blob: 408418e6aa13120224517139c17b5acc3e282842 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Alex Kelly10c28d92012-09-26 21:52:08 -04002#include <linux/slab.h>
3#include <linux/file.h>
4#include <linux/fdtable.h>
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -08005#include <linux/freezer.h>
Alex Kelly10c28d92012-09-26 21:52:08 -04006#include <linux/mm.h>
7#include <linux/stat.h>
8#include <linux/fcntl.h>
9#include <linux/swap.h>
Paul Wise315c6922019-08-02 21:49:05 -070010#include <linux/ctype.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040011#include <linux/string.h>
12#include <linux/init.h>
13#include <linux/pagemap.h>
14#include <linux/perf_event.h>
15#include <linux/highmem.h>
16#include <linux/spinlock.h>
17#include <linux/key.h>
18#include <linux/personality.h>
19#include <linux/binfmts.h>
Alex Kelly179899f2012-10-04 17:15:24 -070020#include <linux/coredump.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010021#include <linux/sched/coredump.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010022#include <linux/sched/signal.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010023#include <linux/sched/task_stack.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040024#include <linux/utsname.h>
25#include <linux/pid_namespace.h>
26#include <linux/module.h>
27#include <linux/namei.h>
28#include <linux/mount.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/tsacct_kern.h>
32#include <linux/cn_proc.h>
33#include <linux/audit.h>
34#include <linux/tracehook.h>
35#include <linux/kmod.h>
36#include <linux/fsnotify.h>
37#include <linux/fs_struct.h>
38#include <linux/pipe_fs_i.h>
39#include <linux/oom.h>
40#include <linux/compat.h>
Jann Horn378c6522016-03-22 14:25:36 -070041#include <linux/fs.h>
42#include <linux/path.h>
Arnd Bergmann03927c82015-11-25 16:22:25 +010043#include <linux/timekeeping.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040044
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080045#include <linux/uaccess.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040046#include <asm/mmu_context.h>
47#include <asm/tlb.h>
48#include <asm/exec.h>
49
50#include <trace/events/task.h>
51#include "internal.h"
Alex Kelly10c28d92012-09-26 21:52:08 -040052
53#include <trace/events/sched.h>
54
55int core_uses_pid;
Alex Kelly10c28d92012-09-26 21:52:08 -040056unsigned int core_pipe_limit;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070057char core_pattern[CORENAME_MAX_SIZE] = "core";
58static int core_name_size = CORENAME_MAX_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -040059
60struct core_name {
61 char *corename;
62 int used, size;
63};
Alex Kelly10c28d92012-09-26 21:52:08 -040064
65/* The maximal length of core_pattern is also specified in sysctl.c */
66
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070067static int expand_corename(struct core_name *cn, int size)
Alex Kelly10c28d92012-09-26 21:52:08 -040068{
Oleg Nesterove7fd1542013-07-03 15:08:16 -070069 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -040070
Oleg Nesterove7fd1542013-07-03 15:08:16 -070071 if (!corename)
Alex Kelly10c28d92012-09-26 21:52:08 -040072 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040073
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070074 if (size > core_name_size) /* racy but harmless */
75 core_name_size = size;
76
77 cn->size = ksize(corename);
Oleg Nesterove7fd1542013-07-03 15:08:16 -070078 cn->corename = corename;
Alex Kelly10c28d92012-09-26 21:52:08 -040079 return 0;
80}
81
Nicolas Ioossb4176b72015-06-25 15:03:53 -070082static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
83 va_list arg)
Alex Kelly10c28d92012-09-26 21:52:08 -040084{
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070085 int free, need;
Eric Dumazet404ca802014-04-19 10:15:07 -070086 va_list arg_copy;
Alex Kelly10c28d92012-09-26 21:52:08 -040087
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070088again:
89 free = cn->size - cn->used;
Eric Dumazet404ca802014-04-19 10:15:07 -070090
91 va_copy(arg_copy, arg);
92 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
93 va_end(arg_copy);
94
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070095 if (need < free) {
96 cn->used += need;
97 return 0;
98 }
Alex Kelly10c28d92012-09-26 21:52:08 -040099
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -0700100 if (!expand_corename(cn, cn->size + need - free + 1))
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -0700101 goto again;
Alex Kelly10c28d92012-09-26 21:52:08 -0400102
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -0700103 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -0400104}
105
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700106static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
Oleg Nesterovbc03c692013-07-03 15:08:17 -0700107{
108 va_list arg;
109 int ret;
110
111 va_start(arg, fmt);
112 ret = cn_vprintf(cn, fmt, arg);
113 va_end(arg);
114
115 return ret;
116}
117
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700118static __printf(2, 3)
119int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
Alex Kelly10c28d92012-09-26 21:52:08 -0400120{
Oleg Nesterov923bed032013-07-03 15:08:20 -0700121 int cur = cn->used;
122 va_list arg;
123 int ret;
124
125 va_start(arg, fmt);
126 ret = cn_vprintf(cn, fmt, arg);
127 va_end(arg);
128
Jann Hornac94b6e2016-01-20 15:00:08 -0800129 if (ret == 0) {
130 /*
131 * Ensure that this coredump name component can't cause the
132 * resulting corefile path to consist of a ".." or ".".
133 */
134 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
135 (cn->used - cur == 2 && cn->corename[cur] == '.'
136 && cn->corename[cur+1] == '.'))
137 cn->corename[cur] = '!';
138
139 /*
140 * Empty names are fishy and could be used to create a "//" in a
141 * corefile name, causing the coredump to happen one directory
142 * level too high. Enforce that all components of the core
143 * pattern are at least one character long.
144 */
145 if (cn->used == cur)
146 ret = cn_printf(cn, "!");
147 }
148
Oleg Nesterov923bed032013-07-03 15:08:20 -0700149 for (; cur < cn->used; ++cur) {
150 if (cn->corename[cur] == '/')
151 cn->corename[cur] = '!';
152 }
153 return ret;
Alex Kelly10c28d92012-09-26 21:52:08 -0400154}
155
156static int cn_print_exe_file(struct core_name *cn)
157{
158 struct file *exe_file;
159 char *pathbuf, *path;
160 int ret;
161
162 exe_file = get_mm_exe_file(current->mm);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700163 if (!exe_file)
164 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400165
Michal Hocko0ee931c2017-09-13 16:28:29 -0700166 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -0400167 if (!pathbuf) {
168 ret = -ENOMEM;
169 goto put_exe_file;
170 }
171
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200172 path = file_path(exe_file, pathbuf, PATH_MAX);
Alex Kelly10c28d92012-09-26 21:52:08 -0400173 if (IS_ERR(path)) {
174 ret = PTR_ERR(path);
175 goto free_buf;
176 }
177
Oleg Nesterov923bed032013-07-03 15:08:20 -0700178 ret = cn_esc_printf(cn, "%s", path);
Alex Kelly10c28d92012-09-26 21:52:08 -0400179
180free_buf:
181 kfree(pathbuf);
182put_exe_file:
183 fput(exe_file);
184 return ret;
185}
186
187/* format_corename will inspect the pattern parameter, and output a
188 * name into corename, which must have space for at least
189 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
190 */
Paul Wise315c6922019-08-02 21:49:05 -0700191static int format_corename(struct core_name *cn, struct coredump_params *cprm,
192 size_t **argv, int *argc)
Alex Kelly10c28d92012-09-26 21:52:08 -0400193{
194 const struct cred *cred = current_cred();
195 const char *pat_ptr = core_pattern;
196 int ispipe = (*pat_ptr == '|');
Paul Wise315c6922019-08-02 21:49:05 -0700197 bool was_space = false;
Alex Kelly10c28d92012-09-26 21:52:08 -0400198 int pid_in_pattern = 0;
199 int err = 0;
200
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700201 cn->used = 0;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -0700202 cn->corename = NULL;
203 if (expand_corename(cn, core_name_size))
Alex Kelly10c28d92012-09-26 21:52:08 -0400204 return -ENOMEM;
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700205 cn->corename[0] = '\0';
206
Paul Wise315c6922019-08-02 21:49:05 -0700207 if (ispipe) {
208 int argvs = sizeof(core_pattern) / 2;
209 (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
210 if (!(*argv))
211 return -ENOMEM;
212 (*argv)[(*argc)++] = 0;
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700213 ++pat_ptr;
Sudip Mukherjeedb973a72020-04-20 18:14:20 -0700214 if (!(*pat_ptr))
215 return -ENOMEM;
Paul Wise315c6922019-08-02 21:49:05 -0700216 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400217
218 /* Repeat as long as we have more pattern to process and more output
219 space */
220 while (*pat_ptr) {
Paul Wise315c6922019-08-02 21:49:05 -0700221 /*
222 * Split on spaces before doing template expansion so that
223 * %e and %E don't get split if they have spaces in them
224 */
225 if (ispipe) {
226 if (isspace(*pat_ptr)) {
227 was_space = true;
228 pat_ptr++;
229 continue;
230 } else if (was_space) {
231 was_space = false;
232 err = cn_printf(cn, "%c", '\0');
233 if (err)
234 return err;
235 (*argv)[(*argc)++] = cn->used;
236 }
237 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400238 if (*pat_ptr != '%') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400239 err = cn_printf(cn, "%c", *pat_ptr++);
240 } else {
241 switch (*++pat_ptr) {
242 /* single % at the end, drop that */
243 case 0:
244 goto out;
245 /* Double percent, output one percent */
246 case '%':
247 err = cn_printf(cn, "%c", '%');
248 break;
249 /* pid */
250 case 'p':
251 pid_in_pattern = 1;
252 err = cn_printf(cn, "%d",
253 task_tgid_vnr(current));
254 break;
Stéphane Graber65aafb12013-09-11 14:24:32 -0700255 /* global pid */
256 case 'P':
257 err = cn_printf(cn, "%d",
258 task_tgid_nr(current));
259 break;
Oleg Nesterovb03023e2014-10-13 15:53:35 -0700260 case 'i':
261 err = cn_printf(cn, "%d",
262 task_pid_vnr(current));
263 break;
264 case 'I':
265 err = cn_printf(cn, "%d",
266 task_pid_nr(current));
267 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400268 /* uid */
269 case 'u':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700270 err = cn_printf(cn, "%u",
271 from_kuid(&init_user_ns,
272 cred->uid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400273 break;
274 /* gid */
275 case 'g':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700276 err = cn_printf(cn, "%u",
277 from_kgid(&init_user_ns,
278 cred->gid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400279 break;
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700280 case 'd':
281 err = cn_printf(cn, "%d",
282 __get_dumpable(cprm->mm_flags));
283 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400284 /* signal that caused the coredump */
285 case 's':
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700286 err = cn_printf(cn, "%d",
287 cprm->siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400288 break;
289 /* UNIX time of coredump */
290 case 't': {
Arnd Bergmann03927c82015-11-25 16:22:25 +0100291 time64_t time;
292
293 time = ktime_get_real_seconds();
294 err = cn_printf(cn, "%lld", time);
Alex Kelly10c28d92012-09-26 21:52:08 -0400295 break;
296 }
297 /* hostname */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700298 case 'h':
Alex Kelly10c28d92012-09-26 21:52:08 -0400299 down_read(&uts_sem);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700300 err = cn_esc_printf(cn, "%s",
Alex Kelly10c28d92012-09-26 21:52:08 -0400301 utsname()->nodename);
302 up_read(&uts_sem);
Alex Kelly10c28d92012-09-26 21:52:08 -0400303 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400304 /* executable */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700305 case 'e':
306 err = cn_esc_printf(cn, "%s", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400307 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400308 case 'E':
309 err = cn_print_exe_file(cn);
310 break;
311 /* core limit size */
312 case 'c':
313 err = cn_printf(cn, "%lu",
314 rlimit(RLIMIT_CORE));
315 break;
316 default:
317 break;
318 }
319 ++pat_ptr;
320 }
321
322 if (err)
323 return err;
324 }
325
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700326out:
Alex Kelly10c28d92012-09-26 21:52:08 -0400327 /* Backward compatibility with core_uses_pid:
328 *
329 * If core_pattern does not include a %p (as is the default)
330 * and core_uses_pid is set, then .%pid will be appended to
331 * the filename. Do not do this for piped commands. */
332 if (!ispipe && !pid_in_pattern && core_uses_pid) {
333 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
334 if (err)
335 return err;
336 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400337 return ispipe;
338}
339
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800340static int zap_process(struct task_struct *start, int exit_code, int flags)
Alex Kelly10c28d92012-09-26 21:52:08 -0400341{
342 struct task_struct *t;
343 int nr = 0;
344
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800345 /* ignore all signals except SIGKILL, see prepare_signal() */
346 start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
Alex Kelly10c28d92012-09-26 21:52:08 -0400347 start->signal->group_exit_code = exit_code;
348 start->signal->group_stop_count = 0;
349
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800350 for_each_thread(start, t) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400351 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
352 if (t != current && t->mm) {
353 sigaddset(&t->pending.signal, SIGKILL);
354 signal_wake_up(t, 1);
355 nr++;
356 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800357 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400358
359 return nr;
360}
361
Oleg Nesterov403bad72013-04-30 15:28:10 -0700362static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
363 struct core_state *core_state, int exit_code)
Alex Kelly10c28d92012-09-26 21:52:08 -0400364{
365 struct task_struct *g, *p;
366 unsigned long flags;
367 int nr = -EAGAIN;
368
369 spin_lock_irq(&tsk->sighand->siglock);
370 if (!signal_group_exit(tsk->signal)) {
371 mm->core_state = core_state;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700372 tsk->signal->group_exit_task = tsk;
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800373 nr = zap_process(tsk, exit_code, 0);
Oleg Nesterov403bad72013-04-30 15:28:10 -0700374 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
Alex Kelly10c28d92012-09-26 21:52:08 -0400375 }
376 spin_unlock_irq(&tsk->sighand->siglock);
377 if (unlikely(nr < 0))
378 return nr;
379
Silesh C Vaed8adb2014-07-23 13:59:59 -0700380 tsk->flags |= PF_DUMPCORE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400381 if (atomic_read(&mm->mm_users) == nr + 1)
382 goto done;
383 /*
384 * We should find and kill all tasks which use this mm, and we should
385 * count them correctly into ->nr_threads. We don't take tasklist
386 * lock, but this is safe wrt:
387 *
388 * fork:
389 * None of sub-threads can fork after zap_process(leader). All
390 * processes which were created before this point should be
391 * visible to zap_threads() because copy_process() adds the new
392 * process to the tail of init_task.tasks list, and lock/unlock
393 * of ->siglock provides a memory barrier.
394 *
395 * do_exit:
396 * The caller holds mm->mmap_sem. This means that the task which
397 * uses this mm can't pass exit_mm(), so it can't exit or clear
398 * its ->mm.
399 *
400 * de_thread:
401 * It does list_replace_rcu(&leader->tasks, &current->tasks),
402 * we must see either old or new leader, this does not matter.
403 * However, it can change p->sighand, so lock_task_sighand(p)
404 * must be used. Since p->mm != NULL and we hold ->mmap_sem
405 * it can't fail.
406 *
407 * Note also that "g" can be the old leader with ->mm == NULL
408 * and already unhashed and thus removed from ->thread_group.
409 * This is OK, __unhash_process()->list_del_rcu() does not
410 * clear the ->next pointer, we will find the new leader via
411 * next_thread().
412 */
413 rcu_read_lock();
414 for_each_process(g) {
415 if (g == tsk->group_leader)
416 continue;
417 if (g->flags & PF_KTHREAD)
418 continue;
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800419
420 for_each_thread(g, p) {
421 if (unlikely(!p->mm))
422 continue;
423 if (unlikely(p->mm == mm)) {
424 lock_task_sighand(p, &flags);
425 nr += zap_process(p, exit_code,
426 SIGNAL_GROUP_EXIT);
427 unlock_task_sighand(p, &flags);
Alex Kelly10c28d92012-09-26 21:52:08 -0400428 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800429 break;
430 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400431 }
432 rcu_read_unlock();
433done:
434 atomic_set(&core_state->nr_threads, nr);
435 return nr;
436}
437
438static int coredump_wait(int exit_code, struct core_state *core_state)
439{
440 struct task_struct *tsk = current;
441 struct mm_struct *mm = tsk->mm;
442 int core_waiters = -EBUSY;
443
444 init_completion(&core_state->startup);
445 core_state->dumper.task = tsk;
446 core_state->dumper.next = NULL;
447
Michal Hocko4136c262016-05-23 16:25:57 -0700448 if (down_write_killable(&mm->mmap_sem))
449 return -EINTR;
450
Alex Kelly10c28d92012-09-26 21:52:08 -0400451 if (!mm->core_state)
452 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
453 up_write(&mm->mmap_sem);
454
455 if (core_waiters > 0) {
456 struct core_thread *ptr;
457
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -0800458 freezer_do_not_count();
Alex Kelly10c28d92012-09-26 21:52:08 -0400459 wait_for_completion(&core_state->startup);
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -0800460 freezer_count();
Alex Kelly10c28d92012-09-26 21:52:08 -0400461 /*
462 * Wait for all the threads to become inactive, so that
463 * all the thread context (extended register state, like
464 * fpu etc) gets copied to the memory.
465 */
466 ptr = core_state->dumper.next;
467 while (ptr != NULL) {
468 wait_task_inactive(ptr->task, 0);
469 ptr = ptr->next;
470 }
471 }
472
473 return core_waiters;
474}
475
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700476static void coredump_finish(struct mm_struct *mm, bool core_dumped)
Alex Kelly10c28d92012-09-26 21:52:08 -0400477{
478 struct core_thread *curr, *next;
479 struct task_struct *task;
480
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700481 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700482 if (core_dumped && !__fatal_signal_pending(current))
483 current->signal->group_exit_code |= 0x80;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700484 current->signal->group_exit_task = NULL;
485 current->signal->flags = SIGNAL_GROUP_EXIT;
486 spin_unlock_irq(&current->sighand->siglock);
487
Alex Kelly10c28d92012-09-26 21:52:08 -0400488 next = mm->core_state->dumper.next;
489 while ((curr = next) != NULL) {
490 next = curr->next;
491 task = curr->task;
492 /*
493 * see exit_mm(), curr->task must not see
494 * ->task == NULL before we read ->next.
495 */
496 smp_mb();
497 curr->task = NULL;
498 wake_up_process(task);
499 }
500
501 mm->core_state = NULL;
502}
503
Oleg Nesterov528f8272013-04-30 15:28:15 -0700504static bool dump_interrupted(void)
505{
506 /*
507 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
508 * can do try_to_freeze() and check __fatal_signal_pending(),
509 * but then we need to teach dump_write() to restart and clear
510 * TIF_SIGPENDING.
511 */
512 return signal_pending(current);
513}
514
Alex Kelly10c28d92012-09-26 21:52:08 -0400515static void wait_for_dump_helpers(struct file *file)
516{
Al Virode32ec42013-03-21 11:16:56 -0400517 struct pipe_inode_info *pipe = file->private_data;
Alex Kelly10c28d92012-09-26 21:52:08 -0400518
519 pipe_lock(pipe);
520 pipe->readers++;
521 pipe->writers--;
Linus Torvalds0ddad212019-12-09 09:48:27 -0800522 wake_up_interruptible_sync(&pipe->rd_wait);
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700523 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
524 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400525
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700526 /*
527 * We actually want wait_event_freezable() but then we need
528 * to clear TIF_SIGPENDING and improve dump_interrupted().
529 */
Linus Torvalds0ddad212019-12-09 09:48:27 -0800530 wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
Alex Kelly10c28d92012-09-26 21:52:08 -0400531
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700532 pipe_lock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400533 pipe->readers--;
534 pipe->writers++;
535 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400536}
537
538/*
539 * umh_pipe_setup
540 * helper function to customize the process used
541 * to collect the core in userspace. Specifically
542 * it sets up a pipe and installs it as fd 0 (stdin)
543 * for the process. Returns 0 on success, or
544 * PTR_ERR on failure.
545 * Note that it also sets the core limit to 1. This
546 * is a special value that we use to trap recursive
547 * core dumps
548 */
549static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
550{
551 struct file *files[2];
552 struct coredump_params *cp = (struct coredump_params *)info->data;
553 int err = create_pipe_files(files, 0);
554 if (err)
555 return err;
556
557 cp->file = files[1];
558
Al Viro45525b22012-10-16 13:30:07 -0400559 err = replace_fd(0, files[0], 0);
560 fput(files[0]);
Alex Kelly10c28d92012-09-26 21:52:08 -0400561 /* and disallow core files too */
562 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
563
Al Viro45525b22012-10-16 13:30:07 -0400564 return err;
Alex Kelly10c28d92012-09-26 21:52:08 -0400565}
566
Eric W. Biedermanae7795b2018-09-25 11:27:20 +0200567void do_coredump(const kernel_siginfo_t *siginfo)
Alex Kelly10c28d92012-09-26 21:52:08 -0400568{
569 struct core_state core_state;
570 struct core_name cn;
571 struct mm_struct *mm = current->mm;
572 struct linux_binfmt * binfmt;
573 const struct cred *old_cred;
574 struct cred *cred;
575 int retval = 0;
Alex Kelly10c28d92012-09-26 21:52:08 -0400576 int ispipe;
Paul Wise315c6922019-08-02 21:49:05 -0700577 size_t *argv = NULL;
578 int argc = 0;
Alex Kelly10c28d92012-09-26 21:52:08 -0400579 struct files_struct *displaced;
Jann Hornfbb18162015-09-09 15:38:28 -0700580 /* require nonrelative corefile path and be extra careful */
581 bool need_suid_safe = false;
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700582 bool core_dumped = false;
Alex Kelly10c28d92012-09-26 21:52:08 -0400583 static atomic_t core_dump_count = ATOMIC_INIT(0);
584 struct coredump_params cprm = {
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700585 .siginfo = siginfo,
Al Viro541880d2012-11-05 13:11:26 -0500586 .regs = signal_pt_regs(),
Alex Kelly10c28d92012-09-26 21:52:08 -0400587 .limit = rlimit(RLIMIT_CORE),
588 /*
589 * We must use the same mm->flags while dumping core to avoid
590 * inconsistency of bit flags, since this flag is not protected
591 * by any locks.
592 */
593 .mm_flags = mm->flags,
594 };
595
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700596 audit_core_dumps(siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400597
598 binfmt = mm->binfmt;
599 if (!binfmt || !binfmt->core_dump)
600 goto fail;
601 if (!__get_dumpable(cprm.mm_flags))
602 goto fail;
603
604 cred = prepare_creds();
605 if (!cred)
606 goto fail;
607 /*
608 * We cannot trust fsuid as being the "true" uid of the process
609 * nor do we know its entire history. We only know it was tainted
610 * so we dump it as root in mode 2, and only into a controlled
611 * environment (pipe handler or fully qualified path).
612 */
Kees Cooke579d2c2013-02-27 17:03:15 -0800613 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400614 /* Setuid core dump mode */
Alex Kelly10c28d92012-09-26 21:52:08 -0400615 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
Jann Hornfbb18162015-09-09 15:38:28 -0700616 need_suid_safe = true;
Alex Kelly10c28d92012-09-26 21:52:08 -0400617 }
618
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700619 retval = coredump_wait(siginfo->si_signo, &core_state);
Alex Kelly10c28d92012-09-26 21:52:08 -0400620 if (retval < 0)
621 goto fail_creds;
622
623 old_cred = override_creds(cred);
624
Paul Wise315c6922019-08-02 21:49:05 -0700625 ispipe = format_corename(&cn, &cprm, &argv, &argc);
Alex Kelly10c28d92012-09-26 21:52:08 -0400626
Lucas De Marchifb96c472013-04-30 15:28:06 -0700627 if (ispipe) {
Paul Wise315c6922019-08-02 21:49:05 -0700628 int argi;
Alex Kelly10c28d92012-09-26 21:52:08 -0400629 int dump_count;
630 char **helper_argv;
Lucas De Marchi907ed132013-04-30 15:28:07 -0700631 struct subprocess_info *sub_info;
Alex Kelly10c28d92012-09-26 21:52:08 -0400632
633 if (ispipe < 0) {
634 printk(KERN_WARNING "format_corename failed\n");
635 printk(KERN_WARNING "Aborting core\n");
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700636 goto fail_unlock;
Alex Kelly10c28d92012-09-26 21:52:08 -0400637 }
638
639 if (cprm.limit == 1) {
640 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
641 *
642 * Normally core limits are irrelevant to pipes, since
643 * we're not writing to the file system, but we use
Bastien Nocerafcbc32b2015-02-05 14:35:05 +0100644 * cprm.limit of 1 here as a special value, this is a
Alex Kelly10c28d92012-09-26 21:52:08 -0400645 * consistent way to catch recursive crashes.
646 * We can still crash if the core_pattern binary sets
647 * RLIM_CORE = !1, but it runs as root, and can do
648 * lots of stupid things.
649 *
650 * Note that we use task_tgid_vnr here to grab the pid
651 * of the process group leader. That way we get the
652 * right pid if a thread in a multi-threaded
653 * core_pattern process dies.
654 */
655 printk(KERN_WARNING
656 "Process %d(%s) has RLIMIT_CORE set to 1\n",
657 task_tgid_vnr(current), current->comm);
658 printk(KERN_WARNING "Aborting core\n");
659 goto fail_unlock;
660 }
661 cprm.limit = RLIM_INFINITY;
662
663 dump_count = atomic_inc_return(&core_dump_count);
664 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
665 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
666 task_tgid_vnr(current), current->comm);
667 printk(KERN_WARNING "Skipping core dump\n");
668 goto fail_dropcount;
669 }
670
Paul Wise315c6922019-08-02 21:49:05 -0700671 helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
672 GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -0400673 if (!helper_argv) {
674 printk(KERN_WARNING "%s failed to allocate memory\n",
675 __func__);
676 goto fail_dropcount;
677 }
Paul Wise315c6922019-08-02 21:49:05 -0700678 for (argi = 0; argi < argc; argi++)
679 helper_argv[argi] = cn.corename + argv[argi];
680 helper_argv[argi] = NULL;
Alex Kelly10c28d92012-09-26 21:52:08 -0400681
Lucas De Marchi907ed132013-04-30 15:28:07 -0700682 retval = -ENOMEM;
683 sub_info = call_usermodehelper_setup(helper_argv[0],
684 helper_argv, NULL, GFP_KERNEL,
685 umh_pipe_setup, NULL, &cprm);
686 if (sub_info)
687 retval = call_usermodehelper_exec(sub_info,
688 UMH_WAIT_EXEC);
689
Paul Wise315c6922019-08-02 21:49:05 -0700690 kfree(helper_argv);
Alex Kelly10c28d92012-09-26 21:52:08 -0400691 if (retval) {
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700692 printk(KERN_INFO "Core dump to |%s pipe failed\n",
Alex Kelly10c28d92012-09-26 21:52:08 -0400693 cn.corename);
694 goto close_fail;
Lucas De Marchifb96c472013-04-30 15:28:06 -0700695 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400696 } else {
697 struct inode *inode;
Jann Horn378c6522016-03-22 14:25:36 -0700698 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
699 O_LARGEFILE | O_EXCL;
Alex Kelly10c28d92012-09-26 21:52:08 -0400700
701 if (cprm.limit < binfmt->min_coredump)
702 goto fail_unlock;
703
Jann Hornfbb18162015-09-09 15:38:28 -0700704 if (need_suid_safe && cn.corename[0] != '/') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400705 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
706 "to fully qualified path!\n",
707 task_tgid_vnr(current), current->comm);
708 printk(KERN_WARNING "Skipping core dump\n");
709 goto fail_unlock;
710 }
711
Jann Hornfbb18162015-09-09 15:38:28 -0700712 /*
713 * Unlink the file if it exists unless this is a SUID
714 * binary - in that case, we're running around with root
715 * privs and don't want to unlink another user's coredump.
716 */
717 if (!need_suid_safe) {
Jann Hornfbb18162015-09-09 15:38:28 -0700718 /*
719 * If it doesn't exist, that's fine. If there's some
720 * other problem, we'll catch it at the filp_open().
721 */
Christoph Hellwig96271652017-11-04 13:44:46 +0300722 do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
Jann Hornfbb18162015-09-09 15:38:28 -0700723 }
724
725 /*
726 * There is a race between unlinking and creating the
727 * file, but if that causes an EEXIST here, that's
728 * fine - another process raced with us while creating
729 * the corefile, and the other process won. To userspace,
730 * what matters is that at least one of the two processes
731 * writes its coredump successfully, not which one.
732 */
Jann Horn378c6522016-03-22 14:25:36 -0700733 if (need_suid_safe) {
734 /*
735 * Using user namespaces, normal user tasks can change
736 * their current->fs->root to point to arbitrary
737 * directories. Since the intention of the "only dump
738 * with a fully qualified path" rule is to control where
739 * coredumps may be placed using root privileges,
740 * current->fs->root must not be used. Instead, use the
741 * root directory of init_task.
742 */
743 struct path root;
744
745 task_lock(&init_task);
746 get_fs_root(init_task.fs, &root);
747 task_unlock(&init_task);
748 cprm.file = file_open_root(root.dentry, root.mnt,
749 cn.corename, open_flags, 0600);
750 path_put(&root);
751 } else {
752 cprm.file = filp_open(cn.corename, open_flags, 0600);
753 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400754 if (IS_ERR(cprm.file))
755 goto fail_unlock;
756
Al Viro496ad9a2013-01-23 17:07:38 -0500757 inode = file_inode(cprm.file);
Alex Kelly10c28d92012-09-26 21:52:08 -0400758 if (inode->i_nlink > 1)
759 goto close_fail;
760 if (d_unhashed(cprm.file->f_path.dentry))
761 goto close_fail;
762 /*
763 * AK: actually i see no reason to not allow this for named
764 * pipes etc, but keep the previous behaviour for now.
765 */
766 if (!S_ISREG(inode->i_mode))
767 goto close_fail;
768 /*
Jann Horn40f705a2015-09-09 15:38:30 -0700769 * Don't dump core if the filesystem changed owner or mode
770 * of the file during file creation. This is an issue when
771 * a process dumps core while its cwd is e.g. on a vfat
772 * filesystem.
Alex Kelly10c28d92012-09-26 21:52:08 -0400773 */
774 if (!uid_eq(inode->i_uid, current_fsuid()))
775 goto close_fail;
Jann Horn40f705a2015-09-09 15:38:30 -0700776 if ((inode->i_mode & 0677) != 0600)
777 goto close_fail;
Al Viro86cc0582015-04-03 15:23:17 -0400778 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
Alex Kelly10c28d92012-09-26 21:52:08 -0400779 goto close_fail;
780 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
781 goto close_fail;
782 }
783
784 /* get us an unshared descriptor table; almost always a no-op */
785 retval = unshare_files(&displaced);
786 if (retval)
787 goto close_fail;
788 if (displaced)
789 put_files_struct(displaced);
Al Viroe86d35c2013-05-04 14:45:54 -0400790 if (!dump_interrupted()) {
791 file_start_write(cprm.file);
792 core_dumped = binfmt->core_dump(&cprm);
793 file_end_write(cprm.file);
794 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400795 if (ispipe && core_pipe_limit)
796 wait_for_dump_helpers(cprm.file);
797close_fail:
798 if (cprm.file)
799 filp_close(cprm.file, NULL);
800fail_dropcount:
801 if (ispipe)
802 atomic_dec(&core_dump_count);
803fail_unlock:
Paul Wise315c6922019-08-02 21:49:05 -0700804 kfree(argv);
Alex Kelly10c28d92012-09-26 21:52:08 -0400805 kfree(cn.corename);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700806 coredump_finish(mm, core_dumped);
Alex Kelly10c28d92012-09-26 21:52:08 -0400807 revert_creds(old_cred);
808fail_creds:
809 put_cred(cred);
810fail:
811 return;
812}
813
814/*
815 * Core dumping helper functions. These are the only things you should
816 * do on a core-file: use only these functions to write out all the
817 * necessary info.
818 */
Al Viroecc8c772013-10-05 15:32:35 -0400819int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
820{
821 struct file *file = cprm->file;
Al Viro2507a4f2013-10-08 09:11:48 -0400822 loff_t pos = file->f_pos;
823 ssize_t n;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700824 if (cprm->written + nr > cprm->limit)
Al Viroecc8c772013-10-05 15:32:35 -0400825 return 0;
Al Viro2507a4f2013-10-08 09:11:48 -0400826 while (nr) {
827 if (dump_interrupted())
828 return 0;
Al Viro52da40a2013-11-15 21:58:33 -0500829 n = __kernel_write(file, addr, nr, &pos);
Al Viro2507a4f2013-10-08 09:11:48 -0400830 if (n <= 0)
831 return 0;
832 file->f_pos = pos;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700833 cprm->written += n;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200834 cprm->pos += n;
Al Viro2507a4f2013-10-08 09:11:48 -0400835 nr -= n;
836 }
Al Viroecc8c772013-10-05 15:32:35 -0400837 return 1;
838}
839EXPORT_SYMBOL(dump_emit);
840
Al Viro9b56d542013-10-08 09:26:08 -0400841int dump_skip(struct coredump_params *cprm, size_t nr)
Alex Kelly10c28d92012-09-26 21:52:08 -0400842{
Al Viro9b56d542013-10-08 09:26:08 -0400843 static char zeroes[PAGE_SIZE];
844 struct file *file = cprm->file;
Alex Kelly10c28d92012-09-26 21:52:08 -0400845 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
Oleg Nesterov528f8272013-04-30 15:28:15 -0700846 if (dump_interrupted() ||
Al Viro9b56d542013-10-08 09:26:08 -0400847 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
Alex Kelly10c28d92012-09-26 21:52:08 -0400848 return 0;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200849 cprm->pos += nr;
Al Viro9b56d542013-10-08 09:26:08 -0400850 return 1;
Alex Kelly10c28d92012-09-26 21:52:08 -0400851 } else {
Al Viro9b56d542013-10-08 09:26:08 -0400852 while (nr > PAGE_SIZE) {
853 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
854 return 0;
855 nr -= PAGE_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400856 }
Al Viro9b56d542013-10-08 09:26:08 -0400857 return dump_emit(cprm, zeroes, nr);
Alex Kelly10c28d92012-09-26 21:52:08 -0400858 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400859}
Al Viro9b56d542013-10-08 09:26:08 -0400860EXPORT_SYMBOL(dump_skip);
Al Viro22a8cb82013-10-08 11:05:01 -0400861
862int dump_align(struct coredump_params *cprm, int align)
863{
Mateusz Guzik1607f092016-06-05 23:14:14 +0200864 unsigned mod = cprm->pos & (align - 1);
Al Viro22a8cb82013-10-08 11:05:01 -0400865 if (align & (align - 1))
Al Virodb512422013-11-15 21:55:52 -0500866 return 0;
867 return mod ? dump_skip(cprm, align - mod) : 1;
Al Viro22a8cb82013-10-08 11:05:01 -0400868}
869EXPORT_SYMBOL(dump_align);
Dave Kleikamp4d22c752017-01-11 13:25:00 -0600870
871/*
872 * Ensures that file size is big enough to contain the current file
873 * postion. This prevents gdb from complaining about a truncated file
874 * if the last "write" to the file was dump_skip.
875 */
876void dump_truncate(struct coredump_params *cprm)
877{
878 struct file *file = cprm->file;
879 loff_t offset;
880
881 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
882 offset = file->f_op->llseek(file, 0, SEEK_CUR);
883 if (i_size_read(file->f_mapping->host) < offset)
884 do_truncate(file->f_path.dentry, offset, 0, file);
885 }
886}
887EXPORT_SYMBOL(dump_truncate);