1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2008-2014 Mathieu Desnoyers
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>
25 extern tracepoint_ptr_t __start___tracepoints_ptrs[];
26 extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
28 DEFINE_SRCU(tracepoint_srcu);
29 EXPORT_SYMBOL_GPL(tracepoint_srcu);
31 enum tp_transition_sync {
32 TP_TRANSITION_SYNC_1_0_1,
33 TP_TRANSITION_SYNC_N_2_1,
35 _NR_TP_TRANSITION_SYNC,
38 struct tp_transition_snapshot {
44 /* Protected by tracepoints_mutex */
45 static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC];
47 static void tp_rcu_get_state(enum tp_transition_sync sync)
49 struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
51 /* Keep the latest get_state snapshot. */
52 snapshot->rcu = get_state_synchronize_rcu();
53 snapshot->srcu = start_poll_synchronize_srcu(&tracepoint_srcu);
54 snapshot->ongoing = true;
57 static void tp_rcu_cond_sync(enum tp_transition_sync sync)
59 struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
61 if (!snapshot->ongoing)
63 cond_synchronize_rcu(snapshot->rcu);
64 if (!poll_state_synchronize_srcu(&tracepoint_srcu, snapshot->srcu))
65 synchronize_srcu(&tracepoint_srcu);
66 snapshot->ongoing = false;
69 /* Set to 1 to enable tracepoint debug output */
70 static const int tracepoint_debug;
74 * Tracepoint module list mutex protects the local module list.
76 static DEFINE_MUTEX(tracepoint_module_list_mutex);
78 /* Local list of struct tp_module */
79 static LIST_HEAD(tracepoint_module_list);
80 #endif /* CONFIG_MODULES */
83 * tracepoints_mutex protects the builtin and module tracepoints.
84 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
86 static DEFINE_MUTEX(tracepoints_mutex);
88 static struct rcu_head *early_probes;
89 static bool ok_to_free_tracepoints;
93 * It is used to delay the free of multiple probes array until a quiescent
98 struct tracepoint_func probes[];
101 /* Called in removal of a func but failed to allocate a new tp_funcs */
102 static void tp_stub_func(void)
107 static inline void *allocate_probes(int count)
109 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
111 return p == NULL ? NULL : p->probes;
114 static void srcu_free_old_probes(struct rcu_head *head)
116 kfree(container_of(head, struct tp_probes, rcu));
119 static void rcu_free_old_probes(struct rcu_head *head)
121 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
124 static __init int release_early_probes(void)
126 struct rcu_head *tmp;
128 ok_to_free_tracepoints = true;
130 while (early_probes) {
132 early_probes = tmp->next;
133 call_rcu(tmp, rcu_free_old_probes);
139 /* SRCU is initialized at core_initcall */
140 postcore_initcall(release_early_probes);
142 static inline void release_probes(struct tracepoint_func *old)
145 struct tp_probes *tp_probes = container_of(old,
146 struct tp_probes, probes[0]);
149 * We can't free probes if SRCU is not initialized yet.
150 * Postpone the freeing till after SRCU is initialized.
152 if (unlikely(!ok_to_free_tracepoints)) {
153 tp_probes->rcu.next = early_probes;
154 early_probes = &tp_probes->rcu;
159 * Tracepoint probes are protected by both sched RCU and SRCU,
160 * by calling the SRCU callback in the sched RCU callback we
161 * cover both cases. So let us chain the SRCU and sched RCU
162 * callbacks to wait for both grace periods.
164 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
168 static void debug_print_probes(struct tracepoint_func *funcs)
172 if (!tracepoint_debug || !funcs)
175 for (i = 0; funcs[i].func; i++)
176 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
179 static struct tracepoint_func *
180 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
183 struct tracepoint_func *old, *new;
188 if (WARN_ON(!tp_func->func))
189 return ERR_PTR(-EINVAL);
191 debug_print_probes(*funcs);
194 /* (N -> N+1), (N != 0, 1) probes */
195 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
196 /* Insert before probes of lower priority */
197 if (pos < 0 && old[nr_probes].prio < prio)
199 if (old[nr_probes].func == tp_func->func &&
200 old[nr_probes].data == tp_func->data)
201 return ERR_PTR(-EEXIST);
202 if (old[nr_probes].func == tp_stub_func)
206 /* + 2 : one for new probe, one for NULL func - stub functions */
207 new = allocate_probes(nr_probes + 2 - stub_funcs);
209 return ERR_PTR(-ENOMEM);
212 /* Need to copy one at a time to remove stubs */
216 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
217 if (old[nr_probes].func == tp_stub_func)
219 if (pos < 0 && old[nr_probes].prio < prio)
221 new[probes++] = old[nr_probes];
227 nr_probes--; /* Account for insertion */
229 } else if (pos < 0) {
231 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
233 /* Copy higher priority probes ahead of the new probe */
234 memcpy(new, old, pos * sizeof(struct tracepoint_func));
235 /* Copy the rest after it. */
236 memcpy(new + pos + 1, old + pos,
237 (nr_probes - pos) * sizeof(struct tracepoint_func));
242 new[nr_probes + 1].func = NULL;
244 debug_print_probes(*funcs);
248 static void *func_remove(struct tracepoint_func **funcs,
249 struct tracepoint_func *tp_func)
251 int nr_probes = 0, nr_del = 0, i;
252 struct tracepoint_func *old, *new;
257 return ERR_PTR(-ENOENT);
259 debug_print_probes(*funcs);
260 /* (N -> M), (N > 1, M >= 0) probes */
262 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
263 if ((old[nr_probes].func == tp_func->func &&
264 old[nr_probes].data == tp_func->data) ||
265 old[nr_probes].func == tp_stub_func)
271 * If probe is NULL, then nr_probes = nr_del = 0, and then the
272 * entire entry will be removed.
274 if (nr_probes - nr_del == 0) {
275 /* N -> 0, (N > 1) */
277 debug_print_probes(*funcs);
281 /* N -> M, (N > 1, M > 0) */
283 new = allocate_probes(nr_probes - nr_del + 1);
285 for (i = 0; old[i].func; i++)
286 if ((old[i].func != tp_func->func
287 || old[i].data != tp_func->data)
288 && old[i].func != tp_stub_func)
290 new[nr_probes - nr_del].func = NULL;
294 * Failed to allocate, replace the old function
295 * with calls to tp_stub_func.
297 for (i = 0; old[i].func; i++)
298 if (old[i].func == tp_func->func &&
299 old[i].data == tp_func->data) {
300 old[i].func = tp_stub_func;
301 /* Set the prio to the next event. */
311 debug_print_probes(*funcs);
316 * Count the number of functions (enum tp_func_state) in a tp_funcs array.
318 static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
322 if (!tp_funcs[1].func)
324 if (!tp_funcs[2].func)
326 return TP_FUNC_N; /* 3 or more */
329 static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
331 void *func = tp->iterator;
333 /* Synthetic events do not have static call sites */
334 if (!tp->static_call_key)
336 if (nr_func_state(tp_funcs) == TP_FUNC_1)
337 func = tp_funcs[0].func;
338 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
342 * Add the probe function to a tracepoint.
344 static int tracepoint_add_func(struct tracepoint *tp,
345 struct tracepoint_func *func, int prio,
348 struct tracepoint_func *old, *tp_funcs;
351 if (tp->regfunc && !static_key_enabled(&tp->key)) {
357 tp_funcs = rcu_dereference_protected(tp->funcs,
358 lockdep_is_held(&tracepoints_mutex));
359 old = func_add(&tp_funcs, func, prio);
361 WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
366 * rcu_assign_pointer has as smp_store_release() which makes sure
367 * that the new probe callbacks array is consistent before setting
368 * a pointer to it. This array is referenced by __DO_TRACE from
369 * include/linux/tracepoint.h using rcu_dereference_sched().
371 switch (nr_func_state(tp_funcs)) {
372 case TP_FUNC_1: /* 0->1 */
374 * Make sure new static func never uses old data after a
375 * 1->0->1 transition sequence.
377 tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
378 /* Set static call to first function */
379 tracepoint_update_call(tp, tp_funcs);
380 /* Both iterator and static call handle NULL tp->funcs */
381 rcu_assign_pointer(tp->funcs, tp_funcs);
382 static_key_enable(&tp->key);
384 case TP_FUNC_2: /* 1->2 */
385 /* Set iterator static call */
386 tracepoint_update_call(tp, tp_funcs);
388 * Iterator callback installed before updating tp->funcs.
389 * Requires ordering between RCU assign/dereference and
390 * static call update/call.
393 case TP_FUNC_N: /* N->N+1 (N>1) */
394 rcu_assign_pointer(tp->funcs, tp_funcs);
396 * Make sure static func never uses incorrect data after a
397 * N->...->2->1 (N>1) transition sequence.
399 if (tp_funcs[0].data != old[0].data)
400 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
412 * Remove a probe function from a tracepoint.
413 * Note: only waiting an RCU period after setting elem->call to the empty
414 * function insures that the original callback is not used anymore. This insured
415 * by preempt_disable around the call site.
417 static int tracepoint_remove_func(struct tracepoint *tp,
418 struct tracepoint_func *func)
420 struct tracepoint_func *old, *tp_funcs;
422 tp_funcs = rcu_dereference_protected(tp->funcs,
423 lockdep_is_held(&tracepoints_mutex));
424 old = func_remove(&tp_funcs, func);
425 if (WARN_ON_ONCE(IS_ERR(old)))
429 /* Failed allocating new tp_funcs, replaced func with stub */
432 switch (nr_func_state(tp_funcs)) {
433 case TP_FUNC_0: /* 1->0 */
434 /* Removed last function */
435 if (tp->unregfunc && static_key_enabled(&tp->key))
438 static_key_disable(&tp->key);
439 /* Set iterator static call */
440 tracepoint_update_call(tp, tp_funcs);
441 /* Both iterator and static call handle NULL tp->funcs */
442 rcu_assign_pointer(tp->funcs, NULL);
444 * Make sure new static func never uses old data after a
445 * 1->0->1 transition sequence.
447 tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
449 case TP_FUNC_1: /* 2->1 */
450 rcu_assign_pointer(tp->funcs, tp_funcs);
452 * Make sure static func never uses incorrect data after a
453 * N->...->2->1 (N>2) transition sequence. If the first
454 * element's data has changed, then force the synchronization
455 * to prevent current readers that have loaded the old data
456 * from calling the new function.
458 if (tp_funcs[0].data != old[0].data)
459 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
460 tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
461 /* Set static call to first function */
462 tracepoint_update_call(tp, tp_funcs);
464 case TP_FUNC_2: /* N->N-1 (N>2) */
467 rcu_assign_pointer(tp->funcs, tp_funcs);
469 * Make sure static func never uses incorrect data after a
470 * N->...->2->1 (N>2) transition sequence.
472 if (tp_funcs[0].data != old[0].data)
473 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
484 * tracepoint_probe_register_prio_may_exist - Connect a probe to a tracepoint with priority
486 * @probe: probe handler
487 * @data: tracepoint data
488 * @prio: priority of this function over other registered functions
490 * Same as tracepoint_probe_register_prio() except that it will not warn
491 * if the tracepoint is already registered.
493 int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
494 void *data, int prio)
496 struct tracepoint_func tp_func;
499 mutex_lock(&tracepoints_mutex);
500 tp_func.func = probe;
503 ret = tracepoint_add_func(tp, &tp_func, prio, false);
504 mutex_unlock(&tracepoints_mutex);
507 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
510 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
512 * @probe: probe handler
513 * @data: tracepoint data
514 * @prio: priority of this function over other registered functions
516 * Returns 0 if ok, error value on error.
517 * Note: if @tp is within a module, the caller is responsible for
518 * unregistering the probe before the module is gone. This can be
519 * performed either with a tracepoint module going notifier, or from
520 * within module exit functions.
522 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
523 void *data, int prio)
525 struct tracepoint_func tp_func;
528 mutex_lock(&tracepoints_mutex);
529 tp_func.func = probe;
532 ret = tracepoint_add_func(tp, &tp_func, prio, true);
533 mutex_unlock(&tracepoints_mutex);
536 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
539 * tracepoint_probe_register - Connect a probe to a tracepoint
541 * @probe: probe handler
542 * @data: tracepoint data
544 * Returns 0 if ok, error value on error.
545 * Note: if @tp is within a module, the caller is responsible for
546 * unregistering the probe before the module is gone. This can be
547 * performed either with a tracepoint module going notifier, or from
548 * within module exit functions.
550 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
552 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
554 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
557 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
559 * @probe: probe function pointer
560 * @data: tracepoint data
562 * Returns 0 if ok, error value on error.
564 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
566 struct tracepoint_func tp_func;
569 mutex_lock(&tracepoints_mutex);
570 tp_func.func = probe;
572 ret = tracepoint_remove_func(tp, &tp_func);
573 mutex_unlock(&tracepoints_mutex);
576 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
578 static void for_each_tracepoint_range(
579 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
580 void (*fct)(struct tracepoint *tp, void *priv),
583 tracepoint_ptr_t *iter;
587 for (iter = begin; iter < end; iter++)
588 fct(tracepoint_ptr_deref(iter), priv);
591 #ifdef CONFIG_MODULES
592 bool trace_module_has_bad_taint(struct module *mod)
594 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
595 (1 << TAINT_UNSIGNED_MODULE));
598 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
601 * register_tracepoint_notifier - register tracepoint coming/going notifier
602 * @nb: notifier block
604 * Notifiers registered with this function are called on module
605 * coming/going with the tracepoint_module_list_mutex held.
606 * The notifier block callback should expect a "struct tp_module" data
609 int register_tracepoint_module_notifier(struct notifier_block *nb)
611 struct tp_module *tp_mod;
614 mutex_lock(&tracepoint_module_list_mutex);
615 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
618 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
619 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
621 mutex_unlock(&tracepoint_module_list_mutex);
624 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
627 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
628 * @nb: notifier block
630 * The notifier block callback should expect a "struct tp_module" data
633 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
635 struct tp_module *tp_mod;
638 mutex_lock(&tracepoint_module_list_mutex);
639 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
642 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
643 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
645 mutex_unlock(&tracepoint_module_list_mutex);
649 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
652 * Ensure the tracer unregistered the module's probes before the module
653 * teardown is performed. Prevents leaks of probe and data pointers.
655 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
657 WARN_ON_ONCE(tp->funcs);
660 static int tracepoint_module_coming(struct module *mod)
662 struct tp_module *tp_mod;
665 if (!mod->num_tracepoints)
669 * We skip modules that taint the kernel, especially those with different
670 * module headers (for forced load), to make sure we don't cause a crash.
671 * Staging, out-of-tree, and unsigned GPL modules are fine.
673 if (trace_module_has_bad_taint(mod))
675 mutex_lock(&tracepoint_module_list_mutex);
676 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
682 list_add_tail(&tp_mod->list, &tracepoint_module_list);
683 blocking_notifier_call_chain(&tracepoint_notify_list,
684 MODULE_STATE_COMING, tp_mod);
686 mutex_unlock(&tracepoint_module_list_mutex);
690 static void tracepoint_module_going(struct module *mod)
692 struct tp_module *tp_mod;
694 if (!mod->num_tracepoints)
697 mutex_lock(&tracepoint_module_list_mutex);
698 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
699 if (tp_mod->mod == mod) {
700 blocking_notifier_call_chain(&tracepoint_notify_list,
701 MODULE_STATE_GOING, tp_mod);
702 list_del(&tp_mod->list);
705 * Called the going notifier before checking for
708 for_each_tracepoint_range(mod->tracepoints_ptrs,
709 mod->tracepoints_ptrs + mod->num_tracepoints,
710 tp_module_going_check_quiescent, NULL);
715 * In the case of modules that were tainted at "coming", we'll simply
716 * walk through the list without finding it. We cannot use the "tainted"
717 * flag on "going", in case a module taints the kernel only after being
720 mutex_unlock(&tracepoint_module_list_mutex);
723 static int tracepoint_module_notify(struct notifier_block *self,
724 unsigned long val, void *data)
726 struct module *mod = data;
730 case MODULE_STATE_COMING:
731 ret = tracepoint_module_coming(mod);
733 case MODULE_STATE_LIVE:
735 case MODULE_STATE_GOING:
736 tracepoint_module_going(mod);
738 case MODULE_STATE_UNFORMED:
741 return notifier_from_errno(ret);
744 static struct notifier_block tracepoint_module_nb = {
745 .notifier_call = tracepoint_module_notify,
749 static __init int init_tracepoints(void)
753 ret = register_module_notifier(&tracepoint_module_nb);
755 pr_warn("Failed to register tracepoint module enter notifier\n");
759 __initcall(init_tracepoints);
760 #endif /* CONFIG_MODULES */
763 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
765 * @priv: private data
767 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
770 for_each_tracepoint_range(__start___tracepoints_ptrs,
771 __stop___tracepoints_ptrs, fct, priv);
773 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
775 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
777 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
778 static int sys_tracepoint_refcount;
780 int syscall_regfunc(void)
782 struct task_struct *p, *t;
784 if (!sys_tracepoint_refcount) {
785 read_lock(&tasklist_lock);
786 for_each_process_thread(p, t) {
787 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
789 read_unlock(&tasklist_lock);
791 sys_tracepoint_refcount++;
796 void syscall_unregfunc(void)
798 struct task_struct *p, *t;
800 sys_tracepoint_refcount--;
801 if (!sys_tracepoint_refcount) {
802 read_lock(&tasklist_lock);
803 for_each_process_thread(p, t) {
804 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
806 read_unlock(&tasklist_lock);