Commit | Line | Data |
---|---|---|
e5a81b62 SR |
1 | /* |
2 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
3 | * | |
4 | */ | |
68db0cf1 | 5 | #include <linux/sched/task_stack.h> |
e5a81b62 SR |
6 | #include <linux/stacktrace.h> |
7 | #include <linux/kallsyms.h> | |
8 | #include <linux/seq_file.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/uaccess.h> | |
e5a81b62 SR |
11 | #include <linux/ftrace.h> |
12 | #include <linux/module.h> | |
f38f1d2a | 13 | #include <linux/sysctl.h> |
e5a81b62 | 14 | #include <linux/init.h> |
762e1207 SR |
15 | |
16 | #include <asm/setup.h> | |
17 | ||
e5a81b62 SR |
18 | #include "trace.h" |
19 | ||
1b6cced6 SR |
20 | static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = |
21 | { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; | |
bb99d8cc | 22 | unsigned stack_trace_index[STACK_TRACE_ENTRIES]; |
1b6cced6 | 23 | |
4df29712 SRRH |
24 | /* |
25 | * Reserve one entry for the passed in ip. This will allow | |
26 | * us to remove most or all of the stack size overhead | |
27 | * added by the stack tracer itself. | |
28 | */ | |
bb99d8cc | 29 | struct stack_trace stack_trace_max = { |
4df29712 | 30 | .max_entries = STACK_TRACE_ENTRIES - 1, |
72ac426a | 31 | .entries = &stack_dump_trace[0], |
e5a81b62 SR |
32 | }; |
33 | ||
bb99d8cc | 34 | unsigned long stack_trace_max_size; |
d332736d | 35 | arch_spinlock_t stack_trace_max_lock = |
edc35bd7 | 36 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
e5a81b62 | 37 | |
8aaf1ee7 | 38 | DEFINE_PER_CPU(int, disable_stack_tracer); |
f38f1d2a SR |
39 | static DEFINE_MUTEX(stack_sysctl_mutex); |
40 | ||
41 | int stack_tracer_enabled; | |
42 | static int last_stack_tracer_enabled; | |
e5a81b62 | 43 | |
bb99d8cc | 44 | void stack_trace_print(void) |
e3172181 MK |
45 | { |
46 | long i; | |
47 | int size; | |
48 | ||
49 | pr_emerg(" Depth Size Location (%d entries)\n" | |
50 | " ----- ---- --------\n", | |
bb99d8cc | 51 | stack_trace_max.nr_entries); |
e3172181 | 52 | |
bb99d8cc | 53 | for (i = 0; i < stack_trace_max.nr_entries; i++) { |
e3172181 MK |
54 | if (stack_dump_trace[i] == ULONG_MAX) |
55 | break; | |
bb99d8cc | 56 | if (i+1 == stack_trace_max.nr_entries || |
e3172181 | 57 | stack_dump_trace[i+1] == ULONG_MAX) |
bb99d8cc | 58 | size = stack_trace_index[i]; |
e3172181 | 59 | else |
bb99d8cc | 60 | size = stack_trace_index[i] - stack_trace_index[i+1]; |
e3172181 | 61 | |
bb99d8cc | 62 | pr_emerg("%3ld) %8d %5d %pS\n", i, stack_trace_index[i], |
e3172181 MK |
63 | size, (void *)stack_dump_trace[i]); |
64 | } | |
65 | } | |
66 | ||
bb99d8cc | 67 | /* |
505d3085 | 68 | * When arch-specific code overrides this function, the following |
d332736d | 69 | * data should be filled up, assuming stack_trace_max_lock is held to |
bb99d8cc AT |
70 | * prevent concurrent updates. |
71 | * stack_trace_index[] | |
72 | * stack_trace_max | |
73 | * stack_trace_max_size | |
74 | */ | |
75 | void __weak | |
d4ecbfc4 | 76 | check_stack(unsigned long ip, unsigned long *stack) |
e5a81b62 | 77 | { |
e3172181 | 78 | unsigned long this_size, flags; unsigned long *p, *top, *start; |
4df29712 SRRH |
79 | static int tracer_frame; |
80 | int frame_size = ACCESS_ONCE(tracer_frame); | |
72ac426a | 81 | int i, x; |
e5a81b62 | 82 | |
87889501 | 83 | this_size = ((unsigned long)stack) & (THREAD_SIZE-1); |
e5a81b62 | 84 | this_size = THREAD_SIZE - this_size; |
4df29712 SRRH |
85 | /* Remove the frame of the tracer */ |
86 | this_size -= frame_size; | |
e5a81b62 | 87 | |
bb99d8cc | 88 | if (this_size <= stack_trace_max_size) |
e5a81b62 SR |
89 | return; |
90 | ||
81520a1b | 91 | /* we do not handle interrupt stacks yet */ |
87889501 | 92 | if (!object_is_on_stack(stack)) |
81520a1b SR |
93 | return; |
94 | ||
1904be1b SRRH |
95 | /* Can't do this from NMI context (can cause deadlocks) */ |
96 | if (in_nmi()) | |
97 | return; | |
98 | ||
a5e25883 | 99 | local_irq_save(flags); |
d332736d | 100 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 101 | |
4df29712 SRRH |
102 | /* In case another CPU set the tracer_frame on us */ |
103 | if (unlikely(!frame_size)) | |
104 | this_size -= tracer_frame; | |
105 | ||
e5a81b62 | 106 | /* a race could have already updated it */ |
bb99d8cc | 107 | if (this_size <= stack_trace_max_size) |
e5a81b62 SR |
108 | goto out; |
109 | ||
bb99d8cc | 110 | stack_trace_max_size = this_size; |
e5a81b62 | 111 | |
bb99d8cc AT |
112 | stack_trace_max.nr_entries = 0; |
113 | stack_trace_max.skip = 3; | |
e5a81b62 | 114 | |
bb99d8cc | 115 | save_stack_trace(&stack_trace_max); |
e5a81b62 | 116 | |
72ac426a | 117 | /* Skip over the overhead of the stack tracer itself */ |
bb99d8cc | 118 | for (i = 0; i < stack_trace_max.nr_entries; i++) { |
72ac426a SRRH |
119 | if (stack_dump_trace[i] == ip) |
120 | break; | |
121 | } | |
d4ecbfc4 | 122 | |
6ccd8371 SR |
123 | /* |
124 | * Some archs may not have the passed in ip in the dump. | |
125 | * If that happens, we need to show everything. | |
126 | */ | |
127 | if (i == stack_trace_max.nr_entries) | |
128 | i = 0; | |
129 | ||
1b6cced6 SR |
130 | /* |
131 | * Now find where in the stack these are. | |
132 | */ | |
72ac426a | 133 | x = 0; |
87889501 | 134 | start = stack; |
1b6cced6 SR |
135 | top = (unsigned long *) |
136 | (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE); | |
137 | ||
138 | /* | |
139 | * Loop through all the entries. One of the entries may | |
140 | * for some reason be missed on the stack, so we may | |
141 | * have to account for them. If they are all there, this | |
142 | * loop will only happen once. This code only takes place | |
143 | * on a new max, so it is far from a fast path. | |
144 | */ | |
bb99d8cc | 145 | while (i < stack_trace_max.nr_entries) { |
0a37119d | 146 | int found = 0; |
1b6cced6 | 147 | |
bb99d8cc | 148 | stack_trace_index[x] = this_size; |
1b6cced6 SR |
149 | p = start; |
150 | ||
bb99d8cc | 151 | for (; p < top && i < stack_trace_max.nr_entries; p++) { |
72ac426a SRRH |
152 | if (stack_dump_trace[i] == ULONG_MAX) |
153 | break; | |
6e22c836 YS |
154 | /* |
155 | * The READ_ONCE_NOCHECK is used to let KASAN know that | |
156 | * this is not a stack-out-of-bounds error. | |
157 | */ | |
158 | if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) { | |
72ac426a | 159 | stack_dump_trace[x] = stack_dump_trace[i++]; |
bb99d8cc | 160 | this_size = stack_trace_index[x++] = |
1b6cced6 | 161 | (top - p) * sizeof(unsigned long); |
0a37119d | 162 | found = 1; |
1b6cced6 SR |
163 | /* Start the search from here */ |
164 | start = p + 1; | |
4df29712 SRRH |
165 | /* |
166 | * We do not want to show the overhead | |
167 | * of the stack tracer stack in the | |
168 | * max stack. If we haven't figured | |
169 | * out what that is, then figure it out | |
170 | * now. | |
171 | */ | |
72ac426a | 172 | if (unlikely(!tracer_frame)) { |
4df29712 SRRH |
173 | tracer_frame = (p - stack) * |
174 | sizeof(unsigned long); | |
bb99d8cc | 175 | stack_trace_max_size -= tracer_frame; |
4df29712 | 176 | } |
1b6cced6 SR |
177 | } |
178 | } | |
179 | ||
0a37119d SR |
180 | if (!found) |
181 | i++; | |
1b6cced6 SR |
182 | } |
183 | ||
bb99d8cc | 184 | stack_trace_max.nr_entries = x; |
72ac426a SRRH |
185 | for (; x < i; x++) |
186 | stack_dump_trace[x] = ULONG_MAX; | |
187 | ||
a70857e4 | 188 | if (task_stack_end_corrupted(current)) { |
bb99d8cc | 189 | stack_trace_print(); |
e3172181 MK |
190 | BUG(); |
191 | } | |
192 | ||
e5a81b62 | 193 | out: |
d332736d | 194 | arch_spin_unlock(&stack_trace_max_lock); |
a5e25883 | 195 | local_irq_restore(flags); |
e5a81b62 SR |
196 | } |
197 | ||
198 | static void | |
a1e2e31d SR |
199 | stack_trace_call(unsigned long ip, unsigned long parent_ip, |
200 | struct ftrace_ops *op, struct pt_regs *pt_regs) | |
e5a81b62 | 201 | { |
87889501 | 202 | unsigned long stack; |
e5a81b62 | 203 | |
5168ae50 | 204 | preempt_disable_notrace(); |
e5a81b62 | 205 | |
e5a81b62 | 206 | /* no atomic needed, we only modify this variable by this cpu */ |
8aaf1ee7 SRV |
207 | __this_cpu_inc(disable_stack_tracer); |
208 | if (__this_cpu_read(disable_stack_tracer) != 1) | |
e5a81b62 SR |
209 | goto out; |
210 | ||
72ac426a | 211 | ip += MCOUNT_INSN_SIZE; |
4df29712 SRRH |
212 | |
213 | check_stack(ip, &stack); | |
e5a81b62 SR |
214 | |
215 | out: | |
8aaf1ee7 | 216 | __this_cpu_dec(disable_stack_tracer); |
e5a81b62 | 217 | /* prevent recursion in schedule */ |
5168ae50 | 218 | preempt_enable_notrace(); |
e5a81b62 SR |
219 | } |
220 | ||
221 | static struct ftrace_ops trace_ops __read_mostly = | |
222 | { | |
223 | .func = stack_trace_call, | |
4740974a | 224 | .flags = FTRACE_OPS_FL_RECURSION_SAFE, |
e5a81b62 SR |
225 | }; |
226 | ||
227 | static ssize_t | |
228 | stack_max_size_read(struct file *filp, char __user *ubuf, | |
229 | size_t count, loff_t *ppos) | |
230 | { | |
231 | unsigned long *ptr = filp->private_data; | |
232 | char buf[64]; | |
233 | int r; | |
234 | ||
235 | r = snprintf(buf, sizeof(buf), "%ld\n", *ptr); | |
236 | if (r > sizeof(buf)) | |
237 | r = sizeof(buf); | |
238 | return simple_read_from_buffer(ubuf, count, ppos, buf, r); | |
239 | } | |
240 | ||
241 | static ssize_t | |
242 | stack_max_size_write(struct file *filp, const char __user *ubuf, | |
243 | size_t count, loff_t *ppos) | |
244 | { | |
245 | long *ptr = filp->private_data; | |
246 | unsigned long val, flags; | |
e5a81b62 SR |
247 | int ret; |
248 | ||
22fe9b54 PH |
249 | ret = kstrtoul_from_user(ubuf, count, 10, &val); |
250 | if (ret) | |
e5a81b62 SR |
251 | return ret; |
252 | ||
a5e25883 | 253 | local_irq_save(flags); |
4f48f8b7 LJ |
254 | |
255 | /* | |
256 | * In case we trace inside arch_spin_lock() or after (NMI), | |
257 | * we will cause circular lock, so we also need to increase | |
8aaf1ee7 | 258 | * the percpu disable_stack_tracer here. |
4f48f8b7 | 259 | */ |
8aaf1ee7 | 260 | __this_cpu_inc(disable_stack_tracer); |
4f48f8b7 | 261 | |
d332736d | 262 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 263 | *ptr = val; |
d332736d | 264 | arch_spin_unlock(&stack_trace_max_lock); |
4f48f8b7 | 265 | |
8aaf1ee7 | 266 | __this_cpu_dec(disable_stack_tracer); |
a5e25883 | 267 | local_irq_restore(flags); |
e5a81b62 SR |
268 | |
269 | return count; | |
270 | } | |
271 | ||
f38f1d2a | 272 | static const struct file_operations stack_max_size_fops = { |
e5a81b62 SR |
273 | .open = tracing_open_generic, |
274 | .read = stack_max_size_read, | |
275 | .write = stack_max_size_write, | |
6038f373 | 276 | .llseek = default_llseek, |
e5a81b62 SR |
277 | }; |
278 | ||
279 | static void * | |
2fc5f0cf | 280 | __next(struct seq_file *m, loff_t *pos) |
e5a81b62 | 281 | { |
2fc5f0cf | 282 | long n = *pos - 1; |
e5a81b62 | 283 | |
bb99d8cc | 284 | if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX) |
e5a81b62 SR |
285 | return NULL; |
286 | ||
2fc5f0cf | 287 | m->private = (void *)n; |
1b6cced6 | 288 | return &m->private; |
e5a81b62 SR |
289 | } |
290 | ||
2fc5f0cf LZ |
291 | static void * |
292 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
e5a81b62 | 293 | { |
2fc5f0cf LZ |
294 | (*pos)++; |
295 | return __next(m, pos); | |
296 | } | |
e5a81b62 | 297 | |
2fc5f0cf LZ |
298 | static void *t_start(struct seq_file *m, loff_t *pos) |
299 | { | |
e5a81b62 | 300 | local_irq_disable(); |
4f48f8b7 | 301 | |
8aaf1ee7 | 302 | __this_cpu_inc(disable_stack_tracer); |
4f48f8b7 | 303 | |
d332736d | 304 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 305 | |
522a110b LW |
306 | if (*pos == 0) |
307 | return SEQ_START_TOKEN; | |
308 | ||
2fc5f0cf | 309 | return __next(m, pos); |
e5a81b62 SR |
310 | } |
311 | ||
312 | static void t_stop(struct seq_file *m, void *p) | |
313 | { | |
d332736d | 314 | arch_spin_unlock(&stack_trace_max_lock); |
4f48f8b7 | 315 | |
8aaf1ee7 | 316 | __this_cpu_dec(disable_stack_tracer); |
4f48f8b7 | 317 | |
e5a81b62 SR |
318 | local_irq_enable(); |
319 | } | |
320 | ||
962e3707 | 321 | static void trace_lookup_stack(struct seq_file *m, long i) |
e5a81b62 | 322 | { |
1b6cced6 | 323 | unsigned long addr = stack_dump_trace[i]; |
e5a81b62 | 324 | |
962e3707 | 325 | seq_printf(m, "%pS\n", (void *)addr); |
e5a81b62 SR |
326 | } |
327 | ||
e447e1df SR |
328 | static void print_disabled(struct seq_file *m) |
329 | { | |
330 | seq_puts(m, "#\n" | |
331 | "# Stack tracer disabled\n" | |
332 | "#\n" | |
333 | "# To enable the stack tracer, either add 'stacktrace' to the\n" | |
334 | "# kernel command line\n" | |
335 | "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n" | |
336 | "#\n"); | |
337 | } | |
338 | ||
e5a81b62 SR |
339 | static int t_show(struct seq_file *m, void *v) |
340 | { | |
522a110b | 341 | long i; |
1b6cced6 SR |
342 | int size; |
343 | ||
522a110b | 344 | if (v == SEQ_START_TOKEN) { |
eb1871f3 | 345 | seq_printf(m, " Depth Size Location" |
1b6cced6 | 346 | " (%d entries)\n" |
eb1871f3 | 347 | " ----- ---- --------\n", |
bb99d8cc | 348 | stack_trace_max.nr_entries); |
e447e1df | 349 | |
bb99d8cc | 350 | if (!stack_tracer_enabled && !stack_trace_max_size) |
e447e1df SR |
351 | print_disabled(m); |
352 | ||
1b6cced6 SR |
353 | return 0; |
354 | } | |
e5a81b62 | 355 | |
522a110b LW |
356 | i = *(long *)v; |
357 | ||
bb99d8cc | 358 | if (i >= stack_trace_max.nr_entries || |
1b6cced6 | 359 | stack_dump_trace[i] == ULONG_MAX) |
e5a81b62 SR |
360 | return 0; |
361 | ||
bb99d8cc | 362 | if (i+1 == stack_trace_max.nr_entries || |
1b6cced6 | 363 | stack_dump_trace[i+1] == ULONG_MAX) |
bb99d8cc | 364 | size = stack_trace_index[i]; |
1b6cced6 | 365 | else |
bb99d8cc | 366 | size = stack_trace_index[i] - stack_trace_index[i+1]; |
1b6cced6 | 367 | |
bb99d8cc | 368 | seq_printf(m, "%3ld) %8d %5d ", i, stack_trace_index[i], size); |
1b6cced6 SR |
369 | |
370 | trace_lookup_stack(m, i); | |
e5a81b62 SR |
371 | |
372 | return 0; | |
373 | } | |
374 | ||
f38f1d2a | 375 | static const struct seq_operations stack_trace_seq_ops = { |
e5a81b62 SR |
376 | .start = t_start, |
377 | .next = t_next, | |
378 | .stop = t_stop, | |
379 | .show = t_show, | |
380 | }; | |
381 | ||
382 | static int stack_trace_open(struct inode *inode, struct file *file) | |
383 | { | |
d8cc1ab7 | 384 | return seq_open(file, &stack_trace_seq_ops); |
e5a81b62 SR |
385 | } |
386 | ||
f38f1d2a | 387 | static const struct file_operations stack_trace_fops = { |
e5a81b62 SR |
388 | .open = stack_trace_open, |
389 | .read = seq_read, | |
390 | .llseek = seq_lseek, | |
d8cc1ab7 | 391 | .release = seq_release, |
e5a81b62 SR |
392 | }; |
393 | ||
bbd1d27d SRV |
394 | #ifdef CONFIG_DYNAMIC_FTRACE |
395 | ||
d2d45c7a SR |
396 | static int |
397 | stack_trace_filter_open(struct inode *inode, struct file *file) | |
398 | { | |
0f179765 SRV |
399 | struct ftrace_ops *ops = inode->i_private; |
400 | ||
401 | return ftrace_regex_open(ops, FTRACE_ITER_FILTER, | |
d2d45c7a SR |
402 | inode, file); |
403 | } | |
404 | ||
405 | static const struct file_operations stack_trace_filter_fops = { | |
406 | .open = stack_trace_filter_open, | |
407 | .read = seq_read, | |
408 | .write = ftrace_filter_write, | |
098c879e | 409 | .llseek = tracing_lseek, |
d2d45c7a SR |
410 | .release = ftrace_regex_release, |
411 | }; | |
412 | ||
bbd1d27d SRV |
413 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
414 | ||
f38f1d2a SR |
415 | int |
416 | stack_trace_sysctl(struct ctl_table *table, int write, | |
8d65af78 | 417 | void __user *buffer, size_t *lenp, |
f38f1d2a SR |
418 | loff_t *ppos) |
419 | { | |
420 | int ret; | |
421 | ||
422 | mutex_lock(&stack_sysctl_mutex); | |
423 | ||
8d65af78 | 424 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
f38f1d2a SR |
425 | |
426 | if (ret || !write || | |
a32c7765 | 427 | (last_stack_tracer_enabled == !!stack_tracer_enabled)) |
f38f1d2a SR |
428 | goto out; |
429 | ||
a32c7765 | 430 | last_stack_tracer_enabled = !!stack_tracer_enabled; |
f38f1d2a SR |
431 | |
432 | if (stack_tracer_enabled) | |
433 | register_ftrace_function(&trace_ops); | |
434 | else | |
435 | unregister_ftrace_function(&trace_ops); | |
436 | ||
437 | out: | |
438 | mutex_unlock(&stack_sysctl_mutex); | |
439 | return ret; | |
440 | } | |
441 | ||
762e1207 SR |
442 | static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata; |
443 | ||
f38f1d2a SR |
444 | static __init int enable_stacktrace(char *str) |
445 | { | |
762e1207 SR |
446 | if (strncmp(str, "_filter=", 8) == 0) |
447 | strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE); | |
448 | ||
e05a43b7 SR |
449 | stack_tracer_enabled = 1; |
450 | last_stack_tracer_enabled = 1; | |
f38f1d2a SR |
451 | return 1; |
452 | } | |
453 | __setup("stacktrace", enable_stacktrace); | |
454 | ||
e5a81b62 SR |
455 | static __init int stack_trace_init(void) |
456 | { | |
457 | struct dentry *d_tracer; | |
e5a81b62 SR |
458 | |
459 | d_tracer = tracing_init_dentry(); | |
14a5ae40 | 460 | if (IS_ERR(d_tracer)) |
ed6f1c99 | 461 | return 0; |
e5a81b62 | 462 | |
5452af66 | 463 | trace_create_file("stack_max_size", 0644, d_tracer, |
bb99d8cc | 464 | &stack_trace_max_size, &stack_max_size_fops); |
e5a81b62 | 465 | |
5452af66 FW |
466 | trace_create_file("stack_trace", 0444, d_tracer, |
467 | NULL, &stack_trace_fops); | |
e5a81b62 | 468 | |
bbd1d27d | 469 | #ifdef CONFIG_DYNAMIC_FTRACE |
d2d45c7a | 470 | trace_create_file("stack_trace_filter", 0444, d_tracer, |
0f179765 | 471 | &trace_ops, &stack_trace_filter_fops); |
bbd1d27d | 472 | #endif |
d2d45c7a | 473 | |
762e1207 SR |
474 | if (stack_trace_filter_buf[0]) |
475 | ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1); | |
476 | ||
e05a43b7 | 477 | if (stack_tracer_enabled) |
f38f1d2a | 478 | register_ftrace_function(&trace_ops); |
e5a81b62 SR |
479 | |
480 | return 0; | |
481 | } | |
482 | ||
483 | device_initcall(stack_trace_init); |