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