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// 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
18enum tp_func_state {
19 TP_FUNC_0,
20 TP_FUNC_1,
21 TP_FUNC_2,
22 TP_FUNC_N,
23};
24
25extern tracepoint_ptr_t __start___tracepoints_ptrs[];
26extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
27
28DEFINE_SRCU(tracepoint_srcu);
29EXPORT_SYMBOL_GPL(tracepoint_srcu);
30
31enum tp_transition_sync {
32 TP_TRANSITION_SYNC_1_0_1,
33 TP_TRANSITION_SYNC_N_2_1,
34
35 _NR_TP_TRANSITION_SYNC,
36};
37
38struct tp_transition_snapshot {
39 unsigned long rcu;
40 unsigned long srcu;
41 bool ongoing;
42};
43
44/* Protected by tracepoints_mutex */
45static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC];
46
47static 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 = start_poll_synchronize_srcu(&tracepoint_srcu);
54 snapshot->ongoing = true;
55}
56
57static 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))
65 synchronize_srcu(&tracepoint_srcu);
66 snapshot->ongoing = false;
67}
68
69/* Set to 1 to enable tracepoint debug output */
70static const int tracepoint_debug;
71
72#ifdef CONFIG_MODULES
73/*
74 * Tracepoint module list mutex protects the local module list.
75 */
76static DEFINE_MUTEX(tracepoint_module_list_mutex);
77
78/* Local list of struct tp_module */
79static 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 */
86static DEFINE_MUTEX(tracepoints_mutex);
87
88static struct rcu_head *early_probes;
89static bool ok_to_free_tracepoints;
90
91/*
92 * Note about RCU :
93 * It is used to delay the free of multiple probes array until a quiescent
94 * state is reached.
95 */
96struct tp_probes {
97 struct rcu_head rcu;
98 struct tracepoint_func probes[];
99};
100
101/* Called in removal of a func but failed to allocate a new tp_funcs */
102static void tp_stub_func(void)
103{
104 return;
105}
106
107static inline void *allocate_probes(int count)
108{
109 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
110 GFP_KERNEL);
111 return p == NULL ? NULL : p->probes;
112}
113
114static void srcu_free_old_probes(struct rcu_head *head)
115{
116 kfree(container_of(head, struct tp_probes, rcu));
117}
118
119static void rcu_free_old_probes(struct rcu_head *head)
120{
121 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
122}
123
124static __init int release_early_probes(void)
125{
126 struct rcu_head *tmp;
127
128 ok_to_free_tracepoints = true;
129
130 while (early_probes) {
131 tmp = early_probes;
132 early_probes = tmp->next;
133 call_rcu(tmp, rcu_free_old_probes);
134 }
135
136 return 0;
137}
138
139/* SRCU is initialized at core_initcall */
140postcore_initcall(release_early_probes);
141
142static inline void release_probes(struct tracepoint_func *old)
143{
144 if (old) {
145 struct tp_probes *tp_probes = container_of(old,
146 struct tp_probes, probes[0]);
147
148 /*
149 * We can't free probes if SRCU is not initialized yet.
150 * Postpone the freeing till after SRCU is initialized.
151 */
152 if (unlikely(!ok_to_free_tracepoints)) {
153 tp_probes->rcu.next = early_probes;
154 early_probes = &tp_probes->rcu;
155 return;
156 }
157
158 /*
159 * Tracepoint probes are protected by both sched RCU and SRCU,
160 * by calling the SRCU callback in the sched RCU callback we
161 * cover both cases. So let us chain the SRCU and sched RCU
162 * callbacks to wait for both grace periods.
163 */
164 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
165 }
166}
167
168static void debug_print_probes(struct tracepoint_func *funcs)
169{
170 int i;
171
172 if (!tracepoint_debug || !funcs)
173 return;
174
175 for (i = 0; funcs[i].func; i++)
176 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
177}
178
179static struct tracepoint_func *
180func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
181 int prio)
182{
183 struct tracepoint_func *old, *new;
184 int iter_probes; /* Iterate over old probe array. */
185 int nr_probes = 0; /* Counter for probes */
186 int pos = -1; /* Insertion position into new array */
187
188 if (WARN_ON(!tp_func->func))
189 return ERR_PTR(-EINVAL);
190
191 debug_print_probes(*funcs);
192 old = *funcs;
193 if (old) {
194 /* (N -> N+1), (N != 0, 1) probes */
195 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
196 if (old[iter_probes].func == tp_stub_func)
197 continue; /* Skip stub functions. */
198 if (old[iter_probes].func == tp_func->func &&
199 old[iter_probes].data == tp_func->data)
200 return ERR_PTR(-EEXIST);
201 nr_probes++;
202 }
203 }
204 /* + 2 : one for new probe, one for NULL func */
205 new = allocate_probes(nr_probes + 2);
206 if (new == NULL)
207 return ERR_PTR(-ENOMEM);
208 if (old) {
209 nr_probes = 0;
210 for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
211 if (old[iter_probes].func == tp_stub_func)
212 continue;
213 /* Insert before probes of lower priority */
214 if (pos < 0 && old[iter_probes].prio < prio)
215 pos = nr_probes++;
216 new[nr_probes++] = old[iter_probes];
217 }
218 if (pos < 0)
219 pos = nr_probes++;
220 /* nr_probes now points to the end of the new array */
221 } else {
222 pos = 0;
223 nr_probes = 1; /* must point at end of array */
224 }
225 new[pos] = *tp_func;
226 new[nr_probes].func = NULL;
227 *funcs = new;
228 debug_print_probes(*funcs);
229 return old;
230}
231
232static void *func_remove(struct tracepoint_func **funcs,
233 struct tracepoint_func *tp_func)
234{
235 int nr_probes = 0, nr_del = 0, i;
236 struct tracepoint_func *old, *new;
237
238 old = *funcs;
239
240 if (!old)
241 return ERR_PTR(-ENOENT);
242
243 debug_print_probes(*funcs);
244 /* (N -> M), (N > 1, M >= 0) probes */
245 if (tp_func->func) {
246 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
247 if ((old[nr_probes].func == tp_func->func &&
248 old[nr_probes].data == tp_func->data) ||
249 old[nr_probes].func == tp_stub_func)
250 nr_del++;
251 }
252 }
253
254 /*
255 * If probe is NULL, then nr_probes = nr_del = 0, and then the
256 * entire entry will be removed.
257 */
258 if (nr_probes - nr_del == 0) {
259 /* N -> 0, (N > 1) */
260 *funcs = NULL;
261 debug_print_probes(*funcs);
262 return old;
263 } else {
264 int j = 0;
265 /* N -> M, (N > 1, M > 0) */
266 /* + 1 for NULL */
267 new = allocate_probes(nr_probes - nr_del + 1);
268 if (new) {
269 for (i = 0; old[i].func; i++) {
270 if ((old[i].func != tp_func->func ||
271 old[i].data != tp_func->data) &&
272 old[i].func != tp_stub_func)
273 new[j++] = old[i];
274 }
275 new[nr_probes - nr_del].func = NULL;
276 *funcs = new;
277 } else {
278 /*
279 * Failed to allocate, replace the old function
280 * with calls to tp_stub_func.
281 */
282 for (i = 0; old[i].func; i++) {
283 if (old[i].func == tp_func->func &&
284 old[i].data == tp_func->data)
285 WRITE_ONCE(old[i].func, tp_stub_func);
286 }
287 *funcs = old;
288 }
289 }
290 debug_print_probes(*funcs);
291 return old;
292}
293
294/*
295 * Count the number of functions (enum tp_func_state) in a tp_funcs array.
296 */
297static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
298{
299 if (!tp_funcs)
300 return TP_FUNC_0;
301 if (!tp_funcs[1].func)
302 return TP_FUNC_1;
303 if (!tp_funcs[2].func)
304 return TP_FUNC_2;
305 return TP_FUNC_N; /* 3 or more */
306}
307
308static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
309{
310 void *func = tp->iterator;
311
312 /* Synthetic events do not have static call sites */
313 if (!tp->static_call_key)
314 return;
315 if (nr_func_state(tp_funcs) == TP_FUNC_1)
316 func = tp_funcs[0].func;
317 __static_call_update(tp->static_call_key, tp->static_call_tramp, func);
318}
319
320/*
321 * Add the probe function to a tracepoint.
322 */
323static int tracepoint_add_func(struct tracepoint *tp,
324 struct tracepoint_func *func, int prio,
325 bool warn)
326{
327 struct tracepoint_func *old, *tp_funcs;
328 int ret;
329
330 if (tp->regfunc && !static_key_enabled(&tp->key)) {
331 ret = tp->regfunc();
332 if (ret < 0)
333 return ret;
334 }
335
336 tp_funcs = rcu_dereference_protected(tp->funcs,
337 lockdep_is_held(&tracepoints_mutex));
338 old = func_add(&tp_funcs, func, prio);
339 if (IS_ERR(old)) {
340 WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
341 return PTR_ERR(old);
342 }
343
344 /*
345 * rcu_assign_pointer has as smp_store_release() which makes sure
346 * that the new probe callbacks array is consistent before setting
347 * a pointer to it. This array is referenced by __DO_TRACE from
348 * include/linux/tracepoint.h using rcu_dereference_sched().
349 */
350 switch (nr_func_state(tp_funcs)) {
351 case TP_FUNC_1: /* 0->1 */
352 /*
353 * Make sure new static func never uses old data after a
354 * 1->0->1 transition sequence.
355 */
356 tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
357 /* Set static call to first function */
358 tracepoint_update_call(tp, tp_funcs);
359 /* Both iterator and static call handle NULL tp->funcs */
360 rcu_assign_pointer(tp->funcs, tp_funcs);
361 static_key_enable(&tp->key);
362 break;
363 case TP_FUNC_2: /* 1->2 */
364 /* Set iterator static call */
365 tracepoint_update_call(tp, tp_funcs);
366 /*
367 * Iterator callback installed before updating tp->funcs.
368 * Requires ordering between RCU assign/dereference and
369 * static call update/call.
370 */
371 fallthrough;
372 case TP_FUNC_N: /* N->N+1 (N>1) */
373 rcu_assign_pointer(tp->funcs, tp_funcs);
374 /*
375 * Make sure static func never uses incorrect data after a
376 * N->...->2->1 (N>1) transition sequence.
377 */
378 if (tp_funcs[0].data != old[0].data)
379 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
380 break;
381 default:
382 WARN_ON_ONCE(1);
383 break;
384 }
385
386 release_probes(old);
387 return 0;
388}
389
390/*
391 * Remove a probe function from a tracepoint.
392 * Note: only waiting an RCU period after setting elem->call to the empty
393 * function insures that the original callback is not used anymore. This insured
394 * by preempt_disable around the call site.
395 */
396static int tracepoint_remove_func(struct tracepoint *tp,
397 struct tracepoint_func *func)
398{
399 struct tracepoint_func *old, *tp_funcs;
400
401 tp_funcs = rcu_dereference_protected(tp->funcs,
402 lockdep_is_held(&tracepoints_mutex));
403 old = func_remove(&tp_funcs, func);
404 if (WARN_ON_ONCE(IS_ERR(old)))
405 return PTR_ERR(old);
406
407 if (tp_funcs == old)
408 /* Failed allocating new tp_funcs, replaced func with stub */
409 return 0;
410
411 switch (nr_func_state(tp_funcs)) {
412 case TP_FUNC_0: /* 1->0 */
413 /* Removed last function */
414 if (tp->unregfunc && static_key_enabled(&tp->key))
415 tp->unregfunc();
416
417 static_key_disable(&tp->key);
418 /* Set iterator static call */
419 tracepoint_update_call(tp, tp_funcs);
420 /* Both iterator and static call handle NULL tp->funcs */
421 rcu_assign_pointer(tp->funcs, NULL);
422 /*
423 * Make sure new static func never uses old data after a
424 * 1->0->1 transition sequence.
425 */
426 tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
427 break;
428 case TP_FUNC_1: /* 2->1 */
429 rcu_assign_pointer(tp->funcs, tp_funcs);
430 /*
431 * Make sure static func never uses incorrect data after a
432 * N->...->2->1 (N>2) transition sequence. If the first
433 * element's data has changed, then force the synchronization
434 * to prevent current readers that have loaded the old data
435 * from calling the new function.
436 */
437 if (tp_funcs[0].data != old[0].data)
438 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
439 tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
440 /* Set static call to first function */
441 tracepoint_update_call(tp, tp_funcs);
442 break;
443 case TP_FUNC_2: /* N->N-1 (N>2) */
444 fallthrough;
445 case TP_FUNC_N:
446 rcu_assign_pointer(tp->funcs, tp_funcs);
447 /*
448 * Make sure static func never uses incorrect data after a
449 * N->...->2->1 (N>2) transition sequence.
450 */
451 if (tp_funcs[0].data != old[0].data)
452 tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
453 break;
454 default:
455 WARN_ON_ONCE(1);
456 break;
457 }
458 release_probes(old);
459 return 0;
460}
461
462/**
463 * tracepoint_probe_register_prio_may_exist - Connect a probe to a tracepoint with priority
464 * @tp: tracepoint
465 * @probe: probe handler
466 * @data: tracepoint data
467 * @prio: priority of this function over other registered functions
468 *
469 * Same as tracepoint_probe_register_prio() except that it will not warn
470 * if the tracepoint is already registered.
471 */
472int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
473 void *data, int prio)
474{
475 struct tracepoint_func tp_func;
476 int ret;
477
478 mutex_lock(&tracepoints_mutex);
479 tp_func.func = probe;
480 tp_func.data = data;
481 tp_func.prio = prio;
482 ret = tracepoint_add_func(tp, &tp_func, prio, false);
483 mutex_unlock(&tracepoints_mutex);
484 return ret;
485}
486EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
487
488/**
489 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
490 * @tp: tracepoint
491 * @probe: probe handler
492 * @data: tracepoint data
493 * @prio: priority of this function over other registered functions
494 *
495 * Returns 0 if ok, error value on error.
496 * Note: if @tp is within a module, the caller is responsible for
497 * unregistering the probe before the module is gone. This can be
498 * performed either with a tracepoint module going notifier, or from
499 * within module exit functions.
500 */
501int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
502 void *data, int prio)
503{
504 struct tracepoint_func tp_func;
505 int ret;
506
507 mutex_lock(&tracepoints_mutex);
508 tp_func.func = probe;
509 tp_func.data = data;
510 tp_func.prio = prio;
511 ret = tracepoint_add_func(tp, &tp_func, prio, true);
512 mutex_unlock(&tracepoints_mutex);
513 return ret;
514}
515EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
516
517/**
518 * tracepoint_probe_register - Connect a probe to a tracepoint
519 * @tp: tracepoint
520 * @probe: probe handler
521 * @data: tracepoint data
522 *
523 * Returns 0 if ok, error value on error.
524 * Note: if @tp is within a module, the caller is responsible for
525 * unregistering the probe before the module is gone. This can be
526 * performed either with a tracepoint module going notifier, or from
527 * within module exit functions.
528 */
529int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
530{
531 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
532}
533EXPORT_SYMBOL_GPL(tracepoint_probe_register);
534
535/**
536 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
537 * @tp: tracepoint
538 * @probe: probe function pointer
539 * @data: tracepoint data
540 *
541 * Returns 0 if ok, error value on error.
542 */
543int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
544{
545 struct tracepoint_func tp_func;
546 int ret;
547
548 mutex_lock(&tracepoints_mutex);
549 tp_func.func = probe;
550 tp_func.data = data;
551 ret = tracepoint_remove_func(tp, &tp_func);
552 mutex_unlock(&tracepoints_mutex);
553 return ret;
554}
555EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
556
557static void for_each_tracepoint_range(
558 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
559 void (*fct)(struct tracepoint *tp, void *priv),
560 void *priv)
561{
562 tracepoint_ptr_t *iter;
563
564 if (!begin)
565 return;
566 for (iter = begin; iter < end; iter++)
567 fct(tracepoint_ptr_deref(iter), priv);
568}
569
570#ifdef CONFIG_MODULES
571bool trace_module_has_bad_taint(struct module *mod)
572{
573 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
574 (1 << TAINT_UNSIGNED_MODULE) | (1 << TAINT_TEST) |
575 (1 << TAINT_LIVEPATCH));
576}
577
578static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
579
580/**
581 * register_tracepoint_module_notifier - register tracepoint coming/going notifier
582 * @nb: notifier block
583 *
584 * Notifiers registered with this function are called on module
585 * coming/going with the tracepoint_module_list_mutex held.
586 * The notifier block callback should expect a "struct tp_module" data
587 * pointer.
588 */
589int register_tracepoint_module_notifier(struct notifier_block *nb)
590{
591 struct tp_module *tp_mod;
592 int ret;
593
594 mutex_lock(&tracepoint_module_list_mutex);
595 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
596 if (ret)
597 goto end;
598 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
599 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
600end:
601 mutex_unlock(&tracepoint_module_list_mutex);
602 return ret;
603}
604EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
605
606/**
607 * unregister_tracepoint_module_notifier - unregister tracepoint coming/going notifier
608 * @nb: notifier block
609 *
610 * The notifier block callback should expect a "struct tp_module" data
611 * pointer.
612 */
613int unregister_tracepoint_module_notifier(struct notifier_block *nb)
614{
615 struct tp_module *tp_mod;
616 int ret;
617
618 mutex_lock(&tracepoint_module_list_mutex);
619 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
620 if (ret)
621 goto end;
622 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
623 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
624end:
625 mutex_unlock(&tracepoint_module_list_mutex);
626 return ret;
627
628}
629EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
630
631/*
632 * Ensure the tracer unregistered the module's probes before the module
633 * teardown is performed. Prevents leaks of probe and data pointers.
634 */
635static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
636{
637 WARN_ON_ONCE(tp->funcs);
638}
639
640static int tracepoint_module_coming(struct module *mod)
641{
642 struct tp_module *tp_mod;
643
644 if (!mod->num_tracepoints)
645 return 0;
646
647 /*
648 * We skip modules that taint the kernel, especially those with different
649 * module headers (for forced load), to make sure we don't cause a crash.
650 * Staging, out-of-tree, unsigned GPL, and test modules are fine.
651 */
652 if (trace_module_has_bad_taint(mod))
653 return 0;
654
655 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
656 if (!tp_mod)
657 return -ENOMEM;
658 tp_mod->mod = mod;
659
660 mutex_lock(&tracepoint_module_list_mutex);
661 list_add_tail(&tp_mod->list, &tracepoint_module_list);
662 blocking_notifier_call_chain(&tracepoint_notify_list,
663 MODULE_STATE_COMING, tp_mod);
664 mutex_unlock(&tracepoint_module_list_mutex);
665 return 0;
666}
667
668static void tracepoint_module_going(struct module *mod)
669{
670 struct tp_module *tp_mod;
671
672 if (!mod->num_tracepoints)
673 return;
674
675 mutex_lock(&tracepoint_module_list_mutex);
676 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
677 if (tp_mod->mod == mod) {
678 blocking_notifier_call_chain(&tracepoint_notify_list,
679 MODULE_STATE_GOING, tp_mod);
680 list_del(&tp_mod->list);
681 kfree(tp_mod);
682 /*
683 * Called the going notifier before checking for
684 * quiescence.
685 */
686 for_each_tracepoint_range(mod->tracepoints_ptrs,
687 mod->tracepoints_ptrs + mod->num_tracepoints,
688 tp_module_going_check_quiescent, NULL);
689 break;
690 }
691 }
692 /*
693 * In the case of modules that were tainted at "coming", we'll simply
694 * walk through the list without finding it. We cannot use the "tainted"
695 * flag on "going", in case a module taints the kernel only after being
696 * loaded.
697 */
698 mutex_unlock(&tracepoint_module_list_mutex);
699}
700
701static int tracepoint_module_notify(struct notifier_block *self,
702 unsigned long val, void *data)
703{
704 struct module *mod = data;
705 int ret = 0;
706
707 switch (val) {
708 case MODULE_STATE_COMING:
709 ret = tracepoint_module_coming(mod);
710 break;
711 case MODULE_STATE_LIVE:
712 break;
713 case MODULE_STATE_GOING:
714 tracepoint_module_going(mod);
715 break;
716 case MODULE_STATE_UNFORMED:
717 break;
718 }
719 return notifier_from_errno(ret);
720}
721
722static struct notifier_block tracepoint_module_nb = {
723 .notifier_call = tracepoint_module_notify,
724 .priority = 0,
725};
726
727static __init int init_tracepoints(void)
728{
729 int ret;
730
731 ret = register_module_notifier(&tracepoint_module_nb);
732 if (ret)
733 pr_warn("Failed to register tracepoint module enter notifier\n");
734
735 return ret;
736}
737__initcall(init_tracepoints);
738#endif /* CONFIG_MODULES */
739
740/**
741 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
742 * @fct: callback
743 * @priv: private data
744 */
745void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
746 void *priv)
747{
748 for_each_tracepoint_range(__start___tracepoints_ptrs,
749 __stop___tracepoints_ptrs, fct, priv);
750}
751EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
752
753#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
754
755/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
756static int sys_tracepoint_refcount;
757
758int syscall_regfunc(void)
759{
760 struct task_struct *p, *t;
761
762 if (!sys_tracepoint_refcount) {
763 read_lock(&tasklist_lock);
764 for_each_process_thread(p, t) {
765 set_task_syscall_work(t, SYSCALL_TRACEPOINT);
766 }
767 read_unlock(&tasklist_lock);
768 }
769 sys_tracepoint_refcount++;
770
771 return 0;
772}
773
774void syscall_unregfunc(void)
775{
776 struct task_struct *p, *t;
777
778 sys_tracepoint_refcount--;
779 if (!sys_tracepoint_refcount) {
780 read_lock(&tasklist_lock);
781 for_each_process_thread(p, t) {
782 clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
783 }
784 read_unlock(&tasklist_lock);
785 }
786}
787#endif