]>
Commit | Line | Data |
---|---|---|
e5a81b62 SR |
1 | /* |
2 | * Copyright (C) 2008 Steven Rostedt <[email protected]> | |
3 | * | |
4 | */ | |
5 | #include <linux/stacktrace.h> | |
6 | #include <linux/kallsyms.h> | |
7 | #include <linux/seq_file.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/uaccess.h> | |
e5a81b62 SR |
10 | #include <linux/ftrace.h> |
11 | #include <linux/module.h> | |
f38f1d2a | 12 | #include <linux/sysctl.h> |
e5a81b62 | 13 | #include <linux/init.h> |
762e1207 SR |
14 | |
15 | #include <asm/setup.h> | |
16 | ||
e5a81b62 SR |
17 | #include "trace.h" |
18 | ||
1b6cced6 SR |
19 | static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = |
20 | { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; | |
bb99d8cc | 21 | unsigned stack_trace_index[STACK_TRACE_ENTRIES]; |
1b6cced6 | 22 | |
4df29712 SRRH |
23 | /* |
24 | * Reserve one entry for the passed in ip. This will allow | |
25 | * us to remove most or all of the stack size overhead | |
26 | * added by the stack tracer itself. | |
27 | */ | |
bb99d8cc | 28 | struct stack_trace stack_trace_max = { |
4df29712 | 29 | .max_entries = STACK_TRACE_ENTRIES - 1, |
72ac426a | 30 | .entries = &stack_dump_trace[0], |
e5a81b62 SR |
31 | }; |
32 | ||
bb99d8cc | 33 | unsigned long stack_trace_max_size; |
d332736d | 34 | arch_spinlock_t stack_trace_max_lock = |
edc35bd7 | 35 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
e5a81b62 | 36 | |
e5a81b62 | 37 | static DEFINE_PER_CPU(int, trace_active); |
f38f1d2a SR |
38 | static DEFINE_MUTEX(stack_sysctl_mutex); |
39 | ||
40 | int stack_tracer_enabled; | |
41 | static int last_stack_tracer_enabled; | |
e5a81b62 | 42 | |
bb99d8cc | 43 | void stack_trace_print(void) |
e3172181 MK |
44 | { |
45 | long i; | |
46 | int size; | |
47 | ||
48 | pr_emerg(" Depth Size Location (%d entries)\n" | |
49 | " ----- ---- --------\n", | |
bb99d8cc | 50 | stack_trace_max.nr_entries); |
e3172181 | 51 | |
bb99d8cc | 52 | for (i = 0; i < stack_trace_max.nr_entries; i++) { |
e3172181 MK |
53 | if (stack_dump_trace[i] == ULONG_MAX) |
54 | break; | |
bb99d8cc | 55 | if (i+1 == stack_trace_max.nr_entries || |
e3172181 | 56 | stack_dump_trace[i+1] == ULONG_MAX) |
bb99d8cc | 57 | size = stack_trace_index[i]; |
e3172181 | 58 | else |
bb99d8cc | 59 | size = stack_trace_index[i] - stack_trace_index[i+1]; |
e3172181 | 60 | |
bb99d8cc | 61 | pr_emerg("%3ld) %8d %5d %pS\n", i, stack_trace_index[i], |
e3172181 MK |
62 | size, (void *)stack_dump_trace[i]); |
63 | } | |
64 | } | |
65 | ||
bb99d8cc AT |
66 | /* |
67 | * When arch-specific code overides this function, the following | |
d332736d | 68 | * data should be filled up, assuming stack_trace_max_lock is held to |
bb99d8cc AT |
69 | * prevent concurrent updates. |
70 | * stack_trace_index[] | |
71 | * stack_trace_max | |
72 | * stack_trace_max_size | |
73 | */ | |
74 | void __weak | |
d4ecbfc4 | 75 | check_stack(unsigned long ip, unsigned long *stack) |
e5a81b62 | 76 | { |
e3172181 | 77 | unsigned long this_size, flags; unsigned long *p, *top, *start; |
4df29712 SRRH |
78 | static int tracer_frame; |
79 | int frame_size = ACCESS_ONCE(tracer_frame); | |
72ac426a | 80 | int i, x; |
e5a81b62 | 81 | |
87889501 | 82 | this_size = ((unsigned long)stack) & (THREAD_SIZE-1); |
e5a81b62 | 83 | this_size = THREAD_SIZE - this_size; |
4df29712 SRRH |
84 | /* Remove the frame of the tracer */ |
85 | this_size -= frame_size; | |
e5a81b62 | 86 | |
bb99d8cc | 87 | if (this_size <= stack_trace_max_size) |
e5a81b62 SR |
88 | return; |
89 | ||
81520a1b | 90 | /* we do not handle interrupt stacks yet */ |
87889501 | 91 | if (!object_is_on_stack(stack)) |
81520a1b SR |
92 | return; |
93 | ||
1904be1b SRRH |
94 | /* Can't do this from NMI context (can cause deadlocks) */ |
95 | if (in_nmi()) | |
96 | return; | |
97 | ||
a5e25883 | 98 | local_irq_save(flags); |
d332736d | 99 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 100 | |
a2d76290 SRRH |
101 | /* |
102 | * RCU may not be watching, make it see us. | |
103 | * The stack trace code uses rcu_sched. | |
104 | */ | |
105 | rcu_irq_enter(); | |
106 | ||
4df29712 SRRH |
107 | /* In case another CPU set the tracer_frame on us */ |
108 | if (unlikely(!frame_size)) | |
109 | this_size -= tracer_frame; | |
110 | ||
e5a81b62 | 111 | /* a race could have already updated it */ |
bb99d8cc | 112 | if (this_size <= stack_trace_max_size) |
e5a81b62 SR |
113 | goto out; |
114 | ||
bb99d8cc | 115 | stack_trace_max_size = this_size; |
e5a81b62 | 116 | |
bb99d8cc AT |
117 | stack_trace_max.nr_entries = 0; |
118 | stack_trace_max.skip = 3; | |
e5a81b62 | 119 | |
bb99d8cc | 120 | save_stack_trace(&stack_trace_max); |
e5a81b62 | 121 | |
72ac426a | 122 | /* Skip over the overhead of the stack tracer itself */ |
bb99d8cc | 123 | for (i = 0; i < stack_trace_max.nr_entries; i++) { |
72ac426a SRRH |
124 | if (stack_dump_trace[i] == ip) |
125 | break; | |
126 | } | |
d4ecbfc4 | 127 | |
6ccd8371 SR |
128 | /* |
129 | * Some archs may not have the passed in ip in the dump. | |
130 | * If that happens, we need to show everything. | |
131 | */ | |
132 | if (i == stack_trace_max.nr_entries) | |
133 | i = 0; | |
134 | ||
1b6cced6 SR |
135 | /* |
136 | * Now find where in the stack these are. | |
137 | */ | |
72ac426a | 138 | x = 0; |
87889501 | 139 | start = stack; |
1b6cced6 SR |
140 | top = (unsigned long *) |
141 | (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE); | |
142 | ||
143 | /* | |
144 | * Loop through all the entries. One of the entries may | |
145 | * for some reason be missed on the stack, so we may | |
146 | * have to account for them. If they are all there, this | |
147 | * loop will only happen once. This code only takes place | |
148 | * on a new max, so it is far from a fast path. | |
149 | */ | |
bb99d8cc | 150 | while (i < stack_trace_max.nr_entries) { |
0a37119d | 151 | int found = 0; |
1b6cced6 | 152 | |
bb99d8cc | 153 | stack_trace_index[x] = this_size; |
1b6cced6 SR |
154 | p = start; |
155 | ||
bb99d8cc | 156 | for (; p < top && i < stack_trace_max.nr_entries; p++) { |
72ac426a SRRH |
157 | if (stack_dump_trace[i] == ULONG_MAX) |
158 | break; | |
6e22c836 YS |
159 | /* |
160 | * The READ_ONCE_NOCHECK is used to let KASAN know that | |
161 | * this is not a stack-out-of-bounds error. | |
162 | */ | |
163 | if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) { | |
72ac426a | 164 | stack_dump_trace[x] = stack_dump_trace[i++]; |
bb99d8cc | 165 | this_size = stack_trace_index[x++] = |
1b6cced6 | 166 | (top - p) * sizeof(unsigned long); |
0a37119d | 167 | found = 1; |
1b6cced6 SR |
168 | /* Start the search from here */ |
169 | start = p + 1; | |
4df29712 SRRH |
170 | /* |
171 | * We do not want to show the overhead | |
172 | * of the stack tracer stack in the | |
173 | * max stack. If we haven't figured | |
174 | * out what that is, then figure it out | |
175 | * now. | |
176 | */ | |
72ac426a | 177 | if (unlikely(!tracer_frame)) { |
4df29712 SRRH |
178 | tracer_frame = (p - stack) * |
179 | sizeof(unsigned long); | |
bb99d8cc | 180 | stack_trace_max_size -= tracer_frame; |
4df29712 | 181 | } |
1b6cced6 SR |
182 | } |
183 | } | |
184 | ||
0a37119d SR |
185 | if (!found) |
186 | i++; | |
1b6cced6 SR |
187 | } |
188 | ||
bb99d8cc | 189 | stack_trace_max.nr_entries = x; |
72ac426a SRRH |
190 | for (; x < i; x++) |
191 | stack_dump_trace[x] = ULONG_MAX; | |
192 | ||
a70857e4 | 193 | if (task_stack_end_corrupted(current)) { |
bb99d8cc | 194 | stack_trace_print(); |
e3172181 MK |
195 | BUG(); |
196 | } | |
197 | ||
e5a81b62 | 198 | out: |
a2d76290 | 199 | rcu_irq_exit(); |
d332736d | 200 | arch_spin_unlock(&stack_trace_max_lock); |
a5e25883 | 201 | local_irq_restore(flags); |
e5a81b62 SR |
202 | } |
203 | ||
204 | static void | |
a1e2e31d SR |
205 | stack_trace_call(unsigned long ip, unsigned long parent_ip, |
206 | struct ftrace_ops *op, struct pt_regs *pt_regs) | |
e5a81b62 | 207 | { |
87889501 | 208 | unsigned long stack; |
5168ae50 | 209 | int cpu; |
e5a81b62 | 210 | |
5168ae50 | 211 | preempt_disable_notrace(); |
e5a81b62 SR |
212 | |
213 | cpu = raw_smp_processor_id(); | |
214 | /* no atomic needed, we only modify this variable by this cpu */ | |
215 | if (per_cpu(trace_active, cpu)++ != 0) | |
216 | goto out; | |
217 | ||
72ac426a | 218 | ip += MCOUNT_INSN_SIZE; |
4df29712 SRRH |
219 | |
220 | check_stack(ip, &stack); | |
e5a81b62 SR |
221 | |
222 | out: | |
223 | per_cpu(trace_active, cpu)--; | |
224 | /* prevent recursion in schedule */ | |
5168ae50 | 225 | preempt_enable_notrace(); |
e5a81b62 SR |
226 | } |
227 | ||
228 | static struct ftrace_ops trace_ops __read_mostly = | |
229 | { | |
230 | .func = stack_trace_call, | |
4740974a | 231 | .flags = FTRACE_OPS_FL_RECURSION_SAFE, |
e5a81b62 SR |
232 | }; |
233 | ||
234 | static ssize_t | |
235 | stack_max_size_read(struct file *filp, char __user *ubuf, | |
236 | size_t count, loff_t *ppos) | |
237 | { | |
238 | unsigned long *ptr = filp->private_data; | |
239 | char buf[64]; | |
240 | int r; | |
241 | ||
242 | r = snprintf(buf, sizeof(buf), "%ld\n", *ptr); | |
243 | if (r > sizeof(buf)) | |
244 | r = sizeof(buf); | |
245 | return simple_read_from_buffer(ubuf, count, ppos, buf, r); | |
246 | } | |
247 | ||
248 | static ssize_t | |
249 | stack_max_size_write(struct file *filp, const char __user *ubuf, | |
250 | size_t count, loff_t *ppos) | |
251 | { | |
252 | long *ptr = filp->private_data; | |
253 | unsigned long val, flags; | |
e5a81b62 | 254 | int ret; |
4f48f8b7 | 255 | int cpu; |
e5a81b62 | 256 | |
22fe9b54 PH |
257 | ret = kstrtoul_from_user(ubuf, count, 10, &val); |
258 | if (ret) | |
e5a81b62 SR |
259 | return ret; |
260 | ||
a5e25883 | 261 | local_irq_save(flags); |
4f48f8b7 LJ |
262 | |
263 | /* | |
264 | * In case we trace inside arch_spin_lock() or after (NMI), | |
265 | * we will cause circular lock, so we also need to increase | |
266 | * the percpu trace_active here. | |
267 | */ | |
268 | cpu = smp_processor_id(); | |
269 | per_cpu(trace_active, cpu)++; | |
270 | ||
d332736d | 271 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 272 | *ptr = val; |
d332736d | 273 | arch_spin_unlock(&stack_trace_max_lock); |
4f48f8b7 LJ |
274 | |
275 | per_cpu(trace_active, cpu)--; | |
a5e25883 | 276 | local_irq_restore(flags); |
e5a81b62 SR |
277 | |
278 | return count; | |
279 | } | |
280 | ||
f38f1d2a | 281 | static const struct file_operations stack_max_size_fops = { |
e5a81b62 SR |
282 | .open = tracing_open_generic, |
283 | .read = stack_max_size_read, | |
284 | .write = stack_max_size_write, | |
6038f373 | 285 | .llseek = default_llseek, |
e5a81b62 SR |
286 | }; |
287 | ||
288 | static void * | |
2fc5f0cf | 289 | __next(struct seq_file *m, loff_t *pos) |
e5a81b62 | 290 | { |
2fc5f0cf | 291 | long n = *pos - 1; |
e5a81b62 | 292 | |
bb99d8cc | 293 | if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX) |
e5a81b62 SR |
294 | return NULL; |
295 | ||
2fc5f0cf | 296 | m->private = (void *)n; |
1b6cced6 | 297 | return &m->private; |
e5a81b62 SR |
298 | } |
299 | ||
2fc5f0cf LZ |
300 | static void * |
301 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
e5a81b62 | 302 | { |
2fc5f0cf LZ |
303 | (*pos)++; |
304 | return __next(m, pos); | |
305 | } | |
e5a81b62 | 306 | |
2fc5f0cf LZ |
307 | static void *t_start(struct seq_file *m, loff_t *pos) |
308 | { | |
4f48f8b7 LJ |
309 | int cpu; |
310 | ||
e5a81b62 | 311 | local_irq_disable(); |
4f48f8b7 LJ |
312 | |
313 | cpu = smp_processor_id(); | |
314 | per_cpu(trace_active, cpu)++; | |
315 | ||
d332736d | 316 | arch_spin_lock(&stack_trace_max_lock); |
e5a81b62 | 317 | |
522a110b LW |
318 | if (*pos == 0) |
319 | return SEQ_START_TOKEN; | |
320 | ||
2fc5f0cf | 321 | return __next(m, pos); |
e5a81b62 SR |
322 | } |
323 | ||
324 | static void t_stop(struct seq_file *m, void *p) | |
325 | { | |
4f48f8b7 LJ |
326 | int cpu; |
327 | ||
d332736d | 328 | arch_spin_unlock(&stack_trace_max_lock); |
4f48f8b7 LJ |
329 | |
330 | cpu = smp_processor_id(); | |
331 | per_cpu(trace_active, cpu)--; | |
332 | ||
e5a81b62 SR |
333 | local_irq_enable(); |
334 | } | |
335 | ||
962e3707 | 336 | static void trace_lookup_stack(struct seq_file *m, long i) |
e5a81b62 | 337 | { |
1b6cced6 | 338 | unsigned long addr = stack_dump_trace[i]; |
e5a81b62 | 339 | |
962e3707 | 340 | seq_printf(m, "%pS\n", (void *)addr); |
e5a81b62 SR |
341 | } |
342 | ||
e447e1df SR |
343 | static void print_disabled(struct seq_file *m) |
344 | { | |
345 | seq_puts(m, "#\n" | |
346 | "# Stack tracer disabled\n" | |
347 | "#\n" | |
348 | "# To enable the stack tracer, either add 'stacktrace' to the\n" | |
349 | "# kernel command line\n" | |
350 | "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n" | |
351 | "#\n"); | |
352 | } | |
353 | ||
e5a81b62 SR |
354 | static int t_show(struct seq_file *m, void *v) |
355 | { | |
522a110b | 356 | long i; |
1b6cced6 SR |
357 | int size; |
358 | ||
522a110b | 359 | if (v == SEQ_START_TOKEN) { |
eb1871f3 | 360 | seq_printf(m, " Depth Size Location" |
1b6cced6 | 361 | " (%d entries)\n" |
eb1871f3 | 362 | " ----- ---- --------\n", |
bb99d8cc | 363 | stack_trace_max.nr_entries); |
e447e1df | 364 | |
bb99d8cc | 365 | if (!stack_tracer_enabled && !stack_trace_max_size) |
e447e1df SR |
366 | print_disabled(m); |
367 | ||
1b6cced6 SR |
368 | return 0; |
369 | } | |
e5a81b62 | 370 | |
522a110b LW |
371 | i = *(long *)v; |
372 | ||
bb99d8cc | 373 | if (i >= stack_trace_max.nr_entries || |
1b6cced6 | 374 | stack_dump_trace[i] == ULONG_MAX) |
e5a81b62 SR |
375 | return 0; |
376 | ||
bb99d8cc | 377 | if (i+1 == stack_trace_max.nr_entries || |
1b6cced6 | 378 | stack_dump_trace[i+1] == ULONG_MAX) |
bb99d8cc | 379 | size = stack_trace_index[i]; |
1b6cced6 | 380 | else |
bb99d8cc | 381 | size = stack_trace_index[i] - stack_trace_index[i+1]; |
1b6cced6 | 382 | |
bb99d8cc | 383 | seq_printf(m, "%3ld) %8d %5d ", i, stack_trace_index[i], size); |
1b6cced6 SR |
384 | |
385 | trace_lookup_stack(m, i); | |
e5a81b62 SR |
386 | |
387 | return 0; | |
388 | } | |
389 | ||
f38f1d2a | 390 | static const struct seq_operations stack_trace_seq_ops = { |
e5a81b62 SR |
391 | .start = t_start, |
392 | .next = t_next, | |
393 | .stop = t_stop, | |
394 | .show = t_show, | |
395 | }; | |
396 | ||
397 | static int stack_trace_open(struct inode *inode, struct file *file) | |
398 | { | |
d8cc1ab7 | 399 | return seq_open(file, &stack_trace_seq_ops); |
e5a81b62 SR |
400 | } |
401 | ||
f38f1d2a | 402 | static const struct file_operations stack_trace_fops = { |
e5a81b62 SR |
403 | .open = stack_trace_open, |
404 | .read = seq_read, | |
405 | .llseek = seq_lseek, | |
d8cc1ab7 | 406 | .release = seq_release, |
e5a81b62 SR |
407 | }; |
408 | ||
d2d45c7a SR |
409 | static int |
410 | stack_trace_filter_open(struct inode *inode, struct file *file) | |
411 | { | |
412 | return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER, | |
413 | inode, file); | |
414 | } | |
415 | ||
416 | static const struct file_operations stack_trace_filter_fops = { | |
417 | .open = stack_trace_filter_open, | |
418 | .read = seq_read, | |
419 | .write = ftrace_filter_write, | |
098c879e | 420 | .llseek = tracing_lseek, |
d2d45c7a SR |
421 | .release = ftrace_regex_release, |
422 | }; | |
423 | ||
f38f1d2a SR |
424 | int |
425 | stack_trace_sysctl(struct ctl_table *table, int write, | |
8d65af78 | 426 | void __user *buffer, size_t *lenp, |
f38f1d2a SR |
427 | loff_t *ppos) |
428 | { | |
429 | int ret; | |
430 | ||
431 | mutex_lock(&stack_sysctl_mutex); | |
432 | ||
8d65af78 | 433 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
f38f1d2a SR |
434 | |
435 | if (ret || !write || | |
a32c7765 | 436 | (last_stack_tracer_enabled == !!stack_tracer_enabled)) |
f38f1d2a SR |
437 | goto out; |
438 | ||
a32c7765 | 439 | last_stack_tracer_enabled = !!stack_tracer_enabled; |
f38f1d2a SR |
440 | |
441 | if (stack_tracer_enabled) | |
442 | register_ftrace_function(&trace_ops); | |
443 | else | |
444 | unregister_ftrace_function(&trace_ops); | |
445 | ||
446 | out: | |
447 | mutex_unlock(&stack_sysctl_mutex); | |
448 | return ret; | |
449 | } | |
450 | ||
762e1207 SR |
451 | static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata; |
452 | ||
f38f1d2a SR |
453 | static __init int enable_stacktrace(char *str) |
454 | { | |
762e1207 SR |
455 | if (strncmp(str, "_filter=", 8) == 0) |
456 | strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE); | |
457 | ||
e05a43b7 SR |
458 | stack_tracer_enabled = 1; |
459 | last_stack_tracer_enabled = 1; | |
f38f1d2a SR |
460 | return 1; |
461 | } | |
462 | __setup("stacktrace", enable_stacktrace); | |
463 | ||
e5a81b62 SR |
464 | static __init int stack_trace_init(void) |
465 | { | |
466 | struct dentry *d_tracer; | |
e5a81b62 SR |
467 | |
468 | d_tracer = tracing_init_dentry(); | |
14a5ae40 | 469 | if (IS_ERR(d_tracer)) |
ed6f1c99 | 470 | return 0; |
e5a81b62 | 471 | |
5452af66 | 472 | trace_create_file("stack_max_size", 0644, d_tracer, |
bb99d8cc | 473 | &stack_trace_max_size, &stack_max_size_fops); |
e5a81b62 | 474 | |
5452af66 FW |
475 | trace_create_file("stack_trace", 0444, d_tracer, |
476 | NULL, &stack_trace_fops); | |
e5a81b62 | 477 | |
d2d45c7a SR |
478 | trace_create_file("stack_trace_filter", 0444, d_tracer, |
479 | NULL, &stack_trace_filter_fops); | |
480 | ||
762e1207 SR |
481 | if (stack_trace_filter_buf[0]) |
482 | ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1); | |
483 | ||
e05a43b7 | 484 | if (stack_tracer_enabled) |
f38f1d2a | 485 | register_ftrace_function(&trace_ops); |
e5a81b62 SR |
486 | |
487 | return 0; | |
488 | } | |
489 | ||
490 | device_initcall(stack_trace_init); |