xref: /openbmc/linux/kernel/trace/trace_events.c (revision c4a11bf4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * event tracer
4  *
5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6  *
7  *  - Added format output of fields of the trace point.
8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
9  *
10  */
11 
12 #define pr_fmt(fmt) fmt
13 
14 #include <linux/workqueue.h>
15 #include <linux/security.h>
16 #include <linux/spinlock.h>
17 #include <linux/kthread.h>
18 #include <linux/tracefs.h>
19 #include <linux/uaccess.h>
20 #include <linux/module.h>
21 #include <linux/ctype.h>
22 #include <linux/sort.h>
23 #include <linux/slab.h>
24 #include <linux/delay.h>
25 
26 #include <trace/events/sched.h>
27 #include <trace/syscall.h>
28 
29 #include <asm/setup.h>
30 
31 #include "trace_output.h"
32 
33 #undef TRACE_SYSTEM
34 #define TRACE_SYSTEM "TRACE_SYSTEM"
35 
36 DEFINE_MUTEX(event_mutex);
37 
38 LIST_HEAD(ftrace_events);
39 static LIST_HEAD(ftrace_generic_fields);
40 static LIST_HEAD(ftrace_common_fields);
41 static bool eventdir_initialized;
42 
43 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
44 
45 static struct kmem_cache *field_cachep;
46 static struct kmem_cache *file_cachep;
47 
48 static inline int system_refcount(struct event_subsystem *system)
49 {
50 	return system->ref_count;
51 }
52 
53 static int system_refcount_inc(struct event_subsystem *system)
54 {
55 	return system->ref_count++;
56 }
57 
58 static int system_refcount_dec(struct event_subsystem *system)
59 {
60 	return --system->ref_count;
61 }
62 
63 /* Double loops, do not use break, only goto's work */
64 #define do_for_each_event_file(tr, file)			\
65 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
66 		list_for_each_entry(file, &tr->events, list)
67 
68 #define do_for_each_event_file_safe(tr, file)			\
69 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
70 		struct trace_event_file *___n;				\
71 		list_for_each_entry_safe(file, ___n, &tr->events, list)
72 
73 #define while_for_each_event_file()		\
74 	}
75 
76 static struct ftrace_event_field *
77 __find_event_field(struct list_head *head, char *name)
78 {
79 	struct ftrace_event_field *field;
80 
81 	list_for_each_entry(field, head, link) {
82 		if (!strcmp(field->name, name))
83 			return field;
84 	}
85 
86 	return NULL;
87 }
88 
89 struct ftrace_event_field *
90 trace_find_event_field(struct trace_event_call *call, char *name)
91 {
92 	struct ftrace_event_field *field;
93 	struct list_head *head;
94 
95 	head = trace_get_fields(call);
96 	field = __find_event_field(head, name);
97 	if (field)
98 		return field;
99 
100 	field = __find_event_field(&ftrace_generic_fields, name);
101 	if (field)
102 		return field;
103 
104 	return __find_event_field(&ftrace_common_fields, name);
105 }
106 
107 static int __trace_define_field(struct list_head *head, const char *type,
108 				const char *name, int offset, int size,
109 				int is_signed, int filter_type)
110 {
111 	struct ftrace_event_field *field;
112 
113 	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
114 	if (!field)
115 		return -ENOMEM;
116 
117 	field->name = name;
118 	field->type = type;
119 
120 	if (filter_type == FILTER_OTHER)
121 		field->filter_type = filter_assign_type(type);
122 	else
123 		field->filter_type = filter_type;
124 
125 	field->offset = offset;
126 	field->size = size;
127 	field->is_signed = is_signed;
128 
129 	list_add(&field->link, head);
130 
131 	return 0;
132 }
133 
134 int trace_define_field(struct trace_event_call *call, const char *type,
135 		       const char *name, int offset, int size, int is_signed,
136 		       int filter_type)
137 {
138 	struct list_head *head;
139 
140 	if (WARN_ON(!call->class))
141 		return 0;
142 
143 	head = trace_get_fields(call);
144 	return __trace_define_field(head, type, name, offset, size,
145 				    is_signed, filter_type);
146 }
147 EXPORT_SYMBOL_GPL(trace_define_field);
148 
149 #define __generic_field(type, item, filter_type)			\
150 	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
151 				   #item, 0, 0, is_signed_type(type),	\
152 				   filter_type);			\
153 	if (ret)							\
154 		return ret;
155 
156 #define __common_field(type, item)					\
157 	ret = __trace_define_field(&ftrace_common_fields, #type,	\
158 				   "common_" #item,			\
159 				   offsetof(typeof(ent), item),		\
160 				   sizeof(ent.item),			\
161 				   is_signed_type(type), FILTER_OTHER);	\
162 	if (ret)							\
163 		return ret;
164 
165 static int trace_define_generic_fields(void)
166 {
167 	int ret;
168 
169 	__generic_field(int, CPU, FILTER_CPU);
170 	__generic_field(int, cpu, FILTER_CPU);
171 	__generic_field(char *, COMM, FILTER_COMM);
172 	__generic_field(char *, comm, FILTER_COMM);
173 
174 	return ret;
175 }
176 
177 static int trace_define_common_fields(void)
178 {
179 	int ret;
180 	struct trace_entry ent;
181 
182 	__common_field(unsigned short, type);
183 	__common_field(unsigned char, flags);
184 	/* Holds both preempt_count and migrate_disable */
185 	__common_field(unsigned char, preempt_count);
186 	__common_field(int, pid);
187 
188 	return ret;
189 }
190 
191 static void trace_destroy_fields(struct trace_event_call *call)
192 {
193 	struct ftrace_event_field *field, *next;
194 	struct list_head *head;
195 
196 	head = trace_get_fields(call);
197 	list_for_each_entry_safe(field, next, head, link) {
198 		list_del(&field->link);
199 		kmem_cache_free(field_cachep, field);
200 	}
201 }
202 
203 /*
204  * run-time version of trace_event_get_offsets_<call>() that returns the last
205  * accessible offset of trace fields excluding __dynamic_array bytes
206  */
207 int trace_event_get_offsets(struct trace_event_call *call)
208 {
209 	struct ftrace_event_field *tail;
210 	struct list_head *head;
211 
212 	head = trace_get_fields(call);
213 	/*
214 	 * head->next points to the last field with the largest offset,
215 	 * since it was added last by trace_define_field()
216 	 */
217 	tail = list_first_entry(head, struct ftrace_event_field, link);
218 	return tail->offset + tail->size;
219 }
220 
221 /*
222  * Check if the referenced field is an array and return true,
223  * as arrays are OK to dereference.
224  */
225 static bool test_field(const char *fmt, struct trace_event_call *call)
226 {
227 	struct trace_event_fields *field = call->class->fields_array;
228 	const char *array_descriptor;
229 	const char *p = fmt;
230 	int len;
231 
232 	if (!(len = str_has_prefix(fmt, "REC->")))
233 		return false;
234 	fmt += len;
235 	for (p = fmt; *p; p++) {
236 		if (!isalnum(*p) && *p != '_')
237 			break;
238 	}
239 	len = p - fmt;
240 
241 	for (; field->type; field++) {
242 		if (strncmp(field->name, fmt, len) ||
243 		    field->name[len])
244 			continue;
245 		array_descriptor = strchr(field->type, '[');
246 		/* This is an array and is OK to dereference. */
247 		return array_descriptor != NULL;
248 	}
249 	return false;
250 }
251 
252 /*
253  * Examine the print fmt of the event looking for unsafe dereference
254  * pointers using %p* that could be recorded in the trace event and
255  * much later referenced after the pointer was freed. Dereferencing
256  * pointers are OK, if it is dereferenced into the event itself.
257  */
258 static void test_event_printk(struct trace_event_call *call)
259 {
260 	u64 dereference_flags = 0;
261 	bool first = true;
262 	const char *fmt, *c, *r, *a;
263 	int parens = 0;
264 	char in_quote = 0;
265 	int start_arg = 0;
266 	int arg = 0;
267 	int i;
268 
269 	fmt = call->print_fmt;
270 
271 	if (!fmt)
272 		return;
273 
274 	for (i = 0; fmt[i]; i++) {
275 		switch (fmt[i]) {
276 		case '\\':
277 			i++;
278 			if (!fmt[i])
279 				return;
280 			continue;
281 		case '"':
282 		case '\'':
283 			/*
284 			 * The print fmt starts with a string that
285 			 * is processed first to find %p* usage,
286 			 * then after the first string, the print fmt
287 			 * contains arguments that are used to check
288 			 * if the dereferenced %p* usage is safe.
289 			 */
290 			if (first) {
291 				if (fmt[i] == '\'')
292 					continue;
293 				if (in_quote) {
294 					arg = 0;
295 					first = false;
296 					/*
297 					 * If there was no %p* uses
298 					 * the fmt is OK.
299 					 */
300 					if (!dereference_flags)
301 						return;
302 				}
303 			}
304 			if (in_quote) {
305 				if (in_quote == fmt[i])
306 					in_quote = 0;
307 			} else {
308 				in_quote = fmt[i];
309 			}
310 			continue;
311 		case '%':
312 			if (!first || !in_quote)
313 				continue;
314 			i++;
315 			if (!fmt[i])
316 				return;
317 			switch (fmt[i]) {
318 			case '%':
319 				continue;
320 			case 'p':
321 				/* Find dereferencing fields */
322 				switch (fmt[i + 1]) {
323 				case 'B': case 'R': case 'r':
324 				case 'b': case 'M': case 'm':
325 				case 'I': case 'i': case 'E':
326 				case 'U': case 'V': case 'N':
327 				case 'a': case 'd': case 'D':
328 				case 'g': case 't': case 'C':
329 				case 'O': case 'f':
330 					if (WARN_ONCE(arg == 63,
331 						      "Too many args for event: %s",
332 						      trace_event_name(call)))
333 						return;
334 					dereference_flags |= 1ULL << arg;
335 				}
336 				break;
337 			default:
338 			{
339 				bool star = false;
340 				int j;
341 
342 				/* Increment arg if %*s exists. */
343 				for (j = 0; fmt[i + j]; j++) {
344 					if (isdigit(fmt[i + j]) ||
345 					    fmt[i + j] == '.')
346 						continue;
347 					if (fmt[i + j] == '*') {
348 						star = true;
349 						continue;
350 					}
351 					if ((fmt[i + j] == 's') && star)
352 						arg++;
353 					break;
354 				}
355 				break;
356 			} /* default */
357 
358 			} /* switch */
359 			arg++;
360 			continue;
361 		case '(':
362 			if (in_quote)
363 				continue;
364 			parens++;
365 			continue;
366 		case ')':
367 			if (in_quote)
368 				continue;
369 			parens--;
370 			if (WARN_ONCE(parens < 0,
371 				      "Paren mismatch for event: %s\narg='%s'\n%*s",
372 				      trace_event_name(call),
373 				      fmt + start_arg,
374 				      (i - start_arg) + 5, "^"))
375 				return;
376 			continue;
377 		case ',':
378 			if (in_quote || parens)
379 				continue;
380 			i++;
381 			while (isspace(fmt[i]))
382 				i++;
383 			start_arg = i;
384 			if (!(dereference_flags & (1ULL << arg)))
385 				goto next_arg;
386 
387 			/* Find the REC-> in the argument */
388 			c = strchr(fmt + i, ',');
389 			r = strstr(fmt + i, "REC->");
390 			if (r && (!c || r < c)) {
391 				/*
392 				 * Addresses of events on the buffer,
393 				 * or an array on the buffer is
394 				 * OK to dereference.
395 				 * There's ways to fool this, but
396 				 * this is to catch common mistakes,
397 				 * not malicious code.
398 				 */
399 				a = strchr(fmt + i, '&');
400 				if ((a && (a < r)) || test_field(r, call))
401 					dereference_flags &= ~(1ULL << arg);
402 			}
403 		next_arg:
404 			i--;
405 			arg++;
406 		}
407 	}
408 
409 	/*
410 	 * If you triggered the below warning, the trace event reported
411 	 * uses an unsafe dereference pointer %p*. As the data stored
412 	 * at the trace event time may no longer exist when the trace
413 	 * event is printed, dereferencing to the original source is
414 	 * unsafe. The source of the dereference must be copied into the
415 	 * event itself, and the dereference must access the copy instead.
416 	 */
417 	if (WARN_ON_ONCE(dereference_flags)) {
418 		arg = 1;
419 		while (!(dereference_flags & 1)) {
420 			dereference_flags >>= 1;
421 			arg++;
422 		}
423 		pr_warn("event %s has unsafe dereference of argument %d\n",
424 			trace_event_name(call), arg);
425 		pr_warn("print_fmt: %s\n", fmt);
426 	}
427 }
428 
429 int trace_event_raw_init(struct trace_event_call *call)
430 {
431 	int id;
432 
433 	id = register_trace_event(&call->event);
434 	if (!id)
435 		return -ENODEV;
436 
437 	test_event_printk(call);
438 
439 	return 0;
440 }
441 EXPORT_SYMBOL_GPL(trace_event_raw_init);
442 
443 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
444 {
445 	struct trace_array *tr = trace_file->tr;
446 	struct trace_array_cpu *data;
447 	struct trace_pid_list *no_pid_list;
448 	struct trace_pid_list *pid_list;
449 
450 	pid_list = rcu_dereference_raw(tr->filtered_pids);
451 	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
452 
453 	if (!pid_list && !no_pid_list)
454 		return false;
455 
456 	data = this_cpu_ptr(tr->array_buffer.data);
457 
458 	return data->ignore_pid;
459 }
460 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
461 
462 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
463 				 struct trace_event_file *trace_file,
464 				 unsigned long len)
465 {
466 	struct trace_event_call *event_call = trace_file->event_call;
467 
468 	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
469 	    trace_event_ignore_this_pid(trace_file))
470 		return NULL;
471 
472 	/*
473 	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
474 	 * preemption (adding one to the preempt_count). Since we are
475 	 * interested in the preempt_count at the time the tracepoint was
476 	 * hit, we need to subtract one to offset the increment.
477 	 */
478 	fbuffer->trace_ctx = tracing_gen_ctx_dec();
479 	fbuffer->trace_file = trace_file;
480 
481 	fbuffer->event =
482 		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
483 						event_call->event.type, len,
484 						fbuffer->trace_ctx);
485 	if (!fbuffer->event)
486 		return NULL;
487 
488 	fbuffer->regs = NULL;
489 	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
490 	return fbuffer->entry;
491 }
492 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
493 
494 int trace_event_reg(struct trace_event_call *call,
495 		    enum trace_reg type, void *data)
496 {
497 	struct trace_event_file *file = data;
498 
499 	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
500 	switch (type) {
501 	case TRACE_REG_REGISTER:
502 		return tracepoint_probe_register(call->tp,
503 						 call->class->probe,
504 						 file);
505 	case TRACE_REG_UNREGISTER:
506 		tracepoint_probe_unregister(call->tp,
507 					    call->class->probe,
508 					    file);
509 		return 0;
510 
511 #ifdef CONFIG_PERF_EVENTS
512 	case TRACE_REG_PERF_REGISTER:
513 		return tracepoint_probe_register(call->tp,
514 						 call->class->perf_probe,
515 						 call);
516 	case TRACE_REG_PERF_UNREGISTER:
517 		tracepoint_probe_unregister(call->tp,
518 					    call->class->perf_probe,
519 					    call);
520 		return 0;
521 	case TRACE_REG_PERF_OPEN:
522 	case TRACE_REG_PERF_CLOSE:
523 	case TRACE_REG_PERF_ADD:
524 	case TRACE_REG_PERF_DEL:
525 		return 0;
526 #endif
527 	}
528 	return 0;
529 }
530 EXPORT_SYMBOL_GPL(trace_event_reg);
531 
532 void trace_event_enable_cmd_record(bool enable)
533 {
534 	struct trace_event_file *file;
535 	struct trace_array *tr;
536 
537 	lockdep_assert_held(&event_mutex);
538 
539 	do_for_each_event_file(tr, file) {
540 
541 		if (!(file->flags & EVENT_FILE_FL_ENABLED))
542 			continue;
543 
544 		if (enable) {
545 			tracing_start_cmdline_record();
546 			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
547 		} else {
548 			tracing_stop_cmdline_record();
549 			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
550 		}
551 	} while_for_each_event_file();
552 }
553 
554 void trace_event_enable_tgid_record(bool enable)
555 {
556 	struct trace_event_file *file;
557 	struct trace_array *tr;
558 
559 	lockdep_assert_held(&event_mutex);
560 
561 	do_for_each_event_file(tr, file) {
562 		if (!(file->flags & EVENT_FILE_FL_ENABLED))
563 			continue;
564 
565 		if (enable) {
566 			tracing_start_tgid_record();
567 			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
568 		} else {
569 			tracing_stop_tgid_record();
570 			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
571 				  &file->flags);
572 		}
573 	} while_for_each_event_file();
574 }
575 
576 static int __ftrace_event_enable_disable(struct trace_event_file *file,
577 					 int enable, int soft_disable)
578 {
579 	struct trace_event_call *call = file->event_call;
580 	struct trace_array *tr = file->tr;
581 	unsigned long file_flags = file->flags;
582 	int ret = 0;
583 	int disable;
584 
585 	switch (enable) {
586 	case 0:
587 		/*
588 		 * When soft_disable is set and enable is cleared, the sm_ref
589 		 * reference counter is decremented. If it reaches 0, we want
590 		 * to clear the SOFT_DISABLED flag but leave the event in the
591 		 * state that it was. That is, if the event was enabled and
592 		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
593 		 * is set we do not want the event to be enabled before we
594 		 * clear the bit.
595 		 *
596 		 * When soft_disable is not set but the SOFT_MODE flag is,
597 		 * we do nothing. Do not disable the tracepoint, otherwise
598 		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
599 		 */
600 		if (soft_disable) {
601 			if (atomic_dec_return(&file->sm_ref) > 0)
602 				break;
603 			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
604 			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
605 		} else
606 			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
607 
608 		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
609 			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
610 			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
611 				tracing_stop_cmdline_record();
612 				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
613 			}
614 
615 			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
616 				tracing_stop_tgid_record();
617 				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
618 			}
619 
620 			call->class->reg(call, TRACE_REG_UNREGISTER, file);
621 		}
622 		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
623 		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
624 			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
625 		else
626 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
627 		break;
628 	case 1:
629 		/*
630 		 * When soft_disable is set and enable is set, we want to
631 		 * register the tracepoint for the event, but leave the event
632 		 * as is. That means, if the event was already enabled, we do
633 		 * nothing (but set SOFT_MODE). If the event is disabled, we
634 		 * set SOFT_DISABLED before enabling the event tracepoint, so
635 		 * it still seems to be disabled.
636 		 */
637 		if (!soft_disable)
638 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
639 		else {
640 			if (atomic_inc_return(&file->sm_ref) > 1)
641 				break;
642 			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
643 		}
644 
645 		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
646 			bool cmd = false, tgid = false;
647 
648 			/* Keep the event disabled, when going to SOFT_MODE. */
649 			if (soft_disable)
650 				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
651 
652 			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
653 				cmd = true;
654 				tracing_start_cmdline_record();
655 				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
656 			}
657 
658 			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
659 				tgid = true;
660 				tracing_start_tgid_record();
661 				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
662 			}
663 
664 			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
665 			if (ret) {
666 				if (cmd)
667 					tracing_stop_cmdline_record();
668 				if (tgid)
669 					tracing_stop_tgid_record();
670 				pr_info("event trace: Could not enable event "
671 					"%s\n", trace_event_name(call));
672 				break;
673 			}
674 			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
675 
676 			/* WAS_ENABLED gets set but never cleared. */
677 			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
678 		}
679 		break;
680 	}
681 
682 	/* Enable or disable use of trace_buffered_event */
683 	if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
684 	    (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
685 		if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
686 			trace_buffered_event_enable();
687 		else
688 			trace_buffered_event_disable();
689 	}
690 
691 	return ret;
692 }
693 
694 int trace_event_enable_disable(struct trace_event_file *file,
695 			       int enable, int soft_disable)
696 {
697 	return __ftrace_event_enable_disable(file, enable, soft_disable);
698 }
699 
700 static int ftrace_event_enable_disable(struct trace_event_file *file,
701 				       int enable)
702 {
703 	return __ftrace_event_enable_disable(file, enable, 0);
704 }
705 
706 static void ftrace_clear_events(struct trace_array *tr)
707 {
708 	struct trace_event_file *file;
709 
710 	mutex_lock(&event_mutex);
711 	list_for_each_entry(file, &tr->events, list) {
712 		ftrace_event_enable_disable(file, 0);
713 	}
714 	mutex_unlock(&event_mutex);
715 }
716 
717 static void
718 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
719 {
720 	struct trace_pid_list *pid_list;
721 	struct trace_array *tr = data;
722 
723 	pid_list = rcu_dereference_raw(tr->filtered_pids);
724 	trace_filter_add_remove_task(pid_list, NULL, task);
725 
726 	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
727 	trace_filter_add_remove_task(pid_list, NULL, task);
728 }
729 
730 static void
731 event_filter_pid_sched_process_fork(void *data,
732 				    struct task_struct *self,
733 				    struct task_struct *task)
734 {
735 	struct trace_pid_list *pid_list;
736 	struct trace_array *tr = data;
737 
738 	pid_list = rcu_dereference_sched(tr->filtered_pids);
739 	trace_filter_add_remove_task(pid_list, self, task);
740 
741 	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
742 	trace_filter_add_remove_task(pid_list, self, task);
743 }
744 
745 void trace_event_follow_fork(struct trace_array *tr, bool enable)
746 {
747 	if (enable) {
748 		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
749 						       tr, INT_MIN);
750 		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
751 						       tr, INT_MAX);
752 	} else {
753 		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
754 						    tr);
755 		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
756 						    tr);
757 	}
758 }
759 
760 static void
761 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
762 		    struct task_struct *prev, struct task_struct *next)
763 {
764 	struct trace_array *tr = data;
765 	struct trace_pid_list *no_pid_list;
766 	struct trace_pid_list *pid_list;
767 	bool ret;
768 
769 	pid_list = rcu_dereference_sched(tr->filtered_pids);
770 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
771 
772 	/*
773 	 * Sched switch is funny, as we only want to ignore it
774 	 * in the notrace case if both prev and next should be ignored.
775 	 */
776 	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
777 		trace_ignore_this_task(NULL, no_pid_list, next);
778 
779 	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
780 		       (trace_ignore_this_task(pid_list, NULL, prev) &&
781 			trace_ignore_this_task(pid_list, NULL, next)));
782 }
783 
784 static void
785 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
786 		    struct task_struct *prev, struct task_struct *next)
787 {
788 	struct trace_array *tr = data;
789 	struct trace_pid_list *no_pid_list;
790 	struct trace_pid_list *pid_list;
791 
792 	pid_list = rcu_dereference_sched(tr->filtered_pids);
793 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
794 
795 	this_cpu_write(tr->array_buffer.data->ignore_pid,
796 		       trace_ignore_this_task(pid_list, no_pid_list, next));
797 }
798 
799 static void
800 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
801 {
802 	struct trace_array *tr = data;
803 	struct trace_pid_list *no_pid_list;
804 	struct trace_pid_list *pid_list;
805 
806 	/* Nothing to do if we are already tracing */
807 	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
808 		return;
809 
810 	pid_list = rcu_dereference_sched(tr->filtered_pids);
811 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
812 
813 	this_cpu_write(tr->array_buffer.data->ignore_pid,
814 		       trace_ignore_this_task(pid_list, no_pid_list, task));
815 }
816 
817 static void
818 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
819 {
820 	struct trace_array *tr = data;
821 	struct trace_pid_list *no_pid_list;
822 	struct trace_pid_list *pid_list;
823 
824 	/* Nothing to do if we are not tracing */
825 	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
826 		return;
827 
828 	pid_list = rcu_dereference_sched(tr->filtered_pids);
829 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
830 
831 	/* Set tracing if current is enabled */
832 	this_cpu_write(tr->array_buffer.data->ignore_pid,
833 		       trace_ignore_this_task(pid_list, no_pid_list, current));
834 }
835 
836 static void unregister_pid_events(struct trace_array *tr)
837 {
838 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
839 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
840 
841 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
842 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
843 
844 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
845 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
846 
847 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
848 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
849 }
850 
851 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
852 {
853 	struct trace_pid_list *pid_list;
854 	struct trace_pid_list *no_pid_list;
855 	struct trace_event_file *file;
856 	int cpu;
857 
858 	pid_list = rcu_dereference_protected(tr->filtered_pids,
859 					     lockdep_is_held(&event_mutex));
860 	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
861 					     lockdep_is_held(&event_mutex));
862 
863 	/* Make sure there's something to do */
864 	if (!pid_type_enabled(type, pid_list, no_pid_list))
865 		return;
866 
867 	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
868 		unregister_pid_events(tr);
869 
870 		list_for_each_entry(file, &tr->events, list) {
871 			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
872 		}
873 
874 		for_each_possible_cpu(cpu)
875 			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
876 	}
877 
878 	if (type & TRACE_PIDS)
879 		rcu_assign_pointer(tr->filtered_pids, NULL);
880 
881 	if (type & TRACE_NO_PIDS)
882 		rcu_assign_pointer(tr->filtered_no_pids, NULL);
883 
884 	/* Wait till all users are no longer using pid filtering */
885 	tracepoint_synchronize_unregister();
886 
887 	if ((type & TRACE_PIDS) && pid_list)
888 		trace_pid_list_free(pid_list);
889 
890 	if ((type & TRACE_NO_PIDS) && no_pid_list)
891 		trace_pid_list_free(no_pid_list);
892 }
893 
894 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
895 {
896 	mutex_lock(&event_mutex);
897 	__ftrace_clear_event_pids(tr, type);
898 	mutex_unlock(&event_mutex);
899 }
900 
901 static void __put_system(struct event_subsystem *system)
902 {
903 	struct event_filter *filter = system->filter;
904 
905 	WARN_ON_ONCE(system_refcount(system) == 0);
906 	if (system_refcount_dec(system))
907 		return;
908 
909 	list_del(&system->list);
910 
911 	if (filter) {
912 		kfree(filter->filter_string);
913 		kfree(filter);
914 	}
915 	kfree_const(system->name);
916 	kfree(system);
917 }
918 
919 static void __get_system(struct event_subsystem *system)
920 {
921 	WARN_ON_ONCE(system_refcount(system) == 0);
922 	system_refcount_inc(system);
923 }
924 
925 static void __get_system_dir(struct trace_subsystem_dir *dir)
926 {
927 	WARN_ON_ONCE(dir->ref_count == 0);
928 	dir->ref_count++;
929 	__get_system(dir->subsystem);
930 }
931 
932 static void __put_system_dir(struct trace_subsystem_dir *dir)
933 {
934 	WARN_ON_ONCE(dir->ref_count == 0);
935 	/* If the subsystem is about to be freed, the dir must be too */
936 	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
937 
938 	__put_system(dir->subsystem);
939 	if (!--dir->ref_count)
940 		kfree(dir);
941 }
942 
943 static void put_system(struct trace_subsystem_dir *dir)
944 {
945 	mutex_lock(&event_mutex);
946 	__put_system_dir(dir);
947 	mutex_unlock(&event_mutex);
948 }
949 
950 static void remove_subsystem(struct trace_subsystem_dir *dir)
951 {
952 	if (!dir)
953 		return;
954 
955 	if (!--dir->nr_events) {
956 		tracefs_remove(dir->entry);
957 		list_del(&dir->list);
958 		__put_system_dir(dir);
959 	}
960 }
961 
962 static void remove_event_file_dir(struct trace_event_file *file)
963 {
964 	struct dentry *dir = file->dir;
965 	struct dentry *child;
966 
967 	if (dir) {
968 		spin_lock(&dir->d_lock);	/* probably unneeded */
969 		list_for_each_entry(child, &dir->d_subdirs, d_child) {
970 			if (d_really_is_positive(child))	/* probably unneeded */
971 				d_inode(child)->i_private = NULL;
972 		}
973 		spin_unlock(&dir->d_lock);
974 
975 		tracefs_remove(dir);
976 	}
977 
978 	list_del(&file->list);
979 	remove_subsystem(file->system);
980 	free_event_filter(file->filter);
981 	kmem_cache_free(file_cachep, file);
982 }
983 
984 /*
985  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
986  */
987 static int
988 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
989 			      const char *sub, const char *event, int set)
990 {
991 	struct trace_event_file *file;
992 	struct trace_event_call *call;
993 	const char *name;
994 	int ret = -EINVAL;
995 	int eret = 0;
996 
997 	list_for_each_entry(file, &tr->events, list) {
998 
999 		call = file->event_call;
1000 		name = trace_event_name(call);
1001 
1002 		if (!name || !call->class || !call->class->reg)
1003 			continue;
1004 
1005 		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1006 			continue;
1007 
1008 		if (match &&
1009 		    strcmp(match, name) != 0 &&
1010 		    strcmp(match, call->class->system) != 0)
1011 			continue;
1012 
1013 		if (sub && strcmp(sub, call->class->system) != 0)
1014 			continue;
1015 
1016 		if (event && strcmp(event, name) != 0)
1017 			continue;
1018 
1019 		ret = ftrace_event_enable_disable(file, set);
1020 
1021 		/*
1022 		 * Save the first error and return that. Some events
1023 		 * may still have been enabled, but let the user
1024 		 * know that something went wrong.
1025 		 */
1026 		if (ret && !eret)
1027 			eret = ret;
1028 
1029 		ret = eret;
1030 	}
1031 
1032 	return ret;
1033 }
1034 
1035 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1036 				  const char *sub, const char *event, int set)
1037 {
1038 	int ret;
1039 
1040 	mutex_lock(&event_mutex);
1041 	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1042 	mutex_unlock(&event_mutex);
1043 
1044 	return ret;
1045 }
1046 
1047 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1048 {
1049 	char *event = NULL, *sub = NULL, *match;
1050 	int ret;
1051 
1052 	if (!tr)
1053 		return -ENOENT;
1054 	/*
1055 	 * The buf format can be <subsystem>:<event-name>
1056 	 *  *:<event-name> means any event by that name.
1057 	 *  :<event-name> is the same.
1058 	 *
1059 	 *  <subsystem>:* means all events in that subsystem
1060 	 *  <subsystem>: means the same.
1061 	 *
1062 	 *  <name> (no ':') means all events in a subsystem with
1063 	 *  the name <name> or any event that matches <name>
1064 	 */
1065 
1066 	match = strsep(&buf, ":");
1067 	if (buf) {
1068 		sub = match;
1069 		event = buf;
1070 		match = NULL;
1071 
1072 		if (!strlen(sub) || strcmp(sub, "*") == 0)
1073 			sub = NULL;
1074 		if (!strlen(event) || strcmp(event, "*") == 0)
1075 			event = NULL;
1076 	}
1077 
1078 	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1079 
1080 	/* Put back the colon to allow this to be called again */
1081 	if (buf)
1082 		*(buf - 1) = ':';
1083 
1084 	return ret;
1085 }
1086 
1087 /**
1088  * trace_set_clr_event - enable or disable an event
1089  * @system: system name to match (NULL for any system)
1090  * @event: event name to match (NULL for all events, within system)
1091  * @set: 1 to enable, 0 to disable
1092  *
1093  * This is a way for other parts of the kernel to enable or disable
1094  * event recording.
1095  *
1096  * Returns 0 on success, -EINVAL if the parameters do not match any
1097  * registered events.
1098  */
1099 int trace_set_clr_event(const char *system, const char *event, int set)
1100 {
1101 	struct trace_array *tr = top_trace_array();
1102 
1103 	if (!tr)
1104 		return -ENODEV;
1105 
1106 	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1107 }
1108 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1109 
1110 /**
1111  * trace_array_set_clr_event - enable or disable an event for a trace array.
1112  * @tr: concerned trace array.
1113  * @system: system name to match (NULL for any system)
1114  * @event: event name to match (NULL for all events, within system)
1115  * @enable: true to enable, false to disable
1116  *
1117  * This is a way for other parts of the kernel to enable or disable
1118  * event recording.
1119  *
1120  * Returns 0 on success, -EINVAL if the parameters do not match any
1121  * registered events.
1122  */
1123 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1124 		const char *event, bool enable)
1125 {
1126 	int set;
1127 
1128 	if (!tr)
1129 		return -ENOENT;
1130 
1131 	set = (enable == true) ? 1 : 0;
1132 	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1133 }
1134 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1135 
1136 /* 128 should be much more than enough */
1137 #define EVENT_BUF_SIZE		127
1138 
1139 static ssize_t
1140 ftrace_event_write(struct file *file, const char __user *ubuf,
1141 		   size_t cnt, loff_t *ppos)
1142 {
1143 	struct trace_parser parser;
1144 	struct seq_file *m = file->private_data;
1145 	struct trace_array *tr = m->private;
1146 	ssize_t read, ret;
1147 
1148 	if (!cnt)
1149 		return 0;
1150 
1151 	ret = tracing_update_buffers();
1152 	if (ret < 0)
1153 		return ret;
1154 
1155 	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1156 		return -ENOMEM;
1157 
1158 	read = trace_get_user(&parser, ubuf, cnt, ppos);
1159 
1160 	if (read >= 0 && trace_parser_loaded((&parser))) {
1161 		int set = 1;
1162 
1163 		if (*parser.buffer == '!')
1164 			set = 0;
1165 
1166 		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1167 		if (ret)
1168 			goto out_put;
1169 	}
1170 
1171 	ret = read;
1172 
1173  out_put:
1174 	trace_parser_put(&parser);
1175 
1176 	return ret;
1177 }
1178 
1179 static void *
1180 t_next(struct seq_file *m, void *v, loff_t *pos)
1181 {
1182 	struct trace_event_file *file = v;
1183 	struct trace_event_call *call;
1184 	struct trace_array *tr = m->private;
1185 
1186 	(*pos)++;
1187 
1188 	list_for_each_entry_continue(file, &tr->events, list) {
1189 		call = file->event_call;
1190 		/*
1191 		 * The ftrace subsystem is for showing formats only.
1192 		 * They can not be enabled or disabled via the event files.
1193 		 */
1194 		if (call->class && call->class->reg &&
1195 		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1196 			return file;
1197 	}
1198 
1199 	return NULL;
1200 }
1201 
1202 static void *t_start(struct seq_file *m, loff_t *pos)
1203 {
1204 	struct trace_event_file *file;
1205 	struct trace_array *tr = m->private;
1206 	loff_t l;
1207 
1208 	mutex_lock(&event_mutex);
1209 
1210 	file = list_entry(&tr->events, struct trace_event_file, list);
1211 	for (l = 0; l <= *pos; ) {
1212 		file = t_next(m, file, &l);
1213 		if (!file)
1214 			break;
1215 	}
1216 	return file;
1217 }
1218 
1219 static void *
1220 s_next(struct seq_file *m, void *v, loff_t *pos)
1221 {
1222 	struct trace_event_file *file = v;
1223 	struct trace_array *tr = m->private;
1224 
1225 	(*pos)++;
1226 
1227 	list_for_each_entry_continue(file, &tr->events, list) {
1228 		if (file->flags & EVENT_FILE_FL_ENABLED)
1229 			return file;
1230 	}
1231 
1232 	return NULL;
1233 }
1234 
1235 static void *s_start(struct seq_file *m, loff_t *pos)
1236 {
1237 	struct trace_event_file *file;
1238 	struct trace_array *tr = m->private;
1239 	loff_t l;
1240 
1241 	mutex_lock(&event_mutex);
1242 
1243 	file = list_entry(&tr->events, struct trace_event_file, list);
1244 	for (l = 0; l <= *pos; ) {
1245 		file = s_next(m, file, &l);
1246 		if (!file)
1247 			break;
1248 	}
1249 	return file;
1250 }
1251 
1252 static int t_show(struct seq_file *m, void *v)
1253 {
1254 	struct trace_event_file *file = v;
1255 	struct trace_event_call *call = file->event_call;
1256 
1257 	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1258 		seq_printf(m, "%s:", call->class->system);
1259 	seq_printf(m, "%s\n", trace_event_name(call));
1260 
1261 	return 0;
1262 }
1263 
1264 static void t_stop(struct seq_file *m, void *p)
1265 {
1266 	mutex_unlock(&event_mutex);
1267 }
1268 
1269 static void *
1270 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1271 {
1272 	struct trace_array *tr = m->private;
1273 	struct trace_pid_list *pid_list;
1274 
1275 	if (type == TRACE_PIDS)
1276 		pid_list = rcu_dereference_sched(tr->filtered_pids);
1277 	else
1278 		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1279 
1280 	return trace_pid_next(pid_list, v, pos);
1281 }
1282 
1283 static void *
1284 p_next(struct seq_file *m, void *v, loff_t *pos)
1285 {
1286 	return __next(m, v, pos, TRACE_PIDS);
1287 }
1288 
1289 static void *
1290 np_next(struct seq_file *m, void *v, loff_t *pos)
1291 {
1292 	return __next(m, v, pos, TRACE_NO_PIDS);
1293 }
1294 
1295 static void *__start(struct seq_file *m, loff_t *pos, int type)
1296 	__acquires(RCU)
1297 {
1298 	struct trace_pid_list *pid_list;
1299 	struct trace_array *tr = m->private;
1300 
1301 	/*
1302 	 * Grab the mutex, to keep calls to p_next() having the same
1303 	 * tr->filtered_pids as p_start() has.
1304 	 * If we just passed the tr->filtered_pids around, then RCU would
1305 	 * have been enough, but doing that makes things more complex.
1306 	 */
1307 	mutex_lock(&event_mutex);
1308 	rcu_read_lock_sched();
1309 
1310 	if (type == TRACE_PIDS)
1311 		pid_list = rcu_dereference_sched(tr->filtered_pids);
1312 	else
1313 		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1314 
1315 	if (!pid_list)
1316 		return NULL;
1317 
1318 	return trace_pid_start(pid_list, pos);
1319 }
1320 
1321 static void *p_start(struct seq_file *m, loff_t *pos)
1322 	__acquires(RCU)
1323 {
1324 	return __start(m, pos, TRACE_PIDS);
1325 }
1326 
1327 static void *np_start(struct seq_file *m, loff_t *pos)
1328 	__acquires(RCU)
1329 {
1330 	return __start(m, pos, TRACE_NO_PIDS);
1331 }
1332 
1333 static void p_stop(struct seq_file *m, void *p)
1334 	__releases(RCU)
1335 {
1336 	rcu_read_unlock_sched();
1337 	mutex_unlock(&event_mutex);
1338 }
1339 
1340 static ssize_t
1341 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1342 		  loff_t *ppos)
1343 {
1344 	struct trace_event_file *file;
1345 	unsigned long flags;
1346 	char buf[4] = "0";
1347 
1348 	mutex_lock(&event_mutex);
1349 	file = event_file_data(filp);
1350 	if (likely(file))
1351 		flags = file->flags;
1352 	mutex_unlock(&event_mutex);
1353 
1354 	if (!file)
1355 		return -ENODEV;
1356 
1357 	if (flags & EVENT_FILE_FL_ENABLED &&
1358 	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1359 		strcpy(buf, "1");
1360 
1361 	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1362 	    flags & EVENT_FILE_FL_SOFT_MODE)
1363 		strcat(buf, "*");
1364 
1365 	strcat(buf, "\n");
1366 
1367 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1368 }
1369 
1370 static ssize_t
1371 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1372 		   loff_t *ppos)
1373 {
1374 	struct trace_event_file *file;
1375 	unsigned long val;
1376 	int ret;
1377 
1378 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1379 	if (ret)
1380 		return ret;
1381 
1382 	ret = tracing_update_buffers();
1383 	if (ret < 0)
1384 		return ret;
1385 
1386 	switch (val) {
1387 	case 0:
1388 	case 1:
1389 		ret = -ENODEV;
1390 		mutex_lock(&event_mutex);
1391 		file = event_file_data(filp);
1392 		if (likely(file))
1393 			ret = ftrace_event_enable_disable(file, val);
1394 		mutex_unlock(&event_mutex);
1395 		break;
1396 
1397 	default:
1398 		return -EINVAL;
1399 	}
1400 
1401 	*ppos += cnt;
1402 
1403 	return ret ? ret : cnt;
1404 }
1405 
1406 static ssize_t
1407 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1408 		   loff_t *ppos)
1409 {
1410 	const char set_to_char[4] = { '?', '0', '1', 'X' };
1411 	struct trace_subsystem_dir *dir = filp->private_data;
1412 	struct event_subsystem *system = dir->subsystem;
1413 	struct trace_event_call *call;
1414 	struct trace_event_file *file;
1415 	struct trace_array *tr = dir->tr;
1416 	char buf[2];
1417 	int set = 0;
1418 	int ret;
1419 
1420 	mutex_lock(&event_mutex);
1421 	list_for_each_entry(file, &tr->events, list) {
1422 		call = file->event_call;
1423 		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1424 		    !trace_event_name(call) || !call->class || !call->class->reg)
1425 			continue;
1426 
1427 		if (system && strcmp(call->class->system, system->name) != 0)
1428 			continue;
1429 
1430 		/*
1431 		 * We need to find out if all the events are set
1432 		 * or if all events or cleared, or if we have
1433 		 * a mixture.
1434 		 */
1435 		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1436 
1437 		/*
1438 		 * If we have a mixture, no need to look further.
1439 		 */
1440 		if (set == 3)
1441 			break;
1442 	}
1443 	mutex_unlock(&event_mutex);
1444 
1445 	buf[0] = set_to_char[set];
1446 	buf[1] = '\n';
1447 
1448 	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1449 
1450 	return ret;
1451 }
1452 
1453 static ssize_t
1454 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1455 		    loff_t *ppos)
1456 {
1457 	struct trace_subsystem_dir *dir = filp->private_data;
1458 	struct event_subsystem *system = dir->subsystem;
1459 	const char *name = NULL;
1460 	unsigned long val;
1461 	ssize_t ret;
1462 
1463 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1464 	if (ret)
1465 		return ret;
1466 
1467 	ret = tracing_update_buffers();
1468 	if (ret < 0)
1469 		return ret;
1470 
1471 	if (val != 0 && val != 1)
1472 		return -EINVAL;
1473 
1474 	/*
1475 	 * Opening of "enable" adds a ref count to system,
1476 	 * so the name is safe to use.
1477 	 */
1478 	if (system)
1479 		name = system->name;
1480 
1481 	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1482 	if (ret)
1483 		goto out;
1484 
1485 	ret = cnt;
1486 
1487 out:
1488 	*ppos += cnt;
1489 
1490 	return ret;
1491 }
1492 
1493 enum {
1494 	FORMAT_HEADER		= 1,
1495 	FORMAT_FIELD_SEPERATOR	= 2,
1496 	FORMAT_PRINTFMT		= 3,
1497 };
1498 
1499 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1500 {
1501 	struct trace_event_call *call = event_file_data(m->private);
1502 	struct list_head *common_head = &ftrace_common_fields;
1503 	struct list_head *head = trace_get_fields(call);
1504 	struct list_head *node = v;
1505 
1506 	(*pos)++;
1507 
1508 	switch ((unsigned long)v) {
1509 	case FORMAT_HEADER:
1510 		node = common_head;
1511 		break;
1512 
1513 	case FORMAT_FIELD_SEPERATOR:
1514 		node = head;
1515 		break;
1516 
1517 	case FORMAT_PRINTFMT:
1518 		/* all done */
1519 		return NULL;
1520 	}
1521 
1522 	node = node->prev;
1523 	if (node == common_head)
1524 		return (void *)FORMAT_FIELD_SEPERATOR;
1525 	else if (node == head)
1526 		return (void *)FORMAT_PRINTFMT;
1527 	else
1528 		return node;
1529 }
1530 
1531 static int f_show(struct seq_file *m, void *v)
1532 {
1533 	struct trace_event_call *call = event_file_data(m->private);
1534 	struct ftrace_event_field *field;
1535 	const char *array_descriptor;
1536 
1537 	switch ((unsigned long)v) {
1538 	case FORMAT_HEADER:
1539 		seq_printf(m, "name: %s\n", trace_event_name(call));
1540 		seq_printf(m, "ID: %d\n", call->event.type);
1541 		seq_puts(m, "format:\n");
1542 		return 0;
1543 
1544 	case FORMAT_FIELD_SEPERATOR:
1545 		seq_putc(m, '\n');
1546 		return 0;
1547 
1548 	case FORMAT_PRINTFMT:
1549 		seq_printf(m, "\nprint fmt: %s\n",
1550 			   call->print_fmt);
1551 		return 0;
1552 	}
1553 
1554 	field = list_entry(v, struct ftrace_event_field, link);
1555 	/*
1556 	 * Smartly shows the array type(except dynamic array).
1557 	 * Normal:
1558 	 *	field:TYPE VAR
1559 	 * If TYPE := TYPE[LEN], it is shown:
1560 	 *	field:TYPE VAR[LEN]
1561 	 */
1562 	array_descriptor = strchr(field->type, '[');
1563 
1564 	if (str_has_prefix(field->type, "__data_loc"))
1565 		array_descriptor = NULL;
1566 
1567 	if (!array_descriptor)
1568 		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1569 			   field->type, field->name, field->offset,
1570 			   field->size, !!field->is_signed);
1571 	else
1572 		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1573 			   (int)(array_descriptor - field->type),
1574 			   field->type, field->name,
1575 			   array_descriptor, field->offset,
1576 			   field->size, !!field->is_signed);
1577 
1578 	return 0;
1579 }
1580 
1581 static void *f_start(struct seq_file *m, loff_t *pos)
1582 {
1583 	void *p = (void *)FORMAT_HEADER;
1584 	loff_t l = 0;
1585 
1586 	/* ->stop() is called even if ->start() fails */
1587 	mutex_lock(&event_mutex);
1588 	if (!event_file_data(m->private))
1589 		return ERR_PTR(-ENODEV);
1590 
1591 	while (l < *pos && p)
1592 		p = f_next(m, p, &l);
1593 
1594 	return p;
1595 }
1596 
1597 static void f_stop(struct seq_file *m, void *p)
1598 {
1599 	mutex_unlock(&event_mutex);
1600 }
1601 
1602 static const struct seq_operations trace_format_seq_ops = {
1603 	.start		= f_start,
1604 	.next		= f_next,
1605 	.stop		= f_stop,
1606 	.show		= f_show,
1607 };
1608 
1609 static int trace_format_open(struct inode *inode, struct file *file)
1610 {
1611 	struct seq_file *m;
1612 	int ret;
1613 
1614 	/* Do we want to hide event format files on tracefs lockdown? */
1615 
1616 	ret = seq_open(file, &trace_format_seq_ops);
1617 	if (ret < 0)
1618 		return ret;
1619 
1620 	m = file->private_data;
1621 	m->private = file;
1622 
1623 	return 0;
1624 }
1625 
1626 static ssize_t
1627 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1628 {
1629 	int id = (long)event_file_data(filp);
1630 	char buf[32];
1631 	int len;
1632 
1633 	if (unlikely(!id))
1634 		return -ENODEV;
1635 
1636 	len = sprintf(buf, "%d\n", id);
1637 
1638 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1639 }
1640 
1641 static ssize_t
1642 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1643 		  loff_t *ppos)
1644 {
1645 	struct trace_event_file *file;
1646 	struct trace_seq *s;
1647 	int r = -ENODEV;
1648 
1649 	if (*ppos)
1650 		return 0;
1651 
1652 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1653 
1654 	if (!s)
1655 		return -ENOMEM;
1656 
1657 	trace_seq_init(s);
1658 
1659 	mutex_lock(&event_mutex);
1660 	file = event_file_data(filp);
1661 	if (file)
1662 		print_event_filter(file, s);
1663 	mutex_unlock(&event_mutex);
1664 
1665 	if (file)
1666 		r = simple_read_from_buffer(ubuf, cnt, ppos,
1667 					    s->buffer, trace_seq_used(s));
1668 
1669 	kfree(s);
1670 
1671 	return r;
1672 }
1673 
1674 static ssize_t
1675 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1676 		   loff_t *ppos)
1677 {
1678 	struct trace_event_file *file;
1679 	char *buf;
1680 	int err = -ENODEV;
1681 
1682 	if (cnt >= PAGE_SIZE)
1683 		return -EINVAL;
1684 
1685 	buf = memdup_user_nul(ubuf, cnt);
1686 	if (IS_ERR(buf))
1687 		return PTR_ERR(buf);
1688 
1689 	mutex_lock(&event_mutex);
1690 	file = event_file_data(filp);
1691 	if (file)
1692 		err = apply_event_filter(file, buf);
1693 	mutex_unlock(&event_mutex);
1694 
1695 	kfree(buf);
1696 	if (err < 0)
1697 		return err;
1698 
1699 	*ppos += cnt;
1700 
1701 	return cnt;
1702 }
1703 
1704 static LIST_HEAD(event_subsystems);
1705 
1706 static int subsystem_open(struct inode *inode, struct file *filp)
1707 {
1708 	struct event_subsystem *system = NULL;
1709 	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1710 	struct trace_array *tr;
1711 	int ret;
1712 
1713 	if (tracing_is_disabled())
1714 		return -ENODEV;
1715 
1716 	/* Make sure the system still exists */
1717 	mutex_lock(&event_mutex);
1718 	mutex_lock(&trace_types_lock);
1719 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1720 		list_for_each_entry(dir, &tr->systems, list) {
1721 			if (dir == inode->i_private) {
1722 				/* Don't open systems with no events */
1723 				if (dir->nr_events) {
1724 					__get_system_dir(dir);
1725 					system = dir->subsystem;
1726 				}
1727 				goto exit_loop;
1728 			}
1729 		}
1730 	}
1731  exit_loop:
1732 	mutex_unlock(&trace_types_lock);
1733 	mutex_unlock(&event_mutex);
1734 
1735 	if (!system)
1736 		return -ENODEV;
1737 
1738 	/* Some versions of gcc think dir can be uninitialized here */
1739 	WARN_ON(!dir);
1740 
1741 	/* Still need to increment the ref count of the system */
1742 	if (trace_array_get(tr) < 0) {
1743 		put_system(dir);
1744 		return -ENODEV;
1745 	}
1746 
1747 	ret = tracing_open_generic(inode, filp);
1748 	if (ret < 0) {
1749 		trace_array_put(tr);
1750 		put_system(dir);
1751 	}
1752 
1753 	return ret;
1754 }
1755 
1756 static int system_tr_open(struct inode *inode, struct file *filp)
1757 {
1758 	struct trace_subsystem_dir *dir;
1759 	struct trace_array *tr = inode->i_private;
1760 	int ret;
1761 
1762 	/* Make a temporary dir that has no system but points to tr */
1763 	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1764 	if (!dir)
1765 		return -ENOMEM;
1766 
1767 	ret = tracing_open_generic_tr(inode, filp);
1768 	if (ret < 0) {
1769 		kfree(dir);
1770 		return ret;
1771 	}
1772 	dir->tr = tr;
1773 	filp->private_data = dir;
1774 
1775 	return 0;
1776 }
1777 
1778 static int subsystem_release(struct inode *inode, struct file *file)
1779 {
1780 	struct trace_subsystem_dir *dir = file->private_data;
1781 
1782 	trace_array_put(dir->tr);
1783 
1784 	/*
1785 	 * If dir->subsystem is NULL, then this is a temporary
1786 	 * descriptor that was made for a trace_array to enable
1787 	 * all subsystems.
1788 	 */
1789 	if (dir->subsystem)
1790 		put_system(dir);
1791 	else
1792 		kfree(dir);
1793 
1794 	return 0;
1795 }
1796 
1797 static ssize_t
1798 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1799 		      loff_t *ppos)
1800 {
1801 	struct trace_subsystem_dir *dir = filp->private_data;
1802 	struct event_subsystem *system = dir->subsystem;
1803 	struct trace_seq *s;
1804 	int r;
1805 
1806 	if (*ppos)
1807 		return 0;
1808 
1809 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1810 	if (!s)
1811 		return -ENOMEM;
1812 
1813 	trace_seq_init(s);
1814 
1815 	print_subsystem_event_filter(system, s);
1816 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1817 				    s->buffer, trace_seq_used(s));
1818 
1819 	kfree(s);
1820 
1821 	return r;
1822 }
1823 
1824 static ssize_t
1825 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1826 		       loff_t *ppos)
1827 {
1828 	struct trace_subsystem_dir *dir = filp->private_data;
1829 	char *buf;
1830 	int err;
1831 
1832 	if (cnt >= PAGE_SIZE)
1833 		return -EINVAL;
1834 
1835 	buf = memdup_user_nul(ubuf, cnt);
1836 	if (IS_ERR(buf))
1837 		return PTR_ERR(buf);
1838 
1839 	err = apply_subsystem_event_filter(dir, buf);
1840 	kfree(buf);
1841 	if (err < 0)
1842 		return err;
1843 
1844 	*ppos += cnt;
1845 
1846 	return cnt;
1847 }
1848 
1849 static ssize_t
1850 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1851 {
1852 	int (*func)(struct trace_seq *s) = filp->private_data;
1853 	struct trace_seq *s;
1854 	int r;
1855 
1856 	if (*ppos)
1857 		return 0;
1858 
1859 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1860 	if (!s)
1861 		return -ENOMEM;
1862 
1863 	trace_seq_init(s);
1864 
1865 	func(s);
1866 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1867 				    s->buffer, trace_seq_used(s));
1868 
1869 	kfree(s);
1870 
1871 	return r;
1872 }
1873 
1874 static void ignore_task_cpu(void *data)
1875 {
1876 	struct trace_array *tr = data;
1877 	struct trace_pid_list *pid_list;
1878 	struct trace_pid_list *no_pid_list;
1879 
1880 	/*
1881 	 * This function is called by on_each_cpu() while the
1882 	 * event_mutex is held.
1883 	 */
1884 	pid_list = rcu_dereference_protected(tr->filtered_pids,
1885 					     mutex_is_locked(&event_mutex));
1886 	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1887 					     mutex_is_locked(&event_mutex));
1888 
1889 	this_cpu_write(tr->array_buffer.data->ignore_pid,
1890 		       trace_ignore_this_task(pid_list, no_pid_list, current));
1891 }
1892 
1893 static void register_pid_events(struct trace_array *tr)
1894 {
1895 	/*
1896 	 * Register a probe that is called before all other probes
1897 	 * to set ignore_pid if next or prev do not match.
1898 	 * Register a probe this is called after all other probes
1899 	 * to only keep ignore_pid set if next pid matches.
1900 	 */
1901 	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1902 					 tr, INT_MAX);
1903 	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1904 					 tr, 0);
1905 
1906 	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1907 					 tr, INT_MAX);
1908 	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1909 					 tr, 0);
1910 
1911 	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1912 					     tr, INT_MAX);
1913 	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1914 					     tr, 0);
1915 
1916 	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1917 					 tr, INT_MAX);
1918 	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1919 					 tr, 0);
1920 }
1921 
1922 static ssize_t
1923 event_pid_write(struct file *filp, const char __user *ubuf,
1924 		size_t cnt, loff_t *ppos, int type)
1925 {
1926 	struct seq_file *m = filp->private_data;
1927 	struct trace_array *tr = m->private;
1928 	struct trace_pid_list *filtered_pids = NULL;
1929 	struct trace_pid_list *other_pids = NULL;
1930 	struct trace_pid_list *pid_list;
1931 	struct trace_event_file *file;
1932 	ssize_t ret;
1933 
1934 	if (!cnt)
1935 		return 0;
1936 
1937 	ret = tracing_update_buffers();
1938 	if (ret < 0)
1939 		return ret;
1940 
1941 	mutex_lock(&event_mutex);
1942 
1943 	if (type == TRACE_PIDS) {
1944 		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1945 							  lockdep_is_held(&event_mutex));
1946 		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1947 							  lockdep_is_held(&event_mutex));
1948 	} else {
1949 		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1950 							  lockdep_is_held(&event_mutex));
1951 		other_pids = rcu_dereference_protected(tr->filtered_pids,
1952 							  lockdep_is_held(&event_mutex));
1953 	}
1954 
1955 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1956 	if (ret < 0)
1957 		goto out;
1958 
1959 	if (type == TRACE_PIDS)
1960 		rcu_assign_pointer(tr->filtered_pids, pid_list);
1961 	else
1962 		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1963 
1964 	list_for_each_entry(file, &tr->events, list) {
1965 		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1966 	}
1967 
1968 	if (filtered_pids) {
1969 		tracepoint_synchronize_unregister();
1970 		trace_pid_list_free(filtered_pids);
1971 	} else if (pid_list && !other_pids) {
1972 		register_pid_events(tr);
1973 	}
1974 
1975 	/*
1976 	 * Ignoring of pids is done at task switch. But we have to
1977 	 * check for those tasks that are currently running.
1978 	 * Always do this in case a pid was appended or removed.
1979 	 */
1980 	on_each_cpu(ignore_task_cpu, tr, 1);
1981 
1982  out:
1983 	mutex_unlock(&event_mutex);
1984 
1985 	if (ret > 0)
1986 		*ppos += ret;
1987 
1988 	return ret;
1989 }
1990 
1991 static ssize_t
1992 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1993 		       size_t cnt, loff_t *ppos)
1994 {
1995 	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1996 }
1997 
1998 static ssize_t
1999 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2000 			size_t cnt, loff_t *ppos)
2001 {
2002 	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2003 }
2004 
2005 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2006 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2007 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2008 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2009 static int ftrace_event_release(struct inode *inode, struct file *file);
2010 
2011 static const struct seq_operations show_event_seq_ops = {
2012 	.start = t_start,
2013 	.next = t_next,
2014 	.show = t_show,
2015 	.stop = t_stop,
2016 };
2017 
2018 static const struct seq_operations show_set_event_seq_ops = {
2019 	.start = s_start,
2020 	.next = s_next,
2021 	.show = t_show,
2022 	.stop = t_stop,
2023 };
2024 
2025 static const struct seq_operations show_set_pid_seq_ops = {
2026 	.start = p_start,
2027 	.next = p_next,
2028 	.show = trace_pid_show,
2029 	.stop = p_stop,
2030 };
2031 
2032 static const struct seq_operations show_set_no_pid_seq_ops = {
2033 	.start = np_start,
2034 	.next = np_next,
2035 	.show = trace_pid_show,
2036 	.stop = p_stop,
2037 };
2038 
2039 static const struct file_operations ftrace_avail_fops = {
2040 	.open = ftrace_event_avail_open,
2041 	.read = seq_read,
2042 	.llseek = seq_lseek,
2043 	.release = seq_release,
2044 };
2045 
2046 static const struct file_operations ftrace_set_event_fops = {
2047 	.open = ftrace_event_set_open,
2048 	.read = seq_read,
2049 	.write = ftrace_event_write,
2050 	.llseek = seq_lseek,
2051 	.release = ftrace_event_release,
2052 };
2053 
2054 static const struct file_operations ftrace_set_event_pid_fops = {
2055 	.open = ftrace_event_set_pid_open,
2056 	.read = seq_read,
2057 	.write = ftrace_event_pid_write,
2058 	.llseek = seq_lseek,
2059 	.release = ftrace_event_release,
2060 };
2061 
2062 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2063 	.open = ftrace_event_set_npid_open,
2064 	.read = seq_read,
2065 	.write = ftrace_event_npid_write,
2066 	.llseek = seq_lseek,
2067 	.release = ftrace_event_release,
2068 };
2069 
2070 static const struct file_operations ftrace_enable_fops = {
2071 	.open = tracing_open_generic,
2072 	.read = event_enable_read,
2073 	.write = event_enable_write,
2074 	.llseek = default_llseek,
2075 };
2076 
2077 static const struct file_operations ftrace_event_format_fops = {
2078 	.open = trace_format_open,
2079 	.read = seq_read,
2080 	.llseek = seq_lseek,
2081 	.release = seq_release,
2082 };
2083 
2084 static const struct file_operations ftrace_event_id_fops = {
2085 	.read = event_id_read,
2086 	.llseek = default_llseek,
2087 };
2088 
2089 static const struct file_operations ftrace_event_filter_fops = {
2090 	.open = tracing_open_generic,
2091 	.read = event_filter_read,
2092 	.write = event_filter_write,
2093 	.llseek = default_llseek,
2094 };
2095 
2096 static const struct file_operations ftrace_subsystem_filter_fops = {
2097 	.open = subsystem_open,
2098 	.read = subsystem_filter_read,
2099 	.write = subsystem_filter_write,
2100 	.llseek = default_llseek,
2101 	.release = subsystem_release,
2102 };
2103 
2104 static const struct file_operations ftrace_system_enable_fops = {
2105 	.open = subsystem_open,
2106 	.read = system_enable_read,
2107 	.write = system_enable_write,
2108 	.llseek = default_llseek,
2109 	.release = subsystem_release,
2110 };
2111 
2112 static const struct file_operations ftrace_tr_enable_fops = {
2113 	.open = system_tr_open,
2114 	.read = system_enable_read,
2115 	.write = system_enable_write,
2116 	.llseek = default_llseek,
2117 	.release = subsystem_release,
2118 };
2119 
2120 static const struct file_operations ftrace_show_header_fops = {
2121 	.open = tracing_open_generic,
2122 	.read = show_header,
2123 	.llseek = default_llseek,
2124 };
2125 
2126 static int
2127 ftrace_event_open(struct inode *inode, struct file *file,
2128 		  const struct seq_operations *seq_ops)
2129 {
2130 	struct seq_file *m;
2131 	int ret;
2132 
2133 	ret = security_locked_down(LOCKDOWN_TRACEFS);
2134 	if (ret)
2135 		return ret;
2136 
2137 	ret = seq_open(file, seq_ops);
2138 	if (ret < 0)
2139 		return ret;
2140 	m = file->private_data;
2141 	/* copy tr over to seq ops */
2142 	m->private = inode->i_private;
2143 
2144 	return ret;
2145 }
2146 
2147 static int ftrace_event_release(struct inode *inode, struct file *file)
2148 {
2149 	struct trace_array *tr = inode->i_private;
2150 
2151 	trace_array_put(tr);
2152 
2153 	return seq_release(inode, file);
2154 }
2155 
2156 static int
2157 ftrace_event_avail_open(struct inode *inode, struct file *file)
2158 {
2159 	const struct seq_operations *seq_ops = &show_event_seq_ops;
2160 
2161 	/* Checks for tracefs lockdown */
2162 	return ftrace_event_open(inode, file, seq_ops);
2163 }
2164 
2165 static int
2166 ftrace_event_set_open(struct inode *inode, struct file *file)
2167 {
2168 	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2169 	struct trace_array *tr = inode->i_private;
2170 	int ret;
2171 
2172 	ret = tracing_check_open_get_tr(tr);
2173 	if (ret)
2174 		return ret;
2175 
2176 	if ((file->f_mode & FMODE_WRITE) &&
2177 	    (file->f_flags & O_TRUNC))
2178 		ftrace_clear_events(tr);
2179 
2180 	ret = ftrace_event_open(inode, file, seq_ops);
2181 	if (ret < 0)
2182 		trace_array_put(tr);
2183 	return ret;
2184 }
2185 
2186 static int
2187 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2188 {
2189 	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2190 	struct trace_array *tr = inode->i_private;
2191 	int ret;
2192 
2193 	ret = tracing_check_open_get_tr(tr);
2194 	if (ret)
2195 		return ret;
2196 
2197 	if ((file->f_mode & FMODE_WRITE) &&
2198 	    (file->f_flags & O_TRUNC))
2199 		ftrace_clear_event_pids(tr, TRACE_PIDS);
2200 
2201 	ret = ftrace_event_open(inode, file, seq_ops);
2202 	if (ret < 0)
2203 		trace_array_put(tr);
2204 	return ret;
2205 }
2206 
2207 static int
2208 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2209 {
2210 	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2211 	struct trace_array *tr = inode->i_private;
2212 	int ret;
2213 
2214 	ret = tracing_check_open_get_tr(tr);
2215 	if (ret)
2216 		return ret;
2217 
2218 	if ((file->f_mode & FMODE_WRITE) &&
2219 	    (file->f_flags & O_TRUNC))
2220 		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2221 
2222 	ret = ftrace_event_open(inode, file, seq_ops);
2223 	if (ret < 0)
2224 		trace_array_put(tr);
2225 	return ret;
2226 }
2227 
2228 static struct event_subsystem *
2229 create_new_subsystem(const char *name)
2230 {
2231 	struct event_subsystem *system;
2232 
2233 	/* need to create new entry */
2234 	system = kmalloc(sizeof(*system), GFP_KERNEL);
2235 	if (!system)
2236 		return NULL;
2237 
2238 	system->ref_count = 1;
2239 
2240 	/* Only allocate if dynamic (kprobes and modules) */
2241 	system->name = kstrdup_const(name, GFP_KERNEL);
2242 	if (!system->name)
2243 		goto out_free;
2244 
2245 	system->filter = NULL;
2246 
2247 	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2248 	if (!system->filter)
2249 		goto out_free;
2250 
2251 	list_add(&system->list, &event_subsystems);
2252 
2253 	return system;
2254 
2255  out_free:
2256 	kfree_const(system->name);
2257 	kfree(system);
2258 	return NULL;
2259 }
2260 
2261 static struct dentry *
2262 event_subsystem_dir(struct trace_array *tr, const char *name,
2263 		    struct trace_event_file *file, struct dentry *parent)
2264 {
2265 	struct trace_subsystem_dir *dir;
2266 	struct event_subsystem *system;
2267 	struct dentry *entry;
2268 
2269 	/* First see if we did not already create this dir */
2270 	list_for_each_entry(dir, &tr->systems, list) {
2271 		system = dir->subsystem;
2272 		if (strcmp(system->name, name) == 0) {
2273 			dir->nr_events++;
2274 			file->system = dir;
2275 			return dir->entry;
2276 		}
2277 	}
2278 
2279 	/* Now see if the system itself exists. */
2280 	list_for_each_entry(system, &event_subsystems, list) {
2281 		if (strcmp(system->name, name) == 0)
2282 			break;
2283 	}
2284 	/* Reset system variable when not found */
2285 	if (&system->list == &event_subsystems)
2286 		system = NULL;
2287 
2288 	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2289 	if (!dir)
2290 		goto out_fail;
2291 
2292 	if (!system) {
2293 		system = create_new_subsystem(name);
2294 		if (!system)
2295 			goto out_free;
2296 	} else
2297 		__get_system(system);
2298 
2299 	dir->entry = tracefs_create_dir(name, parent);
2300 	if (!dir->entry) {
2301 		pr_warn("Failed to create system directory %s\n", name);
2302 		__put_system(system);
2303 		goto out_free;
2304 	}
2305 
2306 	dir->tr = tr;
2307 	dir->ref_count = 1;
2308 	dir->nr_events = 1;
2309 	dir->subsystem = system;
2310 	file->system = dir;
2311 
2312 	/* the ftrace system is special, do not create enable or filter files */
2313 	if (strcmp(name, "ftrace") != 0) {
2314 
2315 		entry = tracefs_create_file("filter", TRACE_MODE_WRITE,
2316 					    dir->entry, dir,
2317 					    &ftrace_subsystem_filter_fops);
2318 		if (!entry) {
2319 			kfree(system->filter);
2320 			system->filter = NULL;
2321 			pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2322 		}
2323 
2324 		trace_create_file("enable", TRACE_MODE_WRITE, dir->entry, dir,
2325 				  &ftrace_system_enable_fops);
2326 	}
2327 
2328 	list_add(&dir->list, &tr->systems);
2329 
2330 	return dir->entry;
2331 
2332  out_free:
2333 	kfree(dir);
2334  out_fail:
2335 	/* Only print this message if failed on memory allocation */
2336 	if (!dir || !system)
2337 		pr_warn("No memory to create event subsystem %s\n", name);
2338 	return NULL;
2339 }
2340 
2341 static int
2342 event_define_fields(struct trace_event_call *call)
2343 {
2344 	struct list_head *head;
2345 	int ret = 0;
2346 
2347 	/*
2348 	 * Other events may have the same class. Only update
2349 	 * the fields if they are not already defined.
2350 	 */
2351 	head = trace_get_fields(call);
2352 	if (list_empty(head)) {
2353 		struct trace_event_fields *field = call->class->fields_array;
2354 		unsigned int offset = sizeof(struct trace_entry);
2355 
2356 		for (; field->type; field++) {
2357 			if (field->type == TRACE_FUNCTION_TYPE) {
2358 				field->define_fields(call);
2359 				break;
2360 			}
2361 
2362 			offset = ALIGN(offset, field->align);
2363 			ret = trace_define_field(call, field->type, field->name,
2364 						 offset, field->size,
2365 						 field->is_signed, field->filter_type);
2366 			if (WARN_ON_ONCE(ret)) {
2367 				pr_err("error code is %d\n", ret);
2368 				break;
2369 			}
2370 
2371 			offset += field->size;
2372 		}
2373 	}
2374 
2375 	return ret;
2376 }
2377 
2378 static int
2379 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2380 {
2381 	struct trace_event_call *call = file->event_call;
2382 	struct trace_array *tr = file->tr;
2383 	struct dentry *d_events;
2384 	const char *name;
2385 	int ret;
2386 
2387 	/*
2388 	 * If the trace point header did not define TRACE_SYSTEM
2389 	 * then the system would be called "TRACE_SYSTEM".
2390 	 */
2391 	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2392 		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2393 		if (!d_events)
2394 			return -ENOMEM;
2395 	} else
2396 		d_events = parent;
2397 
2398 	name = trace_event_name(call);
2399 	file->dir = tracefs_create_dir(name, d_events);
2400 	if (!file->dir) {
2401 		pr_warn("Could not create tracefs '%s' directory\n", name);
2402 		return -1;
2403 	}
2404 
2405 	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2406 		trace_create_file("enable", TRACE_MODE_WRITE, file->dir, file,
2407 				  &ftrace_enable_fops);
2408 
2409 #ifdef CONFIG_PERF_EVENTS
2410 	if (call->event.type && call->class->reg)
2411 		trace_create_file("id", TRACE_MODE_READ, file->dir,
2412 				  (void *)(long)call->event.type,
2413 				  &ftrace_event_id_fops);
2414 #endif
2415 
2416 	ret = event_define_fields(call);
2417 	if (ret < 0) {
2418 		pr_warn("Could not initialize trace point events/%s\n", name);
2419 		return ret;
2420 	}
2421 
2422 	/*
2423 	 * Only event directories that can be enabled should have
2424 	 * triggers or filters.
2425 	 */
2426 	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2427 		trace_create_file("filter", TRACE_MODE_WRITE, file->dir,
2428 				  file, &ftrace_event_filter_fops);
2429 
2430 		trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
2431 				  file, &event_trigger_fops);
2432 	}
2433 
2434 #ifdef CONFIG_HIST_TRIGGERS
2435 	trace_create_file("hist", TRACE_MODE_READ, file->dir, file,
2436 			  &event_hist_fops);
2437 #endif
2438 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2439 	trace_create_file("hist_debug", TRACE_MODE_READ, file->dir, file,
2440 			  &event_hist_debug_fops);
2441 #endif
2442 	trace_create_file("format", TRACE_MODE_READ, file->dir, call,
2443 			  &ftrace_event_format_fops);
2444 
2445 #ifdef CONFIG_TRACE_EVENT_INJECT
2446 	if (call->event.type && call->class->reg)
2447 		trace_create_file("inject", 0200, file->dir, file,
2448 				  &event_inject_fops);
2449 #endif
2450 
2451 	return 0;
2452 }
2453 
2454 static void remove_event_from_tracers(struct trace_event_call *call)
2455 {
2456 	struct trace_event_file *file;
2457 	struct trace_array *tr;
2458 
2459 	do_for_each_event_file_safe(tr, file) {
2460 		if (file->event_call != call)
2461 			continue;
2462 
2463 		remove_event_file_dir(file);
2464 		/*
2465 		 * The do_for_each_event_file_safe() is
2466 		 * a double loop. After finding the call for this
2467 		 * trace_array, we use break to jump to the next
2468 		 * trace_array.
2469 		 */
2470 		break;
2471 	} while_for_each_event_file();
2472 }
2473 
2474 static void event_remove(struct trace_event_call *call)
2475 {
2476 	struct trace_array *tr;
2477 	struct trace_event_file *file;
2478 
2479 	do_for_each_event_file(tr, file) {
2480 		if (file->event_call != call)
2481 			continue;
2482 
2483 		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2484 			tr->clear_trace = true;
2485 
2486 		ftrace_event_enable_disable(file, 0);
2487 		/*
2488 		 * The do_for_each_event_file() is
2489 		 * a double loop. After finding the call for this
2490 		 * trace_array, we use break to jump to the next
2491 		 * trace_array.
2492 		 */
2493 		break;
2494 	} while_for_each_event_file();
2495 
2496 	if (call->event.funcs)
2497 		__unregister_trace_event(&call->event);
2498 	remove_event_from_tracers(call);
2499 	list_del(&call->list);
2500 }
2501 
2502 static int event_init(struct trace_event_call *call)
2503 {
2504 	int ret = 0;
2505 	const char *name;
2506 
2507 	name = trace_event_name(call);
2508 	if (WARN_ON(!name))
2509 		return -EINVAL;
2510 
2511 	if (call->class->raw_init) {
2512 		ret = call->class->raw_init(call);
2513 		if (ret < 0 && ret != -ENOSYS)
2514 			pr_warn("Could not initialize trace events/%s\n", name);
2515 	}
2516 
2517 	return ret;
2518 }
2519 
2520 static int
2521 __register_event(struct trace_event_call *call, struct module *mod)
2522 {
2523 	int ret;
2524 
2525 	ret = event_init(call);
2526 	if (ret < 0)
2527 		return ret;
2528 
2529 	list_add(&call->list, &ftrace_events);
2530 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2531 		atomic_set(&call->refcnt, 0);
2532 	else
2533 		call->module = mod;
2534 
2535 	return 0;
2536 }
2537 
2538 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2539 {
2540 	int rlen;
2541 	int elen;
2542 
2543 	/* Find the length of the eval value as a string */
2544 	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2545 	/* Make sure there's enough room to replace the string with the value */
2546 	if (len < elen)
2547 		return NULL;
2548 
2549 	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2550 
2551 	/* Get the rest of the string of ptr */
2552 	rlen = strlen(ptr + len);
2553 	memmove(ptr + elen, ptr + len, rlen);
2554 	/* Make sure we end the new string */
2555 	ptr[elen + rlen] = 0;
2556 
2557 	return ptr + elen;
2558 }
2559 
2560 static void update_event_printk(struct trace_event_call *call,
2561 				struct trace_eval_map *map)
2562 {
2563 	char *ptr;
2564 	int quote = 0;
2565 	int len = strlen(map->eval_string);
2566 
2567 	for (ptr = call->print_fmt; *ptr; ptr++) {
2568 		if (*ptr == '\\') {
2569 			ptr++;
2570 			/* paranoid */
2571 			if (!*ptr)
2572 				break;
2573 			continue;
2574 		}
2575 		if (*ptr == '"') {
2576 			quote ^= 1;
2577 			continue;
2578 		}
2579 		if (quote)
2580 			continue;
2581 		if (isdigit(*ptr)) {
2582 			/* skip numbers */
2583 			do {
2584 				ptr++;
2585 				/* Check for alpha chars like ULL */
2586 			} while (isalnum(*ptr));
2587 			if (!*ptr)
2588 				break;
2589 			/*
2590 			 * A number must have some kind of delimiter after
2591 			 * it, and we can ignore that too.
2592 			 */
2593 			continue;
2594 		}
2595 		if (isalpha(*ptr) || *ptr == '_') {
2596 			if (strncmp(map->eval_string, ptr, len) == 0 &&
2597 			    !isalnum(ptr[len]) && ptr[len] != '_') {
2598 				ptr = eval_replace(ptr, map, len);
2599 				/* enum/sizeof string smaller than value */
2600 				if (WARN_ON_ONCE(!ptr))
2601 					return;
2602 				/*
2603 				 * No need to decrement here, as eval_replace()
2604 				 * returns the pointer to the character passed
2605 				 * the eval, and two evals can not be placed
2606 				 * back to back without something in between.
2607 				 * We can skip that something in between.
2608 				 */
2609 				continue;
2610 			}
2611 		skip_more:
2612 			do {
2613 				ptr++;
2614 			} while (isalnum(*ptr) || *ptr == '_');
2615 			if (!*ptr)
2616 				break;
2617 			/*
2618 			 * If what comes after this variable is a '.' or
2619 			 * '->' then we can continue to ignore that string.
2620 			 */
2621 			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2622 				ptr += *ptr == '.' ? 1 : 2;
2623 				if (!*ptr)
2624 					break;
2625 				goto skip_more;
2626 			}
2627 			/*
2628 			 * Once again, we can skip the delimiter that came
2629 			 * after the string.
2630 			 */
2631 			continue;
2632 		}
2633 	}
2634 }
2635 
2636 void trace_event_eval_update(struct trace_eval_map **map, int len)
2637 {
2638 	struct trace_event_call *call, *p;
2639 	const char *last_system = NULL;
2640 	bool first = false;
2641 	int last_i;
2642 	int i;
2643 
2644 	down_write(&trace_event_sem);
2645 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2646 		/* events are usually grouped together with systems */
2647 		if (!last_system || call->class->system != last_system) {
2648 			first = true;
2649 			last_i = 0;
2650 			last_system = call->class->system;
2651 		}
2652 
2653 		/*
2654 		 * Since calls are grouped by systems, the likelihood that the
2655 		 * next call in the iteration belongs to the same system as the
2656 		 * previous call is high. As an optimization, we skip searching
2657 		 * for a map[] that matches the call's system if the last call
2658 		 * was from the same system. That's what last_i is for. If the
2659 		 * call has the same system as the previous call, then last_i
2660 		 * will be the index of the first map[] that has a matching
2661 		 * system.
2662 		 */
2663 		for (i = last_i; i < len; i++) {
2664 			if (call->class->system == map[i]->system) {
2665 				/* Save the first system if need be */
2666 				if (first) {
2667 					last_i = i;
2668 					first = false;
2669 				}
2670 				update_event_printk(call, map[i]);
2671 			}
2672 		}
2673 	}
2674 	up_write(&trace_event_sem);
2675 }
2676 
2677 static struct trace_event_file *
2678 trace_create_new_event(struct trace_event_call *call,
2679 		       struct trace_array *tr)
2680 {
2681 	struct trace_event_file *file;
2682 
2683 	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2684 	if (!file)
2685 		return NULL;
2686 
2687 	file->event_call = call;
2688 	file->tr = tr;
2689 	atomic_set(&file->sm_ref, 0);
2690 	atomic_set(&file->tm_ref, 0);
2691 	INIT_LIST_HEAD(&file->triggers);
2692 	list_add(&file->list, &tr->events);
2693 
2694 	return file;
2695 }
2696 
2697 /* Add an event to a trace directory */
2698 static int
2699 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2700 {
2701 	struct trace_event_file *file;
2702 
2703 	file = trace_create_new_event(call, tr);
2704 	if (!file)
2705 		return -ENOMEM;
2706 
2707 	if (eventdir_initialized)
2708 		return event_create_dir(tr->event_dir, file);
2709 	else
2710 		return event_define_fields(call);
2711 }
2712 
2713 /*
2714  * Just create a descriptor for early init. A descriptor is required
2715  * for enabling events at boot. We want to enable events before
2716  * the filesystem is initialized.
2717  */
2718 static int
2719 __trace_early_add_new_event(struct trace_event_call *call,
2720 			    struct trace_array *tr)
2721 {
2722 	struct trace_event_file *file;
2723 
2724 	file = trace_create_new_event(call, tr);
2725 	if (!file)
2726 		return -ENOMEM;
2727 
2728 	return event_define_fields(call);
2729 }
2730 
2731 struct ftrace_module_file_ops;
2732 static void __add_event_to_tracers(struct trace_event_call *call);
2733 
2734 /* Add an additional event_call dynamically */
2735 int trace_add_event_call(struct trace_event_call *call)
2736 {
2737 	int ret;
2738 	lockdep_assert_held(&event_mutex);
2739 
2740 	mutex_lock(&trace_types_lock);
2741 
2742 	ret = __register_event(call, NULL);
2743 	if (ret >= 0)
2744 		__add_event_to_tracers(call);
2745 
2746 	mutex_unlock(&trace_types_lock);
2747 	return ret;
2748 }
2749 
2750 /*
2751  * Must be called under locking of trace_types_lock, event_mutex and
2752  * trace_event_sem.
2753  */
2754 static void __trace_remove_event_call(struct trace_event_call *call)
2755 {
2756 	event_remove(call);
2757 	trace_destroy_fields(call);
2758 	free_event_filter(call->filter);
2759 	call->filter = NULL;
2760 }
2761 
2762 static int probe_remove_event_call(struct trace_event_call *call)
2763 {
2764 	struct trace_array *tr;
2765 	struct trace_event_file *file;
2766 
2767 #ifdef CONFIG_PERF_EVENTS
2768 	if (call->perf_refcount)
2769 		return -EBUSY;
2770 #endif
2771 	do_for_each_event_file(tr, file) {
2772 		if (file->event_call != call)
2773 			continue;
2774 		/*
2775 		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2776 		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2777 		 * TRACE_REG_UNREGISTER.
2778 		 */
2779 		if (file->flags & EVENT_FILE_FL_ENABLED)
2780 			return -EBUSY;
2781 		/*
2782 		 * The do_for_each_event_file_safe() is
2783 		 * a double loop. After finding the call for this
2784 		 * trace_array, we use break to jump to the next
2785 		 * trace_array.
2786 		 */
2787 		break;
2788 	} while_for_each_event_file();
2789 
2790 	__trace_remove_event_call(call);
2791 
2792 	return 0;
2793 }
2794 
2795 /* Remove an event_call */
2796 int trace_remove_event_call(struct trace_event_call *call)
2797 {
2798 	int ret;
2799 
2800 	lockdep_assert_held(&event_mutex);
2801 
2802 	mutex_lock(&trace_types_lock);
2803 	down_write(&trace_event_sem);
2804 	ret = probe_remove_event_call(call);
2805 	up_write(&trace_event_sem);
2806 	mutex_unlock(&trace_types_lock);
2807 
2808 	return ret;
2809 }
2810 
2811 #define for_each_event(event, start, end)			\
2812 	for (event = start;					\
2813 	     (unsigned long)event < (unsigned long)end;		\
2814 	     event++)
2815 
2816 #ifdef CONFIG_MODULES
2817 
2818 static void trace_module_add_events(struct module *mod)
2819 {
2820 	struct trace_event_call **call, **start, **end;
2821 
2822 	if (!mod->num_trace_events)
2823 		return;
2824 
2825 	/* Don't add infrastructure for mods without tracepoints */
2826 	if (trace_module_has_bad_taint(mod)) {
2827 		pr_err("%s: module has bad taint, not creating trace events\n",
2828 		       mod->name);
2829 		return;
2830 	}
2831 
2832 	start = mod->trace_events;
2833 	end = mod->trace_events + mod->num_trace_events;
2834 
2835 	for_each_event(call, start, end) {
2836 		__register_event(*call, mod);
2837 		__add_event_to_tracers(*call);
2838 	}
2839 }
2840 
2841 static void trace_module_remove_events(struct module *mod)
2842 {
2843 	struct trace_event_call *call, *p;
2844 
2845 	down_write(&trace_event_sem);
2846 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2847 		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
2848 			continue;
2849 		if (call->module == mod)
2850 			__trace_remove_event_call(call);
2851 	}
2852 	up_write(&trace_event_sem);
2853 
2854 	/*
2855 	 * It is safest to reset the ring buffer if the module being unloaded
2856 	 * registered any events that were used. The only worry is if
2857 	 * a new module gets loaded, and takes on the same id as the events
2858 	 * of this module. When printing out the buffer, traced events left
2859 	 * over from this module may be passed to the new module events and
2860 	 * unexpected results may occur.
2861 	 */
2862 	tracing_reset_all_online_cpus();
2863 }
2864 
2865 static int trace_module_notify(struct notifier_block *self,
2866 			       unsigned long val, void *data)
2867 {
2868 	struct module *mod = data;
2869 
2870 	mutex_lock(&event_mutex);
2871 	mutex_lock(&trace_types_lock);
2872 	switch (val) {
2873 	case MODULE_STATE_COMING:
2874 		trace_module_add_events(mod);
2875 		break;
2876 	case MODULE_STATE_GOING:
2877 		trace_module_remove_events(mod);
2878 		break;
2879 	}
2880 	mutex_unlock(&trace_types_lock);
2881 	mutex_unlock(&event_mutex);
2882 
2883 	return NOTIFY_OK;
2884 }
2885 
2886 static struct notifier_block trace_module_nb = {
2887 	.notifier_call = trace_module_notify,
2888 	.priority = 1, /* higher than trace.c module notify */
2889 };
2890 #endif /* CONFIG_MODULES */
2891 
2892 /* Create a new event directory structure for a trace directory. */
2893 static void
2894 __trace_add_event_dirs(struct trace_array *tr)
2895 {
2896 	struct trace_event_call *call;
2897 	int ret;
2898 
2899 	list_for_each_entry(call, &ftrace_events, list) {
2900 		ret = __trace_add_new_event(call, tr);
2901 		if (ret < 0)
2902 			pr_warn("Could not create directory for event %s\n",
2903 				trace_event_name(call));
2904 	}
2905 }
2906 
2907 /* Returns any file that matches the system and event */
2908 struct trace_event_file *
2909 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2910 {
2911 	struct trace_event_file *file;
2912 	struct trace_event_call *call;
2913 	const char *name;
2914 
2915 	list_for_each_entry(file, &tr->events, list) {
2916 
2917 		call = file->event_call;
2918 		name = trace_event_name(call);
2919 
2920 		if (!name || !call->class)
2921 			continue;
2922 
2923 		if (strcmp(event, name) == 0 &&
2924 		    strcmp(system, call->class->system) == 0)
2925 			return file;
2926 	}
2927 	return NULL;
2928 }
2929 
2930 /* Returns valid trace event files that match system and event */
2931 struct trace_event_file *
2932 find_event_file(struct trace_array *tr, const char *system, const char *event)
2933 {
2934 	struct trace_event_file *file;
2935 
2936 	file = __find_event_file(tr, system, event);
2937 	if (!file || !file->event_call->class->reg ||
2938 	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2939 		return NULL;
2940 
2941 	return file;
2942 }
2943 
2944 /**
2945  * trace_get_event_file - Find and return a trace event file
2946  * @instance: The name of the trace instance containing the event
2947  * @system: The name of the system containing the event
2948  * @event: The name of the event
2949  *
2950  * Return a trace event file given the trace instance name, trace
2951  * system, and trace event name.  If the instance name is NULL, it
2952  * refers to the top-level trace array.
2953  *
2954  * This function will look it up and return it if found, after calling
2955  * trace_array_get() to prevent the instance from going away, and
2956  * increment the event's module refcount to prevent it from being
2957  * removed.
2958  *
2959  * To release the file, call trace_put_event_file(), which will call
2960  * trace_array_put() and decrement the event's module refcount.
2961  *
2962  * Return: The trace event on success, ERR_PTR otherwise.
2963  */
2964 struct trace_event_file *trace_get_event_file(const char *instance,
2965 					      const char *system,
2966 					      const char *event)
2967 {
2968 	struct trace_array *tr = top_trace_array();
2969 	struct trace_event_file *file = NULL;
2970 	int ret = -EINVAL;
2971 
2972 	if (instance) {
2973 		tr = trace_array_find_get(instance);
2974 		if (!tr)
2975 			return ERR_PTR(-ENOENT);
2976 	} else {
2977 		ret = trace_array_get(tr);
2978 		if (ret)
2979 			return ERR_PTR(ret);
2980 	}
2981 
2982 	mutex_lock(&event_mutex);
2983 
2984 	file = find_event_file(tr, system, event);
2985 	if (!file) {
2986 		trace_array_put(tr);
2987 		ret = -EINVAL;
2988 		goto out;
2989 	}
2990 
2991 	/* Don't let event modules unload while in use */
2992 	ret = trace_event_try_get_ref(file->event_call);
2993 	if (!ret) {
2994 		trace_array_put(tr);
2995 		ret = -EBUSY;
2996 		goto out;
2997 	}
2998 
2999 	ret = 0;
3000  out:
3001 	mutex_unlock(&event_mutex);
3002 
3003 	if (ret)
3004 		file = ERR_PTR(ret);
3005 
3006 	return file;
3007 }
3008 EXPORT_SYMBOL_GPL(trace_get_event_file);
3009 
3010 /**
3011  * trace_put_event_file - Release a file from trace_get_event_file()
3012  * @file: The trace event file
3013  *
3014  * If a file was retrieved using trace_get_event_file(), this should
3015  * be called when it's no longer needed.  It will cancel the previous
3016  * trace_array_get() called by that function, and decrement the
3017  * event's module refcount.
3018  */
3019 void trace_put_event_file(struct trace_event_file *file)
3020 {
3021 	mutex_lock(&event_mutex);
3022 	trace_event_put_ref(file->event_call);
3023 	mutex_unlock(&event_mutex);
3024 
3025 	trace_array_put(file->tr);
3026 }
3027 EXPORT_SYMBOL_GPL(trace_put_event_file);
3028 
3029 #ifdef CONFIG_DYNAMIC_FTRACE
3030 
3031 /* Avoid typos */
3032 #define ENABLE_EVENT_STR	"enable_event"
3033 #define DISABLE_EVENT_STR	"disable_event"
3034 
3035 struct event_probe_data {
3036 	struct trace_event_file	*file;
3037 	unsigned long			count;
3038 	int				ref;
3039 	bool				enable;
3040 };
3041 
3042 static void update_event_probe(struct event_probe_data *data)
3043 {
3044 	if (data->enable)
3045 		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3046 	else
3047 		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3048 }
3049 
3050 static void
3051 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3052 		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3053 		   void *data)
3054 {
3055 	struct ftrace_func_mapper *mapper = data;
3056 	struct event_probe_data *edata;
3057 	void **pdata;
3058 
3059 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3060 	if (!pdata || !*pdata)
3061 		return;
3062 
3063 	edata = *pdata;
3064 	update_event_probe(edata);
3065 }
3066 
3067 static void
3068 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3069 			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3070 			 void *data)
3071 {
3072 	struct ftrace_func_mapper *mapper = data;
3073 	struct event_probe_data *edata;
3074 	void **pdata;
3075 
3076 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3077 	if (!pdata || !*pdata)
3078 		return;
3079 
3080 	edata = *pdata;
3081 
3082 	if (!edata->count)
3083 		return;
3084 
3085 	/* Skip if the event is in a state we want to switch to */
3086 	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3087 		return;
3088 
3089 	if (edata->count != -1)
3090 		(edata->count)--;
3091 
3092 	update_event_probe(edata);
3093 }
3094 
3095 static int
3096 event_enable_print(struct seq_file *m, unsigned long ip,
3097 		   struct ftrace_probe_ops *ops, void *data)
3098 {
3099 	struct ftrace_func_mapper *mapper = data;
3100 	struct event_probe_data *edata;
3101 	void **pdata;
3102 
3103 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3104 
3105 	if (WARN_ON_ONCE(!pdata || !*pdata))
3106 		return 0;
3107 
3108 	edata = *pdata;
3109 
3110 	seq_printf(m, "%ps:", (void *)ip);
3111 
3112 	seq_printf(m, "%s:%s:%s",
3113 		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3114 		   edata->file->event_call->class->system,
3115 		   trace_event_name(edata->file->event_call));
3116 
3117 	if (edata->count == -1)
3118 		seq_puts(m, ":unlimited\n");
3119 	else
3120 		seq_printf(m, ":count=%ld\n", edata->count);
3121 
3122 	return 0;
3123 }
3124 
3125 static int
3126 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3127 		  unsigned long ip, void *init_data, void **data)
3128 {
3129 	struct ftrace_func_mapper *mapper = *data;
3130 	struct event_probe_data *edata = init_data;
3131 	int ret;
3132 
3133 	if (!mapper) {
3134 		mapper = allocate_ftrace_func_mapper();
3135 		if (!mapper)
3136 			return -ENODEV;
3137 		*data = mapper;
3138 	}
3139 
3140 	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3141 	if (ret < 0)
3142 		return ret;
3143 
3144 	edata->ref++;
3145 
3146 	return 0;
3147 }
3148 
3149 static int free_probe_data(void *data)
3150 {
3151 	struct event_probe_data *edata = data;
3152 
3153 	edata->ref--;
3154 	if (!edata->ref) {
3155 		/* Remove the SOFT_MODE flag */
3156 		__ftrace_event_enable_disable(edata->file, 0, 1);
3157 		trace_event_put_ref(edata->file->event_call);
3158 		kfree(edata);
3159 	}
3160 	return 0;
3161 }
3162 
3163 static void
3164 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3165 		  unsigned long ip, void *data)
3166 {
3167 	struct ftrace_func_mapper *mapper = data;
3168 	struct event_probe_data *edata;
3169 
3170 	if (!ip) {
3171 		if (!mapper)
3172 			return;
3173 		free_ftrace_func_mapper(mapper, free_probe_data);
3174 		return;
3175 	}
3176 
3177 	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3178 
3179 	if (WARN_ON_ONCE(!edata))
3180 		return;
3181 
3182 	if (WARN_ON_ONCE(edata->ref <= 0))
3183 		return;
3184 
3185 	free_probe_data(edata);
3186 }
3187 
3188 static struct ftrace_probe_ops event_enable_probe_ops = {
3189 	.func			= event_enable_probe,
3190 	.print			= event_enable_print,
3191 	.init			= event_enable_init,
3192 	.free			= event_enable_free,
3193 };
3194 
3195 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3196 	.func			= event_enable_count_probe,
3197 	.print			= event_enable_print,
3198 	.init			= event_enable_init,
3199 	.free			= event_enable_free,
3200 };
3201 
3202 static struct ftrace_probe_ops event_disable_probe_ops = {
3203 	.func			= event_enable_probe,
3204 	.print			= event_enable_print,
3205 	.init			= event_enable_init,
3206 	.free			= event_enable_free,
3207 };
3208 
3209 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3210 	.func			= event_enable_count_probe,
3211 	.print			= event_enable_print,
3212 	.init			= event_enable_init,
3213 	.free			= event_enable_free,
3214 };
3215 
3216 static int
3217 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3218 		  char *glob, char *cmd, char *param, int enabled)
3219 {
3220 	struct trace_event_file *file;
3221 	struct ftrace_probe_ops *ops;
3222 	struct event_probe_data *data;
3223 	const char *system;
3224 	const char *event;
3225 	char *number;
3226 	bool enable;
3227 	int ret;
3228 
3229 	if (!tr)
3230 		return -ENODEV;
3231 
3232 	/* hash funcs only work with set_ftrace_filter */
3233 	if (!enabled || !param)
3234 		return -EINVAL;
3235 
3236 	system = strsep(&param, ":");
3237 	if (!param)
3238 		return -EINVAL;
3239 
3240 	event = strsep(&param, ":");
3241 
3242 	mutex_lock(&event_mutex);
3243 
3244 	ret = -EINVAL;
3245 	file = find_event_file(tr, system, event);
3246 	if (!file)
3247 		goto out;
3248 
3249 	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3250 
3251 	if (enable)
3252 		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3253 	else
3254 		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3255 
3256 	if (glob[0] == '!') {
3257 		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3258 		goto out;
3259 	}
3260 
3261 	ret = -ENOMEM;
3262 
3263 	data = kzalloc(sizeof(*data), GFP_KERNEL);
3264 	if (!data)
3265 		goto out;
3266 
3267 	data->enable = enable;
3268 	data->count = -1;
3269 	data->file = file;
3270 
3271 	if (!param)
3272 		goto out_reg;
3273 
3274 	number = strsep(&param, ":");
3275 
3276 	ret = -EINVAL;
3277 	if (!strlen(number))
3278 		goto out_free;
3279 
3280 	/*
3281 	 * We use the callback data field (which is a pointer)
3282 	 * as our counter.
3283 	 */
3284 	ret = kstrtoul(number, 0, &data->count);
3285 	if (ret)
3286 		goto out_free;
3287 
3288  out_reg:
3289 	/* Don't let event modules unload while probe registered */
3290 	ret = trace_event_try_get_ref(file->event_call);
3291 	if (!ret) {
3292 		ret = -EBUSY;
3293 		goto out_free;
3294 	}
3295 
3296 	ret = __ftrace_event_enable_disable(file, 1, 1);
3297 	if (ret < 0)
3298 		goto out_put;
3299 
3300 	ret = register_ftrace_function_probe(glob, tr, ops, data);
3301 	/*
3302 	 * The above returns on success the # of functions enabled,
3303 	 * but if it didn't find any functions it returns zero.
3304 	 * Consider no functions a failure too.
3305 	 */
3306 	if (!ret) {
3307 		ret = -ENOENT;
3308 		goto out_disable;
3309 	} else if (ret < 0)
3310 		goto out_disable;
3311 	/* Just return zero, not the number of enabled functions */
3312 	ret = 0;
3313  out:
3314 	mutex_unlock(&event_mutex);
3315 	return ret;
3316 
3317  out_disable:
3318 	__ftrace_event_enable_disable(file, 0, 1);
3319  out_put:
3320 	trace_event_put_ref(file->event_call);
3321  out_free:
3322 	kfree(data);
3323 	goto out;
3324 }
3325 
3326 static struct ftrace_func_command event_enable_cmd = {
3327 	.name			= ENABLE_EVENT_STR,
3328 	.func			= event_enable_func,
3329 };
3330 
3331 static struct ftrace_func_command event_disable_cmd = {
3332 	.name			= DISABLE_EVENT_STR,
3333 	.func			= event_enable_func,
3334 };
3335 
3336 static __init int register_event_cmds(void)
3337 {
3338 	int ret;
3339 
3340 	ret = register_ftrace_command(&event_enable_cmd);
3341 	if (WARN_ON(ret < 0))
3342 		return ret;
3343 	ret = register_ftrace_command(&event_disable_cmd);
3344 	if (WARN_ON(ret < 0))
3345 		unregister_ftrace_command(&event_enable_cmd);
3346 	return ret;
3347 }
3348 #else
3349 static inline int register_event_cmds(void) { return 0; }
3350 #endif /* CONFIG_DYNAMIC_FTRACE */
3351 
3352 /*
3353  * The top level array and trace arrays created by boot-time tracing
3354  * have already had its trace_event_file descriptors created in order
3355  * to allow for early events to be recorded.
3356  * This function is called after the tracefs has been initialized,
3357  * and we now have to create the files associated to the events.
3358  */
3359 static void __trace_early_add_event_dirs(struct trace_array *tr)
3360 {
3361 	struct trace_event_file *file;
3362 	int ret;
3363 
3364 
3365 	list_for_each_entry(file, &tr->events, list) {
3366 		ret = event_create_dir(tr->event_dir, file);
3367 		if (ret < 0)
3368 			pr_warn("Could not create directory for event %s\n",
3369 				trace_event_name(file->event_call));
3370 	}
3371 }
3372 
3373 /*
3374  * For early boot up, the top trace array and the trace arrays created
3375  * by boot-time tracing require to have a list of events that can be
3376  * enabled. This must be done before the filesystem is set up in order
3377  * to allow events to be traced early.
3378  */
3379 void __trace_early_add_events(struct trace_array *tr)
3380 {
3381 	struct trace_event_call *call;
3382 	int ret;
3383 
3384 	list_for_each_entry(call, &ftrace_events, list) {
3385 		/* Early boot up should not have any modules loaded */
3386 		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3387 		    WARN_ON_ONCE(call->module))
3388 			continue;
3389 
3390 		ret = __trace_early_add_new_event(call, tr);
3391 		if (ret < 0)
3392 			pr_warn("Could not create early event %s\n",
3393 				trace_event_name(call));
3394 	}
3395 }
3396 
3397 /* Remove the event directory structure for a trace directory. */
3398 static void
3399 __trace_remove_event_dirs(struct trace_array *tr)
3400 {
3401 	struct trace_event_file *file, *next;
3402 
3403 	list_for_each_entry_safe(file, next, &tr->events, list)
3404 		remove_event_file_dir(file);
3405 }
3406 
3407 static void __add_event_to_tracers(struct trace_event_call *call)
3408 {
3409 	struct trace_array *tr;
3410 
3411 	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3412 		__trace_add_new_event(call, tr);
3413 }
3414 
3415 extern struct trace_event_call *__start_ftrace_events[];
3416 extern struct trace_event_call *__stop_ftrace_events[];
3417 
3418 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3419 
3420 static __init int setup_trace_event(char *str)
3421 {
3422 	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3423 	ring_buffer_expanded = true;
3424 	disable_tracing_selftest("running event tracing");
3425 
3426 	return 1;
3427 }
3428 __setup("trace_event=", setup_trace_event);
3429 
3430 /* Expects to have event_mutex held when called */
3431 static int
3432 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3433 {
3434 	struct dentry *d_events;
3435 	struct dentry *entry;
3436 
3437 	entry = tracefs_create_file("set_event", TRACE_MODE_WRITE, parent,
3438 				    tr, &ftrace_set_event_fops);
3439 	if (!entry) {
3440 		pr_warn("Could not create tracefs 'set_event' entry\n");
3441 		return -ENOMEM;
3442 	}
3443 
3444 	d_events = tracefs_create_dir("events", parent);
3445 	if (!d_events) {
3446 		pr_warn("Could not create tracefs 'events' directory\n");
3447 		return -ENOMEM;
3448 	}
3449 
3450 	entry = trace_create_file("enable", TRACE_MODE_WRITE, d_events,
3451 				  tr, &ftrace_tr_enable_fops);
3452 	if (!entry) {
3453 		pr_warn("Could not create tracefs 'enable' entry\n");
3454 		return -ENOMEM;
3455 	}
3456 
3457 	/* There are not as crucial, just warn if they are not created */
3458 
3459 	entry = tracefs_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3460 				    tr, &ftrace_set_event_pid_fops);
3461 	if (!entry)
3462 		pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3463 
3464 	entry = tracefs_create_file("set_event_notrace_pid",
3465 				    TRACE_MODE_WRITE, parent, tr,
3466 				    &ftrace_set_event_notrace_pid_fops);
3467 	if (!entry)
3468 		pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3469 
3470 	/* ring buffer internal formats */
3471 	entry = trace_create_file("header_page", TRACE_MODE_READ, d_events,
3472 				  ring_buffer_print_page_header,
3473 				  &ftrace_show_header_fops);
3474 	if (!entry)
3475 		pr_warn("Could not create tracefs 'header_page' entry\n");
3476 
3477 	entry = trace_create_file("header_event", TRACE_MODE_READ, d_events,
3478 				  ring_buffer_print_entry_header,
3479 				  &ftrace_show_header_fops);
3480 	if (!entry)
3481 		pr_warn("Could not create tracefs 'header_event' entry\n");
3482 
3483 	tr->event_dir = d_events;
3484 
3485 	return 0;
3486 }
3487 
3488 /**
3489  * event_trace_add_tracer - add a instance of a trace_array to events
3490  * @parent: The parent dentry to place the files/directories for events in
3491  * @tr: The trace array associated with these events
3492  *
3493  * When a new instance is created, it needs to set up its events
3494  * directory, as well as other files associated with events. It also
3495  * creates the event hierarchy in the @parent/events directory.
3496  *
3497  * Returns 0 on success.
3498  *
3499  * Must be called with event_mutex held.
3500  */
3501 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3502 {
3503 	int ret;
3504 
3505 	lockdep_assert_held(&event_mutex);
3506 
3507 	ret = create_event_toplevel_files(parent, tr);
3508 	if (ret)
3509 		goto out;
3510 
3511 	down_write(&trace_event_sem);
3512 	/* If tr already has the event list, it is initialized in early boot. */
3513 	if (unlikely(!list_empty(&tr->events)))
3514 		__trace_early_add_event_dirs(tr);
3515 	else
3516 		__trace_add_event_dirs(tr);
3517 	up_write(&trace_event_sem);
3518 
3519  out:
3520 	return ret;
3521 }
3522 
3523 /*
3524  * The top trace array already had its file descriptors created.
3525  * Now the files themselves need to be created.
3526  */
3527 static __init int
3528 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3529 {
3530 	int ret;
3531 
3532 	mutex_lock(&event_mutex);
3533 
3534 	ret = create_event_toplevel_files(parent, tr);
3535 	if (ret)
3536 		goto out_unlock;
3537 
3538 	down_write(&trace_event_sem);
3539 	__trace_early_add_event_dirs(tr);
3540 	up_write(&trace_event_sem);
3541 
3542  out_unlock:
3543 	mutex_unlock(&event_mutex);
3544 
3545 	return ret;
3546 }
3547 
3548 /* Must be called with event_mutex held */
3549 int event_trace_del_tracer(struct trace_array *tr)
3550 {
3551 	lockdep_assert_held(&event_mutex);
3552 
3553 	/* Disable any event triggers and associated soft-disabled events */
3554 	clear_event_triggers(tr);
3555 
3556 	/* Clear the pid list */
3557 	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3558 
3559 	/* Disable any running events */
3560 	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3561 
3562 	/* Make sure no more events are being executed */
3563 	tracepoint_synchronize_unregister();
3564 
3565 	down_write(&trace_event_sem);
3566 	__trace_remove_event_dirs(tr);
3567 	tracefs_remove(tr->event_dir);
3568 	up_write(&trace_event_sem);
3569 
3570 	tr->event_dir = NULL;
3571 
3572 	return 0;
3573 }
3574 
3575 static __init int event_trace_memsetup(void)
3576 {
3577 	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3578 	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3579 	return 0;
3580 }
3581 
3582 static __init void
3583 early_enable_events(struct trace_array *tr, bool disable_first)
3584 {
3585 	char *buf = bootup_event_buf;
3586 	char *token;
3587 	int ret;
3588 
3589 	while (true) {
3590 		token = strsep(&buf, ",");
3591 
3592 		if (!token)
3593 			break;
3594 
3595 		if (*token) {
3596 			/* Restarting syscalls requires that we stop them first */
3597 			if (disable_first)
3598 				ftrace_set_clr_event(tr, token, 0);
3599 
3600 			ret = ftrace_set_clr_event(tr, token, 1);
3601 			if (ret)
3602 				pr_warn("Failed to enable trace event: %s\n", token);
3603 		}
3604 
3605 		/* Put back the comma to allow this to be called again */
3606 		if (buf)
3607 			*(buf - 1) = ',';
3608 	}
3609 }
3610 
3611 static __init int event_trace_enable(void)
3612 {
3613 	struct trace_array *tr = top_trace_array();
3614 	struct trace_event_call **iter, *call;
3615 	int ret;
3616 
3617 	if (!tr)
3618 		return -ENODEV;
3619 
3620 	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3621 
3622 		call = *iter;
3623 		ret = event_init(call);
3624 		if (!ret)
3625 			list_add(&call->list, &ftrace_events);
3626 	}
3627 
3628 	/*
3629 	 * We need the top trace array to have a working set of trace
3630 	 * points at early init, before the debug files and directories
3631 	 * are created. Create the file entries now, and attach them
3632 	 * to the actual file dentries later.
3633 	 */
3634 	__trace_early_add_events(tr);
3635 
3636 	early_enable_events(tr, false);
3637 
3638 	trace_printk_start_comm();
3639 
3640 	register_event_cmds();
3641 
3642 	register_trigger_cmds();
3643 
3644 	return 0;
3645 }
3646 
3647 /*
3648  * event_trace_enable() is called from trace_event_init() first to
3649  * initialize events and perhaps start any events that are on the
3650  * command line. Unfortunately, there are some events that will not
3651  * start this early, like the system call tracepoints that need
3652  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3653  * event_trace_enable() is called before pid 1 starts, and this flag
3654  * is never set, making the syscall tracepoint never get reached, but
3655  * the event is enabled regardless (and not doing anything).
3656  */
3657 static __init int event_trace_enable_again(void)
3658 {
3659 	struct trace_array *tr;
3660 
3661 	tr = top_trace_array();
3662 	if (!tr)
3663 		return -ENODEV;
3664 
3665 	early_enable_events(tr, true);
3666 
3667 	return 0;
3668 }
3669 
3670 early_initcall(event_trace_enable_again);
3671 
3672 /* Init fields which doesn't related to the tracefs */
3673 static __init int event_trace_init_fields(void)
3674 {
3675 	if (trace_define_generic_fields())
3676 		pr_warn("tracing: Failed to allocated generic fields");
3677 
3678 	if (trace_define_common_fields())
3679 		pr_warn("tracing: Failed to allocate common fields");
3680 
3681 	return 0;
3682 }
3683 
3684 __init int event_trace_init(void)
3685 {
3686 	struct trace_array *tr;
3687 	struct dentry *entry;
3688 	int ret;
3689 
3690 	tr = top_trace_array();
3691 	if (!tr)
3692 		return -ENODEV;
3693 
3694 	entry = tracefs_create_file("available_events", TRACE_MODE_READ,
3695 				    NULL, tr, &ftrace_avail_fops);
3696 	if (!entry)
3697 		pr_warn("Could not create tracefs 'available_events' entry\n");
3698 
3699 	ret = early_event_add_tracer(NULL, tr);
3700 	if (ret)
3701 		return ret;
3702 
3703 #ifdef CONFIG_MODULES
3704 	ret = register_module_notifier(&trace_module_nb);
3705 	if (ret)
3706 		pr_warn("Failed to register trace events module notifier\n");
3707 #endif
3708 
3709 	eventdir_initialized = true;
3710 
3711 	return 0;
3712 }
3713 
3714 void __init trace_event_init(void)
3715 {
3716 	event_trace_memsetup();
3717 	init_ftrace_syscalls();
3718 	event_trace_enable();
3719 	event_trace_init_fields();
3720 }
3721 
3722 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3723 
3724 static DEFINE_SPINLOCK(test_spinlock);
3725 static DEFINE_SPINLOCK(test_spinlock_irq);
3726 static DEFINE_MUTEX(test_mutex);
3727 
3728 static __init void test_work(struct work_struct *dummy)
3729 {
3730 	spin_lock(&test_spinlock);
3731 	spin_lock_irq(&test_spinlock_irq);
3732 	udelay(1);
3733 	spin_unlock_irq(&test_spinlock_irq);
3734 	spin_unlock(&test_spinlock);
3735 
3736 	mutex_lock(&test_mutex);
3737 	msleep(1);
3738 	mutex_unlock(&test_mutex);
3739 }
3740 
3741 static __init int event_test_thread(void *unused)
3742 {
3743 	void *test_malloc;
3744 
3745 	test_malloc = kmalloc(1234, GFP_KERNEL);
3746 	if (!test_malloc)
3747 		pr_info("failed to kmalloc\n");
3748 
3749 	schedule_on_each_cpu(test_work);
3750 
3751 	kfree(test_malloc);
3752 
3753 	set_current_state(TASK_INTERRUPTIBLE);
3754 	while (!kthread_should_stop()) {
3755 		schedule();
3756 		set_current_state(TASK_INTERRUPTIBLE);
3757 	}
3758 	__set_current_state(TASK_RUNNING);
3759 
3760 	return 0;
3761 }
3762 
3763 /*
3764  * Do various things that may trigger events.
3765  */
3766 static __init void event_test_stuff(void)
3767 {
3768 	struct task_struct *test_thread;
3769 
3770 	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3771 	msleep(1);
3772 	kthread_stop(test_thread);
3773 }
3774 
3775 /*
3776  * For every trace event defined, we will test each trace point separately,
3777  * and then by groups, and finally all trace points.
3778  */
3779 static __init void event_trace_self_tests(void)
3780 {
3781 	struct trace_subsystem_dir *dir;
3782 	struct trace_event_file *file;
3783 	struct trace_event_call *call;
3784 	struct event_subsystem *system;
3785 	struct trace_array *tr;
3786 	int ret;
3787 
3788 	tr = top_trace_array();
3789 	if (!tr)
3790 		return;
3791 
3792 	pr_info("Running tests on trace events:\n");
3793 
3794 	list_for_each_entry(file, &tr->events, list) {
3795 
3796 		call = file->event_call;
3797 
3798 		/* Only test those that have a probe */
3799 		if (!call->class || !call->class->probe)
3800 			continue;
3801 
3802 /*
3803  * Testing syscall events here is pretty useless, but
3804  * we still do it if configured. But this is time consuming.
3805  * What we really need is a user thread to perform the
3806  * syscalls as we test.
3807  */
3808 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3809 		if (call->class->system &&
3810 		    strcmp(call->class->system, "syscalls") == 0)
3811 			continue;
3812 #endif
3813 
3814 		pr_info("Testing event %s: ", trace_event_name(call));
3815 
3816 		/*
3817 		 * If an event is already enabled, someone is using
3818 		 * it and the self test should not be on.
3819 		 */
3820 		if (file->flags & EVENT_FILE_FL_ENABLED) {
3821 			pr_warn("Enabled event during self test!\n");
3822 			WARN_ON_ONCE(1);
3823 			continue;
3824 		}
3825 
3826 		ftrace_event_enable_disable(file, 1);
3827 		event_test_stuff();
3828 		ftrace_event_enable_disable(file, 0);
3829 
3830 		pr_cont("OK\n");
3831 	}
3832 
3833 	/* Now test at the sub system level */
3834 
3835 	pr_info("Running tests on trace event systems:\n");
3836 
3837 	list_for_each_entry(dir, &tr->systems, list) {
3838 
3839 		system = dir->subsystem;
3840 
3841 		/* the ftrace system is special, skip it */
3842 		if (strcmp(system->name, "ftrace") == 0)
3843 			continue;
3844 
3845 		pr_info("Testing event system %s: ", system->name);
3846 
3847 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3848 		if (WARN_ON_ONCE(ret)) {
3849 			pr_warn("error enabling system %s\n",
3850 				system->name);
3851 			continue;
3852 		}
3853 
3854 		event_test_stuff();
3855 
3856 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3857 		if (WARN_ON_ONCE(ret)) {
3858 			pr_warn("error disabling system %s\n",
3859 				system->name);
3860 			continue;
3861 		}
3862 
3863 		pr_cont("OK\n");
3864 	}
3865 
3866 	/* Test with all events enabled */
3867 
3868 	pr_info("Running tests on all trace events:\n");
3869 	pr_info("Testing all events: ");
3870 
3871 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3872 	if (WARN_ON_ONCE(ret)) {
3873 		pr_warn("error enabling all events\n");
3874 		return;
3875 	}
3876 
3877 	event_test_stuff();
3878 
3879 	/* reset sysname */
3880 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3881 	if (WARN_ON_ONCE(ret)) {
3882 		pr_warn("error disabling all events\n");
3883 		return;
3884 	}
3885 
3886 	pr_cont("OK\n");
3887 }
3888 
3889 #ifdef CONFIG_FUNCTION_TRACER
3890 
3891 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3892 
3893 static struct trace_event_file event_trace_file __initdata;
3894 
3895 static void __init
3896 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3897 			  struct ftrace_ops *op, struct ftrace_regs *regs)
3898 {
3899 	struct trace_buffer *buffer;
3900 	struct ring_buffer_event *event;
3901 	struct ftrace_entry *entry;
3902 	unsigned int trace_ctx;
3903 	long disabled;
3904 	int cpu;
3905 
3906 	trace_ctx = tracing_gen_ctx();
3907 	preempt_disable_notrace();
3908 	cpu = raw_smp_processor_id();
3909 	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3910 
3911 	if (disabled != 1)
3912 		goto out;
3913 
3914 	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3915 						TRACE_FN, sizeof(*entry),
3916 						trace_ctx);
3917 	if (!event)
3918 		goto out;
3919 	entry	= ring_buffer_event_data(event);
3920 	entry->ip			= ip;
3921 	entry->parent_ip		= parent_ip;
3922 
3923 	event_trigger_unlock_commit(&event_trace_file, buffer, event,
3924 				    entry, trace_ctx);
3925  out:
3926 	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3927 	preempt_enable_notrace();
3928 }
3929 
3930 static struct ftrace_ops trace_ops __initdata  =
3931 {
3932 	.func = function_test_events_call,
3933 };
3934 
3935 static __init void event_trace_self_test_with_function(void)
3936 {
3937 	int ret;
3938 
3939 	event_trace_file.tr = top_trace_array();
3940 	if (WARN_ON(!event_trace_file.tr))
3941 		return;
3942 
3943 	ret = register_ftrace_function(&trace_ops);
3944 	if (WARN_ON(ret < 0)) {
3945 		pr_info("Failed to enable function tracer for event tests\n");
3946 		return;
3947 	}
3948 	pr_info("Running tests again, along with the function tracer\n");
3949 	event_trace_self_tests();
3950 	unregister_ftrace_function(&trace_ops);
3951 }
3952 #else
3953 static __init void event_trace_self_test_with_function(void)
3954 {
3955 }
3956 #endif
3957 
3958 static __init int event_trace_self_tests_init(void)
3959 {
3960 	if (!tracing_selftest_disabled) {
3961 		event_trace_self_tests();
3962 		event_trace_self_test_with_function();
3963 	}
3964 
3965 	return 0;
3966 }
3967 
3968 late_initcall(event_trace_self_tests_init);
3969 
3970 #endif
3971