kernel/extable.c

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