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#include <trace/syscall.h>
28
29#include <asm/setup.h>
30
31#include "trace_output.h"
32
33#undef TRACE_SYSTEM
34#define TRACE_SYSTEM "TRACE_SYSTEM"
35
36DEFINE_MUTEX(event_mutex);
37
38LIST_HEAD(ftrace_events);
39static LIST_HEAD(ftrace_generic_fields);
40static LIST_HEAD(ftrace_common_fields);
41
42#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
43
44static struct kmem_cache *field_cachep;
45static struct kmem_cache *file_cachep;
46
47static inline int system_refcount(struct event_subsystem *system)
48{
49 return system->ref_count;
50}
51
52static int system_refcount_inc(struct event_subsystem *system)
53{
54 return system->ref_count++;
55}
56
57static int system_refcount_dec(struct event_subsystem *system)
58{
59 return --system->ref_count;
60}
61
62/* Double loops, do not use break, only goto's work */
63#define do_for_each_event_file(tr, file) \
64 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
65 list_for_each_entry(file, &tr->events, list)
66
67#define do_for_each_event_file_safe(tr, file) \
68 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
69 struct trace_event_file *___n; \
70 list_for_each_entry_safe(file, ___n, &tr->events, list)
71
72#define while_for_each_event_file() \
73 }
74
75static struct ftrace_event_field *
76__find_event_field(struct list_head *head, char *name)
77{
78 struct ftrace_event_field *field;
79
80 list_for_each_entry(field, head, link) {
81 if (!strcmp(field->name, name))
82 return field;
83 }
84
85 return NULL;
86}
87
88struct ftrace_event_field *
89trace_find_event_field(struct trace_event_call *call, char *name)
90{
91 struct ftrace_event_field *field;
92 struct list_head *head;
93
94 head = trace_get_fields(call);
95 field = __find_event_field(head, name);
96 if (field)
97 return field;
98
99 field = __find_event_field(&ftrace_generic_fields, name);
100 if (field)
101 return field;
102
103 return __find_event_field(&ftrace_common_fields, name);
104}
105
106static int __trace_define_field(struct list_head *head, const char *type,
107 const char *name, int offset, int size,
108 int is_signed, int filter_type)
109{
110 struct ftrace_event_field *field;
111
112 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
113 if (!field)
114 return -ENOMEM;
115
116 field->name = name;
117 field->type = type;
118
119 if (filter_type == FILTER_OTHER)
120 field->filter_type = filter_assign_type(type);
121 else
122 field->filter_type = filter_type;
123
124 field->offset = offset;
125 field->size = size;
126 field->is_signed = is_signed;
127
128 list_add(&field->link, head);
129
130 return 0;
131}
132
133int trace_define_field(struct trace_event_call *call, const char *type,
134 const char *name, int offset, int size, int is_signed,
135 int filter_type)
136{
137 struct list_head *head;
138
139 if (WARN_ON(!call->class))
140 return 0;
141
142 head = trace_get_fields(call);
143 return __trace_define_field(head, type, name, offset, size,
144 is_signed, filter_type);
145}
146EXPORT_SYMBOL_GPL(trace_define_field);
147
148#define __generic_field(type, item, filter_type) \
149 ret = __trace_define_field(&ftrace_generic_fields, #type, \
150 #item, 0, 0, is_signed_type(type), \
151 filter_type); \
152 if (ret) \
153 return ret;
154
155#define __common_field(type, item) \
156 ret = __trace_define_field(&ftrace_common_fields, #type, \
157 "common_" #item, \
158 offsetof(typeof(ent), item), \
159 sizeof(ent.item), \
160 is_signed_type(type), FILTER_OTHER); \
161 if (ret) \
162 return ret;
163
164static int trace_define_generic_fields(void)
165{
166 int ret;
167
168 __generic_field(int, CPU, FILTER_CPU);
169 __generic_field(int, cpu, FILTER_CPU);
170 __generic_field(char *, COMM, FILTER_COMM);
171 __generic_field(char *, comm, FILTER_COMM);
172
173 return ret;
174}
175
176static int trace_define_common_fields(void)
177{
178 int ret;
179 struct trace_entry ent;
180
181 __common_field(unsigned short, type);
182 __common_field(unsigned char, flags);
183 __common_field(unsigned char, preempt_count);
184 __common_field(int, pid);
185
186 return ret;
187}
188
189static void trace_destroy_fields(struct trace_event_call *call)
190{
191 struct ftrace_event_field *field, *next;
192 struct list_head *head;
193
194 head = trace_get_fields(call);
195 list_for_each_entry_safe(field, next, head, link) {
196 list_del(&field->link);
197 kmem_cache_free(field_cachep, field);
198 }
199}
200
201/*
202 * run-time version of trace_event_get_offsets_<call>() that returns the last
203 * accessible offset of trace fields excluding __dynamic_array bytes
204 */
205int trace_event_get_offsets(struct trace_event_call *call)
206{
207 struct ftrace_event_field *tail;
208 struct list_head *head;
209
210 head = trace_get_fields(call);
211 /*
212 * head->next points to the last field with the largest offset,
213 * since it was added last by trace_define_field()
214 */
215 tail = list_first_entry(head, struct ftrace_event_field, link);
216 return tail->offset + tail->size;
217}
218
219int trace_event_raw_init(struct trace_event_call *call)
220{
221 int id;
222
223 id = register_trace_event(&call->event);
224 if (!id)
225 return -ENODEV;
226
227 return 0;
228}
229EXPORT_SYMBOL_GPL(trace_event_raw_init);
230
231bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
232{
233 struct trace_array *tr = trace_file->tr;
234 struct trace_array_cpu *data;
235 struct trace_pid_list *no_pid_list;
236 struct trace_pid_list *pid_list;
237
238 pid_list = rcu_dereference_raw(tr->filtered_pids);
239 no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
240
241 if (!pid_list && !no_pid_list)
242 return false;
243
244 data = this_cpu_ptr(tr->array_buffer.data);
245
246 return data->ignore_pid;
247}
248EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
249
250void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
251 struct trace_event_file *trace_file,
252 unsigned long len)
253{
254 struct trace_event_call *event_call = trace_file->event_call;
255
256 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
257 trace_event_ignore_this_pid(trace_file))
258 return NULL;
259
260 local_save_flags(fbuffer->flags);
261 fbuffer->pc = preempt_count();
262 /*
263 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
264 * preemption (adding one to the preempt_count). Since we are
265 * interested in the preempt_count at the time the tracepoint was
266 * hit, we need to subtract one to offset the increment.
267 */
268 if (IS_ENABLED(CONFIG_PREEMPTION))
269 fbuffer->pc--;
270 fbuffer->trace_file = trace_file;
271
272 fbuffer->event =
273 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
274 event_call->event.type, len,
275 fbuffer->flags, fbuffer->pc);
276 if (!fbuffer->event)
277 return NULL;
278
279 fbuffer->regs = NULL;
280 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
281 return fbuffer->entry;
282}
283EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
284
285int trace_event_reg(struct trace_event_call *call,
286 enum trace_reg type, void *data)
287{
288 struct trace_event_file *file = data;
289
290 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
291 switch (type) {
292 case TRACE_REG_REGISTER:
293 return tracepoint_probe_register(call->tp,
294 call->class->probe,
295 file);
296 case TRACE_REG_UNREGISTER:
297 tracepoint_probe_unregister(call->tp,
298 call->class->probe,
299 file);
300 return 0;
301
302#ifdef CONFIG_PERF_EVENTS
303 case TRACE_REG_PERF_REGISTER:
304 return tracepoint_probe_register(call->tp,
305 call->class->perf_probe,
306 call);
307 case TRACE_REG_PERF_UNREGISTER:
308 tracepoint_probe_unregister(call->tp,
309 call->class->perf_probe,
310 call);
311 return 0;
312 case TRACE_REG_PERF_OPEN:
313 case TRACE_REG_PERF_CLOSE:
314 case TRACE_REG_PERF_ADD:
315 case TRACE_REG_PERF_DEL:
316 return 0;
317#endif
318 }
319 return 0;
320}
321EXPORT_SYMBOL_GPL(trace_event_reg);
322
323void trace_event_enable_cmd_record(bool enable)
324{
325 struct trace_event_file *file;
326 struct trace_array *tr;
327
328 lockdep_assert_held(&event_mutex);
329
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}
344
345void trace_event_enable_tgid_record(bool enable)
346{
347 struct trace_event_file *file;
348 struct trace_array *tr;
349
350 lockdep_assert_held(&event_mutex);
351
352 do_for_each_event_file(tr, file) {
353 if (!(file->flags & EVENT_FILE_FL_ENABLED))
354 continue;
355
356 if (enable) {
357 tracing_start_tgid_record();
358 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
359 } else {
360 tracing_stop_tgid_record();
361 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
362 &file->flags);
363 }
364 } while_for_each_event_file();
365}
366
367static int __ftrace_event_enable_disable(struct trace_event_file *file,
368 int enable, int soft_disable)
369{
370 struct trace_event_call *call = file->event_call;
371 struct trace_array *tr = file->tr;
372 unsigned long file_flags = file->flags;
373 int ret = 0;
374 int disable;
375
376 switch (enable) {
377 case 0:
378 /*
379 * When soft_disable is set and enable is cleared, the sm_ref
380 * reference counter is decremented. If it reaches 0, we want
381 * to clear the SOFT_DISABLED flag but leave the event in the
382 * state that it was. That is, if the event was enabled and
383 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
384 * is set we do not want the event to be enabled before we
385 * clear the bit.
386 *
387 * When soft_disable is not set but the SOFT_MODE flag is,
388 * we do nothing. Do not disable the tracepoint, otherwise
389 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
390 */
391 if (soft_disable) {
392 if (atomic_dec_return(&file->sm_ref) > 0)
393 break;
394 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
395 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
396 } else
397 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
398
399 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
400 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
401 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
402 tracing_stop_cmdline_record();
403 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
404 }
405
406 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
407 tracing_stop_tgid_record();
408 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
409 }
410
411 call->class->reg(call, TRACE_REG_UNREGISTER, file);
412 }
413 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
414 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
415 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
416 else
417 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
418 break;
419 case 1:
420 /*
421 * When soft_disable is set and enable is set, we want to
422 * register the tracepoint for the event, but leave the event
423 * as is. That means, if the event was already enabled, we do
424 * nothing (but set SOFT_MODE). If the event is disabled, we
425 * set SOFT_DISABLED before enabling the event tracepoint, so
426 * it still seems to be disabled.
427 */
428 if (!soft_disable)
429 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
430 else {
431 if (atomic_inc_return(&file->sm_ref) > 1)
432 break;
433 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
434 }
435
436 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
437 bool cmd = false, tgid = false;
438
439 /* Keep the event disabled, when going to SOFT_MODE. */
440 if (soft_disable)
441 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
442
443 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
444 cmd = true;
445 tracing_start_cmdline_record();
446 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
447 }
448
449 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
450 tgid = true;
451 tracing_start_tgid_record();
452 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
453 }
454
455 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
456 if (ret) {
457 if (cmd)
458 tracing_stop_cmdline_record();
459 if (tgid)
460 tracing_stop_tgid_record();
461 pr_info("event trace: Could not enable event "
462 "%s\n", trace_event_name(call));
463 break;
464 }
465 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
466
467 /* WAS_ENABLED gets set but never cleared. */
468 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
469 }
470 break;
471 }
472
473 /* Enable or disable use of trace_buffered_event */
474 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
475 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
476 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
477 trace_buffered_event_enable();
478 else
479 trace_buffered_event_disable();
480 }
481
482 return ret;
483}
484
485int trace_event_enable_disable(struct trace_event_file *file,
486 int enable, int soft_disable)
487{
488 return __ftrace_event_enable_disable(file, enable, soft_disable);
489}
490
491static int ftrace_event_enable_disable(struct trace_event_file *file,
492 int enable)
493{
494 return __ftrace_event_enable_disable(file, enable, 0);
495}
496
497static void ftrace_clear_events(struct trace_array *tr)
498{
499 struct trace_event_file *file;
500
501 mutex_lock(&event_mutex);
502 list_for_each_entry(file, &tr->events, list) {
503 ftrace_event_enable_disable(file, 0);
504 }
505 mutex_unlock(&event_mutex);
506}
507
508static void
509event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
510{
511 struct trace_pid_list *pid_list;
512 struct trace_array *tr = data;
513
514 pid_list = rcu_dereference_raw(tr->filtered_pids);
515 trace_filter_add_remove_task(pid_list, NULL, task);
516
517 pid_list = rcu_dereference_raw(tr->filtered_no_pids);
518 trace_filter_add_remove_task(pid_list, NULL, task);
519}
520
521static void
522event_filter_pid_sched_process_fork(void *data,
523 struct task_struct *self,
524 struct task_struct *task)
525{
526 struct trace_pid_list *pid_list;
527 struct trace_array *tr = data;
528
529 pid_list = rcu_dereference_sched(tr->filtered_pids);
530 trace_filter_add_remove_task(pid_list, self, task);
531
532 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
533 trace_filter_add_remove_task(pid_list, self, task);
534}
535
536void trace_event_follow_fork(struct trace_array *tr, bool enable)
537{
538 if (enable) {
539 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
540 tr, INT_MIN);
541 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
542 tr, INT_MAX);
543 } else {
544 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
545 tr);
546 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
547 tr);
548 }
549}
550
551static void
552event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
553 struct task_struct *prev, struct task_struct *next)
554{
555 struct trace_array *tr = data;
556 struct trace_pid_list *no_pid_list;
557 struct trace_pid_list *pid_list;
558 bool ret;
559
560 pid_list = rcu_dereference_sched(tr->filtered_pids);
561 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
562
563 /*
564 * Sched switch is funny, as we only want to ignore it
565 * in the notrace case if both prev and next should be ignored.
566 */
567 ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
568 trace_ignore_this_task(NULL, no_pid_list, next);
569
570 this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
571 (trace_ignore_this_task(pid_list, NULL, prev) &&
572 trace_ignore_this_task(pid_list, NULL, next)));
573}
574
575static void
576event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
577 struct task_struct *prev, struct task_struct *next)
578{
579 struct trace_array *tr = data;
580 struct trace_pid_list *no_pid_list;
581 struct trace_pid_list *pid_list;
582
583 pid_list = rcu_dereference_sched(tr->filtered_pids);
584 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
585
586 this_cpu_write(tr->array_buffer.data->ignore_pid,
587 trace_ignore_this_task(pid_list, no_pid_list, next));
588}
589
590static void
591event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
592{
593 struct trace_array *tr = data;
594 struct trace_pid_list *no_pid_list;
595 struct trace_pid_list *pid_list;
596
597 /* Nothing to do if we are already tracing */
598 if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
599 return;
600
601 pid_list = rcu_dereference_sched(tr->filtered_pids);
602 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
603
604 this_cpu_write(tr->array_buffer.data->ignore_pid,
605 trace_ignore_this_task(pid_list, no_pid_list, task));
606}
607
608static void
609event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
610{
611 struct trace_array *tr = data;
612 struct trace_pid_list *no_pid_list;
613 struct trace_pid_list *pid_list;
614
615 /* Nothing to do if we are not tracing */
616 if (this_cpu_read(tr->array_buffer.data->ignore_pid))
617 return;
618
619 pid_list = rcu_dereference_sched(tr->filtered_pids);
620 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
621
622 /* Set tracing if current is enabled */
623 this_cpu_write(tr->array_buffer.data->ignore_pid,
624 trace_ignore_this_task(pid_list, no_pid_list, current));
625}
626
627static void unregister_pid_events(struct trace_array *tr)
628{
629 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
630 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
631
632 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
633 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
634
635 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
636 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
637
638 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
639 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
640}
641
642static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
643{
644 struct trace_pid_list *pid_list;
645 struct trace_pid_list *no_pid_list;
646 struct trace_event_file *file;
647 int cpu;
648
649 pid_list = rcu_dereference_protected(tr->filtered_pids,
650 lockdep_is_held(&event_mutex));
651 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
652 lockdep_is_held(&event_mutex));
653
654 /* Make sure there's something to do */
655 if (!pid_type_enabled(type, pid_list, no_pid_list))
656 return;
657
658 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
659 unregister_pid_events(tr);
660
661 list_for_each_entry(file, &tr->events, list) {
662 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
663 }
664
665 for_each_possible_cpu(cpu)
666 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
667 }
668
669 if (type & TRACE_PIDS)
670 rcu_assign_pointer(tr->filtered_pids, NULL);
671
672 if (type & TRACE_NO_PIDS)
673 rcu_assign_pointer(tr->filtered_no_pids, NULL);
674
675 /* Wait till all users are no longer using pid filtering */
676 tracepoint_synchronize_unregister();
677
678 if ((type & TRACE_PIDS) && pid_list)
679 trace_free_pid_list(pid_list);
680
681 if ((type & TRACE_NO_PIDS) && no_pid_list)
682 trace_free_pid_list(no_pid_list);
683}
684
685static void ftrace_clear_event_pids(struct trace_array *tr, int type)
686{
687 mutex_lock(&event_mutex);
688 __ftrace_clear_event_pids(tr, type);
689 mutex_unlock(&event_mutex);
690}
691
692static void __put_system(struct event_subsystem *system)
693{
694 struct event_filter *filter = system->filter;
695
696 WARN_ON_ONCE(system_refcount(system) == 0);
697 if (system_refcount_dec(system))
698 return;
699
700 list_del(&system->list);
701
702 if (filter) {
703 kfree(filter->filter_string);
704 kfree(filter);
705 }
706 kfree_const(system->name);
707 kfree(system);
708}
709
710static void __get_system(struct event_subsystem *system)
711{
712 WARN_ON_ONCE(system_refcount(system) == 0);
713 system_refcount_inc(system);
714}
715
716static void __get_system_dir(struct trace_subsystem_dir *dir)
717{
718 WARN_ON_ONCE(dir->ref_count == 0);
719 dir->ref_count++;
720 __get_system(dir->subsystem);
721}
722
723static void __put_system_dir(struct trace_subsystem_dir *dir)
724{
725 WARN_ON_ONCE(dir->ref_count == 0);
726 /* If the subsystem is about to be freed, the dir must be too */
727 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
728
729 __put_system(dir->subsystem);
730 if (!--dir->ref_count)
731 kfree(dir);
732}
733
734static void put_system(struct trace_subsystem_dir *dir)
735{
736 mutex_lock(&event_mutex);
737 __put_system_dir(dir);
738 mutex_unlock(&event_mutex);
739}
740
741static void remove_subsystem(struct trace_subsystem_dir *dir)
742{
743 if (!dir)
744 return;
745
746 if (!--dir->nr_events) {
747 tracefs_remove(dir->entry);
748 list_del(&dir->list);
749 __put_system_dir(dir);
750 }
751}
752
753static void remove_event_file_dir(struct trace_event_file *file)
754{
755 struct dentry *dir = file->dir;
756 struct dentry *child;
757
758 if (dir) {
759 spin_lock(&dir->d_lock); /* probably unneeded */
760 list_for_each_entry(child, &dir->d_subdirs, d_child) {
761 if (d_really_is_positive(child)) /* probably unneeded */
762 d_inode(child)->i_private = NULL;
763 }
764 spin_unlock(&dir->d_lock);
765
766 tracefs_remove(dir);
767 }
768
769 list_del(&file->list);
770 remove_subsystem(file->system);
771 free_event_filter(file->filter);
772 kmem_cache_free(file_cachep, file);
773}
774
775/*
776 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
777 */
778static int
779__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
780 const char *sub, const char *event, int set)
781{
782 struct trace_event_file *file;
783 struct trace_event_call *call;
784 const char *name;
785 int ret = -EINVAL;
786 int eret = 0;
787
788 list_for_each_entry(file, &tr->events, list) {
789
790 call = file->event_call;
791 name = trace_event_name(call);
792
793 if (!name || !call->class || !call->class->reg)
794 continue;
795
796 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
797 continue;
798
799 if (match &&
800 strcmp(match, name) != 0 &&
801 strcmp(match, call->class->system) != 0)
802 continue;
803
804 if (sub && strcmp(sub, call->class->system) != 0)
805 continue;
806
807 if (event && strcmp(event, name) != 0)
808 continue;
809
810 ret = ftrace_event_enable_disable(file, set);
811
812 /*
813 * Save the first error and return that. Some events
814 * may still have been enabled, but let the user
815 * know that something went wrong.
816 */
817 if (ret && !eret)
818 eret = ret;
819
820 ret = eret;
821 }
822
823 return ret;
824}
825
826static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
827 const char *sub, const char *event, int set)
828{
829 int ret;
830
831 mutex_lock(&event_mutex);
832 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
833 mutex_unlock(&event_mutex);
834
835 return ret;
836}
837
838int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
839{
840 char *event = NULL, *sub = NULL, *match;
841 int ret;
842
843 if (!tr)
844 return -ENOENT;
845 /*
846 * The buf format can be <subsystem>:<event-name>
847 * *:<event-name> means any event by that name.
848 * :<event-name> is the same.
849 *
850 * <subsystem>:* means all events in that subsystem
851 * <subsystem>: means the same.
852 *
853 * <name> (no ':') means all events in a subsystem with
854 * the name <name> or any event that matches <name>
855 */
856
857 match = strsep(&buf, ":");
858 if (buf) {
859 sub = match;
860 event = buf;
861 match = NULL;
862
863 if (!strlen(sub) || strcmp(sub, "*") == 0)
864 sub = NULL;
865 if (!strlen(event) || strcmp(event, "*") == 0)
866 event = NULL;
867 }
868
869 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
870
871 /* Put back the colon to allow this to be called again */
872 if (buf)
873 *(buf - 1) = ':';
874
875 return ret;
876}
877
878/**
879 * trace_set_clr_event - enable or disable an event
880 * @system: system name to match (NULL for any system)
881 * @event: event name to match (NULL for all events, within system)
882 * @set: 1 to enable, 0 to disable
883 *
884 * This is a way for other parts of the kernel to enable or disable
885 * event recording.
886 *
887 * Returns 0 on success, -EINVAL if the parameters do not match any
888 * registered events.
889 */
890int trace_set_clr_event(const char *system, const char *event, int set)
891{
892 struct trace_array *tr = top_trace_array();
893
894 if (!tr)
895 return -ENODEV;
896
897 return __ftrace_set_clr_event(tr, NULL, system, event, set);
898}
899EXPORT_SYMBOL_GPL(trace_set_clr_event);
900
901/**
902 * trace_array_set_clr_event - enable or disable an event for a trace array.
903 * @tr: concerned trace array.
904 * @system: system name to match (NULL for any system)
905 * @event: event name to match (NULL for all events, within system)
906 * @enable: true to enable, false to disable
907 *
908 * This is a way for other parts of the kernel to enable or disable
909 * event recording.
910 *
911 * Returns 0 on success, -EINVAL if the parameters do not match any
912 * registered events.
913 */
914int trace_array_set_clr_event(struct trace_array *tr, const char *system,
915 const char *event, bool enable)
916{
917 int set;
918
919 if (!tr)
920 return -ENOENT;
921
922 set = (enable == true) ? 1 : 0;
923 return __ftrace_set_clr_event(tr, NULL, system, event, set);
924}
925EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
926
927/* 128 should be much more than enough */
928#define EVENT_BUF_SIZE 127
929
930static ssize_t
931ftrace_event_write(struct file *file, const char __user *ubuf,
932 size_t cnt, loff_t *ppos)
933{
934 struct trace_parser parser;
935 struct seq_file *m = file->private_data;
936 struct trace_array *tr = m->private;
937 ssize_t read, ret;
938
939 if (!cnt)
940 return 0;
941
942 ret = tracing_update_buffers();
943 if (ret < 0)
944 return ret;
945
946 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
947 return -ENOMEM;
948
949 read = trace_get_user(&parser, ubuf, cnt, ppos);
950
951 if (read >= 0 && trace_parser_loaded((&parser))) {
952 int set = 1;
953
954 if (*parser.buffer == '!')
955 set = 0;
956
957 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
958 if (ret)
959 goto out_put;
960 }
961
962 ret = read;
963
964 out_put:
965 trace_parser_put(&parser);
966
967 return ret;
968}
969
970static void *
971t_next(struct seq_file *m, void *v, loff_t *pos)
972{
973 struct trace_event_file *file = v;
974 struct trace_event_call *call;
975 struct trace_array *tr = m->private;
976
977 (*pos)++;
978
979 list_for_each_entry_continue(file, &tr->events, list) {
980 call = file->event_call;
981 /*
982 * The ftrace subsystem is for showing formats only.
983 * They can not be enabled or disabled via the event files.
984 */
985 if (call->class && call->class->reg &&
986 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
987 return file;
988 }
989
990 return NULL;
991}
992
993static void *t_start(struct seq_file *m, loff_t *pos)
994{
995 struct trace_event_file *file;
996 struct trace_array *tr = m->private;
997 loff_t l;
998
999 mutex_lock(&event_mutex);
1000
1001 file = list_entry(&tr->events, struct trace_event_file, list);
1002 for (l = 0; l <= *pos; ) {
1003 file = t_next(m, file, &l);
1004 if (!file)
1005 break;
1006 }
1007 return file;
1008}
1009
1010static void *
1011s_next(struct seq_file *m, void *v, loff_t *pos)
1012{
1013 struct trace_event_file *file = v;
1014 struct trace_array *tr = m->private;
1015
1016 (*pos)++;
1017
1018 list_for_each_entry_continue(file, &tr->events, list) {
1019 if (file->flags & EVENT_FILE_FL_ENABLED)
1020 return file;
1021 }
1022
1023 return NULL;
1024}
1025
1026static void *s_start(struct seq_file *m, loff_t *pos)
1027{
1028 struct trace_event_file *file;
1029 struct trace_array *tr = m->private;
1030 loff_t l;
1031
1032 mutex_lock(&event_mutex);
1033
1034 file = list_entry(&tr->events, struct trace_event_file, list);
1035 for (l = 0; l <= *pos; ) {
1036 file = s_next(m, file, &l);
1037 if (!file)
1038 break;
1039 }
1040 return file;
1041}
1042
1043static int t_show(struct seq_file *m, void *v)
1044{
1045 struct trace_event_file *file = v;
1046 struct trace_event_call *call = file->event_call;
1047
1048 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1049 seq_printf(m, "%s:", call->class->system);
1050 seq_printf(m, "%s\n", trace_event_name(call));
1051
1052 return 0;
1053}
1054
1055static void t_stop(struct seq_file *m, void *p)
1056{
1057 mutex_unlock(&event_mutex);
1058}
1059
1060static void *
1061__next(struct seq_file *m, void *v, loff_t *pos, int type)
1062{
1063 struct trace_array *tr = m->private;
1064 struct trace_pid_list *pid_list;
1065
1066 if (type == TRACE_PIDS)
1067 pid_list = rcu_dereference_sched(tr->filtered_pids);
1068 else
1069 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1070
1071 return trace_pid_next(pid_list, v, pos);
1072}
1073
1074static void *
1075p_next(struct seq_file *m, void *v, loff_t *pos)
1076{
1077 return __next(m, v, pos, TRACE_PIDS);
1078}
1079
1080static void *
1081np_next(struct seq_file *m, void *v, loff_t *pos)
1082{
1083 return __next(m, v, pos, TRACE_NO_PIDS);
1084}
1085
1086static void *__start(struct seq_file *m, loff_t *pos, int type)
1087 __acquires(RCU)
1088{
1089 struct trace_pid_list *pid_list;
1090 struct trace_array *tr = m->private;
1091
1092 /*
1093 * Grab the mutex, to keep calls to p_next() having the same
1094 * tr->filtered_pids as p_start() has.
1095 * If we just passed the tr->filtered_pids around, then RCU would
1096 * have been enough, but doing that makes things more complex.
1097 */
1098 mutex_lock(&event_mutex);
1099 rcu_read_lock_sched();
1100
1101 if (type == TRACE_PIDS)
1102 pid_list = rcu_dereference_sched(tr->filtered_pids);
1103 else
1104 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1105
1106 if (!pid_list)
1107 return NULL;
1108
1109 return trace_pid_start(pid_list, pos);
1110}
1111
1112static void *p_start(struct seq_file *m, loff_t *pos)
1113 __acquires(RCU)
1114{
1115 return __start(m, pos, TRACE_PIDS);
1116}
1117
1118static void *np_start(struct seq_file *m, loff_t *pos)
1119 __acquires(RCU)
1120{
1121 return __start(m, pos, TRACE_NO_PIDS);
1122}
1123
1124static void p_stop(struct seq_file *m, void *p)
1125 __releases(RCU)
1126{
1127 rcu_read_unlock_sched();
1128 mutex_unlock(&event_mutex);
1129}
1130
1131static ssize_t
1132event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1133 loff_t *ppos)
1134{
1135 struct trace_event_file *file;
1136 unsigned long flags;
1137 char buf[4] = "0";
1138
1139 mutex_lock(&event_mutex);
1140 file = event_file_data(filp);
1141 if (likely(file))
1142 flags = file->flags;
1143 mutex_unlock(&event_mutex);
1144
1145 if (!file)
1146 return -ENODEV;
1147
1148 if (flags & EVENT_FILE_FL_ENABLED &&
1149 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1150 strcpy(buf, "1");
1151
1152 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1153 flags & EVENT_FILE_FL_SOFT_MODE)
1154 strcat(buf, "*");
1155
1156 strcat(buf, "\n");
1157
1158 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1159}
1160
1161static ssize_t
1162event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1163 loff_t *ppos)
1164{
1165 struct trace_event_file *file;
1166 unsigned long val;
1167 int ret;
1168
1169 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1170 if (ret)
1171 return ret;
1172
1173 ret = tracing_update_buffers();
1174 if (ret < 0)
1175 return ret;
1176
1177 switch (val) {
1178 case 0:
1179 case 1:
1180 ret = -ENODEV;
1181 mutex_lock(&event_mutex);
1182 file = event_file_data(filp);
1183 if (likely(file))
1184 ret = ftrace_event_enable_disable(file, val);
1185 mutex_unlock(&event_mutex);
1186 break;
1187
1188 default:
1189 return -EINVAL;
1190 }
1191
1192 *ppos += cnt;
1193
1194 return ret ? ret : cnt;
1195}
1196
1197static ssize_t
1198system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1199 loff_t *ppos)
1200{
1201 const char set_to_char[4] = { '?', '0', '1', 'X' };
1202 struct trace_subsystem_dir *dir = filp->private_data;
1203 struct event_subsystem *system = dir->subsystem;
1204 struct trace_event_call *call;
1205 struct trace_event_file *file;
1206 struct trace_array *tr = dir->tr;
1207 char buf[2];
1208 int set = 0;
1209 int ret;
1210
1211 mutex_lock(&event_mutex);
1212 list_for_each_entry(file, &tr->events, list) {
1213 call = file->event_call;
1214 if (!trace_event_name(call) || !call->class || !call->class->reg)
1215 continue;
1216
1217 if (system && strcmp(call->class->system, system->name) != 0)
1218 continue;
1219
1220 /*
1221 * We need to find out if all the events are set
1222 * or if all events or cleared, or if we have
1223 * a mixture.
1224 */
1225 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1226
1227 /*
1228 * If we have a mixture, no need to look further.
1229 */
1230 if (set == 3)
1231 break;
1232 }
1233 mutex_unlock(&event_mutex);
1234
1235 buf[0] = set_to_char[set];
1236 buf[1] = '\n';
1237
1238 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1239
1240 return ret;
1241}
1242
1243static ssize_t
1244system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1245 loff_t *ppos)
1246{
1247 struct trace_subsystem_dir *dir = filp->private_data;
1248 struct event_subsystem *system = dir->subsystem;
1249 const char *name = NULL;
1250 unsigned long val;
1251 ssize_t ret;
1252
1253 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1254 if (ret)
1255 return ret;
1256
1257 ret = tracing_update_buffers();
1258 if (ret < 0)
1259 return ret;
1260
1261 if (val != 0 && val != 1)
1262 return -EINVAL;
1263
1264 /*
1265 * Opening of "enable" adds a ref count to system,
1266 * so the name is safe to use.
1267 */
1268 if (system)
1269 name = system->name;
1270
1271 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1272 if (ret)
1273 goto out;
1274
1275 ret = cnt;
1276
1277out:
1278 *ppos += cnt;
1279
1280 return ret;
1281}
1282
1283enum {
1284 FORMAT_HEADER = 1,
1285 FORMAT_FIELD_SEPERATOR = 2,
1286 FORMAT_PRINTFMT = 3,
1287};
1288
1289static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1290{
1291 struct trace_event_call *call = event_file_data(m->private);
1292 struct list_head *common_head = &ftrace_common_fields;
1293 struct list_head *head = trace_get_fields(call);
1294 struct list_head *node = v;
1295
1296 (*pos)++;
1297
1298 switch ((unsigned long)v) {
1299 case FORMAT_HEADER:
1300 node = common_head;
1301 break;
1302
1303 case FORMAT_FIELD_SEPERATOR:
1304 node = head;
1305 break;
1306
1307 case FORMAT_PRINTFMT:
1308 /* all done */
1309 return NULL;
1310 }
1311
1312 node = node->prev;
1313 if (node == common_head)
1314 return (void *)FORMAT_FIELD_SEPERATOR;
1315 else if (node == head)
1316 return (void *)FORMAT_PRINTFMT;
1317 else
1318 return node;
1319}
1320
1321static int f_show(struct seq_file *m, void *v)
1322{
1323 struct trace_event_call *call = event_file_data(m->private);
1324 struct ftrace_event_field *field;
1325 const char *array_descriptor;
1326
1327 switch ((unsigned long)v) {
1328 case FORMAT_HEADER:
1329 seq_printf(m, "name: %s\n", trace_event_name(call));
1330 seq_printf(m, "ID: %d\n", call->event.type);
1331 seq_puts(m, "format:\n");
1332 return 0;
1333
1334 case FORMAT_FIELD_SEPERATOR:
1335 seq_putc(m, '\n');
1336 return 0;
1337
1338 case FORMAT_PRINTFMT:
1339 seq_printf(m, "\nprint fmt: %s\n",
1340 call->print_fmt);
1341 return 0;
1342 }
1343
1344 field = list_entry(v, struct ftrace_event_field, link);
1345 /*
1346 * Smartly shows the array type(except dynamic array).
1347 * Normal:
1348 * field:TYPE VAR
1349 * If TYPE := TYPE[LEN], it is shown:
1350 * field:TYPE VAR[LEN]
1351 */
1352 array_descriptor = strchr(field->type, '[');
1353
1354 if (str_has_prefix(field->type, "__data_loc"))
1355 array_descriptor = NULL;
1356
1357 if (!array_descriptor)
1358 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1359 field->type, field->name, field->offset,
1360 field->size, !!field->is_signed);
1361 else
1362 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1363 (int)(array_descriptor - field->type),
1364 field->type, field->name,
1365 array_descriptor, field->offset,
1366 field->size, !!field->is_signed);
1367
1368 return 0;
1369}
1370
1371static void *f_start(struct seq_file *m, loff_t *pos)
1372{
1373 void *p = (void *)FORMAT_HEADER;
1374 loff_t l = 0;
1375
1376 /* ->stop() is called even if ->start() fails */
1377 mutex_lock(&event_mutex);
1378 if (!event_file_data(m->private))
1379 return ERR_PTR(-ENODEV);
1380
1381 while (l < *pos && p)
1382 p = f_next(m, p, &l);
1383
1384 return p;
1385}
1386
1387static void f_stop(struct seq_file *m, void *p)
1388{
1389 mutex_unlock(&event_mutex);
1390}
1391
1392static const struct seq_operations trace_format_seq_ops = {
1393 .start = f_start,
1394 .next = f_next,
1395 .stop = f_stop,
1396 .show = f_show,
1397};
1398
1399static int trace_format_open(struct inode *inode, struct file *file)
1400{
1401 struct seq_file *m;
1402 int ret;
1403
1404 /* Do we want to hide event format files on tracefs lockdown? */
1405
1406 ret = seq_open(file, &trace_format_seq_ops);
1407 if (ret < 0)
1408 return ret;
1409
1410 m = file->private_data;
1411 m->private = file;
1412
1413 return 0;
1414}
1415
1416static ssize_t
1417event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1418{
1419 int id = (long)event_file_data(filp);
1420 char buf[32];
1421 int len;
1422
1423 if (unlikely(!id))
1424 return -ENODEV;
1425
1426 len = sprintf(buf, "%d\n", id);
1427
1428 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1429}
1430
1431static ssize_t
1432event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1433 loff_t *ppos)
1434{
1435 struct trace_event_file *file;
1436 struct trace_seq *s;
1437 int r = -ENODEV;
1438
1439 if (*ppos)
1440 return 0;
1441
1442 s = kmalloc(sizeof(*s), GFP_KERNEL);
1443
1444 if (!s)
1445 return -ENOMEM;
1446
1447 trace_seq_init(s);
1448
1449 mutex_lock(&event_mutex);
1450 file = event_file_data(filp);
1451 if (file)
1452 print_event_filter(file, s);
1453 mutex_unlock(&event_mutex);
1454
1455 if (file)
1456 r = simple_read_from_buffer(ubuf, cnt, ppos,
1457 s->buffer, trace_seq_used(s));
1458
1459 kfree(s);
1460
1461 return r;
1462}
1463
1464static ssize_t
1465event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1466 loff_t *ppos)
1467{
1468 struct trace_event_file *file;
1469 char *buf;
1470 int err = -ENODEV;
1471
1472 if (cnt >= PAGE_SIZE)
1473 return -EINVAL;
1474
1475 buf = memdup_user_nul(ubuf, cnt);
1476 if (IS_ERR(buf))
1477 return PTR_ERR(buf);
1478
1479 mutex_lock(&event_mutex);
1480 file = event_file_data(filp);
1481 if (file)
1482 err = apply_event_filter(file, buf);
1483 mutex_unlock(&event_mutex);
1484
1485 kfree(buf);
1486 if (err < 0)
1487 return err;
1488
1489 *ppos += cnt;
1490
1491 return cnt;
1492}
1493
1494static LIST_HEAD(event_subsystems);
1495
1496static int subsystem_open(struct inode *inode, struct file *filp)
1497{
1498 struct event_subsystem *system = NULL;
1499 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1500 struct trace_array *tr;
1501 int ret;
1502
1503 if (tracing_is_disabled())
1504 return -ENODEV;
1505
1506 /* Make sure the system still exists */
1507 mutex_lock(&event_mutex);
1508 mutex_lock(&trace_types_lock);
1509 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1510 list_for_each_entry(dir, &tr->systems, list) {
1511 if (dir == inode->i_private) {
1512 /* Don't open systems with no events */
1513 if (dir->nr_events) {
1514 __get_system_dir(dir);
1515 system = dir->subsystem;
1516 }
1517 goto exit_loop;
1518 }
1519 }
1520 }
1521 exit_loop:
1522 mutex_unlock(&trace_types_lock);
1523 mutex_unlock(&event_mutex);
1524
1525 if (!system)
1526 return -ENODEV;
1527
1528 /* Some versions of gcc think dir can be uninitialized here */
1529 WARN_ON(!dir);
1530
1531 /* Still need to increment the ref count of the system */
1532 if (trace_array_get(tr) < 0) {
1533 put_system(dir);
1534 return -ENODEV;
1535 }
1536
1537 ret = tracing_open_generic(inode, filp);
1538 if (ret < 0) {
1539 trace_array_put(tr);
1540 put_system(dir);
1541 }
1542
1543 return ret;
1544}
1545
1546static int system_tr_open(struct inode *inode, struct file *filp)
1547{
1548 struct trace_subsystem_dir *dir;
1549 struct trace_array *tr = inode->i_private;
1550 int ret;
1551
1552 /* Make a temporary dir that has no system but points to tr */
1553 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1554 if (!dir)
1555 return -ENOMEM;
1556
1557 ret = tracing_open_generic_tr(inode, filp);
1558 if (ret < 0) {
1559 kfree(dir);
1560 return ret;
1561 }
1562 dir->tr = tr;
1563 filp->private_data = dir;
1564
1565 return 0;
1566}
1567
1568static int subsystem_release(struct inode *inode, struct file *file)
1569{
1570 struct trace_subsystem_dir *dir = file->private_data;
1571
1572 trace_array_put(dir->tr);
1573
1574 /*
1575 * If dir->subsystem is NULL, then this is a temporary
1576 * descriptor that was made for a trace_array to enable
1577 * all subsystems.
1578 */
1579 if (dir->subsystem)
1580 put_system(dir);
1581 else
1582 kfree(dir);
1583
1584 return 0;
1585}
1586
1587static ssize_t
1588subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1589 loff_t *ppos)
1590{
1591 struct trace_subsystem_dir *dir = filp->private_data;
1592 struct event_subsystem *system = dir->subsystem;
1593 struct trace_seq *s;
1594 int r;
1595
1596 if (*ppos)
1597 return 0;
1598
1599 s = kmalloc(sizeof(*s), GFP_KERNEL);
1600 if (!s)
1601 return -ENOMEM;
1602
1603 trace_seq_init(s);
1604
1605 print_subsystem_event_filter(system, s);
1606 r = simple_read_from_buffer(ubuf, cnt, ppos,
1607 s->buffer, trace_seq_used(s));
1608
1609 kfree(s);
1610
1611 return r;
1612}
1613
1614static ssize_t
1615subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1616 loff_t *ppos)
1617{
1618 struct trace_subsystem_dir *dir = filp->private_data;
1619 char *buf;
1620 int err;
1621
1622 if (cnt >= PAGE_SIZE)
1623 return -EINVAL;
1624
1625 buf = memdup_user_nul(ubuf, cnt);
1626 if (IS_ERR(buf))
1627 return PTR_ERR(buf);
1628
1629 err = apply_subsystem_event_filter(dir, buf);
1630 kfree(buf);
1631 if (err < 0)
1632 return err;
1633
1634 *ppos += cnt;
1635
1636 return cnt;
1637}
1638
1639static ssize_t
1640show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1641{
1642 int (*func)(struct trace_seq *s) = filp->private_data;
1643 struct trace_seq *s;
1644 int r;
1645
1646 if (*ppos)
1647 return 0;
1648
1649 s = kmalloc(sizeof(*s), GFP_KERNEL);
1650 if (!s)
1651 return -ENOMEM;
1652
1653 trace_seq_init(s);
1654
1655 func(s);
1656 r = simple_read_from_buffer(ubuf, cnt, ppos,
1657 s->buffer, trace_seq_used(s));
1658
1659 kfree(s);
1660
1661 return r;
1662}
1663
1664static void ignore_task_cpu(void *data)
1665{
1666 struct trace_array *tr = data;
1667 struct trace_pid_list *pid_list;
1668 struct trace_pid_list *no_pid_list;
1669
1670 /*
1671 * This function is called by on_each_cpu() while the
1672 * event_mutex is held.
1673 */
1674 pid_list = rcu_dereference_protected(tr->filtered_pids,
1675 mutex_is_locked(&event_mutex));
1676 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1677 mutex_is_locked(&event_mutex));
1678
1679 this_cpu_write(tr->array_buffer.data->ignore_pid,
1680 trace_ignore_this_task(pid_list, no_pid_list, current));
1681}
1682
1683static void register_pid_events(struct trace_array *tr)
1684{
1685 /*
1686 * Register a probe that is called before all other probes
1687 * to set ignore_pid if next or prev do not match.
1688 * Register a probe this is called after all other probes
1689 * to only keep ignore_pid set if next pid matches.
1690 */
1691 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1692 tr, INT_MAX);
1693 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1694 tr, 0);
1695
1696 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1697 tr, INT_MAX);
1698 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1699 tr, 0);
1700
1701 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1702 tr, INT_MAX);
1703 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1704 tr, 0);
1705
1706 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1707 tr, INT_MAX);
1708 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1709 tr, 0);
1710}
1711
1712static ssize_t
1713event_pid_write(struct file *filp, const char __user *ubuf,
1714 size_t cnt, loff_t *ppos, int type)
1715{
1716 struct seq_file *m = filp->private_data;
1717 struct trace_array *tr = m->private;
1718 struct trace_pid_list *filtered_pids = NULL;
1719 struct trace_pid_list *other_pids = NULL;
1720 struct trace_pid_list *pid_list;
1721 struct trace_event_file *file;
1722 ssize_t ret;
1723
1724 if (!cnt)
1725 return 0;
1726
1727 ret = tracing_update_buffers();
1728 if (ret < 0)
1729 return ret;
1730
1731 mutex_lock(&event_mutex);
1732
1733 if (type == TRACE_PIDS) {
1734 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1735 lockdep_is_held(&event_mutex));
1736 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1737 lockdep_is_held(&event_mutex));
1738 } else {
1739 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1740 lockdep_is_held(&event_mutex));
1741 other_pids = rcu_dereference_protected(tr->filtered_pids,
1742 lockdep_is_held(&event_mutex));
1743 }
1744
1745 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1746 if (ret < 0)
1747 goto out;
1748
1749 if (type == TRACE_PIDS)
1750 rcu_assign_pointer(tr->filtered_pids, pid_list);
1751 else
1752 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1753
1754 list_for_each_entry(file, &tr->events, list) {
1755 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1756 }
1757
1758 if (filtered_pids) {
1759 tracepoint_synchronize_unregister();
1760 trace_free_pid_list(filtered_pids);
1761 } else if (pid_list && !other_pids) {
1762 register_pid_events(tr);
1763 }
1764
1765 /*
1766 * Ignoring of pids is done at task switch. But we have to
1767 * check for those tasks that are currently running.
1768 * Always do this in case a pid was appended or removed.
1769 */
1770 on_each_cpu(ignore_task_cpu, tr, 1);
1771
1772 out:
1773 mutex_unlock(&event_mutex);
1774
1775 if (ret > 0)
1776 *ppos += ret;
1777
1778 return ret;
1779}
1780
1781static ssize_t
1782ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1783 size_t cnt, loff_t *ppos)
1784{
1785 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1786}
1787
1788static ssize_t
1789ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1790 size_t cnt, loff_t *ppos)
1791{
1792 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
1793}
1794
1795static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1796static int ftrace_event_set_open(struct inode *inode, struct file *file);
1797static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1798static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
1799static int ftrace_event_release(struct inode *inode, struct file *file);
1800
1801static const struct seq_operations show_event_seq_ops = {
1802 .start = t_start,
1803 .next = t_next,
1804 .show = t_show,
1805 .stop = t_stop,
1806};
1807
1808static const struct seq_operations show_set_event_seq_ops = {
1809 .start = s_start,
1810 .next = s_next,
1811 .show = t_show,
1812 .stop = t_stop,
1813};
1814
1815static const struct seq_operations show_set_pid_seq_ops = {
1816 .start = p_start,
1817 .next = p_next,
1818 .show = trace_pid_show,
1819 .stop = p_stop,
1820};
1821
1822static const struct seq_operations show_set_no_pid_seq_ops = {
1823 .start = np_start,
1824 .next = np_next,
1825 .show = trace_pid_show,
1826 .stop = p_stop,
1827};
1828
1829static const struct file_operations ftrace_avail_fops = {
1830 .open = ftrace_event_avail_open,
1831 .read = seq_read,
1832 .llseek = seq_lseek,
1833 .release = seq_release,
1834};
1835
1836static const struct file_operations ftrace_set_event_fops = {
1837 .open = ftrace_event_set_open,
1838 .read = seq_read,
1839 .write = ftrace_event_write,
1840 .llseek = seq_lseek,
1841 .release = ftrace_event_release,
1842};
1843
1844static const struct file_operations ftrace_set_event_pid_fops = {
1845 .open = ftrace_event_set_pid_open,
1846 .read = seq_read,
1847 .write = ftrace_event_pid_write,
1848 .llseek = seq_lseek,
1849 .release = ftrace_event_release,
1850};
1851
1852static const struct file_operations ftrace_set_event_notrace_pid_fops = {
1853 .open = ftrace_event_set_npid_open,
1854 .read = seq_read,
1855 .write = ftrace_event_npid_write,
1856 .llseek = seq_lseek,
1857 .release = ftrace_event_release,
1858};
1859
1860static const struct file_operations ftrace_enable_fops = {
1861 .open = tracing_open_generic,
1862 .read = event_enable_read,
1863 .write = event_enable_write,
1864 .llseek = default_llseek,
1865};
1866
1867static const struct file_operations ftrace_event_format_fops = {
1868 .open = trace_format_open,
1869 .read = seq_read,
1870 .llseek = seq_lseek,
1871 .release = seq_release,
1872};
1873
1874static const struct file_operations ftrace_event_id_fops = {
1875 .read = event_id_read,
1876 .llseek = default_llseek,
1877};
1878
1879static const struct file_operations ftrace_event_filter_fops = {
1880 .open = tracing_open_generic,
1881 .read = event_filter_read,
1882 .write = event_filter_write,
1883 .llseek = default_llseek,
1884};
1885
1886static const struct file_operations ftrace_subsystem_filter_fops = {
1887 .open = subsystem_open,
1888 .read = subsystem_filter_read,
1889 .write = subsystem_filter_write,
1890 .llseek = default_llseek,
1891 .release = subsystem_release,
1892};
1893
1894static const struct file_operations ftrace_system_enable_fops = {
1895 .open = subsystem_open,
1896 .read = system_enable_read,
1897 .write = system_enable_write,
1898 .llseek = default_llseek,
1899 .release = subsystem_release,
1900};
1901
1902static const struct file_operations ftrace_tr_enable_fops = {
1903 .open = system_tr_open,
1904 .read = system_enable_read,
1905 .write = system_enable_write,
1906 .llseek = default_llseek,
1907 .release = subsystem_release,
1908};
1909
1910static const struct file_operations ftrace_show_header_fops = {
1911 .open = tracing_open_generic,
1912 .read = show_header,
1913 .llseek = default_llseek,
1914};
1915
1916static int
1917ftrace_event_open(struct inode *inode, struct file *file,
1918 const struct seq_operations *seq_ops)
1919{
1920 struct seq_file *m;
1921 int ret;
1922
1923 ret = security_locked_down(LOCKDOWN_TRACEFS);
1924 if (ret)
1925 return ret;
1926
1927 ret = seq_open(file, seq_ops);
1928 if (ret < 0)
1929 return ret;
1930 m = file->private_data;
1931 /* copy tr over to seq ops */
1932 m->private = inode->i_private;
1933
1934 return ret;
1935}
1936
1937static int ftrace_event_release(struct inode *inode, struct file *file)
1938{
1939 struct trace_array *tr = inode->i_private;
1940
1941 trace_array_put(tr);
1942
1943 return seq_release(inode, file);
1944}
1945
1946static int
1947ftrace_event_avail_open(struct inode *inode, struct file *file)
1948{
1949 const struct seq_operations *seq_ops = &show_event_seq_ops;
1950
1951 /* Checks for tracefs lockdown */
1952 return ftrace_event_open(inode, file, seq_ops);
1953}
1954
1955static int
1956ftrace_event_set_open(struct inode *inode, struct file *file)
1957{
1958 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1959 struct trace_array *tr = inode->i_private;
1960 int ret;
1961
1962 ret = tracing_check_open_get_tr(tr);
1963 if (ret)
1964 return ret;
1965
1966 if ((file->f_mode & FMODE_WRITE) &&
1967 (file->f_flags & O_TRUNC))
1968 ftrace_clear_events(tr);
1969
1970 ret = ftrace_event_open(inode, file, seq_ops);
1971 if (ret < 0)
1972 trace_array_put(tr);
1973 return ret;
1974}
1975
1976static int
1977ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1978{
1979 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1980 struct trace_array *tr = inode->i_private;
1981 int ret;
1982
1983 ret = tracing_check_open_get_tr(tr);
1984 if (ret)
1985 return ret;
1986
1987 if ((file->f_mode & FMODE_WRITE) &&
1988 (file->f_flags & O_TRUNC))
1989 ftrace_clear_event_pids(tr, TRACE_PIDS);
1990
1991 ret = ftrace_event_open(inode, file, seq_ops);
1992 if (ret < 0)
1993 trace_array_put(tr);
1994 return ret;
1995}
1996
1997static int
1998ftrace_event_set_npid_open(struct inode *inode, struct file *file)
1999{
2000 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2001 struct trace_array *tr = inode->i_private;
2002 int ret;
2003
2004 ret = tracing_check_open_get_tr(tr);
2005 if (ret)
2006 return ret;
2007
2008 if ((file->f_mode & FMODE_WRITE) &&
2009 (file->f_flags & O_TRUNC))
2010 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2011
2012 ret = ftrace_event_open(inode, file, seq_ops);
2013 if (ret < 0)
2014 trace_array_put(tr);
2015 return ret;
2016}
2017
2018static struct event_subsystem *
2019create_new_subsystem(const char *name)
2020{
2021 struct event_subsystem *system;
2022
2023 /* need to create new entry */
2024 system = kmalloc(sizeof(*system), GFP_KERNEL);
2025 if (!system)
2026 return NULL;
2027
2028 system->ref_count = 1;
2029
2030 /* Only allocate if dynamic (kprobes and modules) */
2031 system->name = kstrdup_const(name, GFP_KERNEL);
2032 if (!system->name)
2033 goto out_free;
2034
2035 system->filter = NULL;
2036
2037 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2038 if (!system->filter)
2039 goto out_free;
2040
2041 list_add(&system->list, &event_subsystems);
2042
2043 return system;
2044
2045 out_free:
2046 kfree_const(system->name);
2047 kfree(system);
2048 return NULL;
2049}
2050
2051static struct dentry *
2052event_subsystem_dir(struct trace_array *tr, const char *name,
2053 struct trace_event_file *file, struct dentry *parent)
2054{
2055 struct trace_subsystem_dir *dir;
2056 struct event_subsystem *system;
2057 struct dentry *entry;
2058
2059 /* First see if we did not already create this dir */
2060 list_for_each_entry(dir, &tr->systems, list) {
2061 system = dir->subsystem;
2062 if (strcmp(system->name, name) == 0) {
2063 dir->nr_events++;
2064 file->system = dir;
2065 return dir->entry;
2066 }
2067 }
2068
2069 /* Now see if the system itself exists. */
2070 list_for_each_entry(system, &event_subsystems, list) {
2071 if (strcmp(system->name, name) == 0)
2072 break;
2073 }
2074 /* Reset system variable when not found */
2075 if (&system->list == &event_subsystems)
2076 system = NULL;
2077
2078 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2079 if (!dir)
2080 goto out_fail;
2081
2082 if (!system) {
2083 system = create_new_subsystem(name);
2084 if (!system)
2085 goto out_free;
2086 } else
2087 __get_system(system);
2088
2089 dir->entry = tracefs_create_dir(name, parent);
2090 if (!dir->entry) {
2091 pr_warn("Failed to create system directory %s\n", name);
2092 __put_system(system);
2093 goto out_free;
2094 }
2095
2096 dir->tr = tr;
2097 dir->ref_count = 1;
2098 dir->nr_events = 1;
2099 dir->subsystem = system;
2100 file->system = dir;
2101
2102 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2103 &ftrace_subsystem_filter_fops);
2104 if (!entry) {
2105 kfree(system->filter);
2106 system->filter = NULL;
2107 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2108 }
2109
2110 trace_create_file("enable", 0644, dir->entry, dir,
2111 &ftrace_system_enable_fops);
2112
2113 list_add(&dir->list, &tr->systems);
2114
2115 return dir->entry;
2116
2117 out_free:
2118 kfree(dir);
2119 out_fail:
2120 /* Only print this message if failed on memory allocation */
2121 if (!dir || !system)
2122 pr_warn("No memory to create event subsystem %s\n", name);
2123 return NULL;
2124}
2125
2126static int
2127event_create_dir(struct dentry *parent, struct trace_event_file *file)
2128{
2129 struct trace_event_call *call = file->event_call;
2130 struct trace_array *tr = file->tr;
2131 struct list_head *head;
2132 struct dentry *d_events;
2133 const char *name;
2134 int ret;
2135
2136 /*
2137 * If the trace point header did not define TRACE_SYSTEM
2138 * then the system would be called "TRACE_SYSTEM".
2139 */
2140 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2141 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2142 if (!d_events)
2143 return -ENOMEM;
2144 } else
2145 d_events = parent;
2146
2147 name = trace_event_name(call);
2148 file->dir = tracefs_create_dir(name, d_events);
2149 if (!file->dir) {
2150 pr_warn("Could not create tracefs '%s' directory\n", name);
2151 return -1;
2152 }
2153
2154 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2155 trace_create_file("enable", 0644, file->dir, file,
2156 &ftrace_enable_fops);
2157
2158#ifdef CONFIG_PERF_EVENTS
2159 if (call->event.type && call->class->reg)
2160 trace_create_file("id", 0444, file->dir,
2161 (void *)(long)call->event.type,
2162 &ftrace_event_id_fops);
2163#endif
2164
2165 /*
2166 * Other events may have the same class. Only update
2167 * the fields if they are not already defined.
2168 */
2169 head = trace_get_fields(call);
2170 if (list_empty(head)) {
2171 struct trace_event_fields *field = call->class->fields_array;
2172 unsigned int offset = sizeof(struct trace_entry);
2173
2174 for (; field->type; field++) {
2175 if (field->type == TRACE_FUNCTION_TYPE) {
2176 ret = field->define_fields(call);
2177 break;
2178 }
2179
2180 offset = ALIGN(offset, field->align);
2181 ret = trace_define_field(call, field->type, field->name,
2182 offset, field->size,
2183 field->is_signed, field->filter_type);
2184 if (ret)
2185 break;
2186
2187 offset += field->size;
2188 }
2189 if (ret < 0) {
2190 pr_warn("Could not initialize trace point events/%s\n",
2191 name);
2192 return -1;
2193 }
2194 }
2195
2196 /*
2197 * Only event directories that can be enabled should have
2198 * triggers or filters.
2199 */
2200 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2201 trace_create_file("filter", 0644, file->dir, file,
2202 &ftrace_event_filter_fops);
2203
2204 trace_create_file("trigger", 0644, file->dir, file,
2205 &event_trigger_fops);
2206 }
2207
2208#ifdef CONFIG_HIST_TRIGGERS
2209 trace_create_file("hist", 0444, file->dir, file,
2210 &event_hist_fops);
2211#endif
2212#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2213 trace_create_file("hist_debug", 0444, file->dir, file,
2214 &event_hist_debug_fops);
2215#endif
2216 trace_create_file("format", 0444, file->dir, call,
2217 &ftrace_event_format_fops);
2218
2219#ifdef CONFIG_TRACE_EVENT_INJECT
2220 if (call->event.type && call->class->reg)
2221 trace_create_file("inject", 0200, file->dir, file,
2222 &event_inject_fops);
2223#endif
2224
2225 return 0;
2226}
2227
2228static void remove_event_from_tracers(struct trace_event_call *call)
2229{
2230 struct trace_event_file *file;
2231 struct trace_array *tr;
2232
2233 do_for_each_event_file_safe(tr, file) {
2234 if (file->event_call != call)
2235 continue;
2236
2237 remove_event_file_dir(file);
2238 /*
2239 * The do_for_each_event_file_safe() is
2240 * a double loop. After finding the call for this
2241 * trace_array, we use break to jump to the next
2242 * trace_array.
2243 */
2244 break;
2245 } while_for_each_event_file();
2246}
2247
2248static void event_remove(struct trace_event_call *call)
2249{
2250 struct trace_array *tr;
2251 struct trace_event_file *file;
2252
2253 do_for_each_event_file(tr, file) {
2254 if (file->event_call != call)
2255 continue;
2256
2257 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2258 tr->clear_trace = true;
2259
2260 ftrace_event_enable_disable(file, 0);
2261 /*
2262 * The do_for_each_event_file() is
2263 * a double loop. After finding the call for this
2264 * trace_array, we use break to jump to the next
2265 * trace_array.
2266 */
2267 break;
2268 } while_for_each_event_file();
2269
2270 if (call->event.funcs)
2271 __unregister_trace_event(&call->event);
2272 remove_event_from_tracers(call);
2273 list_del(&call->list);
2274}
2275
2276static int event_init(struct trace_event_call *call)
2277{
2278 int ret = 0;
2279 const char *name;
2280
2281 name = trace_event_name(call);
2282 if (WARN_ON(!name))
2283 return -EINVAL;
2284
2285 if (call->class->raw_init) {
2286 ret = call->class->raw_init(call);
2287 if (ret < 0 && ret != -ENOSYS)
2288 pr_warn("Could not initialize trace events/%s\n", name);
2289 }
2290
2291 return ret;
2292}
2293
2294static int
2295__register_event(struct trace_event_call *call, struct module *mod)
2296{
2297 int ret;
2298
2299 ret = event_init(call);
2300 if (ret < 0)
2301 return ret;
2302
2303 list_add(&call->list, &ftrace_events);
2304 call->mod = mod;
2305
2306 return 0;
2307}
2308
2309static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2310{
2311 int rlen;
2312 int elen;
2313
2314 /* Find the length of the eval value as a string */
2315 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2316 /* Make sure there's enough room to replace the string with the value */
2317 if (len < elen)
2318 return NULL;
2319
2320 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2321
2322 /* Get the rest of the string of ptr */
2323 rlen = strlen(ptr + len);
2324 memmove(ptr + elen, ptr + len, rlen);
2325 /* Make sure we end the new string */
2326 ptr[elen + rlen] = 0;
2327
2328 return ptr + elen;
2329}
2330
2331static void update_event_printk(struct trace_event_call *call,
2332 struct trace_eval_map *map)
2333{
2334 char *ptr;
2335 int quote = 0;
2336 int len = strlen(map->eval_string);
2337
2338 for (ptr = call->print_fmt; *ptr; ptr++) {
2339 if (*ptr == '\\') {
2340 ptr++;
2341 /* paranoid */
2342 if (!*ptr)
2343 break;
2344 continue;
2345 }
2346 if (*ptr == '"') {
2347 quote ^= 1;
2348 continue;
2349 }
2350 if (quote)
2351 continue;
2352 if (isdigit(*ptr)) {
2353 /* skip numbers */
2354 do {
2355 ptr++;
2356 /* Check for alpha chars like ULL */
2357 } while (isalnum(*ptr));
2358 if (!*ptr)
2359 break;
2360 /*
2361 * A number must have some kind of delimiter after
2362 * it, and we can ignore that too.
2363 */
2364 continue;
2365 }
2366 if (isalpha(*ptr) || *ptr == '_') {
2367 if (strncmp(map->eval_string, ptr, len) == 0 &&
2368 !isalnum(ptr[len]) && ptr[len] != '_') {
2369 ptr = eval_replace(ptr, map, len);
2370 /* enum/sizeof string smaller than value */
2371 if (WARN_ON_ONCE(!ptr))
2372 return;
2373 /*
2374 * No need to decrement here, as eval_replace()
2375 * returns the pointer to the character passed
2376 * the eval, and two evals can not be placed
2377 * back to back without something in between.
2378 * We can skip that something in between.
2379 */
2380 continue;
2381 }
2382 skip_more:
2383 do {
2384 ptr++;
2385 } while (isalnum(*ptr) || *ptr == '_');
2386 if (!*ptr)
2387 break;
2388 /*
2389 * If what comes after this variable is a '.' or
2390 * '->' then we can continue to ignore that string.
2391 */
2392 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2393 ptr += *ptr == '.' ? 1 : 2;
2394 if (!*ptr)
2395 break;
2396 goto skip_more;
2397 }
2398 /*
2399 * Once again, we can skip the delimiter that came
2400 * after the string.
2401 */
2402 continue;
2403 }
2404 }
2405}
2406
2407void trace_event_eval_update(struct trace_eval_map **map, int len)
2408{
2409 struct trace_event_call *call, *p;
2410 const char *last_system = NULL;
2411 bool first = false;
2412 int last_i;
2413 int i;
2414
2415 down_write(&trace_event_sem);
2416 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2417 /* events are usually grouped together with systems */
2418 if (!last_system || call->class->system != last_system) {
2419 first = true;
2420 last_i = 0;
2421 last_system = call->class->system;
2422 }
2423
2424 /*
2425 * Since calls are grouped by systems, the likelyhood that the
2426 * next call in the iteration belongs to the same system as the
2427 * previous call is high. As an optimization, we skip seaching
2428 * for a map[] that matches the call's system if the last call
2429 * was from the same system. That's what last_i is for. If the
2430 * call has the same system as the previous call, then last_i
2431 * will be the index of the first map[] that has a matching
2432 * system.
2433 */
2434 for (i = last_i; i < len; i++) {
2435 if (call->class->system == map[i]->system) {
2436 /* Save the first system if need be */
2437 if (first) {
2438 last_i = i;
2439 first = false;
2440 }
2441 update_event_printk(call, map[i]);
2442 }
2443 }
2444 }
2445 up_write(&trace_event_sem);
2446}
2447
2448static struct trace_event_file *
2449trace_create_new_event(struct trace_event_call *call,
2450 struct trace_array *tr)
2451{
2452 struct trace_event_file *file;
2453
2454 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2455 if (!file)
2456 return NULL;
2457
2458 file->event_call = call;
2459 file->tr = tr;
2460 atomic_set(&file->sm_ref, 0);
2461 atomic_set(&file->tm_ref, 0);
2462 INIT_LIST_HEAD(&file->triggers);
2463 list_add(&file->list, &tr->events);
2464
2465 return file;
2466}
2467
2468/* Add an event to a trace directory */
2469static int
2470__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2471{
2472 struct trace_event_file *file;
2473
2474 file = trace_create_new_event(call, tr);
2475 if (!file)
2476 return -ENOMEM;
2477
2478 return event_create_dir(tr->event_dir, file);
2479}
2480
2481/*
2482 * Just create a decriptor for early init. A descriptor is required
2483 * for enabling events at boot. We want to enable events before
2484 * the filesystem is initialized.
2485 */
2486static __init int
2487__trace_early_add_new_event(struct trace_event_call *call,
2488 struct trace_array *tr)
2489{
2490 struct trace_event_file *file;
2491
2492 file = trace_create_new_event(call, tr);
2493 if (!file)
2494 return -ENOMEM;
2495
2496 return 0;
2497}
2498
2499struct ftrace_module_file_ops;
2500static void __add_event_to_tracers(struct trace_event_call *call);
2501
2502/* Add an additional event_call dynamically */
2503int trace_add_event_call(struct trace_event_call *call)
2504{
2505 int ret;
2506 lockdep_assert_held(&event_mutex);
2507
2508 mutex_lock(&trace_types_lock);
2509
2510 ret = __register_event(call, NULL);
2511 if (ret >= 0)
2512 __add_event_to_tracers(call);
2513
2514 mutex_unlock(&trace_types_lock);
2515 return ret;
2516}
2517
2518/*
2519 * Must be called under locking of trace_types_lock, event_mutex and
2520 * trace_event_sem.
2521 */
2522static void __trace_remove_event_call(struct trace_event_call *call)
2523{
2524 event_remove(call);
2525 trace_destroy_fields(call);
2526 free_event_filter(call->filter);
2527 call->filter = NULL;
2528}
2529
2530static int probe_remove_event_call(struct trace_event_call *call)
2531{
2532 struct trace_array *tr;
2533 struct trace_event_file *file;
2534
2535#ifdef CONFIG_PERF_EVENTS
2536 if (call->perf_refcount)
2537 return -EBUSY;
2538#endif
2539 do_for_each_event_file(tr, file) {
2540 if (file->event_call != call)
2541 continue;
2542 /*
2543 * We can't rely on ftrace_event_enable_disable(enable => 0)
2544 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2545 * TRACE_REG_UNREGISTER.
2546 */
2547 if (file->flags & EVENT_FILE_FL_ENABLED)
2548 return -EBUSY;
2549 /*
2550 * The do_for_each_event_file_safe() is
2551 * a double loop. After finding the call for this
2552 * trace_array, we use break to jump to the next
2553 * trace_array.
2554 */
2555 break;
2556 } while_for_each_event_file();
2557
2558 __trace_remove_event_call(call);
2559
2560 return 0;
2561}
2562
2563/* Remove an event_call */
2564int trace_remove_event_call(struct trace_event_call *call)
2565{
2566 int ret;
2567
2568 lockdep_assert_held(&event_mutex);
2569
2570 mutex_lock(&trace_types_lock);
2571 down_write(&trace_event_sem);
2572 ret = probe_remove_event_call(call);
2573 up_write(&trace_event_sem);
2574 mutex_unlock(&trace_types_lock);
2575
2576 return ret;
2577}
2578
2579#define for_each_event(event, start, end) \
2580 for (event = start; \
2581 (unsigned long)event < (unsigned long)end; \
2582 event++)
2583
2584#ifdef CONFIG_MODULES
2585
2586static void trace_module_add_events(struct module *mod)
2587{
2588 struct trace_event_call **call, **start, **end;
2589
2590 if (!mod->num_trace_events)
2591 return;
2592
2593 /* Don't add infrastructure for mods without tracepoints */
2594 if (trace_module_has_bad_taint(mod)) {
2595 pr_err("%s: module has bad taint, not creating trace events\n",
2596 mod->name);
2597 return;
2598 }
2599
2600 start = mod->trace_events;
2601 end = mod->trace_events + mod->num_trace_events;
2602
2603 for_each_event(call, start, end) {
2604 __register_event(*call, mod);
2605 __add_event_to_tracers(*call);
2606 }
2607}
2608
2609static void trace_module_remove_events(struct module *mod)
2610{
2611 struct trace_event_call *call, *p;
2612
2613 down_write(&trace_event_sem);
2614 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2615 if (call->mod == mod)
2616 __trace_remove_event_call(call);
2617 }
2618 up_write(&trace_event_sem);
2619
2620 /*
2621 * It is safest to reset the ring buffer if the module being unloaded
2622 * registered any events that were used. The only worry is if
2623 * a new module gets loaded, and takes on the same id as the events
2624 * of this module. When printing out the buffer, traced events left
2625 * over from this module may be passed to the new module events and
2626 * unexpected results may occur.
2627 */
2628 tracing_reset_all_online_cpus();
2629}
2630
2631static int trace_module_notify(struct notifier_block *self,
2632 unsigned long val, void *data)
2633{
2634 struct module *mod = data;
2635
2636 mutex_lock(&event_mutex);
2637 mutex_lock(&trace_types_lock);
2638 switch (val) {
2639 case MODULE_STATE_COMING:
2640 trace_module_add_events(mod);
2641 break;
2642 case MODULE_STATE_GOING:
2643 trace_module_remove_events(mod);
2644 break;
2645 }
2646 mutex_unlock(&trace_types_lock);
2647 mutex_unlock(&event_mutex);
2648
2649 return 0;
2650}
2651
2652static struct notifier_block trace_module_nb = {
2653 .notifier_call = trace_module_notify,
2654 .priority = 1, /* higher than trace.c module notify */
2655};
2656#endif /* CONFIG_MODULES */
2657
2658/* Create a new event directory structure for a trace directory. */
2659static void
2660__trace_add_event_dirs(struct trace_array *tr)
2661{
2662 struct trace_event_call *call;
2663 int ret;
2664
2665 list_for_each_entry(call, &ftrace_events, list) {
2666 ret = __trace_add_new_event(call, tr);
2667 if (ret < 0)
2668 pr_warn("Could not create directory for event %s\n",
2669 trace_event_name(call));
2670 }
2671}
2672
2673/* Returns any file that matches the system and event */
2674struct trace_event_file *
2675__find_event_file(struct trace_array *tr, const char *system, const char *event)
2676{
2677 struct trace_event_file *file;
2678 struct trace_event_call *call;
2679 const char *name;
2680
2681 list_for_each_entry(file, &tr->events, list) {
2682
2683 call = file->event_call;
2684 name = trace_event_name(call);
2685
2686 if (!name || !call->class)
2687 continue;
2688
2689 if (strcmp(event, name) == 0 &&
2690 strcmp(system, call->class->system) == 0)
2691 return file;
2692 }
2693 return NULL;
2694}
2695
2696/* Returns valid trace event files that match system and event */
2697struct trace_event_file *
2698find_event_file(struct trace_array *tr, const char *system, const char *event)
2699{
2700 struct trace_event_file *file;
2701
2702 file = __find_event_file(tr, system, event);
2703 if (!file || !file->event_call->class->reg ||
2704 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2705 return NULL;
2706
2707 return file;
2708}
2709
2710/**
2711 * trace_get_event_file - Find and return a trace event file
2712 * @instance: The name of the trace instance containing the event
2713 * @system: The name of the system containing the event
2714 * @event: The name of the event
2715 *
2716 * Return a trace event file given the trace instance name, trace
2717 * system, and trace event name. If the instance name is NULL, it
2718 * refers to the top-level trace array.
2719 *
2720 * This function will look it up and return it if found, after calling
2721 * trace_array_get() to prevent the instance from going away, and
2722 * increment the event's module refcount to prevent it from being
2723 * removed.
2724 *
2725 * To release the file, call trace_put_event_file(), which will call
2726 * trace_array_put() and decrement the event's module refcount.
2727 *
2728 * Return: The trace event on success, ERR_PTR otherwise.
2729 */
2730struct trace_event_file *trace_get_event_file(const char *instance,
2731 const char *system,
2732 const char *event)
2733{
2734 struct trace_array *tr = top_trace_array();
2735 struct trace_event_file *file = NULL;
2736 int ret = -EINVAL;
2737
2738 if (instance) {
2739 tr = trace_array_find_get(instance);
2740 if (!tr)
2741 return ERR_PTR(-ENOENT);
2742 } else {
2743 ret = trace_array_get(tr);
2744 if (ret)
2745 return ERR_PTR(ret);
2746 }
2747
2748 mutex_lock(&event_mutex);
2749
2750 file = find_event_file(tr, system, event);
2751 if (!file) {
2752 trace_array_put(tr);
2753 ret = -EINVAL;
2754 goto out;
2755 }
2756
2757 /* Don't let event modules unload while in use */
2758 ret = try_module_get(file->event_call->mod);
2759 if (!ret) {
2760 trace_array_put(tr);
2761 ret = -EBUSY;
2762 goto out;
2763 }
2764
2765 ret = 0;
2766 out:
2767 mutex_unlock(&event_mutex);
2768
2769 if (ret)
2770 file = ERR_PTR(ret);
2771
2772 return file;
2773}
2774EXPORT_SYMBOL_GPL(trace_get_event_file);
2775
2776/**
2777 * trace_put_event_file - Release a file from trace_get_event_file()
2778 * @file: The trace event file
2779 *
2780 * If a file was retrieved using trace_get_event_file(), this should
2781 * be called when it's no longer needed. It will cancel the previous
2782 * trace_array_get() called by that function, and decrement the
2783 * event's module refcount.
2784 */
2785void trace_put_event_file(struct trace_event_file *file)
2786{
2787 mutex_lock(&event_mutex);
2788 module_put(file->event_call->mod);
2789 mutex_unlock(&event_mutex);
2790
2791 trace_array_put(file->tr);
2792}
2793EXPORT_SYMBOL_GPL(trace_put_event_file);
2794
2795#ifdef CONFIG_DYNAMIC_FTRACE
2796
2797/* Avoid typos */
2798#define ENABLE_EVENT_STR "enable_event"
2799#define DISABLE_EVENT_STR "disable_event"
2800
2801struct event_probe_data {
2802 struct trace_event_file *file;
2803 unsigned long count;
2804 int ref;
2805 bool enable;
2806};
2807
2808static void update_event_probe(struct event_probe_data *data)
2809{
2810 if (data->enable)
2811 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2812 else
2813 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2814}
2815
2816static void
2817event_enable_probe(unsigned long ip, unsigned long parent_ip,
2818 struct trace_array *tr, struct ftrace_probe_ops *ops,
2819 void *data)
2820{
2821 struct ftrace_func_mapper *mapper = data;
2822 struct event_probe_data *edata;
2823 void **pdata;
2824
2825 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2826 if (!pdata || !*pdata)
2827 return;
2828
2829 edata = *pdata;
2830 update_event_probe(edata);
2831}
2832
2833static void
2834event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2835 struct trace_array *tr, struct ftrace_probe_ops *ops,
2836 void *data)
2837{
2838 struct ftrace_func_mapper *mapper = data;
2839 struct event_probe_data *edata;
2840 void **pdata;
2841
2842 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2843 if (!pdata || !*pdata)
2844 return;
2845
2846 edata = *pdata;
2847
2848 if (!edata->count)
2849 return;
2850
2851 /* Skip if the event is in a state we want to switch to */
2852 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2853 return;
2854
2855 if (edata->count != -1)
2856 (edata->count)--;
2857
2858 update_event_probe(edata);
2859}
2860
2861static int
2862event_enable_print(struct seq_file *m, unsigned long ip,
2863 struct ftrace_probe_ops *ops, void *data)
2864{
2865 struct ftrace_func_mapper *mapper = data;
2866 struct event_probe_data *edata;
2867 void **pdata;
2868
2869 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2870
2871 if (WARN_ON_ONCE(!pdata || !*pdata))
2872 return 0;
2873
2874 edata = *pdata;
2875
2876 seq_printf(m, "%ps:", (void *)ip);
2877
2878 seq_printf(m, "%s:%s:%s",
2879 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2880 edata->file->event_call->class->system,
2881 trace_event_name(edata->file->event_call));
2882
2883 if (edata->count == -1)
2884 seq_puts(m, ":unlimited\n");
2885 else
2886 seq_printf(m, ":count=%ld\n", edata->count);
2887
2888 return 0;
2889}
2890
2891static int
2892event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2893 unsigned long ip, void *init_data, void **data)
2894{
2895 struct ftrace_func_mapper *mapper = *data;
2896 struct event_probe_data *edata = init_data;
2897 int ret;
2898
2899 if (!mapper) {
2900 mapper = allocate_ftrace_func_mapper();
2901 if (!mapper)
2902 return -ENODEV;
2903 *data = mapper;
2904 }
2905
2906 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2907 if (ret < 0)
2908 return ret;
2909
2910 edata->ref++;
2911
2912 return 0;
2913}
2914
2915static int free_probe_data(void *data)
2916{
2917 struct event_probe_data *edata = data;
2918
2919 edata->ref--;
2920 if (!edata->ref) {
2921 /* Remove the SOFT_MODE flag */
2922 __ftrace_event_enable_disable(edata->file, 0, 1);
2923 module_put(edata->file->event_call->mod);
2924 kfree(edata);
2925 }
2926 return 0;
2927}
2928
2929static void
2930event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2931 unsigned long ip, void *data)
2932{
2933 struct ftrace_func_mapper *mapper = data;
2934 struct event_probe_data *edata;
2935
2936 if (!ip) {
2937 if (!mapper)
2938 return;
2939 free_ftrace_func_mapper(mapper, free_probe_data);
2940 return;
2941 }
2942
2943 edata = ftrace_func_mapper_remove_ip(mapper, ip);
2944
2945 if (WARN_ON_ONCE(!edata))
2946 return;
2947
2948 if (WARN_ON_ONCE(edata->ref <= 0))
2949 return;
2950
2951 free_probe_data(edata);
2952}
2953
2954static struct ftrace_probe_ops event_enable_probe_ops = {
2955 .func = event_enable_probe,
2956 .print = event_enable_print,
2957 .init = event_enable_init,
2958 .free = event_enable_free,
2959};
2960
2961static struct ftrace_probe_ops event_enable_count_probe_ops = {
2962 .func = event_enable_count_probe,
2963 .print = event_enable_print,
2964 .init = event_enable_init,
2965 .free = event_enable_free,
2966};
2967
2968static struct ftrace_probe_ops event_disable_probe_ops = {
2969 .func = event_enable_probe,
2970 .print = event_enable_print,
2971 .init = event_enable_init,
2972 .free = event_enable_free,
2973};
2974
2975static struct ftrace_probe_ops event_disable_count_probe_ops = {
2976 .func = event_enable_count_probe,
2977 .print = event_enable_print,
2978 .init = event_enable_init,
2979 .free = event_enable_free,
2980};
2981
2982static int
2983event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2984 char *glob, char *cmd, char *param, int enabled)
2985{
2986 struct trace_event_file *file;
2987 struct ftrace_probe_ops *ops;
2988 struct event_probe_data *data;
2989 const char *system;
2990 const char *event;
2991 char *number;
2992 bool enable;
2993 int ret;
2994
2995 if (!tr)
2996 return -ENODEV;
2997
2998 /* hash funcs only work with set_ftrace_filter */
2999 if (!enabled || !param)
3000 return -EINVAL;
3001
3002 system = strsep(¶m, ":");
3003 if (!param)
3004 return -EINVAL;
3005
3006 event = strsep(¶m, ":");
3007
3008 mutex_lock(&event_mutex);
3009
3010 ret = -EINVAL;
3011 file = find_event_file(tr, system, event);
3012 if (!file)
3013 goto out;
3014
3015 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3016
3017 if (enable)
3018 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3019 else
3020 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3021
3022 if (glob[0] == '!') {
3023 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3024 goto out;
3025 }
3026
3027 ret = -ENOMEM;
3028
3029 data = kzalloc(sizeof(*data), GFP_KERNEL);
3030 if (!data)
3031 goto out;
3032
3033 data->enable = enable;
3034 data->count = -1;
3035 data->file = file;
3036
3037 if (!param)
3038 goto out_reg;
3039
3040 number = strsep(¶m, ":");
3041
3042 ret = -EINVAL;
3043 if (!strlen(number))
3044 goto out_free;
3045
3046 /*
3047 * We use the callback data field (which is a pointer)
3048 * as our counter.
3049 */
3050 ret = kstrtoul(number, 0, &data->count);
3051 if (ret)
3052 goto out_free;
3053
3054 out_reg:
3055 /* Don't let event modules unload while probe registered */
3056 ret = try_module_get(file->event_call->mod);
3057 if (!ret) {
3058 ret = -EBUSY;
3059 goto out_free;
3060 }
3061
3062 ret = __ftrace_event_enable_disable(file, 1, 1);
3063 if (ret < 0)
3064 goto out_put;
3065
3066 ret = register_ftrace_function_probe(glob, tr, ops, data);
3067 /*
3068 * The above returns on success the # of functions enabled,
3069 * but if it didn't find any functions it returns zero.
3070 * Consider no functions a failure too.
3071 */
3072 if (!ret) {
3073 ret = -ENOENT;
3074 goto out_disable;
3075 } else if (ret < 0)
3076 goto out_disable;
3077 /* Just return zero, not the number of enabled functions */
3078 ret = 0;
3079 out:
3080 mutex_unlock(&event_mutex);
3081 return ret;
3082
3083 out_disable:
3084 __ftrace_event_enable_disable(file, 0, 1);
3085 out_put:
3086 module_put(file->event_call->mod);
3087 out_free:
3088 kfree(data);
3089 goto out;
3090}
3091
3092static struct ftrace_func_command event_enable_cmd = {
3093 .name = ENABLE_EVENT_STR,
3094 .func = event_enable_func,
3095};
3096
3097static struct ftrace_func_command event_disable_cmd = {
3098 .name = DISABLE_EVENT_STR,
3099 .func = event_enable_func,
3100};
3101
3102static __init int register_event_cmds(void)
3103{
3104 int ret;
3105
3106 ret = register_ftrace_command(&event_enable_cmd);
3107 if (WARN_ON(ret < 0))
3108 return ret;
3109 ret = register_ftrace_command(&event_disable_cmd);
3110 if (WARN_ON(ret < 0))
3111 unregister_ftrace_command(&event_enable_cmd);
3112 return ret;
3113}
3114#else
3115static inline int register_event_cmds(void) { return 0; }
3116#endif /* CONFIG_DYNAMIC_FTRACE */
3117
3118/*
3119 * The top level array has already had its trace_event_file
3120 * descriptors created in order to allow for early events to
3121 * be recorded. This function is called after the tracefs has been
3122 * initialized, and we now have to create the files associated
3123 * to the events.
3124 */
3125static __init void
3126__trace_early_add_event_dirs(struct trace_array *tr)
3127{
3128 struct trace_event_file *file;
3129 int ret;
3130
3131
3132 list_for_each_entry(file, &tr->events, list) {
3133 ret = event_create_dir(tr->event_dir, file);
3134 if (ret < 0)
3135 pr_warn("Could not create directory for event %s\n",
3136 trace_event_name(file->event_call));
3137 }
3138}
3139
3140/*
3141 * For early boot up, the top trace array requires to have
3142 * a list of events that can be enabled. This must be done before
3143 * the filesystem is set up in order to allow events to be traced
3144 * early.
3145 */
3146static __init void
3147__trace_early_add_events(struct trace_array *tr)
3148{
3149 struct trace_event_call *call;
3150 int ret;
3151
3152 list_for_each_entry(call, &ftrace_events, list) {
3153 /* Early boot up should not have any modules loaded */
3154 if (WARN_ON_ONCE(call->mod))
3155 continue;
3156
3157 ret = __trace_early_add_new_event(call, tr);
3158 if (ret < 0)
3159 pr_warn("Could not create early event %s\n",
3160 trace_event_name(call));
3161 }
3162}
3163
3164/* Remove the event directory structure for a trace directory. */
3165static void
3166__trace_remove_event_dirs(struct trace_array *tr)
3167{
3168 struct trace_event_file *file, *next;
3169
3170 list_for_each_entry_safe(file, next, &tr->events, list)
3171 remove_event_file_dir(file);
3172}
3173
3174static void __add_event_to_tracers(struct trace_event_call *call)
3175{
3176 struct trace_array *tr;
3177
3178 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3179 __trace_add_new_event(call, tr);
3180}
3181
3182extern struct trace_event_call *__start_ftrace_events[];
3183extern struct trace_event_call *__stop_ftrace_events[];
3184
3185static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3186
3187static __init int setup_trace_event(char *str)
3188{
3189 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3190 ring_buffer_expanded = true;
3191 tracing_selftest_disabled = true;
3192
3193 return 1;
3194}
3195__setup("trace_event=", setup_trace_event);
3196
3197/* Expects to have event_mutex held when called */
3198static int
3199create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3200{
3201 struct dentry *d_events;
3202 struct dentry *entry;
3203
3204 entry = tracefs_create_file("set_event", 0644, parent,
3205 tr, &ftrace_set_event_fops);
3206 if (!entry) {
3207 pr_warn("Could not create tracefs 'set_event' entry\n");
3208 return -ENOMEM;
3209 }
3210
3211 d_events = tracefs_create_dir("events", parent);
3212 if (!d_events) {
3213 pr_warn("Could not create tracefs 'events' directory\n");
3214 return -ENOMEM;
3215 }
3216
3217 entry = trace_create_file("enable", 0644, d_events,
3218 tr, &ftrace_tr_enable_fops);
3219 if (!entry) {
3220 pr_warn("Could not create tracefs 'enable' entry\n");
3221 return -ENOMEM;
3222 }
3223
3224 /* There are not as crucial, just warn if they are not created */
3225
3226 entry = tracefs_create_file("set_event_pid", 0644, parent,
3227 tr, &ftrace_set_event_pid_fops);
3228 if (!entry)
3229 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3230
3231 entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3232 tr, &ftrace_set_event_notrace_pid_fops);
3233 if (!entry)
3234 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3235
3236 /* ring buffer internal formats */
3237 entry = trace_create_file("header_page", 0444, d_events,
3238 ring_buffer_print_page_header,
3239 &ftrace_show_header_fops);
3240 if (!entry)
3241 pr_warn("Could not create tracefs 'header_page' entry\n");
3242
3243 entry = trace_create_file("header_event", 0444, d_events,
3244 ring_buffer_print_entry_header,
3245 &ftrace_show_header_fops);
3246 if (!entry)
3247 pr_warn("Could not create tracefs 'header_event' entry\n");
3248
3249 tr->event_dir = d_events;
3250
3251 return 0;
3252}
3253
3254/**
3255 * event_trace_add_tracer - add a instance of a trace_array to events
3256 * @parent: The parent dentry to place the files/directories for events in
3257 * @tr: The trace array associated with these events
3258 *
3259 * When a new instance is created, it needs to set up its events
3260 * directory, as well as other files associated with events. It also
3261 * creates the event hierachry in the @parent/events directory.
3262 *
3263 * Returns 0 on success.
3264 *
3265 * Must be called with event_mutex held.
3266 */
3267int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3268{
3269 int ret;
3270
3271 lockdep_assert_held(&event_mutex);
3272
3273 ret = create_event_toplevel_files(parent, tr);
3274 if (ret)
3275 goto out;
3276
3277 down_write(&trace_event_sem);
3278 __trace_add_event_dirs(tr);
3279 up_write(&trace_event_sem);
3280
3281 out:
3282 return ret;
3283}
3284
3285/*
3286 * The top trace array already had its file descriptors created.
3287 * Now the files themselves need to be created.
3288 */
3289static __init int
3290early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3291{
3292 int ret;
3293
3294 mutex_lock(&event_mutex);
3295
3296 ret = create_event_toplevel_files(parent, tr);
3297 if (ret)
3298 goto out_unlock;
3299
3300 down_write(&trace_event_sem);
3301 __trace_early_add_event_dirs(tr);
3302 up_write(&trace_event_sem);
3303
3304 out_unlock:
3305 mutex_unlock(&event_mutex);
3306
3307 return ret;
3308}
3309
3310/* Must be called with event_mutex held */
3311int event_trace_del_tracer(struct trace_array *tr)
3312{
3313 lockdep_assert_held(&event_mutex);
3314
3315 /* Disable any event triggers and associated soft-disabled events */
3316 clear_event_triggers(tr);
3317
3318 /* Clear the pid list */
3319 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3320
3321 /* Disable any running events */
3322 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3323
3324 /* Make sure no more events are being executed */
3325 tracepoint_synchronize_unregister();
3326
3327 down_write(&trace_event_sem);
3328 __trace_remove_event_dirs(tr);
3329 tracefs_remove(tr->event_dir);
3330 up_write(&trace_event_sem);
3331
3332 tr->event_dir = NULL;
3333
3334 return 0;
3335}
3336
3337static __init int event_trace_memsetup(void)
3338{
3339 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3340 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3341 return 0;
3342}
3343
3344static __init void
3345early_enable_events(struct trace_array *tr, bool disable_first)
3346{
3347 char *buf = bootup_event_buf;
3348 char *token;
3349 int ret;
3350
3351 while (true) {
3352 token = strsep(&buf, ",");
3353
3354 if (!token)
3355 break;
3356
3357 if (*token) {
3358 /* Restarting syscalls requires that we stop them first */
3359 if (disable_first)
3360 ftrace_set_clr_event(tr, token, 0);
3361
3362 ret = ftrace_set_clr_event(tr, token, 1);
3363 if (ret)
3364 pr_warn("Failed to enable trace event: %s\n", token);
3365 }
3366
3367 /* Put back the comma to allow this to be called again */
3368 if (buf)
3369 *(buf - 1) = ',';
3370 }
3371}
3372
3373static __init int event_trace_enable(void)
3374{
3375 struct trace_array *tr = top_trace_array();
3376 struct trace_event_call **iter, *call;
3377 int ret;
3378
3379 if (!tr)
3380 return -ENODEV;
3381
3382 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3383
3384 call = *iter;
3385 ret = event_init(call);
3386 if (!ret)
3387 list_add(&call->list, &ftrace_events);
3388 }
3389
3390 /*
3391 * We need the top trace array to have a working set of trace
3392 * points at early init, before the debug files and directories
3393 * are created. Create the file entries now, and attach them
3394 * to the actual file dentries later.
3395 */
3396 __trace_early_add_events(tr);
3397
3398 early_enable_events(tr, false);
3399
3400 trace_printk_start_comm();
3401
3402 register_event_cmds();
3403
3404 register_trigger_cmds();
3405
3406 return 0;
3407}
3408
3409/*
3410 * event_trace_enable() is called from trace_event_init() first to
3411 * initialize events and perhaps start any events that are on the
3412 * command line. Unfortunately, there are some events that will not
3413 * start this early, like the system call tracepoints that need
3414 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3415 * is called before pid 1 starts, and this flag is never set, making
3416 * the syscall tracepoint never get reached, but the event is enabled
3417 * regardless (and not doing anything).
3418 */
3419static __init int event_trace_enable_again(void)
3420{
3421 struct trace_array *tr;
3422
3423 tr = top_trace_array();
3424 if (!tr)
3425 return -ENODEV;
3426
3427 early_enable_events(tr, true);
3428
3429 return 0;
3430}
3431
3432early_initcall(event_trace_enable_again);
3433
3434__init int event_trace_init(void)
3435{
3436 struct trace_array *tr;
3437 struct dentry *d_tracer;
3438 struct dentry *entry;
3439 int ret;
3440
3441 tr = top_trace_array();
3442 if (!tr)
3443 return -ENODEV;
3444
3445 d_tracer = tracing_init_dentry();
3446 if (IS_ERR(d_tracer))
3447 return 0;
3448
3449 entry = tracefs_create_file("available_events", 0444, d_tracer,
3450 tr, &ftrace_avail_fops);
3451 if (!entry)
3452 pr_warn("Could not create tracefs 'available_events' entry\n");
3453
3454 if (trace_define_generic_fields())
3455 pr_warn("tracing: Failed to allocated generic fields");
3456
3457 if (trace_define_common_fields())
3458 pr_warn("tracing: Failed to allocate common fields");
3459
3460 ret = early_event_add_tracer(d_tracer, tr);
3461 if (ret)
3462 return ret;
3463
3464#ifdef CONFIG_MODULES
3465 ret = register_module_notifier(&trace_module_nb);
3466 if (ret)
3467 pr_warn("Failed to register trace events module notifier\n");
3468#endif
3469 return 0;
3470}
3471
3472void __init trace_event_init(void)
3473{
3474 event_trace_memsetup();
3475 init_ftrace_syscalls();
3476 event_trace_enable();
3477}
3478
3479#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3480
3481static DEFINE_SPINLOCK(test_spinlock);
3482static DEFINE_SPINLOCK(test_spinlock_irq);
3483static DEFINE_MUTEX(test_mutex);
3484
3485static __init void test_work(struct work_struct *dummy)
3486{
3487 spin_lock(&test_spinlock);
3488 spin_lock_irq(&test_spinlock_irq);
3489 udelay(1);
3490 spin_unlock_irq(&test_spinlock_irq);
3491 spin_unlock(&test_spinlock);
3492
3493 mutex_lock(&test_mutex);
3494 msleep(1);
3495 mutex_unlock(&test_mutex);
3496}
3497
3498static __init int event_test_thread(void *unused)
3499{
3500 void *test_malloc;
3501
3502 test_malloc = kmalloc(1234, GFP_KERNEL);
3503 if (!test_malloc)
3504 pr_info("failed to kmalloc\n");
3505
3506 schedule_on_each_cpu(test_work);
3507
3508 kfree(test_malloc);
3509
3510 set_current_state(TASK_INTERRUPTIBLE);
3511 while (!kthread_should_stop()) {
3512 schedule();
3513 set_current_state(TASK_INTERRUPTIBLE);
3514 }
3515 __set_current_state(TASK_RUNNING);
3516
3517 return 0;
3518}
3519
3520/*
3521 * Do various things that may trigger events.
3522 */
3523static __init void event_test_stuff(void)
3524{
3525 struct task_struct *test_thread;
3526
3527 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3528 msleep(1);
3529 kthread_stop(test_thread);
3530}
3531
3532/*
3533 * For every trace event defined, we will test each trace point separately,
3534 * and then by groups, and finally all trace points.
3535 */
3536static __init void event_trace_self_tests(void)
3537{
3538 struct trace_subsystem_dir *dir;
3539 struct trace_event_file *file;
3540 struct trace_event_call *call;
3541 struct event_subsystem *system;
3542 struct trace_array *tr;
3543 int ret;
3544
3545 tr = top_trace_array();
3546 if (!tr)
3547 return;
3548
3549 pr_info("Running tests on trace events:\n");
3550
3551 list_for_each_entry(file, &tr->events, list) {
3552
3553 call = file->event_call;
3554
3555 /* Only test those that have a probe */
3556 if (!call->class || !call->class->probe)
3557 continue;
3558
3559/*
3560 * Testing syscall events here is pretty useless, but
3561 * we still do it if configured. But this is time consuming.
3562 * What we really need is a user thread to perform the
3563 * syscalls as we test.
3564 */
3565#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3566 if (call->class->system &&
3567 strcmp(call->class->system, "syscalls") == 0)
3568 continue;
3569#endif
3570
3571 pr_info("Testing event %s: ", trace_event_name(call));
3572
3573 /*
3574 * If an event is already enabled, someone is using
3575 * it and the self test should not be on.
3576 */
3577 if (file->flags & EVENT_FILE_FL_ENABLED) {
3578 pr_warn("Enabled event during self test!\n");
3579 WARN_ON_ONCE(1);
3580 continue;
3581 }
3582
3583 ftrace_event_enable_disable(file, 1);
3584 event_test_stuff();
3585 ftrace_event_enable_disable(file, 0);
3586
3587 pr_cont("OK\n");
3588 }
3589
3590 /* Now test at the sub system level */
3591
3592 pr_info("Running tests on trace event systems:\n");
3593
3594 list_for_each_entry(dir, &tr->systems, list) {
3595
3596 system = dir->subsystem;
3597
3598 /* the ftrace system is special, skip it */
3599 if (strcmp(system->name, "ftrace") == 0)
3600 continue;
3601
3602 pr_info("Testing event system %s: ", system->name);
3603
3604 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3605 if (WARN_ON_ONCE(ret)) {
3606 pr_warn("error enabling system %s\n",
3607 system->name);
3608 continue;
3609 }
3610
3611 event_test_stuff();
3612
3613 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3614 if (WARN_ON_ONCE(ret)) {
3615 pr_warn("error disabling system %s\n",
3616 system->name);
3617 continue;
3618 }
3619
3620 pr_cont("OK\n");
3621 }
3622
3623 /* Test with all events enabled */
3624
3625 pr_info("Running tests on all trace events:\n");
3626 pr_info("Testing all events: ");
3627
3628 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3629 if (WARN_ON_ONCE(ret)) {
3630 pr_warn("error enabling all events\n");
3631 return;
3632 }
3633
3634 event_test_stuff();
3635
3636 /* reset sysname */
3637 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3638 if (WARN_ON_ONCE(ret)) {
3639 pr_warn("error disabling all events\n");
3640 return;
3641 }
3642
3643 pr_cont("OK\n");
3644}
3645
3646#ifdef CONFIG_FUNCTION_TRACER
3647
3648static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3649
3650static struct trace_event_file event_trace_file __initdata;
3651
3652static void __init
3653function_test_events_call(unsigned long ip, unsigned long parent_ip,
3654 struct ftrace_ops *op, struct pt_regs *pt_regs)
3655{
3656 struct trace_buffer *buffer;
3657 struct ring_buffer_event *event;
3658 struct ftrace_entry *entry;
3659 unsigned long flags;
3660 long disabled;
3661 int cpu;
3662 int pc;
3663
3664 pc = preempt_count();
3665 preempt_disable_notrace();
3666 cpu = raw_smp_processor_id();
3667 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3668
3669 if (disabled != 1)
3670 goto out;
3671
3672 local_save_flags(flags);
3673
3674 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3675 TRACE_FN, sizeof(*entry),
3676 flags, pc);
3677 if (!event)
3678 goto out;
3679 entry = ring_buffer_event_data(event);
3680 entry->ip = ip;
3681 entry->parent_ip = parent_ip;
3682
3683 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3684 entry, flags, pc);
3685 out:
3686 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3687 preempt_enable_notrace();
3688}
3689
3690static struct ftrace_ops trace_ops __initdata =
3691{
3692 .func = function_test_events_call,
3693 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3694};
3695
3696static __init void event_trace_self_test_with_function(void)
3697{
3698 int ret;
3699
3700 event_trace_file.tr = top_trace_array();
3701 if (WARN_ON(!event_trace_file.tr))
3702 return;
3703
3704 ret = register_ftrace_function(&trace_ops);
3705 if (WARN_ON(ret < 0)) {
3706 pr_info("Failed to enable function tracer for event tests\n");
3707 return;
3708 }
3709 pr_info("Running tests again, along with the function tracer\n");
3710 event_trace_self_tests();
3711 unregister_ftrace_function(&trace_ops);
3712}
3713#else
3714static __init void event_trace_self_test_with_function(void)
3715{
3716}
3717#endif
3718
3719static __init int event_trace_self_tests_init(void)
3720{
3721 if (!tracing_selftest_disabled) {
3722 event_trace_self_tests();
3723 event_trace_self_test_with_function();
3724 }
3725
3726 return 0;
3727}
3728
3729late_initcall(event_trace_self_tests_init);
3730
3731#endif