Loading...
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
18extern tracepoint_ptr_t __start___tracepoints_ptrs[];
19extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
20
21DEFINE_SRCU(tracepoint_srcu);
22EXPORT_SYMBOL_GPL(tracepoint_srcu);
23
24/* Set to 1 to enable tracepoint debug output */
25static const int tracepoint_debug;
26
27#ifdef CONFIG_MODULES
28/*
29 * Tracepoint module list mutex protects the local module list.
30 */
31static DEFINE_MUTEX(tracepoint_module_list_mutex);
32
33/* Local list of struct tp_module */
34static LIST_HEAD(tracepoint_module_list);
35#endif /* CONFIG_MODULES */
36
37/*
38 * tracepoints_mutex protects the builtin and module tracepoints.
39 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
40 */
41static DEFINE_MUTEX(tracepoints_mutex);
42
43static struct rcu_head *early_probes;
44static bool ok_to_free_tracepoints;
45
46/*
47 * Note about RCU :
48 * It is used to delay the free of multiple probes array until a quiescent
49 * state is reached.
50 */
51struct tp_probes {
52 struct rcu_head rcu;
53 struct tracepoint_func probes[0];
54};
55
56static inline void *allocate_probes(int count)
57{
58 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
59 GFP_KERNEL);
60 return p == NULL ? NULL : p->probes;
61}
62
63static void srcu_free_old_probes(struct rcu_head *head)
64{
65 kfree(container_of(head, struct tp_probes, rcu));
66}
67
68static void rcu_free_old_probes(struct rcu_head *head)
69{
70 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
71}
72
73static __init int release_early_probes(void)
74{
75 struct rcu_head *tmp;
76
77 ok_to_free_tracepoints = true;
78
79 while (early_probes) {
80 tmp = early_probes;
81 early_probes = tmp->next;
82 call_rcu(tmp, rcu_free_old_probes);
83 }
84
85 return 0;
86}
87
88/* SRCU is initialized at core_initcall */
89postcore_initcall(release_early_probes);
90
91static inline void release_probes(struct tracepoint_func *old)
92{
93 if (old) {
94 struct tp_probes *tp_probes = container_of(old,
95 struct tp_probes, probes[0]);
96
97 /*
98 * We can't free probes if SRCU is not initialized yet.
99 * Postpone the freeing till after SRCU is initialized.
100 */
101 if (unlikely(!ok_to_free_tracepoints)) {
102 tp_probes->rcu.next = early_probes;
103 early_probes = &tp_probes->rcu;
104 return;
105 }
106
107 /*
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.
112 */
113 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
114 }
115}
116
117static void debug_print_probes(struct tracepoint_func *funcs)
118{
119 int i;
120
121 if (!tracepoint_debug || !funcs)
122 return;
123
124 for (i = 0; funcs[i].func; i++)
125 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
126}
127
128static struct tracepoint_func *
129func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
130 int prio)
131{
132 struct tracepoint_func *old, *new;
133 int nr_probes = 0;
134 int pos = -1;
135
136 if (WARN_ON(!tp_func->func))
137 return ERR_PTR(-EINVAL);
138
139 debug_print_probes(*funcs);
140 old = *funcs;
141 if (old) {
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)
146 pos = nr_probes;
147 if (old[nr_probes].func == tp_func->func &&
148 old[nr_probes].data == tp_func->data)
149 return ERR_PTR(-EEXIST);
150 }
151 }
152 /* + 2 : one for new probe, one for NULL func */
153 new = allocate_probes(nr_probes + 2);
154 if (new == NULL)
155 return ERR_PTR(-ENOMEM);
156 if (old) {
157 if (pos < 0) {
158 pos = nr_probes;
159 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
160 } else {
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));
166 }
167 } else
168 pos = 0;
169 new[pos] = *tp_func;
170 new[nr_probes + 1].func = NULL;
171 *funcs = new;
172 debug_print_probes(*funcs);
173 return old;
174}
175
176static void *func_remove(struct tracepoint_func **funcs,
177 struct tracepoint_func *tp_func)
178{
179 int nr_probes = 0, nr_del = 0, i;
180 struct tracepoint_func *old, *new;
181
182 old = *funcs;
183
184 if (!old)
185 return ERR_PTR(-ENOENT);
186
187 debug_print_probes(*funcs);
188 /* (N -> M), (N > 1, M >= 0) probes */
189 if (tp_func->func) {
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)
193 nr_del++;
194 }
195 }
196
197 /*
198 * If probe is NULL, then nr_probes = nr_del = 0, and then the
199 * entire entry will be removed.
200 */
201 if (nr_probes - nr_del == 0) {
202 /* N -> 0, (N > 1) */
203 *funcs = NULL;
204 debug_print_probes(*funcs);
205 return old;
206 } else {
207 int j = 0;
208 /* N -> M, (N > 1, M > 0) */
209 /* + 1 for NULL */
210 new = allocate_probes(nr_probes - nr_del + 1);
211 if (new == NULL)
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)
216 new[j++] = old[i];
217 new[nr_probes - nr_del].func = NULL;
218 *funcs = new;
219 }
220 debug_print_probes(*funcs);
221 return old;
222}
223
224/*
225 * Add the probe function to a tracepoint.
226 */
227static int tracepoint_add_func(struct tracepoint *tp,
228 struct tracepoint_func *func, int prio)
229{
230 struct tracepoint_func *old, *tp_funcs;
231 int ret;
232
233 if (tp->regfunc && !static_key_enabled(&tp->key)) {
234 ret = tp->regfunc();
235 if (ret < 0)
236 return ret;
237 }
238
239 tp_funcs = rcu_dereference_protected(tp->funcs,
240 lockdep_is_held(&tracepoints_mutex));
241 old = func_add(&tp_funcs, func, prio);
242 if (IS_ERR(old)) {
243 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
244 return PTR_ERR(old);
245 }
246
247 /*
248 * rcu_assign_pointer has as smp_store_release() which makes sure
249 * that the new probe callbacks array is consistent before setting
250 * a pointer to it. This array is referenced by __DO_TRACE from
251 * include/linux/tracepoint.h using rcu_dereference_sched().
252 */
253 rcu_assign_pointer(tp->funcs, tp_funcs);
254 if (!static_key_enabled(&tp->key))
255 static_key_slow_inc(&tp->key);
256 release_probes(old);
257 return 0;
258}
259
260/*
261 * Remove a probe function from a tracepoint.
262 * Note: only waiting an RCU period after setting elem->call to the empty
263 * function insures that the original callback is not used anymore. This insured
264 * by preempt_disable around the call site.
265 */
266static int tracepoint_remove_func(struct tracepoint *tp,
267 struct tracepoint_func *func)
268{
269 struct tracepoint_func *old, *tp_funcs;
270
271 tp_funcs = rcu_dereference_protected(tp->funcs,
272 lockdep_is_held(&tracepoints_mutex));
273 old = func_remove(&tp_funcs, func);
274 if (IS_ERR(old)) {
275 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
276 return PTR_ERR(old);
277 }
278
279 if (!tp_funcs) {
280 /* Removed last function */
281 if (tp->unregfunc && static_key_enabled(&tp->key))
282 tp->unregfunc();
283
284 if (static_key_enabled(&tp->key))
285 static_key_slow_dec(&tp->key);
286 }
287 rcu_assign_pointer(tp->funcs, tp_funcs);
288 release_probes(old);
289 return 0;
290}
291
292/**
293 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
294 * @tp: tracepoint
295 * @probe: probe handler
296 * @data: tracepoint data
297 * @prio: priority of this function over other registered functions
298 *
299 * Returns 0 if ok, error value on error.
300 * Note: if @tp is within a module, the caller is responsible for
301 * unregistering the probe before the module is gone. This can be
302 * performed either with a tracepoint module going notifier, or from
303 * within module exit functions.
304 */
305int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
306 void *data, int prio)
307{
308 struct tracepoint_func tp_func;
309 int ret;
310
311 mutex_lock(&tracepoints_mutex);
312 tp_func.func = probe;
313 tp_func.data = data;
314 tp_func.prio = prio;
315 ret = tracepoint_add_func(tp, &tp_func, prio);
316 mutex_unlock(&tracepoints_mutex);
317 return ret;
318}
319EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
320
321/**
322 * tracepoint_probe_register - Connect a probe to a tracepoint
323 * @tp: tracepoint
324 * @probe: probe handler
325 * @data: tracepoint data
326 *
327 * Returns 0 if ok, error value on error.
328 * Note: if @tp is within a module, the caller is responsible for
329 * unregistering the probe before the module is gone. This can be
330 * performed either with a tracepoint module going notifier, or from
331 * within module exit functions.
332 */
333int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
334{
335 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
336}
337EXPORT_SYMBOL_GPL(tracepoint_probe_register);
338
339/**
340 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
341 * @tp: tracepoint
342 * @probe: probe function pointer
343 * @data: tracepoint data
344 *
345 * Returns 0 if ok, error value on error.
346 */
347int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
348{
349 struct tracepoint_func tp_func;
350 int ret;
351
352 mutex_lock(&tracepoints_mutex);
353 tp_func.func = probe;
354 tp_func.data = data;
355 ret = tracepoint_remove_func(tp, &tp_func);
356 mutex_unlock(&tracepoints_mutex);
357 return ret;
358}
359EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
360
361static void for_each_tracepoint_range(
362 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
363 void (*fct)(struct tracepoint *tp, void *priv),
364 void *priv)
365{
366 tracepoint_ptr_t *iter;
367
368 if (!begin)
369 return;
370 for (iter = begin; iter < end; iter++)
371 fct(tracepoint_ptr_deref(iter), priv);
372}
373
374#ifdef CONFIG_MODULES
375bool trace_module_has_bad_taint(struct module *mod)
376{
377 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
378 (1 << TAINT_UNSIGNED_MODULE));
379}
380
381static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
382
383/**
384 * register_tracepoint_notifier - register tracepoint coming/going notifier
385 * @nb: notifier block
386 *
387 * Notifiers registered with this function are called on module
388 * coming/going with the tracepoint_module_list_mutex held.
389 * The notifier block callback should expect a "struct tp_module" data
390 * pointer.
391 */
392int register_tracepoint_module_notifier(struct notifier_block *nb)
393{
394 struct tp_module *tp_mod;
395 int ret;
396
397 mutex_lock(&tracepoint_module_list_mutex);
398 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
399 if (ret)
400 goto end;
401 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
402 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
403end:
404 mutex_unlock(&tracepoint_module_list_mutex);
405 return ret;
406}
407EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
408
409/**
410 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
411 * @nb: notifier block
412 *
413 * The notifier block callback should expect a "struct tp_module" data
414 * pointer.
415 */
416int unregister_tracepoint_module_notifier(struct notifier_block *nb)
417{
418 struct tp_module *tp_mod;
419 int ret;
420
421 mutex_lock(&tracepoint_module_list_mutex);
422 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
423 if (ret)
424 goto end;
425 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
426 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
427end:
428 mutex_unlock(&tracepoint_module_list_mutex);
429 return ret;
430
431}
432EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
433
434/*
435 * Ensure the tracer unregistered the module's probes before the module
436 * teardown is performed. Prevents leaks of probe and data pointers.
437 */
438static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
439{
440 WARN_ON_ONCE(tp->funcs);
441}
442
443static int tracepoint_module_coming(struct module *mod)
444{
445 struct tp_module *tp_mod;
446 int ret = 0;
447
448 if (!mod->num_tracepoints)
449 return 0;
450
451 /*
452 * We skip modules that taint the kernel, especially those with different
453 * module headers (for forced load), to make sure we don't cause a crash.
454 * Staging, out-of-tree, and unsigned GPL modules are fine.
455 */
456 if (trace_module_has_bad_taint(mod))
457 return 0;
458 mutex_lock(&tracepoint_module_list_mutex);
459 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
460 if (!tp_mod) {
461 ret = -ENOMEM;
462 goto end;
463 }
464 tp_mod->mod = mod;
465 list_add_tail(&tp_mod->list, &tracepoint_module_list);
466 blocking_notifier_call_chain(&tracepoint_notify_list,
467 MODULE_STATE_COMING, tp_mod);
468end:
469 mutex_unlock(&tracepoint_module_list_mutex);
470 return ret;
471}
472
473static void tracepoint_module_going(struct module *mod)
474{
475 struct tp_module *tp_mod;
476
477 if (!mod->num_tracepoints)
478 return;
479
480 mutex_lock(&tracepoint_module_list_mutex);
481 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
482 if (tp_mod->mod == mod) {
483 blocking_notifier_call_chain(&tracepoint_notify_list,
484 MODULE_STATE_GOING, tp_mod);
485 list_del(&tp_mod->list);
486 kfree(tp_mod);
487 /*
488 * Called the going notifier before checking for
489 * quiescence.
490 */
491 for_each_tracepoint_range(mod->tracepoints_ptrs,
492 mod->tracepoints_ptrs + mod->num_tracepoints,
493 tp_module_going_check_quiescent, NULL);
494 break;
495 }
496 }
497 /*
498 * In the case of modules that were tainted at "coming", we'll simply
499 * walk through the list without finding it. We cannot use the "tainted"
500 * flag on "going", in case a module taints the kernel only after being
501 * loaded.
502 */
503 mutex_unlock(&tracepoint_module_list_mutex);
504}
505
506static int tracepoint_module_notify(struct notifier_block *self,
507 unsigned long val, void *data)
508{
509 struct module *mod = data;
510 int ret = 0;
511
512 switch (val) {
513 case MODULE_STATE_COMING:
514 ret = tracepoint_module_coming(mod);
515 break;
516 case MODULE_STATE_LIVE:
517 break;
518 case MODULE_STATE_GOING:
519 tracepoint_module_going(mod);
520 break;
521 case MODULE_STATE_UNFORMED:
522 break;
523 }
524 return ret;
525}
526
527static struct notifier_block tracepoint_module_nb = {
528 .notifier_call = tracepoint_module_notify,
529 .priority = 0,
530};
531
532static __init int init_tracepoints(void)
533{
534 int ret;
535
536 ret = register_module_notifier(&tracepoint_module_nb);
537 if (ret)
538 pr_warn("Failed to register tracepoint module enter notifier\n");
539
540 return ret;
541}
542__initcall(init_tracepoints);
543#endif /* CONFIG_MODULES */
544
545/**
546 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
547 * @fct: callback
548 * @priv: private data
549 */
550void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
551 void *priv)
552{
553 for_each_tracepoint_range(__start___tracepoints_ptrs,
554 __stop___tracepoints_ptrs, fct, priv);
555}
556EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
557
558#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
559
560/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
561static int sys_tracepoint_refcount;
562
563int syscall_regfunc(void)
564{
565 struct task_struct *p, *t;
566
567 if (!sys_tracepoint_refcount) {
568 read_lock(&tasklist_lock);
569 for_each_process_thread(p, t) {
570 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
571 }
572 read_unlock(&tasklist_lock);
573 }
574 sys_tracepoint_refcount++;
575
576 return 0;
577}
578
579void syscall_unregfunc(void)
580{
581 struct task_struct *p, *t;
582
583 sys_tracepoint_refcount--;
584 if (!sys_tracepoint_refcount) {
585 read_lock(&tasklist_lock);
586 for_each_process_thread(p, t) {
587 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
588 }
589 read_unlock(&tasklist_lock);
590 }
591}
592#endif
1/*
2 * Copyright (C) 2008-2014 Mathieu Desnoyers
3 *
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.
8 *
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.
13 *
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.
17 */
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>
30
31extern struct tracepoint * const __start___tracepoints_ptrs[];
32extern struct tracepoint * const __stop___tracepoints_ptrs[];
33
34/* Set to 1 to enable tracepoint debug output */
35static const int tracepoint_debug;
36
37#ifdef CONFIG_MODULES
38/*
39 * Tracepoint module list mutex protects the local module list.
40 */
41static DEFINE_MUTEX(tracepoint_module_list_mutex);
42
43/* Local list of struct tp_module */
44static LIST_HEAD(tracepoint_module_list);
45#endif /* CONFIG_MODULES */
46
47/*
48 * tracepoints_mutex protects the builtin and module tracepoints.
49 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
50 */
51static DEFINE_MUTEX(tracepoints_mutex);
52
53/*
54 * Note about RCU :
55 * It is used to delay the free of multiple probes array until a quiescent
56 * state is reached.
57 */
58struct tp_probes {
59 struct rcu_head rcu;
60 struct tracepoint_func probes[0];
61};
62
63static inline void *allocate_probes(int count)
64{
65 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
66 + sizeof(struct tp_probes), GFP_KERNEL);
67 return p == NULL ? NULL : p->probes;
68}
69
70static void rcu_free_old_probes(struct rcu_head *head)
71{
72 kfree(container_of(head, struct tp_probes, rcu));
73}
74
75static inline void release_probes(struct tracepoint_func *old)
76{
77 if (old) {
78 struct tp_probes *tp_probes = container_of(old,
79 struct tp_probes, probes[0]);
80 call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
81 }
82}
83
84static void debug_print_probes(struct tracepoint_func *funcs)
85{
86 int i;
87
88 if (!tracepoint_debug || !funcs)
89 return;
90
91 for (i = 0; funcs[i].func; i++)
92 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
93}
94
95static struct tracepoint_func *
96func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
97 int prio)
98{
99 struct tracepoint_func *old, *new;
100 int nr_probes = 0;
101 int pos = -1;
102
103 if (WARN_ON(!tp_func->func))
104 return ERR_PTR(-EINVAL);
105
106 debug_print_probes(*funcs);
107 old = *funcs;
108 if (old) {
109 /* (N -> N+1), (N != 0, 1) probes */
110 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
111 /* Insert before probes of lower priority */
112 if (pos < 0 && old[nr_probes].prio < prio)
113 pos = nr_probes;
114 if (old[nr_probes].func == tp_func->func &&
115 old[nr_probes].data == tp_func->data)
116 return ERR_PTR(-EEXIST);
117 }
118 }
119 /* + 2 : one for new probe, one for NULL func */
120 new = allocate_probes(nr_probes + 2);
121 if (new == NULL)
122 return ERR_PTR(-ENOMEM);
123 if (old) {
124 if (pos < 0) {
125 pos = nr_probes;
126 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
127 } else {
128 /* Copy higher priority probes ahead of the new probe */
129 memcpy(new, old, pos * sizeof(struct tracepoint_func));
130 /* Copy the rest after it. */
131 memcpy(new + pos + 1, old + pos,
132 (nr_probes - pos) * sizeof(struct tracepoint_func));
133 }
134 } else
135 pos = 0;
136 new[pos] = *tp_func;
137 new[nr_probes + 1].func = NULL;
138 *funcs = new;
139 debug_print_probes(*funcs);
140 return old;
141}
142
143static void *func_remove(struct tracepoint_func **funcs,
144 struct tracepoint_func *tp_func)
145{
146 int nr_probes = 0, nr_del = 0, i;
147 struct tracepoint_func *old, *new;
148
149 old = *funcs;
150
151 if (!old)
152 return ERR_PTR(-ENOENT);
153
154 debug_print_probes(*funcs);
155 /* (N -> M), (N > 1, M >= 0) probes */
156 if (tp_func->func) {
157 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
158 if (old[nr_probes].func == tp_func->func &&
159 old[nr_probes].data == tp_func->data)
160 nr_del++;
161 }
162 }
163
164 /*
165 * If probe is NULL, then nr_probes = nr_del = 0, and then the
166 * entire entry will be removed.
167 */
168 if (nr_probes - nr_del == 0) {
169 /* N -> 0, (N > 1) */
170 *funcs = NULL;
171 debug_print_probes(*funcs);
172 return old;
173 } else {
174 int j = 0;
175 /* N -> M, (N > 1, M > 0) */
176 /* + 1 for NULL */
177 new = allocate_probes(nr_probes - nr_del + 1);
178 if (new == NULL)
179 return ERR_PTR(-ENOMEM);
180 for (i = 0; old[i].func; i++)
181 if (old[i].func != tp_func->func
182 || old[i].data != tp_func->data)
183 new[j++] = old[i];
184 new[nr_probes - nr_del].func = NULL;
185 *funcs = new;
186 }
187 debug_print_probes(*funcs);
188 return old;
189}
190
191/*
192 * Add the probe function to a tracepoint.
193 */
194static int tracepoint_add_func(struct tracepoint *tp,
195 struct tracepoint_func *func, int prio)
196{
197 struct tracepoint_func *old, *tp_funcs;
198 int ret;
199
200 if (tp->regfunc && !static_key_enabled(&tp->key)) {
201 ret = tp->regfunc();
202 if (ret < 0)
203 return ret;
204 }
205
206 tp_funcs = rcu_dereference_protected(tp->funcs,
207 lockdep_is_held(&tracepoints_mutex));
208 old = func_add(&tp_funcs, func, prio);
209 if (IS_ERR(old)) {
210 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
211 return PTR_ERR(old);
212 }
213
214 /*
215 * rcu_assign_pointer has as smp_store_release() which makes sure
216 * that the new probe callbacks array is consistent before setting
217 * a pointer to it. This array is referenced by __DO_TRACE from
218 * include/linux/tracepoint.h using rcu_dereference_sched().
219 */
220 rcu_assign_pointer(tp->funcs, tp_funcs);
221 if (!static_key_enabled(&tp->key))
222 static_key_slow_inc(&tp->key);
223 release_probes(old);
224 return 0;
225}
226
227/*
228 * Remove a probe function from a tracepoint.
229 * Note: only waiting an RCU period after setting elem->call to the empty
230 * function insures that the original callback is not used anymore. This insured
231 * by preempt_disable around the call site.
232 */
233static int tracepoint_remove_func(struct tracepoint *tp,
234 struct tracepoint_func *func)
235{
236 struct tracepoint_func *old, *tp_funcs;
237
238 tp_funcs = rcu_dereference_protected(tp->funcs,
239 lockdep_is_held(&tracepoints_mutex));
240 old = func_remove(&tp_funcs, func);
241 if (IS_ERR(old)) {
242 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
243 return PTR_ERR(old);
244 }
245
246 if (!tp_funcs) {
247 /* Removed last function */
248 if (tp->unregfunc && static_key_enabled(&tp->key))
249 tp->unregfunc();
250
251 if (static_key_enabled(&tp->key))
252 static_key_slow_dec(&tp->key);
253 }
254 rcu_assign_pointer(tp->funcs, tp_funcs);
255 release_probes(old);
256 return 0;
257}
258
259/**
260 * tracepoint_probe_register - Connect a probe to a tracepoint
261 * @tp: tracepoint
262 * @probe: probe handler
263 * @data: tracepoint data
264 * @prio: priority of this function over other registered functions
265 *
266 * Returns 0 if ok, error value on error.
267 * Note: if @tp is within a module, the caller is responsible for
268 * unregistering the probe before the module is gone. This can be
269 * performed either with a tracepoint module going notifier, or from
270 * within module exit functions.
271 */
272int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
273 void *data, int prio)
274{
275 struct tracepoint_func tp_func;
276 int ret;
277
278 mutex_lock(&tracepoints_mutex);
279 tp_func.func = probe;
280 tp_func.data = data;
281 tp_func.prio = prio;
282 ret = tracepoint_add_func(tp, &tp_func, prio);
283 mutex_unlock(&tracepoints_mutex);
284 return ret;
285}
286EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
287
288/**
289 * tracepoint_probe_register - Connect a probe to a tracepoint
290 * @tp: tracepoint
291 * @probe: probe handler
292 * @data: tracepoint data
293 * @prio: priority of this function over other registered functions
294 *
295 * Returns 0 if ok, error value on error.
296 * Note: if @tp is within a module, the caller is responsible for
297 * unregistering the probe before the module is gone. This can be
298 * performed either with a tracepoint module going notifier, or from
299 * within module exit functions.
300 */
301int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
302{
303 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
304}
305EXPORT_SYMBOL_GPL(tracepoint_probe_register);
306
307/**
308 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
309 * @tp: tracepoint
310 * @probe: probe function pointer
311 * @data: tracepoint data
312 *
313 * Returns 0 if ok, error value on error.
314 */
315int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
316{
317 struct tracepoint_func tp_func;
318 int ret;
319
320 mutex_lock(&tracepoints_mutex);
321 tp_func.func = probe;
322 tp_func.data = data;
323 ret = tracepoint_remove_func(tp, &tp_func);
324 mutex_unlock(&tracepoints_mutex);
325 return ret;
326}
327EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
328
329#ifdef CONFIG_MODULES
330bool trace_module_has_bad_taint(struct module *mod)
331{
332 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
333 (1 << TAINT_UNSIGNED_MODULE));
334}
335
336static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
337
338/**
339 * register_tracepoint_notifier - register tracepoint coming/going notifier
340 * @nb: notifier block
341 *
342 * Notifiers registered with this function are called on module
343 * coming/going with the tracepoint_module_list_mutex held.
344 * The notifier block callback should expect a "struct tp_module" data
345 * pointer.
346 */
347int register_tracepoint_module_notifier(struct notifier_block *nb)
348{
349 struct tp_module *tp_mod;
350 int ret;
351
352 mutex_lock(&tracepoint_module_list_mutex);
353 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
354 if (ret)
355 goto end;
356 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
357 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
358end:
359 mutex_unlock(&tracepoint_module_list_mutex);
360 return ret;
361}
362EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
363
364/**
365 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
366 * @nb: notifier block
367 *
368 * The notifier block callback should expect a "struct tp_module" data
369 * pointer.
370 */
371int unregister_tracepoint_module_notifier(struct notifier_block *nb)
372{
373 struct tp_module *tp_mod;
374 int ret;
375
376 mutex_lock(&tracepoint_module_list_mutex);
377 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
378 if (ret)
379 goto end;
380 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
381 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
382end:
383 mutex_unlock(&tracepoint_module_list_mutex);
384 return ret;
385
386}
387EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
388
389/*
390 * Ensure the tracer unregistered the module's probes before the module
391 * teardown is performed. Prevents leaks of probe and data pointers.
392 */
393static void tp_module_going_check_quiescent(struct tracepoint * const *begin,
394 struct tracepoint * const *end)
395{
396 struct tracepoint * const *iter;
397
398 if (!begin)
399 return;
400 for (iter = begin; iter < end; iter++)
401 WARN_ON_ONCE((*iter)->funcs);
402}
403
404static int tracepoint_module_coming(struct module *mod)
405{
406 struct tp_module *tp_mod;
407 int ret = 0;
408
409 if (!mod->num_tracepoints)
410 return 0;
411
412 /*
413 * We skip modules that taint the kernel, especially those with different
414 * module headers (for forced load), to make sure we don't cause a crash.
415 * Staging, out-of-tree, and unsigned GPL modules are fine.
416 */
417 if (trace_module_has_bad_taint(mod))
418 return 0;
419 mutex_lock(&tracepoint_module_list_mutex);
420 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
421 if (!tp_mod) {
422 ret = -ENOMEM;
423 goto end;
424 }
425 tp_mod->mod = mod;
426 list_add_tail(&tp_mod->list, &tracepoint_module_list);
427 blocking_notifier_call_chain(&tracepoint_notify_list,
428 MODULE_STATE_COMING, tp_mod);
429end:
430 mutex_unlock(&tracepoint_module_list_mutex);
431 return ret;
432}
433
434static void tracepoint_module_going(struct module *mod)
435{
436 struct tp_module *tp_mod;
437
438 if (!mod->num_tracepoints)
439 return;
440
441 mutex_lock(&tracepoint_module_list_mutex);
442 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
443 if (tp_mod->mod == mod) {
444 blocking_notifier_call_chain(&tracepoint_notify_list,
445 MODULE_STATE_GOING, tp_mod);
446 list_del(&tp_mod->list);
447 kfree(tp_mod);
448 /*
449 * Called the going notifier before checking for
450 * quiescence.
451 */
452 tp_module_going_check_quiescent(mod->tracepoints_ptrs,
453 mod->tracepoints_ptrs + mod->num_tracepoints);
454 break;
455 }
456 }
457 /*
458 * In the case of modules that were tainted at "coming", we'll simply
459 * walk through the list without finding it. We cannot use the "tainted"
460 * flag on "going", in case a module taints the kernel only after being
461 * loaded.
462 */
463 mutex_unlock(&tracepoint_module_list_mutex);
464}
465
466static int tracepoint_module_notify(struct notifier_block *self,
467 unsigned long val, void *data)
468{
469 struct module *mod = data;
470 int ret = 0;
471
472 switch (val) {
473 case MODULE_STATE_COMING:
474 ret = tracepoint_module_coming(mod);
475 break;
476 case MODULE_STATE_LIVE:
477 break;
478 case MODULE_STATE_GOING:
479 tracepoint_module_going(mod);
480 break;
481 case MODULE_STATE_UNFORMED:
482 break;
483 }
484 return ret;
485}
486
487static struct notifier_block tracepoint_module_nb = {
488 .notifier_call = tracepoint_module_notify,
489 .priority = 0,
490};
491
492static __init int init_tracepoints(void)
493{
494 int ret;
495
496 ret = register_module_notifier(&tracepoint_module_nb);
497 if (ret)
498 pr_warn("Failed to register tracepoint module enter notifier\n");
499
500 return ret;
501}
502__initcall(init_tracepoints);
503#endif /* CONFIG_MODULES */
504
505static void for_each_tracepoint_range(struct tracepoint * const *begin,
506 struct tracepoint * const *end,
507 void (*fct)(struct tracepoint *tp, void *priv),
508 void *priv)
509{
510 struct tracepoint * const *iter;
511
512 if (!begin)
513 return;
514 for (iter = begin; iter < end; iter++)
515 fct(*iter, priv);
516}
517
518/**
519 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
520 * @fct: callback
521 * @priv: private data
522 */
523void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
524 void *priv)
525{
526 for_each_tracepoint_range(__start___tracepoints_ptrs,
527 __stop___tracepoints_ptrs, fct, priv);
528}
529EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
530
531#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
532
533/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
534static int sys_tracepoint_refcount;
535
536int syscall_regfunc(void)
537{
538 struct task_struct *p, *t;
539
540 if (!sys_tracepoint_refcount) {
541 read_lock(&tasklist_lock);
542 for_each_process_thread(p, t) {
543 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
544 }
545 read_unlock(&tasklist_lock);
546 }
547 sys_tracepoint_refcount++;
548
549 return 0;
550}
551
552void syscall_unregfunc(void)
553{
554 struct task_struct *p, *t;
555
556 sys_tracepoint_refcount--;
557 if (!sys_tracepoint_refcount) {
558 read_lock(&tasklist_lock);
559 for_each_process_thread(p, t) {
560 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
561 }
562 read_unlock(&tasklist_lock);
563 }
564}
565#endif