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 && __stop___ex_table > __start___ex_table) {
38 		pr_notice("Sorting __ex_table...\n");
39 		sort_extable(__start___ex_table, __stop___ex_table);
40 	}
41 }
42 
43 /* Given an address, look for it in the kernel exception table */
44 const
search_kernel_exception_table(unsigned long addr)45 struct exception_table_entry *search_kernel_exception_table(unsigned long addr)
46 {
47 	return search_extable(__start___ex_table,
48 			      __stop___ex_table - __start___ex_table, addr);
49 }
50 
51 /* Given an address, look for it in the exception tables. */
search_exception_tables(unsigned long addr)52 const struct exception_table_entry *search_exception_tables(unsigned long addr)
53 {
54 	const struct exception_table_entry *e;
55 
56 	e = search_kernel_exception_table(addr);
57 	if (!e)
58 		e = search_module_extables(addr);
59 	return e;
60 }
61 
init_kernel_text(unsigned long addr)62 int init_kernel_text(unsigned long addr)
63 {
64 	if (addr >= (unsigned long)_sinittext &&
65 	    addr < (unsigned long)_einittext)
66 		return 1;
67 	return 0;
68 }
69 
core_kernel_text(unsigned long addr)70 int notrace core_kernel_text(unsigned long addr)
71 {
72 	if (addr >= (unsigned long)_stext &&
73 	    addr < (unsigned long)_etext)
74 		return 1;
75 
76 	if (system_state < SYSTEM_RUNNING &&
77 	    init_kernel_text(addr))
78 		return 1;
79 	return 0;
80 }
81 
82 /**
83  * core_kernel_data - tell if addr points to kernel data
84  * @addr: address to test
85  *
86  * Returns true if @addr passed in is from the core kernel data
87  * section.
88  *
89  * Note: On some archs it may return true for core RODATA, and false
90  *  for others. But will always be true for core RW data.
91  */
core_kernel_data(unsigned long addr)92 int core_kernel_data(unsigned long addr)
93 {
94 	if (addr >= (unsigned long)_sdata &&
95 	    addr < (unsigned long)_edata)
96 		return 1;
97 	return 0;
98 }
99 
__kernel_text_address(unsigned long addr)100 int __kernel_text_address(unsigned long addr)
101 {
102 	if (kernel_text_address(addr))
103 		return 1;
104 	/*
105 	 * There might be init symbols in saved stacktraces.
106 	 * Give those symbols a chance to be printed in
107 	 * backtraces (such as lockdep traces).
108 	 *
109 	 * Since we are after the module-symbols check, there's
110 	 * no danger of address overlap:
111 	 */
112 	if (init_kernel_text(addr))
113 		return 1;
114 	return 0;
115 }
116 
kernel_text_address(unsigned long addr)117 int kernel_text_address(unsigned long addr)
118 {
119 	bool no_rcu;
120 	int ret = 1;
121 
122 	if (core_kernel_text(addr))
123 		return 1;
124 
125 	/*
126 	 * If a stack dump happens while RCU is not watching, then
127 	 * RCU needs to be notified that it requires to start
128 	 * watching again. This can happen either by tracing that
129 	 * triggers a stack trace, or a WARN() that happens during
130 	 * coming back from idle, or cpu on or offlining.
131 	 *
132 	 * is_module_text_address() as well as the kprobe slots
133 	 * and is_bpf_text_address() require RCU to be watching.
134 	 */
135 	no_rcu = !rcu_is_watching();
136 
137 	/* Treat this like an NMI as it can happen anywhere */
138 	if (no_rcu)
139 		rcu_nmi_enter();
140 
141 	if (is_module_text_address(addr))
142 		goto out;
143 	if (is_ftrace_trampoline(addr))
144 		goto out;
145 	if (is_kprobe_optinsn_slot(addr) || is_kprobe_insn_slot(addr))
146 		goto out;
147 	if (is_bpf_text_address(addr))
148 		goto out;
149 	ret = 0;
150 out:
151 	if (no_rcu)
152 		rcu_nmi_exit();
153 
154 	return ret;
155 }
156 
157 /*
158  * On some architectures (PPC64, IA64) function pointers
159  * are actually only tokens to some data that then holds the
160  * real function address. As a result, to find if a function
161  * pointer is part of the kernel text, we need to do some
162  * special dereferencing first.
163  */
func_ptr_is_kernel_text(void * ptr)164 int func_ptr_is_kernel_text(void *ptr)
165 {
166 	unsigned long addr;
167 	addr = (unsigned long) dereference_function_descriptor(ptr);
168 	if (core_kernel_text(addr))
169 		return 1;
170 	return is_module_text_address(addr);
171 }
172