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// 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
18extern tracepoint_ptr_t __start___tracepoints_ptrs[];
19extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
20
21DEFINE_SRCU(tracepoint_srcu);
22EXPORT_SYMBOL_GPL(tracepoint_srcu);
23
24/* Set to 1 to enable tracepoint debug output */
25static const int tracepoint_debug;
26
27#ifdef CONFIG_MODULES
28/*
29 * Tracepoint module list mutex protects the local module list.
30 */
31static DEFINE_MUTEX(tracepoint_module_list_mutex);
32
33/* Local list of struct tp_module */
34static LIST_HEAD(tracepoint_module_list);
35#endif /* CONFIG_MODULES */
36
37/*
38 * tracepoints_mutex protects the builtin and module tracepoints.
39 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
40 */
41static DEFINE_MUTEX(tracepoints_mutex);
42
43static struct rcu_head *early_probes;
44static bool ok_to_free_tracepoints;
45
46/*
47 * Note about RCU :
48 * It is used to delay the free of multiple probes array until a quiescent
49 * state is reached.
50 */
51struct tp_probes {
52 struct rcu_head rcu;
53 struct tracepoint_func probes[0];
54};
55
56static inline void *allocate_probes(int count)
57{
58 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
59 GFP_KERNEL);
60 return p == NULL ? NULL : p->probes;
61}
62
63static void srcu_free_old_probes(struct rcu_head *head)
64{
65 kfree(container_of(head, struct tp_probes, rcu));
66}
67
68static void rcu_free_old_probes(struct rcu_head *head)
69{
70 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
71}
72
73static __init int release_early_probes(void)
74{
75 struct rcu_head *tmp;
76
77 ok_to_free_tracepoints = true;
78
79 while (early_probes) {
80 tmp = early_probes;
81 early_probes = tmp->next;
82 call_rcu(tmp, rcu_free_old_probes);
83 }
84
85 return 0;
86}
87
88/* SRCU is initialized at core_initcall */
89postcore_initcall(release_early_probes);
90
91static inline void release_probes(struct tracepoint_func *old)
92{
93 if (old) {
94 struct tp_probes *tp_probes = container_of(old,
95 struct tp_probes, probes[0]);
96
97 /*
98 * We can't free probes if SRCU is not initialized yet.
99 * Postpone the freeing till after SRCU is initialized.
100 */
101 if (unlikely(!ok_to_free_tracepoints)) {
102 tp_probes->rcu.next = early_probes;
103 early_probes = &tp_probes->rcu;
104 return;
105 }
106
107 /*
108 * Tracepoint probes are protected by both sched RCU and SRCU,
109 * by calling the SRCU callback in the sched RCU callback we
110 * cover both cases. So let us chain the SRCU and sched RCU
111 * callbacks to wait for both grace periods.
112 */
113 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
114 }
115}
116
117static void debug_print_probes(struct tracepoint_func *funcs)
118{
119 int i;
120
121 if (!tracepoint_debug || !funcs)
122 return;
123
124 for (i = 0; funcs[i].func; i++)
125 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
126}
127
128static struct tracepoint_func *
129func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
130 int prio)
131{
132 struct tracepoint_func *old, *new;
133 int nr_probes = 0;
134 int pos = -1;
135
136 if (WARN_ON(!tp_func->func))
137 return ERR_PTR(-EINVAL);
138
139 debug_print_probes(*funcs);
140 old = *funcs;
141 if (old) {
142 /* (N -> N+1), (N != 0, 1) probes */
143 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
144 /* Insert before probes of lower priority */
145 if (pos < 0 && old[nr_probes].prio < prio)
146 pos = nr_probes;
147 if (old[nr_probes].func == tp_func->func &&
148 old[nr_probes].data == tp_func->data)
149 return ERR_PTR(-EEXIST);
150 }
151 }
152 /* + 2 : one for new probe, one for NULL func */
153 new = allocate_probes(nr_probes + 2);
154 if (new == NULL)
155 return ERR_PTR(-ENOMEM);
156 if (old) {
157 if (pos < 0) {
158 pos = nr_probes;
159 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
160 } else {
161 /* Copy higher priority probes ahead of the new probe */
162 memcpy(new, old, pos * sizeof(struct tracepoint_func));
163 /* Copy the rest after it. */
164 memcpy(new + pos + 1, old + pos,
165 (nr_probes - pos) * sizeof(struct tracepoint_func));
166 }
167 } else
168 pos = 0;
169 new[pos] = *tp_func;
170 new[nr_probes + 1].func = NULL;
171 *funcs = new;
172 debug_print_probes(*funcs);
173 return old;
174}
175
176static void *func_remove(struct tracepoint_func **funcs,
177 struct tracepoint_func *tp_func)
178{
179 int nr_probes = 0, nr_del = 0, i;
180 struct tracepoint_func *old, *new;
181
182 old = *funcs;
183
184 if (!old)
185 return ERR_PTR(-ENOENT);
186
187 debug_print_probes(*funcs);
188 /* (N -> M), (N > 1, M >= 0) probes */
189 if (tp_func->func) {
190 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
191 if (old[nr_probes].func == tp_func->func &&
192 old[nr_probes].data == tp_func->data)
193 nr_del++;
194 }
195 }
196
197 /*
198 * If probe is NULL, then nr_probes = nr_del = 0, and then the
199 * entire entry will be removed.
200 */
201 if (nr_probes - nr_del == 0) {
202 /* N -> 0, (N > 1) */
203 *funcs = NULL;
204 debug_print_probes(*funcs);
205 return old;
206 } else {
207 int j = 0;
208 /* N -> M, (N > 1, M > 0) */
209 /* + 1 for NULL */
210 new = allocate_probes(nr_probes - nr_del + 1);
211 if (new == NULL)
212 return ERR_PTR(-ENOMEM);
213 for (i = 0; old[i].func; i++)
214 if (old[i].func != tp_func->func
215 || old[i].data != tp_func->data)
216 new[j++] = old[i];
217 new[nr_probes - nr_del].func = NULL;
218 *funcs = new;
219 }
220 debug_print_probes(*funcs);
221 return old;
222}
223
224/*
225 * Add the probe function to a tracepoint.
226 */
227static int tracepoint_add_func(struct tracepoint *tp,
228 struct tracepoint_func *func, int prio)
229{
230 struct tracepoint_func *old, *tp_funcs;
231 int ret;
232
233 if (tp->regfunc && !static_key_enabled(&tp->key)) {
234 ret = tp->regfunc();
235 if (ret < 0)
236 return ret;
237 }
238
239 tp_funcs = rcu_dereference_protected(tp->funcs,
240 lockdep_is_held(&tracepoints_mutex));
241 old = func_add(&tp_funcs, func, prio);
242 if (IS_ERR(old)) {
243 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
244 return PTR_ERR(old);
245 }
246
247 /*
248 * rcu_assign_pointer has as smp_store_release() which makes sure
249 * that the new probe callbacks array is consistent before setting
250 * a pointer to it. This array is referenced by __DO_TRACE from
251 * include/linux/tracepoint.h using rcu_dereference_sched().
252 */
253 rcu_assign_pointer(tp->funcs, tp_funcs);
254 if (!static_key_enabled(&tp->key))
255 static_key_slow_inc(&tp->key);
256 release_probes(old);
257 return 0;
258}
259
260/*
261 * Remove a probe function from a tracepoint.
262 * Note: only waiting an RCU period after setting elem->call to the empty
263 * function insures that the original callback is not used anymore. This insured
264 * by preempt_disable around the call site.
265 */
266static int tracepoint_remove_func(struct tracepoint *tp,
267 struct tracepoint_func *func)
268{
269 struct tracepoint_func *old, *tp_funcs;
270
271 tp_funcs = rcu_dereference_protected(tp->funcs,
272 lockdep_is_held(&tracepoints_mutex));
273 old = func_remove(&tp_funcs, func);
274 if (IS_ERR(old)) {
275 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
276 return PTR_ERR(old);
277 }
278
279 if (!tp_funcs) {
280 /* Removed last function */
281 if (tp->unregfunc && static_key_enabled(&tp->key))
282 tp->unregfunc();
283
284 if (static_key_enabled(&tp->key))
285 static_key_slow_dec(&tp->key);
286 }
287 rcu_assign_pointer(tp->funcs, tp_funcs);
288 release_probes(old);
289 return 0;
290}
291
292/**
293 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
294 * @tp: tracepoint
295 * @probe: probe handler
296 * @data: tracepoint data
297 * @prio: priority of this function over other registered functions
298 *
299 * Returns 0 if ok, error value on error.
300 * Note: if @tp is within a module, the caller is responsible for
301 * unregistering the probe before the module is gone. This can be
302 * performed either with a tracepoint module going notifier, or from
303 * within module exit functions.
304 */
305int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
306 void *data, int prio)
307{
308 struct tracepoint_func tp_func;
309 int ret;
310
311 mutex_lock(&tracepoints_mutex);
312 tp_func.func = probe;
313 tp_func.data = data;
314 tp_func.prio = prio;
315 ret = tracepoint_add_func(tp, &tp_func, prio);
316 mutex_unlock(&tracepoints_mutex);
317 return ret;
318}
319EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
320
321/**
322 * tracepoint_probe_register - Connect a probe to a tracepoint
323 * @tp: tracepoint
324 * @probe: probe handler
325 * @data: tracepoint data
326 *
327 * Returns 0 if ok, error value on error.
328 * Note: if @tp is within a module, the caller is responsible for
329 * unregistering the probe before the module is gone. This can be
330 * performed either with a tracepoint module going notifier, or from
331 * within module exit functions.
332 */
333int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
334{
335 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
336}
337EXPORT_SYMBOL_GPL(tracepoint_probe_register);
338
339/**
340 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
341 * @tp: tracepoint
342 * @probe: probe function pointer
343 * @data: tracepoint data
344 *
345 * Returns 0 if ok, error value on error.
346 */
347int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
348{
349 struct tracepoint_func tp_func;
350 int ret;
351
352 mutex_lock(&tracepoints_mutex);
353 tp_func.func = probe;
354 tp_func.data = data;
355 ret = tracepoint_remove_func(tp, &tp_func);
356 mutex_unlock(&tracepoints_mutex);
357 return ret;
358}
359EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
360
361static void for_each_tracepoint_range(
362 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
363 void (*fct)(struct tracepoint *tp, void *priv),
364 void *priv)
365{
366 tracepoint_ptr_t *iter;
367
368 if (!begin)
369 return;
370 for (iter = begin; iter < end; iter++)
371 fct(tracepoint_ptr_deref(iter), priv);
372}
373
374#ifdef CONFIG_MODULES
375bool trace_module_has_bad_taint(struct module *mod)
376{
377 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
378 (1 << TAINT_UNSIGNED_MODULE));
379}
380
381static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
382
383/**
384 * register_tracepoint_notifier - register tracepoint coming/going notifier
385 * @nb: notifier block
386 *
387 * Notifiers registered with this function are called on module
388 * coming/going with the tracepoint_module_list_mutex held.
389 * The notifier block callback should expect a "struct tp_module" data
390 * pointer.
391 */
392int register_tracepoint_module_notifier(struct notifier_block *nb)
393{
394 struct tp_module *tp_mod;
395 int ret;
396
397 mutex_lock(&tracepoint_module_list_mutex);
398 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
399 if (ret)
400 goto end;
401 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
402 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
403end:
404 mutex_unlock(&tracepoint_module_list_mutex);
405 return ret;
406}
407EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
408
409/**
410 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
411 * @nb: notifier block
412 *
413 * The notifier block callback should expect a "struct tp_module" data
414 * pointer.
415 */
416int unregister_tracepoint_module_notifier(struct notifier_block *nb)
417{
418 struct tp_module *tp_mod;
419 int ret;
420
421 mutex_lock(&tracepoint_module_list_mutex);
422 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
423 if (ret)
424 goto end;
425 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
426 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
427end:
428 mutex_unlock(&tracepoint_module_list_mutex);
429 return ret;
430
431}
432EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
433
434/*
435 * Ensure the tracer unregistered the module's probes before the module
436 * teardown is performed. Prevents leaks of probe and data pointers.
437 */
438static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
439{
440 WARN_ON_ONCE(tp->funcs);
441}
442
443static int tracepoint_module_coming(struct module *mod)
444{
445 struct tp_module *tp_mod;
446 int ret = 0;
447
448 if (!mod->num_tracepoints)
449 return 0;
450
451 /*
452 * We skip modules that taint the kernel, especially those with different
453 * module headers (for forced load), to make sure we don't cause a crash.
454 * Staging, out-of-tree, and unsigned GPL modules are fine.
455 */
456 if (trace_module_has_bad_taint(mod))
457 return 0;
458 mutex_lock(&tracepoint_module_list_mutex);
459 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
460 if (!tp_mod) {
461 ret = -ENOMEM;
462 goto end;
463 }
464 tp_mod->mod = mod;
465 list_add_tail(&tp_mod->list, &tracepoint_module_list);
466 blocking_notifier_call_chain(&tracepoint_notify_list,
467 MODULE_STATE_COMING, tp_mod);
468end:
469 mutex_unlock(&tracepoint_module_list_mutex);
470 return ret;
471}
472
473static void tracepoint_module_going(struct module *mod)
474{
475 struct tp_module *tp_mod;
476
477 if (!mod->num_tracepoints)
478 return;
479
480 mutex_lock(&tracepoint_module_list_mutex);
481 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
482 if (tp_mod->mod == mod) {
483 blocking_notifier_call_chain(&tracepoint_notify_list,
484 MODULE_STATE_GOING, tp_mod);
485 list_del(&tp_mod->list);
486 kfree(tp_mod);
487 /*
488 * Called the going notifier before checking for
489 * quiescence.
490 */
491 for_each_tracepoint_range(mod->tracepoints_ptrs,
492 mod->tracepoints_ptrs + mod->num_tracepoints,
493 tp_module_going_check_quiescent, NULL);
494 break;
495 }
496 }
497 /*
498 * In the case of modules that were tainted at "coming", we'll simply
499 * walk through the list without finding it. We cannot use the "tainted"
500 * flag on "going", in case a module taints the kernel only after being
501 * loaded.
502 */
503 mutex_unlock(&tracepoint_module_list_mutex);
504}
505
506static int tracepoint_module_notify(struct notifier_block *self,
507 unsigned long val, void *data)
508{
509 struct module *mod = data;
510 int ret = 0;
511
512 switch (val) {
513 case MODULE_STATE_COMING:
514 ret = tracepoint_module_coming(mod);
515 break;
516 case MODULE_STATE_LIVE:
517 break;
518 case MODULE_STATE_GOING:
519 tracepoint_module_going(mod);
520 break;
521 case MODULE_STATE_UNFORMED:
522 break;
523 }
524 return ret;
525}
526
527static struct notifier_block tracepoint_module_nb = {
528 .notifier_call = tracepoint_module_notify,
529 .priority = 0,
530};
531
532static __init int init_tracepoints(void)
533{
534 int ret;
535
536 ret = register_module_notifier(&tracepoint_module_nb);
537 if (ret)
538 pr_warn("Failed to register tracepoint module enter notifier\n");
539
540 return ret;
541}
542__initcall(init_tracepoints);
543#endif /* CONFIG_MODULES */
544
545/**
546 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
547 * @fct: callback
548 * @priv: private data
549 */
550void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
551 void *priv)
552{
553 for_each_tracepoint_range(__start___tracepoints_ptrs,
554 __stop___tracepoints_ptrs, fct, priv);
555}
556EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
557
558#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
559
560/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
561static int sys_tracepoint_refcount;
562
563int syscall_regfunc(void)
564{
565 struct task_struct *p, *t;
566
567 if (!sys_tracepoint_refcount) {
568 read_lock(&tasklist_lock);
569 for_each_process_thread(p, t) {
570 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
571 }
572 read_unlock(&tasklist_lock);
573 }
574 sys_tracepoint_refcount++;
575
576 return 0;
577}
578
579void syscall_unregfunc(void)
580{
581 struct task_struct *p, *t;
582
583 sys_tracepoint_refcount--;
584 if (!sys_tracepoint_refcount) {
585 read_lock(&tasklist_lock);
586 for_each_process_thread(p, t) {
587 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
588 }
589 read_unlock(&tasklist_lock);
590 }
591}
592#endif