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 	if (!e)
61 		e = search_bpf_extables(addr);
62 	return e;
63 }
64 
init_kernel_text(unsigned long addr)65 int init_kernel_text(unsigned long addr)
66 {
67 	if (addr >= (unsigned long)_sinittext &&
68 	    addr < (unsigned long)_einittext)
69 		return 1;
70 	return 0;
71 }
72 
core_kernel_text(unsigned long addr)73 int notrace core_kernel_text(unsigned long addr)
74 {
75 	if (addr >= (unsigned long)_stext &&
76 	    addr < (unsigned long)_etext)
77 		return 1;
78 
79 	if (system_state < SYSTEM_RUNNING &&
80 	    init_kernel_text(addr))
81 		return 1;
82 	return 0;
83 }
84 
85 /**
86  * core_kernel_data - tell if addr points to kernel data
87  * @addr: address to test
88  *
89  * Returns true if @addr passed in is from the core kernel data
90  * section.
91  *
92  * Note: On some archs it may return true for core RODATA, and false
93  *  for others. But will always be true for core RW data.
94  */
core_kernel_data(unsigned long addr)95 int core_kernel_data(unsigned long addr)
96 {
97 	if (addr >= (unsigned long)_sdata &&
98 	    addr < (unsigned long)_edata)
99 		return 1;
100 	return 0;
101 }
102 
__kernel_text_address(unsigned long addr)103 int __kernel_text_address(unsigned long addr)
104 {
105 	if (kernel_text_address(addr))
106 		return 1;
107 	/*
108 	 * There might be init symbols in saved stacktraces.
109 	 * Give those symbols a chance to be printed in
110 	 * backtraces (such as lockdep traces).
111 	 *
112 	 * Since we are after the module-symbols check, there's
113 	 * no danger of address overlap:
114 	 */
115 	if (init_kernel_text(addr))
116 		return 1;
117 	return 0;
118 }
119 
kernel_text_address(unsigned long addr)120 int kernel_text_address(unsigned long addr)
121 {
122 	bool no_rcu;
123 	int ret = 1;
124 
125 	if (core_kernel_text(addr))
126 		return 1;
127 
128 	/*
129 	 * If a stack dump happens while RCU is not watching, then
130 	 * RCU needs to be notified that it requires to start
131 	 * watching again. This can happen either by tracing that
132 	 * triggers a stack trace, or a WARN() that happens during
133 	 * coming back from idle, or cpu on or offlining.
134 	 *
135 	 * is_module_text_address() as well as the kprobe slots,
136 	 * is_bpf_text_address() and is_bpf_image_address require
137 	 * RCU to be watching.
138 	 */
139 	no_rcu = !rcu_is_watching();
140 
141 	/* Treat this like an NMI as it can happen anywhere */
142 	if (no_rcu)
143 		rcu_nmi_enter();
144 
145 	if (is_module_text_address(addr))
146 		goto out;
147 	if (is_ftrace_trampoline(addr))
148 		goto out;
149 	if (is_kprobe_optinsn_slot(addr) || is_kprobe_insn_slot(addr))
150 		goto out;
151 	if (is_bpf_text_address(addr))
152 		goto out;
153 	ret = 0;
154 out:
155 	if (no_rcu)
156 		rcu_nmi_exit();
157 
158 	return ret;
159 }
160 
161 /*
162  * On some architectures (PPC64, IA64) function pointers
163  * are actually only tokens to some data that then holds the
164  * real function address. As a result, to find if a function
165  * pointer is part of the kernel text, we need to do some
166  * special dereferencing first.
167  */
func_ptr_is_kernel_text(void * ptr)168 int func_ptr_is_kernel_text(void *ptr)
169 {
170 	unsigned long addr;
171 	addr = (unsigned long) dereference_function_descriptor(ptr);
172 	if (core_kernel_text(addr))
173 		return 1;
174 	return is_module_text_address(addr);
175 }
176