• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Rewritten by Rusty Russell, on the backs of many others...
3    Copyright (C) 2001 Rusty Russell, 2002 Rusty Russell IBM.
4 
5 */
6 #include <linux/ftrace.h>
7 #include <linux/memory.h>
8 #include <linux/extable.h>
9 #include <linux/module.h>
10 #include <linux/mutex.h>
11 #include <linux/init.h>
12 #include <linux/kprobes.h>
13 #include <linux/filter.h>
14 
15 #include <asm/sections.h>
16 #include <linux/uaccess.h>
17 
18 /*
19  * mutex protecting text section modification (dynamic code patching).
20  * some users need to sleep (allocating memory...) while they hold this lock.
21  *
22  * Note: Also protects SMP-alternatives modification on x86.
23  *
24  * NOT exported to modules - patching kernel text is a really delicate matter.
25  */
26 DEFINE_MUTEX(text_mutex);
27 
28 extern struct exception_table_entry __start___ex_table[];
29 extern struct exception_table_entry __stop___ex_table[];
30 
31 /* Cleared by build time tools if the table is already sorted. */
32 u32 __initdata __visible main_extable_sort_needed = 1;
33 
34 /* Sort the kernel's built-in exception table */
sort_main_extable(void)35 void __init sort_main_extable(void)
36 {
37 	if (main_extable_sort_needed &&
38 	    &__stop___ex_table > &__start___ex_table) {
39 		pr_notice("Sorting __ex_table...\n");
40 		sort_extable(__start___ex_table, __stop___ex_table);
41 	}
42 }
43 
44 /* Given an address, look for it in the kernel exception table */
45 const
search_kernel_exception_table(unsigned long addr)46 struct exception_table_entry *search_kernel_exception_table(unsigned long addr)
47 {
48 	return search_extable(__start___ex_table,
49 			      __stop___ex_table - __start___ex_table, addr);
50 }
51 
52 /* Given an address, look for it in the exception tables. */
search_exception_tables(unsigned long addr)53 const struct exception_table_entry *search_exception_tables(unsigned long addr)
54 {
55 	const struct exception_table_entry *e;
56 
57 	e = search_kernel_exception_table(addr);
58 	if (!e)
59 		e = search_module_extables(addr);
60 	return e;
61 }
62 
init_kernel_text(unsigned long addr)63 int init_kernel_text(unsigned long addr)
64 {
65 	if (addr >= (unsigned long)_sinittext &&
66 	    addr < (unsigned long)_einittext)
67 		return 1;
68 	return 0;
69 }
70 
core_kernel_text(unsigned long addr)71 int notrace core_kernel_text(unsigned long addr)
72 {
73 	if (addr >= (unsigned long)_stext &&
74 	    addr < (unsigned long)_etext)
75 		return 1;
76 
77 	if (system_state < SYSTEM_RUNNING &&
78 	    init_kernel_text(addr))
79 		return 1;
80 	return 0;
81 }
82 
83 /**
84  * core_kernel_data - tell if addr points to kernel data
85  * @addr: address to test
86  *
87  * Returns true if @addr passed in is from the core kernel data
88  * section.
89  *
90  * Note: On some archs it may return true for core RODATA, and false
91  *  for others. But will always be true for core RW data.
92  */
core_kernel_data(unsigned long addr)93 int core_kernel_data(unsigned long addr)
94 {
95 	if (addr >= (unsigned long)_sdata &&
96 	    addr < (unsigned long)_edata)
97 		return 1;
98 	return 0;
99 }
100 
__kernel_text_address(unsigned long addr)101 int __kernel_text_address(unsigned long addr)
102 {
103 	if (kernel_text_address(addr))
104 		return 1;
105 	/*
106 	 * There might be init symbols in saved stacktraces.
107 	 * Give those symbols a chance to be printed in
108 	 * backtraces (such as lockdep traces).
109 	 *
110 	 * Since we are after the module-symbols check, there's
111 	 * no danger of address overlap:
112 	 */
113 	if (init_kernel_text(addr))
114 		return 1;
115 	return 0;
116 }
117 
kernel_text_address(unsigned long addr)118 int kernel_text_address(unsigned long addr)
119 {
120 	bool no_rcu;
121 	int ret = 1;
122 
123 	if (core_kernel_text(addr))
124 		return 1;
125 
126 	/*
127 	 * If a stack dump happens while RCU is not watching, then
128 	 * RCU needs to be notified that it requires to start
129 	 * watching again. This can happen either by tracing that
130 	 * triggers a stack trace, or a WARN() that happens during
131 	 * coming back from idle, or cpu on or offlining.
132 	 *
133 	 * is_module_text_address() as well as the kprobe slots
134 	 * and is_bpf_text_address() require RCU to be watching.
135 	 */
136 	no_rcu = !rcu_is_watching();
137 
138 	/* Treat this like an NMI as it can happen anywhere */
139 	if (no_rcu)
140 		rcu_nmi_enter();
141 
142 	if (is_module_text_address(addr))
143 		goto out;
144 	if (is_ftrace_trampoline(addr))
145 		goto out;
146 	if (is_kprobe_optinsn_slot(addr) || is_kprobe_insn_slot(addr))
147 		goto out;
148 	if (is_bpf_text_address(addr))
149 		goto out;
150 	ret = 0;
151 out:
152 	if (no_rcu)
153 		rcu_nmi_exit();
154 
155 	return ret;
156 }
157 
158 /*
159  * On some architectures (PPC64, IA64) function pointers
160  * are actually only tokens to some data that then holds the
161  * real function address. As a result, to find if a function
162  * pointer is part of the kernel text, we need to do some
163  * special dereferencing first.
164  */
func_ptr_is_kernel_text(void * ptr)165 int func_ptr_is_kernel_text(void *ptr)
166 {
167 	unsigned long addr;
168 	addr = (unsigned long) dereference_function_descriptor(ptr);
169 	if (core_kernel_text(addr))
170 		return 1;
171 	return is_module_text_address(addr);
172 }
173