projects/linux·blob v5.12 · refs

646 lines · 16.3 KiB

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[];
54};
55
56/* Called in removal of a func but failed to allocate a new tp_funcs */
57static void tp_stub_func(void)
58{
59 return;
60}
61
62static inline void *allocate_probes(int count)
63{
64 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
65 GFP_KERNEL);
66 return p == NULL ? NULL : p->probes;
67}
68
69static void srcu_free_old_probes(struct rcu_head *head)
70{
71 kfree(container_of(head, struct tp_probes, rcu));
72}
73
74static void rcu_free_old_probes(struct rcu_head *head)
75{
76 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
77}
78
79static __init int release_early_probes(void)
80{
81 struct rcu_head *tmp;
82
83 ok_to_free_tracepoints = true;
84
85 while (early_probes) {
86 tmp = early_probes;
87 early_probes = tmp->next;
88 call_rcu(tmp, rcu_free_old_probes);
89 }
90
91 return 0;
92}
93
94/* SRCU is initialized at core_initcall */
95postcore_initcall(release_early_probes);
96
97static inline void release_probes(struct tracepoint_func *old)
98{
99 if (old) {
100 struct tp_probes *tp_probes = container_of(old,
101 struct tp_probes, probes[0]);
102
103 /*
104 * We can't free probes if SRCU is not initialized yet.
105 * Postpone the freeing till after SRCU is initialized.
106 */
107 if (unlikely(!ok_to_free_tracepoints)) {
108 tp_probes->rcu.next = early_probes;
109 early_probes = &tp_probes->rcu;
110 return;
111 }
112
113 /*
114 * Tracepoint probes are protected by both sched RCU and SRCU,
115 * by calling the SRCU callback in the sched RCU callback we
116 * cover both cases. So let us chain the SRCU and sched RCU
117 * callbacks to wait for both grace periods.
118 */
119 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
120 }
121}
122
123static void debug_print_probes(struct tracepoint_func *funcs)
124{
125 int i;
126
127 if (!tracepoint_debug || !funcs)
128 return;
129
130 for (i = 0; funcs[i].func; i++)
131 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
132}
133
134static struct tracepoint_func *
135func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
136 int prio)
137{
138 struct tracepoint_func *old, *new;
139 int iter_probes; /* Iterate over old probe array. */
140 int nr_probes = 0; /* Counter for probes */
141 int pos = -1; /* Insertion position into new array */
142
143 if (WARN_ON(!tp_func->func))
144 return ERR_PTR(-EINVAL);
145
146 debug_print_probes(*funcs);
147 old = *funcs;
148 if (old) {
149 /* (N -> N+1), (N != 0, 1) probes */
150 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
151 if (old[iter_probes].func == tp_stub_func)
152 continue; /* Skip stub functions. */
153 if (old[iter_probes].func == tp_func->func &&
154 old[iter_probes].data == tp_func->data)
155 return ERR_PTR(-EEXIST);
156 nr_probes++;
157 }
158 }
159 /* + 2 : one for new probe, one for NULL func */
160 new = allocate_probes(nr_probes + 2);
161 if (new == NULL)
162 return ERR_PTR(-ENOMEM);
163 if (old) {
164 nr_probes = 0;
165 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
166 if (old[iter_probes].func == tp_stub_func)
167 continue;
168 /* Insert before probes of lower priority */
169 if (pos < 0 && old[iter_probes].prio < prio)
170 pos = nr_probes++;
171 new[nr_probes++] = old[iter_probes];
172 }
173 if (pos < 0)
174 pos = nr_probes++;
175 /* nr_probes now points to the end of the new array */
176 } else {
177 pos = 0;
178 nr_probes = 1; /* must point at end of array */
179 }
180 new[pos] = *tp_func;
181 new[nr_probes].func = NULL;
182 *funcs = new;
183 debug_print_probes(*funcs);
184 return old;
185}
186
187static void *func_remove(struct tracepoint_func **funcs,
188 struct tracepoint_func *tp_func)
189{
190 int nr_probes = 0, nr_del = 0, i;
191 struct tracepoint_func *old, *new;
192
193 old = *funcs;
194
195 if (!old)
196 return ERR_PTR(-ENOENT);
197
198 debug_print_probes(*funcs);
199 /* (N -> M), (N > 1, M >= 0) probes */
200 if (tp_func->func) {
201 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
202 if ((old[nr_probes].func == tp_func->func &&
203 old[nr_probes].data == tp_func->data) ||
204 old[nr_probes].func == tp_stub_func)
205 nr_del++;
206 }
207 }
208
209 /*
210 * If probe is NULL, then nr_probes = nr_del = 0, and then the
211 * entire entry will be removed.
212 */
213 if (nr_probes - nr_del == 0) {
214 /* N -> 0, (N > 1) */
215 *funcs = NULL;
216 debug_print_probes(*funcs);
217 return old;
218 } else {
219 int j = 0;
220 /* N -> M, (N > 1, M > 0) */
221 /* + 1 for NULL */
222 new = allocate_probes(nr_probes - nr_del + 1);
223 if (new) {
224 for (i = 0; old[i].func; i++) {
225 if ((old[i].func != tp_func->func ||
226 old[i].data != tp_func->data) &&
227 old[i].func != tp_stub_func)
228 new[j++] = old[i];
229 }
230 new[nr_probes - nr_del].func = NULL;
231 *funcs = new;
232 } else {
233 /*
234 * Failed to allocate, replace the old function
235 * with calls to tp_stub_func.
236 */
237 for (i = 0; old[i].func; i++) {
238 if (old[i].func == tp_func->func &&
239 old[i].data == tp_func->data)
240 WRITE_ONCE(old[i].func, tp_stub_func);
241 }
242 *funcs = old;
243 }
244 }
245 debug_print_probes(*funcs);
246 return old;
247}
248
249static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs, bool sync)
250{
251 void *func = tp->iterator;
252
253 /* Synthetic events do not have static call sites */
254 if (!tp->static_call_key)
255 return;
256
257 if (!tp_funcs[1].func) {
258 func = tp_funcs[0].func;
259 /*
260 * If going from the iterator back to a single caller,
261 * we need to synchronize with __DO_TRACE to make sure
262 * that the data passed to the callback is the one that
263 * belongs to that callback.
264 */
265 if (sync)
266 tracepoint_synchronize_unregister();
267 }
268
269 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
270}
271
272/*
273 * Add the probe function to a tracepoint.
274 */
275static int tracepoint_add_func(struct tracepoint *tp,
276 struct tracepoint_func *func, int prio)
277{
278 struct tracepoint_func *old, *tp_funcs;
279 int ret;
280
281 if (tp->regfunc && !static_key_enabled(&tp->key)) {
282 ret = tp->regfunc();
283 if (ret < 0)
284 return ret;
285 }
286
287 tp_funcs = rcu_dereference_protected(tp->funcs,
288 lockdep_is_held(&tracepoints_mutex));
289 old = func_add(&tp_funcs, func, prio);
290 if (IS_ERR(old)) {
291 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
292 return PTR_ERR(old);
293 }
294
295 /*
296 * rcu_assign_pointer has as smp_store_release() which makes sure
297 * that the new probe callbacks array is consistent before setting
298 * a pointer to it. This array is referenced by __DO_TRACE from
299 * include/linux/tracepoint.h using rcu_dereference_sched().
300 */
301 rcu_assign_pointer(tp->funcs, tp_funcs);
302 tracepoint_update_call(tp, tp_funcs, false);
303 static_key_enable(&tp->key);
304
305 release_probes(old);
306 return 0;
307}
308
309/*
310 * Remove a probe function from a tracepoint.
311 * Note: only waiting an RCU period after setting elem->call to the empty
312 * function insures that the original callback is not used anymore. This insured
313 * by preempt_disable around the call site.
314 */
315static int tracepoint_remove_func(struct tracepoint *tp,
316 struct tracepoint_func *func)
317{
318 struct tracepoint_func *old, *tp_funcs;
319
320 tp_funcs = rcu_dereference_protected(tp->funcs,
321 lockdep_is_held(&tracepoints_mutex));
322 old = func_remove(&tp_funcs, func);
323 if (WARN_ON_ONCE(IS_ERR(old)))
324 return PTR_ERR(old);
325
326 if (tp_funcs == old)
327 /* Failed allocating new tp_funcs, replaced func with stub */
328 return 0;
329
330 if (!tp_funcs) {
331 /* Removed last function */
332 if (tp->unregfunc && static_key_enabled(&tp->key))
333 tp->unregfunc();
334
335 static_key_disable(&tp->key);
336 rcu_assign_pointer(tp->funcs, tp_funcs);
337 } else {
338 rcu_assign_pointer(tp->funcs, tp_funcs);
339 tracepoint_update_call(tp, tp_funcs,
340 tp_funcs[0].func != old[0].func);
341 }
342 release_probes(old);
343 return 0;
344}
345
346/**
347 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
348 * @tp: tracepoint
349 * @probe: probe handler
350 * @data: tracepoint data
351 * @prio: priority of this function over other registered functions
352 *
353 * Returns 0 if ok, error value on error.
354 * Note: if @tp is within a module, the caller is responsible for
355 * unregistering the probe before the module is gone. This can be
356 * performed either with a tracepoint module going notifier, or from
357 * within module exit functions.
358 */
359int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
360 void *data, int prio)
361{
362 struct tracepoint_func tp_func;
363 int ret;
364
365 mutex_lock(&tracepoints_mutex);
366 tp_func.func = probe;
367 tp_func.data = data;
368 tp_func.prio = prio;
369 ret = tracepoint_add_func(tp, &tp_func, prio);
370 mutex_unlock(&tracepoints_mutex);
371 return ret;
372}
373EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
374
375/**
376 * tracepoint_probe_register - Connect a probe to a tracepoint
377 * @tp: tracepoint
378 * @probe: probe handler
379 * @data: tracepoint data
380 *
381 * Returns 0 if ok, error value on error.
382 * Note: if @tp is within a module, the caller is responsible for
383 * unregistering the probe before the module is gone. This can be
384 * performed either with a tracepoint module going notifier, or from
385 * within module exit functions.
386 */
387int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
388{
389 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
390}
391EXPORT_SYMBOL_GPL(tracepoint_probe_register);
392
393/**
394 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
395 * @tp: tracepoint
396 * @probe: probe function pointer
397 * @data: tracepoint data
398 *
399 * Returns 0 if ok, error value on error.
400 */
401int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
402{
403 struct tracepoint_func tp_func;
404 int ret;
405
406 mutex_lock(&tracepoints_mutex);
407 tp_func.func = probe;
408 tp_func.data = data;
409 ret = tracepoint_remove_func(tp, &tp_func);
410 mutex_unlock(&tracepoints_mutex);
411 return ret;
412}
413EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
414
415static void for_each_tracepoint_range(
416 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
417 void (*fct)(struct tracepoint *tp, void *priv),
418 void *priv)
419{
420 tracepoint_ptr_t *iter;
421
422 if (!begin)
423 return;
424 for (iter = begin; iter < end; iter++)
425 fct(tracepoint_ptr_deref(iter), priv);
426}
427
428#ifdef CONFIG_MODULES
429bool trace_module_has_bad_taint(struct module *mod)
430{
431 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
432 (1 << TAINT_UNSIGNED_MODULE));
433}
434
435static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
436
437/**
438 * register_tracepoint_notifier - register tracepoint coming/going notifier
439 * @nb: notifier block
440 *
441 * Notifiers registered with this function are called on module
442 * coming/going with the tracepoint_module_list_mutex held.
443 * The notifier block callback should expect a "struct tp_module" data
444 * pointer.
445 */
446int register_tracepoint_module_notifier(struct notifier_block *nb)
447{
448 struct tp_module *tp_mod;
449 int ret;
450
451 mutex_lock(&tracepoint_module_list_mutex);
452 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
453 if (ret)
454 goto end;
455 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
456 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
457end:
458 mutex_unlock(&tracepoint_module_list_mutex);
459 return ret;
460}
461EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
462
463/**
464 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
465 * @nb: notifier block
466 *
467 * The notifier block callback should expect a "struct tp_module" data
468 * pointer.
469 */
470int unregister_tracepoint_module_notifier(struct notifier_block *nb)
471{
472 struct tp_module *tp_mod;
473 int ret;
474
475 mutex_lock(&tracepoint_module_list_mutex);
476 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
477 if (ret)
478 goto end;
479 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
480 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
481end:
482 mutex_unlock(&tracepoint_module_list_mutex);
483 return ret;
484
485}
486EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
487
488/*
489 * Ensure the tracer unregistered the module's probes before the module
490 * teardown is performed. Prevents leaks of probe and data pointers.
491 */
492static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
493{
494 WARN_ON_ONCE(tp->funcs);
495}
496
497static int tracepoint_module_coming(struct module *mod)
498{
499 struct tp_module *tp_mod;
500 int ret = 0;
501
502 if (!mod->num_tracepoints)
503 return 0;
504
505 /*
506 * We skip modules that taint the kernel, especially those with different
507 * module headers (for forced load), to make sure we don't cause a crash.
508 * Staging, out-of-tree, and unsigned GPL modules are fine.
509 */
510 if (trace_module_has_bad_taint(mod))
511 return 0;
512 mutex_lock(&tracepoint_module_list_mutex);
513 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
514 if (!tp_mod) {
515 ret = -ENOMEM;
516 goto end;
517 }
518 tp_mod->mod = mod;
519 list_add_tail(&tp_mod->list, &tracepoint_module_list);
520 blocking_notifier_call_chain(&tracepoint_notify_list,
521 MODULE_STATE_COMING, tp_mod);
522end:
523 mutex_unlock(&tracepoint_module_list_mutex);
524 return ret;
525}
526
527static void tracepoint_module_going(struct module *mod)
528{
529 struct tp_module *tp_mod;
530
531 if (!mod->num_tracepoints)
532 return;
533
534 mutex_lock(&tracepoint_module_list_mutex);
535 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
536 if (tp_mod->mod == mod) {
537 blocking_notifier_call_chain(&tracepoint_notify_list,
538 MODULE_STATE_GOING, tp_mod);
539 list_del(&tp_mod->list);
540 kfree(tp_mod);
541 /*
542 * Called the going notifier before checking for
543 * quiescence.
544 */
545 for_each_tracepoint_range(mod->tracepoints_ptrs,
546 mod->tracepoints_ptrs + mod->num_tracepoints,
547 tp_module_going_check_quiescent, NULL);
548 break;
549 }
550 }
551 /*
552 * In the case of modules that were tainted at "coming", we'll simply
553 * walk through the list without finding it. We cannot use the "tainted"
554 * flag on "going", in case a module taints the kernel only after being
555 * loaded.
556 */
557 mutex_unlock(&tracepoint_module_list_mutex);
558}
559
560static int tracepoint_module_notify(struct notifier_block *self,
561 unsigned long val, void *data)
562{
563 struct module *mod = data;
564 int ret = 0;
565
566 switch (val) {
567 case MODULE_STATE_COMING:
568 ret = tracepoint_module_coming(mod);
569 break;
570 case MODULE_STATE_LIVE:
571 break;
572 case MODULE_STATE_GOING:
573 tracepoint_module_going(mod);
574 break;
575 case MODULE_STATE_UNFORMED:
576 break;
577 }
578 return notifier_from_errno(ret);
579}
580
581static struct notifier_block tracepoint_module_nb = {
582 .notifier_call = tracepoint_module_notify,
583 .priority = 0,
584};
585
586static __init int init_tracepoints(void)
587{
588 int ret;
589
590 ret = register_module_notifier(&tracepoint_module_nb);
591 if (ret)
592 pr_warn("Failed to register tracepoint module enter notifier\n");
593
594 return ret;
595}
596__initcall(init_tracepoints);
597#endif /* CONFIG_MODULES */
598
599/**
600 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
601 * @fct: callback
602 * @priv: private data
603 */
604void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
605 void *priv)
606{
607 for_each_tracepoint_range(__start___tracepoints_ptrs,
608 __stop___tracepoints_ptrs, fct, priv);
609}
610EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
611
612#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
613
614/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
615static int sys_tracepoint_refcount;
616
617int syscall_regfunc(void)
618{
619 struct task_struct *p, *t;
620
621 if (!sys_tracepoint_refcount) {
622 read_lock(&tasklist_lock);
623 for_each_process_thread(p, t) {
624 set_task_syscall_work(t, SYSCALL_TRACEPOINT);
625 }
626 read_unlock(&tasklist_lock);
627 }
628 sys_tracepoint_refcount++;
629
630 return 0;
631}
632
633void syscall_unregfunc(void)
634{
635 struct task_struct *p, *t;
636
637 sys_tracepoint_refcount--;
638 if (!sys_tracepoint_refcount) {
639 read_lock(&tasklist_lock);
640 for_each_process_thread(p, t) {
641 clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
642 }
643 read_unlock(&tasklist_lock);
644 }
645}
646#endif