Merge tag 'pci-v5.9-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
[sfrench/cifs-2.6.git] / kernel / trace / trace_events.c
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
42 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
43
44 static struct kmem_cache *field_cachep;
45 static struct kmem_cache *file_cachep;
46
47 static inline int system_refcount(struct event_subsystem *system)
48 {
49         return system->ref_count;
50 }
51
52 static int system_refcount_inc(struct event_subsystem *system)
53 {
54         return system->ref_count++;
55 }
56
57 static int system_refcount_dec(struct event_subsystem *system)
58 {
59         return --system->ref_count;
60 }
61
62 /* Double loops, do not use break, only goto's work */
63 #define do_for_each_event_file(tr, file)                        \
64         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
65                 list_for_each_entry(file, &tr->events, list)
66
67 #define do_for_each_event_file_safe(tr, file)                   \
68         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
69                 struct trace_event_file *___n;                          \
70                 list_for_each_entry_safe(file, ___n, &tr->events, list)
71
72 #define while_for_each_event_file()             \
73         }
74
75 static struct ftrace_event_field *
76 __find_event_field(struct list_head *head, char *name)
77 {
78         struct ftrace_event_field *field;
79
80         list_for_each_entry(field, head, link) {
81                 if (!strcmp(field->name, name))
82                         return field;
83         }
84
85         return NULL;
86 }
87
88 struct ftrace_event_field *
89 trace_find_event_field(struct trace_event_call *call, char *name)
90 {
91         struct ftrace_event_field *field;
92         struct list_head *head;
93
94         head = trace_get_fields(call);
95         field = __find_event_field(head, name);
96         if (field)
97                 return field;
98
99         field = __find_event_field(&ftrace_generic_fields, name);
100         if (field)
101                 return field;
102
103         return __find_event_field(&ftrace_common_fields, name);
104 }
105
106 static int __trace_define_field(struct list_head *head, const char *type,
107                                 const char *name, int offset, int size,
108                                 int is_signed, int filter_type)
109 {
110         struct ftrace_event_field *field;
111
112         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
113         if (!field)
114                 return -ENOMEM;
115
116         field->name = name;
117         field->type = type;
118
119         if (filter_type == FILTER_OTHER)
120                 field->filter_type = filter_assign_type(type);
121         else
122                 field->filter_type = filter_type;
123
124         field->offset = offset;
125         field->size = size;
126         field->is_signed = is_signed;
127
128         list_add(&field->link, head);
129
130         return 0;
131 }
132
133 int trace_define_field(struct trace_event_call *call, const char *type,
134                        const char *name, int offset, int size, int is_signed,
135                        int filter_type)
136 {
137         struct list_head *head;
138
139         if (WARN_ON(!call->class))
140                 return 0;
141
142         head = trace_get_fields(call);
143         return __trace_define_field(head, type, name, offset, size,
144                                     is_signed, filter_type);
145 }
146 EXPORT_SYMBOL_GPL(trace_define_field);
147
148 #define __generic_field(type, item, filter_type)                        \
149         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
150                                    #item, 0, 0, is_signed_type(type),   \
151                                    filter_type);                        \
152         if (ret)                                                        \
153                 return ret;
154
155 #define __common_field(type, item)                                      \
156         ret = __trace_define_field(&ftrace_common_fields, #type,        \
157                                    "common_" #item,                     \
158                                    offsetof(typeof(ent), item),         \
159                                    sizeof(ent.item),                    \
160                                    is_signed_type(type), FILTER_OTHER); \
161         if (ret)                                                        \
162                 return ret;
163
164 static int trace_define_generic_fields(void)
165 {
166         int ret;
167
168         __generic_field(int, CPU, FILTER_CPU);
169         __generic_field(int, cpu, FILTER_CPU);
170         __generic_field(char *, COMM, FILTER_COMM);
171         __generic_field(char *, comm, FILTER_COMM);
172
173         return ret;
174 }
175
176 static int trace_define_common_fields(void)
177 {
178         int ret;
179         struct trace_entry ent;
180
181         __common_field(unsigned short, type);
182         __common_field(unsigned char, flags);
183         __common_field(unsigned char, preempt_count);
184         __common_field(int, pid);
185
186         return ret;
187 }
188
189 static void trace_destroy_fields(struct trace_event_call *call)
190 {
191         struct ftrace_event_field *field, *next;
192         struct list_head *head;
193
194         head = trace_get_fields(call);
195         list_for_each_entry_safe(field, next, head, link) {
196                 list_del(&field->link);
197                 kmem_cache_free(field_cachep, field);
198         }
199 }
200
201 /*
202  * run-time version of trace_event_get_offsets_<call>() that returns the last
203  * accessible offset of trace fields excluding __dynamic_array bytes
204  */
205 int trace_event_get_offsets(struct trace_event_call *call)
206 {
207         struct ftrace_event_field *tail;
208         struct list_head *head;
209
210         head = trace_get_fields(call);
211         /*
212          * head->next points to the last field with the largest offset,
213          * since it was added last by trace_define_field()
214          */
215         tail = list_first_entry(head, struct ftrace_event_field, link);
216         return tail->offset + tail->size;
217 }
218
219 int trace_event_raw_init(struct trace_event_call *call)
220 {
221         int id;
222
223         id = register_trace_event(&call->event);
224         if (!id)
225                 return -ENODEV;
226
227         return 0;
228 }
229 EXPORT_SYMBOL_GPL(trace_event_raw_init);
230
231 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
232 {
233         struct trace_array *tr = trace_file->tr;
234         struct trace_array_cpu *data;
235         struct trace_pid_list *no_pid_list;
236         struct trace_pid_list *pid_list;
237
238         pid_list = rcu_dereference_raw(tr->filtered_pids);
239         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
240
241         if (!pid_list && !no_pid_list)
242                 return false;
243
244         data = this_cpu_ptr(tr->array_buffer.data);
245
246         return data->ignore_pid;
247 }
248 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
249
250 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
251                                  struct trace_event_file *trace_file,
252                                  unsigned long len)
253 {
254         struct trace_event_call *event_call = trace_file->event_call;
255
256         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
257             trace_event_ignore_this_pid(trace_file))
258                 return NULL;
259
260         local_save_flags(fbuffer->flags);
261         fbuffer->pc = preempt_count();
262         /*
263          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
264          * preemption (adding one to the preempt_count). Since we are
265          * interested in the preempt_count at the time the tracepoint was
266          * hit, we need to subtract one to offset the increment.
267          */
268         if (IS_ENABLED(CONFIG_PREEMPTION))
269                 fbuffer->pc--;
270         fbuffer->trace_file = trace_file;
271
272         fbuffer->event =
273                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
274                                                 event_call->event.type, len,
275                                                 fbuffer->flags, fbuffer->pc);
276         if (!fbuffer->event)
277                 return NULL;
278
279         fbuffer->regs = NULL;
280         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
281         return fbuffer->entry;
282 }
283 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
284
285 int trace_event_reg(struct trace_event_call *call,
286                     enum trace_reg type, void *data)
287 {
288         struct trace_event_file *file = data;
289
290         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
291         switch (type) {
292         case TRACE_REG_REGISTER:
293                 return tracepoint_probe_register(call->tp,
294                                                  call->class->probe,
295                                                  file);
296         case TRACE_REG_UNREGISTER:
297                 tracepoint_probe_unregister(call->tp,
298                                             call->class->probe,
299                                             file);
300                 return 0;
301
302 #ifdef CONFIG_PERF_EVENTS
303         case TRACE_REG_PERF_REGISTER:
304                 return tracepoint_probe_register(call->tp,
305                                                  call->class->perf_probe,
306                                                  call);
307         case TRACE_REG_PERF_UNREGISTER:
308                 tracepoint_probe_unregister(call->tp,
309                                             call->class->perf_probe,
310                                             call);
311                 return 0;
312         case TRACE_REG_PERF_OPEN:
313         case TRACE_REG_PERF_CLOSE:
314         case TRACE_REG_PERF_ADD:
315         case TRACE_REG_PERF_DEL:
316                 return 0;
317 #endif
318         }
319         return 0;
320 }
321 EXPORT_SYMBOL_GPL(trace_event_reg);
322
323 void trace_event_enable_cmd_record(bool enable)
324 {
325         struct trace_event_file *file;
326         struct trace_array *tr;
327
328         lockdep_assert_held(&event_mutex);
329
330         do_for_each_event_file(tr, file) {
331
332                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
333                         continue;
334
335                 if (enable) {
336                         tracing_start_cmdline_record();
337                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
338                 } else {
339                         tracing_stop_cmdline_record();
340                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
341                 }
342         } while_for_each_event_file();
343 }
344
345 void trace_event_enable_tgid_record(bool enable)
346 {
347         struct trace_event_file *file;
348         struct trace_array *tr;
349
350         lockdep_assert_held(&event_mutex);
351
352         do_for_each_event_file(tr, file) {
353                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
354                         continue;
355
356                 if (enable) {
357                         tracing_start_tgid_record();
358                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
359                 } else {
360                         tracing_stop_tgid_record();
361                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
362                                   &file->flags);
363                 }
364         } while_for_each_event_file();
365 }
366
367 static int __ftrace_event_enable_disable(struct trace_event_file *file,
368                                          int enable, int soft_disable)
369 {
370         struct trace_event_call *call = file->event_call;
371         struct trace_array *tr = file->tr;
372         unsigned long file_flags = file->flags;
373         int ret = 0;
374         int disable;
375
376         switch (enable) {
377         case 0:
378                 /*
379                  * When soft_disable is set and enable is cleared, the sm_ref
380                  * reference counter is decremented. If it reaches 0, we want
381                  * to clear the SOFT_DISABLED flag but leave the event in the
382                  * state that it was. That is, if the event was enabled and
383                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
384                  * is set we do not want the event to be enabled before we
385                  * clear the bit.
386                  *
387                  * When soft_disable is not set but the SOFT_MODE flag is,
388                  * we do nothing. Do not disable the tracepoint, otherwise
389                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
390                  */
391                 if (soft_disable) {
392                         if (atomic_dec_return(&file->sm_ref) > 0)
393                                 break;
394                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
395                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
396                 } else
397                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
398
399                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
400                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
401                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
402                                 tracing_stop_cmdline_record();
403                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
404                         }
405
406                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
407                                 tracing_stop_tgid_record();
408                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
409                         }
410
411                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
412                 }
413                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
414                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
415                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
416                 else
417                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
418                 break;
419         case 1:
420                 /*
421                  * When soft_disable is set and enable is set, we want to
422                  * register the tracepoint for the event, but leave the event
423                  * as is. That means, if the event was already enabled, we do
424                  * nothing (but set SOFT_MODE). If the event is disabled, we
425                  * set SOFT_DISABLED before enabling the event tracepoint, so
426                  * it still seems to be disabled.
427                  */
428                 if (!soft_disable)
429                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
430                 else {
431                         if (atomic_inc_return(&file->sm_ref) > 1)
432                                 break;
433                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
434                 }
435
436                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
437                         bool cmd = false, tgid = false;
438
439                         /* Keep the event disabled, when going to SOFT_MODE. */
440                         if (soft_disable)
441                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
442
443                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
444                                 cmd = true;
445                                 tracing_start_cmdline_record();
446                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
447                         }
448
449                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
450                                 tgid = true;
451                                 tracing_start_tgid_record();
452                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
453                         }
454
455                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
456                         if (ret) {
457                                 if (cmd)
458                                         tracing_stop_cmdline_record();
459                                 if (tgid)
460                                         tracing_stop_tgid_record();
461                                 pr_info("event trace: Could not enable event "
462                                         "%s\n", trace_event_name(call));
463                                 break;
464                         }
465                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
466
467                         /* WAS_ENABLED gets set but never cleared. */
468                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
469                 }
470                 break;
471         }
472
473         /* Enable or disable use of trace_buffered_event */
474         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
475             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
476                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
477                         trace_buffered_event_enable();
478                 else
479                         trace_buffered_event_disable();
480         }
481
482         return ret;
483 }
484
485 int trace_event_enable_disable(struct trace_event_file *file,
486                                int enable, int soft_disable)
487 {
488         return __ftrace_event_enable_disable(file, enable, soft_disable);
489 }
490
491 static int ftrace_event_enable_disable(struct trace_event_file *file,
492                                        int enable)
493 {
494         return __ftrace_event_enable_disable(file, enable, 0);
495 }
496
497 static void ftrace_clear_events(struct trace_array *tr)
498 {
499         struct trace_event_file *file;
500
501         mutex_lock(&event_mutex);
502         list_for_each_entry(file, &tr->events, list) {
503                 ftrace_event_enable_disable(file, 0);
504         }
505         mutex_unlock(&event_mutex);
506 }
507
508 static void
509 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
510 {
511         struct trace_pid_list *pid_list;
512         struct trace_array *tr = data;
513
514         pid_list = rcu_dereference_raw(tr->filtered_pids);
515         trace_filter_add_remove_task(pid_list, NULL, task);
516
517         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
518         trace_filter_add_remove_task(pid_list, NULL, task);
519 }
520
521 static void
522 event_filter_pid_sched_process_fork(void *data,
523                                     struct task_struct *self,
524                                     struct task_struct *task)
525 {
526         struct trace_pid_list *pid_list;
527         struct trace_array *tr = data;
528
529         pid_list = rcu_dereference_sched(tr->filtered_pids);
530         trace_filter_add_remove_task(pid_list, self, task);
531
532         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
533         trace_filter_add_remove_task(pid_list, self, task);
534 }
535
536 void trace_event_follow_fork(struct trace_array *tr, bool enable)
537 {
538         if (enable) {
539                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
540                                                        tr, INT_MIN);
541                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
542                                                        tr, INT_MAX);
543         } else {
544                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
545                                                     tr);
546                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
547                                                     tr);
548         }
549 }
550
551 static void
552 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
553                     struct task_struct *prev, struct task_struct *next)
554 {
555         struct trace_array *tr = data;
556         struct trace_pid_list *no_pid_list;
557         struct trace_pid_list *pid_list;
558         bool ret;
559
560         pid_list = rcu_dereference_sched(tr->filtered_pids);
561         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
562
563         /*
564          * Sched switch is funny, as we only want to ignore it
565          * in the notrace case if both prev and next should be ignored.
566          */
567         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
568                 trace_ignore_this_task(NULL, no_pid_list, next);
569
570         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
571                        (trace_ignore_this_task(pid_list, NULL, prev) &&
572                         trace_ignore_this_task(pid_list, NULL, next)));
573 }
574
575 static void
576 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
577                     struct task_struct *prev, struct task_struct *next)
578 {
579         struct trace_array *tr = data;
580         struct trace_pid_list *no_pid_list;
581         struct trace_pid_list *pid_list;
582
583         pid_list = rcu_dereference_sched(tr->filtered_pids);
584         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
585
586         this_cpu_write(tr->array_buffer.data->ignore_pid,
587                        trace_ignore_this_task(pid_list, no_pid_list, next));
588 }
589
590 static void
591 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
592 {
593         struct trace_array *tr = data;
594         struct trace_pid_list *no_pid_list;
595         struct trace_pid_list *pid_list;
596
597         /* Nothing to do if we are already tracing */
598         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
599                 return;
600
601         pid_list = rcu_dereference_sched(tr->filtered_pids);
602         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
603
604         this_cpu_write(tr->array_buffer.data->ignore_pid,
605                        trace_ignore_this_task(pid_list, no_pid_list, task));
606 }
607
608 static void
609 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
610 {
611         struct trace_array *tr = data;
612         struct trace_pid_list *no_pid_list;
613         struct trace_pid_list *pid_list;
614
615         /* Nothing to do if we are not tracing */
616         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
617                 return;
618
619         pid_list = rcu_dereference_sched(tr->filtered_pids);
620         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
621
622         /* Set tracing if current is enabled */
623         this_cpu_write(tr->array_buffer.data->ignore_pid,
624                        trace_ignore_this_task(pid_list, no_pid_list, current));
625 }
626
627 static void unregister_pid_events(struct trace_array *tr)
628 {
629         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
630         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
631
632         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
633         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
634
635         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
636         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
637
638         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
639         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
640 }
641
642 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
643 {
644         struct trace_pid_list *pid_list;
645         struct trace_pid_list *no_pid_list;
646         struct trace_event_file *file;
647         int cpu;
648
649         pid_list = rcu_dereference_protected(tr->filtered_pids,
650                                              lockdep_is_held(&event_mutex));
651         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
652                                              lockdep_is_held(&event_mutex));
653
654         /* Make sure there's something to do */
655         if (!pid_type_enabled(type, pid_list, no_pid_list))
656                 return;
657
658         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
659                 unregister_pid_events(tr);
660
661                 list_for_each_entry(file, &tr->events, list) {
662                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
663                 }
664
665                 for_each_possible_cpu(cpu)
666                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
667         }
668
669         if (type & TRACE_PIDS)
670                 rcu_assign_pointer(tr->filtered_pids, NULL);
671
672         if (type & TRACE_NO_PIDS)
673                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
674
675         /* Wait till all users are no longer using pid filtering */
676         tracepoint_synchronize_unregister();
677
678         if ((type & TRACE_PIDS) && pid_list)
679                 trace_free_pid_list(pid_list);
680
681         if ((type & TRACE_NO_PIDS) && no_pid_list)
682                 trace_free_pid_list(no_pid_list);
683 }
684
685 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
686 {
687         mutex_lock(&event_mutex);
688         __ftrace_clear_event_pids(tr, type);
689         mutex_unlock(&event_mutex);
690 }
691
692 static void __put_system(struct event_subsystem *system)
693 {
694         struct event_filter *filter = system->filter;
695
696         WARN_ON_ONCE(system_refcount(system) == 0);
697         if (system_refcount_dec(system))
698                 return;
699
700         list_del(&system->list);
701
702         if (filter) {
703                 kfree(filter->filter_string);
704                 kfree(filter);
705         }
706         kfree_const(system->name);
707         kfree(system);
708 }
709
710 static void __get_system(struct event_subsystem *system)
711 {
712         WARN_ON_ONCE(system_refcount(system) == 0);
713         system_refcount_inc(system);
714 }
715
716 static void __get_system_dir(struct trace_subsystem_dir *dir)
717 {
718         WARN_ON_ONCE(dir->ref_count == 0);
719         dir->ref_count++;
720         __get_system(dir->subsystem);
721 }
722
723 static void __put_system_dir(struct trace_subsystem_dir *dir)
724 {
725         WARN_ON_ONCE(dir->ref_count == 0);
726         /* If the subsystem is about to be freed, the dir must be too */
727         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
728
729         __put_system(dir->subsystem);
730         if (!--dir->ref_count)
731                 kfree(dir);
732 }
733
734 static void put_system(struct trace_subsystem_dir *dir)
735 {
736         mutex_lock(&event_mutex);
737         __put_system_dir(dir);
738         mutex_unlock(&event_mutex);
739 }
740
741 static void remove_subsystem(struct trace_subsystem_dir *dir)
742 {
743         if (!dir)
744                 return;
745
746         if (!--dir->nr_events) {
747                 tracefs_remove(dir->entry);
748                 list_del(&dir->list);
749                 __put_system_dir(dir);
750         }
751 }
752
753 static void remove_event_file_dir(struct trace_event_file *file)
754 {
755         struct dentry *dir = file->dir;
756         struct dentry *child;
757
758         if (dir) {
759                 spin_lock(&dir->d_lock);        /* probably unneeded */
760                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
761                         if (d_really_is_positive(child))        /* probably unneeded */
762                                 d_inode(child)->i_private = NULL;
763                 }
764                 spin_unlock(&dir->d_lock);
765
766                 tracefs_remove(dir);
767         }
768
769         list_del(&file->list);
770         remove_subsystem(file->system);
771         free_event_filter(file->filter);
772         kmem_cache_free(file_cachep, file);
773 }
774
775 /*
776  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
777  */
778 static int
779 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
780                               const char *sub, const char *event, int set)
781 {
782         struct trace_event_file *file;
783         struct trace_event_call *call;
784         const char *name;
785         int ret = -EINVAL;
786         int eret = 0;
787
788         list_for_each_entry(file, &tr->events, list) {
789
790                 call = file->event_call;
791                 name = trace_event_name(call);
792
793                 if (!name || !call->class || !call->class->reg)
794                         continue;
795
796                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
797                         continue;
798
799                 if (match &&
800                     strcmp(match, name) != 0 &&
801                     strcmp(match, call->class->system) != 0)
802                         continue;
803
804                 if (sub && strcmp(sub, call->class->system) != 0)
805                         continue;
806
807                 if (event && strcmp(event, name) != 0)
808                         continue;
809
810                 ret = ftrace_event_enable_disable(file, set);
811
812                 /*
813                  * Save the first error and return that. Some events
814                  * may still have been enabled, but let the user
815                  * know that something went wrong.
816                  */
817                 if (ret && !eret)
818                         eret = ret;
819
820                 ret = eret;
821         }
822
823         return ret;
824 }
825
826 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
827                                   const char *sub, const char *event, int set)
828 {
829         int ret;
830
831         mutex_lock(&event_mutex);
832         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
833         mutex_unlock(&event_mutex);
834
835         return ret;
836 }
837
838 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
839 {
840         char *event = NULL, *sub = NULL, *match;
841         int ret;
842
843         if (!tr)
844                 return -ENOENT;
845         /*
846          * The buf format can be <subsystem>:<event-name>
847          *  *:<event-name> means any event by that name.
848          *  :<event-name> is the same.
849          *
850          *  <subsystem>:* means all events in that subsystem
851          *  <subsystem>: means the same.
852          *
853          *  <name> (no ':') means all events in a subsystem with
854          *  the name <name> or any event that matches <name>
855          */
856
857         match = strsep(&buf, ":");
858         if (buf) {
859                 sub = match;
860                 event = buf;
861                 match = NULL;
862
863                 if (!strlen(sub) || strcmp(sub, "*") == 0)
864                         sub = NULL;
865                 if (!strlen(event) || strcmp(event, "*") == 0)
866                         event = NULL;
867         }
868
869         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
870
871         /* Put back the colon to allow this to be called again */
872         if (buf)
873                 *(buf - 1) = ':';
874
875         return ret;
876 }
877
878 /**
879  * trace_set_clr_event - enable or disable an event
880  * @system: system name to match (NULL for any system)
881  * @event: event name to match (NULL for all events, within system)
882  * @set: 1 to enable, 0 to disable
883  *
884  * This is a way for other parts of the kernel to enable or disable
885  * event recording.
886  *
887  * Returns 0 on success, -EINVAL if the parameters do not match any
888  * registered events.
889  */
890 int trace_set_clr_event(const char *system, const char *event, int set)
891 {
892         struct trace_array *tr = top_trace_array();
893
894         if (!tr)
895                 return -ENODEV;
896
897         return __ftrace_set_clr_event(tr, NULL, system, event, set);
898 }
899 EXPORT_SYMBOL_GPL(trace_set_clr_event);
900
901 /**
902  * trace_array_set_clr_event - enable or disable an event for a trace array.
903  * @tr: concerned trace array.
904  * @system: system name to match (NULL for any system)
905  * @event: event name to match (NULL for all events, within system)
906  * @enable: true to enable, false to disable
907  *
908  * This is a way for other parts of the kernel to enable or disable
909  * event recording.
910  *
911  * Returns 0 on success, -EINVAL if the parameters do not match any
912  * registered events.
913  */
914 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
915                 const char *event, bool enable)
916 {
917         int set;
918
919         if (!tr)
920                 return -ENOENT;
921
922         set = (enable == true) ? 1 : 0;
923         return __ftrace_set_clr_event(tr, NULL, system, event, set);
924 }
925 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
926
927 /* 128 should be much more than enough */
928 #define EVENT_BUF_SIZE          127
929
930 static ssize_t
931 ftrace_event_write(struct file *file, const char __user *ubuf,
932                    size_t cnt, loff_t *ppos)
933 {
934         struct trace_parser parser;
935         struct seq_file *m = file->private_data;
936         struct trace_array *tr = m->private;
937         ssize_t read, ret;
938
939         if (!cnt)
940                 return 0;
941
942         ret = tracing_update_buffers();
943         if (ret < 0)
944                 return ret;
945
946         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
947                 return -ENOMEM;
948
949         read = trace_get_user(&parser, ubuf, cnt, ppos);
950
951         if (read >= 0 && trace_parser_loaded((&parser))) {
952                 int set = 1;
953
954                 if (*parser.buffer == '!')
955                         set = 0;
956
957                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
958                 if (ret)
959                         goto out_put;
960         }
961
962         ret = read;
963
964  out_put:
965         trace_parser_put(&parser);
966
967         return ret;
968 }
969
970 static void *
971 t_next(struct seq_file *m, void *v, loff_t *pos)
972 {
973         struct trace_event_file *file = v;
974         struct trace_event_call *call;
975         struct trace_array *tr = m->private;
976
977         (*pos)++;
978
979         list_for_each_entry_continue(file, &tr->events, list) {
980                 call = file->event_call;
981                 /*
982                  * The ftrace subsystem is for showing formats only.
983                  * They can not be enabled or disabled via the event files.
984                  */
985                 if (call->class && call->class->reg &&
986                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
987                         return file;
988         }
989
990         return NULL;
991 }
992
993 static void *t_start(struct seq_file *m, loff_t *pos)
994 {
995         struct trace_event_file *file;
996         struct trace_array *tr = m->private;
997         loff_t l;
998
999         mutex_lock(&event_mutex);
1000
1001         file = list_entry(&tr->events, struct trace_event_file, list);
1002         for (l = 0; l <= *pos; ) {
1003                 file = t_next(m, file, &l);
1004                 if (!file)
1005                         break;
1006         }
1007         return file;
1008 }
1009
1010 static void *
1011 s_next(struct seq_file *m, void *v, loff_t *pos)
1012 {
1013         struct trace_event_file *file = v;
1014         struct trace_array *tr = m->private;
1015
1016         (*pos)++;
1017
1018         list_for_each_entry_continue(file, &tr->events, list) {
1019                 if (file->flags & EVENT_FILE_FL_ENABLED)
1020                         return file;
1021         }
1022
1023         return NULL;
1024 }
1025
1026 static void *s_start(struct seq_file *m, loff_t *pos)
1027 {
1028         struct trace_event_file *file;
1029         struct trace_array *tr = m->private;
1030         loff_t l;
1031
1032         mutex_lock(&event_mutex);
1033
1034         file = list_entry(&tr->events, struct trace_event_file, list);
1035         for (l = 0; l <= *pos; ) {
1036                 file = s_next(m, file, &l);
1037                 if (!file)
1038                         break;
1039         }
1040         return file;
1041 }
1042
1043 static int t_show(struct seq_file *m, void *v)
1044 {
1045         struct trace_event_file *file = v;
1046         struct trace_event_call *call = file->event_call;
1047
1048         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1049                 seq_printf(m, "%s:", call->class->system);
1050         seq_printf(m, "%s\n", trace_event_name(call));
1051
1052         return 0;
1053 }
1054
1055 static void t_stop(struct seq_file *m, void *p)
1056 {
1057         mutex_unlock(&event_mutex);
1058 }
1059
1060 static void *
1061 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1062 {
1063         struct trace_array *tr = m->private;
1064         struct trace_pid_list *pid_list;
1065
1066         if (type == TRACE_PIDS)
1067                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1068         else
1069                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1070
1071         return trace_pid_next(pid_list, v, pos);
1072 }
1073
1074 static void *
1075 p_next(struct seq_file *m, void *v, loff_t *pos)
1076 {
1077         return __next(m, v, pos, TRACE_PIDS);
1078 }
1079
1080 static void *
1081 np_next(struct seq_file *m, void *v, loff_t *pos)
1082 {
1083         return __next(m, v, pos, TRACE_NO_PIDS);
1084 }
1085
1086 static void *__start(struct seq_file *m, loff_t *pos, int type)
1087         __acquires(RCU)
1088 {
1089         struct trace_pid_list *pid_list;
1090         struct trace_array *tr = m->private;
1091
1092         /*
1093          * Grab the mutex, to keep calls to p_next() having the same
1094          * tr->filtered_pids as p_start() has.
1095          * If we just passed the tr->filtered_pids around, then RCU would
1096          * have been enough, but doing that makes things more complex.
1097          */
1098         mutex_lock(&event_mutex);
1099         rcu_read_lock_sched();
1100
1101         if (type == TRACE_PIDS)
1102                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1103         else
1104                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1105
1106         if (!pid_list)
1107                 return NULL;
1108
1109         return trace_pid_start(pid_list, pos);
1110 }
1111
1112 static void *p_start(struct seq_file *m, loff_t *pos)
1113         __acquires(RCU)
1114 {
1115         return __start(m, pos, TRACE_PIDS);
1116 }
1117
1118 static void *np_start(struct seq_file *m, loff_t *pos)
1119         __acquires(RCU)
1120 {
1121         return __start(m, pos, TRACE_NO_PIDS);
1122 }
1123
1124 static void p_stop(struct seq_file *m, void *p)
1125         __releases(RCU)
1126 {
1127         rcu_read_unlock_sched();
1128         mutex_unlock(&event_mutex);
1129 }
1130
1131 static ssize_t
1132 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1133                   loff_t *ppos)
1134 {
1135         struct trace_event_file *file;
1136         unsigned long flags;
1137         char buf[4] = "0";
1138
1139         mutex_lock(&event_mutex);
1140         file = event_file_data(filp);
1141         if (likely(file))
1142                 flags = file->flags;
1143         mutex_unlock(&event_mutex);
1144
1145         if (!file)
1146                 return -ENODEV;
1147
1148         if (flags & EVENT_FILE_FL_ENABLED &&
1149             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1150                 strcpy(buf, "1");
1151
1152         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1153             flags & EVENT_FILE_FL_SOFT_MODE)
1154                 strcat(buf, "*");
1155
1156         strcat(buf, "\n");
1157
1158         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1159 }
1160
1161 static ssize_t
1162 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1163                    loff_t *ppos)
1164 {
1165         struct trace_event_file *file;
1166         unsigned long val;
1167         int ret;
1168
1169         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1170         if (ret)
1171                 return ret;
1172
1173         ret = tracing_update_buffers();
1174         if (ret < 0)
1175                 return ret;
1176
1177         switch (val) {
1178         case 0:
1179         case 1:
1180                 ret = -ENODEV;
1181                 mutex_lock(&event_mutex);
1182                 file = event_file_data(filp);
1183                 if (likely(file))
1184                         ret = ftrace_event_enable_disable(file, val);
1185                 mutex_unlock(&event_mutex);
1186                 break;
1187
1188         default:
1189                 return -EINVAL;
1190         }
1191
1192         *ppos += cnt;
1193
1194         return ret ? ret : cnt;
1195 }
1196
1197 static ssize_t
1198 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1199                    loff_t *ppos)
1200 {
1201         const char set_to_char[4] = { '?', '0', '1', 'X' };
1202         struct trace_subsystem_dir *dir = filp->private_data;
1203         struct event_subsystem *system = dir->subsystem;
1204         struct trace_event_call *call;
1205         struct trace_event_file *file;
1206         struct trace_array *tr = dir->tr;
1207         char buf[2];
1208         int set = 0;
1209         int ret;
1210
1211         mutex_lock(&event_mutex);
1212         list_for_each_entry(file, &tr->events, list) {
1213                 call = file->event_call;
1214                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1215                         continue;
1216
1217                 if (system && strcmp(call->class->system, system->name) != 0)
1218                         continue;
1219
1220                 /*
1221                  * We need to find out if all the events are set
1222                  * or if all events or cleared, or if we have
1223                  * a mixture.
1224                  */
1225                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1226
1227                 /*
1228                  * If we have a mixture, no need to look further.
1229                  */
1230                 if (set == 3)
1231                         break;
1232         }
1233         mutex_unlock(&event_mutex);
1234
1235         buf[0] = set_to_char[set];
1236         buf[1] = '\n';
1237
1238         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1239
1240         return ret;
1241 }
1242
1243 static ssize_t
1244 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1245                     loff_t *ppos)
1246 {
1247         struct trace_subsystem_dir *dir = filp->private_data;
1248         struct event_subsystem *system = dir->subsystem;
1249         const char *name = NULL;
1250         unsigned long val;
1251         ssize_t ret;
1252
1253         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1254         if (ret)
1255                 return ret;
1256
1257         ret = tracing_update_buffers();
1258         if (ret < 0)
1259                 return ret;
1260
1261         if (val != 0 && val != 1)
1262                 return -EINVAL;
1263
1264         /*
1265          * Opening of "enable" adds a ref count to system,
1266          * so the name is safe to use.
1267          */
1268         if (system)
1269                 name = system->name;
1270
1271         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1272         if (ret)
1273                 goto out;
1274
1275         ret = cnt;
1276
1277 out:
1278         *ppos += cnt;
1279
1280         return ret;
1281 }
1282
1283 enum {
1284         FORMAT_HEADER           = 1,
1285         FORMAT_FIELD_SEPERATOR  = 2,
1286         FORMAT_PRINTFMT         = 3,
1287 };
1288
1289 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1290 {
1291         struct trace_event_call *call = event_file_data(m->private);
1292         struct list_head *common_head = &ftrace_common_fields;
1293         struct list_head *head = trace_get_fields(call);
1294         struct list_head *node = v;
1295
1296         (*pos)++;
1297
1298         switch ((unsigned long)v) {
1299         case FORMAT_HEADER:
1300                 node = common_head;
1301                 break;
1302
1303         case FORMAT_FIELD_SEPERATOR:
1304                 node = head;
1305                 break;
1306
1307         case FORMAT_PRINTFMT:
1308                 /* all done */
1309                 return NULL;
1310         }
1311
1312         node = node->prev;
1313         if (node == common_head)
1314                 return (void *)FORMAT_FIELD_SEPERATOR;
1315         else if (node == head)
1316                 return (void *)FORMAT_PRINTFMT;
1317         else
1318                 return node;
1319 }
1320
1321 static int f_show(struct seq_file *m, void *v)
1322 {
1323         struct trace_event_call *call = event_file_data(m->private);
1324         struct ftrace_event_field *field;
1325         const char *array_descriptor;
1326
1327         switch ((unsigned long)v) {
1328         case FORMAT_HEADER:
1329                 seq_printf(m, "name: %s\n", trace_event_name(call));
1330                 seq_printf(m, "ID: %d\n", call->event.type);
1331                 seq_puts(m, "format:\n");
1332                 return 0;
1333
1334         case FORMAT_FIELD_SEPERATOR:
1335                 seq_putc(m, '\n');
1336                 return 0;
1337
1338         case FORMAT_PRINTFMT:
1339                 seq_printf(m, "\nprint fmt: %s\n",
1340                            call->print_fmt);
1341                 return 0;
1342         }
1343
1344         field = list_entry(v, struct ftrace_event_field, link);
1345         /*
1346          * Smartly shows the array type(except dynamic array).
1347          * Normal:
1348          *      field:TYPE VAR
1349          * If TYPE := TYPE[LEN], it is shown:
1350          *      field:TYPE VAR[LEN]
1351          */
1352         array_descriptor = strchr(field->type, '[');
1353
1354         if (str_has_prefix(field->type, "__data_loc"))
1355                 array_descriptor = NULL;
1356
1357         if (!array_descriptor)
1358                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1359                            field->type, field->name, field->offset,
1360                            field->size, !!field->is_signed);
1361         else
1362                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1363                            (int)(array_descriptor - field->type),
1364                            field->type, field->name,
1365                            array_descriptor, field->offset,
1366                            field->size, !!field->is_signed);
1367
1368         return 0;
1369 }
1370
1371 static void *f_start(struct seq_file *m, loff_t *pos)
1372 {
1373         void *p = (void *)FORMAT_HEADER;
1374         loff_t l = 0;
1375
1376         /* ->stop() is called even if ->start() fails */
1377         mutex_lock(&event_mutex);
1378         if (!event_file_data(m->private))
1379                 return ERR_PTR(-ENODEV);
1380
1381         while (l < *pos && p)
1382                 p = f_next(m, p, &l);
1383
1384         return p;
1385 }
1386
1387 static void f_stop(struct seq_file *m, void *p)
1388 {
1389         mutex_unlock(&event_mutex);
1390 }
1391
1392 static const struct seq_operations trace_format_seq_ops = {
1393         .start          = f_start,
1394         .next           = f_next,
1395         .stop           = f_stop,
1396         .show           = f_show,
1397 };
1398
1399 static int trace_format_open(struct inode *inode, struct file *file)
1400 {
1401         struct seq_file *m;
1402         int ret;
1403
1404         /* Do we want to hide event format files on tracefs lockdown? */
1405
1406         ret = seq_open(file, &trace_format_seq_ops);
1407         if (ret < 0)
1408                 return ret;
1409
1410         m = file->private_data;
1411         m->private = file;
1412
1413         return 0;
1414 }
1415
1416 static ssize_t
1417 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1418 {
1419         int id = (long)event_file_data(filp);
1420         char buf[32];
1421         int len;
1422
1423         if (unlikely(!id))
1424                 return -ENODEV;
1425
1426         len = sprintf(buf, "%d\n", id);
1427
1428         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1429 }
1430
1431 static ssize_t
1432 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1433                   loff_t *ppos)
1434 {
1435         struct trace_event_file *file;
1436         struct trace_seq *s;
1437         int r = -ENODEV;
1438
1439         if (*ppos)
1440                 return 0;
1441
1442         s = kmalloc(sizeof(*s), GFP_KERNEL);
1443
1444         if (!s)
1445                 return -ENOMEM;
1446
1447         trace_seq_init(s);
1448
1449         mutex_lock(&event_mutex);
1450         file = event_file_data(filp);
1451         if (file)
1452                 print_event_filter(file, s);
1453         mutex_unlock(&event_mutex);
1454
1455         if (file)
1456                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1457                                             s->buffer, trace_seq_used(s));
1458
1459         kfree(s);
1460
1461         return r;
1462 }
1463
1464 static ssize_t
1465 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1466                    loff_t *ppos)
1467 {
1468         struct trace_event_file *file;
1469         char *buf;
1470         int err = -ENODEV;
1471
1472         if (cnt >= PAGE_SIZE)
1473                 return -EINVAL;
1474
1475         buf = memdup_user_nul(ubuf, cnt);
1476         if (IS_ERR(buf))
1477                 return PTR_ERR(buf);
1478
1479         mutex_lock(&event_mutex);
1480         file = event_file_data(filp);
1481         if (file)
1482                 err = apply_event_filter(file, buf);
1483         mutex_unlock(&event_mutex);
1484
1485         kfree(buf);
1486         if (err < 0)
1487                 return err;
1488
1489         *ppos += cnt;
1490
1491         return cnt;
1492 }
1493
1494 static LIST_HEAD(event_subsystems);
1495
1496 static int subsystem_open(struct inode *inode, struct file *filp)
1497 {
1498         struct event_subsystem *system = NULL;
1499         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1500         struct trace_array *tr;
1501         int ret;
1502
1503         if (tracing_is_disabled())
1504                 return -ENODEV;
1505
1506         /* Make sure the system still exists */
1507         mutex_lock(&event_mutex);
1508         mutex_lock(&trace_types_lock);
1509         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1510                 list_for_each_entry(dir, &tr->systems, list) {
1511                         if (dir == inode->i_private) {
1512                                 /* Don't open systems with no events */
1513                                 if (dir->nr_events) {
1514                                         __get_system_dir(dir);
1515                                         system = dir->subsystem;
1516                                 }
1517                                 goto exit_loop;
1518                         }
1519                 }
1520         }
1521  exit_loop:
1522         mutex_unlock(&trace_types_lock);
1523         mutex_unlock(&event_mutex);
1524
1525         if (!system)
1526                 return -ENODEV;
1527
1528         /* Some versions of gcc think dir can be uninitialized here */
1529         WARN_ON(!dir);
1530
1531         /* Still need to increment the ref count of the system */
1532         if (trace_array_get(tr) < 0) {
1533                 put_system(dir);
1534                 return -ENODEV;
1535         }
1536
1537         ret = tracing_open_generic(inode, filp);
1538         if (ret < 0) {
1539                 trace_array_put(tr);
1540                 put_system(dir);
1541         }
1542
1543         return ret;
1544 }
1545
1546 static int system_tr_open(struct inode *inode, struct file *filp)
1547 {
1548         struct trace_subsystem_dir *dir;
1549         struct trace_array *tr = inode->i_private;
1550         int ret;
1551
1552         /* Make a temporary dir that has no system but points to tr */
1553         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1554         if (!dir)
1555                 return -ENOMEM;
1556
1557         ret = tracing_open_generic_tr(inode, filp);
1558         if (ret < 0) {
1559                 kfree(dir);
1560                 return ret;
1561         }
1562         dir->tr = tr;
1563         filp->private_data = dir;
1564
1565         return 0;
1566 }
1567
1568 static int subsystem_release(struct inode *inode, struct file *file)
1569 {
1570         struct trace_subsystem_dir *dir = file->private_data;
1571
1572         trace_array_put(dir->tr);
1573
1574         /*
1575          * If dir->subsystem is NULL, then this is a temporary
1576          * descriptor that was made for a trace_array to enable
1577          * all subsystems.
1578          */
1579         if (dir->subsystem)
1580                 put_system(dir);
1581         else
1582                 kfree(dir);
1583
1584         return 0;
1585 }
1586
1587 static ssize_t
1588 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1589                       loff_t *ppos)
1590 {
1591         struct trace_subsystem_dir *dir = filp->private_data;
1592         struct event_subsystem *system = dir->subsystem;
1593         struct trace_seq *s;
1594         int r;
1595
1596         if (*ppos)
1597                 return 0;
1598
1599         s = kmalloc(sizeof(*s), GFP_KERNEL);
1600         if (!s)
1601                 return -ENOMEM;
1602
1603         trace_seq_init(s);
1604
1605         print_subsystem_event_filter(system, s);
1606         r = simple_read_from_buffer(ubuf, cnt, ppos,
1607                                     s->buffer, trace_seq_used(s));
1608
1609         kfree(s);
1610
1611         return r;
1612 }
1613
1614 static ssize_t
1615 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1616                        loff_t *ppos)
1617 {
1618         struct trace_subsystem_dir *dir = filp->private_data;
1619         char *buf;
1620         int err;
1621
1622         if (cnt >= PAGE_SIZE)
1623                 return -EINVAL;
1624
1625         buf = memdup_user_nul(ubuf, cnt);
1626         if (IS_ERR(buf))
1627                 return PTR_ERR(buf);
1628
1629         err = apply_subsystem_event_filter(dir, buf);
1630         kfree(buf);
1631         if (err < 0)
1632                 return err;
1633
1634         *ppos += cnt;
1635
1636         return cnt;
1637 }
1638
1639 static ssize_t
1640 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1641 {
1642         int (*func)(struct trace_seq *s) = filp->private_data;
1643         struct trace_seq *s;
1644         int r;
1645
1646         if (*ppos)
1647                 return 0;
1648
1649         s = kmalloc(sizeof(*s), GFP_KERNEL);
1650         if (!s)
1651                 return -ENOMEM;
1652
1653         trace_seq_init(s);
1654
1655         func(s);
1656         r = simple_read_from_buffer(ubuf, cnt, ppos,
1657                                     s->buffer, trace_seq_used(s));
1658
1659         kfree(s);
1660
1661         return r;
1662 }
1663
1664 static void ignore_task_cpu(void *data)
1665 {
1666         struct trace_array *tr = data;
1667         struct trace_pid_list *pid_list;
1668         struct trace_pid_list *no_pid_list;
1669
1670         /*
1671          * This function is called by on_each_cpu() while the
1672          * event_mutex is held.
1673          */
1674         pid_list = rcu_dereference_protected(tr->filtered_pids,
1675                                              mutex_is_locked(&event_mutex));
1676         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1677                                              mutex_is_locked(&event_mutex));
1678
1679         this_cpu_write(tr->array_buffer.data->ignore_pid,
1680                        trace_ignore_this_task(pid_list, no_pid_list, current));
1681 }
1682
1683 static void register_pid_events(struct trace_array *tr)
1684 {
1685         /*
1686          * Register a probe that is called before all other probes
1687          * to set ignore_pid if next or prev do not match.
1688          * Register a probe this is called after all other probes
1689          * to only keep ignore_pid set if next pid matches.
1690          */
1691         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1692                                          tr, INT_MAX);
1693         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1694                                          tr, 0);
1695
1696         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1697                                          tr, INT_MAX);
1698         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1699                                          tr, 0);
1700
1701         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1702                                              tr, INT_MAX);
1703         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1704                                              tr, 0);
1705
1706         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1707                                          tr, INT_MAX);
1708         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1709                                          tr, 0);
1710 }
1711
1712 static ssize_t
1713 event_pid_write(struct file *filp, const char __user *ubuf,
1714                 size_t cnt, loff_t *ppos, int type)
1715 {
1716         struct seq_file *m = filp->private_data;
1717         struct trace_array *tr = m->private;
1718         struct trace_pid_list *filtered_pids = NULL;
1719         struct trace_pid_list *other_pids = NULL;
1720         struct trace_pid_list *pid_list;
1721         struct trace_event_file *file;
1722         ssize_t ret;
1723
1724         if (!cnt)
1725                 return 0;
1726
1727         ret = tracing_update_buffers();
1728         if (ret < 0)
1729                 return ret;
1730
1731         mutex_lock(&event_mutex);
1732
1733         if (type == TRACE_PIDS) {
1734                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1735                                                           lockdep_is_held(&event_mutex));
1736                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1737                                                           lockdep_is_held(&event_mutex));
1738         } else {
1739                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1740                                                           lockdep_is_held(&event_mutex));
1741                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1742                                                           lockdep_is_held(&event_mutex));
1743         }
1744
1745         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1746         if (ret < 0)
1747                 goto out;
1748
1749         if (type == TRACE_PIDS)
1750                 rcu_assign_pointer(tr->filtered_pids, pid_list);
1751         else
1752                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1753
1754         list_for_each_entry(file, &tr->events, list) {
1755                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1756         }
1757
1758         if (filtered_pids) {
1759                 tracepoint_synchronize_unregister();
1760                 trace_free_pid_list(filtered_pids);
1761         } else if (pid_list && !other_pids) {
1762                 register_pid_events(tr);
1763         }
1764
1765         /*
1766          * Ignoring of pids is done at task switch. But we have to
1767          * check for those tasks that are currently running.
1768          * Always do this in case a pid was appended or removed.
1769          */
1770         on_each_cpu(ignore_task_cpu, tr, 1);
1771
1772  out:
1773         mutex_unlock(&event_mutex);
1774
1775         if (ret > 0)
1776                 *ppos += ret;
1777
1778         return ret;
1779 }
1780
1781 static ssize_t
1782 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1783                        size_t cnt, loff_t *ppos)
1784 {
1785         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1786 }
1787
1788 static ssize_t
1789 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1790                         size_t cnt, loff_t *ppos)
1791 {
1792         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
1793 }
1794
1795 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1796 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1797 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1798 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
1799 static int ftrace_event_release(struct inode *inode, struct file *file);
1800
1801 static const struct seq_operations show_event_seq_ops = {
1802         .start = t_start,
1803         .next = t_next,
1804         .show = t_show,
1805         .stop = t_stop,
1806 };
1807
1808 static const struct seq_operations show_set_event_seq_ops = {
1809         .start = s_start,
1810         .next = s_next,
1811         .show = t_show,
1812         .stop = t_stop,
1813 };
1814
1815 static const struct seq_operations show_set_pid_seq_ops = {
1816         .start = p_start,
1817         .next = p_next,
1818         .show = trace_pid_show,
1819         .stop = p_stop,
1820 };
1821
1822 static const struct seq_operations show_set_no_pid_seq_ops = {
1823         .start = np_start,
1824         .next = np_next,
1825         .show = trace_pid_show,
1826         .stop = p_stop,
1827 };
1828
1829 static const struct file_operations ftrace_avail_fops = {
1830         .open = ftrace_event_avail_open,
1831         .read = seq_read,
1832         .llseek = seq_lseek,
1833         .release = seq_release,
1834 };
1835
1836 static const struct file_operations ftrace_set_event_fops = {
1837         .open = ftrace_event_set_open,
1838         .read = seq_read,
1839         .write = ftrace_event_write,
1840         .llseek = seq_lseek,
1841         .release = ftrace_event_release,
1842 };
1843
1844 static const struct file_operations ftrace_set_event_pid_fops = {
1845         .open = ftrace_event_set_pid_open,
1846         .read = seq_read,
1847         .write = ftrace_event_pid_write,
1848         .llseek = seq_lseek,
1849         .release = ftrace_event_release,
1850 };
1851
1852 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
1853         .open = ftrace_event_set_npid_open,
1854         .read = seq_read,
1855         .write = ftrace_event_npid_write,
1856         .llseek = seq_lseek,
1857         .release = ftrace_event_release,
1858 };
1859
1860 static const struct file_operations ftrace_enable_fops = {
1861         .open = tracing_open_generic,
1862         .read = event_enable_read,
1863         .write = event_enable_write,
1864         .llseek = default_llseek,
1865 };
1866
1867 static const struct file_operations ftrace_event_format_fops = {
1868         .open = trace_format_open,
1869         .read = seq_read,
1870         .llseek = seq_lseek,
1871         .release = seq_release,
1872 };
1873
1874 static const struct file_operations ftrace_event_id_fops = {
1875         .read = event_id_read,
1876         .llseek = default_llseek,
1877 };
1878
1879 static const struct file_operations ftrace_event_filter_fops = {
1880         .open = tracing_open_generic,
1881         .read = event_filter_read,
1882         .write = event_filter_write,
1883         .llseek = default_llseek,
1884 };
1885
1886 static const struct file_operations ftrace_subsystem_filter_fops = {
1887         .open = subsystem_open,
1888         .read = subsystem_filter_read,
1889         .write = subsystem_filter_write,
1890         .llseek = default_llseek,
1891         .release = subsystem_release,
1892 };
1893
1894 static const struct file_operations ftrace_system_enable_fops = {
1895         .open = subsystem_open,
1896         .read = system_enable_read,
1897         .write = system_enable_write,
1898         .llseek = default_llseek,
1899         .release = subsystem_release,
1900 };
1901
1902 static const struct file_operations ftrace_tr_enable_fops = {
1903         .open = system_tr_open,
1904         .read = system_enable_read,
1905         .write = system_enable_write,
1906         .llseek = default_llseek,
1907         .release = subsystem_release,
1908 };
1909
1910 static const struct file_operations ftrace_show_header_fops = {
1911         .open = tracing_open_generic,
1912         .read = show_header,
1913         .llseek = default_llseek,
1914 };
1915
1916 static int
1917 ftrace_event_open(struct inode *inode, struct file *file,
1918                   const struct seq_operations *seq_ops)
1919 {
1920         struct seq_file *m;
1921         int ret;
1922
1923         ret = security_locked_down(LOCKDOWN_TRACEFS);
1924         if (ret)
1925                 return ret;
1926
1927         ret = seq_open(file, seq_ops);
1928         if (ret < 0)
1929                 return ret;
1930         m = file->private_data;
1931         /* copy tr over to seq ops */
1932         m->private = inode->i_private;
1933
1934         return ret;
1935 }
1936
1937 static int ftrace_event_release(struct inode *inode, struct file *file)
1938 {
1939         struct trace_array *tr = inode->i_private;
1940
1941         trace_array_put(tr);
1942
1943         return seq_release(inode, file);
1944 }
1945
1946 static int
1947 ftrace_event_avail_open(struct inode *inode, struct file *file)
1948 {
1949         const struct seq_operations *seq_ops = &show_event_seq_ops;
1950
1951         /* Checks for tracefs lockdown */
1952         return ftrace_event_open(inode, file, seq_ops);
1953 }
1954
1955 static int
1956 ftrace_event_set_open(struct inode *inode, struct file *file)
1957 {
1958         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1959         struct trace_array *tr = inode->i_private;
1960         int ret;
1961
1962         ret = tracing_check_open_get_tr(tr);
1963         if (ret)
1964                 return ret;
1965
1966         if ((file->f_mode & FMODE_WRITE) &&
1967             (file->f_flags & O_TRUNC))
1968                 ftrace_clear_events(tr);
1969
1970         ret = ftrace_event_open(inode, file, seq_ops);
1971         if (ret < 0)
1972                 trace_array_put(tr);
1973         return ret;
1974 }
1975
1976 static int
1977 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1978 {
1979         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1980         struct trace_array *tr = inode->i_private;
1981         int ret;
1982
1983         ret = tracing_check_open_get_tr(tr);
1984         if (ret)
1985                 return ret;
1986
1987         if ((file->f_mode & FMODE_WRITE) &&
1988             (file->f_flags & O_TRUNC))
1989                 ftrace_clear_event_pids(tr, TRACE_PIDS);
1990
1991         ret = ftrace_event_open(inode, file, seq_ops);
1992         if (ret < 0)
1993                 trace_array_put(tr);
1994         return ret;
1995 }
1996
1997 static int
1998 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
1999 {
2000         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2001         struct trace_array *tr = inode->i_private;
2002         int ret;
2003
2004         ret = tracing_check_open_get_tr(tr);
2005         if (ret)
2006                 return ret;
2007
2008         if ((file->f_mode & FMODE_WRITE) &&
2009             (file->f_flags & O_TRUNC))
2010                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2011
2012         ret = ftrace_event_open(inode, file, seq_ops);
2013         if (ret < 0)
2014                 trace_array_put(tr);
2015         return ret;
2016 }
2017
2018 static struct event_subsystem *
2019 create_new_subsystem(const char *name)
2020 {
2021         struct event_subsystem *system;
2022
2023         /* need to create new entry */
2024         system = kmalloc(sizeof(*system), GFP_KERNEL);
2025         if (!system)
2026                 return NULL;
2027
2028         system->ref_count = 1;
2029
2030         /* Only allocate if dynamic (kprobes and modules) */
2031         system->name = kstrdup_const(name, GFP_KERNEL);
2032         if (!system->name)
2033                 goto out_free;
2034
2035         system->filter = NULL;
2036
2037         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2038         if (!system->filter)
2039                 goto out_free;
2040
2041         list_add(&system->list, &event_subsystems);
2042
2043         return system;
2044
2045  out_free:
2046         kfree_const(system->name);
2047         kfree(system);
2048         return NULL;
2049 }
2050
2051 static struct dentry *
2052 event_subsystem_dir(struct trace_array *tr, const char *name,
2053                     struct trace_event_file *file, struct dentry *parent)
2054 {
2055         struct trace_subsystem_dir *dir;
2056         struct event_subsystem *system;
2057         struct dentry *entry;
2058
2059         /* First see if we did not already create this dir */
2060         list_for_each_entry(dir, &tr->systems, list) {
2061                 system = dir->subsystem;
2062                 if (strcmp(system->name, name) == 0) {
2063                         dir->nr_events++;
2064                         file->system = dir;
2065                         return dir->entry;
2066                 }
2067         }
2068
2069         /* Now see if the system itself exists. */
2070         list_for_each_entry(system, &event_subsystems, list) {
2071                 if (strcmp(system->name, name) == 0)
2072                         break;
2073         }
2074         /* Reset system variable when not found */
2075         if (&system->list == &event_subsystems)
2076                 system = NULL;
2077
2078         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2079         if (!dir)
2080                 goto out_fail;
2081
2082         if (!system) {
2083                 system = create_new_subsystem(name);
2084                 if (!system)
2085                         goto out_free;
2086         } else
2087                 __get_system(system);
2088
2089         dir->entry = tracefs_create_dir(name, parent);
2090         if (!dir->entry) {
2091                 pr_warn("Failed to create system directory %s\n", name);
2092                 __put_system(system);
2093                 goto out_free;
2094         }
2095
2096         dir->tr = tr;
2097         dir->ref_count = 1;
2098         dir->nr_events = 1;
2099         dir->subsystem = system;
2100         file->system = dir;
2101
2102         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2103                                     &ftrace_subsystem_filter_fops);
2104         if (!entry) {
2105                 kfree(system->filter);
2106                 system->filter = NULL;
2107                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2108         }
2109
2110         trace_create_file("enable", 0644, dir->entry, dir,
2111                           &ftrace_system_enable_fops);
2112
2113         list_add(&dir->list, &tr->systems);
2114
2115         return dir->entry;
2116
2117  out_free:
2118         kfree(dir);
2119  out_fail:
2120         /* Only print this message if failed on memory allocation */
2121         if (!dir || !system)
2122                 pr_warn("No memory to create event subsystem %s\n", name);
2123         return NULL;
2124 }
2125
2126 static int
2127 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2128 {
2129         struct trace_event_call *call = file->event_call;
2130         struct trace_array *tr = file->tr;
2131         struct list_head *head;
2132         struct dentry *d_events;
2133         const char *name;
2134         int ret;
2135
2136         /*
2137          * If the trace point header did not define TRACE_SYSTEM
2138          * then the system would be called "TRACE_SYSTEM".
2139          */
2140         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2141                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2142                 if (!d_events)
2143                         return -ENOMEM;
2144         } else
2145                 d_events = parent;
2146
2147         name = trace_event_name(call);
2148         file->dir = tracefs_create_dir(name, d_events);
2149         if (!file->dir) {
2150                 pr_warn("Could not create tracefs '%s' directory\n", name);
2151                 return -1;
2152         }
2153
2154         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2155                 trace_create_file("enable", 0644, file->dir, file,
2156                                   &ftrace_enable_fops);
2157
2158 #ifdef CONFIG_PERF_EVENTS
2159         if (call->event.type && call->class->reg)
2160                 trace_create_file("id", 0444, file->dir,
2161                                   (void *)(long)call->event.type,
2162                                   &ftrace_event_id_fops);
2163 #endif
2164
2165         /*
2166          * Other events may have the same class. Only update
2167          * the fields if they are not already defined.
2168          */
2169         head = trace_get_fields(call);
2170         if (list_empty(head)) {
2171                 struct trace_event_fields *field = call->class->fields_array;
2172                 unsigned int offset = sizeof(struct trace_entry);
2173
2174                 for (; field->type; field++) {
2175                         if (field->type == TRACE_FUNCTION_TYPE) {
2176                                 ret = field->define_fields(call);
2177                                 break;
2178                         }
2179
2180                         offset = ALIGN(offset, field->align);
2181                         ret = trace_define_field(call, field->type, field->name,
2182                                                  offset, field->size,
2183                                                  field->is_signed, field->filter_type);
2184                         if (ret)
2185                                 break;
2186
2187                         offset += field->size;
2188                 }
2189                 if (ret < 0) {
2190                         pr_warn("Could not initialize trace point events/%s\n",
2191                                 name);
2192                         return -1;
2193                 }
2194         }
2195
2196         /*
2197          * Only event directories that can be enabled should have
2198          * triggers or filters.
2199          */
2200         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2201                 trace_create_file("filter", 0644, file->dir, file,
2202                                   &ftrace_event_filter_fops);
2203
2204                 trace_create_file("trigger", 0644, file->dir, file,
2205                                   &event_trigger_fops);
2206         }
2207
2208 #ifdef CONFIG_HIST_TRIGGERS
2209         trace_create_file("hist", 0444, file->dir, file,
2210                           &event_hist_fops);
2211 #endif
2212 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2213         trace_create_file("hist_debug", 0444, file->dir, file,
2214                           &event_hist_debug_fops);
2215 #endif
2216         trace_create_file("format", 0444, file->dir, call,
2217                           &ftrace_event_format_fops);
2218
2219 #ifdef CONFIG_TRACE_EVENT_INJECT
2220         if (call->event.type && call->class->reg)
2221                 trace_create_file("inject", 0200, file->dir, file,
2222                                   &event_inject_fops);
2223 #endif
2224
2225         return 0;
2226 }
2227
2228 static void remove_event_from_tracers(struct trace_event_call *call)
2229 {
2230         struct trace_event_file *file;
2231         struct trace_array *tr;
2232
2233         do_for_each_event_file_safe(tr, file) {
2234                 if (file->event_call != call)
2235                         continue;
2236
2237                 remove_event_file_dir(file);
2238                 /*
2239                  * The do_for_each_event_file_safe() is
2240                  * a double loop. After finding the call for this
2241                  * trace_array, we use break to jump to the next
2242                  * trace_array.
2243                  */
2244                 break;
2245         } while_for_each_event_file();
2246 }
2247
2248 static void event_remove(struct trace_event_call *call)
2249 {
2250         struct trace_array *tr;
2251         struct trace_event_file *file;
2252
2253         do_for_each_event_file(tr, file) {
2254                 if (file->event_call != call)
2255                         continue;
2256
2257                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2258                         tr->clear_trace = true;
2259
2260                 ftrace_event_enable_disable(file, 0);
2261                 /*
2262                  * The do_for_each_event_file() is
2263                  * a double loop. After finding the call for this
2264                  * trace_array, we use break to jump to the next
2265                  * trace_array.
2266                  */
2267                 break;
2268         } while_for_each_event_file();
2269
2270         if (call->event.funcs)
2271                 __unregister_trace_event(&call->event);
2272         remove_event_from_tracers(call);
2273         list_del(&call->list);
2274 }
2275
2276 static int event_init(struct trace_event_call *call)
2277 {
2278         int ret = 0;
2279         const char *name;
2280
2281         name = trace_event_name(call);
2282         if (WARN_ON(!name))
2283                 return -EINVAL;
2284
2285         if (call->class->raw_init) {
2286                 ret = call->class->raw_init(call);
2287                 if (ret < 0 && ret != -ENOSYS)
2288                         pr_warn("Could not initialize trace events/%s\n", name);
2289         }
2290
2291         return ret;
2292 }
2293
2294 static int
2295 __register_event(struct trace_event_call *call, struct module *mod)
2296 {
2297         int ret;
2298
2299         ret = event_init(call);
2300         if (ret < 0)
2301                 return ret;
2302
2303         list_add(&call->list, &ftrace_events);
2304         call->mod = mod;
2305
2306         return 0;
2307 }
2308
2309 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2310 {
2311         int rlen;
2312         int elen;
2313
2314         /* Find the length of the eval value as a string */
2315         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2316         /* Make sure there's enough room to replace the string with the value */
2317         if (len < elen)
2318                 return NULL;
2319
2320         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2321
2322         /* Get the rest of the string of ptr */
2323         rlen = strlen(ptr + len);
2324         memmove(ptr + elen, ptr + len, rlen);
2325         /* Make sure we end the new string */
2326         ptr[elen + rlen] = 0;
2327
2328         return ptr + elen;
2329 }
2330
2331 static void update_event_printk(struct trace_event_call *call,
2332                                 struct trace_eval_map *map)
2333 {
2334         char *ptr;
2335         int quote = 0;
2336         int len = strlen(map->eval_string);
2337
2338         for (ptr = call->print_fmt; *ptr; ptr++) {
2339                 if (*ptr == '\\') {
2340                         ptr++;
2341                         /* paranoid */
2342                         if (!*ptr)
2343                                 break;
2344                         continue;
2345                 }
2346                 if (*ptr == '"') {
2347                         quote ^= 1;
2348                         continue;
2349                 }
2350                 if (quote)
2351                         continue;
2352                 if (isdigit(*ptr)) {
2353                         /* skip numbers */
2354                         do {
2355                                 ptr++;
2356                                 /* Check for alpha chars like ULL */
2357                         } while (isalnum(*ptr));
2358                         if (!*ptr)
2359                                 break;
2360                         /*
2361                          * A number must have some kind of delimiter after
2362                          * it, and we can ignore that too.
2363                          */
2364                         continue;
2365                 }
2366                 if (isalpha(*ptr) || *ptr == '_') {
2367                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2368                             !isalnum(ptr[len]) && ptr[len] != '_') {
2369                                 ptr = eval_replace(ptr, map, len);
2370                                 /* enum/sizeof string smaller than value */
2371                                 if (WARN_ON_ONCE(!ptr))
2372                                         return;
2373                                 /*
2374                                  * No need to decrement here, as eval_replace()
2375                                  * returns the pointer to the character passed
2376                                  * the eval, and two evals can not be placed
2377                                  * back to back without something in between.
2378                                  * We can skip that something in between.
2379                                  */
2380                                 continue;
2381                         }
2382                 skip_more:
2383                         do {
2384                                 ptr++;
2385                         } while (isalnum(*ptr) || *ptr == '_');
2386                         if (!*ptr)
2387                                 break;
2388                         /*
2389                          * If what comes after this variable is a '.' or
2390                          * '->' then we can continue to ignore that string.
2391                          */
2392                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2393                                 ptr += *ptr == '.' ? 1 : 2;
2394                                 if (!*ptr)
2395                                         break;
2396                                 goto skip_more;
2397                         }
2398                         /*
2399                          * Once again, we can skip the delimiter that came
2400                          * after the string.
2401                          */
2402                         continue;
2403                 }
2404         }
2405 }
2406
2407 void trace_event_eval_update(struct trace_eval_map **map, int len)
2408 {
2409         struct trace_event_call *call, *p;
2410         const char *last_system = NULL;
2411         bool first = false;
2412         int last_i;
2413         int i;
2414
2415         down_write(&trace_event_sem);
2416         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2417                 /* events are usually grouped together with systems */
2418                 if (!last_system || call->class->system != last_system) {
2419                         first = true;
2420                         last_i = 0;
2421                         last_system = call->class->system;
2422                 }
2423
2424                 /*
2425                  * Since calls are grouped by systems, the likelyhood that the
2426                  * next call in the iteration belongs to the same system as the
2427                  * previous call is high. As an optimization, we skip seaching
2428                  * for a map[] that matches the call's system if the last call
2429                  * was from the same system. That's what last_i is for. If the
2430                  * call has the same system as the previous call, then last_i
2431                  * will be the index of the first map[] that has a matching
2432                  * system.
2433                  */
2434                 for (i = last_i; i < len; i++) {
2435                         if (call->class->system == map[i]->system) {
2436                                 /* Save the first system if need be */
2437                                 if (first) {
2438                                         last_i = i;
2439                                         first = false;
2440                                 }
2441                                 update_event_printk(call, map[i]);
2442                         }
2443                 }
2444         }
2445         up_write(&trace_event_sem);
2446 }
2447
2448 static struct trace_event_file *
2449 trace_create_new_event(struct trace_event_call *call,
2450                        struct trace_array *tr)
2451 {
2452         struct trace_event_file *file;
2453
2454         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2455         if (!file)
2456                 return NULL;
2457
2458         file->event_call = call;
2459         file->tr = tr;
2460         atomic_set(&file->sm_ref, 0);
2461         atomic_set(&file->tm_ref, 0);
2462         INIT_LIST_HEAD(&file->triggers);
2463         list_add(&file->list, &tr->events);
2464
2465         return file;
2466 }
2467
2468 /* Add an event to a trace directory */
2469 static int
2470 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2471 {
2472         struct trace_event_file *file;
2473
2474         file = trace_create_new_event(call, tr);
2475         if (!file)
2476                 return -ENOMEM;
2477
2478         return event_create_dir(tr->event_dir, file);
2479 }
2480
2481 /*
2482  * Just create a decriptor for early init. A descriptor is required
2483  * for enabling events at boot. We want to enable events before
2484  * the filesystem is initialized.
2485  */
2486 static __init int
2487 __trace_early_add_new_event(struct trace_event_call *call,
2488                             struct trace_array *tr)
2489 {
2490         struct trace_event_file *file;
2491
2492         file = trace_create_new_event(call, tr);
2493         if (!file)
2494                 return -ENOMEM;
2495
2496         return 0;
2497 }
2498
2499 struct ftrace_module_file_ops;
2500 static void __add_event_to_tracers(struct trace_event_call *call);
2501
2502 /* Add an additional event_call dynamically */
2503 int trace_add_event_call(struct trace_event_call *call)
2504 {
2505         int ret;
2506         lockdep_assert_held(&event_mutex);
2507
2508         mutex_lock(&trace_types_lock);
2509
2510         ret = __register_event(call, NULL);
2511         if (ret >= 0)
2512                 __add_event_to_tracers(call);
2513
2514         mutex_unlock(&trace_types_lock);
2515         return ret;
2516 }
2517
2518 /*
2519  * Must be called under locking of trace_types_lock, event_mutex and
2520  * trace_event_sem.
2521  */
2522 static void __trace_remove_event_call(struct trace_event_call *call)
2523 {
2524         event_remove(call);
2525         trace_destroy_fields(call);
2526         free_event_filter(call->filter);
2527         call->filter = NULL;
2528 }
2529
2530 static int probe_remove_event_call(struct trace_event_call *call)
2531 {
2532         struct trace_array *tr;
2533         struct trace_event_file *file;
2534
2535 #ifdef CONFIG_PERF_EVENTS
2536         if (call->perf_refcount)
2537                 return -EBUSY;
2538 #endif
2539         do_for_each_event_file(tr, file) {
2540                 if (file->event_call != call)
2541                         continue;
2542                 /*
2543                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2544                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2545                  * TRACE_REG_UNREGISTER.
2546                  */
2547                 if (file->flags & EVENT_FILE_FL_ENABLED)
2548                         return -EBUSY;
2549                 /*
2550                  * The do_for_each_event_file_safe() is
2551                  * a double loop. After finding the call for this
2552                  * trace_array, we use break to jump to the next
2553                  * trace_array.
2554                  */
2555                 break;
2556         } while_for_each_event_file();
2557
2558         __trace_remove_event_call(call);
2559
2560         return 0;
2561 }
2562
2563 /* Remove an event_call */
2564 int trace_remove_event_call(struct trace_event_call *call)
2565 {
2566         int ret;
2567
2568         lockdep_assert_held(&event_mutex);
2569
2570         mutex_lock(&trace_types_lock);
2571         down_write(&trace_event_sem);
2572         ret = probe_remove_event_call(call);
2573         up_write(&trace_event_sem);
2574         mutex_unlock(&trace_types_lock);
2575
2576         return ret;
2577 }
2578
2579 #define for_each_event(event, start, end)                       \
2580         for (event = start;                                     \
2581              (unsigned long)event < (unsigned long)end;         \
2582              event++)
2583
2584 #ifdef CONFIG_MODULES
2585
2586 static void trace_module_add_events(struct module *mod)
2587 {
2588         struct trace_event_call **call, **start, **end;
2589
2590         if (!mod->num_trace_events)
2591                 return;
2592
2593         /* Don't add infrastructure for mods without tracepoints */
2594         if (trace_module_has_bad_taint(mod)) {
2595                 pr_err("%s: module has bad taint, not creating trace events\n",
2596                        mod->name);
2597                 return;
2598         }
2599
2600         start = mod->trace_events;
2601         end = mod->trace_events + mod->num_trace_events;
2602
2603         for_each_event(call, start, end) {
2604                 __register_event(*call, mod);
2605                 __add_event_to_tracers(*call);
2606         }
2607 }
2608
2609 static void trace_module_remove_events(struct module *mod)
2610 {
2611         struct trace_event_call *call, *p;
2612
2613         down_write(&trace_event_sem);
2614         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2615                 if (call->mod == mod)
2616                         __trace_remove_event_call(call);
2617         }
2618         up_write(&trace_event_sem);
2619
2620         /*
2621          * It is safest to reset the ring buffer if the module being unloaded
2622          * registered any events that were used. The only worry is if
2623          * a new module gets loaded, and takes on the same id as the events
2624          * of this module. When printing out the buffer, traced events left
2625          * over from this module may be passed to the new module events and
2626          * unexpected results may occur.
2627          */
2628         tracing_reset_all_online_cpus();
2629 }
2630
2631 static int trace_module_notify(struct notifier_block *self,
2632                                unsigned long val, void *data)
2633 {
2634         struct module *mod = data;
2635
2636         mutex_lock(&event_mutex);
2637         mutex_lock(&trace_types_lock);
2638         switch (val) {
2639         case MODULE_STATE_COMING:
2640                 trace_module_add_events(mod);
2641                 break;
2642         case MODULE_STATE_GOING:
2643                 trace_module_remove_events(mod);
2644                 break;
2645         }
2646         mutex_unlock(&trace_types_lock);
2647         mutex_unlock(&event_mutex);
2648
2649         return 0;
2650 }
2651
2652 static struct notifier_block trace_module_nb = {
2653         .notifier_call = trace_module_notify,
2654         .priority = 1, /* higher than trace.c module notify */
2655 };
2656 #endif /* CONFIG_MODULES */
2657
2658 /* Create a new event directory structure for a trace directory. */
2659 static void
2660 __trace_add_event_dirs(struct trace_array *tr)
2661 {
2662         struct trace_event_call *call;
2663         int ret;
2664
2665         list_for_each_entry(call, &ftrace_events, list) {
2666                 ret = __trace_add_new_event(call, tr);
2667                 if (ret < 0)
2668                         pr_warn("Could not create directory for event %s\n",
2669                                 trace_event_name(call));
2670         }
2671 }
2672
2673 /* Returns any file that matches the system and event */
2674 struct trace_event_file *
2675 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2676 {
2677         struct trace_event_file *file;
2678         struct trace_event_call *call;
2679         const char *name;
2680
2681         list_for_each_entry(file, &tr->events, list) {
2682
2683                 call = file->event_call;
2684                 name = trace_event_name(call);
2685
2686                 if (!name || !call->class)
2687                         continue;
2688
2689                 if (strcmp(event, name) == 0 &&
2690                     strcmp(system, call->class->system) == 0)
2691                         return file;
2692         }
2693         return NULL;
2694 }
2695
2696 /* Returns valid trace event files that match system and event */
2697 struct trace_event_file *
2698 find_event_file(struct trace_array *tr, const char *system, const char *event)
2699 {
2700         struct trace_event_file *file;
2701
2702         file = __find_event_file(tr, system, event);
2703         if (!file || !file->event_call->class->reg ||
2704             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2705                 return NULL;
2706
2707         return file;
2708 }
2709
2710 /**
2711  * trace_get_event_file - Find and return a trace event file
2712  * @instance: The name of the trace instance containing the event
2713  * @system: The name of the system containing the event
2714  * @event: The name of the event
2715  *
2716  * Return a trace event file given the trace instance name, trace
2717  * system, and trace event name.  If the instance name is NULL, it
2718  * refers to the top-level trace array.
2719  *
2720  * This function will look it up and return it if found, after calling
2721  * trace_array_get() to prevent the instance from going away, and
2722  * increment the event's module refcount to prevent it from being
2723  * removed.
2724  *
2725  * To release the file, call trace_put_event_file(), which will call
2726  * trace_array_put() and decrement the event's module refcount.
2727  *
2728  * Return: The trace event on success, ERR_PTR otherwise.
2729  */
2730 struct trace_event_file *trace_get_event_file(const char *instance,
2731                                               const char *system,
2732                                               const char *event)
2733 {
2734         struct trace_array *tr = top_trace_array();
2735         struct trace_event_file *file = NULL;
2736         int ret = -EINVAL;
2737
2738         if (instance) {
2739                 tr = trace_array_find_get(instance);
2740                 if (!tr)
2741                         return ERR_PTR(-ENOENT);
2742         } else {
2743                 ret = trace_array_get(tr);
2744                 if (ret)
2745                         return ERR_PTR(ret);
2746         }
2747
2748         mutex_lock(&event_mutex);
2749
2750         file = find_event_file(tr, system, event);
2751         if (!file) {
2752                 trace_array_put(tr);
2753                 ret = -EINVAL;
2754                 goto out;
2755         }
2756
2757         /* Don't let event modules unload while in use */
2758         ret = try_module_get(file->event_call->mod);
2759         if (!ret) {
2760                 trace_array_put(tr);
2761                 ret = -EBUSY;
2762                 goto out;
2763         }
2764
2765         ret = 0;
2766  out:
2767         mutex_unlock(&event_mutex);
2768
2769         if (ret)
2770                 file = ERR_PTR(ret);
2771
2772         return file;
2773 }
2774 EXPORT_SYMBOL_GPL(trace_get_event_file);
2775
2776 /**
2777  * trace_put_event_file - Release a file from trace_get_event_file()
2778  * @file: The trace event file
2779  *
2780  * If a file was retrieved using trace_get_event_file(), this should
2781  * be called when it's no longer needed.  It will cancel the previous
2782  * trace_array_get() called by that function, and decrement the
2783  * event's module refcount.
2784  */
2785 void trace_put_event_file(struct trace_event_file *file)
2786 {
2787         mutex_lock(&event_mutex);
2788         module_put(file->event_call->mod);
2789         mutex_unlock(&event_mutex);
2790
2791         trace_array_put(file->tr);
2792 }
2793 EXPORT_SYMBOL_GPL(trace_put_event_file);
2794
2795 #ifdef CONFIG_DYNAMIC_FTRACE
2796
2797 /* Avoid typos */
2798 #define ENABLE_EVENT_STR        "enable_event"
2799 #define DISABLE_EVENT_STR       "disable_event"
2800
2801 struct event_probe_data {
2802         struct trace_event_file *file;
2803         unsigned long                   count;
2804         int                             ref;
2805         bool                            enable;
2806 };
2807
2808 static void update_event_probe(struct event_probe_data *data)
2809 {
2810         if (data->enable)
2811                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2812         else
2813                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2814 }
2815
2816 static void
2817 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2818                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2819                    void *data)
2820 {
2821         struct ftrace_func_mapper *mapper = data;
2822         struct event_probe_data *edata;
2823         void **pdata;
2824
2825         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2826         if (!pdata || !*pdata)
2827                 return;
2828
2829         edata = *pdata;
2830         update_event_probe(edata);
2831 }
2832
2833 static void
2834 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2835                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2836                          void *data)
2837 {
2838         struct ftrace_func_mapper *mapper = data;
2839         struct event_probe_data *edata;
2840         void **pdata;
2841
2842         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2843         if (!pdata || !*pdata)
2844                 return;
2845
2846         edata = *pdata;
2847
2848         if (!edata->count)
2849                 return;
2850
2851         /* Skip if the event is in a state we want to switch to */
2852         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2853                 return;
2854
2855         if (edata->count != -1)
2856                 (edata->count)--;
2857
2858         update_event_probe(edata);
2859 }
2860
2861 static int
2862 event_enable_print(struct seq_file *m, unsigned long ip,
2863                    struct ftrace_probe_ops *ops, void *data)
2864 {
2865         struct ftrace_func_mapper *mapper = data;
2866         struct event_probe_data *edata;
2867         void **pdata;
2868
2869         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2870
2871         if (WARN_ON_ONCE(!pdata || !*pdata))
2872                 return 0;
2873
2874         edata = *pdata;
2875
2876         seq_printf(m, "%ps:", (void *)ip);
2877
2878         seq_printf(m, "%s:%s:%s",
2879                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2880                    edata->file->event_call->class->system,
2881                    trace_event_name(edata->file->event_call));
2882
2883         if (edata->count == -1)
2884                 seq_puts(m, ":unlimited\n");
2885         else
2886                 seq_printf(m, ":count=%ld\n", edata->count);
2887
2888         return 0;
2889 }
2890
2891 static int
2892 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2893                   unsigned long ip, void *init_data, void **data)
2894 {
2895         struct ftrace_func_mapper *mapper = *data;
2896         struct event_probe_data *edata = init_data;
2897         int ret;
2898
2899         if (!mapper) {
2900                 mapper = allocate_ftrace_func_mapper();
2901                 if (!mapper)
2902                         return -ENODEV;
2903                 *data = mapper;
2904         }
2905
2906         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2907         if (ret < 0)
2908                 return ret;
2909
2910         edata->ref++;
2911
2912         return 0;
2913 }
2914
2915 static int free_probe_data(void *data)
2916 {
2917         struct event_probe_data *edata = data;
2918
2919         edata->ref--;
2920         if (!edata->ref) {
2921                 /* Remove the SOFT_MODE flag */
2922                 __ftrace_event_enable_disable(edata->file, 0, 1);
2923                 module_put(edata->file->event_call->mod);
2924                 kfree(edata);
2925         }
2926         return 0;
2927 }
2928
2929 static void
2930 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2931                   unsigned long ip, void *data)
2932 {
2933         struct ftrace_func_mapper *mapper = data;
2934         struct event_probe_data *edata;
2935
2936         if (!ip) {
2937                 if (!mapper)
2938                         return;
2939                 free_ftrace_func_mapper(mapper, free_probe_data);
2940                 return;
2941         }
2942
2943         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2944
2945         if (WARN_ON_ONCE(!edata))
2946                 return;
2947
2948         if (WARN_ON_ONCE(edata->ref <= 0))
2949                 return;
2950
2951         free_probe_data(edata);
2952 }
2953
2954 static struct ftrace_probe_ops event_enable_probe_ops = {
2955         .func                   = event_enable_probe,
2956         .print                  = event_enable_print,
2957         .init                   = event_enable_init,
2958         .free                   = event_enable_free,
2959 };
2960
2961 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2962         .func                   = event_enable_count_probe,
2963         .print                  = event_enable_print,
2964         .init                   = event_enable_init,
2965         .free                   = event_enable_free,
2966 };
2967
2968 static struct ftrace_probe_ops event_disable_probe_ops = {
2969         .func                   = event_enable_probe,
2970         .print                  = event_enable_print,
2971         .init                   = event_enable_init,
2972         .free                   = event_enable_free,
2973 };
2974
2975 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2976         .func                   = event_enable_count_probe,
2977         .print                  = event_enable_print,
2978         .init                   = event_enable_init,
2979         .free                   = event_enable_free,
2980 };
2981
2982 static int
2983 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2984                   char *glob, char *cmd, char *param, int enabled)
2985 {
2986         struct trace_event_file *file;
2987         struct ftrace_probe_ops *ops;
2988         struct event_probe_data *data;
2989         const char *system;
2990         const char *event;
2991         char *number;
2992         bool enable;
2993         int ret;
2994
2995         if (!tr)
2996                 return -ENODEV;
2997
2998         /* hash funcs only work with set_ftrace_filter */
2999         if (!enabled || !param)
3000                 return -EINVAL;
3001
3002         system = strsep(&param, ":");
3003         if (!param)
3004                 return -EINVAL;
3005
3006         event = strsep(&param, ":");
3007
3008         mutex_lock(&event_mutex);
3009
3010         ret = -EINVAL;
3011         file = find_event_file(tr, system, event);
3012         if (!file)
3013                 goto out;
3014
3015         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3016
3017         if (enable)
3018                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3019         else
3020                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3021
3022         if (glob[0] == '!') {
3023                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3024                 goto out;
3025         }
3026
3027         ret = -ENOMEM;
3028
3029         data = kzalloc(sizeof(*data), GFP_KERNEL);
3030         if (!data)
3031                 goto out;
3032
3033         data->enable = enable;
3034         data->count = -1;
3035         data->file = file;
3036
3037         if (!param)
3038                 goto out_reg;
3039
3040         number = strsep(&param, ":");
3041
3042         ret = -EINVAL;
3043         if (!strlen(number))
3044                 goto out_free;
3045
3046         /*
3047          * We use the callback data field (which is a pointer)
3048          * as our counter.
3049          */
3050         ret = kstrtoul(number, 0, &data->count);
3051         if (ret)
3052                 goto out_free;
3053
3054  out_reg:
3055         /* Don't let event modules unload while probe registered */
3056         ret = try_module_get(file->event_call->mod);
3057         if (!ret) {
3058                 ret = -EBUSY;
3059                 goto out_free;
3060         }
3061
3062         ret = __ftrace_event_enable_disable(file, 1, 1);
3063         if (ret < 0)
3064                 goto out_put;
3065
3066         ret = register_ftrace_function_probe(glob, tr, ops, data);
3067         /*
3068          * The above returns on success the # of functions enabled,
3069          * but if it didn't find any functions it returns zero.
3070          * Consider no functions a failure too.
3071          */
3072         if (!ret) {
3073                 ret = -ENOENT;
3074                 goto out_disable;
3075         } else if (ret < 0)
3076                 goto out_disable;
3077         /* Just return zero, not the number of enabled functions */
3078         ret = 0;
3079  out:
3080         mutex_unlock(&event_mutex);
3081         return ret;
3082
3083  out_disable:
3084         __ftrace_event_enable_disable(file, 0, 1);
3085  out_put:
3086         module_put(file->event_call->mod);
3087  out_free:
3088         kfree(data);
3089         goto out;
3090 }
3091
3092 static struct ftrace_func_command event_enable_cmd = {
3093         .name                   = ENABLE_EVENT_STR,
3094         .func                   = event_enable_func,
3095 };
3096
3097 static struct ftrace_func_command event_disable_cmd = {
3098         .name                   = DISABLE_EVENT_STR,
3099         .func                   = event_enable_func,
3100 };
3101
3102 static __init int register_event_cmds(void)
3103 {
3104         int ret;
3105
3106         ret = register_ftrace_command(&event_enable_cmd);
3107         if (WARN_ON(ret < 0))
3108                 return ret;
3109         ret = register_ftrace_command(&event_disable_cmd);
3110         if (WARN_ON(ret < 0))
3111                 unregister_ftrace_command(&event_enable_cmd);
3112         return ret;
3113 }
3114 #else
3115 static inline int register_event_cmds(void) { return 0; }
3116 #endif /* CONFIG_DYNAMIC_FTRACE */
3117
3118 /*
3119  * The top level array has already had its trace_event_file
3120  * descriptors created in order to allow for early events to
3121  * be recorded. This function is called after the tracefs has been
3122  * initialized, and we now have to create the files associated
3123  * to the events.
3124  */
3125 static __init void
3126 __trace_early_add_event_dirs(struct trace_array *tr)
3127 {
3128         struct trace_event_file *file;
3129         int ret;
3130
3131
3132         list_for_each_entry(file, &tr->events, list) {
3133                 ret = event_create_dir(tr->event_dir, file);
3134                 if (ret < 0)
3135                         pr_warn("Could not create directory for event %s\n",
3136                                 trace_event_name(file->event_call));
3137         }
3138 }
3139
3140 /*
3141  * For early boot up, the top trace array requires to have
3142  * a list of events that can be enabled. This must be done before
3143  * the filesystem is set up in order to allow events to be traced
3144  * early.
3145  */
3146 static __init void
3147 __trace_early_add_events(struct trace_array *tr)
3148 {
3149         struct trace_event_call *call;
3150         int ret;
3151
3152         list_for_each_entry(call, &ftrace_events, list) {
3153                 /* Early boot up should not have any modules loaded */
3154                 if (WARN_ON_ONCE(call->mod))
3155                         continue;
3156
3157                 ret = __trace_early_add_new_event(call, tr);
3158                 if (ret < 0)
3159                         pr_warn("Could not create early event %s\n",
3160                                 trace_event_name(call));
3161         }
3162 }
3163
3164 /* Remove the event directory structure for a trace directory. */
3165 static void
3166 __trace_remove_event_dirs(struct trace_array *tr)
3167 {
3168         struct trace_event_file *file, *next;
3169
3170         list_for_each_entry_safe(file, next, &tr->events, list)
3171                 remove_event_file_dir(file);
3172 }
3173
3174 static void __add_event_to_tracers(struct trace_event_call *call)
3175 {
3176         struct trace_array *tr;
3177
3178         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3179                 __trace_add_new_event(call, tr);
3180 }
3181
3182 extern struct trace_event_call *__start_ftrace_events[];
3183 extern struct trace_event_call *__stop_ftrace_events[];
3184
3185 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3186
3187 static __init int setup_trace_event(char *str)
3188 {
3189         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3190         ring_buffer_expanded = true;
3191         tracing_selftest_disabled = true;
3192
3193         return 1;
3194 }
3195 __setup("trace_event=", setup_trace_event);
3196
3197 /* Expects to have event_mutex held when called */
3198 static int
3199 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3200 {
3201         struct dentry *d_events;
3202         struct dentry *entry;
3203
3204         entry = tracefs_create_file("set_event", 0644, parent,
3205                                     tr, &ftrace_set_event_fops);
3206         if (!entry) {
3207                 pr_warn("Could not create tracefs 'set_event' entry\n");
3208                 return -ENOMEM;
3209         }
3210
3211         d_events = tracefs_create_dir("events", parent);
3212         if (!d_events) {
3213                 pr_warn("Could not create tracefs 'events' directory\n");
3214                 return -ENOMEM;
3215         }
3216
3217         entry = trace_create_file("enable", 0644, d_events,
3218                                   tr, &ftrace_tr_enable_fops);
3219         if (!entry) {
3220                 pr_warn("Could not create tracefs 'enable' entry\n");
3221                 return -ENOMEM;
3222         }
3223
3224         /* There are not as crucial, just warn if they are not created */
3225
3226         entry = tracefs_create_file("set_event_pid", 0644, parent,
3227                                     tr, &ftrace_set_event_pid_fops);
3228         if (!entry)
3229                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3230
3231         entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3232                                     tr, &ftrace_set_event_notrace_pid_fops);
3233         if (!entry)
3234                 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3235
3236         /* ring buffer internal formats */
3237         entry = trace_create_file("header_page", 0444, d_events,
3238                                   ring_buffer_print_page_header,
3239                                   &ftrace_show_header_fops);
3240         if (!entry)
3241                 pr_warn("Could not create tracefs 'header_page' entry\n");
3242
3243         entry = trace_create_file("header_event", 0444, d_events,
3244                                   ring_buffer_print_entry_header,
3245                                   &ftrace_show_header_fops);
3246         if (!entry)
3247                 pr_warn("Could not create tracefs 'header_event' entry\n");
3248
3249         tr->event_dir = d_events;
3250
3251         return 0;
3252 }
3253
3254 /**
3255  * event_trace_add_tracer - add a instance of a trace_array to events
3256  * @parent: The parent dentry to place the files/directories for events in
3257  * @tr: The trace array associated with these events
3258  *
3259  * When a new instance is created, it needs to set up its events
3260  * directory, as well as other files associated with events. It also
3261  * creates the event hierachry in the @parent/events directory.
3262  *
3263  * Returns 0 on success.
3264  *
3265  * Must be called with event_mutex held.
3266  */
3267 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3268 {
3269         int ret;
3270
3271         lockdep_assert_held(&event_mutex);
3272
3273         ret = create_event_toplevel_files(parent, tr);
3274         if (ret)
3275                 goto out;
3276
3277         down_write(&trace_event_sem);
3278         __trace_add_event_dirs(tr);
3279         up_write(&trace_event_sem);
3280
3281  out:
3282         return ret;
3283 }
3284
3285 /*
3286  * The top trace array already had its file descriptors created.
3287  * Now the files themselves need to be created.
3288  */
3289 static __init int
3290 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3291 {
3292         int ret;
3293
3294         mutex_lock(&event_mutex);
3295
3296         ret = create_event_toplevel_files(parent, tr);
3297         if (ret)
3298                 goto out_unlock;
3299
3300         down_write(&trace_event_sem);
3301         __trace_early_add_event_dirs(tr);
3302         up_write(&trace_event_sem);
3303
3304  out_unlock:
3305         mutex_unlock(&event_mutex);
3306
3307         return ret;
3308 }
3309
3310 /* Must be called with event_mutex held */
3311 int event_trace_del_tracer(struct trace_array *tr)
3312 {
3313         lockdep_assert_held(&event_mutex);
3314
3315         /* Disable any event triggers and associated soft-disabled events */
3316         clear_event_triggers(tr);
3317
3318         /* Clear the pid list */
3319         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3320
3321         /* Disable any running events */
3322         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3323
3324         /* Make sure no more events are being executed */
3325         tracepoint_synchronize_unregister();
3326
3327         down_write(&trace_event_sem);
3328         __trace_remove_event_dirs(tr);
3329         tracefs_remove(tr->event_dir);
3330         up_write(&trace_event_sem);
3331
3332         tr->event_dir = NULL;
3333
3334         return 0;
3335 }
3336
3337 static __init int event_trace_memsetup(void)
3338 {
3339         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3340         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3341         return 0;
3342 }
3343
3344 static __init void
3345 early_enable_events(struct trace_array *tr, bool disable_first)
3346 {
3347         char *buf = bootup_event_buf;
3348         char *token;
3349         int ret;
3350
3351         while (true) {
3352                 token = strsep(&buf, ",");
3353
3354                 if (!token)
3355                         break;
3356
3357                 if (*token) {
3358                         /* Restarting syscalls requires that we stop them first */
3359                         if (disable_first)
3360                                 ftrace_set_clr_event(tr, token, 0);
3361
3362                         ret = ftrace_set_clr_event(tr, token, 1);
3363                         if (ret)
3364                                 pr_warn("Failed to enable trace event: %s\n", token);
3365                 }
3366
3367                 /* Put back the comma to allow this to be called again */
3368                 if (buf)
3369                         *(buf - 1) = ',';
3370         }
3371 }
3372
3373 static __init int event_trace_enable(void)
3374 {
3375         struct trace_array *tr = top_trace_array();
3376         struct trace_event_call **iter, *call;
3377         int ret;
3378
3379         if (!tr)
3380                 return -ENODEV;
3381
3382         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3383
3384                 call = *iter;
3385                 ret = event_init(call);
3386                 if (!ret)
3387                         list_add(&call->list, &ftrace_events);
3388         }
3389
3390         /*
3391          * We need the top trace array to have a working set of trace
3392          * points at early init, before the debug files and directories
3393          * are created. Create the file entries now, and attach them
3394          * to the actual file dentries later.
3395          */
3396         __trace_early_add_events(tr);
3397
3398         early_enable_events(tr, false);
3399
3400         trace_printk_start_comm();
3401
3402         register_event_cmds();
3403
3404         register_trigger_cmds();
3405
3406         return 0;
3407 }
3408
3409 /*
3410  * event_trace_enable() is called from trace_event_init() first to
3411  * initialize events and perhaps start any events that are on the
3412  * command line. Unfortunately, there are some events that will not
3413  * start this early, like the system call tracepoints that need
3414  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3415  * is called before pid 1 starts, and this flag is never set, making
3416  * the syscall tracepoint never get reached, but the event is enabled
3417  * regardless (and not doing anything).
3418  */
3419 static __init int event_trace_enable_again(void)
3420 {
3421         struct trace_array *tr;
3422
3423         tr = top_trace_array();
3424         if (!tr)
3425                 return -ENODEV;
3426
3427         early_enable_events(tr, true);
3428
3429         return 0;
3430 }
3431
3432 early_initcall(event_trace_enable_again);
3433
3434 __init int event_trace_init(void)
3435 {
3436         struct trace_array *tr;
3437         struct dentry *d_tracer;
3438         struct dentry *entry;
3439         int ret;
3440
3441         tr = top_trace_array();
3442         if (!tr)
3443                 return -ENODEV;
3444
3445         d_tracer = tracing_init_dentry();
3446         if (IS_ERR(d_tracer))
3447                 return 0;
3448
3449         entry = tracefs_create_file("available_events", 0444, d_tracer,
3450                                     tr, &ftrace_avail_fops);
3451         if (!entry)
3452                 pr_warn("Could not create tracefs 'available_events' entry\n");
3453
3454         if (trace_define_generic_fields())
3455                 pr_warn("tracing: Failed to allocated generic fields");
3456
3457         if (trace_define_common_fields())
3458                 pr_warn("tracing: Failed to allocate common fields");
3459
3460         ret = early_event_add_tracer(d_tracer, tr);
3461         if (ret)
3462                 return ret;
3463
3464 #ifdef CONFIG_MODULES
3465         ret = register_module_notifier(&trace_module_nb);
3466         if (ret)
3467                 pr_warn("Failed to register trace events module notifier\n");
3468 #endif
3469         return 0;
3470 }
3471
3472 void __init trace_event_init(void)
3473 {
3474         event_trace_memsetup();
3475         init_ftrace_syscalls();
3476         event_trace_enable();
3477 }
3478
3479 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3480
3481 static DEFINE_SPINLOCK(test_spinlock);
3482 static DEFINE_SPINLOCK(test_spinlock_irq);
3483 static DEFINE_MUTEX(test_mutex);
3484
3485 static __init void test_work(struct work_struct *dummy)
3486 {
3487         spin_lock(&test_spinlock);
3488         spin_lock_irq(&test_spinlock_irq);
3489         udelay(1);
3490         spin_unlock_irq(&test_spinlock_irq);
3491         spin_unlock(&test_spinlock);
3492
3493         mutex_lock(&test_mutex);
3494         msleep(1);
3495         mutex_unlock(&test_mutex);
3496 }
3497
3498 static __init int event_test_thread(void *unused)
3499 {
3500         void *test_malloc;
3501
3502         test_malloc = kmalloc(1234, GFP_KERNEL);
3503         if (!test_malloc)
3504                 pr_info("failed to kmalloc\n");
3505
3506         schedule_on_each_cpu(test_work);
3507
3508         kfree(test_malloc);
3509
3510         set_current_state(TASK_INTERRUPTIBLE);
3511         while (!kthread_should_stop()) {
3512                 schedule();
3513                 set_current_state(TASK_INTERRUPTIBLE);
3514         }
3515         __set_current_state(TASK_RUNNING);
3516
3517         return 0;
3518 }
3519
3520 /*
3521  * Do various things that may trigger events.
3522  */
3523 static __init void event_test_stuff(void)
3524 {
3525         struct task_struct *test_thread;
3526
3527         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3528         msleep(1);
3529         kthread_stop(test_thread);
3530 }
3531
3532 /*
3533  * For every trace event defined, we will test each trace point separately,
3534  * and then by groups, and finally all trace points.
3535  */
3536 static __init void event_trace_self_tests(void)
3537 {
3538         struct trace_subsystem_dir *dir;
3539         struct trace_event_file *file;
3540         struct trace_event_call *call;
3541         struct event_subsystem *system;
3542         struct trace_array *tr;
3543         int ret;
3544
3545         tr = top_trace_array();
3546         if (!tr)
3547                 return;
3548
3549         pr_info("Running tests on trace events:\n");
3550
3551         list_for_each_entry(file, &tr->events, list) {
3552
3553                 call = file->event_call;
3554
3555                 /* Only test those that have a probe */
3556                 if (!call->class || !call->class->probe)
3557                         continue;
3558
3559 /*
3560  * Testing syscall events here is pretty useless, but
3561  * we still do it if configured. But this is time consuming.
3562  * What we really need is a user thread to perform the
3563  * syscalls as we test.
3564  */
3565 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3566                 if (call->class->system &&
3567                     strcmp(call->class->system, "syscalls") == 0)
3568                         continue;
3569 #endif
3570
3571                 pr_info("Testing event %s: ", trace_event_name(call));
3572
3573                 /*
3574                  * If an event is already enabled, someone is using
3575                  * it and the self test should not be on.
3576                  */
3577                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3578                         pr_warn("Enabled event during self test!\n");
3579                         WARN_ON_ONCE(1);
3580                         continue;
3581                 }
3582
3583                 ftrace_event_enable_disable(file, 1);
3584                 event_test_stuff();
3585                 ftrace_event_enable_disable(file, 0);
3586
3587                 pr_cont("OK\n");
3588         }
3589
3590         /* Now test at the sub system level */
3591
3592         pr_info("Running tests on trace event systems:\n");
3593
3594         list_for_each_entry(dir, &tr->systems, list) {
3595
3596                 system = dir->subsystem;
3597
3598                 /* the ftrace system is special, skip it */
3599                 if (strcmp(system->name, "ftrace") == 0)
3600                         continue;
3601
3602                 pr_info("Testing event system %s: ", system->name);
3603
3604                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3605                 if (WARN_ON_ONCE(ret)) {
3606                         pr_warn("error enabling system %s\n",
3607                                 system->name);
3608                         continue;
3609                 }
3610
3611                 event_test_stuff();
3612
3613                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3614                 if (WARN_ON_ONCE(ret)) {
3615                         pr_warn("error disabling system %s\n",
3616                                 system->name);
3617                         continue;
3618                 }
3619
3620                 pr_cont("OK\n");
3621         }
3622
3623         /* Test with all events enabled */
3624
3625         pr_info("Running tests on all trace events:\n");
3626         pr_info("Testing all events: ");
3627
3628         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3629         if (WARN_ON_ONCE(ret)) {
3630                 pr_warn("error enabling all events\n");
3631                 return;
3632         }
3633
3634         event_test_stuff();
3635
3636         /* reset sysname */
3637         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3638         if (WARN_ON_ONCE(ret)) {
3639                 pr_warn("error disabling all events\n");
3640                 return;
3641         }
3642
3643         pr_cont("OK\n");
3644 }
3645
3646 #ifdef CONFIG_FUNCTION_TRACER
3647
3648 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3649
3650 static struct trace_event_file event_trace_file __initdata;
3651
3652 static void __init
3653 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3654                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3655 {
3656         struct trace_buffer *buffer;
3657         struct ring_buffer_event *event;
3658         struct ftrace_entry *entry;
3659         unsigned long flags;
3660         long disabled;
3661         int cpu;
3662         int pc;
3663
3664         pc = preempt_count();
3665         preempt_disable_notrace();
3666         cpu = raw_smp_processor_id();
3667         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3668
3669         if (disabled != 1)
3670                 goto out;
3671
3672         local_save_flags(flags);
3673
3674         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3675                                                 TRACE_FN, sizeof(*entry),
3676                                                 flags, pc);
3677         if (!event)
3678                 goto out;
3679         entry   = ring_buffer_event_data(event);
3680         entry->ip                       = ip;
3681         entry->parent_ip                = parent_ip;
3682
3683         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3684                                     entry, flags, pc);
3685  out:
3686         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3687         preempt_enable_notrace();
3688 }
3689
3690 static struct ftrace_ops trace_ops __initdata  =
3691 {
3692         .func = function_test_events_call,
3693         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3694 };
3695
3696 static __init void event_trace_self_test_with_function(void)
3697 {
3698         int ret;
3699
3700         event_trace_file.tr = top_trace_array();
3701         if (WARN_ON(!event_trace_file.tr))
3702                 return;
3703
3704         ret = register_ftrace_function(&trace_ops);
3705         if (WARN_ON(ret < 0)) {
3706                 pr_info("Failed to enable function tracer for event tests\n");
3707                 return;
3708         }
3709         pr_info("Running tests again, along with the function tracer\n");
3710         event_trace_self_tests();
3711         unregister_ftrace_function(&trace_ops);
3712 }
3713 #else
3714 static __init void event_trace_self_test_with_function(void)
3715 {
3716 }
3717 #endif
3718
3719 static __init int event_trace_self_tests_init(void)
3720 {
3721         if (!tracing_selftest_disabled) {
3722                 event_trace_self_tests();
3723                 event_trace_self_test_with_function();
3724         }
3725
3726         return 0;
3727 }
3728
3729 late_initcall(event_trace_self_tests_init);
3730
3731 #endif