xref: /linux/kernel/tracepoint.c (revision e2683c8868d03382da7e1ce8453b543a043066d1)
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 
18 enum tp_func_state {
19 	TP_FUNC_0,
20 	TP_FUNC_1,
21 	TP_FUNC_2,
22 	TP_FUNC_N,
23 };
24 
25 extern tracepoint_ptr_t __start___tracepoints_ptrs[];
26 extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
27 
28 enum 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 
35 struct tp_transition_snapshot {
36 	unsigned long rcu;
37 	unsigned long srcu_gp;
38 	bool ongoing;
39 };
40 
41 DEFINE_SRCU_FAST(tracepoint_srcu);
42 EXPORT_SYMBOL_GPL(tracepoint_srcu);
43 
44 /* Protected by tracepoints_mutex */
45 static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC];
46 
47 static void tp_rcu_get_state(enum tp_transition_sync sync)
48 {
49 	struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
50 
51 	/* Keep the latest get_state snapshot. */
52 	snapshot->rcu = get_state_synchronize_rcu();
53 	snapshot->srcu_gp = start_poll_synchronize_srcu(&tracepoint_srcu);
54 	snapshot->ongoing = true;
55 }
56 
57 static void tp_rcu_cond_sync(enum tp_transition_sync sync)
58 {
59 	struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
60 
61 	if (!snapshot->ongoing)
62 		return;
63 	cond_synchronize_rcu(snapshot->rcu);
64 	if (!poll_state_synchronize_srcu(&tracepoint_srcu, snapshot->srcu_gp))
65 		synchronize_srcu(&tracepoint_srcu);
66 	snapshot->ongoing = false;
67 }
68 
69 /* Set to 1 to enable tracepoint debug output */
70 static const int tracepoint_debug;
71 
72 #ifdef CONFIG_MODULES
73 /*
74  * Tracepoint module list mutex protects the local module list.
75  */
76 static DEFINE_MUTEX(tracepoint_module_list_mutex);
77 
78 /* Local list of struct tp_module */
79 static LIST_HEAD(tracepoint_module_list);
80 #endif /* CONFIG_MODULES */
81 
82 /*
83  * tracepoints_mutex protects the builtin and module tracepoints.
84  * tracepoints_mutex nests inside tracepoint_module_list_mutex.
85  */
86 static DEFINE_MUTEX(tracepoints_mutex);
87 
88 /*
89  * Note about RCU :
90  * It is used to delay the free of multiple probes array until a quiescent
91  * state is reached.
92  */
93 struct tp_probes {
94 	struct rcu_head rcu;
95 	struct tracepoint_func probes[];
96 };
97 
98 /* Called in removal of a func but failed to allocate a new tp_funcs */
99 static void tp_stub_func(void)
100 {
101 	return;
102 }
103 
104 static inline void *allocate_probes(int count)
105 {
106 	struct tp_probes *p  = kmalloc_flex(*p, probes, count);
107 	return p == NULL ? NULL : p->probes;
108 }
109 
110 static void rcu_free_old_probes(struct rcu_head *head)
111 {
112 	kfree(container_of(head, struct tp_probes, rcu));
113 }
114 
115 static inline void release_probes(struct tracepoint *tp, struct tracepoint_func *old)
116 {
117 	if (old) {
118 		struct tp_probes *tp_probes = container_of(old,
119 			struct tp_probes, probes[0]);
120 
121 		if (tracepoint_is_faultable(tp)) {
122 			call_rcu_tasks_trace(&tp_probes->rcu,
123 					     rcu_free_old_probes);
124 		} else {
125 			call_srcu(&tracepoint_srcu, &tp_probes->rcu,
126 				  rcu_free_old_probes);
127 		}
128 	}
129 }
130 
131 static void debug_print_probes(struct tracepoint_func *funcs)
132 {
133 	int i;
134 
135 	if (!tracepoint_debug || !funcs)
136 		return;
137 
138 	for (i = 0; funcs[i].func; i++)
139 		printk(KERN_DEBUG "Probe %d : %pSb\n", i, funcs[i].func);
140 }
141 
142 static struct tracepoint_func *
143 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
144 	 int prio)
145 {
146 	struct tracepoint_func *old, *new;
147 	int iter_probes;	/* Iterate over old probe array. */
148 	int nr_probes = 0;	/* Counter for probes */
149 	int pos = -1;		/* Insertion position into new array */
150 
151 	if (WARN_ON(!tp_func->func))
152 		return ERR_PTR(-EINVAL);
153 
154 	debug_print_probes(*funcs);
155 	old = *funcs;
156 	if (old) {
157 		/* (N -> N+1), (N != 0, 1) probes */
158 		for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
159 			if (old[iter_probes].func == tp_stub_func)
160 				continue;	/* Skip stub functions. */
161 			if (old[iter_probes].func == tp_func->func &&
162 			    old[iter_probes].data == tp_func->data)
163 				return ERR_PTR(-EEXIST);
164 			nr_probes++;
165 		}
166 	}
167 	/* + 2 : one for new probe, one for NULL func */
168 	new = allocate_probes(nr_probes + 2);
169 	if (new == NULL)
170 		return ERR_PTR(-ENOMEM);
171 	if (old) {
172 		nr_probes = 0;
173 		for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
174 			if (old[iter_probes].func == tp_stub_func)
175 				continue;
176 			/* Insert before probes of lower priority */
177 			if (pos < 0 && old[iter_probes].prio < prio)
178 				pos = nr_probes++;
179 			new[nr_probes++] = old[iter_probes];
180 		}
181 		if (pos < 0)
182 			pos = nr_probes++;
183 		/* nr_probes now points to the end of the new array */
184 	} else {
185 		pos = 0;
186 		nr_probes = 1; /* must point at end of array */
187 	}
188 	new[pos] = *tp_func;
189 	new[nr_probes].func = NULL;
190 	*funcs = new;
191 	debug_print_probes(*funcs);
192 	return old;
193 }
194 
195 static void *func_remove(struct tracepoint_func **funcs,
196 		struct tracepoint_func *tp_func)
197 {
198 	int nr_probes = 0, nr_del = 0, i;
199 	struct tracepoint_func *old, *new;
200 
201 	old = *funcs;
202 
203 	if (!old)
204 		return ERR_PTR(-ENOENT);
205 
206 	debug_print_probes(*funcs);
207 	/* (N -> M), (N > 1, M >= 0) probes */
208 	if (tp_func->func) {
209 		for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
210 			if ((old[nr_probes].func == tp_func->func &&
211 			     old[nr_probes].data == tp_func->data) ||
212 			    old[nr_probes].func == tp_stub_func)
213 				nr_del++;
214 		}
215 	}
216 
217 	/*
218 	 * If probe is NULL, then nr_probes = nr_del = 0, and then the
219 	 * entire entry will be removed.
220 	 */
221 	if (nr_probes - nr_del == 0) {
222 		/* N -> 0, (N > 1) */
223 		*funcs = NULL;
224 		debug_print_probes(*funcs);
225 		return old;
226 	} else {
227 		int j = 0;
228 		/* N -> M, (N > 1, M > 0) */
229 		/* + 1 for NULL */
230 		new = allocate_probes(nr_probes - nr_del + 1);
231 		if (new) {
232 			for (i = 0; old[i].func; i++) {
233 				if ((old[i].func != tp_func->func ||
234 				     old[i].data != tp_func->data) &&
235 				    old[i].func != tp_stub_func)
236 					new[j++] = old[i];
237 			}
238 			new[nr_probes - nr_del].func = NULL;
239 			*funcs = new;
240 		} else {
241 			/*
242 			 * Failed to allocate, replace the old function
243 			 * with calls to tp_stub_func.
244 			 */
245 			for (i = 0; old[i].func; i++) {
246 				if (old[i].func == tp_func->func &&
247 				    old[i].data == tp_func->data)
248 					WRITE_ONCE(old[i].func, tp_stub_func);
249 			}
250 			*funcs = old;
251 		}
252 	}
253 	debug_print_probes(*funcs);
254 	return old;
255 }
256 
257 /*
258  * Count the number of functions (enum tp_func_state) in a tp_funcs array.
259  */
260 static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
261 {
262 	if (!tp_funcs)
263 		return TP_FUNC_0;
264 	if (!tp_funcs[1].func)
265 		return TP_FUNC_1;
266 	if (!tp_funcs[2].func)
267 		return TP_FUNC_2;
268 	return TP_FUNC_N;	/* 3 or more */
269 }
270 
271 static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
272 {
273 	void *func = tp->iterator;
274 
275 	/* Synthetic events do not have static call sites */
276 	if (!tp->static_call_key)
277 		return;
278 	if (nr_func_state(tp_funcs) == TP_FUNC_1)
279 		func = tp_funcs[0].func;
280 	__static_call_update(tp->static_call_key, tp->static_call_tramp, func);
281 }
282 
283 /*
284  * Add the probe function to a tracepoint.
285  */
286 static int tracepoint_add_func(struct tracepoint *tp,
287 			       struct tracepoint_func *func, int prio,
288 			       bool warn)
289 {
290 	struct tracepoint_func *old, *tp_funcs;
291 	int ret;
292 
293 	if (tp->ext && tp->ext->regfunc && !static_key_enabled(&tp->key)) {
294 		ret = tp->ext->regfunc();
295 		if (ret < 0)
296 			return ret;
297 	}
298 
299 	tp_funcs = rcu_dereference_protected(tp->funcs,
300 			lockdep_is_held(&tracepoints_mutex));
301 	old = func_add(&tp_funcs, func, prio);
302 	if (IS_ERR(old)) {
303 		if (tp->ext && tp->ext->unregfunc && !static_key_enabled(&tp->key))
304 			tp->ext->unregfunc();
305 		WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
306 		return PTR_ERR(old);
307 	}
308 
309 	/*
310 	 * rcu_assign_pointer has as smp_store_release() which makes sure
311 	 * that the new probe callbacks array is consistent before setting
312 	 * a pointer to it.  This array is referenced by __DO_TRACE from
313 	 * include/linux/tracepoint.h using rcu_dereference_sched().
314 	 */
315 	switch (nr_func_state(tp_funcs)) {
316 	case TP_FUNC_1:		/* 0->1 */
317 		/*
318 		 * Make sure new static func never uses old data after a
319 		 * 1->0->1 transition sequence.
320 		 */
321 		tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
322 		/* Set static call to first function */
323 		tracepoint_update_call(tp, tp_funcs);
324 		/* Both iterator and static call handle NULL tp->funcs */
325 		rcu_assign_pointer(tp->funcs, tp_funcs);
326 		static_branch_enable(&tp->key);
327 		break;
328 	case TP_FUNC_2:		/* 1->2 */
329 		/* Set iterator static call */
330 		tracepoint_update_call(tp, tp_funcs);
331 		/*
332 		 * Iterator callback installed before updating tp->funcs.
333 		 * Requires ordering between RCU assign/dereference and
334 		 * static call update/call.
335 		 */
336 		fallthrough;
337 	case TP_FUNC_N:		/* N->N+1 (N>1) */
338 		rcu_assign_pointer(tp->funcs, tp_funcs);
339 		/*
340 		 * Make sure static func never uses incorrect data after a
341 		 * N->...->2->1 (N>1) transition sequence.
342 		 */
343 		if (tp_funcs[0].data != old[0].data)
344 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
345 		break;
346 	default:
347 		WARN_ON_ONCE(1);
348 		break;
349 	}
350 
351 	release_probes(tp, old);
352 	return 0;
353 }
354 
355 /*
356  * Remove a probe function from a tracepoint.
357  * Note: only waiting an RCU period after setting elem->call to the empty
358  * function insures that the original callback is not used anymore. This insured
359  * by preempt_disable around the call site.
360  */
361 static int tracepoint_remove_func(struct tracepoint *tp,
362 		struct tracepoint_func *func)
363 {
364 	struct tracepoint_func *old, *tp_funcs;
365 
366 	tp_funcs = rcu_dereference_protected(tp->funcs,
367 			lockdep_is_held(&tracepoints_mutex));
368 	old = func_remove(&tp_funcs, func);
369 	if (WARN_ON_ONCE(IS_ERR(old)))
370 		return PTR_ERR(old);
371 
372 	if (tp_funcs == old)
373 		/* Failed allocating new tp_funcs, replaced func with stub */
374 		return 0;
375 
376 	switch (nr_func_state(tp_funcs)) {
377 	case TP_FUNC_0:		/* 1->0 */
378 		/* Removed last function */
379 		if (tp->ext && tp->ext->unregfunc && static_key_enabled(&tp->key))
380 			tp->ext->unregfunc();
381 		static_branch_disable(&tp->key);
382 		/* Set iterator static call */
383 		tracepoint_update_call(tp, tp_funcs);
384 		/* Both iterator and static call handle NULL tp->funcs */
385 		rcu_assign_pointer(tp->funcs, NULL);
386 		/*
387 		 * Make sure new static func never uses old data after a
388 		 * 1->0->1 transition sequence.
389 		 */
390 		tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
391 		break;
392 	case TP_FUNC_1:		/* 2->1 */
393 		rcu_assign_pointer(tp->funcs, tp_funcs);
394 		/*
395 		 * Make sure static func never uses incorrect data after a
396 		 * N->...->2->1 (N>2) transition sequence. If the first
397 		 * element's data has changed, then force the synchronization
398 		 * to prevent current readers that have loaded the old data
399 		 * from calling the new function.
400 		 */
401 		if (tp_funcs[0].data != old[0].data)
402 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
403 		tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
404 		/* Set static call to first function */
405 		tracepoint_update_call(tp, tp_funcs);
406 		break;
407 	case TP_FUNC_2:		/* N->N-1 (N>2) */
408 		fallthrough;
409 	case TP_FUNC_N:
410 		rcu_assign_pointer(tp->funcs, tp_funcs);
411 		/*
412 		 * Make sure static func never uses incorrect data after a
413 		 * N->...->2->1 (N>2) transition sequence.
414 		 */
415 		if (tp_funcs[0].data != old[0].data)
416 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
417 		break;
418 	default:
419 		WARN_ON_ONCE(1);
420 		break;
421 	}
422 	release_probes(tp, old);
423 	return 0;
424 }
425 
426 /**
427  * tracepoint_probe_register_prio_may_exist -  Connect a probe to a tracepoint with priority
428  * @tp: tracepoint
429  * @probe: probe handler
430  * @data: tracepoint data
431  * @prio: priority of this function over other registered functions
432  *
433  * Same as tracepoint_probe_register_prio() except that it will not warn
434  * if the tracepoint is already registered.
435  */
436 int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
437 					     void *data, int prio)
438 {
439 	struct tracepoint_func tp_func;
440 	int ret;
441 
442 	mutex_lock(&tracepoints_mutex);
443 	tp_func.func = probe;
444 	tp_func.data = data;
445 	tp_func.prio = prio;
446 	ret = tracepoint_add_func(tp, &tp_func, prio, false);
447 	mutex_unlock(&tracepoints_mutex);
448 	return ret;
449 }
450 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
451 
452 /**
453  * tracepoint_probe_register_prio -  Connect a probe to a tracepoint with priority
454  * @tp: tracepoint
455  * @probe: probe handler
456  * @data: tracepoint data
457  * @prio: priority of this function over other registered functions
458  *
459  * Returns 0 if ok, error value on error.
460  * Note: if @tp is within a module, the caller is responsible for
461  * unregistering the probe before the module is gone. This can be
462  * performed either with a tracepoint module going notifier, or from
463  * within module exit functions.
464  */
465 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
466 				   void *data, int prio)
467 {
468 	struct tracepoint_func tp_func;
469 	int ret;
470 
471 	mutex_lock(&tracepoints_mutex);
472 	tp_func.func = probe;
473 	tp_func.data = data;
474 	tp_func.prio = prio;
475 	ret = tracepoint_add_func(tp, &tp_func, prio, true);
476 	mutex_unlock(&tracepoints_mutex);
477 	return ret;
478 }
479 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
480 
481 /**
482  * tracepoint_probe_register -  Connect a probe to a tracepoint
483  * @tp: tracepoint
484  * @probe: probe handler
485  * @data: tracepoint data
486  *
487  * Returns 0 if ok, error value on error.
488  * Note: if @tp is within a module, the caller is responsible for
489  * unregistering the probe before the module is gone. This can be
490  * performed either with a tracepoint module going notifier, or from
491  * within module exit functions.
492  */
493 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
494 {
495 	return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
496 }
497 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
498 
499 /**
500  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
501  * @tp: tracepoint
502  * @probe: probe function pointer
503  * @data: tracepoint data
504  *
505  * Returns 0 if ok, error value on error.
506  */
507 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
508 {
509 	struct tracepoint_func tp_func;
510 	int ret;
511 
512 	mutex_lock(&tracepoints_mutex);
513 	tp_func.func = probe;
514 	tp_func.data = data;
515 	ret = tracepoint_remove_func(tp, &tp_func);
516 	mutex_unlock(&tracepoints_mutex);
517 	return ret;
518 }
519 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
520 
521 static void for_each_tracepoint_range(
522 		tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
523 		void (*fct)(struct tracepoint *tp, void *priv),
524 		void *priv)
525 {
526 	tracepoint_ptr_t *iter;
527 
528 	if (!begin)
529 		return;
530 	for (iter = begin; iter < end; iter++)
531 		fct(tracepoint_ptr_deref(iter), priv);
532 }
533 
534 #ifdef CONFIG_MODULES
535 bool trace_module_has_bad_taint(struct module *mod)
536 {
537 	return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
538 				(1 << TAINT_UNSIGNED_MODULE) | (1 << TAINT_TEST) |
539 				(1 << TAINT_LIVEPATCH));
540 }
541 
542 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
543 
544 /**
545  * register_tracepoint_module_notifier - register tracepoint coming/going notifier
546  * @nb: notifier block
547  *
548  * Notifiers registered with this function are called on module
549  * coming/going with the tracepoint_module_list_mutex held.
550  * The notifier block callback should expect a "struct tp_module" data
551  * pointer.
552  */
553 int register_tracepoint_module_notifier(struct notifier_block *nb)
554 {
555 	struct tp_module *tp_mod;
556 	int ret;
557 
558 	mutex_lock(&tracepoint_module_list_mutex);
559 	ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
560 	if (ret)
561 		goto end;
562 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
563 		(void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
564 end:
565 	mutex_unlock(&tracepoint_module_list_mutex);
566 	return ret;
567 }
568 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
569 
570 /**
571  * unregister_tracepoint_module_notifier - unregister tracepoint coming/going notifier
572  * @nb: notifier block
573  *
574  * The notifier block callback should expect a "struct tp_module" data
575  * pointer.
576  */
577 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
578 {
579 	struct tp_module *tp_mod;
580 	int ret;
581 
582 	mutex_lock(&tracepoint_module_list_mutex);
583 	ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
584 	if (ret)
585 		goto end;
586 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
587 		(void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
588 end:
589 	mutex_unlock(&tracepoint_module_list_mutex);
590 	return ret;
591 
592 }
593 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
594 
595 /*
596  * Ensure the tracer unregistered the module's probes before the module
597  * teardown is performed. Prevents leaks of probe and data pointers.
598  */
599 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
600 {
601 	WARN_ON_ONCE(tp->funcs);
602 }
603 
604 static int tracepoint_module_coming(struct module *mod)
605 {
606 	struct tp_module *tp_mod;
607 
608 	if (!mod->num_tracepoints)
609 		return 0;
610 
611 	/*
612 	 * We skip modules that taint the kernel, especially those with different
613 	 * module headers (for forced load), to make sure we don't cause a crash.
614 	 * Staging, out-of-tree, unsigned GPL, and test modules are fine.
615 	 */
616 	if (trace_module_has_bad_taint(mod))
617 		return 0;
618 
619 	tp_mod = kmalloc_obj(struct tp_module);
620 	if (!tp_mod)
621 		return -ENOMEM;
622 	tp_mod->mod = mod;
623 
624 	mutex_lock(&tracepoint_module_list_mutex);
625 	list_add_tail(&tp_mod->list, &tracepoint_module_list);
626 	blocking_notifier_call_chain(&tracepoint_notify_list,
627 			MODULE_STATE_COMING, tp_mod);
628 	mutex_unlock(&tracepoint_module_list_mutex);
629 	return 0;
630 }
631 
632 static void tracepoint_module_going(struct module *mod)
633 {
634 	struct tp_module *tp_mod;
635 
636 	if (!mod->num_tracepoints)
637 		return;
638 
639 	mutex_lock(&tracepoint_module_list_mutex);
640 	list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
641 		if (tp_mod->mod == mod) {
642 			blocking_notifier_call_chain(&tracepoint_notify_list,
643 					MODULE_STATE_GOING, tp_mod);
644 			list_del(&tp_mod->list);
645 			kfree(tp_mod);
646 			/*
647 			 * Called the going notifier before checking for
648 			 * quiescence.
649 			 */
650 			for_each_tracepoint_range(mod->tracepoints_ptrs,
651 				mod->tracepoints_ptrs + mod->num_tracepoints,
652 				tp_module_going_check_quiescent, NULL);
653 			break;
654 		}
655 	}
656 	/*
657 	 * In the case of modules that were tainted at "coming", we'll simply
658 	 * walk through the list without finding it. We cannot use the "tainted"
659 	 * flag on "going", in case a module taints the kernel only after being
660 	 * loaded.
661 	 */
662 	mutex_unlock(&tracepoint_module_list_mutex);
663 }
664 
665 static int tracepoint_module_notify(struct notifier_block *self,
666 		unsigned long val, void *data)
667 {
668 	struct module *mod = data;
669 	int ret = 0;
670 
671 	switch (val) {
672 	case MODULE_STATE_COMING:
673 		ret = tracepoint_module_coming(mod);
674 		break;
675 	case MODULE_STATE_LIVE:
676 		break;
677 	case MODULE_STATE_GOING:
678 		tracepoint_module_going(mod);
679 		break;
680 	case MODULE_STATE_UNFORMED:
681 		break;
682 	}
683 	return notifier_from_errno(ret);
684 }
685 
686 static struct notifier_block tracepoint_module_nb = {
687 	.notifier_call = tracepoint_module_notify,
688 	.priority = 0,
689 };
690 
691 static __init int init_tracepoints(void)
692 {
693 	int ret;
694 
695 	ret = register_module_notifier(&tracepoint_module_nb);
696 	if (ret)
697 		pr_warn("Failed to register tracepoint module enter notifier\n");
698 
699 	return ret;
700 }
701 __initcall(init_tracepoints);
702 
703 /**
704  * for_each_tracepoint_in_module - iteration on all tracepoints in a module
705  * @mod: module
706  * @fct: callback
707  * @priv: private data
708  */
709 void for_each_tracepoint_in_module(struct module *mod,
710 				   void (*fct)(struct tracepoint *tp,
711 				    struct module *mod, void *priv),
712 				   void *priv)
713 {
714 	tracepoint_ptr_t *begin, *end, *iter;
715 
716 	lockdep_assert_held(&tracepoint_module_list_mutex);
717 
718 	if (!mod)
719 		return;
720 
721 	begin = mod->tracepoints_ptrs;
722 	end = mod->tracepoints_ptrs + mod->num_tracepoints;
723 
724 	for (iter = begin; iter < end; iter++)
725 		fct(tracepoint_ptr_deref(iter), mod, priv);
726 }
727 
728 /**
729  * for_each_module_tracepoint - iteration on all tracepoints in all modules
730  * @fct: callback
731  * @priv: private data
732  */
733 void for_each_module_tracepoint(void (*fct)(struct tracepoint *tp,
734 				 struct module *mod, void *priv),
735 				void *priv)
736 {
737 	struct tp_module *tp_mod;
738 
739 	mutex_lock(&tracepoint_module_list_mutex);
740 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
741 		for_each_tracepoint_in_module(tp_mod->mod, fct, priv);
742 	mutex_unlock(&tracepoint_module_list_mutex);
743 }
744 #endif /* CONFIG_MODULES */
745 
746 /**
747  * for_each_kernel_tracepoint - iteration on all kernel tracepoints
748  * @fct: callback
749  * @priv: private data
750  */
751 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
752 		void *priv)
753 {
754 	for_each_tracepoint_range(__start___tracepoints_ptrs,
755 		__stop___tracepoints_ptrs, fct, priv);
756 }
757 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
758 
759 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
760 
761 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
762 static int sys_tracepoint_refcount;
763 
764 int syscall_regfunc(void)
765 {
766 	struct task_struct *p, *t;
767 
768 	if (!sys_tracepoint_refcount) {
769 		read_lock(&tasklist_lock);
770 		for_each_process_thread(p, t) {
771 			set_task_syscall_work(t, SYSCALL_TRACEPOINT);
772 		}
773 		read_unlock(&tasklist_lock);
774 	}
775 	sys_tracepoint_refcount++;
776 
777 	return 0;
778 }
779 
780 void syscall_unregfunc(void)
781 {
782 	struct task_struct *p, *t;
783 
784 	sys_tracepoint_refcount--;
785 	if (!sys_tracepoint_refcount) {
786 		read_lock(&tasklist_lock);
787 		for_each_process_thread(p, t) {
788 			clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
789 		}
790 		read_unlock(&tasklist_lock);
791 	}
792 }
793 #endif
794