Commit | Line | Data |
---|---|---|
e5a81b62 SR |
1 | /* |
2 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
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 | ||
19 | #define STACK_TRACE_ENTRIES 500 | |
20 | ||
1b6cced6 SR |
21 | static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = |
22 | { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; | |
23 | static unsigned stack_dump_index[STACK_TRACE_ENTRIES]; | |
24 | ||
4df29712 SRRH |
25 | /* |
26 | * Reserve one entry for the passed in ip. This will allow | |
27 | * us to remove most or all of the stack size overhead | |
28 | * added by the stack tracer itself. | |
29 | */ | |
e5a81b62 | 30 | static struct stack_trace max_stack_trace = { |
4df29712 | 31 | .max_entries = STACK_TRACE_ENTRIES - 1, |
72ac426a | 32 | .entries = &stack_dump_trace[0], |
e5a81b62 SR |
33 | }; |
34 | ||
35 | static unsigned long max_stack_size; | |
445c8951 | 36 | static arch_spinlock_t max_stack_lock = |
edc35bd7 | 37 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
e5a81b62 | 38 | |
e5a81b62 | 39 | static DEFINE_PER_CPU(int, trace_active); |
f38f1d2a SR |
40 | static DEFINE_MUTEX(stack_sysctl_mutex); |
41 | ||
42 | int stack_tracer_enabled; | |
43 | static int last_stack_tracer_enabled; | |
e5a81b62 | 44 | |
e3172181 MK |
45 | static inline void print_max_stack(void) |
46 | { | |
47 | long i; | |
48 | int size; | |
49 | ||
50 | pr_emerg(" Depth Size Location (%d entries)\n" | |
51 | " ----- ---- --------\n", | |
72ac426a | 52 | max_stack_trace.nr_entries); |
e3172181 MK |
53 | |
54 | for (i = 0; i < max_stack_trace.nr_entries; i++) { | |
55 | if (stack_dump_trace[i] == ULONG_MAX) | |
56 | break; | |
57 | if (i+1 == max_stack_trace.nr_entries || | |
58 | stack_dump_trace[i+1] == ULONG_MAX) | |
59 | size = stack_dump_index[i]; | |
60 | else | |
61 | size = stack_dump_index[i] - stack_dump_index[i+1]; | |
62 | ||
63 | pr_emerg("%3ld) %8d %5d %pS\n", i, stack_dump_index[i], | |
64 | size, (void *)stack_dump_trace[i]); | |
65 | } | |
66 | } | |
67 | ||
87889501 | 68 | static inline void |
d4ecbfc4 | 69 | check_stack(unsigned long ip, unsigned long *stack) |
e5a81b62 | 70 | { |
e3172181 | 71 | unsigned long this_size, flags; unsigned long *p, *top, *start; |
4df29712 SRRH |
72 | static int tracer_frame; |
73 | int frame_size = ACCESS_ONCE(tracer_frame); | |
72ac426a | 74 | int i, x; |
e5a81b62 | 75 | |
87889501 | 76 | this_size = ((unsigned long)stack) & (THREAD_SIZE-1); |
e5a81b62 | 77 | this_size = THREAD_SIZE - this_size; |
4df29712 SRRH |
78 | /* Remove the frame of the tracer */ |
79 | this_size -= frame_size; | |
e5a81b62 SR |
80 | |
81 | if (this_size <= max_stack_size) | |
82 | return; | |
83 | ||
81520a1b | 84 | /* we do not handle interrupt stacks yet */ |
87889501 | 85 | if (!object_is_on_stack(stack)) |
81520a1b SR |
86 | return; |
87 | ||
a5e25883 | 88 | local_irq_save(flags); |
0199c4e6 | 89 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 90 | |
4df29712 SRRH |
91 | /* In case another CPU set the tracer_frame on us */ |
92 | if (unlikely(!frame_size)) | |
93 | this_size -= tracer_frame; | |
94 | ||
e5a81b62 SR |
95 | /* a race could have already updated it */ |
96 | if (this_size <= max_stack_size) | |
97 | goto out; | |
98 | ||
99 | max_stack_size = this_size; | |
100 | ||
7eea4fce | 101 | max_stack_trace.nr_entries = 0; |
72ac426a | 102 | max_stack_trace.skip = 3; |
e5a81b62 SR |
103 | |
104 | save_stack_trace(&max_stack_trace); | |
105 | ||
72ac426a SRRH |
106 | /* Skip over the overhead of the stack tracer itself */ |
107 | for (i = 0; i < max_stack_trace.nr_entries; i++) { | |
108 | if (stack_dump_trace[i] == ip) | |
109 | break; | |
110 | } | |
d4ecbfc4 | 111 | |
1b6cced6 SR |
112 | /* |
113 | * Now find where in the stack these are. | |
114 | */ | |
72ac426a | 115 | x = 0; |
87889501 | 116 | start = stack; |
1b6cced6 SR |
117 | top = (unsigned long *) |
118 | (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE); | |
119 | ||
120 | /* | |
121 | * Loop through all the entries. One of the entries may | |
122 | * for some reason be missed on the stack, so we may | |
123 | * have to account for them. If they are all there, this | |
124 | * loop will only happen once. This code only takes place | |
125 | * on a new max, so it is far from a fast path. | |
126 | */ | |
127 | while (i < max_stack_trace.nr_entries) { | |
0a37119d | 128 | int found = 0; |
1b6cced6 | 129 | |
72ac426a | 130 | stack_dump_index[x] = this_size; |
1b6cced6 SR |
131 | p = start; |
132 | ||
133 | for (; p < top && i < max_stack_trace.nr_entries; p++) { | |
72ac426a SRRH |
134 | if (stack_dump_trace[i] == ULONG_MAX) |
135 | break; | |
1b6cced6 | 136 | if (*p == stack_dump_trace[i]) { |
72ac426a SRRH |
137 | stack_dump_trace[x] = stack_dump_trace[i++]; |
138 | this_size = stack_dump_index[x++] = | |
1b6cced6 | 139 | (top - p) * sizeof(unsigned long); |
0a37119d | 140 | found = 1; |
1b6cced6 SR |
141 | /* Start the search from here */ |
142 | start = p + 1; | |
4df29712 SRRH |
143 | /* |
144 | * We do not want to show the overhead | |
145 | * of the stack tracer stack in the | |
146 | * max stack. If we haven't figured | |
147 | * out what that is, then figure it out | |
148 | * now. | |
149 | */ | |
72ac426a | 150 | if (unlikely(!tracer_frame)) { |
4df29712 SRRH |
151 | tracer_frame = (p - stack) * |
152 | sizeof(unsigned long); | |
153 | max_stack_size -= tracer_frame; | |
154 | } | |
1b6cced6 SR |
155 | } |
156 | } | |
157 | ||
0a37119d SR |
158 | if (!found) |
159 | i++; | |
1b6cced6 SR |
160 | } |
161 | ||
72ac426a SRRH |
162 | max_stack_trace.nr_entries = x; |
163 | for (; x < i; x++) | |
164 | stack_dump_trace[x] = ULONG_MAX; | |
165 | ||
a70857e4 | 166 | if (task_stack_end_corrupted(current)) { |
e3172181 MK |
167 | print_max_stack(); |
168 | BUG(); | |
169 | } | |
170 | ||
e5a81b62 | 171 | out: |
0199c4e6 | 172 | arch_spin_unlock(&max_stack_lock); |
a5e25883 | 173 | local_irq_restore(flags); |
e5a81b62 SR |
174 | } |
175 | ||
176 | static void | |
a1e2e31d SR |
177 | stack_trace_call(unsigned long ip, unsigned long parent_ip, |
178 | struct ftrace_ops *op, struct pt_regs *pt_regs) | |
e5a81b62 | 179 | { |
87889501 | 180 | unsigned long stack; |
5168ae50 | 181 | int cpu; |
e5a81b62 | 182 | |
5168ae50 | 183 | preempt_disable_notrace(); |
e5a81b62 SR |
184 | |
185 | cpu = raw_smp_processor_id(); | |
186 | /* no atomic needed, we only modify this variable by this cpu */ | |
187 | if (per_cpu(trace_active, cpu)++ != 0) | |
188 | goto out; | |
189 | ||
72ac426a | 190 | ip += MCOUNT_INSN_SIZE; |
4df29712 SRRH |
191 | |
192 | check_stack(ip, &stack); | |
e5a81b62 SR |
193 | |
194 | out: | |
195 | per_cpu(trace_active, cpu)--; | |
196 | /* prevent recursion in schedule */ | |
5168ae50 | 197 | preempt_enable_notrace(); |
e5a81b62 SR |
198 | } |
199 | ||
200 | static struct ftrace_ops trace_ops __read_mostly = | |
201 | { | |
202 | .func = stack_trace_call, | |
4740974a | 203 | .flags = FTRACE_OPS_FL_RECURSION_SAFE, |
e5a81b62 SR |
204 | }; |
205 | ||
206 | static ssize_t | |
207 | stack_max_size_read(struct file *filp, char __user *ubuf, | |
208 | size_t count, loff_t *ppos) | |
209 | { | |
210 | unsigned long *ptr = filp->private_data; | |
211 | char buf[64]; | |
212 | int r; | |
213 | ||
214 | r = snprintf(buf, sizeof(buf), "%ld\n", *ptr); | |
215 | if (r > sizeof(buf)) | |
216 | r = sizeof(buf); | |
217 | return simple_read_from_buffer(ubuf, count, ppos, buf, r); | |
218 | } | |
219 | ||
220 | static ssize_t | |
221 | stack_max_size_write(struct file *filp, const char __user *ubuf, | |
222 | size_t count, loff_t *ppos) | |
223 | { | |
224 | long *ptr = filp->private_data; | |
225 | unsigned long val, flags; | |
e5a81b62 | 226 | int ret; |
4f48f8b7 | 227 | int cpu; |
e5a81b62 | 228 | |
22fe9b54 PH |
229 | ret = kstrtoul_from_user(ubuf, count, 10, &val); |
230 | if (ret) | |
e5a81b62 SR |
231 | return ret; |
232 | ||
a5e25883 | 233 | local_irq_save(flags); |
4f48f8b7 LJ |
234 | |
235 | /* | |
236 | * In case we trace inside arch_spin_lock() or after (NMI), | |
237 | * we will cause circular lock, so we also need to increase | |
238 | * the percpu trace_active here. | |
239 | */ | |
240 | cpu = smp_processor_id(); | |
241 | per_cpu(trace_active, cpu)++; | |
242 | ||
0199c4e6 | 243 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 244 | *ptr = val; |
0199c4e6 | 245 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
246 | |
247 | per_cpu(trace_active, cpu)--; | |
a5e25883 | 248 | local_irq_restore(flags); |
e5a81b62 SR |
249 | |
250 | return count; | |
251 | } | |
252 | ||
f38f1d2a | 253 | static const struct file_operations stack_max_size_fops = { |
e5a81b62 SR |
254 | .open = tracing_open_generic, |
255 | .read = stack_max_size_read, | |
256 | .write = stack_max_size_write, | |
6038f373 | 257 | .llseek = default_llseek, |
e5a81b62 SR |
258 | }; |
259 | ||
260 | static void * | |
2fc5f0cf | 261 | __next(struct seq_file *m, loff_t *pos) |
e5a81b62 | 262 | { |
2fc5f0cf | 263 | long n = *pos - 1; |
e5a81b62 | 264 | |
72ac426a | 265 | if (n > max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX) |
e5a81b62 SR |
266 | return NULL; |
267 | ||
2fc5f0cf | 268 | m->private = (void *)n; |
1b6cced6 | 269 | return &m->private; |
e5a81b62 SR |
270 | } |
271 | ||
2fc5f0cf LZ |
272 | static void * |
273 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
e5a81b62 | 274 | { |
2fc5f0cf LZ |
275 | (*pos)++; |
276 | return __next(m, pos); | |
277 | } | |
e5a81b62 | 278 | |
2fc5f0cf LZ |
279 | static void *t_start(struct seq_file *m, loff_t *pos) |
280 | { | |
4f48f8b7 LJ |
281 | int cpu; |
282 | ||
e5a81b62 | 283 | local_irq_disable(); |
4f48f8b7 LJ |
284 | |
285 | cpu = smp_processor_id(); | |
286 | per_cpu(trace_active, cpu)++; | |
287 | ||
0199c4e6 | 288 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 289 | |
522a110b LW |
290 | if (*pos == 0) |
291 | return SEQ_START_TOKEN; | |
292 | ||
2fc5f0cf | 293 | return __next(m, pos); |
e5a81b62 SR |
294 | } |
295 | ||
296 | static void t_stop(struct seq_file *m, void *p) | |
297 | { | |
4f48f8b7 LJ |
298 | int cpu; |
299 | ||
0199c4e6 | 300 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
301 | |
302 | cpu = smp_processor_id(); | |
303 | per_cpu(trace_active, cpu)--; | |
304 | ||
e5a81b62 SR |
305 | local_irq_enable(); |
306 | } | |
307 | ||
962e3707 | 308 | static void trace_lookup_stack(struct seq_file *m, long i) |
e5a81b62 | 309 | { |
1b6cced6 | 310 | unsigned long addr = stack_dump_trace[i]; |
e5a81b62 | 311 | |
962e3707 | 312 | seq_printf(m, "%pS\n", (void *)addr); |
e5a81b62 SR |
313 | } |
314 | ||
e447e1df SR |
315 | static void print_disabled(struct seq_file *m) |
316 | { | |
317 | seq_puts(m, "#\n" | |
318 | "# Stack tracer disabled\n" | |
319 | "#\n" | |
320 | "# To enable the stack tracer, either add 'stacktrace' to the\n" | |
321 | "# kernel command line\n" | |
322 | "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n" | |
323 | "#\n"); | |
324 | } | |
325 | ||
e5a81b62 SR |
326 | static int t_show(struct seq_file *m, void *v) |
327 | { | |
522a110b | 328 | long i; |
1b6cced6 SR |
329 | int size; |
330 | ||
522a110b | 331 | if (v == SEQ_START_TOKEN) { |
eb1871f3 | 332 | seq_printf(m, " Depth Size Location" |
1b6cced6 | 333 | " (%d entries)\n" |
eb1871f3 | 334 | " ----- ---- --------\n", |
72ac426a | 335 | max_stack_trace.nr_entries); |
e447e1df SR |
336 | |
337 | if (!stack_tracer_enabled && !max_stack_size) | |
338 | print_disabled(m); | |
339 | ||
1b6cced6 SR |
340 | return 0; |
341 | } | |
e5a81b62 | 342 | |
522a110b LW |
343 | i = *(long *)v; |
344 | ||
1b6cced6 SR |
345 | if (i >= max_stack_trace.nr_entries || |
346 | stack_dump_trace[i] == ULONG_MAX) | |
e5a81b62 SR |
347 | return 0; |
348 | ||
1b6cced6 SR |
349 | if (i+1 == max_stack_trace.nr_entries || |
350 | stack_dump_trace[i+1] == ULONG_MAX) | |
351 | size = stack_dump_index[i]; | |
352 | else | |
353 | size = stack_dump_index[i] - stack_dump_index[i+1]; | |
354 | ||
355 | seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size); | |
356 | ||
357 | trace_lookup_stack(m, i); | |
e5a81b62 SR |
358 | |
359 | return 0; | |
360 | } | |
361 | ||
f38f1d2a | 362 | static const struct seq_operations stack_trace_seq_ops = { |
e5a81b62 SR |
363 | .start = t_start, |
364 | .next = t_next, | |
365 | .stop = t_stop, | |
366 | .show = t_show, | |
367 | }; | |
368 | ||
369 | static int stack_trace_open(struct inode *inode, struct file *file) | |
370 | { | |
d8cc1ab7 | 371 | return seq_open(file, &stack_trace_seq_ops); |
e5a81b62 SR |
372 | } |
373 | ||
f38f1d2a | 374 | static const struct file_operations stack_trace_fops = { |
e5a81b62 SR |
375 | .open = stack_trace_open, |
376 | .read = seq_read, | |
377 | .llseek = seq_lseek, | |
d8cc1ab7 | 378 | .release = seq_release, |
e5a81b62 SR |
379 | }; |
380 | ||
d2d45c7a SR |
381 | static int |
382 | stack_trace_filter_open(struct inode *inode, struct file *file) | |
383 | { | |
384 | return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER, | |
385 | inode, file); | |
386 | } | |
387 | ||
388 | static const struct file_operations stack_trace_filter_fops = { | |
389 | .open = stack_trace_filter_open, | |
390 | .read = seq_read, | |
391 | .write = ftrace_filter_write, | |
098c879e | 392 | .llseek = tracing_lseek, |
d2d45c7a SR |
393 | .release = ftrace_regex_release, |
394 | }; | |
395 | ||
f38f1d2a SR |
396 | int |
397 | stack_trace_sysctl(struct ctl_table *table, int write, | |
8d65af78 | 398 | void __user *buffer, size_t *lenp, |
f38f1d2a SR |
399 | loff_t *ppos) |
400 | { | |
401 | int ret; | |
402 | ||
403 | mutex_lock(&stack_sysctl_mutex); | |
404 | ||
8d65af78 | 405 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
f38f1d2a SR |
406 | |
407 | if (ret || !write || | |
a32c7765 | 408 | (last_stack_tracer_enabled == !!stack_tracer_enabled)) |
f38f1d2a SR |
409 | goto out; |
410 | ||
a32c7765 | 411 | last_stack_tracer_enabled = !!stack_tracer_enabled; |
f38f1d2a SR |
412 | |
413 | if (stack_tracer_enabled) | |
414 | register_ftrace_function(&trace_ops); | |
415 | else | |
416 | unregister_ftrace_function(&trace_ops); | |
417 | ||
418 | out: | |
419 | mutex_unlock(&stack_sysctl_mutex); | |
420 | return ret; | |
421 | } | |
422 | ||
762e1207 SR |
423 | static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata; |
424 | ||
f38f1d2a SR |
425 | static __init int enable_stacktrace(char *str) |
426 | { | |
762e1207 SR |
427 | if (strncmp(str, "_filter=", 8) == 0) |
428 | strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE); | |
429 | ||
e05a43b7 SR |
430 | stack_tracer_enabled = 1; |
431 | last_stack_tracer_enabled = 1; | |
f38f1d2a SR |
432 | return 1; |
433 | } | |
434 | __setup("stacktrace", enable_stacktrace); | |
435 | ||
e5a81b62 SR |
436 | static __init int stack_trace_init(void) |
437 | { | |
438 | struct dentry *d_tracer; | |
e5a81b62 SR |
439 | |
440 | d_tracer = tracing_init_dentry(); | |
14a5ae40 | 441 | if (IS_ERR(d_tracer)) |
ed6f1c99 | 442 | return 0; |
e5a81b62 | 443 | |
5452af66 FW |
444 | trace_create_file("stack_max_size", 0644, d_tracer, |
445 | &max_stack_size, &stack_max_size_fops); | |
e5a81b62 | 446 | |
5452af66 FW |
447 | trace_create_file("stack_trace", 0444, d_tracer, |
448 | NULL, &stack_trace_fops); | |
e5a81b62 | 449 | |
d2d45c7a SR |
450 | trace_create_file("stack_trace_filter", 0444, d_tracer, |
451 | NULL, &stack_trace_filter_fops); | |
452 | ||
762e1207 SR |
453 | if (stack_trace_filter_buf[0]) |
454 | ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1); | |
455 | ||
e05a43b7 | 456 | if (stack_tracer_enabled) |
f38f1d2a | 457 | register_ftrace_function(&trace_ops); |
e5a81b62 SR |
458 | |
459 | return 0; | |
460 | } | |
461 | ||
462 | device_initcall(stack_trace_init); |