trace_stack.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469
  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. local_irq_save(flags);
  72. arch_spin_lock(&max_stack_lock);
  73. /*
  74. * RCU may not be watching, make it see us.
  75. * The stack trace code uses rcu_sched.
  76. */
  77. rcu_irq_enter();
  78. /* In case another CPU set the tracer_frame on us */
  79. if (unlikely(!frame_size))
  80. this_size -= tracer_frame;
  81. /* a race could have already updated it */
  82. if (this_size <= max_stack_size)
  83. goto out;
  84. max_stack_size = this_size;
  85. max_stack_trace.nr_entries = 0;
  86. max_stack_trace.skip = 3;
  87. save_stack_trace(&max_stack_trace);
  88. /* Skip over the overhead of the stack tracer itself */
  89. for (i = 0; i < max_stack_trace.nr_entries; i++) {
  90. if (stack_dump_trace[i] == ip)
  91. break;
  92. }
  93. /*
  94. * Now find where in the stack these are.
  95. */
  96. x = 0;
  97. start = stack;
  98. top = (unsigned long *)
  99. (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
  100. /*
  101. * Loop through all the entries. One of the entries may
  102. * for some reason be missed on the stack, so we may
  103. * have to account for them. If they are all there, this
  104. * loop will only happen once. This code only takes place
  105. * on a new max, so it is far from a fast path.
  106. */
  107. while (i < max_stack_trace.nr_entries) {
  108. int found = 0;
  109. stack_dump_index[x] = this_size;
  110. p = start;
  111. for (; p < top && i < max_stack_trace.nr_entries; p++) {
  112. if (stack_dump_trace[i] == ULONG_MAX)
  113. break;
  114. if (*p == stack_dump_trace[i]) {
  115. stack_dump_trace[x] = stack_dump_trace[i++];
  116. this_size = stack_dump_index[x++] =
  117. (top - p) * sizeof(unsigned long);
  118. found = 1;
  119. /* Start the search from here */
  120. start = p + 1;
  121. /*
  122. * We do not want to show the overhead
  123. * of the stack tracer stack in the
  124. * max stack. If we haven't figured
  125. * out what that is, then figure it out
  126. * now.
  127. */
  128. if (unlikely(!tracer_frame)) {
  129. tracer_frame = (p - stack) *
  130. sizeof(unsigned long);
  131. max_stack_size -= tracer_frame;
  132. }
  133. }
  134. }
  135. if (!found)
  136. i++;
  137. }
  138. max_stack_trace.nr_entries = x;
  139. for (; x < i; x++)
  140. stack_dump_trace[x] = ULONG_MAX;
  141. if (task_stack_end_corrupted(current)) {
  142. print_max_stack();
  143. BUG();
  144. }
  145. out:
  146. rcu_irq_exit();
  147. arch_spin_unlock(&max_stack_lock);
  148. local_irq_restore(flags);
  149. }
  150. static void
  151. stack_trace_call(unsigned long ip, unsigned long parent_ip,
  152. struct ftrace_ops *op, struct pt_regs *pt_regs)
  153. {
  154. unsigned long stack;
  155. int cpu;
  156. preempt_disable_notrace();
  157. cpu = raw_smp_processor_id();
  158. /* no atomic needed, we only modify this variable by this cpu */
  159. if (per_cpu(trace_active, cpu)++ != 0)
  160. goto out;
  161. ip += MCOUNT_INSN_SIZE;
  162. check_stack(ip, &stack);
  163. out:
  164. per_cpu(trace_active, cpu)--;
  165. /* prevent recursion in schedule */
  166. preempt_enable_notrace();
  167. }
  168. static struct ftrace_ops trace_ops __read_mostly =
  169. {
  170. .func = stack_trace_call,
  171. .flags = FTRACE_OPS_FL_RECURSION_SAFE,
  172. };
  173. static ssize_t
  174. stack_max_size_read(struct file *filp, char __user *ubuf,
  175. size_t count, loff_t *ppos)
  176. {
  177. unsigned long *ptr = filp->private_data;
  178. char buf[64];
  179. int r;
  180. r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
  181. if (r > sizeof(buf))
  182. r = sizeof(buf);
  183. return simple_read_from_buffer(ubuf, count, ppos, buf, r);
  184. }
  185. static ssize_t
  186. stack_max_size_write(struct file *filp, const char __user *ubuf,
  187. size_t count, loff_t *ppos)
  188. {
  189. long *ptr = filp->private_data;
  190. unsigned long val, flags;
  191. int ret;
  192. int cpu;
  193. ret = kstrtoul_from_user(ubuf, count, 10, &val);
  194. if (ret)
  195. return ret;
  196. local_irq_save(flags);
  197. /*
  198. * In case we trace inside arch_spin_lock() or after (NMI),
  199. * we will cause circular lock, so we also need to increase
  200. * the percpu trace_active here.
  201. */
  202. cpu = smp_processor_id();
  203. per_cpu(trace_active, cpu)++;
  204. arch_spin_lock(&max_stack_lock);
  205. *ptr = val;
  206. arch_spin_unlock(&max_stack_lock);
  207. per_cpu(trace_active, cpu)--;
  208. local_irq_restore(flags);
  209. return count;
  210. }
  211. static const struct file_operations stack_max_size_fops = {
  212. .open = tracing_open_generic,
  213. .read = stack_max_size_read,
  214. .write = stack_max_size_write,
  215. .llseek = default_llseek,
  216. };
  217. static void *
  218. __next(struct seq_file *m, loff_t *pos)
  219. {
  220. long n = *pos - 1;
  221. if (n > max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
  222. return NULL;
  223. m->private = (void *)n;
  224. return &m->private;
  225. }
  226. static void *
  227. t_next(struct seq_file *m, void *v, loff_t *pos)
  228. {
  229. (*pos)++;
  230. return __next(m, pos);
  231. }
  232. static void *t_start(struct seq_file *m, loff_t *pos)
  233. {
  234. int cpu;
  235. local_irq_disable();
  236. cpu = smp_processor_id();
  237. per_cpu(trace_active, cpu)++;
  238. arch_spin_lock(&max_stack_lock);
  239. if (*pos == 0)
  240. return SEQ_START_TOKEN;
  241. return __next(m, pos);
  242. }
  243. static void t_stop(struct seq_file *m, void *p)
  244. {
  245. int cpu;
  246. arch_spin_unlock(&max_stack_lock);
  247. cpu = smp_processor_id();
  248. per_cpu(trace_active, cpu)--;
  249. local_irq_enable();
  250. }
  251. static void trace_lookup_stack(struct seq_file *m, long i)
  252. {
  253. unsigned long addr = stack_dump_trace[i];
  254. seq_printf(m, "%pS\n", (void *)addr);
  255. }
  256. static void print_disabled(struct seq_file *m)
  257. {
  258. seq_puts(m, "#\n"
  259. "# Stack tracer disabled\n"
  260. "#\n"
  261. "# To enable the stack tracer, either add 'stacktrace' to the\n"
  262. "# kernel command line\n"
  263. "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
  264. "#\n");
  265. }
  266. static int t_show(struct seq_file *m, void *v)
  267. {
  268. long i;
  269. int size;
  270. if (v == SEQ_START_TOKEN) {
  271. seq_printf(m, " Depth Size Location"
  272. " (%d entries)\n"
  273. " ----- ---- --------\n",
  274. max_stack_trace.nr_entries);
  275. if (!stack_tracer_enabled && !max_stack_size)
  276. print_disabled(m);
  277. return 0;
  278. }
  279. i = *(long *)v;
  280. if (i >= max_stack_trace.nr_entries ||
  281. stack_dump_trace[i] == ULONG_MAX)
  282. return 0;
  283. if (i+1 == max_stack_trace.nr_entries ||
  284. stack_dump_trace[i+1] == ULONG_MAX)
  285. size = stack_dump_index[i];
  286. else
  287. size = stack_dump_index[i] - stack_dump_index[i+1];
  288. seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size);
  289. trace_lookup_stack(m, i);
  290. return 0;
  291. }
  292. static const struct seq_operations stack_trace_seq_ops = {
  293. .start = t_start,
  294. .next = t_next,
  295. .stop = t_stop,
  296. .show = t_show,
  297. };
  298. static int stack_trace_open(struct inode *inode, struct file *file)
  299. {
  300. return seq_open(file, &stack_trace_seq_ops);
  301. }
  302. static const struct file_operations stack_trace_fops = {
  303. .open = stack_trace_open,
  304. .read = seq_read,
  305. .llseek = seq_lseek,
  306. .release = seq_release,
  307. };
  308. static int
  309. stack_trace_filter_open(struct inode *inode, struct file *file)
  310. {
  311. return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
  312. inode, file);
  313. }
  314. static const struct file_operations stack_trace_filter_fops = {
  315. .open = stack_trace_filter_open,
  316. .read = seq_read,
  317. .write = ftrace_filter_write,
  318. .llseek = tracing_lseek,
  319. .release = ftrace_regex_release,
  320. };
  321. int
  322. stack_trace_sysctl(struct ctl_table *table, int write,
  323. void __user *buffer, size_t *lenp,
  324. loff_t *ppos)
  325. {
  326. int ret;
  327. mutex_lock(&stack_sysctl_mutex);
  328. ret = proc_dointvec(table, write, buffer, lenp, ppos);
  329. if (ret || !write ||
  330. (last_stack_tracer_enabled == !!stack_tracer_enabled))
  331. goto out;
  332. last_stack_tracer_enabled = !!stack_tracer_enabled;
  333. if (stack_tracer_enabled)
  334. register_ftrace_function(&trace_ops);
  335. else
  336. unregister_ftrace_function(&trace_ops);
  337. out:
  338. mutex_unlock(&stack_sysctl_mutex);
  339. return ret;
  340. }
  341. static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
  342. static __init int enable_stacktrace(char *str)
  343. {
  344. if (strncmp(str, "_filter=", 8) == 0)
  345. strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
  346. stack_tracer_enabled = 1;
  347. last_stack_tracer_enabled = 1;
  348. return 1;
  349. }
  350. __setup("stacktrace", enable_stacktrace);
  351. static __init int stack_trace_init(void)
  352. {
  353. struct dentry *d_tracer;
  354. d_tracer = tracing_init_dentry();
  355. if (IS_ERR(d_tracer))
  356. return 0;
  357. trace_create_file("stack_max_size", 0644, d_tracer,
  358. &max_stack_size, &stack_max_size_fops);
  359. trace_create_file("stack_trace", 0444, d_tracer,
  360. NULL, &stack_trace_fops);
  361. trace_create_file("stack_trace_filter", 0444, d_tracer,
  362. NULL, &stack_trace_filter_fops);
  363. if (stack_trace_filter_buf[0])
  364. ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
  365. if (stack_tracer_enabled)
  366. register_ftrace_function(&trace_ops);
  367. return 0;
  368. }
  369. device_initcall(stack_trace_init);