Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v6.13.7
   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
  36DEFINE_MUTEX(event_mutex);
  37
  38LIST_HEAD(ftrace_events);
  39static LIST_HEAD(ftrace_generic_fields);
  40static LIST_HEAD(ftrace_common_fields);
  41static bool eventdir_initialized;
  42
  43static LIST_HEAD(module_strings);
  44
  45struct 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
  53static struct kmem_cache *field_cachep;
  54static struct kmem_cache *file_cachep;
  55
  56static inline int system_refcount(struct event_subsystem *system)
  57{
  58	return system->ref_count;
  59}
  60
  61static int system_refcount_inc(struct event_subsystem *system)
  62{
  63	return system->ref_count++;
  64}
  65
  66static 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
  84static struct ftrace_event_field *
  85__find_event_field(struct list_head *head, const 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
  97struct ftrace_event_field *
  98trace_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
 115static 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				int need_test)
 119{
 120	struct ftrace_event_field *field;
 121
 122	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 123	if (!field)
 124		return -ENOMEM;
 125
 126	field->name = name;
 127	field->type = type;
 128
 129	if (filter_type == FILTER_OTHER)
 130		field->filter_type = filter_assign_type(type);
 131	else
 132		field->filter_type = filter_type;
 133
 134	field->offset = offset;
 135	field->size = size;
 136	field->is_signed = is_signed;
 137	field->needs_test = need_test;
 138	field->len = len;
 139
 140	list_add(&field->link, head);
 141
 142	return 0;
 143}
 144
 145int trace_define_field(struct trace_event_call *call, const char *type,
 146		       const char *name, int offset, int size, int is_signed,
 147		       int filter_type)
 148{
 149	struct list_head *head;
 150
 151	if (WARN_ON(!call->class))
 152		return 0;
 153
 154	head = trace_get_fields(call);
 155	return __trace_define_field(head, type, name, offset, size,
 156				    is_signed, filter_type, 0, 0);
 157}
 158EXPORT_SYMBOL_GPL(trace_define_field);
 159
 160static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 161		       const char *name, int offset, int size, int is_signed,
 162		       int filter_type, int len, int need_test)
 163{
 164	struct list_head *head;
 165
 166	if (WARN_ON(!call->class))
 167		return 0;
 168
 169	head = trace_get_fields(call);
 170	return __trace_define_field(head, type, name, offset, size,
 171				    is_signed, filter_type, len, need_test);
 172}
 173
 174#define __generic_field(type, item, filter_type)			\
 175	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 176				   #item, 0, 0, is_signed_type(type),	\
 177				   filter_type, 0, 0);			\
 178	if (ret)							\
 179		return ret;
 180
 181#define __common_field(type, item)					\
 182	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 183				   "common_" #item,			\
 184				   offsetof(typeof(ent), item),		\
 185				   sizeof(ent.item),			\
 186				   is_signed_type(type), FILTER_OTHER,	\
 187				   0, 0);				\
 188	if (ret)							\
 189		return ret;
 190
 191static int trace_define_generic_fields(void)
 192{
 193	int ret;
 194
 195	__generic_field(int, CPU, FILTER_CPU);
 196	__generic_field(int, cpu, FILTER_CPU);
 197	__generic_field(int, common_cpu, FILTER_CPU);
 198	__generic_field(char *, COMM, FILTER_COMM);
 199	__generic_field(char *, comm, FILTER_COMM);
 200	__generic_field(char *, stacktrace, FILTER_STACKTRACE);
 201	__generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 202
 203	return ret;
 204}
 205
 206static int trace_define_common_fields(void)
 207{
 208	int ret;
 209	struct trace_entry ent;
 210
 211	__common_field(unsigned short, type);
 212	__common_field(unsigned char, flags);
 213	/* Holds both preempt_count and migrate_disable */
 214	__common_field(unsigned char, preempt_count);
 215	__common_field(int, pid);
 216
 217	return ret;
 218}
 219
 220static void trace_destroy_fields(struct trace_event_call *call)
 221{
 222	struct ftrace_event_field *field, *next;
 223	struct list_head *head;
 224
 225	head = trace_get_fields(call);
 226	list_for_each_entry_safe(field, next, head, link) {
 227		list_del(&field->link);
 228		kmem_cache_free(field_cachep, field);
 229	}
 230}
 231
 232/*
 233 * run-time version of trace_event_get_offsets_<call>() that returns the last
 234 * accessible offset of trace fields excluding __dynamic_array bytes
 235 */
 236int trace_event_get_offsets(struct trace_event_call *call)
 237{
 238	struct ftrace_event_field *tail;
 239	struct list_head *head;
 240
 241	head = trace_get_fields(call);
 242	/*
 243	 * head->next points to the last field with the largest offset,
 244	 * since it was added last by trace_define_field()
 245	 */
 246	tail = list_first_entry(head, struct ftrace_event_field, link);
 247	return tail->offset + tail->size;
 248}
 249
 250
 251static struct trace_event_fields *find_event_field(const char *fmt,
 252						   struct trace_event_call *call)
 
 
 253{
 254	struct trace_event_fields *field = call->class->fields_array;
 
 255	const char *p = fmt;
 256	int len;
 257
 258	if (!(len = str_has_prefix(fmt, "REC->")))
 259		return NULL;
 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) || field->name[len])
 
 269			continue;
 270
 271		return field;
 272	}
 273	return NULL;
 274}
 275
 276/*
 277 * Check if the referenced field is an array and return true,
 278 * as arrays are OK to dereference.
 279 */
 280static bool test_field(const char *fmt, struct trace_event_call *call)
 281{
 282	struct trace_event_fields *field;
 283
 284	field = find_event_field(fmt, call);
 285	if (!field)
 286		return false;
 287
 288	/* This is an array and is OK to dereference. */
 289	return strchr(field->type, '[') != NULL;
 290}
 291
 292/* Look for a string within an argument */
 293static bool find_print_string(const char *arg, const char *str, const char *end)
 294{
 295	const char *r;
 296
 297	r = strstr(arg, str);
 298	return r && r < end;
 299}
 300
 301/* Return true if the argument pointer is safe */
 302static bool process_pointer(const char *fmt, int len, struct trace_event_call *call)
 303{
 304	const char *r, *e, *a;
 305
 306	e = fmt + len;
 307
 308	/* Find the REC-> in the argument */
 309	r = strstr(fmt, "REC->");
 310	if (r && r < e) {
 311		/*
 312		 * Addresses of events on the buffer, or an array on the buffer is
 313		 * OK to dereference. There's ways to fool this, but
 314		 * this is to catch common mistakes, not malicious code.
 315		 */
 316		a = strchr(fmt, '&');
 317		if ((a && (a < r)) || test_field(r, call))
 318			return true;
 319	} else if (find_print_string(fmt, "__get_dynamic_array(", e)) {
 320		return true;
 321	} else if (find_print_string(fmt, "__get_rel_dynamic_array(", e)) {
 322		return true;
 323	} else if (find_print_string(fmt, "__get_dynamic_array_len(", e)) {
 324		return true;
 325	} else if (find_print_string(fmt, "__get_rel_dynamic_array_len(", e)) {
 326		return true;
 327	} else if (find_print_string(fmt, "__get_sockaddr(", e)) {
 328		return true;
 329	} else if (find_print_string(fmt, "__get_rel_sockaddr(", e)) {
 330		return true;
 331	}
 332	return false;
 333}
 334
 335/* Return true if the string is safe */
 336static bool process_string(const char *fmt, int len, struct trace_event_call *call)
 337{
 338	struct trace_event_fields *field;
 339	const char *r, *e, *s;
 340
 341	e = fmt + len;
 342
 343	/*
 344	 * There are several helper functions that return strings.
 345	 * If the argument contains a function, then assume its field is valid.
 346	 * It is considered that the argument has a function if it has:
 347	 *   alphanumeric or '_' before a parenthesis.
 348	 */
 349	s = fmt;
 350	do {
 351		r = strstr(s, "(");
 352		if (!r || r >= e)
 353			break;
 354		for (int i = 1; r - i >= s; i++) {
 355			char ch = *(r - i);
 356			if (isspace(ch))
 357				continue;
 358			if (isalnum(ch) || ch == '_')
 359				return true;
 360			/* Anything else, this isn't a function */
 361			break;
 362		}
 363		/* A function could be wrapped in parethesis, try the next one */
 364		s = r + 1;
 365	} while (s < e);
 366
 367	/*
 368	 * Check for arrays. If the argument has: foo[REC->val]
 369	 * then it is very likely that foo is an array of strings
 370	 * that are safe to use.
 371	 */
 372	r = strstr(s, "[");
 373	if (r && r < e) {
 374		r = strstr(r, "REC->");
 375		if (r && r < e)
 376			return true;
 377	}
 378
 379	/*
 380	 * If there's any strings in the argument consider this arg OK as it
 381	 * could be: REC->field ? "foo" : "bar" and we don't want to get into
 382	 * verifying that logic here.
 383	 */
 384	if (find_print_string(fmt, "\"", e))
 385		return true;
 386
 387	/* Dereferenced strings are also valid like any other pointer */
 388	if (process_pointer(fmt, len, call))
 389		return true;
 390
 391	/* Make sure the field is found */
 392	field = find_event_field(fmt, call);
 393	if (!field)
 394		return false;
 395
 396	/* Test this field's string before printing the event */
 397	call->flags |= TRACE_EVENT_FL_TEST_STR;
 398	field->needs_test = 1;
 399
 400	return true;
 401}
 402
 403/*
 404 * Examine the print fmt of the event looking for unsafe dereference
 405 * pointers using %p* that could be recorded in the trace event and
 406 * much later referenced after the pointer was freed. Dereferencing
 407 * pointers are OK, if it is dereferenced into the event itself.
 408 */
 409static void test_event_printk(struct trace_event_call *call)
 410{
 411	u64 dereference_flags = 0;
 412	u64 string_flags = 0;
 413	bool first = true;
 414	const char *fmt;
 415	int parens = 0;
 416	char in_quote = 0;
 417	int start_arg = 0;
 418	int arg = 0;
 419	int i, e;
 420
 421	fmt = call->print_fmt;
 422
 423	if (!fmt)
 424		return;
 425
 426	for (i = 0; fmt[i]; i++) {
 427		switch (fmt[i]) {
 428		case '\\':
 429			i++;
 430			if (!fmt[i])
 431				return;
 432			continue;
 433		case '"':
 434		case '\'':
 435			/*
 436			 * The print fmt starts with a string that
 437			 * is processed first to find %p* usage,
 438			 * then after the first string, the print fmt
 439			 * contains arguments that are used to check
 440			 * if the dereferenced %p* usage is safe.
 441			 */
 442			if (first) {
 443				if (fmt[i] == '\'')
 444					continue;
 445				if (in_quote) {
 446					arg = 0;
 447					first = false;
 448					/*
 449					 * If there was no %p* uses
 450					 * the fmt is OK.
 451					 */
 452					if (!dereference_flags)
 453						return;
 454				}
 455			}
 456			if (in_quote) {
 457				if (in_quote == fmt[i])
 458					in_quote = 0;
 459			} else {
 460				in_quote = fmt[i];
 461			}
 462			continue;
 463		case '%':
 464			if (!first || !in_quote)
 465				continue;
 466			i++;
 467			if (!fmt[i])
 468				return;
 469			switch (fmt[i]) {
 470			case '%':
 471				continue;
 472			case 'p':
 473				/* Find dereferencing fields */
 474				switch (fmt[i + 1]) {
 475				case 'B': case 'R': case 'r':
 476				case 'b': case 'M': case 'm':
 477				case 'I': case 'i': case 'E':
 478				case 'U': case 'V': case 'N':
 479				case 'a': case 'd': case 'D':
 480				case 'g': case 't': case 'C':
 481				case 'O': case 'f':
 482					if (WARN_ONCE(arg == 63,
 483						      "Too many args for event: %s",
 484						      trace_event_name(call)))
 485						return;
 486					dereference_flags |= 1ULL << arg;
 487				}
 488				break;
 489			default:
 490			{
 491				bool star = false;
 492				int j;
 493
 494				/* Increment arg if %*s exists. */
 495				for (j = 0; fmt[i + j]; j++) {
 496					if (isdigit(fmt[i + j]) ||
 497					    fmt[i + j] == '.')
 498						continue;
 499					if (fmt[i + j] == '*') {
 500						star = true;
 501						continue;
 502					}
 503					if ((fmt[i + j] == 's')) {
 504						if (star)
 505							arg++;
 506						if (WARN_ONCE(arg == 63,
 507							      "Too many args for event: %s",
 508							      trace_event_name(call)))
 509							return;
 510						dereference_flags |= 1ULL << arg;
 511						string_flags |= 1ULL << arg;
 512					}
 513					break;
 514				}
 515				break;
 516			} /* default */
 517
 518			} /* switch */
 519			arg++;
 520			continue;
 521		case '(':
 522			if (in_quote)
 523				continue;
 524			parens++;
 525			continue;
 526		case ')':
 527			if (in_quote)
 528				continue;
 529			parens--;
 530			if (WARN_ONCE(parens < 0,
 531				      "Paren mismatch for event: %s\narg='%s'\n%*s",
 532				      trace_event_name(call),
 533				      fmt + start_arg,
 534				      (i - start_arg) + 5, "^"))
 535				return;
 536			continue;
 537		case ',':
 538			if (in_quote || parens)
 539				continue;
 540			e = i;
 541			i++;
 542			while (isspace(fmt[i]))
 543				i++;
 
 
 
 544
 545			/*
 546			 * If start_arg is zero, then this is the start of the
 547			 * first argument. The processing of the argument happens
 548			 * when the end of the argument is found, as it needs to
 549			 * handle paranthesis and such.
 550			 */
 551			if (!start_arg) {
 552				start_arg = i;
 553				/* Balance out the i++ in the for loop */
 554				i--;
 555				continue;
 556			}
 557
 558			if (dereference_flags & (1ULL << arg)) {
 559				if (string_flags & (1ULL << arg)) {
 560					if (process_string(fmt + start_arg, e - start_arg, call))
 561						dereference_flags &= ~(1ULL << arg);
 562				} else if (process_pointer(fmt + start_arg, e - start_arg, call))
 563					dereference_flags &= ~(1ULL << arg);
 
 
 
 
 
 
 564			}
 565
 566			start_arg = i;
 567			arg++;
 568			/* Balance out the i++ in the for loop */
 569			i--;
 
 570		}
 571	}
 572
 573	if (dereference_flags & (1ULL << arg)) {
 574		if (string_flags & (1ULL << arg)) {
 575			if (process_string(fmt + start_arg, i - start_arg, call))
 576				dereference_flags &= ~(1ULL << arg);
 577		} else if (process_pointer(fmt + start_arg, i - start_arg, call))
 578			dereference_flags &= ~(1ULL << arg);
 579	}
 580
 581	/*
 582	 * If you triggered the below warning, the trace event reported
 583	 * uses an unsafe dereference pointer %p*. As the data stored
 584	 * at the trace event time may no longer exist when the trace
 585	 * event is printed, dereferencing to the original source is
 586	 * unsafe. The source of the dereference must be copied into the
 587	 * event itself, and the dereference must access the copy instead.
 588	 */
 589	if (WARN_ON_ONCE(dereference_flags)) {
 590		arg = 1;
 591		while (!(dereference_flags & 1)) {
 592			dereference_flags >>= 1;
 593			arg++;
 594		}
 595		pr_warn("event %s has unsafe dereference of argument %d\n",
 596			trace_event_name(call), arg);
 597		pr_warn("print_fmt: %s\n", fmt);
 598	}
 599}
 600
 601int trace_event_raw_init(struct trace_event_call *call)
 602{
 603	int id;
 604
 605	id = register_trace_event(&call->event);
 606	if (!id)
 607		return -ENODEV;
 608
 609	test_event_printk(call);
 610
 611	return 0;
 612}
 613EXPORT_SYMBOL_GPL(trace_event_raw_init);
 614
 615bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 616{
 617	struct trace_array *tr = trace_file->tr;
 618	struct trace_array_cpu *data;
 619	struct trace_pid_list *no_pid_list;
 620	struct trace_pid_list *pid_list;
 621
 622	pid_list = rcu_dereference_raw(tr->filtered_pids);
 623	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 624
 625	if (!pid_list && !no_pid_list)
 626		return false;
 627
 628	data = this_cpu_ptr(tr->array_buffer.data);
 629
 630	return data->ignore_pid;
 631}
 632EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 633
 634void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 635				 struct trace_event_file *trace_file,
 636				 unsigned long len)
 637{
 638	struct trace_event_call *event_call = trace_file->event_call;
 639
 640	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 641	    trace_event_ignore_this_pid(trace_file))
 642		return NULL;
 643
 644	/*
 645	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 646	 * preemption (adding one to the preempt_count). Since we are
 647	 * interested in the preempt_count at the time the tracepoint was
 648	 * hit, we need to subtract one to offset the increment.
 649	 */
 650	fbuffer->trace_ctx = tracing_gen_ctx_dec();
 651	fbuffer->trace_file = trace_file;
 652
 653	fbuffer->event =
 654		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 655						event_call->event.type, len,
 656						fbuffer->trace_ctx);
 657	if (!fbuffer->event)
 658		return NULL;
 659
 660	fbuffer->regs = NULL;
 661	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 662	return fbuffer->entry;
 663}
 664EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 665
 666int trace_event_reg(struct trace_event_call *call,
 667		    enum trace_reg type, void *data)
 668{
 669	struct trace_event_file *file = data;
 670
 671	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 672	switch (type) {
 673	case TRACE_REG_REGISTER:
 674		return tracepoint_probe_register(call->tp,
 675						 call->class->probe,
 676						 file);
 677	case TRACE_REG_UNREGISTER:
 678		tracepoint_probe_unregister(call->tp,
 679					    call->class->probe,
 680					    file);
 681		return 0;
 682
 683#ifdef CONFIG_PERF_EVENTS
 684	case TRACE_REG_PERF_REGISTER:
 685		return tracepoint_probe_register(call->tp,
 686						 call->class->perf_probe,
 687						 call);
 688	case TRACE_REG_PERF_UNREGISTER:
 689		tracepoint_probe_unregister(call->tp,
 690					    call->class->perf_probe,
 691					    call);
 692		return 0;
 693	case TRACE_REG_PERF_OPEN:
 694	case TRACE_REG_PERF_CLOSE:
 695	case TRACE_REG_PERF_ADD:
 696	case TRACE_REG_PERF_DEL:
 697		return 0;
 698#endif
 699	}
 700	return 0;
 701}
 702EXPORT_SYMBOL_GPL(trace_event_reg);
 703
 704void trace_event_enable_cmd_record(bool enable)
 705{
 706	struct trace_event_file *file;
 707	struct trace_array *tr;
 708
 709	lockdep_assert_held(&event_mutex);
 710
 711	do_for_each_event_file(tr, file) {
 712
 713		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 714			continue;
 715
 716		if (enable) {
 717			tracing_start_cmdline_record();
 718			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 719		} else {
 720			tracing_stop_cmdline_record();
 721			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 722		}
 723	} while_for_each_event_file();
 724}
 725
 726void trace_event_enable_tgid_record(bool enable)
 727{
 728	struct trace_event_file *file;
 729	struct trace_array *tr;
 730
 731	lockdep_assert_held(&event_mutex);
 732
 733	do_for_each_event_file(tr, file) {
 734		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 735			continue;
 736
 737		if (enable) {
 738			tracing_start_tgid_record();
 739			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 740		} else {
 741			tracing_stop_tgid_record();
 742			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 743				  &file->flags);
 744		}
 745	} while_for_each_event_file();
 746}
 747
 748static int __ftrace_event_enable_disable(struct trace_event_file *file,
 749					 int enable, int soft_disable)
 750{
 751	struct trace_event_call *call = file->event_call;
 752	struct trace_array *tr = file->tr;
 753	int ret = 0;
 754	int disable;
 755
 756	switch (enable) {
 757	case 0:
 758		/*
 759		 * When soft_disable is set and enable is cleared, the sm_ref
 760		 * reference counter is decremented. If it reaches 0, we want
 761		 * to clear the SOFT_DISABLED flag but leave the event in the
 762		 * state that it was. That is, if the event was enabled and
 763		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 764		 * is set we do not want the event to be enabled before we
 765		 * clear the bit.
 766		 *
 767		 * When soft_disable is not set but the SOFT_MODE flag is,
 768		 * we do nothing. Do not disable the tracepoint, otherwise
 769		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 770		 */
 771		if (soft_disable) {
 772			if (atomic_dec_return(&file->sm_ref) > 0)
 773				break;
 774			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 775			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 776			/* Disable use of trace_buffered_event */
 777			trace_buffered_event_disable();
 778		} else
 779			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 780
 781		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 782			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 783			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 784				tracing_stop_cmdline_record();
 785				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 786			}
 787
 788			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 789				tracing_stop_tgid_record();
 790				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 791			}
 792
 793			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 794		}
 795		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 796		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 797			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 798		else
 799			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 800		break;
 801	case 1:
 802		/*
 803		 * When soft_disable is set and enable is set, we want to
 804		 * register the tracepoint for the event, but leave the event
 805		 * as is. That means, if the event was already enabled, we do
 806		 * nothing (but set SOFT_MODE). If the event is disabled, we
 807		 * set SOFT_DISABLED before enabling the event tracepoint, so
 808		 * it still seems to be disabled.
 809		 */
 810		if (!soft_disable)
 811			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 812		else {
 813			if (atomic_inc_return(&file->sm_ref) > 1)
 814				break;
 815			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 816			/* Enable use of trace_buffered_event */
 817			trace_buffered_event_enable();
 818		}
 819
 820		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 821			bool cmd = false, tgid = false;
 822
 823			/* Keep the event disabled, when going to SOFT_MODE. */
 824			if (soft_disable)
 825				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 826
 827			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 828				cmd = true;
 829				tracing_start_cmdline_record();
 830				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 831			}
 832
 833			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 834				tgid = true;
 835				tracing_start_tgid_record();
 836				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 837			}
 838
 839			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 840			if (ret) {
 841				if (cmd)
 842					tracing_stop_cmdline_record();
 843				if (tgid)
 844					tracing_stop_tgid_record();
 845				pr_info("event trace: Could not enable event "
 846					"%s\n", trace_event_name(call));
 847				break;
 848			}
 849			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 850
 851			/* WAS_ENABLED gets set but never cleared. */
 852			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 853		}
 854		break;
 855	}
 856
 857	return ret;
 858}
 859
 860int trace_event_enable_disable(struct trace_event_file *file,
 861			       int enable, int soft_disable)
 862{
 863	return __ftrace_event_enable_disable(file, enable, soft_disable);
 864}
 865
 866static int ftrace_event_enable_disable(struct trace_event_file *file,
 867				       int enable)
 868{
 869	return __ftrace_event_enable_disable(file, enable, 0);
 870}
 871
 872static void ftrace_clear_events(struct trace_array *tr)
 873{
 874	struct trace_event_file *file;
 875
 876	mutex_lock(&event_mutex);
 877	list_for_each_entry(file, &tr->events, list) {
 878		ftrace_event_enable_disable(file, 0);
 879	}
 880	mutex_unlock(&event_mutex);
 881}
 882
 883static void
 884event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 885{
 886	struct trace_pid_list *pid_list;
 887	struct trace_array *tr = data;
 888
 889	pid_list = rcu_dereference_raw(tr->filtered_pids);
 890	trace_filter_add_remove_task(pid_list, NULL, task);
 891
 892	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 893	trace_filter_add_remove_task(pid_list, NULL, task);
 894}
 895
 896static void
 897event_filter_pid_sched_process_fork(void *data,
 898				    struct task_struct *self,
 899				    struct task_struct *task)
 900{
 901	struct trace_pid_list *pid_list;
 902	struct trace_array *tr = data;
 903
 904	pid_list = rcu_dereference_sched(tr->filtered_pids);
 905	trace_filter_add_remove_task(pid_list, self, task);
 906
 907	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 908	trace_filter_add_remove_task(pid_list, self, task);
 909}
 910
 911void trace_event_follow_fork(struct trace_array *tr, bool enable)
 912{
 913	if (enable) {
 914		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 915						       tr, INT_MIN);
 916		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 917						       tr, INT_MAX);
 918	} else {
 919		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 920						    tr);
 921		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 922						    tr);
 923	}
 924}
 925
 926static void
 927event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 928					struct task_struct *prev,
 929					struct task_struct *next,
 930					unsigned int prev_state)
 931{
 932	struct trace_array *tr = data;
 933	struct trace_pid_list *no_pid_list;
 934	struct trace_pid_list *pid_list;
 935	bool ret;
 936
 937	pid_list = rcu_dereference_sched(tr->filtered_pids);
 938	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 939
 940	/*
 941	 * Sched switch is funny, as we only want to ignore it
 942	 * in the notrace case if both prev and next should be ignored.
 943	 */
 944	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 945		trace_ignore_this_task(NULL, no_pid_list, next);
 946
 947	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 948		       (trace_ignore_this_task(pid_list, NULL, prev) &&
 949			trace_ignore_this_task(pid_list, NULL, next)));
 950}
 951
 952static void
 953event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 954					 struct task_struct *prev,
 955					 struct task_struct *next,
 956					 unsigned int prev_state)
 957{
 958	struct trace_array *tr = data;
 959	struct trace_pid_list *no_pid_list;
 960	struct trace_pid_list *pid_list;
 961
 962	pid_list = rcu_dereference_sched(tr->filtered_pids);
 963	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 964
 965	this_cpu_write(tr->array_buffer.data->ignore_pid,
 966		       trace_ignore_this_task(pid_list, no_pid_list, next));
 967}
 968
 969static void
 970event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 971{
 972	struct trace_array *tr = data;
 973	struct trace_pid_list *no_pid_list;
 974	struct trace_pid_list *pid_list;
 975
 976	/* Nothing to do if we are already tracing */
 977	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 978		return;
 979
 980	pid_list = rcu_dereference_sched(tr->filtered_pids);
 981	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 982
 983	this_cpu_write(tr->array_buffer.data->ignore_pid,
 984		       trace_ignore_this_task(pid_list, no_pid_list, task));
 985}
 986
 987static void
 988event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 989{
 990	struct trace_array *tr = data;
 991	struct trace_pid_list *no_pid_list;
 992	struct trace_pid_list *pid_list;
 993
 994	/* Nothing to do if we are not tracing */
 995	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 996		return;
 997
 998	pid_list = rcu_dereference_sched(tr->filtered_pids);
 999	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1000
1001	/* Set tracing if current is enabled */
1002	this_cpu_write(tr->array_buffer.data->ignore_pid,
1003		       trace_ignore_this_task(pid_list, no_pid_list, current));
1004}
1005
1006static void unregister_pid_events(struct trace_array *tr)
1007{
1008	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
1009	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
1010
1011	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
1012	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
1013
1014	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
1015	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
1016
1017	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
1018	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
1019}
1020
1021static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
1022{
1023	struct trace_pid_list *pid_list;
1024	struct trace_pid_list *no_pid_list;
1025	struct trace_event_file *file;
1026	int cpu;
1027
1028	pid_list = rcu_dereference_protected(tr->filtered_pids,
1029					     lockdep_is_held(&event_mutex));
1030	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1031					     lockdep_is_held(&event_mutex));
1032
1033	/* Make sure there's something to do */
1034	if (!pid_type_enabled(type, pid_list, no_pid_list))
1035		return;
1036
1037	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
1038		unregister_pid_events(tr);
1039
1040		list_for_each_entry(file, &tr->events, list) {
1041			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1042		}
1043
1044		for_each_possible_cpu(cpu)
1045			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
1046	}
1047
1048	if (type & TRACE_PIDS)
1049		rcu_assign_pointer(tr->filtered_pids, NULL);
1050
1051	if (type & TRACE_NO_PIDS)
1052		rcu_assign_pointer(tr->filtered_no_pids, NULL);
1053
1054	/* Wait till all users are no longer using pid filtering */
1055	tracepoint_synchronize_unregister();
1056
1057	if ((type & TRACE_PIDS) && pid_list)
1058		trace_pid_list_free(pid_list);
1059
1060	if ((type & TRACE_NO_PIDS) && no_pid_list)
1061		trace_pid_list_free(no_pid_list);
1062}
1063
1064static void ftrace_clear_event_pids(struct trace_array *tr, int type)
1065{
1066	mutex_lock(&event_mutex);
1067	__ftrace_clear_event_pids(tr, type);
1068	mutex_unlock(&event_mutex);
1069}
1070
1071static void __put_system(struct event_subsystem *system)
1072{
1073	struct event_filter *filter = system->filter;
1074
1075	WARN_ON_ONCE(system_refcount(system) == 0);
1076	if (system_refcount_dec(system))
1077		return;
1078
1079	list_del(&system->list);
1080
1081	if (filter) {
1082		kfree(filter->filter_string);
1083		kfree(filter);
1084	}
1085	kfree_const(system->name);
1086	kfree(system);
1087}
1088
1089static void __get_system(struct event_subsystem *system)
1090{
1091	WARN_ON_ONCE(system_refcount(system) == 0);
1092	system_refcount_inc(system);
1093}
1094
1095static void __get_system_dir(struct trace_subsystem_dir *dir)
1096{
1097	WARN_ON_ONCE(dir->ref_count == 0);
1098	dir->ref_count++;
1099	__get_system(dir->subsystem);
1100}
1101
1102static void __put_system_dir(struct trace_subsystem_dir *dir)
1103{
1104	WARN_ON_ONCE(dir->ref_count == 0);
1105	/* If the subsystem is about to be freed, the dir must be too */
1106	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
1107
1108	__put_system(dir->subsystem);
1109	if (!--dir->ref_count)
1110		kfree(dir);
1111}
1112
1113static void put_system(struct trace_subsystem_dir *dir)
1114{
1115	mutex_lock(&event_mutex);
1116	__put_system_dir(dir);
1117	mutex_unlock(&event_mutex);
1118}
1119
1120static void remove_subsystem(struct trace_subsystem_dir *dir)
1121{
1122	if (!dir)
1123		return;
1124
1125	if (!--dir->nr_events) {
1126		eventfs_remove_dir(dir->ei);
1127		list_del(&dir->list);
1128		__put_system_dir(dir);
1129	}
1130}
1131
1132void event_file_get(struct trace_event_file *file)
1133{
1134	refcount_inc(&file->ref);
1135}
1136
1137void event_file_put(struct trace_event_file *file)
1138{
1139	if (WARN_ON_ONCE(!refcount_read(&file->ref))) {
1140		if (file->flags & EVENT_FILE_FL_FREED)
1141			kmem_cache_free(file_cachep, file);
1142		return;
1143	}
1144
1145	if (refcount_dec_and_test(&file->ref)) {
1146		/* Count should only go to zero when it is freed */
1147		if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1148			return;
1149		kmem_cache_free(file_cachep, file);
1150	}
1151}
1152
1153static void remove_event_file_dir(struct trace_event_file *file)
1154{
1155	eventfs_remove_dir(file->ei);
1156	list_del(&file->list);
1157	remove_subsystem(file->system);
1158	free_event_filter(file->filter);
1159	file->flags |= EVENT_FILE_FL_FREED;
1160	event_file_put(file);
1161}
1162
1163/*
1164 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1165 */
1166static int
1167__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1168			      const char *sub, const char *event, int set)
1169{
1170	struct trace_event_file *file;
1171	struct trace_event_call *call;
1172	const char *name;
1173	int ret = -EINVAL;
1174	int eret = 0;
1175
1176	list_for_each_entry(file, &tr->events, list) {
1177
1178		call = file->event_call;
1179		name = trace_event_name(call);
1180
1181		if (!name || !call->class || !call->class->reg)
1182			continue;
1183
1184		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1185			continue;
1186
1187		if (match &&
1188		    strcmp(match, name) != 0 &&
1189		    strcmp(match, call->class->system) != 0)
1190			continue;
1191
1192		if (sub && strcmp(sub, call->class->system) != 0)
1193			continue;
1194
1195		if (event && strcmp(event, name) != 0)
1196			continue;
1197
1198		ret = ftrace_event_enable_disable(file, set);
1199
1200		/*
1201		 * Save the first error and return that. Some events
1202		 * may still have been enabled, but let the user
1203		 * know that something went wrong.
1204		 */
1205		if (ret && !eret)
1206			eret = ret;
1207
1208		ret = eret;
1209	}
1210
1211	return ret;
1212}
1213
1214static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1215				  const char *sub, const char *event, int set)
1216{
1217	int ret;
1218
1219	mutex_lock(&event_mutex);
1220	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1221	mutex_unlock(&event_mutex);
1222
1223	return ret;
1224}
1225
1226int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1227{
1228	char *event = NULL, *sub = NULL, *match;
1229	int ret;
1230
1231	if (!tr)
1232		return -ENOENT;
1233	/*
1234	 * The buf format can be <subsystem>:<event-name>
1235	 *  *:<event-name> means any event by that name.
1236	 *  :<event-name> is the same.
1237	 *
1238	 *  <subsystem>:* means all events in that subsystem
1239	 *  <subsystem>: means the same.
1240	 *
1241	 *  <name> (no ':') means all events in a subsystem with
1242	 *  the name <name> or any event that matches <name>
1243	 */
1244
1245	match = strsep(&buf, ":");
1246	if (buf) {
1247		sub = match;
1248		event = buf;
1249		match = NULL;
1250
1251		if (!strlen(sub) || strcmp(sub, "*") == 0)
1252			sub = NULL;
1253		if (!strlen(event) || strcmp(event, "*") == 0)
1254			event = NULL;
1255	}
1256
1257	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1258
1259	/* Put back the colon to allow this to be called again */
1260	if (buf)
1261		*(buf - 1) = ':';
1262
1263	return ret;
1264}
1265
1266/**
1267 * trace_set_clr_event - enable or disable an event
1268 * @system: system name to match (NULL for any system)
1269 * @event: event name to match (NULL for all events, within system)
1270 * @set: 1 to enable, 0 to disable
1271 *
1272 * This is a way for other parts of the kernel to enable or disable
1273 * event recording.
1274 *
1275 * Returns 0 on success, -EINVAL if the parameters do not match any
1276 * registered events.
1277 */
1278int trace_set_clr_event(const char *system, const char *event, int set)
1279{
1280	struct trace_array *tr = top_trace_array();
1281
1282	if (!tr)
1283		return -ENODEV;
1284
1285	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1286}
1287EXPORT_SYMBOL_GPL(trace_set_clr_event);
1288
1289/**
1290 * trace_array_set_clr_event - enable or disable an event for a trace array.
1291 * @tr: concerned trace array.
1292 * @system: system name to match (NULL for any system)
1293 * @event: event name to match (NULL for all events, within system)
1294 * @enable: true to enable, false to disable
1295 *
1296 * This is a way for other parts of the kernel to enable or disable
1297 * event recording.
1298 *
1299 * Returns 0 on success, -EINVAL if the parameters do not match any
1300 * registered events.
1301 */
1302int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1303		const char *event, bool enable)
1304{
1305	int set;
1306
1307	if (!tr)
1308		return -ENOENT;
1309
1310	set = (enable == true) ? 1 : 0;
1311	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1312}
1313EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1314
1315/* 128 should be much more than enough */
1316#define EVENT_BUF_SIZE		127
1317
1318static ssize_t
1319ftrace_event_write(struct file *file, const char __user *ubuf,
1320		   size_t cnt, loff_t *ppos)
1321{
1322	struct trace_parser parser;
1323	struct seq_file *m = file->private_data;
1324	struct trace_array *tr = m->private;
1325	ssize_t read, ret;
1326
1327	if (!cnt)
1328		return 0;
1329
1330	ret = tracing_update_buffers(tr);
1331	if (ret < 0)
1332		return ret;
1333
1334	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1335		return -ENOMEM;
1336
1337	read = trace_get_user(&parser, ubuf, cnt, ppos);
1338
1339	if (read >= 0 && trace_parser_loaded((&parser))) {
1340		int set = 1;
1341
1342		if (*parser.buffer == '!')
1343			set = 0;
1344
1345		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1346		if (ret)
1347			goto out_put;
1348	}
1349
1350	ret = read;
1351
1352 out_put:
1353	trace_parser_put(&parser);
1354
1355	return ret;
1356}
1357
1358static void *
1359t_next(struct seq_file *m, void *v, loff_t *pos)
1360{
1361	struct trace_event_file *file = v;
1362	struct trace_event_call *call;
1363	struct trace_array *tr = m->private;
1364
1365	(*pos)++;
1366
1367	list_for_each_entry_continue(file, &tr->events, list) {
1368		call = file->event_call;
1369		/*
1370		 * The ftrace subsystem is for showing formats only.
1371		 * They can not be enabled or disabled via the event files.
1372		 */
1373		if (call->class && call->class->reg &&
1374		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1375			return file;
1376	}
1377
1378	return NULL;
1379}
1380
1381static void *t_start(struct seq_file *m, loff_t *pos)
1382{
1383	struct trace_event_file *file;
1384	struct trace_array *tr = m->private;
1385	loff_t l;
1386
1387	mutex_lock(&event_mutex);
1388
1389	file = list_entry(&tr->events, struct trace_event_file, list);
1390	for (l = 0; l <= *pos; ) {
1391		file = t_next(m, file, &l);
1392		if (!file)
1393			break;
1394	}
1395	return file;
1396}
1397
1398static void *
1399s_next(struct seq_file *m, void *v, loff_t *pos)
1400{
1401	struct trace_event_file *file = v;
1402	struct trace_array *tr = m->private;
1403
1404	(*pos)++;
1405
1406	list_for_each_entry_continue(file, &tr->events, list) {
1407		if (file->flags & EVENT_FILE_FL_ENABLED)
1408			return file;
1409	}
1410
1411	return NULL;
1412}
1413
1414static void *s_start(struct seq_file *m, loff_t *pos)
1415{
1416	struct trace_event_file *file;
1417	struct trace_array *tr = m->private;
1418	loff_t l;
1419
1420	mutex_lock(&event_mutex);
1421
1422	file = list_entry(&tr->events, struct trace_event_file, list);
1423	for (l = 0; l <= *pos; ) {
1424		file = s_next(m, file, &l);
1425		if (!file)
1426			break;
1427	}
1428	return file;
1429}
1430
1431static int t_show(struct seq_file *m, void *v)
1432{
1433	struct trace_event_file *file = v;
1434	struct trace_event_call *call = file->event_call;
1435
1436	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1437		seq_printf(m, "%s:", call->class->system);
1438	seq_printf(m, "%s\n", trace_event_name(call));
1439
1440	return 0;
1441}
1442
1443static void t_stop(struct seq_file *m, void *p)
1444{
1445	mutex_unlock(&event_mutex);
1446}
1447
1448static void *
1449__next(struct seq_file *m, void *v, loff_t *pos, int type)
1450{
1451	struct trace_array *tr = m->private;
1452	struct trace_pid_list *pid_list;
1453
1454	if (type == TRACE_PIDS)
1455		pid_list = rcu_dereference_sched(tr->filtered_pids);
1456	else
1457		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1458
1459	return trace_pid_next(pid_list, v, pos);
1460}
1461
1462static void *
1463p_next(struct seq_file *m, void *v, loff_t *pos)
1464{
1465	return __next(m, v, pos, TRACE_PIDS);
1466}
1467
1468static void *
1469np_next(struct seq_file *m, void *v, loff_t *pos)
1470{
1471	return __next(m, v, pos, TRACE_NO_PIDS);
1472}
1473
1474static void *__start(struct seq_file *m, loff_t *pos, int type)
1475	__acquires(RCU)
1476{
1477	struct trace_pid_list *pid_list;
1478	struct trace_array *tr = m->private;
1479
1480	/*
1481	 * Grab the mutex, to keep calls to p_next() having the same
1482	 * tr->filtered_pids as p_start() has.
1483	 * If we just passed the tr->filtered_pids around, then RCU would
1484	 * have been enough, but doing that makes things more complex.
1485	 */
1486	mutex_lock(&event_mutex);
1487	rcu_read_lock_sched();
1488
1489	if (type == TRACE_PIDS)
1490		pid_list = rcu_dereference_sched(tr->filtered_pids);
1491	else
1492		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1493
1494	if (!pid_list)
1495		return NULL;
1496
1497	return trace_pid_start(pid_list, pos);
1498}
1499
1500static void *p_start(struct seq_file *m, loff_t *pos)
1501	__acquires(RCU)
1502{
1503	return __start(m, pos, TRACE_PIDS);
1504}
1505
1506static void *np_start(struct seq_file *m, loff_t *pos)
1507	__acquires(RCU)
1508{
1509	return __start(m, pos, TRACE_NO_PIDS);
1510}
1511
1512static void p_stop(struct seq_file *m, void *p)
1513	__releases(RCU)
1514{
1515	rcu_read_unlock_sched();
1516	mutex_unlock(&event_mutex);
1517}
1518
1519static ssize_t
1520event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1521		  loff_t *ppos)
1522{
1523	struct trace_event_file *file;
1524	unsigned long flags;
1525	char buf[4] = "0";
1526
1527	mutex_lock(&event_mutex);
1528	file = event_file_file(filp);
1529	if (likely(file))
1530		flags = file->flags;
1531	mutex_unlock(&event_mutex);
1532
1533	if (!file)
1534		return -ENODEV;
1535
1536	if (flags & EVENT_FILE_FL_ENABLED &&
1537	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1538		strcpy(buf, "1");
1539
1540	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1541	    flags & EVENT_FILE_FL_SOFT_MODE)
1542		strcat(buf, "*");
1543
1544	strcat(buf, "\n");
1545
1546	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1547}
1548
1549static ssize_t
1550event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1551		   loff_t *ppos)
1552{
1553	struct trace_event_file *file;
1554	unsigned long val;
1555	int ret;
1556
1557	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1558	if (ret)
1559		return ret;
1560
1561	switch (val) {
1562	case 0:
1563	case 1:
1564		ret = -ENODEV;
1565		mutex_lock(&event_mutex);
1566		file = event_file_file(filp);
1567		if (likely(file)) {
1568			ret = tracing_update_buffers(file->tr);
1569			if (ret < 0) {
1570				mutex_unlock(&event_mutex);
1571				return ret;
1572			}
1573			ret = ftrace_event_enable_disable(file, val);
1574		}
1575		mutex_unlock(&event_mutex);
1576		break;
1577
1578	default:
1579		return -EINVAL;
1580	}
1581
1582	*ppos += cnt;
1583
1584	return ret ? ret : cnt;
1585}
1586
1587static ssize_t
1588system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1589		   loff_t *ppos)
1590{
1591	const char set_to_char[4] = { '?', '0', '1', 'X' };
1592	struct trace_subsystem_dir *dir = filp->private_data;
1593	struct event_subsystem *system = dir->subsystem;
1594	struct trace_event_call *call;
1595	struct trace_event_file *file;
1596	struct trace_array *tr = dir->tr;
1597	char buf[2];
1598	int set = 0;
1599	int ret;
1600
1601	mutex_lock(&event_mutex);
1602	list_for_each_entry(file, &tr->events, list) {
1603		call = file->event_call;
1604		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1605		    !trace_event_name(call) || !call->class || !call->class->reg)
1606			continue;
1607
1608		if (system && strcmp(call->class->system, system->name) != 0)
1609			continue;
1610
1611		/*
1612		 * We need to find out if all the events are set
1613		 * or if all events or cleared, or if we have
1614		 * a mixture.
1615		 */
1616		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1617
1618		/*
1619		 * If we have a mixture, no need to look further.
1620		 */
1621		if (set == 3)
1622			break;
1623	}
1624	mutex_unlock(&event_mutex);
1625
1626	buf[0] = set_to_char[set];
1627	buf[1] = '\n';
1628
1629	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1630
1631	return ret;
1632}
1633
1634static ssize_t
1635system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1636		    loff_t *ppos)
1637{
1638	struct trace_subsystem_dir *dir = filp->private_data;
1639	struct event_subsystem *system = dir->subsystem;
1640	const char *name = NULL;
1641	unsigned long val;
1642	ssize_t ret;
1643
1644	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1645	if (ret)
1646		return ret;
1647
1648	ret = tracing_update_buffers(dir->tr);
1649	if (ret < 0)
1650		return ret;
1651
1652	if (val != 0 && val != 1)
1653		return -EINVAL;
1654
1655	/*
1656	 * Opening of "enable" adds a ref count to system,
1657	 * so the name is safe to use.
1658	 */
1659	if (system)
1660		name = system->name;
1661
1662	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1663	if (ret)
1664		goto out;
1665
1666	ret = cnt;
1667
1668out:
1669	*ppos += cnt;
1670
1671	return ret;
1672}
1673
1674enum {
1675	FORMAT_HEADER		= 1,
1676	FORMAT_FIELD_SEPERATOR	= 2,
1677	FORMAT_PRINTFMT		= 3,
1678};
1679
1680static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1681{
1682	struct trace_event_file *file = event_file_data(m->private);
1683	struct trace_event_call *call = file->event_call;
1684	struct list_head *common_head = &ftrace_common_fields;
1685	struct list_head *head = trace_get_fields(call);
1686	struct list_head *node = v;
1687
1688	(*pos)++;
1689
1690	switch ((unsigned long)v) {
1691	case FORMAT_HEADER:
1692		node = common_head;
1693		break;
1694
1695	case FORMAT_FIELD_SEPERATOR:
1696		node = head;
1697		break;
1698
1699	case FORMAT_PRINTFMT:
1700		/* all done */
1701		return NULL;
1702	}
1703
1704	node = node->prev;
1705	if (node == common_head)
1706		return (void *)FORMAT_FIELD_SEPERATOR;
1707	else if (node == head)
1708		return (void *)FORMAT_PRINTFMT;
1709	else
1710		return node;
1711}
1712
1713static int f_show(struct seq_file *m, void *v)
1714{
1715	struct trace_event_file *file = event_file_data(m->private);
1716	struct trace_event_call *call = file->event_call;
1717	struct ftrace_event_field *field;
1718	const char *array_descriptor;
1719
1720	switch ((unsigned long)v) {
1721	case FORMAT_HEADER:
1722		seq_printf(m, "name: %s\n", trace_event_name(call));
1723		seq_printf(m, "ID: %d\n", call->event.type);
1724		seq_puts(m, "format:\n");
1725		return 0;
1726
1727	case FORMAT_FIELD_SEPERATOR:
1728		seq_putc(m, '\n');
1729		return 0;
1730
1731	case FORMAT_PRINTFMT:
1732		seq_printf(m, "\nprint fmt: %s\n",
1733			   call->print_fmt);
1734		return 0;
1735	}
1736
1737	field = list_entry(v, struct ftrace_event_field, link);
1738	/*
1739	 * Smartly shows the array type(except dynamic array).
1740	 * Normal:
1741	 *	field:TYPE VAR
1742	 * If TYPE := TYPE[LEN], it is shown:
1743	 *	field:TYPE VAR[LEN]
1744	 */
1745	array_descriptor = strchr(field->type, '[');
1746
1747	if (str_has_prefix(field->type, "__data_loc"))
1748		array_descriptor = NULL;
1749
1750	if (!array_descriptor)
1751		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1752			   field->type, field->name, field->offset,
1753			   field->size, !!field->is_signed);
1754	else if (field->len)
1755		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1756			   (int)(array_descriptor - field->type),
1757			   field->type, field->name,
1758			   field->len, field->offset,
1759			   field->size, !!field->is_signed);
1760	else
1761		seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1762				(int)(array_descriptor - field->type),
1763				field->type, field->name,
1764				field->offset, field->size, !!field->is_signed);
1765
1766	return 0;
1767}
1768
1769static void *f_start(struct seq_file *m, loff_t *pos)
1770{
1771	struct trace_event_file *file;
1772	void *p = (void *)FORMAT_HEADER;
1773	loff_t l = 0;
1774
1775	/* ->stop() is called even if ->start() fails */
1776	mutex_lock(&event_mutex);
1777	file = event_file_file(m->private);
1778	if (!file)
1779		return ERR_PTR(-ENODEV);
1780
1781	while (l < *pos && p)
1782		p = f_next(m, p, &l);
1783
1784	return p;
1785}
1786
1787static void f_stop(struct seq_file *m, void *p)
1788{
1789	mutex_unlock(&event_mutex);
1790}
1791
1792static const struct seq_operations trace_format_seq_ops = {
1793	.start		= f_start,
1794	.next		= f_next,
1795	.stop		= f_stop,
1796	.show		= f_show,
1797};
1798
1799static int trace_format_open(struct inode *inode, struct file *file)
1800{
1801	struct seq_file *m;
1802	int ret;
1803
1804	/* Do we want to hide event format files on tracefs lockdown? */
1805
1806	ret = seq_open(file, &trace_format_seq_ops);
1807	if (ret < 0)
1808		return ret;
1809
1810	m = file->private_data;
1811	m->private = file;
1812
1813	return 0;
1814}
1815
1816#ifdef CONFIG_PERF_EVENTS
1817static ssize_t
1818event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1819{
1820	int id = (long)event_file_data(filp);
1821	char buf[32];
1822	int len;
1823
1824	if (unlikely(!id))
1825		return -ENODEV;
1826
1827	len = sprintf(buf, "%d\n", id);
1828
1829	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1830}
1831#endif
1832
1833static ssize_t
1834event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1835		  loff_t *ppos)
1836{
1837	struct trace_event_file *file;
1838	struct trace_seq *s;
1839	int r = -ENODEV;
1840
1841	if (*ppos)
1842		return 0;
1843
1844	s = kmalloc(sizeof(*s), GFP_KERNEL);
1845
1846	if (!s)
1847		return -ENOMEM;
1848
1849	trace_seq_init(s);
1850
1851	mutex_lock(&event_mutex);
1852	file = event_file_file(filp);
1853	if (file)
1854		print_event_filter(file, s);
1855	mutex_unlock(&event_mutex);
1856
1857	if (file)
1858		r = simple_read_from_buffer(ubuf, cnt, ppos,
1859					    s->buffer, trace_seq_used(s));
1860
1861	kfree(s);
1862
1863	return r;
1864}
1865
1866static ssize_t
1867event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1868		   loff_t *ppos)
1869{
1870	struct trace_event_file *file;
1871	char *buf;
1872	int err = -ENODEV;
1873
1874	if (cnt >= PAGE_SIZE)
1875		return -EINVAL;
1876
1877	buf = memdup_user_nul(ubuf, cnt);
1878	if (IS_ERR(buf))
1879		return PTR_ERR(buf);
1880
1881	mutex_lock(&event_mutex);
1882	file = event_file_file(filp);
1883	if (file) {
1884		if (file->flags & EVENT_FILE_FL_FREED)
1885			err = -ENODEV;
1886		else
1887			err = apply_event_filter(file, buf);
1888	}
1889	mutex_unlock(&event_mutex);
1890
1891	kfree(buf);
1892	if (err < 0)
1893		return err;
1894
1895	*ppos += cnt;
1896
1897	return cnt;
1898}
1899
1900static LIST_HEAD(event_subsystems);
1901
1902static int subsystem_open(struct inode *inode, struct file *filp)
1903{
1904	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1905	struct trace_array *tr = NULL, *iter_tr;
1906	struct event_subsystem *system = NULL;
1907	int ret;
1908
1909	if (tracing_is_disabled())
1910		return -ENODEV;
1911
1912	/* Make sure the system still exists */
1913	mutex_lock(&event_mutex);
1914	mutex_lock(&trace_types_lock);
1915	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1916		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1917			if (iter_dir == inode->i_private) {
1918				/* Don't open systems with no events */
1919				tr = iter_tr;
1920				dir = iter_dir;
1921				if (dir->nr_events) {
1922					__get_system_dir(dir);
1923					system = dir->subsystem;
1924				}
1925				goto exit_loop;
1926			}
1927		}
1928	}
1929 exit_loop:
1930	mutex_unlock(&trace_types_lock);
1931	mutex_unlock(&event_mutex);
1932
1933	if (!system)
1934		return -ENODEV;
1935
1936	/* Still need to increment the ref count of the system */
1937	if (trace_array_get(tr) < 0) {
1938		put_system(dir);
1939		return -ENODEV;
1940	}
1941
1942	ret = tracing_open_generic(inode, filp);
1943	if (ret < 0) {
1944		trace_array_put(tr);
1945		put_system(dir);
1946	}
1947
1948	return ret;
1949}
1950
1951static int system_tr_open(struct inode *inode, struct file *filp)
1952{
1953	struct trace_subsystem_dir *dir;
1954	struct trace_array *tr = inode->i_private;
1955	int ret;
1956
1957	/* Make a temporary dir that has no system but points to tr */
1958	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1959	if (!dir)
1960		return -ENOMEM;
1961
1962	ret = tracing_open_generic_tr(inode, filp);
1963	if (ret < 0) {
1964		kfree(dir);
1965		return ret;
1966	}
1967	dir->tr = tr;
1968	filp->private_data = dir;
1969
1970	return 0;
1971}
1972
1973static int subsystem_release(struct inode *inode, struct file *file)
1974{
1975	struct trace_subsystem_dir *dir = file->private_data;
1976
1977	trace_array_put(dir->tr);
1978
1979	/*
1980	 * If dir->subsystem is NULL, then this is a temporary
1981	 * descriptor that was made for a trace_array to enable
1982	 * all subsystems.
1983	 */
1984	if (dir->subsystem)
1985		put_system(dir);
1986	else
1987		kfree(dir);
1988
1989	return 0;
1990}
1991
1992static ssize_t
1993subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1994		      loff_t *ppos)
1995{
1996	struct trace_subsystem_dir *dir = filp->private_data;
1997	struct event_subsystem *system = dir->subsystem;
1998	struct trace_seq *s;
1999	int r;
2000
2001	if (*ppos)
2002		return 0;
2003
2004	s = kmalloc(sizeof(*s), GFP_KERNEL);
2005	if (!s)
2006		return -ENOMEM;
2007
2008	trace_seq_init(s);
2009
2010	print_subsystem_event_filter(system, s);
2011	r = simple_read_from_buffer(ubuf, cnt, ppos,
2012				    s->buffer, trace_seq_used(s));
2013
2014	kfree(s);
2015
2016	return r;
2017}
2018
2019static ssize_t
2020subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
2021		       loff_t *ppos)
2022{
2023	struct trace_subsystem_dir *dir = filp->private_data;
2024	char *buf;
2025	int err;
2026
2027	if (cnt >= PAGE_SIZE)
2028		return -EINVAL;
2029
2030	buf = memdup_user_nul(ubuf, cnt);
2031	if (IS_ERR(buf))
2032		return PTR_ERR(buf);
2033
2034	err = apply_subsystem_event_filter(dir, buf);
2035	kfree(buf);
2036	if (err < 0)
2037		return err;
2038
2039	*ppos += cnt;
2040
2041	return cnt;
2042}
2043
2044static ssize_t
2045show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
2046{
2047	struct trace_array *tr = filp->private_data;
2048	struct trace_seq *s;
2049	int r;
2050
2051	if (*ppos)
2052		return 0;
2053
2054	s = kmalloc(sizeof(*s), GFP_KERNEL);
2055	if (!s)
2056		return -ENOMEM;
2057
2058	trace_seq_init(s);
2059
2060	ring_buffer_print_page_header(tr->array_buffer.buffer, s);
2061	r = simple_read_from_buffer(ubuf, cnt, ppos,
2062				    s->buffer, trace_seq_used(s));
2063
2064	kfree(s);
2065
2066	return r;
2067}
2068
2069static ssize_t
2070show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
2071{
2072	struct trace_seq *s;
2073	int r;
2074
2075	if (*ppos)
2076		return 0;
2077
2078	s = kmalloc(sizeof(*s), GFP_KERNEL);
2079	if (!s)
2080		return -ENOMEM;
2081
2082	trace_seq_init(s);
2083
2084	ring_buffer_print_entry_header(s);
2085	r = simple_read_from_buffer(ubuf, cnt, ppos,
2086				    s->buffer, trace_seq_used(s));
2087
2088	kfree(s);
2089
2090	return r;
2091}
2092
2093static void ignore_task_cpu(void *data)
2094{
2095	struct trace_array *tr = data;
2096	struct trace_pid_list *pid_list;
2097	struct trace_pid_list *no_pid_list;
2098
2099	/*
2100	 * This function is called by on_each_cpu() while the
2101	 * event_mutex is held.
2102	 */
2103	pid_list = rcu_dereference_protected(tr->filtered_pids,
2104					     mutex_is_locked(&event_mutex));
2105	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2106					     mutex_is_locked(&event_mutex));
2107
2108	this_cpu_write(tr->array_buffer.data->ignore_pid,
2109		       trace_ignore_this_task(pid_list, no_pid_list, current));
2110}
2111
2112static void register_pid_events(struct trace_array *tr)
2113{
2114	/*
2115	 * Register a probe that is called before all other probes
2116	 * to set ignore_pid if next or prev do not match.
2117	 * Register a probe this is called after all other probes
2118	 * to only keep ignore_pid set if next pid matches.
2119	 */
2120	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
2121					 tr, INT_MAX);
2122	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
2123					 tr, 0);
2124
2125	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
2126					 tr, INT_MAX);
2127	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
2128					 tr, 0);
2129
2130	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
2131					     tr, INT_MAX);
2132	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
2133					     tr, 0);
2134
2135	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
2136					 tr, INT_MAX);
2137	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
2138					 tr, 0);
2139}
2140
2141static ssize_t
2142event_pid_write(struct file *filp, const char __user *ubuf,
2143		size_t cnt, loff_t *ppos, int type)
2144{
2145	struct seq_file *m = filp->private_data;
2146	struct trace_array *tr = m->private;
2147	struct trace_pid_list *filtered_pids = NULL;
2148	struct trace_pid_list *other_pids = NULL;
2149	struct trace_pid_list *pid_list;
2150	struct trace_event_file *file;
2151	ssize_t ret;
2152
2153	if (!cnt)
2154		return 0;
2155
2156	ret = tracing_update_buffers(tr);
2157	if (ret < 0)
2158		return ret;
2159
2160	mutex_lock(&event_mutex);
2161
2162	if (type == TRACE_PIDS) {
2163		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2164							  lockdep_is_held(&event_mutex));
2165		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2166							  lockdep_is_held(&event_mutex));
2167	} else {
2168		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2169							  lockdep_is_held(&event_mutex));
2170		other_pids = rcu_dereference_protected(tr->filtered_pids,
2171							  lockdep_is_held(&event_mutex));
2172	}
2173
2174	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2175	if (ret < 0)
2176		goto out;
2177
2178	if (type == TRACE_PIDS)
2179		rcu_assign_pointer(tr->filtered_pids, pid_list);
2180	else
2181		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2182
2183	list_for_each_entry(file, &tr->events, list) {
2184		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2185	}
2186
2187	if (filtered_pids) {
2188		tracepoint_synchronize_unregister();
2189		trace_pid_list_free(filtered_pids);
2190	} else if (pid_list && !other_pids) {
2191		register_pid_events(tr);
2192	}
2193
2194	/*
2195	 * Ignoring of pids is done at task switch. But we have to
2196	 * check for those tasks that are currently running.
2197	 * Always do this in case a pid was appended or removed.
2198	 */
2199	on_each_cpu(ignore_task_cpu, tr, 1);
2200
2201 out:
2202	mutex_unlock(&event_mutex);
2203
2204	if (ret > 0)
2205		*ppos += ret;
2206
2207	return ret;
2208}
2209
2210static ssize_t
2211ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2212		       size_t cnt, loff_t *ppos)
2213{
2214	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2215}
2216
2217static ssize_t
2218ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2219			size_t cnt, loff_t *ppos)
2220{
2221	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2222}
2223
2224static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2225static int ftrace_event_set_open(struct inode *inode, struct file *file);
2226static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2227static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2228static int ftrace_event_release(struct inode *inode, struct file *file);
2229
2230static const struct seq_operations show_event_seq_ops = {
2231	.start = t_start,
2232	.next = t_next,
2233	.show = t_show,
2234	.stop = t_stop,
2235};
2236
2237static const struct seq_operations show_set_event_seq_ops = {
2238	.start = s_start,
2239	.next = s_next,
2240	.show = t_show,
2241	.stop = t_stop,
2242};
2243
2244static const struct seq_operations show_set_pid_seq_ops = {
2245	.start = p_start,
2246	.next = p_next,
2247	.show = trace_pid_show,
2248	.stop = p_stop,
2249};
2250
2251static const struct seq_operations show_set_no_pid_seq_ops = {
2252	.start = np_start,
2253	.next = np_next,
2254	.show = trace_pid_show,
2255	.stop = p_stop,
2256};
2257
2258static const struct file_operations ftrace_avail_fops = {
2259	.open = ftrace_event_avail_open,
2260	.read = seq_read,
2261	.llseek = seq_lseek,
2262	.release = seq_release,
2263};
2264
2265static const struct file_operations ftrace_set_event_fops = {
2266	.open = ftrace_event_set_open,
2267	.read = seq_read,
2268	.write = ftrace_event_write,
2269	.llseek = seq_lseek,
2270	.release = ftrace_event_release,
2271};
2272
2273static const struct file_operations ftrace_set_event_pid_fops = {
2274	.open = ftrace_event_set_pid_open,
2275	.read = seq_read,
2276	.write = ftrace_event_pid_write,
2277	.llseek = seq_lseek,
2278	.release = ftrace_event_release,
2279};
2280
2281static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2282	.open = ftrace_event_set_npid_open,
2283	.read = seq_read,
2284	.write = ftrace_event_npid_write,
2285	.llseek = seq_lseek,
2286	.release = ftrace_event_release,
2287};
2288
2289static const struct file_operations ftrace_enable_fops = {
2290	.open = tracing_open_file_tr,
2291	.read = event_enable_read,
2292	.write = event_enable_write,
2293	.release = tracing_release_file_tr,
2294	.llseek = default_llseek,
2295};
2296
2297static const struct file_operations ftrace_event_format_fops = {
2298	.open = trace_format_open,
2299	.read = seq_read,
2300	.llseek = seq_lseek,
2301	.release = seq_release,
2302};
2303
2304#ifdef CONFIG_PERF_EVENTS
2305static const struct file_operations ftrace_event_id_fops = {
2306	.read = event_id_read,
2307	.llseek = default_llseek,
2308};
2309#endif
2310
2311static const struct file_operations ftrace_event_filter_fops = {
2312	.open = tracing_open_file_tr,
2313	.read = event_filter_read,
2314	.write = event_filter_write,
2315	.release = tracing_release_file_tr,
2316	.llseek = default_llseek,
2317};
2318
2319static const struct file_operations ftrace_subsystem_filter_fops = {
2320	.open = subsystem_open,
2321	.read = subsystem_filter_read,
2322	.write = subsystem_filter_write,
2323	.llseek = default_llseek,
2324	.release = subsystem_release,
2325};
2326
2327static const struct file_operations ftrace_system_enable_fops = {
2328	.open = subsystem_open,
2329	.read = system_enable_read,
2330	.write = system_enable_write,
2331	.llseek = default_llseek,
2332	.release = subsystem_release,
2333};
2334
2335static const struct file_operations ftrace_tr_enable_fops = {
2336	.open = system_tr_open,
2337	.read = system_enable_read,
2338	.write = system_enable_write,
2339	.llseek = default_llseek,
2340	.release = subsystem_release,
2341};
2342
2343static const struct file_operations ftrace_show_header_page_fops = {
2344	.open = tracing_open_generic_tr,
2345	.read = show_header_page_file,
2346	.llseek = default_llseek,
2347	.release = tracing_release_generic_tr,
2348};
2349
2350static const struct file_operations ftrace_show_header_event_fops = {
2351	.open = tracing_open_generic_tr,
2352	.read = show_header_event_file,
2353	.llseek = default_llseek,
2354	.release = tracing_release_generic_tr,
2355};
2356
2357static int
2358ftrace_event_open(struct inode *inode, struct file *file,
2359		  const struct seq_operations *seq_ops)
2360{
2361	struct seq_file *m;
2362	int ret;
2363
2364	ret = security_locked_down(LOCKDOWN_TRACEFS);
2365	if (ret)
2366		return ret;
2367
2368	ret = seq_open(file, seq_ops);
2369	if (ret < 0)
2370		return ret;
2371	m = file->private_data;
2372	/* copy tr over to seq ops */
2373	m->private = inode->i_private;
2374
2375	return ret;
2376}
2377
2378static int ftrace_event_release(struct inode *inode, struct file *file)
2379{
2380	struct trace_array *tr = inode->i_private;
2381
2382	trace_array_put(tr);
2383
2384	return seq_release(inode, file);
2385}
2386
2387static int
2388ftrace_event_avail_open(struct inode *inode, struct file *file)
2389{
2390	const struct seq_operations *seq_ops = &show_event_seq_ops;
2391
2392	/* Checks for tracefs lockdown */
2393	return ftrace_event_open(inode, file, seq_ops);
2394}
2395
2396static int
2397ftrace_event_set_open(struct inode *inode, struct file *file)
2398{
2399	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2400	struct trace_array *tr = inode->i_private;
2401	int ret;
2402
2403	ret = tracing_check_open_get_tr(tr);
2404	if (ret)
2405		return ret;
2406
2407	if ((file->f_mode & FMODE_WRITE) &&
2408	    (file->f_flags & O_TRUNC))
2409		ftrace_clear_events(tr);
2410
2411	ret = ftrace_event_open(inode, file, seq_ops);
2412	if (ret < 0)
2413		trace_array_put(tr);
2414	return ret;
2415}
2416
2417static int
2418ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2419{
2420	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2421	struct trace_array *tr = inode->i_private;
2422	int ret;
2423
2424	ret = tracing_check_open_get_tr(tr);
2425	if (ret)
2426		return ret;
2427
2428	if ((file->f_mode & FMODE_WRITE) &&
2429	    (file->f_flags & O_TRUNC))
2430		ftrace_clear_event_pids(tr, TRACE_PIDS);
2431
2432	ret = ftrace_event_open(inode, file, seq_ops);
2433	if (ret < 0)
2434		trace_array_put(tr);
2435	return ret;
2436}
2437
2438static int
2439ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2440{
2441	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2442	struct trace_array *tr = inode->i_private;
2443	int ret;
2444
2445	ret = tracing_check_open_get_tr(tr);
2446	if (ret)
2447		return ret;
2448
2449	if ((file->f_mode & FMODE_WRITE) &&
2450	    (file->f_flags & O_TRUNC))
2451		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2452
2453	ret = ftrace_event_open(inode, file, seq_ops);
2454	if (ret < 0)
2455		trace_array_put(tr);
2456	return ret;
2457}
2458
2459static struct event_subsystem *
2460create_new_subsystem(const char *name)
2461{
2462	struct event_subsystem *system;
2463
2464	/* need to create new entry */
2465	system = kmalloc(sizeof(*system), GFP_KERNEL);
2466	if (!system)
2467		return NULL;
2468
2469	system->ref_count = 1;
2470
2471	/* Only allocate if dynamic (kprobes and modules) */
2472	system->name = kstrdup_const(name, GFP_KERNEL);
2473	if (!system->name)
2474		goto out_free;
2475
2476	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2477	if (!system->filter)
2478		goto out_free;
2479
2480	list_add(&system->list, &event_subsystems);
2481
2482	return system;
2483
2484 out_free:
2485	kfree_const(system->name);
2486	kfree(system);
2487	return NULL;
2488}
2489
2490static int system_callback(const char *name, umode_t *mode, void **data,
2491		    const struct file_operations **fops)
2492{
2493	if (strcmp(name, "filter") == 0)
2494		*fops = &ftrace_subsystem_filter_fops;
2495
2496	else if (strcmp(name, "enable") == 0)
2497		*fops = &ftrace_system_enable_fops;
2498
2499	else
2500		return 0;
2501
2502	*mode = TRACE_MODE_WRITE;
2503	return 1;
2504}
2505
2506static struct eventfs_inode *
2507event_subsystem_dir(struct trace_array *tr, const char *name,
2508		    struct trace_event_file *file, struct eventfs_inode *parent)
2509{
2510	struct event_subsystem *system, *iter;
2511	struct trace_subsystem_dir *dir;
2512	struct eventfs_inode *ei;
2513	int nr_entries;
2514	static struct eventfs_entry system_entries[] = {
2515		{
2516			.name		= "filter",
2517			.callback	= system_callback,
2518		},
2519		{
2520			.name		= "enable",
2521			.callback	= system_callback,
2522		}
2523	};
2524
2525	/* First see if we did not already create this dir */
2526	list_for_each_entry(dir, &tr->systems, list) {
2527		system = dir->subsystem;
2528		if (strcmp(system->name, name) == 0) {
2529			dir->nr_events++;
2530			file->system = dir;
2531			return dir->ei;
2532		}
2533	}
2534
2535	/* Now see if the system itself exists. */
2536	system = NULL;
2537	list_for_each_entry(iter, &event_subsystems, list) {
2538		if (strcmp(iter->name, name) == 0) {
2539			system = iter;
2540			break;
2541		}
2542	}
2543
2544	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2545	if (!dir)
2546		goto out_fail;
2547
2548	if (!system) {
2549		system = create_new_subsystem(name);
2550		if (!system)
2551			goto out_free;
2552	} else
2553		__get_system(system);
2554
2555	/* ftrace only has directories no files */
2556	if (strcmp(name, "ftrace") == 0)
2557		nr_entries = 0;
2558	else
2559		nr_entries = ARRAY_SIZE(system_entries);
2560
2561	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2562	if (IS_ERR(ei)) {
2563		pr_warn("Failed to create system directory %s\n", name);
2564		__put_system(system);
2565		goto out_free;
2566	}
2567
2568	dir->ei = ei;
2569	dir->tr = tr;
2570	dir->ref_count = 1;
2571	dir->nr_events = 1;
2572	dir->subsystem = system;
2573	file->system = dir;
2574
2575	list_add(&dir->list, &tr->systems);
2576
2577	return dir->ei;
2578
2579 out_free:
2580	kfree(dir);
2581 out_fail:
2582	/* Only print this message if failed on memory allocation */
2583	if (!dir || !system)
2584		pr_warn("No memory to create event subsystem %s\n", name);
2585	return NULL;
2586}
2587
2588static int
2589event_define_fields(struct trace_event_call *call)
2590{
2591	struct list_head *head;
2592	int ret = 0;
2593
2594	/*
2595	 * Other events may have the same class. Only update
2596	 * the fields if they are not already defined.
2597	 */
2598	head = trace_get_fields(call);
2599	if (list_empty(head)) {
2600		struct trace_event_fields *field = call->class->fields_array;
2601		unsigned int offset = sizeof(struct trace_entry);
2602
2603		for (; field->type; field++) {
2604			if (field->type == TRACE_FUNCTION_TYPE) {
2605				field->define_fields(call);
2606				break;
2607			}
2608
2609			offset = ALIGN(offset, field->align);
2610			ret = trace_define_field_ext(call, field->type, field->name,
2611						 offset, field->size,
2612						 field->is_signed, field->filter_type,
2613						 field->len, field->needs_test);
2614			if (WARN_ON_ONCE(ret)) {
2615				pr_err("error code is %d\n", ret);
2616				break;
2617			}
2618
2619			offset += field->size;
2620		}
2621	}
2622
2623	return ret;
2624}
2625
2626static int event_callback(const char *name, umode_t *mode, void **data,
2627			  const struct file_operations **fops)
2628{
2629	struct trace_event_file *file = *data;
2630	struct trace_event_call *call = file->event_call;
2631
2632	if (strcmp(name, "format") == 0) {
2633		*mode = TRACE_MODE_READ;
2634		*fops = &ftrace_event_format_fops;
 
2635		return 1;
2636	}
2637
2638	/*
2639	 * Only event directories that can be enabled should have
2640	 * triggers or filters, with the exception of the "print"
2641	 * event that can have a "trigger" file.
2642	 */
2643	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2644		if (call->class->reg && strcmp(name, "enable") == 0) {
2645			*mode = TRACE_MODE_WRITE;
2646			*fops = &ftrace_enable_fops;
2647			return 1;
2648		}
2649
2650		if (strcmp(name, "filter") == 0) {
2651			*mode = TRACE_MODE_WRITE;
2652			*fops = &ftrace_event_filter_fops;
2653			return 1;
2654		}
2655	}
2656
2657	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2658	    strcmp(trace_event_name(call), "print") == 0) {
2659		if (strcmp(name, "trigger") == 0) {
2660			*mode = TRACE_MODE_WRITE;
2661			*fops = &event_trigger_fops;
2662			return 1;
2663		}
2664	}
2665
2666#ifdef CONFIG_PERF_EVENTS
2667	if (call->event.type && call->class->reg &&
2668	    strcmp(name, "id") == 0) {
2669		*mode = TRACE_MODE_READ;
2670		*data = (void *)(long)call->event.type;
2671		*fops = &ftrace_event_id_fops;
2672		return 1;
2673	}
2674#endif
2675
2676#ifdef CONFIG_HIST_TRIGGERS
2677	if (strcmp(name, "hist") == 0) {
2678		*mode = TRACE_MODE_READ;
2679		*fops = &event_hist_fops;
2680		return 1;
2681	}
2682#endif
2683#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2684	if (strcmp(name, "hist_debug") == 0) {
2685		*mode = TRACE_MODE_READ;
2686		*fops = &event_hist_debug_fops;
2687		return 1;
2688	}
2689#endif
2690#ifdef CONFIG_TRACE_EVENT_INJECT
2691	if (call->event.type && call->class->reg &&
2692	    strcmp(name, "inject") == 0) {
2693		*mode = 0200;
2694		*fops = &event_inject_fops;
2695		return 1;
2696	}
2697#endif
2698	return 0;
2699}
2700
2701/* The file is incremented on creation and freeing the enable file decrements it */
2702static void event_release(const char *name, void *data)
2703{
2704	struct trace_event_file *file = data;
2705
2706	event_file_put(file);
2707}
2708
2709static int
2710event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2711{
2712	struct trace_event_call *call = file->event_call;
2713	struct trace_array *tr = file->tr;
2714	struct eventfs_inode *e_events;
2715	struct eventfs_inode *ei;
2716	const char *name;
2717	int nr_entries;
2718	int ret;
2719	static struct eventfs_entry event_entries[] = {
2720		{
2721			.name		= "enable",
2722			.callback	= event_callback,
2723			.release	= event_release,
2724		},
2725		{
2726			.name		= "filter",
2727			.callback	= event_callback,
2728		},
2729		{
2730			.name		= "trigger",
2731			.callback	= event_callback,
2732		},
2733		{
2734			.name		= "format",
2735			.callback	= event_callback,
2736		},
2737#ifdef CONFIG_PERF_EVENTS
2738		{
2739			.name		= "id",
2740			.callback	= event_callback,
2741		},
2742#endif
2743#ifdef CONFIG_HIST_TRIGGERS
2744		{
2745			.name		= "hist",
2746			.callback	= event_callback,
2747		},
2748#endif
2749#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2750		{
2751			.name		= "hist_debug",
2752			.callback	= event_callback,
2753		},
2754#endif
2755#ifdef CONFIG_TRACE_EVENT_INJECT
2756		{
2757			.name		= "inject",
2758			.callback	= event_callback,
2759		},
2760#endif
2761	};
2762
2763	/*
2764	 * If the trace point header did not define TRACE_SYSTEM
2765	 * then the system would be called "TRACE_SYSTEM". This should
2766	 * never happen.
2767	 */
2768	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2769		return -ENODEV;
2770
2771	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2772	if (!e_events)
2773		return -ENOMEM;
2774
2775	nr_entries = ARRAY_SIZE(event_entries);
2776
2777	name = trace_event_name(call);
2778	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2779	if (IS_ERR(ei)) {
2780		pr_warn("Could not create tracefs '%s' directory\n", name);
2781		return -1;
2782	}
2783
2784	file->ei = ei;
2785
2786	ret = event_define_fields(call);
2787	if (ret < 0) {
2788		pr_warn("Could not initialize trace point events/%s\n", name);
2789		return ret;
2790	}
2791
2792	/* Gets decremented on freeing of the "enable" file */
2793	event_file_get(file);
2794
2795	return 0;
2796}
2797
2798static void remove_event_from_tracers(struct trace_event_call *call)
2799{
2800	struct trace_event_file *file;
2801	struct trace_array *tr;
2802
2803	do_for_each_event_file_safe(tr, file) {
2804		if (file->event_call != call)
2805			continue;
2806
2807		remove_event_file_dir(file);
2808		/*
2809		 * The do_for_each_event_file_safe() is
2810		 * a double loop. After finding the call for this
2811		 * trace_array, we use break to jump to the next
2812		 * trace_array.
2813		 */
2814		break;
2815	} while_for_each_event_file();
2816}
2817
2818static void event_remove(struct trace_event_call *call)
2819{
2820	struct trace_array *tr;
2821	struct trace_event_file *file;
2822
2823	do_for_each_event_file(tr, file) {
2824		if (file->event_call != call)
2825			continue;
2826
2827		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2828			tr->clear_trace = true;
2829
2830		ftrace_event_enable_disable(file, 0);
2831		/*
2832		 * The do_for_each_event_file() is
2833		 * a double loop. After finding the call for this
2834		 * trace_array, we use break to jump to the next
2835		 * trace_array.
2836		 */
2837		break;
2838	} while_for_each_event_file();
2839
2840	if (call->event.funcs)
2841		__unregister_trace_event(&call->event);
2842	remove_event_from_tracers(call);
2843	list_del(&call->list);
2844}
2845
2846static int event_init(struct trace_event_call *call)
2847{
2848	int ret = 0;
2849	const char *name;
2850
2851	name = trace_event_name(call);
2852	if (WARN_ON(!name))
2853		return -EINVAL;
2854
2855	if (call->class->raw_init) {
2856		ret = call->class->raw_init(call);
2857		if (ret < 0 && ret != -ENOSYS)
2858			pr_warn("Could not initialize trace events/%s\n", name);
2859	}
2860
2861	return ret;
2862}
2863
2864static int
2865__register_event(struct trace_event_call *call, struct module *mod)
2866{
2867	int ret;
2868
2869	ret = event_init(call);
2870	if (ret < 0)
2871		return ret;
2872
2873	list_add(&call->list, &ftrace_events);
2874	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2875		atomic_set(&call->refcnt, 0);
2876	else
2877		call->module = mod;
2878
2879	return 0;
2880}
2881
2882static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2883{
2884	int rlen;
2885	int elen;
2886
2887	/* Find the length of the eval value as a string */
2888	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2889	/* Make sure there's enough room to replace the string with the value */
2890	if (len < elen)
2891		return NULL;
2892
2893	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2894
2895	/* Get the rest of the string of ptr */
2896	rlen = strlen(ptr + len);
2897	memmove(ptr + elen, ptr + len, rlen);
2898	/* Make sure we end the new string */
2899	ptr[elen + rlen] = 0;
2900
2901	return ptr + elen;
2902}
2903
2904static void update_event_printk(struct trace_event_call *call,
2905				struct trace_eval_map *map)
2906{
2907	char *ptr;
2908	int quote = 0;
2909	int len = strlen(map->eval_string);
2910
2911	for (ptr = call->print_fmt; *ptr; ptr++) {
2912		if (*ptr == '\\') {
2913			ptr++;
2914			/* paranoid */
2915			if (!*ptr)
2916				break;
2917			continue;
2918		}
2919		if (*ptr == '"') {
2920			quote ^= 1;
2921			continue;
2922		}
2923		if (quote)
2924			continue;
2925		if (isdigit(*ptr)) {
2926			/* skip numbers */
2927			do {
2928				ptr++;
2929				/* Check for alpha chars like ULL */
2930			} while (isalnum(*ptr));
2931			if (!*ptr)
2932				break;
2933			/*
2934			 * A number must have some kind of delimiter after
2935			 * it, and we can ignore that too.
2936			 */
2937			continue;
2938		}
2939		if (isalpha(*ptr) || *ptr == '_') {
2940			if (strncmp(map->eval_string, ptr, len) == 0 &&
2941			    !isalnum(ptr[len]) && ptr[len] != '_') {
2942				ptr = eval_replace(ptr, map, len);
2943				/* enum/sizeof string smaller than value */
2944				if (WARN_ON_ONCE(!ptr))
2945					return;
2946				/*
2947				 * No need to decrement here, as eval_replace()
2948				 * returns the pointer to the character passed
2949				 * the eval, and two evals can not be placed
2950				 * back to back without something in between.
2951				 * We can skip that something in between.
2952				 */
2953				continue;
2954			}
2955		skip_more:
2956			do {
2957				ptr++;
2958			} while (isalnum(*ptr) || *ptr == '_');
2959			if (!*ptr)
2960				break;
2961			/*
2962			 * If what comes after this variable is a '.' or
2963			 * '->' then we can continue to ignore that string.
2964			 */
2965			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2966				ptr += *ptr == '.' ? 1 : 2;
2967				if (!*ptr)
2968					break;
2969				goto skip_more;
2970			}
2971			/*
2972			 * Once again, we can skip the delimiter that came
2973			 * after the string.
2974			 */
2975			continue;
2976		}
2977	}
2978}
2979
2980static void add_str_to_module(struct module *module, char *str)
2981{
2982	struct module_string *modstr;
2983
2984	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2985
2986	/*
2987	 * If we failed to allocate memory here, then we'll just
2988	 * let the str memory leak when the module is removed.
2989	 * If this fails to allocate, there's worse problems than
2990	 * a leaked string on module removal.
2991	 */
2992	if (WARN_ON_ONCE(!modstr))
2993		return;
2994
2995	modstr->module = module;
2996	modstr->str = str;
2997
2998	list_add(&modstr->next, &module_strings);
2999}
3000
3001static void update_event_fields(struct trace_event_call *call,
3002				struct trace_eval_map *map)
3003{
3004	struct ftrace_event_field *field;
3005	struct list_head *head;
3006	char *ptr;
3007	char *str;
3008	int len = strlen(map->eval_string);
3009
3010	/* Dynamic events should never have field maps */
3011	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
3012		return;
3013
3014	head = trace_get_fields(call);
3015	list_for_each_entry(field, head, link) {
3016		ptr = strchr(field->type, '[');
3017		if (!ptr)
3018			continue;
3019		ptr++;
3020
3021		if (!isalpha(*ptr) && *ptr != '_')
3022			continue;
3023
3024		if (strncmp(map->eval_string, ptr, len) != 0)
3025			continue;
3026
3027		str = kstrdup(field->type, GFP_KERNEL);
3028		if (WARN_ON_ONCE(!str))
3029			return;
3030		ptr = str + (ptr - field->type);
3031		ptr = eval_replace(ptr, map, len);
3032		/* enum/sizeof string smaller than value */
3033		if (WARN_ON_ONCE(!ptr)) {
3034			kfree(str);
3035			continue;
3036		}
3037
3038		/*
3039		 * If the event is part of a module, then we need to free the string
3040		 * when the module is removed. Otherwise, it will stay allocated
3041		 * until a reboot.
3042		 */
3043		if (call->module)
3044			add_str_to_module(call->module, str);
3045
3046		field->type = str;
3047	}
3048}
3049
3050void trace_event_eval_update(struct trace_eval_map **map, int len)
3051{
3052	struct trace_event_call *call, *p;
3053	const char *last_system = NULL;
3054	bool first = false;
3055	int last_i;
3056	int i;
3057
3058	down_write(&trace_event_sem);
3059	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3060		/* events are usually grouped together with systems */
3061		if (!last_system || call->class->system != last_system) {
3062			first = true;
3063			last_i = 0;
3064			last_system = call->class->system;
3065		}
3066
3067		/*
3068		 * Since calls are grouped by systems, the likelihood that the
3069		 * next call in the iteration belongs to the same system as the
3070		 * previous call is high. As an optimization, we skip searching
3071		 * for a map[] that matches the call's system if the last call
3072		 * was from the same system. That's what last_i is for. If the
3073		 * call has the same system as the previous call, then last_i
3074		 * will be the index of the first map[] that has a matching
3075		 * system.
3076		 */
3077		for (i = last_i; i < len; i++) {
3078			if (call->class->system == map[i]->system) {
3079				/* Save the first system if need be */
3080				if (first) {
3081					last_i = i;
3082					first = false;
3083				}
3084				update_event_printk(call, map[i]);
3085				update_event_fields(call, map[i]);
3086			}
3087		}
3088		cond_resched();
3089	}
3090	up_write(&trace_event_sem);
3091}
3092
3093static bool event_in_systems(struct trace_event_call *call,
3094			     const char *systems)
3095{
3096	const char *system;
3097	const char *p;
3098
3099	if (!systems)
3100		return true;
3101
3102	system = call->class->system;
3103	p = strstr(systems, system);
3104	if (!p)
3105		return false;
3106
3107	if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
3108		return false;
3109
3110	p += strlen(system);
3111	return !*p || isspace(*p) || *p == ',';
3112}
3113
3114static struct trace_event_file *
3115trace_create_new_event(struct trace_event_call *call,
3116		       struct trace_array *tr)
3117{
3118	struct trace_pid_list *no_pid_list;
3119	struct trace_pid_list *pid_list;
3120	struct trace_event_file *file;
3121	unsigned int first;
3122
3123	if (!event_in_systems(call, tr->system_names))
3124		return NULL;
3125
3126	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
3127	if (!file)
3128		return ERR_PTR(-ENOMEM);
3129
3130	pid_list = rcu_dereference_protected(tr->filtered_pids,
3131					     lockdep_is_held(&event_mutex));
3132	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
3133					     lockdep_is_held(&event_mutex));
3134
3135	if (!trace_pid_list_first(pid_list, &first) ||
3136	    !trace_pid_list_first(no_pid_list, &first))
3137		file->flags |= EVENT_FILE_FL_PID_FILTER;
3138
3139	file->event_call = call;
3140	file->tr = tr;
3141	atomic_set(&file->sm_ref, 0);
3142	atomic_set(&file->tm_ref, 0);
3143	INIT_LIST_HEAD(&file->triggers);
3144	list_add(&file->list, &tr->events);
3145	refcount_set(&file->ref, 1);
3146
3147	return file;
3148}
3149
3150#define MAX_BOOT_TRIGGERS 32
3151
3152static struct boot_triggers {
3153	const char		*event;
3154	char			*trigger;
3155} bootup_triggers[MAX_BOOT_TRIGGERS];
3156
3157static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3158static int nr_boot_triggers;
3159
3160static __init int setup_trace_triggers(char *str)
3161{
3162	char *trigger;
3163	char *buf;
3164	int i;
3165
3166	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3167	trace_set_ring_buffer_expanded(NULL);
3168	disable_tracing_selftest("running event triggers");
3169
3170	buf = bootup_trigger_buf;
3171	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3172		trigger = strsep(&buf, ",");
3173		if (!trigger)
3174			break;
3175		bootup_triggers[i].event = strsep(&trigger, ".");
3176		bootup_triggers[i].trigger = trigger;
3177		if (!bootup_triggers[i].trigger)
3178			break;
3179	}
3180
3181	nr_boot_triggers = i;
3182	return 1;
3183}
3184__setup("trace_trigger=", setup_trace_triggers);
3185
3186/* Add an event to a trace directory */
3187static int
3188__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3189{
3190	struct trace_event_file *file;
3191
3192	file = trace_create_new_event(call, tr);
3193	/*
3194	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3195	 * allocation, or NULL if the event is not part of the tr->system_names.
3196	 * When the event is not part of the tr->system_names, return zero, not
3197	 * an error.
3198	 */
3199	if (!file)
3200		return 0;
3201
3202	if (IS_ERR(file))
3203		return PTR_ERR(file);
3204
3205	if (eventdir_initialized)
3206		return event_create_dir(tr->event_dir, file);
3207	else
3208		return event_define_fields(call);
3209}
3210
3211static void trace_early_triggers(struct trace_event_file *file, const char *name)
3212{
3213	int ret;
3214	int i;
3215
3216	for (i = 0; i < nr_boot_triggers; i++) {
3217		if (strcmp(name, bootup_triggers[i].event))
3218			continue;
3219		mutex_lock(&event_mutex);
3220		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3221		mutex_unlock(&event_mutex);
3222		if (ret)
3223			pr_err("Failed to register trigger '%s' on event %s\n",
3224			       bootup_triggers[i].trigger,
3225			       bootup_triggers[i].event);
3226	}
3227}
3228
3229/*
3230 * Just create a descriptor for early init. A descriptor is required
3231 * for enabling events at boot. We want to enable events before
3232 * the filesystem is initialized.
3233 */
3234static int
3235__trace_early_add_new_event(struct trace_event_call *call,
3236			    struct trace_array *tr)
3237{
3238	struct trace_event_file *file;
3239	int ret;
3240
3241	file = trace_create_new_event(call, tr);
3242	/*
3243	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3244	 * allocation, or NULL if the event is not part of the tr->system_names.
3245	 * When the event is not part of the tr->system_names, return zero, not
3246	 * an error.
3247	 */
3248	if (!file)
3249		return 0;
3250
3251	if (IS_ERR(file))
3252		return PTR_ERR(file);
3253
3254	ret = event_define_fields(call);
3255	if (ret)
3256		return ret;
3257
3258	trace_early_triggers(file, trace_event_name(call));
3259
3260	return 0;
3261}
3262
3263struct ftrace_module_file_ops;
3264static void __add_event_to_tracers(struct trace_event_call *call);
3265
3266/* Add an additional event_call dynamically */
3267int trace_add_event_call(struct trace_event_call *call)
3268{
3269	int ret;
3270	lockdep_assert_held(&event_mutex);
3271
3272	mutex_lock(&trace_types_lock);
3273
3274	ret = __register_event(call, NULL);
3275	if (ret >= 0)
3276		__add_event_to_tracers(call);
3277
3278	mutex_unlock(&trace_types_lock);
3279	return ret;
3280}
3281EXPORT_SYMBOL_GPL(trace_add_event_call);
3282
3283/*
3284 * Must be called under locking of trace_types_lock, event_mutex and
3285 * trace_event_sem.
3286 */
3287static void __trace_remove_event_call(struct trace_event_call *call)
3288{
3289	event_remove(call);
3290	trace_destroy_fields(call);
 
 
3291}
3292
3293static int probe_remove_event_call(struct trace_event_call *call)
3294{
3295	struct trace_array *tr;
3296	struct trace_event_file *file;
3297
3298#ifdef CONFIG_PERF_EVENTS
3299	if (call->perf_refcount)
3300		return -EBUSY;
3301#endif
3302	do_for_each_event_file(tr, file) {
3303		if (file->event_call != call)
3304			continue;
3305		/*
3306		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3307		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3308		 * TRACE_REG_UNREGISTER.
3309		 */
3310		if (file->flags & EVENT_FILE_FL_ENABLED)
3311			goto busy;
3312
3313		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3314			tr->clear_trace = true;
3315		/*
3316		 * The do_for_each_event_file_safe() is
3317		 * a double loop. After finding the call for this
3318		 * trace_array, we use break to jump to the next
3319		 * trace_array.
3320		 */
3321		break;
3322	} while_for_each_event_file();
3323
3324	__trace_remove_event_call(call);
3325
3326	return 0;
3327 busy:
3328	/* No need to clear the trace now */
3329	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3330		tr->clear_trace = false;
3331	}
3332	return -EBUSY;
3333}
3334
3335/* Remove an event_call */
3336int trace_remove_event_call(struct trace_event_call *call)
3337{
3338	int ret;
3339
3340	lockdep_assert_held(&event_mutex);
3341
3342	mutex_lock(&trace_types_lock);
3343	down_write(&trace_event_sem);
3344	ret = probe_remove_event_call(call);
3345	up_write(&trace_event_sem);
3346	mutex_unlock(&trace_types_lock);
3347
3348	return ret;
3349}
3350EXPORT_SYMBOL_GPL(trace_remove_event_call);
3351
3352#define for_each_event(event, start, end)			\
3353	for (event = start;					\
3354	     (unsigned long)event < (unsigned long)end;		\
3355	     event++)
3356
3357#ifdef CONFIG_MODULES
3358
3359static void trace_module_add_events(struct module *mod)
3360{
3361	struct trace_event_call **call, **start, **end;
3362
3363	if (!mod->num_trace_events)
3364		return;
3365
3366	/* Don't add infrastructure for mods without tracepoints */
3367	if (trace_module_has_bad_taint(mod)) {
3368		pr_err("%s: module has bad taint, not creating trace events\n",
3369		       mod->name);
3370		return;
3371	}
3372
3373	start = mod->trace_events;
3374	end = mod->trace_events + mod->num_trace_events;
3375
3376	for_each_event(call, start, end) {
3377		__register_event(*call, mod);
3378		__add_event_to_tracers(*call);
3379	}
3380}
3381
3382static void trace_module_remove_events(struct module *mod)
3383{
3384	struct trace_event_call *call, *p;
3385	struct module_string *modstr, *m;
3386
3387	down_write(&trace_event_sem);
3388	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3389		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3390			continue;
3391		if (call->module == mod)
3392			__trace_remove_event_call(call);
3393	}
3394	/* Check for any strings allocade for this module */
3395	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3396		if (modstr->module != mod)
3397			continue;
3398		list_del(&modstr->next);
3399		kfree(modstr->str);
3400		kfree(modstr);
3401	}
3402	up_write(&trace_event_sem);
3403
3404	/*
3405	 * It is safest to reset the ring buffer if the module being unloaded
3406	 * registered any events that were used. The only worry is if
3407	 * a new module gets loaded, and takes on the same id as the events
3408	 * of this module. When printing out the buffer, traced events left
3409	 * over from this module may be passed to the new module events and
3410	 * unexpected results may occur.
3411	 */
3412	tracing_reset_all_online_cpus_unlocked();
3413}
3414
3415static int trace_module_notify(struct notifier_block *self,
3416			       unsigned long val, void *data)
3417{
3418	struct module *mod = data;
3419
3420	mutex_lock(&event_mutex);
3421	mutex_lock(&trace_types_lock);
3422	switch (val) {
3423	case MODULE_STATE_COMING:
3424		trace_module_add_events(mod);
3425		break;
3426	case MODULE_STATE_GOING:
3427		trace_module_remove_events(mod);
3428		break;
3429	}
3430	mutex_unlock(&trace_types_lock);
3431	mutex_unlock(&event_mutex);
3432
3433	return NOTIFY_OK;
3434}
3435
3436static struct notifier_block trace_module_nb = {
3437	.notifier_call = trace_module_notify,
3438	.priority = 1, /* higher than trace.c module notify */
3439};
3440#endif /* CONFIG_MODULES */
3441
3442/* Create a new event directory structure for a trace directory. */
3443static void
3444__trace_add_event_dirs(struct trace_array *tr)
3445{
3446	struct trace_event_call *call;
3447	int ret;
3448
3449	list_for_each_entry(call, &ftrace_events, list) {
3450		ret = __trace_add_new_event(call, tr);
3451		if (ret < 0)
3452			pr_warn("Could not create directory for event %s\n",
3453				trace_event_name(call));
3454	}
3455}
3456
3457/* Returns any file that matches the system and event */
3458struct trace_event_file *
3459__find_event_file(struct trace_array *tr, const char *system, const char *event)
3460{
3461	struct trace_event_file *file;
3462	struct trace_event_call *call;
3463	const char *name;
3464
3465	list_for_each_entry(file, &tr->events, list) {
3466
3467		call = file->event_call;
3468		name = trace_event_name(call);
3469
3470		if (!name || !call->class)
3471			continue;
3472
3473		if (strcmp(event, name) == 0 &&
3474		    strcmp(system, call->class->system) == 0)
3475			return file;
3476	}
3477	return NULL;
3478}
3479
3480/* Returns valid trace event files that match system and event */
3481struct trace_event_file *
3482find_event_file(struct trace_array *tr, const char *system, const char *event)
3483{
3484	struct trace_event_file *file;
3485
3486	file = __find_event_file(tr, system, event);
3487	if (!file || !file->event_call->class->reg ||
3488	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3489		return NULL;
3490
3491	return file;
3492}
3493
3494/**
3495 * trace_get_event_file - Find and return a trace event file
3496 * @instance: The name of the trace instance containing the event
3497 * @system: The name of the system containing the event
3498 * @event: The name of the event
3499 *
3500 * Return a trace event file given the trace instance name, trace
3501 * system, and trace event name.  If the instance name is NULL, it
3502 * refers to the top-level trace array.
3503 *
3504 * This function will look it up and return it if found, after calling
3505 * trace_array_get() to prevent the instance from going away, and
3506 * increment the event's module refcount to prevent it from being
3507 * removed.
3508 *
3509 * To release the file, call trace_put_event_file(), which will call
3510 * trace_array_put() and decrement the event's module refcount.
3511 *
3512 * Return: The trace event on success, ERR_PTR otherwise.
3513 */
3514struct trace_event_file *trace_get_event_file(const char *instance,
3515					      const char *system,
3516					      const char *event)
3517{
3518	struct trace_array *tr = top_trace_array();
3519	struct trace_event_file *file = NULL;
3520	int ret = -EINVAL;
3521
3522	if (instance) {
3523		tr = trace_array_find_get(instance);
3524		if (!tr)
3525			return ERR_PTR(-ENOENT);
3526	} else {
3527		ret = trace_array_get(tr);
3528		if (ret)
3529			return ERR_PTR(ret);
3530	}
3531
3532	mutex_lock(&event_mutex);
3533
3534	file = find_event_file(tr, system, event);
3535	if (!file) {
3536		trace_array_put(tr);
3537		ret = -EINVAL;
3538		goto out;
3539	}
3540
3541	/* Don't let event modules unload while in use */
3542	ret = trace_event_try_get_ref(file->event_call);
3543	if (!ret) {
3544		trace_array_put(tr);
3545		ret = -EBUSY;
3546		goto out;
3547	}
3548
3549	ret = 0;
3550 out:
3551	mutex_unlock(&event_mutex);
3552
3553	if (ret)
3554		file = ERR_PTR(ret);
3555
3556	return file;
3557}
3558EXPORT_SYMBOL_GPL(trace_get_event_file);
3559
3560/**
3561 * trace_put_event_file - Release a file from trace_get_event_file()
3562 * @file: The trace event file
3563 *
3564 * If a file was retrieved using trace_get_event_file(), this should
3565 * be called when it's no longer needed.  It will cancel the previous
3566 * trace_array_get() called by that function, and decrement the
3567 * event's module refcount.
3568 */
3569void trace_put_event_file(struct trace_event_file *file)
3570{
3571	mutex_lock(&event_mutex);
3572	trace_event_put_ref(file->event_call);
3573	mutex_unlock(&event_mutex);
3574
3575	trace_array_put(file->tr);
3576}
3577EXPORT_SYMBOL_GPL(trace_put_event_file);
3578
3579#ifdef CONFIG_DYNAMIC_FTRACE
3580
3581/* Avoid typos */
3582#define ENABLE_EVENT_STR	"enable_event"
3583#define DISABLE_EVENT_STR	"disable_event"
3584
3585struct event_probe_data {
3586	struct trace_event_file	*file;
3587	unsigned long			count;
3588	int				ref;
3589	bool				enable;
3590};
3591
3592static void update_event_probe(struct event_probe_data *data)
3593{
3594	if (data->enable)
3595		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3596	else
3597		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3598}
3599
3600static void
3601event_enable_probe(unsigned long ip, unsigned long parent_ip,
3602		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3603		   void *data)
3604{
3605	struct ftrace_func_mapper *mapper = data;
3606	struct event_probe_data *edata;
3607	void **pdata;
3608
3609	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3610	if (!pdata || !*pdata)
3611		return;
3612
3613	edata = *pdata;
3614	update_event_probe(edata);
3615}
3616
3617static void
3618event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3619			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3620			 void *data)
3621{
3622	struct ftrace_func_mapper *mapper = data;
3623	struct event_probe_data *edata;
3624	void **pdata;
3625
3626	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3627	if (!pdata || !*pdata)
3628		return;
3629
3630	edata = *pdata;
3631
3632	if (!edata->count)
3633		return;
3634
3635	/* Skip if the event is in a state we want to switch to */
3636	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3637		return;
3638
3639	if (edata->count != -1)
3640		(edata->count)--;
3641
3642	update_event_probe(edata);
3643}
3644
3645static int
3646event_enable_print(struct seq_file *m, unsigned long ip,
3647		   struct ftrace_probe_ops *ops, void *data)
3648{
3649	struct ftrace_func_mapper *mapper = data;
3650	struct event_probe_data *edata;
3651	void **pdata;
3652
3653	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3654
3655	if (WARN_ON_ONCE(!pdata || !*pdata))
3656		return 0;
3657
3658	edata = *pdata;
3659
3660	seq_printf(m, "%ps:", (void *)ip);
3661
3662	seq_printf(m, "%s:%s:%s",
3663		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3664		   edata->file->event_call->class->system,
3665		   trace_event_name(edata->file->event_call));
3666
3667	if (edata->count == -1)
3668		seq_puts(m, ":unlimited\n");
3669	else
3670		seq_printf(m, ":count=%ld\n", edata->count);
3671
3672	return 0;
3673}
3674
3675static int
3676event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3677		  unsigned long ip, void *init_data, void **data)
3678{
3679	struct ftrace_func_mapper *mapper = *data;
3680	struct event_probe_data *edata = init_data;
3681	int ret;
3682
3683	if (!mapper) {
3684		mapper = allocate_ftrace_func_mapper();
3685		if (!mapper)
3686			return -ENODEV;
3687		*data = mapper;
3688	}
3689
3690	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3691	if (ret < 0)
3692		return ret;
3693
3694	edata->ref++;
3695
3696	return 0;
3697}
3698
3699static int free_probe_data(void *data)
3700{
3701	struct event_probe_data *edata = data;
3702
3703	edata->ref--;
3704	if (!edata->ref) {
3705		/* Remove the SOFT_MODE flag */
3706		__ftrace_event_enable_disable(edata->file, 0, 1);
3707		trace_event_put_ref(edata->file->event_call);
3708		kfree(edata);
3709	}
3710	return 0;
3711}
3712
3713static void
3714event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3715		  unsigned long ip, void *data)
3716{
3717	struct ftrace_func_mapper *mapper = data;
3718	struct event_probe_data *edata;
3719
3720	if (!ip) {
3721		if (!mapper)
3722			return;
3723		free_ftrace_func_mapper(mapper, free_probe_data);
3724		return;
3725	}
3726
3727	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3728
3729	if (WARN_ON_ONCE(!edata))
3730		return;
3731
3732	if (WARN_ON_ONCE(edata->ref <= 0))
3733		return;
3734
3735	free_probe_data(edata);
3736}
3737
3738static struct ftrace_probe_ops event_enable_probe_ops = {
3739	.func			= event_enable_probe,
3740	.print			= event_enable_print,
3741	.init			= event_enable_init,
3742	.free			= event_enable_free,
3743};
3744
3745static struct ftrace_probe_ops event_enable_count_probe_ops = {
3746	.func			= event_enable_count_probe,
3747	.print			= event_enable_print,
3748	.init			= event_enable_init,
3749	.free			= event_enable_free,
3750};
3751
3752static struct ftrace_probe_ops event_disable_probe_ops = {
3753	.func			= event_enable_probe,
3754	.print			= event_enable_print,
3755	.init			= event_enable_init,
3756	.free			= event_enable_free,
3757};
3758
3759static struct ftrace_probe_ops event_disable_count_probe_ops = {
3760	.func			= event_enable_count_probe,
3761	.print			= event_enable_print,
3762	.init			= event_enable_init,
3763	.free			= event_enable_free,
3764};
3765
3766static int
3767event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3768		  char *glob, char *cmd, char *param, int enabled)
3769{
3770	struct trace_event_file *file;
3771	struct ftrace_probe_ops *ops;
3772	struct event_probe_data *data;
3773	const char *system;
3774	const char *event;
3775	char *number;
3776	bool enable;
3777	int ret;
3778
3779	if (!tr)
3780		return -ENODEV;
3781
3782	/* hash funcs only work with set_ftrace_filter */
3783	if (!enabled || !param)
3784		return -EINVAL;
3785
3786	system = strsep(&param, ":");
3787	if (!param)
3788		return -EINVAL;
3789
3790	event = strsep(&param, ":");
3791
3792	mutex_lock(&event_mutex);
3793
3794	ret = -EINVAL;
3795	file = find_event_file(tr, system, event);
3796	if (!file)
3797		goto out;
3798
3799	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3800
3801	if (enable)
3802		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3803	else
3804		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3805
3806	if (glob[0] == '!') {
3807		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3808		goto out;
3809	}
3810
3811	ret = -ENOMEM;
3812
3813	data = kzalloc(sizeof(*data), GFP_KERNEL);
3814	if (!data)
3815		goto out;
3816
3817	data->enable = enable;
3818	data->count = -1;
3819	data->file = file;
3820
3821	if (!param)
3822		goto out_reg;
3823
3824	number = strsep(&param, ":");
3825
3826	ret = -EINVAL;
3827	if (!strlen(number))
3828		goto out_free;
3829
3830	/*
3831	 * We use the callback data field (which is a pointer)
3832	 * as our counter.
3833	 */
3834	ret = kstrtoul(number, 0, &data->count);
3835	if (ret)
3836		goto out_free;
3837
3838 out_reg:
3839	/* Don't let event modules unload while probe registered */
3840	ret = trace_event_try_get_ref(file->event_call);
3841	if (!ret) {
3842		ret = -EBUSY;
3843		goto out_free;
3844	}
3845
3846	ret = __ftrace_event_enable_disable(file, 1, 1);
3847	if (ret < 0)
3848		goto out_put;
3849
3850	ret = register_ftrace_function_probe(glob, tr, ops, data);
3851	/*
3852	 * The above returns on success the # of functions enabled,
3853	 * but if it didn't find any functions it returns zero.
3854	 * Consider no functions a failure too.
3855	 */
3856	if (!ret) {
3857		ret = -ENOENT;
3858		goto out_disable;
3859	} else if (ret < 0)
3860		goto out_disable;
3861	/* Just return zero, not the number of enabled functions */
3862	ret = 0;
3863 out:
3864	mutex_unlock(&event_mutex);
3865	return ret;
3866
3867 out_disable:
3868	__ftrace_event_enable_disable(file, 0, 1);
3869 out_put:
3870	trace_event_put_ref(file->event_call);
3871 out_free:
3872	kfree(data);
3873	goto out;
3874}
3875
3876static struct ftrace_func_command event_enable_cmd = {
3877	.name			= ENABLE_EVENT_STR,
3878	.func			= event_enable_func,
3879};
3880
3881static struct ftrace_func_command event_disable_cmd = {
3882	.name			= DISABLE_EVENT_STR,
3883	.func			= event_enable_func,
3884};
3885
3886static __init int register_event_cmds(void)
3887{
3888	int ret;
3889
3890	ret = register_ftrace_command(&event_enable_cmd);
3891	if (WARN_ON(ret < 0))
3892		return ret;
3893	ret = register_ftrace_command(&event_disable_cmd);
3894	if (WARN_ON(ret < 0))
3895		unregister_ftrace_command(&event_enable_cmd);
3896	return ret;
3897}
3898#else
3899static inline int register_event_cmds(void) { return 0; }
3900#endif /* CONFIG_DYNAMIC_FTRACE */
3901
3902/*
3903 * The top level array and trace arrays created by boot-time tracing
3904 * have already had its trace_event_file descriptors created in order
3905 * to allow for early events to be recorded.
3906 * This function is called after the tracefs has been initialized,
3907 * and we now have to create the files associated to the events.
3908 */
3909static void __trace_early_add_event_dirs(struct trace_array *tr)
3910{
3911	struct trace_event_file *file;
3912	int ret;
3913
3914
3915	list_for_each_entry(file, &tr->events, list) {
3916		ret = event_create_dir(tr->event_dir, file);
3917		if (ret < 0)
3918			pr_warn("Could not create directory for event %s\n",
3919				trace_event_name(file->event_call));
3920	}
3921}
3922
3923/*
3924 * For early boot up, the top trace array and the trace arrays created
3925 * by boot-time tracing require to have a list of events that can be
3926 * enabled. This must be done before the filesystem is set up in order
3927 * to allow events to be traced early.
3928 */
3929void __trace_early_add_events(struct trace_array *tr)
3930{
3931	struct trace_event_call *call;
3932	int ret;
3933
3934	list_for_each_entry(call, &ftrace_events, list) {
3935		/* Early boot up should not have any modules loaded */
3936		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3937		    WARN_ON_ONCE(call->module))
3938			continue;
3939
3940		ret = __trace_early_add_new_event(call, tr);
3941		if (ret < 0)
3942			pr_warn("Could not create early event %s\n",
3943				trace_event_name(call));
3944	}
3945}
3946
3947/* Remove the event directory structure for a trace directory. */
3948static void
3949__trace_remove_event_dirs(struct trace_array *tr)
3950{
3951	struct trace_event_file *file, *next;
3952
3953	list_for_each_entry_safe(file, next, &tr->events, list)
3954		remove_event_file_dir(file);
3955}
3956
3957static void __add_event_to_tracers(struct trace_event_call *call)
3958{
3959	struct trace_array *tr;
3960
3961	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3962		__trace_add_new_event(call, tr);
3963}
3964
3965extern struct trace_event_call *__start_ftrace_events[];
3966extern struct trace_event_call *__stop_ftrace_events[];
3967
3968static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3969
3970static __init int setup_trace_event(char *str)
3971{
3972	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3973	trace_set_ring_buffer_expanded(NULL);
3974	disable_tracing_selftest("running event tracing");
3975
3976	return 1;
3977}
3978__setup("trace_event=", setup_trace_event);
3979
3980static int events_callback(const char *name, umode_t *mode, void **data,
3981			   const struct file_operations **fops)
3982{
3983	if (strcmp(name, "enable") == 0) {
3984		*mode = TRACE_MODE_WRITE;
3985		*fops = &ftrace_tr_enable_fops;
3986		return 1;
3987	}
3988
3989	if (strcmp(name, "header_page") == 0) {
3990		*mode = TRACE_MODE_READ;
3991		*fops = &ftrace_show_header_page_fops;
3992
3993	} else if (strcmp(name, "header_event") == 0) {
3994		*mode = TRACE_MODE_READ;
3995		*fops = &ftrace_show_header_event_fops;
3996	} else
3997		return 0;
3998
3999	return 1;
4000}
4001
4002/* Expects to have event_mutex held when called */
4003static int
4004create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
4005{
4006	struct eventfs_inode *e_events;
4007	struct dentry *entry;
4008	int nr_entries;
4009	static struct eventfs_entry events_entries[] = {
4010		{
4011			.name		= "enable",
4012			.callback	= events_callback,
4013		},
4014		{
4015			.name		= "header_page",
4016			.callback	= events_callback,
4017		},
4018		{
4019			.name		= "header_event",
4020			.callback	= events_callback,
4021		},
4022	};
4023
4024	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
4025				  tr, &ftrace_set_event_fops);
4026	if (!entry)
4027		return -ENOMEM;
4028
4029	nr_entries = ARRAY_SIZE(events_entries);
4030
4031	e_events = eventfs_create_events_dir("events", parent, events_entries,
4032					     nr_entries, tr);
4033	if (IS_ERR(e_events)) {
4034		pr_warn("Could not create tracefs 'events' directory\n");
4035		return -ENOMEM;
4036	}
4037
4038	/* There are not as crucial, just warn if they are not created */
4039
4040	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
4041			  tr, &ftrace_set_event_pid_fops);
4042
4043	trace_create_file("set_event_notrace_pid",
4044			  TRACE_MODE_WRITE, parent, tr,
4045			  &ftrace_set_event_notrace_pid_fops);
4046
4047	tr->event_dir = e_events;
4048
4049	return 0;
4050}
4051
4052/**
4053 * event_trace_add_tracer - add a instance of a trace_array to events
4054 * @parent: The parent dentry to place the files/directories for events in
4055 * @tr: The trace array associated with these events
4056 *
4057 * When a new instance is created, it needs to set up its events
4058 * directory, as well as other files associated with events. It also
4059 * creates the event hierarchy in the @parent/events directory.
4060 *
4061 * Returns 0 on success.
4062 *
4063 * Must be called with event_mutex held.
4064 */
4065int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
4066{
4067	int ret;
4068
4069	lockdep_assert_held(&event_mutex);
4070
4071	ret = create_event_toplevel_files(parent, tr);
4072	if (ret)
4073		goto out;
4074
4075	down_write(&trace_event_sem);
4076	/* If tr already has the event list, it is initialized in early boot. */
4077	if (unlikely(!list_empty(&tr->events)))
4078		__trace_early_add_event_dirs(tr);
4079	else
4080		__trace_add_event_dirs(tr);
4081	up_write(&trace_event_sem);
4082
4083 out:
4084	return ret;
4085}
4086
4087/*
4088 * The top trace array already had its file descriptors created.
4089 * Now the files themselves need to be created.
4090 */
4091static __init int
4092early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
4093{
4094	int ret;
4095
4096	mutex_lock(&event_mutex);
4097
4098	ret = create_event_toplevel_files(parent, tr);
4099	if (ret)
4100		goto out_unlock;
4101
4102	down_write(&trace_event_sem);
4103	__trace_early_add_event_dirs(tr);
4104	up_write(&trace_event_sem);
4105
4106 out_unlock:
4107	mutex_unlock(&event_mutex);
4108
4109	return ret;
4110}
4111
4112/* Must be called with event_mutex held */
4113int event_trace_del_tracer(struct trace_array *tr)
4114{
4115	lockdep_assert_held(&event_mutex);
4116
4117	/* Disable any event triggers and associated soft-disabled events */
4118	clear_event_triggers(tr);
4119
4120	/* Clear the pid list */
4121	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
4122
4123	/* Disable any running events */
4124	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
4125
4126	/* Make sure no more events are being executed */
4127	tracepoint_synchronize_unregister();
4128
4129	down_write(&trace_event_sem);
4130	__trace_remove_event_dirs(tr);
4131	eventfs_remove_events_dir(tr->event_dir);
4132	up_write(&trace_event_sem);
4133
4134	tr->event_dir = NULL;
4135
4136	return 0;
4137}
4138
4139static __init int event_trace_memsetup(void)
4140{
4141	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
4142	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
4143	return 0;
4144}
4145
4146__init void
4147early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4148{
4149	char *token;
4150	int ret;
4151
4152	while (true) {
4153		token = strsep(&buf, ",");
4154
4155		if (!token)
4156			break;
4157
4158		if (*token) {
4159			/* Restarting syscalls requires that we stop them first */
4160			if (disable_first)
4161				ftrace_set_clr_event(tr, token, 0);
4162
4163			ret = ftrace_set_clr_event(tr, token, 1);
4164			if (ret)
4165				pr_warn("Failed to enable trace event: %s\n", token);
4166		}
4167
4168		/* Put back the comma to allow this to be called again */
4169		if (buf)
4170			*(buf - 1) = ',';
4171	}
4172}
4173
4174static __init int event_trace_enable(void)
4175{
4176	struct trace_array *tr = top_trace_array();
4177	struct trace_event_call **iter, *call;
4178	int ret;
4179
4180	if (!tr)
4181		return -ENODEV;
4182
4183	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4184
4185		call = *iter;
4186		ret = event_init(call);
4187		if (!ret)
4188			list_add(&call->list, &ftrace_events);
4189	}
4190
4191	register_trigger_cmds();
4192
4193	/*
4194	 * We need the top trace array to have a working set of trace
4195	 * points at early init, before the debug files and directories
4196	 * are created. Create the file entries now, and attach them
4197	 * to the actual file dentries later.
4198	 */
4199	__trace_early_add_events(tr);
4200
4201	early_enable_events(tr, bootup_event_buf, false);
4202
4203	trace_printk_start_comm();
4204
4205	register_event_cmds();
4206
4207
4208	return 0;
4209}
4210
4211/*
4212 * event_trace_enable() is called from trace_event_init() first to
4213 * initialize events and perhaps start any events that are on the
4214 * command line. Unfortunately, there are some events that will not
4215 * start this early, like the system call tracepoints that need
4216 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4217 * event_trace_enable() is called before pid 1 starts, and this flag
4218 * is never set, making the syscall tracepoint never get reached, but
4219 * the event is enabled regardless (and not doing anything).
4220 */
4221static __init int event_trace_enable_again(void)
4222{
4223	struct trace_array *tr;
4224
4225	tr = top_trace_array();
4226	if (!tr)
4227		return -ENODEV;
4228
4229	early_enable_events(tr, bootup_event_buf, true);
4230
4231	return 0;
4232}
4233
4234early_initcall(event_trace_enable_again);
4235
4236/* Init fields which doesn't related to the tracefs */
4237static __init int event_trace_init_fields(void)
4238{
4239	if (trace_define_generic_fields())
4240		pr_warn("tracing: Failed to allocated generic fields");
4241
4242	if (trace_define_common_fields())
4243		pr_warn("tracing: Failed to allocate common fields");
4244
4245	return 0;
4246}
4247
4248__init int event_trace_init(void)
4249{
4250	struct trace_array *tr;
4251	int ret;
4252
4253	tr = top_trace_array();
4254	if (!tr)
4255		return -ENODEV;
4256
4257	trace_create_file("available_events", TRACE_MODE_READ,
4258			  NULL, tr, &ftrace_avail_fops);
4259
4260	ret = early_event_add_tracer(NULL, tr);
4261	if (ret)
4262		return ret;
4263
4264#ifdef CONFIG_MODULES
4265	ret = register_module_notifier(&trace_module_nb);
4266	if (ret)
4267		pr_warn("Failed to register trace events module notifier\n");
4268#endif
4269
4270	eventdir_initialized = true;
4271
4272	return 0;
4273}
4274
4275void __init trace_event_init(void)
4276{
4277	event_trace_memsetup();
4278	init_ftrace_syscalls();
4279	event_trace_enable();
4280	event_trace_init_fields();
4281}
4282
4283#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4284
4285static DEFINE_SPINLOCK(test_spinlock);
4286static DEFINE_SPINLOCK(test_spinlock_irq);
4287static DEFINE_MUTEX(test_mutex);
4288
4289static __init void test_work(struct work_struct *dummy)
4290{
4291	spin_lock(&test_spinlock);
4292	spin_lock_irq(&test_spinlock_irq);
4293	udelay(1);
4294	spin_unlock_irq(&test_spinlock_irq);
4295	spin_unlock(&test_spinlock);
4296
4297	mutex_lock(&test_mutex);
4298	msleep(1);
4299	mutex_unlock(&test_mutex);
4300}
4301
4302static __init int event_test_thread(void *unused)
4303{
4304	void *test_malloc;
4305
4306	test_malloc = kmalloc(1234, GFP_KERNEL);
4307	if (!test_malloc)
4308		pr_info("failed to kmalloc\n");
4309
4310	schedule_on_each_cpu(test_work);
4311
4312	kfree(test_malloc);
4313
4314	set_current_state(TASK_INTERRUPTIBLE);
4315	while (!kthread_should_stop()) {
4316		schedule();
4317		set_current_state(TASK_INTERRUPTIBLE);
4318	}
4319	__set_current_state(TASK_RUNNING);
4320
4321	return 0;
4322}
4323
4324/*
4325 * Do various things that may trigger events.
4326 */
4327static __init void event_test_stuff(void)
4328{
4329	struct task_struct *test_thread;
4330
4331	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4332	msleep(1);
4333	kthread_stop(test_thread);
4334}
4335
4336/*
4337 * For every trace event defined, we will test each trace point separately,
4338 * and then by groups, and finally all trace points.
4339 */
4340static __init void event_trace_self_tests(void)
4341{
4342	struct trace_subsystem_dir *dir;
4343	struct trace_event_file *file;
4344	struct trace_event_call *call;
4345	struct event_subsystem *system;
4346	struct trace_array *tr;
4347	int ret;
4348
4349	tr = top_trace_array();
4350	if (!tr)
4351		return;
4352
4353	pr_info("Running tests on trace events:\n");
4354
4355	list_for_each_entry(file, &tr->events, list) {
4356
4357		call = file->event_call;
4358
4359		/* Only test those that have a probe */
4360		if (!call->class || !call->class->probe)
4361			continue;
4362
4363/*
4364 * Testing syscall events here is pretty useless, but
4365 * we still do it if configured. But this is time consuming.
4366 * What we really need is a user thread to perform the
4367 * syscalls as we test.
4368 */
4369#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4370		if (call->class->system &&
4371		    strcmp(call->class->system, "syscalls") == 0)
4372			continue;
4373#endif
4374
4375		pr_info("Testing event %s: ", trace_event_name(call));
4376
4377		/*
4378		 * If an event is already enabled, someone is using
4379		 * it and the self test should not be on.
4380		 */
4381		if (file->flags & EVENT_FILE_FL_ENABLED) {
4382			pr_warn("Enabled event during self test!\n");
4383			WARN_ON_ONCE(1);
4384			continue;
4385		}
4386
4387		ftrace_event_enable_disable(file, 1);
4388		event_test_stuff();
4389		ftrace_event_enable_disable(file, 0);
4390
4391		pr_cont("OK\n");
4392	}
4393
4394	/* Now test at the sub system level */
4395
4396	pr_info("Running tests on trace event systems:\n");
4397
4398	list_for_each_entry(dir, &tr->systems, list) {
4399
4400		system = dir->subsystem;
4401
4402		/* the ftrace system is special, skip it */
4403		if (strcmp(system->name, "ftrace") == 0)
4404			continue;
4405
4406		pr_info("Testing event system %s: ", system->name);
4407
4408		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4409		if (WARN_ON_ONCE(ret)) {
4410			pr_warn("error enabling system %s\n",
4411				system->name);
4412			continue;
4413		}
4414
4415		event_test_stuff();
4416
4417		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4418		if (WARN_ON_ONCE(ret)) {
4419			pr_warn("error disabling system %s\n",
4420				system->name);
4421			continue;
4422		}
4423
4424		pr_cont("OK\n");
4425	}
4426
4427	/* Test with all events enabled */
4428
4429	pr_info("Running tests on all trace events:\n");
4430	pr_info("Testing all events: ");
4431
4432	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4433	if (WARN_ON_ONCE(ret)) {
4434		pr_warn("error enabling all events\n");
4435		return;
4436	}
4437
4438	event_test_stuff();
4439
4440	/* reset sysname */
4441	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4442	if (WARN_ON_ONCE(ret)) {
4443		pr_warn("error disabling all events\n");
4444		return;
4445	}
4446
4447	pr_cont("OK\n");
4448}
4449
4450#ifdef CONFIG_FUNCTION_TRACER
4451
4452static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4453
4454static struct trace_event_file event_trace_file __initdata;
4455
4456static void __init
4457function_test_events_call(unsigned long ip, unsigned long parent_ip,
4458			  struct ftrace_ops *op, struct ftrace_regs *regs)
4459{
4460	struct trace_buffer *buffer;
4461	struct ring_buffer_event *event;
4462	struct ftrace_entry *entry;
4463	unsigned int trace_ctx;
4464	long disabled;
4465	int cpu;
4466
4467	trace_ctx = tracing_gen_ctx();
4468	preempt_disable_notrace();
4469	cpu = raw_smp_processor_id();
4470	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4471
4472	if (disabled != 1)
4473		goto out;
4474
4475	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4476						TRACE_FN, sizeof(*entry),
4477						trace_ctx);
4478	if (!event)
4479		goto out;
4480	entry	= ring_buffer_event_data(event);
4481	entry->ip			= ip;
4482	entry->parent_ip		= parent_ip;
4483
4484	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4485				    entry, trace_ctx);
4486 out:
4487	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4488	preempt_enable_notrace();
4489}
4490
4491static struct ftrace_ops trace_ops __initdata  =
4492{
4493	.func = function_test_events_call,
4494};
4495
4496static __init void event_trace_self_test_with_function(void)
4497{
4498	int ret;
4499
4500	event_trace_file.tr = top_trace_array();
4501	if (WARN_ON(!event_trace_file.tr))
4502		return;
4503
4504	ret = register_ftrace_function(&trace_ops);
4505	if (WARN_ON(ret < 0)) {
4506		pr_info("Failed to enable function tracer for event tests\n");
4507		return;
4508	}
4509	pr_info("Running tests again, along with the function tracer\n");
4510	event_trace_self_tests();
4511	unregister_ftrace_function(&trace_ops);
4512}
4513#else
4514static __init void event_trace_self_test_with_function(void)
4515{
4516}
4517#endif
4518
4519static __init int event_trace_self_tests_init(void)
4520{
4521	if (!tracing_selftest_disabled) {
4522		event_trace_self_tests();
4523		event_trace_self_test_with_function();
4524	}
4525
4526	return 0;
4527}
4528
4529late_initcall(event_trace_self_tests_init);
4530
4531#endif
v6.9.4
   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
  36DEFINE_MUTEX(event_mutex);
  37
  38LIST_HEAD(ftrace_events);
  39static LIST_HEAD(ftrace_generic_fields);
  40static LIST_HEAD(ftrace_common_fields);
  41static bool eventdir_initialized;
  42
  43static LIST_HEAD(module_strings);
  44
  45struct 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
  53static struct kmem_cache *field_cachep;
  54static struct kmem_cache *file_cachep;
  55
  56static inline int system_refcount(struct event_subsystem *system)
  57{
  58	return system->ref_count;
  59}
  60
  61static int system_refcount_inc(struct event_subsystem *system)
  62{
  63	return system->ref_count++;
  64}
  65
  66static 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
  84static 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
  97struct ftrace_event_field *
  98trace_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
 115static 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
 143int 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}
 156EXPORT_SYMBOL_GPL(trace_define_field);
 157
 158static 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
 188static 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
 203static 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
 217static 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 */
 233int 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 */
 251static 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 */
 284static 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
 462int 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}
 474EXPORT_SYMBOL_GPL(trace_event_raw_init);
 475
 476bool 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}
 493EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 494
 495void *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}
 525EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 526
 527int 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}
 563EXPORT_SYMBOL_GPL(trace_event_reg);
 564
 565void 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
 587void 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
 609static 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
 721int 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
 727static 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
 733static 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
 744static void
 745event_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
 757static void
 758event_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
 772void 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
 787static void
 788event_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
 813static void
 814event_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
 830static void
 831event_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
 848static void
 849event_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
 867static 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
 882static 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
 925static 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
 932static 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
 950static void __get_system(struct event_subsystem *system)
 951{
 952	WARN_ON_ONCE(system_refcount(system) == 0);
 953	system_refcount_inc(system);
 954}
 955
 956static 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
 963static 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
 974static 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
 981static 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
 993void event_file_get(struct trace_event_file *file)
 994{
 995	atomic_inc(&file->ref);
 996}
 997
 998void 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
1014static 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 */
1027static 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
1075static 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
1087int 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 */
1139int 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}
1148EXPORT_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 */
1163int 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}
1174EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176/* 128 should be much more than enough */
1177#define EVENT_BUF_SIZE		127
1178
1179static ssize_t
1180ftrace_event_write(struct file *file, const char __user *ubuf,
1181		   size_t cnt, loff_t *ppos)
1182{
1183	struct trace_parser parser;
1184	struct seq_file *m = file->private_data;
1185	struct trace_array *tr = m->private;
1186	ssize_t read, ret;
1187
1188	if (!cnt)
1189		return 0;
1190
1191	ret = tracing_update_buffers(tr);
1192	if (ret < 0)
1193		return ret;
1194
1195	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196		return -ENOMEM;
1197
1198	read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200	if (read >= 0 && trace_parser_loaded((&parser))) {
1201		int set = 1;
1202
1203		if (*parser.buffer == '!')
1204			set = 0;
1205
1206		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207		if (ret)
1208			goto out_put;
1209	}
1210
1211	ret = read;
1212
1213 out_put:
1214	trace_parser_put(&parser);
1215
1216	return ret;
1217}
1218
1219static void *
1220t_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
1242static 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
1259static void *
1260s_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
1275static 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
1292static 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
1304static void t_stop(struct seq_file *m, void *p)
1305{
1306	mutex_unlock(&event_mutex);
1307}
1308
1309static 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
1323static void *
1324p_next(struct seq_file *m, void *v, loff_t *pos)
1325{
1326	return __next(m, v, pos, TRACE_PIDS);
1327}
1328
1329static void *
1330np_next(struct seq_file *m, void *v, loff_t *pos)
1331{
1332	return __next(m, v, pos, TRACE_NO_PIDS);
1333}
1334
1335static 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
1361static void *p_start(struct seq_file *m, loff_t *pos)
1362	__acquires(RCU)
1363{
1364	return __start(m, pos, TRACE_PIDS);
1365}
1366
1367static void *np_start(struct seq_file *m, loff_t *pos)
1368	__acquires(RCU)
1369{
1370	return __start(m, pos, TRACE_NO_PIDS);
1371}
1372
1373static 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
1380static ssize_t
1381event_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_data(filp);
1390	if (likely(file))
1391		flags = file->flags;
1392	mutex_unlock(&event_mutex);
1393
1394	if (!file || flags & EVENT_FILE_FL_FREED)
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
1410static ssize_t
1411event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412		   loff_t *ppos)
1413{
1414	struct trace_event_file *file;
1415	unsigned long val;
1416	int ret;
1417
1418	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419	if (ret)
1420		return ret;
1421
1422	switch (val) {
1423	case 0:
1424	case 1:
1425		ret = -ENODEV;
1426		mutex_lock(&event_mutex);
1427		file = event_file_data(filp);
1428		if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429			ret = tracing_update_buffers(file->tr);
1430			if (ret < 0) {
1431				mutex_unlock(&event_mutex);
1432				return ret;
1433			}
1434			ret = ftrace_event_enable_disable(file, val);
1435		}
1436		mutex_unlock(&event_mutex);
1437		break;
1438
1439	default:
1440		return -EINVAL;
1441	}
1442
1443	*ppos += cnt;
1444
1445	return ret ? ret : cnt;
1446}
1447
1448static ssize_t
1449system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450		   loff_t *ppos)
1451{
1452	const char set_to_char[4] = { '?', '0', '1', 'X' };
1453	struct trace_subsystem_dir *dir = filp->private_data;
1454	struct event_subsystem *system = dir->subsystem;
1455	struct trace_event_call *call;
1456	struct trace_event_file *file;
1457	struct trace_array *tr = dir->tr;
1458	char buf[2];
1459	int set = 0;
1460	int ret;
1461
1462	mutex_lock(&event_mutex);
1463	list_for_each_entry(file, &tr->events, list) {
1464		call = file->event_call;
1465		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466		    !trace_event_name(call) || !call->class || !call->class->reg)
1467			continue;
1468
1469		if (system && strcmp(call->class->system, system->name) != 0)
1470			continue;
1471
1472		/*
1473		 * We need to find out if all the events are set
1474		 * or if all events or cleared, or if we have
1475		 * a mixture.
1476		 */
1477		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479		/*
1480		 * If we have a mixture, no need to look further.
1481		 */
1482		if (set == 3)
1483			break;
1484	}
1485	mutex_unlock(&event_mutex);
1486
1487	buf[0] = set_to_char[set];
1488	buf[1] = '\n';
1489
1490	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492	return ret;
1493}
1494
1495static ssize_t
1496system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497		    loff_t *ppos)
1498{
1499	struct trace_subsystem_dir *dir = filp->private_data;
1500	struct event_subsystem *system = dir->subsystem;
1501	const char *name = NULL;
1502	unsigned long val;
1503	ssize_t ret;
1504
1505	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506	if (ret)
1507		return ret;
1508
1509	ret = tracing_update_buffers(dir->tr);
1510	if (ret < 0)
1511		return ret;
1512
1513	if (val != 0 && val != 1)
1514		return -EINVAL;
1515
1516	/*
1517	 * Opening of "enable" adds a ref count to system,
1518	 * so the name is safe to use.
1519	 */
1520	if (system)
1521		name = system->name;
1522
1523	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524	if (ret)
1525		goto out;
1526
1527	ret = cnt;
1528
1529out:
1530	*ppos += cnt;
1531
1532	return ret;
1533}
1534
1535enum {
1536	FORMAT_HEADER		= 1,
1537	FORMAT_FIELD_SEPERATOR	= 2,
1538	FORMAT_PRINTFMT		= 3,
1539};
1540
1541static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542{
1543	struct trace_event_call *call = event_file_data(m->private);
 
1544	struct list_head *common_head = &ftrace_common_fields;
1545	struct list_head *head = trace_get_fields(call);
1546	struct list_head *node = v;
1547
1548	(*pos)++;
1549
1550	switch ((unsigned long)v) {
1551	case FORMAT_HEADER:
1552		node = common_head;
1553		break;
1554
1555	case FORMAT_FIELD_SEPERATOR:
1556		node = head;
1557		break;
1558
1559	case FORMAT_PRINTFMT:
1560		/* all done */
1561		return NULL;
1562	}
1563
1564	node = node->prev;
1565	if (node == common_head)
1566		return (void *)FORMAT_FIELD_SEPERATOR;
1567	else if (node == head)
1568		return (void *)FORMAT_PRINTFMT;
1569	else
1570		return node;
1571}
1572
1573static int f_show(struct seq_file *m, void *v)
1574{
1575	struct trace_event_call *call = event_file_data(m->private);
 
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
1628static void *f_start(struct seq_file *m, loff_t *pos)
1629{
 
1630	void *p = (void *)FORMAT_HEADER;
1631	loff_t l = 0;
1632
1633	/* ->stop() is called even if ->start() fails */
1634	mutex_lock(&event_mutex);
1635	if (!event_file_data(m->private))
 
1636		return ERR_PTR(-ENODEV);
1637
1638	while (l < *pos && p)
1639		p = f_next(m, p, &l);
1640
1641	return p;
1642}
1643
1644static void f_stop(struct seq_file *m, void *p)
1645{
1646	mutex_unlock(&event_mutex);
1647}
1648
1649static const struct seq_operations trace_format_seq_ops = {
1650	.start		= f_start,
1651	.next		= f_next,
1652	.stop		= f_stop,
1653	.show		= f_show,
1654};
1655
1656static int trace_format_open(struct inode *inode, struct file *file)
1657{
1658	struct seq_file *m;
1659	int ret;
1660
1661	/* Do we want to hide event format files on tracefs lockdown? */
1662
1663	ret = seq_open(file, &trace_format_seq_ops);
1664	if (ret < 0)
1665		return ret;
1666
1667	m = file->private_data;
1668	m->private = file;
1669
1670	return 0;
1671}
1672
1673#ifdef CONFIG_PERF_EVENTS
1674static ssize_t
1675event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1676{
1677	int id = (long)event_file_data(filp);
1678	char buf[32];
1679	int len;
1680
1681	if (unlikely(!id))
1682		return -ENODEV;
1683
1684	len = sprintf(buf, "%d\n", id);
1685
1686	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687}
1688#endif
1689
1690static ssize_t
1691event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1692		  loff_t *ppos)
1693{
1694	struct trace_event_file *file;
1695	struct trace_seq *s;
1696	int r = -ENODEV;
1697
1698	if (*ppos)
1699		return 0;
1700
1701	s = kmalloc(sizeof(*s), GFP_KERNEL);
1702
1703	if (!s)
1704		return -ENOMEM;
1705
1706	trace_seq_init(s);
1707
1708	mutex_lock(&event_mutex);
1709	file = event_file_data(filp);
1710	if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711		print_event_filter(file, s);
1712	mutex_unlock(&event_mutex);
1713
1714	if (file)
1715		r = simple_read_from_buffer(ubuf, cnt, ppos,
1716					    s->buffer, trace_seq_used(s));
1717
1718	kfree(s);
1719
1720	return r;
1721}
1722
1723static ssize_t
1724event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1725		   loff_t *ppos)
1726{
1727	struct trace_event_file *file;
1728	char *buf;
1729	int err = -ENODEV;
1730
1731	if (cnt >= PAGE_SIZE)
1732		return -EINVAL;
1733
1734	buf = memdup_user_nul(ubuf, cnt);
1735	if (IS_ERR(buf))
1736		return PTR_ERR(buf);
1737
1738	mutex_lock(&event_mutex);
1739	file = event_file_data(filp);
1740	if (file)
1741		err = apply_event_filter(file, buf);
 
 
 
 
1742	mutex_unlock(&event_mutex);
1743
1744	kfree(buf);
1745	if (err < 0)
1746		return err;
1747
1748	*ppos += cnt;
1749
1750	return cnt;
1751}
1752
1753static LIST_HEAD(event_subsystems);
1754
1755static int subsystem_open(struct inode *inode, struct file *filp)
1756{
1757	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758	struct trace_array *tr = NULL, *iter_tr;
1759	struct event_subsystem *system = NULL;
1760	int ret;
1761
1762	if (tracing_is_disabled())
1763		return -ENODEV;
1764
1765	/* Make sure the system still exists */
1766	mutex_lock(&event_mutex);
1767	mutex_lock(&trace_types_lock);
1768	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770			if (iter_dir == inode->i_private) {
1771				/* Don't open systems with no events */
1772				tr = iter_tr;
1773				dir = iter_dir;
1774				if (dir->nr_events) {
1775					__get_system_dir(dir);
1776					system = dir->subsystem;
1777				}
1778				goto exit_loop;
1779			}
1780		}
1781	}
1782 exit_loop:
1783	mutex_unlock(&trace_types_lock);
1784	mutex_unlock(&event_mutex);
1785
1786	if (!system)
1787		return -ENODEV;
1788
1789	/* Still need to increment the ref count of the system */
1790	if (trace_array_get(tr) < 0) {
1791		put_system(dir);
1792		return -ENODEV;
1793	}
1794
1795	ret = tracing_open_generic(inode, filp);
1796	if (ret < 0) {
1797		trace_array_put(tr);
1798		put_system(dir);
1799	}
1800
1801	return ret;
1802}
1803
1804static int system_tr_open(struct inode *inode, struct file *filp)
1805{
1806	struct trace_subsystem_dir *dir;
1807	struct trace_array *tr = inode->i_private;
1808	int ret;
1809
1810	/* Make a temporary dir that has no system but points to tr */
1811	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1812	if (!dir)
1813		return -ENOMEM;
1814
1815	ret = tracing_open_generic_tr(inode, filp);
1816	if (ret < 0) {
1817		kfree(dir);
1818		return ret;
1819	}
1820	dir->tr = tr;
1821	filp->private_data = dir;
1822
1823	return 0;
1824}
1825
1826static int subsystem_release(struct inode *inode, struct file *file)
1827{
1828	struct trace_subsystem_dir *dir = file->private_data;
1829
1830	trace_array_put(dir->tr);
1831
1832	/*
1833	 * If dir->subsystem is NULL, then this is a temporary
1834	 * descriptor that was made for a trace_array to enable
1835	 * all subsystems.
1836	 */
1837	if (dir->subsystem)
1838		put_system(dir);
1839	else
1840		kfree(dir);
1841
1842	return 0;
1843}
1844
1845static ssize_t
1846subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1847		      loff_t *ppos)
1848{
1849	struct trace_subsystem_dir *dir = filp->private_data;
1850	struct event_subsystem *system = dir->subsystem;
1851	struct trace_seq *s;
1852	int r;
1853
1854	if (*ppos)
1855		return 0;
1856
1857	s = kmalloc(sizeof(*s), GFP_KERNEL);
1858	if (!s)
1859		return -ENOMEM;
1860
1861	trace_seq_init(s);
1862
1863	print_subsystem_event_filter(system, s);
1864	r = simple_read_from_buffer(ubuf, cnt, ppos,
1865				    s->buffer, trace_seq_used(s));
1866
1867	kfree(s);
1868
1869	return r;
1870}
1871
1872static ssize_t
1873subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1874		       loff_t *ppos)
1875{
1876	struct trace_subsystem_dir *dir = filp->private_data;
1877	char *buf;
1878	int err;
1879
1880	if (cnt >= PAGE_SIZE)
1881		return -EINVAL;
1882
1883	buf = memdup_user_nul(ubuf, cnt);
1884	if (IS_ERR(buf))
1885		return PTR_ERR(buf);
1886
1887	err = apply_subsystem_event_filter(dir, buf);
1888	kfree(buf);
1889	if (err < 0)
1890		return err;
1891
1892	*ppos += cnt;
1893
1894	return cnt;
1895}
1896
1897static ssize_t
1898show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1899{
1900	struct trace_array *tr = filp->private_data;
1901	struct trace_seq *s;
1902	int r;
1903
1904	if (*ppos)
1905		return 0;
1906
1907	s = kmalloc(sizeof(*s), GFP_KERNEL);
1908	if (!s)
1909		return -ENOMEM;
1910
1911	trace_seq_init(s);
1912
1913	ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914	r = simple_read_from_buffer(ubuf, cnt, ppos,
1915				    s->buffer, trace_seq_used(s));
1916
1917	kfree(s);
1918
1919	return r;
1920}
1921
1922static ssize_t
1923show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1924{
1925	struct trace_seq *s;
1926	int r;
1927
1928	if (*ppos)
1929		return 0;
1930
1931	s = kmalloc(sizeof(*s), GFP_KERNEL);
1932	if (!s)
1933		return -ENOMEM;
1934
1935	trace_seq_init(s);
1936
1937	ring_buffer_print_entry_header(s);
1938	r = simple_read_from_buffer(ubuf, cnt, ppos,
1939				    s->buffer, trace_seq_used(s));
1940
1941	kfree(s);
1942
1943	return r;
1944}
1945
1946static void ignore_task_cpu(void *data)
1947{
1948	struct trace_array *tr = data;
1949	struct trace_pid_list *pid_list;
1950	struct trace_pid_list *no_pid_list;
1951
1952	/*
1953	 * This function is called by on_each_cpu() while the
1954	 * event_mutex is held.
1955	 */
1956	pid_list = rcu_dereference_protected(tr->filtered_pids,
1957					     mutex_is_locked(&event_mutex));
1958	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959					     mutex_is_locked(&event_mutex));
1960
1961	this_cpu_write(tr->array_buffer.data->ignore_pid,
1962		       trace_ignore_this_task(pid_list, no_pid_list, current));
1963}
1964
1965static void register_pid_events(struct trace_array *tr)
1966{
1967	/*
1968	 * Register a probe that is called before all other probes
1969	 * to set ignore_pid if next or prev do not match.
1970	 * Register a probe this is called after all other probes
1971	 * to only keep ignore_pid set if next pid matches.
1972	 */
1973	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1974					 tr, INT_MAX);
1975	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1976					 tr, 0);
1977
1978	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1979					 tr, INT_MAX);
1980	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1981					 tr, 0);
1982
1983	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1984					     tr, INT_MAX);
1985	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1986					     tr, 0);
1987
1988	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1989					 tr, INT_MAX);
1990	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1991					 tr, 0);
1992}
1993
1994static ssize_t
1995event_pid_write(struct file *filp, const char __user *ubuf,
1996		size_t cnt, loff_t *ppos, int type)
1997{
1998	struct seq_file *m = filp->private_data;
1999	struct trace_array *tr = m->private;
2000	struct trace_pid_list *filtered_pids = NULL;
2001	struct trace_pid_list *other_pids = NULL;
2002	struct trace_pid_list *pid_list;
2003	struct trace_event_file *file;
2004	ssize_t ret;
2005
2006	if (!cnt)
2007		return 0;
2008
2009	ret = tracing_update_buffers(tr);
2010	if (ret < 0)
2011		return ret;
2012
2013	mutex_lock(&event_mutex);
2014
2015	if (type == TRACE_PIDS) {
2016		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017							  lockdep_is_held(&event_mutex));
2018		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019							  lockdep_is_held(&event_mutex));
2020	} else {
2021		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022							  lockdep_is_held(&event_mutex));
2023		other_pids = rcu_dereference_protected(tr->filtered_pids,
2024							  lockdep_is_held(&event_mutex));
2025	}
2026
2027	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2028	if (ret < 0)
2029		goto out;
2030
2031	if (type == TRACE_PIDS)
2032		rcu_assign_pointer(tr->filtered_pids, pid_list);
2033	else
2034		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2035
2036	list_for_each_entry(file, &tr->events, list) {
2037		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2038	}
2039
2040	if (filtered_pids) {
2041		tracepoint_synchronize_unregister();
2042		trace_pid_list_free(filtered_pids);
2043	} else if (pid_list && !other_pids) {
2044		register_pid_events(tr);
2045	}
2046
2047	/*
2048	 * Ignoring of pids is done at task switch. But we have to
2049	 * check for those tasks that are currently running.
2050	 * Always do this in case a pid was appended or removed.
2051	 */
2052	on_each_cpu(ignore_task_cpu, tr, 1);
2053
2054 out:
2055	mutex_unlock(&event_mutex);
2056
2057	if (ret > 0)
2058		*ppos += ret;
2059
2060	return ret;
2061}
2062
2063static ssize_t
2064ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065		       size_t cnt, loff_t *ppos)
2066{
2067	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2068}
2069
2070static ssize_t
2071ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072			size_t cnt, loff_t *ppos)
2073{
2074	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2075}
2076
2077static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081static int ftrace_event_release(struct inode *inode, struct file *file);
2082
2083static const struct seq_operations show_event_seq_ops = {
2084	.start = t_start,
2085	.next = t_next,
2086	.show = t_show,
2087	.stop = t_stop,
2088};
2089
2090static const struct seq_operations show_set_event_seq_ops = {
2091	.start = s_start,
2092	.next = s_next,
2093	.show = t_show,
2094	.stop = t_stop,
2095};
2096
2097static const struct seq_operations show_set_pid_seq_ops = {
2098	.start = p_start,
2099	.next = p_next,
2100	.show = trace_pid_show,
2101	.stop = p_stop,
2102};
2103
2104static const struct seq_operations show_set_no_pid_seq_ops = {
2105	.start = np_start,
2106	.next = np_next,
2107	.show = trace_pid_show,
2108	.stop = p_stop,
2109};
2110
2111static const struct file_operations ftrace_avail_fops = {
2112	.open = ftrace_event_avail_open,
2113	.read = seq_read,
2114	.llseek = seq_lseek,
2115	.release = seq_release,
2116};
2117
2118static const struct file_operations ftrace_set_event_fops = {
2119	.open = ftrace_event_set_open,
2120	.read = seq_read,
2121	.write = ftrace_event_write,
2122	.llseek = seq_lseek,
2123	.release = ftrace_event_release,
2124};
2125
2126static const struct file_operations ftrace_set_event_pid_fops = {
2127	.open = ftrace_event_set_pid_open,
2128	.read = seq_read,
2129	.write = ftrace_event_pid_write,
2130	.llseek = seq_lseek,
2131	.release = ftrace_event_release,
2132};
2133
2134static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135	.open = ftrace_event_set_npid_open,
2136	.read = seq_read,
2137	.write = ftrace_event_npid_write,
2138	.llseek = seq_lseek,
2139	.release = ftrace_event_release,
2140};
2141
2142static const struct file_operations ftrace_enable_fops = {
2143	.open = tracing_open_file_tr,
2144	.read = event_enable_read,
2145	.write = event_enable_write,
2146	.release = tracing_release_file_tr,
2147	.llseek = default_llseek,
2148};
2149
2150static const struct file_operations ftrace_event_format_fops = {
2151	.open = trace_format_open,
2152	.read = seq_read,
2153	.llseek = seq_lseek,
2154	.release = seq_release,
2155};
2156
2157#ifdef CONFIG_PERF_EVENTS
2158static const struct file_operations ftrace_event_id_fops = {
2159	.read = event_id_read,
2160	.llseek = default_llseek,
2161};
2162#endif
2163
2164static const struct file_operations ftrace_event_filter_fops = {
2165	.open = tracing_open_file_tr,
2166	.read = event_filter_read,
2167	.write = event_filter_write,
2168	.release = tracing_release_file_tr,
2169	.llseek = default_llseek,
2170};
2171
2172static const struct file_operations ftrace_subsystem_filter_fops = {
2173	.open = subsystem_open,
2174	.read = subsystem_filter_read,
2175	.write = subsystem_filter_write,
2176	.llseek = default_llseek,
2177	.release = subsystem_release,
2178};
2179
2180static const struct file_operations ftrace_system_enable_fops = {
2181	.open = subsystem_open,
2182	.read = system_enable_read,
2183	.write = system_enable_write,
2184	.llseek = default_llseek,
2185	.release = subsystem_release,
2186};
2187
2188static const struct file_operations ftrace_tr_enable_fops = {
2189	.open = system_tr_open,
2190	.read = system_enable_read,
2191	.write = system_enable_write,
2192	.llseek = default_llseek,
2193	.release = subsystem_release,
2194};
2195
2196static const struct file_operations ftrace_show_header_page_fops = {
2197	.open = tracing_open_generic_tr,
2198	.read = show_header_page_file,
2199	.llseek = default_llseek,
2200	.release = tracing_release_generic_tr,
2201};
2202
2203static const struct file_operations ftrace_show_header_event_fops = {
2204	.open = tracing_open_generic_tr,
2205	.read = show_header_event_file,
2206	.llseek = default_llseek,
2207	.release = tracing_release_generic_tr,
2208};
2209
2210static int
2211ftrace_event_open(struct inode *inode, struct file *file,
2212		  const struct seq_operations *seq_ops)
2213{
2214	struct seq_file *m;
2215	int ret;
2216
2217	ret = security_locked_down(LOCKDOWN_TRACEFS);
2218	if (ret)
2219		return ret;
2220
2221	ret = seq_open(file, seq_ops);
2222	if (ret < 0)
2223		return ret;
2224	m = file->private_data;
2225	/* copy tr over to seq ops */
2226	m->private = inode->i_private;
2227
2228	return ret;
2229}
2230
2231static int ftrace_event_release(struct inode *inode, struct file *file)
2232{
2233	struct trace_array *tr = inode->i_private;
2234
2235	trace_array_put(tr);
2236
2237	return seq_release(inode, file);
2238}
2239
2240static int
2241ftrace_event_avail_open(struct inode *inode, struct file *file)
2242{
2243	const struct seq_operations *seq_ops = &show_event_seq_ops;
2244
2245	/* Checks for tracefs lockdown */
2246	return ftrace_event_open(inode, file, seq_ops);
2247}
2248
2249static int
2250ftrace_event_set_open(struct inode *inode, struct file *file)
2251{
2252	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253	struct trace_array *tr = inode->i_private;
2254	int ret;
2255
2256	ret = tracing_check_open_get_tr(tr);
2257	if (ret)
2258		return ret;
2259
2260	if ((file->f_mode & FMODE_WRITE) &&
2261	    (file->f_flags & O_TRUNC))
2262		ftrace_clear_events(tr);
2263
2264	ret = ftrace_event_open(inode, file, seq_ops);
2265	if (ret < 0)
2266		trace_array_put(tr);
2267	return ret;
2268}
2269
2270static int
2271ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2272{
2273	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274	struct trace_array *tr = inode->i_private;
2275	int ret;
2276
2277	ret = tracing_check_open_get_tr(tr);
2278	if (ret)
2279		return ret;
2280
2281	if ((file->f_mode & FMODE_WRITE) &&
2282	    (file->f_flags & O_TRUNC))
2283		ftrace_clear_event_pids(tr, TRACE_PIDS);
2284
2285	ret = ftrace_event_open(inode, file, seq_ops);
2286	if (ret < 0)
2287		trace_array_put(tr);
2288	return ret;
2289}
2290
2291static int
2292ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2293{
2294	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295	struct trace_array *tr = inode->i_private;
2296	int ret;
2297
2298	ret = tracing_check_open_get_tr(tr);
2299	if (ret)
2300		return ret;
2301
2302	if ((file->f_mode & FMODE_WRITE) &&
2303	    (file->f_flags & O_TRUNC))
2304		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2305
2306	ret = ftrace_event_open(inode, file, seq_ops);
2307	if (ret < 0)
2308		trace_array_put(tr);
2309	return ret;
2310}
2311
2312static struct event_subsystem *
2313create_new_subsystem(const char *name)
2314{
2315	struct event_subsystem *system;
2316
2317	/* need to create new entry */
2318	system = kmalloc(sizeof(*system), GFP_KERNEL);
2319	if (!system)
2320		return NULL;
2321
2322	system->ref_count = 1;
2323
2324	/* Only allocate if dynamic (kprobes and modules) */
2325	system->name = kstrdup_const(name, GFP_KERNEL);
2326	if (!system->name)
2327		goto out_free;
2328
2329	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330	if (!system->filter)
2331		goto out_free;
2332
2333	list_add(&system->list, &event_subsystems);
2334
2335	return system;
2336
2337 out_free:
2338	kfree_const(system->name);
2339	kfree(system);
2340	return NULL;
2341}
2342
2343static int system_callback(const char *name, umode_t *mode, void **data,
2344		    const struct file_operations **fops)
2345{
2346	if (strcmp(name, "filter") == 0)
2347		*fops = &ftrace_subsystem_filter_fops;
2348
2349	else if (strcmp(name, "enable") == 0)
2350		*fops = &ftrace_system_enable_fops;
2351
2352	else
2353		return 0;
2354
2355	*mode = TRACE_MODE_WRITE;
2356	return 1;
2357}
2358
2359static struct eventfs_inode *
2360event_subsystem_dir(struct trace_array *tr, const char *name,
2361		    struct trace_event_file *file, struct eventfs_inode *parent)
2362{
2363	struct event_subsystem *system, *iter;
2364	struct trace_subsystem_dir *dir;
2365	struct eventfs_inode *ei;
2366	int nr_entries;
2367	static struct eventfs_entry system_entries[] = {
2368		{
2369			.name		= "filter",
2370			.callback	= system_callback,
2371		},
2372		{
2373			.name		= "enable",
2374			.callback	= system_callback,
2375		}
2376	};
2377
2378	/* First see if we did not already create this dir */
2379	list_for_each_entry(dir, &tr->systems, list) {
2380		system = dir->subsystem;
2381		if (strcmp(system->name, name) == 0) {
2382			dir->nr_events++;
2383			file->system = dir;
2384			return dir->ei;
2385		}
2386	}
2387
2388	/* Now see if the system itself exists. */
2389	system = NULL;
2390	list_for_each_entry(iter, &event_subsystems, list) {
2391		if (strcmp(iter->name, name) == 0) {
2392			system = iter;
2393			break;
2394		}
2395	}
2396
2397	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2398	if (!dir)
2399		goto out_fail;
2400
2401	if (!system) {
2402		system = create_new_subsystem(name);
2403		if (!system)
2404			goto out_free;
2405	} else
2406		__get_system(system);
2407
2408	/* ftrace only has directories no files */
2409	if (strcmp(name, "ftrace") == 0)
2410		nr_entries = 0;
2411	else
2412		nr_entries = ARRAY_SIZE(system_entries);
2413
2414	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2415	if (IS_ERR(ei)) {
2416		pr_warn("Failed to create system directory %s\n", name);
2417		__put_system(system);
2418		goto out_free;
2419	}
2420
2421	dir->ei = ei;
2422	dir->tr = tr;
2423	dir->ref_count = 1;
2424	dir->nr_events = 1;
2425	dir->subsystem = system;
2426	file->system = dir;
2427
2428	list_add(&dir->list, &tr->systems);
2429
2430	return dir->ei;
2431
2432 out_free:
2433	kfree(dir);
2434 out_fail:
2435	/* Only print this message if failed on memory allocation */
2436	if (!dir || !system)
2437		pr_warn("No memory to create event subsystem %s\n", name);
2438	return NULL;
2439}
2440
2441static int
2442event_define_fields(struct trace_event_call *call)
2443{
2444	struct list_head *head;
2445	int ret = 0;
2446
2447	/*
2448	 * Other events may have the same class. Only update
2449	 * the fields if they are not already defined.
2450	 */
2451	head = trace_get_fields(call);
2452	if (list_empty(head)) {
2453		struct trace_event_fields *field = call->class->fields_array;
2454		unsigned int offset = sizeof(struct trace_entry);
2455
2456		for (; field->type; field++) {
2457			if (field->type == TRACE_FUNCTION_TYPE) {
2458				field->define_fields(call);
2459				break;
2460			}
2461
2462			offset = ALIGN(offset, field->align);
2463			ret = trace_define_field_ext(call, field->type, field->name,
2464						 offset, field->size,
2465						 field->is_signed, field->filter_type,
2466						 field->len);
2467			if (WARN_ON_ONCE(ret)) {
2468				pr_err("error code is %d\n", ret);
2469				break;
2470			}
2471
2472			offset += field->size;
2473		}
2474	}
2475
2476	return ret;
2477}
2478
2479static int event_callback(const char *name, umode_t *mode, void **data,
2480			  const struct file_operations **fops)
2481{
2482	struct trace_event_file *file = *data;
2483	struct trace_event_call *call = file->event_call;
2484
2485	if (strcmp(name, "format") == 0) {
2486		*mode = TRACE_MODE_READ;
2487		*fops = &ftrace_event_format_fops;
2488		*data = call;
2489		return 1;
2490	}
2491
2492	/*
2493	 * Only event directories that can be enabled should have
2494	 * triggers or filters, with the exception of the "print"
2495	 * event that can have a "trigger" file.
2496	 */
2497	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498		if (call->class->reg && strcmp(name, "enable") == 0) {
2499			*mode = TRACE_MODE_WRITE;
2500			*fops = &ftrace_enable_fops;
2501			return 1;
2502		}
2503
2504		if (strcmp(name, "filter") == 0) {
2505			*mode = TRACE_MODE_WRITE;
2506			*fops = &ftrace_event_filter_fops;
2507			return 1;
2508		}
2509	}
2510
2511	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512	    strcmp(trace_event_name(call), "print") == 0) {
2513		if (strcmp(name, "trigger") == 0) {
2514			*mode = TRACE_MODE_WRITE;
2515			*fops = &event_trigger_fops;
2516			return 1;
2517		}
2518	}
2519
2520#ifdef CONFIG_PERF_EVENTS
2521	if (call->event.type && call->class->reg &&
2522	    strcmp(name, "id") == 0) {
2523		*mode = TRACE_MODE_READ;
2524		*data = (void *)(long)call->event.type;
2525		*fops = &ftrace_event_id_fops;
2526		return 1;
2527	}
2528#endif
2529
2530#ifdef CONFIG_HIST_TRIGGERS
2531	if (strcmp(name, "hist") == 0) {
2532		*mode = TRACE_MODE_READ;
2533		*fops = &event_hist_fops;
2534		return 1;
2535	}
2536#endif
2537#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538	if (strcmp(name, "hist_debug") == 0) {
2539		*mode = TRACE_MODE_READ;
2540		*fops = &event_hist_debug_fops;
2541		return 1;
2542	}
2543#endif
2544#ifdef CONFIG_TRACE_EVENT_INJECT
2545	if (call->event.type && call->class->reg &&
2546	    strcmp(name, "inject") == 0) {
2547		*mode = 0200;
2548		*fops = &event_inject_fops;
2549		return 1;
2550	}
2551#endif
2552	return 0;
2553}
2554
2555/* The file is incremented on creation and freeing the enable file decrements it */
2556static void event_release(const char *name, void *data)
2557{
2558	struct trace_event_file *file = data;
2559
2560	event_file_put(file);
2561}
2562
2563static int
2564event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2565{
2566	struct trace_event_call *call = file->event_call;
2567	struct trace_array *tr = file->tr;
2568	struct eventfs_inode *e_events;
2569	struct eventfs_inode *ei;
2570	const char *name;
2571	int nr_entries;
2572	int ret;
2573	static struct eventfs_entry event_entries[] = {
2574		{
2575			.name		= "enable",
2576			.callback	= event_callback,
2577			.release	= event_release,
2578		},
2579		{
2580			.name		= "filter",
2581			.callback	= event_callback,
2582		},
2583		{
2584			.name		= "trigger",
2585			.callback	= event_callback,
2586		},
2587		{
2588			.name		= "format",
2589			.callback	= event_callback,
2590		},
2591#ifdef CONFIG_PERF_EVENTS
2592		{
2593			.name		= "id",
2594			.callback	= event_callback,
2595		},
2596#endif
2597#ifdef CONFIG_HIST_TRIGGERS
2598		{
2599			.name		= "hist",
2600			.callback	= event_callback,
2601		},
2602#endif
2603#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2604		{
2605			.name		= "hist_debug",
2606			.callback	= event_callback,
2607		},
2608#endif
2609#ifdef CONFIG_TRACE_EVENT_INJECT
2610		{
2611			.name		= "inject",
2612			.callback	= event_callback,
2613		},
2614#endif
2615	};
2616
2617	/*
2618	 * If the trace point header did not define TRACE_SYSTEM
2619	 * then the system would be called "TRACE_SYSTEM". This should
2620	 * never happen.
2621	 */
2622	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2623		return -ENODEV;
2624
2625	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2626	if (!e_events)
2627		return -ENOMEM;
2628
2629	nr_entries = ARRAY_SIZE(event_entries);
2630
2631	name = trace_event_name(call);
2632	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2633	if (IS_ERR(ei)) {
2634		pr_warn("Could not create tracefs '%s' directory\n", name);
2635		return -1;
2636	}
2637
2638	file->ei = ei;
2639
2640	ret = event_define_fields(call);
2641	if (ret < 0) {
2642		pr_warn("Could not initialize trace point events/%s\n", name);
2643		return ret;
2644	}
2645
2646	/* Gets decremented on freeing of the "enable" file */
2647	event_file_get(file);
2648
2649	return 0;
2650}
2651
2652static void remove_event_from_tracers(struct trace_event_call *call)
2653{
2654	struct trace_event_file *file;
2655	struct trace_array *tr;
2656
2657	do_for_each_event_file_safe(tr, file) {
2658		if (file->event_call != call)
2659			continue;
2660
2661		remove_event_file_dir(file);
2662		/*
2663		 * The do_for_each_event_file_safe() is
2664		 * a double loop. After finding the call for this
2665		 * trace_array, we use break to jump to the next
2666		 * trace_array.
2667		 */
2668		break;
2669	} while_for_each_event_file();
2670}
2671
2672static void event_remove(struct trace_event_call *call)
2673{
2674	struct trace_array *tr;
2675	struct trace_event_file *file;
2676
2677	do_for_each_event_file(tr, file) {
2678		if (file->event_call != call)
2679			continue;
2680
2681		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2682			tr->clear_trace = true;
2683
2684		ftrace_event_enable_disable(file, 0);
2685		/*
2686		 * The do_for_each_event_file() is
2687		 * a double loop. After finding the call for this
2688		 * trace_array, we use break to jump to the next
2689		 * trace_array.
2690		 */
2691		break;
2692	} while_for_each_event_file();
2693
2694	if (call->event.funcs)
2695		__unregister_trace_event(&call->event);
2696	remove_event_from_tracers(call);
2697	list_del(&call->list);
2698}
2699
2700static int event_init(struct trace_event_call *call)
2701{
2702	int ret = 0;
2703	const char *name;
2704
2705	name = trace_event_name(call);
2706	if (WARN_ON(!name))
2707		return -EINVAL;
2708
2709	if (call->class->raw_init) {
2710		ret = call->class->raw_init(call);
2711		if (ret < 0 && ret != -ENOSYS)
2712			pr_warn("Could not initialize trace events/%s\n", name);
2713	}
2714
2715	return ret;
2716}
2717
2718static int
2719__register_event(struct trace_event_call *call, struct module *mod)
2720{
2721	int ret;
2722
2723	ret = event_init(call);
2724	if (ret < 0)
2725		return ret;
2726
2727	list_add(&call->list, &ftrace_events);
2728	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2729		atomic_set(&call->refcnt, 0);
2730	else
2731		call->module = mod;
2732
2733	return 0;
2734}
2735
2736static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2737{
2738	int rlen;
2739	int elen;
2740
2741	/* Find the length of the eval value as a string */
2742	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2743	/* Make sure there's enough room to replace the string with the value */
2744	if (len < elen)
2745		return NULL;
2746
2747	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2748
2749	/* Get the rest of the string of ptr */
2750	rlen = strlen(ptr + len);
2751	memmove(ptr + elen, ptr + len, rlen);
2752	/* Make sure we end the new string */
2753	ptr[elen + rlen] = 0;
2754
2755	return ptr + elen;
2756}
2757
2758static void update_event_printk(struct trace_event_call *call,
2759				struct trace_eval_map *map)
2760{
2761	char *ptr;
2762	int quote = 0;
2763	int len = strlen(map->eval_string);
2764
2765	for (ptr = call->print_fmt; *ptr; ptr++) {
2766		if (*ptr == '\\') {
2767			ptr++;
2768			/* paranoid */
2769			if (!*ptr)
2770				break;
2771			continue;
2772		}
2773		if (*ptr == '"') {
2774			quote ^= 1;
2775			continue;
2776		}
2777		if (quote)
2778			continue;
2779		if (isdigit(*ptr)) {
2780			/* skip numbers */
2781			do {
2782				ptr++;
2783				/* Check for alpha chars like ULL */
2784			} while (isalnum(*ptr));
2785			if (!*ptr)
2786				break;
2787			/*
2788			 * A number must have some kind of delimiter after
2789			 * it, and we can ignore that too.
2790			 */
2791			continue;
2792		}
2793		if (isalpha(*ptr) || *ptr == '_') {
2794			if (strncmp(map->eval_string, ptr, len) == 0 &&
2795			    !isalnum(ptr[len]) && ptr[len] != '_') {
2796				ptr = eval_replace(ptr, map, len);
2797				/* enum/sizeof string smaller than value */
2798				if (WARN_ON_ONCE(!ptr))
2799					return;
2800				/*
2801				 * No need to decrement here, as eval_replace()
2802				 * returns the pointer to the character passed
2803				 * the eval, and two evals can not be placed
2804				 * back to back without something in between.
2805				 * We can skip that something in between.
2806				 */
2807				continue;
2808			}
2809		skip_more:
2810			do {
2811				ptr++;
2812			} while (isalnum(*ptr) || *ptr == '_');
2813			if (!*ptr)
2814				break;
2815			/*
2816			 * If what comes after this variable is a '.' or
2817			 * '->' then we can continue to ignore that string.
2818			 */
2819			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2820				ptr += *ptr == '.' ? 1 : 2;
2821				if (!*ptr)
2822					break;
2823				goto skip_more;
2824			}
2825			/*
2826			 * Once again, we can skip the delimiter that came
2827			 * after the string.
2828			 */
2829			continue;
2830		}
2831	}
2832}
2833
2834static void add_str_to_module(struct module *module, char *str)
2835{
2836	struct module_string *modstr;
2837
2838	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2839
2840	/*
2841	 * If we failed to allocate memory here, then we'll just
2842	 * let the str memory leak when the module is removed.
2843	 * If this fails to allocate, there's worse problems than
2844	 * a leaked string on module removal.
2845	 */
2846	if (WARN_ON_ONCE(!modstr))
2847		return;
2848
2849	modstr->module = module;
2850	modstr->str = str;
2851
2852	list_add(&modstr->next, &module_strings);
2853}
2854
2855static void update_event_fields(struct trace_event_call *call,
2856				struct trace_eval_map *map)
2857{
2858	struct ftrace_event_field *field;
2859	struct list_head *head;
2860	char *ptr;
2861	char *str;
2862	int len = strlen(map->eval_string);
2863
2864	/* Dynamic events should never have field maps */
2865	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2866		return;
2867
2868	head = trace_get_fields(call);
2869	list_for_each_entry(field, head, link) {
2870		ptr = strchr(field->type, '[');
2871		if (!ptr)
2872			continue;
2873		ptr++;
2874
2875		if (!isalpha(*ptr) && *ptr != '_')
2876			continue;
2877
2878		if (strncmp(map->eval_string, ptr, len) != 0)
2879			continue;
2880
2881		str = kstrdup(field->type, GFP_KERNEL);
2882		if (WARN_ON_ONCE(!str))
2883			return;
2884		ptr = str + (ptr - field->type);
2885		ptr = eval_replace(ptr, map, len);
2886		/* enum/sizeof string smaller than value */
2887		if (WARN_ON_ONCE(!ptr)) {
2888			kfree(str);
2889			continue;
2890		}
2891
2892		/*
2893		 * If the event is part of a module, then we need to free the string
2894		 * when the module is removed. Otherwise, it will stay allocated
2895		 * until a reboot.
2896		 */
2897		if (call->module)
2898			add_str_to_module(call->module, str);
2899
2900		field->type = str;
2901	}
2902}
2903
2904void trace_event_eval_update(struct trace_eval_map **map, int len)
2905{
2906	struct trace_event_call *call, *p;
2907	const char *last_system = NULL;
2908	bool first = false;
2909	int last_i;
2910	int i;
2911
2912	down_write(&trace_event_sem);
2913	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2914		/* events are usually grouped together with systems */
2915		if (!last_system || call->class->system != last_system) {
2916			first = true;
2917			last_i = 0;
2918			last_system = call->class->system;
2919		}
2920
2921		/*
2922		 * Since calls are grouped by systems, the likelihood that the
2923		 * next call in the iteration belongs to the same system as the
2924		 * previous call is high. As an optimization, we skip searching
2925		 * for a map[] that matches the call's system if the last call
2926		 * was from the same system. That's what last_i is for. If the
2927		 * call has the same system as the previous call, then last_i
2928		 * will be the index of the first map[] that has a matching
2929		 * system.
2930		 */
2931		for (i = last_i; i < len; i++) {
2932			if (call->class->system == map[i]->system) {
2933				/* Save the first system if need be */
2934				if (first) {
2935					last_i = i;
2936					first = false;
2937				}
2938				update_event_printk(call, map[i]);
2939				update_event_fields(call, map[i]);
2940			}
2941		}
2942		cond_resched();
2943	}
2944	up_write(&trace_event_sem);
2945}
2946
2947static bool event_in_systems(struct trace_event_call *call,
2948			     const char *systems)
2949{
2950	const char *system;
2951	const char *p;
2952
2953	if (!systems)
2954		return true;
2955
2956	system = call->class->system;
2957	p = strstr(systems, system);
2958	if (!p)
2959		return false;
2960
2961	if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2962		return false;
2963
2964	p += strlen(system);
2965	return !*p || isspace(*p) || *p == ',';
2966}
2967
2968static struct trace_event_file *
2969trace_create_new_event(struct trace_event_call *call,
2970		       struct trace_array *tr)
2971{
2972	struct trace_pid_list *no_pid_list;
2973	struct trace_pid_list *pid_list;
2974	struct trace_event_file *file;
2975	unsigned int first;
2976
2977	if (!event_in_systems(call, tr->system_names))
2978		return NULL;
2979
2980	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2981	if (!file)
2982		return ERR_PTR(-ENOMEM);
2983
2984	pid_list = rcu_dereference_protected(tr->filtered_pids,
2985					     lockdep_is_held(&event_mutex));
2986	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2987					     lockdep_is_held(&event_mutex));
2988
2989	if (!trace_pid_list_first(pid_list, &first) ||
2990	    !trace_pid_list_first(no_pid_list, &first))
2991		file->flags |= EVENT_FILE_FL_PID_FILTER;
2992
2993	file->event_call = call;
2994	file->tr = tr;
2995	atomic_set(&file->sm_ref, 0);
2996	atomic_set(&file->tm_ref, 0);
2997	INIT_LIST_HEAD(&file->triggers);
2998	list_add(&file->list, &tr->events);
2999	event_file_get(file);
3000
3001	return file;
3002}
3003
3004#define MAX_BOOT_TRIGGERS 32
3005
3006static struct boot_triggers {
3007	const char		*event;
3008	char			*trigger;
3009} bootup_triggers[MAX_BOOT_TRIGGERS];
3010
3011static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3012static int nr_boot_triggers;
3013
3014static __init int setup_trace_triggers(char *str)
3015{
3016	char *trigger;
3017	char *buf;
3018	int i;
3019
3020	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3021	trace_set_ring_buffer_expanded(NULL);
3022	disable_tracing_selftest("running event triggers");
3023
3024	buf = bootup_trigger_buf;
3025	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3026		trigger = strsep(&buf, ",");
3027		if (!trigger)
3028			break;
3029		bootup_triggers[i].event = strsep(&trigger, ".");
3030		bootup_triggers[i].trigger = trigger;
3031		if (!bootup_triggers[i].trigger)
3032			break;
3033	}
3034
3035	nr_boot_triggers = i;
3036	return 1;
3037}
3038__setup("trace_trigger=", setup_trace_triggers);
3039
3040/* Add an event to a trace directory */
3041static int
3042__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3043{
3044	struct trace_event_file *file;
3045
3046	file = trace_create_new_event(call, tr);
3047	/*
3048	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3049	 * allocation, or NULL if the event is not part of the tr->system_names.
3050	 * When the event is not part of the tr->system_names, return zero, not
3051	 * an error.
3052	 */
3053	if (!file)
3054		return 0;
3055
3056	if (IS_ERR(file))
3057		return PTR_ERR(file);
3058
3059	if (eventdir_initialized)
3060		return event_create_dir(tr->event_dir, file);
3061	else
3062		return event_define_fields(call);
3063}
3064
3065static void trace_early_triggers(struct trace_event_file *file, const char *name)
3066{
3067	int ret;
3068	int i;
3069
3070	for (i = 0; i < nr_boot_triggers; i++) {
3071		if (strcmp(name, bootup_triggers[i].event))
3072			continue;
3073		mutex_lock(&event_mutex);
3074		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3075		mutex_unlock(&event_mutex);
3076		if (ret)
3077			pr_err("Failed to register trigger '%s' on event %s\n",
3078			       bootup_triggers[i].trigger,
3079			       bootup_triggers[i].event);
3080	}
3081}
3082
3083/*
3084 * Just create a descriptor for early init. A descriptor is required
3085 * for enabling events at boot. We want to enable events before
3086 * the filesystem is initialized.
3087 */
3088static int
3089__trace_early_add_new_event(struct trace_event_call *call,
3090			    struct trace_array *tr)
3091{
3092	struct trace_event_file *file;
3093	int ret;
3094
3095	file = trace_create_new_event(call, tr);
3096	/*
3097	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3098	 * allocation, or NULL if the event is not part of the tr->system_names.
3099	 * When the event is not part of the tr->system_names, return zero, not
3100	 * an error.
3101	 */
3102	if (!file)
3103		return 0;
3104
3105	if (IS_ERR(file))
3106		return PTR_ERR(file);
3107
3108	ret = event_define_fields(call);
3109	if (ret)
3110		return ret;
3111
3112	trace_early_triggers(file, trace_event_name(call));
3113
3114	return 0;
3115}
3116
3117struct ftrace_module_file_ops;
3118static void __add_event_to_tracers(struct trace_event_call *call);
3119
3120/* Add an additional event_call dynamically */
3121int trace_add_event_call(struct trace_event_call *call)
3122{
3123	int ret;
3124	lockdep_assert_held(&event_mutex);
3125
3126	mutex_lock(&trace_types_lock);
3127
3128	ret = __register_event(call, NULL);
3129	if (ret >= 0)
3130		__add_event_to_tracers(call);
3131
3132	mutex_unlock(&trace_types_lock);
3133	return ret;
3134}
3135EXPORT_SYMBOL_GPL(trace_add_event_call);
3136
3137/*
3138 * Must be called under locking of trace_types_lock, event_mutex and
3139 * trace_event_sem.
3140 */
3141static void __trace_remove_event_call(struct trace_event_call *call)
3142{
3143	event_remove(call);
3144	trace_destroy_fields(call);
3145	free_event_filter(call->filter);
3146	call->filter = NULL;
3147}
3148
3149static int probe_remove_event_call(struct trace_event_call *call)
3150{
3151	struct trace_array *tr;
3152	struct trace_event_file *file;
3153
3154#ifdef CONFIG_PERF_EVENTS
3155	if (call->perf_refcount)
3156		return -EBUSY;
3157#endif
3158	do_for_each_event_file(tr, file) {
3159		if (file->event_call != call)
3160			continue;
3161		/*
3162		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3163		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3164		 * TRACE_REG_UNREGISTER.
3165		 */
3166		if (file->flags & EVENT_FILE_FL_ENABLED)
3167			goto busy;
3168
3169		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3170			tr->clear_trace = true;
3171		/*
3172		 * The do_for_each_event_file_safe() is
3173		 * a double loop. After finding the call for this
3174		 * trace_array, we use break to jump to the next
3175		 * trace_array.
3176		 */
3177		break;
3178	} while_for_each_event_file();
3179
3180	__trace_remove_event_call(call);
3181
3182	return 0;
3183 busy:
3184	/* No need to clear the trace now */
3185	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3186		tr->clear_trace = false;
3187	}
3188	return -EBUSY;
3189}
3190
3191/* Remove an event_call */
3192int trace_remove_event_call(struct trace_event_call *call)
3193{
3194	int ret;
3195
3196	lockdep_assert_held(&event_mutex);
3197
3198	mutex_lock(&trace_types_lock);
3199	down_write(&trace_event_sem);
3200	ret = probe_remove_event_call(call);
3201	up_write(&trace_event_sem);
3202	mutex_unlock(&trace_types_lock);
3203
3204	return ret;
3205}
3206EXPORT_SYMBOL_GPL(trace_remove_event_call);
3207
3208#define for_each_event(event, start, end)			\
3209	for (event = start;					\
3210	     (unsigned long)event < (unsigned long)end;		\
3211	     event++)
3212
3213#ifdef CONFIG_MODULES
3214
3215static void trace_module_add_events(struct module *mod)
3216{
3217	struct trace_event_call **call, **start, **end;
3218
3219	if (!mod->num_trace_events)
3220		return;
3221
3222	/* Don't add infrastructure for mods without tracepoints */
3223	if (trace_module_has_bad_taint(mod)) {
3224		pr_err("%s: module has bad taint, not creating trace events\n",
3225		       mod->name);
3226		return;
3227	}
3228
3229	start = mod->trace_events;
3230	end = mod->trace_events + mod->num_trace_events;
3231
3232	for_each_event(call, start, end) {
3233		__register_event(*call, mod);
3234		__add_event_to_tracers(*call);
3235	}
3236}
3237
3238static void trace_module_remove_events(struct module *mod)
3239{
3240	struct trace_event_call *call, *p;
3241	struct module_string *modstr, *m;
3242
3243	down_write(&trace_event_sem);
3244	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3245		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3246			continue;
3247		if (call->module == mod)
3248			__trace_remove_event_call(call);
3249	}
3250	/* Check for any strings allocade for this module */
3251	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3252		if (modstr->module != mod)
3253			continue;
3254		list_del(&modstr->next);
3255		kfree(modstr->str);
3256		kfree(modstr);
3257	}
3258	up_write(&trace_event_sem);
3259
3260	/*
3261	 * It is safest to reset the ring buffer if the module being unloaded
3262	 * registered any events that were used. The only worry is if
3263	 * a new module gets loaded, and takes on the same id as the events
3264	 * of this module. When printing out the buffer, traced events left
3265	 * over from this module may be passed to the new module events and
3266	 * unexpected results may occur.
3267	 */
3268	tracing_reset_all_online_cpus_unlocked();
3269}
3270
3271static int trace_module_notify(struct notifier_block *self,
3272			       unsigned long val, void *data)
3273{
3274	struct module *mod = data;
3275
3276	mutex_lock(&event_mutex);
3277	mutex_lock(&trace_types_lock);
3278	switch (val) {
3279	case MODULE_STATE_COMING:
3280		trace_module_add_events(mod);
3281		break;
3282	case MODULE_STATE_GOING:
3283		trace_module_remove_events(mod);
3284		break;
3285	}
3286	mutex_unlock(&trace_types_lock);
3287	mutex_unlock(&event_mutex);
3288
3289	return NOTIFY_OK;
3290}
3291
3292static struct notifier_block trace_module_nb = {
3293	.notifier_call = trace_module_notify,
3294	.priority = 1, /* higher than trace.c module notify */
3295};
3296#endif /* CONFIG_MODULES */
3297
3298/* Create a new event directory structure for a trace directory. */
3299static void
3300__trace_add_event_dirs(struct trace_array *tr)
3301{
3302	struct trace_event_call *call;
3303	int ret;
3304
3305	list_for_each_entry(call, &ftrace_events, list) {
3306		ret = __trace_add_new_event(call, tr);
3307		if (ret < 0)
3308			pr_warn("Could not create directory for event %s\n",
3309				trace_event_name(call));
3310	}
3311}
3312
3313/* Returns any file that matches the system and event */
3314struct trace_event_file *
3315__find_event_file(struct trace_array *tr, const char *system, const char *event)
3316{
3317	struct trace_event_file *file;
3318	struct trace_event_call *call;
3319	const char *name;
3320
3321	list_for_each_entry(file, &tr->events, list) {
3322
3323		call = file->event_call;
3324		name = trace_event_name(call);
3325
3326		if (!name || !call->class)
3327			continue;
3328
3329		if (strcmp(event, name) == 0 &&
3330		    strcmp(system, call->class->system) == 0)
3331			return file;
3332	}
3333	return NULL;
3334}
3335
3336/* Returns valid trace event files that match system and event */
3337struct trace_event_file *
3338find_event_file(struct trace_array *tr, const char *system, const char *event)
3339{
3340	struct trace_event_file *file;
3341
3342	file = __find_event_file(tr, system, event);
3343	if (!file || !file->event_call->class->reg ||
3344	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3345		return NULL;
3346
3347	return file;
3348}
3349
3350/**
3351 * trace_get_event_file - Find and return a trace event file
3352 * @instance: The name of the trace instance containing the event
3353 * @system: The name of the system containing the event
3354 * @event: The name of the event
3355 *
3356 * Return a trace event file given the trace instance name, trace
3357 * system, and trace event name.  If the instance name is NULL, it
3358 * refers to the top-level trace array.
3359 *
3360 * This function will look it up and return it if found, after calling
3361 * trace_array_get() to prevent the instance from going away, and
3362 * increment the event's module refcount to prevent it from being
3363 * removed.
3364 *
3365 * To release the file, call trace_put_event_file(), which will call
3366 * trace_array_put() and decrement the event's module refcount.
3367 *
3368 * Return: The trace event on success, ERR_PTR otherwise.
3369 */
3370struct trace_event_file *trace_get_event_file(const char *instance,
3371					      const char *system,
3372					      const char *event)
3373{
3374	struct trace_array *tr = top_trace_array();
3375	struct trace_event_file *file = NULL;
3376	int ret = -EINVAL;
3377
3378	if (instance) {
3379		tr = trace_array_find_get(instance);
3380		if (!tr)
3381			return ERR_PTR(-ENOENT);
3382	} else {
3383		ret = trace_array_get(tr);
3384		if (ret)
3385			return ERR_PTR(ret);
3386	}
3387
3388	mutex_lock(&event_mutex);
3389
3390	file = find_event_file(tr, system, event);
3391	if (!file) {
3392		trace_array_put(tr);
3393		ret = -EINVAL;
3394		goto out;
3395	}
3396
3397	/* Don't let event modules unload while in use */
3398	ret = trace_event_try_get_ref(file->event_call);
3399	if (!ret) {
3400		trace_array_put(tr);
3401		ret = -EBUSY;
3402		goto out;
3403	}
3404
3405	ret = 0;
3406 out:
3407	mutex_unlock(&event_mutex);
3408
3409	if (ret)
3410		file = ERR_PTR(ret);
3411
3412	return file;
3413}
3414EXPORT_SYMBOL_GPL(trace_get_event_file);
3415
3416/**
3417 * trace_put_event_file - Release a file from trace_get_event_file()
3418 * @file: The trace event file
3419 *
3420 * If a file was retrieved using trace_get_event_file(), this should
3421 * be called when it's no longer needed.  It will cancel the previous
3422 * trace_array_get() called by that function, and decrement the
3423 * event's module refcount.
3424 */
3425void trace_put_event_file(struct trace_event_file *file)
3426{
3427	mutex_lock(&event_mutex);
3428	trace_event_put_ref(file->event_call);
3429	mutex_unlock(&event_mutex);
3430
3431	trace_array_put(file->tr);
3432}
3433EXPORT_SYMBOL_GPL(trace_put_event_file);
3434
3435#ifdef CONFIG_DYNAMIC_FTRACE
3436
3437/* Avoid typos */
3438#define ENABLE_EVENT_STR	"enable_event"
3439#define DISABLE_EVENT_STR	"disable_event"
3440
3441struct event_probe_data {
3442	struct trace_event_file	*file;
3443	unsigned long			count;
3444	int				ref;
3445	bool				enable;
3446};
3447
3448static void update_event_probe(struct event_probe_data *data)
3449{
3450	if (data->enable)
3451		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3452	else
3453		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3454}
3455
3456static void
3457event_enable_probe(unsigned long ip, unsigned long parent_ip,
3458		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3459		   void *data)
3460{
3461	struct ftrace_func_mapper *mapper = data;
3462	struct event_probe_data *edata;
3463	void **pdata;
3464
3465	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3466	if (!pdata || !*pdata)
3467		return;
3468
3469	edata = *pdata;
3470	update_event_probe(edata);
3471}
3472
3473static void
3474event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3475			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3476			 void *data)
3477{
3478	struct ftrace_func_mapper *mapper = data;
3479	struct event_probe_data *edata;
3480	void **pdata;
3481
3482	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3483	if (!pdata || !*pdata)
3484		return;
3485
3486	edata = *pdata;
3487
3488	if (!edata->count)
3489		return;
3490
3491	/* Skip if the event is in a state we want to switch to */
3492	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3493		return;
3494
3495	if (edata->count != -1)
3496		(edata->count)--;
3497
3498	update_event_probe(edata);
3499}
3500
3501static int
3502event_enable_print(struct seq_file *m, unsigned long ip,
3503		   struct ftrace_probe_ops *ops, void *data)
3504{
3505	struct ftrace_func_mapper *mapper = data;
3506	struct event_probe_data *edata;
3507	void **pdata;
3508
3509	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3510
3511	if (WARN_ON_ONCE(!pdata || !*pdata))
3512		return 0;
3513
3514	edata = *pdata;
3515
3516	seq_printf(m, "%ps:", (void *)ip);
3517
3518	seq_printf(m, "%s:%s:%s",
3519		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3520		   edata->file->event_call->class->system,
3521		   trace_event_name(edata->file->event_call));
3522
3523	if (edata->count == -1)
3524		seq_puts(m, ":unlimited\n");
3525	else
3526		seq_printf(m, ":count=%ld\n", edata->count);
3527
3528	return 0;
3529}
3530
3531static int
3532event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3533		  unsigned long ip, void *init_data, void **data)
3534{
3535	struct ftrace_func_mapper *mapper = *data;
3536	struct event_probe_data *edata = init_data;
3537	int ret;
3538
3539	if (!mapper) {
3540		mapper = allocate_ftrace_func_mapper();
3541		if (!mapper)
3542			return -ENODEV;
3543		*data = mapper;
3544	}
3545
3546	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3547	if (ret < 0)
3548		return ret;
3549
3550	edata->ref++;
3551
3552	return 0;
3553}
3554
3555static int free_probe_data(void *data)
3556{
3557	struct event_probe_data *edata = data;
3558
3559	edata->ref--;
3560	if (!edata->ref) {
3561		/* Remove the SOFT_MODE flag */
3562		__ftrace_event_enable_disable(edata->file, 0, 1);
3563		trace_event_put_ref(edata->file->event_call);
3564		kfree(edata);
3565	}
3566	return 0;
3567}
3568
3569static void
3570event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3571		  unsigned long ip, void *data)
3572{
3573	struct ftrace_func_mapper *mapper = data;
3574	struct event_probe_data *edata;
3575
3576	if (!ip) {
3577		if (!mapper)
3578			return;
3579		free_ftrace_func_mapper(mapper, free_probe_data);
3580		return;
3581	}
3582
3583	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3584
3585	if (WARN_ON_ONCE(!edata))
3586		return;
3587
3588	if (WARN_ON_ONCE(edata->ref <= 0))
3589		return;
3590
3591	free_probe_data(edata);
3592}
3593
3594static struct ftrace_probe_ops event_enable_probe_ops = {
3595	.func			= event_enable_probe,
3596	.print			= event_enable_print,
3597	.init			= event_enable_init,
3598	.free			= event_enable_free,
3599};
3600
3601static struct ftrace_probe_ops event_enable_count_probe_ops = {
3602	.func			= event_enable_count_probe,
3603	.print			= event_enable_print,
3604	.init			= event_enable_init,
3605	.free			= event_enable_free,
3606};
3607
3608static struct ftrace_probe_ops event_disable_probe_ops = {
3609	.func			= event_enable_probe,
3610	.print			= event_enable_print,
3611	.init			= event_enable_init,
3612	.free			= event_enable_free,
3613};
3614
3615static struct ftrace_probe_ops event_disable_count_probe_ops = {
3616	.func			= event_enable_count_probe,
3617	.print			= event_enable_print,
3618	.init			= event_enable_init,
3619	.free			= event_enable_free,
3620};
3621
3622static int
3623event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3624		  char *glob, char *cmd, char *param, int enabled)
3625{
3626	struct trace_event_file *file;
3627	struct ftrace_probe_ops *ops;
3628	struct event_probe_data *data;
3629	const char *system;
3630	const char *event;
3631	char *number;
3632	bool enable;
3633	int ret;
3634
3635	if (!tr)
3636		return -ENODEV;
3637
3638	/* hash funcs only work with set_ftrace_filter */
3639	if (!enabled || !param)
3640		return -EINVAL;
3641
3642	system = strsep(&param, ":");
3643	if (!param)
3644		return -EINVAL;
3645
3646	event = strsep(&param, ":");
3647
3648	mutex_lock(&event_mutex);
3649
3650	ret = -EINVAL;
3651	file = find_event_file(tr, system, event);
3652	if (!file)
3653		goto out;
3654
3655	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3656
3657	if (enable)
3658		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3659	else
3660		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3661
3662	if (glob[0] == '!') {
3663		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3664		goto out;
3665	}
3666
3667	ret = -ENOMEM;
3668
3669	data = kzalloc(sizeof(*data), GFP_KERNEL);
3670	if (!data)
3671		goto out;
3672
3673	data->enable = enable;
3674	data->count = -1;
3675	data->file = file;
3676
3677	if (!param)
3678		goto out_reg;
3679
3680	number = strsep(&param, ":");
3681
3682	ret = -EINVAL;
3683	if (!strlen(number))
3684		goto out_free;
3685
3686	/*
3687	 * We use the callback data field (which is a pointer)
3688	 * as our counter.
3689	 */
3690	ret = kstrtoul(number, 0, &data->count);
3691	if (ret)
3692		goto out_free;
3693
3694 out_reg:
3695	/* Don't let event modules unload while probe registered */
3696	ret = trace_event_try_get_ref(file->event_call);
3697	if (!ret) {
3698		ret = -EBUSY;
3699		goto out_free;
3700	}
3701
3702	ret = __ftrace_event_enable_disable(file, 1, 1);
3703	if (ret < 0)
3704		goto out_put;
3705
3706	ret = register_ftrace_function_probe(glob, tr, ops, data);
3707	/*
3708	 * The above returns on success the # of functions enabled,
3709	 * but if it didn't find any functions it returns zero.
3710	 * Consider no functions a failure too.
3711	 */
3712	if (!ret) {
3713		ret = -ENOENT;
3714		goto out_disable;
3715	} else if (ret < 0)
3716		goto out_disable;
3717	/* Just return zero, not the number of enabled functions */
3718	ret = 0;
3719 out:
3720	mutex_unlock(&event_mutex);
3721	return ret;
3722
3723 out_disable:
3724	__ftrace_event_enable_disable(file, 0, 1);
3725 out_put:
3726	trace_event_put_ref(file->event_call);
3727 out_free:
3728	kfree(data);
3729	goto out;
3730}
3731
3732static struct ftrace_func_command event_enable_cmd = {
3733	.name			= ENABLE_EVENT_STR,
3734	.func			= event_enable_func,
3735};
3736
3737static struct ftrace_func_command event_disable_cmd = {
3738	.name			= DISABLE_EVENT_STR,
3739	.func			= event_enable_func,
3740};
3741
3742static __init int register_event_cmds(void)
3743{
3744	int ret;
3745
3746	ret = register_ftrace_command(&event_enable_cmd);
3747	if (WARN_ON(ret < 0))
3748		return ret;
3749	ret = register_ftrace_command(&event_disable_cmd);
3750	if (WARN_ON(ret < 0))
3751		unregister_ftrace_command(&event_enable_cmd);
3752	return ret;
3753}
3754#else
3755static inline int register_event_cmds(void) { return 0; }
3756#endif /* CONFIG_DYNAMIC_FTRACE */
3757
3758/*
3759 * The top level array and trace arrays created by boot-time tracing
3760 * have already had its trace_event_file descriptors created in order
3761 * to allow for early events to be recorded.
3762 * This function is called after the tracefs has been initialized,
3763 * and we now have to create the files associated to the events.
3764 */
3765static void __trace_early_add_event_dirs(struct trace_array *tr)
3766{
3767	struct trace_event_file *file;
3768	int ret;
3769
3770
3771	list_for_each_entry(file, &tr->events, list) {
3772		ret = event_create_dir(tr->event_dir, file);
3773		if (ret < 0)
3774			pr_warn("Could not create directory for event %s\n",
3775				trace_event_name(file->event_call));
3776	}
3777}
3778
3779/*
3780 * For early boot up, the top trace array and the trace arrays created
3781 * by boot-time tracing require to have a list of events that can be
3782 * enabled. This must be done before the filesystem is set up in order
3783 * to allow events to be traced early.
3784 */
3785void __trace_early_add_events(struct trace_array *tr)
3786{
3787	struct trace_event_call *call;
3788	int ret;
3789
3790	list_for_each_entry(call, &ftrace_events, list) {
3791		/* Early boot up should not have any modules loaded */
3792		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3793		    WARN_ON_ONCE(call->module))
3794			continue;
3795
3796		ret = __trace_early_add_new_event(call, tr);
3797		if (ret < 0)
3798			pr_warn("Could not create early event %s\n",
3799				trace_event_name(call));
3800	}
3801}
3802
3803/* Remove the event directory structure for a trace directory. */
3804static void
3805__trace_remove_event_dirs(struct trace_array *tr)
3806{
3807	struct trace_event_file *file, *next;
3808
3809	list_for_each_entry_safe(file, next, &tr->events, list)
3810		remove_event_file_dir(file);
3811}
3812
3813static void __add_event_to_tracers(struct trace_event_call *call)
3814{
3815	struct trace_array *tr;
3816
3817	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3818		__trace_add_new_event(call, tr);
3819}
3820
3821extern struct trace_event_call *__start_ftrace_events[];
3822extern struct trace_event_call *__stop_ftrace_events[];
3823
3824static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3825
3826static __init int setup_trace_event(char *str)
3827{
3828	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3829	trace_set_ring_buffer_expanded(NULL);
3830	disable_tracing_selftest("running event tracing");
3831
3832	return 1;
3833}
3834__setup("trace_event=", setup_trace_event);
3835
3836static int events_callback(const char *name, umode_t *mode, void **data,
3837			   const struct file_operations **fops)
3838{
3839	if (strcmp(name, "enable") == 0) {
3840		*mode = TRACE_MODE_WRITE;
3841		*fops = &ftrace_tr_enable_fops;
3842		return 1;
3843	}
3844
3845	if (strcmp(name, "header_page") == 0) {
3846		*mode = TRACE_MODE_READ;
3847		*fops = &ftrace_show_header_page_fops;
3848
3849	} else if (strcmp(name, "header_event") == 0) {
3850		*mode = TRACE_MODE_READ;
3851		*fops = &ftrace_show_header_event_fops;
3852	} else
3853		return 0;
3854
3855	return 1;
3856}
3857
3858/* Expects to have event_mutex held when called */
3859static int
3860create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3861{
3862	struct eventfs_inode *e_events;
3863	struct dentry *entry;
3864	int nr_entries;
3865	static struct eventfs_entry events_entries[] = {
3866		{
3867			.name		= "enable",
3868			.callback	= events_callback,
3869		},
3870		{
3871			.name		= "header_page",
3872			.callback	= events_callback,
3873		},
3874		{
3875			.name		= "header_event",
3876			.callback	= events_callback,
3877		},
3878	};
3879
3880	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3881				  tr, &ftrace_set_event_fops);
3882	if (!entry)
3883		return -ENOMEM;
3884
3885	nr_entries = ARRAY_SIZE(events_entries);
3886
3887	e_events = eventfs_create_events_dir("events", parent, events_entries,
3888					     nr_entries, tr);
3889	if (IS_ERR(e_events)) {
3890		pr_warn("Could not create tracefs 'events' directory\n");
3891		return -ENOMEM;
3892	}
3893
3894	/* There are not as crucial, just warn if they are not created */
3895
3896	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3897			  tr, &ftrace_set_event_pid_fops);
3898
3899	trace_create_file("set_event_notrace_pid",
3900			  TRACE_MODE_WRITE, parent, tr,
3901			  &ftrace_set_event_notrace_pid_fops);
3902
3903	tr->event_dir = e_events;
3904
3905	return 0;
3906}
3907
3908/**
3909 * event_trace_add_tracer - add a instance of a trace_array to events
3910 * @parent: The parent dentry to place the files/directories for events in
3911 * @tr: The trace array associated with these events
3912 *
3913 * When a new instance is created, it needs to set up its events
3914 * directory, as well as other files associated with events. It also
3915 * creates the event hierarchy in the @parent/events directory.
3916 *
3917 * Returns 0 on success.
3918 *
3919 * Must be called with event_mutex held.
3920 */
3921int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3922{
3923	int ret;
3924
3925	lockdep_assert_held(&event_mutex);
3926
3927	ret = create_event_toplevel_files(parent, tr);
3928	if (ret)
3929		goto out;
3930
3931	down_write(&trace_event_sem);
3932	/* If tr already has the event list, it is initialized in early boot. */
3933	if (unlikely(!list_empty(&tr->events)))
3934		__trace_early_add_event_dirs(tr);
3935	else
3936		__trace_add_event_dirs(tr);
3937	up_write(&trace_event_sem);
3938
3939 out:
3940	return ret;
3941}
3942
3943/*
3944 * The top trace array already had its file descriptors created.
3945 * Now the files themselves need to be created.
3946 */
3947static __init int
3948early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3949{
3950	int ret;
3951
3952	mutex_lock(&event_mutex);
3953
3954	ret = create_event_toplevel_files(parent, tr);
3955	if (ret)
3956		goto out_unlock;
3957
3958	down_write(&trace_event_sem);
3959	__trace_early_add_event_dirs(tr);
3960	up_write(&trace_event_sem);
3961
3962 out_unlock:
3963	mutex_unlock(&event_mutex);
3964
3965	return ret;
3966}
3967
3968/* Must be called with event_mutex held */
3969int event_trace_del_tracer(struct trace_array *tr)
3970{
3971	lockdep_assert_held(&event_mutex);
3972
3973	/* Disable any event triggers and associated soft-disabled events */
3974	clear_event_triggers(tr);
3975
3976	/* Clear the pid list */
3977	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3978
3979	/* Disable any running events */
3980	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3981
3982	/* Make sure no more events are being executed */
3983	tracepoint_synchronize_unregister();
3984
3985	down_write(&trace_event_sem);
3986	__trace_remove_event_dirs(tr);
3987	eventfs_remove_events_dir(tr->event_dir);
3988	up_write(&trace_event_sem);
3989
3990	tr->event_dir = NULL;
3991
3992	return 0;
3993}
3994
3995static __init int event_trace_memsetup(void)
3996{
3997	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3998	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3999	return 0;
4000}
4001
4002__init void
4003early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4004{
4005	char *token;
4006	int ret;
4007
4008	while (true) {
4009		token = strsep(&buf, ",");
4010
4011		if (!token)
4012			break;
4013
4014		if (*token) {
4015			/* Restarting syscalls requires that we stop them first */
4016			if (disable_first)
4017				ftrace_set_clr_event(tr, token, 0);
4018
4019			ret = ftrace_set_clr_event(tr, token, 1);
4020			if (ret)
4021				pr_warn("Failed to enable trace event: %s\n", token);
4022		}
4023
4024		/* Put back the comma to allow this to be called again */
4025		if (buf)
4026			*(buf - 1) = ',';
4027	}
4028}
4029
4030static __init int event_trace_enable(void)
4031{
4032	struct trace_array *tr = top_trace_array();
4033	struct trace_event_call **iter, *call;
4034	int ret;
4035
4036	if (!tr)
4037		return -ENODEV;
4038
4039	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4040
4041		call = *iter;
4042		ret = event_init(call);
4043		if (!ret)
4044			list_add(&call->list, &ftrace_events);
4045	}
4046
4047	register_trigger_cmds();
4048
4049	/*
4050	 * We need the top trace array to have a working set of trace
4051	 * points at early init, before the debug files and directories
4052	 * are created. Create the file entries now, and attach them
4053	 * to the actual file dentries later.
4054	 */
4055	__trace_early_add_events(tr);
4056
4057	early_enable_events(tr, bootup_event_buf, false);
4058
4059	trace_printk_start_comm();
4060
4061	register_event_cmds();
4062
4063
4064	return 0;
4065}
4066
4067/*
4068 * event_trace_enable() is called from trace_event_init() first to
4069 * initialize events and perhaps start any events that are on the
4070 * command line. Unfortunately, there are some events that will not
4071 * start this early, like the system call tracepoints that need
4072 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4073 * event_trace_enable() is called before pid 1 starts, and this flag
4074 * is never set, making the syscall tracepoint never get reached, but
4075 * the event is enabled regardless (and not doing anything).
4076 */
4077static __init int event_trace_enable_again(void)
4078{
4079	struct trace_array *tr;
4080
4081	tr = top_trace_array();
4082	if (!tr)
4083		return -ENODEV;
4084
4085	early_enable_events(tr, bootup_event_buf, true);
4086
4087	return 0;
4088}
4089
4090early_initcall(event_trace_enable_again);
4091
4092/* Init fields which doesn't related to the tracefs */
4093static __init int event_trace_init_fields(void)
4094{
4095	if (trace_define_generic_fields())
4096		pr_warn("tracing: Failed to allocated generic fields");
4097
4098	if (trace_define_common_fields())
4099		pr_warn("tracing: Failed to allocate common fields");
4100
4101	return 0;
4102}
4103
4104__init int event_trace_init(void)
4105{
4106	struct trace_array *tr;
4107	int ret;
4108
4109	tr = top_trace_array();
4110	if (!tr)
4111		return -ENODEV;
4112
4113	trace_create_file("available_events", TRACE_MODE_READ,
4114			  NULL, tr, &ftrace_avail_fops);
4115
4116	ret = early_event_add_tracer(NULL, tr);
4117	if (ret)
4118		return ret;
4119
4120#ifdef CONFIG_MODULES
4121	ret = register_module_notifier(&trace_module_nb);
4122	if (ret)
4123		pr_warn("Failed to register trace events module notifier\n");
4124#endif
4125
4126	eventdir_initialized = true;
4127
4128	return 0;
4129}
4130
4131void __init trace_event_init(void)
4132{
4133	event_trace_memsetup();
4134	init_ftrace_syscalls();
4135	event_trace_enable();
4136	event_trace_init_fields();
4137}
4138
4139#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4140
4141static DEFINE_SPINLOCK(test_spinlock);
4142static DEFINE_SPINLOCK(test_spinlock_irq);
4143static DEFINE_MUTEX(test_mutex);
4144
4145static __init void test_work(struct work_struct *dummy)
4146{
4147	spin_lock(&test_spinlock);
4148	spin_lock_irq(&test_spinlock_irq);
4149	udelay(1);
4150	spin_unlock_irq(&test_spinlock_irq);
4151	spin_unlock(&test_spinlock);
4152
4153	mutex_lock(&test_mutex);
4154	msleep(1);
4155	mutex_unlock(&test_mutex);
4156}
4157
4158static __init int event_test_thread(void *unused)
4159{
4160	void *test_malloc;
4161
4162	test_malloc = kmalloc(1234, GFP_KERNEL);
4163	if (!test_malloc)
4164		pr_info("failed to kmalloc\n");
4165
4166	schedule_on_each_cpu(test_work);
4167
4168	kfree(test_malloc);
4169
4170	set_current_state(TASK_INTERRUPTIBLE);
4171	while (!kthread_should_stop()) {
4172		schedule();
4173		set_current_state(TASK_INTERRUPTIBLE);
4174	}
4175	__set_current_state(TASK_RUNNING);
4176
4177	return 0;
4178}
4179
4180/*
4181 * Do various things that may trigger events.
4182 */
4183static __init void event_test_stuff(void)
4184{
4185	struct task_struct *test_thread;
4186
4187	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4188	msleep(1);
4189	kthread_stop(test_thread);
4190}
4191
4192/*
4193 * For every trace event defined, we will test each trace point separately,
4194 * and then by groups, and finally all trace points.
4195 */
4196static __init void event_trace_self_tests(void)
4197{
4198	struct trace_subsystem_dir *dir;
4199	struct trace_event_file *file;
4200	struct trace_event_call *call;
4201	struct event_subsystem *system;
4202	struct trace_array *tr;
4203	int ret;
4204
4205	tr = top_trace_array();
4206	if (!tr)
4207		return;
4208
4209	pr_info("Running tests on trace events:\n");
4210
4211	list_for_each_entry(file, &tr->events, list) {
4212
4213		call = file->event_call;
4214
4215		/* Only test those that have a probe */
4216		if (!call->class || !call->class->probe)
4217			continue;
4218
4219/*
4220 * Testing syscall events here is pretty useless, but
4221 * we still do it if configured. But this is time consuming.
4222 * What we really need is a user thread to perform the
4223 * syscalls as we test.
4224 */
4225#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4226		if (call->class->system &&
4227		    strcmp(call->class->system, "syscalls") == 0)
4228			continue;
4229#endif
4230
4231		pr_info("Testing event %s: ", trace_event_name(call));
4232
4233		/*
4234		 * If an event is already enabled, someone is using
4235		 * it and the self test should not be on.
4236		 */
4237		if (file->flags & EVENT_FILE_FL_ENABLED) {
4238			pr_warn("Enabled event during self test!\n");
4239			WARN_ON_ONCE(1);
4240			continue;
4241		}
4242
4243		ftrace_event_enable_disable(file, 1);
4244		event_test_stuff();
4245		ftrace_event_enable_disable(file, 0);
4246
4247		pr_cont("OK\n");
4248	}
4249
4250	/* Now test at the sub system level */
4251
4252	pr_info("Running tests on trace event systems:\n");
4253
4254	list_for_each_entry(dir, &tr->systems, list) {
4255
4256		system = dir->subsystem;
4257
4258		/* the ftrace system is special, skip it */
4259		if (strcmp(system->name, "ftrace") == 0)
4260			continue;
4261
4262		pr_info("Testing event system %s: ", system->name);
4263
4264		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4265		if (WARN_ON_ONCE(ret)) {
4266			pr_warn("error enabling system %s\n",
4267				system->name);
4268			continue;
4269		}
4270
4271		event_test_stuff();
4272
4273		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4274		if (WARN_ON_ONCE(ret)) {
4275			pr_warn("error disabling system %s\n",
4276				system->name);
4277			continue;
4278		}
4279
4280		pr_cont("OK\n");
4281	}
4282
4283	/* Test with all events enabled */
4284
4285	pr_info("Running tests on all trace events:\n");
4286	pr_info("Testing all events: ");
4287
4288	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4289	if (WARN_ON_ONCE(ret)) {
4290		pr_warn("error enabling all events\n");
4291		return;
4292	}
4293
4294	event_test_stuff();
4295
4296	/* reset sysname */
4297	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4298	if (WARN_ON_ONCE(ret)) {
4299		pr_warn("error disabling all events\n");
4300		return;
4301	}
4302
4303	pr_cont("OK\n");
4304}
4305
4306#ifdef CONFIG_FUNCTION_TRACER
4307
4308static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4309
4310static struct trace_event_file event_trace_file __initdata;
4311
4312static void __init
4313function_test_events_call(unsigned long ip, unsigned long parent_ip,
4314			  struct ftrace_ops *op, struct ftrace_regs *regs)
4315{
4316	struct trace_buffer *buffer;
4317	struct ring_buffer_event *event;
4318	struct ftrace_entry *entry;
4319	unsigned int trace_ctx;
4320	long disabled;
4321	int cpu;
4322
4323	trace_ctx = tracing_gen_ctx();
4324	preempt_disable_notrace();
4325	cpu = raw_smp_processor_id();
4326	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4327
4328	if (disabled != 1)
4329		goto out;
4330
4331	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4332						TRACE_FN, sizeof(*entry),
4333						trace_ctx);
4334	if (!event)
4335		goto out;
4336	entry	= ring_buffer_event_data(event);
4337	entry->ip			= ip;
4338	entry->parent_ip		= parent_ip;
4339
4340	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4341				    entry, trace_ctx);
4342 out:
4343	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4344	preempt_enable_notrace();
4345}
4346
4347static struct ftrace_ops trace_ops __initdata  =
4348{
4349	.func = function_test_events_call,
4350};
4351
4352static __init void event_trace_self_test_with_function(void)
4353{
4354	int ret;
4355
4356	event_trace_file.tr = top_trace_array();
4357	if (WARN_ON(!event_trace_file.tr))
4358		return;
4359
4360	ret = register_ftrace_function(&trace_ops);
4361	if (WARN_ON(ret < 0)) {
4362		pr_info("Failed to enable function tracer for event tests\n");
4363		return;
4364	}
4365	pr_info("Running tests again, along with the function tracer\n");
4366	event_trace_self_tests();
4367	unregister_ftrace_function(&trace_ops);
4368}
4369#else
4370static __init void event_trace_self_test_with_function(void)
4371{
4372}
4373#endif
4374
4375static __init int event_trace_self_tests_init(void)
4376{
4377	if (!tracing_selftest_disabled) {
4378		event_trace_self_tests();
4379		event_trace_self_test_with_function();
4380	}
4381
4382	return 0;
4383}
4384
4385late_initcall(event_trace_self_tests_init);
4386
4387#endif