Merge branch 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.git] / 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 noinline __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         trace->skip++;
60         __save_stack_trace(trace, current, NULL, false);
61 }
62 EXPORT_SYMBOL_GPL(save_stack_trace);
63
64 void save_stack_trace_regs(struct pt_regs *regs, struct stack_trace *trace)
65 {
66         __save_stack_trace(trace, current, regs, false);
67 }
68
69 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
70 {
71         if (!try_get_task_stack(tsk))
72                 return;
73
74         if (tsk == current)
75                 trace->skip++;
76         __save_stack_trace(trace, tsk, NULL, true);
77
78         put_task_stack(tsk);
79 }
80 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
81
82 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
83
84 #define STACKTRACE_DUMP_ONCE(task) ({                           \
85         static bool __section(.data.unlikely) __dumped;         \
86                                                                 \
87         if (!__dumped) {                                        \
88                 __dumped = true;                                \
89                 WARN_ON(1);                                     \
90                 show_stack(task, NULL);                         \
91         }                                                       \
92 })
93
94 static int __always_inline
95 __save_stack_trace_reliable(struct stack_trace *trace,
96                             struct task_struct *task)
97 {
98         struct unwind_state state;
99         struct pt_regs *regs;
100         unsigned long addr;
101
102         for (unwind_start(&state, task, NULL, NULL); !unwind_done(&state);
103              unwind_next_frame(&state)) {
104
105                 regs = unwind_get_entry_regs(&state, NULL);
106                 if (regs) {
107                         /*
108                          * Kernel mode registers on the stack indicate an
109                          * in-kernel interrupt or exception (e.g., preemption
110                          * or a page fault), which can make frame pointers
111                          * unreliable.
112                          */
113                         if (!user_mode(regs))
114                                 return -EINVAL;
115
116                         /*
117                          * The last frame contains the user mode syscall
118                          * pt_regs.  Skip it and finish the unwind.
119                          */
120                         unwind_next_frame(&state);
121                         if (!unwind_done(&state)) {
122                                 STACKTRACE_DUMP_ONCE(task);
123                                 return -EINVAL;
124                         }
125                         break;
126                 }
127
128                 addr = unwind_get_return_address(&state);
129
130                 /*
131                  * A NULL or invalid return address probably means there's some
132                  * generated code which __kernel_text_address() doesn't know
133                  * about.
134                  */
135                 if (!addr) {
136                         STACKTRACE_DUMP_ONCE(task);
137                         return -EINVAL;
138                 }
139
140                 if (save_stack_address(trace, addr, false))
141                         return -EINVAL;
142         }
143
144         /* Check for stack corruption */
145         if (unwind_error(&state)) {
146                 STACKTRACE_DUMP_ONCE(task);
147                 return -EINVAL;
148         }
149
150         if (trace->nr_entries < trace->max_entries)
151                 trace->entries[trace->nr_entries++] = ULONG_MAX;
152
153         return 0;
154 }
155
156 /*
157  * This function returns an error if it detects any unreliable features of the
158  * stack.  Otherwise it guarantees that the stack trace is reliable.
159  *
160  * If the task is not 'current', the caller *must* ensure the task is inactive.
161  */
162 int save_stack_trace_tsk_reliable(struct task_struct *tsk,
163                                   struct stack_trace *trace)
164 {
165         int ret;
166
167         /*
168          * If the task doesn't have a stack (e.g., a zombie), the stack is
169          * "reliably" empty.
170          */
171         if (!try_get_task_stack(tsk))
172                 return 0;
173
174         ret = __save_stack_trace_reliable(trace, tsk);
175
176         put_task_stack(tsk);
177
178         return ret;
179 }
180 #endif /* CONFIG_HAVE_RELIABLE_STACKTRACE */
181
182 /* Userspace stacktrace - based on kernel/trace/trace_sysprof.c */
183
184 struct stack_frame_user {
185         const void __user       *next_fp;
186         unsigned long           ret_addr;
187 };
188
189 static int
190 copy_stack_frame(const void __user *fp, struct stack_frame_user *frame)
191 {
192         int ret;
193
194         if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
195                 return 0;
196
197         ret = 1;
198         pagefault_disable();
199         if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
200                 ret = 0;
201         pagefault_enable();
202
203         return ret;
204 }
205
206 static inline void __save_stack_trace_user(struct stack_trace *trace)
207 {
208         const struct pt_regs *regs = task_pt_regs(current);
209         const void __user *fp = (const void __user *)regs->bp;
210
211         if (trace->nr_entries < trace->max_entries)
212                 trace->entries[trace->nr_entries++] = regs->ip;
213
214         while (trace->nr_entries < trace->max_entries) {
215                 struct stack_frame_user frame;
216
217                 frame.next_fp = NULL;
218                 frame.ret_addr = 0;
219                 if (!copy_stack_frame(fp, &frame))
220                         break;
221                 if ((unsigned long)fp < regs->sp)
222                         break;
223                 if (frame.ret_addr) {
224                         trace->entries[trace->nr_entries++] =
225                                 frame.ret_addr;
226                 }
227                 if (fp == frame.next_fp)
228                         break;
229                 fp = frame.next_fp;
230         }
231 }
232
233 void save_stack_trace_user(struct stack_trace *trace)
234 {
235         /*
236          * Trace user stack if we are not a kernel thread
237          */
238         if (current->mm) {
239                 __save_stack_trace_user(trace);
240         }
241         if (trace->nr_entries < trace->max_entries)
242                 trace->entries[trace->nr_entries++] = ULONG_MAX;
243 }