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
36 DEFINE_MUTEX(event_mutex);
37
38 LIST_HEAD(ftrace_events);
39 static LIST_HEAD(ftrace_generic_fields);
40 static LIST_HEAD(ftrace_common_fields);
41 static bool eventdir_initialized;
42
43 static LIST_HEAD(module_strings);
44
45 struct 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
53 static struct kmem_cache *field_cachep;
54 static struct kmem_cache *file_cachep;
55
56 static inline int system_refcount(struct event_subsystem *system)
57 {
58 return system->ref_count;
59 }
60
61 static int system_refcount_inc(struct event_subsystem *system)
62 {
63 return system->ref_count++;
64 }
65
66 static 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
84 static 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
97 struct ftrace_event_field *
98 trace_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
115 static 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
143 int 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 }
156 EXPORT_SYMBOL_GPL(trace_define_field);
157
158 static 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
188 static 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
203 static 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
217 static 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 */
233 int 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 */
251 static 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 */
284 static 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
462 int 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 }
474 EXPORT_SYMBOL_GPL(trace_event_raw_init);
475
476 bool 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 }
493 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
494
495 void *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 }
525 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
526
527 int 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 }
563 EXPORT_SYMBOL_GPL(trace_event_reg);
564
565 void 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
587 void 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
609 static 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
721 int 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
727 static 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
733 static 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
744 static void
745 event_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
757 static void
758 event_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
772 void 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
787 static void
788 event_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
813 static void
814 event_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
830 static void
831 event_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
848 static void
849 event_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
867 static 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
882 static 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
925 static 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
932 static 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
950 static void __get_system(struct event_subsystem *system)
951 {
952 WARN_ON_ONCE(system_refcount(system) == 0);
953 system_refcount_inc(system);
954 }
955
956 static 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
963 static 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
974 static 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
981 static 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
993 void event_file_get(struct trace_event_file *file)
994 {
995 refcount_inc(&file->ref);
996 }
997
998 void event_file_put(struct trace_event_file *file)
999 {
1000 if (WARN_ON_ONCE(!refcount_read(&file->ref))) {
1001 if (file->flags & EVENT_FILE_FL_FREED)
1002 kmem_cache_free(file_cachep, file);
1003 return;
1004 }
1005
1006 if (refcount_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
1014 static 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 */
1027 static 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
1075 static 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
1087 int 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 */
1139 int 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 }
1148 EXPORT_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 */
1163 int 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 }
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE 127
1178
1179 static ssize_t
1180 ftrace_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
1219 static void *
1220 t_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
1242 static 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
1259 static void *
1260 s_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
1275 static 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
1292 static 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
1304 static void t_stop(struct seq_file *m, void *p)
1305 {
1306 mutex_unlock(&event_mutex);
1307 }
1308
1309 static 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
1323 static void *
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326 return __next(m, v, pos, TRACE_PIDS);
1327 }
1328
1329 static void *
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1331 {
1332 return __next(m, v, pos, TRACE_NO_PIDS);
1333 }
1334
1335 static 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
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1362 __acquires(RCU)
1363 {
1364 return __start(m, pos, TRACE_PIDS);
1365 }
1366
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1368 __acquires(RCU)
1369 {
1370 return __start(m, pos, TRACE_NO_PIDS);
1371 }
1372
1373 static 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
1380 static ssize_t
1381 event_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] = "";
1387
1388 mutex_lock(&event_mutex);
1389 file = event_file_file(filp);
1390 if (likely(file))
1391 flags = file->flags;
1392 mutex_unlock(&event_mutex);
1393
1394 if (!file)
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
1410 static ssize_t
1411 event_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_file(filp);
1428 if (likely(file)) {
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
1448 static ssize_t
1449 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450 loff_t *ppos)
1451 {
1452 const char set_to_char[4] = { '?', '', '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
1495 static ssize_t
1496 system_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
1529 out:
1530 *ppos += cnt;
1531
1532 return ret;
1533 }
1534
1535 enum {
1536 FORMAT_HEADER = 1,
1537 FORMAT_FIELD_SEPERATOR = 2,
1538 FORMAT_PRINTFMT = 3,
1539 };
1540
1541 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542 {
1543 struct trace_event_file *file = event_file_data(m->private);
1544 struct trace_event_call *call = file->event_call;
1545 struct list_head *common_head = &ftrace_common_fields;
1546 struct list_head *head = trace_get_fields(call);
1547 struct list_head *node = v;
1548
1549 (*pos)++;
1550
1551 switch ((unsigned long)v) {
1552 case FORMAT_HEADER:
1553 node = common_head;
1554 break;
1555
1556 case FORMAT_FIELD_SEPERATOR:
1557 node = head;
1558 break;
1559
1560 case FORMAT_PRINTFMT:
1561 /* all done */
1562 return NULL;
1563 }
1564
1565 node = node->prev;
1566 if (node == common_head)
1567 return (void *)FORMAT_FIELD_SEPERATOR;
1568 else if (node == head)
1569 return (void *)FORMAT_PRINTFMT;
1570 else
1571 return node;
1572 }
1573
1574 static int f_show(struct seq_file *m, void *v)
1575 {
1576 struct trace_event_file *file = event_file_data(m->private);
1577 struct trace_event_call *call = file->event_call;
1578 struct ftrace_event_field *field;
1579 const char *array_descriptor;
1580
1581 switch ((unsigned long)v) {
1582 case FORMAT_HEADER:
1583 seq_printf(m, "name: %s\n", trace_event_name(call));
1584 seq_printf(m, "ID: %d\n", call->event.type);
1585 seq_puts(m, "format:\n");
1586 return 0;
1587
1588 case FORMAT_FIELD_SEPERATOR:
1589 seq_putc(m, '\n');
1590 return 0;
1591
1592 case FORMAT_PRINTFMT:
1593 seq_printf(m, "\nprint fmt: %s\n",
1594 call->print_fmt);
1595 return 0;
1596 }
1597
1598 field = list_entry(v, struct ftrace_event_field, link);
1599 /*
1600 * Smartly shows the array type(except dynamic array).
1601 * Normal:
1602 * field:TYPE VAR
1603 * If TYPE := TYPE[LEN], it is shown:
1604 * field:TYPE VAR[LEN]
1605 */
1606 array_descriptor = strchr(field->type, '[');
1607
1608 if (str_has_prefix(field->type, "__data_loc"))
1609 array_descriptor = NULL;
1610
1611 if (!array_descriptor)
1612 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1613 field->type, field->name, field->offset,
1614 field->size, !!field->is_signed);
1615 else if (field->len)
1616 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1617 (int)(array_descriptor - field->type),
1618 field->type, field->name,
1619 field->len, field->offset,
1620 field->size, !!field->is_signed);
1621 else
1622 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1623 (int)(array_descriptor - field->type),
1624 field->type, field->name,
1625 field->offset, field->size, !!field->is_signed);
1626
1627 return 0;
1628 }
1629
1630 static void *f_start(struct seq_file *m, loff_t *pos)
1631 {
1632 struct trace_event_file *file;
1633 void *p = (void *)FORMAT_HEADER;
1634 loff_t l = 0;
1635
1636 /* ->stop() is called even if ->start() fails */
1637 mutex_lock(&event_mutex);
1638 file = event_file_file(m->private);
1639 if (!file)
1640 return ERR_PTR(-ENODEV);
1641
1642 while (l < *pos && p)
1643 p = f_next(m, p, &l);
1644
1645 return p;
1646 }
1647
1648 static void f_stop(struct seq_file *m, void *p)
1649 {
1650 mutex_unlock(&event_mutex);
1651 }
1652
1653 static const struct seq_operations trace_format_seq_ops = {
1654 .start = f_start,
1655 .next = f_next,
1656 .stop = f_stop,
1657 .show = f_show,
1658 };
1659
1660 static int trace_format_open(struct inode *inode, struct file *file)
1661 {
1662 struct seq_file *m;
1663 int ret;
1664
1665 /* Do we want to hide event format files on tracefs lockdown? */
1666
1667 ret = seq_open(file, &trace_format_seq_ops);
1668 if (ret < 0)
1669 return ret;
1670
1671 m = file->private_data;
1672 m->private = file;
1673
1674 return 0;
1675 }
1676
1677 #ifdef CONFIG_PERF_EVENTS
1678 static ssize_t
1679 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1680 {
1681 int id = (long)event_file_data(filp);
1682 char buf[32];
1683 int len;
1684
1685 if (unlikely(!id))
1686 return -ENODEV;
1687
1688 len = sprintf(buf, "%d\n", id);
1689
1690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1691 }
1692 #endif
1693
1694 static ssize_t
1695 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1696 loff_t *ppos)
1697 {
1698 struct trace_event_file *file;
1699 struct trace_seq *s;
1700 int r = -ENODEV;
1701
1702 if (*ppos)
1703 return 0;
1704
1705 s = kmalloc(sizeof(*s), GFP_KERNEL);
1706
1707 if (!s)
1708 return -ENOMEM;
1709
1710 trace_seq_init(s);
1711
1712 mutex_lock(&event_mutex);
1713 file = event_file_file(filp);
1714 if (file)
1715 print_event_filter(file, s);
1716 mutex_unlock(&event_mutex);
1717
1718 if (file)
1719 r = simple_read_from_buffer(ubuf, cnt, ppos,
1720 s->buffer, trace_seq_used(s));
1721
1722 kfree(s);
1723
1724 return r;
1725 }
1726
1727 static ssize_t
1728 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1729 loff_t *ppos)
1730 {
1731 struct trace_event_file *file;
1732 char *buf;
1733 int err = -ENODEV;
1734
1735 if (cnt >= PAGE_SIZE)
1736 return -EINVAL;
1737
1738 buf = memdup_user_nul(ubuf, cnt);
1739 if (IS_ERR(buf))
1740 return PTR_ERR(buf);
1741
1742 mutex_lock(&event_mutex);
1743 file = event_file_file(filp);
1744 if (file) {
1745 if (file->flags & EVENT_FILE_FL_FREED)
1746 err = -ENODEV;
1747 else
1748 err = apply_event_filter(file, buf);
1749 }
1750 mutex_unlock(&event_mutex);
1751
1752 kfree(buf);
1753 if (err < 0)
1754 return err;
1755
1756 *ppos += cnt;
1757
1758 return cnt;
1759 }
1760
1761 static LIST_HEAD(event_subsystems);
1762
1763 static int subsystem_open(struct inode *inode, struct file *filp)
1764 {
1765 struct trace_subsystem_dir *dir = NULL, *iter_dir;
1766 struct trace_array *tr = NULL, *iter_tr;
1767 struct event_subsystem *system = NULL;
1768 int ret;
1769
1770 if (tracing_is_disabled())
1771 return -ENODEV;
1772
1773 /* Make sure the system still exists */
1774 mutex_lock(&event_mutex);
1775 mutex_lock(&trace_types_lock);
1776 list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1777 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1778 if (iter_dir == inode->i_private) {
1779 /* Don't open systems with no events */
1780 tr = iter_tr;
1781 dir = iter_dir;
1782 if (dir->nr_events) {
1783 __get_system_dir(dir);
1784 system = dir->subsystem;
1785 }
1786 goto exit_loop;
1787 }
1788 }
1789 }
1790 exit_loop:
1791 mutex_unlock(&trace_types_lock);
1792 mutex_unlock(&event_mutex);
1793
1794 if (!system)
1795 return -ENODEV;
1796
1797 /* Still need to increment the ref count of the system */
1798 if (trace_array_get(tr) < 0) {
1799 put_system(dir);
1800 return -ENODEV;
1801 }
1802
1803 ret = tracing_open_generic(inode, filp);
1804 if (ret < 0) {
1805 trace_array_put(tr);
1806 put_system(dir);
1807 }
1808
1809 return ret;
1810 }
1811
1812 static int system_tr_open(struct inode *inode, struct file *filp)
1813 {
1814 struct trace_subsystem_dir *dir;
1815 struct trace_array *tr = inode->i_private;
1816 int ret;
1817
1818 /* Make a temporary dir that has no system but points to tr */
1819 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1820 if (!dir)
1821 return -ENOMEM;
1822
1823 ret = tracing_open_generic_tr(inode, filp);
1824 if (ret < 0) {
1825 kfree(dir);
1826 return ret;
1827 }
1828 dir->tr = tr;
1829 filp->private_data = dir;
1830
1831 return 0;
1832 }
1833
1834 static int subsystem_release(struct inode *inode, struct file *file)
1835 {
1836 struct trace_subsystem_dir *dir = file->private_data;
1837
1838 trace_array_put(dir->tr);
1839
1840 /*
1841 * If dir->subsystem is NULL, then this is a temporary
1842 * descriptor that was made for a trace_array to enable
1843 * all subsystems.
1844 */
1845 if (dir->subsystem)
1846 put_system(dir);
1847 else
1848 kfree(dir);
1849
1850 return 0;
1851 }
1852
1853 static ssize_t
1854 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1855 loff_t *ppos)
1856 {
1857 struct trace_subsystem_dir *dir = filp->private_data;
1858 struct event_subsystem *system = dir->subsystem;
1859 struct trace_seq *s;
1860 int r;
1861
1862 if (*ppos)
1863 return 0;
1864
1865 s = kmalloc(sizeof(*s), GFP_KERNEL);
1866 if (!s)
1867 return -ENOMEM;
1868
1869 trace_seq_init(s);
1870
1871 print_subsystem_event_filter(system, s);
1872 r = simple_read_from_buffer(ubuf, cnt, ppos,
1873 s->buffer, trace_seq_used(s));
1874
1875 kfree(s);
1876
1877 return r;
1878 }
1879
1880 static ssize_t
1881 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1882 loff_t *ppos)
1883 {
1884 struct trace_subsystem_dir *dir = filp->private_data;
1885 char *buf;
1886 int err;
1887
1888 if (cnt >= PAGE_SIZE)
1889 return -EINVAL;
1890
1891 buf = memdup_user_nul(ubuf, cnt);
1892 if (IS_ERR(buf))
1893 return PTR_ERR(buf);
1894
1895 err = apply_subsystem_event_filter(dir, buf);
1896 kfree(buf);
1897 if (err < 0)
1898 return err;
1899
1900 *ppos += cnt;
1901
1902 return cnt;
1903 }
1904
1905 static ssize_t
1906 show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1907 {
1908 struct trace_array *tr = filp->private_data;
1909 struct trace_seq *s;
1910 int r;
1911
1912 if (*ppos)
1913 return 0;
1914
1915 s = kmalloc(sizeof(*s), GFP_KERNEL);
1916 if (!s)
1917 return -ENOMEM;
1918
1919 trace_seq_init(s);
1920
1921 ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1922 r = simple_read_from_buffer(ubuf, cnt, ppos,
1923 s->buffer, trace_seq_used(s));
1924
1925 kfree(s);
1926
1927 return r;
1928 }
1929
1930 static ssize_t
1931 show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1932 {
1933 struct trace_seq *s;
1934 int r;
1935
1936 if (*ppos)
1937 return 0;
1938
1939 s = kmalloc(sizeof(*s), GFP_KERNEL);
1940 if (!s)
1941 return -ENOMEM;
1942
1943 trace_seq_init(s);
1944
1945 ring_buffer_print_entry_header(s);
1946 r = simple_read_from_buffer(ubuf, cnt, ppos,
1947 s->buffer, trace_seq_used(s));
1948
1949 kfree(s);
1950
1951 return r;
1952 }
1953
1954 static void ignore_task_cpu(void *data)
1955 {
1956 struct trace_array *tr = data;
1957 struct trace_pid_list *pid_list;
1958 struct trace_pid_list *no_pid_list;
1959
1960 /*
1961 * This function is called by on_each_cpu() while the
1962 * event_mutex is held.
1963 */
1964 pid_list = rcu_dereference_protected(tr->filtered_pids,
1965 mutex_is_locked(&event_mutex));
1966 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1967 mutex_is_locked(&event_mutex));
1968
1969 this_cpu_write(tr->array_buffer.data->ignore_pid,
1970 trace_ignore_this_task(pid_list, no_pid_list, current));
1971 }
1972
1973 static void register_pid_events(struct trace_array *tr)
1974 {
1975 /*
1976 * Register a probe that is called before all other probes
1977 * to set ignore_pid if next or prev do not match.
1978 * Register a probe this is called after all other probes
1979 * to only keep ignore_pid set if next pid matches.
1980 */
1981 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1982 tr, INT_MAX);
1983 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1984 tr, 0);
1985
1986 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1987 tr, INT_MAX);
1988 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1989 tr, 0);
1990
1991 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1992 tr, INT_MAX);
1993 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1994 tr, 0);
1995
1996 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1997 tr, INT_MAX);
1998 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1999 tr, 0);
2000 }
2001
2002 static ssize_t
2003 event_pid_write(struct file *filp, const char __user *ubuf,
2004 size_t cnt, loff_t *ppos, int type)
2005 {
2006 struct seq_file *m = filp->private_data;
2007 struct trace_array *tr = m->private;
2008 struct trace_pid_list *filtered_pids = NULL;
2009 struct trace_pid_list *other_pids = NULL;
2010 struct trace_pid_list *pid_list;
2011 struct trace_event_file *file;
2012 ssize_t ret;
2013
2014 if (!cnt)
2015 return 0;
2016
2017 ret = tracing_update_buffers(tr);
2018 if (ret < 0)
2019 return ret;
2020
2021 mutex_lock(&event_mutex);
2022
2023 if (type == TRACE_PIDS) {
2024 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2025 lockdep_is_held(&event_mutex));
2026 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2027 lockdep_is_held(&event_mutex));
2028 } else {
2029 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2030 lockdep_is_held(&event_mutex));
2031 other_pids = rcu_dereference_protected(tr->filtered_pids,
2032 lockdep_is_held(&event_mutex));
2033 }
2034
2035 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2036 if (ret < 0)
2037 goto out;
2038
2039 if (type == TRACE_PIDS)
2040 rcu_assign_pointer(tr->filtered_pids, pid_list);
2041 else
2042 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2043
2044 list_for_each_entry(file, &tr->events, list) {
2045 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2046 }
2047
2048 if (filtered_pids) {
2049 tracepoint_synchronize_unregister();
2050 trace_pid_list_free(filtered_pids);
2051 } else if (pid_list && !other_pids) {
2052 register_pid_events(tr);
2053 }
2054
2055 /*
2056 * Ignoring of pids is done at task switch. But we have to
2057 * check for those tasks that are currently running.
2058 * Always do this in case a pid was appended or removed.
2059 */
2060 on_each_cpu(ignore_task_cpu, tr, 1);
2061
2062 out:
2063 mutex_unlock(&event_mutex);
2064
2065 if (ret > 0)
2066 *ppos += ret;
2067
2068 return ret;
2069 }
2070
2071 static ssize_t
2072 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2073 size_t cnt, loff_t *ppos)
2074 {
2075 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2076 }
2077
2078 static ssize_t
2079 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2080 size_t cnt, loff_t *ppos)
2081 {
2082 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2083 }
2084
2085 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2086 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2087 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2088 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2089 static int ftrace_event_release(struct inode *inode, struct file *file);
2090
2091 static const struct seq_operations show_event_seq_ops = {
2092 .start = t_start,
2093 .next = t_next,
2094 .show = t_show,
2095 .stop = t_stop,
2096 };
2097
2098 static const struct seq_operations show_set_event_seq_ops = {
2099 .start = s_start,
2100 .next = s_next,
2101 .show = t_show,
2102 .stop = t_stop,
2103 };
2104
2105 static const struct seq_operations show_set_pid_seq_ops = {
2106 .start = p_start,
2107 .next = p_next,
2108 .show = trace_pid_show,
2109 .stop = p_stop,
2110 };
2111
2112 static const struct seq_operations show_set_no_pid_seq_ops = {
2113 .start = np_start,
2114 .next = np_next,
2115 .show = trace_pid_show,
2116 .stop = p_stop,
2117 };
2118
2119 static const struct file_operations ftrace_avail_fops = {
2120 .open = ftrace_event_avail_open,
2121 .read = seq_read,
2122 .llseek = seq_lseek,
2123 .release = seq_release,
2124 };
2125
2126 static const struct file_operations ftrace_set_event_fops = {
2127 .open = ftrace_event_set_open,
2128 .read = seq_read,
2129 .write = ftrace_event_write,
2130 .llseek = seq_lseek,
2131 .release = ftrace_event_release,
2132 };
2133
2134 static const struct file_operations ftrace_set_event_pid_fops = {
2135 .open = ftrace_event_set_pid_open,
2136 .read = seq_read,
2137 .write = ftrace_event_pid_write,
2138 .llseek = seq_lseek,
2139 .release = ftrace_event_release,
2140 };
2141
2142 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2143 .open = ftrace_event_set_npid_open,
2144 .read = seq_read,
2145 .write = ftrace_event_npid_write,
2146 .llseek = seq_lseek,
2147 .release = ftrace_event_release,
2148 };
2149
2150 static const struct file_operations ftrace_enable_fops = {
2151 .open = tracing_open_file_tr,
2152 .read = event_enable_read,
2153 .write = event_enable_write,
2154 .release = tracing_release_file_tr,
2155 .llseek = default_llseek,
2156 };
2157
2158 static const struct file_operations ftrace_event_format_fops = {
2159 .open = trace_format_open,
2160 .read = seq_read,
2161 .llseek = seq_lseek,
2162 .release = seq_release,
2163 };
2164
2165 #ifdef CONFIG_PERF_EVENTS
2166 static const struct file_operations ftrace_event_id_fops = {
2167 .read = event_id_read,
2168 .llseek = default_llseek,
2169 };
2170 #endif
2171
2172 static const struct file_operations ftrace_event_filter_fops = {
2173 .open = tracing_open_file_tr,
2174 .read = event_filter_read,
2175 .write = event_filter_write,
2176 .release = tracing_release_file_tr,
2177 .llseek = default_llseek,
2178 };
2179
2180 static const struct file_operations ftrace_subsystem_filter_fops = {
2181 .open = subsystem_open,
2182 .read = subsystem_filter_read,
2183 .write = subsystem_filter_write,
2184 .llseek = default_llseek,
2185 .release = subsystem_release,
2186 };
2187
2188 static const struct file_operations ftrace_system_enable_fops = {
2189 .open = subsystem_open,
2190 .read = system_enable_read,
2191 .write = system_enable_write,
2192 .llseek = default_llseek,
2193 .release = subsystem_release,
2194 };
2195
2196 static const struct file_operations ftrace_tr_enable_fops = {
2197 .open = system_tr_open,
2198 .read = system_enable_read,
2199 .write = system_enable_write,
2200 .llseek = default_llseek,
2201 .release = subsystem_release,
2202 };
2203
2204 static const struct file_operations ftrace_show_header_page_fops = {
2205 .open = tracing_open_generic_tr,
2206 .read = show_header_page_file,
2207 .llseek = default_llseek,
2208 .release = tracing_release_generic_tr,
2209 };
2210
2211 static const struct file_operations ftrace_show_header_event_fops = {
2212 .open = tracing_open_generic_tr,
2213 .read = show_header_event_file,
2214 .llseek = default_llseek,
2215 .release = tracing_release_generic_tr,
2216 };
2217
2218 static int
2219 ftrace_event_open(struct inode *inode, struct file *file,
2220 const struct seq_operations *seq_ops)
2221 {
2222 struct seq_file *m;
2223 int ret;
2224
2225 ret = security_locked_down(LOCKDOWN_TRACEFS);
2226 if (ret)
2227 return ret;
2228
2229 ret = seq_open(file, seq_ops);
2230 if (ret < 0)
2231 return ret;
2232 m = file->private_data;
2233 /* copy tr over to seq ops */
2234 m->private = inode->i_private;
2235
2236 return ret;
2237 }
2238
2239 static int ftrace_event_release(struct inode *inode, struct file *file)
2240 {
2241 struct trace_array *tr = inode->i_private;
2242
2243 trace_array_put(tr);
2244
2245 return seq_release(inode, file);
2246 }
2247
2248 static int
2249 ftrace_event_avail_open(struct inode *inode, struct file *file)
2250 {
2251 const struct seq_operations *seq_ops = &show_event_seq_ops;
2252
2253 /* Checks for tracefs lockdown */
2254 return ftrace_event_open(inode, file, seq_ops);
2255 }
2256
2257 static int
2258 ftrace_event_set_open(struct inode *inode, struct file *file)
2259 {
2260 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2261 struct trace_array *tr = inode->i_private;
2262 int ret;
2263
2264 ret = tracing_check_open_get_tr(tr);
2265 if (ret)
2266 return ret;
2267
2268 if ((file->f_mode & FMODE_WRITE) &&
2269 (file->f_flags & O_TRUNC))
2270 ftrace_clear_events(tr);
2271
2272 ret = ftrace_event_open(inode, file, seq_ops);
2273 if (ret < 0)
2274 trace_array_put(tr);
2275 return ret;
2276 }
2277
2278 static int
2279 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2280 {
2281 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2282 struct trace_array *tr = inode->i_private;
2283 int ret;
2284
2285 ret = tracing_check_open_get_tr(tr);
2286 if (ret)
2287 return ret;
2288
2289 if ((file->f_mode & FMODE_WRITE) &&
2290 (file->f_flags & O_TRUNC))
2291 ftrace_clear_event_pids(tr, TRACE_PIDS);
2292
2293 ret = ftrace_event_open(inode, file, seq_ops);
2294 if (ret < 0)
2295 trace_array_put(tr);
2296 return ret;
2297 }
2298
2299 static int
2300 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2301 {
2302 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2303 struct trace_array *tr = inode->i_private;
2304 int ret;
2305
2306 ret = tracing_check_open_get_tr(tr);
2307 if (ret)
2308 return ret;
2309
2310 if ((file->f_mode & FMODE_WRITE) &&
2311 (file->f_flags & O_TRUNC))
2312 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2313
2314 ret = ftrace_event_open(inode, file, seq_ops);
2315 if (ret < 0)
2316 trace_array_put(tr);
2317 return ret;
2318 }
2319
2320 static struct event_subsystem *
2321 create_new_subsystem(const char *name)
2322 {
2323 struct event_subsystem *system;
2324
2325 /* need to create new entry */
2326 system = kmalloc(sizeof(*system), GFP_KERNEL);
2327 if (!system)
2328 return NULL;
2329
2330 system->ref_count = 1;
2331
2332 /* Only allocate if dynamic (kprobes and modules) */
2333 system->name = kstrdup_const(name, GFP_KERNEL);
2334 if (!system->name)
2335 goto out_free;
2336
2337 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2338 if (!system->filter)
2339 goto out_free;
2340
2341 list_add(&system->list, &event_subsystems);
2342
2343 return system;
2344
2345 out_free:
2346 kfree_const(system->name);
2347 kfree(system);
2348 return NULL;
2349 }
2350
2351 static int system_callback(const char *name, umode_t *mode, void **data,
2352 const struct file_operations **fops)
2353 {
2354 if (strcmp(name, "filter") == 0)
2355 *fops = &ftrace_subsystem_filter_fops;
2356
2357 else if (strcmp(name, "enable") == 0)
2358 *fops = &ftrace_system_enable_fops;
2359
2360 else
2361 return 0;
2362
2363 *mode = TRACE_MODE_WRITE;
2364 return 1;
2365 }
2366
2367 static struct eventfs_inode *
2368 event_subsystem_dir(struct trace_array *tr, const char *name,
2369 struct trace_event_file *file, struct eventfs_inode *parent)
2370 {
2371 struct event_subsystem *system, *iter;
2372 struct trace_subsystem_dir *dir;
2373 struct eventfs_inode *ei;
2374 int nr_entries;
2375 static struct eventfs_entry system_entries[] = {
2376 {
2377 .name = "filter",
2378 .callback = system_callback,
2379 },
2380 {
2381 .name = "enable",
2382 .callback = system_callback,
2383 }
2384 };
2385
2386 /* First see if we did not already create this dir */
2387 list_for_each_entry(dir, &tr->systems, list) {
2388 system = dir->subsystem;
2389 if (strcmp(system->name, name) == 0) {
2390 dir->nr_events++;
2391 file->system = dir;
2392 return dir->ei;
2393 }
2394 }
2395
2396 /* Now see if the system itself exists. */
2397 system = NULL;
2398 list_for_each_entry(iter, &event_subsystems, list) {
2399 if (strcmp(iter->name, name) == 0) {
2400 system = iter;
2401 break;
2402 }
2403 }
2404
2405 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2406 if (!dir)
2407 goto out_fail;
2408
2409 if (!system) {
2410 system = create_new_subsystem(name);
2411 if (!system)
2412 goto out_free;
2413 } else
2414 __get_system(system);
2415
2416 /* ftrace only has directories no files */
2417 if (strcmp(name, "ftrace") == 0)
2418 nr_entries = 0;
2419 else
2420 nr_entries = ARRAY_SIZE(system_entries);
2421
2422 ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2423 if (IS_ERR(ei)) {
2424 pr_warn("Failed to create system directory %s\n", name);
2425 __put_system(system);
2426 goto out_free;
2427 }
2428
2429 dir->ei = ei;
2430 dir->tr = tr;
2431 dir->ref_count = 1;
2432 dir->nr_events = 1;
2433 dir->subsystem = system;
2434 file->system = dir;
2435
2436 list_add(&dir->list, &tr->systems);
2437
2438 return dir->ei;
2439
2440 out_free:
2441 kfree(dir);
2442 out_fail:
2443 /* Only print this message if failed on memory allocation */
2444 if (!dir || !system)
2445 pr_warn("No memory to create event subsystem %s\n", name);
2446 return NULL;
2447 }
2448
2449 static int
2450 event_define_fields(struct trace_event_call *call)
2451 {
2452 struct list_head *head;
2453 int ret = 0;
2454
2455 /*
2456 * Other events may have the same class. Only update
2457 * the fields if they are not already defined.
2458 */
2459 head = trace_get_fields(call);
2460 if (list_empty(head)) {
2461 struct trace_event_fields *field = call->class->fields_array;
2462 unsigned int offset = sizeof(struct trace_entry);
2463
2464 for (; field->type; field++) {
2465 if (field->type == TRACE_FUNCTION_TYPE) {
2466 field->define_fields(call);
2467 break;
2468 }
2469
2470 offset = ALIGN(offset, field->align);
2471 ret = trace_define_field_ext(call, field->type, field->name,
2472 offset, field->size,
2473 field->is_signed, field->filter_type,
2474 field->len);
2475 if (WARN_ON_ONCE(ret)) {
2476 pr_err("error code is %d\n", ret);
2477 break;
2478 }
2479
2480 offset += field->size;
2481 }
2482 }
2483
2484 return ret;
2485 }
2486
2487 static int event_callback(const char *name, umode_t *mode, void **data,
2488 const struct file_operations **fops)
2489 {
2490 struct trace_event_file *file = *data;
2491 struct trace_event_call *call = file->event_call;
2492
2493 if (strcmp(name, "format") == 0) {
2494 *mode = TRACE_MODE_READ;
2495 *fops = &ftrace_event_format_fops;
2496 return 1;
2497 }
2498
2499 /*
2500 * Only event directories that can be enabled should have
2501 * triggers or filters, with the exception of the "print"
2502 * event that can have a "trigger" file.
2503 */
2504 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2505 if (call->class->reg && strcmp(name, "enable") == 0) {
2506 *mode = TRACE_MODE_WRITE;
2507 *fops = &ftrace_enable_fops;
2508 return 1;
2509 }
2510
2511 if (strcmp(name, "filter") == 0) {
2512 *mode = TRACE_MODE_WRITE;
2513 *fops = &ftrace_event_filter_fops;
2514 return 1;
2515 }
2516 }
2517
2518 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2519 strcmp(trace_event_name(call), "print") == 0) {
2520 if (strcmp(name, "trigger") == 0) {
2521 *mode = TRACE_MODE_WRITE;
2522 *fops = &event_trigger_fops;
2523 return 1;
2524 }
2525 }
2526
2527 #ifdef CONFIG_PERF_EVENTS
2528 if (call->event.type && call->class->reg &&
2529 strcmp(name, "id") == 0) {
2530 *mode = TRACE_MODE_READ;
2531 *data = (void *)(long)call->event.type;
2532 *fops = &ftrace_event_id_fops;
2533 return 1;
2534 }
2535 #endif
2536
2537 #ifdef CONFIG_HIST_TRIGGERS
2538 if (strcmp(name, "hist") == 0) {
2539 *mode = TRACE_MODE_READ;
2540 *fops = &event_hist_fops;
2541 return 1;
2542 }
2543 #endif
2544 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2545 if (strcmp(name, "hist_debug") == 0) {
2546 *mode = TRACE_MODE_READ;
2547 *fops = &event_hist_debug_fops;
2548 return 1;
2549 }
2550 #endif
2551 #ifdef CONFIG_TRACE_EVENT_INJECT
2552 if (call->event.type && call->class->reg &&
2553 strcmp(name, "inject") == 0) {
2554 *mode = 0200;
2555 *fops = &event_inject_fops;
2556 return 1;
2557 }
2558 #endif
2559 return 0;
2560 }
2561
2562 /* The file is incremented on creation and freeing the enable file decrements it */
2563 static void event_release(const char *name, void *data)
2564 {
2565 struct trace_event_file *file = data;
2566
2567 event_file_put(file);
2568 }
2569
2570 static int
2571 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2572 {
2573 struct trace_event_call *call = file->event_call;
2574 struct trace_array *tr = file->tr;
2575 struct eventfs_inode *e_events;
2576 struct eventfs_inode *ei;
2577 const char *name;
2578 int nr_entries;
2579 int ret;
2580 static struct eventfs_entry event_entries[] = {
2581 {
2582 .name = "enable",
2583 .callback = event_callback,
2584 .release = event_release,
2585 },
2586 {
2587 .name = "filter",
2588 .callback = event_callback,
2589 },
2590 {
2591 .name = "trigger",
2592 .callback = event_callback,
2593 },
2594 {
2595 .name = "format",
2596 .callback = event_callback,
2597 },
2598 #ifdef CONFIG_PERF_EVENTS
2599 {
2600 .name = "id",
2601 .callback = event_callback,
2602 },
2603 #endif
2604 #ifdef CONFIG_HIST_TRIGGERS
2605 {
2606 .name = "hist",
2607 .callback = event_callback,
2608 },
2609 #endif
2610 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2611 {
2612 .name = "hist_debug",
2613 .callback = event_callback,
2614 },
2615 #endif
2616 #ifdef CONFIG_TRACE_EVENT_INJECT
2617 {
2618 .name = "inject",
2619 .callback = event_callback,
2620 },
2621 #endif
2622 };
2623
2624 /*
2625 * If the trace point header did not define TRACE_SYSTEM
2626 * then the system would be called "TRACE_SYSTEM". This should
2627 * never happen.
2628 */
2629 if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2630 return -ENODEV;
2631
2632 e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2633 if (!e_events)
2634 return -ENOMEM;
2635
2636 nr_entries = ARRAY_SIZE(event_entries);
2637
2638 name = trace_event_name(call);
2639 ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2640 if (IS_ERR(ei)) {
2641 pr_warn("Could not create tracefs '%s' directory\n", name);
2642 return -1;
2643 }
2644
2645 file->ei = ei;
2646
2647 ret = event_define_fields(call);
2648 if (ret < 0) {
2649 pr_warn("Could not initialize trace point events/%s\n", name);
2650 return ret;
2651 }
2652
2653 /* Gets decremented on freeing of the "enable" file */
2654 event_file_get(file);
2655
2656 return 0;
2657 }
2658
2659 static void remove_event_from_tracers(struct trace_event_call *call)
2660 {
2661 struct trace_event_file *file;
2662 struct trace_array *tr;
2663
2664 do_for_each_event_file_safe(tr, file) {
2665 if (file->event_call != call)
2666 continue;
2667
2668 remove_event_file_dir(file);
2669 /*
2670 * The do_for_each_event_file_safe() 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
2679 static void event_remove(struct trace_event_call *call)
2680 {
2681 struct trace_array *tr;
2682 struct trace_event_file *file;
2683
2684 do_for_each_event_file(tr, file) {
2685 if (file->event_call != call)
2686 continue;
2687
2688 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2689 tr->clear_trace = true;
2690
2691 ftrace_event_enable_disable(file, 0);
2692 /*
2693 * The do_for_each_event_file() is
2694 * a double loop. After finding the call for this
2695 * trace_array, we use break to jump to the next
2696 * trace_array.
2697 */
2698 break;
2699 } while_for_each_event_file();
2700
2701 if (call->event.funcs)
2702 __unregister_trace_event(&call->event);
2703 remove_event_from_tracers(call);
2704 list_del(&call->list);
2705 }
2706
2707 static int event_init(struct trace_event_call *call)
2708 {
2709 int ret = 0;
2710 const char *name;
2711
2712 name = trace_event_name(call);
2713 if (WARN_ON(!name))
2714 return -EINVAL;
2715
2716 if (call->class->raw_init) {
2717 ret = call->class->raw_init(call);
2718 if (ret < 0 && ret != -ENOSYS)
2719 pr_warn("Could not initialize trace events/%s\n", name);
2720 }
2721
2722 return ret;
2723 }
2724
2725 static int
2726 __register_event(struct trace_event_call *call, struct module *mod)
2727 {
2728 int ret;
2729
2730 ret = event_init(call);
2731 if (ret < 0)
2732 return ret;
2733
2734 list_add(&call->list, &ftrace_events);
2735 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2736 atomic_set(&call->refcnt, 0);
2737 else
2738 call->module = mod;
2739
2740 return 0;
2741 }
2742
2743 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2744 {
2745 int rlen;
2746 int elen;
2747
2748 /* Find the length of the eval value as a string */
2749 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2750 /* Make sure there's enough room to replace the string with the value */
2751 if (len < elen)
2752 return NULL;
2753
2754 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2755
2756 /* Get the rest of the string of ptr */
2757 rlen = strlen(ptr + len);
2758 memmove(ptr + elen, ptr + len, rlen);
2759 /* Make sure we end the new string */
2760 ptr[elen + rlen] = 0;
2761
2762 return ptr + elen;
2763 }
2764
2765 static void update_event_printk(struct trace_event_call *call,
2766 struct trace_eval_map *map)
2767 {
2768 char *ptr;
2769 int quote = 0;
2770 int len = strlen(map->eval_string);
2771
2772 for (ptr = call->print_fmt; *ptr; ptr++) {
2773 if (*ptr == '\\') {
2774 ptr++;
2775 /* paranoid */
2776 if (!*ptr)
2777 break;
2778 continue;
2779 }
2780 if (*ptr == '"') {
2781 quote ^= 1;
2782 continue;
2783 }
2784 if (quote)
2785 continue;
2786 if (isdigit(*ptr)) {
2787 /* skip numbers */
2788 do {
2789 ptr++;
2790 /* Check for alpha chars like ULL */
2791 } while (isalnum(*ptr));
2792 if (!*ptr)
2793 break;
2794 /*
2795 * A number must have some kind of delimiter after
2796 * it, and we can ignore that too.
2797 */
2798 continue;
2799 }
2800 if (isalpha(*ptr) || *ptr == '_') {
2801 if (strncmp(map->eval_string, ptr, len) == 0 &&
2802 !isalnum(ptr[len]) && ptr[len] != '_') {
2803 ptr = eval_replace(ptr, map, len);
2804 /* enum/sizeof string smaller than value */
2805 if (WARN_ON_ONCE(!ptr))
2806 return;
2807 /*
2808 * No need to decrement here, as eval_replace()
2809 * returns the pointer to the character passed
2810 * the eval, and two evals can not be placed
2811 * back to back without something in between.
2812 * We can skip that something in between.
2813 */
2814 continue;
2815 }
2816 skip_more:
2817 do {
2818 ptr++;
2819 } while (isalnum(*ptr) || *ptr == '_');
2820 if (!*ptr)
2821 break;
2822 /*
2823 * If what comes after this variable is a '.' or
2824 * '->' then we can continue to ignore that string.
2825 */
2826 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2827 ptr += *ptr == '.' ? 1 : 2;
2828 if (!*ptr)
2829 break;
2830 goto skip_more;
2831 }
2832 /*
2833 * Once again, we can skip the delimiter that came
2834 * after the string.
2835 */
2836 continue;
2837 }
2838 }
2839 }
2840
2841 static void add_str_to_module(struct module *module, char *str)
2842 {
2843 struct module_string *modstr;
2844
2845 modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2846
2847 /*
2848 * If we failed to allocate memory here, then we'll just
2849 * let the str memory leak when the module is removed.
2850 * If this fails to allocate, there's worse problems than
2851 * a leaked string on module removal.
2852 */
2853 if (WARN_ON_ONCE(!modstr))
2854 return;
2855
2856 modstr->module = module;
2857 modstr->str = str;
2858
2859 list_add(&modstr->next, &module_strings);
2860 }
2861
2862 static void update_event_fields(struct trace_event_call *call,
2863 struct trace_eval_map *map)
2864 {
2865 struct ftrace_event_field *field;
2866 struct list_head *head;
2867 char *ptr;
2868 char *str;
2869 int len = strlen(map->eval_string);
2870
2871 /* Dynamic events should never have field maps */
2872 if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2873 return;
2874
2875 head = trace_get_fields(call);
2876 list_for_each_entry(field, head, link) {
2877 ptr = strchr(field->type, '[');
2878 if (!ptr)
2879 continue;
2880 ptr++;
2881
2882 if (!isalpha(*ptr) && *ptr != '_')
2883 continue;
2884
2885 if (strncmp(map->eval_string, ptr, len) != 0)
2886 continue;
2887
2888 str = kstrdup(field->type, GFP_KERNEL);
2889 if (WARN_ON_ONCE(!str))
2890 return;
2891 ptr = str + (ptr - field->type);
2892 ptr = eval_replace(ptr, map, len);
2893 /* enum/sizeof string smaller than value */
2894 if (WARN_ON_ONCE(!ptr)) {
2895 kfree(str);
2896 continue;
2897 }
2898
2899 /*
2900 * If the event is part of a module, then we need to free the string
2901 * when the module is removed. Otherwise, it will stay allocated
2902 * until a reboot.
2903 */
2904 if (call->module)
2905 add_str_to_module(call->module, str);
2906
2907 field->type = str;
2908 }
2909 }
2910
2911 void trace_event_eval_update(struct trace_eval_map **map, int len)
2912 {
2913 struct trace_event_call *call, *p;
2914 const char *last_system = NULL;
2915 bool first = false;
2916 int last_i;
2917 int i;
2918
2919 down_write(&trace_event_sem);
2920 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2921 /* events are usually grouped together with systems */
2922 if (!last_system || call->class->system != last_system) {
2923 first = true;
2924 last_i = 0;
2925 last_system = call->class->system;
2926 }
2927
2928 /*
2929 * Since calls are grouped by systems, the likelihood that the
2930 * next call in the iteration belongs to the same system as the
2931 * previous call is high. As an optimization, we skip searching
2932 * for a map[] that matches the call's system if the last call
2933 * was from the same system. That's what last_i is for. If the
2934 * call has the same system as the previous call, then last_i
2935 * will be the index of the first map[] that has a matching
2936 * system.
2937 */
2938 for (i = last_i; i < len; i++) {
2939 if (call->class->system == map[i]->system) {
2940 /* Save the first system if need be */
2941 if (first) {
2942 last_i = i;
2943 first = false;
2944 }
2945 update_event_printk(call, map[i]);
2946 update_event_fields(call, map[i]);
2947 }
2948 }
2949 cond_resched();
2950 }
2951 up_write(&trace_event_sem);
2952 }
2953
2954 static bool event_in_systems(struct trace_event_call *call,
2955 const char *systems)
2956 {
2957 const char *system;
2958 const char *p;
2959
2960 if (!systems)
2961 return true;
2962
2963 system = call->class->system;
2964 p = strstr(systems, system);
2965 if (!p)
2966 return false;
2967
2968 if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2969 return false;
2970
2971 p += strlen(system);
2972 return !*p || isspace(*p) || *p == ',';
2973 }
2974
2975 static struct trace_event_file *
2976 trace_create_new_event(struct trace_event_call *call,
2977 struct trace_array *tr)
2978 {
2979 struct trace_pid_list *no_pid_list;
2980 struct trace_pid_list *pid_list;
2981 struct trace_event_file *file;
2982 unsigned int first;
2983
2984 if (!event_in_systems(call, tr->system_names))
2985 return NULL;
2986
2987 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2988 if (!file)
2989 return ERR_PTR(-ENOMEM);
2990
2991 pid_list = rcu_dereference_protected(tr->filtered_pids,
2992 lockdep_is_held(&event_mutex));
2993 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2994 lockdep_is_held(&event_mutex));
2995
2996 if (!trace_pid_list_first(pid_list, &first) ||
2997 !trace_pid_list_first(no_pid_list, &first))
2998 file->flags |= EVENT_FILE_FL_PID_FILTER;
2999
3000 file->event_call = call;
3001 file->tr = tr;
3002 atomic_set(&file->sm_ref, 0);
3003 atomic_set(&file->tm_ref, 0);
3004 INIT_LIST_HEAD(&file->triggers);
3005 list_add(&file->list, &tr->events);
3006 refcount_set(&file->ref, 1);
3007
3008 return file;
3009 }
3010
3011 #define MAX_BOOT_TRIGGERS 32
3012
3013 static struct boot_triggers {
3014 const char *event;
3015 char *trigger;
3016 } bootup_triggers[MAX_BOOT_TRIGGERS];
3017
3018 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3019 static int nr_boot_triggers;
3020
3021 static __init int setup_trace_triggers(char *str)
3022 {
3023 char *trigger;
3024 char *buf;
3025 int i;
3026
3027 strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3028 trace_set_ring_buffer_expanded(NULL);
3029 disable_tracing_selftest("running event triggers");
3030
3031 buf = bootup_trigger_buf;
3032 for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3033 trigger = strsep(&buf, ",");
3034 if (!trigger)
3035 break;
3036 bootup_triggers[i].event = strsep(&trigger, ".");
3037 bootup_triggers[i].trigger = trigger;
3038 if (!bootup_triggers[i].trigger)
3039 break;
3040 }
3041
3042 nr_boot_triggers = i;
3043 return 1;
3044 }
3045 __setup("trace_trigger=", setup_trace_triggers);
3046
3047 /* Add an event to a trace directory */
3048 static int
3049 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3050 {
3051 struct trace_event_file *file;
3052
3053 file = trace_create_new_event(call, tr);
3054 /*
3055 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3056 * allocation, or NULL if the event is not part of the tr->system_names.
3057 * When the event is not part of the tr->system_names, return zero, not
3058 * an error.
3059 */
3060 if (!file)
3061 return 0;
3062
3063 if (IS_ERR(file))
3064 return PTR_ERR(file);
3065
3066 if (eventdir_initialized)
3067 return event_create_dir(tr->event_dir, file);
3068 else
3069 return event_define_fields(call);
3070 }
3071
3072 static void trace_early_triggers(struct trace_event_file *file, const char *name)
3073 {
3074 int ret;
3075 int i;
3076
3077 for (i = 0; i < nr_boot_triggers; i++) {
3078 if (strcmp(name, bootup_triggers[i].event))
3079 continue;
3080 mutex_lock(&event_mutex);
3081 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3082 mutex_unlock(&event_mutex);
3083 if (ret)
3084 pr_err("Failed to register trigger '%s' on event %s\n",
3085 bootup_triggers[i].trigger,
3086 bootup_triggers[i].event);
3087 }
3088 }
3089
3090 /*
3091 * Just create a descriptor for early init. A descriptor is required
3092 * for enabling events at boot. We want to enable events before
3093 * the filesystem is initialized.
3094 */
3095 static int
3096 __trace_early_add_new_event(struct trace_event_call *call,
3097 struct trace_array *tr)
3098 {
3099 struct trace_event_file *file;
3100 int ret;
3101
3102 file = trace_create_new_event(call, tr);
3103 /*
3104 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3105 * allocation, or NULL if the event is not part of the tr->system_names.
3106 * When the event is not part of the tr->system_names, return zero, not
3107 * an error.
3108 */
3109 if (!file)
3110 return 0;
3111
3112 if (IS_ERR(file))
3113 return PTR_ERR(file);
3114
3115 ret = event_define_fields(call);
3116 if (ret)
3117 return ret;
3118
3119 trace_early_triggers(file, trace_event_name(call));
3120
3121 return 0;
3122 }
3123
3124 struct ftrace_module_file_ops;
3125 static void __add_event_to_tracers(struct trace_event_call *call);
3126
3127 /* Add an additional event_call dynamically */
3128 int trace_add_event_call(struct trace_event_call *call)
3129 {
3130 int ret;
3131 lockdep_assert_held(&event_mutex);
3132
3133 mutex_lock(&trace_types_lock);
3134
3135 ret = __register_event(call, NULL);
3136 if (ret >= 0)
3137 __add_event_to_tracers(call);
3138
3139 mutex_unlock(&trace_types_lock);
3140 return ret;
3141 }
3142 EXPORT_SYMBOL_GPL(trace_add_event_call);
3143
3144 /*
3145 * Must be called under locking of trace_types_lock, event_mutex and
3146 * trace_event_sem.
3147 */
3148 static void __trace_remove_event_call(struct trace_event_call *call)
3149 {
3150 event_remove(call);
3151 trace_destroy_fields(call);
3152 free_event_filter(call->filter);
3153 call->filter = NULL;
3154 }
3155
3156 static int probe_remove_event_call(struct trace_event_call *call)
3157 {
3158 struct trace_array *tr;
3159 struct trace_event_file *file;
3160
3161 #ifdef CONFIG_PERF_EVENTS
3162 if (call->perf_refcount)
3163 return -EBUSY;
3164 #endif
3165 do_for_each_event_file(tr, file) {
3166 if (file->event_call != call)
3167 continue;
3168 /*
3169 * We can't rely on ftrace_event_enable_disable(enable => 0)
3170 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3171 * TRACE_REG_UNREGISTER.
3172 */
3173 if (file->flags & EVENT_FILE_FL_ENABLED)
3174 goto busy;
3175
3176 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3177 tr->clear_trace = true;
3178 /*
3179 * The do_for_each_event_file_safe() is
3180 * a double loop. After finding the call for this
3181 * trace_array, we use break to jump to the next
3182 * trace_array.
3183 */
3184 break;
3185 } while_for_each_event_file();
3186
3187 __trace_remove_event_call(call);
3188
3189 return 0;
3190 busy:
3191 /* No need to clear the trace now */
3192 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3193 tr->clear_trace = false;
3194 }
3195 return -EBUSY;
3196 }
3197
3198 /* Remove an event_call */
3199 int trace_remove_event_call(struct trace_event_call *call)
3200 {
3201 int ret;
3202
3203 lockdep_assert_held(&event_mutex);
3204
3205 mutex_lock(&trace_types_lock);
3206 down_write(&trace_event_sem);
3207 ret = probe_remove_event_call(call);
3208 up_write(&trace_event_sem);
3209 mutex_unlock(&trace_types_lock);
3210
3211 return ret;
3212 }
3213 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3214
3215 #define for_each_event(event, start, end) \
3216 for (event = start; \
3217 (unsigned long)event < (unsigned long)end; \
3218 event++)
3219
3220 #ifdef CONFIG_MODULES
3221
3222 static void trace_module_add_events(struct module *mod)
3223 {
3224 struct trace_event_call **call, **start, **end;
3225
3226 if (!mod->num_trace_events)
3227 return;
3228
3229 /* Don't add infrastructure for mods without tracepoints */
3230 if (trace_module_has_bad_taint(mod)) {
3231 pr_err("%s: module has bad taint, not creating trace events\n",
3232 mod->name);
3233 return;
3234 }
3235
3236 start = mod->trace_events;
3237 end = mod->trace_events + mod->num_trace_events;
3238
3239 for_each_event(call, start, end) {
3240 __register_event(*call, mod);
3241 __add_event_to_tracers(*call);
3242 }
3243 }
3244
3245 static void trace_module_remove_events(struct module *mod)
3246 {
3247 struct trace_event_call *call, *p;
3248 struct module_string *modstr, *m;
3249
3250 down_write(&trace_event_sem);
3251 list_for_each_entry_safe(call, p, &ftrace_events, list) {
3252 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3253 continue;
3254 if (call->module == mod)
3255 __trace_remove_event_call(call);
3256 }
3257 /* Check for any strings allocade for this module */
3258 list_for_each_entry_safe(modstr, m, &module_strings, next) {
3259 if (modstr->module != mod)
3260 continue;
3261 list_del(&modstr->next);
3262 kfree(modstr->str);
3263 kfree(modstr);
3264 }
3265 up_write(&trace_event_sem);
3266
3267 /*
3268 * It is safest to reset the ring buffer if the module being unloaded
3269 * registered any events that were used. The only worry is if
3270 * a new module gets loaded, and takes on the same id as the events
3271 * of this module. When printing out the buffer, traced events left
3272 * over from this module may be passed to the new module events and
3273 * unexpected results may occur.
3274 */
3275 tracing_reset_all_online_cpus_unlocked();
3276 }
3277
3278 static int trace_module_notify(struct notifier_block *self,
3279 unsigned long val, void *data)
3280 {
3281 struct module *mod = data;
3282
3283 mutex_lock(&event_mutex);
3284 mutex_lock(&trace_types_lock);
3285 switch (val) {
3286 case MODULE_STATE_COMING:
3287 trace_module_add_events(mod);
3288 break;
3289 case MODULE_STATE_GOING:
3290 trace_module_remove_events(mod);
3291 break;
3292 }
3293 mutex_unlock(&trace_types_lock);
3294 mutex_unlock(&event_mutex);
3295
3296 return NOTIFY_OK;
3297 }
3298
3299 static struct notifier_block trace_module_nb = {
3300 .notifier_call = trace_module_notify,
3301 .priority = 1, /* higher than trace.c module notify */
3302 };
3303 #endif /* CONFIG_MODULES */
3304
3305 /* Create a new event directory structure for a trace directory. */
3306 static void
3307 __trace_add_event_dirs(struct trace_array *tr)
3308 {
3309 struct trace_event_call *call;
3310 int ret;
3311
3312 list_for_each_entry(call, &ftrace_events, list) {
3313 ret = __trace_add_new_event(call, tr);
3314 if (ret < 0)
3315 pr_warn("Could not create directory for event %s\n",
3316 trace_event_name(call));
3317 }
3318 }
3319
3320 /* Returns any file that matches the system and event */
3321 struct trace_event_file *
3322 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3323 {
3324 struct trace_event_file *file;
3325 struct trace_event_call *call;
3326 const char *name;
3327
3328 list_for_each_entry(file, &tr->events, list) {
3329
3330 call = file->event_call;
3331 name = trace_event_name(call);
3332
3333 if (!name || !call->class)
3334 continue;
3335
3336 if (strcmp(event, name) == 0 &&
3337 strcmp(system, call->class->system) == 0)
3338 return file;
3339 }
3340 return NULL;
3341 }
3342
3343 /* Returns valid trace event files that match system and event */
3344 struct trace_event_file *
3345 find_event_file(struct trace_array *tr, const char *system, const char *event)
3346 {
3347 struct trace_event_file *file;
3348
3349 file = __find_event_file(tr, system, event);
3350 if (!file || !file->event_call->class->reg ||
3351 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3352 return NULL;
3353
3354 return file;
3355 }
3356
3357 /**
3358 * trace_get_event_file - Find and return a trace event file
3359 * @instance: The name of the trace instance containing the event
3360 * @system: The name of the system containing the event
3361 * @event: The name of the event
3362 *
3363 * Return a trace event file given the trace instance name, trace
3364 * system, and trace event name. If the instance name is NULL, it
3365 * refers to the top-level trace array.
3366 *
3367 * This function will look it up and return it if found, after calling
3368 * trace_array_get() to prevent the instance from going away, and
3369 * increment the event's module refcount to prevent it from being
3370 * removed.
3371 *
3372 * To release the file, call trace_put_event_file(), which will call
3373 * trace_array_put() and decrement the event's module refcount.
3374 *
3375 * Return: The trace event on success, ERR_PTR otherwise.
3376 */
3377 struct trace_event_file *trace_get_event_file(const char *instance,
3378 const char *system,
3379 const char *event)
3380 {
3381 struct trace_array *tr = top_trace_array();
3382 struct trace_event_file *file = NULL;
3383 int ret = -EINVAL;
3384
3385 if (instance) {
3386 tr = trace_array_find_get(instance);
3387 if (!tr)
3388 return ERR_PTR(-ENOENT);
3389 } else {
3390 ret = trace_array_get(tr);
3391 if (ret)
3392 return ERR_PTR(ret);
3393 }
3394
3395 mutex_lock(&event_mutex);
3396
3397 file = find_event_file(tr, system, event);
3398 if (!file) {
3399 trace_array_put(tr);
3400 ret = -EINVAL;
3401 goto out;
3402 }
3403
3404 /* Don't let event modules unload while in use */
3405 ret = trace_event_try_get_ref(file->event_call);
3406 if (!ret) {
3407 trace_array_put(tr);
3408 ret = -EBUSY;
3409 goto out;
3410 }
3411
3412 ret = 0;
3413 out:
3414 mutex_unlock(&event_mutex);
3415
3416 if (ret)
3417 file = ERR_PTR(ret);
3418
3419 return file;
3420 }
3421 EXPORT_SYMBOL_GPL(trace_get_event_file);
3422
3423 /**
3424 * trace_put_event_file - Release a file from trace_get_event_file()
3425 * @file: The trace event file
3426 *
3427 * If a file was retrieved using trace_get_event_file(), this should
3428 * be called when it's no longer needed. It will cancel the previous
3429 * trace_array_get() called by that function, and decrement the
3430 * event's module refcount.
3431 */
3432 void trace_put_event_file(struct trace_event_file *file)
3433 {
3434 mutex_lock(&event_mutex);
3435 trace_event_put_ref(file->event_call);
3436 mutex_unlock(&event_mutex);
3437
3438 trace_array_put(file->tr);
3439 }
3440 EXPORT_SYMBOL_GPL(trace_put_event_file);
3441
3442 #ifdef CONFIG_DYNAMIC_FTRACE
3443
3444 /* Avoid typos */
3445 #define ENABLE_EVENT_STR "enable_event"
3446 #define DISABLE_EVENT_STR "disable_event"
3447
3448 struct event_probe_data {
3449 struct trace_event_file *file;
3450 unsigned long count;
3451 int ref;
3452 bool enable;
3453 };
3454
3455 static void update_event_probe(struct event_probe_data *data)
3456 {
3457 if (data->enable)
3458 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3459 else
3460 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3461 }
3462
3463 static void
3464 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3465 struct trace_array *tr, struct ftrace_probe_ops *ops,
3466 void *data)
3467 {
3468 struct ftrace_func_mapper *mapper = data;
3469 struct event_probe_data *edata;
3470 void **pdata;
3471
3472 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3473 if (!pdata || !*pdata)
3474 return;
3475
3476 edata = *pdata;
3477 update_event_probe(edata);
3478 }
3479
3480 static void
3481 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3482 struct trace_array *tr, struct ftrace_probe_ops *ops,
3483 void *data)
3484 {
3485 struct ftrace_func_mapper *mapper = data;
3486 struct event_probe_data *edata;
3487 void **pdata;
3488
3489 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3490 if (!pdata || !*pdata)
3491 return;
3492
3493 edata = *pdata;
3494
3495 if (!edata->count)
3496 return;
3497
3498 /* Skip if the event is in a state we want to switch to */
3499 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3500 return;
3501
3502 if (edata->count != -1)
3503 (edata->count)--;
3504
3505 update_event_probe(edata);
3506 }
3507
3508 static int
3509 event_enable_print(struct seq_file *m, unsigned long ip,
3510 struct ftrace_probe_ops *ops, void *data)
3511 {
3512 struct ftrace_func_mapper *mapper = data;
3513 struct event_probe_data *edata;
3514 void **pdata;
3515
3516 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3517
3518 if (WARN_ON_ONCE(!pdata || !*pdata))
3519 return 0;
3520
3521 edata = *pdata;
3522
3523 seq_printf(m, "%ps:", (void *)ip);
3524
3525 seq_printf(m, "%s:%s:%s",
3526 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3527 edata->file->event_call->class->system,
3528 trace_event_name(edata->file->event_call));
3529
3530 if (edata->count == -1)
3531 seq_puts(m, ":unlimited\n");
3532 else
3533 seq_printf(m, ":count=%ld\n", edata->count);
3534
3535 return 0;
3536 }
3537
3538 static int
3539 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3540 unsigned long ip, void *init_data, void **data)
3541 {
3542 struct ftrace_func_mapper *mapper = *data;
3543 struct event_probe_data *edata = init_data;
3544 int ret;
3545
3546 if (!mapper) {
3547 mapper = allocate_ftrace_func_mapper();
3548 if (!mapper)
3549 return -ENODEV;
3550 *data = mapper;
3551 }
3552
3553 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3554 if (ret < 0)
3555 return ret;
3556
3557 edata->ref++;
3558
3559 return 0;
3560 }
3561
3562 static int free_probe_data(void *data)
3563 {
3564 struct event_probe_data *edata = data;
3565
3566 edata->ref--;
3567 if (!edata->ref) {
3568 /* Remove the SOFT_MODE flag */
3569 __ftrace_event_enable_disable(edata->file, 0, 1);
3570 trace_event_put_ref(edata->file->event_call);
3571 kfree(edata);
3572 }
3573 return 0;
3574 }
3575
3576 static void
3577 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3578 unsigned long ip, void *data)
3579 {
3580 struct ftrace_func_mapper *mapper = data;
3581 struct event_probe_data *edata;
3582
3583 if (!ip) {
3584 if (!mapper)
3585 return;
3586 free_ftrace_func_mapper(mapper, free_probe_data);
3587 return;
3588 }
3589
3590 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3591
3592 if (WARN_ON_ONCE(!edata))
3593 return;
3594
3595 if (WARN_ON_ONCE(edata->ref <= 0))
3596 return;
3597
3598 free_probe_data(edata);
3599 }
3600
3601 static struct ftrace_probe_ops event_enable_probe_ops = {
3602 .func = event_enable_probe,
3603 .print = event_enable_print,
3604 .init = event_enable_init,
3605 .free = event_enable_free,
3606 };
3607
3608 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3609 .func = event_enable_count_probe,
3610 .print = event_enable_print,
3611 .init = event_enable_init,
3612 .free = event_enable_free,
3613 };
3614
3615 static struct ftrace_probe_ops event_disable_probe_ops = {
3616 .func = event_enable_probe,
3617 .print = event_enable_print,
3618 .init = event_enable_init,
3619 .free = event_enable_free,
3620 };
3621
3622 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3623 .func = event_enable_count_probe,
3624 .print = event_enable_print,
3625 .init = event_enable_init,
3626 .free = event_enable_free,
3627 };
3628
3629 static int
3630 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3631 char *glob, char *cmd, char *param, int enabled)
3632 {
3633 struct trace_event_file *file;
3634 struct ftrace_probe_ops *ops;
3635 struct event_probe_data *data;
3636 const char *system;
3637 const char *event;
3638 char *number;
3639 bool enable;
3640 int ret;
3641
3642 if (!tr)
3643 return -ENODEV;
3644
3645 /* hash funcs only work with set_ftrace_filter */
3646 if (!enabled || !param)
3647 return -EINVAL;
3648
3649 system = strsep(¶m, ":");
3650 if (!param)
3651 return -EINVAL;
3652
3653 event = strsep(¶m, ":");
3654
3655 mutex_lock(&event_mutex);
3656
3657 ret = -EINVAL;
3658 file = find_event_file(tr, system, event);
3659 if (!file)
3660 goto out;
3661
3662 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3663
3664 if (enable)
3665 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3666 else
3667 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3668
3669 if (glob[0] == '!') {
3670 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3671 goto out;
3672 }
3673
3674 ret = -ENOMEM;
3675
3676 data = kzalloc(sizeof(*data), GFP_KERNEL);
3677 if (!data)
3678 goto out;
3679
3680 data->enable = enable;
3681 data->count = -1;
3682 data->file = file;
3683
3684 if (!param)
3685 goto out_reg;
3686
3687 number = strsep(¶m, ":");
3688
3689 ret = -EINVAL;
3690 if (!strlen(number))
3691 goto out_free;
3692
3693 /*
3694 * We use the callback data field (which is a pointer)
3695 * as our counter.
3696 */
3697 ret = kstrtoul(number, 0, &data->count);
3698 if (ret)
3699 goto out_free;
3700
3701 out_reg:
3702 /* Don't let event modules unload while probe registered */
3703 ret = trace_event_try_get_ref(file->event_call);
3704 if (!ret) {
3705 ret = -EBUSY;
3706 goto out_free;
3707 }
3708
3709 ret = __ftrace_event_enable_disable(file, 1, 1);
3710 if (ret < 0)
3711 goto out_put;
3712
3713 ret = register_ftrace_function_probe(glob, tr, ops, data);
3714 /*
3715 * The above returns on success the # of functions enabled,
3716 * but if it didn't find any functions it returns zero.
3717 * Consider no functions a failure too.
3718 */
3719 if (!ret) {
3720 ret = -ENOENT;
3721 goto out_disable;
3722 } else if (ret < 0)
3723 goto out_disable;
3724 /* Just return zero, not the number of enabled functions */
3725 ret = 0;
3726 out:
3727 mutex_unlock(&event_mutex);
3728 return ret;
3729
3730 out_disable:
3731 __ftrace_event_enable_disable(file, 0, 1);
3732 out_put:
3733 trace_event_put_ref(file->event_call);
3734 out_free:
3735 kfree(data);
3736 goto out;
3737 }
3738
3739 static struct ftrace_func_command event_enable_cmd = {
3740 .name = ENABLE_EVENT_STR,
3741 .func = event_enable_func,
3742 };
3743
3744 static struct ftrace_func_command event_disable_cmd = {
3745 .name = DISABLE_EVENT_STR,
3746 .func = event_enable_func,
3747 };
3748
3749 static __init int register_event_cmds(void)
3750 {
3751 int ret;
3752
3753 ret = register_ftrace_command(&event_enable_cmd);
3754 if (WARN_ON(ret < 0))
3755 return ret;
3756 ret = register_ftrace_command(&event_disable_cmd);
3757 if (WARN_ON(ret < 0))
3758 unregister_ftrace_command(&event_enable_cmd);
3759 return ret;
3760 }
3761 #else
3762 static inline int register_event_cmds(void) { return 0; }
3763 #endif /* CONFIG_DYNAMIC_FTRACE */
3764
3765 /*
3766 * The top level array and trace arrays created by boot-time tracing
3767 * have already had its trace_event_file descriptors created in order
3768 * to allow for early events to be recorded.
3769 * This function is called after the tracefs has been initialized,
3770 * and we now have to create the files associated to the events.
3771 */
3772 static void __trace_early_add_event_dirs(struct trace_array *tr)
3773 {
3774 struct trace_event_file *file;
3775 int ret;
3776
3777
3778 list_for_each_entry(file, &tr->events, list) {
3779 ret = event_create_dir(tr->event_dir, file);
3780 if (ret < 0)
3781 pr_warn("Could not create directory for event %s\n",
3782 trace_event_name(file->event_call));
3783 }
3784 }
3785
3786 /*
3787 * For early boot up, the top trace array and the trace arrays created
3788 * by boot-time tracing require to have a list of events that can be
3789 * enabled. This must be done before the filesystem is set up in order
3790 * to allow events to be traced early.
3791 */
3792 void __trace_early_add_events(struct trace_array *tr)
3793 {
3794 struct trace_event_call *call;
3795 int ret;
3796
3797 list_for_each_entry(call, &ftrace_events, list) {
3798 /* Early boot up should not have any modules loaded */
3799 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3800 WARN_ON_ONCE(call->module))
3801 continue;
3802
3803 ret = __trace_early_add_new_event(call, tr);
3804 if (ret < 0)
3805 pr_warn("Could not create early event %s\n",
3806 trace_event_name(call));
3807 }
3808 }
3809
3810 /* Remove the event directory structure for a trace directory. */
3811 static void
3812 __trace_remove_event_dirs(struct trace_array *tr)
3813 {
3814 struct trace_event_file *file, *next;
3815
3816 list_for_each_entry_safe(file, next, &tr->events, list)
3817 remove_event_file_dir(file);
3818 }
3819
3820 static void __add_event_to_tracers(struct trace_event_call *call)
3821 {
3822 struct trace_array *tr;
3823
3824 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3825 __trace_add_new_event(call, tr);
3826 }
3827
3828 extern struct trace_event_call *__start_ftrace_events[];
3829 extern struct trace_event_call *__stop_ftrace_events[];
3830
3831 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3832
3833 static __init int setup_trace_event(char *str)
3834 {
3835 strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3836 trace_set_ring_buffer_expanded(NULL);
3837 disable_tracing_selftest("running event tracing");
3838
3839 return 1;
3840 }
3841 __setup("trace_event=", setup_trace_event);
3842
3843 static int events_callback(const char *name, umode_t *mode, void **data,
3844 const struct file_operations **fops)
3845 {
3846 if (strcmp(name, "enable") == 0) {
3847 *mode = TRACE_MODE_WRITE;
3848 *fops = &ftrace_tr_enable_fops;
3849 return 1;
3850 }
3851
3852 if (strcmp(name, "header_page") == 0) {
3853 *mode = TRACE_MODE_READ;
3854 *fops = &ftrace_show_header_page_fops;
3855
3856 } else if (strcmp(name, "header_event") == 0) {
3857 *mode = TRACE_MODE_READ;
3858 *fops = &ftrace_show_header_event_fops;
3859 } else
3860 return 0;
3861
3862 return 1;
3863 }
3864
3865 /* Expects to have event_mutex held when called */
3866 static int
3867 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3868 {
3869 struct eventfs_inode *e_events;
3870 struct dentry *entry;
3871 int nr_entries;
3872 static struct eventfs_entry events_entries[] = {
3873 {
3874 .name = "enable",
3875 .callback = events_callback,
3876 },
3877 {
3878 .name = "header_page",
3879 .callback = events_callback,
3880 },
3881 {
3882 .name = "header_event",
3883 .callback = events_callback,
3884 },
3885 };
3886
3887 entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3888 tr, &ftrace_set_event_fops);
3889 if (!entry)
3890 return -ENOMEM;
3891
3892 nr_entries = ARRAY_SIZE(events_entries);
3893
3894 e_events = eventfs_create_events_dir("events", parent, events_entries,
3895 nr_entries, tr);
3896 if (IS_ERR(e_events)) {
3897 pr_warn("Could not create tracefs 'events' directory\n");
3898 return -ENOMEM;
3899 }
3900
3901 /* There are not as crucial, just warn if they are not created */
3902
3903 trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3904 tr, &ftrace_set_event_pid_fops);
3905
3906 trace_create_file("set_event_notrace_pid",
3907 TRACE_MODE_WRITE, parent, tr,
3908 &ftrace_set_event_notrace_pid_fops);
3909
3910 tr->event_dir = e_events;
3911
3912 return 0;
3913 }
3914
3915 /**
3916 * event_trace_add_tracer - add a instance of a trace_array to events
3917 * @parent: The parent dentry to place the files/directories for events in
3918 * @tr: The trace array associated with these events
3919 *
3920 * When a new instance is created, it needs to set up its events
3921 * directory, as well as other files associated with events. It also
3922 * creates the event hierarchy in the @parent/events directory.
3923 *
3924 * Returns 0 on success.
3925 *
3926 * Must be called with event_mutex held.
3927 */
3928 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3929 {
3930 int ret;
3931
3932 lockdep_assert_held(&event_mutex);
3933
3934 ret = create_event_toplevel_files(parent, tr);
3935 if (ret)
3936 goto out;
3937
3938 down_write(&trace_event_sem);
3939 /* If tr already has the event list, it is initialized in early boot. */
3940 if (unlikely(!list_empty(&tr->events)))
3941 __trace_early_add_event_dirs(tr);
3942 else
3943 __trace_add_event_dirs(tr);
3944 up_write(&trace_event_sem);
3945
3946 out:
3947 return ret;
3948 }
3949
3950 /*
3951 * The top trace array already had its file descriptors created.
3952 * Now the files themselves need to be created.
3953 */
3954 static __init int
3955 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3956 {
3957 int ret;
3958
3959 mutex_lock(&event_mutex);
3960
3961 ret = create_event_toplevel_files(parent, tr);
3962 if (ret)
3963 goto out_unlock;
3964
3965 down_write(&trace_event_sem);
3966 __trace_early_add_event_dirs(tr);
3967 up_write(&trace_event_sem);
3968
3969 out_unlock:
3970 mutex_unlock(&event_mutex);
3971
3972 return ret;
3973 }
3974
3975 /* Must be called with event_mutex held */
3976 int event_trace_del_tracer(struct trace_array *tr)
3977 {
3978 lockdep_assert_held(&event_mutex);
3979
3980 /* Disable any event triggers and associated soft-disabled events */
3981 clear_event_triggers(tr);
3982
3983 /* Clear the pid list */
3984 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3985
3986 /* Disable any running events */
3987 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3988
3989 /* Make sure no more events are being executed */
3990 tracepoint_synchronize_unregister();
3991
3992 down_write(&trace_event_sem);
3993 __trace_remove_event_dirs(tr);
3994 eventfs_remove_events_dir(tr->event_dir);
3995 up_write(&trace_event_sem);
3996
3997 tr->event_dir = NULL;
3998
3999 return 0;
4000 }
4001
4002 static __init int event_trace_memsetup(void)
4003 {
4004 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
4005 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
4006 return 0;
4007 }
4008
4009 __init void
4010 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4011 {
4012 char *token;
4013 int ret;
4014
4015 while (true) {
4016 token = strsep(&buf, ",");
4017
4018 if (!token)
4019 break;
4020
4021 if (*token) {
4022 /* Restarting syscalls requires that we stop them first */
4023 if (disable_first)
4024 ftrace_set_clr_event(tr, token, 0);
4025
4026 ret = ftrace_set_clr_event(tr, token, 1);
4027 if (ret)
4028 pr_warn("Failed to enable trace event: %s\n", token);
4029 }
4030
4031 /* Put back the comma to allow this to be called again */
4032 if (buf)
4033 *(buf - 1) = ',';
4034 }
4035 }
4036
4037 static __init int event_trace_enable(void)
4038 {
4039 struct trace_array *tr = top_trace_array();
4040 struct trace_event_call **iter, *call;
4041 int ret;
4042
4043 if (!tr)
4044 return -ENODEV;
4045
4046 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4047
4048 call = *iter;
4049 ret = event_init(call);
4050 if (!ret)
4051 list_add(&call->list, &ftrace_events);
4052 }
4053
4054 register_trigger_cmds();
4055
4056 /*
4057 * We need the top trace array to have a working set of trace
4058 * points at early init, before the debug files and directories
4059 * are created. Create the file entries now, and attach them
4060 * to the actual file dentries later.
4061 */
4062 __trace_early_add_events(tr);
4063
4064 early_enable_events(tr, bootup_event_buf, false);
4065
4066 trace_printk_start_comm();
4067
4068 register_event_cmds();
4069
4070
4071 return 0;
4072 }
4073
4074 /*
4075 * event_trace_enable() is called from trace_event_init() first to
4076 * initialize events and perhaps start any events that are on the
4077 * command line. Unfortunately, there are some events that will not
4078 * start this early, like the system call tracepoints that need
4079 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4080 * event_trace_enable() is called before pid 1 starts, and this flag
4081 * is never set, making the syscall tracepoint never get reached, but
4082 * the event is enabled regardless (and not doing anything).
4083 */
4084 static __init int event_trace_enable_again(void)
4085 {
4086 struct trace_array *tr;
4087
4088 tr = top_trace_array();
4089 if (!tr)
4090 return -ENODEV;
4091
4092 early_enable_events(tr, bootup_event_buf, true);
4093
4094 return 0;
4095 }
4096
4097 early_initcall(event_trace_enable_again);
4098
4099 /* Init fields which doesn't related to the tracefs */
4100 static __init int event_trace_init_fields(void)
4101 {
4102 if (trace_define_generic_fields())
4103 pr_warn("tracing: Failed to allocated generic fields");
4104
4105 if (trace_define_common_fields())
4106 pr_warn("tracing: Failed to allocate common fields");
4107
4108 return 0;
4109 }
4110
4111 __init int event_trace_init(void)
4112 {
4113 struct trace_array *tr;
4114 int ret;
4115
4116 tr = top_trace_array();
4117 if (!tr)
4118 return -ENODEV;
4119
4120 trace_create_file("available_events", TRACE_MODE_READ,
4121 NULL, tr, &ftrace_avail_fops);
4122
4123 ret = early_event_add_tracer(NULL, tr);
4124 if (ret)
4125 return ret;
4126
4127 #ifdef CONFIG_MODULES
4128 ret = register_module_notifier(&trace_module_nb);
4129 if (ret)
4130 pr_warn("Failed to register trace events module notifier\n");
4131 #endif
4132
4133 eventdir_initialized = true;
4134
4135 return 0;
4136 }
4137
4138 void __init trace_event_init(void)
4139 {
4140 event_trace_memsetup();
4141 init_ftrace_syscalls();
4142 event_trace_enable();
4143 event_trace_init_fields();
4144 }
4145
4146 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4147
4148 static DEFINE_SPINLOCK(test_spinlock);
4149 static DEFINE_SPINLOCK(test_spinlock_irq);
4150 static DEFINE_MUTEX(test_mutex);
4151
4152 static __init void test_work(struct work_struct *dummy)
4153 {
4154 spin_lock(&test_spinlock);
4155 spin_lock_irq(&test_spinlock_irq);
4156 udelay(1);
4157 spin_unlock_irq(&test_spinlock_irq);
4158 spin_unlock(&test_spinlock);
4159
4160 mutex_lock(&test_mutex);
4161 msleep(1);
4162 mutex_unlock(&test_mutex);
4163 }
4164
4165 static __init int event_test_thread(void *unused)
4166 {
4167 void *test_malloc;
4168
4169 test_malloc = kmalloc(1234, GFP_KERNEL);
4170 if (!test_malloc)
4171 pr_info("failed to kmalloc\n");
4172
4173 schedule_on_each_cpu(test_work);
4174
4175 kfree(test_malloc);
4176
4177 set_current_state(TASK_INTERRUPTIBLE);
4178 while (!kthread_should_stop()) {
4179 schedule();
4180 set_current_state(TASK_INTERRUPTIBLE);
4181 }
4182 __set_current_state(TASK_RUNNING);
4183
4184 return 0;
4185 }
4186
4187 /*
4188 * Do various things that may trigger events.
4189 */
4190 static __init void event_test_stuff(void)
4191 {
4192 struct task_struct *test_thread;
4193
4194 test_thread = kthread_run(event_test_thread, NULL, "test-events");
4195 msleep(1);
4196 kthread_stop(test_thread);
4197 }
4198
4199 /*
4200 * For every trace event defined, we will test each trace point separately,
4201 * and then by groups, and finally all trace points.
4202 */
4203 static __init void event_trace_self_tests(void)
4204 {
4205 struct trace_subsystem_dir *dir;
4206 struct trace_event_file *file;
4207 struct trace_event_call *call;
4208 struct event_subsystem *system;
4209 struct trace_array *tr;
4210 int ret;
4211
4212 tr = top_trace_array();
4213 if (!tr)
4214 return;
4215
4216 pr_info("Running tests on trace events:\n");
4217
4218 list_for_each_entry(file, &tr->events, list) {
4219
4220 call = file->event_call;
4221
4222 /* Only test those that have a probe */
4223 if (!call->class || !call->class->probe)
4224 continue;
4225
4226 /*
4227 * Testing syscall events here is pretty useless, but
4228 * we still do it if configured. But this is time consuming.
4229 * What we really need is a user thread to perform the
4230 * syscalls as we test.
4231 */
4232 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4233 if (call->class->system &&
4234 strcmp(call->class->system, "syscalls") == 0)
4235 continue;
4236 #endif
4237
4238 pr_info("Testing event %s: ", trace_event_name(call));
4239
4240 /*
4241 * If an event is already enabled, someone is using
4242 * it and the self test should not be on.
4243 */
4244 if (file->flags & EVENT_FILE_FL_ENABLED) {
4245 pr_warn("Enabled event during self test!\n");
4246 WARN_ON_ONCE(1);
4247 continue;
4248 }
4249
4250 ftrace_event_enable_disable(file, 1);
4251 event_test_stuff();
4252 ftrace_event_enable_disable(file, 0);
4253
4254 pr_cont("OK\n");
4255 }
4256
4257 /* Now test at the sub system level */
4258
4259 pr_info("Running tests on trace event systems:\n");
4260
4261 list_for_each_entry(dir, &tr->systems, list) {
4262
4263 system = dir->subsystem;
4264
4265 /* the ftrace system is special, skip it */
4266 if (strcmp(system->name, "ftrace") == 0)
4267 continue;
4268
4269 pr_info("Testing event system %s: ", system->name);
4270
4271 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4272 if (WARN_ON_ONCE(ret)) {
4273 pr_warn("error enabling system %s\n",
4274 system->name);
4275 continue;
4276 }
4277
4278 event_test_stuff();
4279
4280 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4281 if (WARN_ON_ONCE(ret)) {
4282 pr_warn("error disabling system %s\n",
4283 system->name);
4284 continue;
4285 }
4286
4287 pr_cont("OK\n");
4288 }
4289
4290 /* Test with all events enabled */
4291
4292 pr_info("Running tests on all trace events:\n");
4293 pr_info("Testing all events: ");
4294
4295 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4296 if (WARN_ON_ONCE(ret)) {
4297 pr_warn("error enabling all events\n");
4298 return;
4299 }
4300
4301 event_test_stuff();
4302
4303 /* reset sysname */
4304 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4305 if (WARN_ON_ONCE(ret)) {
4306 pr_warn("error disabling all events\n");
4307 return;
4308 }
4309
4310 pr_cont("OK\n");
4311 }
4312
4313 #ifdef CONFIG_FUNCTION_TRACER
4314
4315 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4316
4317 static struct trace_event_file event_trace_file __initdata;
4318
4319 static void __init
4320 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4321 struct ftrace_ops *op, struct ftrace_regs *regs)
4322 {
4323 struct trace_buffer *buffer;
4324 struct ring_buffer_event *event;
4325 struct ftrace_entry *entry;
4326 unsigned int trace_ctx;
4327 long disabled;
4328 int cpu;
4329
4330 trace_ctx = tracing_gen_ctx();
4331 preempt_disable_notrace();
4332 cpu = raw_smp_processor_id();
4333 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4334
4335 if (disabled != 1)
4336 goto out;
4337
4338 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4339 TRACE_FN, sizeof(*entry),
4340 trace_ctx);
4341 if (!event)
4342 goto out;
4343 entry = ring_buffer_event_data(event);
4344 entry->ip = ip;
4345 entry->parent_ip = parent_ip;
4346
4347 event_trigger_unlock_commit(&event_trace_file, buffer, event,
4348 entry, trace_ctx);
4349 out:
4350 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4351 preempt_enable_notrace();
4352 }
4353
4354 static struct ftrace_ops trace_ops __initdata =
4355 {
4356 .func = function_test_events_call,
4357 };
4358
4359 static __init void event_trace_self_test_with_function(void)
4360 {
4361 int ret;
4362
4363 event_trace_file.tr = top_trace_array();
4364 if (WARN_ON(!event_trace_file.tr))
4365 return;
4366
4367 ret = register_ftrace_function(&trace_ops);
4368 if (WARN_ON(ret < 0)) {
4369 pr_info("Failed to enable function tracer for event tests\n");
4370 return;
4371 }
4372 pr_info("Running tests again, along with the function tracer\n");
4373 event_trace_self_tests();
4374 unregister_ftrace_function(&trace_ops);
4375 }
4376 #else
4377 static __init void event_trace_self_test_with_function(void)
4378 {
4379 }
4380 #endif
4381
4382 static __init int event_trace_self_tests_init(void)
4383 {
4384 if (!tracing_selftest_disabled) {
4385 event_trace_self_tests();
4386 event_trace_self_test_with_function();
4387 }
4388
4389 return 0;
4390 }
4391
4392 late_initcall(event_trace_self_tests_init);
4393
4394 #endif
4395
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.