]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2008-2014 Mathieu Desnoyers | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | */ | |
18 | #include <linux/module.h> | |
19 | #include <linux/mutex.h> | |
20 | #include <linux/types.h> | |
21 | #include <linux/jhash.h> | |
22 | #include <linux/list.h> | |
23 | #include <linux/rcupdate.h> | |
24 | #include <linux/tracepoint.h> | |
25 | #include <linux/err.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/sched/signal.h> | |
28 | #include <linux/static_key.h> | |
29 | ||
30 | extern struct tracepoint * const __start___tracepoints_ptrs[]; | |
31 | extern struct tracepoint * const __stop___tracepoints_ptrs[]; | |
32 | ||
33 | /* Set to 1 to enable tracepoint debug output */ | |
34 | static const int tracepoint_debug; | |
35 | ||
36 | #ifdef CONFIG_MODULES | |
37 | /* | |
38 | * Tracepoint module list mutex protects the local module list. | |
39 | */ | |
40 | static DEFINE_MUTEX(tracepoint_module_list_mutex); | |
41 | ||
42 | /* Local list of struct tp_module */ | |
43 | static LIST_HEAD(tracepoint_module_list); | |
44 | #endif /* CONFIG_MODULES */ | |
45 | ||
46 | /* | |
47 | * tracepoints_mutex protects the builtin and module tracepoints. | |
48 | * tracepoints_mutex nests inside tracepoint_module_list_mutex. | |
49 | */ | |
50 | static DEFINE_MUTEX(tracepoints_mutex); | |
51 | ||
52 | /* | |
53 | * Note about RCU : | |
54 | * It is used to delay the free of multiple probes array until a quiescent | |
55 | * state is reached. | |
56 | */ | |
57 | struct tp_probes { | |
58 | struct rcu_head rcu; | |
59 | struct tracepoint_func probes[0]; | |
60 | }; | |
61 | ||
62 | static inline void *allocate_probes(int count) | |
63 | { | |
64 | struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func) | |
65 | + sizeof(struct tp_probes), GFP_KERNEL); | |
66 | return p == NULL ? NULL : p->probes; | |
67 | } | |
68 | ||
69 | static void rcu_free_old_probes(struct rcu_head *head) | |
70 | { | |
71 | kfree(container_of(head, struct tp_probes, rcu)); | |
72 | } | |
73 | ||
74 | static inline void release_probes(struct tracepoint_func *old) | |
75 | { | |
76 | if (old) { | |
77 | struct tp_probes *tp_probes = container_of(old, | |
78 | struct tp_probes, probes[0]); | |
79 | call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes); | |
80 | } | |
81 | } | |
82 | ||
83 | static void debug_print_probes(struct tracepoint_func *funcs) | |
84 | { | |
85 | int i; | |
86 | ||
87 | if (!tracepoint_debug || !funcs) | |
88 | return; | |
89 | ||
90 | for (i = 0; funcs[i].func; i++) | |
91 | printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func); | |
92 | } | |
93 | ||
94 | static struct tracepoint_func * | |
95 | func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func, | |
96 | int prio) | |
97 | { | |
98 | struct tracepoint_func *old, *new; | |
99 | int nr_probes = 0; | |
100 | int pos = -1; | |
101 | ||
102 | if (WARN_ON(!tp_func->func)) | |
103 | return ERR_PTR(-EINVAL); | |
104 | ||
105 | debug_print_probes(*funcs); | |
106 | old = *funcs; | |
107 | if (old) { | |
108 | /* (N -> N+1), (N != 0, 1) probes */ | |
109 | for (nr_probes = 0; old[nr_probes].func; nr_probes++) { | |
110 | /* Insert before probes of lower priority */ | |
111 | if (pos < 0 && old[nr_probes].prio < prio) | |
112 | pos = nr_probes; | |
113 | if (old[nr_probes].func == tp_func->func && | |
114 | old[nr_probes].data == tp_func->data) | |
115 | return ERR_PTR(-EEXIST); | |
116 | } | |
117 | } | |
118 | /* + 2 : one for new probe, one for NULL func */ | |
119 | new = allocate_probes(nr_probes + 2); | |
120 | if (new == NULL) | |
121 | return ERR_PTR(-ENOMEM); | |
122 | if (old) { | |
123 | if (pos < 0) { | |
124 | pos = nr_probes; | |
125 | memcpy(new, old, nr_probes * sizeof(struct tracepoint_func)); | |
126 | } else { | |
127 | /* Copy higher priority probes ahead of the new probe */ | |
128 | memcpy(new, old, pos * sizeof(struct tracepoint_func)); | |
129 | /* Copy the rest after it. */ | |
130 | memcpy(new + pos + 1, old + pos, | |
131 | (nr_probes - pos) * sizeof(struct tracepoint_func)); | |
132 | } | |
133 | } else | |
134 | pos = 0; | |
135 | new[pos] = *tp_func; | |
136 | new[nr_probes + 1].func = NULL; | |
137 | *funcs = new; | |
138 | debug_print_probes(*funcs); | |
139 | return old; | |
140 | } | |
141 | ||
142 | static void *func_remove(struct tracepoint_func **funcs, | |
143 | struct tracepoint_func *tp_func) | |
144 | { | |
145 | int nr_probes = 0, nr_del = 0, i; | |
146 | struct tracepoint_func *old, *new; | |
147 | ||
148 | old = *funcs; | |
149 | ||
150 | if (!old) | |
151 | return ERR_PTR(-ENOENT); | |
152 | ||
153 | debug_print_probes(*funcs); | |
154 | /* (N -> M), (N > 1, M >= 0) probes */ | |
155 | if (tp_func->func) { | |
156 | for (nr_probes = 0; old[nr_probes].func; nr_probes++) { | |
157 | if (old[nr_probes].func == tp_func->func && | |
158 | old[nr_probes].data == tp_func->data) | |
159 | nr_del++; | |
160 | } | |
161 | } | |
162 | ||
163 | /* | |
164 | * If probe is NULL, then nr_probes = nr_del = 0, and then the | |
165 | * entire entry will be removed. | |
166 | */ | |
167 | if (nr_probes - nr_del == 0) { | |
168 | /* N -> 0, (N > 1) */ | |
169 | *funcs = NULL; | |
170 | debug_print_probes(*funcs); | |
171 | return old; | |
172 | } else { | |
173 | int j = 0; | |
174 | /* N -> M, (N > 1, M > 0) */ | |
175 | /* + 1 for NULL */ | |
176 | new = allocate_probes(nr_probes - nr_del + 1); | |
177 | if (new == NULL) | |
178 | return ERR_PTR(-ENOMEM); | |
179 | for (i = 0; old[i].func; i++) | |
180 | if (old[i].func != tp_func->func | |
181 | || old[i].data != tp_func->data) | |
182 | new[j++] = old[i]; | |
183 | new[nr_probes - nr_del].func = NULL; | |
184 | *funcs = new; | |
185 | } | |
186 | debug_print_probes(*funcs); | |
187 | return old; | |
188 | } | |
189 | ||
190 | /* | |
191 | * Add the probe function to a tracepoint. | |
192 | */ | |
193 | static int tracepoint_add_func(struct tracepoint *tp, | |
194 | struct tracepoint_func *func, int prio) | |
195 | { | |
196 | struct tracepoint_func *old, *tp_funcs; | |
197 | int ret; | |
198 | ||
199 | if (tp->regfunc && !static_key_enabled(&tp->key)) { | |
200 | ret = tp->regfunc(); | |
201 | if (ret < 0) | |
202 | return ret; | |
203 | } | |
204 | ||
205 | tp_funcs = rcu_dereference_protected(tp->funcs, | |
206 | lockdep_is_held(&tracepoints_mutex)); | |
207 | old = func_add(&tp_funcs, func, prio); | |
208 | if (IS_ERR(old)) { | |
209 | WARN_ON_ONCE(1); | |
210 | return PTR_ERR(old); | |
211 | } | |
212 | ||
213 | /* | |
214 | * rcu_assign_pointer has a smp_wmb() which makes sure that the new | |
215 | * probe callbacks array is consistent before setting a pointer to it. | |
216 | * This array is referenced by __DO_TRACE from | |
217 | * include/linux/tracepoints.h. A matching smp_read_barrier_depends() | |
218 | * is used. | |
219 | */ | |
220 | rcu_assign_pointer(tp->funcs, tp_funcs); | |
221 | if (!static_key_enabled(&tp->key)) | |
222 | static_key_slow_inc(&tp->key); | |
223 | release_probes(old); | |
224 | return 0; | |
225 | } | |
226 | ||
227 | /* | |
228 | * Remove a probe function from a tracepoint. | |
229 | * Note: only waiting an RCU period after setting elem->call to the empty | |
230 | * function insures that the original callback is not used anymore. This insured | |
231 | * by preempt_disable around the call site. | |
232 | */ | |
233 | static int tracepoint_remove_func(struct tracepoint *tp, | |
234 | struct tracepoint_func *func) | |
235 | { | |
236 | struct tracepoint_func *old, *tp_funcs; | |
237 | ||
238 | tp_funcs = rcu_dereference_protected(tp->funcs, | |
239 | lockdep_is_held(&tracepoints_mutex)); | |
240 | old = func_remove(&tp_funcs, func); | |
241 | if (IS_ERR(old)) { | |
242 | WARN_ON_ONCE(1); | |
243 | return PTR_ERR(old); | |
244 | } | |
245 | ||
246 | if (!tp_funcs) { | |
247 | /* Removed last function */ | |
248 | if (tp->unregfunc && static_key_enabled(&tp->key)) | |
249 | tp->unregfunc(); | |
250 | ||
251 | if (static_key_enabled(&tp->key)) | |
252 | static_key_slow_dec(&tp->key); | |
253 | } | |
254 | rcu_assign_pointer(tp->funcs, tp_funcs); | |
255 | release_probes(old); | |
256 | return 0; | |
257 | } | |
258 | ||
259 | /** | |
260 | * tracepoint_probe_register - Connect a probe to a tracepoint | |
261 | * @tp: tracepoint | |
262 | * @probe: probe handler | |
263 | * @data: tracepoint data | |
264 | * @prio: priority of this function over other registered functions | |
265 | * | |
266 | * Returns 0 if ok, error value on error. | |
267 | * Note: if @tp is within a module, the caller is responsible for | |
268 | * unregistering the probe before the module is gone. This can be | |
269 | * performed either with a tracepoint module going notifier, or from | |
270 | * within module exit functions. | |
271 | */ | |
272 | int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, | |
273 | void *data, int prio) | |
274 | { | |
275 | struct tracepoint_func tp_func; | |
276 | int ret; | |
277 | ||
278 | mutex_lock(&tracepoints_mutex); | |
279 | tp_func.func = probe; | |
280 | tp_func.data = data; | |
281 | tp_func.prio = prio; | |
282 | ret = tracepoint_add_func(tp, &tp_func, prio); | |
283 | mutex_unlock(&tracepoints_mutex); | |
284 | return ret; | |
285 | } | |
286 | EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio); | |
287 | ||
288 | /** | |
289 | * tracepoint_probe_register - Connect a probe to a tracepoint | |
290 | * @tp: tracepoint | |
291 | * @probe: probe handler | |
292 | * @data: tracepoint data | |
293 | * @prio: priority of this function over other registered functions | |
294 | * | |
295 | * Returns 0 if ok, error value on error. | |
296 | * Note: if @tp is within a module, the caller is responsible for | |
297 | * unregistering the probe before the module is gone. This can be | |
298 | * performed either with a tracepoint module going notifier, or from | |
299 | * within module exit functions. | |
300 | */ | |
301 | int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data) | |
302 | { | |
303 | return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO); | |
304 | } | |
305 | EXPORT_SYMBOL_GPL(tracepoint_probe_register); | |
306 | ||
307 | /** | |
308 | * tracepoint_probe_unregister - Disconnect a probe from a tracepoint | |
309 | * @tp: tracepoint | |
310 | * @probe: probe function pointer | |
311 | * @data: tracepoint data | |
312 | * | |
313 | * Returns 0 if ok, error value on error. | |
314 | */ | |
315 | int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data) | |
316 | { | |
317 | struct tracepoint_func tp_func; | |
318 | int ret; | |
319 | ||
320 | mutex_lock(&tracepoints_mutex); | |
321 | tp_func.func = probe; | |
322 | tp_func.data = data; | |
323 | ret = tracepoint_remove_func(tp, &tp_func); | |
324 | mutex_unlock(&tracepoints_mutex); | |
325 | return ret; | |
326 | } | |
327 | EXPORT_SYMBOL_GPL(tracepoint_probe_unregister); | |
328 | ||
329 | #ifdef CONFIG_MODULES | |
330 | bool trace_module_has_bad_taint(struct module *mod) | |
331 | { | |
332 | return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) | | |
333 | (1 << TAINT_UNSIGNED_MODULE)); | |
334 | } | |
335 | ||
336 | static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list); | |
337 | ||
338 | /** | |
339 | * register_tracepoint_notifier - register tracepoint coming/going notifier | |
340 | * @nb: notifier block | |
341 | * | |
342 | * Notifiers registered with this function are called on module | |
343 | * coming/going with the tracepoint_module_list_mutex held. | |
344 | * The notifier block callback should expect a "struct tp_module" data | |
345 | * pointer. | |
346 | */ | |
347 | int register_tracepoint_module_notifier(struct notifier_block *nb) | |
348 | { | |
349 | struct tp_module *tp_mod; | |
350 | int ret; | |
351 | ||
352 | mutex_lock(&tracepoint_module_list_mutex); | |
353 | ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb); | |
354 | if (ret) | |
355 | goto end; | |
356 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) | |
357 | (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod); | |
358 | end: | |
359 | mutex_unlock(&tracepoint_module_list_mutex); | |
360 | return ret; | |
361 | } | |
362 | EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier); | |
363 | ||
364 | /** | |
365 | * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier | |
366 | * @nb: notifier block | |
367 | * | |
368 | * The notifier block callback should expect a "struct tp_module" data | |
369 | * pointer. | |
370 | */ | |
371 | int unregister_tracepoint_module_notifier(struct notifier_block *nb) | |
372 | { | |
373 | struct tp_module *tp_mod; | |
374 | int ret; | |
375 | ||
376 | mutex_lock(&tracepoint_module_list_mutex); | |
377 | ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb); | |
378 | if (ret) | |
379 | goto end; | |
380 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) | |
381 | (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod); | |
382 | end: | |
383 | mutex_unlock(&tracepoint_module_list_mutex); | |
384 | return ret; | |
385 | ||
386 | } | |
387 | EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier); | |
388 | ||
389 | /* | |
390 | * Ensure the tracer unregistered the module's probes before the module | |
391 | * teardown is performed. Prevents leaks of probe and data pointers. | |
392 | */ | |
393 | static void tp_module_going_check_quiescent(struct tracepoint * const *begin, | |
394 | struct tracepoint * const *end) | |
395 | { | |
396 | struct tracepoint * const *iter; | |
397 | ||
398 | if (!begin) | |
399 | return; | |
400 | for (iter = begin; iter < end; iter++) | |
401 | WARN_ON_ONCE((*iter)->funcs); | |
402 | } | |
403 | ||
404 | static int tracepoint_module_coming(struct module *mod) | |
405 | { | |
406 | struct tp_module *tp_mod; | |
407 | int ret = 0; | |
408 | ||
409 | if (!mod->num_tracepoints) | |
410 | return 0; | |
411 | ||
412 | /* | |
413 | * We skip modules that taint the kernel, especially those with different | |
414 | * module headers (for forced load), to make sure we don't cause a crash. | |
415 | * Staging, out-of-tree, and unsigned GPL modules are fine. | |
416 | */ | |
417 | if (trace_module_has_bad_taint(mod)) | |
418 | return 0; | |
419 | mutex_lock(&tracepoint_module_list_mutex); | |
420 | tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL); | |
421 | if (!tp_mod) { | |
422 | ret = -ENOMEM; | |
423 | goto end; | |
424 | } | |
425 | tp_mod->mod = mod; | |
426 | list_add_tail(&tp_mod->list, &tracepoint_module_list); | |
427 | blocking_notifier_call_chain(&tracepoint_notify_list, | |
428 | MODULE_STATE_COMING, tp_mod); | |
429 | end: | |
430 | mutex_unlock(&tracepoint_module_list_mutex); | |
431 | return ret; | |
432 | } | |
433 | ||
434 | static void tracepoint_module_going(struct module *mod) | |
435 | { | |
436 | struct tp_module *tp_mod; | |
437 | ||
438 | if (!mod->num_tracepoints) | |
439 | return; | |
440 | ||
441 | mutex_lock(&tracepoint_module_list_mutex); | |
442 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) { | |
443 | if (tp_mod->mod == mod) { | |
444 | blocking_notifier_call_chain(&tracepoint_notify_list, | |
445 | MODULE_STATE_GOING, tp_mod); | |
446 | list_del(&tp_mod->list); | |
447 | kfree(tp_mod); | |
448 | /* | |
449 | * Called the going notifier before checking for | |
450 | * quiescence. | |
451 | */ | |
452 | tp_module_going_check_quiescent(mod->tracepoints_ptrs, | |
453 | mod->tracepoints_ptrs + mod->num_tracepoints); | |
454 | break; | |
455 | } | |
456 | } | |
457 | /* | |
458 | * In the case of modules that were tainted at "coming", we'll simply | |
459 | * walk through the list without finding it. We cannot use the "tainted" | |
460 | * flag on "going", in case a module taints the kernel only after being | |
461 | * loaded. | |
462 | */ | |
463 | mutex_unlock(&tracepoint_module_list_mutex); | |
464 | } | |
465 | ||
466 | static int tracepoint_module_notify(struct notifier_block *self, | |
467 | unsigned long val, void *data) | |
468 | { | |
469 | struct module *mod = data; | |
470 | int ret = 0; | |
471 | ||
472 | switch (val) { | |
473 | case MODULE_STATE_COMING: | |
474 | ret = tracepoint_module_coming(mod); | |
475 | break; | |
476 | case MODULE_STATE_LIVE: | |
477 | break; | |
478 | case MODULE_STATE_GOING: | |
479 | tracepoint_module_going(mod); | |
480 | break; | |
481 | case MODULE_STATE_UNFORMED: | |
482 | break; | |
483 | } | |
484 | return ret; | |
485 | } | |
486 | ||
487 | static struct notifier_block tracepoint_module_nb = { | |
488 | .notifier_call = tracepoint_module_notify, | |
489 | .priority = 0, | |
490 | }; | |
491 | ||
492 | static __init int init_tracepoints(void) | |
493 | { | |
494 | int ret; | |
495 | ||
496 | ret = register_module_notifier(&tracepoint_module_nb); | |
497 | if (ret) | |
498 | pr_warn("Failed to register tracepoint module enter notifier\n"); | |
499 | ||
500 | return ret; | |
501 | } | |
502 | __initcall(init_tracepoints); | |
503 | #endif /* CONFIG_MODULES */ | |
504 | ||
505 | static void for_each_tracepoint_range(struct tracepoint * const *begin, | |
506 | struct tracepoint * const *end, | |
507 | void (*fct)(struct tracepoint *tp, void *priv), | |
508 | void *priv) | |
509 | { | |
510 | struct tracepoint * const *iter; | |
511 | ||
512 | if (!begin) | |
513 | return; | |
514 | for (iter = begin; iter < end; iter++) | |
515 | fct(*iter, priv); | |
516 | } | |
517 | ||
518 | /** | |
519 | * for_each_kernel_tracepoint - iteration on all kernel tracepoints | |
520 | * @fct: callback | |
521 | * @priv: private data | |
522 | */ | |
523 | void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), | |
524 | void *priv) | |
525 | { | |
526 | for_each_tracepoint_range(__start___tracepoints_ptrs, | |
527 | __stop___tracepoints_ptrs, fct, priv); | |
528 | } | |
529 | EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint); | |
530 | ||
531 | #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS | |
532 | ||
533 | /* NB: reg/unreg are called while guarded with the tracepoints_mutex */ | |
534 | static int sys_tracepoint_refcount; | |
535 | ||
536 | int syscall_regfunc(void) | |
537 | { | |
538 | struct task_struct *p, *t; | |
539 | ||
540 | if (!sys_tracepoint_refcount) { | |
541 | read_lock(&tasklist_lock); | |
542 | for_each_process_thread(p, t) { | |
543 | set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT); | |
544 | } | |
545 | read_unlock(&tasklist_lock); | |
546 | } | |
547 | sys_tracepoint_refcount++; | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | void syscall_unregfunc(void) | |
553 | { | |
554 | struct task_struct *p, *t; | |
555 | ||
556 | sys_tracepoint_refcount--; | |
557 | if (!sys_tracepoint_refcount) { | |
558 | read_lock(&tasklist_lock); | |
559 | for_each_process_thread(p, t) { | |
560 | clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT); | |
561 | } | |
562 | read_unlock(&tasklist_lock); | |
563 | } | |
564 | } | |
565 | #endif |