-09-29 16:50:44 -070011277 }
11278
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011279 datasec_id = find_btf_percpu_datasec(btf);
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011280 if (datasec_id > 0) {
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011281 datasec = btf_type_by_id(btf, datasec_id);
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011282 for_each_vsi(i, datasec, vsi) {
11283 if (vsi->type == id) {
11284 percpu = true;
11285 break;
11286 }
11287 }
11288 }
11289
Hao Luo4976b712020-09-29 16:50:44 -070011290 insn[0].imm = (u32)addr;
11291 insn[1].imm = addr >> 32;
11292
11293 type = t->type;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011294 t = btf_type_skip_modifiers(btf, type, NULL);
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011295 if (percpu) {
11296 aux->btf_var.reg_type = PTR_TO_PERCPU_BTF_ID;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011297 aux->btf_var.btf = btf;
Hao Luoeaa6bcb2020-09-29 16:50:47 -070011298 aux->btf_var.btf_id = type;
11299 } else if (!btf_type_is_struct(t)) {
Hao Luo4976b712020-09-29 16:50:44 -070011300 const struct btf_type *ret;
11301 const char *tname;
11302 u32 tsize;
11303
11304 /* resolve the type size of ksym. */
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011305 ret = btf_resolve_size(btf, t, &tsize);
Hao Luo4976b712020-09-29 16:50:44 -070011306 if (IS_ERR(ret)) {
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011307 tname = btf_name_by_offset(btf, t->name_off);
Hao Luo4976b712020-09-29 16:50:44 -070011308 verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n",
11309 tname, PTR_ERR(ret));
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011310 err = -EINVAL;
11311 goto err_put;
Hao Luo4976b712020-09-29 16:50:44 -070011312 }
11313 aux->btf_var.reg_type = PTR_TO_MEM;
11314 aux->btf_var.mem_size = tsize;
11315 } else {
11316 aux->btf_var.reg_type = PTR_TO_BTF_ID;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011317 aux->btf_var.btf = btf;
Hao Luo4976b712020-09-29 16:50:44 -070011318 aux->btf_var.btf_id = type;
11319 }
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011320
11321 /* check whether we recorded this BTF (and maybe module) already */
11322 for (i = 0; i < env->used_btf_cnt; i++) {
11323 if (env->used_btfs[i].btf == btf) {
11324 btf_put(btf);
11325 return 0;
11326 }
11327 }
11328
11329 if (env->used_btf_cnt >= MAX_USED_BTFS) {
11330 err = -E2BIG;
11331 goto err_put;
11332 }
11333
11334 btf_mod = &env->used_btfs[env->used_btf_cnt];
11335 btf_mod->btf = btf;
11336 btf_mod->module = NULL;
11337
11338 /* if we reference variables from kernel module, bump its refcount */
11339 if (btf_is_module(btf)) {
11340 btf_mod->module = btf_try_get_module(btf);
11341 if (!btf_mod->module) {
11342 err = -ENXIO;
11343 goto err_put;
11344 }
11345 }
11346
11347 env->used_btf_cnt++;
11348
Hao Luo4976b712020-09-29 16:50:44 -070011349 return 0;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011350err_put:
11351 btf_put(btf);
11352 return err;
Hao Luo4976b712020-09-29 16:50:44 -070011353}
11354
Martin KaFai Lau56f668d2017-03-22 10:00:33 -070011355static int check_map_prealloc(struct bpf_map *map)
11356{
11357 return (map->map_type != BPF_MAP_TYPE_HASH &&
Martin KaFai Laubcc6b1b2017-03-22 10:00:34 -070011358 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
11359 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
Martin KaFai Lau56f668d2017-03-22 10:00:33 -070011360 !(map->map_flags & BPF_F_NO_PREALLOC);
11361}
11362
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080011363static bool is_tracing_prog_type(enum bpf_prog_type type)
11364{
11365 switch (type) {
11366 case BPF_PROG_TYPE_KPROBE:
11367 case BPF_PROG_TYPE_TRACEPOINT:
11368 case BPF_PROG_TYPE_PERF_EVENT:
11369 case BPF_PROG_TYPE_RAW_TRACEPOINT:
11370 return true;
11371 default:
11372 return false;
11373 }
11374}
11375
Thomas Gleixner94dacdb2020-02-24 15:01:32 +010011376static bool is_preallocated_map(struct bpf_map *map)
11377{
11378 if (!check_map_prealloc(map))
11379 return false;
11380 if (map->inner_map_meta && !check_map_prealloc(map->inner_map_meta))
11381 return false;
11382 return true;
11383}
11384
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011385static int check_map_prog_compatibility(struct bpf_verifier_env *env,
11386 struct bpf_map *map,
Alexei Starovoitovfdc15d32016-09-01 18:37:23 -070011387 struct bpf_prog *prog)
11388
11389{
Udip Pant7e407812020-08-25 16:20:00 -070011390 enum bpf_prog_type prog_type = resolve_prog_type(prog);
Thomas Gleixner94dacdb2020-02-24 15:01:32 +010011391 /*
11392 * Validate that trace type programs use preallocated hash maps.
11393 *
11394 * For programs attached to PERF events this is mandatory as the
11395 * perf NMI can hit any arbitrary code sequence.
11396 *
11397 * All other trace types using preallocated hash maps are unsafe as
11398 * well because tracepoint or kprobes can be inside locked regions
11399 * of the memory allocator or at a place where a recursion into the
11400 * memory allocator would see inconsistent state.
11401 *
Thomas Gleixner2ed905c2020-02-24 15:01:33 +010011402 * On RT enabled kernels run-time allocation of all trace type
11403 * programs is strictly prohibited due to lock type constraints. On
11404 * !RT kernels it is allowed for backwards compatibility reasons for
11405 * now, but warnings are emitted so developers are made aware of
11406 * the unsafety and can fix their programs before this is enforced.
Martin KaFai Lau56f668d2017-03-22 10:00:33 -070011407 */
Udip Pant7e407812020-08-25 16:20:00 -070011408 if (is_tracing_prog_type(prog_type) && !is_preallocated_map(map)) {
11409 if (prog_type == BPF_PROG_TYPE_PERF_EVENT) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011410 verbose(env, "perf_event programs can only use preallocated hash map\n");
Martin KaFai Lau56f668d2017-03-22 10:00:33 -070011411 return -EINVAL;
11412 }
Thomas Gleixner2ed905c2020-02-24 15:01:33 +010011413 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
11414 verbose(env, "trace type programs can only use preallocated hash map\n");
11415 return -EINVAL;
11416 }
Thomas Gleixner94dacdb2020-02-24 15:01:32 +010011417 WARN_ONCE(1, "trace type BPF program uses run-time allocation\n");
11418 verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n");
Alexei Starovoitovfdc15d32016-09-01 18:37:23 -070011419 }
Jakub Kicinskia3884572018-01-11 20:29:09 -080011420
KP Singh9e7a4d92020-11-06 10:37:39 +000011421 if (map_value_has_spin_lock(map)) {
11422 if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) {
11423 verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n");
11424 return -EINVAL;
11425 }
11426
11427 if (is_tracing_prog_type(prog_type)) {
11428 verbose(env, "tracing progs cannot use bpf_spin_lock yet\n");
11429 return -EINVAL;
11430 }
11431
11432 if (prog->aux->sleepable) {
11433 verbose(env, "sleepable progs cannot use bpf_spin_lock yet\n");
11434 return -EINVAL;
11435 }
Alexei Starovoitovd83525c2019-01-31 15:40:04 -080011436 }
11437
Jakub Kicinskia3884572018-01-11 20:29:09 -080011438 if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
Jakub Kicinski09728262018-07-17 10:53:23 -070011439 !bpf_offload_prog_map_match(prog, map)) {
Jakub Kicinskia3884572018-01-11 20:29:09 -080011440 verbose(env, "offload device mismatch between prog and map\n");
11441 return -EINVAL;
11442 }
11443
Martin KaFai Lau85d33df2020-01-08 16:35:05 -080011444 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
11445 verbose(env, "bpf_struct_ops map cannot be used in prog\n");
11446 return -EINVAL;
11447 }
11448
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070011449 if (prog->aux->sleepable)
11450 switch (map->map_type) {
11451 case BPF_MAP_TYPE_HASH:
11452 case BPF_MAP_TYPE_LRU_HASH:
11453 case BPF_MAP_TYPE_ARRAY:
Alexei Starovoitov638e4b82021-02-09 19:36:33 -080011454 case BPF_MAP_TYPE_PERCPU_HASH:
11455 case BPF_MAP_TYPE_PERCPU_ARRAY:
11456 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
11457 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
11458 case BPF_MAP_TYPE_HASH_OF_MAPS:
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070011459 if (!is_preallocated_map(map)) {
11460 verbose(env,
Alexei Starovoitov638e4b82021-02-09 19:36:33 -080011461 "Sleepable programs can only use preallocated maps\n");
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070011462 return -EINVAL;
11463 }
11464 break;
KP Singhba90c2c2021-02-04 19:36:21 +000011465 case BPF_MAP_TYPE_RINGBUF:
11466 break;
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070011467 default:
11468 verbose(env,
KP Singhba90c2c2021-02-04 19:36:21 +000011469 "Sleepable programs can only use array, hash, and ringbuf maps\n");
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070011470 return -EINVAL;
11471 }
11472
Alexei Starovoitovfdc15d32016-09-01 18:37:23 -070011473 return 0;
11474}
11475
Roman Gushchinb741f162018-09-28 14:45:43 +000011476static bool bpf_map_is_cgroup_storage(struct bpf_map *map)
11477{
11478 return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE ||
11479 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
11480}
11481
Hao Luo4976b712020-09-29 16:50:44 -070011482/* find and rewrite pseudo imm in ld_imm64 instructions:
11483 *
11484 * 1. if it accesses map FD, replace it with actual map pointer.
11485 * 2. if it accesses btf_id of a VAR, replace it with pointer to the var.
11486 *
11487 * NOTE: btf_vmlinux is required for converting pseudo btf_id.
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011488 */
Hao Luo4976b712020-09-29 16:50:44 -070011489static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env)
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011490{
11491 struct bpf_insn *insn = env->prog->insnsi;
11492 int insn_cnt = env->prog->len;
Alexei Starovoitovfdc15d32016-09-01 18:37:23 -070011493 int i, j, err;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011494
Daniel Borkmannf1f77142017-01-13 23:38:15 +010011495 err = bpf_prog_calc_tag(env->prog);
Daniel Borkmannaafe6ae2016-12-18 01:52:57 +010011496 if (err)
11497 return err;
11498
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011499 for (i = 0; i < insn_cnt; i++, insn++) {
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070011500 if (BPF_CLASS(insn->code) == BPF_LDX &&
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070011501 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011502 verbose(env, "BPF_LDX uses reserved fields\n");
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070011503 return -EINVAL;
11504 }
11505
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011506 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011507 struct bpf_insn_aux_data *aux;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011508 struct bpf_map *map;
11509 struct fd f;
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011510 u64 addr;
Alexei Starovoitov387544b2021-05-13 17:36:10 -070011511 u32 fd;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011512
11513 if (i == insn_cnt - 1 || insn[1].code != 0 ||
11514 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
11515 insn[1].off != 0) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011516 verbose(env, "invalid bpf_ld_imm64 insn\n");
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011517 return -EINVAL;
11518 }
11519
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011520 if (insn[0].src_reg == 0)
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011521 /* valid generic load 64-bit imm */
11522 goto next_insn;
11523
Hao Luo4976b712020-09-29 16:50:44 -070011524 if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) {
11525 aux = &env->insn_aux_data[i];
11526 err = check_pseudo_btf_id(env, insn, aux);
11527 if (err)
11528 return err;
11529 goto next_insn;
11530 }
11531
Yonghong Song69c087b2021-02-26 12:49:25 -080011532 if (insn[0].src_reg == BPF_PSEUDO_FUNC) {
11533 aux = &env->insn_aux_data[i];
11534 aux->ptr_type = PTR_TO_FUNC;
11535 goto next_insn;
11536 }
11537
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011538 /* In final convert_pseudo_ld_imm64() step, this is
11539 * converted into regular 64-bit imm load insn.
11540 */
Alexei Starovoitov387544b2021-05-13 17:36:10 -070011541 switch (insn[0].src_reg) {
11542 case BPF_PSEUDO_MAP_VALUE:
11543 case BPF_PSEUDO_MAP_IDX_VALUE:
11544 break;
11545 case BPF_PSEUDO_MAP_FD:
11546 case BPF_PSEUDO_MAP_IDX:
11547 if (insn[1].imm == 0)
11548 break;
11549 fallthrough;
11550 default:
11551 verbose(env, "unrecognized bpf_ld_imm64 insn\n");
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011552 return -EINVAL;
11553 }
11554
Alexei Starovoitov387544b2021-05-13 17:36:10 -070011555 switch (insn[0].src_reg) {
11556 case BPF_PSEUDO_MAP_IDX_VALUE:
11557 case BPF_PSEUDO_MAP_IDX:
11558 if (bpfptr_is_null(env->fd_array)) {
11559 verbose(env, "fd_idx without fd_array is invalid\n");
11560 return -EPROTO;
11561 }
11562 if (copy_from_bpfptr_offset(&fd, env->fd_array,
11563 insn[0].imm * sizeof(fd),
11564 sizeof(fd)))
11565 return -EFAULT;
11566 break;
11567 default:
11568 fd = insn[0].imm;
11569 break;
11570 }
11571
11572 f = fdget(fd);
Daniel Borkmannc2101292015-10-29 14:58:07 +010011573 map = __bpf_map_get(f);
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011574 if (IS_ERR(map)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011575 verbose(env, "fd %d is not pointing to valid bpf_map\n",
Daniel Borkmann20182392019-03-04 21:08:53 +010011576 insn[0].imm);
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011577 return PTR_ERR(map);
11578 }
11579
Jakub Kicinski61bd5212017-10-09 10:30:11 -070011580 err = check_map_prog_compatibility(env, map, env->prog);
Alexei Starovoitovfdc15d32016-09-01 18:37:23 -070011581 if (err) {
11582 fdput(f);
11583 return err;
11584 }
11585
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011586 aux = &env->insn_aux_data[i];
Alexei Starovoitov387544b2021-05-13 17:36:10 -070011587 if (insn[0].src_reg == BPF_PSEUDO_MAP_FD ||
11588 insn[0].src_reg == BPF_PSEUDO_MAP_IDX) {
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011589 addr = (unsigned long)map;
11590 } else {
11591 u32 off = insn[1].imm;
11592
11593 if (off >= BPF_MAX_VAR_OFF) {
11594 verbose(env, "direct value offset of %u is not allowed\n", off);
11595 fdput(f);
11596 return -EINVAL;
11597 }
11598
11599 if (!map->ops->map_direct_value_addr) {
11600 verbose(env, "no direct value access support for this map type\n");
11601 fdput(f);
11602 return -EINVAL;
11603 }
11604
11605 err = map->ops->map_direct_value_addr(map, &addr, off);
11606 if (err) {
11607 verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n",
11608 map->value_size, off);
11609 fdput(f);
11610 return err;
11611 }
11612
11613 aux->map_off = off;
11614 addr += off;
11615 }
11616
11617 insn[0].imm = (u32)addr;
11618 insn[1].imm = addr >> 32;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011619
11620 /* check whether we recorded this map already */
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011621 for (j = 0; j < env->used_map_cnt; j++) {
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011622 if (env->used_maps[j] == map) {
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011623 aux->map_index = j;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011624 fdput(f);
11625 goto next_insn;
11626 }
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011627 }
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011628
11629 if (env->used_map_cnt >= MAX_USED_MAPS) {
11630 fdput(f);
11631 return -E2BIG;
11632 }
11633
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011634 /* hold the map. If the program is rejected by verifier,
11635 * the map will be released by release_maps() or it
11636 * will be used by the valid program until it's unloaded
Jakub Kicinskiab7f5bf2018-05-03 18:37:17 -070011637 * and all maps are released in free_used_maps()
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011638 */
Andrii Nakryiko1e0bd5a2019-11-17 09:28:02 -080011639 bpf_map_inc(map);
Daniel Borkmannd8eca5b2019-04-09 23:20:03 +020011640
11641 aux->map_index = env->used_map_cnt;
Alexei Starovoitov92117d82016-04-27 18:56:20 -070011642 env->used_maps[env->used_map_cnt++] = map;
11643
Roman Gushchinb741f162018-09-28 14:45:43 +000011644 if (bpf_map_is_cgroup_storage(map) &&
Daniel Borkmanne4730422019-12-17 13:28:16 +010011645 bpf_cgroup_storage_assign(env->prog->aux, map)) {
Roman Gushchinb741f162018-09-28 14:45:43 +000011646 verbose(env, "only one cgroup storage of each type is allowed\n");
Roman Gushchinde9cbba2018-08-02 14:27:18 -070011647 fdput(f);
11648 return -EBUSY;
11649 }
11650
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011651 fdput(f);
11652next_insn:
11653 insn++;
11654 i++;
Daniel Borkmann5e581da2018-01-26 23:33:38 +010011655 continue;
11656 }
11657
11658 /* Basic sanity check before we invest more work here. */
11659 if (!bpf_opcode_in_insntable(insn->code)) {
11660 verbose(env, "unknown opcode %02x\n", insn->code);
11661 return -EINVAL;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011662 }
11663 }
11664
11665 /* now all pseudo BPF_LD_IMM64 instructions load valid
11666 * 'struct bpf_map *' into a register instead of user map_fd.
11667 * These pointers will be used later by verifier to validate map access.
11668 */
11669 return 0;
11670}
11671
11672/* drop refcnt of maps used by the rejected program */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010011673static void release_maps(struct bpf_verifier_env *env)
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011674{
Daniel Borkmanna2ea0742019-12-16 17:49:00 +010011675 __bpf_free_used_maps(env->prog->aux, env->used_maps,
11676 env->used_map_cnt);
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011677}
11678
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080011679/* drop refcnt of maps used by the rejected program */
11680static void release_btfs(struct bpf_verifier_env *env)
11681{
11682 __bpf_free_used_btfs(env->prog->aux, env->used_btfs,
11683 env->used_btf_cnt);
11684}
11685
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011686/* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010011687static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011688{
11689 struct bpf_insn *insn = env->prog->insnsi;
11690 int insn_cnt = env->prog->len;
11691 int i;
11692
Yonghong Song69c087b2021-02-26 12:49:25 -080011693 for (i = 0; i < insn_cnt; i++, insn++) {
11694 if (insn->code != (BPF_LD | BPF_IMM | BPF_DW))
11695 continue;
11696 if (insn->src_reg == BPF_PSEUDO_FUNC)
11697 continue;
11698 insn->src_reg = 0;
11699 }
Alexei Starovoitov0246e642014-09-26 00:17:04 -070011700}
11701
Alexei Starovoitov80419022017-03-15 18:26:41 -070011702/* single env->prog->insni[off] instruction was replaced with the range
11703 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
11704 * [0, off) and [off, end) to new locations, so the patched range stays zero
11705 */
He Fengqing75f0fc72021-07-14 10:18:15 +000011706static void adjust_insn_aux_data(struct bpf_verifier_env *env,
11707 struct bpf_insn_aux_data *new_data,
11708 struct bpf_prog *new_prog, u32 off, u32 cnt)
Alexei Starovoitov80419022017-03-15 18:26:41 -070011709{
He Fengqing75f0fc72021-07-14 10:18:15 +000011710 struct bpf_insn_aux_data *old_data = env->insn_aux_data;
Jiong Wangb325fbc2019-05-24 23:25:13 +010011711 struct bpf_insn *insn = new_prog->insnsi;
Daniel Borkmannd203b0f2021-05-28 13:03:30 +000011712 u32 old_seen = old_data[off].seen;
Jiong Wangb325fbc2019-05-24 23:25:13 +010011713 u32 prog_len;
Alexei Starovoitovc1311872017-11-22 16:42:05 -080011714 int i;
Alexei Starovoitov80419022017-03-15 18:26:41 -070011715
Jiong Wangb325fbc2019-05-24 23:25:13 +010011716 /* aux info at OFF always needs adjustment, no matter fast path
11717 * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the
11718 * original insn at old prog.
11719 */
11720 old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1);
11721
Alexei Starovoitov80419022017-03-15 18:26:41 -070011722 if (cnt == 1)
He Fengqing75f0fc72021-07-14 10:18:15 +000011723 return;
Jiong Wangb325fbc2019-05-24 23:25:13 +010011724 prog_len = new_prog->len;
He Fengqing75f0fc72021-07-14 10:18:15 +000011725
Alexei Starovoitov80419022017-03-15 18:26:41 -070011726 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
11727 memcpy(new_data + off + cnt - 1, old_data + off,
11728 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
Jiong Wangb325fbc2019-05-24 23:25:13 +010011729 for (i = off; i < off + cnt - 1; i++) {
Daniel Borkmannd203b0f2021-05-28 13:03:30 +000011730 /* Expand insni[off]'s seen count to the patched range. */
11731 new_data[i].seen = old_seen;
Jiong Wangb325fbc2019-05-24 23:25:13 +010011732 new_data[i].zext_dst = insn_has_def32(env, insn + i);
11733 }
Alexei Starovoitov80419022017-03-15 18:26:41 -070011734 env->insn_aux_data = new_data;
11735 vfree(old_data);
Alexei Starovoitov80419022017-03-15 18:26:41 -070011736}
11737
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080011738static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len)
11739{
11740 int i;
11741
11742 if (len == 1)
11743 return;
Jiong Wang4cb3d992018-05-02 16:17:19 -040011744 /* NOTE: fake 'exit' subprog should be updated as well. */
11745 for (i = 0; i <= env->subprog_cnt; i++) {
Edward Creeafd59422018-11-16 12:00:07 +000011746 if (env->subprog_info[i].start <= off)
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080011747 continue;
Jiong Wang9c8105b2018-05-02 16:17:18 -040011748 env->subprog_info[i].start += len - 1;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080011749 }
11750}
11751
John Fastabend7506d212021-06-16 15:55:00 -070011752static void adjust_poke_descs(struct bpf_prog *prog, u32 off, u32 len)
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020011753{
11754 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
11755 int i, sz = prog->aux->size_poke_tab;
11756 struct bpf_jit_poke_descriptor *desc;
11757
11758 for (i = 0; i < sz; i++) {
11759 desc = &tab[i];
John Fastabend7506d212021-06-16 15:55:00 -070011760 if (desc->insn_idx <= off)
11761 continue;
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020011762 desc->insn_idx += len - 1;
11763 }
11764}
11765
Alexei Starovoitov80419022017-03-15 18:26:41 -070011766static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
11767 const struct bpf_insn *patch, u32 len)
11768{
11769 struct bpf_prog *new_prog;
He Fengqing75f0fc72021-07-14 10:18:15 +000011770 struct bpf_insn_aux_data *new_data = NULL;
11771
11772 if (len > 1) {
11773 new_data = vzalloc(array_size(env->prog->len + len - 1,
11774 sizeof(struct bpf_insn_aux_data)));
11775 if (!new_data)
11776 return NULL;
11777 }
Alexei Starovoitov80419022017-03-15 18:26:41 -070011778
11779 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
Alexei Starovoitov4f733792019-04-01 21:27:44 -070011780 if (IS_ERR(new_prog)) {
11781 if (PTR_ERR(new_prog) == -ERANGE)
11782 verbose(env,
11783 "insn %d cannot be patched due to 16-bit range\n",
11784 env->insn_aux_data[off].orig_idx);
He Fengqing75f0fc72021-07-14 10:18:15 +000011785 vfree(new_data);
Alexei Starovoitov80419022017-03-15 18:26:41 -070011786 return NULL;
Alexei Starovoitov4f733792019-04-01 21:27:44 -070011787 }
He Fengqing75f0fc72021-07-14 10:18:15 +000011788 adjust_insn_aux_data(env, new_data, new_prog, off, len);
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080011789 adjust_subprog_starts(env, off, len);
John Fastabend7506d212021-06-16 15:55:00 -070011790 adjust_poke_descs(new_prog, off, len);
Alexei Starovoitov80419022017-03-15 18:26:41 -070011791 return new_prog;
11792}
11793
Jakub Kicinski52875a02019-01-22 22:45:20 -080011794static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env,
11795 u32 off, u32 cnt)
11796{
11797 int i, j;
11798
11799 /* find first prog starting at or after off (first to remove) */
11800 for (i = 0; i < env->subprog_cnt; i++)
11801 if (env->subprog_info[i].start >= off)
11802 break;
11803 /* find first prog starting at or after off + cnt (first to stay) */
11804 for (j = i; j < env->subprog_cnt; j++)
11805 if (env->subprog_info[j].start >= off + cnt)
11806 break;
11807 /* if j doesn't start exactly at off + cnt, we are just removing
11808 * the front of previous prog
11809 */
11810 if (env->subprog_info[j].start != off + cnt)
11811 j--;
11812
11813 if (j > i) {
11814 struct bpf_prog_aux *aux = env->prog->aux;
11815 int move;
11816
11817 /* move fake 'exit' subprog as well */
11818 move = env->subprog_cnt + 1 - j;
11819
11820 memmove(env->subprog_info + i,
11821 env->subprog_info + j,
11822 sizeof(*env->subprog_info) * move);
11823 env->subprog_cnt -= j - i;
11824
11825 /* remove func_info */
11826 if (aux->func_info) {
11827 move = aux->func_info_cnt - j;
11828
11829 memmove(aux->func_info + i,
11830 aux->func_info + j,
11831 sizeof(*aux->func_info) * move);
11832 aux->func_info_cnt -= j - i;
11833 /* func_info->insn_off is set after all code rewrites,
11834 * in adjust_btf_func() - no need to adjust
11835 */
11836 }
11837 } else {
11838 /* convert i from "first prog to remove" to "first to adjust" */
11839 if (env->subprog_info[i].start == off)
11840 i++;
11841 }
11842
11843 /* update fake 'exit' subprog as well */
11844 for (; i <= env->subprog_cnt; i++)
11845 env->subprog_info[i].start -= cnt;
11846
11847 return 0;
11848}
11849
11850static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off,
11851 u32 cnt)
11852{
11853 struct bpf_prog *prog = env->prog;
11854 u32 i, l_off, l_cnt, nr_linfo;
11855 struct bpf_line_info *linfo;
11856
11857 nr_linfo = prog->aux->nr_linfo;
11858 if (!nr_linfo)
11859 return 0;
11860
11861 linfo = prog->aux->linfo;
11862
11863 /* find first line info to remove, count lines to be removed */
11864 for (i = 0; i < nr_linfo; i++)
11865 if (linfo[i].insn_off >= off)
11866 break;
11867
11868 l_off = i;
11869 l_cnt = 0;
11870 for (; i < nr_linfo; i++)
11871 if (linfo[i].insn_off < off + cnt)
11872 l_cnt++;
11873 else
11874 break;
11875
11876 /* First live insn doesn't match first live linfo, it needs to "inherit"
11877 * last removed linfo. prog is already modified, so prog->len == off
11878 * means no live instructions after (tail of the program was removed).
11879 */
11880 if (prog->len != off && l_cnt &&
11881 (i == nr_linfo || linfo[i].insn_off != off + cnt)) {
11882 l_cnt--;
11883 linfo[--i].insn_off = off + cnt;
11884 }
11885
11886 /* remove the line info which refer to the removed instructions */
11887 if (l_cnt) {
11888 memmove(linfo + l_off, linfo + i,
11889 sizeof(*linfo) * (nr_linfo - i));
11890
11891 prog->aux->nr_linfo -= l_cnt;
11892 nr_linfo = prog->aux->nr_linfo;
11893 }
11894
11895 /* pull all linfo[i].insn_off >= off + cnt in by cnt */
11896 for (i = l_off; i < nr_linfo; i++)
11897 linfo[i].insn_off -= cnt;
11898
11899 /* fix up all subprogs (incl. 'exit') which start >= off */
11900 for (i = 0; i <= env->subprog_cnt; i++)
11901 if (env->subprog_info[i].linfo_idx > l_off) {
11902 /* program may have started in the removed region but
11903 * may not be fully removed
11904 */
11905 if (env->subprog_info[i].linfo_idx >= l_off + l_cnt)
11906 env->subprog_info[i].linfo_idx -= l_cnt;
11907 else
11908 env->subprog_info[i].linfo_idx = l_off;
11909 }
11910
11911 return 0;
11912}
11913
11914static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt)
11915{
11916 struct bpf_insn_aux_data *aux_data = env->insn_aux_data;
11917 unsigned int orig_prog_len = env->prog->len;
11918 int err;
11919
Jakub Kicinski08ca90a2019-01-22 22:45:24 -080011920 if (bpf_prog_is_dev_bound(env->prog->aux))
11921 bpf_prog_offload_remove_insns(env, off, cnt);
11922
Jakub Kicinski52875a02019-01-22 22:45:20 -080011923 err = bpf_remove_insns(env->prog, off, cnt);
11924 if (err)
11925 return err;
11926
11927 err = adjust_subprog_starts_after_remove(env, off, cnt);
11928 if (err)
11929 return err;
11930
11931 err = bpf_adj_linfo_after_remove(env, off, cnt);
11932 if (err)
11933 return err;
11934
11935 memmove(aux_data + off, aux_data + off + cnt,
11936 sizeof(*aux_data) * (orig_prog_len - off - cnt));
11937
11938 return 0;
11939}
11940
Daniel Borkmann2a5418a2018-01-26 23:33:37 +010011941/* The verifier does more data flow analysis than llvm and will not
11942 * explore branches that are dead at run time. Malicious programs can
11943 * have dead code too. Therefore replace all dead at-run-time code
11944 * with 'ja -1'.
11945 *
11946 * Just nops are not optimal, e.g. if they would sit at the end of the
11947 * program and through another bug we would manage to jump there, then
11948 * we'd execute beyond program memory otherwise. Returning exception
11949 * code also wouldn't work since we can have subprogs where the dead
11950 * code could be located.
Alexei Starovoitovc1311872017-11-22 16:42:05 -080011951 */
11952static void sanitize_dead_code(struct bpf_verifier_env *env)
11953{
11954 struct bpf_insn_aux_data *aux_data = env->insn_aux_data;
Daniel Borkmann2a5418a2018-01-26 23:33:37 +010011955 struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1);
Alexei Starovoitovc1311872017-11-22 16:42:05 -080011956 struct bpf_insn *insn = env->prog->insnsi;
11957 const int insn_cnt = env->prog->len;
11958 int i;
11959
11960 for (i = 0; i < insn_cnt; i++) {
11961 if (aux_data[i].seen)
11962 continue;
Daniel Borkmann2a5418a2018-01-26 23:33:37 +010011963 memcpy(insn + i, &trap, sizeof(trap));
Ilya Leoshkevich45c709f2021-08-12 17:18:10 +020011964 aux_data[i].zext_dst = false;
Alexei Starovoitovc1311872017-11-22 16:42:05 -080011965 }
11966}
11967
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080011968static bool insn_is_cond_jump(u8 code)
11969{
11970 u8 op;
11971
Jiong Wang092ed092019-01-26 12:26:01 -050011972 if (BPF_CLASS(code) == BPF_JMP32)
11973 return true;
11974
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080011975 if (BPF_CLASS(code) != BPF_JMP)
11976 return false;
11977
11978 op = BPF_OP(code);
11979 return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL;
11980}
11981
11982static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env)
11983{
11984 struct bpf_insn_aux_data *aux_data = env->insn_aux_data;
11985 struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0);
11986 struct bpf_insn *insn = env->prog->insnsi;
11987 const int insn_cnt = env->prog->len;
11988 int i;
11989
11990 for (i = 0; i < insn_cnt; i++, insn++) {
11991 if (!insn_is_cond_jump(insn->code))
11992 continue;
11993
11994 if (!aux_data[i + 1].seen)
11995 ja.off = insn->off;
11996 else if (!aux_data[i + 1 + insn->off].seen)
11997 ja.off = 0;
11998 else
11999 continue;
12000
Jakub Kicinski08ca90a2019-01-22 22:45:24 -080012001 if (bpf_prog_is_dev_bound(env->prog->aux))
12002 bpf_prog_offload_replace_insn(env, i, &ja);
12003
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080012004 memcpy(insn, &ja, sizeof(ja));
12005 }
12006}
12007
Jakub Kicinski52875a02019-01-22 22:45:20 -080012008static int opt_remove_dead_code(struct bpf_verifier_env *env)
12009{
12010 struct bpf_insn_aux_data *aux_data = env->insn_aux_data;
12011 int insn_cnt = env->prog->len;
12012 int i, err;
12013
12014 for (i = 0; i < insn_cnt; i++) {
12015 int j;
12016
12017 j = 0;
12018 while (i + j < insn_cnt && !aux_data[i + j].seen)
12019 j++;
12020 if (!j)
12021 continue;
12022
12023 err = verifier_remove_insns(env, i, j);
12024 if (err)
12025 return err;
12026 insn_cnt = env->prog->len;
12027 }
12028
12029 return 0;
12030}
12031
Jakub Kicinskia1b14ab2019-01-22 22:45:21 -080012032static int opt_remove_nops(struct bpf_verifier_env *env)
12033{
12034 const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0);
12035 struct bpf_insn *insn = env->prog->insnsi;
12036 int insn_cnt = env->prog->len;
12037 int i, err;
12038
12039 for (i = 0; i < insn_cnt; i++) {
12040 if (memcmp(&insn[i], &ja, sizeof(ja)))
12041 continue;
12042
12043 err = verifier_remove_insns(env, i, 1);
12044 if (err)
12045 return err;
12046 insn_cnt--;
12047 i--;
12048 }
12049
12050 return 0;
12051}
12052
Jiong Wangd6c23082019-05-24 23:25:18 +010012053static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
12054 const union bpf_attr *attr)
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012055{
Jiong Wangd6c23082019-05-24 23:25:18 +010012056 struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4];
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012057 struct bpf_insn_aux_data *aux = env->insn_aux_data;
Jiong Wangd6c23082019-05-24 23:25:18 +010012058 int i, patch_len, delta = 0, len = env->prog->len;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012059 struct bpf_insn *insns = env->prog->insnsi;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012060 struct bpf_prog *new_prog;
Jiong Wangd6c23082019-05-24 23:25:18 +010012061 bool rnd_hi32;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012062
Jiong Wangd6c23082019-05-24 23:25:18 +010012063 rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012064 zext_patch[1] = BPF_ZEXT_REG(0);
Jiong Wangd6c23082019-05-24 23:25:18 +010012065 rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0);
12066 rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
12067 rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX);
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012068 for (i = 0; i < len; i++) {
12069 int adj_idx = i + delta;
12070 struct bpf_insn insn;
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012071 int load_reg;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012072
Jiong Wangd6c23082019-05-24 23:25:18 +010012073 insn = insns[adj_idx];
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012074 load_reg = insn_def_regno(&insn);
Jiong Wangd6c23082019-05-24 23:25:18 +010012075 if (!aux[adj_idx].zext_dst) {
12076 u8 code, class;
12077 u32 imm_rnd;
12078
12079 if (!rnd_hi32)
12080 continue;
12081
12082 code = insn.code;
12083 class = BPF_CLASS(code);
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012084 if (load_reg == -1)
Jiong Wangd6c23082019-05-24 23:25:18 +010012085 continue;
12086
12087 /* NOTE: arg "reg" (the fourth one) is only used for
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012088 * BPF_STX + SRC_OP, so it is safe to pass NULL
12089 * here.
Jiong Wangd6c23082019-05-24 23:25:18 +010012090 */
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012091 if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) {
Jiong Wangd6c23082019-05-24 23:25:18 +010012092 if (class == BPF_LD &&
12093 BPF_MODE(code) == BPF_IMM)
12094 i++;
12095 continue;
12096 }
12097
12098 /* ctx load could be transformed into wider load. */
12099 if (class == BPF_LDX &&
12100 aux[adj_idx].ptr_type == PTR_TO_CTX)
12101 continue;
12102
12103 imm_rnd = get_random_int();
12104 rnd_hi32_patch[0] = insn;
12105 rnd_hi32_patch[1].imm = imm_rnd;
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012106 rnd_hi32_patch[3].dst_reg = load_reg;
Jiong Wangd6c23082019-05-24 23:25:18 +010012107 patch = rnd_hi32_patch;
12108 patch_len = 4;
12109 goto apply_patch_buffer;
12110 }
12111
Brendan Jackman39491862021-03-04 18:56:46 -080012112 /* Add in an zero-extend instruction if a) the JIT has requested
12113 * it or b) it's a CMPXCHG.
12114 *
12115 * The latter is because: BPF_CMPXCHG always loads a value into
12116 * R0, therefore always zero-extends. However some archs'
12117 * equivalent instruction only does this load when the
12118 * comparison is successful. This detail of CMPXCHG is
12119 * orthogonal to the general zero-extension behaviour of the
12120 * CPU, so it's treated independently of bpf_jit_needs_zext.
12121 */
12122 if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn))
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012123 continue;
12124
Ilya Leoshkevich83a28812021-03-01 16:40:19 +010012125 if (WARN_ON(load_reg == -1)) {
12126 verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n");
12127 return -EFAULT;
Ilya Leoshkevichb2e37a72021-02-10 21:45:02 +010012128 }
12129
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012130 zext_patch[0] = insn;
Ilya Leoshkevichb2e37a72021-02-10 21:45:02 +010012131 zext_patch[1].dst_reg = load_reg;
12132 zext_patch[1].src_reg = load_reg;
Jiong Wangd6c23082019-05-24 23:25:18 +010012133 patch = zext_patch;
12134 patch_len = 2;
12135apply_patch_buffer:
12136 new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len);
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012137 if (!new_prog)
12138 return -ENOMEM;
12139 env->prog = new_prog;
12140 insns = new_prog->insnsi;
12141 aux = env->insn_aux_data;
Jiong Wangd6c23082019-05-24 23:25:18 +010012142 delta += patch_len - 1;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010012143 }
12144
12145 return 0;
12146}
12147
Joe Stringerc64b7982018-10-02 13:35:33 -070012148/* convert load instructions that access fields of a context type into a
12149 * sequence of instructions that access fields of the underlying structure:
12150 * struct __sk_buff -> struct sk_buff
12151 * struct bpf_sock_ops -> struct sock
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012152 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010012153static int convert_ctx_accesses(struct bpf_verifier_env *env)
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012154{
Jakub Kicinski00176a32017-10-16 16:40:54 -070012155 const struct bpf_verifier_ops *ops = env->ops;
Daniel Borkmannf96da092017-07-02 02:13:27 +020012156 int i, cnt, size, ctx_field_size, delta = 0;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010012157 const int insn_cnt = env->prog->len;
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012158 struct bpf_insn insn_buf[16], *insn;
Andrey Ignatov46f53a62018-11-10 22:15:13 -080012159 u32 target_size, size_default, off;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012160 struct bpf_prog *new_prog;
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070012161 enum bpf_access_type type;
Daniel Borkmannf96da092017-07-02 02:13:27 +020012162 bool is_narrower_load;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012163
Daniel Borkmannb09928b2018-10-24 22:05:49 +020012164 if (ops->gen_prologue || env->seen_direct_write) {
12165 if (!ops->gen_prologue) {
12166 verbose(env, "bpf verifier is misconfigured\n");
12167 return -EINVAL;
12168 }
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012169 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
12170 env->prog);
12171 if (cnt >= ARRAY_SIZE(insn_buf)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070012172 verbose(env, "bpf verifier is misconfigured\n");
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012173 return -EINVAL;
12174 } else if (cnt) {
Alexei Starovoitov80419022017-03-15 18:26:41 -070012175 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012176 if (!new_prog)
12177 return -ENOMEM;
Alexei Starovoitov80419022017-03-15 18:26:41 -070012178
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012179 env->prog = new_prog;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010012180 delta += cnt - 1;
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012181 }
12182 }
12183
Joe Stringerc64b7982018-10-02 13:35:33 -070012184 if (bpf_prog_is_dev_bound(env->prog->aux))
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012185 return 0;
12186
Jakub Kicinski3df126f2016-09-21 11:43:56 +010012187 insn = env->prog->insnsi + delta;
Daniel Borkmann36bbef52016-09-20 00:26:13 +020012188
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012189 for (i = 0; i < insn_cnt; i++, insn++) {
Joe Stringerc64b7982018-10-02 13:35:33 -070012190 bpf_convert_ctx_access_t convert_ctx_access;
Daniel Borkmann2039f262021-07-13 08:18:31 +000012191 bool ctx_access;
Joe Stringerc64b7982018-10-02 13:35:33 -070012192
Daniel Borkmann62c79892017-01-12 11:51:33 +010012193 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
12194 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
12195 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
Daniel Borkmann2039f262021-07-13 08:18:31 +000012196 insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) {
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070012197 type = BPF_READ;
Daniel Borkmann2039f262021-07-13 08:18:31 +000012198 ctx_access = true;
12199 } else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
12200 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
12201 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
12202 insn->code == (BPF_STX | BPF_MEM | BPF_DW) ||
12203 insn->code == (BPF_ST | BPF_MEM | BPF_B) ||
12204 insn->code == (BPF_ST | BPF_MEM | BPF_H) ||
12205 insn->code == (BPF_ST | BPF_MEM | BPF_W) ||
12206 insn->code == (BPF_ST | BPF_MEM | BPF_DW)) {
Alexei Starovoitovd691f9e2015-06-04 10:11:54 -070012207 type = BPF_WRITE;
Daniel Borkmann2039f262021-07-13 08:18:31 +000012208 ctx_access = BPF_CLASS(insn->code) == BPF_STX;
12209 } else {
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012210 continue;
Daniel Borkmann2039f262021-07-13 08:18:31 +000012211 }
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012212
Alexei Starovoitovaf86ca42018-05-15 09:27:05 -070012213 if (type == BPF_WRITE &&
Daniel Borkmann2039f262021-07-13 08:18:31 +000012214 env->insn_aux_data[i + delta].sanitize_stack_spill) {
Alexei Starovoitovaf86ca42018-05-15 09:27:05 -070012215 struct bpf_insn patch[] = {
Alexei Starovoitovaf86ca42018-05-15 09:27:05 -070012216 *insn,
Daniel Borkmann2039f262021-07-13 08:18:31 +000012217 BPF_ST_NOSPEC(),
Alexei Starovoitovaf86ca42018-05-15 09:27:05 -070012218 };
12219
12220 cnt = ARRAY_SIZE(patch);
12221 new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt);
12222 if (!new_prog)
12223 return -ENOMEM;
12224
12225 delta += cnt - 1;
12226 env->prog = new_prog;
12227 insn = new_prog->insnsi + i + delta;
12228 continue;
12229 }
12230
Daniel Borkmann2039f262021-07-13 08:18:31 +000012231 if (!ctx_access)
12232 continue;
12233
Joe Stringerc64b7982018-10-02 13:35:33 -070012234 switch (env->insn_aux_data[i + delta].ptr_type) {
12235 case PTR_TO_CTX:
12236 if (!ops->convert_ctx_access)
12237 continue;
12238 convert_ctx_access = ops->convert_ctx_access;
12239 break;
12240 case PTR_TO_SOCKET:
Martin KaFai Lau46f8bc92019-02-09 23:22:20 -080012241 case PTR_TO_SOCK_COMMON:
Joe Stringerc64b7982018-10-02 13:35:33 -070012242 convert_ctx_access = bpf_sock_convert_ctx_access;
12243 break;
Martin KaFai Lau655a51e2019-02-09 23:22:24 -080012244 case PTR_TO_TCP_SOCK:
12245 convert_ctx_access = bpf_tcp_sock_convert_ctx_access;
12246 break;
Jonathan Lemonfada7fd2019-06-06 13:59:40 -070012247 case PTR_TO_XDP_SOCK:
12248 convert_ctx_access = bpf_xdp_sock_convert_ctx_access;
12249 break;
Alexei Starovoitov2a027592019-10-15 20:25:02 -070012250 case PTR_TO_BTF_ID:
Martin KaFai Lau27ae79972020-01-08 16:35:03 -080012251 if (type == BPF_READ) {
12252 insn->code = BPF_LDX | BPF_PROBE_MEM |
12253 BPF_SIZE((insn)->code);
12254 env->prog->aux->num_exentries++;
Udip Pant7e407812020-08-25 16:20:00 -070012255 } else if (resolve_prog_type(env->prog) != BPF_PROG_TYPE_STRUCT_OPS) {
Alexei Starovoitov2a027592019-10-15 20:25:02 -070012256 verbose(env, "Writes through BTF pointers are not allowed\n");
12257 return -EINVAL;
12258 }
Alexei Starovoitov2a027592019-10-15 20:25:02 -070012259 continue;
Joe Stringerc64b7982018-10-02 13:35:33 -070012260 default:
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012261 continue;
Joe Stringerc64b7982018-10-02 13:35:33 -070012262 }
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012263
Yonghong Song31fd8582017-06-13 15:52:13 -070012264 ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size;
Daniel Borkmannf96da092017-07-02 02:13:27 +020012265 size = BPF_LDST_BYTES(insn);
Yonghong Song31fd8582017-06-13 15:52:13 -070012266
12267 /* If the read access is a narrower load of the field,
12268 * convert to a 4/8-byte load, to minimum program type specific
12269 * convert_ctx_access changes. If conversion is successful,
12270 * we will apply proper mask to the result.
12271 */
Daniel Borkmannf96da092017-07-02 02:13:27 +020012272 is_narrower_load = size < ctx_field_size;
Andrey Ignatov46f53a62018-11-10 22:15:13 -080012273 size_default = bpf_ctx_off_adjust_machine(ctx_field_size);
12274 off = insn->off;
Yonghong Song31fd8582017-06-13 15:52:13 -070012275 if (is_narrower_load) {
Daniel Borkmannf96da092017-07-02 02:13:27 +020012276 u8 size_code;
Yonghong Song31fd8582017-06-13 15:52:13 -070012277
Daniel Borkmannf96da092017-07-02 02:13:27 +020012278 if (type == BPF_WRITE) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070012279 verbose(env, "bpf verifier narrow ctx access misconfigured\n");
Daniel Borkmannf96da092017-07-02 02:13:27 +020012280 return -EINVAL;
12281 }
12282
12283 size_code = BPF_H;
Yonghong Song31fd8582017-06-13 15:52:13 -070012284 if (ctx_field_size == 4)
12285 size_code = BPF_W;
12286 else if (ctx_field_size == 8)
12287 size_code = BPF_DW;
Daniel Borkmannf96da092017-07-02 02:13:27 +020012288
Daniel Borkmannbc231052018-06-02 23:06:39 +020012289 insn->off = off & ~(size_default - 1);
Yonghong Song31fd8582017-06-13 15:52:13 -070012290 insn->code = BPF_LDX | BPF_MEM | size_code;
12291 }
Daniel Borkmannf96da092017-07-02 02:13:27 +020012292
12293 target_size = 0;
Joe Stringerc64b7982018-10-02 13:35:33 -070012294 cnt = convert_ctx_access(type, insn, insn_buf, env->prog,
12295 &target_size);
Daniel Borkmannf96da092017-07-02 02:13:27 +020012296 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) ||
12297 (ctx_field_size && !target_size)) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070012298 verbose(env, "bpf verifier is misconfigured\n");
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012299 return -EINVAL;
12300 }
Daniel Borkmannf96da092017-07-02 02:13:27 +020012301
12302 if (is_narrower_load && size < target_size) {
Ilya Leoshkevichd895a0f2019-08-16 12:53:00 +020012303 u8 shift = bpf_ctx_narrow_access_offset(
12304 off, size, size_default) * 8;
Andrey Ignatovd7af7e42021-08-20 09:39:35 -070012305 if (shift && cnt + 1 >= ARRAY_SIZE(insn_buf)) {
12306 verbose(env, "bpf verifier narrow ctx load misconfigured\n");
12307 return -EINVAL;
12308 }
Andrey Ignatov46f53a62018-11-10 22:15:13 -080012309 if (ctx_field_size <= 4) {
12310 if (shift)
12311 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH,
12312 insn->dst_reg,
12313 shift);
Yonghong Song31fd8582017-06-13 15:52:13 -070012314 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg,
Daniel Borkmannf96da092017-07-02 02:13:27 +020012315 (1 << size * 8) - 1);
Andrey Ignatov46f53a62018-11-10 22:15:13 -080012316 } else {
12317 if (shift)
12318 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH,
12319 insn->dst_reg,
12320 shift);
Yonghong Song31fd8582017-06-13 15:52:13 -070012321 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg,
Krzesimir Nowake2f7fc02019-05-08 18:08:58 +020012322 (1ULL << size * 8) - 1);
Andrey Ignatov46f53a62018-11-10 22:15:13 -080012323 }
Yonghong Song31fd8582017-06-13 15:52:13 -070012324 }
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012325
Alexei Starovoitov80419022017-03-15 18:26:41 -070012326 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012327 if (!new_prog)
12328 return -ENOMEM;
12329
Jakub Kicinski3df126f2016-09-21 11:43:56 +010012330 delta += cnt - 1;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012331
12332 /* keep walking new program and skip insns we just inserted */
12333 env->prog = new_prog;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010012334 insn = new_prog->insnsi + i + delta;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070012335 }
12336
12337 return 0;
12338}
12339
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012340static int jit_subprogs(struct bpf_verifier_env *env)
12341{
12342 struct bpf_prog *prog = env->prog, **func, *tmp;
12343 int i, j, subprog_start, subprog_end = 0, len, subprog;
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012344 struct bpf_map *map_ptr;
Daniel Borkmann7105e822017-12-20 13:42:57 +010012345 struct bpf_insn *insn;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012346 void *old_bpf_func;
Yonghong Songc4c0bdc2020-06-23 17:10:54 -070012347 int err, num_exentries;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012348
Jiong Wangf910cef2018-05-02 16:17:17 -040012349 if (env->subprog_cnt <= 1)
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012350 return 0;
12351
Daniel Borkmann7105e822017-12-20 13:42:57 +010012352 for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
Yonghong Song69c087b2021-02-26 12:49:25 -080012353 if (bpf_pseudo_func(insn)) {
12354 env->insn_aux_data[i].call_imm = insn->imm;
12355 /* subprog is encoded in insn[1].imm */
12356 continue;
12357 }
12358
Yonghong Song23a2d702021-02-04 15:48:27 -080012359 if (!bpf_pseudo_call(insn))
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012360 continue;
Daniel Borkmannc7a89782018-07-12 21:44:28 +020012361 /* Upon error here we cannot fall back to interpreter but
12362 * need a hard reject of the program. Thus -EFAULT is
12363 * propagated in any case.
12364 */
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012365 subprog = find_subprog(env, i + insn->imm + 1);
12366 if (subprog < 0) {
12367 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
12368 i + insn->imm + 1);
12369 return -EFAULT;
12370 }
12371 /* temporarily remember subprog id inside insn instead of
12372 * aux_data, since next loop will split up all insns into funcs
12373 */
Jiong Wangf910cef2018-05-02 16:17:17 -040012374 insn->off = subprog;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012375 /* remember original imm in case JIT fails and fallback
12376 * to interpreter will be needed
12377 */
12378 env->insn_aux_data[i].call_imm = insn->imm;
12379 /* point imm to __bpf_call_base+1 from JITs point of view */
12380 insn->imm = 1;
12381 }
12382
Martin KaFai Lauc454a462018-12-07 16:42:25 -080012383 err = bpf_prog_alloc_jited_linfo(prog);
12384 if (err)
12385 goto out_undo_insn;
12386
12387 err = -ENOMEM;
Kees Cook6396bb22018-06-12 14:03:40 -070012388 func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL);
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012389 if (!func)
Daniel Borkmannc7a89782018-07-12 21:44:28 +020012390 goto out_undo_insn;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012391
Jiong Wangf910cef2018-05-02 16:17:17 -040012392 for (i = 0; i < env->subprog_cnt; i++) {
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012393 subprog_start = subprog_end;
Jiong Wang4cb3d992018-05-02 16:17:19 -040012394 subprog_end = env->subprog_info[i + 1].start;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012395
12396 len = subprog_end - subprog_start;
Andrii Nakryikofb7dd8b2021-08-15 00:05:54 -070012397 /* bpf_prog_run() doesn't call subprogs directly,
Alexei Starovoitov492ecee2019-02-25 14:28:39 -080012398 * hence main prog stats include the runtime of subprogs.
12399 * subprogs don't have IDs and not reachable via prog_get_next_id
Alexei Starovoitov700d4792021-02-09 19:36:26 -080012400 * func[i]->stats will never be accessed and stays NULL
Alexei Starovoitov492ecee2019-02-25 14:28:39 -080012401 */
12402 func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER);
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012403 if (!func[i])
12404 goto out_free;
12405 memcpy(func[i]->insnsi, &prog->insnsi[subprog_start],
12406 len * sizeof(struct bpf_insn));
Daniel Borkmann4f74d802017-12-20 13:42:56 +010012407 func[i]->type = prog->type;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012408 func[i]->len = len;
Daniel Borkmann4f74d802017-12-20 13:42:56 +010012409 if (bpf_prog_calc_tag(func[i]))
12410 goto out_free;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012411 func[i]->is_func = 1;
Yonghong Songba64e7d2018-11-24 23:20:44 -080012412 func[i]->aux->func_idx = i;
John Fastabendf263a812021-07-07 15:38:47 -070012413 /* Below members will be freed only at prog->aux */
Yonghong Songba64e7d2018-11-24 23:20:44 -080012414 func[i]->aux->btf = prog->aux->btf;
12415 func[i]->aux->func_info = prog->aux->func_info;
John Fastabendf263a812021-07-07 15:38:47 -070012416 func[i]->aux->poke_tab = prog->aux->poke_tab;
12417 func[i]->aux->size_poke_tab = prog->aux->size_poke_tab;
Yonghong Songba64e7d2018-11-24 23:20:44 -080012418
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012419 for (j = 0; j < prog->aux->size_poke_tab; j++) {
John Fastabendf263a812021-07-07 15:38:47 -070012420 struct bpf_jit_poke_descriptor *poke;
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012421
John Fastabendf263a812021-07-07 15:38:47 -070012422 poke = &prog->aux->poke_tab[j];
12423 if (poke->insn_idx < subprog_end &&
12424 poke->insn_idx >= subprog_start)
12425 poke->aux = func[i]->aux;
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012426 }
12427
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012428 /* Use bpf_prog_F_tag to indicate functions in stack traces.
12429 * Long term would need debug info to populate names
12430 */
12431 func[i]->aux->name[0] = 'F';
Jiong Wang9c8105b2018-05-02 16:17:18 -040012432 func[i]->aux->stack_depth = env->subprog_info[i].stack_depth;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012433 func[i]->jit_requested = 1;
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070012434 func[i]->aux->kfunc_tab = prog->aux->kfunc_tab;
Martin KaFai Lauc454a462018-12-07 16:42:25 -080012435 func[i]->aux->linfo = prog->aux->linfo;
12436 func[i]->aux->nr_linfo = prog->aux->nr_linfo;
12437 func[i]->aux->jited_linfo = prog->aux->jited_linfo;
12438 func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx;
Yonghong Songc4c0bdc2020-06-23 17:10:54 -070012439 num_exentries = 0;
12440 insn = func[i]->insnsi;
12441 for (j = 0; j < func[i]->len; j++, insn++) {
12442 if (BPF_CLASS(insn->code) == BPF_LDX &&
12443 BPF_MODE(insn->code) == BPF_PROBE_MEM)
12444 num_exentries++;
12445 }
12446 func[i]->aux->num_exentries = num_exentries;
Maciej Fijalkowskiebf7d1f2020-09-16 23:10:08 +020012447 func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012448 func[i] = bpf_int_jit_compile(func[i]);
12449 if (!func[i]->jited) {
12450 err = -ENOTSUPP;
12451 goto out_free;
12452 }
12453 cond_resched();
12454 }
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012455
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012456 /* at this point all bpf functions were successfully JITed
12457 * now populate all bpf_calls with correct addresses and
12458 * run last pass of JIT
12459 */
Jiong Wangf910cef2018-05-02 16:17:17 -040012460 for (i = 0; i < env->subprog_cnt; i++) {
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012461 insn = func[i]->insnsi;
12462 for (j = 0; j < func[i]->len; j++, insn++) {
Yonghong Song69c087b2021-02-26 12:49:25 -080012463 if (bpf_pseudo_func(insn)) {
12464 subprog = insn[1].imm;
12465 insn[0].imm = (u32)(long)func[subprog]->bpf_func;
12466 insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32;
12467 continue;
12468 }
Yonghong Song23a2d702021-02-04 15:48:27 -080012469 if (!bpf_pseudo_call(insn))
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012470 continue;
12471 subprog = insn->off;
Prashant Bhole0d306c32019-04-16 18:13:01 +090012472 insn->imm = BPF_CAST_CALL(func[subprog]->bpf_func) -
12473 __bpf_call_base;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012474 }
Sandipan Das2162fed2018-05-24 12:26:45 +053012475
12476 /* we use the aux data to keep a list of the start addresses
12477 * of the JITed images for each function in the program
12478 *
12479 * for some architectures, such as powerpc64, the imm field
12480 * might not be large enough to hold the offset of the start
12481 * address of the callee's JITed image from __bpf_call_base
12482 *
12483 * in such cases, we can lookup the start address of a callee
12484 * by using its subprog id, available from the off field of
12485 * the call instruction, as an index for this list
12486 */
12487 func[i]->aux->func = func;
12488 func[i]->aux->func_cnt = env->subprog_cnt;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012489 }
Jiong Wangf910cef2018-05-02 16:17:17 -040012490 for (i = 0; i < env->subprog_cnt; i++) {
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012491 old_bpf_func = func[i]->bpf_func;
12492 tmp = bpf_int_jit_compile(func[i]);
12493 if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) {
12494 verbose(env, "JIT doesn't support bpf-to-bpf calls\n");
Daniel Borkmannc7a89782018-07-12 21:44:28 +020012495 err = -ENOTSUPP;
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012496 goto out_free;
12497 }
12498 cond_resched();
12499 }
12500
12501 /* finally lock prog and jit images for all functions and
12502 * populate kallsysm
12503 */
Jiong Wangf910cef2018-05-02 16:17:17 -040012504 for (i = 0; i < env->subprog_cnt; i++) {
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012505 bpf_prog_lock_ro(func[i]);
12506 bpf_prog_kallsyms_add(func[i]);
12507 }
Daniel Borkmann7105e822017-12-20 13:42:57 +010012508
12509 /* Last step: make now unused interpreter insns from main
12510 * prog consistent for later dump requests, so they can
12511 * later look the same as if they were interpreted only.
12512 */
12513 for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
Yonghong Song69c087b2021-02-26 12:49:25 -080012514 if (bpf_pseudo_func(insn)) {
12515 insn[0].imm = env->insn_aux_data[i].call_imm;
12516 insn[1].imm = find_subprog(env, i + insn[0].imm + 1);
12517 continue;
12518 }
Yonghong Song23a2d702021-02-04 15:48:27 -080012519 if (!bpf_pseudo_call(insn))
Daniel Borkmann7105e822017-12-20 13:42:57 +010012520 continue;
12521 insn->off = env->insn_aux_data[i].call_imm;
12522 subprog = find_subprog(env, i + insn->off + 1);
Sandipan Dasdbecd732018-05-24 12:26:48 +053012523 insn->imm = subprog;
Daniel Borkmann7105e822017-12-20 13:42:57 +010012524 }
12525
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012526 prog->jited = 1;
12527 prog->bpf_func = func[0]->bpf_func;
12528 prog->aux->func = func;
Jiong Wangf910cef2018-05-02 16:17:17 -040012529 prog->aux->func_cnt = env->subprog_cnt;
Martin KaFai Laue16301f2021-03-24 18:51:30 -070012530 bpf_prog_jit_attempt_done(prog);
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012531 return 0;
12532out_free:
John Fastabendf263a812021-07-07 15:38:47 -070012533 /* We failed JIT'ing, so at this point we need to unregister poke
12534 * descriptors from subprogs, so that kernel is not attempting to
12535 * patch it anymore as we're freeing the subprog JIT memory.
12536 */
12537 for (i = 0; i < prog->aux->size_poke_tab; i++) {
12538 map_ptr = prog->aux->poke_tab[i].tail_call.map;
12539 map_ptr->ops->map_poke_untrack(map_ptr, prog->aux);
12540 }
12541 /* At this point we're guaranteed that poke descriptors are not
12542 * live anymore. We can just unlink its descriptor table as it's
12543 * released with the main prog.
12544 */
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012545 for (i = 0; i < env->subprog_cnt; i++) {
12546 if (!func[i])
12547 continue;
John Fastabendf263a812021-07-07 15:38:47 -070012548 func[i]->aux->poke_tab = NULL;
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012549 bpf_jit_free(func[i]);
12550 }
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012551 kfree(func);
Daniel Borkmannc7a89782018-07-12 21:44:28 +020012552out_undo_insn:
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012553 /* cleanup main prog to be interpreted */
12554 prog->jit_requested = 0;
12555 for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) {
Yonghong Song23a2d702021-02-04 15:48:27 -080012556 if (!bpf_pseudo_call(insn))
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012557 continue;
12558 insn->off = 0;
12559 insn->imm = env->insn_aux_data[i].call_imm;
12560 }
Martin KaFai Laue16301f2021-03-24 18:51:30 -070012561 bpf_prog_jit_attempt_done(prog);
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012562 return err;
12563}
12564
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012565static int fixup_call_args(struct bpf_verifier_env *env)
12566{
David S. Miller19d28fb2018-01-11 21:27:54 -050012567#ifndef CONFIG_BPF_JIT_ALWAYS_ON
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012568 struct bpf_prog *prog = env->prog;
12569 struct bpf_insn *insn = prog->insnsi;
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070012570 bool has_kfunc_call = bpf_prog_has_kfunc_call(prog);
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012571 int i, depth;
David S. Miller19d28fb2018-01-11 21:27:54 -050012572#endif
Quentin Monnete4052d02018-10-07 12:56:58 +010012573 int err = 0;
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012574
Quentin Monnete4052d02018-10-07 12:56:58 +010012575 if (env->prog->jit_requested &&
12576 !bpf_prog_is_dev_bound(env->prog->aux)) {
David S. Miller19d28fb2018-01-11 21:27:54 -050012577 err = jit_subprogs(env);
12578 if (err == 0)
Alexei Starovoitov1c2a0882017-12-14 17:55:15 -080012579 return 0;
Daniel Borkmannc7a89782018-07-12 21:44:28 +020012580 if (err == -EFAULT)
12581 return err;
David S. Miller19d28fb2018-01-11 21:27:54 -050012582 }
12583#ifndef CONFIG_BPF_JIT_ALWAYS_ON
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070012584 if (has_kfunc_call) {
12585 verbose(env, "calling kernel functions are not allowed in non-JITed programs\n");
12586 return -EINVAL;
12587 }
Maciej Fijalkowskie4119012020-09-16 23:10:09 +020012588 if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) {
12589 /* When JIT fails the progs with bpf2bpf calls and tail_calls
12590 * have to be rejected, since interpreter doesn't support them yet.
12591 */
12592 verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
12593 return -EINVAL;
12594 }
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012595 for (i = 0; i < prog->len; i++, insn++) {
Yonghong Song69c087b2021-02-26 12:49:25 -080012596 if (bpf_pseudo_func(insn)) {
12597 /* When JIT fails the progs with callback calls
12598 * have to be rejected, since interpreter doesn't support them yet.
12599 */
12600 verbose(env, "callbacks are not allowed in non-JITed programs\n");
12601 return -EINVAL;
12602 }
12603
Yonghong Song23a2d702021-02-04 15:48:27 -080012604 if (!bpf_pseudo_call(insn))
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012605 continue;
12606 depth = get_callee_stack_depth(env, insn, i);
12607 if (depth < 0)
12608 return depth;
12609 bpf_patch_call_args(insn, depth);
12610 }
David S. Miller19d28fb2018-01-11 21:27:54 -050012611 err = 0;
12612#endif
12613 return err;
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080012614}
12615
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070012616static int fixup_kfunc_call(struct bpf_verifier_env *env,
12617 struct bpf_insn *insn)
12618{
12619 const struct bpf_kfunc_desc *desc;
12620
12621 /* insn->imm has the btf func_id. Replace it with
12622 * an address (relative to __bpf_base_call).
12623 */
12624 desc = find_kfunc_desc(env->prog, insn->imm);
12625 if (!desc) {
12626 verbose(env, "verifier internal error: kernel function descriptor not found for func_id %u\n",
12627 insn->imm);
12628 return -EFAULT;
12629 }
12630
12631 insn->imm = desc->imm;
12632
12633 return 0;
12634}
12635
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012636/* Do various post-verification rewrites in a single program pass.
12637 * These rewrites simplify JIT and interpreter implementations.
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012638 */
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012639static int do_misc_fixups(struct bpf_verifier_env *env)
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012640{
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012641 struct bpf_prog *prog = env->prog;
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012642 bool expect_blinding = bpf_jit_blinding_enabled(prog);
Jiri Olsa9b99edc2021-07-14 11:43:55 +020012643 enum bpf_prog_type prog_type = resolve_prog_type(prog);
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012644 struct bpf_insn *insn = prog->insnsi;
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012645 const struct bpf_func_proto *fn;
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012646 const int insn_cnt = prog->len;
Daniel Borkmann09772d92018-06-02 23:06:35 +020012647 const struct bpf_map_ops *ops;
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012648 struct bpf_insn_aux_data *aux;
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070012649 struct bpf_insn insn_buf[16];
12650 struct bpf_prog *new_prog;
12651 struct bpf_map *map_ptr;
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012652 int i, ret, cnt, delta = 0;
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012653
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012654 for (i = 0; i < insn_cnt; i++, insn++) {
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012655 /* Make divide-by-zero exceptions impossible. */
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012656 if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) ||
12657 insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) ||
12658 insn->code == (BPF_ALU | BPF_MOD | BPF_X) ||
Alexei Starovoitov68fda452018-01-12 18:59:52 -080012659 insn->code == (BPF_ALU | BPF_DIV | BPF_X)) {
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012660 bool is64 = BPF_CLASS(insn->code) == BPF_ALU64;
Daniel Borkmanne88b2c62021-02-09 18:46:10 +000012661 bool isdiv = BPF_OP(insn->code) == BPF_DIV;
12662 struct bpf_insn *patchlet;
12663 struct bpf_insn chk_and_div[] = {
Daniel Borkmann9b00f1b2021-02-10 14:14:42 +010012664 /* [R,W]x div 0 -> 0 */
Daniel Borkmanne88b2c62021-02-09 18:46:10 +000012665 BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) |
12666 BPF_JNE | BPF_K, insn->src_reg,
12667 0, 2, 0),
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012668 BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg),
12669 BPF_JMP_IMM(BPF_JA, 0, 0, 1),
12670 *insn,
12671 };
Daniel Borkmanne88b2c62021-02-09 18:46:10 +000012672 struct bpf_insn chk_and_mod[] = {
Daniel Borkmann9b00f1b2021-02-10 14:14:42 +010012673 /* [R,W]x mod 0 -> [R,W]x */
Daniel Borkmanne88b2c62021-02-09 18:46:10 +000012674 BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) |
12675 BPF_JEQ | BPF_K, insn->src_reg,
Daniel Borkmann9b00f1b2021-02-10 14:14:42 +010012676 0, 1 + (is64 ? 0 : 1), 0),
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012677 *insn,
Daniel Borkmann9b00f1b2021-02-10 14:14:42 +010012678 BPF_JMP_IMM(BPF_JA, 0, 0, 1),
12679 BPF_MOV32_REG(insn->dst_reg, insn->dst_reg),
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012680 };
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012681
Daniel Borkmanne88b2c62021-02-09 18:46:10 +000012682 patchlet = isdiv ? chk_and_div : chk_and_mod;
12683 cnt = isdiv ? ARRAY_SIZE(chk_and_div) :
Daniel Borkmann9b00f1b2021-02-10 14:14:42 +010012684 ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0);
Daniel Borkmannf6b1b3b2018-01-26 23:33:39 +010012685
12686 new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt);
Alexei Starovoitov68fda452018-01-12 18:59:52 -080012687 if (!new_prog)
12688 return -ENOMEM;
12689
12690 delta += cnt - 1;
12691 env->prog = prog = new_prog;
12692 insn = new_prog->insnsi + i + delta;
12693 continue;
12694 }
12695
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012696 /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */
Daniel Borkmanne0cea7c2018-05-04 01:08:14 +020012697 if (BPF_CLASS(insn->code) == BPF_LD &&
12698 (BPF_MODE(insn->code) == BPF_ABS ||
12699 BPF_MODE(insn->code) == BPF_IND)) {
12700 cnt = env->ops->gen_ld_abs(insn, insn_buf);
12701 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
12702 verbose(env, "bpf verifier is misconfigured\n");
12703 return -EINVAL;
12704 }
12705
12706 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
12707 if (!new_prog)
12708 return -ENOMEM;
12709
12710 delta += cnt - 1;
12711 env->prog = prog = new_prog;
12712 insn = new_prog->insnsi + i + delta;
12713 continue;
12714 }
12715
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012716 /* Rewrite pointer arithmetic to mitigate speculation attacks. */
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012717 if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) ||
12718 insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) {
12719 const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X;
12720 const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X;
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012721 struct bpf_insn *patch = &insn_buf[0];
Daniel Borkmann801c6052021-04-29 15:19:37 +000012722 bool issrc, isneg, isimm;
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012723 u32 off_reg;
12724
12725 aux = &env->insn_aux_data[i + delta];
Daniel Borkmann3612af72019-03-01 22:05:29 +010012726 if (!aux->alu_state ||
12727 aux->alu_state == BPF_ALU_NON_POINTER)
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012728 continue;
12729
12730 isneg = aux->alu_state & BPF_ALU_NEG_VALUE;
12731 issrc = (aux->alu_state & BPF_ALU_SANITIZE) ==
12732 BPF_ALU_SANITIZE_SRC;
Daniel Borkmann801c6052021-04-29 15:19:37 +000012733 isimm = aux->alu_state & BPF_ALU_IMMEDIATE;
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012734
12735 off_reg = issrc ? insn->src_reg : insn->dst_reg;
Daniel Borkmann801c6052021-04-29 15:19:37 +000012736 if (isimm) {
12737 *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit);
12738 } else {
12739 if (isneg)
12740 *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1);
12741 *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit);
12742 *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg);
12743 *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg);
12744 *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0);
12745 *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63);
12746 *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, off_reg);
12747 }
Daniel Borkmannb9b34dd2021-04-30 16:21:46 +020012748 if (!issrc)
12749 *patch++ = BPF_MOV64_REG(insn->dst_reg, insn->src_reg);
12750 insn->src_reg = BPF_REG_AX;
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012751 if (isneg)
12752 insn->code = insn->code == code_add ?
12753 code_sub : code_add;
12754 *patch++ = *insn;
Daniel Borkmann801c6052021-04-29 15:19:37 +000012755 if (issrc && isneg && !isimm)
Daniel Borkmann979d63d2019-01-03 00:58:34 +010012756 *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1);
12757 cnt = patch - insn_buf;
12758
12759 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
12760 if (!new_prog)
12761 return -ENOMEM;
12762
12763 delta += cnt - 1;
12764 env->prog = prog = new_prog;
12765 insn = new_prog->insnsi + i + delta;
12766 continue;
12767 }
12768
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012769 if (insn->code != (BPF_JMP | BPF_CALL))
12770 continue;
Alexei Starovoitovcc8b0b92017-12-14 17:55:05 -080012771 if (insn->src_reg == BPF_PSEUDO_CALL)
12772 continue;
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070012773 if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) {
12774 ret = fixup_kfunc_call(env, insn);
12775 if (ret)
12776 return ret;
12777 continue;
12778 }
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012779
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012780 if (insn->imm == BPF_FUNC_get_route_realm)
12781 prog->dst_needed = 1;
12782 if (insn->imm == BPF_FUNC_get_prandom_u32)
12783 bpf_user_rnd_init_once();
Josef Bacik9802d862017-12-11 11:36:48 -050012784 if (insn->imm == BPF_FUNC_override_return)
12785 prog->kprobe_override = 1;
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012786 if (insn->imm == BPF_FUNC_tail_call) {
David S. Miller7b9f6da2017-04-20 10:35:33 -040012787 /* If we tail call into other programs, we
12788 * cannot make any assumptions since they can
12789 * be replaced dynamically during runtime in
12790 * the program array.
12791 */
12792 prog->cb_access = 1;
Maciej Fijalkowskie4119012020-09-16 23:10:09 +020012793 if (!allow_tail_call_in_subprogs(env))
12794 prog->aux->stack_depth = MAX_BPF_STACK;
12795 prog->aux->max_pkt_offset = MAX_PACKET_OFF;
David S. Miller7b9f6da2017-04-20 10:35:33 -040012796
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012797 /* mark bpf_tail_call as different opcode to avoid
Zhen Lei8fb33b62021-05-25 10:56:59 +080012798 * conditional branch in the interpreter for every normal
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012799 * call and to prevent accidental JITing by JIT compiler
12800 * that doesn't support bpf_tail_call yet
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012801 */
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012802 insn->imm = 0;
Alexei Starovoitov71189fa2017-05-30 13:31:27 -070012803 insn->code = BPF_JMP | BPF_TAIL_CALL;
Alexei Starovoitovb2157392018-01-07 17:33:02 -080012804
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012805 aux = &env->insn_aux_data[i + delta];
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -070012806 if (env->bpf_capable && !expect_blinding &&
Daniel Borkmanncc52d912019-12-19 22:19:50 +010012807 prog->jit_requested &&
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012808 !bpf_map_key_poisoned(aux) &&
12809 !bpf_map_ptr_poisoned(aux) &&
12810 !bpf_map_ptr_unpriv(aux)) {
12811 struct bpf_jit_poke_descriptor desc = {
12812 .reason = BPF_POKE_REASON_TAIL_CALL,
12813 .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state),
12814 .tail_call.key = bpf_map_key_immediate(aux),
Maciej Fijalkowskia748c692020-09-16 23:10:05 +020012815 .insn_idx = i + delta,
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012816 };
12817
12818 ret = bpf_jit_add_poke_descriptor(prog, &desc);
12819 if (ret < 0) {
12820 verbose(env, "adding tail call poke descriptor failed\n");
12821 return ret;
12822 }
12823
12824 insn->imm = ret + 1;
12825 continue;
12826 }
12827
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012828 if (!bpf_map_ptr_unpriv(aux))
12829 continue;
12830
Alexei Starovoitovb2157392018-01-07 17:33:02 -080012831 /* instead of changing every JIT dealing with tail_call
12832 * emit two extra insns:
12833 * if (index >= max_entries) goto out;
12834 * index &= array->index_mask;
12835 * to avoid out-of-bounds cpu speculation
12836 */
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012837 if (bpf_map_ptr_poisoned(aux)) {
Colin Ian King40950342018-01-10 09:20:54 +000012838 verbose(env, "tail_call abusing map_ptr\n");
Alexei Starovoitovb2157392018-01-07 17:33:02 -080012839 return -EINVAL;
12840 }
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012841
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012842 map_ptr = BPF_MAP_PTR(aux->map_ptr_state);
Alexei Starovoitovb2157392018-01-07 17:33:02 -080012843 insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3,
12844 map_ptr->max_entries, 2);
12845 insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3,
12846 container_of(map_ptr,
12847 struct bpf_array,
12848 map)->index_mask);
12849 insn_buf[2] = *insn;
12850 cnt = 3;
12851 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
12852 if (!new_prog)
12853 return -ENOMEM;
12854
12855 delta += cnt - 1;
12856 env->prog = prog = new_prog;
12857 insn = new_prog->insnsi + i + delta;
Alexei Starovoitov79741b32017-03-15 18:26:40 -070012858 continue;
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012859 }
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070012860
Alexei Starovoitovb00628b2021-07-14 17:54:09 -070012861 if (insn->imm == BPF_FUNC_timer_set_callback) {
12862 /* The verifier will process callback_fn as many times as necessary
12863 * with different maps and the register states prepared by
12864 * set_timer_callback_state will be accurate.
12865 *
12866 * The following use case is valid:
12867 * map1 is shared by prog1, prog2, prog3.
12868 * prog1 calls bpf_timer_init for some map1 elements
12869 * prog2 calls bpf_timer_set_callback for some map1 elements.
12870 * Those that were not bpf_timer_init-ed will return -EINVAL.
12871 * prog3 calls bpf_timer_start for some map1 elements.
12872 * Those that were not both bpf_timer_init-ed and
12873 * bpf_timer_set_callback-ed will return -EINVAL.
12874 */
12875 struct bpf_insn ld_addrs[2] = {
12876 BPF_LD_IMM64(BPF_REG_3, (long)prog->aux),
12877 };
12878
12879 insn_buf[0] = ld_addrs[0];
12880 insn_buf[1] = ld_addrs[1];
12881 insn_buf[2] = *insn;
12882 cnt = 3;
12883
12884 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
12885 if (!new_prog)
12886 return -ENOMEM;
12887
12888 delta += cnt - 1;
12889 env->prog = prog = new_prog;
12890 insn = new_prog->insnsi + i + delta;
12891 goto patch_call_imm;
12892 }
12893
Daniel Borkmann89c63072017-08-19 03:12:45 +020012894 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
Daniel Borkmann09772d92018-06-02 23:06:35 +020012895 * and other inlining handlers are currently limited to 64 bit
12896 * only.
Daniel Borkmann89c63072017-08-19 03:12:45 +020012897 */
Alexei Starovoitov60b58afc2017-12-14 17:55:14 -080012898 if (prog->jit_requested && BITS_PER_LONG == 64 &&
Daniel Borkmann09772d92018-06-02 23:06:35 +020012899 (insn->imm == BPF_FUNC_map_lookup_elem ||
12900 insn->imm == BPF_FUNC_map_update_elem ||
Daniel Borkmann84430d42018-10-21 02:09:27 +020012901 insn->imm == BPF_FUNC_map_delete_elem ||
12902 insn->imm == BPF_FUNC_map_push_elem ||
12903 insn->imm == BPF_FUNC_map_pop_elem ||
Björn Töpele6a47502021-03-08 12:29:06 +010012904 insn->imm == BPF_FUNC_map_peek_elem ||
12905 insn->imm == BPF_FUNC_redirect_map)) {
Daniel Borkmannc93552c2018-05-24 02:32:53 +020012906 aux = &env->insn_aux_data[i + delta];
12907 if (bpf_map_ptr_poisoned(aux))
12908 goto patch_call_imm;
12909
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010012910 map_ptr = BPF_MAP_PTR(aux->map_ptr_state);
Daniel Borkmann09772d92018-06-02 23:06:35 +020012911 ops = map_ptr->ops;
12912 if (insn->imm == BPF_FUNC_map_lookup_elem &&
12913 ops->map_gen_lookup) {
12914 cnt = ops->map_gen_lookup(map_ptr, insn_buf);
Daniel Borkmann4a8f87e2020-10-11 01:40:03 +020012915 if (cnt == -EOPNOTSUPP)
12916 goto patch_map_ops_generic;
12917 if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) {
Daniel Borkmann09772d92018-06-02 23:06:35 +020012918 verbose(env, "bpf verifier is misconfigured\n");
12919 return -EINVAL;
12920 }
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070012921
Daniel Borkmann09772d92018-06-02 23:06:35 +020012922 new_prog = bpf_patch_insn_data(env, i + delta,
12923 insn_buf, cnt);
12924 if (!new_prog)
12925 return -ENOMEM;
12926
12927 delta += cnt - 1;
12928 env->prog = prog = new_prog;
12929 insn = new_prog->insnsi + i + delta;
12930 continue;
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070012931 }
12932
Daniel Borkmann09772d92018-06-02 23:06:35 +020012933 BUILD_BUG_ON(!__same_type(ops->map_lookup_elem,
12934 (void *(*)(struct bpf_map *map, void *key))NULL));
12935 BUILD_BUG_ON(!__same_type(ops->map_delete_elem,
12936 (int (*)(struct bpf_map *map, void *key))NULL));
12937 BUILD_BUG_ON(!__same_type(ops->map_update_elem,
12938 (int (*)(struct bpf_map *map, void *key, void *value,
12939 u64 flags))NULL));
Daniel Borkmann84430d42018-10-21 02:09:27 +020012940 BUILD_BUG_ON(!__same_type(ops->map_push_elem,
12941 (int (*)(struct bpf_map *map, void *value,
12942 u64 flags))NULL));
12943 BUILD_BUG_ON(!__same_type(ops->map_pop_elem,
12944 (int (*)(struct bpf_map *map, void *value))NULL));
12945 BUILD_BUG_ON(!__same_type(ops->map_peek_elem,
12946 (int (*)(struct bpf_map *map, void *value))NULL));
Björn Töpele6a47502021-03-08 12:29:06 +010012947 BUILD_BUG_ON(!__same_type(ops->map_redirect,
12948 (int (*)(struct bpf_map *map, u32 ifindex, u64 flags))NULL));
12949
Daniel Borkmann4a8f87e2020-10-11 01:40:03 +020012950patch_map_ops_generic:
Daniel Borkmann09772d92018-06-02 23:06:35 +020012951 switch (insn->imm) {
12952 case BPF_FUNC_map_lookup_elem:
12953 insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) -
12954 __bpf_call_base;
12955 continue;
12956 case BPF_FUNC_map_update_elem:
12957 insn->imm = BPF_CAST_CALL(ops->map_update_elem) -
12958 __bpf_call_base;
12959 continue;
12960 case BPF_FUNC_map_delete_elem:
12961 insn->imm = BPF_CAST_CALL(ops->map_delete_elem) -
12962 __bpf_call_base;
12963 continue;
Daniel Borkmann84430d42018-10-21 02:09:27 +020012964 case BPF_FUNC_map_push_elem:
12965 insn->imm = BPF_CAST_CALL(ops->map_push_elem) -
12966 __bpf_call_base;
12967 continue;
12968 case BPF_FUNC_map_pop_elem:
12969 insn->imm = BPF_CAST_CALL(ops->map_pop_elem) -
12970 __bpf_call_base;
12971 continue;
12972 case BPF_FUNC_map_peek_elem:
12973 insn->imm = BPF_CAST_CALL(ops->map_peek_elem) -
12974 __bpf_call_base;
12975 continue;
Björn Töpele6a47502021-03-08 12:29:06 +010012976 case BPF_FUNC_redirect_map:
12977 insn->imm = BPF_CAST_CALL(ops->map_redirect) -
12978 __bpf_call_base;
12979 continue;
Daniel Borkmann09772d92018-06-02 23:06:35 +020012980 }
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070012981
Daniel Borkmann09772d92018-06-02 23:06:35 +020012982 goto patch_call_imm;
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070012983 }
12984
Brendan Jackmane6ac5932021-02-17 10:45:09 +000012985 /* Implement bpf_jiffies64 inline. */
Martin KaFai Lau5576b992020-01-22 15:36:46 -080012986 if (prog->jit_requested && BITS_PER_LONG == 64 &&
12987 insn->imm == BPF_FUNC_jiffies64) {
12988 struct bpf_insn ld_jiffies_addr[2] = {
12989 BPF_LD_IMM64(BPF_REG_0,
12990 (unsigned long)&jiffies),
12991 };
12992
12993 insn_buf[0] = ld_jiffies_addr[0];
12994 insn_buf[1] = ld_jiffies_addr[1];
12995 insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0,
12996 BPF_REG_0, 0);
12997 cnt = 3;
12998
12999 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
13000 cnt);
13001 if (!new_prog)
13002 return -ENOMEM;
13003
13004 delta += cnt - 1;
13005 env->prog = prog = new_prog;
13006 insn = new_prog->insnsi + i + delta;
13007 continue;
13008 }
13009
Jiri Olsa9b99edc2021-07-14 11:43:55 +020013010 /* Implement bpf_get_func_ip inline. */
13011 if (prog_type == BPF_PROG_TYPE_TRACING &&
13012 insn->imm == BPF_FUNC_get_func_ip) {
13013 /* Load IP address from ctx - 8 */
13014 insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
13015
13016 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1);
13017 if (!new_prog)
13018 return -ENOMEM;
13019
13020 env->prog = prog = new_prog;
13021 insn = new_prog->insnsi + i + delta;
13022 continue;
13023 }
13024
Alexei Starovoitov81ed18a2017-03-15 18:26:42 -070013025patch_call_imm:
Andrey Ignatov5e43f892018-03-30 15:08:00 -070013026 fn = env->ops->get_func_proto(insn->imm, env->prog);
Alexei Starovoitov79741b32017-03-15 18:26:40 -070013027 /* all functions that have prototype and verifier allowed
13028 * programs to call them, must be real in-kernel functions
13029 */
13030 if (!fn->func) {
Jakub Kicinski61bd5212017-10-09 10:30:11 -070013031 verbose(env,
13032 "kernel subsystem misconfigured func %s#%d\n",
Alexei Starovoitov79741b32017-03-15 18:26:40 -070013033 func_id_name(insn->imm), insn->imm);
13034 return -EFAULT;
13035 }
13036 insn->imm = fn->func - __bpf_call_base;
13037 }
13038
Daniel Borkmannd2e4c1e2019-11-22 21:07:59 +010013039 /* Since poke tab is now finalized, publish aux to tracker. */
13040 for (i = 0; i < prog->aux->size_poke_tab; i++) {
13041 map_ptr = prog->aux->poke_tab[i].tail_call.map;
13042 if (!map_ptr->ops->map_poke_track ||
13043 !map_ptr->ops->map_poke_untrack ||
13044 !map_ptr->ops->map_poke_run) {
13045 verbose(env, "bpf verifier is misconfigured\n");
13046 return -EINVAL;
13047 }
13048
13049 ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux);
13050 if (ret < 0) {
13051 verbose(env, "tracking tail call prog failed\n");
13052 return ret;
13053 }
13054 }
13055
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070013056 sort_kfunc_descs_by_imm(env->prog);
13057
Alexei Starovoitov79741b32017-03-15 18:26:40 -070013058 return 0;
13059}
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070013060
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013061static void free_states(struct bpf_verifier_env *env)
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013062{
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013063 struct bpf_verifier_state_list *sl, *sln;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013064 int i;
13065
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070013066 sl = env->free_list;
13067 while (sl) {
13068 sln = sl->next;
13069 free_verifier_state(&sl->state, false);
13070 kfree(sl);
13071 sl = sln;
13072 }
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013073 env->free_list = NULL;
Alexei Starovoitov9f4686c2019-04-01 21:27:41 -070013074
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013075 if (!env->explored_states)
13076 return;
13077
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070013078 for (i = 0; i < state_htab_size(env); i++) {
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013079 sl = env->explored_states[i];
13080
Alexei Starovoitova8f500a2019-05-21 20:17:06 -070013081 while (sl) {
13082 sln = sl->next;
13083 free_verifier_state(&sl->state, false);
13084 kfree(sl);
13085 sl = sln;
13086 }
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013087 env->explored_states[i] = NULL;
13088 }
13089}
13090
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013091static int do_check_common(struct bpf_verifier_env *env, int subprog)
13092{
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -070013093 bool pop_log = !(env->log.level & BPF_LOG_LEVEL2);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013094 struct bpf_verifier_state *state;
13095 struct bpf_reg_state *regs;
13096 int ret, i;
13097
13098 env->prev_linfo = NULL;
13099 env->pass_cnt++;
13100
13101 state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL);
13102 if (!state)
13103 return -ENOMEM;
13104 state->curframe = 0;
13105 state->speculative = false;
13106 state->branches = 1;
13107 state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL);
13108 if (!state->frame[0]) {
13109 kfree(state);
13110 return -ENOMEM;
13111 }
13112 env->cur_state = state;
13113 init_func_state(env, state->frame[0],
13114 BPF_MAIN_FUNC /* callsite */,
13115 0 /* frameno */,
13116 subprog);
13117
13118 regs = state->frame[state->curframe]->regs;
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013119 if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) {
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013120 ret = btf_prepare_func_args(env, subprog, regs);
13121 if (ret)
13122 goto out;
13123 for (i = BPF_REG_1; i <= BPF_REG_5; i++) {
13124 if (regs[i].type == PTR_TO_CTX)
13125 mark_reg_known_zero(env, regs, i);
13126 else if (regs[i].type == SCALAR_VALUE)
13127 mark_reg_unknown(env, regs, i);
Dmitrii Banshchikove5069b9c2021-02-13 00:56:41 +040013128 else if (regs[i].type == PTR_TO_MEM_OR_NULL) {
13129 const u32 mem_size = regs[i].mem_size;
13130
13131 mark_reg_known_zero(env, regs, i);
13132 regs[i].mem_size = mem_size;
13133 regs[i].id = ++env->id_gen;
13134 }
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013135 }
13136 } else {
13137 /* 1st arg to a function */
13138 regs[BPF_REG_1].type = PTR_TO_CTX;
13139 mark_reg_known_zero(env, regs, BPF_REG_1);
Martin KaFai Lau34747c42021-03-24 18:51:36 -070013140 ret = btf_check_subprog_arg_match(env, subprog, regs);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013141 if (ret == -EFAULT)
13142 /* unlikely verifier bug. abort.
13143 * ret == 0 and ret < 0 are sadly acceptable for
13144 * main() function due to backward compatibility.
13145 * Like socket filter program may be written as:
13146 * int bpf_prog(struct pt_regs *ctx)
13147 * and never dereference that ctx in the program.
13148 * 'struct pt_regs' is a type mismatch for socket
13149 * filter that should be using 'struct __sk_buff'.
13150 */
13151 goto out;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013152 }
13153
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013154 ret = do_check(env);
13155out:
Alexei Starovoitovf59bbfc2020-01-21 18:41:38 -080013156 /* check for NULL is necessary, since cur_state can be freed inside
13157 * do_check() under memory pressure.
13158 */
13159 if (env->cur_state) {
13160 free_verifier_state(env->cur_state, true);
13161 env->cur_state = NULL;
13162 }
Andrii Nakryiko6f8a57c2020-04-23 12:58:50 -070013163 while (!pop_stack(env, NULL, NULL, false));
13164 if (!ret && pop_log)
13165 bpf_vlog_reset(&env->log, 0);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013166 free_states(env);
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013167 return ret;
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013168}
13169
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013170/* Verify all global functions in a BPF program one by one based on their BTF.
13171 * All global functions must pass verification. Otherwise the whole program is rejected.
13172 * Consider:
13173 * int bar(int);
13174 * int foo(int f)
13175 * {
13176 * return bar(f);
13177 * }
13178 * int bar(int b)
13179 * {
13180 * ...
13181 * }
13182 * foo() will be verified first for R1=any_scalar_value. During verification it
13183 * will be assumed that bar() already verified successfully and call to bar()
13184 * from foo() will be checked for type match only. Later bar() will be verified
13185 * independently to check that it's safe for R1=any_scalar_value.
13186 */
13187static int do_check_subprogs(struct bpf_verifier_env *env)
13188{
13189 struct bpf_prog_aux *aux = env->prog->aux;
13190 int i, ret;
13191
13192 if (!aux->func_info)
13193 return 0;
13194
13195 for (i = 1; i < env->subprog_cnt; i++) {
13196 if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL)
13197 continue;
13198 env->insn_idx = env->subprog_info[i].start;
13199 WARN_ON_ONCE(env->insn_idx == 0);
13200 ret = do_check_common(env, i);
13201 if (ret) {
13202 return ret;
13203 } else if (env->log.level & BPF_LOG_LEVEL) {
13204 verbose(env,
13205 "Func#%d is safe for any args that match its prototype\n",
13206 i);
13207 }
13208 }
13209 return 0;
13210}
13211
13212static int do_check_main(struct bpf_verifier_env *env)
13213{
13214 int ret;
13215
13216 env->insn_idx = 0;
13217 ret = do_check_common(env, 0);
13218 if (!ret)
13219 env->prog->aux->stack_depth = env->subprog_info[0].stack_depth;
13220 return ret;
13221}
13222
13223
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070013224static void print_verification_stats(struct bpf_verifier_env *env)
13225{
13226 int i;
13227
13228 if (env->log.level & BPF_LOG_STATS) {
13229 verbose(env, "verification time %lld usec\n",
13230 div_u64(env->verification_time, 1000));
13231 verbose(env, "stack depth ");
13232 for (i = 0; i < env->subprog_cnt; i++) {
13233 u32 depth = env->subprog_info[i].stack_depth;
13234
13235 verbose(env, "%d", depth);
13236 if (i + 1 < env->subprog_cnt)
13237 verbose(env, "+");
13238 }
13239 verbose(env, "\n");
13240 }
13241 verbose(env, "processed %d insns (limit %d) max_states_per_insn %d "
13242 "total_states %d peak_states %d mark_read %d\n",
13243 env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS,
13244 env->max_states_per_insn, env->total_states,
13245 env->peak_states, env->longest_mark_read_walk);
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013246}
13247
Martin KaFai Lau27ae79972020-01-08 16:35:03 -080013248static int check_struct_ops_btf_id(struct bpf_verifier_env *env)
13249{
13250 const struct btf_type *t, *func_proto;
13251 const struct bpf_struct_ops *st_ops;
13252 const struct btf_member *member;
13253 struct bpf_prog *prog = env->prog;
13254 u32 btf_id, member_idx;
13255 const char *mname;
13256
Toke Høiland-Jørgensen12aa8a92021-03-26 11:03:13 +010013257 if (!prog->gpl_compatible) {
13258 verbose(env, "struct ops programs must have a GPL compatible license\n");
13259 return -EINVAL;
13260 }
13261
Martin KaFai Lau27ae79972020-01-08 16:35:03 -080013262 btf_id = prog->aux->attach_btf_id;
13263 st_ops = bpf_struct_ops_find(btf_id);
13264 if (!st_ops) {
13265 verbose(env, "attach_btf_id %u is not a supported struct\n",
13266 btf_id);
13267 return -ENOTSUPP;
13268 }
13269
13270 t = st_ops->type;
13271 member_idx = prog->expected_attach_type;
13272 if (member_idx >= btf_type_vlen(t)) {
13273 verbose(env, "attach to invalid member idx %u of struct %s\n",
13274 member_idx, st_ops->name);
13275 return -EINVAL;
13276 }
13277
13278 member = &btf_type_member(t)[member_idx];
13279 mname = btf_name_by_offset(btf_vmlinux, member->name_off);
13280 func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type,
13281 NULL);
13282 if (!func_proto) {
13283 verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n",
13284 mname, member_idx, st_ops->name);
13285 return -EINVAL;
13286 }
13287
13288 if (st_ops->check_member) {
13289 int err = st_ops->check_member(t, member);
13290
13291 if (err) {
13292 verbose(env, "attach to unsupported member %s of struct %s\n",
13293 mname, st_ops->name);
13294 return err;
13295 }
13296 }
13297
13298 prog->aux->attach_func_proto = func_proto;
13299 prog->aux->attach_func_name = mname;
13300 env->ops = st_ops->verifier_ops;
13301
13302 return 0;
13303}
KP Singh6ba43b72020-03-04 20:18:50 +010013304#define SECURITY_PREFIX "security_"
13305
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013306static int check_attach_modify_return(unsigned long addr, const char *func_name)
KP Singh6ba43b72020-03-04 20:18:50 +010013307{
KP Singh69191752020-03-05 21:49:55 +010013308 if (within_error_injection_list(addr) ||
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013309 !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1))
KP Singh6ba43b72020-03-04 20:18:50 +010013310 return 0;
KP Singh6ba43b72020-03-04 20:18:50 +010013311
KP Singh6ba43b72020-03-04 20:18:50 +010013312 return -EINVAL;
13313}
Martin KaFai Lau27ae79972020-01-08 16:35:03 -080013314
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070013315/* list of non-sleepable functions that are otherwise on
13316 * ALLOW_ERROR_INJECTION list
13317 */
13318BTF_SET_START(btf_non_sleepable_error_inject)
13319/* Three functions below can be called from sleepable and non-sleepable context.
13320 * Assume non-sleepable from bpf safety point of view.
13321 */
13322BTF_ID(func, __add_to_page_cache_locked)
13323BTF_ID(func, should_fail_alloc_page)
13324BTF_ID(func, should_failslab)
13325BTF_SET_END(btf_non_sleepable_error_inject)
13326
13327static int check_non_sleepable_error_inject(u32 btf_id)
13328{
13329 return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id);
13330}
13331
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013332int bpf_check_attach_target(struct bpf_verifier_log *log,
13333 const struct bpf_prog *prog,
13334 const struct bpf_prog *tgt_prog,
13335 u32 btf_id,
13336 struct bpf_attach_target_info *tgt_info)
Martin KaFai Lau38207292019-10-24 17:18:11 -070013337{
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013338 bool prog_extension = prog->type == BPF_PROG_TYPE_EXT;
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013339 const char prefix[] = "btf_trace_";
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013340 int ret = 0, subprog = -1, i;
Martin KaFai Lau38207292019-10-24 17:18:11 -070013341 const struct btf_type *t;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013342 bool conservative = true;
Martin KaFai Lau38207292019-10-24 17:18:11 -070013343 const char *tname;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013344 struct btf *btf;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013345 long addr = 0;
Martin KaFai Lau38207292019-10-24 17:18:11 -070013346
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013347 if (!btf_id) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013348 bpf_log(log, "Tracing programs must provide btf_id\n");
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013349 return -EINVAL;
13350 }
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -080013351 btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013352 if (!btf) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013353 bpf_log(log,
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013354 "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n");
13355 return -EINVAL;
13356 }
13357 t = btf_type_by_id(btf, btf_id);
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013358 if (!t) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013359 bpf_log(log, "attach_btf_id %u is invalid\n", btf_id);
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013360 return -EINVAL;
13361 }
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013362 tname = btf_name_by_offset(btf, t->name_off);
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013363 if (!tname) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013364 bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id);
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013365 return -EINVAL;
13366 }
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013367 if (tgt_prog) {
13368 struct bpf_prog_aux *aux = tgt_prog->aux;
13369
13370 for (i = 0; i < aux->func_info_cnt; i++)
13371 if (aux->func_info[i].type_id == btf_id) {
13372 subprog = i;
13373 break;
13374 }
13375 if (subprog == -1) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013376 bpf_log(log, "Subprog %s doesn't exist\n", tname);
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013377 return -EINVAL;
13378 }
13379 conservative = aux->func_info_aux[subprog].unreliable;
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013380 if (prog_extension) {
13381 if (conservative) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013382 bpf_log(log,
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013383 "Cannot replace static functions\n");
13384 return -EINVAL;
13385 }
13386 if (!prog->jit_requested) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013387 bpf_log(log,
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013388 "Extension programs should be JITed\n");
13389 return -EINVAL;
13390 }
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013391 }
13392 if (!tgt_prog->jited) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013393 bpf_log(log, "Can attach to only JITed progs\n");
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013394 return -EINVAL;
13395 }
13396 if (tgt_prog->type == prog->type) {
13397 /* Cannot fentry/fexit another fentry/fexit program.
13398 * Cannot attach program extension to another extension.
13399 * It's ok to attach fentry/fexit to extension program.
13400 */
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013401 bpf_log(log, "Cannot recursively attach\n");
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013402 return -EINVAL;
13403 }
13404 if (tgt_prog->type == BPF_PROG_TYPE_TRACING &&
13405 prog_extension &&
13406 (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY ||
13407 tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) {
13408 /* Program extensions can extend all program types
13409 * except fentry/fexit. The reason is the following.
13410 * The fentry/fexit programs are used for performance
13411 * analysis, stats and can be attached to any program
13412 * type except themselves. When extension program is
13413 * replacing XDP function it is necessary to allow
13414 * performance analysis of all functions. Both original
13415 * XDP program and its program extension. Hence
13416 * attaching fentry/fexit to BPF_PROG_TYPE_EXT is
13417 * allowed. If extending of fentry/fexit was allowed it
13418 * would be possible to create long call chain
13419 * fentry->extension->fentry->extension beyond
13420 * reasonable stack size. Hence extending fentry is not
13421 * allowed.
13422 */
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013423 bpf_log(log, "Cannot extend fentry/fexit\n");
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013424 return -EINVAL;
13425 }
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013426 } else {
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013427 if (prog_extension) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013428 bpf_log(log, "Cannot replace kernel functions\n");
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013429 return -EINVAL;
13430 }
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013431 }
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013432
13433 switch (prog->expected_attach_type) {
13434 case BPF_TRACE_RAW_TP:
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013435 if (tgt_prog) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013436 bpf_log(log,
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013437 "Only FENTRY/FEXIT progs are attachable to another BPF prog\n");
13438 return -EINVAL;
13439 }
Martin KaFai Lau38207292019-10-24 17:18:11 -070013440 if (!btf_type_is_typedef(t)) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013441 bpf_log(log, "attach_btf_id %u is not a typedef\n",
Martin KaFai Lau38207292019-10-24 17:18:11 -070013442 btf_id);
13443 return -EINVAL;
13444 }
Alexei Starovoitovf1b95092019-10-30 15:32:11 -070013445 if (strncmp(prefix, tname, sizeof(prefix) - 1)) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013446 bpf_log(log, "attach_btf_id %u points to wrong type name %s\n",
Martin KaFai Lau38207292019-10-24 17:18:11 -070013447 btf_id, tname);
13448 return -EINVAL;
13449 }
13450 tname += sizeof(prefix) - 1;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013451 t = btf_type_by_id(btf, t->type);
Martin KaFai Lau38207292019-10-24 17:18:11 -070013452 if (!btf_type_is_ptr(t))
13453 /* should never happen in valid vmlinux build */
13454 return -EINVAL;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013455 t = btf_type_by_id(btf, t->type);
Martin KaFai Lau38207292019-10-24 17:18:11 -070013456 if (!btf_type_is_func_proto(t))
13457 /* should never happen in valid vmlinux build */
13458 return -EINVAL;
13459
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013460 break;
Yonghong Song15d83c42020-05-09 10:59:00 -070013461 case BPF_TRACE_ITER:
13462 if (!btf_type_is_func(t)) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013463 bpf_log(log, "attach_btf_id %u is not a function\n",
Yonghong Song15d83c42020-05-09 10:59:00 -070013464 btf_id);
13465 return -EINVAL;
13466 }
13467 t = btf_type_by_id(btf, t->type);
13468 if (!btf_type_is_func_proto(t))
13469 return -EINVAL;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013470 ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel);
13471 if (ret)
13472 return ret;
13473 break;
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013474 default:
13475 if (!prog_extension)
13476 return -EINVAL;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -050013477 fallthrough;
KP Singhae240822020-03-04 20:18:49 +010013478 case BPF_MODIFY_RETURN:
KP Singh9e4e01d2020-03-29 01:43:52 +010013479 case BPF_LSM_MAC:
Alexei Starovoitovfec56f52019-11-14 10:57:04 -080013480 case BPF_TRACE_FENTRY:
13481 case BPF_TRACE_FEXIT:
13482 if (!btf_type_is_func(t)) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013483 bpf_log(log, "attach_btf_id %u is not a function\n",
Alexei Starovoitovfec56f52019-11-14 10:57:04 -080013484 btf_id);
13485 return -EINVAL;
13486 }
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013487 if (prog_extension &&
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013488 btf_check_type_match(log, prog, btf, t))
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013489 return -EINVAL;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013490 t = btf_type_by_id(btf, t->type);
Alexei Starovoitovfec56f52019-11-14 10:57:04 -080013491 if (!btf_type_is_func_proto(t))
13492 return -EINVAL;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013493
Toke Høiland-Jørgensen4a1e7c02020-09-29 14:45:51 +020013494 if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) &&
13495 (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type ||
13496 prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type))
13497 return -EINVAL;
13498
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013499 if (tgt_prog && conservative)
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013500 t = NULL;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013501
13502 ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel);
Alexei Starovoitovfec56f52019-11-14 10:57:04 -080013503 if (ret < 0)
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013504 return ret;
13505
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013506 if (tgt_prog) {
Yonghong Songe9eeec52019-12-04 17:06:06 -080013507 if (subprog == 0)
13508 addr = (long) tgt_prog->bpf_func;
13509 else
13510 addr = (long) tgt_prog->aux->func[subprog]->bpf_func;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013511 } else {
13512 addr = kallsyms_lookup_name(tname);
13513 if (!addr) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013514 bpf_log(log,
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013515 "The address of function %s cannot be found\n",
13516 tname);
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013517 return -ENOENT;
Alexei Starovoitov5b92a282019-11-14 10:57:17 -080013518 }
Alexei Starovoitovfec56f52019-11-14 10:57:04 -080013519 }
Alexei Starovoitov18644ce2020-05-28 21:38:36 -070013520
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070013521 if (prog->aux->sleepable) {
13522 ret = -EINVAL;
13523 switch (prog->type) {
13524 case BPF_PROG_TYPE_TRACING:
13525 /* fentry/fexit/fmod_ret progs can be sleepable only if they are
13526 * attached to ALLOW_ERROR_INJECTION and are not in denylist.
13527 */
13528 if (!check_non_sleepable_error_inject(btf_id) &&
13529 within_error_injection_list(addr))
13530 ret = 0;
13531 break;
13532 case BPF_PROG_TYPE_LSM:
13533 /* LSM progs check that they are attached to bpf_lsm_*() funcs.
13534 * Only some of them are sleepable.
13535 */
KP Singh423f1612020-11-13 00:59:29 +000013536 if (bpf_lsm_is_sleepable_hook(btf_id))
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070013537 ret = 0;
13538 break;
13539 default:
13540 break;
13541 }
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013542 if (ret) {
13543 bpf_log(log, "%s is not sleepable\n", tname);
13544 return ret;
13545 }
Alexei Starovoitov1e6c62a2020-08-27 15:01:11 -070013546 } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) {
Toke Høiland-Jørgensen1af92702020-09-25 23:25:00 +020013547 if (tgt_prog) {
Toke Høiland-Jørgensenefc68152020-09-25 23:25:01 +020013548 bpf_log(log, "can't modify return codes of BPF programs\n");
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013549 return -EINVAL;
Toke Høiland-Jørgensen1af92702020-09-25 23:25:00 +020013550 }
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013551 ret = check_attach_modify_return(addr, tname);
13552 if (ret) {
13553 bpf_log(log, "%s() is not modifiable\n", tname);
13554 return ret;
13555 }
Alexei Starovoitov18644ce2020-05-28 21:38:36 -070013556 }
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013557
13558 break;
Martin KaFai Lau38207292019-10-24 17:18:11 -070013559 }
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013560 tgt_info->tgt_addr = addr;
13561 tgt_info->tgt_name = tname;
13562 tgt_info->tgt_type = t;
13563 return 0;
13564}
13565
Jiri Olsa35e38152021-04-29 13:47:12 +020013566BTF_SET_START(btf_id_deny)
13567BTF_ID_UNUSED
13568#ifdef CONFIG_SMP
13569BTF_ID(func, migrate_disable)
13570BTF_ID(func, migrate_enable)
13571#endif
13572#if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU
13573BTF_ID(func, rcu_read_unlock_strict)
13574#endif
13575BTF_SET_END(btf_id_deny)
13576
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013577static int check_attach_btf_id(struct bpf_verifier_env *env)
13578{
13579 struct bpf_prog *prog = env->prog;
Toke Høiland-Jørgensen3aac1ea2020-09-29 14:45:50 +020013580 struct bpf_prog *tgt_prog = prog->aux->dst_prog;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013581 struct bpf_attach_target_info tgt_info = {};
13582 u32 btf_id = prog->aux->attach_btf_id;
13583 struct bpf_trampoline *tr;
13584 int ret;
13585 u64 key;
13586
Alexei Starovoitov79a7f8b2021-05-13 17:36:03 -070013587 if (prog->type == BPF_PROG_TYPE_SYSCALL) {
13588 if (prog->aux->sleepable)
13589 /* attach_btf_id checked to be zero already */
13590 return 0;
13591 verbose(env, "Syscall programs can only be sleepable\n");
13592 return -EINVAL;
13593 }
13594
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013595 if (prog->aux->sleepable && prog->type != BPF_PROG_TYPE_TRACING &&
13596 prog->type != BPF_PROG_TYPE_LSM) {
13597 verbose(env, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n");
13598 return -EINVAL;
13599 }
13600
13601 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS)
13602 return check_struct_ops_btf_id(env);
13603
13604 if (prog->type != BPF_PROG_TYPE_TRACING &&
13605 prog->type != BPF_PROG_TYPE_LSM &&
13606 prog->type != BPF_PROG_TYPE_EXT)
13607 return 0;
13608
13609 ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info);
13610 if (ret)
13611 return ret;
13612
13613 if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) {
Toke Høiland-Jørgensen3aac1ea2020-09-29 14:45:50 +020013614 /* to make freplace equivalent to their targets, they need to
13615 * inherit env->ops and expected_attach_type for the rest of the
13616 * verification
13617 */
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013618 env->ops = bpf_verifier_ops[tgt_prog->type];
13619 prog->expected_attach_type = tgt_prog->expected_attach_type;
13620 }
13621
13622 /* store info about the attachment target that will be used later */
13623 prog->aux->attach_func_proto = tgt_info.tgt_type;
13624 prog->aux->attach_func_name = tgt_info.tgt_name;
13625
Toke Høiland-Jørgensen4a1e7c02020-09-29 14:45:51 +020013626 if (tgt_prog) {
13627 prog->aux->saved_dst_prog_type = tgt_prog->type;
13628 prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type;
13629 }
13630
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013631 if (prog->expected_attach_type == BPF_TRACE_RAW_TP) {
13632 prog->aux->attach_btf_trace = true;
13633 return 0;
13634 } else if (prog->expected_attach_type == BPF_TRACE_ITER) {
13635 if (!bpf_iter_prog_supported(prog))
13636 return -EINVAL;
13637 return 0;
13638 }
13639
13640 if (prog->type == BPF_PROG_TYPE_LSM) {
13641 ret = bpf_lsm_verify_prog(&env->log, prog);
13642 if (ret < 0)
13643 return ret;
Jiri Olsa35e38152021-04-29 13:47:12 +020013644 } else if (prog->type == BPF_PROG_TYPE_TRACING &&
13645 btf_id_set_contains(&btf_id_deny, btf_id)) {
13646 return -EINVAL;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013647 }
13648
Andrii Nakryiko22dc4a02020-12-03 12:46:29 -080013649 key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id);
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013650 tr = bpf_trampoline_get(key, &tgt_info);
13651 if (!tr)
13652 return -ENOMEM;
13653
Toke Høiland-Jørgensen3aac1ea2020-09-29 14:45:50 +020013654 prog->aux->dst_trampoline = tr;
Toke Høiland-Jørgensenf7b12b62020-09-25 23:25:02 +020013655 return 0;
Martin KaFai Lau38207292019-10-24 17:18:11 -070013656}
13657
Alan Maguire76654e62020-09-28 12:31:03 +010013658struct btf *bpf_get_btf_vmlinux(void)
13659{
13660 if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) {
13661 mutex_lock(&bpf_verifier_lock);
13662 if (!btf_vmlinux)
13663 btf_vmlinux = btf_parse_vmlinux();
13664 mutex_unlock(&bpf_verifier_lock);
13665 }
13666 return btf_vmlinux;
13667}
13668
Alexei Starovoitovaf2ac3e2021-05-13 17:36:05 -070013669int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr)
Alexei Starovoitov51580e72014-09-26 00:17:02 -070013670{
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070013671 u64 start_time = ktime_get_ns();
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013672 struct bpf_verifier_env *env;
Martin KaFai Laub9193c12018-03-24 11:44:22 -070013673 struct bpf_verifier_log *log;
Jakub Kicinski9e4c24e2019-01-22 22:45:23 -080013674 int i, len, ret = -EINVAL;
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080013675 bool is_priv;
Alexei Starovoitov51580e72014-09-26 00:17:02 -070013676
Arnd Bergmanneba0c922017-11-02 12:05:52 +010013677 /* no program is valid */
13678 if (ARRAY_SIZE(bpf_verifier_ops) == 0)
13679 return -EINVAL;
13680
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013681 /* 'struct bpf_verifier_env' can be global, but since it's not small,
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013682 * allocate/free it every time bpf_check() is called
13683 */
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013684 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013685 if (!env)
13686 return -ENOMEM;
Jakub Kicinski61bd5212017-10-09 10:30:11 -070013687 log = &env->log;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013688
Jakub Kicinski9e4c24e2019-01-22 22:45:23 -080013689 len = (*prog)->len;
Kees Cookfad953c2018-06-12 14:27:37 -070013690 env->insn_aux_data =
Jakub Kicinski9e4c24e2019-01-22 22:45:23 -080013691 vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len));
Jakub Kicinski3df126f2016-09-21 11:43:56 +010013692 ret = -ENOMEM;
13693 if (!env->insn_aux_data)
13694 goto err_free_env;
Jakub Kicinski9e4c24e2019-01-22 22:45:23 -080013695 for (i = 0; i < len; i++)
13696 env->insn_aux_data[i].orig_idx = i;
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013697 env->prog = *prog;
Jakub Kicinski00176a32017-10-16 16:40:54 -070013698 env->ops = bpf_verifier_ops[env->prog->type];
Alexei Starovoitov387544b2021-05-13 17:36:10 -070013699 env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel);
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -070013700 is_priv = bpf_capable();
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013701
Alan Maguire76654e62020-09-28 12:31:03 +010013702 bpf_get_btf_vmlinux();
Alexei Starovoitov8580ac92019-10-15 20:24:57 -070013703
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013704 /* grab the mutex to protect few globals used by verifier */
Alexei Starovoitov45a73c12019-04-19 07:44:55 -070013705 if (!is_priv)
13706 mutex_lock(&bpf_verifier_lock);
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013707
13708 if (attr->log_level || attr->log_buf || attr->log_size) {
13709 /* user requested verbose verifier output
13710 * and supplied buffer to store the verification trace
13711 */
Jakub Kicinskie7bf8242017-10-09 10:30:10 -070013712 log->level = attr->log_level;
13713 log->ubuf = (char __user *) (unsigned long) attr->log_buf;
13714 log->len_total = attr->log_size;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013715
13716 ret = -EINVAL;
Jakub Kicinskie7bf8242017-10-09 10:30:10 -070013717 /* log attributes have to be sane */
Alexei Starovoitov7a9f5c62019-04-01 21:27:46 -070013718 if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 ||
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070013719 !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK)
Jakub Kicinski3df126f2016-09-21 11:43:56 +010013720 goto err_unlock;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013721 }
Daniel Borkmann1ad2f582017-05-25 01:05:05 +020013722
Alexei Starovoitov8580ac92019-10-15 20:24:57 -070013723 if (IS_ERR(btf_vmlinux)) {
13724 /* Either gcc or pahole or kernel are broken. */
13725 verbose(env, "in-kernel BTF is malformed\n");
13726 ret = PTR_ERR(btf_vmlinux);
Martin KaFai Lau38207292019-10-24 17:18:11 -070013727 goto skip_full_check;
Alexei Starovoitov8580ac92019-10-15 20:24:57 -070013728 }
13729
Daniel Borkmann1ad2f582017-05-25 01:05:05 +020013730 env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT);
13731 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
David S. Millere07b98d2017-05-10 11:38:07 -070013732 env->strict_alignment = true;
David Millere9ee9ef2018-11-30 21:08:14 -080013733 if (attr->prog_flags & BPF_F_ANY_ALIGNMENT)
13734 env->strict_alignment = false;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013735
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -070013736 env->allow_ptr_leaks = bpf_allow_ptr_leaks();
Andrei Matei01f810a2021-02-06 20:10:24 -050013737 env->allow_uninit_stack = bpf_allow_uninit_stack();
Andrey Ignatov41c48f32020-06-19 14:11:43 -070013738 env->allow_ptr_to_map_access = bpf_allow_ptr_to_map_access();
Alexei Starovoitov2c78ee82020-05-13 16:03:54 -070013739 env->bypass_spec_v1 = bpf_bypass_spec_v1();
13740 env->bypass_spec_v4 = bpf_bypass_spec_v4();
13741 env->bpf_capable = bpf_capable();
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080013742
Alexei Starovoitov10d274e2019-08-22 22:52:12 -070013743 if (is_priv)
13744 env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ;
13745
Alexei Starovoitovdc2a4eb2019-05-21 20:17:07 -070013746 env->explored_states = kvcalloc(state_htab_size(env),
Jakub Kicinski58e2af8b2016-09-21 11:43:57 +010013747 sizeof(struct bpf_verifier_state_list *),
Alexei Starovoitovf1bca822014-09-29 18:50:01 -070013748 GFP_USER);
13749 ret = -ENOMEM;
13750 if (!env->explored_states)
13751 goto skip_full_check;
13752
Martin KaFai Laue6ac2452021-03-24 18:51:42 -070013753 ret = add_subprog_and_kfunc(env);
13754 if (ret < 0)
13755 goto skip_full_check;
13756
Martin KaFai Laud9762e82018-12-13 10:41:48 -080013757 ret = check_subprogs(env);
Alexei Starovoitov475fb782014-09-26 00:17:05 -070013758 if (ret < 0)
13759 goto skip_full_check;
13760
Martin KaFai Lauc454a462018-12-07 16:42:25 -080013761 ret = check_btf_info(env, attr, uattr);
Yonghong Song838e9692018-11-19 15:29:11 -080013762 if (ret < 0)
13763 goto skip_full_check;
13764
Alexei Starovoitovbe8704f2020-01-20 16:53:46 -080013765 ret = check_attach_btf_id(env);
13766 if (ret)
13767 goto skip_full_check;
13768
Hao Luo4976b712020-09-29 16:50:44 -070013769 ret = resolve_pseudo_ldimm64(env);
13770 if (ret < 0)
13771 goto skip_full_check;
13772
Yinjun Zhangceb11672021-05-20 10:58:34 +020013773 if (bpf_prog_is_dev_bound(env->prog->aux)) {
13774 ret = bpf_prog_offload_verifier_prep(env->prog);
13775 if (ret)
13776 goto skip_full_check;
13777 }
13778
Martin KaFai Laud9762e82018-12-13 10:41:48 -080013779 ret = check_cfg(env);
13780 if (ret < 0)
13781 goto skip_full_check;
13782
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013783 ret = do_check_subprogs(env);
13784 ret = ret ?: do_check_main(env);
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013785
Quentin Monnetc941ce92018-10-07 12:56:47 +010013786 if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux))
13787 ret = bpf_prog_offload_finalize(env);
13788
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013789skip_full_check:
Alexei Starovoitov51c39bb2020-01-09 22:41:20 -080013790 kvfree(env->explored_states);
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013791
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013792 if (ret == 0)
Alexei Starovoitov70a87ff2017-12-25 13:15:40 -080013793 ret = check_max_stack_depth(env);
13794
Jakub Kicinski9b38c402018-12-19 22:13:06 -080013795 /* instruction rewrites happen after this point */
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080013796 if (is_priv) {
13797 if (ret == 0)
13798 opt_hard_wire_dead_code_branches(env);
Jakub Kicinski52875a02019-01-22 22:45:20 -080013799 if (ret == 0)
13800 ret = opt_remove_dead_code(env);
Jakub Kicinskia1b14ab2019-01-22 22:45:21 -080013801 if (ret == 0)
13802 ret = opt_remove_nops(env);
Jakub Kicinski52875a02019-01-22 22:45:20 -080013803 } else {
13804 if (ret == 0)
13805 sanitize_dead_code(env);
Jakub Kicinskie2ae4ca2019-01-22 22:45:19 -080013806 }
13807
Jakub Kicinski9b38c402018-12-19 22:13:06 -080013808 if (ret == 0)
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013809 /* program is valid, convert *(u32*)(ctx + off) accesses */
13810 ret = convert_ctx_accesses(env);
13811
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070013812 if (ret == 0)
Brendan Jackmane6ac5932021-02-17 10:45:09 +000013813 ret = do_misc_fixups(env);
Alexei Starovoitove245c5c62017-03-15 18:26:39 -070013814
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010013815 /* do 32-bit optimization after insn patching has done so those patched
13816 * insns could be handled correctly.
13817 */
Jiong Wangd6c23082019-05-24 23:25:18 +010013818 if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) {
13819 ret = opt_subreg_zext_lo32_rnd_hi32(env, attr);
13820 env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret
13821 : false;
Jiong Wanga4b1d3c2019-05-24 23:25:15 +010013822 }
13823
Alexei Starovoitov1ea47e02017-12-14 17:55:13 -080013824 if (ret == 0)
13825 ret = fixup_call_args(env);
13826
Alexei Starovoitov06ee7112019-04-01 21:27:40 -070013827 env->verification_time = ktime_get_ns() - start_time;
13828 print_verification_stats(env);
13829
Jakub Kicinskia2a7d572017-10-09 10:30:15 -070013830 if (log->level && bpf_verifier_log_full(log))
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013831 ret = -ENOSPC;
Jakub Kicinskia2a7d572017-10-09 10:30:15 -070013832 if (log->level && !log->ubuf) {
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013833 ret = -EFAULT;
Jakub Kicinskia2a7d572017-10-09 10:30:15 -070013834 goto err_release_maps;
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013835 }
13836
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080013837 if (ret)
13838 goto err_release_maps;
13839
13840 if (env->used_map_cnt) {
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013841 /* if program passed verifier, update used_maps in bpf_prog_info */
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013842 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
13843 sizeof(env->used_maps[0]),
13844 GFP_KERNEL);
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013845
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013846 if (!env->prog->aux->used_maps) {
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013847 ret = -ENOMEM;
Jakub Kicinskia2a7d572017-10-09 10:30:15 -070013848 goto err_release_maps;
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013849 }
13850
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013851 memcpy(env->prog->aux->used_maps, env->used_maps,
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013852 sizeof(env->used_maps[0]) * env->used_map_cnt);
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013853 env->prog->aux->used_map_cnt = env->used_map_cnt;
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080013854 }
13855 if (env->used_btf_cnt) {
13856 /* if program passed verifier, update used_btfs in bpf_prog_aux */
13857 env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt,
13858 sizeof(env->used_btfs[0]),
13859 GFP_KERNEL);
13860 if (!env->prog->aux->used_btfs) {
13861 ret = -ENOMEM;
13862 goto err_release_maps;
13863 }
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013864
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080013865 memcpy(env->prog->aux->used_btfs, env->used_btfs,
13866 sizeof(env->used_btfs[0]) * env->used_btf_cnt);
13867 env->prog->aux->used_btf_cnt = env->used_btf_cnt;
13868 }
13869 if (env->used_map_cnt || env->used_btf_cnt) {
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013870 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
13871 * bpf_ld_imm64 instructions
13872 */
13873 convert_pseudo_ld_imm64(env);
13874 }
Alexei Starovoitovcbd35702014-09-26 00:17:03 -070013875
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080013876 adjust_btf_func(env);
Yonghong Songba64e7d2018-11-24 23:20:44 -080013877
Jakub Kicinskia2a7d572017-10-09 10:30:15 -070013878err_release_maps:
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013879 if (!env->prog->aux->used_maps)
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013880 /* if we didn't copy map pointers into bpf_prog_info, release
Jakub Kicinskiab7f5bf2018-05-03 18:37:17 -070013881 * them now. Otherwise free_used_maps() will release them.
Alexei Starovoitov0246e642014-09-26 00:17:04 -070013882 */
13883 release_maps(env);
Andrii Nakryiko541c3ba2021-01-11 23:55:18 -080013884 if (!env->prog->aux->used_btfs)
13885 release_btfs(env);
Toke Høiland-Jørgensen03f87c02020-04-24 15:34:27 +020013886
13887 /* extension progs temporarily inherit the attach_type of their targets
13888 for verification purposes, so set it back to zero before returning
13889 */
13890 if (env->prog->type == BPF_PROG_TYPE_EXT)
13891 env->prog->expected_attach_type = 0;
13892
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -070013893 *prog = env->prog;
Jakub Kicinski3df126f2016-09-21 11:43:56 +010013894err_unlock:
Alexei Starovoitov45a73c12019-04-19 07:44:55 -070013895 if (!is_priv)
13896 mutex_unlock(&bpf_verifier_lock);
Jakub Kicinski3df126f2016-09-21 11:43:56 +010013897 vfree(env->insn_aux_data);
13898err_free_env:
13899 kfree(env);
Alexei Starovoitov51580e72014-09-26 00:17:02 -070013900 return ret;
13901}