trace_stack.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473
  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>
  10. #include <linux/ftrace.h>
  11. #include <linux/module.h>
  12. #include <linux/sysctl.h>
  13. #include <linux/init.h>
  14. #include <asm/setup.h>
  15. #include "trace.h"
  16. #define STACK_TRACE_ENTRIES 500
  17. static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
  18. { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
  19. static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
  20. /*
  21. * Reserve one entry for the passed in ip. This will allow
  22. * us to remove most or all of the stack size overhead
  23. * added by the stack tracer itself.
  24. */
  25. static struct stack_trace max_stack_trace = {
  26. .max_entries = STACK_TRACE_ENTRIES - 1,
  27. .entries = &stack_dump_trace[0],
  28. };
  29. static unsigned long max_stack_size;
  30. static arch_spinlock_t max_stack_lock =
  31. (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  32. static DEFINE_PER_CPU(int, trace_active);
  33. static DEFINE_MUTEX(stack_sysctl_mutex);
  34. int stack_tracer_enabled;
  35. static int last_stack_tracer_enabled;
  36. static inline void print_max_stack(void)
  37. {
  38. long i;
  39. int size;
  40. pr_emerg(" Depth Size Location (%d entries)\n"
  41. " ----- ---- --------\n",
  42. max_stack_trace.nr_entries);
  43. for (i = 0; i < max_stack_trace.nr_entries; i++) {
  44. if (stack_dump_trace[i] == ULONG_MAX)
  45. break;
  46. if (i+1 == max_stack_trace.nr_entries ||
  47. stack_dump_trace[i+1] == ULONG_MAX)
  48. size = stack_dump_index[i];
  49. else
  50. size = stack_dump_index[i] - stack_dump_index[i+1];
  51. pr_emerg("%3ld) %8d %5d %pS\n", i, stack_dump_index[i],
  52. size, (void *)stack_dump_trace[i]);
  53. }
  54. }
  55. static inline void
  56. check_stack(unsigned long ip, unsigned long *stack)
  57. {
  58. unsigned long this_size, flags; unsigned long *p, *top, *start;
  59. static int tracer_frame;
  60. int frame_size = ACCESS_ONCE(tracer_frame);
  61. int i, x;
  62. this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
  63. this_size = THREAD_SIZE - this_size;
  64. /* Remove the frame of the tracer */
  65. this_size -= frame_size;
  66. if (this_size <= max_stack_size)
  67. return;
  68. /* we do not handle interrupt stacks yet */
  69. if (!object_is_on_stack(stack))
  70. return;
  71. /* Can't do this from NMI context (can cause deadlocks) */
  72. if (in_nmi())
  73. return;
  74. local_irq_save(flags);
  75. arch_spin_lock(&max_stack_lock);
  76. /*
  77. * RCU may not be watching, make it see us.
  78. * The stack trace code uses rcu_sched.
  79. */
  80. rcu_irq_enter();
  81. /* In case another CPU set the tracer_frame on us */
  82. if (unlikely(!frame_size))
  83. this_size -= tracer_frame;
  84. /* a race could have already updated it */
  85. if (this_size <= max_stack_size)
  86. goto out;
  87. max_stack_size = this_size;
  88. max_stack_trace.nr_entries = 0;
  89. max_stack_trace.skip = 3;
  90. save_stack_trace(&max_stack_trace);
  91. /* Skip over the overhead of the stack tracer itself */
  92. for (i = 0; i < max_stack_trace.nr_entries; i++) {
  93. if (stack_dump_trace[i] == ip)
  94. break;
  95. }
  96. /*
  97. * Now find where in the stack these are.
  98. */
  99. x = 0;
  100. start = stack;
  101. top = (unsigned long *)
  102. (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
  103. /*
  104. * Loop through all the entries. One of the entries may
  105. * for some reason be missed on the stack, so we may
  106. * have to account for them. If they are all there, this
  107. * loop will only happen once. This code only takes place
  108. * on a new max, so it is far from a fast path.
  109. */
  110. while (i < max_stack_trace.nr_entries) {
  111. int found = 0;
  112. stack_dump_index[x] = this_size;
  113. p = start;
  114. for (; p < top && i < max_stack_trace.nr_entries; p++) {
  115. if (stack_dump_trace[i] == ULONG_MAX)
  116. break;
  117. if (*p == stack_dump_trace[i]) {
  118. stack_dump_trace[x] = stack_dump_trace[i++];
  119. this_size = stack_dump_index[x++] =
  120. (top - p) * sizeof(unsigned long);
  121. found = 1;
  122. /* Start the search from here */
  123. start = p + 1;
  124. /*
  125. * We do not want to show the overhead
  126. * of the stack tracer stack in the
  127. * max stack. If we haven't figured
  128. * out what that is, then figure it out
  129. * now.
  130. */
  131. if (unlikely(!tracer_frame)) {
  132. tracer_frame = (p - stack) *
  133. sizeof(unsigned long);
  134. max_stack_size -= tracer_frame;
  135. }
  136. }
  137. }
  138. if (!found)
  139. i++;
  140. }
  141. max_stack_trace.nr_entries = x;
  142. for (; x < i; x++)
  143. stack_dump_trace[x] = ULONG_MAX;
  144. if (task_stack_end_corrupted(current)) {
  145. print_max_stack();
  146. BUG();
  147. }
  148. out:
  149. rcu_irq_exit();
  150. arch_spin_unlock(&max_stack_lock);
  151. local_irq_restore(flags);
  152. }
  153. static void
  154. stack_trace_call(unsigned long ip, unsigned long parent_ip,
  155. struct ftrace_ops *op, struct pt_regs *pt_regs)
  156. {
  157. unsigned long stack;
  158. int cpu;
  159. preempt_disable_notrace();
  160. cpu = raw_smp_processor_id();
  161. /* no atomic needed, we only modify this variable by this cpu */
  162. if (per_cpu(trace_active, cpu)++ != 0)
  163. goto out;
  164. ip += MCOUNT_INSN_SIZE;
  165. check_stack(ip, &stack);
  166. out:
  167. per_cpu(trace_active, cpu)--;
  168. /* prevent recursion in schedule */
  169. preempt_enable_notrace();
  170. }
  171. static struct ftrace_ops trace_ops __read_mostly =
  172. {
  173. .func = stack_trace_call,
  174. .flags = FTRACE_OPS_FL_RECURSION_SAFE,
  175. };
  176. static ssize_t
  177. stack_max_size_read(struct file *filp, char __user *ubuf,
  178. size_t count, loff_t *ppos)
  179. {
  180. unsigned long *ptr = filp->private_data;
  181. char buf[64];
  182. int r;
  183. r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
  184. if (r > sizeof(buf))
  185. r = sizeof(buf);
  186. return simple_read_from_buffer(ubuf, count, ppos, buf, r);
  187. }
  188. static ssize_t
  189. stack_max_size_write(struct file *filp, const char __user *ubuf,
  190. size_t count, loff_t *ppos)
  191. {
  192. long *ptr = filp->private_data;
  193. unsigned long val, flags;
  194. int ret;
  195. int cpu;
  196. ret = kstrtoul_from_user(ubuf, count, 10, &val);
  197. if (ret)
  198. return ret;
  199. local_irq_save(flags);
  200. /*
  201. * In case we trace inside arch_spin_lock() or after (NMI),
  202. * we will cause circular lock, so we also need to increase
  203. * the percpu trace_active here.
  204. */
  205. cpu = smp_processor_id();
  206. per_cpu(trace_active, cpu)++;
  207. arch_spin_lock(&max_stack_lock);
  208. *ptr = val;
  209. arch_spin_unlock(&max_stack_lock);
  210. per_cpu(trace_active, cpu)--;
  211. local_irq_restore(flags);
  212. return count;
  213. }
  214. static const struct file_operations stack_max_size_fops = {
  215. .open = tracing_open_generic,
  216. .read = stack_max_size_read,
  217. .write = stack_max_size_write,
  218. .llseek = default_llseek,
  219. };
  220. static void *
  221. __next(struct seq_file *m, loff_t *pos)
  222. {
  223. long n = *pos - 1;
  224. if (n > max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
  225. return NULL;
  226. m->private = (void *)n;
  227. return &m->private;
  228. }
  229. static void *
  230. t_next(struct seq_file *m, void *v, loff_t *pos)
  231. {
  232. (*pos)++;
  233. return __next(m, pos);
  234. }
  235. static void *t_start(struct seq_file *m, loff_t *pos)
  236. {
  237. int cpu;
  238. local_irq_disable();
  239. cpu = smp_processor_id();
  240. per_cpu(trace_active, cpu)++;
  241. arch_spin_lock(&max_stack_lock);
  242. if (*pos == 0)
  243. return SEQ_START_TOKEN;
  244. return __next(m, pos);
  245. }
  246. static void t_stop(struct seq_file *m, void *p)
  247. {
  248. int cpu;
  249. arch_spin_unlock(&max_stack_lock);
  250. cpu = smp_processor_id();
  251. per_cpu(trace_active, cpu)--;
  252. local_irq_enable();
  253. }
  254. static void trace_lookup_stack(struct seq_file *m, long i)
  255. {
  256. unsigned long addr = stack_dump_trace[i];
  257. seq_printf(m, "%pS\n", (void *)addr);
  258. }
  259. static void print_disabled(struct seq_file *m)
  260. {
  261. seq_puts(m, "#\n"
  262. "# Stack tracer disabled\n"
  263. "#\n"
  264. "# To enable the stack tracer, either add 'stacktrace' to the\n"
  265. "# kernel command line\n"
  266. "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
  267. "#\n");
  268. }
  269. static int t_show(struct seq_file *m, void *v)
  270. {
  271. long i;
  272. int size;
  273. if (v == SEQ_START_TOKEN) {
  274. seq_printf(m, " Depth Size Location"
  275. " (%d entries)\n"
  276. " ----- ---- --------\n",
  277. max_stack_trace.nr_entries);
  278. if (!stack_tracer_enabled && !max_stack_size)
  279. print_disabled(m);
  280. return 0;
  281. }
  282. i = *(long *)v;
  283. if (i >= max_stack_trace.nr_entries ||
  284. stack_dump_trace[i] == ULONG_MAX)
  285. return 0;
  286. if (i+1 == max_stack_trace.nr_entries ||
  287. stack_dump_trace[i+1] == ULONG_MAX)
  288. size = stack_dump_index[i];
  289. else
  290. size = stack_dump_index[i] - stack_dump_index[i+1];
  291. seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size);
  292. trace_lookup_stack(m, i);
  293. return 0;
  294. }
  295. static const struct seq_operations stack_trace_seq_ops = {
  296. .start = t_start,
  297. .next = t_next,
  298. .stop = t_stop,
  299. .show = t_show,
  300. };
  301. static int stack_trace_open(struct inode *inode, struct file *file)
  302. {
  303. return seq_open(file, &stack_trace_seq_ops);
  304. }
  305. static const struct file_operations stack_trace_fops = {
  306. .open = stack_trace_open,
  307. .read = seq_read,
  308. .llseek = seq_lseek,
  309. .release = seq_release,
  310. };
  311. static int
  312. stack_trace_filter_open(struct inode *inode, struct file *file)
  313. {
  314. return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
  315. inode, file);
  316. }
  317. static const struct file_operations stack_trace_filter_fops = {
  318. .open = stack_trace_filter_open,
  319. .read = seq_read,
  320. .write = ftrace_filter_write,
  321. .llseek = tracing_lseek,
  322. .release = ftrace_regex_release,
  323. };
  324. int
  325. stack_trace_sysctl(struct ctl_table *table, int write,
  326. void __user *buffer, size_t *lenp,
  327. loff_t *ppos)
  328. {
  329. int ret;
  330. mutex_lock(&stack_sysctl_mutex);
  331. ret = proc_dointvec(table, write, buffer, lenp, ppos);
  332. if (ret || !write ||
  333. (last_stack_tracer_enabled == !!stack_tracer_enabled))
  334. goto out;
  335. last_stack_tracer_enabled = !!stack_tracer_enabled;
  336. if (stack_tracer_enabled)
  337. register_ftrace_function(&trace_ops);
  338. else
  339. unregister_ftrace_function(&trace_ops);
  340. out:
  341. mutex_unlock(&stack_sysctl_mutex);
  342. return ret;
  343. }
  344. static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
  345. static __init int enable_stacktrace(char *str)
  346. {
  347. if (strncmp(str, "_filter=", 8) == 0)
  348. strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
  349. stack_tracer_enabled = 1;
  350. last_stack_tracer_enabled = 1;
  351. return 1;
  352. }
  353. __setup("stacktrace", enable_stacktrace);
  354. static __init int stack_trace_init(void)
  355. {
  356. struct dentry *d_tracer;
  357. d_tracer = tracing_init_dentry();
  358. if (IS_ERR(d_tracer))
  359. return 0;
  360. trace_create_file("stack_max_size", 0644, d_tracer,
  361. &max_stack_size, &stack_max_size_fops);
  362. trace_create_file("stack_trace", 0444, d_tracer,
  363. NULL, &stack_trace_fops);
  364. trace_create_file("stack_trace_filter", 0444, d_tracer,
  365. NULL, &stack_trace_filter_fops);
  366. if (stack_trace_filter_buf[0])
  367. ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
  368. if (stack_tracer_enabled)
  369. register_ftrace_function(&trace_ops);
  370. return 0;
  371. }
  372. device_initcall(stack_trace_init);