1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
7 * - Added format output of fields of the trace point.
8 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
12 #define pr_fmt(fmt) fmt
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>
26 #include <trace/events/sched.h>
27 #include <trace/syscall.h>
29 #include <asm/setup.h>
31 #include "trace_output.h"
34 #define TRACE_SYSTEM "TRACE_SYSTEM"
36 DEFINE_MUTEX(event_mutex);
38 LIST_HEAD(ftrace_events);
39 static LIST_HEAD(ftrace_generic_fields);
40 static LIST_HEAD(ftrace_common_fields);
41 static bool eventdir_initialized;
43 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
45 static struct kmem_cache *field_cachep;
46 static struct kmem_cache *file_cachep;
48 static inline int system_refcount(struct event_subsystem *system)
50 return system->ref_count;
53 static int system_refcount_inc(struct event_subsystem *system)
55 return system->ref_count++;
58 static int system_refcount_dec(struct event_subsystem *system)
60 return --system->ref_count;
63 /* Double loops, do not use break, only goto's work */
64 #define do_for_each_event_file(tr, file) \
65 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
66 list_for_each_entry(file, &tr->events, list)
68 #define do_for_each_event_file_safe(tr, file) \
69 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
70 struct trace_event_file *___n; \
71 list_for_each_entry_safe(file, ___n, &tr->events, list)
73 #define while_for_each_event_file() \
76 static struct ftrace_event_field *
77 __find_event_field(struct list_head *head, char *name)
79 struct ftrace_event_field *field;
81 list_for_each_entry(field, head, link) {
82 if (!strcmp(field->name, name))
89 struct ftrace_event_field *
90 trace_find_event_field(struct trace_event_call *call, char *name)
92 struct ftrace_event_field *field;
93 struct list_head *head;
95 head = trace_get_fields(call);
96 field = __find_event_field(head, name);
100 field = __find_event_field(&ftrace_generic_fields, name);
104 return __find_event_field(&ftrace_common_fields, name);
107 static int __trace_define_field(struct list_head *head, const char *type,
108 const char *name, int offset, int size,
109 int is_signed, int filter_type)
111 struct ftrace_event_field *field;
113 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
120 if (filter_type == FILTER_OTHER)
121 field->filter_type = filter_assign_type(type);
123 field->filter_type = filter_type;
125 field->offset = offset;
127 field->is_signed = is_signed;
129 list_add(&field->link, head);
134 int trace_define_field(struct trace_event_call *call, const char *type,
135 const char *name, int offset, int size, int is_signed,
138 struct list_head *head;
140 if (WARN_ON(!call->class))
143 head = trace_get_fields(call);
144 return __trace_define_field(head, type, name, offset, size,
145 is_signed, filter_type);
147 EXPORT_SYMBOL_GPL(trace_define_field);
149 #define __generic_field(type, item, filter_type) \
150 ret = __trace_define_field(&ftrace_generic_fields, #type, \
151 #item, 0, 0, is_signed_type(type), \
156 #define __common_field(type, item) \
157 ret = __trace_define_field(&ftrace_common_fields, #type, \
159 offsetof(typeof(ent), item), \
161 is_signed_type(type), FILTER_OTHER); \
165 static int trace_define_generic_fields(void)
169 __generic_field(int, CPU, FILTER_CPU);
170 __generic_field(int, cpu, FILTER_CPU);
171 __generic_field(char *, COMM, FILTER_COMM);
172 __generic_field(char *, comm, FILTER_COMM);
177 static int trace_define_common_fields(void)
180 struct trace_entry ent;
182 __common_field(unsigned short, type);
183 __common_field(unsigned char, flags);
184 /* Holds both preempt_count and migrate_disable */
185 __common_field(unsigned char, preempt_count);
186 __common_field(int, pid);
191 static void trace_destroy_fields(struct trace_event_call *call)
193 struct ftrace_event_field *field, *next;
194 struct list_head *head;
196 head = trace_get_fields(call);
197 list_for_each_entry_safe(field, next, head, link) {
198 list_del(&field->link);
199 kmem_cache_free(field_cachep, field);
204 * run-time version of trace_event_get_offsets_<call>() that returns the last
205 * accessible offset of trace fields excluding __dynamic_array bytes
207 int trace_event_get_offsets(struct trace_event_call *call)
209 struct ftrace_event_field *tail;
210 struct list_head *head;
212 head = trace_get_fields(call);
214 * head->next points to the last field with the largest offset,
215 * since it was added last by trace_define_field()
217 tail = list_first_entry(head, struct ftrace_event_field, link);
218 return tail->offset + tail->size;
222 * Check if the referenced field is an array and return true,
223 * as arrays are OK to dereference.
225 static bool test_field(const char *fmt, struct trace_event_call *call)
227 struct trace_event_fields *field = call->class->fields_array;
228 const char *array_descriptor;
232 if (!(len = str_has_prefix(fmt, "REC->")))
235 for (p = fmt; *p; p++) {
236 if (!isalnum(*p) && *p != '_')
241 for (; field->type; field++) {
242 if (strncmp(field->name, fmt, len) ||
245 array_descriptor = strchr(field->type, '[');
246 /* This is an array and is OK to dereference. */
247 return array_descriptor != NULL;
253 * Examine the print fmt of the event looking for unsafe dereference
254 * pointers using %p* that could be recorded in the trace event and
255 * much later referenced after the pointer was freed. Dereferencing
256 * pointers are OK, if it is dereferenced into the event itself.
258 static void test_event_printk(struct trace_event_call *call)
260 u64 dereference_flags = 0;
262 const char *fmt, *c, *r, *a;
269 fmt = call->print_fmt;
274 for (i = 0; fmt[i]; i++) {
284 * The print fmt starts with a string that
285 * is processed first to find %p* usage,
286 * then after the first string, the print fmt
287 * contains arguments that are used to check
288 * if the dereferenced %p* usage is safe.
297 * If there was no %p* uses
300 if (!dereference_flags)
305 if (in_quote == fmt[i])
312 if (!first || !in_quote)
321 /* Find dereferencing fields */
322 switch (fmt[i + 1]) {
323 case 'B': case 'R': case 'r':
324 case 'b': case 'M': case 'm':
325 case 'I': case 'i': case 'E':
326 case 'U': case 'V': case 'N':
327 case 'a': case 'd': case 'D':
328 case 'g': case 't': case 'C':
330 if (WARN_ONCE(arg == 63,
331 "Too many args for event: %s",
332 trace_event_name(call)))
334 dereference_flags |= 1ULL << arg;
342 /* Increment arg if %*s exists. */
343 for (j = 0; fmt[i + j]; j++) {
344 if (isdigit(fmt[i + j]) ||
347 if (fmt[i + j] == '*') {
351 if ((fmt[i + j] == 's') && star)
370 if (WARN_ONCE(parens < 0,
371 "Paren mismatch for event: %s\narg='%s'\n%*s",
372 trace_event_name(call),
374 (i - start_arg) + 5, "^"))
378 if (in_quote || parens)
381 while (isspace(fmt[i]))
384 if (!(dereference_flags & (1ULL << arg)))
387 /* Find the REC-> in the argument */
388 c = strchr(fmt + i, ',');
389 r = strstr(fmt + i, "REC->");
390 if (r && (!c || r < c)) {
392 * Addresses of events on the buffer,
393 * or an array on the buffer is
395 * There's ways to fool this, but
396 * this is to catch common mistakes,
397 * not malicious code.
399 a = strchr(fmt + i, '&');
400 if ((a && (a < r)) || test_field(r, call))
401 dereference_flags &= ~(1ULL << arg);
410 * If you triggered the below warning, the trace event reported
411 * uses an unsafe dereference pointer %p*. As the data stored
412 * at the trace event time may no longer exist when the trace
413 * event is printed, dereferencing to the original source is
414 * unsafe. The source of the dereference must be copied into the
415 * event itself, and the dereference must access the copy instead.
417 if (WARN_ON_ONCE(dereference_flags)) {
419 while (!(dereference_flags & 1)) {
420 dereference_flags >>= 1;
423 pr_warn("event %s has unsafe dereference of argument %d\n",
424 trace_event_name(call), arg);
425 pr_warn("print_fmt: %s\n", fmt);
429 int trace_event_raw_init(struct trace_event_call *call)
433 id = register_trace_event(&call->event);
437 test_event_printk(call);
441 EXPORT_SYMBOL_GPL(trace_event_raw_init);
443 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
445 struct trace_array *tr = trace_file->tr;
446 struct trace_array_cpu *data;
447 struct trace_pid_list *no_pid_list;
448 struct trace_pid_list *pid_list;
450 pid_list = rcu_dereference_raw(tr->filtered_pids);
451 no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
453 if (!pid_list && !no_pid_list)
456 data = this_cpu_ptr(tr->array_buffer.data);
458 return data->ignore_pid;
460 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
462 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
463 struct trace_event_file *trace_file,
466 struct trace_event_call *event_call = trace_file->event_call;
468 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
469 trace_event_ignore_this_pid(trace_file))
473 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
474 * preemption (adding one to the preempt_count). Since we are
475 * interested in the preempt_count at the time the tracepoint was
476 * hit, we need to subtract one to offset the increment.
478 fbuffer->trace_ctx = tracing_gen_ctx_dec();
479 fbuffer->trace_file = trace_file;
482 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
483 event_call->event.type, len,
488 fbuffer->regs = NULL;
489 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
490 return fbuffer->entry;
492 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
494 int trace_event_reg(struct trace_event_call *call,
495 enum trace_reg type, void *data)
497 struct trace_event_file *file = data;
499 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
501 case TRACE_REG_REGISTER:
502 return tracepoint_probe_register(call->tp,
505 case TRACE_REG_UNREGISTER:
506 tracepoint_probe_unregister(call->tp,
511 #ifdef CONFIG_PERF_EVENTS
512 case TRACE_REG_PERF_REGISTER:
513 return tracepoint_probe_register(call->tp,
514 call->class->perf_probe,
516 case TRACE_REG_PERF_UNREGISTER:
517 tracepoint_probe_unregister(call->tp,
518 call->class->perf_probe,
521 case TRACE_REG_PERF_OPEN:
522 case TRACE_REG_PERF_CLOSE:
523 case TRACE_REG_PERF_ADD:
524 case TRACE_REG_PERF_DEL:
530 EXPORT_SYMBOL_GPL(trace_event_reg);
532 void trace_event_enable_cmd_record(bool enable)
534 struct trace_event_file *file;
535 struct trace_array *tr;
537 lockdep_assert_held(&event_mutex);
539 do_for_each_event_file(tr, file) {
541 if (!(file->flags & EVENT_FILE_FL_ENABLED))
545 tracing_start_cmdline_record();
546 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
548 tracing_stop_cmdline_record();
549 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
551 } while_for_each_event_file();
554 void trace_event_enable_tgid_record(bool enable)
556 struct trace_event_file *file;
557 struct trace_array *tr;
559 lockdep_assert_held(&event_mutex);
561 do_for_each_event_file(tr, file) {
562 if (!(file->flags & EVENT_FILE_FL_ENABLED))
566 tracing_start_tgid_record();
567 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
569 tracing_stop_tgid_record();
570 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
573 } while_for_each_event_file();
576 static int __ftrace_event_enable_disable(struct trace_event_file *file,
577 int enable, int soft_disable)
579 struct trace_event_call *call = file->event_call;
580 struct trace_array *tr = file->tr;
581 unsigned long file_flags = file->flags;
588 * When soft_disable is set and enable is cleared, the sm_ref
589 * reference counter is decremented. If it reaches 0, we want
590 * to clear the SOFT_DISABLED flag but leave the event in the
591 * state that it was. That is, if the event was enabled and
592 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
593 * is set we do not want the event to be enabled before we
596 * When soft_disable is not set but the SOFT_MODE flag is,
597 * we do nothing. Do not disable the tracepoint, otherwise
598 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
601 if (atomic_dec_return(&file->sm_ref) > 0)
603 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
604 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
606 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
608 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
609 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
610 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
611 tracing_stop_cmdline_record();
612 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
615 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
616 tracing_stop_tgid_record();
617 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
620 call->class->reg(call, TRACE_REG_UNREGISTER, file);
622 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
623 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
624 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
626 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
630 * When soft_disable is set and enable is set, we want to
631 * register the tracepoint for the event, but leave the event
632 * as is. That means, if the event was already enabled, we do
633 * nothing (but set SOFT_MODE). If the event is disabled, we
634 * set SOFT_DISABLED before enabling the event tracepoint, so
635 * it still seems to be disabled.
638 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
640 if (atomic_inc_return(&file->sm_ref) > 1)
642 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
645 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
646 bool cmd = false, tgid = false;
648 /* Keep the event disabled, when going to SOFT_MODE. */
650 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
652 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
654 tracing_start_cmdline_record();
655 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
658 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
660 tracing_start_tgid_record();
661 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
664 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
667 tracing_stop_cmdline_record();
669 tracing_stop_tgid_record();
670 pr_info("event trace: Could not enable event "
671 "%s\n", trace_event_name(call));
674 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
676 /* WAS_ENABLED gets set but never cleared. */
677 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
682 /* Enable or disable use of trace_buffered_event */
683 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
684 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
685 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
686 trace_buffered_event_enable();
688 trace_buffered_event_disable();
694 int trace_event_enable_disable(struct trace_event_file *file,
695 int enable, int soft_disable)
697 return __ftrace_event_enable_disable(file, enable, soft_disable);
700 static int ftrace_event_enable_disable(struct trace_event_file *file,
703 return __ftrace_event_enable_disable(file, enable, 0);
706 static void ftrace_clear_events(struct trace_array *tr)
708 struct trace_event_file *file;
710 mutex_lock(&event_mutex);
711 list_for_each_entry(file, &tr->events, list) {
712 ftrace_event_enable_disable(file, 0);
714 mutex_unlock(&event_mutex);
718 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
720 struct trace_pid_list *pid_list;
721 struct trace_array *tr = data;
723 pid_list = rcu_dereference_raw(tr->filtered_pids);
724 trace_filter_add_remove_task(pid_list, NULL, task);
726 pid_list = rcu_dereference_raw(tr->filtered_no_pids);
727 trace_filter_add_remove_task(pid_list, NULL, task);
731 event_filter_pid_sched_process_fork(void *data,
732 struct task_struct *self,
733 struct task_struct *task)
735 struct trace_pid_list *pid_list;
736 struct trace_array *tr = data;
738 pid_list = rcu_dereference_sched(tr->filtered_pids);
739 trace_filter_add_remove_task(pid_list, self, task);
741 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
742 trace_filter_add_remove_task(pid_list, self, task);
745 void trace_event_follow_fork(struct trace_array *tr, bool enable)
748 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
750 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
753 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
755 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
761 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
762 struct task_struct *prev, struct task_struct *next)
764 struct trace_array *tr = data;
765 struct trace_pid_list *no_pid_list;
766 struct trace_pid_list *pid_list;
769 pid_list = rcu_dereference_sched(tr->filtered_pids);
770 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
773 * Sched switch is funny, as we only want to ignore it
774 * in the notrace case if both prev and next should be ignored.
776 ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
777 trace_ignore_this_task(NULL, no_pid_list, next);
779 this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
780 (trace_ignore_this_task(pid_list, NULL, prev) &&
781 trace_ignore_this_task(pid_list, NULL, next)));
785 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
786 struct task_struct *prev, struct task_struct *next)
788 struct trace_array *tr = data;
789 struct trace_pid_list *no_pid_list;
790 struct trace_pid_list *pid_list;
792 pid_list = rcu_dereference_sched(tr->filtered_pids);
793 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
795 this_cpu_write(tr->array_buffer.data->ignore_pid,
796 trace_ignore_this_task(pid_list, no_pid_list, next));
800 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
802 struct trace_array *tr = data;
803 struct trace_pid_list *no_pid_list;
804 struct trace_pid_list *pid_list;
806 /* Nothing to do if we are already tracing */
807 if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
810 pid_list = rcu_dereference_sched(tr->filtered_pids);
811 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
813 this_cpu_write(tr->array_buffer.data->ignore_pid,
814 trace_ignore_this_task(pid_list, no_pid_list, task));
818 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
820 struct trace_array *tr = data;
821 struct trace_pid_list *no_pid_list;
822 struct trace_pid_list *pid_list;
824 /* Nothing to do if we are not tracing */
825 if (this_cpu_read(tr->array_buffer.data->ignore_pid))
828 pid_list = rcu_dereference_sched(tr->filtered_pids);
829 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
831 /* Set tracing if current is enabled */
832 this_cpu_write(tr->array_buffer.data->ignore_pid,
833 trace_ignore_this_task(pid_list, no_pid_list, current));
836 static void unregister_pid_events(struct trace_array *tr)
838 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
839 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
841 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
842 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
844 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
845 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
847 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
848 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
851 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
853 struct trace_pid_list *pid_list;
854 struct trace_pid_list *no_pid_list;
855 struct trace_event_file *file;
858 pid_list = rcu_dereference_protected(tr->filtered_pids,
859 lockdep_is_held(&event_mutex));
860 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
861 lockdep_is_held(&event_mutex));
863 /* Make sure there's something to do */
864 if (!pid_type_enabled(type, pid_list, no_pid_list))
867 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
868 unregister_pid_events(tr);
870 list_for_each_entry(file, &tr->events, list) {
871 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
874 for_each_possible_cpu(cpu)
875 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
878 if (type & TRACE_PIDS)
879 rcu_assign_pointer(tr->filtered_pids, NULL);
881 if (type & TRACE_NO_PIDS)
882 rcu_assign_pointer(tr->filtered_no_pids, NULL);
884 /* Wait till all users are no longer using pid filtering */
885 tracepoint_synchronize_unregister();
887 if ((type & TRACE_PIDS) && pid_list)
888 trace_pid_list_free(pid_list);
890 if ((type & TRACE_NO_PIDS) && no_pid_list)
891 trace_pid_list_free(no_pid_list);
894 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
896 mutex_lock(&event_mutex);
897 __ftrace_clear_event_pids(tr, type);
898 mutex_unlock(&event_mutex);
901 static void __put_system(struct event_subsystem *system)
903 struct event_filter *filter = system->filter;
905 WARN_ON_ONCE(system_refcount(system) == 0);
906 if (system_refcount_dec(system))
909 list_del(&system->list);
912 kfree(filter->filter_string);
915 kfree_const(system->name);
919 static void __get_system(struct event_subsystem *system)
921 WARN_ON_ONCE(system_refcount(system) == 0);
922 system_refcount_inc(system);
925 static void __get_system_dir(struct trace_subsystem_dir *dir)
927 WARN_ON_ONCE(dir->ref_count == 0);
929 __get_system(dir->subsystem);
932 static void __put_system_dir(struct trace_subsystem_dir *dir)
934 WARN_ON_ONCE(dir->ref_count == 0);
935 /* If the subsystem is about to be freed, the dir must be too */
936 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
938 __put_system(dir->subsystem);
939 if (!--dir->ref_count)
943 static void put_system(struct trace_subsystem_dir *dir)
945 mutex_lock(&event_mutex);
946 __put_system_dir(dir);
947 mutex_unlock(&event_mutex);
950 static void remove_subsystem(struct trace_subsystem_dir *dir)
955 if (!--dir->nr_events) {
956 tracefs_remove(dir->entry);
957 list_del(&dir->list);
958 __put_system_dir(dir);
962 static void remove_event_file_dir(struct trace_event_file *file)
964 struct dentry *dir = file->dir;
965 struct dentry *child;
968 spin_lock(&dir->d_lock); /* probably unneeded */
969 list_for_each_entry(child, &dir->d_subdirs, d_child) {
970 if (d_really_is_positive(child)) /* probably unneeded */
971 d_inode(child)->i_private = NULL;
973 spin_unlock(&dir->d_lock);
978 list_del(&file->list);
979 remove_subsystem(file->system);
980 free_event_filter(file->filter);
981 kmem_cache_free(file_cachep, file);
985 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
988 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
989 const char *sub, const char *event, int set)
991 struct trace_event_file *file;
992 struct trace_event_call *call;
997 list_for_each_entry(file, &tr->events, list) {
999 call = file->event_call;
1000 name = trace_event_name(call);
1002 if (!name || !call->class || !call->class->reg)
1005 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1009 strcmp(match, name) != 0 &&
1010 strcmp(match, call->class->system) != 0)
1013 if (sub && strcmp(sub, call->class->system) != 0)
1016 if (event && strcmp(event, name) != 0)
1019 ret = ftrace_event_enable_disable(file, set);
1022 * Save the first error and return that. Some events
1023 * may still have been enabled, but let the user
1024 * know that something went wrong.
1035 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1036 const char *sub, const char *event, int set)
1040 mutex_lock(&event_mutex);
1041 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1042 mutex_unlock(&event_mutex);
1047 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1049 char *event = NULL, *sub = NULL, *match;
1055 * The buf format can be <subsystem>:<event-name>
1056 * *:<event-name> means any event by that name.
1057 * :<event-name> is the same.
1059 * <subsystem>:* means all events in that subsystem
1060 * <subsystem>: means the same.
1062 * <name> (no ':') means all events in a subsystem with
1063 * the name <name> or any event that matches <name>
1066 match = strsep(&buf, ":");
1072 if (!strlen(sub) || strcmp(sub, "*") == 0)
1074 if (!strlen(event) || strcmp(event, "*") == 0)
1078 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1080 /* Put back the colon to allow this to be called again */
1088 * trace_set_clr_event - enable or disable an event
1089 * @system: system name to match (NULL for any system)
1090 * @event: event name to match (NULL for all events, within system)
1091 * @set: 1 to enable, 0 to disable
1093 * This is a way for other parts of the kernel to enable or disable
1096 * Returns 0 on success, -EINVAL if the parameters do not match any
1097 * registered events.
1099 int trace_set_clr_event(const char *system, const char *event, int set)
1101 struct trace_array *tr = top_trace_array();
1106 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1108 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1111 * trace_array_set_clr_event - enable or disable an event for a trace array.
1112 * @tr: concerned trace array.
1113 * @system: system name to match (NULL for any system)
1114 * @event: event name to match (NULL for all events, within system)
1115 * @enable: true to enable, false to disable
1117 * This is a way for other parts of the kernel to enable or disable
1120 * Returns 0 on success, -EINVAL if the parameters do not match any
1121 * registered events.
1123 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1124 const char *event, bool enable)
1131 set = (enable == true) ? 1 : 0;
1132 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1134 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1136 /* 128 should be much more than enough */
1137 #define EVENT_BUF_SIZE 127
1140 ftrace_event_write(struct file *file, const char __user *ubuf,
1141 size_t cnt, loff_t *ppos)
1143 struct trace_parser parser;
1144 struct seq_file *m = file->private_data;
1145 struct trace_array *tr = m->private;
1151 ret = tracing_update_buffers();
1155 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1158 read = trace_get_user(&parser, ubuf, cnt, ppos);
1160 if (read >= 0 && trace_parser_loaded((&parser))) {
1163 if (*parser.buffer == '!')
1166 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1174 trace_parser_put(&parser);
1180 t_next(struct seq_file *m, void *v, loff_t *pos)
1182 struct trace_event_file *file = v;
1183 struct trace_event_call *call;
1184 struct trace_array *tr = m->private;
1188 list_for_each_entry_continue(file, &tr->events, list) {
1189 call = file->event_call;
1191 * The ftrace subsystem is for showing formats only.
1192 * They can not be enabled or disabled via the event files.
1194 if (call->class && call->class->reg &&
1195 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1202 static void *t_start(struct seq_file *m, loff_t *pos)
1204 struct trace_event_file *file;
1205 struct trace_array *tr = m->private;
1208 mutex_lock(&event_mutex);
1210 file = list_entry(&tr->events, struct trace_event_file, list);
1211 for (l = 0; l <= *pos; ) {
1212 file = t_next(m, file, &l);
1220 s_next(struct seq_file *m, void *v, loff_t *pos)
1222 struct trace_event_file *file = v;
1223 struct trace_array *tr = m->private;
1227 list_for_each_entry_continue(file, &tr->events, list) {
1228 if (file->flags & EVENT_FILE_FL_ENABLED)
1235 static void *s_start(struct seq_file *m, loff_t *pos)
1237 struct trace_event_file *file;
1238 struct trace_array *tr = m->private;
1241 mutex_lock(&event_mutex);
1243 file = list_entry(&tr->events, struct trace_event_file, list);
1244 for (l = 0; l <= *pos; ) {
1245 file = s_next(m, file, &l);
1252 static int t_show(struct seq_file *m, void *v)
1254 struct trace_event_file *file = v;
1255 struct trace_event_call *call = file->event_call;
1257 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1258 seq_printf(m, "%s:", call->class->system);
1259 seq_printf(m, "%s\n", trace_event_name(call));
1264 static void t_stop(struct seq_file *m, void *p)
1266 mutex_unlock(&event_mutex);
1270 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1272 struct trace_array *tr = m->private;
1273 struct trace_pid_list *pid_list;
1275 if (type == TRACE_PIDS)
1276 pid_list = rcu_dereference_sched(tr->filtered_pids);
1278 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1280 return trace_pid_next(pid_list, v, pos);
1284 p_next(struct seq_file *m, void *v, loff_t *pos)
1286 return __next(m, v, pos, TRACE_PIDS);
1290 np_next(struct seq_file *m, void *v, loff_t *pos)
1292 return __next(m, v, pos, TRACE_NO_PIDS);
1295 static void *__start(struct seq_file *m, loff_t *pos, int type)
1298 struct trace_pid_list *pid_list;
1299 struct trace_array *tr = m->private;
1302 * Grab the mutex, to keep calls to p_next() having the same
1303 * tr->filtered_pids as p_start() has.
1304 * If we just passed the tr->filtered_pids around, then RCU would
1305 * have been enough, but doing that makes things more complex.
1307 mutex_lock(&event_mutex);
1308 rcu_read_lock_sched();
1310 if (type == TRACE_PIDS)
1311 pid_list = rcu_dereference_sched(tr->filtered_pids);
1313 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1318 return trace_pid_start(pid_list, pos);
1321 static void *p_start(struct seq_file *m, loff_t *pos)
1324 return __start(m, pos, TRACE_PIDS);
1327 static void *np_start(struct seq_file *m, loff_t *pos)
1330 return __start(m, pos, TRACE_NO_PIDS);
1333 static void p_stop(struct seq_file *m, void *p)
1336 rcu_read_unlock_sched();
1337 mutex_unlock(&event_mutex);
1341 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1344 struct trace_event_file *file;
1345 unsigned long flags;
1348 mutex_lock(&event_mutex);
1349 file = event_file_data(filp);
1351 flags = file->flags;
1352 mutex_unlock(&event_mutex);
1357 if (flags & EVENT_FILE_FL_ENABLED &&
1358 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1361 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1362 flags & EVENT_FILE_FL_SOFT_MODE)
1367 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1371 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1374 struct trace_event_file *file;
1378 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1382 ret = tracing_update_buffers();
1390 mutex_lock(&event_mutex);
1391 file = event_file_data(filp);
1393 ret = ftrace_event_enable_disable(file, val);
1394 mutex_unlock(&event_mutex);
1403 return ret ? ret : cnt;
1407 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1410 const char set_to_char[4] = { '?', '0', '1', 'X' };
1411 struct trace_subsystem_dir *dir = filp->private_data;
1412 struct event_subsystem *system = dir->subsystem;
1413 struct trace_event_call *call;
1414 struct trace_event_file *file;
1415 struct trace_array *tr = dir->tr;
1420 mutex_lock(&event_mutex);
1421 list_for_each_entry(file, &tr->events, list) {
1422 call = file->event_call;
1423 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1424 !trace_event_name(call) || !call->class || !call->class->reg)
1427 if (system && strcmp(call->class->system, system->name) != 0)
1431 * We need to find out if all the events are set
1432 * or if all events or cleared, or if we have
1435 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1438 * If we have a mixture, no need to look further.
1443 mutex_unlock(&event_mutex);
1445 buf[0] = set_to_char[set];
1448 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1454 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1457 struct trace_subsystem_dir *dir = filp->private_data;
1458 struct event_subsystem *system = dir->subsystem;
1459 const char *name = NULL;
1463 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1467 ret = tracing_update_buffers();
1471 if (val != 0 && val != 1)
1475 * Opening of "enable" adds a ref count to system,
1476 * so the name is safe to use.
1479 name = system->name;
1481 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1495 FORMAT_FIELD_SEPERATOR = 2,
1496 FORMAT_PRINTFMT = 3,
1499 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1501 struct trace_event_call *call = event_file_data(m->private);
1502 struct list_head *common_head = &ftrace_common_fields;
1503 struct list_head *head = trace_get_fields(call);
1504 struct list_head *node = v;
1508 switch ((unsigned long)v) {
1513 case FORMAT_FIELD_SEPERATOR:
1517 case FORMAT_PRINTFMT:
1523 if (node == common_head)
1524 return (void *)FORMAT_FIELD_SEPERATOR;
1525 else if (node == head)
1526 return (void *)FORMAT_PRINTFMT;
1531 static int f_show(struct seq_file *m, void *v)
1533 struct trace_event_call *call = event_file_data(m->private);
1534 struct ftrace_event_field *field;
1535 const char *array_descriptor;
1537 switch ((unsigned long)v) {
1539 seq_printf(m, "name: %s\n", trace_event_name(call));
1540 seq_printf(m, "ID: %d\n", call->event.type);
1541 seq_puts(m, "format:\n");
1544 case FORMAT_FIELD_SEPERATOR:
1548 case FORMAT_PRINTFMT:
1549 seq_printf(m, "\nprint fmt: %s\n",
1554 field = list_entry(v, struct ftrace_event_field, link);
1556 * Smartly shows the array type(except dynamic array).
1559 * If TYPE := TYPE[LEN], it is shown:
1560 * field:TYPE VAR[LEN]
1562 array_descriptor = strchr(field->type, '[');
1564 if (str_has_prefix(field->type, "__data_loc"))
1565 array_descriptor = NULL;
1567 if (!array_descriptor)
1568 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1569 field->type, field->name, field->offset,
1570 field->size, !!field->is_signed);
1572 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1573 (int)(array_descriptor - field->type),
1574 field->type, field->name,
1575 array_descriptor, field->offset,
1576 field->size, !!field->is_signed);
1581 static void *f_start(struct seq_file *m, loff_t *pos)
1583 void *p = (void *)FORMAT_HEADER;
1586 /* ->stop() is called even if ->start() fails */
1587 mutex_lock(&event_mutex);
1588 if (!event_file_data(m->private))
1589 return ERR_PTR(-ENODEV);
1591 while (l < *pos && p)
1592 p = f_next(m, p, &l);
1597 static void f_stop(struct seq_file *m, void *p)
1599 mutex_unlock(&event_mutex);
1602 static const struct seq_operations trace_format_seq_ops = {
1609 static int trace_format_open(struct inode *inode, struct file *file)
1614 /* Do we want to hide event format files on tracefs lockdown? */
1616 ret = seq_open(file, &trace_format_seq_ops);
1620 m = file->private_data;
1627 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1629 int id = (long)event_file_data(filp);
1636 len = sprintf(buf, "%d\n", id);
1638 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1642 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1645 struct trace_event_file *file;
1646 struct trace_seq *s;
1652 s = kmalloc(sizeof(*s), GFP_KERNEL);
1659 mutex_lock(&event_mutex);
1660 file = event_file_data(filp);
1662 print_event_filter(file, s);
1663 mutex_unlock(&event_mutex);
1666 r = simple_read_from_buffer(ubuf, cnt, ppos,
1667 s->buffer, trace_seq_used(s));
1675 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1678 struct trace_event_file *file;
1682 if (cnt >= PAGE_SIZE)
1685 buf = memdup_user_nul(ubuf, cnt);
1687 return PTR_ERR(buf);
1689 mutex_lock(&event_mutex);
1690 file = event_file_data(filp);
1692 err = apply_event_filter(file, buf);
1693 mutex_unlock(&event_mutex);
1704 static LIST_HEAD(event_subsystems);
1706 static int subsystem_open(struct inode *inode, struct file *filp)
1708 struct event_subsystem *system = NULL;
1709 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1710 struct trace_array *tr;
1713 if (tracing_is_disabled())
1716 /* Make sure the system still exists */
1717 mutex_lock(&event_mutex);
1718 mutex_lock(&trace_types_lock);
1719 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1720 list_for_each_entry(dir, &tr->systems, list) {
1721 if (dir == inode->i_private) {
1722 /* Don't open systems with no events */
1723 if (dir->nr_events) {
1724 __get_system_dir(dir);
1725 system = dir->subsystem;
1732 mutex_unlock(&trace_types_lock);
1733 mutex_unlock(&event_mutex);
1738 /* Some versions of gcc think dir can be uninitialized here */
1741 /* Still need to increment the ref count of the system */
1742 if (trace_array_get(tr) < 0) {
1747 ret = tracing_open_generic(inode, filp);
1749 trace_array_put(tr);
1756 static int system_tr_open(struct inode *inode, struct file *filp)
1758 struct trace_subsystem_dir *dir;
1759 struct trace_array *tr = inode->i_private;
1762 /* Make a temporary dir that has no system but points to tr */
1763 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1767 ret = tracing_open_generic_tr(inode, filp);
1773 filp->private_data = dir;
1778 static int subsystem_release(struct inode *inode, struct file *file)
1780 struct trace_subsystem_dir *dir = file->private_data;
1782 trace_array_put(dir->tr);
1785 * If dir->subsystem is NULL, then this is a temporary
1786 * descriptor that was made for a trace_array to enable
1798 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1801 struct trace_subsystem_dir *dir = filp->private_data;
1802 struct event_subsystem *system = dir->subsystem;
1803 struct trace_seq *s;
1809 s = kmalloc(sizeof(*s), GFP_KERNEL);
1815 print_subsystem_event_filter(system, s);
1816 r = simple_read_from_buffer(ubuf, cnt, ppos,
1817 s->buffer, trace_seq_used(s));
1825 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1828 struct trace_subsystem_dir *dir = filp->private_data;
1832 if (cnt >= PAGE_SIZE)
1835 buf = memdup_user_nul(ubuf, cnt);
1837 return PTR_ERR(buf);
1839 err = apply_subsystem_event_filter(dir, buf);
1850 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1852 int (*func)(struct trace_seq *s) = filp->private_data;
1853 struct trace_seq *s;
1859 s = kmalloc(sizeof(*s), GFP_KERNEL);
1866 r = simple_read_from_buffer(ubuf, cnt, ppos,
1867 s->buffer, trace_seq_used(s));
1874 static void ignore_task_cpu(void *data)
1876 struct trace_array *tr = data;
1877 struct trace_pid_list *pid_list;
1878 struct trace_pid_list *no_pid_list;
1881 * This function is called by on_each_cpu() while the
1882 * event_mutex is held.
1884 pid_list = rcu_dereference_protected(tr->filtered_pids,
1885 mutex_is_locked(&event_mutex));
1886 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1887 mutex_is_locked(&event_mutex));
1889 this_cpu_write(tr->array_buffer.data->ignore_pid,
1890 trace_ignore_this_task(pid_list, no_pid_list, current));
1893 static void register_pid_events(struct trace_array *tr)
1896 * Register a probe that is called before all other probes
1897 * to set ignore_pid if next or prev do not match.
1898 * Register a probe this is called after all other probes
1899 * to only keep ignore_pid set if next pid matches.
1901 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1903 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1906 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1908 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1911 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1913 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1916 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1918 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1923 event_pid_write(struct file *filp, const char __user *ubuf,
1924 size_t cnt, loff_t *ppos, int type)
1926 struct seq_file *m = filp->private_data;
1927 struct trace_array *tr = m->private;
1928 struct trace_pid_list *filtered_pids = NULL;
1929 struct trace_pid_list *other_pids = NULL;
1930 struct trace_pid_list *pid_list;
1931 struct trace_event_file *file;
1937 ret = tracing_update_buffers();
1941 mutex_lock(&event_mutex);
1943 if (type == TRACE_PIDS) {
1944 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1945 lockdep_is_held(&event_mutex));
1946 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1947 lockdep_is_held(&event_mutex));
1949 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1950 lockdep_is_held(&event_mutex));
1951 other_pids = rcu_dereference_protected(tr->filtered_pids,
1952 lockdep_is_held(&event_mutex));
1955 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1959 if (type == TRACE_PIDS)
1960 rcu_assign_pointer(tr->filtered_pids, pid_list);
1962 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1964 list_for_each_entry(file, &tr->events, list) {
1965 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1968 if (filtered_pids) {
1969 tracepoint_synchronize_unregister();
1970 trace_pid_list_free(filtered_pids);
1971 } else if (pid_list && !other_pids) {
1972 register_pid_events(tr);
1976 * Ignoring of pids is done at task switch. But we have to
1977 * check for those tasks that are currently running.
1978 * Always do this in case a pid was appended or removed.
1980 on_each_cpu(ignore_task_cpu, tr, 1);
1983 mutex_unlock(&event_mutex);
1992 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1993 size_t cnt, loff_t *ppos)
1995 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1999 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2000 size_t cnt, loff_t *ppos)
2002 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2005 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2006 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2007 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2008 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2009 static int ftrace_event_release(struct inode *inode, struct file *file);
2011 static const struct seq_operations show_event_seq_ops = {
2018 static const struct seq_operations show_set_event_seq_ops = {
2025 static const struct seq_operations show_set_pid_seq_ops = {
2028 .show = trace_pid_show,
2032 static const struct seq_operations show_set_no_pid_seq_ops = {
2035 .show = trace_pid_show,
2039 static const struct file_operations ftrace_avail_fops = {
2040 .open = ftrace_event_avail_open,
2042 .llseek = seq_lseek,
2043 .release = seq_release,
2046 static const struct file_operations ftrace_set_event_fops = {
2047 .open = ftrace_event_set_open,
2049 .write = ftrace_event_write,
2050 .llseek = seq_lseek,
2051 .release = ftrace_event_release,
2054 static const struct file_operations ftrace_set_event_pid_fops = {
2055 .open = ftrace_event_set_pid_open,
2057 .write = ftrace_event_pid_write,
2058 .llseek = seq_lseek,
2059 .release = ftrace_event_release,
2062 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2063 .open = ftrace_event_set_npid_open,
2065 .write = ftrace_event_npid_write,
2066 .llseek = seq_lseek,
2067 .release = ftrace_event_release,
2070 static const struct file_operations ftrace_enable_fops = {
2071 .open = tracing_open_generic,
2072 .read = event_enable_read,
2073 .write = event_enable_write,
2074 .llseek = default_llseek,
2077 static const struct file_operations ftrace_event_format_fops = {
2078 .open = trace_format_open,
2080 .llseek = seq_lseek,
2081 .release = seq_release,
2084 static const struct file_operations ftrace_event_id_fops = {
2085 .read = event_id_read,
2086 .llseek = default_llseek,
2089 static const struct file_operations ftrace_event_filter_fops = {
2090 .open = tracing_open_generic,
2091 .read = event_filter_read,
2092 .write = event_filter_write,
2093 .llseek = default_llseek,
2096 static const struct file_operations ftrace_subsystem_filter_fops = {
2097 .open = subsystem_open,
2098 .read = subsystem_filter_read,
2099 .write = subsystem_filter_write,
2100 .llseek = default_llseek,
2101 .release = subsystem_release,
2104 static const struct file_operations ftrace_system_enable_fops = {
2105 .open = subsystem_open,
2106 .read = system_enable_read,
2107 .write = system_enable_write,
2108 .llseek = default_llseek,
2109 .release = subsystem_release,
2112 static const struct file_operations ftrace_tr_enable_fops = {
2113 .open = system_tr_open,
2114 .read = system_enable_read,
2115 .write = system_enable_write,
2116 .llseek = default_llseek,
2117 .release = subsystem_release,
2120 static const struct file_operations ftrace_show_header_fops = {
2121 .open = tracing_open_generic,
2122 .read = show_header,
2123 .llseek = default_llseek,
2127 ftrace_event_open(struct inode *inode, struct file *file,
2128 const struct seq_operations *seq_ops)
2133 ret = security_locked_down(LOCKDOWN_TRACEFS);
2137 ret = seq_open(file, seq_ops);
2140 m = file->private_data;
2141 /* copy tr over to seq ops */
2142 m->private = inode->i_private;
2147 static int ftrace_event_release(struct inode *inode, struct file *file)
2149 struct trace_array *tr = inode->i_private;
2151 trace_array_put(tr);
2153 return seq_release(inode, file);
2157 ftrace_event_avail_open(struct inode *inode, struct file *file)
2159 const struct seq_operations *seq_ops = &show_event_seq_ops;
2161 /* Checks for tracefs lockdown */
2162 return ftrace_event_open(inode, file, seq_ops);
2166 ftrace_event_set_open(struct inode *inode, struct file *file)
2168 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2169 struct trace_array *tr = inode->i_private;
2172 ret = tracing_check_open_get_tr(tr);
2176 if ((file->f_mode & FMODE_WRITE) &&
2177 (file->f_flags & O_TRUNC))
2178 ftrace_clear_events(tr);
2180 ret = ftrace_event_open(inode, file, seq_ops);
2182 trace_array_put(tr);
2187 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2189 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2190 struct trace_array *tr = inode->i_private;
2193 ret = tracing_check_open_get_tr(tr);
2197 if ((file->f_mode & FMODE_WRITE) &&
2198 (file->f_flags & O_TRUNC))
2199 ftrace_clear_event_pids(tr, TRACE_PIDS);
2201 ret = ftrace_event_open(inode, file, seq_ops);
2203 trace_array_put(tr);
2208 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2210 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2211 struct trace_array *tr = inode->i_private;
2214 ret = tracing_check_open_get_tr(tr);
2218 if ((file->f_mode & FMODE_WRITE) &&
2219 (file->f_flags & O_TRUNC))
2220 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2222 ret = ftrace_event_open(inode, file, seq_ops);
2224 trace_array_put(tr);
2228 static struct event_subsystem *
2229 create_new_subsystem(const char *name)
2231 struct event_subsystem *system;
2233 /* need to create new entry */
2234 system = kmalloc(sizeof(*system), GFP_KERNEL);
2238 system->ref_count = 1;
2240 /* Only allocate if dynamic (kprobes and modules) */
2241 system->name = kstrdup_const(name, GFP_KERNEL);
2245 system->filter = NULL;
2247 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2248 if (!system->filter)
2251 list_add(&system->list, &event_subsystems);
2256 kfree_const(system->name);
2261 static struct dentry *
2262 event_subsystem_dir(struct trace_array *tr, const char *name,
2263 struct trace_event_file *file, struct dentry *parent)
2265 struct trace_subsystem_dir *dir;
2266 struct event_subsystem *system;
2267 struct dentry *entry;
2269 /* First see if we did not already create this dir */
2270 list_for_each_entry(dir, &tr->systems, list) {
2271 system = dir->subsystem;
2272 if (strcmp(system->name, name) == 0) {
2279 /* Now see if the system itself exists. */
2280 list_for_each_entry(system, &event_subsystems, list) {
2281 if (strcmp(system->name, name) == 0)
2284 /* Reset system variable when not found */
2285 if (&system->list == &event_subsystems)
2288 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2293 system = create_new_subsystem(name);
2297 __get_system(system);
2299 dir->entry = tracefs_create_dir(name, parent);
2301 pr_warn("Failed to create system directory %s\n", name);
2302 __put_system(system);
2309 dir->subsystem = system;
2312 /* the ftrace system is special, do not create enable or filter files */
2313 if (strcmp(name, "ftrace") != 0) {
2315 entry = tracefs_create_file("filter", TRACE_MODE_WRITE,
2317 &ftrace_subsystem_filter_fops);
2319 kfree(system->filter);
2320 system->filter = NULL;
2321 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2324 trace_create_file("enable", TRACE_MODE_WRITE, dir->entry, dir,
2325 &ftrace_system_enable_fops);
2328 list_add(&dir->list, &tr->systems);
2335 /* Only print this message if failed on memory allocation */
2336 if (!dir || !system)
2337 pr_warn("No memory to create event subsystem %s\n", name);
2342 event_define_fields(struct trace_event_call *call)
2344 struct list_head *head;
2348 * Other events may have the same class. Only update
2349 * the fields if they are not already defined.
2351 head = trace_get_fields(call);
2352 if (list_empty(head)) {
2353 struct trace_event_fields *field = call->class->fields_array;
2354 unsigned int offset = sizeof(struct trace_entry);
2356 for (; field->type; field++) {
2357 if (field->type == TRACE_FUNCTION_TYPE) {
2358 field->define_fields(call);
2362 offset = ALIGN(offset, field->align);
2363 ret = trace_define_field(call, field->type, field->name,
2364 offset, field->size,
2365 field->is_signed, field->filter_type);
2366 if (WARN_ON_ONCE(ret)) {
2367 pr_err("error code is %d\n", ret);
2371 offset += field->size;
2379 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2381 struct trace_event_call *call = file->event_call;
2382 struct trace_array *tr = file->tr;
2383 struct dentry *d_events;
2388 * If the trace point header did not define TRACE_SYSTEM
2389 * then the system would be called "TRACE_SYSTEM".
2391 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2392 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2398 name = trace_event_name(call);
2399 file->dir = tracefs_create_dir(name, d_events);
2401 pr_warn("Could not create tracefs '%s' directory\n", name);
2405 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2406 trace_create_file("enable", TRACE_MODE_WRITE, file->dir, file,
2407 &ftrace_enable_fops);
2409 #ifdef CONFIG_PERF_EVENTS
2410 if (call->event.type && call->class->reg)
2411 trace_create_file("id", TRACE_MODE_READ, file->dir,
2412 (void *)(long)call->event.type,
2413 &ftrace_event_id_fops);
2416 ret = event_define_fields(call);
2418 pr_warn("Could not initialize trace point events/%s\n", name);
2423 * Only event directories that can be enabled should have
2424 * triggers or filters.
2426 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2427 trace_create_file("filter", TRACE_MODE_WRITE, file->dir,
2428 file, &ftrace_event_filter_fops);
2430 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
2431 file, &event_trigger_fops);
2434 #ifdef CONFIG_HIST_TRIGGERS
2435 trace_create_file("hist", TRACE_MODE_READ, file->dir, file,
2438 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2439 trace_create_file("hist_debug", TRACE_MODE_READ, file->dir, file,
2440 &event_hist_debug_fops);
2442 trace_create_file("format", TRACE_MODE_READ, file->dir, call,
2443 &ftrace_event_format_fops);
2445 #ifdef CONFIG_TRACE_EVENT_INJECT
2446 if (call->event.type && call->class->reg)
2447 trace_create_file("inject", 0200, file->dir, file,
2448 &event_inject_fops);
2454 static void remove_event_from_tracers(struct trace_event_call *call)
2456 struct trace_event_file *file;
2457 struct trace_array *tr;
2459 do_for_each_event_file_safe(tr, file) {
2460 if (file->event_call != call)
2463 remove_event_file_dir(file);
2465 * The do_for_each_event_file_safe() is
2466 * a double loop. After finding the call for this
2467 * trace_array, we use break to jump to the next
2471 } while_for_each_event_file();
2474 static void event_remove(struct trace_event_call *call)
2476 struct trace_array *tr;
2477 struct trace_event_file *file;
2479 do_for_each_event_file(tr, file) {
2480 if (file->event_call != call)
2483 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2484 tr->clear_trace = true;
2486 ftrace_event_enable_disable(file, 0);
2488 * The do_for_each_event_file() is
2489 * a double loop. After finding the call for this
2490 * trace_array, we use break to jump to the next
2494 } while_for_each_event_file();
2496 if (call->event.funcs)
2497 __unregister_trace_event(&call->event);
2498 remove_event_from_tracers(call);
2499 list_del(&call->list);
2502 static int event_init(struct trace_event_call *call)
2507 name = trace_event_name(call);
2511 if (call->class->raw_init) {
2512 ret = call->class->raw_init(call);
2513 if (ret < 0 && ret != -ENOSYS)
2514 pr_warn("Could not initialize trace events/%s\n", name);
2521 __register_event(struct trace_event_call *call, struct module *mod)
2525 ret = event_init(call);
2529 list_add(&call->list, &ftrace_events);
2530 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2531 atomic_set(&call->refcnt, 0);
2538 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2543 /* Find the length of the eval value as a string */
2544 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2545 /* Make sure there's enough room to replace the string with the value */
2549 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2551 /* Get the rest of the string of ptr */
2552 rlen = strlen(ptr + len);
2553 memmove(ptr + elen, ptr + len, rlen);
2554 /* Make sure we end the new string */
2555 ptr[elen + rlen] = 0;
2560 static void update_event_printk(struct trace_event_call *call,
2561 struct trace_eval_map *map)
2565 int len = strlen(map->eval_string);
2567 for (ptr = call->print_fmt; *ptr; ptr++) {
2581 if (isdigit(*ptr)) {
2585 /* Check for alpha chars like ULL */
2586 } while (isalnum(*ptr));
2590 * A number must have some kind of delimiter after
2591 * it, and we can ignore that too.
2595 if (isalpha(*ptr) || *ptr == '_') {
2596 if (strncmp(map->eval_string, ptr, len) == 0 &&
2597 !isalnum(ptr[len]) && ptr[len] != '_') {
2598 ptr = eval_replace(ptr, map, len);
2599 /* enum/sizeof string smaller than value */
2600 if (WARN_ON_ONCE(!ptr))
2603 * No need to decrement here, as eval_replace()
2604 * returns the pointer to the character passed
2605 * the eval, and two evals can not be placed
2606 * back to back without something in between.
2607 * We can skip that something in between.
2614 } while (isalnum(*ptr) || *ptr == '_');
2618 * If what comes after this variable is a '.' or
2619 * '->' then we can continue to ignore that string.
2621 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2622 ptr += *ptr == '.' ? 1 : 2;
2628 * Once again, we can skip the delimiter that came
2636 void trace_event_eval_update(struct trace_eval_map **map, int len)
2638 struct trace_event_call *call, *p;
2639 const char *last_system = NULL;
2644 down_write(&trace_event_sem);
2645 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2646 /* events are usually grouped together with systems */
2647 if (!last_system || call->class->system != last_system) {
2650 last_system = call->class->system;
2654 * Since calls are grouped by systems, the likelihood that the
2655 * next call in the iteration belongs to the same system as the
2656 * previous call is high. As an optimization, we skip searching
2657 * for a map[] that matches the call's system if the last call
2658 * was from the same system. That's what last_i is for. If the
2659 * call has the same system as the previous call, then last_i
2660 * will be the index of the first map[] that has a matching
2663 for (i = last_i; i < len; i++) {
2664 if (call->class->system == map[i]->system) {
2665 /* Save the first system if need be */
2670 update_event_printk(call, map[i]);
2674 up_write(&trace_event_sem);
2677 static struct trace_event_file *
2678 trace_create_new_event(struct trace_event_call *call,
2679 struct trace_array *tr)
2681 struct trace_pid_list *no_pid_list;
2682 struct trace_pid_list *pid_list;
2683 struct trace_event_file *file;
2686 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2690 pid_list = rcu_dereference_protected(tr->filtered_pids,
2691 lockdep_is_held(&event_mutex));
2692 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2693 lockdep_is_held(&event_mutex));
2695 if (!trace_pid_list_first(pid_list, &first) ||
2696 !trace_pid_list_first(no_pid_list, &first))
2697 file->flags |= EVENT_FILE_FL_PID_FILTER;
2699 file->event_call = call;
2701 atomic_set(&file->sm_ref, 0);
2702 atomic_set(&file->tm_ref, 0);
2703 INIT_LIST_HEAD(&file->triggers);
2704 list_add(&file->list, &tr->events);
2709 /* Add an event to a trace directory */
2711 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2713 struct trace_event_file *file;
2715 file = trace_create_new_event(call, tr);
2719 if (eventdir_initialized)
2720 return event_create_dir(tr->event_dir, file);
2722 return event_define_fields(call);
2726 * Just create a descriptor for early init. A descriptor is required
2727 * for enabling events at boot. We want to enable events before
2728 * the filesystem is initialized.
2731 __trace_early_add_new_event(struct trace_event_call *call,
2732 struct trace_array *tr)
2734 struct trace_event_file *file;
2736 file = trace_create_new_event(call, tr);
2740 return event_define_fields(call);
2743 struct ftrace_module_file_ops;
2744 static void __add_event_to_tracers(struct trace_event_call *call);
2746 /* Add an additional event_call dynamically */
2747 int trace_add_event_call(struct trace_event_call *call)
2750 lockdep_assert_held(&event_mutex);
2752 mutex_lock(&trace_types_lock);
2754 ret = __register_event(call, NULL);
2756 __add_event_to_tracers(call);
2758 mutex_unlock(&trace_types_lock);
2763 * Must be called under locking of trace_types_lock, event_mutex and
2766 static void __trace_remove_event_call(struct trace_event_call *call)
2769 trace_destroy_fields(call);
2770 free_event_filter(call->filter);
2771 call->filter = NULL;
2774 static int probe_remove_event_call(struct trace_event_call *call)
2776 struct trace_array *tr;
2777 struct trace_event_file *file;
2779 #ifdef CONFIG_PERF_EVENTS
2780 if (call->perf_refcount)
2783 do_for_each_event_file(tr, file) {
2784 if (file->event_call != call)
2787 * We can't rely on ftrace_event_enable_disable(enable => 0)
2788 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2789 * TRACE_REG_UNREGISTER.
2791 if (file->flags & EVENT_FILE_FL_ENABLED)
2794 * The do_for_each_event_file_safe() is
2795 * a double loop. After finding the call for this
2796 * trace_array, we use break to jump to the next
2800 } while_for_each_event_file();
2802 __trace_remove_event_call(call);
2807 /* Remove an event_call */
2808 int trace_remove_event_call(struct trace_event_call *call)
2812 lockdep_assert_held(&event_mutex);
2814 mutex_lock(&trace_types_lock);
2815 down_write(&trace_event_sem);
2816 ret = probe_remove_event_call(call);
2817 up_write(&trace_event_sem);
2818 mutex_unlock(&trace_types_lock);
2823 #define for_each_event(event, start, end) \
2824 for (event = start; \
2825 (unsigned long)event < (unsigned long)end; \
2828 #ifdef CONFIG_MODULES
2830 static void trace_module_add_events(struct module *mod)
2832 struct trace_event_call **call, **start, **end;
2834 if (!mod->num_trace_events)
2837 /* Don't add infrastructure for mods without tracepoints */
2838 if (trace_module_has_bad_taint(mod)) {
2839 pr_err("%s: module has bad taint, not creating trace events\n",
2844 start = mod->trace_events;
2845 end = mod->trace_events + mod->num_trace_events;
2847 for_each_event(call, start, end) {
2848 __register_event(*call, mod);
2849 __add_event_to_tracers(*call);
2853 static void trace_module_remove_events(struct module *mod)
2855 struct trace_event_call *call, *p;
2857 down_write(&trace_event_sem);
2858 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2859 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
2861 if (call->module == mod)
2862 __trace_remove_event_call(call);
2864 up_write(&trace_event_sem);
2867 * It is safest to reset the ring buffer if the module being unloaded
2868 * registered any events that were used. The only worry is if
2869 * a new module gets loaded, and takes on the same id as the events
2870 * of this module. When printing out the buffer, traced events left
2871 * over from this module may be passed to the new module events and
2872 * unexpected results may occur.
2874 tracing_reset_all_online_cpus();
2877 static int trace_module_notify(struct notifier_block *self,
2878 unsigned long val, void *data)
2880 struct module *mod = data;
2882 mutex_lock(&event_mutex);
2883 mutex_lock(&trace_types_lock);
2885 case MODULE_STATE_COMING:
2886 trace_module_add_events(mod);
2888 case MODULE_STATE_GOING:
2889 trace_module_remove_events(mod);
2892 mutex_unlock(&trace_types_lock);
2893 mutex_unlock(&event_mutex);
2898 static struct notifier_block trace_module_nb = {
2899 .notifier_call = trace_module_notify,
2900 .priority = 1, /* higher than trace.c module notify */
2902 #endif /* CONFIG_MODULES */
2904 /* Create a new event directory structure for a trace directory. */
2906 __trace_add_event_dirs(struct trace_array *tr)
2908 struct trace_event_call *call;
2911 list_for_each_entry(call, &ftrace_events, list) {
2912 ret = __trace_add_new_event(call, tr);
2914 pr_warn("Could not create directory for event %s\n",
2915 trace_event_name(call));
2919 /* Returns any file that matches the system and event */
2920 struct trace_event_file *
2921 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2923 struct trace_event_file *file;
2924 struct trace_event_call *call;
2927 list_for_each_entry(file, &tr->events, list) {
2929 call = file->event_call;
2930 name = trace_event_name(call);
2932 if (!name || !call->class)
2935 if (strcmp(event, name) == 0 &&
2936 strcmp(system, call->class->system) == 0)
2942 /* Returns valid trace event files that match system and event */
2943 struct trace_event_file *
2944 find_event_file(struct trace_array *tr, const char *system, const char *event)
2946 struct trace_event_file *file;
2948 file = __find_event_file(tr, system, event);
2949 if (!file || !file->event_call->class->reg ||
2950 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2957 * trace_get_event_file - Find and return a trace event file
2958 * @instance: The name of the trace instance containing the event
2959 * @system: The name of the system containing the event
2960 * @event: The name of the event
2962 * Return a trace event file given the trace instance name, trace
2963 * system, and trace event name. If the instance name is NULL, it
2964 * refers to the top-level trace array.
2966 * This function will look it up and return it if found, after calling
2967 * trace_array_get() to prevent the instance from going away, and
2968 * increment the event's module refcount to prevent it from being
2971 * To release the file, call trace_put_event_file(), which will call
2972 * trace_array_put() and decrement the event's module refcount.
2974 * Return: The trace event on success, ERR_PTR otherwise.
2976 struct trace_event_file *trace_get_event_file(const char *instance,
2980 struct trace_array *tr = top_trace_array();
2981 struct trace_event_file *file = NULL;
2985 tr = trace_array_find_get(instance);
2987 return ERR_PTR(-ENOENT);
2989 ret = trace_array_get(tr);
2991 return ERR_PTR(ret);
2994 mutex_lock(&event_mutex);
2996 file = find_event_file(tr, system, event);
2998 trace_array_put(tr);
3003 /* Don't let event modules unload while in use */
3004 ret = trace_event_try_get_ref(file->event_call);
3006 trace_array_put(tr);
3013 mutex_unlock(&event_mutex);
3016 file = ERR_PTR(ret);
3020 EXPORT_SYMBOL_GPL(trace_get_event_file);
3023 * trace_put_event_file - Release a file from trace_get_event_file()
3024 * @file: The trace event file
3026 * If a file was retrieved using trace_get_event_file(), this should
3027 * be called when it's no longer needed. It will cancel the previous
3028 * trace_array_get() called by that function, and decrement the
3029 * event's module refcount.
3031 void trace_put_event_file(struct trace_event_file *file)
3033 mutex_lock(&event_mutex);
3034 trace_event_put_ref(file->event_call);
3035 mutex_unlock(&event_mutex);
3037 trace_array_put(file->tr);
3039 EXPORT_SYMBOL_GPL(trace_put_event_file);
3041 #ifdef CONFIG_DYNAMIC_FTRACE
3044 #define ENABLE_EVENT_STR "enable_event"
3045 #define DISABLE_EVENT_STR "disable_event"
3047 struct event_probe_data {
3048 struct trace_event_file *file;
3049 unsigned long count;
3054 static void update_event_probe(struct event_probe_data *data)
3057 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3059 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3063 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3064 struct trace_array *tr, struct ftrace_probe_ops *ops,
3067 struct ftrace_func_mapper *mapper = data;
3068 struct event_probe_data *edata;
3071 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3072 if (!pdata || !*pdata)
3076 update_event_probe(edata);
3080 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3081 struct trace_array *tr, struct ftrace_probe_ops *ops,
3084 struct ftrace_func_mapper *mapper = data;
3085 struct event_probe_data *edata;
3088 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3089 if (!pdata || !*pdata)
3097 /* Skip if the event is in a state we want to switch to */
3098 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3101 if (edata->count != -1)
3104 update_event_probe(edata);
3108 event_enable_print(struct seq_file *m, unsigned long ip,
3109 struct ftrace_probe_ops *ops, void *data)
3111 struct ftrace_func_mapper *mapper = data;
3112 struct event_probe_data *edata;
3115 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3117 if (WARN_ON_ONCE(!pdata || !*pdata))
3122 seq_printf(m, "%ps:", (void *)ip);
3124 seq_printf(m, "%s:%s:%s",
3125 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3126 edata->file->event_call->class->system,
3127 trace_event_name(edata->file->event_call));
3129 if (edata->count == -1)
3130 seq_puts(m, ":unlimited\n");
3132 seq_printf(m, ":count=%ld\n", edata->count);
3138 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3139 unsigned long ip, void *init_data, void **data)
3141 struct ftrace_func_mapper *mapper = *data;
3142 struct event_probe_data *edata = init_data;
3146 mapper = allocate_ftrace_func_mapper();
3152 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3161 static int free_probe_data(void *data)
3163 struct event_probe_data *edata = data;
3167 /* Remove the SOFT_MODE flag */
3168 __ftrace_event_enable_disable(edata->file, 0, 1);
3169 trace_event_put_ref(edata->file->event_call);
3176 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3177 unsigned long ip, void *data)
3179 struct ftrace_func_mapper *mapper = data;
3180 struct event_probe_data *edata;
3185 free_ftrace_func_mapper(mapper, free_probe_data);
3189 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3191 if (WARN_ON_ONCE(!edata))
3194 if (WARN_ON_ONCE(edata->ref <= 0))
3197 free_probe_data(edata);
3200 static struct ftrace_probe_ops event_enable_probe_ops = {
3201 .func = event_enable_probe,
3202 .print = event_enable_print,
3203 .init = event_enable_init,
3204 .free = event_enable_free,
3207 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3208 .func = event_enable_count_probe,
3209 .print = event_enable_print,
3210 .init = event_enable_init,
3211 .free = event_enable_free,
3214 static struct ftrace_probe_ops event_disable_probe_ops = {
3215 .func = event_enable_probe,
3216 .print = event_enable_print,
3217 .init = event_enable_init,
3218 .free = event_enable_free,
3221 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3222 .func = event_enable_count_probe,
3223 .print = event_enable_print,
3224 .init = event_enable_init,
3225 .free = event_enable_free,
3229 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3230 char *glob, char *cmd, char *param, int enabled)
3232 struct trace_event_file *file;
3233 struct ftrace_probe_ops *ops;
3234 struct event_probe_data *data;
3244 /* hash funcs only work with set_ftrace_filter */
3245 if (!enabled || !param)
3248 system = strsep(¶m, ":");
3252 event = strsep(¶m, ":");
3254 mutex_lock(&event_mutex);
3257 file = find_event_file(tr, system, event);
3261 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3264 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3266 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3268 if (glob[0] == '!') {
3269 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3275 data = kzalloc(sizeof(*data), GFP_KERNEL);
3279 data->enable = enable;
3286 number = strsep(¶m, ":");
3289 if (!strlen(number))
3293 * We use the callback data field (which is a pointer)
3296 ret = kstrtoul(number, 0, &data->count);
3301 /* Don't let event modules unload while probe registered */
3302 ret = trace_event_try_get_ref(file->event_call);
3308 ret = __ftrace_event_enable_disable(file, 1, 1);
3312 ret = register_ftrace_function_probe(glob, tr, ops, data);
3314 * The above returns on success the # of functions enabled,
3315 * but if it didn't find any functions it returns zero.
3316 * Consider no functions a failure too.
3323 /* Just return zero, not the number of enabled functions */
3326 mutex_unlock(&event_mutex);
3330 __ftrace_event_enable_disable(file, 0, 1);
3332 trace_event_put_ref(file->event_call);
3338 static struct ftrace_func_command event_enable_cmd = {
3339 .name = ENABLE_EVENT_STR,
3340 .func = event_enable_func,
3343 static struct ftrace_func_command event_disable_cmd = {
3344 .name = DISABLE_EVENT_STR,
3345 .func = event_enable_func,
3348 static __init int register_event_cmds(void)
3352 ret = register_ftrace_command(&event_enable_cmd);
3353 if (WARN_ON(ret < 0))
3355 ret = register_ftrace_command(&event_disable_cmd);
3356 if (WARN_ON(ret < 0))
3357 unregister_ftrace_command(&event_enable_cmd);
3361 static inline int register_event_cmds(void) { return 0; }
3362 #endif /* CONFIG_DYNAMIC_FTRACE */
3365 * The top level array and trace arrays created by boot-time tracing
3366 * have already had its trace_event_file descriptors created in order
3367 * to allow for early events to be recorded.
3368 * This function is called after the tracefs has been initialized,
3369 * and we now have to create the files associated to the events.
3371 static void __trace_early_add_event_dirs(struct trace_array *tr)
3373 struct trace_event_file *file;
3377 list_for_each_entry(file, &tr->events, list) {
3378 ret = event_create_dir(tr->event_dir, file);
3380 pr_warn("Could not create directory for event %s\n",
3381 trace_event_name(file->event_call));
3386 * For early boot up, the top trace array and the trace arrays created
3387 * by boot-time tracing require to have a list of events that can be
3388 * enabled. This must be done before the filesystem is set up in order
3389 * to allow events to be traced early.
3391 void __trace_early_add_events(struct trace_array *tr)
3393 struct trace_event_call *call;
3396 list_for_each_entry(call, &ftrace_events, list) {
3397 /* Early boot up should not have any modules loaded */
3398 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3399 WARN_ON_ONCE(call->module))
3402 ret = __trace_early_add_new_event(call, tr);
3404 pr_warn("Could not create early event %s\n",
3405 trace_event_name(call));
3409 /* Remove the event directory structure for a trace directory. */
3411 __trace_remove_event_dirs(struct trace_array *tr)
3413 struct trace_event_file *file, *next;
3415 list_for_each_entry_safe(file, next, &tr->events, list)
3416 remove_event_file_dir(file);
3419 static void __add_event_to_tracers(struct trace_event_call *call)
3421 struct trace_array *tr;
3423 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3424 __trace_add_new_event(call, tr);
3427 extern struct trace_event_call *__start_ftrace_events[];
3428 extern struct trace_event_call *__stop_ftrace_events[];
3430 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3432 static __init int setup_trace_event(char *str)
3434 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3435 ring_buffer_expanded = true;
3436 disable_tracing_selftest("running event tracing");
3440 __setup("trace_event=", setup_trace_event);
3442 /* Expects to have event_mutex held when called */
3444 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3446 struct dentry *d_events;
3447 struct dentry *entry;
3449 entry = tracefs_create_file("set_event", TRACE_MODE_WRITE, parent,
3450 tr, &ftrace_set_event_fops);
3452 pr_warn("Could not create tracefs 'set_event' entry\n");
3456 d_events = tracefs_create_dir("events", parent);
3458 pr_warn("Could not create tracefs 'events' directory\n");
3462 entry = trace_create_file("enable", TRACE_MODE_WRITE, d_events,
3463 tr, &ftrace_tr_enable_fops);
3467 /* There are not as crucial, just warn if they are not created */
3469 entry = tracefs_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3470 tr, &ftrace_set_event_pid_fops);
3472 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3474 entry = tracefs_create_file("set_event_notrace_pid",
3475 TRACE_MODE_WRITE, parent, tr,
3476 &ftrace_set_event_notrace_pid_fops);
3478 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3480 /* ring buffer internal formats */
3481 trace_create_file("header_page", TRACE_MODE_READ, d_events,
3482 ring_buffer_print_page_header,
3483 &ftrace_show_header_fops);
3485 trace_create_file("header_event", TRACE_MODE_READ, d_events,
3486 ring_buffer_print_entry_header,
3487 &ftrace_show_header_fops);
3489 tr->event_dir = d_events;
3495 * event_trace_add_tracer - add a instance of a trace_array to events
3496 * @parent: The parent dentry to place the files/directories for events in
3497 * @tr: The trace array associated with these events
3499 * When a new instance is created, it needs to set up its events
3500 * directory, as well as other files associated with events. It also
3501 * creates the event hierarchy in the @parent/events directory.
3503 * Returns 0 on success.
3505 * Must be called with event_mutex held.
3507 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3511 lockdep_assert_held(&event_mutex);
3513 ret = create_event_toplevel_files(parent, tr);
3517 down_write(&trace_event_sem);
3518 /* If tr already has the event list, it is initialized in early boot. */
3519 if (unlikely(!list_empty(&tr->events)))
3520 __trace_early_add_event_dirs(tr);
3522 __trace_add_event_dirs(tr);
3523 up_write(&trace_event_sem);
3530 * The top trace array already had its file descriptors created.
3531 * Now the files themselves need to be created.
3534 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3538 mutex_lock(&event_mutex);
3540 ret = create_event_toplevel_files(parent, tr);
3544 down_write(&trace_event_sem);
3545 __trace_early_add_event_dirs(tr);
3546 up_write(&trace_event_sem);
3549 mutex_unlock(&event_mutex);
3554 /* Must be called with event_mutex held */
3555 int event_trace_del_tracer(struct trace_array *tr)
3557 lockdep_assert_held(&event_mutex);
3559 /* Disable any event triggers and associated soft-disabled events */
3560 clear_event_triggers(tr);
3562 /* Clear the pid list */
3563 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3565 /* Disable any running events */
3566 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3568 /* Make sure no more events are being executed */
3569 tracepoint_synchronize_unregister();
3571 down_write(&trace_event_sem);
3572 __trace_remove_event_dirs(tr);
3573 tracefs_remove(tr->event_dir);
3574 up_write(&trace_event_sem);
3576 tr->event_dir = NULL;
3581 static __init int event_trace_memsetup(void)
3583 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3584 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3589 early_enable_events(struct trace_array *tr, bool disable_first)
3591 char *buf = bootup_event_buf;
3596 token = strsep(&buf, ",");
3602 /* Restarting syscalls requires that we stop them first */
3604 ftrace_set_clr_event(tr, token, 0);
3606 ret = ftrace_set_clr_event(tr, token, 1);
3608 pr_warn("Failed to enable trace event: %s\n", token);
3611 /* Put back the comma to allow this to be called again */
3617 static __init int event_trace_enable(void)
3619 struct trace_array *tr = top_trace_array();
3620 struct trace_event_call **iter, *call;
3626 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3629 ret = event_init(call);
3631 list_add(&call->list, &ftrace_events);
3635 * We need the top trace array to have a working set of trace
3636 * points at early init, before the debug files and directories
3637 * are created. Create the file entries now, and attach them
3638 * to the actual file dentries later.
3640 __trace_early_add_events(tr);
3642 early_enable_events(tr, false);
3644 trace_printk_start_comm();
3646 register_event_cmds();
3648 register_trigger_cmds();
3654 * event_trace_enable() is called from trace_event_init() first to
3655 * initialize events and perhaps start any events that are on the
3656 * command line. Unfortunately, there are some events that will not
3657 * start this early, like the system call tracepoints that need
3658 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3659 * event_trace_enable() is called before pid 1 starts, and this flag
3660 * is never set, making the syscall tracepoint never get reached, but
3661 * the event is enabled regardless (and not doing anything).
3663 static __init int event_trace_enable_again(void)
3665 struct trace_array *tr;
3667 tr = top_trace_array();
3671 early_enable_events(tr, true);
3676 early_initcall(event_trace_enable_again);
3678 /* Init fields which doesn't related to the tracefs */
3679 static __init int event_trace_init_fields(void)
3681 if (trace_define_generic_fields())
3682 pr_warn("tracing: Failed to allocated generic fields");
3684 if (trace_define_common_fields())
3685 pr_warn("tracing: Failed to allocate common fields");
3690 __init int event_trace_init(void)
3692 struct trace_array *tr;
3693 struct dentry *entry;
3696 tr = top_trace_array();
3700 entry = tracefs_create_file("available_events", TRACE_MODE_READ,
3701 NULL, tr, &ftrace_avail_fops);
3703 pr_warn("Could not create tracefs 'available_events' entry\n");
3705 ret = early_event_add_tracer(NULL, tr);
3709 #ifdef CONFIG_MODULES
3710 ret = register_module_notifier(&trace_module_nb);
3712 pr_warn("Failed to register trace events module notifier\n");
3715 eventdir_initialized = true;
3720 void __init trace_event_init(void)
3722 event_trace_memsetup();
3723 init_ftrace_syscalls();
3724 event_trace_enable();
3725 event_trace_init_fields();
3728 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3730 static DEFINE_SPINLOCK(test_spinlock);
3731 static DEFINE_SPINLOCK(test_spinlock_irq);
3732 static DEFINE_MUTEX(test_mutex);
3734 static __init void test_work(struct work_struct *dummy)
3736 spin_lock(&test_spinlock);
3737 spin_lock_irq(&test_spinlock_irq);
3739 spin_unlock_irq(&test_spinlock_irq);
3740 spin_unlock(&test_spinlock);
3742 mutex_lock(&test_mutex);
3744 mutex_unlock(&test_mutex);
3747 static __init int event_test_thread(void *unused)
3751 test_malloc = kmalloc(1234, GFP_KERNEL);
3753 pr_info("failed to kmalloc\n");
3755 schedule_on_each_cpu(test_work);
3759 set_current_state(TASK_INTERRUPTIBLE);
3760 while (!kthread_should_stop()) {
3762 set_current_state(TASK_INTERRUPTIBLE);
3764 __set_current_state(TASK_RUNNING);
3770 * Do various things that may trigger events.
3772 static __init void event_test_stuff(void)
3774 struct task_struct *test_thread;
3776 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3778 kthread_stop(test_thread);
3782 * For every trace event defined, we will test each trace point separately,
3783 * and then by groups, and finally all trace points.
3785 static __init void event_trace_self_tests(void)
3787 struct trace_subsystem_dir *dir;
3788 struct trace_event_file *file;
3789 struct trace_event_call *call;
3790 struct event_subsystem *system;
3791 struct trace_array *tr;
3794 tr = top_trace_array();
3798 pr_info("Running tests on trace events:\n");
3800 list_for_each_entry(file, &tr->events, list) {
3802 call = file->event_call;
3804 /* Only test those that have a probe */
3805 if (!call->class || !call->class->probe)
3809 * Testing syscall events here is pretty useless, but
3810 * we still do it if configured. But this is time consuming.
3811 * What we really need is a user thread to perform the
3812 * syscalls as we test.
3814 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3815 if (call->class->system &&
3816 strcmp(call->class->system, "syscalls") == 0)
3820 pr_info("Testing event %s: ", trace_event_name(call));
3823 * If an event is already enabled, someone is using
3824 * it and the self test should not be on.
3826 if (file->flags & EVENT_FILE_FL_ENABLED) {
3827 pr_warn("Enabled event during self test!\n");
3832 ftrace_event_enable_disable(file, 1);
3834 ftrace_event_enable_disable(file, 0);
3839 /* Now test at the sub system level */
3841 pr_info("Running tests on trace event systems:\n");
3843 list_for_each_entry(dir, &tr->systems, list) {
3845 system = dir->subsystem;
3847 /* the ftrace system is special, skip it */
3848 if (strcmp(system->name, "ftrace") == 0)
3851 pr_info("Testing event system %s: ", system->name);
3853 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3854 if (WARN_ON_ONCE(ret)) {
3855 pr_warn("error enabling system %s\n",
3862 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3863 if (WARN_ON_ONCE(ret)) {
3864 pr_warn("error disabling system %s\n",
3872 /* Test with all events enabled */
3874 pr_info("Running tests on all trace events:\n");
3875 pr_info("Testing all events: ");
3877 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3878 if (WARN_ON_ONCE(ret)) {
3879 pr_warn("error enabling all events\n");
3886 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3887 if (WARN_ON_ONCE(ret)) {
3888 pr_warn("error disabling all events\n");
3895 #ifdef CONFIG_FUNCTION_TRACER
3897 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3899 static struct trace_event_file event_trace_file __initdata;
3902 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3903 struct ftrace_ops *op, struct ftrace_regs *regs)
3905 struct trace_buffer *buffer;
3906 struct ring_buffer_event *event;
3907 struct ftrace_entry *entry;
3908 unsigned int trace_ctx;
3912 trace_ctx = tracing_gen_ctx();
3913 preempt_disable_notrace();
3914 cpu = raw_smp_processor_id();
3915 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3920 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3921 TRACE_FN, sizeof(*entry),
3925 entry = ring_buffer_event_data(event);
3927 entry->parent_ip = parent_ip;
3929 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3932 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3933 preempt_enable_notrace();
3936 static struct ftrace_ops trace_ops __initdata =
3938 .func = function_test_events_call,
3941 static __init void event_trace_self_test_with_function(void)
3945 event_trace_file.tr = top_trace_array();
3946 if (WARN_ON(!event_trace_file.tr))
3949 ret = register_ftrace_function(&trace_ops);
3950 if (WARN_ON(ret < 0)) {
3951 pr_info("Failed to enable function tracer for event tests\n");
3954 pr_info("Running tests again, along with the function tracer\n");
3955 event_trace_self_tests();
3956 unregister_ftrace_function(&trace_ops);
3959 static __init void event_trace_self_test_with_function(void)
3964 static __init int event_trace_self_tests_init(void)
3966 if (!tracing_selftest_disabled) {
3967 event_trace_self_tests();
3968 event_trace_self_test_with_function();
3974 late_initcall(event_trace_self_tests_init);