Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2008-2014 Mathieu Desnoyers
4 */
5#include <linux/module.h>
6#include <linux/mutex.h>
7#include <linux/types.h>
8#include <linux/jhash.h>
9#include <linux/list.h>
10#include <linux/rcupdate.h>
11#include <linux/tracepoint.h>
12#include <linux/err.h>
13#include <linux/slab.h>
14#include <linux/sched/signal.h>
15#include <linux/sched/task.h>
16#include <linux/static_key.h>
17
18enum tp_func_state {
19 TP_FUNC_0,
20 TP_FUNC_1,
21 TP_FUNC_2,
22 TP_FUNC_N,
23};
24
25extern tracepoint_ptr_t __start___tracepoints_ptrs[];
26extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
27
28enum tp_transition_sync {
29 TP_TRANSITION_SYNC_1_0_1,
30 TP_TRANSITION_SYNC_N_2_1,
31
32 _NR_TP_TRANSITION_SYNC,
33};
34
35struct tp_transition_snapshot {
36 unsigned long rcu;
37 bool ongoing;
38};
39
40/* Protected by tracepoints_mutex */
41static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC];
42
43static void tp_rcu_get_state(enum tp_transition_sync sync)
44{
45 struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
46
47 /* Keep the latest get_state snapshot. */
48 snapshot->rcu = get_state_synchronize_rcu();
49 snapshot->ongoing = true;
50}
51
52static void tp_rcu_cond_sync(enum tp_transition_sync sync)
53{
54 struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
55
56 if (!snapshot->ongoing)
57 return;
58 cond_synchronize_rcu(snapshot->rcu);
59 snapshot->ongoing = false;
60}
61
62/* Set to 1 to enable tracepoint debug output */
63static const int tracepoint_debug;
64
65#ifdef CONFIG_MODULES
66/*
67 * Tracepoint module list mutex protects the local module list.
68 */
69static DEFINE_MUTEX(tracepoint_module_list_mutex);
70
71/* Local list of struct tp_module */
72static LIST_HEAD(tracepoint_module_list);
73#endif /* CONFIG_MODULES */
74
75/*
76 * tracepoints_mutex protects the builtin and module tracepoints.
77 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
78 */
79static DEFINE_MUTEX(tracepoints_mutex);
80
81/*
82 * Note about RCU :
83 * It is used to delay the free of multiple probes array until a quiescent
84 * state is reached.
85 */
86struct tp_probes {
87 struct rcu_head rcu;
88 struct tracepoint_func probes[];
89};
90
91/* Called in removal of a func but failed to allocate a new tp_funcs */
92static void tp_stub_func(void)
93{
94 return;
95}
96
97static inline void *allocate_probes(int count)
98{
99 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
100 GFP_KERNEL);
101 return p == NULL ? NULL : p->probes;
102}
103
104static void rcu_free_old_probes(struct rcu_head *head)
105{
106 kfree(container_of(head, struct tp_probes, rcu));
107}
108
109static inline void release_probes(struct tracepoint *tp, struct tracepoint_func *old)
110{
111 if (old) {
112 struct tp_probes *tp_probes = container_of(old,
113 struct tp_probes, probes[0]);
114
115 if (tracepoint_is_faultable(tp))
116 call_rcu_tasks_trace(&tp_probes->rcu, rcu_free_old_probes);
117 else
118 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
119 }
120}
121
122static void debug_print_probes(struct tracepoint_func *funcs)
123{
124 int i;
125
126 if (!tracepoint_debug || !funcs)
127 return;
128
129 for (i = 0; funcs[i].func; i++)
130 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
131}
132
133static struct tracepoint_func *
134func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
135 int prio)
136{
137 struct tracepoint_func *old, *new;
138 int iter_probes; /* Iterate over old probe array. */
139 int nr_probes = 0; /* Counter for probes */
140 int pos = -1; /* Insertion position into new array */
141
142 if (WARN_ON(!tp_func->func))
143 return ERR_PTR(-EINVAL);
144
145 debug_print_probes(*funcs);
146 old = *funcs;
147 if (old) {
148 /* (N -> N+1), (N != 0, 1) probes */
149 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
150 if (old[iter_probes].func == tp_stub_func)
151 continue; /* Skip stub functions. */
152 if (old[iter_probes].func == tp_func->func &&
153 old[iter_probes].data == tp_func->data)
154 return ERR_PTR(-EEXIST);
155 nr_probes++;
156 }
157 }
158 /* + 2 : one for new probe, one for NULL func */
159 new = allocate_probes(nr_probes + 2);
160 if (new == NULL)
161 return ERR_PTR(-ENOMEM);
162 if (old) {
163 nr_probes = 0;
164 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
165 if (old[iter_probes].func == tp_stub_func)
166 continue;
167 /* Insert before probes of lower priority */
168 if (pos < 0 && old[iter_probes].prio < prio)
169 pos = nr_probes++;
170 new[nr_probes++] = old[iter_probes];
171 }
172 if (pos < 0)
173 pos = nr_probes++;
174 /* nr_probes now points to the end of the new array */
175 } else {
176 pos = 0;
177 nr_probes = 1; /* must point at end of array */
178 }
179 new[pos] = *tp_func;
180 new[nr_probes].func = NULL;
181 *funcs = new;
182 debug_print_probes(*funcs);
183 return old;
184}
185
186static void *func_remove(struct tracepoint_func **funcs,
187 struct tracepoint_func *tp_func)
188{
189 int nr_probes = 0, nr_del = 0, i;
190 struct tracepoint_func *old, *new;
191
192 old = *funcs;
193
194 if (!old)
195 return ERR_PTR(-ENOENT);
196
197 debug_print_probes(*funcs);
198 /* (N -> M), (N > 1, M >= 0) probes */
199 if (tp_func->func) {
200 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
201 if ((old[nr_probes].func == tp_func->func &&
202 old[nr_probes].data == tp_func->data) ||
203 old[nr_probes].func == tp_stub_func)
204 nr_del++;
205 }
206 }
207
208 /*
209 * If probe is NULL, then nr_probes = nr_del = 0, and then the
210 * entire entry will be removed.
211 */
212 if (nr_probes - nr_del == 0) {
213 /* N -> 0, (N > 1) */
214 *funcs = NULL;
215 debug_print_probes(*funcs);
216 return old;
217 } else {
218 int j = 0;
219 /* N -> M, (N > 1, M > 0) */
220 /* + 1 for NULL */
221 new = allocate_probes(nr_probes - nr_del + 1);
222 if (new) {
223 for (i = 0; old[i].func; i++) {
224 if ((old[i].func != tp_func->func ||
225 old[i].data != tp_func->data) &&
226 old[i].func != tp_stub_func)
227 new[j++] = old[i];
228 }
229 new[nr_probes - nr_del].func = NULL;
230 *funcs = new;
231 } else {
232 /*
233 * Failed to allocate, replace the old function
234 * with calls to tp_stub_func.
235 */
236 for (i = 0; old[i].func; i++) {
237 if (old[i].func == tp_func->func &&
238 old[i].data == tp_func->data)
239 WRITE_ONCE(old[i].func, tp_stub_func);
240 }
241 *funcs = old;
242 }
243 }
244 debug_print_probes(*funcs);
245 return old;
246}
247
248/*
249 * Count the number of functions (enum tp_func_state) in a tp_funcs array.
250 */
251static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
252{
253 if (!tp_funcs)
254 return TP_FUNC_0;
255 if (!tp_funcs[1].func)
256 return TP_FUNC_1;
257 if (!tp_funcs[2].func)
258 return TP_FUNC_2;
259 return TP_FUNC_N; /* 3 or more */
260}
261
262static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
263{
264 void *func = tp->iterator;
265
266 /* Synthetic events do not have static call sites */
267 if (!tp->static_call_key)
268 return;
269 if (nr_func_state(tp_funcs) == TP_FUNC_1)
270 func = tp_funcs[0].func;
271 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
272}
273
274/*
275 * Add the probe function to a tracepoint.
276 */
277static int tracepoint_add_func(struct tracepoint *tp,
278 struct tracepoint_func *func, int prio,
279 bool warn)
280{
281 struct tracepoint_func *old, *tp_funcs;
282 int ret;
283
284 if (tp->ext && tp->ext->regfunc && !static_key_enabled(&tp->key)) {
285 ret = tp->ext->regfunc();
286 if (ret < 0)
287 return ret;
288 }
289
290 tp_funcs = rcu_dereference_protected(tp->funcs,
291 lockdep_is_held(&tracepoints_mutex));
292 old = func_add(&tp_funcs, func, prio);
293 if (IS_ERR(old)) {
294 WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
295 return PTR_ERR(old);
296 }
297
298 /*
299 * rcu_assign_pointer has as smp_store_release() which makes sure
300 * that the new probe callbacks array is consistent before setting
301 * a pointer to it. This array is referenced by __DO_TRACE from
302 * include/linux/tracepoint.h using rcu_dereference_sched().
303 */
304 switch (nr_func_state(tp_funcs)) {
305 case TP_FUNC_1: /* 0->1 */
306 /*
307 * Make sure new static func never uses old data after a
308 * 1->0->1 transition sequence.
309 */
310 tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
311 /* Set static call to first function */
312 tracepoint_update_call(tp, tp_funcs);
313 /* Both iterator and static call handle NULL tp->funcs */
314 rcu_assign_pointer(tp->funcs, tp_funcs);
315 static_branch_enable(&tp->key);
316 break;
317 case TP_FUNC_2: /* 1->2 */
318 /* Set iterator static call */
319 tracepoint_update_call(tp, tp_funcs);
320 /*
321 * Iterator callback installed before updating tp->funcs.
322 * Requires ordering between RCU assign/dereference and
323 * static call update/call.
324 */
325 fallthrough;
326 case TP_FUNC_N: /* N->N+1 (N>1) */
327 rcu_assign_pointer(tp->funcs, tp_funcs);
328 /*
329 * Make sure static func never uses incorrect data after a
330 * N->...->2->1 (N>1) transition sequence.
331 */
332 if (tp_funcs[0].data != old[0].data)
333 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
334 break;
335 default:
336 WARN_ON_ONCE(1);
337 break;
338 }
339
340 release_probes(tp, old);
341 return 0;
342}
343
344/*
345 * Remove a probe function from a tracepoint.
346 * Note: only waiting an RCU period after setting elem->call to the empty
347 * function insures that the original callback is not used anymore. This insured
348 * by preempt_disable around the call site.
349 */
350static int tracepoint_remove_func(struct tracepoint *tp,
351 struct tracepoint_func *func)
352{
353 struct tracepoint_func *old, *tp_funcs;
354
355 tp_funcs = rcu_dereference_protected(tp->funcs,
356 lockdep_is_held(&tracepoints_mutex));
357 old = func_remove(&tp_funcs, func);
358 if (WARN_ON_ONCE(IS_ERR(old)))
359 return PTR_ERR(old);
360
361 if (tp_funcs == old)
362 /* Failed allocating new tp_funcs, replaced func with stub */
363 return 0;
364
365 switch (nr_func_state(tp_funcs)) {
366 case TP_FUNC_0: /* 1->0 */
367 /* Removed last function */
368 if (tp->ext && tp->ext->unregfunc && static_key_enabled(&tp->key))
369 tp->ext->unregfunc();
370 static_branch_disable(&tp->key);
371 /* Set iterator static call */
372 tracepoint_update_call(tp, tp_funcs);
373 /* Both iterator and static call handle NULL tp->funcs */
374 rcu_assign_pointer(tp->funcs, NULL);
375 /*
376 * Make sure new static func never uses old data after a
377 * 1->0->1 transition sequence.
378 */
379 tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
380 break;
381 case TP_FUNC_1: /* 2->1 */
382 rcu_assign_pointer(tp->funcs, tp_funcs);
383 /*
384 * Make sure static func never uses incorrect data after a
385 * N->...->2->1 (N>2) transition sequence. If the first
386 * element's data has changed, then force the synchronization
387 * to prevent current readers that have loaded the old data
388 * from calling the new function.
389 */
390 if (tp_funcs[0].data != old[0].data)
391 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
392 tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
393 /* Set static call to first function */
394 tracepoint_update_call(tp, tp_funcs);
395 break;
396 case TP_FUNC_2: /* N->N-1 (N>2) */
397 fallthrough;
398 case TP_FUNC_N:
399 rcu_assign_pointer(tp->funcs, tp_funcs);
400 /*
401 * Make sure static func never uses incorrect data after a
402 * N->...->2->1 (N>2) transition sequence.
403 */
404 if (tp_funcs[0].data != old[0].data)
405 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
406 break;
407 default:
408 WARN_ON_ONCE(1);
409 break;
410 }
411 release_probes(tp, old);
412 return 0;
413}
414
415/**
416 * tracepoint_probe_register_prio_may_exist - Connect a probe to a tracepoint with priority
417 * @tp: tracepoint
418 * @probe: probe handler
419 * @data: tracepoint data
420 * @prio: priority of this function over other registered functions
421 *
422 * Same as tracepoint_probe_register_prio() except that it will not warn
423 * if the tracepoint is already registered.
424 */
425int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
426 void *data, int prio)
427{
428 struct tracepoint_func tp_func;
429 int ret;
430
431 mutex_lock(&tracepoints_mutex);
432 tp_func.func = probe;
433 tp_func.data = data;
434 tp_func.prio = prio;
435 ret = tracepoint_add_func(tp, &tp_func, prio, false);
436 mutex_unlock(&tracepoints_mutex);
437 return ret;
438}
439EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
440
441/**
442 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
443 * @tp: tracepoint
444 * @probe: probe handler
445 * @data: tracepoint data
446 * @prio: priority of this function over other registered functions
447 *
448 * Returns 0 if ok, error value on error.
449 * Note: if @tp is within a module, the caller is responsible for
450 * unregistering the probe before the module is gone. This can be
451 * performed either with a tracepoint module going notifier, or from
452 * within module exit functions.
453 */
454int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
455 void *data, int prio)
456{
457 struct tracepoint_func tp_func;
458 int ret;
459
460 mutex_lock(&tracepoints_mutex);
461 tp_func.func = probe;
462 tp_func.data = data;
463 tp_func.prio = prio;
464 ret = tracepoint_add_func(tp, &tp_func, prio, true);
465 mutex_unlock(&tracepoints_mutex);
466 return ret;
467}
468EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
469
470/**
471 * tracepoint_probe_register - Connect a probe to a tracepoint
472 * @tp: tracepoint
473 * @probe: probe handler
474 * @data: tracepoint data
475 *
476 * Returns 0 if ok, error value on error.
477 * Note: if @tp is within a module, the caller is responsible for
478 * unregistering the probe before the module is gone. This can be
479 * performed either with a tracepoint module going notifier, or from
480 * within module exit functions.
481 */
482int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
483{
484 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
485}
486EXPORT_SYMBOL_GPL(tracepoint_probe_register);
487
488/**
489 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
490 * @tp: tracepoint
491 * @probe: probe function pointer
492 * @data: tracepoint data
493 *
494 * Returns 0 if ok, error value on error.
495 */
496int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
497{
498 struct tracepoint_func tp_func;
499 int ret;
500
501 mutex_lock(&tracepoints_mutex);
502 tp_func.func = probe;
503 tp_func.data = data;
504 ret = tracepoint_remove_func(tp, &tp_func);
505 mutex_unlock(&tracepoints_mutex);
506 return ret;
507}
508EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
509
510static void for_each_tracepoint_range(
511 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
512 void (*fct)(struct tracepoint *tp, void *priv),
513 void *priv)
514{
515 tracepoint_ptr_t *iter;
516
517 if (!begin)
518 return;
519 for (iter = begin; iter < end; iter++)
520 fct(tracepoint_ptr_deref(iter), priv);
521}
522
523#ifdef CONFIG_MODULES
524bool trace_module_has_bad_taint(struct module *mod)
525{
526 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
527 (1 << TAINT_UNSIGNED_MODULE) | (1 << TAINT_TEST) |
528 (1 << TAINT_LIVEPATCH));
529}
530
531static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
532
533/**
534 * register_tracepoint_module_notifier - register tracepoint coming/going notifier
535 * @nb: notifier block
536 *
537 * Notifiers registered with this function are called on module
538 * coming/going with the tracepoint_module_list_mutex held.
539 * The notifier block callback should expect a "struct tp_module" data
540 * pointer.
541 */
542int register_tracepoint_module_notifier(struct notifier_block *nb)
543{
544 struct tp_module *tp_mod;
545 int ret;
546
547 mutex_lock(&tracepoint_module_list_mutex);
548 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
549 if (ret)
550 goto end;
551 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
552 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
553end:
554 mutex_unlock(&tracepoint_module_list_mutex);
555 return ret;
556}
557EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
558
559/**
560 * unregister_tracepoint_module_notifier - unregister tracepoint coming/going notifier
561 * @nb: notifier block
562 *
563 * The notifier block callback should expect a "struct tp_module" data
564 * pointer.
565 */
566int unregister_tracepoint_module_notifier(struct notifier_block *nb)
567{
568 struct tp_module *tp_mod;
569 int ret;
570
571 mutex_lock(&tracepoint_module_list_mutex);
572 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
573 if (ret)
574 goto end;
575 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
576 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
577end:
578 mutex_unlock(&tracepoint_module_list_mutex);
579 return ret;
580
581}
582EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
583
584/*
585 * Ensure the tracer unregistered the module's probes before the module
586 * teardown is performed. Prevents leaks of probe and data pointers.
587 */
588static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
589{
590 WARN_ON_ONCE(tp->funcs);
591}
592
593static int tracepoint_module_coming(struct module *mod)
594{
595 struct tp_module *tp_mod;
596
597 if (!mod->num_tracepoints)
598 return 0;
599
600 /*
601 * We skip modules that taint the kernel, especially those with different
602 * module headers (for forced load), to make sure we don't cause a crash.
603 * Staging, out-of-tree, unsigned GPL, and test modules are fine.
604 */
605 if (trace_module_has_bad_taint(mod))
606 return 0;
607
608 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
609 if (!tp_mod)
610 return -ENOMEM;
611 tp_mod->mod = mod;
612
613 mutex_lock(&tracepoint_module_list_mutex);
614 list_add_tail(&tp_mod->list, &tracepoint_module_list);
615 blocking_notifier_call_chain(&tracepoint_notify_list,
616 MODULE_STATE_COMING, tp_mod);
617 mutex_unlock(&tracepoint_module_list_mutex);
618 return 0;
619}
620
621static void tracepoint_module_going(struct module *mod)
622{
623 struct tp_module *tp_mod;
624
625 if (!mod->num_tracepoints)
626 return;
627
628 mutex_lock(&tracepoint_module_list_mutex);
629 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
630 if (tp_mod->mod == mod) {
631 blocking_notifier_call_chain(&tracepoint_notify_list,
632 MODULE_STATE_GOING, tp_mod);
633 list_del(&tp_mod->list);
634 kfree(tp_mod);
635 /*
636 * Called the going notifier before checking for
637 * quiescence.
638 */
639 for_each_tracepoint_range(mod->tracepoints_ptrs,
640 mod->tracepoints_ptrs + mod->num_tracepoints,
641 tp_module_going_check_quiescent, NULL);
642 break;
643 }
644 }
645 /*
646 * In the case of modules that were tainted at "coming", we'll simply
647 * walk through the list without finding it. We cannot use the "tainted"
648 * flag on "going", in case a module taints the kernel only after being
649 * loaded.
650 */
651 mutex_unlock(&tracepoint_module_list_mutex);
652}
653
654static int tracepoint_module_notify(struct notifier_block *self,
655 unsigned long val, void *data)
656{
657 struct module *mod = data;
658 int ret = 0;
659
660 switch (val) {
661 case MODULE_STATE_COMING:
662 ret = tracepoint_module_coming(mod);
663 break;
664 case MODULE_STATE_LIVE:
665 break;
666 case MODULE_STATE_GOING:
667 tracepoint_module_going(mod);
668 break;
669 case MODULE_STATE_UNFORMED:
670 break;
671 }
672 return notifier_from_errno(ret);
673}
674
675static struct notifier_block tracepoint_module_nb = {
676 .notifier_call = tracepoint_module_notify,
677 .priority = 0,
678};
679
680static __init int init_tracepoints(void)
681{
682 int ret;
683
684 ret = register_module_notifier(&tracepoint_module_nb);
685 if (ret)
686 pr_warn("Failed to register tracepoint module enter notifier\n");
687
688 return ret;
689}
690__initcall(init_tracepoints);
691
692/**
693 * for_each_tracepoint_in_module - iteration on all tracepoints in a module
694 * @mod: module
695 * @fct: callback
696 * @priv: private data
697 */
698void for_each_tracepoint_in_module(struct module *mod,
699 void (*fct)(struct tracepoint *tp,
700 struct module *mod, void *priv),
701 void *priv)
702{
703 tracepoint_ptr_t *begin, *end, *iter;
704
705 lockdep_assert_held(&tracepoint_module_list_mutex);
706
707 if (!mod)
708 return;
709
710 begin = mod->tracepoints_ptrs;
711 end = mod->tracepoints_ptrs + mod->num_tracepoints;
712
713 for (iter = begin; iter < end; iter++)
714 fct(tracepoint_ptr_deref(iter), mod, priv);
715}
716
717/**
718 * for_each_module_tracepoint - iteration on all tracepoints in all modules
719 * @fct: callback
720 * @priv: private data
721 */
722void for_each_module_tracepoint(void (*fct)(struct tracepoint *tp,
723 struct module *mod, void *priv),
724 void *priv)
725{
726 struct tp_module *tp_mod;
727
728 mutex_lock(&tracepoint_module_list_mutex);
729 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
730 for_each_tracepoint_in_module(tp_mod->mod, fct, priv);
731 mutex_unlock(&tracepoint_module_list_mutex);
732}
733#endif /* CONFIG_MODULES */
734
735/**
736 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
737 * @fct: callback
738 * @priv: private data
739 */
740void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
741 void *priv)
742{
743 for_each_tracepoint_range(__start___tracepoints_ptrs,
744 __stop___tracepoints_ptrs, fct, priv);
745}
746EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
747
748#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
749
750/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
751static int sys_tracepoint_refcount;
752
753int syscall_regfunc(void)
754{
755 struct task_struct *p, *t;
756
757 if (!sys_tracepoint_refcount) {
758 read_lock(&tasklist_lock);
759 for_each_process_thread(p, t) {
760 set_task_syscall_work(t, SYSCALL_TRACEPOINT);
761 }
762 read_unlock(&tasklist_lock);
763 }
764 sys_tracepoint_refcount++;
765
766 return 0;
767}
768
769void syscall_unregfunc(void)
770{
771 struct task_struct *p, *t;
772
773 sys_tracepoint_refcount--;
774 if (!sys_tracepoint_refcount) {
775 read_lock(&tasklist_lock);
776 for_each_process_thread(p, t) {
777 clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
778 }
779 read_unlock(&tasklist_lock);
780 }
781}
782#endif
1/*
2 * Copyright (C) 2008 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.h>
28#include <linux/jump_label.h>
29
30extern struct tracepoint * const __start___tracepoints_ptrs[];
31extern struct tracepoint * const __stop___tracepoints_ptrs[];
32
33/* Set to 1 to enable tracepoint debug output */
34static const int tracepoint_debug;
35
36/*
37 * tracepoints_mutex nests inside module_mutex. Tracepoints mutex protects the
38 * builtin and module tracepoints and the hash table.
39 */
40static DEFINE_MUTEX(tracepoints_mutex);
41
42/*
43 * Tracepoint hash table, containing the active tracepoints.
44 * Protected by tracepoints_mutex.
45 */
46#define TRACEPOINT_HASH_BITS 6
47#define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
48static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
49
50/*
51 * Note about RCU :
52 * It is used to delay the free of multiple probes array until a quiescent
53 * state is reached.
54 * Tracepoint entries modifications are protected by the tracepoints_mutex.
55 */
56struct tracepoint_entry {
57 struct hlist_node hlist;
58 struct tracepoint_func *funcs;
59 int refcount; /* Number of times armed. 0 if disarmed. */
60 char name[0];
61};
62
63struct tp_probes {
64 union {
65 struct rcu_head rcu;
66 struct list_head list;
67 } u;
68 struct tracepoint_func probes[0];
69};
70
71static inline void *allocate_probes(int count)
72{
73 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
74 + sizeof(struct tp_probes), GFP_KERNEL);
75 return p == NULL ? NULL : p->probes;
76}
77
78static void rcu_free_old_probes(struct rcu_head *head)
79{
80 kfree(container_of(head, struct tp_probes, u.rcu));
81}
82
83static inline void release_probes(struct tracepoint_func *old)
84{
85 if (old) {
86 struct tp_probes *tp_probes = container_of(old,
87 struct tp_probes, probes[0]);
88 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
89 }
90}
91
92static void debug_print_probes(struct tracepoint_entry *entry)
93{
94 int i;
95
96 if (!tracepoint_debug || !entry->funcs)
97 return;
98
99 for (i = 0; entry->funcs[i].func; i++)
100 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
101}
102
103static struct tracepoint_func *
104tracepoint_entry_add_probe(struct tracepoint_entry *entry,
105 void *probe, void *data)
106{
107 int nr_probes = 0;
108 struct tracepoint_func *old, *new;
109
110 WARN_ON(!probe);
111
112 debug_print_probes(entry);
113 old = entry->funcs;
114 if (old) {
115 /* (N -> N+1), (N != 0, 1) probes */
116 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
117 if (old[nr_probes].func == probe &&
118 old[nr_probes].data == data)
119 return ERR_PTR(-EEXIST);
120 }
121 /* + 2 : one for new probe, one for NULL func */
122 new = allocate_probes(nr_probes + 2);
123 if (new == NULL)
124 return ERR_PTR(-ENOMEM);
125 if (old)
126 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
127 new[nr_probes].func = probe;
128 new[nr_probes].data = data;
129 new[nr_probes + 1].func = NULL;
130 entry->refcount = nr_probes + 1;
131 entry->funcs = new;
132 debug_print_probes(entry);
133 return old;
134}
135
136static void *
137tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
138 void *probe, void *data)
139{
140 int nr_probes = 0, nr_del = 0, i;
141 struct tracepoint_func *old, *new;
142
143 old = entry->funcs;
144
145 if (!old)
146 return ERR_PTR(-ENOENT);
147
148 debug_print_probes(entry);
149 /* (N -> M), (N > 1, M >= 0) probes */
150 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
151 if (!probe ||
152 (old[nr_probes].func == probe &&
153 old[nr_probes].data == data))
154 nr_del++;
155 }
156
157 if (nr_probes - nr_del == 0) {
158 /* N -> 0, (N > 1) */
159 entry->funcs = NULL;
160 entry->refcount = 0;
161 debug_print_probes(entry);
162 return old;
163 } else {
164 int j = 0;
165 /* N -> M, (N > 1, M > 0) */
166 /* + 1 for NULL */
167 new = allocate_probes(nr_probes - nr_del + 1);
168 if (new == NULL)
169 return ERR_PTR(-ENOMEM);
170 for (i = 0; old[i].func; i++)
171 if (probe &&
172 (old[i].func != probe || old[i].data != data))
173 new[j++] = old[i];
174 new[nr_probes - nr_del].func = NULL;
175 entry->refcount = nr_probes - nr_del;
176 entry->funcs = new;
177 }
178 debug_print_probes(entry);
179 return old;
180}
181
182/*
183 * Get tracepoint if the tracepoint is present in the tracepoint hash table.
184 * Must be called with tracepoints_mutex held.
185 * Returns NULL if not present.
186 */
187static struct tracepoint_entry *get_tracepoint(const char *name)
188{
189 struct hlist_head *head;
190 struct hlist_node *node;
191 struct tracepoint_entry *e;
192 u32 hash = jhash(name, strlen(name), 0);
193
194 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
195 hlist_for_each_entry(e, node, head, hlist) {
196 if (!strcmp(name, e->name))
197 return e;
198 }
199 return NULL;
200}
201
202/*
203 * Add the tracepoint to the tracepoint hash table. Must be called with
204 * tracepoints_mutex held.
205 */
206static struct tracepoint_entry *add_tracepoint(const char *name)
207{
208 struct hlist_head *head;
209 struct hlist_node *node;
210 struct tracepoint_entry *e;
211 size_t name_len = strlen(name) + 1;
212 u32 hash = jhash(name, name_len-1, 0);
213
214 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
215 hlist_for_each_entry(e, node, head, hlist) {
216 if (!strcmp(name, e->name)) {
217 printk(KERN_NOTICE
218 "tracepoint %s busy\n", name);
219 return ERR_PTR(-EEXIST); /* Already there */
220 }
221 }
222 /*
223 * Using kmalloc here to allocate a variable length element. Could
224 * cause some memory fragmentation if overused.
225 */
226 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
227 if (!e)
228 return ERR_PTR(-ENOMEM);
229 memcpy(&e->name[0], name, name_len);
230 e->funcs = NULL;
231 e->refcount = 0;
232 hlist_add_head(&e->hlist, head);
233 return e;
234}
235
236/*
237 * Remove the tracepoint from the tracepoint hash table. Must be called with
238 * mutex_lock held.
239 */
240static inline void remove_tracepoint(struct tracepoint_entry *e)
241{
242 hlist_del(&e->hlist);
243 kfree(e);
244}
245
246/*
247 * Sets the probe callback corresponding to one tracepoint.
248 */
249static void set_tracepoint(struct tracepoint_entry **entry,
250 struct tracepoint *elem, int active)
251{
252 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
253
254 if (elem->regfunc && !jump_label_enabled(&elem->key) && active)
255 elem->regfunc();
256 else if (elem->unregfunc && jump_label_enabled(&elem->key) && !active)
257 elem->unregfunc();
258
259 /*
260 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
261 * probe callbacks array is consistent before setting a pointer to it.
262 * This array is referenced by __DO_TRACE from
263 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
264 * is used.
265 */
266 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
267 if (active && !jump_label_enabled(&elem->key))
268 jump_label_inc(&elem->key);
269 else if (!active && jump_label_enabled(&elem->key))
270 jump_label_dec(&elem->key);
271}
272
273/*
274 * Disable a tracepoint and its probe callback.
275 * Note: only waiting an RCU period after setting elem->call to the empty
276 * function insures that the original callback is not used anymore. This insured
277 * by preempt_disable around the call site.
278 */
279static void disable_tracepoint(struct tracepoint *elem)
280{
281 if (elem->unregfunc && jump_label_enabled(&elem->key))
282 elem->unregfunc();
283
284 if (jump_label_enabled(&elem->key))
285 jump_label_dec(&elem->key);
286 rcu_assign_pointer(elem->funcs, NULL);
287}
288
289/**
290 * tracepoint_update_probe_range - Update a probe range
291 * @begin: beginning of the range
292 * @end: end of the range
293 *
294 * Updates the probe callback corresponding to a range of tracepoints.
295 */
296void tracepoint_update_probe_range(struct tracepoint * const *begin,
297 struct tracepoint * const *end)
298{
299 struct tracepoint * const *iter;
300 struct tracepoint_entry *mark_entry;
301
302 if (!begin)
303 return;
304
305 mutex_lock(&tracepoints_mutex);
306 for (iter = begin; iter < end; iter++) {
307 mark_entry = get_tracepoint((*iter)->name);
308 if (mark_entry) {
309 set_tracepoint(&mark_entry, *iter,
310 !!mark_entry->refcount);
311 } else {
312 disable_tracepoint(*iter);
313 }
314 }
315 mutex_unlock(&tracepoints_mutex);
316}
317
318/*
319 * Update probes, removing the faulty probes.
320 */
321static void tracepoint_update_probes(void)
322{
323 /* Core kernel tracepoints */
324 tracepoint_update_probe_range(__start___tracepoints_ptrs,
325 __stop___tracepoints_ptrs);
326 /* tracepoints in modules. */
327 module_update_tracepoints();
328}
329
330static struct tracepoint_func *
331tracepoint_add_probe(const char *name, void *probe, void *data)
332{
333 struct tracepoint_entry *entry;
334 struct tracepoint_func *old;
335
336 entry = get_tracepoint(name);
337 if (!entry) {
338 entry = add_tracepoint(name);
339 if (IS_ERR(entry))
340 return (struct tracepoint_func *)entry;
341 }
342 old = tracepoint_entry_add_probe(entry, probe, data);
343 if (IS_ERR(old) && !entry->refcount)
344 remove_tracepoint(entry);
345 return old;
346}
347
348/**
349 * tracepoint_probe_register - Connect a probe to a tracepoint
350 * @name: tracepoint name
351 * @probe: probe handler
352 *
353 * Returns 0 if ok, error value on error.
354 * The probe address must at least be aligned on the architecture pointer size.
355 */
356int tracepoint_probe_register(const char *name, void *probe, void *data)
357{
358 struct tracepoint_func *old;
359
360 mutex_lock(&tracepoints_mutex);
361 old = tracepoint_add_probe(name, probe, data);
362 mutex_unlock(&tracepoints_mutex);
363 if (IS_ERR(old))
364 return PTR_ERR(old);
365
366 tracepoint_update_probes(); /* may update entry */
367 release_probes(old);
368 return 0;
369}
370EXPORT_SYMBOL_GPL(tracepoint_probe_register);
371
372static struct tracepoint_func *
373tracepoint_remove_probe(const char *name, void *probe, void *data)
374{
375 struct tracepoint_entry *entry;
376 struct tracepoint_func *old;
377
378 entry = get_tracepoint(name);
379 if (!entry)
380 return ERR_PTR(-ENOENT);
381 old = tracepoint_entry_remove_probe(entry, probe, data);
382 if (IS_ERR(old))
383 return old;
384 if (!entry->refcount)
385 remove_tracepoint(entry);
386 return old;
387}
388
389/**
390 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
391 * @name: tracepoint name
392 * @probe: probe function pointer
393 *
394 * We do not need to call a synchronize_sched to make sure the probes have
395 * finished running before doing a module unload, because the module unload
396 * itself uses stop_machine(), which insures that every preempt disabled section
397 * have finished.
398 */
399int tracepoint_probe_unregister(const char *name, void *probe, void *data)
400{
401 struct tracepoint_func *old;
402
403 mutex_lock(&tracepoints_mutex);
404 old = tracepoint_remove_probe(name, probe, data);
405 mutex_unlock(&tracepoints_mutex);
406 if (IS_ERR(old))
407 return PTR_ERR(old);
408
409 tracepoint_update_probes(); /* may update entry */
410 release_probes(old);
411 return 0;
412}
413EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
414
415static LIST_HEAD(old_probes);
416static int need_update;
417
418static void tracepoint_add_old_probes(void *old)
419{
420 need_update = 1;
421 if (old) {
422 struct tp_probes *tp_probes = container_of(old,
423 struct tp_probes, probes[0]);
424 list_add(&tp_probes->u.list, &old_probes);
425 }
426}
427
428/**
429 * tracepoint_probe_register_noupdate - register a probe but not connect
430 * @name: tracepoint name
431 * @probe: probe handler
432 *
433 * caller must call tracepoint_probe_update_all()
434 */
435int tracepoint_probe_register_noupdate(const char *name, void *probe,
436 void *data)
437{
438 struct tracepoint_func *old;
439
440 mutex_lock(&tracepoints_mutex);
441 old = tracepoint_add_probe(name, probe, data);
442 if (IS_ERR(old)) {
443 mutex_unlock(&tracepoints_mutex);
444 return PTR_ERR(old);
445 }
446 tracepoint_add_old_probes(old);
447 mutex_unlock(&tracepoints_mutex);
448 return 0;
449}
450EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
451
452/**
453 * tracepoint_probe_unregister_noupdate - remove a probe but not disconnect
454 * @name: tracepoint name
455 * @probe: probe function pointer
456 *
457 * caller must call tracepoint_probe_update_all()
458 */
459int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
460 void *data)
461{
462 struct tracepoint_func *old;
463
464 mutex_lock(&tracepoints_mutex);
465 old = tracepoint_remove_probe(name, probe, data);
466 if (IS_ERR(old)) {
467 mutex_unlock(&tracepoints_mutex);
468 return PTR_ERR(old);
469 }
470 tracepoint_add_old_probes(old);
471 mutex_unlock(&tracepoints_mutex);
472 return 0;
473}
474EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
475
476/**
477 * tracepoint_probe_update_all - update tracepoints
478 */
479void tracepoint_probe_update_all(void)
480{
481 LIST_HEAD(release_probes);
482 struct tp_probes *pos, *next;
483
484 mutex_lock(&tracepoints_mutex);
485 if (!need_update) {
486 mutex_unlock(&tracepoints_mutex);
487 return;
488 }
489 if (!list_empty(&old_probes))
490 list_replace_init(&old_probes, &release_probes);
491 need_update = 0;
492 mutex_unlock(&tracepoints_mutex);
493
494 tracepoint_update_probes();
495 list_for_each_entry_safe(pos, next, &release_probes, u.list) {
496 list_del(&pos->u.list);
497 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
498 }
499}
500EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
501
502/**
503 * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
504 * @tracepoint: current tracepoints (in), next tracepoint (out)
505 * @begin: beginning of the range
506 * @end: end of the range
507 *
508 * Returns whether a next tracepoint has been found (1) or not (0).
509 * Will return the first tracepoint in the range if the input tracepoint is
510 * NULL.
511 */
512int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
513 struct tracepoint * const *begin, struct tracepoint * const *end)
514{
515 if (!*tracepoint && begin != end) {
516 *tracepoint = begin;
517 return 1;
518 }
519 if (*tracepoint >= begin && *tracepoint < end)
520 return 1;
521 return 0;
522}
523EXPORT_SYMBOL_GPL(tracepoint_get_iter_range);
524
525static void tracepoint_get_iter(struct tracepoint_iter *iter)
526{
527 int found = 0;
528
529 /* Core kernel tracepoints */
530 if (!iter->module) {
531 found = tracepoint_get_iter_range(&iter->tracepoint,
532 __start___tracepoints_ptrs,
533 __stop___tracepoints_ptrs);
534 if (found)
535 goto end;
536 }
537 /* tracepoints in modules. */
538 found = module_get_iter_tracepoints(iter);
539end:
540 if (!found)
541 tracepoint_iter_reset(iter);
542}
543
544void tracepoint_iter_start(struct tracepoint_iter *iter)
545{
546 tracepoint_get_iter(iter);
547}
548EXPORT_SYMBOL_GPL(tracepoint_iter_start);
549
550void tracepoint_iter_next(struct tracepoint_iter *iter)
551{
552 iter->tracepoint++;
553 /*
554 * iter->tracepoint may be invalid because we blindly incremented it.
555 * Make sure it is valid by marshalling on the tracepoints, getting the
556 * tracepoints from following modules if necessary.
557 */
558 tracepoint_get_iter(iter);
559}
560EXPORT_SYMBOL_GPL(tracepoint_iter_next);
561
562void tracepoint_iter_stop(struct tracepoint_iter *iter)
563{
564}
565EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
566
567void tracepoint_iter_reset(struct tracepoint_iter *iter)
568{
569 iter->module = NULL;
570 iter->tracepoint = NULL;
571}
572EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
573
574#ifdef CONFIG_MODULES
575
576int tracepoint_module_notify(struct notifier_block *self,
577 unsigned long val, void *data)
578{
579 struct module *mod = data;
580
581 switch (val) {
582 case MODULE_STATE_COMING:
583 case MODULE_STATE_GOING:
584 tracepoint_update_probe_range(mod->tracepoints_ptrs,
585 mod->tracepoints_ptrs + mod->num_tracepoints);
586 break;
587 }
588 return 0;
589}
590
591struct notifier_block tracepoint_module_nb = {
592 .notifier_call = tracepoint_module_notify,
593 .priority = 0,
594};
595
596static int init_tracepoints(void)
597{
598 return register_module_notifier(&tracepoint_module_nb);
599}
600__initcall(init_tracepoints);
601
602#endif /* CONFIG_MODULES */
603
604#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
605
606/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
607static int sys_tracepoint_refcount;
608
609void syscall_regfunc(void)
610{
611 unsigned long flags;
612 struct task_struct *g, *t;
613
614 if (!sys_tracepoint_refcount) {
615 read_lock_irqsave(&tasklist_lock, flags);
616 do_each_thread(g, t) {
617 /* Skip kernel threads. */
618 if (t->mm)
619 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
620 } while_each_thread(g, t);
621 read_unlock_irqrestore(&tasklist_lock, flags);
622 }
623 sys_tracepoint_refcount++;
624}
625
626void syscall_unregfunc(void)
627{
628 unsigned long flags;
629 struct task_struct *g, *t;
630
631 sys_tracepoint_refcount--;
632 if (!sys_tracepoint_refcount) {
633 read_lock_irqsave(&tasklist_lock, flags);
634 do_each_thread(g, t) {
635 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
636 } while_each_thread(g, t);
637 read_unlock_irqrestore(&tasklist_lock, flags);
638 }
639}
640#endif