blob: e23cce6e6092af76569e882efdee87b10638efd9 [file] [log] [blame]
Thomas Gleixner1a59d1b82019-05-27 08:55:05 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/* Rewritten by Rusty Russell, on the backs of many others...
3 Copyright (C) 2001 Rusty Russell, 2002 Rusty Russell IBM.
4
Linus Torvalds1da177e2005-04-16 15:20:36 -07005*/
Frederic Weisbecker8b96f012008-12-06 03:40:00 +01006#include <linux/ftrace.h>
Dmitri Vorobievf80d2d72009-03-22 19:11:10 +02007#include <linux/memory.h>
Paul Gortmaker8a293be2016-07-23 14:01:45 -04008#include <linux/extable.h>
Ingo Molnar505f2b92009-03-20 11:05:04 +01009#include <linux/module.h>
10#include <linux/mutex.h>
11#include <linux/init.h>
Masami Hiramatsu5b485622017-01-08 23:58:09 +090012#include <linux/kprobes.h>
Daniel Borkmann74451e662017-02-16 22:24:50 +010013#include <linux/filter.h>
Ingo Molnar505f2b92009-03-20 11:05:04 +010014
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <asm/sections.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080016#include <linux/uaccess.h>
Ingo Molnar505f2b92009-03-20 11:05:04 +010017
18/*
19 * mutex protecting text section modification (dynamic code patching).
20 * some users need to sleep (allocating memory...) while they hold this lock.
21 *
Zhou Chengminge846d132017-11-02 09:18:21 +080022 * Note: Also protects SMP-alternatives modification on x86.
23 *
Ingo Molnar505f2b92009-03-20 11:05:04 +010024 * NOT exported to modules - patching kernel text is a really delicate matter.
25 */
26DEFINE_MUTEX(text_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28extern struct exception_table_entry __start___ex_table[];
29extern struct exception_table_entry __stop___ex_table[];
30
David Daneyd219e2e2012-04-19 14:59:56 -070031/* Cleared by build time tools if the table is already sorted. */
Andi Kleen00b71032014-02-08 08:52:04 +010032u32 __initdata __visible main_extable_sort_needed = 1;
David Daneyd219e2e2012-04-19 14:59:56 -070033
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/* Sort the kernel's built-in exception table */
35void __init sort_main_extable(void)
36{
Uwe Kleine-Könige656a632013-09-11 14:23:27 -070037 if (main_extable_sort_needed && __stop___ex_table > __start___ex_table) {
Borislav Petkovbec1b9e2013-04-15 12:51:49 +020038 pr_notice("Sorting __ex_table...\n");
David Daneyd219e2e2012-04-19 14:59:56 -070039 sort_extable(__start___ex_table, __stop___ex_table);
Borislav Petkovbec1b9e2013-04-15 12:51:49 +020040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070041}
42
43/* Given an address, look for it in the exception tables. */
44const struct exception_table_entry *search_exception_tables(unsigned long addr)
45{
46 const struct exception_table_entry *e;
47
Thomas Meyera94c33d2017-07-10 15:51:58 -070048 e = search_extable(__start___ex_table,
49 __stop___ex_table - __start___ex_table, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 if (!e)
51 e = search_module_extables(addr);
52 return e;
53}
54
Josh Poimboeuf9fbcc572018-02-20 11:37:53 -060055int init_kernel_text(unsigned long addr)
Ingo Molnar4a44bac2009-03-19 13:21:44 +010056{
57 if (addr >= (unsigned long)_sinittext &&
Helge Deller5ecbe3c2013-11-28 09:16:33 +010058 addr < (unsigned long)_einittext)
Ingo Molnar4a44bac2009-03-19 13:21:44 +010059 return 1;
60 return 0;
61}
62
Marcin Nowakowskic0d80dd2017-07-06 15:35:31 -070063int notrace core_kernel_text(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064{
65 if (addr >= (unsigned long)_stext &&
Helge Deller5ecbe3c2013-11-28 09:16:33 +010066 addr < (unsigned long)_etext)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 return 1;
68
Thomas Gleixner05947292017-05-16 20:42:44 +020069 if (system_state < SYSTEM_RUNNING &&
Ingo Molnar4a44bac2009-03-19 13:21:44 +010070 init_kernel_text(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 return 1;
72 return 0;
73}
74
Steven Rostedta2d063a2011-05-19 21:34:58 -040075/**
76 * core_kernel_data - tell if addr points to kernel data
77 * @addr: address to test
78 *
79 * Returns true if @addr passed in is from the core kernel data
80 * section.
81 *
82 * Note: On some archs it may return true for core RODATA, and false
83 * for others. But will always be true for core RW data.
84 */
Steven Rostedtcdbe61b2011-05-05 21:14:55 -040085int core_kernel_data(unsigned long addr)
86{
Steven Rostedta2d063a2011-05-19 21:34:58 -040087 if (addr >= (unsigned long)_sdata &&
Steven Rostedtcdbe61b2011-05-05 21:14:55 -040088 addr < (unsigned long)_edata)
89 return 1;
90 return 0;
91}
92
Frederic Weisbecker3861a172009-02-08 00:04:02 +010093int __kernel_text_address(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Steven Rostedt (VMware)9aadde92017-09-22 17:22:19 -040095 if (kernel_text_address(addr))
Daniel Borkmann74451e662017-02-16 22:24:50 +010096 return 1;
Ingo Molnar4a44bac2009-03-19 13:21:44 +010097 /*
98 * There might be init symbols in saved stacktraces.
99 * Give those symbols a chance to be printed in
100 * backtraces (such as lockdep traces).
101 *
102 * Since we are after the module-symbols check, there's
103 * no danger of address overlap:
104 */
105 if (init_kernel_text(addr))
106 return 1;
107 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
110int kernel_text_address(unsigned long addr)
111{
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400112 bool no_rcu;
113 int ret = 1;
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 if (core_kernel_text(addr))
116 return 1;
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400117
118 /*
119 * If a stack dump happens while RCU is not watching, then
120 * RCU needs to be notified that it requires to start
121 * watching again. This can happen either by tracing that
122 * triggers a stack trace, or a WARN() that happens during
123 * coming back from idle, or cpu on or offlining.
124 *
125 * is_module_text_address() as well as the kprobe slots
126 * and is_bpf_text_address() require RCU to be watching.
127 */
128 no_rcu = !rcu_is_watching();
129
130 /* Treat this like an NMI as it can happen anywhere */
131 if (no_rcu)
132 rcu_nmi_enter();
133
Steven Rostedt (Red Hat)aec0be22014-11-18 21:14:11 -0500134 if (is_module_text_address(addr))
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400135 goto out;
Masami Hiramatsu5b485622017-01-08 23:58:09 +0900136 if (is_ftrace_trampoline(addr))
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400137 goto out;
Masami Hiramatsu5b485622017-01-08 23:58:09 +0900138 if (is_kprobe_optinsn_slot(addr) || is_kprobe_insn_slot(addr))
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400139 goto out;
Daniel Borkmann74451e662017-02-16 22:24:50 +0100140 if (is_bpf_text_address(addr))
Steven Rostedt (VMware)e8cac8b2017-09-22 17:36:32 -0400141 goto out;
142 ret = 0;
143out:
144 if (no_rcu)
145 rcu_nmi_exit();
146
147 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
Arjan van de Venab7476c2008-08-15 15:29:38 -0700149
150/*
151 * On some architectures (PPC64, IA64) function pointers
152 * are actually only tokens to some data that then holds the
153 * real function address. As a result, to find if a function
154 * pointer is part of the kernel text, we need to do some
155 * special dereferencing first.
156 */
157int func_ptr_is_kernel_text(void *ptr)
158{
159 unsigned long addr;
160 addr = (unsigned long) dereference_function_descriptor(ptr);
161 if (core_kernel_text(addr))
162 return 1;
Rusty Russella6e6abd2009-03-31 13:05:31 -0600163 return is_module_text_address(addr);
Arjan van de Venab7476c2008-08-15 15:29:38 -0700164}