2 * Copyright (C) 2008-2014 Mathieu Desnoyers
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.
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.
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.
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/signal.h>
28 #include <linux/sched/task.h>
29 #include <linux/static_key.h>
31 extern struct tracepoint * const __start___tracepoints_ptrs[];
32 extern struct tracepoint * const __stop___tracepoints_ptrs[];
34 /* Set to 1 to enable tracepoint debug output */
35 static const int tracepoint_debug;
39 * Tracepoint module list mutex protects the local module list.
41 static DEFINE_MUTEX(tracepoint_module_list_mutex);
43 /* Local list of struct tp_module */
44 static LIST_HEAD(tracepoint_module_list);
45 #endif /* CONFIG_MODULES */
48 * tracepoints_mutex protects the builtin and module tracepoints.
49 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
51 static DEFINE_MUTEX(tracepoints_mutex);
55 * It is used to delay the free of multiple probes array until a quiescent
60 struct tracepoint_func probes[0];
63 /* Called in removal of a func but failed to allocate a new tp_funcs */
64 static void tp_stub_func(void)
69 static inline void *allocate_probes(int count)
71 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
72 + sizeof(struct tp_probes), GFP_KERNEL);
73 return p == NULL ? NULL : p->probes;
76 static void rcu_free_old_probes(struct rcu_head *head)
78 kfree(container_of(head, struct tp_probes, rcu));
81 static inline void release_probes(struct tracepoint_func *old)
84 struct tp_probes *tp_probes = container_of(old,
85 struct tp_probes, probes[0]);
86 call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
90 static void debug_print_probes(struct tracepoint_func *funcs)
94 if (!tracepoint_debug || !funcs)
97 for (i = 0; funcs[i].func; i++)
98 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
101 static struct tracepoint_func *
102 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
105 struct tracepoint_func *old, *new;
110 if (WARN_ON(!tp_func->func))
111 return ERR_PTR(-EINVAL);
113 debug_print_probes(*funcs);
116 /* (N -> N+1), (N != 0, 1) probes */
117 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
118 /* Insert before probes of lower priority */
119 if (pos < 0 && old[nr_probes].prio < prio)
121 if (old[nr_probes].func == tp_func->func &&
122 old[nr_probes].data == tp_func->data)
123 return ERR_PTR(-EEXIST);
124 if (old[nr_probes].func == tp_stub_func)
128 /* + 2 : one for new probe, one for NULL func - stub functions */
129 new = allocate_probes(nr_probes + 2 - stub_funcs);
131 return ERR_PTR(-ENOMEM);
134 /* Need to copy one at a time to remove stubs */
138 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
139 if (old[nr_probes].func == tp_stub_func)
141 if (pos < 0 && old[nr_probes].prio < prio)
143 new[probes++] = old[nr_probes];
149 nr_probes--; /* Account for insertion */
151 } else if (pos < 0) {
153 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
155 /* Copy higher priority probes ahead of the new probe */
156 memcpy(new, old, pos * sizeof(struct tracepoint_func));
157 /* Copy the rest after it. */
158 memcpy(new + pos + 1, old + pos,
159 (nr_probes - pos) * sizeof(struct tracepoint_func));
164 new[nr_probes + 1].func = NULL;
166 debug_print_probes(*funcs);
170 static void *func_remove(struct tracepoint_func **funcs,
171 struct tracepoint_func *tp_func)
173 int nr_probes = 0, nr_del = 0, i;
174 struct tracepoint_func *old, *new;
179 return ERR_PTR(-ENOENT);
181 debug_print_probes(*funcs);
182 /* (N -> M), (N > 1, M >= 0) probes */
184 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
185 if ((old[nr_probes].func == tp_func->func &&
186 old[nr_probes].data == tp_func->data) ||
187 old[nr_probes].func == tp_stub_func)
193 * If probe is NULL, then nr_probes = nr_del = 0, and then the
194 * entire entry will be removed.
196 if (nr_probes - nr_del == 0) {
197 /* N -> 0, (N > 1) */
199 debug_print_probes(*funcs);
203 /* N -> M, (N > 1, M > 0) */
205 new = allocate_probes(nr_probes - nr_del + 1);
207 for (i = 0; old[i].func; i++)
208 if ((old[i].func != tp_func->func
209 || old[i].data != tp_func->data)
210 && old[i].func != tp_stub_func)
212 new[nr_probes - nr_del].func = NULL;
216 * Failed to allocate, replace the old function
217 * with calls to tp_stub_func.
219 for (i = 0; old[i].func; i++)
220 if (old[i].func == tp_func->func &&
221 old[i].data == tp_func->data) {
222 old[i].func = tp_stub_func;
223 /* Set the prio to the next event. */
233 debug_print_probes(*funcs);
238 * Add the probe function to a tracepoint.
240 static int tracepoint_add_func(struct tracepoint *tp,
241 struct tracepoint_func *func, int prio)
243 struct tracepoint_func *old, *tp_funcs;
246 if (tp->regfunc && !static_key_enabled(&tp->key)) {
252 tp_funcs = rcu_dereference_protected(tp->funcs,
253 lockdep_is_held(&tracepoints_mutex));
254 old = func_add(&tp_funcs, func, prio);
256 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
261 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
262 * probe callbacks array is consistent before setting a pointer to it.
263 * This array is referenced by __DO_TRACE from
264 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
267 rcu_assign_pointer(tp->funcs, tp_funcs);
268 if (!static_key_enabled(&tp->key))
269 static_key_slow_inc(&tp->key);
275 * Remove a probe function from a tracepoint.
276 * Note: only waiting an RCU period after setting elem->call to the empty
277 * function insures that the original callback is not used anymore. This insured
278 * by preempt_disable around the call site.
280 static int tracepoint_remove_func(struct tracepoint *tp,
281 struct tracepoint_func *func)
283 struct tracepoint_func *old, *tp_funcs;
285 tp_funcs = rcu_dereference_protected(tp->funcs,
286 lockdep_is_held(&tracepoints_mutex));
287 old = func_remove(&tp_funcs, func);
288 if (WARN_ON_ONCE(IS_ERR(old)))
292 /* Failed allocating new tp_funcs, replaced func with stub */
296 /* Removed last function */
297 if (tp->unregfunc && static_key_enabled(&tp->key))
300 if (static_key_enabled(&tp->key))
301 static_key_slow_dec(&tp->key);
303 rcu_assign_pointer(tp->funcs, tp_funcs);
309 * tracepoint_probe_register - Connect a probe to a tracepoint
311 * @probe: probe handler
312 * @data: tracepoint data
313 * @prio: priority of this function over other registered functions
315 * Returns 0 if ok, error value on error.
316 * Note: if @tp is within a module, the caller is responsible for
317 * unregistering the probe before the module is gone. This can be
318 * performed either with a tracepoint module going notifier, or from
319 * within module exit functions.
321 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
322 void *data, int prio)
324 struct tracepoint_func tp_func;
327 mutex_lock(&tracepoints_mutex);
328 tp_func.func = probe;
331 ret = tracepoint_add_func(tp, &tp_func, prio);
332 mutex_unlock(&tracepoints_mutex);
335 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
338 * tracepoint_probe_register - Connect a probe to a tracepoint
340 * @probe: probe handler
341 * @data: tracepoint data
342 * @prio: priority of this function over other registered functions
344 * Returns 0 if ok, error value on error.
345 * Note: if @tp is within a module, the caller is responsible for
346 * unregistering the probe before the module is gone. This can be
347 * performed either with a tracepoint module going notifier, or from
348 * within module exit functions.
350 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
352 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
354 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
357 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
359 * @probe: probe function pointer
360 * @data: tracepoint data
362 * Returns 0 if ok, error value on error.
364 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
366 struct tracepoint_func tp_func;
369 mutex_lock(&tracepoints_mutex);
370 tp_func.func = probe;
372 ret = tracepoint_remove_func(tp, &tp_func);
373 mutex_unlock(&tracepoints_mutex);
376 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
378 #ifdef CONFIG_MODULES
379 bool trace_module_has_bad_taint(struct module *mod)
381 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
382 (1 << TAINT_UNSIGNED_MODULE));
385 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
388 * register_tracepoint_notifier - register tracepoint coming/going notifier
389 * @nb: notifier block
391 * Notifiers registered with this function are called on module
392 * coming/going with the tracepoint_module_list_mutex held.
393 * The notifier block callback should expect a "struct tp_module" data
396 int register_tracepoint_module_notifier(struct notifier_block *nb)
398 struct tp_module *tp_mod;
401 mutex_lock(&tracepoint_module_list_mutex);
402 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
405 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
406 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
408 mutex_unlock(&tracepoint_module_list_mutex);
411 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
414 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
415 * @nb: notifier block
417 * The notifier block callback should expect a "struct tp_module" data
420 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
422 struct tp_module *tp_mod;
425 mutex_lock(&tracepoint_module_list_mutex);
426 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
429 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
430 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
432 mutex_unlock(&tracepoint_module_list_mutex);
436 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
439 * Ensure the tracer unregistered the module's probes before the module
440 * teardown is performed. Prevents leaks of probe and data pointers.
442 static void tp_module_going_check_quiescent(struct tracepoint * const *begin,
443 struct tracepoint * const *end)
445 struct tracepoint * const *iter;
449 for (iter = begin; iter < end; iter++)
450 WARN_ON_ONCE((*iter)->funcs);
453 static int tracepoint_module_coming(struct module *mod)
455 struct tp_module *tp_mod;
458 if (!mod->num_tracepoints)
462 * We skip modules that taint the kernel, especially those with different
463 * module headers (for forced load), to make sure we don't cause a crash.
464 * Staging, out-of-tree, and unsigned GPL modules are fine.
466 if (trace_module_has_bad_taint(mod))
468 mutex_lock(&tracepoint_module_list_mutex);
469 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
475 list_add_tail(&tp_mod->list, &tracepoint_module_list);
476 blocking_notifier_call_chain(&tracepoint_notify_list,
477 MODULE_STATE_COMING, tp_mod);
479 mutex_unlock(&tracepoint_module_list_mutex);
483 static void tracepoint_module_going(struct module *mod)
485 struct tp_module *tp_mod;
487 if (!mod->num_tracepoints)
490 mutex_lock(&tracepoint_module_list_mutex);
491 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
492 if (tp_mod->mod == mod) {
493 blocking_notifier_call_chain(&tracepoint_notify_list,
494 MODULE_STATE_GOING, tp_mod);
495 list_del(&tp_mod->list);
498 * Called the going notifier before checking for
501 tp_module_going_check_quiescent(mod->tracepoints_ptrs,
502 mod->tracepoints_ptrs + mod->num_tracepoints);
507 * In the case of modules that were tainted at "coming", we'll simply
508 * walk through the list without finding it. We cannot use the "tainted"
509 * flag on "going", in case a module taints the kernel only after being
512 mutex_unlock(&tracepoint_module_list_mutex);
515 static int tracepoint_module_notify(struct notifier_block *self,
516 unsigned long val, void *data)
518 struct module *mod = data;
522 case MODULE_STATE_COMING:
523 ret = tracepoint_module_coming(mod);
525 case MODULE_STATE_LIVE:
527 case MODULE_STATE_GOING:
528 tracepoint_module_going(mod);
530 case MODULE_STATE_UNFORMED:
536 static struct notifier_block tracepoint_module_nb = {
537 .notifier_call = tracepoint_module_notify,
541 static __init int init_tracepoints(void)
545 ret = register_module_notifier(&tracepoint_module_nb);
547 pr_warn("Failed to register tracepoint module enter notifier\n");
551 __initcall(init_tracepoints);
552 #endif /* CONFIG_MODULES */
554 static void for_each_tracepoint_range(struct tracepoint * const *begin,
555 struct tracepoint * const *end,
556 void (*fct)(struct tracepoint *tp, void *priv),
559 struct tracepoint * const *iter;
563 for (iter = begin; iter < end; iter++)
568 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
570 * @priv: private data
572 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
575 for_each_tracepoint_range(__start___tracepoints_ptrs,
576 __stop___tracepoints_ptrs, fct, priv);
578 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
580 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
582 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
583 static int sys_tracepoint_refcount;
585 int syscall_regfunc(void)
587 struct task_struct *p, *t;
589 if (!sys_tracepoint_refcount) {
590 read_lock(&tasklist_lock);
591 for_each_process_thread(p, t) {
592 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
594 read_unlock(&tasklist_lock);
596 sys_tracepoint_refcount++;
601 void syscall_unregfunc(void)
603 struct task_struct *p, *t;
605 sys_tracepoint_refcount--;
606 if (!sys_tracepoint_refcount) {
607 read_lock(&tasklist_lock);
608 for_each_process_thread(p, t) {
609 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
611 read_unlock(&tasklist_lock);