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>
18 extern tracepoint_ptr_t __start___tracepoints_ptrs[];
19 extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
21 DEFINE_SRCU(tracepoint_srcu);
22 EXPORT_SYMBOL_GPL(tracepoint_srcu);
24 /* Set to 1 to enable tracepoint debug output */
25 static const int tracepoint_debug;
29 * Tracepoint module list mutex protects the local module list.
31 static DEFINE_MUTEX(tracepoint_module_list_mutex);
33 /* Local list of struct tp_module */
34 static LIST_HEAD(tracepoint_module_list);
35 #endif /* CONFIG_MODULES */
38 * tracepoints_mutex protects the builtin and module tracepoints.
39 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
41 static DEFINE_MUTEX(tracepoints_mutex);
43 static struct rcu_head *early_probes;
44 static bool ok_to_free_tracepoints;
48 * It is used to delay the free of multiple probes array until a quiescent
53 struct tracepoint_func probes[];
56 static inline void *allocate_probes(int count)
58 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
60 return p == NULL ? NULL : p->probes;
63 static void srcu_free_old_probes(struct rcu_head *head)
65 kfree(container_of(head, struct tp_probes, rcu));
68 static void rcu_free_old_probes(struct rcu_head *head)
70 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
73 static __init int release_early_probes(void)
77 ok_to_free_tracepoints = true;
79 while (early_probes) {
81 early_probes = tmp->next;
82 call_rcu(tmp, rcu_free_old_probes);
88 /* SRCU is initialized at core_initcall */
89 postcore_initcall(release_early_probes);
91 static inline void release_probes(struct tracepoint_func *old)
94 struct tp_probes *tp_probes = container_of(old,
95 struct tp_probes, probes[0]);
98 * We can't free probes if SRCU is not initialized yet.
99 * Postpone the freeing till after SRCU is initialized.
101 if (unlikely(!ok_to_free_tracepoints)) {
102 tp_probes->rcu.next = early_probes;
103 early_probes = &tp_probes->rcu;
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.
113 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
117 static void debug_print_probes(struct tracepoint_func *funcs)
121 if (!tracepoint_debug || !funcs)
124 for (i = 0; funcs[i].func; i++)
125 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
128 static struct tracepoint_func *
129 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
132 struct tracepoint_func *old, *new;
136 if (WARN_ON(!tp_func->func))
137 return ERR_PTR(-EINVAL);
139 debug_print_probes(*funcs);
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)
147 if (old[nr_probes].func == tp_func->func &&
148 old[nr_probes].data == tp_func->data)
149 return ERR_PTR(-EEXIST);
152 /* + 2 : one for new probe, one for NULL func */
153 new = allocate_probes(nr_probes + 2);
155 return ERR_PTR(-ENOMEM);
159 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
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));
170 new[nr_probes + 1].func = NULL;
172 debug_print_probes(*funcs);
176 static void *func_remove(struct tracepoint_func **funcs,
177 struct tracepoint_func *tp_func)
179 int nr_probes = 0, nr_del = 0, i;
180 struct tracepoint_func *old, *new;
185 return ERR_PTR(-ENOENT);
187 debug_print_probes(*funcs);
188 /* (N -> M), (N > 1, M >= 0) probes */
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)
198 * If probe is NULL, then nr_probes = nr_del = 0, and then the
199 * entire entry will be removed.
201 if (nr_probes - nr_del == 0) {
202 /* N -> 0, (N > 1) */
204 debug_print_probes(*funcs);
208 /* N -> M, (N > 1, M > 0) */
210 new = allocate_probes(nr_probes - nr_del + 1);
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)
217 new[nr_probes - nr_del].func = NULL;
220 debug_print_probes(*funcs);
224 static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs, bool sync)
226 void *func = tp->iterator;
228 /* Synthetic events do not have static call sites */
229 if (!tp->static_call_key)
232 if (!tp_funcs[1].func) {
233 func = tp_funcs[0].func;
235 * If going from the iterator back to a single caller,
236 * we need to synchronize with __DO_TRACE to make sure
237 * that the data passed to the callback is the one that
238 * belongs to that callback.
241 tracepoint_synchronize_unregister();
244 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
248 * Add the probe function to a tracepoint.
250 static int tracepoint_add_func(struct tracepoint *tp,
251 struct tracepoint_func *func, int prio)
253 struct tracepoint_func *old, *tp_funcs;
256 if (tp->regfunc && !static_key_enabled(&tp->key)) {
262 tp_funcs = rcu_dereference_protected(tp->funcs,
263 lockdep_is_held(&tracepoints_mutex));
264 old = func_add(&tp_funcs, func, prio);
266 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
271 * rcu_assign_pointer has as smp_store_release() which makes sure
272 * that the new probe callbacks array is consistent before setting
273 * a pointer to it. This array is referenced by __DO_TRACE from
274 * include/linux/tracepoint.h using rcu_dereference_sched().
276 rcu_assign_pointer(tp->funcs, tp_funcs);
277 tracepoint_update_call(tp, tp_funcs, false);
278 static_key_enable(&tp->key);
285 * Remove a probe function from a tracepoint.
286 * Note: only waiting an RCU period after setting elem->call to the empty
287 * function insures that the original callback is not used anymore. This insured
288 * by preempt_disable around the call site.
290 static int tracepoint_remove_func(struct tracepoint *tp,
291 struct tracepoint_func *func)
293 struct tracepoint_func *old, *tp_funcs;
295 tp_funcs = rcu_dereference_protected(tp->funcs,
296 lockdep_is_held(&tracepoints_mutex));
297 old = func_remove(&tp_funcs, func);
299 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
304 /* Removed last function */
305 if (tp->unregfunc && static_key_enabled(&tp->key))
308 static_key_disable(&tp->key);
309 rcu_assign_pointer(tp->funcs, tp_funcs);
311 rcu_assign_pointer(tp->funcs, tp_funcs);
312 tracepoint_update_call(tp, tp_funcs,
313 tp_funcs[0].func != old[0].func);
320 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
322 * @probe: probe handler
323 * @data: tracepoint data
324 * @prio: priority of this function over other registered functions
326 * Returns 0 if ok, error value on error.
327 * Note: if @tp is within a module, the caller is responsible for
328 * unregistering the probe before the module is gone. This can be
329 * performed either with a tracepoint module going notifier, or from
330 * within module exit functions.
332 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
333 void *data, int prio)
335 struct tracepoint_func tp_func;
338 mutex_lock(&tracepoints_mutex);
339 tp_func.func = probe;
342 ret = tracepoint_add_func(tp, &tp_func, prio);
343 mutex_unlock(&tracepoints_mutex);
346 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
349 * tracepoint_probe_register - Connect a probe to a tracepoint
351 * @probe: probe handler
352 * @data: tracepoint data
354 * Returns 0 if ok, error value on error.
355 * Note: if @tp is within a module, the caller is responsible for
356 * unregistering the probe before the module is gone. This can be
357 * performed either with a tracepoint module going notifier, or from
358 * within module exit functions.
360 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
362 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
364 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
367 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
369 * @probe: probe function pointer
370 * @data: tracepoint data
372 * Returns 0 if ok, error value on error.
374 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
376 struct tracepoint_func tp_func;
379 mutex_lock(&tracepoints_mutex);
380 tp_func.func = probe;
382 ret = tracepoint_remove_func(tp, &tp_func);
383 mutex_unlock(&tracepoints_mutex);
386 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
388 static void for_each_tracepoint_range(
389 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
390 void (*fct)(struct tracepoint *tp, void *priv),
393 tracepoint_ptr_t *iter;
397 for (iter = begin; iter < end; iter++)
398 fct(tracepoint_ptr_deref(iter), priv);
401 #ifdef CONFIG_MODULES
402 bool trace_module_has_bad_taint(struct module *mod)
404 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
405 (1 << TAINT_UNSIGNED_MODULE));
408 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
411 * register_tracepoint_notifier - register tracepoint coming/going notifier
412 * @nb: notifier block
414 * Notifiers registered with this function are called on module
415 * coming/going with the tracepoint_module_list_mutex held.
416 * The notifier block callback should expect a "struct tp_module" data
419 int register_tracepoint_module_notifier(struct notifier_block *nb)
421 struct tp_module *tp_mod;
424 mutex_lock(&tracepoint_module_list_mutex);
425 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
428 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
429 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
431 mutex_unlock(&tracepoint_module_list_mutex);
434 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
437 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
438 * @nb: notifier block
440 * The notifier block callback should expect a "struct tp_module" data
443 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
445 struct tp_module *tp_mod;
448 mutex_lock(&tracepoint_module_list_mutex);
449 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
452 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
453 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
455 mutex_unlock(&tracepoint_module_list_mutex);
459 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
462 * Ensure the tracer unregistered the module's probes before the module
463 * teardown is performed. Prevents leaks of probe and data pointers.
465 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
467 WARN_ON_ONCE(tp->funcs);
470 static int tracepoint_module_coming(struct module *mod)
472 struct tp_module *tp_mod;
475 if (!mod->num_tracepoints)
479 * We skip modules that taint the kernel, especially those with different
480 * module headers (for forced load), to make sure we don't cause a crash.
481 * Staging, out-of-tree, and unsigned GPL modules are fine.
483 if (trace_module_has_bad_taint(mod))
485 mutex_lock(&tracepoint_module_list_mutex);
486 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
492 list_add_tail(&tp_mod->list, &tracepoint_module_list);
493 blocking_notifier_call_chain(&tracepoint_notify_list,
494 MODULE_STATE_COMING, tp_mod);
496 mutex_unlock(&tracepoint_module_list_mutex);
500 static void tracepoint_module_going(struct module *mod)
502 struct tp_module *tp_mod;
504 if (!mod->num_tracepoints)
507 mutex_lock(&tracepoint_module_list_mutex);
508 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
509 if (tp_mod->mod == mod) {
510 blocking_notifier_call_chain(&tracepoint_notify_list,
511 MODULE_STATE_GOING, tp_mod);
512 list_del(&tp_mod->list);
515 * Called the going notifier before checking for
518 for_each_tracepoint_range(mod->tracepoints_ptrs,
519 mod->tracepoints_ptrs + mod->num_tracepoints,
520 tp_module_going_check_quiescent, NULL);
525 * In the case of modules that were tainted at "coming", we'll simply
526 * walk through the list without finding it. We cannot use the "tainted"
527 * flag on "going", in case a module taints the kernel only after being
530 mutex_unlock(&tracepoint_module_list_mutex);
533 static int tracepoint_module_notify(struct notifier_block *self,
534 unsigned long val, void *data)
536 struct module *mod = data;
540 case MODULE_STATE_COMING:
541 ret = tracepoint_module_coming(mod);
543 case MODULE_STATE_LIVE:
545 case MODULE_STATE_GOING:
546 tracepoint_module_going(mod);
548 case MODULE_STATE_UNFORMED:
551 return notifier_from_errno(ret);
554 static struct notifier_block tracepoint_module_nb = {
555 .notifier_call = tracepoint_module_notify,
559 static __init int init_tracepoints(void)
563 ret = register_module_notifier(&tracepoint_module_nb);
565 pr_warn("Failed to register tracepoint module enter notifier\n");
569 __initcall(init_tracepoints);
570 #endif /* CONFIG_MODULES */
573 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
575 * @priv: private data
577 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
580 for_each_tracepoint_range(__start___tracepoints_ptrs,
581 __stop___tracepoints_ptrs, fct, priv);
583 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
585 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
587 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
588 static int sys_tracepoint_refcount;
590 int syscall_regfunc(void)
592 struct task_struct *p, *t;
594 if (!sys_tracepoint_refcount) {
595 read_lock(&tasklist_lock);
596 for_each_process_thread(p, t) {
597 set_task_syscall_work(t, SYSCALL_TRACEPOINT);
599 read_unlock(&tasklist_lock);
601 sys_tracepoint_refcount++;
606 void syscall_unregfunc(void)
608 struct task_struct *p, *t;
610 sys_tracepoint_refcount--;
611 if (!sys_tracepoint_refcount) {
612 read_lock(&tasklist_lock);
613 for_each_process_thread(p, t) {
614 clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
616 read_unlock(&tasklist_lock);