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;
184 int iter_probes; /* Iterate over old probe array. */
185 int nr_probes = 0; /* Counter for probes */
186 int pos = -1; /* Insertion position into new array */
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 (iter_probes = 0; old[iter_probes].func; iter_probes++) {
196 if (old[iter_probes].func == tp_stub_func)
197 continue; /* Skip stub functions. */
198 if (old[iter_probes].func == tp_func->func &&
199 old[iter_probes].data == tp_func->data)
200 return ERR_PTR(-EEXIST);
204 /* + 2 : one for new probe, one for NULL func */
205 new = allocate_probes(nr_probes + 2);
207 return ERR_PTR(-ENOMEM);
210 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
211 if (old[iter_probes].func == tp_stub_func)
213 /* Insert before probes of lower priority */
214 if (pos < 0 && old[iter_probes].prio < prio)
216 new[nr_probes++] = old[iter_probes];
220 /* nr_probes now points to the end of the new array */
223 nr_probes = 1; /* must point at end of array */
226 new[nr_probes].func = NULL;
228 debug_print_probes(*funcs);
232 static void *func_remove(struct tracepoint_func **funcs,
233 struct tracepoint_func *tp_func)
235 int nr_probes = 0, nr_del = 0, i;
236 struct tracepoint_func *old, *new;
241 return ERR_PTR(-ENOENT);
243 debug_print_probes(*funcs);
244 /* (N -> M), (N > 1, M >= 0) probes */
246 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
247 if ((old[nr_probes].func == tp_func->func &&
248 old[nr_probes].data == tp_func->data) ||
249 old[nr_probes].func == tp_stub_func)
255 * If probe is NULL, then nr_probes = nr_del = 0, and then the
256 * entire entry will be removed.
258 if (nr_probes - nr_del == 0) {
259 /* N -> 0, (N > 1) */
261 debug_print_probes(*funcs);
265 /* N -> M, (N > 1, M > 0) */
267 new = allocate_probes(nr_probes - nr_del + 1);
269 for (i = 0; old[i].func; i++) {
270 if ((old[i].func != tp_func->func ||
271 old[i].data != tp_func->data) &&
272 old[i].func != tp_stub_func)
275 new[nr_probes - nr_del].func = NULL;
279 * Failed to allocate, replace the old function
280 * with calls to tp_stub_func.
282 for (i = 0; old[i].func; i++) {
283 if (old[i].func == tp_func->func &&
284 old[i].data == tp_func->data)
285 WRITE_ONCE(old[i].func, tp_stub_func);
290 debug_print_probes(*funcs);
295 * Count the number of functions (enum tp_func_state) in a tp_funcs array.
297 static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
301 if (!tp_funcs[1].func)
303 if (!tp_funcs[2].func)
305 return TP_FUNC_N; /* 3 or more */
308 static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
310 void *func = tp->iterator;
312 /* Synthetic events do not have static call sites */
313 if (!tp->static_call_key)
315 if (nr_func_state(tp_funcs) == TP_FUNC_1)
316 func = tp_funcs[0].func;
317 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
321 * Add the probe function to a tracepoint.
323 static int tracepoint_add_func(struct tracepoint *tp,
324 struct tracepoint_func *func, int prio,
327 struct tracepoint_func *old, *tp_funcs;
330 if (tp->regfunc && !static_key_enabled(&tp->key)) {
336 tp_funcs = rcu_dereference_protected(tp->funcs,
337 lockdep_is_held(&tracepoints_mutex));
338 old = func_add(&tp_funcs, func, prio);
340 WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
345 * rcu_assign_pointer has as smp_store_release() which makes sure
346 * that the new probe callbacks array is consistent before setting
347 * a pointer to it. This array is referenced by __DO_TRACE from
348 * include/linux/tracepoint.h using rcu_dereference_sched().
350 switch (nr_func_state(tp_funcs)) {
351 case TP_FUNC_1: /* 0->1 */
353 * Make sure new static func never uses old data after a
354 * 1->0->1 transition sequence.
356 tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
357 /* Set static call to first function */
358 tracepoint_update_call(tp, tp_funcs);
359 /* Both iterator and static call handle NULL tp->funcs */
360 rcu_assign_pointer(tp->funcs, tp_funcs);
361 static_key_enable(&tp->key);
363 case TP_FUNC_2: /* 1->2 */
364 /* Set iterator static call */
365 tracepoint_update_call(tp, tp_funcs);
367 * Iterator callback installed before updating tp->funcs.
368 * Requires ordering between RCU assign/dereference and
369 * static call update/call.
372 case TP_FUNC_N: /* N->N+1 (N>1) */
373 rcu_assign_pointer(tp->funcs, tp_funcs);
375 * Make sure static func never uses incorrect data after a
376 * N->...->2->1 (N>1) transition sequence.
378 if (tp_funcs[0].data != old[0].data)
379 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
391 * Remove a probe function from a tracepoint.
392 * Note: only waiting an RCU period after setting elem->call to the empty
393 * function insures that the original callback is not used anymore. This insured
394 * by preempt_disable around the call site.
396 static int tracepoint_remove_func(struct tracepoint *tp,
397 struct tracepoint_func *func)
399 struct tracepoint_func *old, *tp_funcs;
401 tp_funcs = rcu_dereference_protected(tp->funcs,
402 lockdep_is_held(&tracepoints_mutex));
403 old = func_remove(&tp_funcs, func);
404 if (WARN_ON_ONCE(IS_ERR(old)))
408 /* Failed allocating new tp_funcs, replaced func with stub */
411 switch (nr_func_state(tp_funcs)) {
412 case TP_FUNC_0: /* 1->0 */
413 /* Removed last function */
414 if (tp->unregfunc && static_key_enabled(&tp->key))
417 static_key_disable(&tp->key);
418 /* Set iterator static call */
419 tracepoint_update_call(tp, tp_funcs);
420 /* Both iterator and static call handle NULL tp->funcs */
421 rcu_assign_pointer(tp->funcs, NULL);
423 * Make sure new static func never uses old data after a
424 * 1->0->1 transition sequence.
426 tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
428 case TP_FUNC_1: /* 2->1 */
429 rcu_assign_pointer(tp->funcs, tp_funcs);
431 * Make sure static func never uses incorrect data after a
432 * N->...->2->1 (N>2) transition sequence. If the first
433 * element's data has changed, then force the synchronization
434 * to prevent current readers that have loaded the old data
435 * from calling the new function.
437 if (tp_funcs[0].data != old[0].data)
438 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
439 tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
440 /* Set static call to first function */
441 tracepoint_update_call(tp, tp_funcs);
443 case TP_FUNC_2: /* N->N-1 (N>2) */
446 rcu_assign_pointer(tp->funcs, tp_funcs);
448 * Make sure static func never uses incorrect data after a
449 * N->...->2->1 (N>2) transition sequence.
451 if (tp_funcs[0].data != old[0].data)
452 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
463 * tracepoint_probe_register_prio_may_exist - Connect a probe to a tracepoint with priority
465 * @probe: probe handler
466 * @data: tracepoint data
467 * @prio: priority of this function over other registered functions
469 * Same as tracepoint_probe_register_prio() except that it will not warn
470 * if the tracepoint is already registered.
472 int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
473 void *data, int prio)
475 struct tracepoint_func tp_func;
478 mutex_lock(&tracepoints_mutex);
479 tp_func.func = probe;
482 ret = tracepoint_add_func(tp, &tp_func, prio, false);
483 mutex_unlock(&tracepoints_mutex);
486 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
489 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
491 * @probe: probe handler
492 * @data: tracepoint data
493 * @prio: priority of this function over other registered functions
495 * Returns 0 if ok, error value on error.
496 * Note: if @tp is within a module, the caller is responsible for
497 * unregistering the probe before the module is gone. This can be
498 * performed either with a tracepoint module going notifier, or from
499 * within module exit functions.
501 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
502 void *data, int prio)
504 struct tracepoint_func tp_func;
507 mutex_lock(&tracepoints_mutex);
508 tp_func.func = probe;
511 ret = tracepoint_add_func(tp, &tp_func, prio, true);
512 mutex_unlock(&tracepoints_mutex);
515 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
518 * tracepoint_probe_register - Connect a probe to a tracepoint
520 * @probe: probe handler
521 * @data: tracepoint data
523 * Returns 0 if ok, error value on error.
524 * Note: if @tp is within a module, the caller is responsible for
525 * unregistering the probe before the module is gone. This can be
526 * performed either with a tracepoint module going notifier, or from
527 * within module exit functions.
529 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
531 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
533 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
536 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
538 * @probe: probe function pointer
539 * @data: tracepoint data
541 * Returns 0 if ok, error value on error.
543 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
545 struct tracepoint_func tp_func;
548 mutex_lock(&tracepoints_mutex);
549 tp_func.func = probe;
551 ret = tracepoint_remove_func(tp, &tp_func);
552 mutex_unlock(&tracepoints_mutex);
555 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
557 static void for_each_tracepoint_range(
558 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
559 void (*fct)(struct tracepoint *tp, void *priv),
562 tracepoint_ptr_t *iter;
566 for (iter = begin; iter < end; iter++)
567 fct(tracepoint_ptr_deref(iter), priv);
570 #ifdef CONFIG_MODULES
571 bool trace_module_has_bad_taint(struct module *mod)
573 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
574 (1 << TAINT_UNSIGNED_MODULE) | (1 << TAINT_TEST) |
575 (1 << TAINT_LIVEPATCH));
578 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
581 * register_tracepoint_module_notifier - register tracepoint coming/going notifier
582 * @nb: notifier block
584 * Notifiers registered with this function are called on module
585 * coming/going with the tracepoint_module_list_mutex held.
586 * The notifier block callback should expect a "struct tp_module" data
589 int register_tracepoint_module_notifier(struct notifier_block *nb)
591 struct tp_module *tp_mod;
594 mutex_lock(&tracepoint_module_list_mutex);
595 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
598 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
599 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
601 mutex_unlock(&tracepoint_module_list_mutex);
604 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
607 * unregister_tracepoint_module_notifier - unregister tracepoint coming/going notifier
608 * @nb: notifier block
610 * The notifier block callback should expect a "struct tp_module" data
613 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
615 struct tp_module *tp_mod;
618 mutex_lock(&tracepoint_module_list_mutex);
619 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
622 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
623 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
625 mutex_unlock(&tracepoint_module_list_mutex);
629 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
632 * Ensure the tracer unregistered the module's probes before the module
633 * teardown is performed. Prevents leaks of probe and data pointers.
635 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
637 WARN_ON_ONCE(tp->funcs);
640 static int tracepoint_module_coming(struct module *mod)
642 struct tp_module *tp_mod;
644 if (!mod->num_tracepoints)
648 * We skip modules that taint the kernel, especially those with different
649 * module headers (for forced load), to make sure we don't cause a crash.
650 * Staging, out-of-tree, unsigned GPL, and test modules are fine.
652 if (trace_module_has_bad_taint(mod))
655 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
660 mutex_lock(&tracepoint_module_list_mutex);
661 list_add_tail(&tp_mod->list, &tracepoint_module_list);
662 blocking_notifier_call_chain(&tracepoint_notify_list,
663 MODULE_STATE_COMING, tp_mod);
664 mutex_unlock(&tracepoint_module_list_mutex);
668 static void tracepoint_module_going(struct module *mod)
670 struct tp_module *tp_mod;
672 if (!mod->num_tracepoints)
675 mutex_lock(&tracepoint_module_list_mutex);
676 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
677 if (tp_mod->mod == mod) {
678 blocking_notifier_call_chain(&tracepoint_notify_list,
679 MODULE_STATE_GOING, tp_mod);
680 list_del(&tp_mod->list);
683 * Called the going notifier before checking for
686 for_each_tracepoint_range(mod->tracepoints_ptrs,
687 mod->tracepoints_ptrs + mod->num_tracepoints,
688 tp_module_going_check_quiescent, NULL);
693 * In the case of modules that were tainted at "coming", we'll simply
694 * walk through the list without finding it. We cannot use the "tainted"
695 * flag on "going", in case a module taints the kernel only after being
698 mutex_unlock(&tracepoint_module_list_mutex);
701 static int tracepoint_module_notify(struct notifier_block *self,
702 unsigned long val, void *data)
704 struct module *mod = data;
708 case MODULE_STATE_COMING:
709 ret = tracepoint_module_coming(mod);
711 case MODULE_STATE_LIVE:
713 case MODULE_STATE_GOING:
714 tracepoint_module_going(mod);
716 case MODULE_STATE_UNFORMED:
719 return notifier_from_errno(ret);
722 static struct notifier_block tracepoint_module_nb = {
723 .notifier_call = tracepoint_module_notify,
727 static __init int init_tracepoints(void)
731 ret = register_module_notifier(&tracepoint_module_nb);
733 pr_warn("Failed to register tracepoint module enter notifier\n");
737 __initcall(init_tracepoints);
738 #endif /* CONFIG_MODULES */
741 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
743 * @priv: private data
745 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
748 for_each_tracepoint_range(__start___tracepoints_ptrs,
749 __stop___tracepoints_ptrs, fct, priv);
751 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
753 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
755 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
756 static int sys_tracepoint_refcount;
758 int syscall_regfunc(void)
760 struct task_struct *p, *t;
762 if (!sys_tracepoint_refcount) {
763 read_lock(&tasklist_lock);
764 for_each_process_thread(p, t) {
765 set_task_syscall_work(t, SYSCALL_TRACEPOINT);
767 read_unlock(&tasklist_lock);
769 sys_tracepoint_refcount++;
774 void syscall_unregfunc(void)
776 struct task_struct *p, *t;
778 sys_tracepoint_refcount--;
779 if (!sys_tracepoint_refcount) {
780 read_lock(&tasklist_lock);
781 for_each_process_thread(p, t) {
782 clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
784 read_unlock(&tasklist_lock);