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