xref: /openbmc/linux/kernel/trace/trace_events.c (revision 2f4e3926)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * event tracer
4  *
5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6  *
7  *  - Added format output of fields of the trace point.
8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
9  *
10  */
11 
12 #define pr_fmt(fmt) fmt
13 
14 #include <linux/workqueue.h>
15 #include <linux/security.h>
16 #include <linux/spinlock.h>
17 #include <linux/kthread.h>
18 #include <linux/tracefs.h>
19 #include <linux/uaccess.h>
20 #include <linux/module.h>
21 #include <linux/ctype.h>
22 #include <linux/sort.h>
23 #include <linux/slab.h>
24 #include <linux/delay.h>
25 
26 #include <trace/events/sched.h>
27 #include <trace/syscall.h>
28 
29 #include <asm/setup.h>
30 
31 #include "trace_output.h"
32 
33 #undef TRACE_SYSTEM
34 #define TRACE_SYSTEM "TRACE_SYSTEM"
35 
36 DEFINE_MUTEX(event_mutex);
37 
38 LIST_HEAD(ftrace_events);
39 static LIST_HEAD(ftrace_generic_fields);
40 static LIST_HEAD(ftrace_common_fields);
41 static bool eventdir_initialized;
42 
43 static LIST_HEAD(module_strings);
44 
45 struct module_string {
46 	struct list_head	next;
47 	struct module		*module;
48 	char			*str;
49 };
50 
51 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
52 
53 static struct kmem_cache *field_cachep;
54 static struct kmem_cache *file_cachep;
55 
56 static inline int system_refcount(struct event_subsystem *system)
57 {
58 	return system->ref_count;
59 }
60 
61 static int system_refcount_inc(struct event_subsystem *system)
62 {
63 	return system->ref_count++;
64 }
65 
66 static int system_refcount_dec(struct event_subsystem *system)
67 {
68 	return --system->ref_count;
69 }
70 
71 /* Double loops, do not use break, only goto's work */
72 #define do_for_each_event_file(tr, file)			\
73 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
74 		list_for_each_entry(file, &tr->events, list)
75 
76 #define do_for_each_event_file_safe(tr, file)			\
77 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
78 		struct trace_event_file *___n;				\
79 		list_for_each_entry_safe(file, ___n, &tr->events, list)
80 
81 #define while_for_each_event_file()		\
82 	}
83 
84 static struct ftrace_event_field *
85 __find_event_field(struct list_head *head, char *name)
86 {
87 	struct ftrace_event_field *field;
88 
89 	list_for_each_entry(field, head, link) {
90 		if (!strcmp(field->name, name))
91 			return field;
92 	}
93 
94 	return NULL;
95 }
96 
97 struct ftrace_event_field *
98 trace_find_event_field(struct trace_event_call *call, char *name)
99 {
100 	struct ftrace_event_field *field;
101 	struct list_head *head;
102 
103 	head = trace_get_fields(call);
104 	field = __find_event_field(head, name);
105 	if (field)
106 		return field;
107 
108 	field = __find_event_field(&ftrace_generic_fields, name);
109 	if (field)
110 		return field;
111 
112 	return __find_event_field(&ftrace_common_fields, name);
113 }
114 
115 static int __trace_define_field(struct list_head *head, const char *type,
116 				const char *name, int offset, int size,
117 				int is_signed, int filter_type, int len)
118 {
119 	struct ftrace_event_field *field;
120 
121 	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
122 	if (!field)
123 		return -ENOMEM;
124 
125 	field->name = name;
126 	field->type = type;
127 
128 	if (filter_type == FILTER_OTHER)
129 		field->filter_type = filter_assign_type(type);
130 	else
131 		field->filter_type = filter_type;
132 
133 	field->offset = offset;
134 	field->size = size;
135 	field->is_signed = is_signed;
136 	field->len = len;
137 
138 	list_add(&field->link, head);
139 
140 	return 0;
141 }
142 
143 int trace_define_field(struct trace_event_call *call, const char *type,
144 		       const char *name, int offset, int size, int is_signed,
145 		       int filter_type)
146 {
147 	struct list_head *head;
148 
149 	if (WARN_ON(!call->class))
150 		return 0;
151 
152 	head = trace_get_fields(call);
153 	return __trace_define_field(head, type, name, offset, size,
154 				    is_signed, filter_type, 0);
155 }
156 EXPORT_SYMBOL_GPL(trace_define_field);
157 
158 static int trace_define_field_ext(struct trace_event_call *call, const char *type,
159 		       const char *name, int offset, int size, int is_signed,
160 		       int filter_type, int len)
161 {
162 	struct list_head *head;
163 
164 	if (WARN_ON(!call->class))
165 		return 0;
166 
167 	head = trace_get_fields(call);
168 	return __trace_define_field(head, type, name, offset, size,
169 				    is_signed, filter_type, len);
170 }
171 
172 #define __generic_field(type, item, filter_type)			\
173 	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
174 				   #item, 0, 0, is_signed_type(type),	\
175 				   filter_type, 0);			\
176 	if (ret)							\
177 		return ret;
178 
179 #define __common_field(type, item)					\
180 	ret = __trace_define_field(&ftrace_common_fields, #type,	\
181 				   "common_" #item,			\
182 				   offsetof(typeof(ent), item),		\
183 				   sizeof(ent.item),			\
184 				   is_signed_type(type), FILTER_OTHER, 0);	\
185 	if (ret)							\
186 		return ret;
187 
188 static int trace_define_generic_fields(void)
189 {
190 	int ret;
191 
192 	__generic_field(int, CPU, FILTER_CPU);
193 	__generic_field(int, cpu, FILTER_CPU);
194 	__generic_field(int, common_cpu, FILTER_CPU);
195 	__generic_field(char *, COMM, FILTER_COMM);
196 	__generic_field(char *, comm, FILTER_COMM);
197 	__generic_field(char *, stacktrace, FILTER_STACKTRACE);
198 	__generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
199 
200 	return ret;
201 }
202 
203 static int trace_define_common_fields(void)
204 {
205 	int ret;
206 	struct trace_entry ent;
207 
208 	__common_field(unsigned short, type);
209 	__common_field(unsigned char, flags);
210 	/* Holds both preempt_count and migrate_disable */
211 	__common_field(unsigned char, preempt_count);
212 	__common_field(int, pid);
213 
214 	return ret;
215 }
216 
217 static void trace_destroy_fields(struct trace_event_call *call)
218 {
219 	struct ftrace_event_field *field, *next;
220 	struct list_head *head;
221 
222 	head = trace_get_fields(call);
223 	list_for_each_entry_safe(field, next, head, link) {
224 		list_del(&field->link);
225 		kmem_cache_free(field_cachep, field);
226 	}
227 }
228 
229 /*
230  * run-time version of trace_event_get_offsets_<call>() that returns the last
231  * accessible offset of trace fields excluding __dynamic_array bytes
232  */
233 int trace_event_get_offsets(struct trace_event_call *call)
234 {
235 	struct ftrace_event_field *tail;
236 	struct list_head *head;
237 
238 	head = trace_get_fields(call);
239 	/*
240 	 * head->next points to the last field with the largest offset,
241 	 * since it was added last by trace_define_field()
242 	 */
243 	tail = list_first_entry(head, struct ftrace_event_field, link);
244 	return tail->offset + tail->size;
245 }
246 
247 /*
248  * Check if the referenced field is an array and return true,
249  * as arrays are OK to dereference.
250  */
251 static bool test_field(const char *fmt, struct trace_event_call *call)
252 {
253 	struct trace_event_fields *field = call->class->fields_array;
254 	const char *array_descriptor;
255 	const char *p = fmt;
256 	int len;
257 
258 	if (!(len = str_has_prefix(fmt, "REC->")))
259 		return false;
260 	fmt += len;
261 	for (p = fmt; *p; p++) {
262 		if (!isalnum(*p) && *p != '_')
263 			break;
264 	}
265 	len = p - fmt;
266 
267 	for (; field->type; field++) {
268 		if (strncmp(field->name, fmt, len) ||
269 		    field->name[len])
270 			continue;
271 		array_descriptor = strchr(field->type, '[');
272 		/* This is an array and is OK to dereference. */
273 		return array_descriptor != NULL;
274 	}
275 	return false;
276 }
277 
278 /*
279  * Examine the print fmt of the event looking for unsafe dereference
280  * pointers using %p* that could be recorded in the trace event and
281  * much later referenced after the pointer was freed. Dereferencing
282  * pointers are OK, if it is dereferenced into the event itself.
283  */
284 static void test_event_printk(struct trace_event_call *call)
285 {
286 	u64 dereference_flags = 0;
287 	bool first = true;
288 	const char *fmt, *c, *r, *a;
289 	int parens = 0;
290 	char in_quote = 0;
291 	int start_arg = 0;
292 	int arg = 0;
293 	int i;
294 
295 	fmt = call->print_fmt;
296 
297 	if (!fmt)
298 		return;
299 
300 	for (i = 0; fmt[i]; i++) {
301 		switch (fmt[i]) {
302 		case '\\':
303 			i++;
304 			if (!fmt[i])
305 				return;
306 			continue;
307 		case '"':
308 		case '\'':
309 			/*
310 			 * The print fmt starts with a string that
311 			 * is processed first to find %p* usage,
312 			 * then after the first string, the print fmt
313 			 * contains arguments that are used to check
314 			 * if the dereferenced %p* usage is safe.
315 			 */
316 			if (first) {
317 				if (fmt[i] == '\'')
318 					continue;
319 				if (in_quote) {
320 					arg = 0;
321 					first = false;
322 					/*
323 					 * If there was no %p* uses
324 					 * the fmt is OK.
325 					 */
326 					if (!dereference_flags)
327 						return;
328 				}
329 			}
330 			if (in_quote) {
331 				if (in_quote == fmt[i])
332 					in_quote = 0;
333 			} else {
334 				in_quote = fmt[i];
335 			}
336 			continue;
337 		case '%':
338 			if (!first || !in_quote)
339 				continue;
340 			i++;
341 			if (!fmt[i])
342 				return;
343 			switch (fmt[i]) {
344 			case '%':
345 				continue;
346 			case 'p':
347 				/* Find dereferencing fields */
348 				switch (fmt[i + 1]) {
349 				case 'B': case 'R': case 'r':
350 				case 'b': case 'M': case 'm':
351 				case 'I': case 'i': case 'E':
352 				case 'U': case 'V': case 'N':
353 				case 'a': case 'd': case 'D':
354 				case 'g': case 't': case 'C':
355 				case 'O': case 'f':
356 					if (WARN_ONCE(arg == 63,
357 						      "Too many args for event: %s",
358 						      trace_event_name(call)))
359 						return;
360 					dereference_flags |= 1ULL << arg;
361 				}
362 				break;
363 			default:
364 			{
365 				bool star = false;
366 				int j;
367 
368 				/* Increment arg if %*s exists. */
369 				for (j = 0; fmt[i + j]; j++) {
370 					if (isdigit(fmt[i + j]) ||
371 					    fmt[i + j] == '.')
372 						continue;
373 					if (fmt[i + j] == '*') {
374 						star = true;
375 						continue;
376 					}
377 					if ((fmt[i + j] == 's') && star)
378 						arg++;
379 					break;
380 				}
381 				break;
382 			} /* default */
383 
384 			} /* switch */
385 			arg++;
386 			continue;
387 		case '(':
388 			if (in_quote)
389 				continue;
390 			parens++;
391 			continue;
392 		case ')':
393 			if (in_quote)
394 				continue;
395 			parens--;
396 			if (WARN_ONCE(parens < 0,
397 				      "Paren mismatch for event: %s\narg='%s'\n%*s",
398 				      trace_event_name(call),
399 				      fmt + start_arg,
400 				      (i - start_arg) + 5, "^"))
401 				return;
402 			continue;
403 		case ',':
404 			if (in_quote || parens)
405 				continue;
406 			i++;
407 			while (isspace(fmt[i]))
408 				i++;
409 			start_arg = i;
410 			if (!(dereference_flags & (1ULL << arg)))
411 				goto next_arg;
412 
413 			/* Find the REC-> in the argument */
414 			c = strchr(fmt + i, ',');
415 			r = strstr(fmt + i, "REC->");
416 			if (r && (!c || r < c)) {
417 				/*
418 				 * Addresses of events on the buffer,
419 				 * or an array on the buffer is
420 				 * OK to dereference.
421 				 * There's ways to fool this, but
422 				 * this is to catch common mistakes,
423 				 * not malicious code.
424 				 */
425 				a = strchr(fmt + i, '&');
426 				if ((a && (a < r)) || test_field(r, call))
427 					dereference_flags &= ~(1ULL << arg);
428 			} else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
429 				   (!c || r < c)) {
430 				dereference_flags &= ~(1ULL << arg);
431 			} else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
432 				   (!c || r < c)) {
433 				dereference_flags &= ~(1ULL << arg);
434 			}
435 
436 		next_arg:
437 			i--;
438 			arg++;
439 		}
440 	}
441 
442 	/*
443 	 * If you triggered the below warning, the trace event reported
444 	 * uses an unsafe dereference pointer %p*. As the data stored
445 	 * at the trace event time may no longer exist when the trace
446 	 * event is printed, dereferencing to the original source is
447 	 * unsafe. The source of the dereference must be copied into the
448 	 * event itself, and the dereference must access the copy instead.
449 	 */
450 	if (WARN_ON_ONCE(dereference_flags)) {
451 		arg = 1;
452 		while (!(dereference_flags & 1)) {
453 			dereference_flags >>= 1;
454 			arg++;
455 		}
456 		pr_warn("event %s has unsafe dereference of argument %d\n",
457 			trace_event_name(call), arg);
458 		pr_warn("print_fmt: %s\n", fmt);
459 	}
460 }
461 
462 int trace_event_raw_init(struct trace_event_call *call)
463 {
464 	int id;
465 
466 	id = register_trace_event(&call->event);
467 	if (!id)
468 		return -ENODEV;
469 
470 	test_event_printk(call);
471 
472 	return 0;
473 }
474 EXPORT_SYMBOL_GPL(trace_event_raw_init);
475 
476 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
477 {
478 	struct trace_array *tr = trace_file->tr;
479 	struct trace_array_cpu *data;
480 	struct trace_pid_list *no_pid_list;
481 	struct trace_pid_list *pid_list;
482 
483 	pid_list = rcu_dereference_raw(tr->filtered_pids);
484 	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
485 
486 	if (!pid_list && !no_pid_list)
487 		return false;
488 
489 	data = this_cpu_ptr(tr->array_buffer.data);
490 
491 	return data->ignore_pid;
492 }
493 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
494 
495 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
496 				 struct trace_event_file *trace_file,
497 				 unsigned long len)
498 {
499 	struct trace_event_call *event_call = trace_file->event_call;
500 
501 	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
502 	    trace_event_ignore_this_pid(trace_file))
503 		return NULL;
504 
505 	/*
506 	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
507 	 * preemption (adding one to the preempt_count). Since we are
508 	 * interested in the preempt_count at the time the tracepoint was
509 	 * hit, we need to subtract one to offset the increment.
510 	 */
511 	fbuffer->trace_ctx = tracing_gen_ctx_dec();
512 	fbuffer->trace_file = trace_file;
513 
514 	fbuffer->event =
515 		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
516 						event_call->event.type, len,
517 						fbuffer->trace_ctx);
518 	if (!fbuffer->event)
519 		return NULL;
520 
521 	fbuffer->regs = NULL;
522 	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
523 	return fbuffer->entry;
524 }
525 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
526 
527 int trace_event_reg(struct trace_event_call *call,
528 		    enum trace_reg type, void *data)
529 {
530 	struct trace_event_file *file = data;
531 
532 	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
533 	switch (type) {
534 	case TRACE_REG_REGISTER:
535 		return tracepoint_probe_register(call->tp,
536 						 call->class->probe,
537 						 file);
538 	case TRACE_REG_UNREGISTER:
539 		tracepoint_probe_unregister(call->tp,
540 					    call->class->probe,
541 					    file);
542 		return 0;
543 
544 #ifdef CONFIG_PERF_EVENTS
545 	case TRACE_REG_PERF_REGISTER:
546 		return tracepoint_probe_register(call->tp,
547 						 call->class->perf_probe,
548 						 call);
549 	case TRACE_REG_PERF_UNREGISTER:
550 		tracepoint_probe_unregister(call->tp,
551 					    call->class->perf_probe,
552 					    call);
553 		return 0;
554 	case TRACE_REG_PERF_OPEN:
555 	case TRACE_REG_PERF_CLOSE:
556 	case TRACE_REG_PERF_ADD:
557 	case TRACE_REG_PERF_DEL:
558 		return 0;
559 #endif
560 	}
561 	return 0;
562 }
563 EXPORT_SYMBOL_GPL(trace_event_reg);
564 
565 void trace_event_enable_cmd_record(bool enable)
566 {
567 	struct trace_event_file *file;
568 	struct trace_array *tr;
569 
570 	lockdep_assert_held(&event_mutex);
571 
572 	do_for_each_event_file(tr, file) {
573 
574 		if (!(file->flags & EVENT_FILE_FL_ENABLED))
575 			continue;
576 
577 		if (enable) {
578 			tracing_start_cmdline_record();
579 			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
580 		} else {
581 			tracing_stop_cmdline_record();
582 			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
583 		}
584 	} while_for_each_event_file();
585 }
586 
587 void trace_event_enable_tgid_record(bool enable)
588 {
589 	struct trace_event_file *file;
590 	struct trace_array *tr;
591 
592 	lockdep_assert_held(&event_mutex);
593 
594 	do_for_each_event_file(tr, file) {
595 		if (!(file->flags & EVENT_FILE_FL_ENABLED))
596 			continue;
597 
598 		if (enable) {
599 			tracing_start_tgid_record();
600 			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
601 		} else {
602 			tracing_stop_tgid_record();
603 			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
604 				  &file->flags);
605 		}
606 	} while_for_each_event_file();
607 }
608 
609 static int __ftrace_event_enable_disable(struct trace_event_file *file,
610 					 int enable, int soft_disable)
611 {
612 	struct trace_event_call *call = file->event_call;
613 	struct trace_array *tr = file->tr;
614 	int ret = 0;
615 	int disable;
616 
617 	switch (enable) {
618 	case 0:
619 		/*
620 		 * When soft_disable is set and enable is cleared, the sm_ref
621 		 * reference counter is decremented. If it reaches 0, we want
622 		 * to clear the SOFT_DISABLED flag but leave the event in the
623 		 * state that it was. That is, if the event was enabled and
624 		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
625 		 * is set we do not want the event to be enabled before we
626 		 * clear the bit.
627 		 *
628 		 * When soft_disable is not set but the SOFT_MODE flag is,
629 		 * we do nothing. Do not disable the tracepoint, otherwise
630 		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
631 		 */
632 		if (soft_disable) {
633 			if (atomic_dec_return(&file->sm_ref) > 0)
634 				break;
635 			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
636 			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
637 			/* Disable use of trace_buffered_event */
638 			trace_buffered_event_disable();
639 		} else
640 			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
641 
642 		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
643 			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
644 			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
645 				tracing_stop_cmdline_record();
646 				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
647 			}
648 
649 			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
650 				tracing_stop_tgid_record();
651 				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
652 			}
653 
654 			call->class->reg(call, TRACE_REG_UNREGISTER, file);
655 		}
656 		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
657 		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
658 			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
659 		else
660 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
661 		break;
662 	case 1:
663 		/*
664 		 * When soft_disable is set and enable is set, we want to
665 		 * register the tracepoint for the event, but leave the event
666 		 * as is. That means, if the event was already enabled, we do
667 		 * nothing (but set SOFT_MODE). If the event is disabled, we
668 		 * set SOFT_DISABLED before enabling the event tracepoint, so
669 		 * it still seems to be disabled.
670 		 */
671 		if (!soft_disable)
672 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
673 		else {
674 			if (atomic_inc_return(&file->sm_ref) > 1)
675 				break;
676 			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
677 			/* Enable use of trace_buffered_event */
678 			trace_buffered_event_enable();
679 		}
680 
681 		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
682 			bool cmd = false, tgid = false;
683 
684 			/* Keep the event disabled, when going to SOFT_MODE. */
685 			if (soft_disable)
686 				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
687 
688 			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
689 				cmd = true;
690 				tracing_start_cmdline_record();
691 				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
692 			}
693 
694 			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
695 				tgid = true;
696 				tracing_start_tgid_record();
697 				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
698 			}
699 
700 			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
701 			if (ret) {
702 				if (cmd)
703 					tracing_stop_cmdline_record();
704 				if (tgid)
705 					tracing_stop_tgid_record();
706 				pr_info("event trace: Could not enable event "
707 					"%s\n", trace_event_name(call));
708 				break;
709 			}
710 			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
711 
712 			/* WAS_ENABLED gets set but never cleared. */
713 			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
714 		}
715 		break;
716 	}
717 
718 	return ret;
719 }
720 
721 int trace_event_enable_disable(struct trace_event_file *file,
722 			       int enable, int soft_disable)
723 {
724 	return __ftrace_event_enable_disable(file, enable, soft_disable);
725 }
726 
727 static int ftrace_event_enable_disable(struct trace_event_file *file,
728 				       int enable)
729 {
730 	return __ftrace_event_enable_disable(file, enable, 0);
731 }
732 
733 static void ftrace_clear_events(struct trace_array *tr)
734 {
735 	struct trace_event_file *file;
736 
737 	mutex_lock(&event_mutex);
738 	list_for_each_entry(file, &tr->events, list) {
739 		ftrace_event_enable_disable(file, 0);
740 	}
741 	mutex_unlock(&event_mutex);
742 }
743 
744 static void
745 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
746 {
747 	struct trace_pid_list *pid_list;
748 	struct trace_array *tr = data;
749 
750 	pid_list = rcu_dereference_raw(tr->filtered_pids);
751 	trace_filter_add_remove_task(pid_list, NULL, task);
752 
753 	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
754 	trace_filter_add_remove_task(pid_list, NULL, task);
755 }
756 
757 static void
758 event_filter_pid_sched_process_fork(void *data,
759 				    struct task_struct *self,
760 				    struct task_struct *task)
761 {
762 	struct trace_pid_list *pid_list;
763 	struct trace_array *tr = data;
764 
765 	pid_list = rcu_dereference_sched(tr->filtered_pids);
766 	trace_filter_add_remove_task(pid_list, self, task);
767 
768 	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
769 	trace_filter_add_remove_task(pid_list, self, task);
770 }
771 
772 void trace_event_follow_fork(struct trace_array *tr, bool enable)
773 {
774 	if (enable) {
775 		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
776 						       tr, INT_MIN);
777 		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
778 						       tr, INT_MAX);
779 	} else {
780 		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
781 						    tr);
782 		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
783 						    tr);
784 	}
785 }
786 
787 static void
788 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
789 					struct task_struct *prev,
790 					struct task_struct *next,
791 					unsigned int prev_state)
792 {
793 	struct trace_array *tr = data;
794 	struct trace_pid_list *no_pid_list;
795 	struct trace_pid_list *pid_list;
796 	bool ret;
797 
798 	pid_list = rcu_dereference_sched(tr->filtered_pids);
799 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
800 
801 	/*
802 	 * Sched switch is funny, as we only want to ignore it
803 	 * in the notrace case if both prev and next should be ignored.
804 	 */
805 	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
806 		trace_ignore_this_task(NULL, no_pid_list, next);
807 
808 	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
809 		       (trace_ignore_this_task(pid_list, NULL, prev) &&
810 			trace_ignore_this_task(pid_list, NULL, next)));
811 }
812 
813 static void
814 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
815 					 struct task_struct *prev,
816 					 struct task_struct *next,
817 					 unsigned int prev_state)
818 {
819 	struct trace_array *tr = data;
820 	struct trace_pid_list *no_pid_list;
821 	struct trace_pid_list *pid_list;
822 
823 	pid_list = rcu_dereference_sched(tr->filtered_pids);
824 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
825 
826 	this_cpu_write(tr->array_buffer.data->ignore_pid,
827 		       trace_ignore_this_task(pid_list, no_pid_list, next));
828 }
829 
830 static void
831 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
832 {
833 	struct trace_array *tr = data;
834 	struct trace_pid_list *no_pid_list;
835 	struct trace_pid_list *pid_list;
836 
837 	/* Nothing to do if we are already tracing */
838 	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
839 		return;
840 
841 	pid_list = rcu_dereference_sched(tr->filtered_pids);
842 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
843 
844 	this_cpu_write(tr->array_buffer.data->ignore_pid,
845 		       trace_ignore_this_task(pid_list, no_pid_list, task));
846 }
847 
848 static void
849 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
850 {
851 	struct trace_array *tr = data;
852 	struct trace_pid_list *no_pid_list;
853 	struct trace_pid_list *pid_list;
854 
855 	/* Nothing to do if we are not tracing */
856 	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
857 		return;
858 
859 	pid_list = rcu_dereference_sched(tr->filtered_pids);
860 	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
861 
862 	/* Set tracing if current is enabled */
863 	this_cpu_write(tr->array_buffer.data->ignore_pid,
864 		       trace_ignore_this_task(pid_list, no_pid_list, current));
865 }
866 
867 static void unregister_pid_events(struct trace_array *tr)
868 {
869 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
870 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
871 
872 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
873 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
874 
875 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
876 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
877 
878 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
879 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
880 }
881 
882 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
883 {
884 	struct trace_pid_list *pid_list;
885 	struct trace_pid_list *no_pid_list;
886 	struct trace_event_file *file;
887 	int cpu;
888 
889 	pid_list = rcu_dereference_protected(tr->filtered_pids,
890 					     lockdep_is_held(&event_mutex));
891 	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
892 					     lockdep_is_held(&event_mutex));
893 
894 	/* Make sure there's something to do */
895 	if (!pid_type_enabled(type, pid_list, no_pid_list))
896 		return;
897 
898 	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
899 		unregister_pid_events(tr);
900 
901 		list_for_each_entry(file, &tr->events, list) {
902 			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
903 		}
904 
905 		for_each_possible_cpu(cpu)
906 			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
907 	}
908 
909 	if (type & TRACE_PIDS)
910 		rcu_assign_pointer(tr->filtered_pids, NULL);
911 
912 	if (type & TRACE_NO_PIDS)
913 		rcu_assign_pointer(tr->filtered_no_pids, NULL);
914 
915 	/* Wait till all users are no longer using pid filtering */
916 	tracepoint_synchronize_unregister();
917 
918 	if ((type & TRACE_PIDS) && pid_list)
919 		trace_pid_list_free(pid_list);
920 
921 	if ((type & TRACE_NO_PIDS) && no_pid_list)
922 		trace_pid_list_free(no_pid_list);
923 }
924 
925 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
926 {
927 	mutex_lock(&event_mutex);
928 	__ftrace_clear_event_pids(tr, type);
929 	mutex_unlock(&event_mutex);
930 }
931 
932 static void __put_system(struct event_subsystem *system)
933 {
934 	struct event_filter *filter = system->filter;
935 
936 	WARN_ON_ONCE(system_refcount(system) == 0);
937 	if (system_refcount_dec(system))
938 		return;
939 
940 	list_del(&system->list);
941 
942 	if (filter) {
943 		kfree(filter->filter_string);
944 		kfree(filter);
945 	}
946 	kfree_const(system->name);
947 	kfree(system);
948 }
949 
950 static void __get_system(struct event_subsystem *system)
951 {
952 	WARN_ON_ONCE(system_refcount(system) == 0);
953 	system_refcount_inc(system);
954 }
955 
956 static void __get_system_dir(struct trace_subsystem_dir *dir)
957 {
958 	WARN_ON_ONCE(dir->ref_count == 0);
959 	dir->ref_count++;
960 	__get_system(dir->subsystem);
961 }
962 
963 static void __put_system_dir(struct trace_subsystem_dir *dir)
964 {
965 	WARN_ON_ONCE(dir->ref_count == 0);
966 	/* If the subsystem is about to be freed, the dir must be too */
967 	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
968 
969 	__put_system(dir->subsystem);
970 	if (!--dir->ref_count)
971 		kfree(dir);
972 }
973 
974 static void put_system(struct trace_subsystem_dir *dir)
975 {
976 	mutex_lock(&event_mutex);
977 	__put_system_dir(dir);
978 	mutex_unlock(&event_mutex);
979 }
980 
981 static void remove_subsystem(struct trace_subsystem_dir *dir)
982 {
983 	if (!dir)
984 		return;
985 
986 	if (!--dir->nr_events) {
987 		eventfs_remove_dir(dir->ei);
988 		list_del(&dir->list);
989 		__put_system_dir(dir);
990 	}
991 }
992 
993 void event_file_get(struct trace_event_file *file)
994 {
995 	atomic_inc(&file->ref);
996 }
997 
998 void event_file_put(struct trace_event_file *file)
999 {
1000 	if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001 		if (file->flags & EVENT_FILE_FL_FREED)
1002 			kmem_cache_free(file_cachep, file);
1003 		return;
1004 	}
1005 
1006 	if (atomic_dec_and_test(&file->ref)) {
1007 		/* Count should only go to zero when it is freed */
1008 		if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009 			return;
1010 		kmem_cache_free(file_cachep, file);
1011 	}
1012 }
1013 
1014 static void remove_event_file_dir(struct trace_event_file *file)
1015 {
1016 	eventfs_remove_dir(file->ei);
1017 	list_del(&file->list);
1018 	remove_subsystem(file->system);
1019 	free_event_filter(file->filter);
1020 	file->flags |= EVENT_FILE_FL_FREED;
1021 	event_file_put(file);
1022 }
1023 
1024 /*
1025  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026  */
1027 static int
1028 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029 			      const char *sub, const char *event, int set)
1030 {
1031 	struct trace_event_file *file;
1032 	struct trace_event_call *call;
1033 	const char *name;
1034 	int ret = -EINVAL;
1035 	int eret = 0;
1036 
1037 	list_for_each_entry(file, &tr->events, list) {
1038 
1039 		call = file->event_call;
1040 		name = trace_event_name(call);
1041 
1042 		if (!name || !call->class || !call->class->reg)
1043 			continue;
1044 
1045 		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046 			continue;
1047 
1048 		if (match &&
1049 		    strcmp(match, name) != 0 &&
1050 		    strcmp(match, call->class->system) != 0)
1051 			continue;
1052 
1053 		if (sub && strcmp(sub, call->class->system) != 0)
1054 			continue;
1055 
1056 		if (event && strcmp(event, name) != 0)
1057 			continue;
1058 
1059 		ret = ftrace_event_enable_disable(file, set);
1060 
1061 		/*
1062 		 * Save the first error and return that. Some events
1063 		 * may still have been enabled, but let the user
1064 		 * know that something went wrong.
1065 		 */
1066 		if (ret && !eret)
1067 			eret = ret;
1068 
1069 		ret = eret;
1070 	}
1071 
1072 	return ret;
1073 }
1074 
1075 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076 				  const char *sub, const char *event, int set)
1077 {
1078 	int ret;
1079 
1080 	mutex_lock(&event_mutex);
1081 	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082 	mutex_unlock(&event_mutex);
1083 
1084 	return ret;
1085 }
1086 
1087 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088 {
1089 	char *event = NULL, *sub = NULL, *match;
1090 	int ret;
1091 
1092 	if (!tr)
1093 		return -ENOENT;
1094 	/*
1095 	 * The buf format can be <subsystem>:<event-name>
1096 	 *  *:<event-name> means any event by that name.
1097 	 *  :<event-name> is the same.
1098 	 *
1099 	 *  <subsystem>:* means all events in that subsystem
1100 	 *  <subsystem>: means the same.
1101 	 *
1102 	 *  <name> (no ':') means all events in a subsystem with
1103 	 *  the name <name> or any event that matches <name>
1104 	 */
1105 
1106 	match = strsep(&buf, ":");
1107 	if (buf) {
1108 		sub = match;
1109 		event = buf;
1110 		match = NULL;
1111 
1112 		if (!strlen(sub) || strcmp(sub, "*") == 0)
1113 			sub = NULL;
1114 		if (!strlen(event) || strcmp(event, "*") == 0)
1115 			event = NULL;
1116 	}
1117 
1118 	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119 
1120 	/* Put back the colon to allow this to be called again */
1121 	if (buf)
1122 		*(buf - 1) = ':';
1123 
1124 	return ret;
1125 }
1126 
1127 /**
1128  * trace_set_clr_event - enable or disable an event
1129  * @system: system name to match (NULL for any system)
1130  * @event: event name to match (NULL for all events, within system)
1131  * @set: 1 to enable, 0 to disable
1132  *
1133  * This is a way for other parts of the kernel to enable or disable
1134  * event recording.
1135  *
1136  * Returns 0 on success, -EINVAL if the parameters do not match any
1137  * registered events.
1138  */
1139 int trace_set_clr_event(const char *system, const char *event, int set)
1140 {
1141 	struct trace_array *tr = top_trace_array();
1142 
1143 	if (!tr)
1144 		return -ENODEV;
1145 
1146 	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147 }
1148 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149 
1150 /**
1151  * trace_array_set_clr_event - enable or disable an event for a trace array.
1152  * @tr: concerned trace array.
1153  * @system: system name to match (NULL for any system)
1154  * @event: event name to match (NULL for all events, within system)
1155  * @enable: true to enable, false to disable
1156  *
1157  * This is a way for other parts of the kernel to enable or disable
1158  * event recording.
1159  *
1160  * Returns 0 on success, -EINVAL if the parameters do not match any
1161  * registered events.
1162  */
1163 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164 		const char *event, bool enable)
1165 {
1166 	int set;
1167 
1168 	if (!tr)
1169 		return -ENOENT;
1170 
1171 	set = (enable == true) ? 1 : 0;
1172 	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173 }
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175 
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE		127
1178 
1179 static ssize_t
1180 ftrace_event_write(struct file *file, const char __user *ubuf,
1181 		   size_t cnt, loff_t *ppos)
1182 {
1183 	struct trace_parser parser;
1184 	struct seq_file *m = file->private_data;
1185 	struct trace_array *tr = m->private;
1186 	ssize_t read, ret;
1187 
1188 	if (!cnt)
1189 		return 0;
1190 
1191 	ret = tracing_update_buffers();
1192 	if (ret < 0)
1193 		return ret;
1194 
1195 	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196 		return -ENOMEM;
1197 
1198 	read = trace_get_user(&parser, ubuf, cnt, ppos);
1199 
1200 	if (read >= 0 && trace_parser_loaded((&parser))) {
1201 		int set = 1;
1202 
1203 		if (*parser.buffer == '!')
1204 			set = 0;
1205 
1206 		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207 		if (ret)
1208 			goto out_put;
1209 	}
1210 
1211 	ret = read;
1212 
1213  out_put:
1214 	trace_parser_put(&parser);
1215 
1216 	return ret;
1217 }
1218 
1219 static void *
1220 t_next(struct seq_file *m, void *v, loff_t *pos)
1221 {
1222 	struct trace_event_file *file = v;
1223 	struct trace_event_call *call;
1224 	struct trace_array *tr = m->private;
1225 
1226 	(*pos)++;
1227 
1228 	list_for_each_entry_continue(file, &tr->events, list) {
1229 		call = file->event_call;
1230 		/*
1231 		 * The ftrace subsystem is for showing formats only.
1232 		 * They can not be enabled or disabled via the event files.
1233 		 */
1234 		if (call->class && call->class->reg &&
1235 		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236 			return file;
1237 	}
1238 
1239 	return NULL;
1240 }
1241 
1242 static void *t_start(struct seq_file *m, loff_t *pos)
1243 {
1244 	struct trace_event_file *file;
1245 	struct trace_array *tr = m->private;
1246 	loff_t l;
1247 
1248 	mutex_lock(&event_mutex);
1249 
1250 	file = list_entry(&tr->events, struct trace_event_file, list);
1251 	for (l = 0; l <= *pos; ) {
1252 		file = t_next(m, file, &l);
1253 		if (!file)
1254 			break;
1255 	}
1256 	return file;
1257 }
1258 
1259 static void *
1260 s_next(struct seq_file *m, void *v, loff_t *pos)
1261 {
1262 	struct trace_event_file *file = v;
1263 	struct trace_array *tr = m->private;
1264 
1265 	(*pos)++;
1266 
1267 	list_for_each_entry_continue(file, &tr->events, list) {
1268 		if (file->flags & EVENT_FILE_FL_ENABLED)
1269 			return file;
1270 	}
1271 
1272 	return NULL;
1273 }
1274 
1275 static void *s_start(struct seq_file *m, loff_t *pos)
1276 {
1277 	struct trace_event_file *file;
1278 	struct trace_array *tr = m->private;
1279 	loff_t l;
1280 
1281 	mutex_lock(&event_mutex);
1282 
1283 	file = list_entry(&tr->events, struct trace_event_file, list);
1284 	for (l = 0; l <= *pos; ) {
1285 		file = s_next(m, file, &l);
1286 		if (!file)
1287 			break;
1288 	}
1289 	return file;
1290 }
1291 
1292 static int t_show(struct seq_file *m, void *v)
1293 {
1294 	struct trace_event_file *file = v;
1295 	struct trace_event_call *call = file->event_call;
1296 
1297 	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298 		seq_printf(m, "%s:", call->class->system);
1299 	seq_printf(m, "%s\n", trace_event_name(call));
1300 
1301 	return 0;
1302 }
1303 
1304 static void t_stop(struct seq_file *m, void *p)
1305 {
1306 	mutex_unlock(&event_mutex);
1307 }
1308 
1309 static void *
1310 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1311 {
1312 	struct trace_array *tr = m->private;
1313 	struct trace_pid_list *pid_list;
1314 
1315 	if (type == TRACE_PIDS)
1316 		pid_list = rcu_dereference_sched(tr->filtered_pids);
1317 	else
1318 		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319 
1320 	return trace_pid_next(pid_list, v, pos);
1321 }
1322 
1323 static void *
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326 	return __next(m, v, pos, TRACE_PIDS);
1327 }
1328 
1329 static void *
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1331 {
1332 	return __next(m, v, pos, TRACE_NO_PIDS);
1333 }
1334 
1335 static void *__start(struct seq_file *m, loff_t *pos, int type)
1336 	__acquires(RCU)
1337 {
1338 	struct trace_pid_list *pid_list;
1339 	struct trace_array *tr = m->private;
1340 
1341 	/*
1342 	 * Grab the mutex, to keep calls to p_next() having the same
1343 	 * tr->filtered_pids as p_start() has.
1344 	 * If we just passed the tr->filtered_pids around, then RCU would
1345 	 * have been enough, but doing that makes things more complex.
1346 	 */
1347 	mutex_lock(&event_mutex);
1348 	rcu_read_lock_sched();
1349 
1350 	if (type == TRACE_PIDS)
1351 		pid_list = rcu_dereference_sched(tr->filtered_pids);
1352 	else
1353 		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354 
1355 	if (!pid_list)
1356 		return NULL;
1357 
1358 	return trace_pid_start(pid_list, pos);
1359 }
1360 
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1362 	__acquires(RCU)
1363 {
1364 	return __start(m, pos, TRACE_PIDS);
1365 }
1366 
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1368 	__acquires(RCU)
1369 {
1370 	return __start(m, pos, TRACE_NO_PIDS);
1371 }
1372 
1373 static void p_stop(struct seq_file *m, void *p)
1374 	__releases(RCU)
1375 {
1376 	rcu_read_unlock_sched();
1377 	mutex_unlock(&event_mutex);
1378 }
1379 
1380 static ssize_t
1381 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382 		  loff_t *ppos)
1383 {
1384 	struct trace_event_file *file;
1385 	unsigned long flags;
1386 	char buf[4] = "0";
1387 
1388 	mutex_lock(&event_mutex);
1389 	file = event_file_file(filp);
1390 	if (likely(file))
1391 		flags = file->flags;
1392 	mutex_unlock(&event_mutex);
1393 
1394 	if (!file)
1395 		return -ENODEV;
1396 
1397 	if (flags & EVENT_FILE_FL_ENABLED &&
1398 	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399 		strcpy(buf, "1");
1400 
1401 	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402 	    flags & EVENT_FILE_FL_SOFT_MODE)
1403 		strcat(buf, "*");
1404 
1405 	strcat(buf, "\n");
1406 
1407 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408 }
1409 
1410 static ssize_t
1411 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412 		   loff_t *ppos)
1413 {
1414 	struct trace_event_file *file;
1415 	unsigned long val;
1416 	int ret;
1417 
1418 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419 	if (ret)
1420 		return ret;
1421 
1422 	ret = tracing_update_buffers();
1423 	if (ret < 0)
1424 		return ret;
1425 
1426 	switch (val) {
1427 	case 0:
1428 	case 1:
1429 		ret = -ENODEV;
1430 		mutex_lock(&event_mutex);
1431 		file = event_file_file(filp);
1432 		if (likely(file))
1433 			ret = ftrace_event_enable_disable(file, val);
1434 		mutex_unlock(&event_mutex);
1435 		break;
1436 
1437 	default:
1438 		return -EINVAL;
1439 	}
1440 
1441 	*ppos += cnt;
1442 
1443 	return ret ? ret : cnt;
1444 }
1445 
1446 static ssize_t
1447 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1448 		   loff_t *ppos)
1449 {
1450 	const char set_to_char[4] = { '?', '0', '1', 'X' };
1451 	struct trace_subsystem_dir *dir = filp->private_data;
1452 	struct event_subsystem *system = dir->subsystem;
1453 	struct trace_event_call *call;
1454 	struct trace_event_file *file;
1455 	struct trace_array *tr = dir->tr;
1456 	char buf[2];
1457 	int set = 0;
1458 	int ret;
1459 
1460 	mutex_lock(&event_mutex);
1461 	list_for_each_entry(file, &tr->events, list) {
1462 		call = file->event_call;
1463 		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1464 		    !trace_event_name(call) || !call->class || !call->class->reg)
1465 			continue;
1466 
1467 		if (system && strcmp(call->class->system, system->name) != 0)
1468 			continue;
1469 
1470 		/*
1471 		 * We need to find out if all the events are set
1472 		 * or if all events or cleared, or if we have
1473 		 * a mixture.
1474 		 */
1475 		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1476 
1477 		/*
1478 		 * If we have a mixture, no need to look further.
1479 		 */
1480 		if (set == 3)
1481 			break;
1482 	}
1483 	mutex_unlock(&event_mutex);
1484 
1485 	buf[0] = set_to_char[set];
1486 	buf[1] = '\n';
1487 
1488 	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1489 
1490 	return ret;
1491 }
1492 
1493 static ssize_t
1494 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1495 		    loff_t *ppos)
1496 {
1497 	struct trace_subsystem_dir *dir = filp->private_data;
1498 	struct event_subsystem *system = dir->subsystem;
1499 	const char *name = NULL;
1500 	unsigned long val;
1501 	ssize_t ret;
1502 
1503 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1504 	if (ret)
1505 		return ret;
1506 
1507 	ret = tracing_update_buffers();
1508 	if (ret < 0)
1509 		return ret;
1510 
1511 	if (val != 0 && val != 1)
1512 		return -EINVAL;
1513 
1514 	/*
1515 	 * Opening of "enable" adds a ref count to system,
1516 	 * so the name is safe to use.
1517 	 */
1518 	if (system)
1519 		name = system->name;
1520 
1521 	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1522 	if (ret)
1523 		goto out;
1524 
1525 	ret = cnt;
1526 
1527 out:
1528 	*ppos += cnt;
1529 
1530 	return ret;
1531 }
1532 
1533 enum {
1534 	FORMAT_HEADER		= 1,
1535 	FORMAT_FIELD_SEPERATOR	= 2,
1536 	FORMAT_PRINTFMT		= 3,
1537 };
1538 
1539 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1540 {
1541 	struct trace_event_file *file = event_file_data(m->private);
1542 	struct trace_event_call *call = file->event_call;
1543 	struct list_head *common_head = &ftrace_common_fields;
1544 	struct list_head *head = trace_get_fields(call);
1545 	struct list_head *node = v;
1546 
1547 	(*pos)++;
1548 
1549 	switch ((unsigned long)v) {
1550 	case FORMAT_HEADER:
1551 		node = common_head;
1552 		break;
1553 
1554 	case FORMAT_FIELD_SEPERATOR:
1555 		node = head;
1556 		break;
1557 
1558 	case FORMAT_PRINTFMT:
1559 		/* all done */
1560 		return NULL;
1561 	}
1562 
1563 	node = node->prev;
1564 	if (node == common_head)
1565 		return (void *)FORMAT_FIELD_SEPERATOR;
1566 	else if (node == head)
1567 		return (void *)FORMAT_PRINTFMT;
1568 	else
1569 		return node;
1570 }
1571 
1572 static int f_show(struct seq_file *m, void *v)
1573 {
1574 	struct trace_event_file *file = event_file_data(m->private);
1575 	struct trace_event_call *call = file->event_call;
1576 	struct ftrace_event_field *field;
1577 	const char *array_descriptor;
1578 
1579 	switch ((unsigned long)v) {
1580 	case FORMAT_HEADER:
1581 		seq_printf(m, "name: %s\n", trace_event_name(call));
1582 		seq_printf(m, "ID: %d\n", call->event.type);
1583 		seq_puts(m, "format:\n");
1584 		return 0;
1585 
1586 	case FORMAT_FIELD_SEPERATOR:
1587 		seq_putc(m, '\n');
1588 		return 0;
1589 
1590 	case FORMAT_PRINTFMT:
1591 		seq_printf(m, "\nprint fmt: %s\n",
1592 			   call->print_fmt);
1593 		return 0;
1594 	}
1595 
1596 	field = list_entry(v, struct ftrace_event_field, link);
1597 	/*
1598 	 * Smartly shows the array type(except dynamic array).
1599 	 * Normal:
1600 	 *	field:TYPE VAR
1601 	 * If TYPE := TYPE[LEN], it is shown:
1602 	 *	field:TYPE VAR[LEN]
1603 	 */
1604 	array_descriptor = strchr(field->type, '[');
1605 
1606 	if (str_has_prefix(field->type, "__data_loc"))
1607 		array_descriptor = NULL;
1608 
1609 	if (!array_descriptor)
1610 		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611 			   field->type, field->name, field->offset,
1612 			   field->size, !!field->is_signed);
1613 	else if (field->len)
1614 		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615 			   (int)(array_descriptor - field->type),
1616 			   field->type, field->name,
1617 			   field->len, field->offset,
1618 			   field->size, !!field->is_signed);
1619 	else
1620 		seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621 				(int)(array_descriptor - field->type),
1622 				field->type, field->name,
1623 				field->offset, field->size, !!field->is_signed);
1624 
1625 	return 0;
1626 }
1627 
1628 static void *f_start(struct seq_file *m, loff_t *pos)
1629 {
1630 	struct trace_event_file *file;
1631 	void *p = (void *)FORMAT_HEADER;
1632 	loff_t l = 0;
1633 
1634 	/* ->stop() is called even if ->start() fails */
1635 	mutex_lock(&event_mutex);
1636 	file = event_file_file(m->private);
1637 	if (!file)
1638 		return ERR_PTR(-ENODEV);
1639 
1640 	while (l < *pos && p)
1641 		p = f_next(m, p, &l);
1642 
1643 	return p;
1644 }
1645 
1646 static void f_stop(struct seq_file *m, void *p)
1647 {
1648 	mutex_unlock(&event_mutex);
1649 }
1650 
1651 static const struct seq_operations trace_format_seq_ops = {
1652 	.start		= f_start,
1653 	.next		= f_next,
1654 	.stop		= f_stop,
1655 	.show		= f_show,
1656 };
1657 
1658 static int trace_format_open(struct inode *inode, struct file *file)
1659 {
1660 	struct seq_file *m;
1661 	int ret;
1662 
1663 	/* Do we want to hide event format files on tracefs lockdown? */
1664 
1665 	ret = seq_open(file, &trace_format_seq_ops);
1666 	if (ret < 0)
1667 		return ret;
1668 
1669 	m = file->private_data;
1670 	m->private = file;
1671 
1672 	return 0;
1673 }
1674 
1675 #ifdef CONFIG_PERF_EVENTS
1676 static ssize_t
1677 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1678 {
1679 	int id = (long)event_file_data(filp);
1680 	char buf[32];
1681 	int len;
1682 
1683 	if (unlikely(!id))
1684 		return -ENODEV;
1685 
1686 	len = sprintf(buf, "%d\n", id);
1687 
1688 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1689 }
1690 #endif
1691 
1692 static ssize_t
1693 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1694 		  loff_t *ppos)
1695 {
1696 	struct trace_event_file *file;
1697 	struct trace_seq *s;
1698 	int r = -ENODEV;
1699 
1700 	if (*ppos)
1701 		return 0;
1702 
1703 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1704 
1705 	if (!s)
1706 		return -ENOMEM;
1707 
1708 	trace_seq_init(s);
1709 
1710 	mutex_lock(&event_mutex);
1711 	file = event_file_file(filp);
1712 	if (file)
1713 		print_event_filter(file, s);
1714 	mutex_unlock(&event_mutex);
1715 
1716 	if (file)
1717 		r = simple_read_from_buffer(ubuf, cnt, ppos,
1718 					    s->buffer, trace_seq_used(s));
1719 
1720 	kfree(s);
1721 
1722 	return r;
1723 }
1724 
1725 static ssize_t
1726 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1727 		   loff_t *ppos)
1728 {
1729 	struct trace_event_file *file;
1730 	char *buf;
1731 	int err = -ENODEV;
1732 
1733 	if (cnt >= PAGE_SIZE)
1734 		return -EINVAL;
1735 
1736 	buf = memdup_user_nul(ubuf, cnt);
1737 	if (IS_ERR(buf))
1738 		return PTR_ERR(buf);
1739 
1740 	mutex_lock(&event_mutex);
1741 	file = event_file_file(filp);
1742 	if (file) {
1743 		if (file->flags & EVENT_FILE_FL_FREED)
1744 			err = -ENODEV;
1745 		else
1746 			err = apply_event_filter(file, buf);
1747 	}
1748 	mutex_unlock(&event_mutex);
1749 
1750 	kfree(buf);
1751 	if (err < 0)
1752 		return err;
1753 
1754 	*ppos += cnt;
1755 
1756 	return cnt;
1757 }
1758 
1759 static LIST_HEAD(event_subsystems);
1760 
1761 static int subsystem_open(struct inode *inode, struct file *filp)
1762 {
1763 	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1764 	struct trace_array *tr = NULL, *iter_tr;
1765 	struct event_subsystem *system = NULL;
1766 	int ret;
1767 
1768 	if (tracing_is_disabled())
1769 		return -ENODEV;
1770 
1771 	/* Make sure the system still exists */
1772 	mutex_lock(&event_mutex);
1773 	mutex_lock(&trace_types_lock);
1774 	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1775 		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1776 			if (iter_dir == inode->i_private) {
1777 				/* Don't open systems with no events */
1778 				tr = iter_tr;
1779 				dir = iter_dir;
1780 				if (dir->nr_events) {
1781 					__get_system_dir(dir);
1782 					system = dir->subsystem;
1783 				}
1784 				goto exit_loop;
1785 			}
1786 		}
1787 	}
1788  exit_loop:
1789 	mutex_unlock(&trace_types_lock);
1790 	mutex_unlock(&event_mutex);
1791 
1792 	if (!system)
1793 		return -ENODEV;
1794 
1795 	/* Still need to increment the ref count of the system */
1796 	if (trace_array_get(tr) < 0) {
1797 		put_system(dir);
1798 		return -ENODEV;
1799 	}
1800 
1801 	ret = tracing_open_generic(inode, filp);
1802 	if (ret < 0) {
1803 		trace_array_put(tr);
1804 		put_system(dir);
1805 	}
1806 
1807 	return ret;
1808 }
1809 
1810 static int system_tr_open(struct inode *inode, struct file *filp)
1811 {
1812 	struct trace_subsystem_dir *dir;
1813 	struct trace_array *tr = inode->i_private;
1814 	int ret;
1815 
1816 	/* Make a temporary dir that has no system but points to tr */
1817 	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1818 	if (!dir)
1819 		return -ENOMEM;
1820 
1821 	ret = tracing_open_generic_tr(inode, filp);
1822 	if (ret < 0) {
1823 		kfree(dir);
1824 		return ret;
1825 	}
1826 	dir->tr = tr;
1827 	filp->private_data = dir;
1828 
1829 	return 0;
1830 }
1831 
1832 static int subsystem_release(struct inode *inode, struct file *file)
1833 {
1834 	struct trace_subsystem_dir *dir = file->private_data;
1835 
1836 	trace_array_put(dir->tr);
1837 
1838 	/*
1839 	 * If dir->subsystem is NULL, then this is a temporary
1840 	 * descriptor that was made for a trace_array to enable
1841 	 * all subsystems.
1842 	 */
1843 	if (dir->subsystem)
1844 		put_system(dir);
1845 	else
1846 		kfree(dir);
1847 
1848 	return 0;
1849 }
1850 
1851 static ssize_t
1852 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1853 		      loff_t *ppos)
1854 {
1855 	struct trace_subsystem_dir *dir = filp->private_data;
1856 	struct event_subsystem *system = dir->subsystem;
1857 	struct trace_seq *s;
1858 	int r;
1859 
1860 	if (*ppos)
1861 		return 0;
1862 
1863 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1864 	if (!s)
1865 		return -ENOMEM;
1866 
1867 	trace_seq_init(s);
1868 
1869 	print_subsystem_event_filter(system, s);
1870 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1871 				    s->buffer, trace_seq_used(s));
1872 
1873 	kfree(s);
1874 
1875 	return r;
1876 }
1877 
1878 static ssize_t
1879 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1880 		       loff_t *ppos)
1881 {
1882 	struct trace_subsystem_dir *dir = filp->private_data;
1883 	char *buf;
1884 	int err;
1885 
1886 	if (cnt >= PAGE_SIZE)
1887 		return -EINVAL;
1888 
1889 	buf = memdup_user_nul(ubuf, cnt);
1890 	if (IS_ERR(buf))
1891 		return PTR_ERR(buf);
1892 
1893 	err = apply_subsystem_event_filter(dir, buf);
1894 	kfree(buf);
1895 	if (err < 0)
1896 		return err;
1897 
1898 	*ppos += cnt;
1899 
1900 	return cnt;
1901 }
1902 
1903 static ssize_t
1904 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1905 {
1906 	int (*func)(struct trace_seq *s) = filp->private_data;
1907 	struct trace_seq *s;
1908 	int r;
1909 
1910 	if (*ppos)
1911 		return 0;
1912 
1913 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1914 	if (!s)
1915 		return -ENOMEM;
1916 
1917 	trace_seq_init(s);
1918 
1919 	func(s);
1920 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1921 				    s->buffer, trace_seq_used(s));
1922 
1923 	kfree(s);
1924 
1925 	return r;
1926 }
1927 
1928 static void ignore_task_cpu(void *data)
1929 {
1930 	struct trace_array *tr = data;
1931 	struct trace_pid_list *pid_list;
1932 	struct trace_pid_list *no_pid_list;
1933 
1934 	/*
1935 	 * This function is called by on_each_cpu() while the
1936 	 * event_mutex is held.
1937 	 */
1938 	pid_list = rcu_dereference_protected(tr->filtered_pids,
1939 					     mutex_is_locked(&event_mutex));
1940 	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1941 					     mutex_is_locked(&event_mutex));
1942 
1943 	this_cpu_write(tr->array_buffer.data->ignore_pid,
1944 		       trace_ignore_this_task(pid_list, no_pid_list, current));
1945 }
1946 
1947 static void register_pid_events(struct trace_array *tr)
1948 {
1949 	/*
1950 	 * Register a probe that is called before all other probes
1951 	 * to set ignore_pid if next or prev do not match.
1952 	 * Register a probe this is called after all other probes
1953 	 * to only keep ignore_pid set if next pid matches.
1954 	 */
1955 	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1956 					 tr, INT_MAX);
1957 	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1958 					 tr, 0);
1959 
1960 	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1961 					 tr, INT_MAX);
1962 	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1963 					 tr, 0);
1964 
1965 	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1966 					     tr, INT_MAX);
1967 	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1968 					     tr, 0);
1969 
1970 	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1971 					 tr, INT_MAX);
1972 	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1973 					 tr, 0);
1974 }
1975 
1976 static ssize_t
1977 event_pid_write(struct file *filp, const char __user *ubuf,
1978 		size_t cnt, loff_t *ppos, int type)
1979 {
1980 	struct seq_file *m = filp->private_data;
1981 	struct trace_array *tr = m->private;
1982 	struct trace_pid_list *filtered_pids = NULL;
1983 	struct trace_pid_list *other_pids = NULL;
1984 	struct trace_pid_list *pid_list;
1985 	struct trace_event_file *file;
1986 	ssize_t ret;
1987 
1988 	if (!cnt)
1989 		return 0;
1990 
1991 	ret = tracing_update_buffers();
1992 	if (ret < 0)
1993 		return ret;
1994 
1995 	mutex_lock(&event_mutex);
1996 
1997 	if (type == TRACE_PIDS) {
1998 		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1999 							  lockdep_is_held(&event_mutex));
2000 		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2001 							  lockdep_is_held(&event_mutex));
2002 	} else {
2003 		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2004 							  lockdep_is_held(&event_mutex));
2005 		other_pids = rcu_dereference_protected(tr->filtered_pids,
2006 							  lockdep_is_held(&event_mutex));
2007 	}
2008 
2009 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2010 	if (ret < 0)
2011 		goto out;
2012 
2013 	if (type == TRACE_PIDS)
2014 		rcu_assign_pointer(tr->filtered_pids, pid_list);
2015 	else
2016 		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2017 
2018 	list_for_each_entry(file, &tr->events, list) {
2019 		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2020 	}
2021 
2022 	if (filtered_pids) {
2023 		tracepoint_synchronize_unregister();
2024 		trace_pid_list_free(filtered_pids);
2025 	} else if (pid_list && !other_pids) {
2026 		register_pid_events(tr);
2027 	}
2028 
2029 	/*
2030 	 * Ignoring of pids is done at task switch. But we have to
2031 	 * check for those tasks that are currently running.
2032 	 * Always do this in case a pid was appended or removed.
2033 	 */
2034 	on_each_cpu(ignore_task_cpu, tr, 1);
2035 
2036  out:
2037 	mutex_unlock(&event_mutex);
2038 
2039 	if (ret > 0)
2040 		*ppos += ret;
2041 
2042 	return ret;
2043 }
2044 
2045 static ssize_t
2046 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2047 		       size_t cnt, loff_t *ppos)
2048 {
2049 	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2050 }
2051 
2052 static ssize_t
2053 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2054 			size_t cnt, loff_t *ppos)
2055 {
2056 	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2057 }
2058 
2059 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2060 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2061 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2062 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2063 static int ftrace_event_release(struct inode *inode, struct file *file);
2064 
2065 static const struct seq_operations show_event_seq_ops = {
2066 	.start = t_start,
2067 	.next = t_next,
2068 	.show = t_show,
2069 	.stop = t_stop,
2070 };
2071 
2072 static const struct seq_operations show_set_event_seq_ops = {
2073 	.start = s_start,
2074 	.next = s_next,
2075 	.show = t_show,
2076 	.stop = t_stop,
2077 };
2078 
2079 static const struct seq_operations show_set_pid_seq_ops = {
2080 	.start = p_start,
2081 	.next = p_next,
2082 	.show = trace_pid_show,
2083 	.stop = p_stop,
2084 };
2085 
2086 static const struct seq_operations show_set_no_pid_seq_ops = {
2087 	.start = np_start,
2088 	.next = np_next,
2089 	.show = trace_pid_show,
2090 	.stop = p_stop,
2091 };
2092 
2093 static const struct file_operations ftrace_avail_fops = {
2094 	.open = ftrace_event_avail_open,
2095 	.read = seq_read,
2096 	.llseek = seq_lseek,
2097 	.release = seq_release,
2098 };
2099 
2100 static const struct file_operations ftrace_set_event_fops = {
2101 	.open = ftrace_event_set_open,
2102 	.read = seq_read,
2103 	.write = ftrace_event_write,
2104 	.llseek = seq_lseek,
2105 	.release = ftrace_event_release,
2106 };
2107 
2108 static const struct file_operations ftrace_set_event_pid_fops = {
2109 	.open = ftrace_event_set_pid_open,
2110 	.read = seq_read,
2111 	.write = ftrace_event_pid_write,
2112 	.llseek = seq_lseek,
2113 	.release = ftrace_event_release,
2114 };
2115 
2116 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2117 	.open = ftrace_event_set_npid_open,
2118 	.read = seq_read,
2119 	.write = ftrace_event_npid_write,
2120 	.llseek = seq_lseek,
2121 	.release = ftrace_event_release,
2122 };
2123 
2124 static const struct file_operations ftrace_enable_fops = {
2125 	.open = tracing_open_file_tr,
2126 	.read = event_enable_read,
2127 	.write = event_enable_write,
2128 	.release = tracing_release_file_tr,
2129 	.llseek = default_llseek,
2130 };
2131 
2132 static const struct file_operations ftrace_event_format_fops = {
2133 	.open = trace_format_open,
2134 	.read = seq_read,
2135 	.llseek = seq_lseek,
2136 	.release = seq_release,
2137 };
2138 
2139 #ifdef CONFIG_PERF_EVENTS
2140 static const struct file_operations ftrace_event_id_fops = {
2141 	.read = event_id_read,
2142 	.llseek = default_llseek,
2143 };
2144 #endif
2145 
2146 static const struct file_operations ftrace_event_filter_fops = {
2147 	.open = tracing_open_file_tr,
2148 	.read = event_filter_read,
2149 	.write = event_filter_write,
2150 	.release = tracing_release_file_tr,
2151 	.llseek = default_llseek,
2152 };
2153 
2154 static const struct file_operations ftrace_subsystem_filter_fops = {
2155 	.open = subsystem_open,
2156 	.read = subsystem_filter_read,
2157 	.write = subsystem_filter_write,
2158 	.llseek = default_llseek,
2159 	.release = subsystem_release,
2160 };
2161 
2162 static const struct file_operations ftrace_system_enable_fops = {
2163 	.open = subsystem_open,
2164 	.read = system_enable_read,
2165 	.write = system_enable_write,
2166 	.llseek = default_llseek,
2167 	.release = subsystem_release,
2168 };
2169 
2170 static const struct file_operations ftrace_tr_enable_fops = {
2171 	.open = system_tr_open,
2172 	.read = system_enable_read,
2173 	.write = system_enable_write,
2174 	.llseek = default_llseek,
2175 	.release = subsystem_release,
2176 };
2177 
2178 static const struct file_operations ftrace_show_header_fops = {
2179 	.open = tracing_open_generic,
2180 	.read = show_header,
2181 	.llseek = default_llseek,
2182 };
2183 
2184 static int
2185 ftrace_event_open(struct inode *inode, struct file *file,
2186 		  const struct seq_operations *seq_ops)
2187 {
2188 	struct seq_file *m;
2189 	int ret;
2190 
2191 	ret = security_locked_down(LOCKDOWN_TRACEFS);
2192 	if (ret)
2193 		return ret;
2194 
2195 	ret = seq_open(file, seq_ops);
2196 	if (ret < 0)
2197 		return ret;
2198 	m = file->private_data;
2199 	/* copy tr over to seq ops */
2200 	m->private = inode->i_private;
2201 
2202 	return ret;
2203 }
2204 
2205 static int ftrace_event_release(struct inode *inode, struct file *file)
2206 {
2207 	struct trace_array *tr = inode->i_private;
2208 
2209 	trace_array_put(tr);
2210 
2211 	return seq_release(inode, file);
2212 }
2213 
2214 static int
2215 ftrace_event_avail_open(struct inode *inode, struct file *file)
2216 {
2217 	const struct seq_operations *seq_ops = &show_event_seq_ops;
2218 
2219 	/* Checks for tracefs lockdown */
2220 	return ftrace_event_open(inode, file, seq_ops);
2221 }
2222 
2223 static int
2224 ftrace_event_set_open(struct inode *inode, struct file *file)
2225 {
2226 	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2227 	struct trace_array *tr = inode->i_private;
2228 	int ret;
2229 
2230 	ret = tracing_check_open_get_tr(tr);
2231 	if (ret)
2232 		return ret;
2233 
2234 	if ((file->f_mode & FMODE_WRITE) &&
2235 	    (file->f_flags & O_TRUNC))
2236 		ftrace_clear_events(tr);
2237 
2238 	ret = ftrace_event_open(inode, file, seq_ops);
2239 	if (ret < 0)
2240 		trace_array_put(tr);
2241 	return ret;
2242 }
2243 
2244 static int
2245 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2246 {
2247 	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2248 	struct trace_array *tr = inode->i_private;
2249 	int ret;
2250 
2251 	ret = tracing_check_open_get_tr(tr);
2252 	if (ret)
2253 		return ret;
2254 
2255 	if ((file->f_mode & FMODE_WRITE) &&
2256 	    (file->f_flags & O_TRUNC))
2257 		ftrace_clear_event_pids(tr, TRACE_PIDS);
2258 
2259 	ret = ftrace_event_open(inode, file, seq_ops);
2260 	if (ret < 0)
2261 		trace_array_put(tr);
2262 	return ret;
2263 }
2264 
2265 static int
2266 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2267 {
2268 	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2269 	struct trace_array *tr = inode->i_private;
2270 	int ret;
2271 
2272 	ret = tracing_check_open_get_tr(tr);
2273 	if (ret)
2274 		return ret;
2275 
2276 	if ((file->f_mode & FMODE_WRITE) &&
2277 	    (file->f_flags & O_TRUNC))
2278 		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2279 
2280 	ret = ftrace_event_open(inode, file, seq_ops);
2281 	if (ret < 0)
2282 		trace_array_put(tr);
2283 	return ret;
2284 }
2285 
2286 static struct event_subsystem *
2287 create_new_subsystem(const char *name)
2288 {
2289 	struct event_subsystem *system;
2290 
2291 	/* need to create new entry */
2292 	system = kmalloc(sizeof(*system), GFP_KERNEL);
2293 	if (!system)
2294 		return NULL;
2295 
2296 	system->ref_count = 1;
2297 
2298 	/* Only allocate if dynamic (kprobes and modules) */
2299 	system->name = kstrdup_const(name, GFP_KERNEL);
2300 	if (!system->name)
2301 		goto out_free;
2302 
2303 	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2304 	if (!system->filter)
2305 		goto out_free;
2306 
2307 	list_add(&system->list, &event_subsystems);
2308 
2309 	return system;
2310 
2311  out_free:
2312 	kfree_const(system->name);
2313 	kfree(system);
2314 	return NULL;
2315 }
2316 
2317 static int system_callback(const char *name, umode_t *mode, void **data,
2318 		    const struct file_operations **fops)
2319 {
2320 	if (strcmp(name, "filter") == 0)
2321 		*fops = &ftrace_subsystem_filter_fops;
2322 
2323 	else if (strcmp(name, "enable") == 0)
2324 		*fops = &ftrace_system_enable_fops;
2325 
2326 	else
2327 		return 0;
2328 
2329 	*mode = TRACE_MODE_WRITE;
2330 	return 1;
2331 }
2332 
2333 static struct eventfs_inode *
2334 event_subsystem_dir(struct trace_array *tr, const char *name,
2335 		    struct trace_event_file *file, struct eventfs_inode *parent)
2336 {
2337 	struct event_subsystem *system, *iter;
2338 	struct trace_subsystem_dir *dir;
2339 	struct eventfs_inode *ei;
2340 	int nr_entries;
2341 	static struct eventfs_entry system_entries[] = {
2342 		{
2343 			.name		= "filter",
2344 			.callback	= system_callback,
2345 		},
2346 		{
2347 			.name		= "enable",
2348 			.callback	= system_callback,
2349 		}
2350 	};
2351 
2352 	/* First see if we did not already create this dir */
2353 	list_for_each_entry(dir, &tr->systems, list) {
2354 		system = dir->subsystem;
2355 		if (strcmp(system->name, name) == 0) {
2356 			dir->nr_events++;
2357 			file->system = dir;
2358 			return dir->ei;
2359 		}
2360 	}
2361 
2362 	/* Now see if the system itself exists. */
2363 	system = NULL;
2364 	list_for_each_entry(iter, &event_subsystems, list) {
2365 		if (strcmp(iter->name, name) == 0) {
2366 			system = iter;
2367 			break;
2368 		}
2369 	}
2370 
2371 	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2372 	if (!dir)
2373 		goto out_fail;
2374 
2375 	if (!system) {
2376 		system = create_new_subsystem(name);
2377 		if (!system)
2378 			goto out_free;
2379 	} else
2380 		__get_system(system);
2381 
2382 	/* ftrace only has directories no files */
2383 	if (strcmp(name, "ftrace") == 0)
2384 		nr_entries = 0;
2385 	else
2386 		nr_entries = ARRAY_SIZE(system_entries);
2387 
2388 	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2389 	if (IS_ERR(ei)) {
2390 		pr_warn("Failed to create system directory %s\n", name);
2391 		__put_system(system);
2392 		goto out_free;
2393 	}
2394 
2395 	dir->ei = ei;
2396 	dir->tr = tr;
2397 	dir->ref_count = 1;
2398 	dir->nr_events = 1;
2399 	dir->subsystem = system;
2400 	file->system = dir;
2401 
2402 	list_add(&dir->list, &tr->systems);
2403 
2404 	return dir->ei;
2405 
2406  out_free:
2407 	kfree(dir);
2408  out_fail:
2409 	/* Only print this message if failed on memory allocation */
2410 	if (!dir || !system)
2411 		pr_warn("No memory to create event subsystem %s\n", name);
2412 	return NULL;
2413 }
2414 
2415 static int
2416 event_define_fields(struct trace_event_call *call)
2417 {
2418 	struct list_head *head;
2419 	int ret = 0;
2420 
2421 	/*
2422 	 * Other events may have the same class. Only update
2423 	 * the fields if they are not already defined.
2424 	 */
2425 	head = trace_get_fields(call);
2426 	if (list_empty(head)) {
2427 		struct trace_event_fields *field = call->class->fields_array;
2428 		unsigned int offset = sizeof(struct trace_entry);
2429 
2430 		for (; field->type; field++) {
2431 			if (field->type == TRACE_FUNCTION_TYPE) {
2432 				field->define_fields(call);
2433 				break;
2434 			}
2435 
2436 			offset = ALIGN(offset, field->align);
2437 			ret = trace_define_field_ext(call, field->type, field->name,
2438 						 offset, field->size,
2439 						 field->is_signed, field->filter_type,
2440 						 field->len);
2441 			if (WARN_ON_ONCE(ret)) {
2442 				pr_err("error code is %d\n", ret);
2443 				break;
2444 			}
2445 
2446 			offset += field->size;
2447 		}
2448 	}
2449 
2450 	return ret;
2451 }
2452 
2453 static int event_callback(const char *name, umode_t *mode, void **data,
2454 			  const struct file_operations **fops)
2455 {
2456 	struct trace_event_file *file = *data;
2457 	struct trace_event_call *call = file->event_call;
2458 
2459 	if (strcmp(name, "format") == 0) {
2460 		*mode = TRACE_MODE_READ;
2461 		*fops = &ftrace_event_format_fops;
2462 		return 1;
2463 	}
2464 
2465 	/*
2466 	 * Only event directories that can be enabled should have
2467 	 * triggers or filters, with the exception of the "print"
2468 	 * event that can have a "trigger" file.
2469 	 */
2470 	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2471 		if (call->class->reg && strcmp(name, "enable") == 0) {
2472 			*mode = TRACE_MODE_WRITE;
2473 			*fops = &ftrace_enable_fops;
2474 			return 1;
2475 		}
2476 
2477 		if (strcmp(name, "filter") == 0) {
2478 			*mode = TRACE_MODE_WRITE;
2479 			*fops = &ftrace_event_filter_fops;
2480 			return 1;
2481 		}
2482 	}
2483 
2484 	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2485 	    strcmp(trace_event_name(call), "print") == 0) {
2486 		if (strcmp(name, "trigger") == 0) {
2487 			*mode = TRACE_MODE_WRITE;
2488 			*fops = &event_trigger_fops;
2489 			return 1;
2490 		}
2491 	}
2492 
2493 #ifdef CONFIG_PERF_EVENTS
2494 	if (call->event.type && call->class->reg &&
2495 	    strcmp(name, "id") == 0) {
2496 		*mode = TRACE_MODE_READ;
2497 		*data = (void *)(long)call->event.type;
2498 		*fops = &ftrace_event_id_fops;
2499 		return 1;
2500 	}
2501 #endif
2502 
2503 #ifdef CONFIG_HIST_TRIGGERS
2504 	if (strcmp(name, "hist") == 0) {
2505 		*mode = TRACE_MODE_READ;
2506 		*fops = &event_hist_fops;
2507 		return 1;
2508 	}
2509 #endif
2510 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2511 	if (strcmp(name, "hist_debug") == 0) {
2512 		*mode = TRACE_MODE_READ;
2513 		*fops = &event_hist_debug_fops;
2514 		return 1;
2515 	}
2516 #endif
2517 #ifdef CONFIG_TRACE_EVENT_INJECT
2518 	if (call->event.type && call->class->reg &&
2519 	    strcmp(name, "inject") == 0) {
2520 		*mode = 0200;
2521 		*fops = &event_inject_fops;
2522 		return 1;
2523 	}
2524 #endif
2525 	return 0;
2526 }
2527 
2528 /* The file is incremented on creation and freeing the enable file decrements it */
2529 static void event_release(const char *name, void *data)
2530 {
2531 	struct trace_event_file *file = data;
2532 
2533 	event_file_put(file);
2534 }
2535 
2536 static int
2537 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2538 {
2539 	struct trace_event_call *call = file->event_call;
2540 	struct trace_array *tr = file->tr;
2541 	struct eventfs_inode *e_events;
2542 	struct eventfs_inode *ei;
2543 	const char *name;
2544 	int nr_entries;
2545 	int ret;
2546 	static struct eventfs_entry event_entries[] = {
2547 		{
2548 			.name		= "enable",
2549 			.callback	= event_callback,
2550 			.release	= event_release,
2551 		},
2552 		{
2553 			.name		= "filter",
2554 			.callback	= event_callback,
2555 		},
2556 		{
2557 			.name		= "trigger",
2558 			.callback	= event_callback,
2559 		},
2560 		{
2561 			.name		= "format",
2562 			.callback	= event_callback,
2563 		},
2564 #ifdef CONFIG_PERF_EVENTS
2565 		{
2566 			.name		= "id",
2567 			.callback	= event_callback,
2568 		},
2569 #endif
2570 #ifdef CONFIG_HIST_TRIGGERS
2571 		{
2572 			.name		= "hist",
2573 			.callback	= event_callback,
2574 		},
2575 #endif
2576 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2577 		{
2578 			.name		= "hist_debug",
2579 			.callback	= event_callback,
2580 		},
2581 #endif
2582 #ifdef CONFIG_TRACE_EVENT_INJECT
2583 		{
2584 			.name		= "inject",
2585 			.callback	= event_callback,
2586 		},
2587 #endif
2588 	};
2589 
2590 	/*
2591 	 * If the trace point header did not define TRACE_SYSTEM
2592 	 * then the system would be called "TRACE_SYSTEM". This should
2593 	 * never happen.
2594 	 */
2595 	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2596 		return -ENODEV;
2597 
2598 	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2599 	if (!e_events)
2600 		return -ENOMEM;
2601 
2602 	nr_entries = ARRAY_SIZE(event_entries);
2603 
2604 	name = trace_event_name(call);
2605 	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2606 	if (IS_ERR(ei)) {
2607 		pr_warn("Could not create tracefs '%s' directory\n", name);
2608 		return -1;
2609 	}
2610 
2611 	file->ei = ei;
2612 
2613 	ret = event_define_fields(call);
2614 	if (ret < 0) {
2615 		pr_warn("Could not initialize trace point events/%s\n", name);
2616 		return ret;
2617 	}
2618 
2619 	/* Gets decremented on freeing of the "enable" file */
2620 	event_file_get(file);
2621 
2622 	return 0;
2623 }
2624 
2625 static void remove_event_from_tracers(struct trace_event_call *call)
2626 {
2627 	struct trace_event_file *file;
2628 	struct trace_array *tr;
2629 
2630 	do_for_each_event_file_safe(tr, file) {
2631 		if (file->event_call != call)
2632 			continue;
2633 
2634 		remove_event_file_dir(file);
2635 		/*
2636 		 * The do_for_each_event_file_safe() is
2637 		 * a double loop. After finding the call for this
2638 		 * trace_array, we use break to jump to the next
2639 		 * trace_array.
2640 		 */
2641 		break;
2642 	} while_for_each_event_file();
2643 }
2644 
2645 static void event_remove(struct trace_event_call *call)
2646 {
2647 	struct trace_array *tr;
2648 	struct trace_event_file *file;
2649 
2650 	do_for_each_event_file(tr, file) {
2651 		if (file->event_call != call)
2652 			continue;
2653 
2654 		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2655 			tr->clear_trace = true;
2656 
2657 		ftrace_event_enable_disable(file, 0);
2658 		/*
2659 		 * The do_for_each_event_file() is
2660 		 * a double loop. After finding the call for this
2661 		 * trace_array, we use break to jump to the next
2662 		 * trace_array.
2663 		 */
2664 		break;
2665 	} while_for_each_event_file();
2666 
2667 	if (call->event.funcs)
2668 		__unregister_trace_event(&call->event);
2669 	remove_event_from_tracers(call);
2670 	list_del(&call->list);
2671 }
2672 
2673 static int event_init(struct trace_event_call *call)
2674 {
2675 	int ret = 0;
2676 	const char *name;
2677 
2678 	name = trace_event_name(call);
2679 	if (WARN_ON(!name))
2680 		return -EINVAL;
2681 
2682 	if (call->class->raw_init) {
2683 		ret = call->class->raw_init(call);
2684 		if (ret < 0 && ret != -ENOSYS)
2685 			pr_warn("Could not initialize trace events/%s\n", name);
2686 	}
2687 
2688 	return ret;
2689 }
2690 
2691 static int
2692 __register_event(struct trace_event_call *call, struct module *mod)
2693 {
2694 	int ret;
2695 
2696 	ret = event_init(call);
2697 	if (ret < 0)
2698 		return ret;
2699 
2700 	list_add(&call->list, &ftrace_events);
2701 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2702 		atomic_set(&call->refcnt, 0);
2703 	else
2704 		call->module = mod;
2705 
2706 	return 0;
2707 }
2708 
2709 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2710 {
2711 	int rlen;
2712 	int elen;
2713 
2714 	/* Find the length of the eval value as a string */
2715 	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2716 	/* Make sure there's enough room to replace the string with the value */
2717 	if (len < elen)
2718 		return NULL;
2719 
2720 	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2721 
2722 	/* Get the rest of the string of ptr */
2723 	rlen = strlen(ptr + len);
2724 	memmove(ptr + elen, ptr + len, rlen);
2725 	/* Make sure we end the new string */
2726 	ptr[elen + rlen] = 0;
2727 
2728 	return ptr + elen;
2729 }
2730 
2731 static void update_event_printk(struct trace_event_call *call,
2732 				struct trace_eval_map *map)
2733 {
2734 	char *ptr;
2735 	int quote = 0;
2736 	int len = strlen(map->eval_string);
2737 
2738 	for (ptr = call->print_fmt; *ptr; ptr++) {
2739 		if (*ptr == '\\') {
2740 			ptr++;
2741 			/* paranoid */
2742 			if (!*ptr)
2743 				break;
2744 			continue;
2745 		}
2746 		if (*ptr == '"') {
2747 			quote ^= 1;
2748 			continue;
2749 		}
2750 		if (quote)
2751 			continue;
2752 		if (isdigit(*ptr)) {
2753 			/* skip numbers */
2754 			do {
2755 				ptr++;
2756 				/* Check for alpha chars like ULL */
2757 			} while (isalnum(*ptr));
2758 			if (!*ptr)
2759 				break;
2760 			/*
2761 			 * A number must have some kind of delimiter after
2762 			 * it, and we can ignore that too.
2763 			 */
2764 			continue;
2765 		}
2766 		if (isalpha(*ptr) || *ptr == '_') {
2767 			if (strncmp(map->eval_string, ptr, len) == 0 &&
2768 			    !isalnum(ptr[len]) && ptr[len] != '_') {
2769 				ptr = eval_replace(ptr, map, len);
2770 				/* enum/sizeof string smaller than value */
2771 				if (WARN_ON_ONCE(!ptr))
2772 					return;
2773 				/*
2774 				 * No need to decrement here, as eval_replace()
2775 				 * returns the pointer to the character passed
2776 				 * the eval, and two evals can not be placed
2777 				 * back to back without something in between.
2778 				 * We can skip that something in between.
2779 				 */
2780 				continue;
2781 			}
2782 		skip_more:
2783 			do {
2784 				ptr++;
2785 			} while (isalnum(*ptr) || *ptr == '_');
2786 			if (!*ptr)
2787 				break;
2788 			/*
2789 			 * If what comes after this variable is a '.' or
2790 			 * '->' then we can continue to ignore that string.
2791 			 */
2792 			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2793 				ptr += *ptr == '.' ? 1 : 2;
2794 				if (!*ptr)
2795 					break;
2796 				goto skip_more;
2797 			}
2798 			/*
2799 			 * Once again, we can skip the delimiter that came
2800 			 * after the string.
2801 			 */
2802 			continue;
2803 		}
2804 	}
2805 }
2806 
2807 static void add_str_to_module(struct module *module, char *str)
2808 {
2809 	struct module_string *modstr;
2810 
2811 	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2812 
2813 	/*
2814 	 * If we failed to allocate memory here, then we'll just
2815 	 * let the str memory leak when the module is removed.
2816 	 * If this fails to allocate, there's worse problems than
2817 	 * a leaked string on module removal.
2818 	 */
2819 	if (WARN_ON_ONCE(!modstr))
2820 		return;
2821 
2822 	modstr->module = module;
2823 	modstr->str = str;
2824 
2825 	list_add(&modstr->next, &module_strings);
2826 }
2827 
2828 static void update_event_fields(struct trace_event_call *call,
2829 				struct trace_eval_map *map)
2830 {
2831 	struct ftrace_event_field *field;
2832 	struct list_head *head;
2833 	char *ptr;
2834 	char *str;
2835 	int len = strlen(map->eval_string);
2836 
2837 	/* Dynamic events should never have field maps */
2838 	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2839 		return;
2840 
2841 	head = trace_get_fields(call);
2842 	list_for_each_entry(field, head, link) {
2843 		ptr = strchr(field->type, '[');
2844 		if (!ptr)
2845 			continue;
2846 		ptr++;
2847 
2848 		if (!isalpha(*ptr) && *ptr != '_')
2849 			continue;
2850 
2851 		if (strncmp(map->eval_string, ptr, len) != 0)
2852 			continue;
2853 
2854 		str = kstrdup(field->type, GFP_KERNEL);
2855 		if (WARN_ON_ONCE(!str))
2856 			return;
2857 		ptr = str + (ptr - field->type);
2858 		ptr = eval_replace(ptr, map, len);
2859 		/* enum/sizeof string smaller than value */
2860 		if (WARN_ON_ONCE(!ptr)) {
2861 			kfree(str);
2862 			continue;
2863 		}
2864 
2865 		/*
2866 		 * If the event is part of a module, then we need to free the string
2867 		 * when the module is removed. Otherwise, it will stay allocated
2868 		 * until a reboot.
2869 		 */
2870 		if (call->module)
2871 			add_str_to_module(call->module, str);
2872 
2873 		field->type = str;
2874 	}
2875 }
2876 
2877 void trace_event_eval_update(struct trace_eval_map **map, int len)
2878 {
2879 	struct trace_event_call *call, *p;
2880 	const char *last_system = NULL;
2881 	bool first = false;
2882 	int last_i;
2883 	int i;
2884 
2885 	down_write(&trace_event_sem);
2886 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2887 		/* events are usually grouped together with systems */
2888 		if (!last_system || call->class->system != last_system) {
2889 			first = true;
2890 			last_i = 0;
2891 			last_system = call->class->system;
2892 		}
2893 
2894 		/*
2895 		 * Since calls are grouped by systems, the likelihood that the
2896 		 * next call in the iteration belongs to the same system as the
2897 		 * previous call is high. As an optimization, we skip searching
2898 		 * for a map[] that matches the call's system if the last call
2899 		 * was from the same system. That's what last_i is for. If the
2900 		 * call has the same system as the previous call, then last_i
2901 		 * will be the index of the first map[] that has a matching
2902 		 * system.
2903 		 */
2904 		for (i = last_i; i < len; i++) {
2905 			if (call->class->system == map[i]->system) {
2906 				/* Save the first system if need be */
2907 				if (first) {
2908 					last_i = i;
2909 					first = false;
2910 				}
2911 				update_event_printk(call, map[i]);
2912 				update_event_fields(call, map[i]);
2913 			}
2914 		}
2915 		cond_resched();
2916 	}
2917 	up_write(&trace_event_sem);
2918 }
2919 
2920 static struct trace_event_file *
2921 trace_create_new_event(struct trace_event_call *call,
2922 		       struct trace_array *tr)
2923 {
2924 	struct trace_pid_list *no_pid_list;
2925 	struct trace_pid_list *pid_list;
2926 	struct trace_event_file *file;
2927 	unsigned int first;
2928 
2929 	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2930 	if (!file)
2931 		return NULL;
2932 
2933 	pid_list = rcu_dereference_protected(tr->filtered_pids,
2934 					     lockdep_is_held(&event_mutex));
2935 	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2936 					     lockdep_is_held(&event_mutex));
2937 
2938 	if (!trace_pid_list_first(pid_list, &first) ||
2939 	    !trace_pid_list_first(no_pid_list, &first))
2940 		file->flags |= EVENT_FILE_FL_PID_FILTER;
2941 
2942 	file->event_call = call;
2943 	file->tr = tr;
2944 	atomic_set(&file->sm_ref, 0);
2945 	atomic_set(&file->tm_ref, 0);
2946 	INIT_LIST_HEAD(&file->triggers);
2947 	list_add(&file->list, &tr->events);
2948 	event_file_get(file);
2949 
2950 	return file;
2951 }
2952 
2953 #define MAX_BOOT_TRIGGERS 32
2954 
2955 static struct boot_triggers {
2956 	const char		*event;
2957 	char			*trigger;
2958 } bootup_triggers[MAX_BOOT_TRIGGERS];
2959 
2960 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2961 static int nr_boot_triggers;
2962 
2963 static __init int setup_trace_triggers(char *str)
2964 {
2965 	char *trigger;
2966 	char *buf;
2967 	int i;
2968 
2969 	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
2970 	ring_buffer_expanded = true;
2971 	disable_tracing_selftest("running event triggers");
2972 
2973 	buf = bootup_trigger_buf;
2974 	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
2975 		trigger = strsep(&buf, ",");
2976 		if (!trigger)
2977 			break;
2978 		bootup_triggers[i].event = strsep(&trigger, ".");
2979 		bootup_triggers[i].trigger = trigger;
2980 		if (!bootup_triggers[i].trigger)
2981 			break;
2982 	}
2983 
2984 	nr_boot_triggers = i;
2985 	return 1;
2986 }
2987 __setup("trace_trigger=", setup_trace_triggers);
2988 
2989 /* Add an event to a trace directory */
2990 static int
2991 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2992 {
2993 	struct trace_event_file *file;
2994 
2995 	file = trace_create_new_event(call, tr);
2996 	if (!file)
2997 		return -ENOMEM;
2998 
2999 	if (eventdir_initialized)
3000 		return event_create_dir(tr->event_dir, file);
3001 	else
3002 		return event_define_fields(call);
3003 }
3004 
3005 static void trace_early_triggers(struct trace_event_file *file, const char *name)
3006 {
3007 	int ret;
3008 	int i;
3009 
3010 	for (i = 0; i < nr_boot_triggers; i++) {
3011 		if (strcmp(name, bootup_triggers[i].event))
3012 			continue;
3013 		mutex_lock(&event_mutex);
3014 		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3015 		mutex_unlock(&event_mutex);
3016 		if (ret)
3017 			pr_err("Failed to register trigger '%s' on event %s\n",
3018 			       bootup_triggers[i].trigger,
3019 			       bootup_triggers[i].event);
3020 	}
3021 }
3022 
3023 /*
3024  * Just create a descriptor for early init. A descriptor is required
3025  * for enabling events at boot. We want to enable events before
3026  * the filesystem is initialized.
3027  */
3028 static int
3029 __trace_early_add_new_event(struct trace_event_call *call,
3030 			    struct trace_array *tr)
3031 {
3032 	struct trace_event_file *file;
3033 	int ret;
3034 
3035 	file = trace_create_new_event(call, tr);
3036 	if (!file)
3037 		return -ENOMEM;
3038 
3039 	ret = event_define_fields(call);
3040 	if (ret)
3041 		return ret;
3042 
3043 	trace_early_triggers(file, trace_event_name(call));
3044 
3045 	return 0;
3046 }
3047 
3048 struct ftrace_module_file_ops;
3049 static void __add_event_to_tracers(struct trace_event_call *call);
3050 
3051 /* Add an additional event_call dynamically */
3052 int trace_add_event_call(struct trace_event_call *call)
3053 {
3054 	int ret;
3055 	lockdep_assert_held(&event_mutex);
3056 
3057 	mutex_lock(&trace_types_lock);
3058 
3059 	ret = __register_event(call, NULL);
3060 	if (ret >= 0)
3061 		__add_event_to_tracers(call);
3062 
3063 	mutex_unlock(&trace_types_lock);
3064 	return ret;
3065 }
3066 EXPORT_SYMBOL_GPL(trace_add_event_call);
3067 
3068 /*
3069  * Must be called under locking of trace_types_lock, event_mutex and
3070  * trace_event_sem.
3071  */
3072 static void __trace_remove_event_call(struct trace_event_call *call)
3073 {
3074 	event_remove(call);
3075 	trace_destroy_fields(call);
3076 	free_event_filter(call->filter);
3077 	call->filter = NULL;
3078 }
3079 
3080 static int probe_remove_event_call(struct trace_event_call *call)
3081 {
3082 	struct trace_array *tr;
3083 	struct trace_event_file *file;
3084 
3085 #ifdef CONFIG_PERF_EVENTS
3086 	if (call->perf_refcount)
3087 		return -EBUSY;
3088 #endif
3089 	do_for_each_event_file(tr, file) {
3090 		if (file->event_call != call)
3091 			continue;
3092 		/*
3093 		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3094 		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3095 		 * TRACE_REG_UNREGISTER.
3096 		 */
3097 		if (file->flags & EVENT_FILE_FL_ENABLED)
3098 			goto busy;
3099 
3100 		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3101 			tr->clear_trace = true;
3102 		/*
3103 		 * The do_for_each_event_file_safe() is
3104 		 * a double loop. After finding the call for this
3105 		 * trace_array, we use break to jump to the next
3106 		 * trace_array.
3107 		 */
3108 		break;
3109 	} while_for_each_event_file();
3110 
3111 	__trace_remove_event_call(call);
3112 
3113 	return 0;
3114  busy:
3115 	/* No need to clear the trace now */
3116 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3117 		tr->clear_trace = false;
3118 	}
3119 	return -EBUSY;
3120 }
3121 
3122 /* Remove an event_call */
3123 int trace_remove_event_call(struct trace_event_call *call)
3124 {
3125 	int ret;
3126 
3127 	lockdep_assert_held(&event_mutex);
3128 
3129 	mutex_lock(&trace_types_lock);
3130 	down_write(&trace_event_sem);
3131 	ret = probe_remove_event_call(call);
3132 	up_write(&trace_event_sem);
3133 	mutex_unlock(&trace_types_lock);
3134 
3135 	return ret;
3136 }
3137 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3138 
3139 #define for_each_event(event, start, end)			\
3140 	for (event = start;					\
3141 	     (unsigned long)event < (unsigned long)end;		\
3142 	     event++)
3143 
3144 #ifdef CONFIG_MODULES
3145 
3146 static void trace_module_add_events(struct module *mod)
3147 {
3148 	struct trace_event_call **call, **start, **end;
3149 
3150 	if (!mod->num_trace_events)
3151 		return;
3152 
3153 	/* Don't add infrastructure for mods without tracepoints */
3154 	if (trace_module_has_bad_taint(mod)) {
3155 		pr_err("%s: module has bad taint, not creating trace events\n",
3156 		       mod->name);
3157 		return;
3158 	}
3159 
3160 	start = mod->trace_events;
3161 	end = mod->trace_events + mod->num_trace_events;
3162 
3163 	for_each_event(call, start, end) {
3164 		__register_event(*call, mod);
3165 		__add_event_to_tracers(*call);
3166 	}
3167 }
3168 
3169 static void trace_module_remove_events(struct module *mod)
3170 {
3171 	struct trace_event_call *call, *p;
3172 	struct module_string *modstr, *m;
3173 
3174 	down_write(&trace_event_sem);
3175 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3176 		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3177 			continue;
3178 		if (call->module == mod)
3179 			__trace_remove_event_call(call);
3180 	}
3181 	/* Check for any strings allocade for this module */
3182 	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3183 		if (modstr->module != mod)
3184 			continue;
3185 		list_del(&modstr->next);
3186 		kfree(modstr->str);
3187 		kfree(modstr);
3188 	}
3189 	up_write(&trace_event_sem);
3190 
3191 	/*
3192 	 * It is safest to reset the ring buffer if the module being unloaded
3193 	 * registered any events that were used. The only worry is if
3194 	 * a new module gets loaded, and takes on the same id as the events
3195 	 * of this module. When printing out the buffer, traced events left
3196 	 * over from this module may be passed to the new module events and
3197 	 * unexpected results may occur.
3198 	 */
3199 	tracing_reset_all_online_cpus_unlocked();
3200 }
3201 
3202 static int trace_module_notify(struct notifier_block *self,
3203 			       unsigned long val, void *data)
3204 {
3205 	struct module *mod = data;
3206 
3207 	mutex_lock(&event_mutex);
3208 	mutex_lock(&trace_types_lock);
3209 	switch (val) {
3210 	case MODULE_STATE_COMING:
3211 		trace_module_add_events(mod);
3212 		break;
3213 	case MODULE_STATE_GOING:
3214 		trace_module_remove_events(mod);
3215 		break;
3216 	}
3217 	mutex_unlock(&trace_types_lock);
3218 	mutex_unlock(&event_mutex);
3219 
3220 	return NOTIFY_OK;
3221 }
3222 
3223 static struct notifier_block trace_module_nb = {
3224 	.notifier_call = trace_module_notify,
3225 	.priority = 1, /* higher than trace.c module notify */
3226 };
3227 #endif /* CONFIG_MODULES */
3228 
3229 /* Create a new event directory structure for a trace directory. */
3230 static void
3231 __trace_add_event_dirs(struct trace_array *tr)
3232 {
3233 	struct trace_event_call *call;
3234 	int ret;
3235 
3236 	list_for_each_entry(call, &ftrace_events, list) {
3237 		ret = __trace_add_new_event(call, tr);
3238 		if (ret < 0)
3239 			pr_warn("Could not create directory for event %s\n",
3240 				trace_event_name(call));
3241 	}
3242 }
3243 
3244 /* Returns any file that matches the system and event */
3245 struct trace_event_file *
3246 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3247 {
3248 	struct trace_event_file *file;
3249 	struct trace_event_call *call;
3250 	const char *name;
3251 
3252 	list_for_each_entry(file, &tr->events, list) {
3253 
3254 		call = file->event_call;
3255 		name = trace_event_name(call);
3256 
3257 		if (!name || !call->class)
3258 			continue;
3259 
3260 		if (strcmp(event, name) == 0 &&
3261 		    strcmp(system, call->class->system) == 0)
3262 			return file;
3263 	}
3264 	return NULL;
3265 }
3266 
3267 /* Returns valid trace event files that match system and event */
3268 struct trace_event_file *
3269 find_event_file(struct trace_array *tr, const char *system, const char *event)
3270 {
3271 	struct trace_event_file *file;
3272 
3273 	file = __find_event_file(tr, system, event);
3274 	if (!file || !file->event_call->class->reg ||
3275 	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3276 		return NULL;
3277 
3278 	return file;
3279 }
3280 
3281 /**
3282  * trace_get_event_file - Find and return a trace event file
3283  * @instance: The name of the trace instance containing the event
3284  * @system: The name of the system containing the event
3285  * @event: The name of the event
3286  *
3287  * Return a trace event file given the trace instance name, trace
3288  * system, and trace event name.  If the instance name is NULL, it
3289  * refers to the top-level trace array.
3290  *
3291  * This function will look it up and return it if found, after calling
3292  * trace_array_get() to prevent the instance from going away, and
3293  * increment the event's module refcount to prevent it from being
3294  * removed.
3295  *
3296  * To release the file, call trace_put_event_file(), which will call
3297  * trace_array_put() and decrement the event's module refcount.
3298  *
3299  * Return: The trace event on success, ERR_PTR otherwise.
3300  */
3301 struct trace_event_file *trace_get_event_file(const char *instance,
3302 					      const char *system,
3303 					      const char *event)
3304 {
3305 	struct trace_array *tr = top_trace_array();
3306 	struct trace_event_file *file = NULL;
3307 	int ret = -EINVAL;
3308 
3309 	if (instance) {
3310 		tr = trace_array_find_get(instance);
3311 		if (!tr)
3312 			return ERR_PTR(-ENOENT);
3313 	} else {
3314 		ret = trace_array_get(tr);
3315 		if (ret)
3316 			return ERR_PTR(ret);
3317 	}
3318 
3319 	mutex_lock(&event_mutex);
3320 
3321 	file = find_event_file(tr, system, event);
3322 	if (!file) {
3323 		trace_array_put(tr);
3324 		ret = -EINVAL;
3325 		goto out;
3326 	}
3327 
3328 	/* Don't let event modules unload while in use */
3329 	ret = trace_event_try_get_ref(file->event_call);
3330 	if (!ret) {
3331 		trace_array_put(tr);
3332 		ret = -EBUSY;
3333 		goto out;
3334 	}
3335 
3336 	ret = 0;
3337  out:
3338 	mutex_unlock(&event_mutex);
3339 
3340 	if (ret)
3341 		file = ERR_PTR(ret);
3342 
3343 	return file;
3344 }
3345 EXPORT_SYMBOL_GPL(trace_get_event_file);
3346 
3347 /**
3348  * trace_put_event_file - Release a file from trace_get_event_file()
3349  * @file: The trace event file
3350  *
3351  * If a file was retrieved using trace_get_event_file(), this should
3352  * be called when it's no longer needed.  It will cancel the previous
3353  * trace_array_get() called by that function, and decrement the
3354  * event's module refcount.
3355  */
3356 void trace_put_event_file(struct trace_event_file *file)
3357 {
3358 	mutex_lock(&event_mutex);
3359 	trace_event_put_ref(file->event_call);
3360 	mutex_unlock(&event_mutex);
3361 
3362 	trace_array_put(file->tr);
3363 }
3364 EXPORT_SYMBOL_GPL(trace_put_event_file);
3365 
3366 #ifdef CONFIG_DYNAMIC_FTRACE
3367 
3368 /* Avoid typos */
3369 #define ENABLE_EVENT_STR	"enable_event"
3370 #define DISABLE_EVENT_STR	"disable_event"
3371 
3372 struct event_probe_data {
3373 	struct trace_event_file	*file;
3374 	unsigned long			count;
3375 	int				ref;
3376 	bool				enable;
3377 };
3378 
3379 static void update_event_probe(struct event_probe_data *data)
3380 {
3381 	if (data->enable)
3382 		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3383 	else
3384 		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3385 }
3386 
3387 static void
3388 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3389 		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3390 		   void *data)
3391 {
3392 	struct ftrace_func_mapper *mapper = data;
3393 	struct event_probe_data *edata;
3394 	void **pdata;
3395 
3396 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3397 	if (!pdata || !*pdata)
3398 		return;
3399 
3400 	edata = *pdata;
3401 	update_event_probe(edata);
3402 }
3403 
3404 static void
3405 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3406 			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3407 			 void *data)
3408 {
3409 	struct ftrace_func_mapper *mapper = data;
3410 	struct event_probe_data *edata;
3411 	void **pdata;
3412 
3413 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3414 	if (!pdata || !*pdata)
3415 		return;
3416 
3417 	edata = *pdata;
3418 
3419 	if (!edata->count)
3420 		return;
3421 
3422 	/* Skip if the event is in a state we want to switch to */
3423 	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3424 		return;
3425 
3426 	if (edata->count != -1)
3427 		(edata->count)--;
3428 
3429 	update_event_probe(edata);
3430 }
3431 
3432 static int
3433 event_enable_print(struct seq_file *m, unsigned long ip,
3434 		   struct ftrace_probe_ops *ops, void *data)
3435 {
3436 	struct ftrace_func_mapper *mapper = data;
3437 	struct event_probe_data *edata;
3438 	void **pdata;
3439 
3440 	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3441 
3442 	if (WARN_ON_ONCE(!pdata || !*pdata))
3443 		return 0;
3444 
3445 	edata = *pdata;
3446 
3447 	seq_printf(m, "%ps:", (void *)ip);
3448 
3449 	seq_printf(m, "%s:%s:%s",
3450 		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3451 		   edata->file->event_call->class->system,
3452 		   trace_event_name(edata->file->event_call));
3453 
3454 	if (edata->count == -1)
3455 		seq_puts(m, ":unlimited\n");
3456 	else
3457 		seq_printf(m, ":count=%ld\n", edata->count);
3458 
3459 	return 0;
3460 }
3461 
3462 static int
3463 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3464 		  unsigned long ip, void *init_data, void **data)
3465 {
3466 	struct ftrace_func_mapper *mapper = *data;
3467 	struct event_probe_data *edata = init_data;
3468 	int ret;
3469 
3470 	if (!mapper) {
3471 		mapper = allocate_ftrace_func_mapper();
3472 		if (!mapper)
3473 			return -ENODEV;
3474 		*data = mapper;
3475 	}
3476 
3477 	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3478 	if (ret < 0)
3479 		return ret;
3480 
3481 	edata->ref++;
3482 
3483 	return 0;
3484 }
3485 
3486 static int free_probe_data(void *data)
3487 {
3488 	struct event_probe_data *edata = data;
3489 
3490 	edata->ref--;
3491 	if (!edata->ref) {
3492 		/* Remove the SOFT_MODE flag */
3493 		__ftrace_event_enable_disable(edata->file, 0, 1);
3494 		trace_event_put_ref(edata->file->event_call);
3495 		kfree(edata);
3496 	}
3497 	return 0;
3498 }
3499 
3500 static void
3501 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3502 		  unsigned long ip, void *data)
3503 {
3504 	struct ftrace_func_mapper *mapper = data;
3505 	struct event_probe_data *edata;
3506 
3507 	if (!ip) {
3508 		if (!mapper)
3509 			return;
3510 		free_ftrace_func_mapper(mapper, free_probe_data);
3511 		return;
3512 	}
3513 
3514 	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3515 
3516 	if (WARN_ON_ONCE(!edata))
3517 		return;
3518 
3519 	if (WARN_ON_ONCE(edata->ref <= 0))
3520 		return;
3521 
3522 	free_probe_data(edata);
3523 }
3524 
3525 static struct ftrace_probe_ops event_enable_probe_ops = {
3526 	.func			= event_enable_probe,
3527 	.print			= event_enable_print,
3528 	.init			= event_enable_init,
3529 	.free			= event_enable_free,
3530 };
3531 
3532 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3533 	.func			= event_enable_count_probe,
3534 	.print			= event_enable_print,
3535 	.init			= event_enable_init,
3536 	.free			= event_enable_free,
3537 };
3538 
3539 static struct ftrace_probe_ops event_disable_probe_ops = {
3540 	.func			= event_enable_probe,
3541 	.print			= event_enable_print,
3542 	.init			= event_enable_init,
3543 	.free			= event_enable_free,
3544 };
3545 
3546 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3547 	.func			= event_enable_count_probe,
3548 	.print			= event_enable_print,
3549 	.init			= event_enable_init,
3550 	.free			= event_enable_free,
3551 };
3552 
3553 static int
3554 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3555 		  char *glob, char *cmd, char *param, int enabled)
3556 {
3557 	struct trace_event_file *file;
3558 	struct ftrace_probe_ops *ops;
3559 	struct event_probe_data *data;
3560 	const char *system;
3561 	const char *event;
3562 	char *number;
3563 	bool enable;
3564 	int ret;
3565 
3566 	if (!tr)
3567 		return -ENODEV;
3568 
3569 	/* hash funcs only work with set_ftrace_filter */
3570 	if (!enabled || !param)
3571 		return -EINVAL;
3572 
3573 	system = strsep(&param, ":");
3574 	if (!param)
3575 		return -EINVAL;
3576 
3577 	event = strsep(&param, ":");
3578 
3579 	mutex_lock(&event_mutex);
3580 
3581 	ret = -EINVAL;
3582 	file = find_event_file(tr, system, event);
3583 	if (!file)
3584 		goto out;
3585 
3586 	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3587 
3588 	if (enable)
3589 		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3590 	else
3591 		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3592 
3593 	if (glob[0] == '!') {
3594 		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3595 		goto out;
3596 	}
3597 
3598 	ret = -ENOMEM;
3599 
3600 	data = kzalloc(sizeof(*data), GFP_KERNEL);
3601 	if (!data)
3602 		goto out;
3603 
3604 	data->enable = enable;
3605 	data->count = -1;
3606 	data->file = file;
3607 
3608 	if (!param)
3609 		goto out_reg;
3610 
3611 	number = strsep(&param, ":");
3612 
3613 	ret = -EINVAL;
3614 	if (!strlen(number))
3615 		goto out_free;
3616 
3617 	/*
3618 	 * We use the callback data field (which is a pointer)
3619 	 * as our counter.
3620 	 */
3621 	ret = kstrtoul(number, 0, &data->count);
3622 	if (ret)
3623 		goto out_free;
3624 
3625  out_reg:
3626 	/* Don't let event modules unload while probe registered */
3627 	ret = trace_event_try_get_ref(file->event_call);
3628 	if (!ret) {
3629 		ret = -EBUSY;
3630 		goto out_free;
3631 	}
3632 
3633 	ret = __ftrace_event_enable_disable(file, 1, 1);
3634 	if (ret < 0)
3635 		goto out_put;
3636 
3637 	ret = register_ftrace_function_probe(glob, tr, ops, data);
3638 	/*
3639 	 * The above returns on success the # of functions enabled,
3640 	 * but if it didn't find any functions it returns zero.
3641 	 * Consider no functions a failure too.
3642 	 */
3643 	if (!ret) {
3644 		ret = -ENOENT;
3645 		goto out_disable;
3646 	} else if (ret < 0)
3647 		goto out_disable;
3648 	/* Just return zero, not the number of enabled functions */
3649 	ret = 0;
3650  out:
3651 	mutex_unlock(&event_mutex);
3652 	return ret;
3653 
3654  out_disable:
3655 	__ftrace_event_enable_disable(file, 0, 1);
3656  out_put:
3657 	trace_event_put_ref(file->event_call);
3658  out_free:
3659 	kfree(data);
3660 	goto out;
3661 }
3662 
3663 static struct ftrace_func_command event_enable_cmd = {
3664 	.name			= ENABLE_EVENT_STR,
3665 	.func			= event_enable_func,
3666 };
3667 
3668 static struct ftrace_func_command event_disable_cmd = {
3669 	.name			= DISABLE_EVENT_STR,
3670 	.func			= event_enable_func,
3671 };
3672 
3673 static __init int register_event_cmds(void)
3674 {
3675 	int ret;
3676 
3677 	ret = register_ftrace_command(&event_enable_cmd);
3678 	if (WARN_ON(ret < 0))
3679 		return ret;
3680 	ret = register_ftrace_command(&event_disable_cmd);
3681 	if (WARN_ON(ret < 0))
3682 		unregister_ftrace_command(&event_enable_cmd);
3683 	return ret;
3684 }
3685 #else
3686 static inline int register_event_cmds(void) { return 0; }
3687 #endif /* CONFIG_DYNAMIC_FTRACE */
3688 
3689 /*
3690  * The top level array and trace arrays created by boot-time tracing
3691  * have already had its trace_event_file descriptors created in order
3692  * to allow for early events to be recorded.
3693  * This function is called after the tracefs has been initialized,
3694  * and we now have to create the files associated to the events.
3695  */
3696 static void __trace_early_add_event_dirs(struct trace_array *tr)
3697 {
3698 	struct trace_event_file *file;
3699 	int ret;
3700 
3701 
3702 	list_for_each_entry(file, &tr->events, list) {
3703 		ret = event_create_dir(tr->event_dir, file);
3704 		if (ret < 0)
3705 			pr_warn("Could not create directory for event %s\n",
3706 				trace_event_name(file->event_call));
3707 	}
3708 }
3709 
3710 /*
3711  * For early boot up, the top trace array and the trace arrays created
3712  * by boot-time tracing require to have a list of events that can be
3713  * enabled. This must be done before the filesystem is set up in order
3714  * to allow events to be traced early.
3715  */
3716 void __trace_early_add_events(struct trace_array *tr)
3717 {
3718 	struct trace_event_call *call;
3719 	int ret;
3720 
3721 	list_for_each_entry(call, &ftrace_events, list) {
3722 		/* Early boot up should not have any modules loaded */
3723 		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3724 		    WARN_ON_ONCE(call->module))
3725 			continue;
3726 
3727 		ret = __trace_early_add_new_event(call, tr);
3728 		if (ret < 0)
3729 			pr_warn("Could not create early event %s\n",
3730 				trace_event_name(call));
3731 	}
3732 }
3733 
3734 /* Remove the event directory structure for a trace directory. */
3735 static void
3736 __trace_remove_event_dirs(struct trace_array *tr)
3737 {
3738 	struct trace_event_file *file, *next;
3739 
3740 	list_for_each_entry_safe(file, next, &tr->events, list)
3741 		remove_event_file_dir(file);
3742 }
3743 
3744 static void __add_event_to_tracers(struct trace_event_call *call)
3745 {
3746 	struct trace_array *tr;
3747 
3748 	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3749 		__trace_add_new_event(call, tr);
3750 }
3751 
3752 extern struct trace_event_call *__start_ftrace_events[];
3753 extern struct trace_event_call *__stop_ftrace_events[];
3754 
3755 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3756 
3757 static __init int setup_trace_event(char *str)
3758 {
3759 	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3760 	ring_buffer_expanded = true;
3761 	disable_tracing_selftest("running event tracing");
3762 
3763 	return 1;
3764 }
3765 __setup("trace_event=", setup_trace_event);
3766 
3767 static int events_callback(const char *name, umode_t *mode, void **data,
3768 			   const struct file_operations **fops)
3769 {
3770 	if (strcmp(name, "enable") == 0) {
3771 		*mode = TRACE_MODE_WRITE;
3772 		*fops = &ftrace_tr_enable_fops;
3773 		return 1;
3774 	}
3775 
3776 	if (strcmp(name, "header_page") == 0)
3777 		*data = ring_buffer_print_page_header;
3778 
3779 	else if (strcmp(name, "header_event") == 0)
3780 		*data = ring_buffer_print_entry_header;
3781 
3782 	else
3783 		return 0;
3784 
3785 	*mode = TRACE_MODE_READ;
3786 	*fops = &ftrace_show_header_fops;
3787 	return 1;
3788 }
3789 
3790 /* Expects to have event_mutex held when called */
3791 static int
3792 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3793 {
3794 	struct eventfs_inode *e_events;
3795 	struct dentry *entry;
3796 	int nr_entries;
3797 	static struct eventfs_entry events_entries[] = {
3798 		{
3799 			.name		= "enable",
3800 			.callback	= events_callback,
3801 		},
3802 		{
3803 			.name		= "header_page",
3804 			.callback	= events_callback,
3805 		},
3806 		{
3807 			.name		= "header_event",
3808 			.callback	= events_callback,
3809 		},
3810 	};
3811 
3812 	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3813 				  tr, &ftrace_set_event_fops);
3814 	if (!entry)
3815 		return -ENOMEM;
3816 
3817 	nr_entries = ARRAY_SIZE(events_entries);
3818 
3819 	e_events = eventfs_create_events_dir("events", parent, events_entries,
3820 					     nr_entries, tr);
3821 	if (IS_ERR(e_events)) {
3822 		pr_warn("Could not create tracefs 'events' directory\n");
3823 		return -ENOMEM;
3824 	}
3825 
3826 	/* There are not as crucial, just warn if they are not created */
3827 
3828 	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3829 			  tr, &ftrace_set_event_pid_fops);
3830 
3831 	trace_create_file("set_event_notrace_pid",
3832 			  TRACE_MODE_WRITE, parent, tr,
3833 			  &ftrace_set_event_notrace_pid_fops);
3834 
3835 	tr->event_dir = e_events;
3836 
3837 	return 0;
3838 }
3839 
3840 /**
3841  * event_trace_add_tracer - add a instance of a trace_array to events
3842  * @parent: The parent dentry to place the files/directories for events in
3843  * @tr: The trace array associated with these events
3844  *
3845  * When a new instance is created, it needs to set up its events
3846  * directory, as well as other files associated with events. It also
3847  * creates the event hierarchy in the @parent/events directory.
3848  *
3849  * Returns 0 on success.
3850  *
3851  * Must be called with event_mutex held.
3852  */
3853 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3854 {
3855 	int ret;
3856 
3857 	lockdep_assert_held(&event_mutex);
3858 
3859 	ret = create_event_toplevel_files(parent, tr);
3860 	if (ret)
3861 		goto out;
3862 
3863 	down_write(&trace_event_sem);
3864 	/* If tr already has the event list, it is initialized in early boot. */
3865 	if (unlikely(!list_empty(&tr->events)))
3866 		__trace_early_add_event_dirs(tr);
3867 	else
3868 		__trace_add_event_dirs(tr);
3869 	up_write(&trace_event_sem);
3870 
3871  out:
3872 	return ret;
3873 }
3874 
3875 /*
3876  * The top trace array already had its file descriptors created.
3877  * Now the files themselves need to be created.
3878  */
3879 static __init int
3880 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3881 {
3882 	int ret;
3883 
3884 	mutex_lock(&event_mutex);
3885 
3886 	ret = create_event_toplevel_files(parent, tr);
3887 	if (ret)
3888 		goto out_unlock;
3889 
3890 	down_write(&trace_event_sem);
3891 	__trace_early_add_event_dirs(tr);
3892 	up_write(&trace_event_sem);
3893 
3894  out_unlock:
3895 	mutex_unlock(&event_mutex);
3896 
3897 	return ret;
3898 }
3899 
3900 /* Must be called with event_mutex held */
3901 int event_trace_del_tracer(struct trace_array *tr)
3902 {
3903 	lockdep_assert_held(&event_mutex);
3904 
3905 	/* Disable any event triggers and associated soft-disabled events */
3906 	clear_event_triggers(tr);
3907 
3908 	/* Clear the pid list */
3909 	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3910 
3911 	/* Disable any running events */
3912 	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3913 
3914 	/* Make sure no more events are being executed */
3915 	tracepoint_synchronize_unregister();
3916 
3917 	down_write(&trace_event_sem);
3918 	__trace_remove_event_dirs(tr);
3919 	eventfs_remove_events_dir(tr->event_dir);
3920 	up_write(&trace_event_sem);
3921 
3922 	tr->event_dir = NULL;
3923 
3924 	return 0;
3925 }
3926 
3927 static __init int event_trace_memsetup(void)
3928 {
3929 	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3930 	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3931 	return 0;
3932 }
3933 
3934 __init void
3935 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3936 {
3937 	char *token;
3938 	int ret;
3939 
3940 	while (true) {
3941 		token = strsep(&buf, ",");
3942 
3943 		if (!token)
3944 			break;
3945 
3946 		if (*token) {
3947 			/* Restarting syscalls requires that we stop them first */
3948 			if (disable_first)
3949 				ftrace_set_clr_event(tr, token, 0);
3950 
3951 			ret = ftrace_set_clr_event(tr, token, 1);
3952 			if (ret)
3953 				pr_warn("Failed to enable trace event: %s\n", token);
3954 		}
3955 
3956 		/* Put back the comma to allow this to be called again */
3957 		if (buf)
3958 			*(buf - 1) = ',';
3959 	}
3960 }
3961 
3962 static __init int event_trace_enable(void)
3963 {
3964 	struct trace_array *tr = top_trace_array();
3965 	struct trace_event_call **iter, *call;
3966 	int ret;
3967 
3968 	if (!tr)
3969 		return -ENODEV;
3970 
3971 	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3972 
3973 		call = *iter;
3974 		ret = event_init(call);
3975 		if (!ret)
3976 			list_add(&call->list, &ftrace_events);
3977 	}
3978 
3979 	register_trigger_cmds();
3980 
3981 	/*
3982 	 * We need the top trace array to have a working set of trace
3983 	 * points at early init, before the debug files and directories
3984 	 * are created. Create the file entries now, and attach them
3985 	 * to the actual file dentries later.
3986 	 */
3987 	__trace_early_add_events(tr);
3988 
3989 	early_enable_events(tr, bootup_event_buf, false);
3990 
3991 	trace_printk_start_comm();
3992 
3993 	register_event_cmds();
3994 
3995 
3996 	return 0;
3997 }
3998 
3999 /*
4000  * event_trace_enable() is called from trace_event_init() first to
4001  * initialize events and perhaps start any events that are on the
4002  * command line. Unfortunately, there are some events that will not
4003  * start this early, like the system call tracepoints that need
4004  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4005  * event_trace_enable() is called before pid 1 starts, and this flag
4006  * is never set, making the syscall tracepoint never get reached, but
4007  * the event is enabled regardless (and not doing anything).
4008  */
4009 static __init int event_trace_enable_again(void)
4010 {
4011 	struct trace_array *tr;
4012 
4013 	tr = top_trace_array();
4014 	if (!tr)
4015 		return -ENODEV;
4016 
4017 	early_enable_events(tr, bootup_event_buf, true);
4018 
4019 	return 0;
4020 }
4021 
4022 early_initcall(event_trace_enable_again);
4023 
4024 /* Init fields which doesn't related to the tracefs */
4025 static __init int event_trace_init_fields(void)
4026 {
4027 	if (trace_define_generic_fields())
4028 		pr_warn("tracing: Failed to allocated generic fields");
4029 
4030 	if (trace_define_common_fields())
4031 		pr_warn("tracing: Failed to allocate common fields");
4032 
4033 	return 0;
4034 }
4035 
4036 __init int event_trace_init(void)
4037 {
4038 	struct trace_array *tr;
4039 	int ret;
4040 
4041 	tr = top_trace_array();
4042 	if (!tr)
4043 		return -ENODEV;
4044 
4045 	trace_create_file("available_events", TRACE_MODE_READ,
4046 			  NULL, tr, &ftrace_avail_fops);
4047 
4048 	ret = early_event_add_tracer(NULL, tr);
4049 	if (ret)
4050 		return ret;
4051 
4052 #ifdef CONFIG_MODULES
4053 	ret = register_module_notifier(&trace_module_nb);
4054 	if (ret)
4055 		pr_warn("Failed to register trace events module notifier\n");
4056 #endif
4057 
4058 	eventdir_initialized = true;
4059 
4060 	return 0;
4061 }
4062 
4063 void __init trace_event_init(void)
4064 {
4065 	event_trace_memsetup();
4066 	init_ftrace_syscalls();
4067 	event_trace_enable();
4068 	event_trace_init_fields();
4069 }
4070 
4071 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4072 
4073 static DEFINE_SPINLOCK(test_spinlock);
4074 static DEFINE_SPINLOCK(test_spinlock_irq);
4075 static DEFINE_MUTEX(test_mutex);
4076 
4077 static __init void test_work(struct work_struct *dummy)
4078 {
4079 	spin_lock(&test_spinlock);
4080 	spin_lock_irq(&test_spinlock_irq);
4081 	udelay(1);
4082 	spin_unlock_irq(&test_spinlock_irq);
4083 	spin_unlock(&test_spinlock);
4084 
4085 	mutex_lock(&test_mutex);
4086 	msleep(1);
4087 	mutex_unlock(&test_mutex);
4088 }
4089 
4090 static __init int event_test_thread(void *unused)
4091 {
4092 	void *test_malloc;
4093 
4094 	test_malloc = kmalloc(1234, GFP_KERNEL);
4095 	if (!test_malloc)
4096 		pr_info("failed to kmalloc\n");
4097 
4098 	schedule_on_each_cpu(test_work);
4099 
4100 	kfree(test_malloc);
4101 
4102 	set_current_state(TASK_INTERRUPTIBLE);
4103 	while (!kthread_should_stop()) {
4104 		schedule();
4105 		set_current_state(TASK_INTERRUPTIBLE);
4106 	}
4107 	__set_current_state(TASK_RUNNING);
4108 
4109 	return 0;
4110 }
4111 
4112 /*
4113  * Do various things that may trigger events.
4114  */
4115 static __init void event_test_stuff(void)
4116 {
4117 	struct task_struct *test_thread;
4118 
4119 	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4120 	msleep(1);
4121 	kthread_stop(test_thread);
4122 }
4123 
4124 /*
4125  * For every trace event defined, we will test each trace point separately,
4126  * and then by groups, and finally all trace points.
4127  */
4128 static __init void event_trace_self_tests(void)
4129 {
4130 	struct trace_subsystem_dir *dir;
4131 	struct trace_event_file *file;
4132 	struct trace_event_call *call;
4133 	struct event_subsystem *system;
4134 	struct trace_array *tr;
4135 	int ret;
4136 
4137 	tr = top_trace_array();
4138 	if (!tr)
4139 		return;
4140 
4141 	pr_info("Running tests on trace events:\n");
4142 
4143 	list_for_each_entry(file, &tr->events, list) {
4144 
4145 		call = file->event_call;
4146 
4147 		/* Only test those that have a probe */
4148 		if (!call->class || !call->class->probe)
4149 			continue;
4150 
4151 /*
4152  * Testing syscall events here is pretty useless, but
4153  * we still do it if configured. But this is time consuming.
4154  * What we really need is a user thread to perform the
4155  * syscalls as we test.
4156  */
4157 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4158 		if (call->class->system &&
4159 		    strcmp(call->class->system, "syscalls") == 0)
4160 			continue;
4161 #endif
4162 
4163 		pr_info("Testing event %s: ", trace_event_name(call));
4164 
4165 		/*
4166 		 * If an event is already enabled, someone is using
4167 		 * it and the self test should not be on.
4168 		 */
4169 		if (file->flags & EVENT_FILE_FL_ENABLED) {
4170 			pr_warn("Enabled event during self test!\n");
4171 			WARN_ON_ONCE(1);
4172 			continue;
4173 		}
4174 
4175 		ftrace_event_enable_disable(file, 1);
4176 		event_test_stuff();
4177 		ftrace_event_enable_disable(file, 0);
4178 
4179 		pr_cont("OK\n");
4180 	}
4181 
4182 	/* Now test at the sub system level */
4183 
4184 	pr_info("Running tests on trace event systems:\n");
4185 
4186 	list_for_each_entry(dir, &tr->systems, list) {
4187 
4188 		system = dir->subsystem;
4189 
4190 		/* the ftrace system is special, skip it */
4191 		if (strcmp(system->name, "ftrace") == 0)
4192 			continue;
4193 
4194 		pr_info("Testing event system %s: ", system->name);
4195 
4196 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4197 		if (WARN_ON_ONCE(ret)) {
4198 			pr_warn("error enabling system %s\n",
4199 				system->name);
4200 			continue;
4201 		}
4202 
4203 		event_test_stuff();
4204 
4205 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4206 		if (WARN_ON_ONCE(ret)) {
4207 			pr_warn("error disabling system %s\n",
4208 				system->name);
4209 			continue;
4210 		}
4211 
4212 		pr_cont("OK\n");
4213 	}
4214 
4215 	/* Test with all events enabled */
4216 
4217 	pr_info("Running tests on all trace events:\n");
4218 	pr_info("Testing all events: ");
4219 
4220 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4221 	if (WARN_ON_ONCE(ret)) {
4222 		pr_warn("error enabling all events\n");
4223 		return;
4224 	}
4225 
4226 	event_test_stuff();
4227 
4228 	/* reset sysname */
4229 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4230 	if (WARN_ON_ONCE(ret)) {
4231 		pr_warn("error disabling all events\n");
4232 		return;
4233 	}
4234 
4235 	pr_cont("OK\n");
4236 }
4237 
4238 #ifdef CONFIG_FUNCTION_TRACER
4239 
4240 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4241 
4242 static struct trace_event_file event_trace_file __initdata;
4243 
4244 static void __init
4245 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4246 			  struct ftrace_ops *op, struct ftrace_regs *regs)
4247 {
4248 	struct trace_buffer *buffer;
4249 	struct ring_buffer_event *event;
4250 	struct ftrace_entry *entry;
4251 	unsigned int trace_ctx;
4252 	long disabled;
4253 	int cpu;
4254 
4255 	trace_ctx = tracing_gen_ctx();
4256 	preempt_disable_notrace();
4257 	cpu = raw_smp_processor_id();
4258 	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4259 
4260 	if (disabled != 1)
4261 		goto out;
4262 
4263 	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4264 						TRACE_FN, sizeof(*entry),
4265 						trace_ctx);
4266 	if (!event)
4267 		goto out;
4268 	entry	= ring_buffer_event_data(event);
4269 	entry->ip			= ip;
4270 	entry->parent_ip		= parent_ip;
4271 
4272 	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4273 				    entry, trace_ctx);
4274  out:
4275 	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4276 	preempt_enable_notrace();
4277 }
4278 
4279 static struct ftrace_ops trace_ops __initdata  =
4280 {
4281 	.func = function_test_events_call,
4282 };
4283 
4284 static __init void event_trace_self_test_with_function(void)
4285 {
4286 	int ret;
4287 
4288 	event_trace_file.tr = top_trace_array();
4289 	if (WARN_ON(!event_trace_file.tr))
4290 		return;
4291 
4292 	ret = register_ftrace_function(&trace_ops);
4293 	if (WARN_ON(ret < 0)) {
4294 		pr_info("Failed to enable function tracer for event tests\n");
4295 		return;
4296 	}
4297 	pr_info("Running tests again, along with the function tracer\n");
4298 	event_trace_self_tests();
4299 	unregister_ftrace_function(&trace_ops);
4300 }
4301 #else
4302 static __init void event_trace_self_test_with_function(void)
4303 {
4304 }
4305 #endif
4306 
4307 static __init int event_trace_self_tests_init(void)
4308 {
4309 	if (!tracing_selftest_disabled) {
4310 		event_trace_self_tests();
4311 		event_trace_self_test_with_function();
4312 	}
4313 
4314 	return 0;
4315 }
4316 
4317 late_initcall(event_trace_self_tests_init);
4318 
4319 #endif
4320