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
1673static ssize_t
1674event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1675{
1676 int id = (long)event_file_data(filp);
1677 char buf[32];
1678 int len;
1679
1680 if (unlikely(!id))
1681 return -ENODEV;
1682
1683 len = sprintf(buf, "%d\n", id);
1684
1685 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1686}
1687
1688static ssize_t
1689event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1690 loff_t *ppos)
1691{
1692 struct trace_event_file *file;
1693 struct trace_seq *s;
1694 int r = -ENODEV;
1695
1696 if (*ppos)
1697 return 0;
1698
1699 s = kmalloc(sizeof(*s), GFP_KERNEL);
1700
1701 if (!s)
1702 return -ENOMEM;
1703
1704 trace_seq_init(s);
1705
1706 mutex_lock(&event_mutex);
1707 file = event_file_data(filp);
1708 if (file && !(file->flags & EVENT_FILE_FL_FREED))
1709 print_event_filter(file, s);
1710 mutex_unlock(&event_mutex);
1711
1712 if (file)
1713 r = simple_read_from_buffer(ubuf, cnt, ppos,
1714 s->buffer, trace_seq_used(s));
1715
1716 kfree(s);
1717
1718 return r;
1719}
1720
1721static ssize_t
1722event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1723 loff_t *ppos)
1724{
1725 struct trace_event_file *file;
1726 char *buf;
1727 int err = -ENODEV;
1728
1729 if (cnt >= PAGE_SIZE)
1730 return -EINVAL;
1731
1732 buf = memdup_user_nul(ubuf, cnt);
1733 if (IS_ERR(buf))
1734 return PTR_ERR(buf);
1735
1736 mutex_lock(&event_mutex);
1737 file = event_file_data(filp);
1738 if (file)
1739 err = apply_event_filter(file, buf);
1740 mutex_unlock(&event_mutex);
1741
1742 kfree(buf);
1743 if (err < 0)
1744 return err;
1745
1746 *ppos += cnt;
1747
1748 return cnt;
1749}
1750
1751static LIST_HEAD(event_subsystems);
1752
1753static int subsystem_open(struct inode *inode, struct file *filp)
1754{
1755 struct trace_subsystem_dir *dir = NULL, *iter_dir;
1756 struct trace_array *tr = NULL, *iter_tr;
1757 struct event_subsystem *system = NULL;
1758 int ret;
1759
1760 if (tracing_is_disabled())
1761 return -ENODEV;
1762
1763 /* Make sure the system still exists */
1764 mutex_lock(&event_mutex);
1765 mutex_lock(&trace_types_lock);
1766 list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1767 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1768 if (iter_dir == inode->i_private) {
1769 /* Don't open systems with no events */
1770 tr = iter_tr;
1771 dir = iter_dir;
1772 if (dir->nr_events) {
1773 __get_system_dir(dir);
1774 system = dir->subsystem;
1775 }
1776 goto exit_loop;
1777 }
1778 }
1779 }
1780 exit_loop:
1781 mutex_unlock(&trace_types_lock);
1782 mutex_unlock(&event_mutex);
1783
1784 if (!system)
1785 return -ENODEV;
1786
1787 /* Still need to increment the ref count of the system */
1788 if (trace_array_get(tr) < 0) {
1789 put_system(dir);
1790 return -ENODEV;
1791 }
1792
1793 ret = tracing_open_generic(inode, filp);
1794 if (ret < 0) {
1795 trace_array_put(tr);
1796 put_system(dir);
1797 }
1798
1799 return ret;
1800}
1801
1802static int system_tr_open(struct inode *inode, struct file *filp)
1803{
1804 struct trace_subsystem_dir *dir;
1805 struct trace_array *tr = inode->i_private;
1806 int ret;
1807
1808 /* Make a temporary dir that has no system but points to tr */
1809 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1810 if (!dir)
1811 return -ENOMEM;
1812
1813 ret = tracing_open_generic_tr(inode, filp);
1814 if (ret < 0) {
1815 kfree(dir);
1816 return ret;
1817 }
1818 dir->tr = tr;
1819 filp->private_data = dir;
1820
1821 return 0;
1822}
1823
1824static int subsystem_release(struct inode *inode, struct file *file)
1825{
1826 struct trace_subsystem_dir *dir = file->private_data;
1827
1828 trace_array_put(dir->tr);
1829
1830 /*
1831 * If dir->subsystem is NULL, then this is a temporary
1832 * descriptor that was made for a trace_array to enable
1833 * all subsystems.
1834 */
1835 if (dir->subsystem)
1836 put_system(dir);
1837 else
1838 kfree(dir);
1839
1840 return 0;
1841}
1842
1843static ssize_t
1844subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1845 loff_t *ppos)
1846{
1847 struct trace_subsystem_dir *dir = filp->private_data;
1848 struct event_subsystem *system = dir->subsystem;
1849 struct trace_seq *s;
1850 int r;
1851
1852 if (*ppos)
1853 return 0;
1854
1855 s = kmalloc(sizeof(*s), GFP_KERNEL);
1856 if (!s)
1857 return -ENOMEM;
1858
1859 trace_seq_init(s);
1860
1861 print_subsystem_event_filter(system, s);
1862 r = simple_read_from_buffer(ubuf, cnt, ppos,
1863 s->buffer, trace_seq_used(s));
1864
1865 kfree(s);
1866
1867 return r;
1868}
1869
1870static ssize_t
1871subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1872 loff_t *ppos)
1873{
1874 struct trace_subsystem_dir *dir = filp->private_data;
1875 char *buf;
1876 int err;
1877
1878 if (cnt >= PAGE_SIZE)
1879 return -EINVAL;
1880
1881 buf = memdup_user_nul(ubuf, cnt);
1882 if (IS_ERR(buf))
1883 return PTR_ERR(buf);
1884
1885 err = apply_subsystem_event_filter(dir, buf);
1886 kfree(buf);
1887 if (err < 0)
1888 return err;
1889
1890 *ppos += cnt;
1891
1892 return cnt;
1893}
1894
1895static ssize_t
1896show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1897{
1898 struct trace_array *tr = filp->private_data;
1899 struct trace_seq *s;
1900 int r;
1901
1902 if (*ppos)
1903 return 0;
1904
1905 s = kmalloc(sizeof(*s), GFP_KERNEL);
1906 if (!s)
1907 return -ENOMEM;
1908
1909 trace_seq_init(s);
1910
1911 ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1912 r = simple_read_from_buffer(ubuf, cnt, ppos,
1913 s->buffer, trace_seq_used(s));
1914
1915 kfree(s);
1916
1917 return r;
1918}
1919
1920static ssize_t
1921show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1922{
1923 struct trace_seq *s;
1924 int r;
1925
1926 if (*ppos)
1927 return 0;
1928
1929 s = kmalloc(sizeof(*s), GFP_KERNEL);
1930 if (!s)
1931 return -ENOMEM;
1932
1933 trace_seq_init(s);
1934
1935 ring_buffer_print_entry_header(s);
1936 r = simple_read_from_buffer(ubuf, cnt, ppos,
1937 s->buffer, trace_seq_used(s));
1938
1939 kfree(s);
1940
1941 return r;
1942}
1943
1944static void ignore_task_cpu(void *data)
1945{
1946 struct trace_array *tr = data;
1947 struct trace_pid_list *pid_list;
1948 struct trace_pid_list *no_pid_list;
1949
1950 /*
1951 * This function is called by on_each_cpu() while the
1952 * event_mutex is held.
1953 */
1954 pid_list = rcu_dereference_protected(tr->filtered_pids,
1955 mutex_is_locked(&event_mutex));
1956 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1957 mutex_is_locked(&event_mutex));
1958
1959 this_cpu_write(tr->array_buffer.data->ignore_pid,
1960 trace_ignore_this_task(pid_list, no_pid_list, current));
1961}
1962
1963static void register_pid_events(struct trace_array *tr)
1964{
1965 /*
1966 * Register a probe that is called before all other probes
1967 * to set ignore_pid if next or prev do not match.
1968 * Register a probe this is called after all other probes
1969 * to only keep ignore_pid set if next pid matches.
1970 */
1971 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1972 tr, INT_MAX);
1973 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1974 tr, 0);
1975
1976 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1977 tr, INT_MAX);
1978 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1979 tr, 0);
1980
1981 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1982 tr, INT_MAX);
1983 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1984 tr, 0);
1985
1986 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1987 tr, INT_MAX);
1988 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1989 tr, 0);
1990}
1991
1992static ssize_t
1993event_pid_write(struct file *filp, const char __user *ubuf,
1994 size_t cnt, loff_t *ppos, int type)
1995{
1996 struct seq_file *m = filp->private_data;
1997 struct trace_array *tr = m->private;
1998 struct trace_pid_list *filtered_pids = NULL;
1999 struct trace_pid_list *other_pids = NULL;
2000 struct trace_pid_list *pid_list;
2001 struct trace_event_file *file;
2002 ssize_t ret;
2003
2004 if (!cnt)
2005 return 0;
2006
2007 ret = tracing_update_buffers(tr);
2008 if (ret < 0)
2009 return ret;
2010
2011 mutex_lock(&event_mutex);
2012
2013 if (type == TRACE_PIDS) {
2014 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2015 lockdep_is_held(&event_mutex));
2016 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2017 lockdep_is_held(&event_mutex));
2018 } else {
2019 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2020 lockdep_is_held(&event_mutex));
2021 other_pids = rcu_dereference_protected(tr->filtered_pids,
2022 lockdep_is_held(&event_mutex));
2023 }
2024
2025 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2026 if (ret < 0)
2027 goto out;
2028
2029 if (type == TRACE_PIDS)
2030 rcu_assign_pointer(tr->filtered_pids, pid_list);
2031 else
2032 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2033
2034 list_for_each_entry(file, &tr->events, list) {
2035 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2036 }
2037
2038 if (filtered_pids) {
2039 tracepoint_synchronize_unregister();
2040 trace_pid_list_free(filtered_pids);
2041 } else if (pid_list && !other_pids) {
2042 register_pid_events(tr);
2043 }
2044
2045 /*
2046 * Ignoring of pids is done at task switch. But we have to
2047 * check for those tasks that are currently running.
2048 * Always do this in case a pid was appended or removed.
2049 */
2050 on_each_cpu(ignore_task_cpu, tr, 1);
2051
2052 out:
2053 mutex_unlock(&event_mutex);
2054
2055 if (ret > 0)
2056 *ppos += ret;
2057
2058 return ret;
2059}
2060
2061static ssize_t
2062ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2063 size_t cnt, loff_t *ppos)
2064{
2065 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2066}
2067
2068static ssize_t
2069ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2070 size_t cnt, loff_t *ppos)
2071{
2072 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2073}
2074
2075static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2076static int ftrace_event_set_open(struct inode *inode, struct file *file);
2077static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2078static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2079static int ftrace_event_release(struct inode *inode, struct file *file);
2080
2081static const struct seq_operations show_event_seq_ops = {
2082 .start = t_start,
2083 .next = t_next,
2084 .show = t_show,
2085 .stop = t_stop,
2086};
2087
2088static const struct seq_operations show_set_event_seq_ops = {
2089 .start = s_start,
2090 .next = s_next,
2091 .show = t_show,
2092 .stop = t_stop,
2093};
2094
2095static const struct seq_operations show_set_pid_seq_ops = {
2096 .start = p_start,
2097 .next = p_next,
2098 .show = trace_pid_show,
2099 .stop = p_stop,
2100};
2101
2102static const struct seq_operations show_set_no_pid_seq_ops = {
2103 .start = np_start,
2104 .next = np_next,
2105 .show = trace_pid_show,
2106 .stop = p_stop,
2107};
2108
2109static const struct file_operations ftrace_avail_fops = {
2110 .open = ftrace_event_avail_open,
2111 .read = seq_read,
2112 .llseek = seq_lseek,
2113 .release = seq_release,
2114};
2115
2116static const struct file_operations ftrace_set_event_fops = {
2117 .open = ftrace_event_set_open,
2118 .read = seq_read,
2119 .write = ftrace_event_write,
2120 .llseek = seq_lseek,
2121 .release = ftrace_event_release,
2122};
2123
2124static const struct file_operations ftrace_set_event_pid_fops = {
2125 .open = ftrace_event_set_pid_open,
2126 .read = seq_read,
2127 .write = ftrace_event_pid_write,
2128 .llseek = seq_lseek,
2129 .release = ftrace_event_release,
2130};
2131
2132static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2133 .open = ftrace_event_set_npid_open,
2134 .read = seq_read,
2135 .write = ftrace_event_npid_write,
2136 .llseek = seq_lseek,
2137 .release = ftrace_event_release,
2138};
2139
2140static const struct file_operations ftrace_enable_fops = {
2141 .open = tracing_open_file_tr,
2142 .read = event_enable_read,
2143 .write = event_enable_write,
2144 .release = tracing_release_file_tr,
2145 .llseek = default_llseek,
2146};
2147
2148static const struct file_operations ftrace_event_format_fops = {
2149 .open = trace_format_open,
2150 .read = seq_read,
2151 .llseek = seq_lseek,
2152 .release = seq_release,
2153};
2154
2155static const struct file_operations ftrace_event_id_fops = {
2156 .read = event_id_read,
2157 .llseek = default_llseek,
2158};
2159
2160static const struct file_operations ftrace_event_filter_fops = {
2161 .open = tracing_open_file_tr,
2162 .read = event_filter_read,
2163 .write = event_filter_write,
2164 .release = tracing_release_file_tr,
2165 .llseek = default_llseek,
2166};
2167
2168static const struct file_operations ftrace_subsystem_filter_fops = {
2169 .open = subsystem_open,
2170 .read = subsystem_filter_read,
2171 .write = subsystem_filter_write,
2172 .llseek = default_llseek,
2173 .release = subsystem_release,
2174};
2175
2176static const struct file_operations ftrace_system_enable_fops = {
2177 .open = subsystem_open,
2178 .read = system_enable_read,
2179 .write = system_enable_write,
2180 .llseek = default_llseek,
2181 .release = subsystem_release,
2182};
2183
2184static const struct file_operations ftrace_tr_enable_fops = {
2185 .open = system_tr_open,
2186 .read = system_enable_read,
2187 .write = system_enable_write,
2188 .llseek = default_llseek,
2189 .release = subsystem_release,
2190};
2191
2192static const struct file_operations ftrace_show_header_page_fops = {
2193 .open = tracing_open_generic_tr,
2194 .read = show_header_page_file,
2195 .llseek = default_llseek,
2196 .release = tracing_release_generic_tr,
2197};
2198
2199static const struct file_operations ftrace_show_header_event_fops = {
2200 .open = tracing_open_generic_tr,
2201 .read = show_header_event_file,
2202 .llseek = default_llseek,
2203 .release = tracing_release_generic_tr,
2204};
2205
2206static int
2207ftrace_event_open(struct inode *inode, struct file *file,
2208 const struct seq_operations *seq_ops)
2209{
2210 struct seq_file *m;
2211 int ret;
2212
2213 ret = security_locked_down(LOCKDOWN_TRACEFS);
2214 if (ret)
2215 return ret;
2216
2217 ret = seq_open(file, seq_ops);
2218 if (ret < 0)
2219 return ret;
2220 m = file->private_data;
2221 /* copy tr over to seq ops */
2222 m->private = inode->i_private;
2223
2224 return ret;
2225}
2226
2227static int ftrace_event_release(struct inode *inode, struct file *file)
2228{
2229 struct trace_array *tr = inode->i_private;
2230
2231 trace_array_put(tr);
2232
2233 return seq_release(inode, file);
2234}
2235
2236static int
2237ftrace_event_avail_open(struct inode *inode, struct file *file)
2238{
2239 const struct seq_operations *seq_ops = &show_event_seq_ops;
2240
2241 /* Checks for tracefs lockdown */
2242 return ftrace_event_open(inode, file, seq_ops);
2243}
2244
2245static int
2246ftrace_event_set_open(struct inode *inode, struct file *file)
2247{
2248 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2249 struct trace_array *tr = inode->i_private;
2250 int ret;
2251
2252 ret = tracing_check_open_get_tr(tr);
2253 if (ret)
2254 return ret;
2255
2256 if ((file->f_mode & FMODE_WRITE) &&
2257 (file->f_flags & O_TRUNC))
2258 ftrace_clear_events(tr);
2259
2260 ret = ftrace_event_open(inode, file, seq_ops);
2261 if (ret < 0)
2262 trace_array_put(tr);
2263 return ret;
2264}
2265
2266static int
2267ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2268{
2269 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2270 struct trace_array *tr = inode->i_private;
2271 int ret;
2272
2273 ret = tracing_check_open_get_tr(tr);
2274 if (ret)
2275 return ret;
2276
2277 if ((file->f_mode & FMODE_WRITE) &&
2278 (file->f_flags & O_TRUNC))
2279 ftrace_clear_event_pids(tr, TRACE_PIDS);
2280
2281 ret = ftrace_event_open(inode, file, seq_ops);
2282 if (ret < 0)
2283 trace_array_put(tr);
2284 return ret;
2285}
2286
2287static int
2288ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2289{
2290 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2291 struct trace_array *tr = inode->i_private;
2292 int ret;
2293
2294 ret = tracing_check_open_get_tr(tr);
2295 if (ret)
2296 return ret;
2297
2298 if ((file->f_mode & FMODE_WRITE) &&
2299 (file->f_flags & O_TRUNC))
2300 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2301
2302 ret = ftrace_event_open(inode, file, seq_ops);
2303 if (ret < 0)
2304 trace_array_put(tr);
2305 return ret;
2306}
2307
2308static struct event_subsystem *
2309create_new_subsystem(const char *name)
2310{
2311 struct event_subsystem *system;
2312
2313 /* need to create new entry */
2314 system = kmalloc(sizeof(*system), GFP_KERNEL);
2315 if (!system)
2316 return NULL;
2317
2318 system->ref_count = 1;
2319
2320 /* Only allocate if dynamic (kprobes and modules) */
2321 system->name = kstrdup_const(name, GFP_KERNEL);
2322 if (!system->name)
2323 goto out_free;
2324
2325 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2326 if (!system->filter)
2327 goto out_free;
2328
2329 list_add(&system->list, &event_subsystems);
2330
2331 return system;
2332
2333 out_free:
2334 kfree_const(system->name);
2335 kfree(system);
2336 return NULL;
2337}
2338
2339static int system_callback(const char *name, umode_t *mode, void **data,
2340 const struct file_operations **fops)
2341{
2342 if (strcmp(name, "filter") == 0)
2343 *fops = &ftrace_subsystem_filter_fops;
2344
2345 else if (strcmp(name, "enable") == 0)
2346 *fops = &ftrace_system_enable_fops;
2347
2348 else
2349 return 0;
2350
2351 *mode = TRACE_MODE_WRITE;
2352 return 1;
2353}
2354
2355static struct eventfs_inode *
2356event_subsystem_dir(struct trace_array *tr, const char *name,
2357 struct trace_event_file *file, struct eventfs_inode *parent)
2358{
2359 struct event_subsystem *system, *iter;
2360 struct trace_subsystem_dir *dir;
2361 struct eventfs_inode *ei;
2362 int nr_entries;
2363 static struct eventfs_entry system_entries[] = {
2364 {
2365 .name = "filter",
2366 .callback = system_callback,
2367 },
2368 {
2369 .name = "enable",
2370 .callback = system_callback,
2371 }
2372 };
2373
2374 /* First see if we did not already create this dir */
2375 list_for_each_entry(dir, &tr->systems, list) {
2376 system = dir->subsystem;
2377 if (strcmp(system->name, name) == 0) {
2378 dir->nr_events++;
2379 file->system = dir;
2380 return dir->ei;
2381 }
2382 }
2383
2384 /* Now see if the system itself exists. */
2385 system = NULL;
2386 list_for_each_entry(iter, &event_subsystems, list) {
2387 if (strcmp(iter->name, name) == 0) {
2388 system = iter;
2389 break;
2390 }
2391 }
2392
2393 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2394 if (!dir)
2395 goto out_fail;
2396
2397 if (!system) {
2398 system = create_new_subsystem(name);
2399 if (!system)
2400 goto out_free;
2401 } else
2402 __get_system(system);
2403
2404 /* ftrace only has directories no files */
2405 if (strcmp(name, "ftrace") == 0)
2406 nr_entries = 0;
2407 else
2408 nr_entries = ARRAY_SIZE(system_entries);
2409
2410 ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2411 if (IS_ERR(ei)) {
2412 pr_warn("Failed to create system directory %s\n", name);
2413 __put_system(system);
2414 goto out_free;
2415 }
2416
2417 dir->ei = ei;
2418 dir->tr = tr;
2419 dir->ref_count = 1;
2420 dir->nr_events = 1;
2421 dir->subsystem = system;
2422 file->system = dir;
2423
2424 list_add(&dir->list, &tr->systems);
2425
2426 return dir->ei;
2427
2428 out_free:
2429 kfree(dir);
2430 out_fail:
2431 /* Only print this message if failed on memory allocation */
2432 if (!dir || !system)
2433 pr_warn("No memory to create event subsystem %s\n", name);
2434 return NULL;
2435}
2436
2437static int
2438event_define_fields(struct trace_event_call *call)
2439{
2440 struct list_head *head;
2441 int ret = 0;
2442
2443 /*
2444 * Other events may have the same class. Only update
2445 * the fields if they are not already defined.
2446 */
2447 head = trace_get_fields(call);
2448 if (list_empty(head)) {
2449 struct trace_event_fields *field = call->class->fields_array;
2450 unsigned int offset = sizeof(struct trace_entry);
2451
2452 for (; field->type; field++) {
2453 if (field->type == TRACE_FUNCTION_TYPE) {
2454 field->define_fields(call);
2455 break;
2456 }
2457
2458 offset = ALIGN(offset, field->align);
2459 ret = trace_define_field_ext(call, field->type, field->name,
2460 offset, field->size,
2461 field->is_signed, field->filter_type,
2462 field->len);
2463 if (WARN_ON_ONCE(ret)) {
2464 pr_err("error code is %d\n", ret);
2465 break;
2466 }
2467
2468 offset += field->size;
2469 }
2470 }
2471
2472 return ret;
2473}
2474
2475static int event_callback(const char *name, umode_t *mode, void **data,
2476 const struct file_operations **fops)
2477{
2478 struct trace_event_file *file = *data;
2479 struct trace_event_call *call = file->event_call;
2480
2481 if (strcmp(name, "format") == 0) {
2482 *mode = TRACE_MODE_READ;
2483 *fops = &ftrace_event_format_fops;
2484 *data = call;
2485 return 1;
2486 }
2487
2488 /*
2489 * Only event directories that can be enabled should have
2490 * triggers or filters, with the exception of the "print"
2491 * event that can have a "trigger" file.
2492 */
2493 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2494 if (call->class->reg && strcmp(name, "enable") == 0) {
2495 *mode = TRACE_MODE_WRITE;
2496 *fops = &ftrace_enable_fops;
2497 return 1;
2498 }
2499
2500 if (strcmp(name, "filter") == 0) {
2501 *mode = TRACE_MODE_WRITE;
2502 *fops = &ftrace_event_filter_fops;
2503 return 1;
2504 }
2505 }
2506
2507 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2508 strcmp(trace_event_name(call), "print") == 0) {
2509 if (strcmp(name, "trigger") == 0) {
2510 *mode = TRACE_MODE_WRITE;
2511 *fops = &event_trigger_fops;
2512 return 1;
2513 }
2514 }
2515
2516#ifdef CONFIG_PERF_EVENTS
2517 if (call->event.type && call->class->reg &&
2518 strcmp(name, "id") == 0) {
2519 *mode = TRACE_MODE_READ;
2520 *data = (void *)(long)call->event.type;
2521 *fops = &ftrace_event_id_fops;
2522 return 1;
2523 }
2524#endif
2525
2526#ifdef CONFIG_HIST_TRIGGERS
2527 if (strcmp(name, "hist") == 0) {
2528 *mode = TRACE_MODE_READ;
2529 *fops = &event_hist_fops;
2530 return 1;
2531 }
2532#endif
2533#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2534 if (strcmp(name, "hist_debug") == 0) {
2535 *mode = TRACE_MODE_READ;
2536 *fops = &event_hist_debug_fops;
2537 return 1;
2538 }
2539#endif
2540#ifdef CONFIG_TRACE_EVENT_INJECT
2541 if (call->event.type && call->class->reg &&
2542 strcmp(name, "inject") == 0) {
2543 *mode = 0200;
2544 *fops = &event_inject_fops;
2545 return 1;
2546 }
2547#endif
2548 return 0;
2549}
2550
2551static int
2552event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2553{
2554 struct trace_event_call *call = file->event_call;
2555 struct trace_array *tr = file->tr;
2556 struct eventfs_inode *e_events;
2557 struct eventfs_inode *ei;
2558 const char *name;
2559 int nr_entries;
2560 int ret;
2561 static struct eventfs_entry event_entries[] = {
2562 {
2563 .name = "enable",
2564 .callback = event_callback,
2565 },
2566 {
2567 .name = "filter",
2568 .callback = event_callback,
2569 },
2570 {
2571 .name = "trigger",
2572 .callback = event_callback,
2573 },
2574 {
2575 .name = "format",
2576 .callback = event_callback,
2577 },
2578#ifdef CONFIG_PERF_EVENTS
2579 {
2580 .name = "id",
2581 .callback = event_callback,
2582 },
2583#endif
2584#ifdef CONFIG_HIST_TRIGGERS
2585 {
2586 .name = "hist",
2587 .callback = event_callback,
2588 },
2589#endif
2590#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2591 {
2592 .name = "hist_debug",
2593 .callback = event_callback,
2594 },
2595#endif
2596#ifdef CONFIG_TRACE_EVENT_INJECT
2597 {
2598 .name = "inject",
2599 .callback = event_callback,
2600 },
2601#endif
2602 };
2603
2604 /*
2605 * If the trace point header did not define TRACE_SYSTEM
2606 * then the system would be called "TRACE_SYSTEM". This should
2607 * never happen.
2608 */
2609 if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2610 return -ENODEV;
2611
2612 e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2613 if (!e_events)
2614 return -ENOMEM;
2615
2616 nr_entries = ARRAY_SIZE(event_entries);
2617
2618 name = trace_event_name(call);
2619 ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2620 if (IS_ERR(ei)) {
2621 pr_warn("Could not create tracefs '%s' directory\n", name);
2622 return -1;
2623 }
2624
2625 file->ei = ei;
2626
2627 ret = event_define_fields(call);
2628 if (ret < 0) {
2629 pr_warn("Could not initialize trace point events/%s\n", name);
2630 return ret;
2631 }
2632
2633 return 0;
2634}
2635
2636static void remove_event_from_tracers(struct trace_event_call *call)
2637{
2638 struct trace_event_file *file;
2639 struct trace_array *tr;
2640
2641 do_for_each_event_file_safe(tr, file) {
2642 if (file->event_call != call)
2643 continue;
2644
2645 remove_event_file_dir(file);
2646 /*
2647 * The do_for_each_event_file_safe() is
2648 * a double loop. After finding the call for this
2649 * trace_array, we use break to jump to the next
2650 * trace_array.
2651 */
2652 break;
2653 } while_for_each_event_file();
2654}
2655
2656static void event_remove(struct trace_event_call *call)
2657{
2658 struct trace_array *tr;
2659 struct trace_event_file *file;
2660
2661 do_for_each_event_file(tr, file) {
2662 if (file->event_call != call)
2663 continue;
2664
2665 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2666 tr->clear_trace = true;
2667
2668 ftrace_event_enable_disable(file, 0);
2669 /*
2670 * The do_for_each_event_file() is
2671 * a double loop. After finding the call for this
2672 * trace_array, we use break to jump to the next
2673 * trace_array.
2674 */
2675 break;
2676 } while_for_each_event_file();
2677
2678 if (call->event.funcs)
2679 __unregister_trace_event(&call->event);
2680 remove_event_from_tracers(call);
2681 list_del(&call->list);
2682}
2683
2684static int event_init(struct trace_event_call *call)
2685{
2686 int ret = 0;
2687 const char *name;
2688
2689 name = trace_event_name(call);
2690 if (WARN_ON(!name))
2691 return -EINVAL;
2692
2693 if (call->class->raw_init) {
2694 ret = call->class->raw_init(call);
2695 if (ret < 0 && ret != -ENOSYS)
2696 pr_warn("Could not initialize trace events/%s\n", name);
2697 }
2698
2699 return ret;
2700}
2701
2702static int
2703__register_event(struct trace_event_call *call, struct module *mod)
2704{
2705 int ret;
2706
2707 ret = event_init(call);
2708 if (ret < 0)
2709 return ret;
2710
2711 list_add(&call->list, &ftrace_events);
2712 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2713 atomic_set(&call->refcnt, 0);
2714 else
2715 call->module = mod;
2716
2717 return 0;
2718}
2719
2720static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2721{
2722 int rlen;
2723 int elen;
2724
2725 /* Find the length of the eval value as a string */
2726 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2727 /* Make sure there's enough room to replace the string with the value */
2728 if (len < elen)
2729 return NULL;
2730
2731 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2732
2733 /* Get the rest of the string of ptr */
2734 rlen = strlen(ptr + len);
2735 memmove(ptr + elen, ptr + len, rlen);
2736 /* Make sure we end the new string */
2737 ptr[elen + rlen] = 0;
2738
2739 return ptr + elen;
2740}
2741
2742static void update_event_printk(struct trace_event_call *call,
2743 struct trace_eval_map *map)
2744{
2745 char *ptr;
2746 int quote = 0;
2747 int len = strlen(map->eval_string);
2748
2749 for (ptr = call->print_fmt; *ptr; ptr++) {
2750 if (*ptr == '\\') {
2751 ptr++;
2752 /* paranoid */
2753 if (!*ptr)
2754 break;
2755 continue;
2756 }
2757 if (*ptr == '"') {
2758 quote ^= 1;
2759 continue;
2760 }
2761 if (quote)
2762 continue;
2763 if (isdigit(*ptr)) {
2764 /* skip numbers */
2765 do {
2766 ptr++;
2767 /* Check for alpha chars like ULL */
2768 } while (isalnum(*ptr));
2769 if (!*ptr)
2770 break;
2771 /*
2772 * A number must have some kind of delimiter after
2773 * it, and we can ignore that too.
2774 */
2775 continue;
2776 }
2777 if (isalpha(*ptr) || *ptr == '_') {
2778 if (strncmp(map->eval_string, ptr, len) == 0 &&
2779 !isalnum(ptr[len]) && ptr[len] != '_') {
2780 ptr = eval_replace(ptr, map, len);
2781 /* enum/sizeof string smaller than value */
2782 if (WARN_ON_ONCE(!ptr))
2783 return;
2784 /*
2785 * No need to decrement here, as eval_replace()
2786 * returns the pointer to the character passed
2787 * the eval, and two evals can not be placed
2788 * back to back without something in between.
2789 * We can skip that something in between.
2790 */
2791 continue;
2792 }
2793 skip_more:
2794 do {
2795 ptr++;
2796 } while (isalnum(*ptr) || *ptr == '_');
2797 if (!*ptr)
2798 break;
2799 /*
2800 * If what comes after this variable is a '.' or
2801 * '->' then we can continue to ignore that string.
2802 */
2803 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2804 ptr += *ptr == '.' ? 1 : 2;
2805 if (!*ptr)
2806 break;
2807 goto skip_more;
2808 }
2809 /*
2810 * Once again, we can skip the delimiter that came
2811 * after the string.
2812 */
2813 continue;
2814 }
2815 }
2816}
2817
2818static void add_str_to_module(struct module *module, char *str)
2819{
2820 struct module_string *modstr;
2821
2822 modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2823
2824 /*
2825 * If we failed to allocate memory here, then we'll just
2826 * let the str memory leak when the module is removed.
2827 * If this fails to allocate, there's worse problems than
2828 * a leaked string on module removal.
2829 */
2830 if (WARN_ON_ONCE(!modstr))
2831 return;
2832
2833 modstr->module = module;
2834 modstr->str = str;
2835
2836 list_add(&modstr->next, &module_strings);
2837}
2838
2839static void update_event_fields(struct trace_event_call *call,
2840 struct trace_eval_map *map)
2841{
2842 struct ftrace_event_field *field;
2843 struct list_head *head;
2844 char *ptr;
2845 char *str;
2846 int len = strlen(map->eval_string);
2847
2848 /* Dynamic events should never have field maps */
2849 if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2850 return;
2851
2852 head = trace_get_fields(call);
2853 list_for_each_entry(field, head, link) {
2854 ptr = strchr(field->type, '[');
2855 if (!ptr)
2856 continue;
2857 ptr++;
2858
2859 if (!isalpha(*ptr) && *ptr != '_')
2860 continue;
2861
2862 if (strncmp(map->eval_string, ptr, len) != 0)
2863 continue;
2864
2865 str = kstrdup(field->type, GFP_KERNEL);
2866 if (WARN_ON_ONCE(!str))
2867 return;
2868 ptr = str + (ptr - field->type);
2869 ptr = eval_replace(ptr, map, len);
2870 /* enum/sizeof string smaller than value */
2871 if (WARN_ON_ONCE(!ptr)) {
2872 kfree(str);
2873 continue;
2874 }
2875
2876 /*
2877 * If the event is part of a module, then we need to free the string
2878 * when the module is removed. Otherwise, it will stay allocated
2879 * until a reboot.
2880 */
2881 if (call->module)
2882 add_str_to_module(call->module, str);
2883
2884 field->type = str;
2885 }
2886}
2887
2888void trace_event_eval_update(struct trace_eval_map **map, int len)
2889{
2890 struct trace_event_call *call, *p;
2891 const char *last_system = NULL;
2892 bool first = false;
2893 int last_i;
2894 int i;
2895
2896 down_write(&trace_event_sem);
2897 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2898 /* events are usually grouped together with systems */
2899 if (!last_system || call->class->system != last_system) {
2900 first = true;
2901 last_i = 0;
2902 last_system = call->class->system;
2903 }
2904
2905 /*
2906 * Since calls are grouped by systems, the likelihood that the
2907 * next call in the iteration belongs to the same system as the
2908 * previous call is high. As an optimization, we skip searching
2909 * for a map[] that matches the call's system if the last call
2910 * was from the same system. That's what last_i is for. If the
2911 * call has the same system as the previous call, then last_i
2912 * will be the index of the first map[] that has a matching
2913 * system.
2914 */
2915 for (i = last_i; i < len; i++) {
2916 if (call->class->system == map[i]->system) {
2917 /* Save the first system if need be */
2918 if (first) {
2919 last_i = i;
2920 first = false;
2921 }
2922 update_event_printk(call, map[i]);
2923 update_event_fields(call, map[i]);
2924 }
2925 }
2926 cond_resched();
2927 }
2928 up_write(&trace_event_sem);
2929}
2930
2931static bool event_in_systems(struct trace_event_call *call,
2932 const char *systems)
2933{
2934 const char *system;
2935 const char *p;
2936
2937 if (!systems)
2938 return true;
2939
2940 system = call->class->system;
2941 p = strstr(systems, system);
2942 if (!p)
2943 return false;
2944
2945 if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2946 return false;
2947
2948 p += strlen(system);
2949 return !*p || isspace(*p) || *p == ',';
2950}
2951
2952static struct trace_event_file *
2953trace_create_new_event(struct trace_event_call *call,
2954 struct trace_array *tr)
2955{
2956 struct trace_pid_list *no_pid_list;
2957 struct trace_pid_list *pid_list;
2958 struct trace_event_file *file;
2959 unsigned int first;
2960
2961 if (!event_in_systems(call, tr->system_names))
2962 return NULL;
2963
2964 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2965 if (!file)
2966 return ERR_PTR(-ENOMEM);
2967
2968 pid_list = rcu_dereference_protected(tr->filtered_pids,
2969 lockdep_is_held(&event_mutex));
2970 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2971 lockdep_is_held(&event_mutex));
2972
2973 if (!trace_pid_list_first(pid_list, &first) ||
2974 !trace_pid_list_first(no_pid_list, &first))
2975 file->flags |= EVENT_FILE_FL_PID_FILTER;
2976
2977 file->event_call = call;
2978 file->tr = tr;
2979 atomic_set(&file->sm_ref, 0);
2980 atomic_set(&file->tm_ref, 0);
2981 INIT_LIST_HEAD(&file->triggers);
2982 list_add(&file->list, &tr->events);
2983 event_file_get(file);
2984
2985 return file;
2986}
2987
2988#define MAX_BOOT_TRIGGERS 32
2989
2990static struct boot_triggers {
2991 const char *event;
2992 char *trigger;
2993} bootup_triggers[MAX_BOOT_TRIGGERS];
2994
2995static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2996static int nr_boot_triggers;
2997
2998static __init int setup_trace_triggers(char *str)
2999{
3000 char *trigger;
3001 char *buf;
3002 int i;
3003
3004 strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3005 trace_set_ring_buffer_expanded(NULL);
3006 disable_tracing_selftest("running event triggers");
3007
3008 buf = bootup_trigger_buf;
3009 for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3010 trigger = strsep(&buf, ",");
3011 if (!trigger)
3012 break;
3013 bootup_triggers[i].event = strsep(&trigger, ".");
3014 bootup_triggers[i].trigger = trigger;
3015 if (!bootup_triggers[i].trigger)
3016 break;
3017 }
3018
3019 nr_boot_triggers = i;
3020 return 1;
3021}
3022__setup("trace_trigger=", setup_trace_triggers);
3023
3024/* Add an event to a trace directory */
3025static int
3026__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3027{
3028 struct trace_event_file *file;
3029
3030 file = trace_create_new_event(call, tr);
3031 /*
3032 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3033 * allocation, or NULL if the event is not part of the tr->system_names.
3034 * When the event is not part of the tr->system_names, return zero, not
3035 * an error.
3036 */
3037 if (!file)
3038 return 0;
3039
3040 if (IS_ERR(file))
3041 return PTR_ERR(file);
3042
3043 if (eventdir_initialized)
3044 return event_create_dir(tr->event_dir, file);
3045 else
3046 return event_define_fields(call);
3047}
3048
3049static void trace_early_triggers(struct trace_event_file *file, const char *name)
3050{
3051 int ret;
3052 int i;
3053
3054 for (i = 0; i < nr_boot_triggers; i++) {
3055 if (strcmp(name, bootup_triggers[i].event))
3056 continue;
3057 mutex_lock(&event_mutex);
3058 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3059 mutex_unlock(&event_mutex);
3060 if (ret)
3061 pr_err("Failed to register trigger '%s' on event %s\n",
3062 bootup_triggers[i].trigger,
3063 bootup_triggers[i].event);
3064 }
3065}
3066
3067/*
3068 * Just create a descriptor for early init. A descriptor is required
3069 * for enabling events at boot. We want to enable events before
3070 * the filesystem is initialized.
3071 */
3072static int
3073__trace_early_add_new_event(struct trace_event_call *call,
3074 struct trace_array *tr)
3075{
3076 struct trace_event_file *file;
3077 int ret;
3078
3079 file = trace_create_new_event(call, tr);
3080 /*
3081 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3082 * allocation, or NULL if the event is not part of the tr->system_names.
3083 * When the event is not part of the tr->system_names, return zero, not
3084 * an error.
3085 */
3086 if (!file)
3087 return 0;
3088
3089 if (IS_ERR(file))
3090 return PTR_ERR(file);
3091
3092 ret = event_define_fields(call);
3093 if (ret)
3094 return ret;
3095
3096 trace_early_triggers(file, trace_event_name(call));
3097
3098 return 0;
3099}
3100
3101struct ftrace_module_file_ops;
3102static void __add_event_to_tracers(struct trace_event_call *call);
3103
3104/* Add an additional event_call dynamically */
3105int trace_add_event_call(struct trace_event_call *call)
3106{
3107 int ret;
3108 lockdep_assert_held(&event_mutex);
3109
3110 mutex_lock(&trace_types_lock);
3111
3112 ret = __register_event(call, NULL);
3113 if (ret >= 0)
3114 __add_event_to_tracers(call);
3115
3116 mutex_unlock(&trace_types_lock);
3117 return ret;
3118}
3119EXPORT_SYMBOL_GPL(trace_add_event_call);
3120
3121/*
3122 * Must be called under locking of trace_types_lock, event_mutex and
3123 * trace_event_sem.
3124 */
3125static void __trace_remove_event_call(struct trace_event_call *call)
3126{
3127 event_remove(call);
3128 trace_destroy_fields(call);
3129 free_event_filter(call->filter);
3130 call->filter = NULL;
3131}
3132
3133static int probe_remove_event_call(struct trace_event_call *call)
3134{
3135 struct trace_array *tr;
3136 struct trace_event_file *file;
3137
3138#ifdef CONFIG_PERF_EVENTS
3139 if (call->perf_refcount)
3140 return -EBUSY;
3141#endif
3142 do_for_each_event_file(tr, file) {
3143 if (file->event_call != call)
3144 continue;
3145 /*
3146 * We can't rely on ftrace_event_enable_disable(enable => 0)
3147 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3148 * TRACE_REG_UNREGISTER.
3149 */
3150 if (file->flags & EVENT_FILE_FL_ENABLED)
3151 goto busy;
3152
3153 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3154 tr->clear_trace = true;
3155 /*
3156 * The do_for_each_event_file_safe() is
3157 * a double loop. After finding the call for this
3158 * trace_array, we use break to jump to the next
3159 * trace_array.
3160 */
3161 break;
3162 } while_for_each_event_file();
3163
3164 __trace_remove_event_call(call);
3165
3166 return 0;
3167 busy:
3168 /* No need to clear the trace now */
3169 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3170 tr->clear_trace = false;
3171 }
3172 return -EBUSY;
3173}
3174
3175/* Remove an event_call */
3176int trace_remove_event_call(struct trace_event_call *call)
3177{
3178 int ret;
3179
3180 lockdep_assert_held(&event_mutex);
3181
3182 mutex_lock(&trace_types_lock);
3183 down_write(&trace_event_sem);
3184 ret = probe_remove_event_call(call);
3185 up_write(&trace_event_sem);
3186 mutex_unlock(&trace_types_lock);
3187
3188 return ret;
3189}
3190EXPORT_SYMBOL_GPL(trace_remove_event_call);
3191
3192#define for_each_event(event, start, end) \
3193 for (event = start; \
3194 (unsigned long)event < (unsigned long)end; \
3195 event++)
3196
3197#ifdef CONFIG_MODULES
3198
3199static void trace_module_add_events(struct module *mod)
3200{
3201 struct trace_event_call **call, **start, **end;
3202
3203 if (!mod->num_trace_events)
3204 return;
3205
3206 /* Don't add infrastructure for mods without tracepoints */
3207 if (trace_module_has_bad_taint(mod)) {
3208 pr_err("%s: module has bad taint, not creating trace events\n",
3209 mod->name);
3210 return;
3211 }
3212
3213 start = mod->trace_events;
3214 end = mod->trace_events + mod->num_trace_events;
3215
3216 for_each_event(call, start, end) {
3217 __register_event(*call, mod);
3218 __add_event_to_tracers(*call);
3219 }
3220}
3221
3222static void trace_module_remove_events(struct module *mod)
3223{
3224 struct trace_event_call *call, *p;
3225 struct module_string *modstr, *m;
3226
3227 down_write(&trace_event_sem);
3228 list_for_each_entry_safe(call, p, &ftrace_events, list) {
3229 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3230 continue;
3231 if (call->module == mod)
3232 __trace_remove_event_call(call);
3233 }
3234 /* Check for any strings allocade for this module */
3235 list_for_each_entry_safe(modstr, m, &module_strings, next) {
3236 if (modstr->module != mod)
3237 continue;
3238 list_del(&modstr->next);
3239 kfree(modstr->str);
3240 kfree(modstr);
3241 }
3242 up_write(&trace_event_sem);
3243
3244 /*
3245 * It is safest to reset the ring buffer if the module being unloaded
3246 * registered any events that were used. The only worry is if
3247 * a new module gets loaded, and takes on the same id as the events
3248 * of this module. When printing out the buffer, traced events left
3249 * over from this module may be passed to the new module events and
3250 * unexpected results may occur.
3251 */
3252 tracing_reset_all_online_cpus_unlocked();
3253}
3254
3255static int trace_module_notify(struct notifier_block *self,
3256 unsigned long val, void *data)
3257{
3258 struct module *mod = data;
3259
3260 mutex_lock(&event_mutex);
3261 mutex_lock(&trace_types_lock);
3262 switch (val) {
3263 case MODULE_STATE_COMING:
3264 trace_module_add_events(mod);
3265 break;
3266 case MODULE_STATE_GOING:
3267 trace_module_remove_events(mod);
3268 break;
3269 }
3270 mutex_unlock(&trace_types_lock);
3271 mutex_unlock(&event_mutex);
3272
3273 return NOTIFY_OK;
3274}
3275
3276static struct notifier_block trace_module_nb = {
3277 .notifier_call = trace_module_notify,
3278 .priority = 1, /* higher than trace.c module notify */
3279};
3280#endif /* CONFIG_MODULES */
3281
3282/* Create a new event directory structure for a trace directory. */
3283static void
3284__trace_add_event_dirs(struct trace_array *tr)
3285{
3286 struct trace_event_call *call;
3287 int ret;
3288
3289 list_for_each_entry(call, &ftrace_events, list) {
3290 ret = __trace_add_new_event(call, tr);
3291 if (ret < 0)
3292 pr_warn("Could not create directory for event %s\n",
3293 trace_event_name(call));
3294 }
3295}
3296
3297/* Returns any file that matches the system and event */
3298struct trace_event_file *
3299__find_event_file(struct trace_array *tr, const char *system, const char *event)
3300{
3301 struct trace_event_file *file;
3302 struct trace_event_call *call;
3303 const char *name;
3304
3305 list_for_each_entry(file, &tr->events, list) {
3306
3307 call = file->event_call;
3308 name = trace_event_name(call);
3309
3310 if (!name || !call->class)
3311 continue;
3312
3313 if (strcmp(event, name) == 0 &&
3314 strcmp(system, call->class->system) == 0)
3315 return file;
3316 }
3317 return NULL;
3318}
3319
3320/* Returns valid trace event files that match system and event */
3321struct trace_event_file *
3322find_event_file(struct trace_array *tr, const char *system, const char *event)
3323{
3324 struct trace_event_file *file;
3325
3326 file = __find_event_file(tr, system, event);
3327 if (!file || !file->event_call->class->reg ||
3328 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3329 return NULL;
3330
3331 return file;
3332}
3333
3334/**
3335 * trace_get_event_file - Find and return a trace event file
3336 * @instance: The name of the trace instance containing the event
3337 * @system: The name of the system containing the event
3338 * @event: The name of the event
3339 *
3340 * Return a trace event file given the trace instance name, trace
3341 * system, and trace event name. If the instance name is NULL, it
3342 * refers to the top-level trace array.
3343 *
3344 * This function will look it up and return it if found, after calling
3345 * trace_array_get() to prevent the instance from going away, and
3346 * increment the event's module refcount to prevent it from being
3347 * removed.
3348 *
3349 * To release the file, call trace_put_event_file(), which will call
3350 * trace_array_put() and decrement the event's module refcount.
3351 *
3352 * Return: The trace event on success, ERR_PTR otherwise.
3353 */
3354struct trace_event_file *trace_get_event_file(const char *instance,
3355 const char *system,
3356 const char *event)
3357{
3358 struct trace_array *tr = top_trace_array();
3359 struct trace_event_file *file = NULL;
3360 int ret = -EINVAL;
3361
3362 if (instance) {
3363 tr = trace_array_find_get(instance);
3364 if (!tr)
3365 return ERR_PTR(-ENOENT);
3366 } else {
3367 ret = trace_array_get(tr);
3368 if (ret)
3369 return ERR_PTR(ret);
3370 }
3371
3372 mutex_lock(&event_mutex);
3373
3374 file = find_event_file(tr, system, event);
3375 if (!file) {
3376 trace_array_put(tr);
3377 ret = -EINVAL;
3378 goto out;
3379 }
3380
3381 /* Don't let event modules unload while in use */
3382 ret = trace_event_try_get_ref(file->event_call);
3383 if (!ret) {
3384 trace_array_put(tr);
3385 ret = -EBUSY;
3386 goto out;
3387 }
3388
3389 ret = 0;
3390 out:
3391 mutex_unlock(&event_mutex);
3392
3393 if (ret)
3394 file = ERR_PTR(ret);
3395
3396 return file;
3397}
3398EXPORT_SYMBOL_GPL(trace_get_event_file);
3399
3400/**
3401 * trace_put_event_file - Release a file from trace_get_event_file()
3402 * @file: The trace event file
3403 *
3404 * If a file was retrieved using trace_get_event_file(), this should
3405 * be called when it's no longer needed. It will cancel the previous
3406 * trace_array_get() called by that function, and decrement the
3407 * event's module refcount.
3408 */
3409void trace_put_event_file(struct trace_event_file *file)
3410{
3411 mutex_lock(&event_mutex);
3412 trace_event_put_ref(file->event_call);
3413 mutex_unlock(&event_mutex);
3414
3415 trace_array_put(file->tr);
3416}
3417EXPORT_SYMBOL_GPL(trace_put_event_file);
3418
3419#ifdef CONFIG_DYNAMIC_FTRACE
3420
3421/* Avoid typos */
3422#define ENABLE_EVENT_STR "enable_event"
3423#define DISABLE_EVENT_STR "disable_event"
3424
3425struct event_probe_data {
3426 struct trace_event_file *file;
3427 unsigned long count;
3428 int ref;
3429 bool enable;
3430};
3431
3432static void update_event_probe(struct event_probe_data *data)
3433{
3434 if (data->enable)
3435 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3436 else
3437 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3438}
3439
3440static void
3441event_enable_probe(unsigned long ip, unsigned long parent_ip,
3442 struct trace_array *tr, struct ftrace_probe_ops *ops,
3443 void *data)
3444{
3445 struct ftrace_func_mapper *mapper = data;
3446 struct event_probe_data *edata;
3447 void **pdata;
3448
3449 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3450 if (!pdata || !*pdata)
3451 return;
3452
3453 edata = *pdata;
3454 update_event_probe(edata);
3455}
3456
3457static void
3458event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3459 struct trace_array *tr, struct ftrace_probe_ops *ops,
3460 void *data)
3461{
3462 struct ftrace_func_mapper *mapper = data;
3463 struct event_probe_data *edata;
3464 void **pdata;
3465
3466 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3467 if (!pdata || !*pdata)
3468 return;
3469
3470 edata = *pdata;
3471
3472 if (!edata->count)
3473 return;
3474
3475 /* Skip if the event is in a state we want to switch to */
3476 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3477 return;
3478
3479 if (edata->count != -1)
3480 (edata->count)--;
3481
3482 update_event_probe(edata);
3483}
3484
3485static int
3486event_enable_print(struct seq_file *m, unsigned long ip,
3487 struct ftrace_probe_ops *ops, void *data)
3488{
3489 struct ftrace_func_mapper *mapper = data;
3490 struct event_probe_data *edata;
3491 void **pdata;
3492
3493 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3494
3495 if (WARN_ON_ONCE(!pdata || !*pdata))
3496 return 0;
3497
3498 edata = *pdata;
3499
3500 seq_printf(m, "%ps:", (void *)ip);
3501
3502 seq_printf(m, "%s:%s:%s",
3503 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3504 edata->file->event_call->class->system,
3505 trace_event_name(edata->file->event_call));
3506
3507 if (edata->count == -1)
3508 seq_puts(m, ":unlimited\n");
3509 else
3510 seq_printf(m, ":count=%ld\n", edata->count);
3511
3512 return 0;
3513}
3514
3515static int
3516event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3517 unsigned long ip, void *init_data, void **data)
3518{
3519 struct ftrace_func_mapper *mapper = *data;
3520 struct event_probe_data *edata = init_data;
3521 int ret;
3522
3523 if (!mapper) {
3524 mapper = allocate_ftrace_func_mapper();
3525 if (!mapper)
3526 return -ENODEV;
3527 *data = mapper;
3528 }
3529
3530 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3531 if (ret < 0)
3532 return ret;
3533
3534 edata->ref++;
3535
3536 return 0;
3537}
3538
3539static int free_probe_data(void *data)
3540{
3541 struct event_probe_data *edata = data;
3542
3543 edata->ref--;
3544 if (!edata->ref) {
3545 /* Remove the SOFT_MODE flag */
3546 __ftrace_event_enable_disable(edata->file, 0, 1);
3547 trace_event_put_ref(edata->file->event_call);
3548 kfree(edata);
3549 }
3550 return 0;
3551}
3552
3553static void
3554event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3555 unsigned long ip, void *data)
3556{
3557 struct ftrace_func_mapper *mapper = data;
3558 struct event_probe_data *edata;
3559
3560 if (!ip) {
3561 if (!mapper)
3562 return;
3563 free_ftrace_func_mapper(mapper, free_probe_data);
3564 return;
3565 }
3566
3567 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3568
3569 if (WARN_ON_ONCE(!edata))
3570 return;
3571
3572 if (WARN_ON_ONCE(edata->ref <= 0))
3573 return;
3574
3575 free_probe_data(edata);
3576}
3577
3578static struct ftrace_probe_ops event_enable_probe_ops = {
3579 .func = event_enable_probe,
3580 .print = event_enable_print,
3581 .init = event_enable_init,
3582 .free = event_enable_free,
3583};
3584
3585static struct ftrace_probe_ops event_enable_count_probe_ops = {
3586 .func = event_enable_count_probe,
3587 .print = event_enable_print,
3588 .init = event_enable_init,
3589 .free = event_enable_free,
3590};
3591
3592static struct ftrace_probe_ops event_disable_probe_ops = {
3593 .func = event_enable_probe,
3594 .print = event_enable_print,
3595 .init = event_enable_init,
3596 .free = event_enable_free,
3597};
3598
3599static struct ftrace_probe_ops event_disable_count_probe_ops = {
3600 .func = event_enable_count_probe,
3601 .print = event_enable_print,
3602 .init = event_enable_init,
3603 .free = event_enable_free,
3604};
3605
3606static int
3607event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3608 char *glob, char *cmd, char *param, int enabled)
3609{
3610 struct trace_event_file *file;
3611 struct ftrace_probe_ops *ops;
3612 struct event_probe_data *data;
3613 const char *system;
3614 const char *event;
3615 char *number;
3616 bool enable;
3617 int ret;
3618
3619 if (!tr)
3620 return -ENODEV;
3621
3622 /* hash funcs only work with set_ftrace_filter */
3623 if (!enabled || !param)
3624 return -EINVAL;
3625
3626 system = strsep(¶m, ":");
3627 if (!param)
3628 return -EINVAL;
3629
3630 event = strsep(¶m, ":");
3631
3632 mutex_lock(&event_mutex);
3633
3634 ret = -EINVAL;
3635 file = find_event_file(tr, system, event);
3636 if (!file)
3637 goto out;
3638
3639 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3640
3641 if (enable)
3642 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3643 else
3644 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3645
3646 if (glob[0] == '!') {
3647 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3648 goto out;
3649 }
3650
3651 ret = -ENOMEM;
3652
3653 data = kzalloc(sizeof(*data), GFP_KERNEL);
3654 if (!data)
3655 goto out;
3656
3657 data->enable = enable;
3658 data->count = -1;
3659 data->file = file;
3660
3661 if (!param)
3662 goto out_reg;
3663
3664 number = strsep(¶m, ":");
3665
3666 ret = -EINVAL;
3667 if (!strlen(number))
3668 goto out_free;
3669
3670 /*
3671 * We use the callback data field (which is a pointer)
3672 * as our counter.
3673 */
3674 ret = kstrtoul(number, 0, &data->count);
3675 if (ret)
3676 goto out_free;
3677
3678 out_reg:
3679 /* Don't let event modules unload while probe registered */
3680 ret = trace_event_try_get_ref(file->event_call);
3681 if (!ret) {
3682 ret = -EBUSY;
3683 goto out_free;
3684 }
3685
3686 ret = __ftrace_event_enable_disable(file, 1, 1);
3687 if (ret < 0)
3688 goto out_put;
3689
3690 ret = register_ftrace_function_probe(glob, tr, ops, data);
3691 /*
3692 * The above returns on success the # of functions enabled,
3693 * but if it didn't find any functions it returns zero.
3694 * Consider no functions a failure too.
3695 */
3696 if (!ret) {
3697 ret = -ENOENT;
3698 goto out_disable;
3699 } else if (ret < 0)
3700 goto out_disable;
3701 /* Just return zero, not the number of enabled functions */
3702 ret = 0;
3703 out:
3704 mutex_unlock(&event_mutex);
3705 return ret;
3706
3707 out_disable:
3708 __ftrace_event_enable_disable(file, 0, 1);
3709 out_put:
3710 trace_event_put_ref(file->event_call);
3711 out_free:
3712 kfree(data);
3713 goto out;
3714}
3715
3716static struct ftrace_func_command event_enable_cmd = {
3717 .name = ENABLE_EVENT_STR,
3718 .func = event_enable_func,
3719};
3720
3721static struct ftrace_func_command event_disable_cmd = {
3722 .name = DISABLE_EVENT_STR,
3723 .func = event_enable_func,
3724};
3725
3726static __init int register_event_cmds(void)
3727{
3728 int ret;
3729
3730 ret = register_ftrace_command(&event_enable_cmd);
3731 if (WARN_ON(ret < 0))
3732 return ret;
3733 ret = register_ftrace_command(&event_disable_cmd);
3734 if (WARN_ON(ret < 0))
3735 unregister_ftrace_command(&event_enable_cmd);
3736 return ret;
3737}
3738#else
3739static inline int register_event_cmds(void) { return 0; }
3740#endif /* CONFIG_DYNAMIC_FTRACE */
3741
3742/*
3743 * The top level array and trace arrays created by boot-time tracing
3744 * have already had its trace_event_file descriptors created in order
3745 * to allow for early events to be recorded.
3746 * This function is called after the tracefs has been initialized,
3747 * and we now have to create the files associated to the events.
3748 */
3749static void __trace_early_add_event_dirs(struct trace_array *tr)
3750{
3751 struct trace_event_file *file;
3752 int ret;
3753
3754
3755 list_for_each_entry(file, &tr->events, list) {
3756 ret = event_create_dir(tr->event_dir, file);
3757 if (ret < 0)
3758 pr_warn("Could not create directory for event %s\n",
3759 trace_event_name(file->event_call));
3760 }
3761}
3762
3763/*
3764 * For early boot up, the top trace array and the trace arrays created
3765 * by boot-time tracing require to have a list of events that can be
3766 * enabled. This must be done before the filesystem is set up in order
3767 * to allow events to be traced early.
3768 */
3769void __trace_early_add_events(struct trace_array *tr)
3770{
3771 struct trace_event_call *call;
3772 int ret;
3773
3774 list_for_each_entry(call, &ftrace_events, list) {
3775 /* Early boot up should not have any modules loaded */
3776 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3777 WARN_ON_ONCE(call->module))
3778 continue;
3779
3780 ret = __trace_early_add_new_event(call, tr);
3781 if (ret < 0)
3782 pr_warn("Could not create early event %s\n",
3783 trace_event_name(call));
3784 }
3785}
3786
3787/* Remove the event directory structure for a trace directory. */
3788static void
3789__trace_remove_event_dirs(struct trace_array *tr)
3790{
3791 struct trace_event_file *file, *next;
3792
3793 list_for_each_entry_safe(file, next, &tr->events, list)
3794 remove_event_file_dir(file);
3795}
3796
3797static void __add_event_to_tracers(struct trace_event_call *call)
3798{
3799 struct trace_array *tr;
3800
3801 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3802 __trace_add_new_event(call, tr);
3803}
3804
3805extern struct trace_event_call *__start_ftrace_events[];
3806extern struct trace_event_call *__stop_ftrace_events[];
3807
3808static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3809
3810static __init int setup_trace_event(char *str)
3811{
3812 strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3813 trace_set_ring_buffer_expanded(NULL);
3814 disable_tracing_selftest("running event tracing");
3815
3816 return 1;
3817}
3818__setup("trace_event=", setup_trace_event);
3819
3820static int events_callback(const char *name, umode_t *mode, void **data,
3821 const struct file_operations **fops)
3822{
3823 if (strcmp(name, "enable") == 0) {
3824 *mode = TRACE_MODE_WRITE;
3825 *fops = &ftrace_tr_enable_fops;
3826 return 1;
3827 }
3828
3829 if (strcmp(name, "header_page") == 0) {
3830 *mode = TRACE_MODE_READ;
3831 *fops = &ftrace_show_header_page_fops;
3832
3833 } else if (strcmp(name, "header_event") == 0) {
3834 *mode = TRACE_MODE_READ;
3835 *fops = &ftrace_show_header_event_fops;
3836 } else
3837 return 0;
3838
3839 return 1;
3840}
3841
3842/* Expects to have event_mutex held when called */
3843static int
3844create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3845{
3846 struct eventfs_inode *e_events;
3847 struct dentry *entry;
3848 int nr_entries;
3849 static struct eventfs_entry events_entries[] = {
3850 {
3851 .name = "enable",
3852 .callback = events_callback,
3853 },
3854 {
3855 .name = "header_page",
3856 .callback = events_callback,
3857 },
3858 {
3859 .name = "header_event",
3860 .callback = events_callback,
3861 },
3862 };
3863
3864 entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3865 tr, &ftrace_set_event_fops);
3866 if (!entry)
3867 return -ENOMEM;
3868
3869 nr_entries = ARRAY_SIZE(events_entries);
3870
3871 e_events = eventfs_create_events_dir("events", parent, events_entries,
3872 nr_entries, tr);
3873 if (IS_ERR(e_events)) {
3874 pr_warn("Could not create tracefs 'events' directory\n");
3875 return -ENOMEM;
3876 }
3877
3878 /* There are not as crucial, just warn if they are not created */
3879
3880 trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3881 tr, &ftrace_set_event_pid_fops);
3882
3883 trace_create_file("set_event_notrace_pid",
3884 TRACE_MODE_WRITE, parent, tr,
3885 &ftrace_set_event_notrace_pid_fops);
3886
3887 tr->event_dir = e_events;
3888
3889 return 0;
3890}
3891
3892/**
3893 * event_trace_add_tracer - add a instance of a trace_array to events
3894 * @parent: The parent dentry to place the files/directories for events in
3895 * @tr: The trace array associated with these events
3896 *
3897 * When a new instance is created, it needs to set up its events
3898 * directory, as well as other files associated with events. It also
3899 * creates the event hierarchy in the @parent/events directory.
3900 *
3901 * Returns 0 on success.
3902 *
3903 * Must be called with event_mutex held.
3904 */
3905int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3906{
3907 int ret;
3908
3909 lockdep_assert_held(&event_mutex);
3910
3911 ret = create_event_toplevel_files(parent, tr);
3912 if (ret)
3913 goto out;
3914
3915 down_write(&trace_event_sem);
3916 /* If tr already has the event list, it is initialized in early boot. */
3917 if (unlikely(!list_empty(&tr->events)))
3918 __trace_early_add_event_dirs(tr);
3919 else
3920 __trace_add_event_dirs(tr);
3921 up_write(&trace_event_sem);
3922
3923 out:
3924 return ret;
3925}
3926
3927/*
3928 * The top trace array already had its file descriptors created.
3929 * Now the files themselves need to be created.
3930 */
3931static __init int
3932early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3933{
3934 int ret;
3935
3936 mutex_lock(&event_mutex);
3937
3938 ret = create_event_toplevel_files(parent, tr);
3939 if (ret)
3940 goto out_unlock;
3941
3942 down_write(&trace_event_sem);
3943 __trace_early_add_event_dirs(tr);
3944 up_write(&trace_event_sem);
3945
3946 out_unlock:
3947 mutex_unlock(&event_mutex);
3948
3949 return ret;
3950}
3951
3952/* Must be called with event_mutex held */
3953int event_trace_del_tracer(struct trace_array *tr)
3954{
3955 lockdep_assert_held(&event_mutex);
3956
3957 /* Disable any event triggers and associated soft-disabled events */
3958 clear_event_triggers(tr);
3959
3960 /* Clear the pid list */
3961 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3962
3963 /* Disable any running events */
3964 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3965
3966 /* Make sure no more events are being executed */
3967 tracepoint_synchronize_unregister();
3968
3969 down_write(&trace_event_sem);
3970 __trace_remove_event_dirs(tr);
3971 eventfs_remove_events_dir(tr->event_dir);
3972 up_write(&trace_event_sem);
3973
3974 tr->event_dir = NULL;
3975
3976 return 0;
3977}
3978
3979static __init int event_trace_memsetup(void)
3980{
3981 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3982 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3983 return 0;
3984}
3985
3986__init void
3987early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3988{
3989 char *token;
3990 int ret;
3991
3992 while (true) {
3993 token = strsep(&buf, ",");
3994
3995 if (!token)
3996 break;
3997
3998 if (*token) {
3999 /* Restarting syscalls requires that we stop them first */
4000 if (disable_first)
4001 ftrace_set_clr_event(tr, token, 0);
4002
4003 ret = ftrace_set_clr_event(tr, token, 1);
4004 if (ret)
4005 pr_warn("Failed to enable trace event: %s\n", token);
4006 }
4007
4008 /* Put back the comma to allow this to be called again */
4009 if (buf)
4010 *(buf - 1) = ',';
4011 }
4012}
4013
4014static __init int event_trace_enable(void)
4015{
4016 struct trace_array *tr = top_trace_array();
4017 struct trace_event_call **iter, *call;
4018 int ret;
4019
4020 if (!tr)
4021 return -ENODEV;
4022
4023 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4024
4025 call = *iter;
4026 ret = event_init(call);
4027 if (!ret)
4028 list_add(&call->list, &ftrace_events);
4029 }
4030
4031 register_trigger_cmds();
4032
4033 /*
4034 * We need the top trace array to have a working set of trace
4035 * points at early init, before the debug files and directories
4036 * are created. Create the file entries now, and attach them
4037 * to the actual file dentries later.
4038 */
4039 __trace_early_add_events(tr);
4040
4041 early_enable_events(tr, bootup_event_buf, false);
4042
4043 trace_printk_start_comm();
4044
4045 register_event_cmds();
4046
4047
4048 return 0;
4049}
4050
4051/*
4052 * event_trace_enable() is called from trace_event_init() first to
4053 * initialize events and perhaps start any events that are on the
4054 * command line. Unfortunately, there are some events that will not
4055 * start this early, like the system call tracepoints that need
4056 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4057 * event_trace_enable() is called before pid 1 starts, and this flag
4058 * is never set, making the syscall tracepoint never get reached, but
4059 * the event is enabled regardless (and not doing anything).
4060 */
4061static __init int event_trace_enable_again(void)
4062{
4063 struct trace_array *tr;
4064
4065 tr = top_trace_array();
4066 if (!tr)
4067 return -ENODEV;
4068
4069 early_enable_events(tr, bootup_event_buf, true);
4070
4071 return 0;
4072}
4073
4074early_initcall(event_trace_enable_again);
4075
4076/* Init fields which doesn't related to the tracefs */
4077static __init int event_trace_init_fields(void)
4078{
4079 if (trace_define_generic_fields())
4080 pr_warn("tracing: Failed to allocated generic fields");
4081
4082 if (trace_define_common_fields())
4083 pr_warn("tracing: Failed to allocate common fields");
4084
4085 return 0;
4086}
4087
4088__init int event_trace_init(void)
4089{
4090 struct trace_array *tr;
4091 int ret;
4092
4093 tr = top_trace_array();
4094 if (!tr)
4095 return -ENODEV;
4096
4097 trace_create_file("available_events", TRACE_MODE_READ,
4098 NULL, tr, &ftrace_avail_fops);
4099
4100 ret = early_event_add_tracer(NULL, tr);
4101 if (ret)
4102 return ret;
4103
4104#ifdef CONFIG_MODULES
4105 ret = register_module_notifier(&trace_module_nb);
4106 if (ret)
4107 pr_warn("Failed to register trace events module notifier\n");
4108#endif
4109
4110 eventdir_initialized = true;
4111
4112 return 0;
4113}
4114
4115void __init trace_event_init(void)
4116{
4117 event_trace_memsetup();
4118 init_ftrace_syscalls();
4119 event_trace_enable();
4120 event_trace_init_fields();
4121}
4122
4123#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4124
4125static DEFINE_SPINLOCK(test_spinlock);
4126static DEFINE_SPINLOCK(test_spinlock_irq);
4127static DEFINE_MUTEX(test_mutex);
4128
4129static __init void test_work(struct work_struct *dummy)
4130{
4131 spin_lock(&test_spinlock);
4132 spin_lock_irq(&test_spinlock_irq);
4133 udelay(1);
4134 spin_unlock_irq(&test_spinlock_irq);
4135 spin_unlock(&test_spinlock);
4136
4137 mutex_lock(&test_mutex);
4138 msleep(1);
4139 mutex_unlock(&test_mutex);
4140}
4141
4142static __init int event_test_thread(void *unused)
4143{
4144 void *test_malloc;
4145
4146 test_malloc = kmalloc(1234, GFP_KERNEL);
4147 if (!test_malloc)
4148 pr_info("failed to kmalloc\n");
4149
4150 schedule_on_each_cpu(test_work);
4151
4152 kfree(test_malloc);
4153
4154 set_current_state(TASK_INTERRUPTIBLE);
4155 while (!kthread_should_stop()) {
4156 schedule();
4157 set_current_state(TASK_INTERRUPTIBLE);
4158 }
4159 __set_current_state(TASK_RUNNING);
4160
4161 return 0;
4162}
4163
4164/*
4165 * Do various things that may trigger events.
4166 */
4167static __init void event_test_stuff(void)
4168{
4169 struct task_struct *test_thread;
4170
4171 test_thread = kthread_run(event_test_thread, NULL, "test-events");
4172 msleep(1);
4173 kthread_stop(test_thread);
4174}
4175
4176/*
4177 * For every trace event defined, we will test each trace point separately,
4178 * and then by groups, and finally all trace points.
4179 */
4180static __init void event_trace_self_tests(void)
4181{
4182 struct trace_subsystem_dir *dir;
4183 struct trace_event_file *file;
4184 struct trace_event_call *call;
4185 struct event_subsystem *system;
4186 struct trace_array *tr;
4187 int ret;
4188
4189 tr = top_trace_array();
4190 if (!tr)
4191 return;
4192
4193 pr_info("Running tests on trace events:\n");
4194
4195 list_for_each_entry(file, &tr->events, list) {
4196
4197 call = file->event_call;
4198
4199 /* Only test those that have a probe */
4200 if (!call->class || !call->class->probe)
4201 continue;
4202
4203/*
4204 * Testing syscall events here is pretty useless, but
4205 * we still do it if configured. But this is time consuming.
4206 * What we really need is a user thread to perform the
4207 * syscalls as we test.
4208 */
4209#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4210 if (call->class->system &&
4211 strcmp(call->class->system, "syscalls") == 0)
4212 continue;
4213#endif
4214
4215 pr_info("Testing event %s: ", trace_event_name(call));
4216
4217 /*
4218 * If an event is already enabled, someone is using
4219 * it and the self test should not be on.
4220 */
4221 if (file->flags & EVENT_FILE_FL_ENABLED) {
4222 pr_warn("Enabled event during self test!\n");
4223 WARN_ON_ONCE(1);
4224 continue;
4225 }
4226
4227 ftrace_event_enable_disable(file, 1);
4228 event_test_stuff();
4229 ftrace_event_enable_disable(file, 0);
4230
4231 pr_cont("OK\n");
4232 }
4233
4234 /* Now test at the sub system level */
4235
4236 pr_info("Running tests on trace event systems:\n");
4237
4238 list_for_each_entry(dir, &tr->systems, list) {
4239
4240 system = dir->subsystem;
4241
4242 /* the ftrace system is special, skip it */
4243 if (strcmp(system->name, "ftrace") == 0)
4244 continue;
4245
4246 pr_info("Testing event system %s: ", system->name);
4247
4248 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4249 if (WARN_ON_ONCE(ret)) {
4250 pr_warn("error enabling system %s\n",
4251 system->name);
4252 continue;
4253 }
4254
4255 event_test_stuff();
4256
4257 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4258 if (WARN_ON_ONCE(ret)) {
4259 pr_warn("error disabling system %s\n",
4260 system->name);
4261 continue;
4262 }
4263
4264 pr_cont("OK\n");
4265 }
4266
4267 /* Test with all events enabled */
4268
4269 pr_info("Running tests on all trace events:\n");
4270 pr_info("Testing all events: ");
4271
4272 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4273 if (WARN_ON_ONCE(ret)) {
4274 pr_warn("error enabling all events\n");
4275 return;
4276 }
4277
4278 event_test_stuff();
4279
4280 /* reset sysname */
4281 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4282 if (WARN_ON_ONCE(ret)) {
4283 pr_warn("error disabling all events\n");
4284 return;
4285 }
4286
4287 pr_cont("OK\n");
4288}
4289
4290#ifdef CONFIG_FUNCTION_TRACER
4291
4292static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4293
4294static struct trace_event_file event_trace_file __initdata;
4295
4296static void __init
4297function_test_events_call(unsigned long ip, unsigned long parent_ip,
4298 struct ftrace_ops *op, struct ftrace_regs *regs)
4299{
4300 struct trace_buffer *buffer;
4301 struct ring_buffer_event *event;
4302 struct ftrace_entry *entry;
4303 unsigned int trace_ctx;
4304 long disabled;
4305 int cpu;
4306
4307 trace_ctx = tracing_gen_ctx();
4308 preempt_disable_notrace();
4309 cpu = raw_smp_processor_id();
4310 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4311
4312 if (disabled != 1)
4313 goto out;
4314
4315 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4316 TRACE_FN, sizeof(*entry),
4317 trace_ctx);
4318 if (!event)
4319 goto out;
4320 entry = ring_buffer_event_data(event);
4321 entry->ip = ip;
4322 entry->parent_ip = parent_ip;
4323
4324 event_trigger_unlock_commit(&event_trace_file, buffer, event,
4325 entry, trace_ctx);
4326 out:
4327 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4328 preempt_enable_notrace();
4329}
4330
4331static struct ftrace_ops trace_ops __initdata =
4332{
4333 .func = function_test_events_call,
4334};
4335
4336static __init void event_trace_self_test_with_function(void)
4337{
4338 int ret;
4339
4340 event_trace_file.tr = top_trace_array();
4341 if (WARN_ON(!event_trace_file.tr))
4342 return;
4343
4344 ret = register_ftrace_function(&trace_ops);
4345 if (WARN_ON(ret < 0)) {
4346 pr_info("Failed to enable function tracer for event tests\n");
4347 return;
4348 }
4349 pr_info("Running tests again, along with the function tracer\n");
4350 event_trace_self_tests();
4351 unregister_ftrace_function(&trace_ops);
4352}
4353#else
4354static __init void event_trace_self_test_with_function(void)
4355{
4356}
4357#endif
4358
4359static __init int event_trace_self_tests_init(void)
4360{
4361 if (!tracing_selftest_disabled) {
4362 event_trace_self_tests();
4363 event_trace_self_test_with_function();
4364 }
4365
4366 return 0;
4367}
4368
4369late_initcall(event_trace_self_tests_init);
4370
4371#endif
1/*
2 * event tracer
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
8 *
9 */
10
11#define pr_fmt(fmt) fmt
12
13#include <linux/workqueue.h>
14#include <linux/spinlock.h>
15#include <linux/kthread.h>
16#include <linux/tracefs.h>
17#include <linux/uaccess.h>
18#include <linux/module.h>
19#include <linux/ctype.h>
20#include <linux/sort.h>
21#include <linux/slab.h>
22#include <linux/delay.h>
23
24#include <trace/events/sched.h>
25
26#include <asm/setup.h>
27
28#include "trace_output.h"
29
30#undef TRACE_SYSTEM
31#define TRACE_SYSTEM "TRACE_SYSTEM"
32
33DEFINE_MUTEX(event_mutex);
34
35LIST_HEAD(ftrace_events);
36static LIST_HEAD(ftrace_generic_fields);
37static LIST_HEAD(ftrace_common_fields);
38
39#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
40
41static struct kmem_cache *field_cachep;
42static struct kmem_cache *file_cachep;
43
44static inline int system_refcount(struct event_subsystem *system)
45{
46 return system->ref_count;
47}
48
49static int system_refcount_inc(struct event_subsystem *system)
50{
51 return system->ref_count++;
52}
53
54static int system_refcount_dec(struct event_subsystem *system)
55{
56 return --system->ref_count;
57}
58
59/* Double loops, do not use break, only goto's work */
60#define do_for_each_event_file(tr, file) \
61 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
62 list_for_each_entry(file, &tr->events, list)
63
64#define do_for_each_event_file_safe(tr, file) \
65 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
66 struct trace_event_file *___n; \
67 list_for_each_entry_safe(file, ___n, &tr->events, list)
68
69#define while_for_each_event_file() \
70 }
71
72static struct list_head *
73trace_get_fields(struct trace_event_call *event_call)
74{
75 if (!event_call->class->get_fields)
76 return &event_call->class->fields;
77 return event_call->class->get_fields(event_call);
78}
79
80static struct ftrace_event_field *
81__find_event_field(struct list_head *head, char *name)
82{
83 struct ftrace_event_field *field;
84
85 list_for_each_entry(field, head, link) {
86 if (!strcmp(field->name, name))
87 return field;
88 }
89
90 return NULL;
91}
92
93struct ftrace_event_field *
94trace_find_event_field(struct trace_event_call *call, char *name)
95{
96 struct ftrace_event_field *field;
97 struct list_head *head;
98
99 head = trace_get_fields(call);
100 field = __find_event_field(head, name);
101 if (field)
102 return field;
103
104 field = __find_event_field(&ftrace_generic_fields, name);
105 if (field)
106 return field;
107
108 return __find_event_field(&ftrace_common_fields, name);
109}
110
111static int __trace_define_field(struct list_head *head, const char *type,
112 const char *name, int offset, int size,
113 int is_signed, int filter_type)
114{
115 struct ftrace_event_field *field;
116
117 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
118 if (!field)
119 return -ENOMEM;
120
121 field->name = name;
122 field->type = type;
123
124 if (filter_type == FILTER_OTHER)
125 field->filter_type = filter_assign_type(type);
126 else
127 field->filter_type = filter_type;
128
129 field->offset = offset;
130 field->size = size;
131 field->is_signed = is_signed;
132
133 list_add(&field->link, head);
134
135 return 0;
136}
137
138int trace_define_field(struct trace_event_call *call, const char *type,
139 const char *name, int offset, int size, int is_signed,
140 int filter_type)
141{
142 struct list_head *head;
143
144 if (WARN_ON(!call->class))
145 return 0;
146
147 head = trace_get_fields(call);
148 return __trace_define_field(head, type, name, offset, size,
149 is_signed, filter_type);
150}
151EXPORT_SYMBOL_GPL(trace_define_field);
152
153#define __generic_field(type, item, filter_type) \
154 ret = __trace_define_field(&ftrace_generic_fields, #type, \
155 #item, 0, 0, is_signed_type(type), \
156 filter_type); \
157 if (ret) \
158 return ret;
159
160#define __common_field(type, item) \
161 ret = __trace_define_field(&ftrace_common_fields, #type, \
162 "common_" #item, \
163 offsetof(typeof(ent), item), \
164 sizeof(ent.item), \
165 is_signed_type(type), FILTER_OTHER); \
166 if (ret) \
167 return ret;
168
169static int trace_define_generic_fields(void)
170{
171 int ret;
172
173 __generic_field(int, CPU, FILTER_CPU);
174 __generic_field(int, cpu, FILTER_CPU);
175 __generic_field(char *, COMM, FILTER_COMM);
176 __generic_field(char *, comm, FILTER_COMM);
177
178 return ret;
179}
180
181static int trace_define_common_fields(void)
182{
183 int ret;
184 struct trace_entry ent;
185
186 __common_field(unsigned short, type);
187 __common_field(unsigned char, flags);
188 __common_field(unsigned char, preempt_count);
189 __common_field(int, pid);
190
191 return ret;
192}
193
194static void trace_destroy_fields(struct trace_event_call *call)
195{
196 struct ftrace_event_field *field, *next;
197 struct list_head *head;
198
199 head = trace_get_fields(call);
200 list_for_each_entry_safe(field, next, head, link) {
201 list_del(&field->link);
202 kmem_cache_free(field_cachep, field);
203 }
204}
205
206/*
207 * run-time version of trace_event_get_offsets_<call>() that returns the last
208 * accessible offset of trace fields excluding __dynamic_array bytes
209 */
210int trace_event_get_offsets(struct trace_event_call *call)
211{
212 struct ftrace_event_field *tail;
213 struct list_head *head;
214
215 head = trace_get_fields(call);
216 /*
217 * head->next points to the last field with the largest offset,
218 * since it was added last by trace_define_field()
219 */
220 tail = list_first_entry(head, struct ftrace_event_field, link);
221 return tail->offset + tail->size;
222}
223
224int trace_event_raw_init(struct trace_event_call *call)
225{
226 int id;
227
228 id = register_trace_event(&call->event);
229 if (!id)
230 return -ENODEV;
231
232 return 0;
233}
234EXPORT_SYMBOL_GPL(trace_event_raw_init);
235
236bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
237{
238 struct trace_array *tr = trace_file->tr;
239 struct trace_array_cpu *data;
240 struct trace_pid_list *pid_list;
241
242 pid_list = rcu_dereference_sched(tr->filtered_pids);
243 if (!pid_list)
244 return false;
245
246 data = this_cpu_ptr(tr->trace_buffer.data);
247
248 return data->ignore_pid;
249}
250EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
251
252void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
253 struct trace_event_file *trace_file,
254 unsigned long len)
255{
256 struct trace_event_call *event_call = trace_file->event_call;
257
258 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
259 trace_event_ignore_this_pid(trace_file))
260 return NULL;
261
262 local_save_flags(fbuffer->flags);
263 fbuffer->pc = preempt_count();
264 /*
265 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
266 * preemption (adding one to the preempt_count). Since we are
267 * interested in the preempt_count at the time the tracepoint was
268 * hit, we need to subtract one to offset the increment.
269 */
270 if (IS_ENABLED(CONFIG_PREEMPT))
271 fbuffer->pc--;
272 fbuffer->trace_file = trace_file;
273
274 fbuffer->event =
275 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
276 event_call->event.type, len,
277 fbuffer->flags, fbuffer->pc);
278 if (!fbuffer->event)
279 return NULL;
280
281 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
282 return fbuffer->entry;
283}
284EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
285
286int trace_event_reg(struct trace_event_call *call,
287 enum trace_reg type, void *data)
288{
289 struct trace_event_file *file = data;
290
291 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
292 switch (type) {
293 case TRACE_REG_REGISTER:
294 return tracepoint_probe_register(call->tp,
295 call->class->probe,
296 file);
297 case TRACE_REG_UNREGISTER:
298 tracepoint_probe_unregister(call->tp,
299 call->class->probe,
300 file);
301 return 0;
302
303#ifdef CONFIG_PERF_EVENTS
304 case TRACE_REG_PERF_REGISTER:
305 return tracepoint_probe_register(call->tp,
306 call->class->perf_probe,
307 call);
308 case TRACE_REG_PERF_UNREGISTER:
309 tracepoint_probe_unregister(call->tp,
310 call->class->perf_probe,
311 call);
312 return 0;
313 case TRACE_REG_PERF_OPEN:
314 case TRACE_REG_PERF_CLOSE:
315 case TRACE_REG_PERF_ADD:
316 case TRACE_REG_PERF_DEL:
317 return 0;
318#endif
319 }
320 return 0;
321}
322EXPORT_SYMBOL_GPL(trace_event_reg);
323
324void trace_event_enable_cmd_record(bool enable)
325{
326 struct trace_event_file *file;
327 struct trace_array *tr;
328
329 mutex_lock(&event_mutex);
330 do_for_each_event_file(tr, file) {
331
332 if (!(file->flags & EVENT_FILE_FL_ENABLED))
333 continue;
334
335 if (enable) {
336 tracing_start_cmdline_record();
337 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
338 } else {
339 tracing_stop_cmdline_record();
340 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
341 }
342 } while_for_each_event_file();
343 mutex_unlock(&event_mutex);
344}
345
346static int __ftrace_event_enable_disable(struct trace_event_file *file,
347 int enable, int soft_disable)
348{
349 struct trace_event_call *call = file->event_call;
350 struct trace_array *tr = file->tr;
351 unsigned long file_flags = file->flags;
352 int ret = 0;
353 int disable;
354
355 switch (enable) {
356 case 0:
357 /*
358 * When soft_disable is set and enable is cleared, the sm_ref
359 * reference counter is decremented. If it reaches 0, we want
360 * to clear the SOFT_DISABLED flag but leave the event in the
361 * state that it was. That is, if the event was enabled and
362 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
363 * is set we do not want the event to be enabled before we
364 * clear the bit.
365 *
366 * When soft_disable is not set but the SOFT_MODE flag is,
367 * we do nothing. Do not disable the tracepoint, otherwise
368 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
369 */
370 if (soft_disable) {
371 if (atomic_dec_return(&file->sm_ref) > 0)
372 break;
373 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
374 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
375 } else
376 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
377
378 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
379 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
380 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
381 tracing_stop_cmdline_record();
382 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
383 }
384 call->class->reg(call, TRACE_REG_UNREGISTER, file);
385 }
386 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
387 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
388 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
389 else
390 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
391 break;
392 case 1:
393 /*
394 * When soft_disable is set and enable is set, we want to
395 * register the tracepoint for the event, but leave the event
396 * as is. That means, if the event was already enabled, we do
397 * nothing (but set SOFT_MODE). If the event is disabled, we
398 * set SOFT_DISABLED before enabling the event tracepoint, so
399 * it still seems to be disabled.
400 */
401 if (!soft_disable)
402 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
403 else {
404 if (atomic_inc_return(&file->sm_ref) > 1)
405 break;
406 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
407 }
408
409 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
410
411 /* Keep the event disabled, when going to SOFT_MODE. */
412 if (soft_disable)
413 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
414
415 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
416 tracing_start_cmdline_record();
417 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
418 }
419 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
420 if (ret) {
421 tracing_stop_cmdline_record();
422 pr_info("event trace: Could not enable event "
423 "%s\n", trace_event_name(call));
424 break;
425 }
426 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
427
428 /* WAS_ENABLED gets set but never cleared. */
429 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
430 }
431 break;
432 }
433
434 /* Enable or disable use of trace_buffered_event */
435 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
436 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
437 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
438 trace_buffered_event_enable();
439 else
440 trace_buffered_event_disable();
441 }
442
443 return ret;
444}
445
446int trace_event_enable_disable(struct trace_event_file *file,
447 int enable, int soft_disable)
448{
449 return __ftrace_event_enable_disable(file, enable, soft_disable);
450}
451
452static int ftrace_event_enable_disable(struct trace_event_file *file,
453 int enable)
454{
455 return __ftrace_event_enable_disable(file, enable, 0);
456}
457
458static void ftrace_clear_events(struct trace_array *tr)
459{
460 struct trace_event_file *file;
461
462 mutex_lock(&event_mutex);
463 list_for_each_entry(file, &tr->events, list) {
464 ftrace_event_enable_disable(file, 0);
465 }
466 mutex_unlock(&event_mutex);
467}
468
469static void
470event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
471{
472 struct trace_pid_list *pid_list;
473 struct trace_array *tr = data;
474
475 pid_list = rcu_dereference_sched(tr->filtered_pids);
476 trace_filter_add_remove_task(pid_list, NULL, task);
477}
478
479static void
480event_filter_pid_sched_process_fork(void *data,
481 struct task_struct *self,
482 struct task_struct *task)
483{
484 struct trace_pid_list *pid_list;
485 struct trace_array *tr = data;
486
487 pid_list = rcu_dereference_sched(tr->filtered_pids);
488 trace_filter_add_remove_task(pid_list, self, task);
489}
490
491void trace_event_follow_fork(struct trace_array *tr, bool enable)
492{
493 if (enable) {
494 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
495 tr, INT_MIN);
496 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
497 tr, INT_MAX);
498 } else {
499 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
500 tr);
501 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
502 tr);
503 }
504}
505
506static void
507event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
508 struct task_struct *prev, struct task_struct *next)
509{
510 struct trace_array *tr = data;
511 struct trace_pid_list *pid_list;
512
513 pid_list = rcu_dereference_sched(tr->filtered_pids);
514
515 this_cpu_write(tr->trace_buffer.data->ignore_pid,
516 trace_ignore_this_task(pid_list, prev) &&
517 trace_ignore_this_task(pid_list, next));
518}
519
520static void
521event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
522 struct task_struct *prev, struct task_struct *next)
523{
524 struct trace_array *tr = data;
525 struct trace_pid_list *pid_list;
526
527 pid_list = rcu_dereference_sched(tr->filtered_pids);
528
529 this_cpu_write(tr->trace_buffer.data->ignore_pid,
530 trace_ignore_this_task(pid_list, next));
531}
532
533static void
534event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
535{
536 struct trace_array *tr = data;
537 struct trace_pid_list *pid_list;
538
539 /* Nothing to do if we are already tracing */
540 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
541 return;
542
543 pid_list = rcu_dereference_sched(tr->filtered_pids);
544
545 this_cpu_write(tr->trace_buffer.data->ignore_pid,
546 trace_ignore_this_task(pid_list, task));
547}
548
549static void
550event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
551{
552 struct trace_array *tr = data;
553 struct trace_pid_list *pid_list;
554
555 /* Nothing to do if we are not tracing */
556 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
557 return;
558
559 pid_list = rcu_dereference_sched(tr->filtered_pids);
560
561 /* Set tracing if current is enabled */
562 this_cpu_write(tr->trace_buffer.data->ignore_pid,
563 trace_ignore_this_task(pid_list, current));
564}
565
566static void __ftrace_clear_event_pids(struct trace_array *tr)
567{
568 struct trace_pid_list *pid_list;
569 struct trace_event_file *file;
570 int cpu;
571
572 pid_list = rcu_dereference_protected(tr->filtered_pids,
573 lockdep_is_held(&event_mutex));
574 if (!pid_list)
575 return;
576
577 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
578 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
579
580 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
581 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
582
583 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
584 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
585
586 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
587 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
588
589 list_for_each_entry(file, &tr->events, list) {
590 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
591 }
592
593 for_each_possible_cpu(cpu)
594 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
595
596 rcu_assign_pointer(tr->filtered_pids, NULL);
597
598 /* Wait till all users are no longer using pid filtering */
599 synchronize_sched();
600
601 trace_free_pid_list(pid_list);
602}
603
604static void ftrace_clear_event_pids(struct trace_array *tr)
605{
606 mutex_lock(&event_mutex);
607 __ftrace_clear_event_pids(tr);
608 mutex_unlock(&event_mutex);
609}
610
611static void __put_system(struct event_subsystem *system)
612{
613 struct event_filter *filter = system->filter;
614
615 WARN_ON_ONCE(system_refcount(system) == 0);
616 if (system_refcount_dec(system))
617 return;
618
619 list_del(&system->list);
620
621 if (filter) {
622 kfree(filter->filter_string);
623 kfree(filter);
624 }
625 kfree_const(system->name);
626 kfree(system);
627}
628
629static void __get_system(struct event_subsystem *system)
630{
631 WARN_ON_ONCE(system_refcount(system) == 0);
632 system_refcount_inc(system);
633}
634
635static void __get_system_dir(struct trace_subsystem_dir *dir)
636{
637 WARN_ON_ONCE(dir->ref_count == 0);
638 dir->ref_count++;
639 __get_system(dir->subsystem);
640}
641
642static void __put_system_dir(struct trace_subsystem_dir *dir)
643{
644 WARN_ON_ONCE(dir->ref_count == 0);
645 /* If the subsystem is about to be freed, the dir must be too */
646 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
647
648 __put_system(dir->subsystem);
649 if (!--dir->ref_count)
650 kfree(dir);
651}
652
653static void put_system(struct trace_subsystem_dir *dir)
654{
655 mutex_lock(&event_mutex);
656 __put_system_dir(dir);
657 mutex_unlock(&event_mutex);
658}
659
660static void remove_subsystem(struct trace_subsystem_dir *dir)
661{
662 if (!dir)
663 return;
664
665 if (!--dir->nr_events) {
666 tracefs_remove_recursive(dir->entry);
667 list_del(&dir->list);
668 __put_system_dir(dir);
669 }
670}
671
672static void remove_event_file_dir(struct trace_event_file *file)
673{
674 struct dentry *dir = file->dir;
675 struct dentry *child;
676
677 if (dir) {
678 spin_lock(&dir->d_lock); /* probably unneeded */
679 list_for_each_entry(child, &dir->d_subdirs, d_child) {
680 if (d_really_is_positive(child)) /* probably unneeded */
681 d_inode(child)->i_private = NULL;
682 }
683 spin_unlock(&dir->d_lock);
684
685 tracefs_remove_recursive(dir);
686 }
687
688 list_del(&file->list);
689 remove_subsystem(file->system);
690 free_event_filter(file->filter);
691 kmem_cache_free(file_cachep, file);
692}
693
694/*
695 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
696 */
697static int
698__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
699 const char *sub, const char *event, int set)
700{
701 struct trace_event_file *file;
702 struct trace_event_call *call;
703 const char *name;
704 int ret = -EINVAL;
705 int eret = 0;
706
707 list_for_each_entry(file, &tr->events, list) {
708
709 call = file->event_call;
710 name = trace_event_name(call);
711
712 if (!name || !call->class || !call->class->reg)
713 continue;
714
715 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
716 continue;
717
718 if (match &&
719 strcmp(match, name) != 0 &&
720 strcmp(match, call->class->system) != 0)
721 continue;
722
723 if (sub && strcmp(sub, call->class->system) != 0)
724 continue;
725
726 if (event && strcmp(event, name) != 0)
727 continue;
728
729 ret = ftrace_event_enable_disable(file, set);
730
731 /*
732 * Save the first error and return that. Some events
733 * may still have been enabled, but let the user
734 * know that something went wrong.
735 */
736 if (ret && !eret)
737 eret = ret;
738
739 ret = eret;
740 }
741
742 return ret;
743}
744
745static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
746 const char *sub, const char *event, int set)
747{
748 int ret;
749
750 mutex_lock(&event_mutex);
751 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
752 mutex_unlock(&event_mutex);
753
754 return ret;
755}
756
757static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
758{
759 char *event = NULL, *sub = NULL, *match;
760 int ret;
761
762 /*
763 * The buf format can be <subsystem>:<event-name>
764 * *:<event-name> means any event by that name.
765 * :<event-name> is the same.
766 *
767 * <subsystem>:* means all events in that subsystem
768 * <subsystem>: means the same.
769 *
770 * <name> (no ':') means all events in a subsystem with
771 * the name <name> or any event that matches <name>
772 */
773
774 match = strsep(&buf, ":");
775 if (buf) {
776 sub = match;
777 event = buf;
778 match = NULL;
779
780 if (!strlen(sub) || strcmp(sub, "*") == 0)
781 sub = NULL;
782 if (!strlen(event) || strcmp(event, "*") == 0)
783 event = NULL;
784 }
785
786 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
787
788 /* Put back the colon to allow this to be called again */
789 if (buf)
790 *(buf - 1) = ':';
791
792 return ret;
793}
794
795/**
796 * trace_set_clr_event - enable or disable an event
797 * @system: system name to match (NULL for any system)
798 * @event: event name to match (NULL for all events, within system)
799 * @set: 1 to enable, 0 to disable
800 *
801 * This is a way for other parts of the kernel to enable or disable
802 * event recording.
803 *
804 * Returns 0 on success, -EINVAL if the parameters do not match any
805 * registered events.
806 */
807int trace_set_clr_event(const char *system, const char *event, int set)
808{
809 struct trace_array *tr = top_trace_array();
810
811 if (!tr)
812 return -ENODEV;
813
814 return __ftrace_set_clr_event(tr, NULL, system, event, set);
815}
816EXPORT_SYMBOL_GPL(trace_set_clr_event);
817
818/* 128 should be much more than enough */
819#define EVENT_BUF_SIZE 127
820
821static ssize_t
822ftrace_event_write(struct file *file, const char __user *ubuf,
823 size_t cnt, loff_t *ppos)
824{
825 struct trace_parser parser;
826 struct seq_file *m = file->private_data;
827 struct trace_array *tr = m->private;
828 ssize_t read, ret;
829
830 if (!cnt)
831 return 0;
832
833 ret = tracing_update_buffers();
834 if (ret < 0)
835 return ret;
836
837 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
838 return -ENOMEM;
839
840 read = trace_get_user(&parser, ubuf, cnt, ppos);
841
842 if (read >= 0 && trace_parser_loaded((&parser))) {
843 int set = 1;
844
845 if (*parser.buffer == '!')
846 set = 0;
847
848 parser.buffer[parser.idx] = 0;
849
850 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
851 if (ret)
852 goto out_put;
853 }
854
855 ret = read;
856
857 out_put:
858 trace_parser_put(&parser);
859
860 return ret;
861}
862
863static void *
864t_next(struct seq_file *m, void *v, loff_t *pos)
865{
866 struct trace_event_file *file = v;
867 struct trace_event_call *call;
868 struct trace_array *tr = m->private;
869
870 (*pos)++;
871
872 list_for_each_entry_continue(file, &tr->events, list) {
873 call = file->event_call;
874 /*
875 * The ftrace subsystem is for showing formats only.
876 * They can not be enabled or disabled via the event files.
877 */
878 if (call->class && call->class->reg &&
879 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
880 return file;
881 }
882
883 return NULL;
884}
885
886static void *t_start(struct seq_file *m, loff_t *pos)
887{
888 struct trace_event_file *file;
889 struct trace_array *tr = m->private;
890 loff_t l;
891
892 mutex_lock(&event_mutex);
893
894 file = list_entry(&tr->events, struct trace_event_file, list);
895 for (l = 0; l <= *pos; ) {
896 file = t_next(m, file, &l);
897 if (!file)
898 break;
899 }
900 return file;
901}
902
903static void *
904s_next(struct seq_file *m, void *v, loff_t *pos)
905{
906 struct trace_event_file *file = v;
907 struct trace_array *tr = m->private;
908
909 (*pos)++;
910
911 list_for_each_entry_continue(file, &tr->events, list) {
912 if (file->flags & EVENT_FILE_FL_ENABLED)
913 return file;
914 }
915
916 return NULL;
917}
918
919static void *s_start(struct seq_file *m, loff_t *pos)
920{
921 struct trace_event_file *file;
922 struct trace_array *tr = m->private;
923 loff_t l;
924
925 mutex_lock(&event_mutex);
926
927 file = list_entry(&tr->events, struct trace_event_file, list);
928 for (l = 0; l <= *pos; ) {
929 file = s_next(m, file, &l);
930 if (!file)
931 break;
932 }
933 return file;
934}
935
936static int t_show(struct seq_file *m, void *v)
937{
938 struct trace_event_file *file = v;
939 struct trace_event_call *call = file->event_call;
940
941 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
942 seq_printf(m, "%s:", call->class->system);
943 seq_printf(m, "%s\n", trace_event_name(call));
944
945 return 0;
946}
947
948static void t_stop(struct seq_file *m, void *p)
949{
950 mutex_unlock(&event_mutex);
951}
952
953static void *
954p_next(struct seq_file *m, void *v, loff_t *pos)
955{
956 struct trace_array *tr = m->private;
957 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
958
959 return trace_pid_next(pid_list, v, pos);
960}
961
962static void *p_start(struct seq_file *m, loff_t *pos)
963 __acquires(RCU)
964{
965 struct trace_pid_list *pid_list;
966 struct trace_array *tr = m->private;
967
968 /*
969 * Grab the mutex, to keep calls to p_next() having the same
970 * tr->filtered_pids as p_start() has.
971 * If we just passed the tr->filtered_pids around, then RCU would
972 * have been enough, but doing that makes things more complex.
973 */
974 mutex_lock(&event_mutex);
975 rcu_read_lock_sched();
976
977 pid_list = rcu_dereference_sched(tr->filtered_pids);
978
979 if (!pid_list)
980 return NULL;
981
982 return trace_pid_start(pid_list, pos);
983}
984
985static void p_stop(struct seq_file *m, void *p)
986 __releases(RCU)
987{
988 rcu_read_unlock_sched();
989 mutex_unlock(&event_mutex);
990}
991
992static ssize_t
993event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
994 loff_t *ppos)
995{
996 struct trace_event_file *file;
997 unsigned long flags;
998 char buf[4] = "0";
999
1000 mutex_lock(&event_mutex);
1001 file = event_file_data(filp);
1002 if (likely(file))
1003 flags = file->flags;
1004 mutex_unlock(&event_mutex);
1005
1006 if (!file)
1007 return -ENODEV;
1008
1009 if (flags & EVENT_FILE_FL_ENABLED &&
1010 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1011 strcpy(buf, "1");
1012
1013 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1014 flags & EVENT_FILE_FL_SOFT_MODE)
1015 strcat(buf, "*");
1016
1017 strcat(buf, "\n");
1018
1019 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1020}
1021
1022static ssize_t
1023event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1024 loff_t *ppos)
1025{
1026 struct trace_event_file *file;
1027 unsigned long val;
1028 int ret;
1029
1030 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1031 if (ret)
1032 return ret;
1033
1034 ret = tracing_update_buffers();
1035 if (ret < 0)
1036 return ret;
1037
1038 switch (val) {
1039 case 0:
1040 case 1:
1041 ret = -ENODEV;
1042 mutex_lock(&event_mutex);
1043 file = event_file_data(filp);
1044 if (likely(file))
1045 ret = ftrace_event_enable_disable(file, val);
1046 mutex_unlock(&event_mutex);
1047 break;
1048
1049 default:
1050 return -EINVAL;
1051 }
1052
1053 *ppos += cnt;
1054
1055 return ret ? ret : cnt;
1056}
1057
1058static ssize_t
1059system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1060 loff_t *ppos)
1061{
1062 const char set_to_char[4] = { '?', '0', '1', 'X' };
1063 struct trace_subsystem_dir *dir = filp->private_data;
1064 struct event_subsystem *system = dir->subsystem;
1065 struct trace_event_call *call;
1066 struct trace_event_file *file;
1067 struct trace_array *tr = dir->tr;
1068 char buf[2];
1069 int set = 0;
1070 int ret;
1071
1072 mutex_lock(&event_mutex);
1073 list_for_each_entry(file, &tr->events, list) {
1074 call = file->event_call;
1075 if (!trace_event_name(call) || !call->class || !call->class->reg)
1076 continue;
1077
1078 if (system && strcmp(call->class->system, system->name) != 0)
1079 continue;
1080
1081 /*
1082 * We need to find out if all the events are set
1083 * or if all events or cleared, or if we have
1084 * a mixture.
1085 */
1086 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1087
1088 /*
1089 * If we have a mixture, no need to look further.
1090 */
1091 if (set == 3)
1092 break;
1093 }
1094 mutex_unlock(&event_mutex);
1095
1096 buf[0] = set_to_char[set];
1097 buf[1] = '\n';
1098
1099 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1100
1101 return ret;
1102}
1103
1104static ssize_t
1105system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1106 loff_t *ppos)
1107{
1108 struct trace_subsystem_dir *dir = filp->private_data;
1109 struct event_subsystem *system = dir->subsystem;
1110 const char *name = NULL;
1111 unsigned long val;
1112 ssize_t ret;
1113
1114 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1115 if (ret)
1116 return ret;
1117
1118 ret = tracing_update_buffers();
1119 if (ret < 0)
1120 return ret;
1121
1122 if (val != 0 && val != 1)
1123 return -EINVAL;
1124
1125 /*
1126 * Opening of "enable" adds a ref count to system,
1127 * so the name is safe to use.
1128 */
1129 if (system)
1130 name = system->name;
1131
1132 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1133 if (ret)
1134 goto out;
1135
1136 ret = cnt;
1137
1138out:
1139 *ppos += cnt;
1140
1141 return ret;
1142}
1143
1144enum {
1145 FORMAT_HEADER = 1,
1146 FORMAT_FIELD_SEPERATOR = 2,
1147 FORMAT_PRINTFMT = 3,
1148};
1149
1150static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1151{
1152 struct trace_event_call *call = event_file_data(m->private);
1153 struct list_head *common_head = &ftrace_common_fields;
1154 struct list_head *head = trace_get_fields(call);
1155 struct list_head *node = v;
1156
1157 (*pos)++;
1158
1159 switch ((unsigned long)v) {
1160 case FORMAT_HEADER:
1161 node = common_head;
1162 break;
1163
1164 case FORMAT_FIELD_SEPERATOR:
1165 node = head;
1166 break;
1167
1168 case FORMAT_PRINTFMT:
1169 /* all done */
1170 return NULL;
1171 }
1172
1173 node = node->prev;
1174 if (node == common_head)
1175 return (void *)FORMAT_FIELD_SEPERATOR;
1176 else if (node == head)
1177 return (void *)FORMAT_PRINTFMT;
1178 else
1179 return node;
1180}
1181
1182static int f_show(struct seq_file *m, void *v)
1183{
1184 struct trace_event_call *call = event_file_data(m->private);
1185 struct ftrace_event_field *field;
1186 const char *array_descriptor;
1187
1188 switch ((unsigned long)v) {
1189 case FORMAT_HEADER:
1190 seq_printf(m, "name: %s\n", trace_event_name(call));
1191 seq_printf(m, "ID: %d\n", call->event.type);
1192 seq_puts(m, "format:\n");
1193 return 0;
1194
1195 case FORMAT_FIELD_SEPERATOR:
1196 seq_putc(m, '\n');
1197 return 0;
1198
1199 case FORMAT_PRINTFMT:
1200 seq_printf(m, "\nprint fmt: %s\n",
1201 call->print_fmt);
1202 return 0;
1203 }
1204
1205 field = list_entry(v, struct ftrace_event_field, link);
1206 /*
1207 * Smartly shows the array type(except dynamic array).
1208 * Normal:
1209 * field:TYPE VAR
1210 * If TYPE := TYPE[LEN], it is shown:
1211 * field:TYPE VAR[LEN]
1212 */
1213 array_descriptor = strchr(field->type, '[');
1214
1215 if (!strncmp(field->type, "__data_loc", 10))
1216 array_descriptor = NULL;
1217
1218 if (!array_descriptor)
1219 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1220 field->type, field->name, field->offset,
1221 field->size, !!field->is_signed);
1222 else
1223 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1224 (int)(array_descriptor - field->type),
1225 field->type, field->name,
1226 array_descriptor, field->offset,
1227 field->size, !!field->is_signed);
1228
1229 return 0;
1230}
1231
1232static void *f_start(struct seq_file *m, loff_t *pos)
1233{
1234 void *p = (void *)FORMAT_HEADER;
1235 loff_t l = 0;
1236
1237 /* ->stop() is called even if ->start() fails */
1238 mutex_lock(&event_mutex);
1239 if (!event_file_data(m->private))
1240 return ERR_PTR(-ENODEV);
1241
1242 while (l < *pos && p)
1243 p = f_next(m, p, &l);
1244
1245 return p;
1246}
1247
1248static void f_stop(struct seq_file *m, void *p)
1249{
1250 mutex_unlock(&event_mutex);
1251}
1252
1253static const struct seq_operations trace_format_seq_ops = {
1254 .start = f_start,
1255 .next = f_next,
1256 .stop = f_stop,
1257 .show = f_show,
1258};
1259
1260static int trace_format_open(struct inode *inode, struct file *file)
1261{
1262 struct seq_file *m;
1263 int ret;
1264
1265 ret = seq_open(file, &trace_format_seq_ops);
1266 if (ret < 0)
1267 return ret;
1268
1269 m = file->private_data;
1270 m->private = file;
1271
1272 return 0;
1273}
1274
1275static ssize_t
1276event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1277{
1278 int id = (long)event_file_data(filp);
1279 char buf[32];
1280 int len;
1281
1282 if (*ppos)
1283 return 0;
1284
1285 if (unlikely(!id))
1286 return -ENODEV;
1287
1288 len = sprintf(buf, "%d\n", id);
1289
1290 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1291}
1292
1293static ssize_t
1294event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1295 loff_t *ppos)
1296{
1297 struct trace_event_file *file;
1298 struct trace_seq *s;
1299 int r = -ENODEV;
1300
1301 if (*ppos)
1302 return 0;
1303
1304 s = kmalloc(sizeof(*s), GFP_KERNEL);
1305
1306 if (!s)
1307 return -ENOMEM;
1308
1309 trace_seq_init(s);
1310
1311 mutex_lock(&event_mutex);
1312 file = event_file_data(filp);
1313 if (file)
1314 print_event_filter(file, s);
1315 mutex_unlock(&event_mutex);
1316
1317 if (file)
1318 r = simple_read_from_buffer(ubuf, cnt, ppos,
1319 s->buffer, trace_seq_used(s));
1320
1321 kfree(s);
1322
1323 return r;
1324}
1325
1326static ssize_t
1327event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1328 loff_t *ppos)
1329{
1330 struct trace_event_file *file;
1331 char *buf;
1332 int err = -ENODEV;
1333
1334 if (cnt >= PAGE_SIZE)
1335 return -EINVAL;
1336
1337 buf = memdup_user_nul(ubuf, cnt);
1338 if (IS_ERR(buf))
1339 return PTR_ERR(buf);
1340
1341 mutex_lock(&event_mutex);
1342 file = event_file_data(filp);
1343 if (file)
1344 err = apply_event_filter(file, buf);
1345 mutex_unlock(&event_mutex);
1346
1347 kfree(buf);
1348 if (err < 0)
1349 return err;
1350
1351 *ppos += cnt;
1352
1353 return cnt;
1354}
1355
1356static LIST_HEAD(event_subsystems);
1357
1358static int subsystem_open(struct inode *inode, struct file *filp)
1359{
1360 struct event_subsystem *system = NULL;
1361 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1362 struct trace_array *tr;
1363 int ret;
1364
1365 if (tracing_is_disabled())
1366 return -ENODEV;
1367
1368 /* Make sure the system still exists */
1369 mutex_lock(&trace_types_lock);
1370 mutex_lock(&event_mutex);
1371 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1372 list_for_each_entry(dir, &tr->systems, list) {
1373 if (dir == inode->i_private) {
1374 /* Don't open systems with no events */
1375 if (dir->nr_events) {
1376 __get_system_dir(dir);
1377 system = dir->subsystem;
1378 }
1379 goto exit_loop;
1380 }
1381 }
1382 }
1383 exit_loop:
1384 mutex_unlock(&event_mutex);
1385 mutex_unlock(&trace_types_lock);
1386
1387 if (!system)
1388 return -ENODEV;
1389
1390 /* Some versions of gcc think dir can be uninitialized here */
1391 WARN_ON(!dir);
1392
1393 /* Still need to increment the ref count of the system */
1394 if (trace_array_get(tr) < 0) {
1395 put_system(dir);
1396 return -ENODEV;
1397 }
1398
1399 ret = tracing_open_generic(inode, filp);
1400 if (ret < 0) {
1401 trace_array_put(tr);
1402 put_system(dir);
1403 }
1404
1405 return ret;
1406}
1407
1408static int system_tr_open(struct inode *inode, struct file *filp)
1409{
1410 struct trace_subsystem_dir *dir;
1411 struct trace_array *tr = inode->i_private;
1412 int ret;
1413
1414 if (tracing_is_disabled())
1415 return -ENODEV;
1416
1417 if (trace_array_get(tr) < 0)
1418 return -ENODEV;
1419
1420 /* Make a temporary dir that has no system but points to tr */
1421 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1422 if (!dir) {
1423 trace_array_put(tr);
1424 return -ENOMEM;
1425 }
1426
1427 dir->tr = tr;
1428
1429 ret = tracing_open_generic(inode, filp);
1430 if (ret < 0) {
1431 trace_array_put(tr);
1432 kfree(dir);
1433 return ret;
1434 }
1435
1436 filp->private_data = dir;
1437
1438 return 0;
1439}
1440
1441static int subsystem_release(struct inode *inode, struct file *file)
1442{
1443 struct trace_subsystem_dir *dir = file->private_data;
1444
1445 trace_array_put(dir->tr);
1446
1447 /*
1448 * If dir->subsystem is NULL, then this is a temporary
1449 * descriptor that was made for a trace_array to enable
1450 * all subsystems.
1451 */
1452 if (dir->subsystem)
1453 put_system(dir);
1454 else
1455 kfree(dir);
1456
1457 return 0;
1458}
1459
1460static ssize_t
1461subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1462 loff_t *ppos)
1463{
1464 struct trace_subsystem_dir *dir = filp->private_data;
1465 struct event_subsystem *system = dir->subsystem;
1466 struct trace_seq *s;
1467 int r;
1468
1469 if (*ppos)
1470 return 0;
1471
1472 s = kmalloc(sizeof(*s), GFP_KERNEL);
1473 if (!s)
1474 return -ENOMEM;
1475
1476 trace_seq_init(s);
1477
1478 print_subsystem_event_filter(system, s);
1479 r = simple_read_from_buffer(ubuf, cnt, ppos,
1480 s->buffer, trace_seq_used(s));
1481
1482 kfree(s);
1483
1484 return r;
1485}
1486
1487static ssize_t
1488subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1489 loff_t *ppos)
1490{
1491 struct trace_subsystem_dir *dir = filp->private_data;
1492 char *buf;
1493 int err;
1494
1495 if (cnt >= PAGE_SIZE)
1496 return -EINVAL;
1497
1498 buf = memdup_user_nul(ubuf, cnt);
1499 if (IS_ERR(buf))
1500 return PTR_ERR(buf);
1501
1502 err = apply_subsystem_event_filter(dir, buf);
1503 kfree(buf);
1504 if (err < 0)
1505 return err;
1506
1507 *ppos += cnt;
1508
1509 return cnt;
1510}
1511
1512static ssize_t
1513show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1514{
1515 int (*func)(struct trace_seq *s) = filp->private_data;
1516 struct trace_seq *s;
1517 int r;
1518
1519 if (*ppos)
1520 return 0;
1521
1522 s = kmalloc(sizeof(*s), GFP_KERNEL);
1523 if (!s)
1524 return -ENOMEM;
1525
1526 trace_seq_init(s);
1527
1528 func(s);
1529 r = simple_read_from_buffer(ubuf, cnt, ppos,
1530 s->buffer, trace_seq_used(s));
1531
1532 kfree(s);
1533
1534 return r;
1535}
1536
1537static void ignore_task_cpu(void *data)
1538{
1539 struct trace_array *tr = data;
1540 struct trace_pid_list *pid_list;
1541
1542 /*
1543 * This function is called by on_each_cpu() while the
1544 * event_mutex is held.
1545 */
1546 pid_list = rcu_dereference_protected(tr->filtered_pids,
1547 mutex_is_locked(&event_mutex));
1548
1549 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1550 trace_ignore_this_task(pid_list, current));
1551}
1552
1553static ssize_t
1554ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1555 size_t cnt, loff_t *ppos)
1556{
1557 struct seq_file *m = filp->private_data;
1558 struct trace_array *tr = m->private;
1559 struct trace_pid_list *filtered_pids = NULL;
1560 struct trace_pid_list *pid_list;
1561 struct trace_event_file *file;
1562 ssize_t ret;
1563
1564 if (!cnt)
1565 return 0;
1566
1567 ret = tracing_update_buffers();
1568 if (ret < 0)
1569 return ret;
1570
1571 mutex_lock(&event_mutex);
1572
1573 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1574 lockdep_is_held(&event_mutex));
1575
1576 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1577 if (ret < 0)
1578 goto out;
1579
1580 rcu_assign_pointer(tr->filtered_pids, pid_list);
1581
1582 list_for_each_entry(file, &tr->events, list) {
1583 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1584 }
1585
1586 if (filtered_pids) {
1587 synchronize_sched();
1588 trace_free_pid_list(filtered_pids);
1589 } else if (pid_list) {
1590 /*
1591 * Register a probe that is called before all other probes
1592 * to set ignore_pid if next or prev do not match.
1593 * Register a probe this is called after all other probes
1594 * to only keep ignore_pid set if next pid matches.
1595 */
1596 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1597 tr, INT_MAX);
1598 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1599 tr, 0);
1600
1601 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1602 tr, INT_MAX);
1603 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1604 tr, 0);
1605
1606 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1607 tr, INT_MAX);
1608 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1609 tr, 0);
1610
1611 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1612 tr, INT_MAX);
1613 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1614 tr, 0);
1615 }
1616
1617 /*
1618 * Ignoring of pids is done at task switch. But we have to
1619 * check for those tasks that are currently running.
1620 * Always do this in case a pid was appended or removed.
1621 */
1622 on_each_cpu(ignore_task_cpu, tr, 1);
1623
1624 out:
1625 mutex_unlock(&event_mutex);
1626
1627 if (ret > 0)
1628 *ppos += ret;
1629
1630 return ret;
1631}
1632
1633static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1634static int ftrace_event_set_open(struct inode *inode, struct file *file);
1635static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1636static int ftrace_event_release(struct inode *inode, struct file *file);
1637
1638static const struct seq_operations show_event_seq_ops = {
1639 .start = t_start,
1640 .next = t_next,
1641 .show = t_show,
1642 .stop = t_stop,
1643};
1644
1645static const struct seq_operations show_set_event_seq_ops = {
1646 .start = s_start,
1647 .next = s_next,
1648 .show = t_show,
1649 .stop = t_stop,
1650};
1651
1652static const struct seq_operations show_set_pid_seq_ops = {
1653 .start = p_start,
1654 .next = p_next,
1655 .show = trace_pid_show,
1656 .stop = p_stop,
1657};
1658
1659static const struct file_operations ftrace_avail_fops = {
1660 .open = ftrace_event_avail_open,
1661 .read = seq_read,
1662 .llseek = seq_lseek,
1663 .release = seq_release,
1664};
1665
1666static const struct file_operations ftrace_set_event_fops = {
1667 .open = ftrace_event_set_open,
1668 .read = seq_read,
1669 .write = ftrace_event_write,
1670 .llseek = seq_lseek,
1671 .release = ftrace_event_release,
1672};
1673
1674static const struct file_operations ftrace_set_event_pid_fops = {
1675 .open = ftrace_event_set_pid_open,
1676 .read = seq_read,
1677 .write = ftrace_event_pid_write,
1678 .llseek = seq_lseek,
1679 .release = ftrace_event_release,
1680};
1681
1682static const struct file_operations ftrace_enable_fops = {
1683 .open = tracing_open_generic,
1684 .read = event_enable_read,
1685 .write = event_enable_write,
1686 .llseek = default_llseek,
1687};
1688
1689static const struct file_operations ftrace_event_format_fops = {
1690 .open = trace_format_open,
1691 .read = seq_read,
1692 .llseek = seq_lseek,
1693 .release = seq_release,
1694};
1695
1696static const struct file_operations ftrace_event_id_fops = {
1697 .read = event_id_read,
1698 .llseek = default_llseek,
1699};
1700
1701static const struct file_operations ftrace_event_filter_fops = {
1702 .open = tracing_open_generic,
1703 .read = event_filter_read,
1704 .write = event_filter_write,
1705 .llseek = default_llseek,
1706};
1707
1708static const struct file_operations ftrace_subsystem_filter_fops = {
1709 .open = subsystem_open,
1710 .read = subsystem_filter_read,
1711 .write = subsystem_filter_write,
1712 .llseek = default_llseek,
1713 .release = subsystem_release,
1714};
1715
1716static const struct file_operations ftrace_system_enable_fops = {
1717 .open = subsystem_open,
1718 .read = system_enable_read,
1719 .write = system_enable_write,
1720 .llseek = default_llseek,
1721 .release = subsystem_release,
1722};
1723
1724static const struct file_operations ftrace_tr_enable_fops = {
1725 .open = system_tr_open,
1726 .read = system_enable_read,
1727 .write = system_enable_write,
1728 .llseek = default_llseek,
1729 .release = subsystem_release,
1730};
1731
1732static const struct file_operations ftrace_show_header_fops = {
1733 .open = tracing_open_generic,
1734 .read = show_header,
1735 .llseek = default_llseek,
1736};
1737
1738static int
1739ftrace_event_open(struct inode *inode, struct file *file,
1740 const struct seq_operations *seq_ops)
1741{
1742 struct seq_file *m;
1743 int ret;
1744
1745 ret = seq_open(file, seq_ops);
1746 if (ret < 0)
1747 return ret;
1748 m = file->private_data;
1749 /* copy tr over to seq ops */
1750 m->private = inode->i_private;
1751
1752 return ret;
1753}
1754
1755static int ftrace_event_release(struct inode *inode, struct file *file)
1756{
1757 struct trace_array *tr = inode->i_private;
1758
1759 trace_array_put(tr);
1760
1761 return seq_release(inode, file);
1762}
1763
1764static int
1765ftrace_event_avail_open(struct inode *inode, struct file *file)
1766{
1767 const struct seq_operations *seq_ops = &show_event_seq_ops;
1768
1769 return ftrace_event_open(inode, file, seq_ops);
1770}
1771
1772static int
1773ftrace_event_set_open(struct inode *inode, struct file *file)
1774{
1775 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1776 struct trace_array *tr = inode->i_private;
1777 int ret;
1778
1779 if (trace_array_get(tr) < 0)
1780 return -ENODEV;
1781
1782 if ((file->f_mode & FMODE_WRITE) &&
1783 (file->f_flags & O_TRUNC))
1784 ftrace_clear_events(tr);
1785
1786 ret = ftrace_event_open(inode, file, seq_ops);
1787 if (ret < 0)
1788 trace_array_put(tr);
1789 return ret;
1790}
1791
1792static int
1793ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1794{
1795 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1796 struct trace_array *tr = inode->i_private;
1797 int ret;
1798
1799 if (trace_array_get(tr) < 0)
1800 return -ENODEV;
1801
1802 if ((file->f_mode & FMODE_WRITE) &&
1803 (file->f_flags & O_TRUNC))
1804 ftrace_clear_event_pids(tr);
1805
1806 ret = ftrace_event_open(inode, file, seq_ops);
1807 if (ret < 0)
1808 trace_array_put(tr);
1809 return ret;
1810}
1811
1812static struct event_subsystem *
1813create_new_subsystem(const char *name)
1814{
1815 struct event_subsystem *system;
1816
1817 /* need to create new entry */
1818 system = kmalloc(sizeof(*system), GFP_KERNEL);
1819 if (!system)
1820 return NULL;
1821
1822 system->ref_count = 1;
1823
1824 /* Only allocate if dynamic (kprobes and modules) */
1825 system->name = kstrdup_const(name, GFP_KERNEL);
1826 if (!system->name)
1827 goto out_free;
1828
1829 system->filter = NULL;
1830
1831 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1832 if (!system->filter)
1833 goto out_free;
1834
1835 list_add(&system->list, &event_subsystems);
1836
1837 return system;
1838
1839 out_free:
1840 kfree_const(system->name);
1841 kfree(system);
1842 return NULL;
1843}
1844
1845static struct dentry *
1846event_subsystem_dir(struct trace_array *tr, const char *name,
1847 struct trace_event_file *file, struct dentry *parent)
1848{
1849 struct trace_subsystem_dir *dir;
1850 struct event_subsystem *system;
1851 struct dentry *entry;
1852
1853 /* First see if we did not already create this dir */
1854 list_for_each_entry(dir, &tr->systems, list) {
1855 system = dir->subsystem;
1856 if (strcmp(system->name, name) == 0) {
1857 dir->nr_events++;
1858 file->system = dir;
1859 return dir->entry;
1860 }
1861 }
1862
1863 /* Now see if the system itself exists. */
1864 list_for_each_entry(system, &event_subsystems, list) {
1865 if (strcmp(system->name, name) == 0)
1866 break;
1867 }
1868 /* Reset system variable when not found */
1869 if (&system->list == &event_subsystems)
1870 system = NULL;
1871
1872 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1873 if (!dir)
1874 goto out_fail;
1875
1876 if (!system) {
1877 system = create_new_subsystem(name);
1878 if (!system)
1879 goto out_free;
1880 } else
1881 __get_system(system);
1882
1883 dir->entry = tracefs_create_dir(name, parent);
1884 if (!dir->entry) {
1885 pr_warn("Failed to create system directory %s\n", name);
1886 __put_system(system);
1887 goto out_free;
1888 }
1889
1890 dir->tr = tr;
1891 dir->ref_count = 1;
1892 dir->nr_events = 1;
1893 dir->subsystem = system;
1894 file->system = dir;
1895
1896 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1897 &ftrace_subsystem_filter_fops);
1898 if (!entry) {
1899 kfree(system->filter);
1900 system->filter = NULL;
1901 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1902 }
1903
1904 trace_create_file("enable", 0644, dir->entry, dir,
1905 &ftrace_system_enable_fops);
1906
1907 list_add(&dir->list, &tr->systems);
1908
1909 return dir->entry;
1910
1911 out_free:
1912 kfree(dir);
1913 out_fail:
1914 /* Only print this message if failed on memory allocation */
1915 if (!dir || !system)
1916 pr_warn("No memory to create event subsystem %s\n", name);
1917 return NULL;
1918}
1919
1920static int
1921event_create_dir(struct dentry *parent, struct trace_event_file *file)
1922{
1923 struct trace_event_call *call = file->event_call;
1924 struct trace_array *tr = file->tr;
1925 struct list_head *head;
1926 struct dentry *d_events;
1927 const char *name;
1928 int ret;
1929
1930 /*
1931 * If the trace point header did not define TRACE_SYSTEM
1932 * then the system would be called "TRACE_SYSTEM".
1933 */
1934 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1935 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1936 if (!d_events)
1937 return -ENOMEM;
1938 } else
1939 d_events = parent;
1940
1941 name = trace_event_name(call);
1942 file->dir = tracefs_create_dir(name, d_events);
1943 if (!file->dir) {
1944 pr_warn("Could not create tracefs '%s' directory\n", name);
1945 return -1;
1946 }
1947
1948 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1949 trace_create_file("enable", 0644, file->dir, file,
1950 &ftrace_enable_fops);
1951
1952#ifdef CONFIG_PERF_EVENTS
1953 if (call->event.type && call->class->reg)
1954 trace_create_file("id", 0444, file->dir,
1955 (void *)(long)call->event.type,
1956 &ftrace_event_id_fops);
1957#endif
1958
1959 /*
1960 * Other events may have the same class. Only update
1961 * the fields if they are not already defined.
1962 */
1963 head = trace_get_fields(call);
1964 if (list_empty(head)) {
1965 ret = call->class->define_fields(call);
1966 if (ret < 0) {
1967 pr_warn("Could not initialize trace point events/%s\n",
1968 name);
1969 return -1;
1970 }
1971 }
1972 trace_create_file("filter", 0644, file->dir, file,
1973 &ftrace_event_filter_fops);
1974
1975 /*
1976 * Only event directories that can be enabled should have
1977 * triggers.
1978 */
1979 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1980 trace_create_file("trigger", 0644, file->dir, file,
1981 &event_trigger_fops);
1982
1983#ifdef CONFIG_HIST_TRIGGERS
1984 trace_create_file("hist", 0444, file->dir, file,
1985 &event_hist_fops);
1986#endif
1987 trace_create_file("format", 0444, file->dir, call,
1988 &ftrace_event_format_fops);
1989
1990 return 0;
1991}
1992
1993static void remove_event_from_tracers(struct trace_event_call *call)
1994{
1995 struct trace_event_file *file;
1996 struct trace_array *tr;
1997
1998 do_for_each_event_file_safe(tr, file) {
1999 if (file->event_call != call)
2000 continue;
2001
2002 remove_event_file_dir(file);
2003 /*
2004 * The do_for_each_event_file_safe() is
2005 * a double loop. After finding the call for this
2006 * trace_array, we use break to jump to the next
2007 * trace_array.
2008 */
2009 break;
2010 } while_for_each_event_file();
2011}
2012
2013static void event_remove(struct trace_event_call *call)
2014{
2015 struct trace_array *tr;
2016 struct trace_event_file *file;
2017
2018 do_for_each_event_file(tr, file) {
2019 if (file->event_call != call)
2020 continue;
2021 ftrace_event_enable_disable(file, 0);
2022 /*
2023 * The do_for_each_event_file() is
2024 * a double loop. After finding the call for this
2025 * trace_array, we use break to jump to the next
2026 * trace_array.
2027 */
2028 break;
2029 } while_for_each_event_file();
2030
2031 if (call->event.funcs)
2032 __unregister_trace_event(&call->event);
2033 remove_event_from_tracers(call);
2034 list_del(&call->list);
2035}
2036
2037static int event_init(struct trace_event_call *call)
2038{
2039 int ret = 0;
2040 const char *name;
2041
2042 name = trace_event_name(call);
2043 if (WARN_ON(!name))
2044 return -EINVAL;
2045
2046 if (call->class->raw_init) {
2047 ret = call->class->raw_init(call);
2048 if (ret < 0 && ret != -ENOSYS)
2049 pr_warn("Could not initialize trace events/%s\n", name);
2050 }
2051
2052 return ret;
2053}
2054
2055static int
2056__register_event(struct trace_event_call *call, struct module *mod)
2057{
2058 int ret;
2059
2060 ret = event_init(call);
2061 if (ret < 0)
2062 return ret;
2063
2064 list_add(&call->list, &ftrace_events);
2065 call->mod = mod;
2066
2067 return 0;
2068}
2069
2070static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2071{
2072 int rlen;
2073 int elen;
2074
2075 /* Find the length of the enum value as a string */
2076 elen = snprintf(ptr, 0, "%ld", map->enum_value);
2077 /* Make sure there's enough room to replace the string with the value */
2078 if (len < elen)
2079 return NULL;
2080
2081 snprintf(ptr, elen + 1, "%ld", map->enum_value);
2082
2083 /* Get the rest of the string of ptr */
2084 rlen = strlen(ptr + len);
2085 memmove(ptr + elen, ptr + len, rlen);
2086 /* Make sure we end the new string */
2087 ptr[elen + rlen] = 0;
2088
2089 return ptr + elen;
2090}
2091
2092static void update_event_printk(struct trace_event_call *call,
2093 struct trace_enum_map *map)
2094{
2095 char *ptr;
2096 int quote = 0;
2097 int len = strlen(map->enum_string);
2098
2099 for (ptr = call->print_fmt; *ptr; ptr++) {
2100 if (*ptr == '\\') {
2101 ptr++;
2102 /* paranoid */
2103 if (!*ptr)
2104 break;
2105 continue;
2106 }
2107 if (*ptr == '"') {
2108 quote ^= 1;
2109 continue;
2110 }
2111 if (quote)
2112 continue;
2113 if (isdigit(*ptr)) {
2114 /* skip numbers */
2115 do {
2116 ptr++;
2117 /* Check for alpha chars like ULL */
2118 } while (isalnum(*ptr));
2119 if (!*ptr)
2120 break;
2121 /*
2122 * A number must have some kind of delimiter after
2123 * it, and we can ignore that too.
2124 */
2125 continue;
2126 }
2127 if (isalpha(*ptr) || *ptr == '_') {
2128 if (strncmp(map->enum_string, ptr, len) == 0 &&
2129 !isalnum(ptr[len]) && ptr[len] != '_') {
2130 ptr = enum_replace(ptr, map, len);
2131 /* Hmm, enum string smaller than value */
2132 if (WARN_ON_ONCE(!ptr))
2133 return;
2134 /*
2135 * No need to decrement here, as enum_replace()
2136 * returns the pointer to the character passed
2137 * the enum, and two enums can not be placed
2138 * back to back without something in between.
2139 * We can skip that something in between.
2140 */
2141 continue;
2142 }
2143 skip_more:
2144 do {
2145 ptr++;
2146 } while (isalnum(*ptr) || *ptr == '_');
2147 if (!*ptr)
2148 break;
2149 /*
2150 * If what comes after this variable is a '.' or
2151 * '->' then we can continue to ignore that string.
2152 */
2153 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2154 ptr += *ptr == '.' ? 1 : 2;
2155 if (!*ptr)
2156 break;
2157 goto skip_more;
2158 }
2159 /*
2160 * Once again, we can skip the delimiter that came
2161 * after the string.
2162 */
2163 continue;
2164 }
2165 }
2166}
2167
2168void trace_event_enum_update(struct trace_enum_map **map, int len)
2169{
2170 struct trace_event_call *call, *p;
2171 const char *last_system = NULL;
2172 int last_i;
2173 int i;
2174
2175 down_write(&trace_event_sem);
2176 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2177 /* events are usually grouped together with systems */
2178 if (!last_system || call->class->system != last_system) {
2179 last_i = 0;
2180 last_system = call->class->system;
2181 }
2182
2183 for (i = last_i; i < len; i++) {
2184 if (call->class->system == map[i]->system) {
2185 /* Save the first system if need be */
2186 if (!last_i)
2187 last_i = i;
2188 update_event_printk(call, map[i]);
2189 }
2190 }
2191 }
2192 up_write(&trace_event_sem);
2193}
2194
2195static struct trace_event_file *
2196trace_create_new_event(struct trace_event_call *call,
2197 struct trace_array *tr)
2198{
2199 struct trace_event_file *file;
2200
2201 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2202 if (!file)
2203 return NULL;
2204
2205 file->event_call = call;
2206 file->tr = tr;
2207 atomic_set(&file->sm_ref, 0);
2208 atomic_set(&file->tm_ref, 0);
2209 INIT_LIST_HEAD(&file->triggers);
2210 list_add(&file->list, &tr->events);
2211
2212 return file;
2213}
2214
2215/* Add an event to a trace directory */
2216static int
2217__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2218{
2219 struct trace_event_file *file;
2220
2221 file = trace_create_new_event(call, tr);
2222 if (!file)
2223 return -ENOMEM;
2224
2225 return event_create_dir(tr->event_dir, file);
2226}
2227
2228/*
2229 * Just create a decriptor for early init. A descriptor is required
2230 * for enabling events at boot. We want to enable events before
2231 * the filesystem is initialized.
2232 */
2233static __init int
2234__trace_early_add_new_event(struct trace_event_call *call,
2235 struct trace_array *tr)
2236{
2237 struct trace_event_file *file;
2238
2239 file = trace_create_new_event(call, tr);
2240 if (!file)
2241 return -ENOMEM;
2242
2243 return 0;
2244}
2245
2246struct ftrace_module_file_ops;
2247static void __add_event_to_tracers(struct trace_event_call *call);
2248
2249/* Add an additional event_call dynamically */
2250int trace_add_event_call(struct trace_event_call *call)
2251{
2252 int ret;
2253 mutex_lock(&trace_types_lock);
2254 mutex_lock(&event_mutex);
2255
2256 ret = __register_event(call, NULL);
2257 if (ret >= 0)
2258 __add_event_to_tracers(call);
2259
2260 mutex_unlock(&event_mutex);
2261 mutex_unlock(&trace_types_lock);
2262 return ret;
2263}
2264
2265/*
2266 * Must be called under locking of trace_types_lock, event_mutex and
2267 * trace_event_sem.
2268 */
2269static void __trace_remove_event_call(struct trace_event_call *call)
2270{
2271 event_remove(call);
2272 trace_destroy_fields(call);
2273 free_event_filter(call->filter);
2274 call->filter = NULL;
2275}
2276
2277static int probe_remove_event_call(struct trace_event_call *call)
2278{
2279 struct trace_array *tr;
2280 struct trace_event_file *file;
2281
2282#ifdef CONFIG_PERF_EVENTS
2283 if (call->perf_refcount)
2284 return -EBUSY;
2285#endif
2286 do_for_each_event_file(tr, file) {
2287 if (file->event_call != call)
2288 continue;
2289 /*
2290 * We can't rely on ftrace_event_enable_disable(enable => 0)
2291 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2292 * TRACE_REG_UNREGISTER.
2293 */
2294 if (file->flags & EVENT_FILE_FL_ENABLED)
2295 return -EBUSY;
2296 /*
2297 * The do_for_each_event_file_safe() is
2298 * a double loop. After finding the call for this
2299 * trace_array, we use break to jump to the next
2300 * trace_array.
2301 */
2302 break;
2303 } while_for_each_event_file();
2304
2305 __trace_remove_event_call(call);
2306
2307 return 0;
2308}
2309
2310/* Remove an event_call */
2311int trace_remove_event_call(struct trace_event_call *call)
2312{
2313 int ret;
2314
2315 mutex_lock(&trace_types_lock);
2316 mutex_lock(&event_mutex);
2317 down_write(&trace_event_sem);
2318 ret = probe_remove_event_call(call);
2319 up_write(&trace_event_sem);
2320 mutex_unlock(&event_mutex);
2321 mutex_unlock(&trace_types_lock);
2322
2323 return ret;
2324}
2325
2326#define for_each_event(event, start, end) \
2327 for (event = start; \
2328 (unsigned long)event < (unsigned long)end; \
2329 event++)
2330
2331#ifdef CONFIG_MODULES
2332
2333static void trace_module_add_events(struct module *mod)
2334{
2335 struct trace_event_call **call, **start, **end;
2336
2337 if (!mod->num_trace_events)
2338 return;
2339
2340 /* Don't add infrastructure for mods without tracepoints */
2341 if (trace_module_has_bad_taint(mod)) {
2342 pr_err("%s: module has bad taint, not creating trace events\n",
2343 mod->name);
2344 return;
2345 }
2346
2347 start = mod->trace_events;
2348 end = mod->trace_events + mod->num_trace_events;
2349
2350 for_each_event(call, start, end) {
2351 __register_event(*call, mod);
2352 __add_event_to_tracers(*call);
2353 }
2354}
2355
2356static void trace_module_remove_events(struct module *mod)
2357{
2358 struct trace_event_call *call, *p;
2359 bool clear_trace = false;
2360
2361 down_write(&trace_event_sem);
2362 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2363 if (call->mod == mod) {
2364 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2365 clear_trace = true;
2366 __trace_remove_event_call(call);
2367 }
2368 }
2369 up_write(&trace_event_sem);
2370
2371 /*
2372 * It is safest to reset the ring buffer if the module being unloaded
2373 * registered any events that were used. The only worry is if
2374 * a new module gets loaded, and takes on the same id as the events
2375 * of this module. When printing out the buffer, traced events left
2376 * over from this module may be passed to the new module events and
2377 * unexpected results may occur.
2378 */
2379 if (clear_trace)
2380 tracing_reset_all_online_cpus();
2381}
2382
2383static int trace_module_notify(struct notifier_block *self,
2384 unsigned long val, void *data)
2385{
2386 struct module *mod = data;
2387
2388 mutex_lock(&trace_types_lock);
2389 mutex_lock(&event_mutex);
2390 switch (val) {
2391 case MODULE_STATE_COMING:
2392 trace_module_add_events(mod);
2393 break;
2394 case MODULE_STATE_GOING:
2395 trace_module_remove_events(mod);
2396 break;
2397 }
2398 mutex_unlock(&event_mutex);
2399 mutex_unlock(&trace_types_lock);
2400
2401 return 0;
2402}
2403
2404static struct notifier_block trace_module_nb = {
2405 .notifier_call = trace_module_notify,
2406 .priority = 1, /* higher than trace.c module notify */
2407};
2408#endif /* CONFIG_MODULES */
2409
2410/* Create a new event directory structure for a trace directory. */
2411static void
2412__trace_add_event_dirs(struct trace_array *tr)
2413{
2414 struct trace_event_call *call;
2415 int ret;
2416
2417 list_for_each_entry(call, &ftrace_events, list) {
2418 ret = __trace_add_new_event(call, tr);
2419 if (ret < 0)
2420 pr_warn("Could not create directory for event %s\n",
2421 trace_event_name(call));
2422 }
2423}
2424
2425struct trace_event_file *
2426find_event_file(struct trace_array *tr, const char *system, const char *event)
2427{
2428 struct trace_event_file *file;
2429 struct trace_event_call *call;
2430 const char *name;
2431
2432 list_for_each_entry(file, &tr->events, list) {
2433
2434 call = file->event_call;
2435 name = trace_event_name(call);
2436
2437 if (!name || !call->class || !call->class->reg)
2438 continue;
2439
2440 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2441 continue;
2442
2443 if (strcmp(event, name) == 0 &&
2444 strcmp(system, call->class->system) == 0)
2445 return file;
2446 }
2447 return NULL;
2448}
2449
2450#ifdef CONFIG_DYNAMIC_FTRACE
2451
2452/* Avoid typos */
2453#define ENABLE_EVENT_STR "enable_event"
2454#define DISABLE_EVENT_STR "disable_event"
2455
2456struct event_probe_data {
2457 struct trace_event_file *file;
2458 unsigned long count;
2459 int ref;
2460 bool enable;
2461};
2462
2463static void
2464event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2465{
2466 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2467 struct event_probe_data *data = *pdata;
2468
2469 if (!data)
2470 return;
2471
2472 if (data->enable)
2473 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2474 else
2475 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2476}
2477
2478static void
2479event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2480{
2481 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2482 struct event_probe_data *data = *pdata;
2483
2484 if (!data)
2485 return;
2486
2487 if (!data->count)
2488 return;
2489
2490 /* Skip if the event is in a state we want to switch to */
2491 if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2492 return;
2493
2494 if (data->count != -1)
2495 (data->count)--;
2496
2497 event_enable_probe(ip, parent_ip, _data);
2498}
2499
2500static int
2501event_enable_print(struct seq_file *m, unsigned long ip,
2502 struct ftrace_probe_ops *ops, void *_data)
2503{
2504 struct event_probe_data *data = _data;
2505
2506 seq_printf(m, "%ps:", (void *)ip);
2507
2508 seq_printf(m, "%s:%s:%s",
2509 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2510 data->file->event_call->class->system,
2511 trace_event_name(data->file->event_call));
2512
2513 if (data->count == -1)
2514 seq_puts(m, ":unlimited\n");
2515 else
2516 seq_printf(m, ":count=%ld\n", data->count);
2517
2518 return 0;
2519}
2520
2521static int
2522event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2523 void **_data)
2524{
2525 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2526 struct event_probe_data *data = *pdata;
2527
2528 data->ref++;
2529 return 0;
2530}
2531
2532static void
2533event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2534 void **_data)
2535{
2536 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2537 struct event_probe_data *data = *pdata;
2538
2539 if (WARN_ON_ONCE(data->ref <= 0))
2540 return;
2541
2542 data->ref--;
2543 if (!data->ref) {
2544 /* Remove the SOFT_MODE flag */
2545 __ftrace_event_enable_disable(data->file, 0, 1);
2546 module_put(data->file->event_call->mod);
2547 kfree(data);
2548 }
2549 *pdata = NULL;
2550}
2551
2552static struct ftrace_probe_ops event_enable_probe_ops = {
2553 .func = event_enable_probe,
2554 .print = event_enable_print,
2555 .init = event_enable_init,
2556 .free = event_enable_free,
2557};
2558
2559static struct ftrace_probe_ops event_enable_count_probe_ops = {
2560 .func = event_enable_count_probe,
2561 .print = event_enable_print,
2562 .init = event_enable_init,
2563 .free = event_enable_free,
2564};
2565
2566static struct ftrace_probe_ops event_disable_probe_ops = {
2567 .func = event_enable_probe,
2568 .print = event_enable_print,
2569 .init = event_enable_init,
2570 .free = event_enable_free,
2571};
2572
2573static struct ftrace_probe_ops event_disable_count_probe_ops = {
2574 .func = event_enable_count_probe,
2575 .print = event_enable_print,
2576 .init = event_enable_init,
2577 .free = event_enable_free,
2578};
2579
2580static int
2581event_enable_func(struct ftrace_hash *hash,
2582 char *glob, char *cmd, char *param, int enabled)
2583{
2584 struct trace_array *tr = top_trace_array();
2585 struct trace_event_file *file;
2586 struct ftrace_probe_ops *ops;
2587 struct event_probe_data *data;
2588 const char *system;
2589 const char *event;
2590 char *number;
2591 bool enable;
2592 int ret;
2593
2594 if (!tr)
2595 return -ENODEV;
2596
2597 /* hash funcs only work with set_ftrace_filter */
2598 if (!enabled || !param)
2599 return -EINVAL;
2600
2601 system = strsep(¶m, ":");
2602 if (!param)
2603 return -EINVAL;
2604
2605 event = strsep(¶m, ":");
2606
2607 mutex_lock(&event_mutex);
2608
2609 ret = -EINVAL;
2610 file = find_event_file(tr, system, event);
2611 if (!file)
2612 goto out;
2613
2614 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2615
2616 if (enable)
2617 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2618 else
2619 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2620
2621 if (glob[0] == '!') {
2622 unregister_ftrace_function_probe_func(glob+1, ops);
2623 ret = 0;
2624 goto out;
2625 }
2626
2627 ret = -ENOMEM;
2628 data = kzalloc(sizeof(*data), GFP_KERNEL);
2629 if (!data)
2630 goto out;
2631
2632 data->enable = enable;
2633 data->count = -1;
2634 data->file = file;
2635
2636 if (!param)
2637 goto out_reg;
2638
2639 number = strsep(¶m, ":");
2640
2641 ret = -EINVAL;
2642 if (!strlen(number))
2643 goto out_free;
2644
2645 /*
2646 * We use the callback data field (which is a pointer)
2647 * as our counter.
2648 */
2649 ret = kstrtoul(number, 0, &data->count);
2650 if (ret)
2651 goto out_free;
2652
2653 out_reg:
2654 /* Don't let event modules unload while probe registered */
2655 ret = try_module_get(file->event_call->mod);
2656 if (!ret) {
2657 ret = -EBUSY;
2658 goto out_free;
2659 }
2660
2661 ret = __ftrace_event_enable_disable(file, 1, 1);
2662 if (ret < 0)
2663 goto out_put;
2664 ret = register_ftrace_function_probe(glob, ops, data);
2665 /*
2666 * The above returns on success the # of functions enabled,
2667 * but if it didn't find any functions it returns zero.
2668 * Consider no functions a failure too.
2669 */
2670 if (!ret) {
2671 ret = -ENOENT;
2672 goto out_disable;
2673 } else if (ret < 0)
2674 goto out_disable;
2675 /* Just return zero, not the number of enabled functions */
2676 ret = 0;
2677 out:
2678 mutex_unlock(&event_mutex);
2679 return ret;
2680
2681 out_disable:
2682 __ftrace_event_enable_disable(file, 0, 1);
2683 out_put:
2684 module_put(file->event_call->mod);
2685 out_free:
2686 kfree(data);
2687 goto out;
2688}
2689
2690static struct ftrace_func_command event_enable_cmd = {
2691 .name = ENABLE_EVENT_STR,
2692 .func = event_enable_func,
2693};
2694
2695static struct ftrace_func_command event_disable_cmd = {
2696 .name = DISABLE_EVENT_STR,
2697 .func = event_enable_func,
2698};
2699
2700static __init int register_event_cmds(void)
2701{
2702 int ret;
2703
2704 ret = register_ftrace_command(&event_enable_cmd);
2705 if (WARN_ON(ret < 0))
2706 return ret;
2707 ret = register_ftrace_command(&event_disable_cmd);
2708 if (WARN_ON(ret < 0))
2709 unregister_ftrace_command(&event_enable_cmd);
2710 return ret;
2711}
2712#else
2713static inline int register_event_cmds(void) { return 0; }
2714#endif /* CONFIG_DYNAMIC_FTRACE */
2715
2716/*
2717 * The top level array has already had its trace_event_file
2718 * descriptors created in order to allow for early events to
2719 * be recorded. This function is called after the tracefs has been
2720 * initialized, and we now have to create the files associated
2721 * to the events.
2722 */
2723static __init void
2724__trace_early_add_event_dirs(struct trace_array *tr)
2725{
2726 struct trace_event_file *file;
2727 int ret;
2728
2729
2730 list_for_each_entry(file, &tr->events, list) {
2731 ret = event_create_dir(tr->event_dir, file);
2732 if (ret < 0)
2733 pr_warn("Could not create directory for event %s\n",
2734 trace_event_name(file->event_call));
2735 }
2736}
2737
2738/*
2739 * For early boot up, the top trace array requires to have
2740 * a list of events that can be enabled. This must be done before
2741 * the filesystem is set up in order to allow events to be traced
2742 * early.
2743 */
2744static __init void
2745__trace_early_add_events(struct trace_array *tr)
2746{
2747 struct trace_event_call *call;
2748 int ret;
2749
2750 list_for_each_entry(call, &ftrace_events, list) {
2751 /* Early boot up should not have any modules loaded */
2752 if (WARN_ON_ONCE(call->mod))
2753 continue;
2754
2755 ret = __trace_early_add_new_event(call, tr);
2756 if (ret < 0)
2757 pr_warn("Could not create early event %s\n",
2758 trace_event_name(call));
2759 }
2760}
2761
2762/* Remove the event directory structure for a trace directory. */
2763static void
2764__trace_remove_event_dirs(struct trace_array *tr)
2765{
2766 struct trace_event_file *file, *next;
2767
2768 list_for_each_entry_safe(file, next, &tr->events, list)
2769 remove_event_file_dir(file);
2770}
2771
2772static void __add_event_to_tracers(struct trace_event_call *call)
2773{
2774 struct trace_array *tr;
2775
2776 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2777 __trace_add_new_event(call, tr);
2778}
2779
2780extern struct trace_event_call *__start_ftrace_events[];
2781extern struct trace_event_call *__stop_ftrace_events[];
2782
2783static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2784
2785static __init int setup_trace_event(char *str)
2786{
2787 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2788 ring_buffer_expanded = true;
2789 tracing_selftest_disabled = true;
2790
2791 return 1;
2792}
2793__setup("trace_event=", setup_trace_event);
2794
2795/* Expects to have event_mutex held when called */
2796static int
2797create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2798{
2799 struct dentry *d_events;
2800 struct dentry *entry;
2801
2802 entry = tracefs_create_file("set_event", 0644, parent,
2803 tr, &ftrace_set_event_fops);
2804 if (!entry) {
2805 pr_warn("Could not create tracefs 'set_event' entry\n");
2806 return -ENOMEM;
2807 }
2808
2809 d_events = tracefs_create_dir("events", parent);
2810 if (!d_events) {
2811 pr_warn("Could not create tracefs 'events' directory\n");
2812 return -ENOMEM;
2813 }
2814
2815 entry = trace_create_file("enable", 0644, d_events,
2816 tr, &ftrace_tr_enable_fops);
2817 if (!entry) {
2818 pr_warn("Could not create tracefs 'enable' entry\n");
2819 return -ENOMEM;
2820 }
2821
2822 /* There are not as crucial, just warn if they are not created */
2823
2824 entry = tracefs_create_file("set_event_pid", 0644, parent,
2825 tr, &ftrace_set_event_pid_fops);
2826 if (!entry)
2827 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2828
2829 /* ring buffer internal formats */
2830 entry = trace_create_file("header_page", 0444, d_events,
2831 ring_buffer_print_page_header,
2832 &ftrace_show_header_fops);
2833 if (!entry)
2834 pr_warn("Could not create tracefs 'header_page' entry\n");
2835
2836 entry = trace_create_file("header_event", 0444, d_events,
2837 ring_buffer_print_entry_header,
2838 &ftrace_show_header_fops);
2839 if (!entry)
2840 pr_warn("Could not create tracefs 'header_event' entry\n");
2841
2842 tr->event_dir = d_events;
2843
2844 return 0;
2845}
2846
2847/**
2848 * event_trace_add_tracer - add a instance of a trace_array to events
2849 * @parent: The parent dentry to place the files/directories for events in
2850 * @tr: The trace array associated with these events
2851 *
2852 * When a new instance is created, it needs to set up its events
2853 * directory, as well as other files associated with events. It also
2854 * creates the event hierachry in the @parent/events directory.
2855 *
2856 * Returns 0 on success.
2857 */
2858int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2859{
2860 int ret;
2861
2862 mutex_lock(&event_mutex);
2863
2864 ret = create_event_toplevel_files(parent, tr);
2865 if (ret)
2866 goto out_unlock;
2867
2868 down_write(&trace_event_sem);
2869 __trace_add_event_dirs(tr);
2870 up_write(&trace_event_sem);
2871
2872 out_unlock:
2873 mutex_unlock(&event_mutex);
2874
2875 return ret;
2876}
2877
2878/*
2879 * The top trace array already had its file descriptors created.
2880 * Now the files themselves need to be created.
2881 */
2882static __init int
2883early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2884{
2885 int ret;
2886
2887 mutex_lock(&event_mutex);
2888
2889 ret = create_event_toplevel_files(parent, tr);
2890 if (ret)
2891 goto out_unlock;
2892
2893 down_write(&trace_event_sem);
2894 __trace_early_add_event_dirs(tr);
2895 up_write(&trace_event_sem);
2896
2897 out_unlock:
2898 mutex_unlock(&event_mutex);
2899
2900 return ret;
2901}
2902
2903int event_trace_del_tracer(struct trace_array *tr)
2904{
2905 mutex_lock(&event_mutex);
2906
2907 /* Disable any event triggers and associated soft-disabled events */
2908 clear_event_triggers(tr);
2909
2910 /* Clear the pid list */
2911 __ftrace_clear_event_pids(tr);
2912
2913 /* Disable any running events */
2914 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2915
2916 /* Access to events are within rcu_read_lock_sched() */
2917 synchronize_sched();
2918
2919 down_write(&trace_event_sem);
2920 __trace_remove_event_dirs(tr);
2921 tracefs_remove_recursive(tr->event_dir);
2922 up_write(&trace_event_sem);
2923
2924 tr->event_dir = NULL;
2925
2926 mutex_unlock(&event_mutex);
2927
2928 return 0;
2929}
2930
2931static __init int event_trace_memsetup(void)
2932{
2933 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2934 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
2935 return 0;
2936}
2937
2938static __init void
2939early_enable_events(struct trace_array *tr, bool disable_first)
2940{
2941 char *buf = bootup_event_buf;
2942 char *token;
2943 int ret;
2944
2945 while (true) {
2946 token = strsep(&buf, ",");
2947
2948 if (!token)
2949 break;
2950
2951 if (*token) {
2952 /* Restarting syscalls requires that we stop them first */
2953 if (disable_first)
2954 ftrace_set_clr_event(tr, token, 0);
2955
2956 ret = ftrace_set_clr_event(tr, token, 1);
2957 if (ret)
2958 pr_warn("Failed to enable trace event: %s\n", token);
2959 }
2960
2961 /* Put back the comma to allow this to be called again */
2962 if (buf)
2963 *(buf - 1) = ',';
2964 }
2965}
2966
2967static __init int event_trace_enable(void)
2968{
2969 struct trace_array *tr = top_trace_array();
2970 struct trace_event_call **iter, *call;
2971 int ret;
2972
2973 if (!tr)
2974 return -ENODEV;
2975
2976 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2977
2978 call = *iter;
2979 ret = event_init(call);
2980 if (!ret)
2981 list_add(&call->list, &ftrace_events);
2982 }
2983
2984 /*
2985 * We need the top trace array to have a working set of trace
2986 * points at early init, before the debug files and directories
2987 * are created. Create the file entries now, and attach them
2988 * to the actual file dentries later.
2989 */
2990 __trace_early_add_events(tr);
2991
2992 early_enable_events(tr, false);
2993
2994 trace_printk_start_comm();
2995
2996 register_event_cmds();
2997
2998 register_trigger_cmds();
2999
3000 return 0;
3001}
3002
3003/*
3004 * event_trace_enable() is called from trace_event_init() first to
3005 * initialize events and perhaps start any events that are on the
3006 * command line. Unfortunately, there are some events that will not
3007 * start this early, like the system call tracepoints that need
3008 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3009 * is called before pid 1 starts, and this flag is never set, making
3010 * the syscall tracepoint never get reached, but the event is enabled
3011 * regardless (and not doing anything).
3012 */
3013static __init int event_trace_enable_again(void)
3014{
3015 struct trace_array *tr;
3016
3017 tr = top_trace_array();
3018 if (!tr)
3019 return -ENODEV;
3020
3021 early_enable_events(tr, true);
3022
3023 return 0;
3024}
3025
3026early_initcall(event_trace_enable_again);
3027
3028static __init int event_trace_init(void)
3029{
3030 struct trace_array *tr;
3031 struct dentry *d_tracer;
3032 struct dentry *entry;
3033 int ret;
3034
3035 tr = top_trace_array();
3036 if (!tr)
3037 return -ENODEV;
3038
3039 d_tracer = tracing_init_dentry();
3040 if (IS_ERR(d_tracer))
3041 return 0;
3042
3043 entry = tracefs_create_file("available_events", 0444, d_tracer,
3044 tr, &ftrace_avail_fops);
3045 if (!entry)
3046 pr_warn("Could not create tracefs 'available_events' entry\n");
3047
3048 if (trace_define_generic_fields())
3049 pr_warn("tracing: Failed to allocated generic fields");
3050
3051 if (trace_define_common_fields())
3052 pr_warn("tracing: Failed to allocate common fields");
3053
3054 ret = early_event_add_tracer(d_tracer, tr);
3055 if (ret)
3056 return ret;
3057
3058#ifdef CONFIG_MODULES
3059 ret = register_module_notifier(&trace_module_nb);
3060 if (ret)
3061 pr_warn("Failed to register trace events module notifier\n");
3062#endif
3063 return 0;
3064}
3065
3066void __init trace_event_init(void)
3067{
3068 event_trace_memsetup();
3069 init_ftrace_syscalls();
3070 event_trace_enable();
3071}
3072
3073fs_initcall(event_trace_init);
3074
3075#ifdef CONFIG_FTRACE_STARTUP_TEST
3076
3077static DEFINE_SPINLOCK(test_spinlock);
3078static DEFINE_SPINLOCK(test_spinlock_irq);
3079static DEFINE_MUTEX(test_mutex);
3080
3081static __init void test_work(struct work_struct *dummy)
3082{
3083 spin_lock(&test_spinlock);
3084 spin_lock_irq(&test_spinlock_irq);
3085 udelay(1);
3086 spin_unlock_irq(&test_spinlock_irq);
3087 spin_unlock(&test_spinlock);
3088
3089 mutex_lock(&test_mutex);
3090 msleep(1);
3091 mutex_unlock(&test_mutex);
3092}
3093
3094static __init int event_test_thread(void *unused)
3095{
3096 void *test_malloc;
3097
3098 test_malloc = kmalloc(1234, GFP_KERNEL);
3099 if (!test_malloc)
3100 pr_info("failed to kmalloc\n");
3101
3102 schedule_on_each_cpu(test_work);
3103
3104 kfree(test_malloc);
3105
3106 set_current_state(TASK_INTERRUPTIBLE);
3107 while (!kthread_should_stop()) {
3108 schedule();
3109 set_current_state(TASK_INTERRUPTIBLE);
3110 }
3111 __set_current_state(TASK_RUNNING);
3112
3113 return 0;
3114}
3115
3116/*
3117 * Do various things that may trigger events.
3118 */
3119static __init void event_test_stuff(void)
3120{
3121 struct task_struct *test_thread;
3122
3123 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3124 msleep(1);
3125 kthread_stop(test_thread);
3126}
3127
3128/*
3129 * For every trace event defined, we will test each trace point separately,
3130 * and then by groups, and finally all trace points.
3131 */
3132static __init void event_trace_self_tests(void)
3133{
3134 struct trace_subsystem_dir *dir;
3135 struct trace_event_file *file;
3136 struct trace_event_call *call;
3137 struct event_subsystem *system;
3138 struct trace_array *tr;
3139 int ret;
3140
3141 tr = top_trace_array();
3142 if (!tr)
3143 return;
3144
3145 pr_info("Running tests on trace events:\n");
3146
3147 list_for_each_entry(file, &tr->events, list) {
3148
3149 call = file->event_call;
3150
3151 /* Only test those that have a probe */
3152 if (!call->class || !call->class->probe)
3153 continue;
3154
3155/*
3156 * Testing syscall events here is pretty useless, but
3157 * we still do it if configured. But this is time consuming.
3158 * What we really need is a user thread to perform the
3159 * syscalls as we test.
3160 */
3161#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3162 if (call->class->system &&
3163 strcmp(call->class->system, "syscalls") == 0)
3164 continue;
3165#endif
3166
3167 pr_info("Testing event %s: ", trace_event_name(call));
3168
3169 /*
3170 * If an event is already enabled, someone is using
3171 * it and the self test should not be on.
3172 */
3173 if (file->flags & EVENT_FILE_FL_ENABLED) {
3174 pr_warn("Enabled event during self test!\n");
3175 WARN_ON_ONCE(1);
3176 continue;
3177 }
3178
3179 ftrace_event_enable_disable(file, 1);
3180 event_test_stuff();
3181 ftrace_event_enable_disable(file, 0);
3182
3183 pr_cont("OK\n");
3184 }
3185
3186 /* Now test at the sub system level */
3187
3188 pr_info("Running tests on trace event systems:\n");
3189
3190 list_for_each_entry(dir, &tr->systems, list) {
3191
3192 system = dir->subsystem;
3193
3194 /* the ftrace system is special, skip it */
3195 if (strcmp(system->name, "ftrace") == 0)
3196 continue;
3197
3198 pr_info("Testing event system %s: ", system->name);
3199
3200 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3201 if (WARN_ON_ONCE(ret)) {
3202 pr_warn("error enabling system %s\n",
3203 system->name);
3204 continue;
3205 }
3206
3207 event_test_stuff();
3208
3209 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3210 if (WARN_ON_ONCE(ret)) {
3211 pr_warn("error disabling system %s\n",
3212 system->name);
3213 continue;
3214 }
3215
3216 pr_cont("OK\n");
3217 }
3218
3219 /* Test with all events enabled */
3220
3221 pr_info("Running tests on all trace events:\n");
3222 pr_info("Testing all events: ");
3223
3224 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3225 if (WARN_ON_ONCE(ret)) {
3226 pr_warn("error enabling all events\n");
3227 return;
3228 }
3229
3230 event_test_stuff();
3231
3232 /* reset sysname */
3233 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3234 if (WARN_ON_ONCE(ret)) {
3235 pr_warn("error disabling all events\n");
3236 return;
3237 }
3238
3239 pr_cont("OK\n");
3240}
3241
3242#ifdef CONFIG_FUNCTION_TRACER
3243
3244static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3245
3246static struct trace_event_file event_trace_file __initdata;
3247
3248static void __init
3249function_test_events_call(unsigned long ip, unsigned long parent_ip,
3250 struct ftrace_ops *op, struct pt_regs *pt_regs)
3251{
3252 struct ring_buffer_event *event;
3253 struct ring_buffer *buffer;
3254 struct ftrace_entry *entry;
3255 unsigned long flags;
3256 long disabled;
3257 int cpu;
3258 int pc;
3259
3260 pc = preempt_count();
3261 preempt_disable_notrace();
3262 cpu = raw_smp_processor_id();
3263 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3264
3265 if (disabled != 1)
3266 goto out;
3267
3268 local_save_flags(flags);
3269
3270 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3271 TRACE_FN, sizeof(*entry),
3272 flags, pc);
3273 if (!event)
3274 goto out;
3275 entry = ring_buffer_event_data(event);
3276 entry->ip = ip;
3277 entry->parent_ip = parent_ip;
3278
3279 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3280 entry, flags, pc);
3281 out:
3282 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3283 preempt_enable_notrace();
3284}
3285
3286static struct ftrace_ops trace_ops __initdata =
3287{
3288 .func = function_test_events_call,
3289 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3290};
3291
3292static __init void event_trace_self_test_with_function(void)
3293{
3294 int ret;
3295
3296 event_trace_file.tr = top_trace_array();
3297 if (WARN_ON(!event_trace_file.tr))
3298 return;
3299
3300 ret = register_ftrace_function(&trace_ops);
3301 if (WARN_ON(ret < 0)) {
3302 pr_info("Failed to enable function tracer for event tests\n");
3303 return;
3304 }
3305 pr_info("Running tests again, along with the function tracer\n");
3306 event_trace_self_tests();
3307 unregister_ftrace_function(&trace_ops);
3308}
3309#else
3310static __init void event_trace_self_test_with_function(void)
3311{
3312}
3313#endif
3314
3315static __init int event_trace_self_tests_init(void)
3316{
3317 if (!tracing_selftest_disabled) {
3318 event_trace_self_tests();
3319 event_trace_self_test_with_function();
3320 }
3321
3322 return 0;
3323}
3324
3325late_initcall(event_trace_self_tests_init);
3326
3327#endif