Loading...
1/*
2 * event tracer
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
8 *
9 */
10
11#define pr_fmt(fmt) fmt
12
13#include <linux/workqueue.h>
14#include <linux/spinlock.h>
15#include <linux/kthread.h>
16#include <linux/tracefs.h>
17#include <linux/uaccess.h>
18#include <linux/module.h>
19#include <linux/ctype.h>
20#include <linux/sort.h>
21#include <linux/slab.h>
22#include <linux/delay.h>
23
24#include <trace/events/sched.h>
25
26#include <asm/setup.h>
27
28#include "trace_output.h"
29
30#undef TRACE_SYSTEM
31#define TRACE_SYSTEM "TRACE_SYSTEM"
32
33DEFINE_MUTEX(event_mutex);
34
35LIST_HEAD(ftrace_events);
36static LIST_HEAD(ftrace_generic_fields);
37static LIST_HEAD(ftrace_common_fields);
38
39#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
40
41static struct kmem_cache *field_cachep;
42static struct kmem_cache *file_cachep;
43
44static inline int system_refcount(struct event_subsystem *system)
45{
46 return system->ref_count;
47}
48
49static int system_refcount_inc(struct event_subsystem *system)
50{
51 return system->ref_count++;
52}
53
54static int system_refcount_dec(struct event_subsystem *system)
55{
56 return --system->ref_count;
57}
58
59/* Double loops, do not use break, only goto's work */
60#define do_for_each_event_file(tr, file) \
61 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
62 list_for_each_entry(file, &tr->events, list)
63
64#define do_for_each_event_file_safe(tr, file) \
65 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
66 struct trace_event_file *___n; \
67 list_for_each_entry_safe(file, ___n, &tr->events, list)
68
69#define while_for_each_event_file() \
70 }
71
72static struct list_head *
73trace_get_fields(struct trace_event_call *event_call)
74{
75 if (!event_call->class->get_fields)
76 return &event_call->class->fields;
77 return event_call->class->get_fields(event_call);
78}
79
80static struct ftrace_event_field *
81__find_event_field(struct list_head *head, char *name)
82{
83 struct ftrace_event_field *field;
84
85 list_for_each_entry(field, head, link) {
86 if (!strcmp(field->name, name))
87 return field;
88 }
89
90 return NULL;
91}
92
93struct ftrace_event_field *
94trace_find_event_field(struct trace_event_call *call, char *name)
95{
96 struct ftrace_event_field *field;
97 struct list_head *head;
98
99 head = trace_get_fields(call);
100 field = __find_event_field(head, name);
101 if (field)
102 return field;
103
104 field = __find_event_field(&ftrace_generic_fields, name);
105 if (field)
106 return field;
107
108 return __find_event_field(&ftrace_common_fields, name);
109}
110
111static int __trace_define_field(struct list_head *head, const char *type,
112 const char *name, int offset, int size,
113 int is_signed, int filter_type)
114{
115 struct ftrace_event_field *field;
116
117 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
118 if (!field)
119 return -ENOMEM;
120
121 field->name = name;
122 field->type = type;
123
124 if (filter_type == FILTER_OTHER)
125 field->filter_type = filter_assign_type(type);
126 else
127 field->filter_type = filter_type;
128
129 field->offset = offset;
130 field->size = size;
131 field->is_signed = is_signed;
132
133 list_add(&field->link, head);
134
135 return 0;
136}
137
138int trace_define_field(struct trace_event_call *call, const char *type,
139 const char *name, int offset, int size, int is_signed,
140 int filter_type)
141{
142 struct list_head *head;
143
144 if (WARN_ON(!call->class))
145 return 0;
146
147 head = trace_get_fields(call);
148 return __trace_define_field(head, type, name, offset, size,
149 is_signed, filter_type);
150}
151EXPORT_SYMBOL_GPL(trace_define_field);
152
153#define __generic_field(type, item, filter_type) \
154 ret = __trace_define_field(&ftrace_generic_fields, #type, \
155 #item, 0, 0, is_signed_type(type), \
156 filter_type); \
157 if (ret) \
158 return ret;
159
160#define __common_field(type, item) \
161 ret = __trace_define_field(&ftrace_common_fields, #type, \
162 "common_" #item, \
163 offsetof(typeof(ent), item), \
164 sizeof(ent.item), \
165 is_signed_type(type), FILTER_OTHER); \
166 if (ret) \
167 return ret;
168
169static int trace_define_generic_fields(void)
170{
171 int ret;
172
173 __generic_field(int, CPU, FILTER_CPU);
174 __generic_field(int, cpu, FILTER_CPU);
175 __generic_field(char *, COMM, FILTER_COMM);
176 __generic_field(char *, comm, FILTER_COMM);
177
178 return ret;
179}
180
181static int trace_define_common_fields(void)
182{
183 int ret;
184 struct trace_entry ent;
185
186 __common_field(unsigned short, type);
187 __common_field(unsigned char, flags);
188 __common_field(unsigned char, preempt_count);
189 __common_field(int, pid);
190
191 return ret;
192}
193
194static void trace_destroy_fields(struct trace_event_call *call)
195{
196 struct ftrace_event_field *field, *next;
197 struct list_head *head;
198
199 head = trace_get_fields(call);
200 list_for_each_entry_safe(field, next, head, link) {
201 list_del(&field->link);
202 kmem_cache_free(field_cachep, field);
203 }
204}
205
206/*
207 * run-time version of trace_event_get_offsets_<call>() that returns the last
208 * accessible offset of trace fields excluding __dynamic_array bytes
209 */
210int trace_event_get_offsets(struct trace_event_call *call)
211{
212 struct ftrace_event_field *tail;
213 struct list_head *head;
214
215 head = trace_get_fields(call);
216 /*
217 * head->next points to the last field with the largest offset,
218 * since it was added last by trace_define_field()
219 */
220 tail = list_first_entry(head, struct ftrace_event_field, link);
221 return tail->offset + tail->size;
222}
223
224int trace_event_raw_init(struct trace_event_call *call)
225{
226 int id;
227
228 id = register_trace_event(&call->event);
229 if (!id)
230 return -ENODEV;
231
232 return 0;
233}
234EXPORT_SYMBOL_GPL(trace_event_raw_init);
235
236bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
237{
238 struct trace_array *tr = trace_file->tr;
239 struct trace_array_cpu *data;
240 struct trace_pid_list *pid_list;
241
242 pid_list = rcu_dereference_sched(tr->filtered_pids);
243 if (!pid_list)
244 return false;
245
246 data = this_cpu_ptr(tr->trace_buffer.data);
247
248 return data->ignore_pid;
249}
250EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
251
252void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
253 struct trace_event_file *trace_file,
254 unsigned long len)
255{
256 struct trace_event_call *event_call = trace_file->event_call;
257
258 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
259 trace_event_ignore_this_pid(trace_file))
260 return NULL;
261
262 local_save_flags(fbuffer->flags);
263 fbuffer->pc = preempt_count();
264 /*
265 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
266 * preemption (adding one to the preempt_count). Since we are
267 * interested in the preempt_count at the time the tracepoint was
268 * hit, we need to subtract one to offset the increment.
269 */
270 if (IS_ENABLED(CONFIG_PREEMPT))
271 fbuffer->pc--;
272 fbuffer->trace_file = trace_file;
273
274 fbuffer->event =
275 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
276 event_call->event.type, len,
277 fbuffer->flags, fbuffer->pc);
278 if (!fbuffer->event)
279 return NULL;
280
281 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
282 return fbuffer->entry;
283}
284EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
285
286int trace_event_reg(struct trace_event_call *call,
287 enum trace_reg type, void *data)
288{
289 struct trace_event_file *file = data;
290
291 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
292 switch (type) {
293 case TRACE_REG_REGISTER:
294 return tracepoint_probe_register(call->tp,
295 call->class->probe,
296 file);
297 case TRACE_REG_UNREGISTER:
298 tracepoint_probe_unregister(call->tp,
299 call->class->probe,
300 file);
301 return 0;
302
303#ifdef CONFIG_PERF_EVENTS
304 case TRACE_REG_PERF_REGISTER:
305 return tracepoint_probe_register(call->tp,
306 call->class->perf_probe,
307 call);
308 case TRACE_REG_PERF_UNREGISTER:
309 tracepoint_probe_unregister(call->tp,
310 call->class->perf_probe,
311 call);
312 return 0;
313 case TRACE_REG_PERF_OPEN:
314 case TRACE_REG_PERF_CLOSE:
315 case TRACE_REG_PERF_ADD:
316 case TRACE_REG_PERF_DEL:
317 return 0;
318#endif
319 }
320 return 0;
321}
322EXPORT_SYMBOL_GPL(trace_event_reg);
323
324void trace_event_enable_cmd_record(bool enable)
325{
326 struct trace_event_file *file;
327 struct trace_array *tr;
328
329 mutex_lock(&event_mutex);
330 do_for_each_event_file(tr, file) {
331
332 if (!(file->flags & EVENT_FILE_FL_ENABLED))
333 continue;
334
335 if (enable) {
336 tracing_start_cmdline_record();
337 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
338 } else {
339 tracing_stop_cmdline_record();
340 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
341 }
342 } while_for_each_event_file();
343 mutex_unlock(&event_mutex);
344}
345
346static int __ftrace_event_enable_disable(struct trace_event_file *file,
347 int enable, int soft_disable)
348{
349 struct trace_event_call *call = file->event_call;
350 struct trace_array *tr = file->tr;
351 unsigned long file_flags = file->flags;
352 int ret = 0;
353 int disable;
354
355 switch (enable) {
356 case 0:
357 /*
358 * When soft_disable is set and enable is cleared, the sm_ref
359 * reference counter is decremented. If it reaches 0, we want
360 * to clear the SOFT_DISABLED flag but leave the event in the
361 * state that it was. That is, if the event was enabled and
362 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
363 * is set we do not want the event to be enabled before we
364 * clear the bit.
365 *
366 * When soft_disable is not set but the SOFT_MODE flag is,
367 * we do nothing. Do not disable the tracepoint, otherwise
368 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
369 */
370 if (soft_disable) {
371 if (atomic_dec_return(&file->sm_ref) > 0)
372 break;
373 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
374 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
375 } else
376 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
377
378 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
379 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
380 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
381 tracing_stop_cmdline_record();
382 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
383 }
384 call->class->reg(call, TRACE_REG_UNREGISTER, file);
385 }
386 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
387 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
388 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
389 else
390 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
391 break;
392 case 1:
393 /*
394 * When soft_disable is set and enable is set, we want to
395 * register the tracepoint for the event, but leave the event
396 * as is. That means, if the event was already enabled, we do
397 * nothing (but set SOFT_MODE). If the event is disabled, we
398 * set SOFT_DISABLED before enabling the event tracepoint, so
399 * it still seems to be disabled.
400 */
401 if (!soft_disable)
402 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
403 else {
404 if (atomic_inc_return(&file->sm_ref) > 1)
405 break;
406 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
407 }
408
409 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
410
411 /* Keep the event disabled, when going to SOFT_MODE. */
412 if (soft_disable)
413 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
414
415 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
416 tracing_start_cmdline_record();
417 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
418 }
419 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
420 if (ret) {
421 tracing_stop_cmdline_record();
422 pr_info("event trace: Could not enable event "
423 "%s\n", trace_event_name(call));
424 break;
425 }
426 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
427
428 /* WAS_ENABLED gets set but never cleared. */
429 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
430 }
431 break;
432 }
433
434 /* Enable or disable use of trace_buffered_event */
435 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
436 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
437 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
438 trace_buffered_event_enable();
439 else
440 trace_buffered_event_disable();
441 }
442
443 return ret;
444}
445
446int trace_event_enable_disable(struct trace_event_file *file,
447 int enable, int soft_disable)
448{
449 return __ftrace_event_enable_disable(file, enable, soft_disable);
450}
451
452static int ftrace_event_enable_disable(struct trace_event_file *file,
453 int enable)
454{
455 return __ftrace_event_enable_disable(file, enable, 0);
456}
457
458static void ftrace_clear_events(struct trace_array *tr)
459{
460 struct trace_event_file *file;
461
462 mutex_lock(&event_mutex);
463 list_for_each_entry(file, &tr->events, list) {
464 ftrace_event_enable_disable(file, 0);
465 }
466 mutex_unlock(&event_mutex);
467}
468
469static void
470event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
471{
472 struct trace_pid_list *pid_list;
473 struct trace_array *tr = data;
474
475 pid_list = rcu_dereference_sched(tr->filtered_pids);
476 trace_filter_add_remove_task(pid_list, NULL, task);
477}
478
479static void
480event_filter_pid_sched_process_fork(void *data,
481 struct task_struct *self,
482 struct task_struct *task)
483{
484 struct trace_pid_list *pid_list;
485 struct trace_array *tr = data;
486
487 pid_list = rcu_dereference_sched(tr->filtered_pids);
488 trace_filter_add_remove_task(pid_list, self, task);
489}
490
491void trace_event_follow_fork(struct trace_array *tr, bool enable)
492{
493 if (enable) {
494 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
495 tr, INT_MIN);
496 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
497 tr, INT_MAX);
498 } else {
499 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
500 tr);
501 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
502 tr);
503 }
504}
505
506static void
507event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
508 struct task_struct *prev, struct task_struct *next)
509{
510 struct trace_array *tr = data;
511 struct trace_pid_list *pid_list;
512
513 pid_list = rcu_dereference_sched(tr->filtered_pids);
514
515 this_cpu_write(tr->trace_buffer.data->ignore_pid,
516 trace_ignore_this_task(pid_list, prev) &&
517 trace_ignore_this_task(pid_list, next));
518}
519
520static void
521event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
522 struct task_struct *prev, struct task_struct *next)
523{
524 struct trace_array *tr = data;
525 struct trace_pid_list *pid_list;
526
527 pid_list = rcu_dereference_sched(tr->filtered_pids);
528
529 this_cpu_write(tr->trace_buffer.data->ignore_pid,
530 trace_ignore_this_task(pid_list, next));
531}
532
533static void
534event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
535{
536 struct trace_array *tr = data;
537 struct trace_pid_list *pid_list;
538
539 /* Nothing to do if we are already tracing */
540 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
541 return;
542
543 pid_list = rcu_dereference_sched(tr->filtered_pids);
544
545 this_cpu_write(tr->trace_buffer.data->ignore_pid,
546 trace_ignore_this_task(pid_list, task));
547}
548
549static void
550event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
551{
552 struct trace_array *tr = data;
553 struct trace_pid_list *pid_list;
554
555 /* Nothing to do if we are not tracing */
556 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
557 return;
558
559 pid_list = rcu_dereference_sched(tr->filtered_pids);
560
561 /* Set tracing if current is enabled */
562 this_cpu_write(tr->trace_buffer.data->ignore_pid,
563 trace_ignore_this_task(pid_list, current));
564}
565
566static void __ftrace_clear_event_pids(struct trace_array *tr)
567{
568 struct trace_pid_list *pid_list;
569 struct trace_event_file *file;
570 int cpu;
571
572 pid_list = rcu_dereference_protected(tr->filtered_pids,
573 lockdep_is_held(&event_mutex));
574 if (!pid_list)
575 return;
576
577 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
578 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
579
580 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
581 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
582
583 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
584 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
585
586 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
587 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
588
589 list_for_each_entry(file, &tr->events, list) {
590 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
591 }
592
593 for_each_possible_cpu(cpu)
594 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
595
596 rcu_assign_pointer(tr->filtered_pids, NULL);
597
598 /* Wait till all users are no longer using pid filtering */
599 synchronize_sched();
600
601 trace_free_pid_list(pid_list);
602}
603
604static void ftrace_clear_event_pids(struct trace_array *tr)
605{
606 mutex_lock(&event_mutex);
607 __ftrace_clear_event_pids(tr);
608 mutex_unlock(&event_mutex);
609}
610
611static void __put_system(struct event_subsystem *system)
612{
613 struct event_filter *filter = system->filter;
614
615 WARN_ON_ONCE(system_refcount(system) == 0);
616 if (system_refcount_dec(system))
617 return;
618
619 list_del(&system->list);
620
621 if (filter) {
622 kfree(filter->filter_string);
623 kfree(filter);
624 }
625 kfree_const(system->name);
626 kfree(system);
627}
628
629static void __get_system(struct event_subsystem *system)
630{
631 WARN_ON_ONCE(system_refcount(system) == 0);
632 system_refcount_inc(system);
633}
634
635static void __get_system_dir(struct trace_subsystem_dir *dir)
636{
637 WARN_ON_ONCE(dir->ref_count == 0);
638 dir->ref_count++;
639 __get_system(dir->subsystem);
640}
641
642static void __put_system_dir(struct trace_subsystem_dir *dir)
643{
644 WARN_ON_ONCE(dir->ref_count == 0);
645 /* If the subsystem is about to be freed, the dir must be too */
646 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
647
648 __put_system(dir->subsystem);
649 if (!--dir->ref_count)
650 kfree(dir);
651}
652
653static void put_system(struct trace_subsystem_dir *dir)
654{
655 mutex_lock(&event_mutex);
656 __put_system_dir(dir);
657 mutex_unlock(&event_mutex);
658}
659
660static void remove_subsystem(struct trace_subsystem_dir *dir)
661{
662 if (!dir)
663 return;
664
665 if (!--dir->nr_events) {
666 tracefs_remove_recursive(dir->entry);
667 list_del(&dir->list);
668 __put_system_dir(dir);
669 }
670}
671
672static void remove_event_file_dir(struct trace_event_file *file)
673{
674 struct dentry *dir = file->dir;
675 struct dentry *child;
676
677 if (dir) {
678 spin_lock(&dir->d_lock); /* probably unneeded */
679 list_for_each_entry(child, &dir->d_subdirs, d_child) {
680 if (d_really_is_positive(child)) /* probably unneeded */
681 d_inode(child)->i_private = NULL;
682 }
683 spin_unlock(&dir->d_lock);
684
685 tracefs_remove_recursive(dir);
686 }
687
688 list_del(&file->list);
689 remove_subsystem(file->system);
690 free_event_filter(file->filter);
691 kmem_cache_free(file_cachep, file);
692}
693
694/*
695 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
696 */
697static int
698__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
699 const char *sub, const char *event, int set)
700{
701 struct trace_event_file *file;
702 struct trace_event_call *call;
703 const char *name;
704 int ret = -EINVAL;
705 int eret = 0;
706
707 list_for_each_entry(file, &tr->events, list) {
708
709 call = file->event_call;
710 name = trace_event_name(call);
711
712 if (!name || !call->class || !call->class->reg)
713 continue;
714
715 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
716 continue;
717
718 if (match &&
719 strcmp(match, name) != 0 &&
720 strcmp(match, call->class->system) != 0)
721 continue;
722
723 if (sub && strcmp(sub, call->class->system) != 0)
724 continue;
725
726 if (event && strcmp(event, name) != 0)
727 continue;
728
729 ret = ftrace_event_enable_disable(file, set);
730
731 /*
732 * Save the first error and return that. Some events
733 * may still have been enabled, but let the user
734 * know that something went wrong.
735 */
736 if (ret && !eret)
737 eret = ret;
738
739 ret = eret;
740 }
741
742 return ret;
743}
744
745static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
746 const char *sub, const char *event, int set)
747{
748 int ret;
749
750 mutex_lock(&event_mutex);
751 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
752 mutex_unlock(&event_mutex);
753
754 return ret;
755}
756
757static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
758{
759 char *event = NULL, *sub = NULL, *match;
760 int ret;
761
762 /*
763 * The buf format can be <subsystem>:<event-name>
764 * *:<event-name> means any event by that name.
765 * :<event-name> is the same.
766 *
767 * <subsystem>:* means all events in that subsystem
768 * <subsystem>: means the same.
769 *
770 * <name> (no ':') means all events in a subsystem with
771 * the name <name> or any event that matches <name>
772 */
773
774 match = strsep(&buf, ":");
775 if (buf) {
776 sub = match;
777 event = buf;
778 match = NULL;
779
780 if (!strlen(sub) || strcmp(sub, "*") == 0)
781 sub = NULL;
782 if (!strlen(event) || strcmp(event, "*") == 0)
783 event = NULL;
784 }
785
786 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
787
788 /* Put back the colon to allow this to be called again */
789 if (buf)
790 *(buf - 1) = ':';
791
792 return ret;
793}
794
795/**
796 * trace_set_clr_event - enable or disable an event
797 * @system: system name to match (NULL for any system)
798 * @event: event name to match (NULL for all events, within system)
799 * @set: 1 to enable, 0 to disable
800 *
801 * This is a way for other parts of the kernel to enable or disable
802 * event recording.
803 *
804 * Returns 0 on success, -EINVAL if the parameters do not match any
805 * registered events.
806 */
807int trace_set_clr_event(const char *system, const char *event, int set)
808{
809 struct trace_array *tr = top_trace_array();
810
811 if (!tr)
812 return -ENODEV;
813
814 return __ftrace_set_clr_event(tr, NULL, system, event, set);
815}
816EXPORT_SYMBOL_GPL(trace_set_clr_event);
817
818/* 128 should be much more than enough */
819#define EVENT_BUF_SIZE 127
820
821static ssize_t
822ftrace_event_write(struct file *file, const char __user *ubuf,
823 size_t cnt, loff_t *ppos)
824{
825 struct trace_parser parser;
826 struct seq_file *m = file->private_data;
827 struct trace_array *tr = m->private;
828 ssize_t read, ret;
829
830 if (!cnt)
831 return 0;
832
833 ret = tracing_update_buffers();
834 if (ret < 0)
835 return ret;
836
837 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
838 return -ENOMEM;
839
840 read = trace_get_user(&parser, ubuf, cnt, ppos);
841
842 if (read >= 0 && trace_parser_loaded((&parser))) {
843 int set = 1;
844
845 if (*parser.buffer == '!')
846 set = 0;
847
848 parser.buffer[parser.idx] = 0;
849
850 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
851 if (ret)
852 goto out_put;
853 }
854
855 ret = read;
856
857 out_put:
858 trace_parser_put(&parser);
859
860 return ret;
861}
862
863static void *
864t_next(struct seq_file *m, void *v, loff_t *pos)
865{
866 struct trace_event_file *file = v;
867 struct trace_event_call *call;
868 struct trace_array *tr = m->private;
869
870 (*pos)++;
871
872 list_for_each_entry_continue(file, &tr->events, list) {
873 call = file->event_call;
874 /*
875 * The ftrace subsystem is for showing formats only.
876 * They can not be enabled or disabled via the event files.
877 */
878 if (call->class && call->class->reg &&
879 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
880 return file;
881 }
882
883 return NULL;
884}
885
886static void *t_start(struct seq_file *m, loff_t *pos)
887{
888 struct trace_event_file *file;
889 struct trace_array *tr = m->private;
890 loff_t l;
891
892 mutex_lock(&event_mutex);
893
894 file = list_entry(&tr->events, struct trace_event_file, list);
895 for (l = 0; l <= *pos; ) {
896 file = t_next(m, file, &l);
897 if (!file)
898 break;
899 }
900 return file;
901}
902
903static void *
904s_next(struct seq_file *m, void *v, loff_t *pos)
905{
906 struct trace_event_file *file = v;
907 struct trace_array *tr = m->private;
908
909 (*pos)++;
910
911 list_for_each_entry_continue(file, &tr->events, list) {
912 if (file->flags & EVENT_FILE_FL_ENABLED)
913 return file;
914 }
915
916 return NULL;
917}
918
919static void *s_start(struct seq_file *m, loff_t *pos)
920{
921 struct trace_event_file *file;
922 struct trace_array *tr = m->private;
923 loff_t l;
924
925 mutex_lock(&event_mutex);
926
927 file = list_entry(&tr->events, struct trace_event_file, list);
928 for (l = 0; l <= *pos; ) {
929 file = s_next(m, file, &l);
930 if (!file)
931 break;
932 }
933 return file;
934}
935
936static int t_show(struct seq_file *m, void *v)
937{
938 struct trace_event_file *file = v;
939 struct trace_event_call *call = file->event_call;
940
941 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
942 seq_printf(m, "%s:", call->class->system);
943 seq_printf(m, "%s\n", trace_event_name(call));
944
945 return 0;
946}
947
948static void t_stop(struct seq_file *m, void *p)
949{
950 mutex_unlock(&event_mutex);
951}
952
953static void *
954p_next(struct seq_file *m, void *v, loff_t *pos)
955{
956 struct trace_array *tr = m->private;
957 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
958
959 return trace_pid_next(pid_list, v, pos);
960}
961
962static void *p_start(struct seq_file *m, loff_t *pos)
963 __acquires(RCU)
964{
965 struct trace_pid_list *pid_list;
966 struct trace_array *tr = m->private;
967
968 /*
969 * Grab the mutex, to keep calls to p_next() having the same
970 * tr->filtered_pids as p_start() has.
971 * If we just passed the tr->filtered_pids around, then RCU would
972 * have been enough, but doing that makes things more complex.
973 */
974 mutex_lock(&event_mutex);
975 rcu_read_lock_sched();
976
977 pid_list = rcu_dereference_sched(tr->filtered_pids);
978
979 if (!pid_list)
980 return NULL;
981
982 return trace_pid_start(pid_list, pos);
983}
984
985static void p_stop(struct seq_file *m, void *p)
986 __releases(RCU)
987{
988 rcu_read_unlock_sched();
989 mutex_unlock(&event_mutex);
990}
991
992static ssize_t
993event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
994 loff_t *ppos)
995{
996 struct trace_event_file *file;
997 unsigned long flags;
998 char buf[4] = "0";
999
1000 mutex_lock(&event_mutex);
1001 file = event_file_data(filp);
1002 if (likely(file))
1003 flags = file->flags;
1004 mutex_unlock(&event_mutex);
1005
1006 if (!file)
1007 return -ENODEV;
1008
1009 if (flags & EVENT_FILE_FL_ENABLED &&
1010 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1011 strcpy(buf, "1");
1012
1013 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1014 flags & EVENT_FILE_FL_SOFT_MODE)
1015 strcat(buf, "*");
1016
1017 strcat(buf, "\n");
1018
1019 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1020}
1021
1022static ssize_t
1023event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1024 loff_t *ppos)
1025{
1026 struct trace_event_file *file;
1027 unsigned long val;
1028 int ret;
1029
1030 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1031 if (ret)
1032 return ret;
1033
1034 ret = tracing_update_buffers();
1035 if (ret < 0)
1036 return ret;
1037
1038 switch (val) {
1039 case 0:
1040 case 1:
1041 ret = -ENODEV;
1042 mutex_lock(&event_mutex);
1043 file = event_file_data(filp);
1044 if (likely(file))
1045 ret = ftrace_event_enable_disable(file, val);
1046 mutex_unlock(&event_mutex);
1047 break;
1048
1049 default:
1050 return -EINVAL;
1051 }
1052
1053 *ppos += cnt;
1054
1055 return ret ? ret : cnt;
1056}
1057
1058static ssize_t
1059system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1060 loff_t *ppos)
1061{
1062 const char set_to_char[4] = { '?', '0', '1', 'X' };
1063 struct trace_subsystem_dir *dir = filp->private_data;
1064 struct event_subsystem *system = dir->subsystem;
1065 struct trace_event_call *call;
1066 struct trace_event_file *file;
1067 struct trace_array *tr = dir->tr;
1068 char buf[2];
1069 int set = 0;
1070 int ret;
1071
1072 mutex_lock(&event_mutex);
1073 list_for_each_entry(file, &tr->events, list) {
1074 call = file->event_call;
1075 if (!trace_event_name(call) || !call->class || !call->class->reg)
1076 continue;
1077
1078 if (system && strcmp(call->class->system, system->name) != 0)
1079 continue;
1080
1081 /*
1082 * We need to find out if all the events are set
1083 * or if all events or cleared, or if we have
1084 * a mixture.
1085 */
1086 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1087
1088 /*
1089 * If we have a mixture, no need to look further.
1090 */
1091 if (set == 3)
1092 break;
1093 }
1094 mutex_unlock(&event_mutex);
1095
1096 buf[0] = set_to_char[set];
1097 buf[1] = '\n';
1098
1099 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1100
1101 return ret;
1102}
1103
1104static ssize_t
1105system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1106 loff_t *ppos)
1107{
1108 struct trace_subsystem_dir *dir = filp->private_data;
1109 struct event_subsystem *system = dir->subsystem;
1110 const char *name = NULL;
1111 unsigned long val;
1112 ssize_t ret;
1113
1114 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1115 if (ret)
1116 return ret;
1117
1118 ret = tracing_update_buffers();
1119 if (ret < 0)
1120 return ret;
1121
1122 if (val != 0 && val != 1)
1123 return -EINVAL;
1124
1125 /*
1126 * Opening of "enable" adds a ref count to system,
1127 * so the name is safe to use.
1128 */
1129 if (system)
1130 name = system->name;
1131
1132 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1133 if (ret)
1134 goto out;
1135
1136 ret = cnt;
1137
1138out:
1139 *ppos += cnt;
1140
1141 return ret;
1142}
1143
1144enum {
1145 FORMAT_HEADER = 1,
1146 FORMAT_FIELD_SEPERATOR = 2,
1147 FORMAT_PRINTFMT = 3,
1148};
1149
1150static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1151{
1152 struct trace_event_call *call = event_file_data(m->private);
1153 struct list_head *common_head = &ftrace_common_fields;
1154 struct list_head *head = trace_get_fields(call);
1155 struct list_head *node = v;
1156
1157 (*pos)++;
1158
1159 switch ((unsigned long)v) {
1160 case FORMAT_HEADER:
1161 node = common_head;
1162 break;
1163
1164 case FORMAT_FIELD_SEPERATOR:
1165 node = head;
1166 break;
1167
1168 case FORMAT_PRINTFMT:
1169 /* all done */
1170 return NULL;
1171 }
1172
1173 node = node->prev;
1174 if (node == common_head)
1175 return (void *)FORMAT_FIELD_SEPERATOR;
1176 else if (node == head)
1177 return (void *)FORMAT_PRINTFMT;
1178 else
1179 return node;
1180}
1181
1182static int f_show(struct seq_file *m, void *v)
1183{
1184 struct trace_event_call *call = event_file_data(m->private);
1185 struct ftrace_event_field *field;
1186 const char *array_descriptor;
1187
1188 switch ((unsigned long)v) {
1189 case FORMAT_HEADER:
1190 seq_printf(m, "name: %s\n", trace_event_name(call));
1191 seq_printf(m, "ID: %d\n", call->event.type);
1192 seq_puts(m, "format:\n");
1193 return 0;
1194
1195 case FORMAT_FIELD_SEPERATOR:
1196 seq_putc(m, '\n');
1197 return 0;
1198
1199 case FORMAT_PRINTFMT:
1200 seq_printf(m, "\nprint fmt: %s\n",
1201 call->print_fmt);
1202 return 0;
1203 }
1204
1205 field = list_entry(v, struct ftrace_event_field, link);
1206 /*
1207 * Smartly shows the array type(except dynamic array).
1208 * Normal:
1209 * field:TYPE VAR
1210 * If TYPE := TYPE[LEN], it is shown:
1211 * field:TYPE VAR[LEN]
1212 */
1213 array_descriptor = strchr(field->type, '[');
1214
1215 if (!strncmp(field->type, "__data_loc", 10))
1216 array_descriptor = NULL;
1217
1218 if (!array_descriptor)
1219 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1220 field->type, field->name, field->offset,
1221 field->size, !!field->is_signed);
1222 else
1223 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1224 (int)(array_descriptor - field->type),
1225 field->type, field->name,
1226 array_descriptor, field->offset,
1227 field->size, !!field->is_signed);
1228
1229 return 0;
1230}
1231
1232static void *f_start(struct seq_file *m, loff_t *pos)
1233{
1234 void *p = (void *)FORMAT_HEADER;
1235 loff_t l = 0;
1236
1237 /* ->stop() is called even if ->start() fails */
1238 mutex_lock(&event_mutex);
1239 if (!event_file_data(m->private))
1240 return ERR_PTR(-ENODEV);
1241
1242 while (l < *pos && p)
1243 p = f_next(m, p, &l);
1244
1245 return p;
1246}
1247
1248static void f_stop(struct seq_file *m, void *p)
1249{
1250 mutex_unlock(&event_mutex);
1251}
1252
1253static const struct seq_operations trace_format_seq_ops = {
1254 .start = f_start,
1255 .next = f_next,
1256 .stop = f_stop,
1257 .show = f_show,
1258};
1259
1260static int trace_format_open(struct inode *inode, struct file *file)
1261{
1262 struct seq_file *m;
1263 int ret;
1264
1265 ret = seq_open(file, &trace_format_seq_ops);
1266 if (ret < 0)
1267 return ret;
1268
1269 m = file->private_data;
1270 m->private = file;
1271
1272 return 0;
1273}
1274
1275static ssize_t
1276event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1277{
1278 int id = (long)event_file_data(filp);
1279 char buf[32];
1280 int len;
1281
1282 if (*ppos)
1283 return 0;
1284
1285 if (unlikely(!id))
1286 return -ENODEV;
1287
1288 len = sprintf(buf, "%d\n", id);
1289
1290 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1291}
1292
1293static ssize_t
1294event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1295 loff_t *ppos)
1296{
1297 struct trace_event_file *file;
1298 struct trace_seq *s;
1299 int r = -ENODEV;
1300
1301 if (*ppos)
1302 return 0;
1303
1304 s = kmalloc(sizeof(*s), GFP_KERNEL);
1305
1306 if (!s)
1307 return -ENOMEM;
1308
1309 trace_seq_init(s);
1310
1311 mutex_lock(&event_mutex);
1312 file = event_file_data(filp);
1313 if (file)
1314 print_event_filter(file, s);
1315 mutex_unlock(&event_mutex);
1316
1317 if (file)
1318 r = simple_read_from_buffer(ubuf, cnt, ppos,
1319 s->buffer, trace_seq_used(s));
1320
1321 kfree(s);
1322
1323 return r;
1324}
1325
1326static ssize_t
1327event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1328 loff_t *ppos)
1329{
1330 struct trace_event_file *file;
1331 char *buf;
1332 int err = -ENODEV;
1333
1334 if (cnt >= PAGE_SIZE)
1335 return -EINVAL;
1336
1337 buf = memdup_user_nul(ubuf, cnt);
1338 if (IS_ERR(buf))
1339 return PTR_ERR(buf);
1340
1341 mutex_lock(&event_mutex);
1342 file = event_file_data(filp);
1343 if (file)
1344 err = apply_event_filter(file, buf);
1345 mutex_unlock(&event_mutex);
1346
1347 kfree(buf);
1348 if (err < 0)
1349 return err;
1350
1351 *ppos += cnt;
1352
1353 return cnt;
1354}
1355
1356static LIST_HEAD(event_subsystems);
1357
1358static int subsystem_open(struct inode *inode, struct file *filp)
1359{
1360 struct event_subsystem *system = NULL;
1361 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1362 struct trace_array *tr;
1363 int ret;
1364
1365 if (tracing_is_disabled())
1366 return -ENODEV;
1367
1368 /* Make sure the system still exists */
1369 mutex_lock(&trace_types_lock);
1370 mutex_lock(&event_mutex);
1371 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1372 list_for_each_entry(dir, &tr->systems, list) {
1373 if (dir == inode->i_private) {
1374 /* Don't open systems with no events */
1375 if (dir->nr_events) {
1376 __get_system_dir(dir);
1377 system = dir->subsystem;
1378 }
1379 goto exit_loop;
1380 }
1381 }
1382 }
1383 exit_loop:
1384 mutex_unlock(&event_mutex);
1385 mutex_unlock(&trace_types_lock);
1386
1387 if (!system)
1388 return -ENODEV;
1389
1390 /* Some versions of gcc think dir can be uninitialized here */
1391 WARN_ON(!dir);
1392
1393 /* Still need to increment the ref count of the system */
1394 if (trace_array_get(tr) < 0) {
1395 put_system(dir);
1396 return -ENODEV;
1397 }
1398
1399 ret = tracing_open_generic(inode, filp);
1400 if (ret < 0) {
1401 trace_array_put(tr);
1402 put_system(dir);
1403 }
1404
1405 return ret;
1406}
1407
1408static int system_tr_open(struct inode *inode, struct file *filp)
1409{
1410 struct trace_subsystem_dir *dir;
1411 struct trace_array *tr = inode->i_private;
1412 int ret;
1413
1414 if (tracing_is_disabled())
1415 return -ENODEV;
1416
1417 if (trace_array_get(tr) < 0)
1418 return -ENODEV;
1419
1420 /* Make a temporary dir that has no system but points to tr */
1421 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1422 if (!dir) {
1423 trace_array_put(tr);
1424 return -ENOMEM;
1425 }
1426
1427 dir->tr = tr;
1428
1429 ret = tracing_open_generic(inode, filp);
1430 if (ret < 0) {
1431 trace_array_put(tr);
1432 kfree(dir);
1433 return ret;
1434 }
1435
1436 filp->private_data = dir;
1437
1438 return 0;
1439}
1440
1441static int subsystem_release(struct inode *inode, struct file *file)
1442{
1443 struct trace_subsystem_dir *dir = file->private_data;
1444
1445 trace_array_put(dir->tr);
1446
1447 /*
1448 * If dir->subsystem is NULL, then this is a temporary
1449 * descriptor that was made for a trace_array to enable
1450 * all subsystems.
1451 */
1452 if (dir->subsystem)
1453 put_system(dir);
1454 else
1455 kfree(dir);
1456
1457 return 0;
1458}
1459
1460static ssize_t
1461subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1462 loff_t *ppos)
1463{
1464 struct trace_subsystem_dir *dir = filp->private_data;
1465 struct event_subsystem *system = dir->subsystem;
1466 struct trace_seq *s;
1467 int r;
1468
1469 if (*ppos)
1470 return 0;
1471
1472 s = kmalloc(sizeof(*s), GFP_KERNEL);
1473 if (!s)
1474 return -ENOMEM;
1475
1476 trace_seq_init(s);
1477
1478 print_subsystem_event_filter(system, s);
1479 r = simple_read_from_buffer(ubuf, cnt, ppos,
1480 s->buffer, trace_seq_used(s));
1481
1482 kfree(s);
1483
1484 return r;
1485}
1486
1487static ssize_t
1488subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1489 loff_t *ppos)
1490{
1491 struct trace_subsystem_dir *dir = filp->private_data;
1492 char *buf;
1493 int err;
1494
1495 if (cnt >= PAGE_SIZE)
1496 return -EINVAL;
1497
1498 buf = memdup_user_nul(ubuf, cnt);
1499 if (IS_ERR(buf))
1500 return PTR_ERR(buf);
1501
1502 err = apply_subsystem_event_filter(dir, buf);
1503 kfree(buf);
1504 if (err < 0)
1505 return err;
1506
1507 *ppos += cnt;
1508
1509 return cnt;
1510}
1511
1512static ssize_t
1513show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1514{
1515 int (*func)(struct trace_seq *s) = filp->private_data;
1516 struct trace_seq *s;
1517 int r;
1518
1519 if (*ppos)
1520 return 0;
1521
1522 s = kmalloc(sizeof(*s), GFP_KERNEL);
1523 if (!s)
1524 return -ENOMEM;
1525
1526 trace_seq_init(s);
1527
1528 func(s);
1529 r = simple_read_from_buffer(ubuf, cnt, ppos,
1530 s->buffer, trace_seq_used(s));
1531
1532 kfree(s);
1533
1534 return r;
1535}
1536
1537static void ignore_task_cpu(void *data)
1538{
1539 struct trace_array *tr = data;
1540 struct trace_pid_list *pid_list;
1541
1542 /*
1543 * This function is called by on_each_cpu() while the
1544 * event_mutex is held.
1545 */
1546 pid_list = rcu_dereference_protected(tr->filtered_pids,
1547 mutex_is_locked(&event_mutex));
1548
1549 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1550 trace_ignore_this_task(pid_list, current));
1551}
1552
1553static ssize_t
1554ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1555 size_t cnt, loff_t *ppos)
1556{
1557 struct seq_file *m = filp->private_data;
1558 struct trace_array *tr = m->private;
1559 struct trace_pid_list *filtered_pids = NULL;
1560 struct trace_pid_list *pid_list;
1561 struct trace_event_file *file;
1562 ssize_t ret;
1563
1564 if (!cnt)
1565 return 0;
1566
1567 ret = tracing_update_buffers();
1568 if (ret < 0)
1569 return ret;
1570
1571 mutex_lock(&event_mutex);
1572
1573 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1574 lockdep_is_held(&event_mutex));
1575
1576 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1577 if (ret < 0)
1578 goto out;
1579
1580 rcu_assign_pointer(tr->filtered_pids, pid_list);
1581
1582 list_for_each_entry(file, &tr->events, list) {
1583 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1584 }
1585
1586 if (filtered_pids) {
1587 synchronize_sched();
1588 trace_free_pid_list(filtered_pids);
1589 } else if (pid_list) {
1590 /*
1591 * Register a probe that is called before all other probes
1592 * to set ignore_pid if next or prev do not match.
1593 * Register a probe this is called after all other probes
1594 * to only keep ignore_pid set if next pid matches.
1595 */
1596 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1597 tr, INT_MAX);
1598 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1599 tr, 0);
1600
1601 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1602 tr, INT_MAX);
1603 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1604 tr, 0);
1605
1606 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1607 tr, INT_MAX);
1608 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1609 tr, 0);
1610
1611 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1612 tr, INT_MAX);
1613 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1614 tr, 0);
1615 }
1616
1617 /*
1618 * Ignoring of pids is done at task switch. But we have to
1619 * check for those tasks that are currently running.
1620 * Always do this in case a pid was appended or removed.
1621 */
1622 on_each_cpu(ignore_task_cpu, tr, 1);
1623
1624 out:
1625 mutex_unlock(&event_mutex);
1626
1627 if (ret > 0)
1628 *ppos += ret;
1629
1630 return ret;
1631}
1632
1633static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1634static int ftrace_event_set_open(struct inode *inode, struct file *file);
1635static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1636static int ftrace_event_release(struct inode *inode, struct file *file);
1637
1638static const struct seq_operations show_event_seq_ops = {
1639 .start = t_start,
1640 .next = t_next,
1641 .show = t_show,
1642 .stop = t_stop,
1643};
1644
1645static const struct seq_operations show_set_event_seq_ops = {
1646 .start = s_start,
1647 .next = s_next,
1648 .show = t_show,
1649 .stop = t_stop,
1650};
1651
1652static const struct seq_operations show_set_pid_seq_ops = {
1653 .start = p_start,
1654 .next = p_next,
1655 .show = trace_pid_show,
1656 .stop = p_stop,
1657};
1658
1659static const struct file_operations ftrace_avail_fops = {
1660 .open = ftrace_event_avail_open,
1661 .read = seq_read,
1662 .llseek = seq_lseek,
1663 .release = seq_release,
1664};
1665
1666static const struct file_operations ftrace_set_event_fops = {
1667 .open = ftrace_event_set_open,
1668 .read = seq_read,
1669 .write = ftrace_event_write,
1670 .llseek = seq_lseek,
1671 .release = ftrace_event_release,
1672};
1673
1674static const struct file_operations ftrace_set_event_pid_fops = {
1675 .open = ftrace_event_set_pid_open,
1676 .read = seq_read,
1677 .write = ftrace_event_pid_write,
1678 .llseek = seq_lseek,
1679 .release = ftrace_event_release,
1680};
1681
1682static const struct file_operations ftrace_enable_fops = {
1683 .open = tracing_open_generic,
1684 .read = event_enable_read,
1685 .write = event_enable_write,
1686 .llseek = default_llseek,
1687};
1688
1689static const struct file_operations ftrace_event_format_fops = {
1690 .open = trace_format_open,
1691 .read = seq_read,
1692 .llseek = seq_lseek,
1693 .release = seq_release,
1694};
1695
1696static const struct file_operations ftrace_event_id_fops = {
1697 .read = event_id_read,
1698 .llseek = default_llseek,
1699};
1700
1701static const struct file_operations ftrace_event_filter_fops = {
1702 .open = tracing_open_generic,
1703 .read = event_filter_read,
1704 .write = event_filter_write,
1705 .llseek = default_llseek,
1706};
1707
1708static const struct file_operations ftrace_subsystem_filter_fops = {
1709 .open = subsystem_open,
1710 .read = subsystem_filter_read,
1711 .write = subsystem_filter_write,
1712 .llseek = default_llseek,
1713 .release = subsystem_release,
1714};
1715
1716static const struct file_operations ftrace_system_enable_fops = {
1717 .open = subsystem_open,
1718 .read = system_enable_read,
1719 .write = system_enable_write,
1720 .llseek = default_llseek,
1721 .release = subsystem_release,
1722};
1723
1724static const struct file_operations ftrace_tr_enable_fops = {
1725 .open = system_tr_open,
1726 .read = system_enable_read,
1727 .write = system_enable_write,
1728 .llseek = default_llseek,
1729 .release = subsystem_release,
1730};
1731
1732static const struct file_operations ftrace_show_header_fops = {
1733 .open = tracing_open_generic,
1734 .read = show_header,
1735 .llseek = default_llseek,
1736};
1737
1738static int
1739ftrace_event_open(struct inode *inode, struct file *file,
1740 const struct seq_operations *seq_ops)
1741{
1742 struct seq_file *m;
1743 int ret;
1744
1745 ret = seq_open(file, seq_ops);
1746 if (ret < 0)
1747 return ret;
1748 m = file->private_data;
1749 /* copy tr over to seq ops */
1750 m->private = inode->i_private;
1751
1752 return ret;
1753}
1754
1755static int ftrace_event_release(struct inode *inode, struct file *file)
1756{
1757 struct trace_array *tr = inode->i_private;
1758
1759 trace_array_put(tr);
1760
1761 return seq_release(inode, file);
1762}
1763
1764static int
1765ftrace_event_avail_open(struct inode *inode, struct file *file)
1766{
1767 const struct seq_operations *seq_ops = &show_event_seq_ops;
1768
1769 return ftrace_event_open(inode, file, seq_ops);
1770}
1771
1772static int
1773ftrace_event_set_open(struct inode *inode, struct file *file)
1774{
1775 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1776 struct trace_array *tr = inode->i_private;
1777 int ret;
1778
1779 if (trace_array_get(tr) < 0)
1780 return -ENODEV;
1781
1782 if ((file->f_mode & FMODE_WRITE) &&
1783 (file->f_flags & O_TRUNC))
1784 ftrace_clear_events(tr);
1785
1786 ret = ftrace_event_open(inode, file, seq_ops);
1787 if (ret < 0)
1788 trace_array_put(tr);
1789 return ret;
1790}
1791
1792static int
1793ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1794{
1795 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1796 struct trace_array *tr = inode->i_private;
1797 int ret;
1798
1799 if (trace_array_get(tr) < 0)
1800 return -ENODEV;
1801
1802 if ((file->f_mode & FMODE_WRITE) &&
1803 (file->f_flags & O_TRUNC))
1804 ftrace_clear_event_pids(tr);
1805
1806 ret = ftrace_event_open(inode, file, seq_ops);
1807 if (ret < 0)
1808 trace_array_put(tr);
1809 return ret;
1810}
1811
1812static struct event_subsystem *
1813create_new_subsystem(const char *name)
1814{
1815 struct event_subsystem *system;
1816
1817 /* need to create new entry */
1818 system = kmalloc(sizeof(*system), GFP_KERNEL);
1819 if (!system)
1820 return NULL;
1821
1822 system->ref_count = 1;
1823
1824 /* Only allocate if dynamic (kprobes and modules) */
1825 system->name = kstrdup_const(name, GFP_KERNEL);
1826 if (!system->name)
1827 goto out_free;
1828
1829 system->filter = NULL;
1830
1831 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1832 if (!system->filter)
1833 goto out_free;
1834
1835 list_add(&system->list, &event_subsystems);
1836
1837 return system;
1838
1839 out_free:
1840 kfree_const(system->name);
1841 kfree(system);
1842 return NULL;
1843}
1844
1845static struct dentry *
1846event_subsystem_dir(struct trace_array *tr, const char *name,
1847 struct trace_event_file *file, struct dentry *parent)
1848{
1849 struct trace_subsystem_dir *dir;
1850 struct event_subsystem *system;
1851 struct dentry *entry;
1852
1853 /* First see if we did not already create this dir */
1854 list_for_each_entry(dir, &tr->systems, list) {
1855 system = dir->subsystem;
1856 if (strcmp(system->name, name) == 0) {
1857 dir->nr_events++;
1858 file->system = dir;
1859 return dir->entry;
1860 }
1861 }
1862
1863 /* Now see if the system itself exists. */
1864 list_for_each_entry(system, &event_subsystems, list) {
1865 if (strcmp(system->name, name) == 0)
1866 break;
1867 }
1868 /* Reset system variable when not found */
1869 if (&system->list == &event_subsystems)
1870 system = NULL;
1871
1872 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1873 if (!dir)
1874 goto out_fail;
1875
1876 if (!system) {
1877 system = create_new_subsystem(name);
1878 if (!system)
1879 goto out_free;
1880 } else
1881 __get_system(system);
1882
1883 dir->entry = tracefs_create_dir(name, parent);
1884 if (!dir->entry) {
1885 pr_warn("Failed to create system directory %s\n", name);
1886 __put_system(system);
1887 goto out_free;
1888 }
1889
1890 dir->tr = tr;
1891 dir->ref_count = 1;
1892 dir->nr_events = 1;
1893 dir->subsystem = system;
1894 file->system = dir;
1895
1896 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1897 &ftrace_subsystem_filter_fops);
1898 if (!entry) {
1899 kfree(system->filter);
1900 system->filter = NULL;
1901 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1902 }
1903
1904 trace_create_file("enable", 0644, dir->entry, dir,
1905 &ftrace_system_enable_fops);
1906
1907 list_add(&dir->list, &tr->systems);
1908
1909 return dir->entry;
1910
1911 out_free:
1912 kfree(dir);
1913 out_fail:
1914 /* Only print this message if failed on memory allocation */
1915 if (!dir || !system)
1916 pr_warn("No memory to create event subsystem %s\n", name);
1917 return NULL;
1918}
1919
1920static int
1921event_create_dir(struct dentry *parent, struct trace_event_file *file)
1922{
1923 struct trace_event_call *call = file->event_call;
1924 struct trace_array *tr = file->tr;
1925 struct list_head *head;
1926 struct dentry *d_events;
1927 const char *name;
1928 int ret;
1929
1930 /*
1931 * If the trace point header did not define TRACE_SYSTEM
1932 * then the system would be called "TRACE_SYSTEM".
1933 */
1934 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1935 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1936 if (!d_events)
1937 return -ENOMEM;
1938 } else
1939 d_events = parent;
1940
1941 name = trace_event_name(call);
1942 file->dir = tracefs_create_dir(name, d_events);
1943 if (!file->dir) {
1944 pr_warn("Could not create tracefs '%s' directory\n", name);
1945 return -1;
1946 }
1947
1948 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1949 trace_create_file("enable", 0644, file->dir, file,
1950 &ftrace_enable_fops);
1951
1952#ifdef CONFIG_PERF_EVENTS
1953 if (call->event.type && call->class->reg)
1954 trace_create_file("id", 0444, file->dir,
1955 (void *)(long)call->event.type,
1956 &ftrace_event_id_fops);
1957#endif
1958
1959 /*
1960 * Other events may have the same class. Only update
1961 * the fields if they are not already defined.
1962 */
1963 head = trace_get_fields(call);
1964 if (list_empty(head)) {
1965 ret = call->class->define_fields(call);
1966 if (ret < 0) {
1967 pr_warn("Could not initialize trace point events/%s\n",
1968 name);
1969 return -1;
1970 }
1971 }
1972 trace_create_file("filter", 0644, file->dir, file,
1973 &ftrace_event_filter_fops);
1974
1975 /*
1976 * Only event directories that can be enabled should have
1977 * triggers.
1978 */
1979 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1980 trace_create_file("trigger", 0644, file->dir, file,
1981 &event_trigger_fops);
1982
1983#ifdef CONFIG_HIST_TRIGGERS
1984 trace_create_file("hist", 0444, file->dir, file,
1985 &event_hist_fops);
1986#endif
1987 trace_create_file("format", 0444, file->dir, call,
1988 &ftrace_event_format_fops);
1989
1990 return 0;
1991}
1992
1993static void remove_event_from_tracers(struct trace_event_call *call)
1994{
1995 struct trace_event_file *file;
1996 struct trace_array *tr;
1997
1998 do_for_each_event_file_safe(tr, file) {
1999 if (file->event_call != call)
2000 continue;
2001
2002 remove_event_file_dir(file);
2003 /*
2004 * The do_for_each_event_file_safe() is
2005 * a double loop. After finding the call for this
2006 * trace_array, we use break to jump to the next
2007 * trace_array.
2008 */
2009 break;
2010 } while_for_each_event_file();
2011}
2012
2013static void event_remove(struct trace_event_call *call)
2014{
2015 struct trace_array *tr;
2016 struct trace_event_file *file;
2017
2018 do_for_each_event_file(tr, file) {
2019 if (file->event_call != call)
2020 continue;
2021 ftrace_event_enable_disable(file, 0);
2022 /*
2023 * The do_for_each_event_file() is
2024 * a double loop. After finding the call for this
2025 * trace_array, we use break to jump to the next
2026 * trace_array.
2027 */
2028 break;
2029 } while_for_each_event_file();
2030
2031 if (call->event.funcs)
2032 __unregister_trace_event(&call->event);
2033 remove_event_from_tracers(call);
2034 list_del(&call->list);
2035}
2036
2037static int event_init(struct trace_event_call *call)
2038{
2039 int ret = 0;
2040 const char *name;
2041
2042 name = trace_event_name(call);
2043 if (WARN_ON(!name))
2044 return -EINVAL;
2045
2046 if (call->class->raw_init) {
2047 ret = call->class->raw_init(call);
2048 if (ret < 0 && ret != -ENOSYS)
2049 pr_warn("Could not initialize trace events/%s\n", name);
2050 }
2051
2052 return ret;
2053}
2054
2055static int
2056__register_event(struct trace_event_call *call, struct module *mod)
2057{
2058 int ret;
2059
2060 ret = event_init(call);
2061 if (ret < 0)
2062 return ret;
2063
2064 list_add(&call->list, &ftrace_events);
2065 call->mod = mod;
2066
2067 return 0;
2068}
2069
2070static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2071{
2072 int rlen;
2073 int elen;
2074
2075 /* Find the length of the enum value as a string */
2076 elen = snprintf(ptr, 0, "%ld", map->enum_value);
2077 /* Make sure there's enough room to replace the string with the value */
2078 if (len < elen)
2079 return NULL;
2080
2081 snprintf(ptr, elen + 1, "%ld", map->enum_value);
2082
2083 /* Get the rest of the string of ptr */
2084 rlen = strlen(ptr + len);
2085 memmove(ptr + elen, ptr + len, rlen);
2086 /* Make sure we end the new string */
2087 ptr[elen + rlen] = 0;
2088
2089 return ptr + elen;
2090}
2091
2092static void update_event_printk(struct trace_event_call *call,
2093 struct trace_enum_map *map)
2094{
2095 char *ptr;
2096 int quote = 0;
2097 int len = strlen(map->enum_string);
2098
2099 for (ptr = call->print_fmt; *ptr; ptr++) {
2100 if (*ptr == '\\') {
2101 ptr++;
2102 /* paranoid */
2103 if (!*ptr)
2104 break;
2105 continue;
2106 }
2107 if (*ptr == '"') {
2108 quote ^= 1;
2109 continue;
2110 }
2111 if (quote)
2112 continue;
2113 if (isdigit(*ptr)) {
2114 /* skip numbers */
2115 do {
2116 ptr++;
2117 /* Check for alpha chars like ULL */
2118 } while (isalnum(*ptr));
2119 if (!*ptr)
2120 break;
2121 /*
2122 * A number must have some kind of delimiter after
2123 * it, and we can ignore that too.
2124 */
2125 continue;
2126 }
2127 if (isalpha(*ptr) || *ptr == '_') {
2128 if (strncmp(map->enum_string, ptr, len) == 0 &&
2129 !isalnum(ptr[len]) && ptr[len] != '_') {
2130 ptr = enum_replace(ptr, map, len);
2131 /* Hmm, enum string smaller than value */
2132 if (WARN_ON_ONCE(!ptr))
2133 return;
2134 /*
2135 * No need to decrement here, as enum_replace()
2136 * returns the pointer to the character passed
2137 * the enum, and two enums can not be placed
2138 * back to back without something in between.
2139 * We can skip that something in between.
2140 */
2141 continue;
2142 }
2143 skip_more:
2144 do {
2145 ptr++;
2146 } while (isalnum(*ptr) || *ptr == '_');
2147 if (!*ptr)
2148 break;
2149 /*
2150 * If what comes after this variable is a '.' or
2151 * '->' then we can continue to ignore that string.
2152 */
2153 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2154 ptr += *ptr == '.' ? 1 : 2;
2155 if (!*ptr)
2156 break;
2157 goto skip_more;
2158 }
2159 /*
2160 * Once again, we can skip the delimiter that came
2161 * after the string.
2162 */
2163 continue;
2164 }
2165 }
2166}
2167
2168void trace_event_enum_update(struct trace_enum_map **map, int len)
2169{
2170 struct trace_event_call *call, *p;
2171 const char *last_system = NULL;
2172 int last_i;
2173 int i;
2174
2175 down_write(&trace_event_sem);
2176 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2177 /* events are usually grouped together with systems */
2178 if (!last_system || call->class->system != last_system) {
2179 last_i = 0;
2180 last_system = call->class->system;
2181 }
2182
2183 for (i = last_i; i < len; i++) {
2184 if (call->class->system == map[i]->system) {
2185 /* Save the first system if need be */
2186 if (!last_i)
2187 last_i = i;
2188 update_event_printk(call, map[i]);
2189 }
2190 }
2191 }
2192 up_write(&trace_event_sem);
2193}
2194
2195static struct trace_event_file *
2196trace_create_new_event(struct trace_event_call *call,
2197 struct trace_array *tr)
2198{
2199 struct trace_event_file *file;
2200
2201 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2202 if (!file)
2203 return NULL;
2204
2205 file->event_call = call;
2206 file->tr = tr;
2207 atomic_set(&file->sm_ref, 0);
2208 atomic_set(&file->tm_ref, 0);
2209 INIT_LIST_HEAD(&file->triggers);
2210 list_add(&file->list, &tr->events);
2211
2212 return file;
2213}
2214
2215/* Add an event to a trace directory */
2216static int
2217__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2218{
2219 struct trace_event_file *file;
2220
2221 file = trace_create_new_event(call, tr);
2222 if (!file)
2223 return -ENOMEM;
2224
2225 return event_create_dir(tr->event_dir, file);
2226}
2227
2228/*
2229 * Just create a decriptor for early init. A descriptor is required
2230 * for enabling events at boot. We want to enable events before
2231 * the filesystem is initialized.
2232 */
2233static __init int
2234__trace_early_add_new_event(struct trace_event_call *call,
2235 struct trace_array *tr)
2236{
2237 struct trace_event_file *file;
2238
2239 file = trace_create_new_event(call, tr);
2240 if (!file)
2241 return -ENOMEM;
2242
2243 return 0;
2244}
2245
2246struct ftrace_module_file_ops;
2247static void __add_event_to_tracers(struct trace_event_call *call);
2248
2249/* Add an additional event_call dynamically */
2250int trace_add_event_call(struct trace_event_call *call)
2251{
2252 int ret;
2253 mutex_lock(&trace_types_lock);
2254 mutex_lock(&event_mutex);
2255
2256 ret = __register_event(call, NULL);
2257 if (ret >= 0)
2258 __add_event_to_tracers(call);
2259
2260 mutex_unlock(&event_mutex);
2261 mutex_unlock(&trace_types_lock);
2262 return ret;
2263}
2264
2265/*
2266 * Must be called under locking of trace_types_lock, event_mutex and
2267 * trace_event_sem.
2268 */
2269static void __trace_remove_event_call(struct trace_event_call *call)
2270{
2271 event_remove(call);
2272 trace_destroy_fields(call);
2273 free_event_filter(call->filter);
2274 call->filter = NULL;
2275}
2276
2277static int probe_remove_event_call(struct trace_event_call *call)
2278{
2279 struct trace_array *tr;
2280 struct trace_event_file *file;
2281
2282#ifdef CONFIG_PERF_EVENTS
2283 if (call->perf_refcount)
2284 return -EBUSY;
2285#endif
2286 do_for_each_event_file(tr, file) {
2287 if (file->event_call != call)
2288 continue;
2289 /*
2290 * We can't rely on ftrace_event_enable_disable(enable => 0)
2291 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2292 * TRACE_REG_UNREGISTER.
2293 */
2294 if (file->flags & EVENT_FILE_FL_ENABLED)
2295 return -EBUSY;
2296 /*
2297 * The do_for_each_event_file_safe() is
2298 * a double loop. After finding the call for this
2299 * trace_array, we use break to jump to the next
2300 * trace_array.
2301 */
2302 break;
2303 } while_for_each_event_file();
2304
2305 __trace_remove_event_call(call);
2306
2307 return 0;
2308}
2309
2310/* Remove an event_call */
2311int trace_remove_event_call(struct trace_event_call *call)
2312{
2313 int ret;
2314
2315 mutex_lock(&trace_types_lock);
2316 mutex_lock(&event_mutex);
2317 down_write(&trace_event_sem);
2318 ret = probe_remove_event_call(call);
2319 up_write(&trace_event_sem);
2320 mutex_unlock(&event_mutex);
2321 mutex_unlock(&trace_types_lock);
2322
2323 return ret;
2324}
2325
2326#define for_each_event(event, start, end) \
2327 for (event = start; \
2328 (unsigned long)event < (unsigned long)end; \
2329 event++)
2330
2331#ifdef CONFIG_MODULES
2332
2333static void trace_module_add_events(struct module *mod)
2334{
2335 struct trace_event_call **call, **start, **end;
2336
2337 if (!mod->num_trace_events)
2338 return;
2339
2340 /* Don't add infrastructure for mods without tracepoints */
2341 if (trace_module_has_bad_taint(mod)) {
2342 pr_err("%s: module has bad taint, not creating trace events\n",
2343 mod->name);
2344 return;
2345 }
2346
2347 start = mod->trace_events;
2348 end = mod->trace_events + mod->num_trace_events;
2349
2350 for_each_event(call, start, end) {
2351 __register_event(*call, mod);
2352 __add_event_to_tracers(*call);
2353 }
2354}
2355
2356static void trace_module_remove_events(struct module *mod)
2357{
2358 struct trace_event_call *call, *p;
2359 bool clear_trace = false;
2360
2361 down_write(&trace_event_sem);
2362 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2363 if (call->mod == mod) {
2364 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2365 clear_trace = true;
2366 __trace_remove_event_call(call);
2367 }
2368 }
2369 up_write(&trace_event_sem);
2370
2371 /*
2372 * It is safest to reset the ring buffer if the module being unloaded
2373 * registered any events that were used. The only worry is if
2374 * a new module gets loaded, and takes on the same id as the events
2375 * of this module. When printing out the buffer, traced events left
2376 * over from this module may be passed to the new module events and
2377 * unexpected results may occur.
2378 */
2379 if (clear_trace)
2380 tracing_reset_all_online_cpus();
2381}
2382
2383static int trace_module_notify(struct notifier_block *self,
2384 unsigned long val, void *data)
2385{
2386 struct module *mod = data;
2387
2388 mutex_lock(&trace_types_lock);
2389 mutex_lock(&event_mutex);
2390 switch (val) {
2391 case MODULE_STATE_COMING:
2392 trace_module_add_events(mod);
2393 break;
2394 case MODULE_STATE_GOING:
2395 trace_module_remove_events(mod);
2396 break;
2397 }
2398 mutex_unlock(&event_mutex);
2399 mutex_unlock(&trace_types_lock);
2400
2401 return 0;
2402}
2403
2404static struct notifier_block trace_module_nb = {
2405 .notifier_call = trace_module_notify,
2406 .priority = 1, /* higher than trace.c module notify */
2407};
2408#endif /* CONFIG_MODULES */
2409
2410/* Create a new event directory structure for a trace directory. */
2411static void
2412__trace_add_event_dirs(struct trace_array *tr)
2413{
2414 struct trace_event_call *call;
2415 int ret;
2416
2417 list_for_each_entry(call, &ftrace_events, list) {
2418 ret = __trace_add_new_event(call, tr);
2419 if (ret < 0)
2420 pr_warn("Could not create directory for event %s\n",
2421 trace_event_name(call));
2422 }
2423}
2424
2425struct trace_event_file *
2426find_event_file(struct trace_array *tr, const char *system, const char *event)
2427{
2428 struct trace_event_file *file;
2429 struct trace_event_call *call;
2430 const char *name;
2431
2432 list_for_each_entry(file, &tr->events, list) {
2433
2434 call = file->event_call;
2435 name = trace_event_name(call);
2436
2437 if (!name || !call->class || !call->class->reg)
2438 continue;
2439
2440 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2441 continue;
2442
2443 if (strcmp(event, name) == 0 &&
2444 strcmp(system, call->class->system) == 0)
2445 return file;
2446 }
2447 return NULL;
2448}
2449
2450#ifdef CONFIG_DYNAMIC_FTRACE
2451
2452/* Avoid typos */
2453#define ENABLE_EVENT_STR "enable_event"
2454#define DISABLE_EVENT_STR "disable_event"
2455
2456struct event_probe_data {
2457 struct trace_event_file *file;
2458 unsigned long count;
2459 int ref;
2460 bool enable;
2461};
2462
2463static void
2464event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2465{
2466 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2467 struct event_probe_data *data = *pdata;
2468
2469 if (!data)
2470 return;
2471
2472 if (data->enable)
2473 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2474 else
2475 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2476}
2477
2478static void
2479event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2480{
2481 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2482 struct event_probe_data *data = *pdata;
2483
2484 if (!data)
2485 return;
2486
2487 if (!data->count)
2488 return;
2489
2490 /* Skip if the event is in a state we want to switch to */
2491 if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2492 return;
2493
2494 if (data->count != -1)
2495 (data->count)--;
2496
2497 event_enable_probe(ip, parent_ip, _data);
2498}
2499
2500static int
2501event_enable_print(struct seq_file *m, unsigned long ip,
2502 struct ftrace_probe_ops *ops, void *_data)
2503{
2504 struct event_probe_data *data = _data;
2505
2506 seq_printf(m, "%ps:", (void *)ip);
2507
2508 seq_printf(m, "%s:%s:%s",
2509 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2510 data->file->event_call->class->system,
2511 trace_event_name(data->file->event_call));
2512
2513 if (data->count == -1)
2514 seq_puts(m, ":unlimited\n");
2515 else
2516 seq_printf(m, ":count=%ld\n", data->count);
2517
2518 return 0;
2519}
2520
2521static int
2522event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2523 void **_data)
2524{
2525 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2526 struct event_probe_data *data = *pdata;
2527
2528 data->ref++;
2529 return 0;
2530}
2531
2532static void
2533event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2534 void **_data)
2535{
2536 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2537 struct event_probe_data *data = *pdata;
2538
2539 if (WARN_ON_ONCE(data->ref <= 0))
2540 return;
2541
2542 data->ref--;
2543 if (!data->ref) {
2544 /* Remove the SOFT_MODE flag */
2545 __ftrace_event_enable_disable(data->file, 0, 1);
2546 module_put(data->file->event_call->mod);
2547 kfree(data);
2548 }
2549 *pdata = NULL;
2550}
2551
2552static struct ftrace_probe_ops event_enable_probe_ops = {
2553 .func = event_enable_probe,
2554 .print = event_enable_print,
2555 .init = event_enable_init,
2556 .free = event_enable_free,
2557};
2558
2559static struct ftrace_probe_ops event_enable_count_probe_ops = {
2560 .func = event_enable_count_probe,
2561 .print = event_enable_print,
2562 .init = event_enable_init,
2563 .free = event_enable_free,
2564};
2565
2566static struct ftrace_probe_ops event_disable_probe_ops = {
2567 .func = event_enable_probe,
2568 .print = event_enable_print,
2569 .init = event_enable_init,
2570 .free = event_enable_free,
2571};
2572
2573static struct ftrace_probe_ops event_disable_count_probe_ops = {
2574 .func = event_enable_count_probe,
2575 .print = event_enable_print,
2576 .init = event_enable_init,
2577 .free = event_enable_free,
2578};
2579
2580static int
2581event_enable_func(struct ftrace_hash *hash,
2582 char *glob, char *cmd, char *param, int enabled)
2583{
2584 struct trace_array *tr = top_trace_array();
2585 struct trace_event_file *file;
2586 struct ftrace_probe_ops *ops;
2587 struct event_probe_data *data;
2588 const char *system;
2589 const char *event;
2590 char *number;
2591 bool enable;
2592 int ret;
2593
2594 if (!tr)
2595 return -ENODEV;
2596
2597 /* hash funcs only work with set_ftrace_filter */
2598 if (!enabled || !param)
2599 return -EINVAL;
2600
2601 system = strsep(¶m, ":");
2602 if (!param)
2603 return -EINVAL;
2604
2605 event = strsep(¶m, ":");
2606
2607 mutex_lock(&event_mutex);
2608
2609 ret = -EINVAL;
2610 file = find_event_file(tr, system, event);
2611 if (!file)
2612 goto out;
2613
2614 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2615
2616 if (enable)
2617 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2618 else
2619 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2620
2621 if (glob[0] == '!') {
2622 unregister_ftrace_function_probe_func(glob+1, ops);
2623 ret = 0;
2624 goto out;
2625 }
2626
2627 ret = -ENOMEM;
2628 data = kzalloc(sizeof(*data), GFP_KERNEL);
2629 if (!data)
2630 goto out;
2631
2632 data->enable = enable;
2633 data->count = -1;
2634 data->file = file;
2635
2636 if (!param)
2637 goto out_reg;
2638
2639 number = strsep(¶m, ":");
2640
2641 ret = -EINVAL;
2642 if (!strlen(number))
2643 goto out_free;
2644
2645 /*
2646 * We use the callback data field (which is a pointer)
2647 * as our counter.
2648 */
2649 ret = kstrtoul(number, 0, &data->count);
2650 if (ret)
2651 goto out_free;
2652
2653 out_reg:
2654 /* Don't let event modules unload while probe registered */
2655 ret = try_module_get(file->event_call->mod);
2656 if (!ret) {
2657 ret = -EBUSY;
2658 goto out_free;
2659 }
2660
2661 ret = __ftrace_event_enable_disable(file, 1, 1);
2662 if (ret < 0)
2663 goto out_put;
2664 ret = register_ftrace_function_probe(glob, ops, data);
2665 /*
2666 * The above returns on success the # of functions enabled,
2667 * but if it didn't find any functions it returns zero.
2668 * Consider no functions a failure too.
2669 */
2670 if (!ret) {
2671 ret = -ENOENT;
2672 goto out_disable;
2673 } else if (ret < 0)
2674 goto out_disable;
2675 /* Just return zero, not the number of enabled functions */
2676 ret = 0;
2677 out:
2678 mutex_unlock(&event_mutex);
2679 return ret;
2680
2681 out_disable:
2682 __ftrace_event_enable_disable(file, 0, 1);
2683 out_put:
2684 module_put(file->event_call->mod);
2685 out_free:
2686 kfree(data);
2687 goto out;
2688}
2689
2690static struct ftrace_func_command event_enable_cmd = {
2691 .name = ENABLE_EVENT_STR,
2692 .func = event_enable_func,
2693};
2694
2695static struct ftrace_func_command event_disable_cmd = {
2696 .name = DISABLE_EVENT_STR,
2697 .func = event_enable_func,
2698};
2699
2700static __init int register_event_cmds(void)
2701{
2702 int ret;
2703
2704 ret = register_ftrace_command(&event_enable_cmd);
2705 if (WARN_ON(ret < 0))
2706 return ret;
2707 ret = register_ftrace_command(&event_disable_cmd);
2708 if (WARN_ON(ret < 0))
2709 unregister_ftrace_command(&event_enable_cmd);
2710 return ret;
2711}
2712#else
2713static inline int register_event_cmds(void) { return 0; }
2714#endif /* CONFIG_DYNAMIC_FTRACE */
2715
2716/*
2717 * The top level array has already had its trace_event_file
2718 * descriptors created in order to allow for early events to
2719 * be recorded. This function is called after the tracefs has been
2720 * initialized, and we now have to create the files associated
2721 * to the events.
2722 */
2723static __init void
2724__trace_early_add_event_dirs(struct trace_array *tr)
2725{
2726 struct trace_event_file *file;
2727 int ret;
2728
2729
2730 list_for_each_entry(file, &tr->events, list) {
2731 ret = event_create_dir(tr->event_dir, file);
2732 if (ret < 0)
2733 pr_warn("Could not create directory for event %s\n",
2734 trace_event_name(file->event_call));
2735 }
2736}
2737
2738/*
2739 * For early boot up, the top trace array requires to have
2740 * a list of events that can be enabled. This must be done before
2741 * the filesystem is set up in order to allow events to be traced
2742 * early.
2743 */
2744static __init void
2745__trace_early_add_events(struct trace_array *tr)
2746{
2747 struct trace_event_call *call;
2748 int ret;
2749
2750 list_for_each_entry(call, &ftrace_events, list) {
2751 /* Early boot up should not have any modules loaded */
2752 if (WARN_ON_ONCE(call->mod))
2753 continue;
2754
2755 ret = __trace_early_add_new_event(call, tr);
2756 if (ret < 0)
2757 pr_warn("Could not create early event %s\n",
2758 trace_event_name(call));
2759 }
2760}
2761
2762/* Remove the event directory structure for a trace directory. */
2763static void
2764__trace_remove_event_dirs(struct trace_array *tr)
2765{
2766 struct trace_event_file *file, *next;
2767
2768 list_for_each_entry_safe(file, next, &tr->events, list)
2769 remove_event_file_dir(file);
2770}
2771
2772static void __add_event_to_tracers(struct trace_event_call *call)
2773{
2774 struct trace_array *tr;
2775
2776 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2777 __trace_add_new_event(call, tr);
2778}
2779
2780extern struct trace_event_call *__start_ftrace_events[];
2781extern struct trace_event_call *__stop_ftrace_events[];
2782
2783static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2784
2785static __init int setup_trace_event(char *str)
2786{
2787 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2788 ring_buffer_expanded = true;
2789 tracing_selftest_disabled = true;
2790
2791 return 1;
2792}
2793__setup("trace_event=", setup_trace_event);
2794
2795/* Expects to have event_mutex held when called */
2796static int
2797create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2798{
2799 struct dentry *d_events;
2800 struct dentry *entry;
2801
2802 entry = tracefs_create_file("set_event", 0644, parent,
2803 tr, &ftrace_set_event_fops);
2804 if (!entry) {
2805 pr_warn("Could not create tracefs 'set_event' entry\n");
2806 return -ENOMEM;
2807 }
2808
2809 d_events = tracefs_create_dir("events", parent);
2810 if (!d_events) {
2811 pr_warn("Could not create tracefs 'events' directory\n");
2812 return -ENOMEM;
2813 }
2814
2815 entry = trace_create_file("enable", 0644, d_events,
2816 tr, &ftrace_tr_enable_fops);
2817 if (!entry) {
2818 pr_warn("Could not create tracefs 'enable' entry\n");
2819 return -ENOMEM;
2820 }
2821
2822 /* There are not as crucial, just warn if they are not created */
2823
2824 entry = tracefs_create_file("set_event_pid", 0644, parent,
2825 tr, &ftrace_set_event_pid_fops);
2826 if (!entry)
2827 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2828
2829 /* ring buffer internal formats */
2830 entry = trace_create_file("header_page", 0444, d_events,
2831 ring_buffer_print_page_header,
2832 &ftrace_show_header_fops);
2833 if (!entry)
2834 pr_warn("Could not create tracefs 'header_page' entry\n");
2835
2836 entry = trace_create_file("header_event", 0444, d_events,
2837 ring_buffer_print_entry_header,
2838 &ftrace_show_header_fops);
2839 if (!entry)
2840 pr_warn("Could not create tracefs 'header_event' entry\n");
2841
2842 tr->event_dir = d_events;
2843
2844 return 0;
2845}
2846
2847/**
2848 * event_trace_add_tracer - add a instance of a trace_array to events
2849 * @parent: The parent dentry to place the files/directories for events in
2850 * @tr: The trace array associated with these events
2851 *
2852 * When a new instance is created, it needs to set up its events
2853 * directory, as well as other files associated with events. It also
2854 * creates the event hierachry in the @parent/events directory.
2855 *
2856 * Returns 0 on success.
2857 */
2858int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2859{
2860 int ret;
2861
2862 mutex_lock(&event_mutex);
2863
2864 ret = create_event_toplevel_files(parent, tr);
2865 if (ret)
2866 goto out_unlock;
2867
2868 down_write(&trace_event_sem);
2869 __trace_add_event_dirs(tr);
2870 up_write(&trace_event_sem);
2871
2872 out_unlock:
2873 mutex_unlock(&event_mutex);
2874
2875 return ret;
2876}
2877
2878/*
2879 * The top trace array already had its file descriptors created.
2880 * Now the files themselves need to be created.
2881 */
2882static __init int
2883early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2884{
2885 int ret;
2886
2887 mutex_lock(&event_mutex);
2888
2889 ret = create_event_toplevel_files(parent, tr);
2890 if (ret)
2891 goto out_unlock;
2892
2893 down_write(&trace_event_sem);
2894 __trace_early_add_event_dirs(tr);
2895 up_write(&trace_event_sem);
2896
2897 out_unlock:
2898 mutex_unlock(&event_mutex);
2899
2900 return ret;
2901}
2902
2903int event_trace_del_tracer(struct trace_array *tr)
2904{
2905 mutex_lock(&event_mutex);
2906
2907 /* Disable any event triggers and associated soft-disabled events */
2908 clear_event_triggers(tr);
2909
2910 /* Clear the pid list */
2911 __ftrace_clear_event_pids(tr);
2912
2913 /* Disable any running events */
2914 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2915
2916 /* Access to events are within rcu_read_lock_sched() */
2917 synchronize_sched();
2918
2919 down_write(&trace_event_sem);
2920 __trace_remove_event_dirs(tr);
2921 tracefs_remove_recursive(tr->event_dir);
2922 up_write(&trace_event_sem);
2923
2924 tr->event_dir = NULL;
2925
2926 mutex_unlock(&event_mutex);
2927
2928 return 0;
2929}
2930
2931static __init int event_trace_memsetup(void)
2932{
2933 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2934 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
2935 return 0;
2936}
2937
2938static __init void
2939early_enable_events(struct trace_array *tr, bool disable_first)
2940{
2941 char *buf = bootup_event_buf;
2942 char *token;
2943 int ret;
2944
2945 while (true) {
2946 token = strsep(&buf, ",");
2947
2948 if (!token)
2949 break;
2950
2951 if (*token) {
2952 /* Restarting syscalls requires that we stop them first */
2953 if (disable_first)
2954 ftrace_set_clr_event(tr, token, 0);
2955
2956 ret = ftrace_set_clr_event(tr, token, 1);
2957 if (ret)
2958 pr_warn("Failed to enable trace event: %s\n", token);
2959 }
2960
2961 /* Put back the comma to allow this to be called again */
2962 if (buf)
2963 *(buf - 1) = ',';
2964 }
2965}
2966
2967static __init int event_trace_enable(void)
2968{
2969 struct trace_array *tr = top_trace_array();
2970 struct trace_event_call **iter, *call;
2971 int ret;
2972
2973 if (!tr)
2974 return -ENODEV;
2975
2976 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2977
2978 call = *iter;
2979 ret = event_init(call);
2980 if (!ret)
2981 list_add(&call->list, &ftrace_events);
2982 }
2983
2984 /*
2985 * We need the top trace array to have a working set of trace
2986 * points at early init, before the debug files and directories
2987 * are created. Create the file entries now, and attach them
2988 * to the actual file dentries later.
2989 */
2990 __trace_early_add_events(tr);
2991
2992 early_enable_events(tr, false);
2993
2994 trace_printk_start_comm();
2995
2996 register_event_cmds();
2997
2998 register_trigger_cmds();
2999
3000 return 0;
3001}
3002
3003/*
3004 * event_trace_enable() is called from trace_event_init() first to
3005 * initialize events and perhaps start any events that are on the
3006 * command line. Unfortunately, there are some events that will not
3007 * start this early, like the system call tracepoints that need
3008 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3009 * is called before pid 1 starts, and this flag is never set, making
3010 * the syscall tracepoint never get reached, but the event is enabled
3011 * regardless (and not doing anything).
3012 */
3013static __init int event_trace_enable_again(void)
3014{
3015 struct trace_array *tr;
3016
3017 tr = top_trace_array();
3018 if (!tr)
3019 return -ENODEV;
3020
3021 early_enable_events(tr, true);
3022
3023 return 0;
3024}
3025
3026early_initcall(event_trace_enable_again);
3027
3028static __init int event_trace_init(void)
3029{
3030 struct trace_array *tr;
3031 struct dentry *d_tracer;
3032 struct dentry *entry;
3033 int ret;
3034
3035 tr = top_trace_array();
3036 if (!tr)
3037 return -ENODEV;
3038
3039 d_tracer = tracing_init_dentry();
3040 if (IS_ERR(d_tracer))
3041 return 0;
3042
3043 entry = tracefs_create_file("available_events", 0444, d_tracer,
3044 tr, &ftrace_avail_fops);
3045 if (!entry)
3046 pr_warn("Could not create tracefs 'available_events' entry\n");
3047
3048 if (trace_define_generic_fields())
3049 pr_warn("tracing: Failed to allocated generic fields");
3050
3051 if (trace_define_common_fields())
3052 pr_warn("tracing: Failed to allocate common fields");
3053
3054 ret = early_event_add_tracer(d_tracer, tr);
3055 if (ret)
3056 return ret;
3057
3058#ifdef CONFIG_MODULES
3059 ret = register_module_notifier(&trace_module_nb);
3060 if (ret)
3061 pr_warn("Failed to register trace events module notifier\n");
3062#endif
3063 return 0;
3064}
3065
3066void __init trace_event_init(void)
3067{
3068 event_trace_memsetup();
3069 init_ftrace_syscalls();
3070 event_trace_enable();
3071}
3072
3073fs_initcall(event_trace_init);
3074
3075#ifdef CONFIG_FTRACE_STARTUP_TEST
3076
3077static DEFINE_SPINLOCK(test_spinlock);
3078static DEFINE_SPINLOCK(test_spinlock_irq);
3079static DEFINE_MUTEX(test_mutex);
3080
3081static __init void test_work(struct work_struct *dummy)
3082{
3083 spin_lock(&test_spinlock);
3084 spin_lock_irq(&test_spinlock_irq);
3085 udelay(1);
3086 spin_unlock_irq(&test_spinlock_irq);
3087 spin_unlock(&test_spinlock);
3088
3089 mutex_lock(&test_mutex);
3090 msleep(1);
3091 mutex_unlock(&test_mutex);
3092}
3093
3094static __init int event_test_thread(void *unused)
3095{
3096 void *test_malloc;
3097
3098 test_malloc = kmalloc(1234, GFP_KERNEL);
3099 if (!test_malloc)
3100 pr_info("failed to kmalloc\n");
3101
3102 schedule_on_each_cpu(test_work);
3103
3104 kfree(test_malloc);
3105
3106 set_current_state(TASK_INTERRUPTIBLE);
3107 while (!kthread_should_stop()) {
3108 schedule();
3109 set_current_state(TASK_INTERRUPTIBLE);
3110 }
3111 __set_current_state(TASK_RUNNING);
3112
3113 return 0;
3114}
3115
3116/*
3117 * Do various things that may trigger events.
3118 */
3119static __init void event_test_stuff(void)
3120{
3121 struct task_struct *test_thread;
3122
3123 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3124 msleep(1);
3125 kthread_stop(test_thread);
3126}
3127
3128/*
3129 * For every trace event defined, we will test each trace point separately,
3130 * and then by groups, and finally all trace points.
3131 */
3132static __init void event_trace_self_tests(void)
3133{
3134 struct trace_subsystem_dir *dir;
3135 struct trace_event_file *file;
3136 struct trace_event_call *call;
3137 struct event_subsystem *system;
3138 struct trace_array *tr;
3139 int ret;
3140
3141 tr = top_trace_array();
3142 if (!tr)
3143 return;
3144
3145 pr_info("Running tests on trace events:\n");
3146
3147 list_for_each_entry(file, &tr->events, list) {
3148
3149 call = file->event_call;
3150
3151 /* Only test those that have a probe */
3152 if (!call->class || !call->class->probe)
3153 continue;
3154
3155/*
3156 * Testing syscall events here is pretty useless, but
3157 * we still do it if configured. But this is time consuming.
3158 * What we really need is a user thread to perform the
3159 * syscalls as we test.
3160 */
3161#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3162 if (call->class->system &&
3163 strcmp(call->class->system, "syscalls") == 0)
3164 continue;
3165#endif
3166
3167 pr_info("Testing event %s: ", trace_event_name(call));
3168
3169 /*
3170 * If an event is already enabled, someone is using
3171 * it and the self test should not be on.
3172 */
3173 if (file->flags & EVENT_FILE_FL_ENABLED) {
3174 pr_warn("Enabled event during self test!\n");
3175 WARN_ON_ONCE(1);
3176 continue;
3177 }
3178
3179 ftrace_event_enable_disable(file, 1);
3180 event_test_stuff();
3181 ftrace_event_enable_disable(file, 0);
3182
3183 pr_cont("OK\n");
3184 }
3185
3186 /* Now test at the sub system level */
3187
3188 pr_info("Running tests on trace event systems:\n");
3189
3190 list_for_each_entry(dir, &tr->systems, list) {
3191
3192 system = dir->subsystem;
3193
3194 /* the ftrace system is special, skip it */
3195 if (strcmp(system->name, "ftrace") == 0)
3196 continue;
3197
3198 pr_info("Testing event system %s: ", system->name);
3199
3200 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3201 if (WARN_ON_ONCE(ret)) {
3202 pr_warn("error enabling system %s\n",
3203 system->name);
3204 continue;
3205 }
3206
3207 event_test_stuff();
3208
3209 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3210 if (WARN_ON_ONCE(ret)) {
3211 pr_warn("error disabling system %s\n",
3212 system->name);
3213 continue;
3214 }
3215
3216 pr_cont("OK\n");
3217 }
3218
3219 /* Test with all events enabled */
3220
3221 pr_info("Running tests on all trace events:\n");
3222 pr_info("Testing all events: ");
3223
3224 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3225 if (WARN_ON_ONCE(ret)) {
3226 pr_warn("error enabling all events\n");
3227 return;
3228 }
3229
3230 event_test_stuff();
3231
3232 /* reset sysname */
3233 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3234 if (WARN_ON_ONCE(ret)) {
3235 pr_warn("error disabling all events\n");
3236 return;
3237 }
3238
3239 pr_cont("OK\n");
3240}
3241
3242#ifdef CONFIG_FUNCTION_TRACER
3243
3244static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3245
3246static struct trace_event_file event_trace_file __initdata;
3247
3248static void __init
3249function_test_events_call(unsigned long ip, unsigned long parent_ip,
3250 struct ftrace_ops *op, struct pt_regs *pt_regs)
3251{
3252 struct ring_buffer_event *event;
3253 struct ring_buffer *buffer;
3254 struct ftrace_entry *entry;
3255 unsigned long flags;
3256 long disabled;
3257 int cpu;
3258 int pc;
3259
3260 pc = preempt_count();
3261 preempt_disable_notrace();
3262 cpu = raw_smp_processor_id();
3263 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3264
3265 if (disabled != 1)
3266 goto out;
3267
3268 local_save_flags(flags);
3269
3270 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3271 TRACE_FN, sizeof(*entry),
3272 flags, pc);
3273 if (!event)
3274 goto out;
3275 entry = ring_buffer_event_data(event);
3276 entry->ip = ip;
3277 entry->parent_ip = parent_ip;
3278
3279 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3280 entry, flags, pc);
3281 out:
3282 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3283 preempt_enable_notrace();
3284}
3285
3286static struct ftrace_ops trace_ops __initdata =
3287{
3288 .func = function_test_events_call,
3289 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3290};
3291
3292static __init void event_trace_self_test_with_function(void)
3293{
3294 int ret;
3295
3296 event_trace_file.tr = top_trace_array();
3297 if (WARN_ON(!event_trace_file.tr))
3298 return;
3299
3300 ret = register_ftrace_function(&trace_ops);
3301 if (WARN_ON(ret < 0)) {
3302 pr_info("Failed to enable function tracer for event tests\n");
3303 return;
3304 }
3305 pr_info("Running tests again, along with the function tracer\n");
3306 event_trace_self_tests();
3307 unregister_ftrace_function(&trace_ops);
3308}
3309#else
3310static __init void event_trace_self_test_with_function(void)
3311{
3312}
3313#endif
3314
3315static __init int event_trace_self_tests_init(void)
3316{
3317 if (!tracing_selftest_disabled) {
3318 event_trace_self_tests();
3319 event_trace_self_test_with_function();
3320 }
3321
3322 return 0;
3323}
3324
3325late_initcall(event_trace_self_tests_init);
3326
3327#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * event tracer
4 *
5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6 *
7 * - Added format output of fields of the trace point.
8 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
9 *
10 */
11
12#define pr_fmt(fmt) fmt
13
14#include <linux/workqueue.h>
15#include <linux/security.h>
16#include <linux/spinlock.h>
17#include <linux/kthread.h>
18#include <linux/tracefs.h>
19#include <linux/uaccess.h>
20#include <linux/module.h>
21#include <linux/ctype.h>
22#include <linux/sort.h>
23#include <linux/slab.h>
24#include <linux/delay.h>
25
26#include <trace/events/sched.h>
27
28#include <asm/setup.h>
29
30#include "trace_output.h"
31
32#undef TRACE_SYSTEM
33#define TRACE_SYSTEM "TRACE_SYSTEM"
34
35DEFINE_MUTEX(event_mutex);
36
37LIST_HEAD(ftrace_events);
38static LIST_HEAD(ftrace_generic_fields);
39static LIST_HEAD(ftrace_common_fields);
40
41#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
42
43static struct kmem_cache *field_cachep;
44static struct kmem_cache *file_cachep;
45
46static inline int system_refcount(struct event_subsystem *system)
47{
48 return system->ref_count;
49}
50
51static int system_refcount_inc(struct event_subsystem *system)
52{
53 return system->ref_count++;
54}
55
56static int system_refcount_dec(struct event_subsystem *system)
57{
58 return --system->ref_count;
59}
60
61/* Double loops, do not use break, only goto's work */
62#define do_for_each_event_file(tr, file) \
63 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
64 list_for_each_entry(file, &tr->events, list)
65
66#define do_for_each_event_file_safe(tr, file) \
67 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
68 struct trace_event_file *___n; \
69 list_for_each_entry_safe(file, ___n, &tr->events, list)
70
71#define while_for_each_event_file() \
72 }
73
74static struct ftrace_event_field *
75__find_event_field(struct list_head *head, char *name)
76{
77 struct ftrace_event_field *field;
78
79 list_for_each_entry(field, head, link) {
80 if (!strcmp(field->name, name))
81 return field;
82 }
83
84 return NULL;
85}
86
87struct ftrace_event_field *
88trace_find_event_field(struct trace_event_call *call, char *name)
89{
90 struct ftrace_event_field *field;
91 struct list_head *head;
92
93 head = trace_get_fields(call);
94 field = __find_event_field(head, name);
95 if (field)
96 return field;
97
98 field = __find_event_field(&ftrace_generic_fields, name);
99 if (field)
100 return field;
101
102 return __find_event_field(&ftrace_common_fields, name);
103}
104
105static int __trace_define_field(struct list_head *head, const char *type,
106 const char *name, int offset, int size,
107 int is_signed, int filter_type)
108{
109 struct ftrace_event_field *field;
110
111 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
112 if (!field)
113 return -ENOMEM;
114
115 field->name = name;
116 field->type = type;
117
118 if (filter_type == FILTER_OTHER)
119 field->filter_type = filter_assign_type(type);
120 else
121 field->filter_type = filter_type;
122
123 field->offset = offset;
124 field->size = size;
125 field->is_signed = is_signed;
126
127 list_add(&field->link, head);
128
129 return 0;
130}
131
132int trace_define_field(struct trace_event_call *call, const char *type,
133 const char *name, int offset, int size, int is_signed,
134 int filter_type)
135{
136 struct list_head *head;
137
138 if (WARN_ON(!call->class))
139 return 0;
140
141 head = trace_get_fields(call);
142 return __trace_define_field(head, type, name, offset, size,
143 is_signed, filter_type);
144}
145EXPORT_SYMBOL_GPL(trace_define_field);
146
147#define __generic_field(type, item, filter_type) \
148 ret = __trace_define_field(&ftrace_generic_fields, #type, \
149 #item, 0, 0, is_signed_type(type), \
150 filter_type); \
151 if (ret) \
152 return ret;
153
154#define __common_field(type, item) \
155 ret = __trace_define_field(&ftrace_common_fields, #type, \
156 "common_" #item, \
157 offsetof(typeof(ent), item), \
158 sizeof(ent.item), \
159 is_signed_type(type), FILTER_OTHER); \
160 if (ret) \
161 return ret;
162
163static int trace_define_generic_fields(void)
164{
165 int ret;
166
167 __generic_field(int, CPU, FILTER_CPU);
168 __generic_field(int, cpu, FILTER_CPU);
169 __generic_field(char *, COMM, FILTER_COMM);
170 __generic_field(char *, comm, FILTER_COMM);
171
172 return ret;
173}
174
175static int trace_define_common_fields(void)
176{
177 int ret;
178 struct trace_entry ent;
179
180 __common_field(unsigned short, type);
181 __common_field(unsigned char, flags);
182 __common_field(unsigned char, preempt_count);
183 __common_field(int, pid);
184
185 return ret;
186}
187
188static void trace_destroy_fields(struct trace_event_call *call)
189{
190 struct ftrace_event_field *field, *next;
191 struct list_head *head;
192
193 head = trace_get_fields(call);
194 list_for_each_entry_safe(field, next, head, link) {
195 list_del(&field->link);
196 kmem_cache_free(field_cachep, field);
197 }
198}
199
200/*
201 * run-time version of trace_event_get_offsets_<call>() that returns the last
202 * accessible offset of trace fields excluding __dynamic_array bytes
203 */
204int trace_event_get_offsets(struct trace_event_call *call)
205{
206 struct ftrace_event_field *tail;
207 struct list_head *head;
208
209 head = trace_get_fields(call);
210 /*
211 * head->next points to the last field with the largest offset,
212 * since it was added last by trace_define_field()
213 */
214 tail = list_first_entry(head, struct ftrace_event_field, link);
215 return tail->offset + tail->size;
216}
217
218int trace_event_raw_init(struct trace_event_call *call)
219{
220 int id;
221
222 id = register_trace_event(&call->event);
223 if (!id)
224 return -ENODEV;
225
226 return 0;
227}
228EXPORT_SYMBOL_GPL(trace_event_raw_init);
229
230bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
231{
232 struct trace_array *tr = trace_file->tr;
233 struct trace_array_cpu *data;
234 struct trace_pid_list *pid_list;
235
236 pid_list = rcu_dereference_raw(tr->filtered_pids);
237 if (!pid_list)
238 return false;
239
240 data = this_cpu_ptr(tr->trace_buffer.data);
241
242 return data->ignore_pid;
243}
244EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
245
246void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
247 struct trace_event_file *trace_file,
248 unsigned long len)
249{
250 struct trace_event_call *event_call = trace_file->event_call;
251
252 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
253 trace_event_ignore_this_pid(trace_file))
254 return NULL;
255
256 local_save_flags(fbuffer->flags);
257 fbuffer->pc = preempt_count();
258 /*
259 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
260 * preemption (adding one to the preempt_count). Since we are
261 * interested in the preempt_count at the time the tracepoint was
262 * hit, we need to subtract one to offset the increment.
263 */
264 if (IS_ENABLED(CONFIG_PREEMPTION))
265 fbuffer->pc--;
266 fbuffer->trace_file = trace_file;
267
268 fbuffer->event =
269 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
270 event_call->event.type, len,
271 fbuffer->flags, fbuffer->pc);
272 if (!fbuffer->event)
273 return NULL;
274
275 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
276 return fbuffer->entry;
277}
278EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
279
280int trace_event_reg(struct trace_event_call *call,
281 enum trace_reg type, void *data)
282{
283 struct trace_event_file *file = data;
284
285 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
286 switch (type) {
287 case TRACE_REG_REGISTER:
288 return tracepoint_probe_register(call->tp,
289 call->class->probe,
290 file);
291 case TRACE_REG_UNREGISTER:
292 tracepoint_probe_unregister(call->tp,
293 call->class->probe,
294 file);
295 return 0;
296
297#ifdef CONFIG_PERF_EVENTS
298 case TRACE_REG_PERF_REGISTER:
299 return tracepoint_probe_register(call->tp,
300 call->class->perf_probe,
301 call);
302 case TRACE_REG_PERF_UNREGISTER:
303 tracepoint_probe_unregister(call->tp,
304 call->class->perf_probe,
305 call);
306 return 0;
307 case TRACE_REG_PERF_OPEN:
308 case TRACE_REG_PERF_CLOSE:
309 case TRACE_REG_PERF_ADD:
310 case TRACE_REG_PERF_DEL:
311 return 0;
312#endif
313 }
314 return 0;
315}
316EXPORT_SYMBOL_GPL(trace_event_reg);
317
318void trace_event_enable_cmd_record(bool enable)
319{
320 struct trace_event_file *file;
321 struct trace_array *tr;
322
323 mutex_lock(&event_mutex);
324 do_for_each_event_file(tr, file) {
325
326 if (!(file->flags & EVENT_FILE_FL_ENABLED))
327 continue;
328
329 if (enable) {
330 tracing_start_cmdline_record();
331 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
332 } else {
333 tracing_stop_cmdline_record();
334 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
335 }
336 } while_for_each_event_file();
337 mutex_unlock(&event_mutex);
338}
339
340void trace_event_enable_tgid_record(bool enable)
341{
342 struct trace_event_file *file;
343 struct trace_array *tr;
344
345 mutex_lock(&event_mutex);
346 do_for_each_event_file(tr, file) {
347 if (!(file->flags & EVENT_FILE_FL_ENABLED))
348 continue;
349
350 if (enable) {
351 tracing_start_tgid_record();
352 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
353 } else {
354 tracing_stop_tgid_record();
355 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
356 &file->flags);
357 }
358 } while_for_each_event_file();
359 mutex_unlock(&event_mutex);
360}
361
362static int __ftrace_event_enable_disable(struct trace_event_file *file,
363 int enable, int soft_disable)
364{
365 struct trace_event_call *call = file->event_call;
366 struct trace_array *tr = file->tr;
367 unsigned long file_flags = file->flags;
368 int ret = 0;
369 int disable;
370
371 switch (enable) {
372 case 0:
373 /*
374 * When soft_disable is set and enable is cleared, the sm_ref
375 * reference counter is decremented. If it reaches 0, we want
376 * to clear the SOFT_DISABLED flag but leave the event in the
377 * state that it was. That is, if the event was enabled and
378 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
379 * is set we do not want the event to be enabled before we
380 * clear the bit.
381 *
382 * When soft_disable is not set but the SOFT_MODE flag is,
383 * we do nothing. Do not disable the tracepoint, otherwise
384 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
385 */
386 if (soft_disable) {
387 if (atomic_dec_return(&file->sm_ref) > 0)
388 break;
389 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
390 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
391 } else
392 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
393
394 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
395 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
396 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
397 tracing_stop_cmdline_record();
398 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
399 }
400
401 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
402 tracing_stop_tgid_record();
403 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
404 }
405
406 call->class->reg(call, TRACE_REG_UNREGISTER, file);
407 }
408 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
409 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
410 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
411 else
412 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
413 break;
414 case 1:
415 /*
416 * When soft_disable is set and enable is set, we want to
417 * register the tracepoint for the event, but leave the event
418 * as is. That means, if the event was already enabled, we do
419 * nothing (but set SOFT_MODE). If the event is disabled, we
420 * set SOFT_DISABLED before enabling the event tracepoint, so
421 * it still seems to be disabled.
422 */
423 if (!soft_disable)
424 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
425 else {
426 if (atomic_inc_return(&file->sm_ref) > 1)
427 break;
428 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
429 }
430
431 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
432 bool cmd = false, tgid = false;
433
434 /* Keep the event disabled, when going to SOFT_MODE. */
435 if (soft_disable)
436 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
437
438 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
439 cmd = true;
440 tracing_start_cmdline_record();
441 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
442 }
443
444 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
445 tgid = true;
446 tracing_start_tgid_record();
447 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
448 }
449
450 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
451 if (ret) {
452 if (cmd)
453 tracing_stop_cmdline_record();
454 if (tgid)
455 tracing_stop_tgid_record();
456 pr_info("event trace: Could not enable event "
457 "%s\n", trace_event_name(call));
458 break;
459 }
460 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
461
462 /* WAS_ENABLED gets set but never cleared. */
463 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
464 }
465 break;
466 }
467
468 /* Enable or disable use of trace_buffered_event */
469 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
470 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
471 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
472 trace_buffered_event_enable();
473 else
474 trace_buffered_event_disable();
475 }
476
477 return ret;
478}
479
480int trace_event_enable_disable(struct trace_event_file *file,
481 int enable, int soft_disable)
482{
483 return __ftrace_event_enable_disable(file, enable, soft_disable);
484}
485
486static int ftrace_event_enable_disable(struct trace_event_file *file,
487 int enable)
488{
489 return __ftrace_event_enable_disable(file, enable, 0);
490}
491
492static void ftrace_clear_events(struct trace_array *tr)
493{
494 struct trace_event_file *file;
495
496 mutex_lock(&event_mutex);
497 list_for_each_entry(file, &tr->events, list) {
498 ftrace_event_enable_disable(file, 0);
499 }
500 mutex_unlock(&event_mutex);
501}
502
503static void
504event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
505{
506 struct trace_pid_list *pid_list;
507 struct trace_array *tr = data;
508
509 pid_list = rcu_dereference_raw(tr->filtered_pids);
510 trace_filter_add_remove_task(pid_list, NULL, task);
511}
512
513static void
514event_filter_pid_sched_process_fork(void *data,
515 struct task_struct *self,
516 struct task_struct *task)
517{
518 struct trace_pid_list *pid_list;
519 struct trace_array *tr = data;
520
521 pid_list = rcu_dereference_sched(tr->filtered_pids);
522 trace_filter_add_remove_task(pid_list, self, task);
523}
524
525void trace_event_follow_fork(struct trace_array *tr, bool enable)
526{
527 if (enable) {
528 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
529 tr, INT_MIN);
530 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
531 tr, INT_MAX);
532 } else {
533 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
534 tr);
535 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
536 tr);
537 }
538}
539
540static void
541event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
542 struct task_struct *prev, struct task_struct *next)
543{
544 struct trace_array *tr = data;
545 struct trace_pid_list *pid_list;
546
547 pid_list = rcu_dereference_sched(tr->filtered_pids);
548
549 this_cpu_write(tr->trace_buffer.data->ignore_pid,
550 trace_ignore_this_task(pid_list, prev) &&
551 trace_ignore_this_task(pid_list, next));
552}
553
554static void
555event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
556 struct task_struct *prev, struct task_struct *next)
557{
558 struct trace_array *tr = data;
559 struct trace_pid_list *pid_list;
560
561 pid_list = rcu_dereference_sched(tr->filtered_pids);
562
563 this_cpu_write(tr->trace_buffer.data->ignore_pid,
564 trace_ignore_this_task(pid_list, next));
565}
566
567static void
568event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
569{
570 struct trace_array *tr = data;
571 struct trace_pid_list *pid_list;
572
573 /* Nothing to do if we are already tracing */
574 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
575 return;
576
577 pid_list = rcu_dereference_sched(tr->filtered_pids);
578
579 this_cpu_write(tr->trace_buffer.data->ignore_pid,
580 trace_ignore_this_task(pid_list, task));
581}
582
583static void
584event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
585{
586 struct trace_array *tr = data;
587 struct trace_pid_list *pid_list;
588
589 /* Nothing to do if we are not tracing */
590 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
591 return;
592
593 pid_list = rcu_dereference_sched(tr->filtered_pids);
594
595 /* Set tracing if current is enabled */
596 this_cpu_write(tr->trace_buffer.data->ignore_pid,
597 trace_ignore_this_task(pid_list, current));
598}
599
600static void __ftrace_clear_event_pids(struct trace_array *tr)
601{
602 struct trace_pid_list *pid_list;
603 struct trace_event_file *file;
604 int cpu;
605
606 pid_list = rcu_dereference_protected(tr->filtered_pids,
607 lockdep_is_held(&event_mutex));
608 if (!pid_list)
609 return;
610
611 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
612 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
613
614 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
615 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
616
617 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
618 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
619
620 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
621 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
622
623 list_for_each_entry(file, &tr->events, list) {
624 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
625 }
626
627 for_each_possible_cpu(cpu)
628 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
629
630 rcu_assign_pointer(tr->filtered_pids, NULL);
631
632 /* Wait till all users are no longer using pid filtering */
633 tracepoint_synchronize_unregister();
634
635 trace_free_pid_list(pid_list);
636}
637
638static void ftrace_clear_event_pids(struct trace_array *tr)
639{
640 mutex_lock(&event_mutex);
641 __ftrace_clear_event_pids(tr);
642 mutex_unlock(&event_mutex);
643}
644
645static void __put_system(struct event_subsystem *system)
646{
647 struct event_filter *filter = system->filter;
648
649 WARN_ON_ONCE(system_refcount(system) == 0);
650 if (system_refcount_dec(system))
651 return;
652
653 list_del(&system->list);
654
655 if (filter) {
656 kfree(filter->filter_string);
657 kfree(filter);
658 }
659 kfree_const(system->name);
660 kfree(system);
661}
662
663static void __get_system(struct event_subsystem *system)
664{
665 WARN_ON_ONCE(system_refcount(system) == 0);
666 system_refcount_inc(system);
667}
668
669static void __get_system_dir(struct trace_subsystem_dir *dir)
670{
671 WARN_ON_ONCE(dir->ref_count == 0);
672 dir->ref_count++;
673 __get_system(dir->subsystem);
674}
675
676static void __put_system_dir(struct trace_subsystem_dir *dir)
677{
678 WARN_ON_ONCE(dir->ref_count == 0);
679 /* If the subsystem is about to be freed, the dir must be too */
680 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
681
682 __put_system(dir->subsystem);
683 if (!--dir->ref_count)
684 kfree(dir);
685}
686
687static void put_system(struct trace_subsystem_dir *dir)
688{
689 mutex_lock(&event_mutex);
690 __put_system_dir(dir);
691 mutex_unlock(&event_mutex);
692}
693
694static void remove_subsystem(struct trace_subsystem_dir *dir)
695{
696 if (!dir)
697 return;
698
699 if (!--dir->nr_events) {
700 tracefs_remove_recursive(dir->entry);
701 list_del(&dir->list);
702 __put_system_dir(dir);
703 }
704}
705
706static void remove_event_file_dir(struct trace_event_file *file)
707{
708 struct dentry *dir = file->dir;
709 struct dentry *child;
710
711 if (dir) {
712 spin_lock(&dir->d_lock); /* probably unneeded */
713 list_for_each_entry(child, &dir->d_subdirs, d_child) {
714 if (d_really_is_positive(child)) /* probably unneeded */
715 d_inode(child)->i_private = NULL;
716 }
717 spin_unlock(&dir->d_lock);
718
719 tracefs_remove_recursive(dir);
720 }
721
722 list_del(&file->list);
723 remove_subsystem(file->system);
724 free_event_filter(file->filter);
725 kmem_cache_free(file_cachep, file);
726}
727
728/*
729 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
730 */
731static int
732__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
733 const char *sub, const char *event, int set)
734{
735 struct trace_event_file *file;
736 struct trace_event_call *call;
737 const char *name;
738 int ret = -EINVAL;
739 int eret = 0;
740
741 list_for_each_entry(file, &tr->events, list) {
742
743 call = file->event_call;
744 name = trace_event_name(call);
745
746 if (!name || !call->class || !call->class->reg)
747 continue;
748
749 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
750 continue;
751
752 if (match &&
753 strcmp(match, name) != 0 &&
754 strcmp(match, call->class->system) != 0)
755 continue;
756
757 if (sub && strcmp(sub, call->class->system) != 0)
758 continue;
759
760 if (event && strcmp(event, name) != 0)
761 continue;
762
763 ret = ftrace_event_enable_disable(file, set);
764
765 /*
766 * Save the first error and return that. Some events
767 * may still have been enabled, but let the user
768 * know that something went wrong.
769 */
770 if (ret && !eret)
771 eret = ret;
772
773 ret = eret;
774 }
775
776 return ret;
777}
778
779static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
780 const char *sub, const char *event, int set)
781{
782 int ret;
783
784 mutex_lock(&event_mutex);
785 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
786 mutex_unlock(&event_mutex);
787
788 return ret;
789}
790
791int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
792{
793 char *event = NULL, *sub = NULL, *match;
794 int ret;
795
796 /*
797 * The buf format can be <subsystem>:<event-name>
798 * *:<event-name> means any event by that name.
799 * :<event-name> is the same.
800 *
801 * <subsystem>:* means all events in that subsystem
802 * <subsystem>: means the same.
803 *
804 * <name> (no ':') means all events in a subsystem with
805 * the name <name> or any event that matches <name>
806 */
807
808 match = strsep(&buf, ":");
809 if (buf) {
810 sub = match;
811 event = buf;
812 match = NULL;
813
814 if (!strlen(sub) || strcmp(sub, "*") == 0)
815 sub = NULL;
816 if (!strlen(event) || strcmp(event, "*") == 0)
817 event = NULL;
818 }
819
820 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
821
822 /* Put back the colon to allow this to be called again */
823 if (buf)
824 *(buf - 1) = ':';
825
826 return ret;
827}
828EXPORT_SYMBOL_GPL(ftrace_set_clr_event);
829
830/**
831 * trace_set_clr_event - enable or disable an event
832 * @system: system name to match (NULL for any system)
833 * @event: event name to match (NULL for all events, within system)
834 * @set: 1 to enable, 0 to disable
835 *
836 * This is a way for other parts of the kernel to enable or disable
837 * event recording.
838 *
839 * Returns 0 on success, -EINVAL if the parameters do not match any
840 * registered events.
841 */
842int trace_set_clr_event(const char *system, const char *event, int set)
843{
844 struct trace_array *tr = top_trace_array();
845
846 if (!tr)
847 return -ENODEV;
848
849 return __ftrace_set_clr_event(tr, NULL, system, event, set);
850}
851EXPORT_SYMBOL_GPL(trace_set_clr_event);
852
853/* 128 should be much more than enough */
854#define EVENT_BUF_SIZE 127
855
856static ssize_t
857ftrace_event_write(struct file *file, const char __user *ubuf,
858 size_t cnt, loff_t *ppos)
859{
860 struct trace_parser parser;
861 struct seq_file *m = file->private_data;
862 struct trace_array *tr = m->private;
863 ssize_t read, ret;
864
865 if (!cnt)
866 return 0;
867
868 ret = tracing_update_buffers();
869 if (ret < 0)
870 return ret;
871
872 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
873 return -ENOMEM;
874
875 read = trace_get_user(&parser, ubuf, cnt, ppos);
876
877 if (read >= 0 && trace_parser_loaded((&parser))) {
878 int set = 1;
879
880 if (*parser.buffer == '!')
881 set = 0;
882
883 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
884 if (ret)
885 goto out_put;
886 }
887
888 ret = read;
889
890 out_put:
891 trace_parser_put(&parser);
892
893 return ret;
894}
895
896static void *
897t_next(struct seq_file *m, void *v, loff_t *pos)
898{
899 struct trace_event_file *file = v;
900 struct trace_event_call *call;
901 struct trace_array *tr = m->private;
902
903 (*pos)++;
904
905 list_for_each_entry_continue(file, &tr->events, list) {
906 call = file->event_call;
907 /*
908 * The ftrace subsystem is for showing formats only.
909 * They can not be enabled or disabled via the event files.
910 */
911 if (call->class && call->class->reg &&
912 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
913 return file;
914 }
915
916 return NULL;
917}
918
919static void *t_start(struct seq_file *m, loff_t *pos)
920{
921 struct trace_event_file *file;
922 struct trace_array *tr = m->private;
923 loff_t l;
924
925 mutex_lock(&event_mutex);
926
927 file = list_entry(&tr->events, struct trace_event_file, list);
928 for (l = 0; l <= *pos; ) {
929 file = t_next(m, file, &l);
930 if (!file)
931 break;
932 }
933 return file;
934}
935
936static void *
937s_next(struct seq_file *m, void *v, loff_t *pos)
938{
939 struct trace_event_file *file = v;
940 struct trace_array *tr = m->private;
941
942 (*pos)++;
943
944 list_for_each_entry_continue(file, &tr->events, list) {
945 if (file->flags & EVENT_FILE_FL_ENABLED)
946 return file;
947 }
948
949 return NULL;
950}
951
952static void *s_start(struct seq_file *m, loff_t *pos)
953{
954 struct trace_event_file *file;
955 struct trace_array *tr = m->private;
956 loff_t l;
957
958 mutex_lock(&event_mutex);
959
960 file = list_entry(&tr->events, struct trace_event_file, list);
961 for (l = 0; l <= *pos; ) {
962 file = s_next(m, file, &l);
963 if (!file)
964 break;
965 }
966 return file;
967}
968
969static int t_show(struct seq_file *m, void *v)
970{
971 struct trace_event_file *file = v;
972 struct trace_event_call *call = file->event_call;
973
974 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
975 seq_printf(m, "%s:", call->class->system);
976 seq_printf(m, "%s\n", trace_event_name(call));
977
978 return 0;
979}
980
981static void t_stop(struct seq_file *m, void *p)
982{
983 mutex_unlock(&event_mutex);
984}
985
986static void *
987p_next(struct seq_file *m, void *v, loff_t *pos)
988{
989 struct trace_array *tr = m->private;
990 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
991
992 return trace_pid_next(pid_list, v, pos);
993}
994
995static void *p_start(struct seq_file *m, loff_t *pos)
996 __acquires(RCU)
997{
998 struct trace_pid_list *pid_list;
999 struct trace_array *tr = m->private;
1000
1001 /*
1002 * Grab the mutex, to keep calls to p_next() having the same
1003 * tr->filtered_pids as p_start() has.
1004 * If we just passed the tr->filtered_pids around, then RCU would
1005 * have been enough, but doing that makes things more complex.
1006 */
1007 mutex_lock(&event_mutex);
1008 rcu_read_lock_sched();
1009
1010 pid_list = rcu_dereference_sched(tr->filtered_pids);
1011
1012 if (!pid_list)
1013 return NULL;
1014
1015 return trace_pid_start(pid_list, pos);
1016}
1017
1018static void p_stop(struct seq_file *m, void *p)
1019 __releases(RCU)
1020{
1021 rcu_read_unlock_sched();
1022 mutex_unlock(&event_mutex);
1023}
1024
1025static ssize_t
1026event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1027 loff_t *ppos)
1028{
1029 struct trace_event_file *file;
1030 unsigned long flags;
1031 char buf[4] = "0";
1032
1033 mutex_lock(&event_mutex);
1034 file = event_file_data(filp);
1035 if (likely(file))
1036 flags = file->flags;
1037 mutex_unlock(&event_mutex);
1038
1039 if (!file)
1040 return -ENODEV;
1041
1042 if (flags & EVENT_FILE_FL_ENABLED &&
1043 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1044 strcpy(buf, "1");
1045
1046 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1047 flags & EVENT_FILE_FL_SOFT_MODE)
1048 strcat(buf, "*");
1049
1050 strcat(buf, "\n");
1051
1052 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1053}
1054
1055static ssize_t
1056event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1057 loff_t *ppos)
1058{
1059 struct trace_event_file *file;
1060 unsigned long val;
1061 int ret;
1062
1063 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1064 if (ret)
1065 return ret;
1066
1067 ret = tracing_update_buffers();
1068 if (ret < 0)
1069 return ret;
1070
1071 switch (val) {
1072 case 0:
1073 case 1:
1074 ret = -ENODEV;
1075 mutex_lock(&event_mutex);
1076 file = event_file_data(filp);
1077 if (likely(file))
1078 ret = ftrace_event_enable_disable(file, val);
1079 mutex_unlock(&event_mutex);
1080 break;
1081
1082 default:
1083 return -EINVAL;
1084 }
1085
1086 *ppos += cnt;
1087
1088 return ret ? ret : cnt;
1089}
1090
1091static ssize_t
1092system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1093 loff_t *ppos)
1094{
1095 const char set_to_char[4] = { '?', '0', '1', 'X' };
1096 struct trace_subsystem_dir *dir = filp->private_data;
1097 struct event_subsystem *system = dir->subsystem;
1098 struct trace_event_call *call;
1099 struct trace_event_file *file;
1100 struct trace_array *tr = dir->tr;
1101 char buf[2];
1102 int set = 0;
1103 int ret;
1104
1105 mutex_lock(&event_mutex);
1106 list_for_each_entry(file, &tr->events, list) {
1107 call = file->event_call;
1108 if (!trace_event_name(call) || !call->class || !call->class->reg)
1109 continue;
1110
1111 if (system && strcmp(call->class->system, system->name) != 0)
1112 continue;
1113
1114 /*
1115 * We need to find out if all the events are set
1116 * or if all events or cleared, or if we have
1117 * a mixture.
1118 */
1119 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1120
1121 /*
1122 * If we have a mixture, no need to look further.
1123 */
1124 if (set == 3)
1125 break;
1126 }
1127 mutex_unlock(&event_mutex);
1128
1129 buf[0] = set_to_char[set];
1130 buf[1] = '\n';
1131
1132 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1133
1134 return ret;
1135}
1136
1137static ssize_t
1138system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1139 loff_t *ppos)
1140{
1141 struct trace_subsystem_dir *dir = filp->private_data;
1142 struct event_subsystem *system = dir->subsystem;
1143 const char *name = NULL;
1144 unsigned long val;
1145 ssize_t ret;
1146
1147 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1148 if (ret)
1149 return ret;
1150
1151 ret = tracing_update_buffers();
1152 if (ret < 0)
1153 return ret;
1154
1155 if (val != 0 && val != 1)
1156 return -EINVAL;
1157
1158 /*
1159 * Opening of "enable" adds a ref count to system,
1160 * so the name is safe to use.
1161 */
1162 if (system)
1163 name = system->name;
1164
1165 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1166 if (ret)
1167 goto out;
1168
1169 ret = cnt;
1170
1171out:
1172 *ppos += cnt;
1173
1174 return ret;
1175}
1176
1177enum {
1178 FORMAT_HEADER = 1,
1179 FORMAT_FIELD_SEPERATOR = 2,
1180 FORMAT_PRINTFMT = 3,
1181};
1182
1183static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1184{
1185 struct trace_event_call *call = event_file_data(m->private);
1186 struct list_head *common_head = &ftrace_common_fields;
1187 struct list_head *head = trace_get_fields(call);
1188 struct list_head *node = v;
1189
1190 (*pos)++;
1191
1192 switch ((unsigned long)v) {
1193 case FORMAT_HEADER:
1194 node = common_head;
1195 break;
1196
1197 case FORMAT_FIELD_SEPERATOR:
1198 node = head;
1199 break;
1200
1201 case FORMAT_PRINTFMT:
1202 /* all done */
1203 return NULL;
1204 }
1205
1206 node = node->prev;
1207 if (node == common_head)
1208 return (void *)FORMAT_FIELD_SEPERATOR;
1209 else if (node == head)
1210 return (void *)FORMAT_PRINTFMT;
1211 else
1212 return node;
1213}
1214
1215static int f_show(struct seq_file *m, void *v)
1216{
1217 struct trace_event_call *call = event_file_data(m->private);
1218 struct ftrace_event_field *field;
1219 const char *array_descriptor;
1220
1221 switch ((unsigned long)v) {
1222 case FORMAT_HEADER:
1223 seq_printf(m, "name: %s\n", trace_event_name(call));
1224 seq_printf(m, "ID: %d\n", call->event.type);
1225 seq_puts(m, "format:\n");
1226 return 0;
1227
1228 case FORMAT_FIELD_SEPERATOR:
1229 seq_putc(m, '\n');
1230 return 0;
1231
1232 case FORMAT_PRINTFMT:
1233 seq_printf(m, "\nprint fmt: %s\n",
1234 call->print_fmt);
1235 return 0;
1236 }
1237
1238 field = list_entry(v, struct ftrace_event_field, link);
1239 /*
1240 * Smartly shows the array type(except dynamic array).
1241 * Normal:
1242 * field:TYPE VAR
1243 * If TYPE := TYPE[LEN], it is shown:
1244 * field:TYPE VAR[LEN]
1245 */
1246 array_descriptor = strchr(field->type, '[');
1247
1248 if (str_has_prefix(field->type, "__data_loc"))
1249 array_descriptor = NULL;
1250
1251 if (!array_descriptor)
1252 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1253 field->type, field->name, field->offset,
1254 field->size, !!field->is_signed);
1255 else
1256 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1257 (int)(array_descriptor - field->type),
1258 field->type, field->name,
1259 array_descriptor, field->offset,
1260 field->size, !!field->is_signed);
1261
1262 return 0;
1263}
1264
1265static void *f_start(struct seq_file *m, loff_t *pos)
1266{
1267 void *p = (void *)FORMAT_HEADER;
1268 loff_t l = 0;
1269
1270 /* ->stop() is called even if ->start() fails */
1271 mutex_lock(&event_mutex);
1272 if (!event_file_data(m->private))
1273 return ERR_PTR(-ENODEV);
1274
1275 while (l < *pos && p)
1276 p = f_next(m, p, &l);
1277
1278 return p;
1279}
1280
1281static void f_stop(struct seq_file *m, void *p)
1282{
1283 mutex_unlock(&event_mutex);
1284}
1285
1286static const struct seq_operations trace_format_seq_ops = {
1287 .start = f_start,
1288 .next = f_next,
1289 .stop = f_stop,
1290 .show = f_show,
1291};
1292
1293static int trace_format_open(struct inode *inode, struct file *file)
1294{
1295 struct seq_file *m;
1296 int ret;
1297
1298 /* Do we want to hide event format files on tracefs lockdown? */
1299
1300 ret = seq_open(file, &trace_format_seq_ops);
1301 if (ret < 0)
1302 return ret;
1303
1304 m = file->private_data;
1305 m->private = file;
1306
1307 return 0;
1308}
1309
1310static ssize_t
1311event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1312{
1313 int id = (long)event_file_data(filp);
1314 char buf[32];
1315 int len;
1316
1317 if (unlikely(!id))
1318 return -ENODEV;
1319
1320 len = sprintf(buf, "%d\n", id);
1321
1322 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1323}
1324
1325static ssize_t
1326event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1327 loff_t *ppos)
1328{
1329 struct trace_event_file *file;
1330 struct trace_seq *s;
1331 int r = -ENODEV;
1332
1333 if (*ppos)
1334 return 0;
1335
1336 s = kmalloc(sizeof(*s), GFP_KERNEL);
1337
1338 if (!s)
1339 return -ENOMEM;
1340
1341 trace_seq_init(s);
1342
1343 mutex_lock(&event_mutex);
1344 file = event_file_data(filp);
1345 if (file)
1346 print_event_filter(file, s);
1347 mutex_unlock(&event_mutex);
1348
1349 if (file)
1350 r = simple_read_from_buffer(ubuf, cnt, ppos,
1351 s->buffer, trace_seq_used(s));
1352
1353 kfree(s);
1354
1355 return r;
1356}
1357
1358static ssize_t
1359event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1360 loff_t *ppos)
1361{
1362 struct trace_event_file *file;
1363 char *buf;
1364 int err = -ENODEV;
1365
1366 if (cnt >= PAGE_SIZE)
1367 return -EINVAL;
1368
1369 buf = memdup_user_nul(ubuf, cnt);
1370 if (IS_ERR(buf))
1371 return PTR_ERR(buf);
1372
1373 mutex_lock(&event_mutex);
1374 file = event_file_data(filp);
1375 if (file)
1376 err = apply_event_filter(file, buf);
1377 mutex_unlock(&event_mutex);
1378
1379 kfree(buf);
1380 if (err < 0)
1381 return err;
1382
1383 *ppos += cnt;
1384
1385 return cnt;
1386}
1387
1388static LIST_HEAD(event_subsystems);
1389
1390static int subsystem_open(struct inode *inode, struct file *filp)
1391{
1392 struct event_subsystem *system = NULL;
1393 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1394 struct trace_array *tr;
1395 int ret;
1396
1397 if (tracing_is_disabled())
1398 return -ENODEV;
1399
1400 /* Make sure the system still exists */
1401 mutex_lock(&event_mutex);
1402 mutex_lock(&trace_types_lock);
1403 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1404 list_for_each_entry(dir, &tr->systems, list) {
1405 if (dir == inode->i_private) {
1406 /* Don't open systems with no events */
1407 if (dir->nr_events) {
1408 __get_system_dir(dir);
1409 system = dir->subsystem;
1410 }
1411 goto exit_loop;
1412 }
1413 }
1414 }
1415 exit_loop:
1416 mutex_unlock(&trace_types_lock);
1417 mutex_unlock(&event_mutex);
1418
1419 if (!system)
1420 return -ENODEV;
1421
1422 /* Some versions of gcc think dir can be uninitialized here */
1423 WARN_ON(!dir);
1424
1425 /* Still need to increment the ref count of the system */
1426 if (trace_array_get(tr) < 0) {
1427 put_system(dir);
1428 return -ENODEV;
1429 }
1430
1431 ret = tracing_open_generic(inode, filp);
1432 if (ret < 0) {
1433 trace_array_put(tr);
1434 put_system(dir);
1435 }
1436
1437 return ret;
1438}
1439
1440static int system_tr_open(struct inode *inode, struct file *filp)
1441{
1442 struct trace_subsystem_dir *dir;
1443 struct trace_array *tr = inode->i_private;
1444 int ret;
1445
1446 /* Make a temporary dir that has no system but points to tr */
1447 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1448 if (!dir)
1449 return -ENOMEM;
1450
1451 ret = tracing_open_generic_tr(inode, filp);
1452 if (ret < 0) {
1453 kfree(dir);
1454 return ret;
1455 }
1456 dir->tr = tr;
1457 filp->private_data = dir;
1458
1459 return 0;
1460}
1461
1462static int subsystem_release(struct inode *inode, struct file *file)
1463{
1464 struct trace_subsystem_dir *dir = file->private_data;
1465
1466 trace_array_put(dir->tr);
1467
1468 /*
1469 * If dir->subsystem is NULL, then this is a temporary
1470 * descriptor that was made for a trace_array to enable
1471 * all subsystems.
1472 */
1473 if (dir->subsystem)
1474 put_system(dir);
1475 else
1476 kfree(dir);
1477
1478 return 0;
1479}
1480
1481static ssize_t
1482subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1483 loff_t *ppos)
1484{
1485 struct trace_subsystem_dir *dir = filp->private_data;
1486 struct event_subsystem *system = dir->subsystem;
1487 struct trace_seq *s;
1488 int r;
1489
1490 if (*ppos)
1491 return 0;
1492
1493 s = kmalloc(sizeof(*s), GFP_KERNEL);
1494 if (!s)
1495 return -ENOMEM;
1496
1497 trace_seq_init(s);
1498
1499 print_subsystem_event_filter(system, s);
1500 r = simple_read_from_buffer(ubuf, cnt, ppos,
1501 s->buffer, trace_seq_used(s));
1502
1503 kfree(s);
1504
1505 return r;
1506}
1507
1508static ssize_t
1509subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1510 loff_t *ppos)
1511{
1512 struct trace_subsystem_dir *dir = filp->private_data;
1513 char *buf;
1514 int err;
1515
1516 if (cnt >= PAGE_SIZE)
1517 return -EINVAL;
1518
1519 buf = memdup_user_nul(ubuf, cnt);
1520 if (IS_ERR(buf))
1521 return PTR_ERR(buf);
1522
1523 err = apply_subsystem_event_filter(dir, buf);
1524 kfree(buf);
1525 if (err < 0)
1526 return err;
1527
1528 *ppos += cnt;
1529
1530 return cnt;
1531}
1532
1533static ssize_t
1534show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1535{
1536 int (*func)(struct trace_seq *s) = filp->private_data;
1537 struct trace_seq *s;
1538 int r;
1539
1540 if (*ppos)
1541 return 0;
1542
1543 s = kmalloc(sizeof(*s), GFP_KERNEL);
1544 if (!s)
1545 return -ENOMEM;
1546
1547 trace_seq_init(s);
1548
1549 func(s);
1550 r = simple_read_from_buffer(ubuf, cnt, ppos,
1551 s->buffer, trace_seq_used(s));
1552
1553 kfree(s);
1554
1555 return r;
1556}
1557
1558static void ignore_task_cpu(void *data)
1559{
1560 struct trace_array *tr = data;
1561 struct trace_pid_list *pid_list;
1562
1563 /*
1564 * This function is called by on_each_cpu() while the
1565 * event_mutex is held.
1566 */
1567 pid_list = rcu_dereference_protected(tr->filtered_pids,
1568 mutex_is_locked(&event_mutex));
1569
1570 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1571 trace_ignore_this_task(pid_list, current));
1572}
1573
1574static ssize_t
1575ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1576 size_t cnt, loff_t *ppos)
1577{
1578 struct seq_file *m = filp->private_data;
1579 struct trace_array *tr = m->private;
1580 struct trace_pid_list *filtered_pids = NULL;
1581 struct trace_pid_list *pid_list;
1582 struct trace_event_file *file;
1583 ssize_t ret;
1584
1585 if (!cnt)
1586 return 0;
1587
1588 ret = tracing_update_buffers();
1589 if (ret < 0)
1590 return ret;
1591
1592 mutex_lock(&event_mutex);
1593
1594 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1595 lockdep_is_held(&event_mutex));
1596
1597 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1598 if (ret < 0)
1599 goto out;
1600
1601 rcu_assign_pointer(tr->filtered_pids, pid_list);
1602
1603 list_for_each_entry(file, &tr->events, list) {
1604 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1605 }
1606
1607 if (filtered_pids) {
1608 tracepoint_synchronize_unregister();
1609 trace_free_pid_list(filtered_pids);
1610 } else if (pid_list) {
1611 /*
1612 * Register a probe that is called before all other probes
1613 * to set ignore_pid if next or prev do not match.
1614 * Register a probe this is called after all other probes
1615 * to only keep ignore_pid set if next pid matches.
1616 */
1617 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1618 tr, INT_MAX);
1619 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1620 tr, 0);
1621
1622 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1623 tr, INT_MAX);
1624 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1625 tr, 0);
1626
1627 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1628 tr, INT_MAX);
1629 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1630 tr, 0);
1631
1632 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1633 tr, INT_MAX);
1634 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1635 tr, 0);
1636 }
1637
1638 /*
1639 * Ignoring of pids is done at task switch. But we have to
1640 * check for those tasks that are currently running.
1641 * Always do this in case a pid was appended or removed.
1642 */
1643 on_each_cpu(ignore_task_cpu, tr, 1);
1644
1645 out:
1646 mutex_unlock(&event_mutex);
1647
1648 if (ret > 0)
1649 *ppos += ret;
1650
1651 return ret;
1652}
1653
1654static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1655static int ftrace_event_set_open(struct inode *inode, struct file *file);
1656static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1657static int ftrace_event_release(struct inode *inode, struct file *file);
1658
1659static const struct seq_operations show_event_seq_ops = {
1660 .start = t_start,
1661 .next = t_next,
1662 .show = t_show,
1663 .stop = t_stop,
1664};
1665
1666static const struct seq_operations show_set_event_seq_ops = {
1667 .start = s_start,
1668 .next = s_next,
1669 .show = t_show,
1670 .stop = t_stop,
1671};
1672
1673static const struct seq_operations show_set_pid_seq_ops = {
1674 .start = p_start,
1675 .next = p_next,
1676 .show = trace_pid_show,
1677 .stop = p_stop,
1678};
1679
1680static const struct file_operations ftrace_avail_fops = {
1681 .open = ftrace_event_avail_open,
1682 .read = seq_read,
1683 .llseek = seq_lseek,
1684 .release = seq_release,
1685};
1686
1687static const struct file_operations ftrace_set_event_fops = {
1688 .open = ftrace_event_set_open,
1689 .read = seq_read,
1690 .write = ftrace_event_write,
1691 .llseek = seq_lseek,
1692 .release = ftrace_event_release,
1693};
1694
1695static const struct file_operations ftrace_set_event_pid_fops = {
1696 .open = ftrace_event_set_pid_open,
1697 .read = seq_read,
1698 .write = ftrace_event_pid_write,
1699 .llseek = seq_lseek,
1700 .release = ftrace_event_release,
1701};
1702
1703static const struct file_operations ftrace_enable_fops = {
1704 .open = tracing_open_generic,
1705 .read = event_enable_read,
1706 .write = event_enable_write,
1707 .llseek = default_llseek,
1708};
1709
1710static const struct file_operations ftrace_event_format_fops = {
1711 .open = trace_format_open,
1712 .read = seq_read,
1713 .llseek = seq_lseek,
1714 .release = seq_release,
1715};
1716
1717static const struct file_operations ftrace_event_id_fops = {
1718 .read = event_id_read,
1719 .llseek = default_llseek,
1720};
1721
1722static const struct file_operations ftrace_event_filter_fops = {
1723 .open = tracing_open_generic,
1724 .read = event_filter_read,
1725 .write = event_filter_write,
1726 .llseek = default_llseek,
1727};
1728
1729static const struct file_operations ftrace_subsystem_filter_fops = {
1730 .open = subsystem_open,
1731 .read = subsystem_filter_read,
1732 .write = subsystem_filter_write,
1733 .llseek = default_llseek,
1734 .release = subsystem_release,
1735};
1736
1737static const struct file_operations ftrace_system_enable_fops = {
1738 .open = subsystem_open,
1739 .read = system_enable_read,
1740 .write = system_enable_write,
1741 .llseek = default_llseek,
1742 .release = subsystem_release,
1743};
1744
1745static const struct file_operations ftrace_tr_enable_fops = {
1746 .open = system_tr_open,
1747 .read = system_enable_read,
1748 .write = system_enable_write,
1749 .llseek = default_llseek,
1750 .release = subsystem_release,
1751};
1752
1753static const struct file_operations ftrace_show_header_fops = {
1754 .open = tracing_open_generic,
1755 .read = show_header,
1756 .llseek = default_llseek,
1757};
1758
1759static int
1760ftrace_event_open(struct inode *inode, struct file *file,
1761 const struct seq_operations *seq_ops)
1762{
1763 struct seq_file *m;
1764 int ret;
1765
1766 ret = security_locked_down(LOCKDOWN_TRACEFS);
1767 if (ret)
1768 return ret;
1769
1770 ret = seq_open(file, seq_ops);
1771 if (ret < 0)
1772 return ret;
1773 m = file->private_data;
1774 /* copy tr over to seq ops */
1775 m->private = inode->i_private;
1776
1777 return ret;
1778}
1779
1780static int ftrace_event_release(struct inode *inode, struct file *file)
1781{
1782 struct trace_array *tr = inode->i_private;
1783
1784 trace_array_put(tr);
1785
1786 return seq_release(inode, file);
1787}
1788
1789static int
1790ftrace_event_avail_open(struct inode *inode, struct file *file)
1791{
1792 const struct seq_operations *seq_ops = &show_event_seq_ops;
1793
1794 /* Checks for tracefs lockdown */
1795 return ftrace_event_open(inode, file, seq_ops);
1796}
1797
1798static int
1799ftrace_event_set_open(struct inode *inode, struct file *file)
1800{
1801 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1802 struct trace_array *tr = inode->i_private;
1803 int ret;
1804
1805 ret = tracing_check_open_get_tr(tr);
1806 if (ret)
1807 return ret;
1808
1809 if ((file->f_mode & FMODE_WRITE) &&
1810 (file->f_flags & O_TRUNC))
1811 ftrace_clear_events(tr);
1812
1813 ret = ftrace_event_open(inode, file, seq_ops);
1814 if (ret < 0)
1815 trace_array_put(tr);
1816 return ret;
1817}
1818
1819static int
1820ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1821{
1822 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1823 struct trace_array *tr = inode->i_private;
1824 int ret;
1825
1826 ret = tracing_check_open_get_tr(tr);
1827 if (ret)
1828 return ret;
1829
1830 if ((file->f_mode & FMODE_WRITE) &&
1831 (file->f_flags & O_TRUNC))
1832 ftrace_clear_event_pids(tr);
1833
1834 ret = ftrace_event_open(inode, file, seq_ops);
1835 if (ret < 0)
1836 trace_array_put(tr);
1837 return ret;
1838}
1839
1840static struct event_subsystem *
1841create_new_subsystem(const char *name)
1842{
1843 struct event_subsystem *system;
1844
1845 /* need to create new entry */
1846 system = kmalloc(sizeof(*system), GFP_KERNEL);
1847 if (!system)
1848 return NULL;
1849
1850 system->ref_count = 1;
1851
1852 /* Only allocate if dynamic (kprobes and modules) */
1853 system->name = kstrdup_const(name, GFP_KERNEL);
1854 if (!system->name)
1855 goto out_free;
1856
1857 system->filter = NULL;
1858
1859 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1860 if (!system->filter)
1861 goto out_free;
1862
1863 list_add(&system->list, &event_subsystems);
1864
1865 return system;
1866
1867 out_free:
1868 kfree_const(system->name);
1869 kfree(system);
1870 return NULL;
1871}
1872
1873static struct dentry *
1874event_subsystem_dir(struct trace_array *tr, const char *name,
1875 struct trace_event_file *file, struct dentry *parent)
1876{
1877 struct trace_subsystem_dir *dir;
1878 struct event_subsystem *system;
1879 struct dentry *entry;
1880
1881 /* First see if we did not already create this dir */
1882 list_for_each_entry(dir, &tr->systems, list) {
1883 system = dir->subsystem;
1884 if (strcmp(system->name, name) == 0) {
1885 dir->nr_events++;
1886 file->system = dir;
1887 return dir->entry;
1888 }
1889 }
1890
1891 /* Now see if the system itself exists. */
1892 list_for_each_entry(system, &event_subsystems, list) {
1893 if (strcmp(system->name, name) == 0)
1894 break;
1895 }
1896 /* Reset system variable when not found */
1897 if (&system->list == &event_subsystems)
1898 system = NULL;
1899
1900 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1901 if (!dir)
1902 goto out_fail;
1903
1904 if (!system) {
1905 system = create_new_subsystem(name);
1906 if (!system)
1907 goto out_free;
1908 } else
1909 __get_system(system);
1910
1911 dir->entry = tracefs_create_dir(name, parent);
1912 if (!dir->entry) {
1913 pr_warn("Failed to create system directory %s\n", name);
1914 __put_system(system);
1915 goto out_free;
1916 }
1917
1918 dir->tr = tr;
1919 dir->ref_count = 1;
1920 dir->nr_events = 1;
1921 dir->subsystem = system;
1922 file->system = dir;
1923
1924 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1925 &ftrace_subsystem_filter_fops);
1926 if (!entry) {
1927 kfree(system->filter);
1928 system->filter = NULL;
1929 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1930 }
1931
1932 trace_create_file("enable", 0644, dir->entry, dir,
1933 &ftrace_system_enable_fops);
1934
1935 list_add(&dir->list, &tr->systems);
1936
1937 return dir->entry;
1938
1939 out_free:
1940 kfree(dir);
1941 out_fail:
1942 /* Only print this message if failed on memory allocation */
1943 if (!dir || !system)
1944 pr_warn("No memory to create event subsystem %s\n", name);
1945 return NULL;
1946}
1947
1948static int
1949event_create_dir(struct dentry *parent, struct trace_event_file *file)
1950{
1951 struct trace_event_call *call = file->event_call;
1952 struct trace_array *tr = file->tr;
1953 struct list_head *head;
1954 struct dentry *d_events;
1955 const char *name;
1956 int ret;
1957
1958 /*
1959 * If the trace point header did not define TRACE_SYSTEM
1960 * then the system would be called "TRACE_SYSTEM".
1961 */
1962 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1963 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1964 if (!d_events)
1965 return -ENOMEM;
1966 } else
1967 d_events = parent;
1968
1969 name = trace_event_name(call);
1970 file->dir = tracefs_create_dir(name, d_events);
1971 if (!file->dir) {
1972 pr_warn("Could not create tracefs '%s' directory\n", name);
1973 return -1;
1974 }
1975
1976 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1977 trace_create_file("enable", 0644, file->dir, file,
1978 &ftrace_enable_fops);
1979
1980#ifdef CONFIG_PERF_EVENTS
1981 if (call->event.type && call->class->reg)
1982 trace_create_file("id", 0444, file->dir,
1983 (void *)(long)call->event.type,
1984 &ftrace_event_id_fops);
1985#endif
1986
1987 /*
1988 * Other events may have the same class. Only update
1989 * the fields if they are not already defined.
1990 */
1991 head = trace_get_fields(call);
1992 if (list_empty(head)) {
1993 ret = call->class->define_fields(call);
1994 if (ret < 0) {
1995 pr_warn("Could not initialize trace point events/%s\n",
1996 name);
1997 return -1;
1998 }
1999 }
2000
2001 /*
2002 * Only event directories that can be enabled should have
2003 * triggers or filters.
2004 */
2005 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2006 trace_create_file("filter", 0644, file->dir, file,
2007 &ftrace_event_filter_fops);
2008
2009 trace_create_file("trigger", 0644, file->dir, file,
2010 &event_trigger_fops);
2011 }
2012
2013#ifdef CONFIG_HIST_TRIGGERS
2014 trace_create_file("hist", 0444, file->dir, file,
2015 &event_hist_fops);
2016#endif
2017 trace_create_file("format", 0444, file->dir, call,
2018 &ftrace_event_format_fops);
2019
2020 return 0;
2021}
2022
2023static void remove_event_from_tracers(struct trace_event_call *call)
2024{
2025 struct trace_event_file *file;
2026 struct trace_array *tr;
2027
2028 do_for_each_event_file_safe(tr, file) {
2029 if (file->event_call != call)
2030 continue;
2031
2032 remove_event_file_dir(file);
2033 /*
2034 * The do_for_each_event_file_safe() is
2035 * a double loop. After finding the call for this
2036 * trace_array, we use break to jump to the next
2037 * trace_array.
2038 */
2039 break;
2040 } while_for_each_event_file();
2041}
2042
2043static void event_remove(struct trace_event_call *call)
2044{
2045 struct trace_array *tr;
2046 struct trace_event_file *file;
2047
2048 do_for_each_event_file(tr, file) {
2049 if (file->event_call != call)
2050 continue;
2051
2052 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2053 tr->clear_trace = true;
2054
2055 ftrace_event_enable_disable(file, 0);
2056 /*
2057 * The do_for_each_event_file() is
2058 * a double loop. After finding the call for this
2059 * trace_array, we use break to jump to the next
2060 * trace_array.
2061 */
2062 break;
2063 } while_for_each_event_file();
2064
2065 if (call->event.funcs)
2066 __unregister_trace_event(&call->event);
2067 remove_event_from_tracers(call);
2068 list_del(&call->list);
2069}
2070
2071static int event_init(struct trace_event_call *call)
2072{
2073 int ret = 0;
2074 const char *name;
2075
2076 name = trace_event_name(call);
2077 if (WARN_ON(!name))
2078 return -EINVAL;
2079
2080 if (call->class->raw_init) {
2081 ret = call->class->raw_init(call);
2082 if (ret < 0 && ret != -ENOSYS)
2083 pr_warn("Could not initialize trace events/%s\n", name);
2084 }
2085
2086 return ret;
2087}
2088
2089static int
2090__register_event(struct trace_event_call *call, struct module *mod)
2091{
2092 int ret;
2093
2094 ret = event_init(call);
2095 if (ret < 0)
2096 return ret;
2097
2098 list_add(&call->list, &ftrace_events);
2099 call->mod = mod;
2100
2101 return 0;
2102}
2103
2104static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2105{
2106 int rlen;
2107 int elen;
2108
2109 /* Find the length of the eval value as a string */
2110 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2111 /* Make sure there's enough room to replace the string with the value */
2112 if (len < elen)
2113 return NULL;
2114
2115 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2116
2117 /* Get the rest of the string of ptr */
2118 rlen = strlen(ptr + len);
2119 memmove(ptr + elen, ptr + len, rlen);
2120 /* Make sure we end the new string */
2121 ptr[elen + rlen] = 0;
2122
2123 return ptr + elen;
2124}
2125
2126static void update_event_printk(struct trace_event_call *call,
2127 struct trace_eval_map *map)
2128{
2129 char *ptr;
2130 int quote = 0;
2131 int len = strlen(map->eval_string);
2132
2133 for (ptr = call->print_fmt; *ptr; ptr++) {
2134 if (*ptr == '\\') {
2135 ptr++;
2136 /* paranoid */
2137 if (!*ptr)
2138 break;
2139 continue;
2140 }
2141 if (*ptr == '"') {
2142 quote ^= 1;
2143 continue;
2144 }
2145 if (quote)
2146 continue;
2147 if (isdigit(*ptr)) {
2148 /* skip numbers */
2149 do {
2150 ptr++;
2151 /* Check for alpha chars like ULL */
2152 } while (isalnum(*ptr));
2153 if (!*ptr)
2154 break;
2155 /*
2156 * A number must have some kind of delimiter after
2157 * it, and we can ignore that too.
2158 */
2159 continue;
2160 }
2161 if (isalpha(*ptr) || *ptr == '_') {
2162 if (strncmp(map->eval_string, ptr, len) == 0 &&
2163 !isalnum(ptr[len]) && ptr[len] != '_') {
2164 ptr = eval_replace(ptr, map, len);
2165 /* enum/sizeof string smaller than value */
2166 if (WARN_ON_ONCE(!ptr))
2167 return;
2168 /*
2169 * No need to decrement here, as eval_replace()
2170 * returns the pointer to the character passed
2171 * the eval, and two evals can not be placed
2172 * back to back without something in between.
2173 * We can skip that something in between.
2174 */
2175 continue;
2176 }
2177 skip_more:
2178 do {
2179 ptr++;
2180 } while (isalnum(*ptr) || *ptr == '_');
2181 if (!*ptr)
2182 break;
2183 /*
2184 * If what comes after this variable is a '.' or
2185 * '->' then we can continue to ignore that string.
2186 */
2187 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2188 ptr += *ptr == '.' ? 1 : 2;
2189 if (!*ptr)
2190 break;
2191 goto skip_more;
2192 }
2193 /*
2194 * Once again, we can skip the delimiter that came
2195 * after the string.
2196 */
2197 continue;
2198 }
2199 }
2200}
2201
2202void trace_event_eval_update(struct trace_eval_map **map, int len)
2203{
2204 struct trace_event_call *call, *p;
2205 const char *last_system = NULL;
2206 bool first = false;
2207 int last_i;
2208 int i;
2209
2210 down_write(&trace_event_sem);
2211 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2212 /* events are usually grouped together with systems */
2213 if (!last_system || call->class->system != last_system) {
2214 first = true;
2215 last_i = 0;
2216 last_system = call->class->system;
2217 }
2218
2219 /*
2220 * Since calls are grouped by systems, the likelyhood that the
2221 * next call in the iteration belongs to the same system as the
2222 * previous call is high. As an optimization, we skip seaching
2223 * for a map[] that matches the call's system if the last call
2224 * was from the same system. That's what last_i is for. If the
2225 * call has the same system as the previous call, then last_i
2226 * will be the index of the first map[] that has a matching
2227 * system.
2228 */
2229 for (i = last_i; i < len; i++) {
2230 if (call->class->system == map[i]->system) {
2231 /* Save the first system if need be */
2232 if (first) {
2233 last_i = i;
2234 first = false;
2235 }
2236 update_event_printk(call, map[i]);
2237 }
2238 }
2239 }
2240 up_write(&trace_event_sem);
2241}
2242
2243static struct trace_event_file *
2244trace_create_new_event(struct trace_event_call *call,
2245 struct trace_array *tr)
2246{
2247 struct trace_event_file *file;
2248
2249 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2250 if (!file)
2251 return NULL;
2252
2253 file->event_call = call;
2254 file->tr = tr;
2255 atomic_set(&file->sm_ref, 0);
2256 atomic_set(&file->tm_ref, 0);
2257 INIT_LIST_HEAD(&file->triggers);
2258 list_add(&file->list, &tr->events);
2259
2260 return file;
2261}
2262
2263/* Add an event to a trace directory */
2264static int
2265__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2266{
2267 struct trace_event_file *file;
2268
2269 file = trace_create_new_event(call, tr);
2270 if (!file)
2271 return -ENOMEM;
2272
2273 return event_create_dir(tr->event_dir, file);
2274}
2275
2276/*
2277 * Just create a decriptor for early init. A descriptor is required
2278 * for enabling events at boot. We want to enable events before
2279 * the filesystem is initialized.
2280 */
2281static __init int
2282__trace_early_add_new_event(struct trace_event_call *call,
2283 struct trace_array *tr)
2284{
2285 struct trace_event_file *file;
2286
2287 file = trace_create_new_event(call, tr);
2288 if (!file)
2289 return -ENOMEM;
2290
2291 return 0;
2292}
2293
2294struct ftrace_module_file_ops;
2295static void __add_event_to_tracers(struct trace_event_call *call);
2296
2297/* Add an additional event_call dynamically */
2298int trace_add_event_call(struct trace_event_call *call)
2299{
2300 int ret;
2301 lockdep_assert_held(&event_mutex);
2302
2303 mutex_lock(&trace_types_lock);
2304
2305 ret = __register_event(call, NULL);
2306 if (ret >= 0)
2307 __add_event_to_tracers(call);
2308
2309 mutex_unlock(&trace_types_lock);
2310 return ret;
2311}
2312
2313/*
2314 * Must be called under locking of trace_types_lock, event_mutex and
2315 * trace_event_sem.
2316 */
2317static void __trace_remove_event_call(struct trace_event_call *call)
2318{
2319 event_remove(call);
2320 trace_destroy_fields(call);
2321 free_event_filter(call->filter);
2322 call->filter = NULL;
2323}
2324
2325static int probe_remove_event_call(struct trace_event_call *call)
2326{
2327 struct trace_array *tr;
2328 struct trace_event_file *file;
2329
2330#ifdef CONFIG_PERF_EVENTS
2331 if (call->perf_refcount)
2332 return -EBUSY;
2333#endif
2334 do_for_each_event_file(tr, file) {
2335 if (file->event_call != call)
2336 continue;
2337 /*
2338 * We can't rely on ftrace_event_enable_disable(enable => 0)
2339 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2340 * TRACE_REG_UNREGISTER.
2341 */
2342 if (file->flags & EVENT_FILE_FL_ENABLED)
2343 return -EBUSY;
2344 /*
2345 * The do_for_each_event_file_safe() is
2346 * a double loop. After finding the call for this
2347 * trace_array, we use break to jump to the next
2348 * trace_array.
2349 */
2350 break;
2351 } while_for_each_event_file();
2352
2353 __trace_remove_event_call(call);
2354
2355 return 0;
2356}
2357
2358/* Remove an event_call */
2359int trace_remove_event_call(struct trace_event_call *call)
2360{
2361 int ret;
2362
2363 lockdep_assert_held(&event_mutex);
2364
2365 mutex_lock(&trace_types_lock);
2366 down_write(&trace_event_sem);
2367 ret = probe_remove_event_call(call);
2368 up_write(&trace_event_sem);
2369 mutex_unlock(&trace_types_lock);
2370
2371 return ret;
2372}
2373
2374#define for_each_event(event, start, end) \
2375 for (event = start; \
2376 (unsigned long)event < (unsigned long)end; \
2377 event++)
2378
2379#ifdef CONFIG_MODULES
2380
2381static void trace_module_add_events(struct module *mod)
2382{
2383 struct trace_event_call **call, **start, **end;
2384
2385 if (!mod->num_trace_events)
2386 return;
2387
2388 /* Don't add infrastructure for mods without tracepoints */
2389 if (trace_module_has_bad_taint(mod)) {
2390 pr_err("%s: module has bad taint, not creating trace events\n",
2391 mod->name);
2392 return;
2393 }
2394
2395 start = mod->trace_events;
2396 end = mod->trace_events + mod->num_trace_events;
2397
2398 for_each_event(call, start, end) {
2399 __register_event(*call, mod);
2400 __add_event_to_tracers(*call);
2401 }
2402}
2403
2404static void trace_module_remove_events(struct module *mod)
2405{
2406 struct trace_event_call *call, *p;
2407
2408 down_write(&trace_event_sem);
2409 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2410 if (call->mod == mod)
2411 __trace_remove_event_call(call);
2412 }
2413 up_write(&trace_event_sem);
2414
2415 /*
2416 * It is safest to reset the ring buffer if the module being unloaded
2417 * registered any events that were used. The only worry is if
2418 * a new module gets loaded, and takes on the same id as the events
2419 * of this module. When printing out the buffer, traced events left
2420 * over from this module may be passed to the new module events and
2421 * unexpected results may occur.
2422 */
2423 tracing_reset_all_online_cpus();
2424}
2425
2426static int trace_module_notify(struct notifier_block *self,
2427 unsigned long val, void *data)
2428{
2429 struct module *mod = data;
2430
2431 mutex_lock(&event_mutex);
2432 mutex_lock(&trace_types_lock);
2433 switch (val) {
2434 case MODULE_STATE_COMING:
2435 trace_module_add_events(mod);
2436 break;
2437 case MODULE_STATE_GOING:
2438 trace_module_remove_events(mod);
2439 break;
2440 }
2441 mutex_unlock(&trace_types_lock);
2442 mutex_unlock(&event_mutex);
2443
2444 return 0;
2445}
2446
2447static struct notifier_block trace_module_nb = {
2448 .notifier_call = trace_module_notify,
2449 .priority = 1, /* higher than trace.c module notify */
2450};
2451#endif /* CONFIG_MODULES */
2452
2453/* Create a new event directory structure for a trace directory. */
2454static void
2455__trace_add_event_dirs(struct trace_array *tr)
2456{
2457 struct trace_event_call *call;
2458 int ret;
2459
2460 list_for_each_entry(call, &ftrace_events, list) {
2461 ret = __trace_add_new_event(call, tr);
2462 if (ret < 0)
2463 pr_warn("Could not create directory for event %s\n",
2464 trace_event_name(call));
2465 }
2466}
2467
2468/* Returns any file that matches the system and event */
2469struct trace_event_file *
2470__find_event_file(struct trace_array *tr, const char *system, const char *event)
2471{
2472 struct trace_event_file *file;
2473 struct trace_event_call *call;
2474 const char *name;
2475
2476 list_for_each_entry(file, &tr->events, list) {
2477
2478 call = file->event_call;
2479 name = trace_event_name(call);
2480
2481 if (!name || !call->class)
2482 continue;
2483
2484 if (strcmp(event, name) == 0 &&
2485 strcmp(system, call->class->system) == 0)
2486 return file;
2487 }
2488 return NULL;
2489}
2490
2491/* Returns valid trace event files that match system and event */
2492struct trace_event_file *
2493find_event_file(struct trace_array *tr, const char *system, const char *event)
2494{
2495 struct trace_event_file *file;
2496
2497 file = __find_event_file(tr, system, event);
2498 if (!file || !file->event_call->class->reg ||
2499 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2500 return NULL;
2501
2502 return file;
2503}
2504
2505#ifdef CONFIG_DYNAMIC_FTRACE
2506
2507/* Avoid typos */
2508#define ENABLE_EVENT_STR "enable_event"
2509#define DISABLE_EVENT_STR "disable_event"
2510
2511struct event_probe_data {
2512 struct trace_event_file *file;
2513 unsigned long count;
2514 int ref;
2515 bool enable;
2516};
2517
2518static void update_event_probe(struct event_probe_data *data)
2519{
2520 if (data->enable)
2521 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2522 else
2523 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2524}
2525
2526static void
2527event_enable_probe(unsigned long ip, unsigned long parent_ip,
2528 struct trace_array *tr, struct ftrace_probe_ops *ops,
2529 void *data)
2530{
2531 struct ftrace_func_mapper *mapper = data;
2532 struct event_probe_data *edata;
2533 void **pdata;
2534
2535 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2536 if (!pdata || !*pdata)
2537 return;
2538
2539 edata = *pdata;
2540 update_event_probe(edata);
2541}
2542
2543static void
2544event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2545 struct trace_array *tr, struct ftrace_probe_ops *ops,
2546 void *data)
2547{
2548 struct ftrace_func_mapper *mapper = data;
2549 struct event_probe_data *edata;
2550 void **pdata;
2551
2552 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2553 if (!pdata || !*pdata)
2554 return;
2555
2556 edata = *pdata;
2557
2558 if (!edata->count)
2559 return;
2560
2561 /* Skip if the event is in a state we want to switch to */
2562 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2563 return;
2564
2565 if (edata->count != -1)
2566 (edata->count)--;
2567
2568 update_event_probe(edata);
2569}
2570
2571static int
2572event_enable_print(struct seq_file *m, unsigned long ip,
2573 struct ftrace_probe_ops *ops, void *data)
2574{
2575 struct ftrace_func_mapper *mapper = data;
2576 struct event_probe_data *edata;
2577 void **pdata;
2578
2579 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2580
2581 if (WARN_ON_ONCE(!pdata || !*pdata))
2582 return 0;
2583
2584 edata = *pdata;
2585
2586 seq_printf(m, "%ps:", (void *)ip);
2587
2588 seq_printf(m, "%s:%s:%s",
2589 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2590 edata->file->event_call->class->system,
2591 trace_event_name(edata->file->event_call));
2592
2593 if (edata->count == -1)
2594 seq_puts(m, ":unlimited\n");
2595 else
2596 seq_printf(m, ":count=%ld\n", edata->count);
2597
2598 return 0;
2599}
2600
2601static int
2602event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2603 unsigned long ip, void *init_data, void **data)
2604{
2605 struct ftrace_func_mapper *mapper = *data;
2606 struct event_probe_data *edata = init_data;
2607 int ret;
2608
2609 if (!mapper) {
2610 mapper = allocate_ftrace_func_mapper();
2611 if (!mapper)
2612 return -ENODEV;
2613 *data = mapper;
2614 }
2615
2616 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2617 if (ret < 0)
2618 return ret;
2619
2620 edata->ref++;
2621
2622 return 0;
2623}
2624
2625static int free_probe_data(void *data)
2626{
2627 struct event_probe_data *edata = data;
2628
2629 edata->ref--;
2630 if (!edata->ref) {
2631 /* Remove the SOFT_MODE flag */
2632 __ftrace_event_enable_disable(edata->file, 0, 1);
2633 module_put(edata->file->event_call->mod);
2634 kfree(edata);
2635 }
2636 return 0;
2637}
2638
2639static void
2640event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2641 unsigned long ip, void *data)
2642{
2643 struct ftrace_func_mapper *mapper = data;
2644 struct event_probe_data *edata;
2645
2646 if (!ip) {
2647 if (!mapper)
2648 return;
2649 free_ftrace_func_mapper(mapper, free_probe_data);
2650 return;
2651 }
2652
2653 edata = ftrace_func_mapper_remove_ip(mapper, ip);
2654
2655 if (WARN_ON_ONCE(!edata))
2656 return;
2657
2658 if (WARN_ON_ONCE(edata->ref <= 0))
2659 return;
2660
2661 free_probe_data(edata);
2662}
2663
2664static struct ftrace_probe_ops event_enable_probe_ops = {
2665 .func = event_enable_probe,
2666 .print = event_enable_print,
2667 .init = event_enable_init,
2668 .free = event_enable_free,
2669};
2670
2671static struct ftrace_probe_ops event_enable_count_probe_ops = {
2672 .func = event_enable_count_probe,
2673 .print = event_enable_print,
2674 .init = event_enable_init,
2675 .free = event_enable_free,
2676};
2677
2678static struct ftrace_probe_ops event_disable_probe_ops = {
2679 .func = event_enable_probe,
2680 .print = event_enable_print,
2681 .init = event_enable_init,
2682 .free = event_enable_free,
2683};
2684
2685static struct ftrace_probe_ops event_disable_count_probe_ops = {
2686 .func = event_enable_count_probe,
2687 .print = event_enable_print,
2688 .init = event_enable_init,
2689 .free = event_enable_free,
2690};
2691
2692static int
2693event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2694 char *glob, char *cmd, char *param, int enabled)
2695{
2696 struct trace_event_file *file;
2697 struct ftrace_probe_ops *ops;
2698 struct event_probe_data *data;
2699 const char *system;
2700 const char *event;
2701 char *number;
2702 bool enable;
2703 int ret;
2704
2705 if (!tr)
2706 return -ENODEV;
2707
2708 /* hash funcs only work with set_ftrace_filter */
2709 if (!enabled || !param)
2710 return -EINVAL;
2711
2712 system = strsep(¶m, ":");
2713 if (!param)
2714 return -EINVAL;
2715
2716 event = strsep(¶m, ":");
2717
2718 mutex_lock(&event_mutex);
2719
2720 ret = -EINVAL;
2721 file = find_event_file(tr, system, event);
2722 if (!file)
2723 goto out;
2724
2725 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2726
2727 if (enable)
2728 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2729 else
2730 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2731
2732 if (glob[0] == '!') {
2733 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2734 goto out;
2735 }
2736
2737 ret = -ENOMEM;
2738
2739 data = kzalloc(sizeof(*data), GFP_KERNEL);
2740 if (!data)
2741 goto out;
2742
2743 data->enable = enable;
2744 data->count = -1;
2745 data->file = file;
2746
2747 if (!param)
2748 goto out_reg;
2749
2750 number = strsep(¶m, ":");
2751
2752 ret = -EINVAL;
2753 if (!strlen(number))
2754 goto out_free;
2755
2756 /*
2757 * We use the callback data field (which is a pointer)
2758 * as our counter.
2759 */
2760 ret = kstrtoul(number, 0, &data->count);
2761 if (ret)
2762 goto out_free;
2763
2764 out_reg:
2765 /* Don't let event modules unload while probe registered */
2766 ret = try_module_get(file->event_call->mod);
2767 if (!ret) {
2768 ret = -EBUSY;
2769 goto out_free;
2770 }
2771
2772 ret = __ftrace_event_enable_disable(file, 1, 1);
2773 if (ret < 0)
2774 goto out_put;
2775
2776 ret = register_ftrace_function_probe(glob, tr, ops, data);
2777 /*
2778 * The above returns on success the # of functions enabled,
2779 * but if it didn't find any functions it returns zero.
2780 * Consider no functions a failure too.
2781 */
2782 if (!ret) {
2783 ret = -ENOENT;
2784 goto out_disable;
2785 } else if (ret < 0)
2786 goto out_disable;
2787 /* Just return zero, not the number of enabled functions */
2788 ret = 0;
2789 out:
2790 mutex_unlock(&event_mutex);
2791 return ret;
2792
2793 out_disable:
2794 __ftrace_event_enable_disable(file, 0, 1);
2795 out_put:
2796 module_put(file->event_call->mod);
2797 out_free:
2798 kfree(data);
2799 goto out;
2800}
2801
2802static struct ftrace_func_command event_enable_cmd = {
2803 .name = ENABLE_EVENT_STR,
2804 .func = event_enable_func,
2805};
2806
2807static struct ftrace_func_command event_disable_cmd = {
2808 .name = DISABLE_EVENT_STR,
2809 .func = event_enable_func,
2810};
2811
2812static __init int register_event_cmds(void)
2813{
2814 int ret;
2815
2816 ret = register_ftrace_command(&event_enable_cmd);
2817 if (WARN_ON(ret < 0))
2818 return ret;
2819 ret = register_ftrace_command(&event_disable_cmd);
2820 if (WARN_ON(ret < 0))
2821 unregister_ftrace_command(&event_enable_cmd);
2822 return ret;
2823}
2824#else
2825static inline int register_event_cmds(void) { return 0; }
2826#endif /* CONFIG_DYNAMIC_FTRACE */
2827
2828/*
2829 * The top level array has already had its trace_event_file
2830 * descriptors created in order to allow for early events to
2831 * be recorded. This function is called after the tracefs has been
2832 * initialized, and we now have to create the files associated
2833 * to the events.
2834 */
2835static __init void
2836__trace_early_add_event_dirs(struct trace_array *tr)
2837{
2838 struct trace_event_file *file;
2839 int ret;
2840
2841
2842 list_for_each_entry(file, &tr->events, list) {
2843 ret = event_create_dir(tr->event_dir, file);
2844 if (ret < 0)
2845 pr_warn("Could not create directory for event %s\n",
2846 trace_event_name(file->event_call));
2847 }
2848}
2849
2850/*
2851 * For early boot up, the top trace array requires to have
2852 * a list of events that can be enabled. This must be done before
2853 * the filesystem is set up in order to allow events to be traced
2854 * early.
2855 */
2856static __init void
2857__trace_early_add_events(struct trace_array *tr)
2858{
2859 struct trace_event_call *call;
2860 int ret;
2861
2862 list_for_each_entry(call, &ftrace_events, list) {
2863 /* Early boot up should not have any modules loaded */
2864 if (WARN_ON_ONCE(call->mod))
2865 continue;
2866
2867 ret = __trace_early_add_new_event(call, tr);
2868 if (ret < 0)
2869 pr_warn("Could not create early event %s\n",
2870 trace_event_name(call));
2871 }
2872}
2873
2874/* Remove the event directory structure for a trace directory. */
2875static void
2876__trace_remove_event_dirs(struct trace_array *tr)
2877{
2878 struct trace_event_file *file, *next;
2879
2880 list_for_each_entry_safe(file, next, &tr->events, list)
2881 remove_event_file_dir(file);
2882}
2883
2884static void __add_event_to_tracers(struct trace_event_call *call)
2885{
2886 struct trace_array *tr;
2887
2888 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2889 __trace_add_new_event(call, tr);
2890}
2891
2892extern struct trace_event_call *__start_ftrace_events[];
2893extern struct trace_event_call *__stop_ftrace_events[];
2894
2895static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2896
2897static __init int setup_trace_event(char *str)
2898{
2899 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2900 ring_buffer_expanded = true;
2901 tracing_selftest_disabled = true;
2902
2903 return 1;
2904}
2905__setup("trace_event=", setup_trace_event);
2906
2907/* Expects to have event_mutex held when called */
2908static int
2909create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2910{
2911 struct dentry *d_events;
2912 struct dentry *entry;
2913
2914 entry = tracefs_create_file("set_event", 0644, parent,
2915 tr, &ftrace_set_event_fops);
2916 if (!entry) {
2917 pr_warn("Could not create tracefs 'set_event' entry\n");
2918 return -ENOMEM;
2919 }
2920
2921 d_events = tracefs_create_dir("events", parent);
2922 if (!d_events) {
2923 pr_warn("Could not create tracefs 'events' directory\n");
2924 return -ENOMEM;
2925 }
2926
2927 entry = trace_create_file("enable", 0644, d_events,
2928 tr, &ftrace_tr_enable_fops);
2929 if (!entry) {
2930 pr_warn("Could not create tracefs 'enable' entry\n");
2931 return -ENOMEM;
2932 }
2933
2934 /* There are not as crucial, just warn if they are not created */
2935
2936 entry = tracefs_create_file("set_event_pid", 0644, parent,
2937 tr, &ftrace_set_event_pid_fops);
2938 if (!entry)
2939 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2940
2941 /* ring buffer internal formats */
2942 entry = trace_create_file("header_page", 0444, d_events,
2943 ring_buffer_print_page_header,
2944 &ftrace_show_header_fops);
2945 if (!entry)
2946 pr_warn("Could not create tracefs 'header_page' entry\n");
2947
2948 entry = trace_create_file("header_event", 0444, d_events,
2949 ring_buffer_print_entry_header,
2950 &ftrace_show_header_fops);
2951 if (!entry)
2952 pr_warn("Could not create tracefs 'header_event' entry\n");
2953
2954 tr->event_dir = d_events;
2955
2956 return 0;
2957}
2958
2959/**
2960 * event_trace_add_tracer - add a instance of a trace_array to events
2961 * @parent: The parent dentry to place the files/directories for events in
2962 * @tr: The trace array associated with these events
2963 *
2964 * When a new instance is created, it needs to set up its events
2965 * directory, as well as other files associated with events. It also
2966 * creates the event hierachry in the @parent/events directory.
2967 *
2968 * Returns 0 on success.
2969 *
2970 * Must be called with event_mutex held.
2971 */
2972int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2973{
2974 int ret;
2975
2976 lockdep_assert_held(&event_mutex);
2977
2978 ret = create_event_toplevel_files(parent, tr);
2979 if (ret)
2980 goto out;
2981
2982 down_write(&trace_event_sem);
2983 __trace_add_event_dirs(tr);
2984 up_write(&trace_event_sem);
2985
2986 out:
2987 return ret;
2988}
2989
2990/*
2991 * The top trace array already had its file descriptors created.
2992 * Now the files themselves need to be created.
2993 */
2994static __init int
2995early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2996{
2997 int ret;
2998
2999 mutex_lock(&event_mutex);
3000
3001 ret = create_event_toplevel_files(parent, tr);
3002 if (ret)
3003 goto out_unlock;
3004
3005 down_write(&trace_event_sem);
3006 __trace_early_add_event_dirs(tr);
3007 up_write(&trace_event_sem);
3008
3009 out_unlock:
3010 mutex_unlock(&event_mutex);
3011
3012 return ret;
3013}
3014
3015/* Must be called with event_mutex held */
3016int event_trace_del_tracer(struct trace_array *tr)
3017{
3018 lockdep_assert_held(&event_mutex);
3019
3020 /* Disable any event triggers and associated soft-disabled events */
3021 clear_event_triggers(tr);
3022
3023 /* Clear the pid list */
3024 __ftrace_clear_event_pids(tr);
3025
3026 /* Disable any running events */
3027 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3028
3029 /* Make sure no more events are being executed */
3030 tracepoint_synchronize_unregister();
3031
3032 down_write(&trace_event_sem);
3033 __trace_remove_event_dirs(tr);
3034 tracefs_remove_recursive(tr->event_dir);
3035 up_write(&trace_event_sem);
3036
3037 tr->event_dir = NULL;
3038
3039 return 0;
3040}
3041
3042static __init int event_trace_memsetup(void)
3043{
3044 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3045 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3046 return 0;
3047}
3048
3049static __init void
3050early_enable_events(struct trace_array *tr, bool disable_first)
3051{
3052 char *buf = bootup_event_buf;
3053 char *token;
3054 int ret;
3055
3056 while (true) {
3057 token = strsep(&buf, ",");
3058
3059 if (!token)
3060 break;
3061
3062 if (*token) {
3063 /* Restarting syscalls requires that we stop them first */
3064 if (disable_first)
3065 ftrace_set_clr_event(tr, token, 0);
3066
3067 ret = ftrace_set_clr_event(tr, token, 1);
3068 if (ret)
3069 pr_warn("Failed to enable trace event: %s\n", token);
3070 }
3071
3072 /* Put back the comma to allow this to be called again */
3073 if (buf)
3074 *(buf - 1) = ',';
3075 }
3076}
3077
3078static __init int event_trace_enable(void)
3079{
3080 struct trace_array *tr = top_trace_array();
3081 struct trace_event_call **iter, *call;
3082 int ret;
3083
3084 if (!tr)
3085 return -ENODEV;
3086
3087 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3088
3089 call = *iter;
3090 ret = event_init(call);
3091 if (!ret)
3092 list_add(&call->list, &ftrace_events);
3093 }
3094
3095 /*
3096 * We need the top trace array to have a working set of trace
3097 * points at early init, before the debug files and directories
3098 * are created. Create the file entries now, and attach them
3099 * to the actual file dentries later.
3100 */
3101 __trace_early_add_events(tr);
3102
3103 early_enable_events(tr, false);
3104
3105 trace_printk_start_comm();
3106
3107 register_event_cmds();
3108
3109 register_trigger_cmds();
3110
3111 return 0;
3112}
3113
3114/*
3115 * event_trace_enable() is called from trace_event_init() first to
3116 * initialize events and perhaps start any events that are on the
3117 * command line. Unfortunately, there are some events that will not
3118 * start this early, like the system call tracepoints that need
3119 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3120 * is called before pid 1 starts, and this flag is never set, making
3121 * the syscall tracepoint never get reached, but the event is enabled
3122 * regardless (and not doing anything).
3123 */
3124static __init int event_trace_enable_again(void)
3125{
3126 struct trace_array *tr;
3127
3128 tr = top_trace_array();
3129 if (!tr)
3130 return -ENODEV;
3131
3132 early_enable_events(tr, true);
3133
3134 return 0;
3135}
3136
3137early_initcall(event_trace_enable_again);
3138
3139__init int event_trace_init(void)
3140{
3141 struct trace_array *tr;
3142 struct dentry *d_tracer;
3143 struct dentry *entry;
3144 int ret;
3145
3146 tr = top_trace_array();
3147 if (!tr)
3148 return -ENODEV;
3149
3150 d_tracer = tracing_init_dentry();
3151 if (IS_ERR(d_tracer))
3152 return 0;
3153
3154 entry = tracefs_create_file("available_events", 0444, d_tracer,
3155 tr, &ftrace_avail_fops);
3156 if (!entry)
3157 pr_warn("Could not create tracefs 'available_events' entry\n");
3158
3159 if (trace_define_generic_fields())
3160 pr_warn("tracing: Failed to allocated generic fields");
3161
3162 if (trace_define_common_fields())
3163 pr_warn("tracing: Failed to allocate common fields");
3164
3165 ret = early_event_add_tracer(d_tracer, tr);
3166 if (ret)
3167 return ret;
3168
3169#ifdef CONFIG_MODULES
3170 ret = register_module_notifier(&trace_module_nb);
3171 if (ret)
3172 pr_warn("Failed to register trace events module notifier\n");
3173#endif
3174 return 0;
3175}
3176
3177void __init trace_event_init(void)
3178{
3179 event_trace_memsetup();
3180 init_ftrace_syscalls();
3181 event_trace_enable();
3182}
3183
3184#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3185
3186static DEFINE_SPINLOCK(test_spinlock);
3187static DEFINE_SPINLOCK(test_spinlock_irq);
3188static DEFINE_MUTEX(test_mutex);
3189
3190static __init void test_work(struct work_struct *dummy)
3191{
3192 spin_lock(&test_spinlock);
3193 spin_lock_irq(&test_spinlock_irq);
3194 udelay(1);
3195 spin_unlock_irq(&test_spinlock_irq);
3196 spin_unlock(&test_spinlock);
3197
3198 mutex_lock(&test_mutex);
3199 msleep(1);
3200 mutex_unlock(&test_mutex);
3201}
3202
3203static __init int event_test_thread(void *unused)
3204{
3205 void *test_malloc;
3206
3207 test_malloc = kmalloc(1234, GFP_KERNEL);
3208 if (!test_malloc)
3209 pr_info("failed to kmalloc\n");
3210
3211 schedule_on_each_cpu(test_work);
3212
3213 kfree(test_malloc);
3214
3215 set_current_state(TASK_INTERRUPTIBLE);
3216 while (!kthread_should_stop()) {
3217 schedule();
3218 set_current_state(TASK_INTERRUPTIBLE);
3219 }
3220 __set_current_state(TASK_RUNNING);
3221
3222 return 0;
3223}
3224
3225/*
3226 * Do various things that may trigger events.
3227 */
3228static __init void event_test_stuff(void)
3229{
3230 struct task_struct *test_thread;
3231
3232 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3233 msleep(1);
3234 kthread_stop(test_thread);
3235}
3236
3237/*
3238 * For every trace event defined, we will test each trace point separately,
3239 * and then by groups, and finally all trace points.
3240 */
3241static __init void event_trace_self_tests(void)
3242{
3243 struct trace_subsystem_dir *dir;
3244 struct trace_event_file *file;
3245 struct trace_event_call *call;
3246 struct event_subsystem *system;
3247 struct trace_array *tr;
3248 int ret;
3249
3250 tr = top_trace_array();
3251 if (!tr)
3252 return;
3253
3254 pr_info("Running tests on trace events:\n");
3255
3256 list_for_each_entry(file, &tr->events, list) {
3257
3258 call = file->event_call;
3259
3260 /* Only test those that have a probe */
3261 if (!call->class || !call->class->probe)
3262 continue;
3263
3264/*
3265 * Testing syscall events here is pretty useless, but
3266 * we still do it if configured. But this is time consuming.
3267 * What we really need is a user thread to perform the
3268 * syscalls as we test.
3269 */
3270#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3271 if (call->class->system &&
3272 strcmp(call->class->system, "syscalls") == 0)
3273 continue;
3274#endif
3275
3276 pr_info("Testing event %s: ", trace_event_name(call));
3277
3278 /*
3279 * If an event is already enabled, someone is using
3280 * it and the self test should not be on.
3281 */
3282 if (file->flags & EVENT_FILE_FL_ENABLED) {
3283 pr_warn("Enabled event during self test!\n");
3284 WARN_ON_ONCE(1);
3285 continue;
3286 }
3287
3288 ftrace_event_enable_disable(file, 1);
3289 event_test_stuff();
3290 ftrace_event_enable_disable(file, 0);
3291
3292 pr_cont("OK\n");
3293 }
3294
3295 /* Now test at the sub system level */
3296
3297 pr_info("Running tests on trace event systems:\n");
3298
3299 list_for_each_entry(dir, &tr->systems, list) {
3300
3301 system = dir->subsystem;
3302
3303 /* the ftrace system is special, skip it */
3304 if (strcmp(system->name, "ftrace") == 0)
3305 continue;
3306
3307 pr_info("Testing event system %s: ", system->name);
3308
3309 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3310 if (WARN_ON_ONCE(ret)) {
3311 pr_warn("error enabling system %s\n",
3312 system->name);
3313 continue;
3314 }
3315
3316 event_test_stuff();
3317
3318 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3319 if (WARN_ON_ONCE(ret)) {
3320 pr_warn("error disabling system %s\n",
3321 system->name);
3322 continue;
3323 }
3324
3325 pr_cont("OK\n");
3326 }
3327
3328 /* Test with all events enabled */
3329
3330 pr_info("Running tests on all trace events:\n");
3331 pr_info("Testing all events: ");
3332
3333 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3334 if (WARN_ON_ONCE(ret)) {
3335 pr_warn("error enabling all events\n");
3336 return;
3337 }
3338
3339 event_test_stuff();
3340
3341 /* reset sysname */
3342 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3343 if (WARN_ON_ONCE(ret)) {
3344 pr_warn("error disabling all events\n");
3345 return;
3346 }
3347
3348 pr_cont("OK\n");
3349}
3350
3351#ifdef CONFIG_FUNCTION_TRACER
3352
3353static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3354
3355static struct trace_event_file event_trace_file __initdata;
3356
3357static void __init
3358function_test_events_call(unsigned long ip, unsigned long parent_ip,
3359 struct ftrace_ops *op, struct pt_regs *pt_regs)
3360{
3361 struct ring_buffer_event *event;
3362 struct ring_buffer *buffer;
3363 struct ftrace_entry *entry;
3364 unsigned long flags;
3365 long disabled;
3366 int cpu;
3367 int pc;
3368
3369 pc = preempt_count();
3370 preempt_disable_notrace();
3371 cpu = raw_smp_processor_id();
3372 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3373
3374 if (disabled != 1)
3375 goto out;
3376
3377 local_save_flags(flags);
3378
3379 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3380 TRACE_FN, sizeof(*entry),
3381 flags, pc);
3382 if (!event)
3383 goto out;
3384 entry = ring_buffer_event_data(event);
3385 entry->ip = ip;
3386 entry->parent_ip = parent_ip;
3387
3388 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3389 entry, flags, pc);
3390 out:
3391 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3392 preempt_enable_notrace();
3393}
3394
3395static struct ftrace_ops trace_ops __initdata =
3396{
3397 .func = function_test_events_call,
3398 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3399};
3400
3401static __init void event_trace_self_test_with_function(void)
3402{
3403 int ret;
3404
3405 event_trace_file.tr = top_trace_array();
3406 if (WARN_ON(!event_trace_file.tr))
3407 return;
3408
3409 ret = register_ftrace_function(&trace_ops);
3410 if (WARN_ON(ret < 0)) {
3411 pr_info("Failed to enable function tracer for event tests\n");
3412 return;
3413 }
3414 pr_info("Running tests again, along with the function tracer\n");
3415 event_trace_self_tests();
3416 unregister_ftrace_function(&trace_ops);
3417}
3418#else
3419static __init void event_trace_self_test_with_function(void)
3420{
3421}
3422#endif
3423
3424static __init int event_trace_self_tests_init(void)
3425{
3426 if (!tracing_selftest_disabled) {
3427 event_trace_self_tests();
3428 event_trace_self_test_with_function();
3429 }
3430
3431 return 0;
3432}
3433
3434late_initcall(event_trace_self_tests_init);
3435
3436#endif