Loading...
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);
41static bool eventdir_initialized;
42
43static LIST_HEAD(module_strings);
44
45struct module_string {
46 struct list_head next;
47 struct module *module;
48 char *str;
49};
50
51#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
52
53static struct kmem_cache *field_cachep;
54static struct kmem_cache *file_cachep;
55
56static inline int system_refcount(struct event_subsystem *system)
57{
58 return system->ref_count;
59}
60
61static int system_refcount_inc(struct event_subsystem *system)
62{
63 return system->ref_count++;
64}
65
66static int system_refcount_dec(struct event_subsystem *system)
67{
68 return --system->ref_count;
69}
70
71/* Double loops, do not use break, only goto's work */
72#define do_for_each_event_file(tr, file) \
73 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
74 list_for_each_entry(file, &tr->events, list)
75
76#define do_for_each_event_file_safe(tr, file) \
77 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
78 struct trace_event_file *___n; \
79 list_for_each_entry_safe(file, ___n, &tr->events, list)
80
81#define while_for_each_event_file() \
82 }
83
84static struct ftrace_event_field *
85__find_event_field(struct list_head *head, char *name)
86{
87 struct ftrace_event_field *field;
88
89 list_for_each_entry(field, head, link) {
90 if (!strcmp(field->name, name))
91 return field;
92 }
93
94 return NULL;
95}
96
97struct ftrace_event_field *
98trace_find_event_field(struct trace_event_call *call, char *name)
99{
100 struct ftrace_event_field *field;
101 struct list_head *head;
102
103 head = trace_get_fields(call);
104 field = __find_event_field(head, name);
105 if (field)
106 return field;
107
108 field = __find_event_field(&ftrace_generic_fields, name);
109 if (field)
110 return field;
111
112 return __find_event_field(&ftrace_common_fields, name);
113}
114
115static int __trace_define_field(struct list_head *head, const char *type,
116 const char *name, int offset, int size,
117 int is_signed, int filter_type, int len)
118{
119 struct ftrace_event_field *field;
120
121 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
122 if (!field)
123 return -ENOMEM;
124
125 field->name = name;
126 field->type = type;
127
128 if (filter_type == FILTER_OTHER)
129 field->filter_type = filter_assign_type(type);
130 else
131 field->filter_type = filter_type;
132
133 field->offset = offset;
134 field->size = size;
135 field->is_signed = is_signed;
136 field->len = len;
137
138 list_add(&field->link, head);
139
140 return 0;
141}
142
143int trace_define_field(struct trace_event_call *call, const char *type,
144 const char *name, int offset, int size, int is_signed,
145 int filter_type)
146{
147 struct list_head *head;
148
149 if (WARN_ON(!call->class))
150 return 0;
151
152 head = trace_get_fields(call);
153 return __trace_define_field(head, type, name, offset, size,
154 is_signed, filter_type, 0);
155}
156EXPORT_SYMBOL_GPL(trace_define_field);
157
158static int trace_define_field_ext(struct trace_event_call *call, const char *type,
159 const char *name, int offset, int size, int is_signed,
160 int filter_type, int len)
161{
162 struct list_head *head;
163
164 if (WARN_ON(!call->class))
165 return 0;
166
167 head = trace_get_fields(call);
168 return __trace_define_field(head, type, name, offset, size,
169 is_signed, filter_type, len);
170}
171
172#define __generic_field(type, item, filter_type) \
173 ret = __trace_define_field(&ftrace_generic_fields, #type, \
174 #item, 0, 0, is_signed_type(type), \
175 filter_type, 0); \
176 if (ret) \
177 return ret;
178
179#define __common_field(type, item) \
180 ret = __trace_define_field(&ftrace_common_fields, #type, \
181 "common_" #item, \
182 offsetof(typeof(ent), item), \
183 sizeof(ent.item), \
184 is_signed_type(type), FILTER_OTHER, 0); \
185 if (ret) \
186 return ret;
187
188static int trace_define_generic_fields(void)
189{
190 int ret;
191
192 __generic_field(int, CPU, FILTER_CPU);
193 __generic_field(int, cpu, FILTER_CPU);
194 __generic_field(int, common_cpu, FILTER_CPU);
195 __generic_field(char *, COMM, FILTER_COMM);
196 __generic_field(char *, comm, FILTER_COMM);
197 __generic_field(char *, stacktrace, FILTER_STACKTRACE);
198 __generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
199
200 return ret;
201}
202
203static int trace_define_common_fields(void)
204{
205 int ret;
206 struct trace_entry ent;
207
208 __common_field(unsigned short, type);
209 __common_field(unsigned char, flags);
210 /* Holds both preempt_count and migrate_disable */
211 __common_field(unsigned char, preempt_count);
212 __common_field(int, pid);
213
214 return ret;
215}
216
217static void trace_destroy_fields(struct trace_event_call *call)
218{
219 struct ftrace_event_field *field, *next;
220 struct list_head *head;
221
222 head = trace_get_fields(call);
223 list_for_each_entry_safe(field, next, head, link) {
224 list_del(&field->link);
225 kmem_cache_free(field_cachep, field);
226 }
227}
228
229/*
230 * run-time version of trace_event_get_offsets_<call>() that returns the last
231 * accessible offset of trace fields excluding __dynamic_array bytes
232 */
233int trace_event_get_offsets(struct trace_event_call *call)
234{
235 struct ftrace_event_field *tail;
236 struct list_head *head;
237
238 head = trace_get_fields(call);
239 /*
240 * head->next points to the last field with the largest offset,
241 * since it was added last by trace_define_field()
242 */
243 tail = list_first_entry(head, struct ftrace_event_field, link);
244 return tail->offset + tail->size;
245}
246
247/*
248 * Check if the referenced field is an array and return true,
249 * as arrays are OK to dereference.
250 */
251static bool test_field(const char *fmt, struct trace_event_call *call)
252{
253 struct trace_event_fields *field = call->class->fields_array;
254 const char *array_descriptor;
255 const char *p = fmt;
256 int len;
257
258 if (!(len = str_has_prefix(fmt, "REC->")))
259 return false;
260 fmt += len;
261 for (p = fmt; *p; p++) {
262 if (!isalnum(*p) && *p != '_')
263 break;
264 }
265 len = p - fmt;
266
267 for (; field->type; field++) {
268 if (strncmp(field->name, fmt, len) ||
269 field->name[len])
270 continue;
271 array_descriptor = strchr(field->type, '[');
272 /* This is an array and is OK to dereference. */
273 return array_descriptor != NULL;
274 }
275 return false;
276}
277
278/*
279 * Examine the print fmt of the event looking for unsafe dereference
280 * pointers using %p* that could be recorded in the trace event and
281 * much later referenced after the pointer was freed. Dereferencing
282 * pointers are OK, if it is dereferenced into the event itself.
283 */
284static void test_event_printk(struct trace_event_call *call)
285{
286 u64 dereference_flags = 0;
287 bool first = true;
288 const char *fmt, *c, *r, *a;
289 int parens = 0;
290 char in_quote = 0;
291 int start_arg = 0;
292 int arg = 0;
293 int i;
294
295 fmt = call->print_fmt;
296
297 if (!fmt)
298 return;
299
300 for (i = 0; fmt[i]; i++) {
301 switch (fmt[i]) {
302 case '\\':
303 i++;
304 if (!fmt[i])
305 return;
306 continue;
307 case '"':
308 case '\'':
309 /*
310 * The print fmt starts with a string that
311 * is processed first to find %p* usage,
312 * then after the first string, the print fmt
313 * contains arguments that are used to check
314 * if the dereferenced %p* usage is safe.
315 */
316 if (first) {
317 if (fmt[i] == '\'')
318 continue;
319 if (in_quote) {
320 arg = 0;
321 first = false;
322 /*
323 * If there was no %p* uses
324 * the fmt is OK.
325 */
326 if (!dereference_flags)
327 return;
328 }
329 }
330 if (in_quote) {
331 if (in_quote == fmt[i])
332 in_quote = 0;
333 } else {
334 in_quote = fmt[i];
335 }
336 continue;
337 case '%':
338 if (!first || !in_quote)
339 continue;
340 i++;
341 if (!fmt[i])
342 return;
343 switch (fmt[i]) {
344 case '%':
345 continue;
346 case 'p':
347 /* Find dereferencing fields */
348 switch (fmt[i + 1]) {
349 case 'B': case 'R': case 'r':
350 case 'b': case 'M': case 'm':
351 case 'I': case 'i': case 'E':
352 case 'U': case 'V': case 'N':
353 case 'a': case 'd': case 'D':
354 case 'g': case 't': case 'C':
355 case 'O': case 'f':
356 if (WARN_ONCE(arg == 63,
357 "Too many args for event: %s",
358 trace_event_name(call)))
359 return;
360 dereference_flags |= 1ULL << arg;
361 }
362 break;
363 default:
364 {
365 bool star = false;
366 int j;
367
368 /* Increment arg if %*s exists. */
369 for (j = 0; fmt[i + j]; j++) {
370 if (isdigit(fmt[i + j]) ||
371 fmt[i + j] == '.')
372 continue;
373 if (fmt[i + j] == '*') {
374 star = true;
375 continue;
376 }
377 if ((fmt[i + j] == 's') && star)
378 arg++;
379 break;
380 }
381 break;
382 } /* default */
383
384 } /* switch */
385 arg++;
386 continue;
387 case '(':
388 if (in_quote)
389 continue;
390 parens++;
391 continue;
392 case ')':
393 if (in_quote)
394 continue;
395 parens--;
396 if (WARN_ONCE(parens < 0,
397 "Paren mismatch for event: %s\narg='%s'\n%*s",
398 trace_event_name(call),
399 fmt + start_arg,
400 (i - start_arg) + 5, "^"))
401 return;
402 continue;
403 case ',':
404 if (in_quote || parens)
405 continue;
406 i++;
407 while (isspace(fmt[i]))
408 i++;
409 start_arg = i;
410 if (!(dereference_flags & (1ULL << arg)))
411 goto next_arg;
412
413 /* Find the REC-> in the argument */
414 c = strchr(fmt + i, ',');
415 r = strstr(fmt + i, "REC->");
416 if (r && (!c || r < c)) {
417 /*
418 * Addresses of events on the buffer,
419 * or an array on the buffer is
420 * OK to dereference.
421 * There's ways to fool this, but
422 * this is to catch common mistakes,
423 * not malicious code.
424 */
425 a = strchr(fmt + i, '&');
426 if ((a && (a < r)) || test_field(r, call))
427 dereference_flags &= ~(1ULL << arg);
428 } else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
429 (!c || r < c)) {
430 dereference_flags &= ~(1ULL << arg);
431 } else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
432 (!c || r < c)) {
433 dereference_flags &= ~(1ULL << arg);
434 }
435
436 next_arg:
437 i--;
438 arg++;
439 }
440 }
441
442 /*
443 * If you triggered the below warning, the trace event reported
444 * uses an unsafe dereference pointer %p*. As the data stored
445 * at the trace event time may no longer exist when the trace
446 * event is printed, dereferencing to the original source is
447 * unsafe. The source of the dereference must be copied into the
448 * event itself, and the dereference must access the copy instead.
449 */
450 if (WARN_ON_ONCE(dereference_flags)) {
451 arg = 1;
452 while (!(dereference_flags & 1)) {
453 dereference_flags >>= 1;
454 arg++;
455 }
456 pr_warn("event %s has unsafe dereference of argument %d\n",
457 trace_event_name(call), arg);
458 pr_warn("print_fmt: %s\n", fmt);
459 }
460}
461
462int trace_event_raw_init(struct trace_event_call *call)
463{
464 int id;
465
466 id = register_trace_event(&call->event);
467 if (!id)
468 return -ENODEV;
469
470 test_event_printk(call);
471
472 return 0;
473}
474EXPORT_SYMBOL_GPL(trace_event_raw_init);
475
476bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
477{
478 struct trace_array *tr = trace_file->tr;
479 struct trace_array_cpu *data;
480 struct trace_pid_list *no_pid_list;
481 struct trace_pid_list *pid_list;
482
483 pid_list = rcu_dereference_raw(tr->filtered_pids);
484 no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
485
486 if (!pid_list && !no_pid_list)
487 return false;
488
489 data = this_cpu_ptr(tr->array_buffer.data);
490
491 return data->ignore_pid;
492}
493EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
494
495void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
496 struct trace_event_file *trace_file,
497 unsigned long len)
498{
499 struct trace_event_call *event_call = trace_file->event_call;
500
501 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
502 trace_event_ignore_this_pid(trace_file))
503 return NULL;
504
505 /*
506 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
507 * preemption (adding one to the preempt_count). Since we are
508 * interested in the preempt_count at the time the tracepoint was
509 * hit, we need to subtract one to offset the increment.
510 */
511 fbuffer->trace_ctx = tracing_gen_ctx_dec();
512 fbuffer->trace_file = trace_file;
513
514 fbuffer->event =
515 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
516 event_call->event.type, len,
517 fbuffer->trace_ctx);
518 if (!fbuffer->event)
519 return NULL;
520
521 fbuffer->regs = NULL;
522 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
523 return fbuffer->entry;
524}
525EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
526
527int trace_event_reg(struct trace_event_call *call,
528 enum trace_reg type, void *data)
529{
530 struct trace_event_file *file = data;
531
532 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
533 switch (type) {
534 case TRACE_REG_REGISTER:
535 return tracepoint_probe_register(call->tp,
536 call->class->probe,
537 file);
538 case TRACE_REG_UNREGISTER:
539 tracepoint_probe_unregister(call->tp,
540 call->class->probe,
541 file);
542 return 0;
543
544#ifdef CONFIG_PERF_EVENTS
545 case TRACE_REG_PERF_REGISTER:
546 return tracepoint_probe_register(call->tp,
547 call->class->perf_probe,
548 call);
549 case TRACE_REG_PERF_UNREGISTER:
550 tracepoint_probe_unregister(call->tp,
551 call->class->perf_probe,
552 call);
553 return 0;
554 case TRACE_REG_PERF_OPEN:
555 case TRACE_REG_PERF_CLOSE:
556 case TRACE_REG_PERF_ADD:
557 case TRACE_REG_PERF_DEL:
558 return 0;
559#endif
560 }
561 return 0;
562}
563EXPORT_SYMBOL_GPL(trace_event_reg);
564
565void trace_event_enable_cmd_record(bool enable)
566{
567 struct trace_event_file *file;
568 struct trace_array *tr;
569
570 lockdep_assert_held(&event_mutex);
571
572 do_for_each_event_file(tr, file) {
573
574 if (!(file->flags & EVENT_FILE_FL_ENABLED))
575 continue;
576
577 if (enable) {
578 tracing_start_cmdline_record();
579 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
580 } else {
581 tracing_stop_cmdline_record();
582 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
583 }
584 } while_for_each_event_file();
585}
586
587void trace_event_enable_tgid_record(bool enable)
588{
589 struct trace_event_file *file;
590 struct trace_array *tr;
591
592 lockdep_assert_held(&event_mutex);
593
594 do_for_each_event_file(tr, file) {
595 if (!(file->flags & EVENT_FILE_FL_ENABLED))
596 continue;
597
598 if (enable) {
599 tracing_start_tgid_record();
600 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
601 } else {
602 tracing_stop_tgid_record();
603 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
604 &file->flags);
605 }
606 } while_for_each_event_file();
607}
608
609static int __ftrace_event_enable_disable(struct trace_event_file *file,
610 int enable, int soft_disable)
611{
612 struct trace_event_call *call = file->event_call;
613 struct trace_array *tr = file->tr;
614 int ret = 0;
615 int disable;
616
617 switch (enable) {
618 case 0:
619 /*
620 * When soft_disable is set and enable is cleared, the sm_ref
621 * reference counter is decremented. If it reaches 0, we want
622 * to clear the SOFT_DISABLED flag but leave the event in the
623 * state that it was. That is, if the event was enabled and
624 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
625 * is set we do not want the event to be enabled before we
626 * clear the bit.
627 *
628 * When soft_disable is not set but the SOFT_MODE flag is,
629 * we do nothing. Do not disable the tracepoint, otherwise
630 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
631 */
632 if (soft_disable) {
633 if (atomic_dec_return(&file->sm_ref) > 0)
634 break;
635 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
636 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
637 /* Disable use of trace_buffered_event */
638 trace_buffered_event_disable();
639 } else
640 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
641
642 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
643 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
644 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
645 tracing_stop_cmdline_record();
646 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
647 }
648
649 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
650 tracing_stop_tgid_record();
651 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
652 }
653
654 call->class->reg(call, TRACE_REG_UNREGISTER, file);
655 }
656 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
657 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
658 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
659 else
660 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
661 break;
662 case 1:
663 /*
664 * When soft_disable is set and enable is set, we want to
665 * register the tracepoint for the event, but leave the event
666 * as is. That means, if the event was already enabled, we do
667 * nothing (but set SOFT_MODE). If the event is disabled, we
668 * set SOFT_DISABLED before enabling the event tracepoint, so
669 * it still seems to be disabled.
670 */
671 if (!soft_disable)
672 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
673 else {
674 if (atomic_inc_return(&file->sm_ref) > 1)
675 break;
676 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
677 /* Enable use of trace_buffered_event */
678 trace_buffered_event_enable();
679 }
680
681 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
682 bool cmd = false, tgid = false;
683
684 /* Keep the event disabled, when going to SOFT_MODE. */
685 if (soft_disable)
686 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
687
688 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
689 cmd = true;
690 tracing_start_cmdline_record();
691 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
692 }
693
694 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
695 tgid = true;
696 tracing_start_tgid_record();
697 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
698 }
699
700 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
701 if (ret) {
702 if (cmd)
703 tracing_stop_cmdline_record();
704 if (tgid)
705 tracing_stop_tgid_record();
706 pr_info("event trace: Could not enable event "
707 "%s\n", trace_event_name(call));
708 break;
709 }
710 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
711
712 /* WAS_ENABLED gets set but never cleared. */
713 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
714 }
715 break;
716 }
717
718 return ret;
719}
720
721int trace_event_enable_disable(struct trace_event_file *file,
722 int enable, int soft_disable)
723{
724 return __ftrace_event_enable_disable(file, enable, soft_disable);
725}
726
727static int ftrace_event_enable_disable(struct trace_event_file *file,
728 int enable)
729{
730 return __ftrace_event_enable_disable(file, enable, 0);
731}
732
733static void ftrace_clear_events(struct trace_array *tr)
734{
735 struct trace_event_file *file;
736
737 mutex_lock(&event_mutex);
738 list_for_each_entry(file, &tr->events, list) {
739 ftrace_event_enable_disable(file, 0);
740 }
741 mutex_unlock(&event_mutex);
742}
743
744static void
745event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
746{
747 struct trace_pid_list *pid_list;
748 struct trace_array *tr = data;
749
750 pid_list = rcu_dereference_raw(tr->filtered_pids);
751 trace_filter_add_remove_task(pid_list, NULL, task);
752
753 pid_list = rcu_dereference_raw(tr->filtered_no_pids);
754 trace_filter_add_remove_task(pid_list, NULL, task);
755}
756
757static void
758event_filter_pid_sched_process_fork(void *data,
759 struct task_struct *self,
760 struct task_struct *task)
761{
762 struct trace_pid_list *pid_list;
763 struct trace_array *tr = data;
764
765 pid_list = rcu_dereference_sched(tr->filtered_pids);
766 trace_filter_add_remove_task(pid_list, self, task);
767
768 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
769 trace_filter_add_remove_task(pid_list, self, task);
770}
771
772void trace_event_follow_fork(struct trace_array *tr, bool enable)
773{
774 if (enable) {
775 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
776 tr, INT_MIN);
777 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
778 tr, INT_MAX);
779 } else {
780 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
781 tr);
782 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
783 tr);
784 }
785}
786
787static void
788event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
789 struct task_struct *prev,
790 struct task_struct *next,
791 unsigned int prev_state)
792{
793 struct trace_array *tr = data;
794 struct trace_pid_list *no_pid_list;
795 struct trace_pid_list *pid_list;
796 bool ret;
797
798 pid_list = rcu_dereference_sched(tr->filtered_pids);
799 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
800
801 /*
802 * Sched switch is funny, as we only want to ignore it
803 * in the notrace case if both prev and next should be ignored.
804 */
805 ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
806 trace_ignore_this_task(NULL, no_pid_list, next);
807
808 this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
809 (trace_ignore_this_task(pid_list, NULL, prev) &&
810 trace_ignore_this_task(pid_list, NULL, next)));
811}
812
813static void
814event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
815 struct task_struct *prev,
816 struct task_struct *next,
817 unsigned int prev_state)
818{
819 struct trace_array *tr = data;
820 struct trace_pid_list *no_pid_list;
821 struct trace_pid_list *pid_list;
822
823 pid_list = rcu_dereference_sched(tr->filtered_pids);
824 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
825
826 this_cpu_write(tr->array_buffer.data->ignore_pid,
827 trace_ignore_this_task(pid_list, no_pid_list, next));
828}
829
830static void
831event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
832{
833 struct trace_array *tr = data;
834 struct trace_pid_list *no_pid_list;
835 struct trace_pid_list *pid_list;
836
837 /* Nothing to do if we are already tracing */
838 if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
839 return;
840
841 pid_list = rcu_dereference_sched(tr->filtered_pids);
842 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
843
844 this_cpu_write(tr->array_buffer.data->ignore_pid,
845 trace_ignore_this_task(pid_list, no_pid_list, task));
846}
847
848static void
849event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
850{
851 struct trace_array *tr = data;
852 struct trace_pid_list *no_pid_list;
853 struct trace_pid_list *pid_list;
854
855 /* Nothing to do if we are not tracing */
856 if (this_cpu_read(tr->array_buffer.data->ignore_pid))
857 return;
858
859 pid_list = rcu_dereference_sched(tr->filtered_pids);
860 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
861
862 /* Set tracing if current is enabled */
863 this_cpu_write(tr->array_buffer.data->ignore_pid,
864 trace_ignore_this_task(pid_list, no_pid_list, current));
865}
866
867static void unregister_pid_events(struct trace_array *tr)
868{
869 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
870 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
871
872 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
873 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
874
875 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
876 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
877
878 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
879 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
880}
881
882static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
883{
884 struct trace_pid_list *pid_list;
885 struct trace_pid_list *no_pid_list;
886 struct trace_event_file *file;
887 int cpu;
888
889 pid_list = rcu_dereference_protected(tr->filtered_pids,
890 lockdep_is_held(&event_mutex));
891 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
892 lockdep_is_held(&event_mutex));
893
894 /* Make sure there's something to do */
895 if (!pid_type_enabled(type, pid_list, no_pid_list))
896 return;
897
898 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
899 unregister_pid_events(tr);
900
901 list_for_each_entry(file, &tr->events, list) {
902 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
903 }
904
905 for_each_possible_cpu(cpu)
906 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
907 }
908
909 if (type & TRACE_PIDS)
910 rcu_assign_pointer(tr->filtered_pids, NULL);
911
912 if (type & TRACE_NO_PIDS)
913 rcu_assign_pointer(tr->filtered_no_pids, NULL);
914
915 /* Wait till all users are no longer using pid filtering */
916 tracepoint_synchronize_unregister();
917
918 if ((type & TRACE_PIDS) && pid_list)
919 trace_pid_list_free(pid_list);
920
921 if ((type & TRACE_NO_PIDS) && no_pid_list)
922 trace_pid_list_free(no_pid_list);
923}
924
925static void ftrace_clear_event_pids(struct trace_array *tr, int type)
926{
927 mutex_lock(&event_mutex);
928 __ftrace_clear_event_pids(tr, type);
929 mutex_unlock(&event_mutex);
930}
931
932static void __put_system(struct event_subsystem *system)
933{
934 struct event_filter *filter = system->filter;
935
936 WARN_ON_ONCE(system_refcount(system) == 0);
937 if (system_refcount_dec(system))
938 return;
939
940 list_del(&system->list);
941
942 if (filter) {
943 kfree(filter->filter_string);
944 kfree(filter);
945 }
946 kfree_const(system->name);
947 kfree(system);
948}
949
950static void __get_system(struct event_subsystem *system)
951{
952 WARN_ON_ONCE(system_refcount(system) == 0);
953 system_refcount_inc(system);
954}
955
956static void __get_system_dir(struct trace_subsystem_dir *dir)
957{
958 WARN_ON_ONCE(dir->ref_count == 0);
959 dir->ref_count++;
960 __get_system(dir->subsystem);
961}
962
963static void __put_system_dir(struct trace_subsystem_dir *dir)
964{
965 WARN_ON_ONCE(dir->ref_count == 0);
966 /* If the subsystem is about to be freed, the dir must be too */
967 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
968
969 __put_system(dir->subsystem);
970 if (!--dir->ref_count)
971 kfree(dir);
972}
973
974static void put_system(struct trace_subsystem_dir *dir)
975{
976 mutex_lock(&event_mutex);
977 __put_system_dir(dir);
978 mutex_unlock(&event_mutex);
979}
980
981static void remove_subsystem(struct trace_subsystem_dir *dir)
982{
983 if (!dir)
984 return;
985
986 if (!--dir->nr_events) {
987 eventfs_remove_dir(dir->ei);
988 list_del(&dir->list);
989 __put_system_dir(dir);
990 }
991}
992
993void event_file_get(struct trace_event_file *file)
994{
995 atomic_inc(&file->ref);
996}
997
998void event_file_put(struct trace_event_file *file)
999{
1000 if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001 if (file->flags & EVENT_FILE_FL_FREED)
1002 kmem_cache_free(file_cachep, file);
1003 return;
1004 }
1005
1006 if (atomic_dec_and_test(&file->ref)) {
1007 /* Count should only go to zero when it is freed */
1008 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009 return;
1010 kmem_cache_free(file_cachep, file);
1011 }
1012}
1013
1014static void remove_event_file_dir(struct trace_event_file *file)
1015{
1016 eventfs_remove_dir(file->ei);
1017 list_del(&file->list);
1018 remove_subsystem(file->system);
1019 free_event_filter(file->filter);
1020 file->flags |= EVENT_FILE_FL_FREED;
1021 event_file_put(file);
1022}
1023
1024/*
1025 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026 */
1027static int
1028__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029 const char *sub, const char *event, int set)
1030{
1031 struct trace_event_file *file;
1032 struct trace_event_call *call;
1033 const char *name;
1034 int ret = -EINVAL;
1035 int eret = 0;
1036
1037 list_for_each_entry(file, &tr->events, list) {
1038
1039 call = file->event_call;
1040 name = trace_event_name(call);
1041
1042 if (!name || !call->class || !call->class->reg)
1043 continue;
1044
1045 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046 continue;
1047
1048 if (match &&
1049 strcmp(match, name) != 0 &&
1050 strcmp(match, call->class->system) != 0)
1051 continue;
1052
1053 if (sub && strcmp(sub, call->class->system) != 0)
1054 continue;
1055
1056 if (event && strcmp(event, name) != 0)
1057 continue;
1058
1059 ret = ftrace_event_enable_disable(file, set);
1060
1061 /*
1062 * Save the first error and return that. Some events
1063 * may still have been enabled, but let the user
1064 * know that something went wrong.
1065 */
1066 if (ret && !eret)
1067 eret = ret;
1068
1069 ret = eret;
1070 }
1071
1072 return ret;
1073}
1074
1075static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076 const char *sub, const char *event, int set)
1077{
1078 int ret;
1079
1080 mutex_lock(&event_mutex);
1081 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082 mutex_unlock(&event_mutex);
1083
1084 return ret;
1085}
1086
1087int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088{
1089 char *event = NULL, *sub = NULL, *match;
1090 int ret;
1091
1092 if (!tr)
1093 return -ENOENT;
1094 /*
1095 * The buf format can be <subsystem>:<event-name>
1096 * *:<event-name> means any event by that name.
1097 * :<event-name> is the same.
1098 *
1099 * <subsystem>:* means all events in that subsystem
1100 * <subsystem>: means the same.
1101 *
1102 * <name> (no ':') means all events in a subsystem with
1103 * the name <name> or any event that matches <name>
1104 */
1105
1106 match = strsep(&buf, ":");
1107 if (buf) {
1108 sub = match;
1109 event = buf;
1110 match = NULL;
1111
1112 if (!strlen(sub) || strcmp(sub, "*") == 0)
1113 sub = NULL;
1114 if (!strlen(event) || strcmp(event, "*") == 0)
1115 event = NULL;
1116 }
1117
1118 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120 /* Put back the colon to allow this to be called again */
1121 if (buf)
1122 *(buf - 1) = ':';
1123
1124 return ret;
1125}
1126
1127/**
1128 * trace_set_clr_event - enable or disable an event
1129 * @system: system name to match (NULL for any system)
1130 * @event: event name to match (NULL for all events, within system)
1131 * @set: 1 to enable, 0 to disable
1132 *
1133 * This is a way for other parts of the kernel to enable or disable
1134 * event recording.
1135 *
1136 * Returns 0 on success, -EINVAL if the parameters do not match any
1137 * registered events.
1138 */
1139int trace_set_clr_event(const char *system, const char *event, int set)
1140{
1141 struct trace_array *tr = top_trace_array();
1142
1143 if (!tr)
1144 return -ENODEV;
1145
1146 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147}
1148EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150/**
1151 * trace_array_set_clr_event - enable or disable an event for a trace array.
1152 * @tr: concerned trace array.
1153 * @system: system name to match (NULL for any system)
1154 * @event: event name to match (NULL for all events, within system)
1155 * @enable: true to enable, false to disable
1156 *
1157 * This is a way for other parts of the kernel to enable or disable
1158 * event recording.
1159 *
1160 * Returns 0 on success, -EINVAL if the parameters do not match any
1161 * registered events.
1162 */
1163int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164 const char *event, bool enable)
1165{
1166 int set;
1167
1168 if (!tr)
1169 return -ENOENT;
1170
1171 set = (enable == true) ? 1 : 0;
1172 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173}
1174EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176/* 128 should be much more than enough */
1177#define EVENT_BUF_SIZE 127
1178
1179static ssize_t
1180ftrace_event_write(struct file *file, const char __user *ubuf,
1181 size_t cnt, loff_t *ppos)
1182{
1183 struct trace_parser parser;
1184 struct seq_file *m = file->private_data;
1185 struct trace_array *tr = m->private;
1186 ssize_t read, ret;
1187
1188 if (!cnt)
1189 return 0;
1190
1191 ret = tracing_update_buffers(tr);
1192 if (ret < 0)
1193 return ret;
1194
1195 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196 return -ENOMEM;
1197
1198 read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200 if (read >= 0 && trace_parser_loaded((&parser))) {
1201 int set = 1;
1202
1203 if (*parser.buffer == '!')
1204 set = 0;
1205
1206 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207 if (ret)
1208 goto out_put;
1209 }
1210
1211 ret = read;
1212
1213 out_put:
1214 trace_parser_put(&parser);
1215
1216 return ret;
1217}
1218
1219static void *
1220t_next(struct seq_file *m, void *v, loff_t *pos)
1221{
1222 struct trace_event_file *file = v;
1223 struct trace_event_call *call;
1224 struct trace_array *tr = m->private;
1225
1226 (*pos)++;
1227
1228 list_for_each_entry_continue(file, &tr->events, list) {
1229 call = file->event_call;
1230 /*
1231 * The ftrace subsystem is for showing formats only.
1232 * They can not be enabled or disabled via the event files.
1233 */
1234 if (call->class && call->class->reg &&
1235 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236 return file;
1237 }
1238
1239 return NULL;
1240}
1241
1242static void *t_start(struct seq_file *m, loff_t *pos)
1243{
1244 struct trace_event_file *file;
1245 struct trace_array *tr = m->private;
1246 loff_t l;
1247
1248 mutex_lock(&event_mutex);
1249
1250 file = list_entry(&tr->events, struct trace_event_file, list);
1251 for (l = 0; l <= *pos; ) {
1252 file = t_next(m, file, &l);
1253 if (!file)
1254 break;
1255 }
1256 return file;
1257}
1258
1259static void *
1260s_next(struct seq_file *m, void *v, loff_t *pos)
1261{
1262 struct trace_event_file *file = v;
1263 struct trace_array *tr = m->private;
1264
1265 (*pos)++;
1266
1267 list_for_each_entry_continue(file, &tr->events, list) {
1268 if (file->flags & EVENT_FILE_FL_ENABLED)
1269 return file;
1270 }
1271
1272 return NULL;
1273}
1274
1275static void *s_start(struct seq_file *m, loff_t *pos)
1276{
1277 struct trace_event_file *file;
1278 struct trace_array *tr = m->private;
1279 loff_t l;
1280
1281 mutex_lock(&event_mutex);
1282
1283 file = list_entry(&tr->events, struct trace_event_file, list);
1284 for (l = 0; l <= *pos; ) {
1285 file = s_next(m, file, &l);
1286 if (!file)
1287 break;
1288 }
1289 return file;
1290}
1291
1292static int t_show(struct seq_file *m, void *v)
1293{
1294 struct trace_event_file *file = v;
1295 struct trace_event_call *call = file->event_call;
1296
1297 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298 seq_printf(m, "%s:", call->class->system);
1299 seq_printf(m, "%s\n", trace_event_name(call));
1300
1301 return 0;
1302}
1303
1304static void t_stop(struct seq_file *m, void *p)
1305{
1306 mutex_unlock(&event_mutex);
1307}
1308
1309static void *
1310__next(struct seq_file *m, void *v, loff_t *pos, int type)
1311{
1312 struct trace_array *tr = m->private;
1313 struct trace_pid_list *pid_list;
1314
1315 if (type == TRACE_PIDS)
1316 pid_list = rcu_dereference_sched(tr->filtered_pids);
1317 else
1318 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320 return trace_pid_next(pid_list, v, pos);
1321}
1322
1323static void *
1324p_next(struct seq_file *m, void *v, loff_t *pos)
1325{
1326 return __next(m, v, pos, TRACE_PIDS);
1327}
1328
1329static void *
1330np_next(struct seq_file *m, void *v, loff_t *pos)
1331{
1332 return __next(m, v, pos, TRACE_NO_PIDS);
1333}
1334
1335static void *__start(struct seq_file *m, loff_t *pos, int type)
1336 __acquires(RCU)
1337{
1338 struct trace_pid_list *pid_list;
1339 struct trace_array *tr = m->private;
1340
1341 /*
1342 * Grab the mutex, to keep calls to p_next() having the same
1343 * tr->filtered_pids as p_start() has.
1344 * If we just passed the tr->filtered_pids around, then RCU would
1345 * have been enough, but doing that makes things more complex.
1346 */
1347 mutex_lock(&event_mutex);
1348 rcu_read_lock_sched();
1349
1350 if (type == TRACE_PIDS)
1351 pid_list = rcu_dereference_sched(tr->filtered_pids);
1352 else
1353 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355 if (!pid_list)
1356 return NULL;
1357
1358 return trace_pid_start(pid_list, pos);
1359}
1360
1361static void *p_start(struct seq_file *m, loff_t *pos)
1362 __acquires(RCU)
1363{
1364 return __start(m, pos, TRACE_PIDS);
1365}
1366
1367static void *np_start(struct seq_file *m, loff_t *pos)
1368 __acquires(RCU)
1369{
1370 return __start(m, pos, TRACE_NO_PIDS);
1371}
1372
1373static void p_stop(struct seq_file *m, void *p)
1374 __releases(RCU)
1375{
1376 rcu_read_unlock_sched();
1377 mutex_unlock(&event_mutex);
1378}
1379
1380static ssize_t
1381event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382 loff_t *ppos)
1383{
1384 struct trace_event_file *file;
1385 unsigned long flags;
1386 char buf[4] = "0";
1387
1388 mutex_lock(&event_mutex);
1389 file = event_file_data(filp);
1390 if (likely(file))
1391 flags = file->flags;
1392 mutex_unlock(&event_mutex);
1393
1394 if (!file || flags & EVENT_FILE_FL_FREED)
1395 return -ENODEV;
1396
1397 if (flags & EVENT_FILE_FL_ENABLED &&
1398 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399 strcpy(buf, "1");
1400
1401 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402 flags & EVENT_FILE_FL_SOFT_MODE)
1403 strcat(buf, "*");
1404
1405 strcat(buf, "\n");
1406
1407 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408}
1409
1410static ssize_t
1411event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412 loff_t *ppos)
1413{
1414 struct trace_event_file *file;
1415 unsigned long val;
1416 int ret;
1417
1418 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419 if (ret)
1420 return ret;
1421
1422 switch (val) {
1423 case 0:
1424 case 1:
1425 ret = -ENODEV;
1426 mutex_lock(&event_mutex);
1427 file = event_file_data(filp);
1428 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429 ret = tracing_update_buffers(file->tr);
1430 if (ret < 0) {
1431 mutex_unlock(&event_mutex);
1432 return ret;
1433 }
1434 ret = ftrace_event_enable_disable(file, val);
1435 }
1436 mutex_unlock(&event_mutex);
1437 break;
1438
1439 default:
1440 return -EINVAL;
1441 }
1442
1443 *ppos += cnt;
1444
1445 return ret ? ret : cnt;
1446}
1447
1448static ssize_t
1449system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450 loff_t *ppos)
1451{
1452 const char set_to_char[4] = { '?', '0', '1', 'X' };
1453 struct trace_subsystem_dir *dir = filp->private_data;
1454 struct event_subsystem *system = dir->subsystem;
1455 struct trace_event_call *call;
1456 struct trace_event_file *file;
1457 struct trace_array *tr = dir->tr;
1458 char buf[2];
1459 int set = 0;
1460 int ret;
1461
1462 mutex_lock(&event_mutex);
1463 list_for_each_entry(file, &tr->events, list) {
1464 call = file->event_call;
1465 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466 !trace_event_name(call) || !call->class || !call->class->reg)
1467 continue;
1468
1469 if (system && strcmp(call->class->system, system->name) != 0)
1470 continue;
1471
1472 /*
1473 * We need to find out if all the events are set
1474 * or if all events or cleared, or if we have
1475 * a mixture.
1476 */
1477 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479 /*
1480 * If we have a mixture, no need to look further.
1481 */
1482 if (set == 3)
1483 break;
1484 }
1485 mutex_unlock(&event_mutex);
1486
1487 buf[0] = set_to_char[set];
1488 buf[1] = '\n';
1489
1490 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492 return ret;
1493}
1494
1495static ssize_t
1496system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497 loff_t *ppos)
1498{
1499 struct trace_subsystem_dir *dir = filp->private_data;
1500 struct event_subsystem *system = dir->subsystem;
1501 const char *name = NULL;
1502 unsigned long val;
1503 ssize_t ret;
1504
1505 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506 if (ret)
1507 return ret;
1508
1509 ret = tracing_update_buffers(dir->tr);
1510 if (ret < 0)
1511 return ret;
1512
1513 if (val != 0 && val != 1)
1514 return -EINVAL;
1515
1516 /*
1517 * Opening of "enable" adds a ref count to system,
1518 * so the name is safe to use.
1519 */
1520 if (system)
1521 name = system->name;
1522
1523 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524 if (ret)
1525 goto out;
1526
1527 ret = cnt;
1528
1529out:
1530 *ppos += cnt;
1531
1532 return ret;
1533}
1534
1535enum {
1536 FORMAT_HEADER = 1,
1537 FORMAT_FIELD_SEPERATOR = 2,
1538 FORMAT_PRINTFMT = 3,
1539};
1540
1541static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542{
1543 struct trace_event_call *call = event_file_data(m->private);
1544 struct list_head *common_head = &ftrace_common_fields;
1545 struct list_head *head = trace_get_fields(call);
1546 struct list_head *node = v;
1547
1548 (*pos)++;
1549
1550 switch ((unsigned long)v) {
1551 case FORMAT_HEADER:
1552 node = common_head;
1553 break;
1554
1555 case FORMAT_FIELD_SEPERATOR:
1556 node = head;
1557 break;
1558
1559 case FORMAT_PRINTFMT:
1560 /* all done */
1561 return NULL;
1562 }
1563
1564 node = node->prev;
1565 if (node == common_head)
1566 return (void *)FORMAT_FIELD_SEPERATOR;
1567 else if (node == head)
1568 return (void *)FORMAT_PRINTFMT;
1569 else
1570 return node;
1571}
1572
1573static int f_show(struct seq_file *m, void *v)
1574{
1575 struct trace_event_call *call = event_file_data(m->private);
1576 struct ftrace_event_field *field;
1577 const char *array_descriptor;
1578
1579 switch ((unsigned long)v) {
1580 case FORMAT_HEADER:
1581 seq_printf(m, "name: %s\n", trace_event_name(call));
1582 seq_printf(m, "ID: %d\n", call->event.type);
1583 seq_puts(m, "format:\n");
1584 return 0;
1585
1586 case FORMAT_FIELD_SEPERATOR:
1587 seq_putc(m, '\n');
1588 return 0;
1589
1590 case FORMAT_PRINTFMT:
1591 seq_printf(m, "\nprint fmt: %s\n",
1592 call->print_fmt);
1593 return 0;
1594 }
1595
1596 field = list_entry(v, struct ftrace_event_field, link);
1597 /*
1598 * Smartly shows the array type(except dynamic array).
1599 * Normal:
1600 * field:TYPE VAR
1601 * If TYPE := TYPE[LEN], it is shown:
1602 * field:TYPE VAR[LEN]
1603 */
1604 array_descriptor = strchr(field->type, '[');
1605
1606 if (str_has_prefix(field->type, "__data_loc"))
1607 array_descriptor = NULL;
1608
1609 if (!array_descriptor)
1610 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611 field->type, field->name, field->offset,
1612 field->size, !!field->is_signed);
1613 else if (field->len)
1614 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615 (int)(array_descriptor - field->type),
1616 field->type, field->name,
1617 field->len, field->offset,
1618 field->size, !!field->is_signed);
1619 else
1620 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621 (int)(array_descriptor - field->type),
1622 field->type, field->name,
1623 field->offset, field->size, !!field->is_signed);
1624
1625 return 0;
1626}
1627
1628static void *f_start(struct seq_file *m, loff_t *pos)
1629{
1630 void *p = (void *)FORMAT_HEADER;
1631 loff_t l = 0;
1632
1633 /* ->stop() is called even if ->start() fails */
1634 mutex_lock(&event_mutex);
1635 if (!event_file_data(m->private))
1636 return ERR_PTR(-ENODEV);
1637
1638 while (l < *pos && p)
1639 p = f_next(m, p, &l);
1640
1641 return p;
1642}
1643
1644static void f_stop(struct seq_file *m, void *p)
1645{
1646 mutex_unlock(&event_mutex);
1647}
1648
1649static const struct seq_operations trace_format_seq_ops = {
1650 .start = f_start,
1651 .next = f_next,
1652 .stop = f_stop,
1653 .show = f_show,
1654};
1655
1656static int trace_format_open(struct inode *inode, struct file *file)
1657{
1658 struct seq_file *m;
1659 int ret;
1660
1661 /* Do we want to hide event format files on tracefs lockdown? */
1662
1663 ret = seq_open(file, &trace_format_seq_ops);
1664 if (ret < 0)
1665 return ret;
1666
1667 m = file->private_data;
1668 m->private = file;
1669
1670 return 0;
1671}
1672
1673#ifdef CONFIG_PERF_EVENTS
1674static ssize_t
1675event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1676{
1677 int id = (long)event_file_data(filp);
1678 char buf[32];
1679 int len;
1680
1681 if (unlikely(!id))
1682 return -ENODEV;
1683
1684 len = sprintf(buf, "%d\n", id);
1685
1686 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687}
1688#endif
1689
1690static ssize_t
1691event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1692 loff_t *ppos)
1693{
1694 struct trace_event_file *file;
1695 struct trace_seq *s;
1696 int r = -ENODEV;
1697
1698 if (*ppos)
1699 return 0;
1700
1701 s = kmalloc(sizeof(*s), GFP_KERNEL);
1702
1703 if (!s)
1704 return -ENOMEM;
1705
1706 trace_seq_init(s);
1707
1708 mutex_lock(&event_mutex);
1709 file = event_file_data(filp);
1710 if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711 print_event_filter(file, s);
1712 mutex_unlock(&event_mutex);
1713
1714 if (file)
1715 r = simple_read_from_buffer(ubuf, cnt, ppos,
1716 s->buffer, trace_seq_used(s));
1717
1718 kfree(s);
1719
1720 return r;
1721}
1722
1723static ssize_t
1724event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1725 loff_t *ppos)
1726{
1727 struct trace_event_file *file;
1728 char *buf;
1729 int err = -ENODEV;
1730
1731 if (cnt >= PAGE_SIZE)
1732 return -EINVAL;
1733
1734 buf = memdup_user_nul(ubuf, cnt);
1735 if (IS_ERR(buf))
1736 return PTR_ERR(buf);
1737
1738 mutex_lock(&event_mutex);
1739 file = event_file_data(filp);
1740 if (file)
1741 err = apply_event_filter(file, buf);
1742 mutex_unlock(&event_mutex);
1743
1744 kfree(buf);
1745 if (err < 0)
1746 return err;
1747
1748 *ppos += cnt;
1749
1750 return cnt;
1751}
1752
1753static LIST_HEAD(event_subsystems);
1754
1755static int subsystem_open(struct inode *inode, struct file *filp)
1756{
1757 struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758 struct trace_array *tr = NULL, *iter_tr;
1759 struct event_subsystem *system = NULL;
1760 int ret;
1761
1762 if (tracing_is_disabled())
1763 return -ENODEV;
1764
1765 /* Make sure the system still exists */
1766 mutex_lock(&event_mutex);
1767 mutex_lock(&trace_types_lock);
1768 list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770 if (iter_dir == inode->i_private) {
1771 /* Don't open systems with no events */
1772 tr = iter_tr;
1773 dir = iter_dir;
1774 if (dir->nr_events) {
1775 __get_system_dir(dir);
1776 system = dir->subsystem;
1777 }
1778 goto exit_loop;
1779 }
1780 }
1781 }
1782 exit_loop:
1783 mutex_unlock(&trace_types_lock);
1784 mutex_unlock(&event_mutex);
1785
1786 if (!system)
1787 return -ENODEV;
1788
1789 /* Still need to increment the ref count of the system */
1790 if (trace_array_get(tr) < 0) {
1791 put_system(dir);
1792 return -ENODEV;
1793 }
1794
1795 ret = tracing_open_generic(inode, filp);
1796 if (ret < 0) {
1797 trace_array_put(tr);
1798 put_system(dir);
1799 }
1800
1801 return ret;
1802}
1803
1804static int system_tr_open(struct inode *inode, struct file *filp)
1805{
1806 struct trace_subsystem_dir *dir;
1807 struct trace_array *tr = inode->i_private;
1808 int ret;
1809
1810 /* Make a temporary dir that has no system but points to tr */
1811 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1812 if (!dir)
1813 return -ENOMEM;
1814
1815 ret = tracing_open_generic_tr(inode, filp);
1816 if (ret < 0) {
1817 kfree(dir);
1818 return ret;
1819 }
1820 dir->tr = tr;
1821 filp->private_data = dir;
1822
1823 return 0;
1824}
1825
1826static int subsystem_release(struct inode *inode, struct file *file)
1827{
1828 struct trace_subsystem_dir *dir = file->private_data;
1829
1830 trace_array_put(dir->tr);
1831
1832 /*
1833 * If dir->subsystem is NULL, then this is a temporary
1834 * descriptor that was made for a trace_array to enable
1835 * all subsystems.
1836 */
1837 if (dir->subsystem)
1838 put_system(dir);
1839 else
1840 kfree(dir);
1841
1842 return 0;
1843}
1844
1845static ssize_t
1846subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1847 loff_t *ppos)
1848{
1849 struct trace_subsystem_dir *dir = filp->private_data;
1850 struct event_subsystem *system = dir->subsystem;
1851 struct trace_seq *s;
1852 int r;
1853
1854 if (*ppos)
1855 return 0;
1856
1857 s = kmalloc(sizeof(*s), GFP_KERNEL);
1858 if (!s)
1859 return -ENOMEM;
1860
1861 trace_seq_init(s);
1862
1863 print_subsystem_event_filter(system, s);
1864 r = simple_read_from_buffer(ubuf, cnt, ppos,
1865 s->buffer, trace_seq_used(s));
1866
1867 kfree(s);
1868
1869 return r;
1870}
1871
1872static ssize_t
1873subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1874 loff_t *ppos)
1875{
1876 struct trace_subsystem_dir *dir = filp->private_data;
1877 char *buf;
1878 int err;
1879
1880 if (cnt >= PAGE_SIZE)
1881 return -EINVAL;
1882
1883 buf = memdup_user_nul(ubuf, cnt);
1884 if (IS_ERR(buf))
1885 return PTR_ERR(buf);
1886
1887 err = apply_subsystem_event_filter(dir, buf);
1888 kfree(buf);
1889 if (err < 0)
1890 return err;
1891
1892 *ppos += cnt;
1893
1894 return cnt;
1895}
1896
1897static ssize_t
1898show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1899{
1900 struct trace_array *tr = filp->private_data;
1901 struct trace_seq *s;
1902 int r;
1903
1904 if (*ppos)
1905 return 0;
1906
1907 s = kmalloc(sizeof(*s), GFP_KERNEL);
1908 if (!s)
1909 return -ENOMEM;
1910
1911 trace_seq_init(s);
1912
1913 ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914 r = simple_read_from_buffer(ubuf, cnt, ppos,
1915 s->buffer, trace_seq_used(s));
1916
1917 kfree(s);
1918
1919 return r;
1920}
1921
1922static ssize_t
1923show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1924{
1925 struct trace_seq *s;
1926 int r;
1927
1928 if (*ppos)
1929 return 0;
1930
1931 s = kmalloc(sizeof(*s), GFP_KERNEL);
1932 if (!s)
1933 return -ENOMEM;
1934
1935 trace_seq_init(s);
1936
1937 ring_buffer_print_entry_header(s);
1938 r = simple_read_from_buffer(ubuf, cnt, ppos,
1939 s->buffer, trace_seq_used(s));
1940
1941 kfree(s);
1942
1943 return r;
1944}
1945
1946static void ignore_task_cpu(void *data)
1947{
1948 struct trace_array *tr = data;
1949 struct trace_pid_list *pid_list;
1950 struct trace_pid_list *no_pid_list;
1951
1952 /*
1953 * This function is called by on_each_cpu() while the
1954 * event_mutex is held.
1955 */
1956 pid_list = rcu_dereference_protected(tr->filtered_pids,
1957 mutex_is_locked(&event_mutex));
1958 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959 mutex_is_locked(&event_mutex));
1960
1961 this_cpu_write(tr->array_buffer.data->ignore_pid,
1962 trace_ignore_this_task(pid_list, no_pid_list, current));
1963}
1964
1965static void register_pid_events(struct trace_array *tr)
1966{
1967 /*
1968 * Register a probe that is called before all other probes
1969 * to set ignore_pid if next or prev do not match.
1970 * Register a probe this is called after all other probes
1971 * to only keep ignore_pid set if next pid matches.
1972 */
1973 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1974 tr, INT_MAX);
1975 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1976 tr, 0);
1977
1978 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1979 tr, INT_MAX);
1980 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1981 tr, 0);
1982
1983 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1984 tr, INT_MAX);
1985 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1986 tr, 0);
1987
1988 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1989 tr, INT_MAX);
1990 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1991 tr, 0);
1992}
1993
1994static ssize_t
1995event_pid_write(struct file *filp, const char __user *ubuf,
1996 size_t cnt, loff_t *ppos, int type)
1997{
1998 struct seq_file *m = filp->private_data;
1999 struct trace_array *tr = m->private;
2000 struct trace_pid_list *filtered_pids = NULL;
2001 struct trace_pid_list *other_pids = NULL;
2002 struct trace_pid_list *pid_list;
2003 struct trace_event_file *file;
2004 ssize_t ret;
2005
2006 if (!cnt)
2007 return 0;
2008
2009 ret = tracing_update_buffers(tr);
2010 if (ret < 0)
2011 return ret;
2012
2013 mutex_lock(&event_mutex);
2014
2015 if (type == TRACE_PIDS) {
2016 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017 lockdep_is_held(&event_mutex));
2018 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019 lockdep_is_held(&event_mutex));
2020 } else {
2021 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022 lockdep_is_held(&event_mutex));
2023 other_pids = rcu_dereference_protected(tr->filtered_pids,
2024 lockdep_is_held(&event_mutex));
2025 }
2026
2027 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2028 if (ret < 0)
2029 goto out;
2030
2031 if (type == TRACE_PIDS)
2032 rcu_assign_pointer(tr->filtered_pids, pid_list);
2033 else
2034 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2035
2036 list_for_each_entry(file, &tr->events, list) {
2037 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2038 }
2039
2040 if (filtered_pids) {
2041 tracepoint_synchronize_unregister();
2042 trace_pid_list_free(filtered_pids);
2043 } else if (pid_list && !other_pids) {
2044 register_pid_events(tr);
2045 }
2046
2047 /*
2048 * Ignoring of pids is done at task switch. But we have to
2049 * check for those tasks that are currently running.
2050 * Always do this in case a pid was appended or removed.
2051 */
2052 on_each_cpu(ignore_task_cpu, tr, 1);
2053
2054 out:
2055 mutex_unlock(&event_mutex);
2056
2057 if (ret > 0)
2058 *ppos += ret;
2059
2060 return ret;
2061}
2062
2063static ssize_t
2064ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065 size_t cnt, loff_t *ppos)
2066{
2067 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2068}
2069
2070static ssize_t
2071ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072 size_t cnt, loff_t *ppos)
2073{
2074 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2075}
2076
2077static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081static int ftrace_event_release(struct inode *inode, struct file *file);
2082
2083static const struct seq_operations show_event_seq_ops = {
2084 .start = t_start,
2085 .next = t_next,
2086 .show = t_show,
2087 .stop = t_stop,
2088};
2089
2090static const struct seq_operations show_set_event_seq_ops = {
2091 .start = s_start,
2092 .next = s_next,
2093 .show = t_show,
2094 .stop = t_stop,
2095};
2096
2097static const struct seq_operations show_set_pid_seq_ops = {
2098 .start = p_start,
2099 .next = p_next,
2100 .show = trace_pid_show,
2101 .stop = p_stop,
2102};
2103
2104static const struct seq_operations show_set_no_pid_seq_ops = {
2105 .start = np_start,
2106 .next = np_next,
2107 .show = trace_pid_show,
2108 .stop = p_stop,
2109};
2110
2111static const struct file_operations ftrace_avail_fops = {
2112 .open = ftrace_event_avail_open,
2113 .read = seq_read,
2114 .llseek = seq_lseek,
2115 .release = seq_release,
2116};
2117
2118static const struct file_operations ftrace_set_event_fops = {
2119 .open = ftrace_event_set_open,
2120 .read = seq_read,
2121 .write = ftrace_event_write,
2122 .llseek = seq_lseek,
2123 .release = ftrace_event_release,
2124};
2125
2126static const struct file_operations ftrace_set_event_pid_fops = {
2127 .open = ftrace_event_set_pid_open,
2128 .read = seq_read,
2129 .write = ftrace_event_pid_write,
2130 .llseek = seq_lseek,
2131 .release = ftrace_event_release,
2132};
2133
2134static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135 .open = ftrace_event_set_npid_open,
2136 .read = seq_read,
2137 .write = ftrace_event_npid_write,
2138 .llseek = seq_lseek,
2139 .release = ftrace_event_release,
2140};
2141
2142static const struct file_operations ftrace_enable_fops = {
2143 .open = tracing_open_file_tr,
2144 .read = event_enable_read,
2145 .write = event_enable_write,
2146 .release = tracing_release_file_tr,
2147 .llseek = default_llseek,
2148};
2149
2150static const struct file_operations ftrace_event_format_fops = {
2151 .open = trace_format_open,
2152 .read = seq_read,
2153 .llseek = seq_lseek,
2154 .release = seq_release,
2155};
2156
2157#ifdef CONFIG_PERF_EVENTS
2158static const struct file_operations ftrace_event_id_fops = {
2159 .read = event_id_read,
2160 .llseek = default_llseek,
2161};
2162#endif
2163
2164static const struct file_operations ftrace_event_filter_fops = {
2165 .open = tracing_open_file_tr,
2166 .read = event_filter_read,
2167 .write = event_filter_write,
2168 .release = tracing_release_file_tr,
2169 .llseek = default_llseek,
2170};
2171
2172static const struct file_operations ftrace_subsystem_filter_fops = {
2173 .open = subsystem_open,
2174 .read = subsystem_filter_read,
2175 .write = subsystem_filter_write,
2176 .llseek = default_llseek,
2177 .release = subsystem_release,
2178};
2179
2180static const struct file_operations ftrace_system_enable_fops = {
2181 .open = subsystem_open,
2182 .read = system_enable_read,
2183 .write = system_enable_write,
2184 .llseek = default_llseek,
2185 .release = subsystem_release,
2186};
2187
2188static const struct file_operations ftrace_tr_enable_fops = {
2189 .open = system_tr_open,
2190 .read = system_enable_read,
2191 .write = system_enable_write,
2192 .llseek = default_llseek,
2193 .release = subsystem_release,
2194};
2195
2196static const struct file_operations ftrace_show_header_page_fops = {
2197 .open = tracing_open_generic_tr,
2198 .read = show_header_page_file,
2199 .llseek = default_llseek,
2200 .release = tracing_release_generic_tr,
2201};
2202
2203static const struct file_operations ftrace_show_header_event_fops = {
2204 .open = tracing_open_generic_tr,
2205 .read = show_header_event_file,
2206 .llseek = default_llseek,
2207 .release = tracing_release_generic_tr,
2208};
2209
2210static int
2211ftrace_event_open(struct inode *inode, struct file *file,
2212 const struct seq_operations *seq_ops)
2213{
2214 struct seq_file *m;
2215 int ret;
2216
2217 ret = security_locked_down(LOCKDOWN_TRACEFS);
2218 if (ret)
2219 return ret;
2220
2221 ret = seq_open(file, seq_ops);
2222 if (ret < 0)
2223 return ret;
2224 m = file->private_data;
2225 /* copy tr over to seq ops */
2226 m->private = inode->i_private;
2227
2228 return ret;
2229}
2230
2231static int ftrace_event_release(struct inode *inode, struct file *file)
2232{
2233 struct trace_array *tr = inode->i_private;
2234
2235 trace_array_put(tr);
2236
2237 return seq_release(inode, file);
2238}
2239
2240static int
2241ftrace_event_avail_open(struct inode *inode, struct file *file)
2242{
2243 const struct seq_operations *seq_ops = &show_event_seq_ops;
2244
2245 /* Checks for tracefs lockdown */
2246 return ftrace_event_open(inode, file, seq_ops);
2247}
2248
2249static int
2250ftrace_event_set_open(struct inode *inode, struct file *file)
2251{
2252 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253 struct trace_array *tr = inode->i_private;
2254 int ret;
2255
2256 ret = tracing_check_open_get_tr(tr);
2257 if (ret)
2258 return ret;
2259
2260 if ((file->f_mode & FMODE_WRITE) &&
2261 (file->f_flags & O_TRUNC))
2262 ftrace_clear_events(tr);
2263
2264 ret = ftrace_event_open(inode, file, seq_ops);
2265 if (ret < 0)
2266 trace_array_put(tr);
2267 return ret;
2268}
2269
2270static int
2271ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2272{
2273 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274 struct trace_array *tr = inode->i_private;
2275 int ret;
2276
2277 ret = tracing_check_open_get_tr(tr);
2278 if (ret)
2279 return ret;
2280
2281 if ((file->f_mode & FMODE_WRITE) &&
2282 (file->f_flags & O_TRUNC))
2283 ftrace_clear_event_pids(tr, TRACE_PIDS);
2284
2285 ret = ftrace_event_open(inode, file, seq_ops);
2286 if (ret < 0)
2287 trace_array_put(tr);
2288 return ret;
2289}
2290
2291static int
2292ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2293{
2294 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295 struct trace_array *tr = inode->i_private;
2296 int ret;
2297
2298 ret = tracing_check_open_get_tr(tr);
2299 if (ret)
2300 return ret;
2301
2302 if ((file->f_mode & FMODE_WRITE) &&
2303 (file->f_flags & O_TRUNC))
2304 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2305
2306 ret = ftrace_event_open(inode, file, seq_ops);
2307 if (ret < 0)
2308 trace_array_put(tr);
2309 return ret;
2310}
2311
2312static struct event_subsystem *
2313create_new_subsystem(const char *name)
2314{
2315 struct event_subsystem *system;
2316
2317 /* need to create new entry */
2318 system = kmalloc(sizeof(*system), GFP_KERNEL);
2319 if (!system)
2320 return NULL;
2321
2322 system->ref_count = 1;
2323
2324 /* Only allocate if dynamic (kprobes and modules) */
2325 system->name = kstrdup_const(name, GFP_KERNEL);
2326 if (!system->name)
2327 goto out_free;
2328
2329 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330 if (!system->filter)
2331 goto out_free;
2332
2333 list_add(&system->list, &event_subsystems);
2334
2335 return system;
2336
2337 out_free:
2338 kfree_const(system->name);
2339 kfree(system);
2340 return NULL;
2341}
2342
2343static int system_callback(const char *name, umode_t *mode, void **data,
2344 const struct file_operations **fops)
2345{
2346 if (strcmp(name, "filter") == 0)
2347 *fops = &ftrace_subsystem_filter_fops;
2348
2349 else if (strcmp(name, "enable") == 0)
2350 *fops = &ftrace_system_enable_fops;
2351
2352 else
2353 return 0;
2354
2355 *mode = TRACE_MODE_WRITE;
2356 return 1;
2357}
2358
2359static struct eventfs_inode *
2360event_subsystem_dir(struct trace_array *tr, const char *name,
2361 struct trace_event_file *file, struct eventfs_inode *parent)
2362{
2363 struct event_subsystem *system, *iter;
2364 struct trace_subsystem_dir *dir;
2365 struct eventfs_inode *ei;
2366 int nr_entries;
2367 static struct eventfs_entry system_entries[] = {
2368 {
2369 .name = "filter",
2370 .callback = system_callback,
2371 },
2372 {
2373 .name = "enable",
2374 .callback = system_callback,
2375 }
2376 };
2377
2378 /* First see if we did not already create this dir */
2379 list_for_each_entry(dir, &tr->systems, list) {
2380 system = dir->subsystem;
2381 if (strcmp(system->name, name) == 0) {
2382 dir->nr_events++;
2383 file->system = dir;
2384 return dir->ei;
2385 }
2386 }
2387
2388 /* Now see if the system itself exists. */
2389 system = NULL;
2390 list_for_each_entry(iter, &event_subsystems, list) {
2391 if (strcmp(iter->name, name) == 0) {
2392 system = iter;
2393 break;
2394 }
2395 }
2396
2397 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2398 if (!dir)
2399 goto out_fail;
2400
2401 if (!system) {
2402 system = create_new_subsystem(name);
2403 if (!system)
2404 goto out_free;
2405 } else
2406 __get_system(system);
2407
2408 /* ftrace only has directories no files */
2409 if (strcmp(name, "ftrace") == 0)
2410 nr_entries = 0;
2411 else
2412 nr_entries = ARRAY_SIZE(system_entries);
2413
2414 ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2415 if (IS_ERR(ei)) {
2416 pr_warn("Failed to create system directory %s\n", name);
2417 __put_system(system);
2418 goto out_free;
2419 }
2420
2421 dir->ei = ei;
2422 dir->tr = tr;
2423 dir->ref_count = 1;
2424 dir->nr_events = 1;
2425 dir->subsystem = system;
2426 file->system = dir;
2427
2428 list_add(&dir->list, &tr->systems);
2429
2430 return dir->ei;
2431
2432 out_free:
2433 kfree(dir);
2434 out_fail:
2435 /* Only print this message if failed on memory allocation */
2436 if (!dir || !system)
2437 pr_warn("No memory to create event subsystem %s\n", name);
2438 return NULL;
2439}
2440
2441static int
2442event_define_fields(struct trace_event_call *call)
2443{
2444 struct list_head *head;
2445 int ret = 0;
2446
2447 /*
2448 * Other events may have the same class. Only update
2449 * the fields if they are not already defined.
2450 */
2451 head = trace_get_fields(call);
2452 if (list_empty(head)) {
2453 struct trace_event_fields *field = call->class->fields_array;
2454 unsigned int offset = sizeof(struct trace_entry);
2455
2456 for (; field->type; field++) {
2457 if (field->type == TRACE_FUNCTION_TYPE) {
2458 field->define_fields(call);
2459 break;
2460 }
2461
2462 offset = ALIGN(offset, field->align);
2463 ret = trace_define_field_ext(call, field->type, field->name,
2464 offset, field->size,
2465 field->is_signed, field->filter_type,
2466 field->len);
2467 if (WARN_ON_ONCE(ret)) {
2468 pr_err("error code is %d\n", ret);
2469 break;
2470 }
2471
2472 offset += field->size;
2473 }
2474 }
2475
2476 return ret;
2477}
2478
2479static int event_callback(const char *name, umode_t *mode, void **data,
2480 const struct file_operations **fops)
2481{
2482 struct trace_event_file *file = *data;
2483 struct trace_event_call *call = file->event_call;
2484
2485 if (strcmp(name, "format") == 0) {
2486 *mode = TRACE_MODE_READ;
2487 *fops = &ftrace_event_format_fops;
2488 *data = call;
2489 return 1;
2490 }
2491
2492 /*
2493 * Only event directories that can be enabled should have
2494 * triggers or filters, with the exception of the "print"
2495 * event that can have a "trigger" file.
2496 */
2497 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498 if (call->class->reg && strcmp(name, "enable") == 0) {
2499 *mode = TRACE_MODE_WRITE;
2500 *fops = &ftrace_enable_fops;
2501 return 1;
2502 }
2503
2504 if (strcmp(name, "filter") == 0) {
2505 *mode = TRACE_MODE_WRITE;
2506 *fops = &ftrace_event_filter_fops;
2507 return 1;
2508 }
2509 }
2510
2511 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512 strcmp(trace_event_name(call), "print") == 0) {
2513 if (strcmp(name, "trigger") == 0) {
2514 *mode = TRACE_MODE_WRITE;
2515 *fops = &event_trigger_fops;
2516 return 1;
2517 }
2518 }
2519
2520#ifdef CONFIG_PERF_EVENTS
2521 if (call->event.type && call->class->reg &&
2522 strcmp(name, "id") == 0) {
2523 *mode = TRACE_MODE_READ;
2524 *data = (void *)(long)call->event.type;
2525 *fops = &ftrace_event_id_fops;
2526 return 1;
2527 }
2528#endif
2529
2530#ifdef CONFIG_HIST_TRIGGERS
2531 if (strcmp(name, "hist") == 0) {
2532 *mode = TRACE_MODE_READ;
2533 *fops = &event_hist_fops;
2534 return 1;
2535 }
2536#endif
2537#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538 if (strcmp(name, "hist_debug") == 0) {
2539 *mode = TRACE_MODE_READ;
2540 *fops = &event_hist_debug_fops;
2541 return 1;
2542 }
2543#endif
2544#ifdef CONFIG_TRACE_EVENT_INJECT
2545 if (call->event.type && call->class->reg &&
2546 strcmp(name, "inject") == 0) {
2547 *mode = 0200;
2548 *fops = &event_inject_fops;
2549 return 1;
2550 }
2551#endif
2552 return 0;
2553}
2554
2555/* The file is incremented on creation and freeing the enable file decrements it */
2556static void event_release(const char *name, void *data)
2557{
2558 struct trace_event_file *file = data;
2559
2560 event_file_put(file);
2561}
2562
2563static int
2564event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2565{
2566 struct trace_event_call *call = file->event_call;
2567 struct trace_array *tr = file->tr;
2568 struct eventfs_inode *e_events;
2569 struct eventfs_inode *ei;
2570 const char *name;
2571 int nr_entries;
2572 int ret;
2573 static struct eventfs_entry event_entries[] = {
2574 {
2575 .name = "enable",
2576 .callback = event_callback,
2577 .release = event_release,
2578 },
2579 {
2580 .name = "filter",
2581 .callback = event_callback,
2582 },
2583 {
2584 .name = "trigger",
2585 .callback = event_callback,
2586 },
2587 {
2588 .name = "format",
2589 .callback = event_callback,
2590 },
2591#ifdef CONFIG_PERF_EVENTS
2592 {
2593 .name = "id",
2594 .callback = event_callback,
2595 },
2596#endif
2597#ifdef CONFIG_HIST_TRIGGERS
2598 {
2599 .name = "hist",
2600 .callback = event_callback,
2601 },
2602#endif
2603#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2604 {
2605 .name = "hist_debug",
2606 .callback = event_callback,
2607 },
2608#endif
2609#ifdef CONFIG_TRACE_EVENT_INJECT
2610 {
2611 .name = "inject",
2612 .callback = event_callback,
2613 },
2614#endif
2615 };
2616
2617 /*
2618 * If the trace point header did not define TRACE_SYSTEM
2619 * then the system would be called "TRACE_SYSTEM". This should
2620 * never happen.
2621 */
2622 if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2623 return -ENODEV;
2624
2625 e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2626 if (!e_events)
2627 return -ENOMEM;
2628
2629 nr_entries = ARRAY_SIZE(event_entries);
2630
2631 name = trace_event_name(call);
2632 ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2633 if (IS_ERR(ei)) {
2634 pr_warn("Could not create tracefs '%s' directory\n", name);
2635 return -1;
2636 }
2637
2638 file->ei = ei;
2639
2640 ret = event_define_fields(call);
2641 if (ret < 0) {
2642 pr_warn("Could not initialize trace point events/%s\n", name);
2643 return ret;
2644 }
2645
2646 /* Gets decremented on freeing of the "enable" file */
2647 event_file_get(file);
2648
2649 return 0;
2650}
2651
2652static void remove_event_from_tracers(struct trace_event_call *call)
2653{
2654 struct trace_event_file *file;
2655 struct trace_array *tr;
2656
2657 do_for_each_event_file_safe(tr, file) {
2658 if (file->event_call != call)
2659 continue;
2660
2661 remove_event_file_dir(file);
2662 /*
2663 * The do_for_each_event_file_safe() is
2664 * a double loop. After finding the call for this
2665 * trace_array, we use break to jump to the next
2666 * trace_array.
2667 */
2668 break;
2669 } while_for_each_event_file();
2670}
2671
2672static void event_remove(struct trace_event_call *call)
2673{
2674 struct trace_array *tr;
2675 struct trace_event_file *file;
2676
2677 do_for_each_event_file(tr, file) {
2678 if (file->event_call != call)
2679 continue;
2680
2681 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2682 tr->clear_trace = true;
2683
2684 ftrace_event_enable_disable(file, 0);
2685 /*
2686 * The do_for_each_event_file() is
2687 * a double loop. After finding the call for this
2688 * trace_array, we use break to jump to the next
2689 * trace_array.
2690 */
2691 break;
2692 } while_for_each_event_file();
2693
2694 if (call->event.funcs)
2695 __unregister_trace_event(&call->event);
2696 remove_event_from_tracers(call);
2697 list_del(&call->list);
2698}
2699
2700static int event_init(struct trace_event_call *call)
2701{
2702 int ret = 0;
2703 const char *name;
2704
2705 name = trace_event_name(call);
2706 if (WARN_ON(!name))
2707 return -EINVAL;
2708
2709 if (call->class->raw_init) {
2710 ret = call->class->raw_init(call);
2711 if (ret < 0 && ret != -ENOSYS)
2712 pr_warn("Could not initialize trace events/%s\n", name);
2713 }
2714
2715 return ret;
2716}
2717
2718static int
2719__register_event(struct trace_event_call *call, struct module *mod)
2720{
2721 int ret;
2722
2723 ret = event_init(call);
2724 if (ret < 0)
2725 return ret;
2726
2727 list_add(&call->list, &ftrace_events);
2728 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2729 atomic_set(&call->refcnt, 0);
2730 else
2731 call->module = mod;
2732
2733 return 0;
2734}
2735
2736static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2737{
2738 int rlen;
2739 int elen;
2740
2741 /* Find the length of the eval value as a string */
2742 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2743 /* Make sure there's enough room to replace the string with the value */
2744 if (len < elen)
2745 return NULL;
2746
2747 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2748
2749 /* Get the rest of the string of ptr */
2750 rlen = strlen(ptr + len);
2751 memmove(ptr + elen, ptr + len, rlen);
2752 /* Make sure we end the new string */
2753 ptr[elen + rlen] = 0;
2754
2755 return ptr + elen;
2756}
2757
2758static void update_event_printk(struct trace_event_call *call,
2759 struct trace_eval_map *map)
2760{
2761 char *ptr;
2762 int quote = 0;
2763 int len = strlen(map->eval_string);
2764
2765 for (ptr = call->print_fmt; *ptr; ptr++) {
2766 if (*ptr == '\\') {
2767 ptr++;
2768 /* paranoid */
2769 if (!*ptr)
2770 break;
2771 continue;
2772 }
2773 if (*ptr == '"') {
2774 quote ^= 1;
2775 continue;
2776 }
2777 if (quote)
2778 continue;
2779 if (isdigit(*ptr)) {
2780 /* skip numbers */
2781 do {
2782 ptr++;
2783 /* Check for alpha chars like ULL */
2784 } while (isalnum(*ptr));
2785 if (!*ptr)
2786 break;
2787 /*
2788 * A number must have some kind of delimiter after
2789 * it, and we can ignore that too.
2790 */
2791 continue;
2792 }
2793 if (isalpha(*ptr) || *ptr == '_') {
2794 if (strncmp(map->eval_string, ptr, len) == 0 &&
2795 !isalnum(ptr[len]) && ptr[len] != '_') {
2796 ptr = eval_replace(ptr, map, len);
2797 /* enum/sizeof string smaller than value */
2798 if (WARN_ON_ONCE(!ptr))
2799 return;
2800 /*
2801 * No need to decrement here, as eval_replace()
2802 * returns the pointer to the character passed
2803 * the eval, and two evals can not be placed
2804 * back to back without something in between.
2805 * We can skip that something in between.
2806 */
2807 continue;
2808 }
2809 skip_more:
2810 do {
2811 ptr++;
2812 } while (isalnum(*ptr) || *ptr == '_');
2813 if (!*ptr)
2814 break;
2815 /*
2816 * If what comes after this variable is a '.' or
2817 * '->' then we can continue to ignore that string.
2818 */
2819 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2820 ptr += *ptr == '.' ? 1 : 2;
2821 if (!*ptr)
2822 break;
2823 goto skip_more;
2824 }
2825 /*
2826 * Once again, we can skip the delimiter that came
2827 * after the string.
2828 */
2829 continue;
2830 }
2831 }
2832}
2833
2834static void add_str_to_module(struct module *module, char *str)
2835{
2836 struct module_string *modstr;
2837
2838 modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2839
2840 /*
2841 * If we failed to allocate memory here, then we'll just
2842 * let the str memory leak when the module is removed.
2843 * If this fails to allocate, there's worse problems than
2844 * a leaked string on module removal.
2845 */
2846 if (WARN_ON_ONCE(!modstr))
2847 return;
2848
2849 modstr->module = module;
2850 modstr->str = str;
2851
2852 list_add(&modstr->next, &module_strings);
2853}
2854
2855static void update_event_fields(struct trace_event_call *call,
2856 struct trace_eval_map *map)
2857{
2858 struct ftrace_event_field *field;
2859 struct list_head *head;
2860 char *ptr;
2861 char *str;
2862 int len = strlen(map->eval_string);
2863
2864 /* Dynamic events should never have field maps */
2865 if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2866 return;
2867
2868 head = trace_get_fields(call);
2869 list_for_each_entry(field, head, link) {
2870 ptr = strchr(field->type, '[');
2871 if (!ptr)
2872 continue;
2873 ptr++;
2874
2875 if (!isalpha(*ptr) && *ptr != '_')
2876 continue;
2877
2878 if (strncmp(map->eval_string, ptr, len) != 0)
2879 continue;
2880
2881 str = kstrdup(field->type, GFP_KERNEL);
2882 if (WARN_ON_ONCE(!str))
2883 return;
2884 ptr = str + (ptr - field->type);
2885 ptr = eval_replace(ptr, map, len);
2886 /* enum/sizeof string smaller than value */
2887 if (WARN_ON_ONCE(!ptr)) {
2888 kfree(str);
2889 continue;
2890 }
2891
2892 /*
2893 * If the event is part of a module, then we need to free the string
2894 * when the module is removed. Otherwise, it will stay allocated
2895 * until a reboot.
2896 */
2897 if (call->module)
2898 add_str_to_module(call->module, str);
2899
2900 field->type = str;
2901 }
2902}
2903
2904void trace_event_eval_update(struct trace_eval_map **map, int len)
2905{
2906 struct trace_event_call *call, *p;
2907 const char *last_system = NULL;
2908 bool first = false;
2909 int last_i;
2910 int i;
2911
2912 down_write(&trace_event_sem);
2913 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2914 /* events are usually grouped together with systems */
2915 if (!last_system || call->class->system != last_system) {
2916 first = true;
2917 last_i = 0;
2918 last_system = call->class->system;
2919 }
2920
2921 /*
2922 * Since calls are grouped by systems, the likelihood that the
2923 * next call in the iteration belongs to the same system as the
2924 * previous call is high. As an optimization, we skip searching
2925 * for a map[] that matches the call's system if the last call
2926 * was from the same system. That's what last_i is for. If the
2927 * call has the same system as the previous call, then last_i
2928 * will be the index of the first map[] that has a matching
2929 * system.
2930 */
2931 for (i = last_i; i < len; i++) {
2932 if (call->class->system == map[i]->system) {
2933 /* Save the first system if need be */
2934 if (first) {
2935 last_i = i;
2936 first = false;
2937 }
2938 update_event_printk(call, map[i]);
2939 update_event_fields(call, map[i]);
2940 }
2941 }
2942 cond_resched();
2943 }
2944 up_write(&trace_event_sem);
2945}
2946
2947static bool event_in_systems(struct trace_event_call *call,
2948 const char *systems)
2949{
2950 const char *system;
2951 const char *p;
2952
2953 if (!systems)
2954 return true;
2955
2956 system = call->class->system;
2957 p = strstr(systems, system);
2958 if (!p)
2959 return false;
2960
2961 if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2962 return false;
2963
2964 p += strlen(system);
2965 return !*p || isspace(*p) || *p == ',';
2966}
2967
2968static struct trace_event_file *
2969trace_create_new_event(struct trace_event_call *call,
2970 struct trace_array *tr)
2971{
2972 struct trace_pid_list *no_pid_list;
2973 struct trace_pid_list *pid_list;
2974 struct trace_event_file *file;
2975 unsigned int first;
2976
2977 if (!event_in_systems(call, tr->system_names))
2978 return NULL;
2979
2980 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2981 if (!file)
2982 return ERR_PTR(-ENOMEM);
2983
2984 pid_list = rcu_dereference_protected(tr->filtered_pids,
2985 lockdep_is_held(&event_mutex));
2986 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2987 lockdep_is_held(&event_mutex));
2988
2989 if (!trace_pid_list_first(pid_list, &first) ||
2990 !trace_pid_list_first(no_pid_list, &first))
2991 file->flags |= EVENT_FILE_FL_PID_FILTER;
2992
2993 file->event_call = call;
2994 file->tr = tr;
2995 atomic_set(&file->sm_ref, 0);
2996 atomic_set(&file->tm_ref, 0);
2997 INIT_LIST_HEAD(&file->triggers);
2998 list_add(&file->list, &tr->events);
2999 event_file_get(file);
3000
3001 return file;
3002}
3003
3004#define MAX_BOOT_TRIGGERS 32
3005
3006static struct boot_triggers {
3007 const char *event;
3008 char *trigger;
3009} bootup_triggers[MAX_BOOT_TRIGGERS];
3010
3011static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3012static int nr_boot_triggers;
3013
3014static __init int setup_trace_triggers(char *str)
3015{
3016 char *trigger;
3017 char *buf;
3018 int i;
3019
3020 strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3021 trace_set_ring_buffer_expanded(NULL);
3022 disable_tracing_selftest("running event triggers");
3023
3024 buf = bootup_trigger_buf;
3025 for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3026 trigger = strsep(&buf, ",");
3027 if (!trigger)
3028 break;
3029 bootup_triggers[i].event = strsep(&trigger, ".");
3030 bootup_triggers[i].trigger = trigger;
3031 if (!bootup_triggers[i].trigger)
3032 break;
3033 }
3034
3035 nr_boot_triggers = i;
3036 return 1;
3037}
3038__setup("trace_trigger=", setup_trace_triggers);
3039
3040/* Add an event to a trace directory */
3041static int
3042__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3043{
3044 struct trace_event_file *file;
3045
3046 file = trace_create_new_event(call, tr);
3047 /*
3048 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3049 * allocation, or NULL if the event is not part of the tr->system_names.
3050 * When the event is not part of the tr->system_names, return zero, not
3051 * an error.
3052 */
3053 if (!file)
3054 return 0;
3055
3056 if (IS_ERR(file))
3057 return PTR_ERR(file);
3058
3059 if (eventdir_initialized)
3060 return event_create_dir(tr->event_dir, file);
3061 else
3062 return event_define_fields(call);
3063}
3064
3065static void trace_early_triggers(struct trace_event_file *file, const char *name)
3066{
3067 int ret;
3068 int i;
3069
3070 for (i = 0; i < nr_boot_triggers; i++) {
3071 if (strcmp(name, bootup_triggers[i].event))
3072 continue;
3073 mutex_lock(&event_mutex);
3074 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3075 mutex_unlock(&event_mutex);
3076 if (ret)
3077 pr_err("Failed to register trigger '%s' on event %s\n",
3078 bootup_triggers[i].trigger,
3079 bootup_triggers[i].event);
3080 }
3081}
3082
3083/*
3084 * Just create a descriptor for early init. A descriptor is required
3085 * for enabling events at boot. We want to enable events before
3086 * the filesystem is initialized.
3087 */
3088static int
3089__trace_early_add_new_event(struct trace_event_call *call,
3090 struct trace_array *tr)
3091{
3092 struct trace_event_file *file;
3093 int ret;
3094
3095 file = trace_create_new_event(call, tr);
3096 /*
3097 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3098 * allocation, or NULL if the event is not part of the tr->system_names.
3099 * When the event is not part of the tr->system_names, return zero, not
3100 * an error.
3101 */
3102 if (!file)
3103 return 0;
3104
3105 if (IS_ERR(file))
3106 return PTR_ERR(file);
3107
3108 ret = event_define_fields(call);
3109 if (ret)
3110 return ret;
3111
3112 trace_early_triggers(file, trace_event_name(call));
3113
3114 return 0;
3115}
3116
3117struct ftrace_module_file_ops;
3118static void __add_event_to_tracers(struct trace_event_call *call);
3119
3120/* Add an additional event_call dynamically */
3121int trace_add_event_call(struct trace_event_call *call)
3122{
3123 int ret;
3124 lockdep_assert_held(&event_mutex);
3125
3126 mutex_lock(&trace_types_lock);
3127
3128 ret = __register_event(call, NULL);
3129 if (ret >= 0)
3130 __add_event_to_tracers(call);
3131
3132 mutex_unlock(&trace_types_lock);
3133 return ret;
3134}
3135EXPORT_SYMBOL_GPL(trace_add_event_call);
3136
3137/*
3138 * Must be called under locking of trace_types_lock, event_mutex and
3139 * trace_event_sem.
3140 */
3141static void __trace_remove_event_call(struct trace_event_call *call)
3142{
3143 event_remove(call);
3144 trace_destroy_fields(call);
3145 free_event_filter(call->filter);
3146 call->filter = NULL;
3147}
3148
3149static int probe_remove_event_call(struct trace_event_call *call)
3150{
3151 struct trace_array *tr;
3152 struct trace_event_file *file;
3153
3154#ifdef CONFIG_PERF_EVENTS
3155 if (call->perf_refcount)
3156 return -EBUSY;
3157#endif
3158 do_for_each_event_file(tr, file) {
3159 if (file->event_call != call)
3160 continue;
3161 /*
3162 * We can't rely on ftrace_event_enable_disable(enable => 0)
3163 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3164 * TRACE_REG_UNREGISTER.
3165 */
3166 if (file->flags & EVENT_FILE_FL_ENABLED)
3167 goto busy;
3168
3169 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3170 tr->clear_trace = true;
3171 /*
3172 * The do_for_each_event_file_safe() is
3173 * a double loop. After finding the call for this
3174 * trace_array, we use break to jump to the next
3175 * trace_array.
3176 */
3177 break;
3178 } while_for_each_event_file();
3179
3180 __trace_remove_event_call(call);
3181
3182 return 0;
3183 busy:
3184 /* No need to clear the trace now */
3185 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3186 tr->clear_trace = false;
3187 }
3188 return -EBUSY;
3189}
3190
3191/* Remove an event_call */
3192int trace_remove_event_call(struct trace_event_call *call)
3193{
3194 int ret;
3195
3196 lockdep_assert_held(&event_mutex);
3197
3198 mutex_lock(&trace_types_lock);
3199 down_write(&trace_event_sem);
3200 ret = probe_remove_event_call(call);
3201 up_write(&trace_event_sem);
3202 mutex_unlock(&trace_types_lock);
3203
3204 return ret;
3205}
3206EXPORT_SYMBOL_GPL(trace_remove_event_call);
3207
3208#define for_each_event(event, start, end) \
3209 for (event = start; \
3210 (unsigned long)event < (unsigned long)end; \
3211 event++)
3212
3213#ifdef CONFIG_MODULES
3214
3215static void trace_module_add_events(struct module *mod)
3216{
3217 struct trace_event_call **call, **start, **end;
3218
3219 if (!mod->num_trace_events)
3220 return;
3221
3222 /* Don't add infrastructure for mods without tracepoints */
3223 if (trace_module_has_bad_taint(mod)) {
3224 pr_err("%s: module has bad taint, not creating trace events\n",
3225 mod->name);
3226 return;
3227 }
3228
3229 start = mod->trace_events;
3230 end = mod->trace_events + mod->num_trace_events;
3231
3232 for_each_event(call, start, end) {
3233 __register_event(*call, mod);
3234 __add_event_to_tracers(*call);
3235 }
3236}
3237
3238static void trace_module_remove_events(struct module *mod)
3239{
3240 struct trace_event_call *call, *p;
3241 struct module_string *modstr, *m;
3242
3243 down_write(&trace_event_sem);
3244 list_for_each_entry_safe(call, p, &ftrace_events, list) {
3245 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3246 continue;
3247 if (call->module == mod)
3248 __trace_remove_event_call(call);
3249 }
3250 /* Check for any strings allocade for this module */
3251 list_for_each_entry_safe(modstr, m, &module_strings, next) {
3252 if (modstr->module != mod)
3253 continue;
3254 list_del(&modstr->next);
3255 kfree(modstr->str);
3256 kfree(modstr);
3257 }
3258 up_write(&trace_event_sem);
3259
3260 /*
3261 * It is safest to reset the ring buffer if the module being unloaded
3262 * registered any events that were used. The only worry is if
3263 * a new module gets loaded, and takes on the same id as the events
3264 * of this module. When printing out the buffer, traced events left
3265 * over from this module may be passed to the new module events and
3266 * unexpected results may occur.
3267 */
3268 tracing_reset_all_online_cpus_unlocked();
3269}
3270
3271static int trace_module_notify(struct notifier_block *self,
3272 unsigned long val, void *data)
3273{
3274 struct module *mod = data;
3275
3276 mutex_lock(&event_mutex);
3277 mutex_lock(&trace_types_lock);
3278 switch (val) {
3279 case MODULE_STATE_COMING:
3280 trace_module_add_events(mod);
3281 break;
3282 case MODULE_STATE_GOING:
3283 trace_module_remove_events(mod);
3284 break;
3285 }
3286 mutex_unlock(&trace_types_lock);
3287 mutex_unlock(&event_mutex);
3288
3289 return NOTIFY_OK;
3290}
3291
3292static struct notifier_block trace_module_nb = {
3293 .notifier_call = trace_module_notify,
3294 .priority = 1, /* higher than trace.c module notify */
3295};
3296#endif /* CONFIG_MODULES */
3297
3298/* Create a new event directory structure for a trace directory. */
3299static void
3300__trace_add_event_dirs(struct trace_array *tr)
3301{
3302 struct trace_event_call *call;
3303 int ret;
3304
3305 list_for_each_entry(call, &ftrace_events, list) {
3306 ret = __trace_add_new_event(call, tr);
3307 if (ret < 0)
3308 pr_warn("Could not create directory for event %s\n",
3309 trace_event_name(call));
3310 }
3311}
3312
3313/* Returns any file that matches the system and event */
3314struct trace_event_file *
3315__find_event_file(struct trace_array *tr, const char *system, const char *event)
3316{
3317 struct trace_event_file *file;
3318 struct trace_event_call *call;
3319 const char *name;
3320
3321 list_for_each_entry(file, &tr->events, list) {
3322
3323 call = file->event_call;
3324 name = trace_event_name(call);
3325
3326 if (!name || !call->class)
3327 continue;
3328
3329 if (strcmp(event, name) == 0 &&
3330 strcmp(system, call->class->system) == 0)
3331 return file;
3332 }
3333 return NULL;
3334}
3335
3336/* Returns valid trace event files that match system and event */
3337struct trace_event_file *
3338find_event_file(struct trace_array *tr, const char *system, const char *event)
3339{
3340 struct trace_event_file *file;
3341
3342 file = __find_event_file(tr, system, event);
3343 if (!file || !file->event_call->class->reg ||
3344 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3345 return NULL;
3346
3347 return file;
3348}
3349
3350/**
3351 * trace_get_event_file - Find and return a trace event file
3352 * @instance: The name of the trace instance containing the event
3353 * @system: The name of the system containing the event
3354 * @event: The name of the event
3355 *
3356 * Return a trace event file given the trace instance name, trace
3357 * system, and trace event name. If the instance name is NULL, it
3358 * refers to the top-level trace array.
3359 *
3360 * This function will look it up and return it if found, after calling
3361 * trace_array_get() to prevent the instance from going away, and
3362 * increment the event's module refcount to prevent it from being
3363 * removed.
3364 *
3365 * To release the file, call trace_put_event_file(), which will call
3366 * trace_array_put() and decrement the event's module refcount.
3367 *
3368 * Return: The trace event on success, ERR_PTR otherwise.
3369 */
3370struct trace_event_file *trace_get_event_file(const char *instance,
3371 const char *system,
3372 const char *event)
3373{
3374 struct trace_array *tr = top_trace_array();
3375 struct trace_event_file *file = NULL;
3376 int ret = -EINVAL;
3377
3378 if (instance) {
3379 tr = trace_array_find_get(instance);
3380 if (!tr)
3381 return ERR_PTR(-ENOENT);
3382 } else {
3383 ret = trace_array_get(tr);
3384 if (ret)
3385 return ERR_PTR(ret);
3386 }
3387
3388 mutex_lock(&event_mutex);
3389
3390 file = find_event_file(tr, system, event);
3391 if (!file) {
3392 trace_array_put(tr);
3393 ret = -EINVAL;
3394 goto out;
3395 }
3396
3397 /* Don't let event modules unload while in use */
3398 ret = trace_event_try_get_ref(file->event_call);
3399 if (!ret) {
3400 trace_array_put(tr);
3401 ret = -EBUSY;
3402 goto out;
3403 }
3404
3405 ret = 0;
3406 out:
3407 mutex_unlock(&event_mutex);
3408
3409 if (ret)
3410 file = ERR_PTR(ret);
3411
3412 return file;
3413}
3414EXPORT_SYMBOL_GPL(trace_get_event_file);
3415
3416/**
3417 * trace_put_event_file - Release a file from trace_get_event_file()
3418 * @file: The trace event file
3419 *
3420 * If a file was retrieved using trace_get_event_file(), this should
3421 * be called when it's no longer needed. It will cancel the previous
3422 * trace_array_get() called by that function, and decrement the
3423 * event's module refcount.
3424 */
3425void trace_put_event_file(struct trace_event_file *file)
3426{
3427 mutex_lock(&event_mutex);
3428 trace_event_put_ref(file->event_call);
3429 mutex_unlock(&event_mutex);
3430
3431 trace_array_put(file->tr);
3432}
3433EXPORT_SYMBOL_GPL(trace_put_event_file);
3434
3435#ifdef CONFIG_DYNAMIC_FTRACE
3436
3437/* Avoid typos */
3438#define ENABLE_EVENT_STR "enable_event"
3439#define DISABLE_EVENT_STR "disable_event"
3440
3441struct event_probe_data {
3442 struct trace_event_file *file;
3443 unsigned long count;
3444 int ref;
3445 bool enable;
3446};
3447
3448static void update_event_probe(struct event_probe_data *data)
3449{
3450 if (data->enable)
3451 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3452 else
3453 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3454}
3455
3456static void
3457event_enable_probe(unsigned long ip, unsigned long parent_ip,
3458 struct trace_array *tr, struct ftrace_probe_ops *ops,
3459 void *data)
3460{
3461 struct ftrace_func_mapper *mapper = data;
3462 struct event_probe_data *edata;
3463 void **pdata;
3464
3465 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3466 if (!pdata || !*pdata)
3467 return;
3468
3469 edata = *pdata;
3470 update_event_probe(edata);
3471}
3472
3473static void
3474event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3475 struct trace_array *tr, struct ftrace_probe_ops *ops,
3476 void *data)
3477{
3478 struct ftrace_func_mapper *mapper = data;
3479 struct event_probe_data *edata;
3480 void **pdata;
3481
3482 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3483 if (!pdata || !*pdata)
3484 return;
3485
3486 edata = *pdata;
3487
3488 if (!edata->count)
3489 return;
3490
3491 /* Skip if the event is in a state we want to switch to */
3492 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3493 return;
3494
3495 if (edata->count != -1)
3496 (edata->count)--;
3497
3498 update_event_probe(edata);
3499}
3500
3501static int
3502event_enable_print(struct seq_file *m, unsigned long ip,
3503 struct ftrace_probe_ops *ops, void *data)
3504{
3505 struct ftrace_func_mapper *mapper = data;
3506 struct event_probe_data *edata;
3507 void **pdata;
3508
3509 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3510
3511 if (WARN_ON_ONCE(!pdata || !*pdata))
3512 return 0;
3513
3514 edata = *pdata;
3515
3516 seq_printf(m, "%ps:", (void *)ip);
3517
3518 seq_printf(m, "%s:%s:%s",
3519 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3520 edata->file->event_call->class->system,
3521 trace_event_name(edata->file->event_call));
3522
3523 if (edata->count == -1)
3524 seq_puts(m, ":unlimited\n");
3525 else
3526 seq_printf(m, ":count=%ld\n", edata->count);
3527
3528 return 0;
3529}
3530
3531static int
3532event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3533 unsigned long ip, void *init_data, void **data)
3534{
3535 struct ftrace_func_mapper *mapper = *data;
3536 struct event_probe_data *edata = init_data;
3537 int ret;
3538
3539 if (!mapper) {
3540 mapper = allocate_ftrace_func_mapper();
3541 if (!mapper)
3542 return -ENODEV;
3543 *data = mapper;
3544 }
3545
3546 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3547 if (ret < 0)
3548 return ret;
3549
3550 edata->ref++;
3551
3552 return 0;
3553}
3554
3555static int free_probe_data(void *data)
3556{
3557 struct event_probe_data *edata = data;
3558
3559 edata->ref--;
3560 if (!edata->ref) {
3561 /* Remove the SOFT_MODE flag */
3562 __ftrace_event_enable_disable(edata->file, 0, 1);
3563 trace_event_put_ref(edata->file->event_call);
3564 kfree(edata);
3565 }
3566 return 0;
3567}
3568
3569static void
3570event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3571 unsigned long ip, void *data)
3572{
3573 struct ftrace_func_mapper *mapper = data;
3574 struct event_probe_data *edata;
3575
3576 if (!ip) {
3577 if (!mapper)
3578 return;
3579 free_ftrace_func_mapper(mapper, free_probe_data);
3580 return;
3581 }
3582
3583 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3584
3585 if (WARN_ON_ONCE(!edata))
3586 return;
3587
3588 if (WARN_ON_ONCE(edata->ref <= 0))
3589 return;
3590
3591 free_probe_data(edata);
3592}
3593
3594static struct ftrace_probe_ops event_enable_probe_ops = {
3595 .func = event_enable_probe,
3596 .print = event_enable_print,
3597 .init = event_enable_init,
3598 .free = event_enable_free,
3599};
3600
3601static struct ftrace_probe_ops event_enable_count_probe_ops = {
3602 .func = event_enable_count_probe,
3603 .print = event_enable_print,
3604 .init = event_enable_init,
3605 .free = event_enable_free,
3606};
3607
3608static struct ftrace_probe_ops event_disable_probe_ops = {
3609 .func = event_enable_probe,
3610 .print = event_enable_print,
3611 .init = event_enable_init,
3612 .free = event_enable_free,
3613};
3614
3615static struct ftrace_probe_ops event_disable_count_probe_ops = {
3616 .func = event_enable_count_probe,
3617 .print = event_enable_print,
3618 .init = event_enable_init,
3619 .free = event_enable_free,
3620};
3621
3622static int
3623event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3624 char *glob, char *cmd, char *param, int enabled)
3625{
3626 struct trace_event_file *file;
3627 struct ftrace_probe_ops *ops;
3628 struct event_probe_data *data;
3629 const char *system;
3630 const char *event;
3631 char *number;
3632 bool enable;
3633 int ret;
3634
3635 if (!tr)
3636 return -ENODEV;
3637
3638 /* hash funcs only work with set_ftrace_filter */
3639 if (!enabled || !param)
3640 return -EINVAL;
3641
3642 system = strsep(¶m, ":");
3643 if (!param)
3644 return -EINVAL;
3645
3646 event = strsep(¶m, ":");
3647
3648 mutex_lock(&event_mutex);
3649
3650 ret = -EINVAL;
3651 file = find_event_file(tr, system, event);
3652 if (!file)
3653 goto out;
3654
3655 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3656
3657 if (enable)
3658 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3659 else
3660 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3661
3662 if (glob[0] == '!') {
3663 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3664 goto out;
3665 }
3666
3667 ret = -ENOMEM;
3668
3669 data = kzalloc(sizeof(*data), GFP_KERNEL);
3670 if (!data)
3671 goto out;
3672
3673 data->enable = enable;
3674 data->count = -1;
3675 data->file = file;
3676
3677 if (!param)
3678 goto out_reg;
3679
3680 number = strsep(¶m, ":");
3681
3682 ret = -EINVAL;
3683 if (!strlen(number))
3684 goto out_free;
3685
3686 /*
3687 * We use the callback data field (which is a pointer)
3688 * as our counter.
3689 */
3690 ret = kstrtoul(number, 0, &data->count);
3691 if (ret)
3692 goto out_free;
3693
3694 out_reg:
3695 /* Don't let event modules unload while probe registered */
3696 ret = trace_event_try_get_ref(file->event_call);
3697 if (!ret) {
3698 ret = -EBUSY;
3699 goto out_free;
3700 }
3701
3702 ret = __ftrace_event_enable_disable(file, 1, 1);
3703 if (ret < 0)
3704 goto out_put;
3705
3706 ret = register_ftrace_function_probe(glob, tr, ops, data);
3707 /*
3708 * The above returns on success the # of functions enabled,
3709 * but if it didn't find any functions it returns zero.
3710 * Consider no functions a failure too.
3711 */
3712 if (!ret) {
3713 ret = -ENOENT;
3714 goto out_disable;
3715 } else if (ret < 0)
3716 goto out_disable;
3717 /* Just return zero, not the number of enabled functions */
3718 ret = 0;
3719 out:
3720 mutex_unlock(&event_mutex);
3721 return ret;
3722
3723 out_disable:
3724 __ftrace_event_enable_disable(file, 0, 1);
3725 out_put:
3726 trace_event_put_ref(file->event_call);
3727 out_free:
3728 kfree(data);
3729 goto out;
3730}
3731
3732static struct ftrace_func_command event_enable_cmd = {
3733 .name = ENABLE_EVENT_STR,
3734 .func = event_enable_func,
3735};
3736
3737static struct ftrace_func_command event_disable_cmd = {
3738 .name = DISABLE_EVENT_STR,
3739 .func = event_enable_func,
3740};
3741
3742static __init int register_event_cmds(void)
3743{
3744 int ret;
3745
3746 ret = register_ftrace_command(&event_enable_cmd);
3747 if (WARN_ON(ret < 0))
3748 return ret;
3749 ret = register_ftrace_command(&event_disable_cmd);
3750 if (WARN_ON(ret < 0))
3751 unregister_ftrace_command(&event_enable_cmd);
3752 return ret;
3753}
3754#else
3755static inline int register_event_cmds(void) { return 0; }
3756#endif /* CONFIG_DYNAMIC_FTRACE */
3757
3758/*
3759 * The top level array and trace arrays created by boot-time tracing
3760 * have already had its trace_event_file descriptors created in order
3761 * to allow for early events to be recorded.
3762 * This function is called after the tracefs has been initialized,
3763 * and we now have to create the files associated to the events.
3764 */
3765static void __trace_early_add_event_dirs(struct trace_array *tr)
3766{
3767 struct trace_event_file *file;
3768 int ret;
3769
3770
3771 list_for_each_entry(file, &tr->events, list) {
3772 ret = event_create_dir(tr->event_dir, file);
3773 if (ret < 0)
3774 pr_warn("Could not create directory for event %s\n",
3775 trace_event_name(file->event_call));
3776 }
3777}
3778
3779/*
3780 * For early boot up, the top trace array and the trace arrays created
3781 * by boot-time tracing require to have a list of events that can be
3782 * enabled. This must be done before the filesystem is set up in order
3783 * to allow events to be traced early.
3784 */
3785void __trace_early_add_events(struct trace_array *tr)
3786{
3787 struct trace_event_call *call;
3788 int ret;
3789
3790 list_for_each_entry(call, &ftrace_events, list) {
3791 /* Early boot up should not have any modules loaded */
3792 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3793 WARN_ON_ONCE(call->module))
3794 continue;
3795
3796 ret = __trace_early_add_new_event(call, tr);
3797 if (ret < 0)
3798 pr_warn("Could not create early event %s\n",
3799 trace_event_name(call));
3800 }
3801}
3802
3803/* Remove the event directory structure for a trace directory. */
3804static void
3805__trace_remove_event_dirs(struct trace_array *tr)
3806{
3807 struct trace_event_file *file, *next;
3808
3809 list_for_each_entry_safe(file, next, &tr->events, list)
3810 remove_event_file_dir(file);
3811}
3812
3813static void __add_event_to_tracers(struct trace_event_call *call)
3814{
3815 struct trace_array *tr;
3816
3817 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3818 __trace_add_new_event(call, tr);
3819}
3820
3821extern struct trace_event_call *__start_ftrace_events[];
3822extern struct trace_event_call *__stop_ftrace_events[];
3823
3824static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3825
3826static __init int setup_trace_event(char *str)
3827{
3828 strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3829 trace_set_ring_buffer_expanded(NULL);
3830 disable_tracing_selftest("running event tracing");
3831
3832 return 1;
3833}
3834__setup("trace_event=", setup_trace_event);
3835
3836static int events_callback(const char *name, umode_t *mode, void **data,
3837 const struct file_operations **fops)
3838{
3839 if (strcmp(name, "enable") == 0) {
3840 *mode = TRACE_MODE_WRITE;
3841 *fops = &ftrace_tr_enable_fops;
3842 return 1;
3843 }
3844
3845 if (strcmp(name, "header_page") == 0) {
3846 *mode = TRACE_MODE_READ;
3847 *fops = &ftrace_show_header_page_fops;
3848
3849 } else if (strcmp(name, "header_event") == 0) {
3850 *mode = TRACE_MODE_READ;
3851 *fops = &ftrace_show_header_event_fops;
3852 } else
3853 return 0;
3854
3855 return 1;
3856}
3857
3858/* Expects to have event_mutex held when called */
3859static int
3860create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3861{
3862 struct eventfs_inode *e_events;
3863 struct dentry *entry;
3864 int nr_entries;
3865 static struct eventfs_entry events_entries[] = {
3866 {
3867 .name = "enable",
3868 .callback = events_callback,
3869 },
3870 {
3871 .name = "header_page",
3872 .callback = events_callback,
3873 },
3874 {
3875 .name = "header_event",
3876 .callback = events_callback,
3877 },
3878 };
3879
3880 entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3881 tr, &ftrace_set_event_fops);
3882 if (!entry)
3883 return -ENOMEM;
3884
3885 nr_entries = ARRAY_SIZE(events_entries);
3886
3887 e_events = eventfs_create_events_dir("events", parent, events_entries,
3888 nr_entries, tr);
3889 if (IS_ERR(e_events)) {
3890 pr_warn("Could not create tracefs 'events' directory\n");
3891 return -ENOMEM;
3892 }
3893
3894 /* There are not as crucial, just warn if they are not created */
3895
3896 trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3897 tr, &ftrace_set_event_pid_fops);
3898
3899 trace_create_file("set_event_notrace_pid",
3900 TRACE_MODE_WRITE, parent, tr,
3901 &ftrace_set_event_notrace_pid_fops);
3902
3903 tr->event_dir = e_events;
3904
3905 return 0;
3906}
3907
3908/**
3909 * event_trace_add_tracer - add a instance of a trace_array to events
3910 * @parent: The parent dentry to place the files/directories for events in
3911 * @tr: The trace array associated with these events
3912 *
3913 * When a new instance is created, it needs to set up its events
3914 * directory, as well as other files associated with events. It also
3915 * creates the event hierarchy in the @parent/events directory.
3916 *
3917 * Returns 0 on success.
3918 *
3919 * Must be called with event_mutex held.
3920 */
3921int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3922{
3923 int ret;
3924
3925 lockdep_assert_held(&event_mutex);
3926
3927 ret = create_event_toplevel_files(parent, tr);
3928 if (ret)
3929 goto out;
3930
3931 down_write(&trace_event_sem);
3932 /* If tr already has the event list, it is initialized in early boot. */
3933 if (unlikely(!list_empty(&tr->events)))
3934 __trace_early_add_event_dirs(tr);
3935 else
3936 __trace_add_event_dirs(tr);
3937 up_write(&trace_event_sem);
3938
3939 out:
3940 return ret;
3941}
3942
3943/*
3944 * The top trace array already had its file descriptors created.
3945 * Now the files themselves need to be created.
3946 */
3947static __init int
3948early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3949{
3950 int ret;
3951
3952 mutex_lock(&event_mutex);
3953
3954 ret = create_event_toplevel_files(parent, tr);
3955 if (ret)
3956 goto out_unlock;
3957
3958 down_write(&trace_event_sem);
3959 __trace_early_add_event_dirs(tr);
3960 up_write(&trace_event_sem);
3961
3962 out_unlock:
3963 mutex_unlock(&event_mutex);
3964
3965 return ret;
3966}
3967
3968/* Must be called with event_mutex held */
3969int event_trace_del_tracer(struct trace_array *tr)
3970{
3971 lockdep_assert_held(&event_mutex);
3972
3973 /* Disable any event triggers and associated soft-disabled events */
3974 clear_event_triggers(tr);
3975
3976 /* Clear the pid list */
3977 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3978
3979 /* Disable any running events */
3980 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3981
3982 /* Make sure no more events are being executed */
3983 tracepoint_synchronize_unregister();
3984
3985 down_write(&trace_event_sem);
3986 __trace_remove_event_dirs(tr);
3987 eventfs_remove_events_dir(tr->event_dir);
3988 up_write(&trace_event_sem);
3989
3990 tr->event_dir = NULL;
3991
3992 return 0;
3993}
3994
3995static __init int event_trace_memsetup(void)
3996{
3997 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3998 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3999 return 0;
4000}
4001
4002__init void
4003early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4004{
4005 char *token;
4006 int ret;
4007
4008 while (true) {
4009 token = strsep(&buf, ",");
4010
4011 if (!token)
4012 break;
4013
4014 if (*token) {
4015 /* Restarting syscalls requires that we stop them first */
4016 if (disable_first)
4017 ftrace_set_clr_event(tr, token, 0);
4018
4019 ret = ftrace_set_clr_event(tr, token, 1);
4020 if (ret)
4021 pr_warn("Failed to enable trace event: %s\n", token);
4022 }
4023
4024 /* Put back the comma to allow this to be called again */
4025 if (buf)
4026 *(buf - 1) = ',';
4027 }
4028}
4029
4030static __init int event_trace_enable(void)
4031{
4032 struct trace_array *tr = top_trace_array();
4033 struct trace_event_call **iter, *call;
4034 int ret;
4035
4036 if (!tr)
4037 return -ENODEV;
4038
4039 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4040
4041 call = *iter;
4042 ret = event_init(call);
4043 if (!ret)
4044 list_add(&call->list, &ftrace_events);
4045 }
4046
4047 register_trigger_cmds();
4048
4049 /*
4050 * We need the top trace array to have a working set of trace
4051 * points at early init, before the debug files and directories
4052 * are created. Create the file entries now, and attach them
4053 * to the actual file dentries later.
4054 */
4055 __trace_early_add_events(tr);
4056
4057 early_enable_events(tr, bootup_event_buf, false);
4058
4059 trace_printk_start_comm();
4060
4061 register_event_cmds();
4062
4063
4064 return 0;
4065}
4066
4067/*
4068 * event_trace_enable() is called from trace_event_init() first to
4069 * initialize events and perhaps start any events that are on the
4070 * command line. Unfortunately, there are some events that will not
4071 * start this early, like the system call tracepoints that need
4072 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4073 * event_trace_enable() is called before pid 1 starts, and this flag
4074 * is never set, making the syscall tracepoint never get reached, but
4075 * the event is enabled regardless (and not doing anything).
4076 */
4077static __init int event_trace_enable_again(void)
4078{
4079 struct trace_array *tr;
4080
4081 tr = top_trace_array();
4082 if (!tr)
4083 return -ENODEV;
4084
4085 early_enable_events(tr, bootup_event_buf, true);
4086
4087 return 0;
4088}
4089
4090early_initcall(event_trace_enable_again);
4091
4092/* Init fields which doesn't related to the tracefs */
4093static __init int event_trace_init_fields(void)
4094{
4095 if (trace_define_generic_fields())
4096 pr_warn("tracing: Failed to allocated generic fields");
4097
4098 if (trace_define_common_fields())
4099 pr_warn("tracing: Failed to allocate common fields");
4100
4101 return 0;
4102}
4103
4104__init int event_trace_init(void)
4105{
4106 struct trace_array *tr;
4107 int ret;
4108
4109 tr = top_trace_array();
4110 if (!tr)
4111 return -ENODEV;
4112
4113 trace_create_file("available_events", TRACE_MODE_READ,
4114 NULL, tr, &ftrace_avail_fops);
4115
4116 ret = early_event_add_tracer(NULL, tr);
4117 if (ret)
4118 return ret;
4119
4120#ifdef CONFIG_MODULES
4121 ret = register_module_notifier(&trace_module_nb);
4122 if (ret)
4123 pr_warn("Failed to register trace events module notifier\n");
4124#endif
4125
4126 eventdir_initialized = true;
4127
4128 return 0;
4129}
4130
4131void __init trace_event_init(void)
4132{
4133 event_trace_memsetup();
4134 init_ftrace_syscalls();
4135 event_trace_enable();
4136 event_trace_init_fields();
4137}
4138
4139#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4140
4141static DEFINE_SPINLOCK(test_spinlock);
4142static DEFINE_SPINLOCK(test_spinlock_irq);
4143static DEFINE_MUTEX(test_mutex);
4144
4145static __init void test_work(struct work_struct *dummy)
4146{
4147 spin_lock(&test_spinlock);
4148 spin_lock_irq(&test_spinlock_irq);
4149 udelay(1);
4150 spin_unlock_irq(&test_spinlock_irq);
4151 spin_unlock(&test_spinlock);
4152
4153 mutex_lock(&test_mutex);
4154 msleep(1);
4155 mutex_unlock(&test_mutex);
4156}
4157
4158static __init int event_test_thread(void *unused)
4159{
4160 void *test_malloc;
4161
4162 test_malloc = kmalloc(1234, GFP_KERNEL);
4163 if (!test_malloc)
4164 pr_info("failed to kmalloc\n");
4165
4166 schedule_on_each_cpu(test_work);
4167
4168 kfree(test_malloc);
4169
4170 set_current_state(TASK_INTERRUPTIBLE);
4171 while (!kthread_should_stop()) {
4172 schedule();
4173 set_current_state(TASK_INTERRUPTIBLE);
4174 }
4175 __set_current_state(TASK_RUNNING);
4176
4177 return 0;
4178}
4179
4180/*
4181 * Do various things that may trigger events.
4182 */
4183static __init void event_test_stuff(void)
4184{
4185 struct task_struct *test_thread;
4186
4187 test_thread = kthread_run(event_test_thread, NULL, "test-events");
4188 msleep(1);
4189 kthread_stop(test_thread);
4190}
4191
4192/*
4193 * For every trace event defined, we will test each trace point separately,
4194 * and then by groups, and finally all trace points.
4195 */
4196static __init void event_trace_self_tests(void)
4197{
4198 struct trace_subsystem_dir *dir;
4199 struct trace_event_file *file;
4200 struct trace_event_call *call;
4201 struct event_subsystem *system;
4202 struct trace_array *tr;
4203 int ret;
4204
4205 tr = top_trace_array();
4206 if (!tr)
4207 return;
4208
4209 pr_info("Running tests on trace events:\n");
4210
4211 list_for_each_entry(file, &tr->events, list) {
4212
4213 call = file->event_call;
4214
4215 /* Only test those that have a probe */
4216 if (!call->class || !call->class->probe)
4217 continue;
4218
4219/*
4220 * Testing syscall events here is pretty useless, but
4221 * we still do it if configured. But this is time consuming.
4222 * What we really need is a user thread to perform the
4223 * syscalls as we test.
4224 */
4225#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4226 if (call->class->system &&
4227 strcmp(call->class->system, "syscalls") == 0)
4228 continue;
4229#endif
4230
4231 pr_info("Testing event %s: ", trace_event_name(call));
4232
4233 /*
4234 * If an event is already enabled, someone is using
4235 * it and the self test should not be on.
4236 */
4237 if (file->flags & EVENT_FILE_FL_ENABLED) {
4238 pr_warn("Enabled event during self test!\n");
4239 WARN_ON_ONCE(1);
4240 continue;
4241 }
4242
4243 ftrace_event_enable_disable(file, 1);
4244 event_test_stuff();
4245 ftrace_event_enable_disable(file, 0);
4246
4247 pr_cont("OK\n");
4248 }
4249
4250 /* Now test at the sub system level */
4251
4252 pr_info("Running tests on trace event systems:\n");
4253
4254 list_for_each_entry(dir, &tr->systems, list) {
4255
4256 system = dir->subsystem;
4257
4258 /* the ftrace system is special, skip it */
4259 if (strcmp(system->name, "ftrace") == 0)
4260 continue;
4261
4262 pr_info("Testing event system %s: ", system->name);
4263
4264 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4265 if (WARN_ON_ONCE(ret)) {
4266 pr_warn("error enabling system %s\n",
4267 system->name);
4268 continue;
4269 }
4270
4271 event_test_stuff();
4272
4273 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4274 if (WARN_ON_ONCE(ret)) {
4275 pr_warn("error disabling system %s\n",
4276 system->name);
4277 continue;
4278 }
4279
4280 pr_cont("OK\n");
4281 }
4282
4283 /* Test with all events enabled */
4284
4285 pr_info("Running tests on all trace events:\n");
4286 pr_info("Testing all events: ");
4287
4288 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4289 if (WARN_ON_ONCE(ret)) {
4290 pr_warn("error enabling all events\n");
4291 return;
4292 }
4293
4294 event_test_stuff();
4295
4296 /* reset sysname */
4297 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4298 if (WARN_ON_ONCE(ret)) {
4299 pr_warn("error disabling all events\n");
4300 return;
4301 }
4302
4303 pr_cont("OK\n");
4304}
4305
4306#ifdef CONFIG_FUNCTION_TRACER
4307
4308static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4309
4310static struct trace_event_file event_trace_file __initdata;
4311
4312static void __init
4313function_test_events_call(unsigned long ip, unsigned long parent_ip,
4314 struct ftrace_ops *op, struct ftrace_regs *regs)
4315{
4316 struct trace_buffer *buffer;
4317 struct ring_buffer_event *event;
4318 struct ftrace_entry *entry;
4319 unsigned int trace_ctx;
4320 long disabled;
4321 int cpu;
4322
4323 trace_ctx = tracing_gen_ctx();
4324 preempt_disable_notrace();
4325 cpu = raw_smp_processor_id();
4326 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4327
4328 if (disabled != 1)
4329 goto out;
4330
4331 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4332 TRACE_FN, sizeof(*entry),
4333 trace_ctx);
4334 if (!event)
4335 goto out;
4336 entry = ring_buffer_event_data(event);
4337 entry->ip = ip;
4338 entry->parent_ip = parent_ip;
4339
4340 event_trigger_unlock_commit(&event_trace_file, buffer, event,
4341 entry, trace_ctx);
4342 out:
4343 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4344 preempt_enable_notrace();
4345}
4346
4347static struct ftrace_ops trace_ops __initdata =
4348{
4349 .func = function_test_events_call,
4350};
4351
4352static __init void event_trace_self_test_with_function(void)
4353{
4354 int ret;
4355
4356 event_trace_file.tr = top_trace_array();
4357 if (WARN_ON(!event_trace_file.tr))
4358 return;
4359
4360 ret = register_ftrace_function(&trace_ops);
4361 if (WARN_ON(ret < 0)) {
4362 pr_info("Failed to enable function tracer for event tests\n");
4363 return;
4364 }
4365 pr_info("Running tests again, along with the function tracer\n");
4366 event_trace_self_tests();
4367 unregister_ftrace_function(&trace_ops);
4368}
4369#else
4370static __init void event_trace_self_test_with_function(void)
4371{
4372}
4373#endif
4374
4375static __init int event_trace_self_tests_init(void)
4376{
4377 if (!tracing_selftest_disabled) {
4378 event_trace_self_tests();
4379 event_trace_self_test_with_function();
4380 }
4381
4382 return 0;
4383}
4384
4385late_initcall(event_trace_self_tests_init);
4386
4387#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