trace_stack.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491
  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. static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
  17. { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
  18. unsigned stack_trace_index[STACK_TRACE_ENTRIES];
  19. /*
  20. * Reserve one entry for the passed in ip. This will allow
  21. * us to remove most or all of the stack size overhead
  22. * added by the stack tracer itself.
  23. */
  24. struct stack_trace stack_trace_max = {
  25. .max_entries = STACK_TRACE_ENTRIES - 1,
  26. .entries = &stack_dump_trace[0],
  27. };
  28. unsigned long stack_trace_max_size;
  29. arch_spinlock_t stack_trace_max_lock =
  30. (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  31. static DEFINE_PER_CPU(int, trace_active);
  32. static DEFINE_MUTEX(stack_sysctl_mutex);
  33. int stack_tracer_enabled;
  34. static int last_stack_tracer_enabled;
  35. void stack_trace_print(void)
  36. {
  37. long i;
  38. int size;
  39. pr_emerg(" Depth Size Location (%d entries)\n"
  40. " ----- ---- --------\n",
  41. stack_trace_max.nr_entries);
  42. for (i = 0; i < stack_trace_max.nr_entries; i++) {
  43. if (stack_dump_trace[i] == ULONG_MAX)
  44. break;
  45. if (i+1 == stack_trace_max.nr_entries ||
  46. stack_dump_trace[i+1] == ULONG_MAX)
  47. size = stack_trace_index[i];
  48. else
  49. size = stack_trace_index[i] - stack_trace_index[i+1];
  50. pr_emerg("%3ld) %8d %5d %pS\n", i, stack_trace_index[i],
  51. size, (void *)stack_dump_trace[i]);
  52. }
  53. }
  54. /*
  55. * When arch-specific code overides this function, the following
  56. * data should be filled up, assuming stack_trace_max_lock is held to
  57. * prevent concurrent updates.
  58. * stack_trace_index[]
  59. * stack_trace_max
  60. * stack_trace_max_size
  61. */
  62. void __weak
  63. check_stack(unsigned long ip, unsigned long *stack)
  64. {
  65. unsigned long this_size, flags; unsigned long *p, *top, *start;
  66. static int tracer_frame;
  67. int frame_size = ACCESS_ONCE(tracer_frame);
  68. int i, x;
  69. this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
  70. this_size = THREAD_SIZE - this_size;
  71. /* Remove the frame of the tracer */
  72. this_size -= frame_size;
  73. if (this_size <= stack_trace_max_size)
  74. return;
  75. /* we do not handle interrupt stacks yet */
  76. if (!object_is_on_stack(stack))
  77. return;
  78. /* Can't do this from NMI context (can cause deadlocks) */
  79. if (in_nmi())
  80. return;
  81. local_irq_save(flags);
  82. arch_spin_lock(&stack_trace_max_lock);
  83. /*
  84. * RCU may not be watching, make it see us.
  85. * The stack trace code uses rcu_sched.
  86. */
  87. rcu_irq_enter();
  88. /* In case another CPU set the tracer_frame on us */
  89. if (unlikely(!frame_size))
  90. this_size -= tracer_frame;
  91. /* a race could have already updated it */
  92. if (this_size <= stack_trace_max_size)
  93. goto out;
  94. stack_trace_max_size = this_size;
  95. stack_trace_max.nr_entries = 0;
  96. stack_trace_max.skip = 3;
  97. save_stack_trace(&stack_trace_max);
  98. /* Skip over the overhead of the stack tracer itself */
  99. for (i = 0; i < stack_trace_max.nr_entries; i++) {
  100. if (stack_dump_trace[i] == ip)
  101. break;
  102. }
  103. /*
  104. * Some archs may not have the passed in ip in the dump.
  105. * If that happens, we need to show everything.
  106. */
  107. if (i == stack_trace_max.nr_entries)
  108. i = 0;
  109. /*
  110. * Now find where in the stack these are.
  111. */
  112. x = 0;
  113. start = stack;
  114. top = (unsigned long *)
  115. (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
  116. /*
  117. * Loop through all the entries. One of the entries may
  118. * for some reason be missed on the stack, so we may
  119. * have to account for them. If they are all there, this
  120. * loop will only happen once. This code only takes place
  121. * on a new max, so it is far from a fast path.
  122. */
  123. while (i < stack_trace_max.nr_entries) {
  124. int found = 0;
  125. stack_trace_index[x] = this_size;
  126. p = start;
  127. for (; p < top && i < stack_trace_max.nr_entries; p++) {
  128. if (stack_dump_trace[i] == ULONG_MAX)
  129. break;
  130. /*
  131. * The READ_ONCE_NOCHECK is used to let KASAN know that
  132. * this is not a stack-out-of-bounds error.
  133. */
  134. if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) {
  135. stack_dump_trace[x] = stack_dump_trace[i++];
  136. this_size = stack_trace_index[x++] =
  137. (top - p) * sizeof(unsigned long);
  138. found = 1;
  139. /* Start the search from here */
  140. start = p + 1;
  141. /*
  142. * We do not want to show the overhead
  143. * of the stack tracer stack in the
  144. * max stack. If we haven't figured
  145. * out what that is, then figure it out
  146. * now.
  147. */
  148. if (unlikely(!tracer_frame)) {
  149. tracer_frame = (p - stack) *
  150. sizeof(unsigned long);
  151. stack_trace_max_size -= tracer_frame;
  152. }
  153. }
  154. }
  155. if (!found)
  156. i++;
  157. }
  158. stack_trace_max.nr_entries = x;
  159. for (; x < i; x++)
  160. stack_dump_trace[x] = ULONG_MAX;
  161. if (task_stack_end_corrupted(current)) {
  162. stack_trace_print();
  163. BUG();
  164. }
  165. out:
  166. rcu_irq_exit();
  167. arch_spin_unlock(&stack_trace_max_lock);
  168. local_irq_restore(flags);
  169. }
  170. static void
  171. stack_trace_call(unsigned long ip, unsigned long parent_ip,
  172. struct ftrace_ops *op, struct pt_regs *pt_regs)
  173. {
  174. unsigned long stack;
  175. int cpu;
  176. preempt_disable_notrace();
  177. cpu = raw_smp_processor_id();
  178. /* no atomic needed, we only modify this variable by this cpu */
  179. if (per_cpu(trace_active, cpu)++ != 0)
  180. goto out;
  181. ip += MCOUNT_INSN_SIZE;
  182. check_stack(ip, &stack);
  183. out:
  184. per_cpu(trace_active, cpu)--;
  185. /* prevent recursion in schedule */
  186. preempt_enable_notrace();
  187. }
  188. static struct ftrace_ops trace_ops __read_mostly =
  189. {
  190. .func = stack_trace_call,
  191. .flags = FTRACE_OPS_FL_RECURSION_SAFE,
  192. };
  193. static ssize_t
  194. stack_max_size_read(struct file *filp, char __user *ubuf,
  195. size_t count, loff_t *ppos)
  196. {
  197. unsigned long *ptr = filp->private_data;
  198. char buf[64];
  199. int r;
  200. r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
  201. if (r > sizeof(buf))
  202. r = sizeof(buf);
  203. return simple_read_from_buffer(ubuf, count, ppos, buf, r);
  204. }
  205. static ssize_t
  206. stack_max_size_write(struct file *filp, const char __user *ubuf,
  207. size_t count, loff_t *ppos)
  208. {
  209. long *ptr = filp->private_data;
  210. unsigned long val, flags;
  211. int ret;
  212. int cpu;
  213. ret = kstrtoul_from_user(ubuf, count, 10, &val);
  214. if (ret)
  215. return ret;
  216. local_irq_save(flags);
  217. /*
  218. * In case we trace inside arch_spin_lock() or after (NMI),
  219. * we will cause circular lock, so we also need to increase
  220. * the percpu trace_active here.
  221. */
  222. cpu = smp_processor_id();
  223. per_cpu(trace_active, cpu)++;
  224. arch_spin_lock(&stack_trace_max_lock);
  225. *ptr = val;
  226. arch_spin_unlock(&stack_trace_max_lock);
  227. per_cpu(trace_active, cpu)--;
  228. local_irq_restore(flags);
  229. return count;
  230. }
  231. static const struct file_operations stack_max_size_fops = {
  232. .open = tracing_open_generic,
  233. .read = stack_max_size_read,
  234. .write = stack_max_size_write,
  235. .llseek = default_llseek,
  236. };
  237. static void *
  238. __next(struct seq_file *m, loff_t *pos)
  239. {
  240. long n = *pos - 1;
  241. if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
  242. return NULL;
  243. m->private = (void *)n;
  244. return &m->private;
  245. }
  246. static void *
  247. t_next(struct seq_file *m, void *v, loff_t *pos)
  248. {
  249. (*pos)++;
  250. return __next(m, pos);
  251. }
  252. static void *t_start(struct seq_file *m, loff_t *pos)
  253. {
  254. int cpu;
  255. local_irq_disable();
  256. cpu = smp_processor_id();
  257. per_cpu(trace_active, cpu)++;
  258. arch_spin_lock(&stack_trace_max_lock);
  259. if (*pos == 0)
  260. return SEQ_START_TOKEN;
  261. return __next(m, pos);
  262. }
  263. static void t_stop(struct seq_file *m, void *p)
  264. {
  265. int cpu;
  266. arch_spin_unlock(&stack_trace_max_lock);
  267. cpu = smp_processor_id();
  268. per_cpu(trace_active, cpu)--;
  269. local_irq_enable();
  270. }
  271. static void trace_lookup_stack(struct seq_file *m, long i)
  272. {
  273. unsigned long addr = stack_dump_trace[i];
  274. seq_printf(m, "%pS\n", (void *)addr);
  275. }
  276. static void print_disabled(struct seq_file *m)
  277. {
  278. seq_puts(m, "#\n"
  279. "# Stack tracer disabled\n"
  280. "#\n"
  281. "# To enable the stack tracer, either add 'stacktrace' to the\n"
  282. "# kernel command line\n"
  283. "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
  284. "#\n");
  285. }
  286. static int t_show(struct seq_file *m, void *v)
  287. {
  288. long i;
  289. int size;
  290. if (v == SEQ_START_TOKEN) {
  291. seq_printf(m, " Depth Size Location"
  292. " (%d entries)\n"
  293. " ----- ---- --------\n",
  294. stack_trace_max.nr_entries);
  295. if (!stack_tracer_enabled && !stack_trace_max_size)
  296. print_disabled(m);
  297. return 0;
  298. }
  299. i = *(long *)v;
  300. if (i >= stack_trace_max.nr_entries ||
  301. stack_dump_trace[i] == ULONG_MAX)
  302. return 0;
  303. if (i+1 == stack_trace_max.nr_entries ||
  304. stack_dump_trace[i+1] == ULONG_MAX)
  305. size = stack_trace_index[i];
  306. else
  307. size = stack_trace_index[i] - stack_trace_index[i+1];
  308. seq_printf(m, "%3ld) %8d %5d ", i, stack_trace_index[i], size);
  309. trace_lookup_stack(m, i);
  310. return 0;
  311. }
  312. static const struct seq_operations stack_trace_seq_ops = {
  313. .start = t_start,
  314. .next = t_next,
  315. .stop = t_stop,
  316. .show = t_show,
  317. };
  318. static int stack_trace_open(struct inode *inode, struct file *file)
  319. {
  320. return seq_open(file, &stack_trace_seq_ops);
  321. }
  322. static const struct file_operations stack_trace_fops = {
  323. .open = stack_trace_open,
  324. .read = seq_read,
  325. .llseek = seq_lseek,
  326. .release = seq_release,
  327. };
  328. static int
  329. stack_trace_filter_open(struct inode *inode, struct file *file)
  330. {
  331. return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
  332. inode, file);
  333. }
  334. static const struct file_operations stack_trace_filter_fops = {
  335. .open = stack_trace_filter_open,
  336. .read = seq_read,
  337. .write = ftrace_filter_write,
  338. .llseek = tracing_lseek,
  339. .release = ftrace_regex_release,
  340. };
  341. int
  342. stack_trace_sysctl(struct ctl_table *table, int write,
  343. void __user *buffer, size_t *lenp,
  344. loff_t *ppos)
  345. {
  346. int ret;
  347. mutex_lock(&stack_sysctl_mutex);
  348. ret = proc_dointvec(table, write, buffer, lenp, ppos);
  349. if (ret || !write ||
  350. (last_stack_tracer_enabled == !!stack_tracer_enabled))
  351. goto out;
  352. last_stack_tracer_enabled = !!stack_tracer_enabled;
  353. if (stack_tracer_enabled)
  354. register_ftrace_function(&trace_ops);
  355. else
  356. unregister_ftrace_function(&trace_ops);
  357. out:
  358. mutex_unlock(&stack_sysctl_mutex);
  359. return ret;
  360. }
  361. static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
  362. static __init int enable_stacktrace(char *str)
  363. {
  364. if (strncmp(str, "_filter=", 8) == 0)
  365. strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
  366. stack_tracer_enabled = 1;
  367. last_stack_tracer_enabled = 1;
  368. return 1;
  369. }
  370. __setup("stacktrace", enable_stacktrace);
  371. static __init int stack_trace_init(void)
  372. {
  373. struct dentry *d_tracer;
  374. d_tracer = tracing_init_dentry();
  375. if (IS_ERR(d_tracer))
  376. return 0;
  377. trace_create_file("stack_max_size", 0644, d_tracer,
  378. &stack_trace_max_size, &stack_max_size_fops);
  379. trace_create_file("stack_trace", 0444, d_tracer,
  380. NULL, &stack_trace_fops);
  381. trace_create_file("stack_trace_filter", 0444, d_tracer,
  382. NULL, &stack_trace_filter_fops);
  383. if (stack_trace_filter_buf[0])
  384. ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
  385. if (stack_tracer_enabled)
  386. register_ftrace_function(&trace_ops);
  387. return 0;
  388. }
  389. device_initcall(stack_trace_init);