arch/x86/kernel/stacktrace.c

   1 /*
   2  * Stack trace management functions
   3  *
   4  *  Copyright (C) 2006-2009 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
   5  */
   6 #include <linux/sched.h>
   7 #include <linux/sched/debug.h>
   8 #include <linux/sched/task_stack.h>
   9 #include <linux/stacktrace.h>
  10 #include <linux/export.h>
  11 #include <linux/uaccess.h>
  12 #include <asm/stacktrace.h>
  13 #include <asm/unwind.h>
  14
  15 static int save_stack_address(struct stack_trace *trace, unsigned long addr,
  16                               bool nosched)
  17 {
  18         if (nosched && in_sched_functions(addr))
  19                 return 0;
  20
  21         if (trace->skip > 0) {
  22                 trace->skip--;
  23                 return 0;
  24         }
  25
  26         if (trace->nr_entries >= trace->max_entries)
  27                 return -1;
  28
  29         trace->entries[trace->nr_entries++] = addr;
  30         return 0;
  31 }
  32
  33 static void __save_stack_trace(struct stack_trace *trace,
  34                                struct task_struct *task, struct pt_regs *regs,
  35                                bool nosched)
  36 {
  37         struct unwind_state state;
  38         unsigned long addr;
  39
  40         if (regs)
  41                 save_stack_address(trace, regs->ip, nosched);
  42
  43         for (unwind_start(&state, task, regs, NULL); !unwind_done(&state);
  44              unwind_next_frame(&state)) {
  45                 addr = unwind_get_return_address(&state);
  46                 if (!addr || save_stack_address(trace, addr, nosched))
  47                         break;
  48         }
  49
  50         if (trace->nr_entries < trace->max_entries)
  51                 trace->entries[trace->nr_entries++] = ULONG_MAX;
  52 }
  53
  54 /*
  55  * Save stack-backtrace addresses into a stack_trace buffer.
  56  */
  57 void save_stack_trace(struct stack_trace *trace)
  58 {
  59         __save_stack_trace(trace, current, NULL, false);
  60 }
  61 EXPORT_SYMBOL_GPL(save_stack_trace);
  62
  63 void save_stack_trace_regs(struct pt_regs *regs, struct stack_trace *trace)
  64 {
  65         __save_stack_trace(trace, current, regs, false);
  66 }
  67
  68 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
  69 {
  70         if (!try_get_task_stack(tsk))
  71                 return;
  72
  73         __save_stack_trace(trace, tsk, NULL, true);
  74
  75         put_task_stack(tsk);
  76 }
  77 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
  78
  79 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
  80
  81 #define STACKTRACE_DUMP_ONCE(task) ({                           \
  82         static bool __section(.data.unlikely) __dumped;         \
  83                                                                 \
  84         if (!__dumped) {                                        \
  85                 __dumped = true;                                \
  86                 WARN_ON(1);                                     \
  87                 show_stack(task, NULL);                         \
  88         }                                                       \
  89 })
  90
  91 static int __save_stack_trace_reliable(struct stack_trace *trace,
  92                                        struct task_struct *task)
  93 {
  94         struct unwind_state state;
  95         struct pt_regs *regs;
  96         unsigned long addr;
  97
  98         for (unwind_start(&state, task, NULL, NULL); !unwind_done(&state);
  99              unwind_next_frame(&state)) {
 100
 101                 regs = unwind_get_entry_regs(&state);
 102                 if (regs) {
 103                         /*
 104                          * Kernel mode registers on the stack indicate an
 105                          * in-kernel interrupt or exception (e.g., preemption
 106                          * or a page fault), which can make frame pointers
 107                          * unreliable.
 108                          */
 109                         if (!user_mode(regs))
 110                                 return -EINVAL;
 111
 112                         /*
 113                          * The last frame contains the user mode syscall
 114                          * pt_regs.  Skip it and finish the unwind.
 115                          */
 116                         unwind_next_frame(&state);
 117                         if (!unwind_done(&state)) {
 118                                 STACKTRACE_DUMP_ONCE(task);
 119                                 return -EINVAL;
 120                         }
 121                         break;
 122                 }
 123
 124                 addr = unwind_get_return_address(&state);
 125
 126                 /*
 127                  * A NULL or invalid return address probably means there's some
 128                  * generated code which __kernel_text_address() doesn't know
 129                  * about.
 130                  */
 131                 if (!addr) {
 132                         STACKTRACE_DUMP_ONCE(task);
 133                         return -EINVAL;
 134                 }
 135
 136                 if (save_stack_address(trace, addr, false))
 137                         return -EINVAL;
 138         }
 139
 140         /* Check for stack corruption */
 141         if (unwind_error(&state)) {
 142                 STACKTRACE_DUMP_ONCE(task);
 143                 return -EINVAL;
 144         }
 145
 146         if (trace->nr_entries < trace->max_entries)
 147                 trace->entries[trace->nr_entries++] = ULONG_MAX;
 148
 149         return 0;
 150 }
 151
 152 /*
 153  * This function returns an error if it detects any unreliable features of the
 154  * stack.  Otherwise it guarantees that the stack trace is reliable.
 155  *
 156  * If the task is not 'current', the caller *must* ensure the task is inactive.
 157  */
 158 int save_stack_trace_tsk_reliable(struct task_struct *tsk,
 159                                   struct stack_trace *trace)
 160 {
 161         int ret;
 162
 163         if (!try_get_task_stack(tsk))
 164                 return -EINVAL;
 165
 166         ret = __save_stack_trace_reliable(trace, tsk);
 167
 168         put_task_stack(tsk);
 169
 170         return ret;
 171 }
 172 #endif /* CONFIG_HAVE_RELIABLE_STACKTRACE */
 173
 174 /* Userspace stacktrace - based on kernel/trace/trace_sysprof.c */
 175
 176 struct stack_frame_user {
 177         const void __user       *next_fp;
 178         unsigned long           ret_addr;
 179 };
 180
 181 static int
 182 copy_stack_frame(const void __user *fp, struct stack_frame_user *frame)
 183 {
 184         int ret;
 185
 186         if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
 187                 return 0;
 188
 189         ret = 1;
 190         pagefault_disable();
 191         if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
 192                 ret = 0;
 193         pagefault_enable();
 194
 195         return ret;
 196 }
 197
 198 static inline void __save_stack_trace_user(struct stack_trace *trace)
 199 {
 200         const struct pt_regs *regs = task_pt_regs(current);
 201         const void __user *fp = (const void __user *)regs->bp;
 202
 203         if (trace->nr_entries < trace->max_entries)
 204                 trace->entries[trace->nr_entries++] = regs->ip;
 205
 206         while (trace->nr_entries < trace->max_entries) {
 207                 struct stack_frame_user frame;
 208
 209                 frame.next_fp = NULL;
 210                 frame.ret_addr = 0;
 211                 if (!copy_stack_frame(fp, &frame))
 212                         break;
 213                 if ((unsigned long)fp < regs->sp)
 214                         break;
 215                 if (frame.ret_addr) {
 216                         trace->entries[trace->nr_entries++] =
 217                                 frame.ret_addr;
 218                 }
 219                 if (fp == frame.next_fp)
 220                         break;
 221                 fp = frame.next_fp;
 222         }
 223 }
 224
 225 void save_stack_trace_user(struct stack_trace *trace)
 226 {
 227         /*
 228          * Trace user stack if we are not a kernel thread
 229          */
 230         if (current->mm) {
 231                 __save_stack_trace_user(trace);
 232         }
 233         if (trace->nr_entries < trace->max_entries)
 234                 trace->entries[trace->nr_entries++] = ULONG_MAX;
 235 }