Linux Audio

Check our new training course

Loading...
v4.17
   1/*
   2 * event tracer
   3 *
   4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5 *
   6 *  - Added format output of fields of the trace point.
   7 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8 *
   9 */
  10
  11#define pr_fmt(fmt) fmt
  12
  13#include <linux/workqueue.h>
  14#include <linux/spinlock.h>
  15#include <linux/kthread.h>
  16#include <linux/tracefs.h>
  17#include <linux/uaccess.h>
  18#include <linux/module.h>
  19#include <linux/ctype.h>
  20#include <linux/sort.h>
  21#include <linux/slab.h>
  22#include <linux/delay.h>
  23
  24#include <trace/events/sched.h>
  25
  26#include <asm/setup.h>
  27
  28#include "trace_output.h"
  29
  30#undef TRACE_SYSTEM
  31#define TRACE_SYSTEM "TRACE_SYSTEM"
  32
  33DEFINE_MUTEX(event_mutex);
  34
  35LIST_HEAD(ftrace_events);
  36static LIST_HEAD(ftrace_generic_fields);
  37static LIST_HEAD(ftrace_common_fields);
  38
  39#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41static struct kmem_cache *field_cachep;
  42static struct kmem_cache *file_cachep;
  43
  44static inline int system_refcount(struct event_subsystem *system)
  45{
  46	return system->ref_count;
  47}
  48
  49static int system_refcount_inc(struct event_subsystem *system)
  50{
  51	return system->ref_count++;
  52}
  53
  54static int system_refcount_dec(struct event_subsystem *system)
  55{
  56	return --system->ref_count;
  57}
  58
  59/* Double loops, do not use break, only goto's work */
  60#define do_for_each_event_file(tr, file)			\
  61	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  62		list_for_each_entry(file, &tr->events, list)
  63
  64#define do_for_each_event_file_safe(tr, file)			\
  65	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  66		struct trace_event_file *___n;				\
  67		list_for_each_entry_safe(file, ___n, &tr->events, list)
  68
  69#define while_for_each_event_file()		\
  70	}
  71
  72static struct list_head *
  73trace_get_fields(struct trace_event_call *event_call)
  74{
  75	if (!event_call->class->get_fields)
  76		return &event_call->class->fields;
  77	return event_call->class->get_fields(event_call);
  78}
  79
  80static struct ftrace_event_field *
  81__find_event_field(struct list_head *head, char *name)
  82{
  83	struct ftrace_event_field *field;
  84
  85	list_for_each_entry(field, head, link) {
  86		if (!strcmp(field->name, name))
  87			return field;
  88	}
  89
  90	return NULL;
  91}
  92
  93struct ftrace_event_field *
  94trace_find_event_field(struct trace_event_call *call, char *name)
  95{
  96	struct ftrace_event_field *field;
  97	struct list_head *head;
  98
  99	head = trace_get_fields(call);
 100	field = __find_event_field(head, name);
 101	if (field)
 102		return field;
 103
 104	field = __find_event_field(&ftrace_generic_fields, name);
 105	if (field)
 106		return field;
 107
 108	return __find_event_field(&ftrace_common_fields, name);
 109}
 110
 111static int __trace_define_field(struct list_head *head, const char *type,
 112				const char *name, int offset, int size,
 113				int is_signed, int filter_type)
 114{
 115	struct ftrace_event_field *field;
 116
 117	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 118	if (!field)
 119		return -ENOMEM;
 120
 121	field->name = name;
 122	field->type = type;
 123
 124	if (filter_type == FILTER_OTHER)
 125		field->filter_type = filter_assign_type(type);
 126	else
 127		field->filter_type = filter_type;
 128
 129	field->offset = offset;
 130	field->size = size;
 131	field->is_signed = is_signed;
 132
 133	list_add(&field->link, head);
 134
 135	return 0;
 136}
 137
 138int trace_define_field(struct trace_event_call *call, const char *type,
 139		       const char *name, int offset, int size, int is_signed,
 140		       int filter_type)
 141{
 142	struct list_head *head;
 143
 144	if (WARN_ON(!call->class))
 145		return 0;
 146
 147	head = trace_get_fields(call);
 148	return __trace_define_field(head, type, name, offset, size,
 149				    is_signed, filter_type);
 150}
 151EXPORT_SYMBOL_GPL(trace_define_field);
 152
 153#define __generic_field(type, item, filter_type)			\
 154	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 155				   #item, 0, 0, is_signed_type(type),	\
 156				   filter_type);			\
 157	if (ret)							\
 158		return ret;
 159
 160#define __common_field(type, item)					\
 161	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 162				   "common_" #item,			\
 163				   offsetof(typeof(ent), item),		\
 164				   sizeof(ent.item),			\
 165				   is_signed_type(type), FILTER_OTHER);	\
 166	if (ret)							\
 167		return ret;
 168
 169static int trace_define_generic_fields(void)
 170{
 171	int ret;
 172
 173	__generic_field(int, CPU, FILTER_CPU);
 174	__generic_field(int, cpu, FILTER_CPU);
 175	__generic_field(char *, COMM, FILTER_COMM);
 176	__generic_field(char *, comm, FILTER_COMM);
 177
 178	return ret;
 179}
 180
 181static int trace_define_common_fields(void)
 182{
 183	int ret;
 184	struct trace_entry ent;
 185
 186	__common_field(unsigned short, type);
 187	__common_field(unsigned char, flags);
 188	__common_field(unsigned char, preempt_count);
 189	__common_field(int, pid);
 190
 191	return ret;
 192}
 193
 194static void trace_destroy_fields(struct trace_event_call *call)
 195{
 196	struct ftrace_event_field *field, *next;
 197	struct list_head *head;
 198
 199	head = trace_get_fields(call);
 200	list_for_each_entry_safe(field, next, head, link) {
 201		list_del(&field->link);
 202		kmem_cache_free(field_cachep, field);
 203	}
 204}
 205
 206/*
 207 * run-time version of trace_event_get_offsets_<call>() that returns the last
 208 * accessible offset of trace fields excluding __dynamic_array bytes
 209 */
 210int trace_event_get_offsets(struct trace_event_call *call)
 211{
 212	struct ftrace_event_field *tail;
 213	struct list_head *head;
 214
 215	head = trace_get_fields(call);
 216	/*
 217	 * head->next points to the last field with the largest offset,
 218	 * since it was added last by trace_define_field()
 219	 */
 220	tail = list_first_entry(head, struct ftrace_event_field, link);
 221	return tail->offset + tail->size;
 222}
 223
 224int trace_event_raw_init(struct trace_event_call *call)
 225{
 226	int id;
 227
 228	id = register_trace_event(&call->event);
 229	if (!id)
 230		return -ENODEV;
 231
 232	return 0;
 233}
 234EXPORT_SYMBOL_GPL(trace_event_raw_init);
 235
 236bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 237{
 238	struct trace_array *tr = trace_file->tr;
 239	struct trace_array_cpu *data;
 240	struct trace_pid_list *pid_list;
 241
 242	pid_list = rcu_dereference_sched(tr->filtered_pids);
 243	if (!pid_list)
 244		return false;
 245
 246	data = this_cpu_ptr(tr->trace_buffer.data);
 247
 248	return data->ignore_pid;
 249}
 250EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 251
 252void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 253				 struct trace_event_file *trace_file,
 254				 unsigned long len)
 255{
 256	struct trace_event_call *event_call = trace_file->event_call;
 257
 258	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 259	    trace_event_ignore_this_pid(trace_file))
 260		return NULL;
 261
 262	local_save_flags(fbuffer->flags);
 263	fbuffer->pc = preempt_count();
 264	/*
 265	 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
 266	 * preemption (adding one to the preempt_count). Since we are
 267	 * interested in the preempt_count at the time the tracepoint was
 268	 * hit, we need to subtract one to offset the increment.
 269	 */
 270	if (IS_ENABLED(CONFIG_PREEMPT))
 271		fbuffer->pc--;
 272	fbuffer->trace_file = trace_file;
 273
 274	fbuffer->event =
 275		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 276						event_call->event.type, len,
 277						fbuffer->flags, fbuffer->pc);
 278	if (!fbuffer->event)
 279		return NULL;
 280
 281	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 282	return fbuffer->entry;
 283}
 284EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 285
 286int trace_event_reg(struct trace_event_call *call,
 287		    enum trace_reg type, void *data)
 288{
 289	struct trace_event_file *file = data;
 290
 291	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 292	switch (type) {
 293	case TRACE_REG_REGISTER:
 294		return tracepoint_probe_register(call->tp,
 295						 call->class->probe,
 296						 file);
 297	case TRACE_REG_UNREGISTER:
 298		tracepoint_probe_unregister(call->tp,
 299					    call->class->probe,
 300					    file);
 301		return 0;
 302
 303#ifdef CONFIG_PERF_EVENTS
 304	case TRACE_REG_PERF_REGISTER:
 305		return tracepoint_probe_register(call->tp,
 306						 call->class->perf_probe,
 307						 call);
 308	case TRACE_REG_PERF_UNREGISTER:
 309		tracepoint_probe_unregister(call->tp,
 310					    call->class->perf_probe,
 311					    call);
 312		return 0;
 313	case TRACE_REG_PERF_OPEN:
 314	case TRACE_REG_PERF_CLOSE:
 315	case TRACE_REG_PERF_ADD:
 316	case TRACE_REG_PERF_DEL:
 317		return 0;
 318#endif
 319	}
 320	return 0;
 321}
 322EXPORT_SYMBOL_GPL(trace_event_reg);
 323
 324void trace_event_enable_cmd_record(bool enable)
 325{
 326	struct trace_event_file *file;
 327	struct trace_array *tr;
 328
 329	mutex_lock(&event_mutex);
 330	do_for_each_event_file(tr, file) {
 331
 332		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 333			continue;
 334
 335		if (enable) {
 336			tracing_start_cmdline_record();
 337			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338		} else {
 339			tracing_stop_cmdline_record();
 340			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 341		}
 342	} while_for_each_event_file();
 343	mutex_unlock(&event_mutex);
 344}
 345
 346void trace_event_enable_tgid_record(bool enable)
 347{
 348	struct trace_event_file *file;
 349	struct trace_array *tr;
 350
 351	mutex_lock(&event_mutex);
 352	do_for_each_event_file(tr, file) {
 353		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 354			continue;
 355
 356		if (enable) {
 357			tracing_start_tgid_record();
 358			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 359		} else {
 360			tracing_stop_tgid_record();
 361			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 362				  &file->flags);
 363		}
 364	} while_for_each_event_file();
 365	mutex_unlock(&event_mutex);
 366}
 367
 368static int __ftrace_event_enable_disable(struct trace_event_file *file,
 369					 int enable, int soft_disable)
 370{
 371	struct trace_event_call *call = file->event_call;
 372	struct trace_array *tr = file->tr;
 373	unsigned long file_flags = file->flags;
 374	int ret = 0;
 375	int disable;
 376
 377	switch (enable) {
 378	case 0:
 379		/*
 380		 * When soft_disable is set and enable is cleared, the sm_ref
 381		 * reference counter is decremented. If it reaches 0, we want
 382		 * to clear the SOFT_DISABLED flag but leave the event in the
 383		 * state that it was. That is, if the event was enabled and
 384		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 385		 * is set we do not want the event to be enabled before we
 386		 * clear the bit.
 387		 *
 388		 * When soft_disable is not set but the SOFT_MODE flag is,
 389		 * we do nothing. Do not disable the tracepoint, otherwise
 390		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 391		 */
 392		if (soft_disable) {
 393			if (atomic_dec_return(&file->sm_ref) > 0)
 394				break;
 395			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 396			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 397		} else
 398			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 399
 400		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 401			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 402			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 403				tracing_stop_cmdline_record();
 404				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 405			}
 406
 407			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 408				tracing_stop_tgid_record();
 409				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 410			}
 411
 412			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 413		}
 414		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 415		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 416			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 417		else
 418			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 419		break;
 420	case 1:
 421		/*
 422		 * When soft_disable is set and enable is set, we want to
 423		 * register the tracepoint for the event, but leave the event
 424		 * as is. That means, if the event was already enabled, we do
 425		 * nothing (but set SOFT_MODE). If the event is disabled, we
 426		 * set SOFT_DISABLED before enabling the event tracepoint, so
 427		 * it still seems to be disabled.
 428		 */
 429		if (!soft_disable)
 430			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 431		else {
 432			if (atomic_inc_return(&file->sm_ref) > 1)
 433				break;
 434			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 435		}
 436
 437		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 438			bool cmd = false, tgid = false;
 439
 440			/* Keep the event disabled, when going to SOFT_MODE. */
 441			if (soft_disable)
 442				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 443
 444			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 445				cmd = true;
 446				tracing_start_cmdline_record();
 447				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 448			}
 449
 450			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 451				tgid = true;
 452				tracing_start_tgid_record();
 453				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 454			}
 455
 456			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 457			if (ret) {
 458				if (cmd)
 459					tracing_stop_cmdline_record();
 460				if (tgid)
 461					tracing_stop_tgid_record();
 462				pr_info("event trace: Could not enable event "
 463					"%s\n", trace_event_name(call));
 464				break;
 465			}
 466			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 467
 468			/* WAS_ENABLED gets set but never cleared. */
 469			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 470		}
 471		break;
 472	}
 473
 474	/* Enable or disable use of trace_buffered_event */
 475	if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 476	    (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 477		if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 478			trace_buffered_event_enable();
 479		else
 480			trace_buffered_event_disable();
 481	}
 482
 483	return ret;
 484}
 485
 486int trace_event_enable_disable(struct trace_event_file *file,
 487			       int enable, int soft_disable)
 488{
 489	return __ftrace_event_enable_disable(file, enable, soft_disable);
 490}
 491
 492static int ftrace_event_enable_disable(struct trace_event_file *file,
 493				       int enable)
 494{
 495	return __ftrace_event_enable_disable(file, enable, 0);
 496}
 497
 498static void ftrace_clear_events(struct trace_array *tr)
 499{
 500	struct trace_event_file *file;
 501
 502	mutex_lock(&event_mutex);
 503	list_for_each_entry(file, &tr->events, list) {
 504		ftrace_event_enable_disable(file, 0);
 505	}
 506	mutex_unlock(&event_mutex);
 507}
 508
 509static void
 510event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 511{
 512	struct trace_pid_list *pid_list;
 513	struct trace_array *tr = data;
 514
 515	pid_list = rcu_dereference_sched(tr->filtered_pids);
 516	trace_filter_add_remove_task(pid_list, NULL, task);
 517}
 518
 519static void
 520event_filter_pid_sched_process_fork(void *data,
 521				    struct task_struct *self,
 522				    struct task_struct *task)
 523{
 524	struct trace_pid_list *pid_list;
 525	struct trace_array *tr = data;
 526
 527	pid_list = rcu_dereference_sched(tr->filtered_pids);
 528	trace_filter_add_remove_task(pid_list, self, task);
 529}
 530
 531void trace_event_follow_fork(struct trace_array *tr, bool enable)
 532{
 533	if (enable) {
 534		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 535						       tr, INT_MIN);
 536		register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 537						       tr, INT_MAX);
 538	} else {
 539		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 540						    tr);
 541		unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 542						    tr);
 543	}
 544}
 545
 546static void
 547event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 548		    struct task_struct *prev, struct task_struct *next)
 549{
 550	struct trace_array *tr = data;
 551	struct trace_pid_list *pid_list;
 552
 553	pid_list = rcu_dereference_sched(tr->filtered_pids);
 554
 555	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 556		       trace_ignore_this_task(pid_list, prev) &&
 557		       trace_ignore_this_task(pid_list, next));
 558}
 559
 560static void
 561event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 562		    struct task_struct *prev, struct task_struct *next)
 563{
 564	struct trace_array *tr = data;
 565	struct trace_pid_list *pid_list;
 566
 567	pid_list = rcu_dereference_sched(tr->filtered_pids);
 568
 569	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 570		       trace_ignore_this_task(pid_list, next));
 571}
 572
 573static void
 574event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 575{
 576	struct trace_array *tr = data;
 577	struct trace_pid_list *pid_list;
 578
 579	/* Nothing to do if we are already tracing */
 580	if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 581		return;
 582
 583	pid_list = rcu_dereference_sched(tr->filtered_pids);
 584
 585	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 586		       trace_ignore_this_task(pid_list, task));
 587}
 588
 589static void
 590event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 591{
 592	struct trace_array *tr = data;
 593	struct trace_pid_list *pid_list;
 594
 595	/* Nothing to do if we are not tracing */
 596	if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 597		return;
 598
 599	pid_list = rcu_dereference_sched(tr->filtered_pids);
 600
 601	/* Set tracing if current is enabled */
 602	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 603		       trace_ignore_this_task(pid_list, current));
 604}
 605
 606static void __ftrace_clear_event_pids(struct trace_array *tr)
 607{
 608	struct trace_pid_list *pid_list;
 609	struct trace_event_file *file;
 610	int cpu;
 611
 612	pid_list = rcu_dereference_protected(tr->filtered_pids,
 613					     lockdep_is_held(&event_mutex));
 614	if (!pid_list)
 615		return;
 616
 617	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 618	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 619
 620	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 621	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 622
 623	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 624	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 625
 626	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 627	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 628
 629	list_for_each_entry(file, &tr->events, list) {
 630		clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 631	}
 632
 633	for_each_possible_cpu(cpu)
 634		per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 635
 636	rcu_assign_pointer(tr->filtered_pids, NULL);
 637
 638	/* Wait till all users are no longer using pid filtering */
 639	synchronize_sched();
 640
 641	trace_free_pid_list(pid_list);
 642}
 643
 644static void ftrace_clear_event_pids(struct trace_array *tr)
 645{
 646	mutex_lock(&event_mutex);
 647	__ftrace_clear_event_pids(tr);
 648	mutex_unlock(&event_mutex);
 649}
 650
 651static void __put_system(struct event_subsystem *system)
 652{
 653	struct event_filter *filter = system->filter;
 654
 655	WARN_ON_ONCE(system_refcount(system) == 0);
 656	if (system_refcount_dec(system))
 657		return;
 658
 659	list_del(&system->list);
 660
 661	if (filter) {
 662		kfree(filter->filter_string);
 663		kfree(filter);
 664	}
 665	kfree_const(system->name);
 666	kfree(system);
 667}
 668
 669static void __get_system(struct event_subsystem *system)
 670{
 671	WARN_ON_ONCE(system_refcount(system) == 0);
 672	system_refcount_inc(system);
 673}
 674
 675static void __get_system_dir(struct trace_subsystem_dir *dir)
 676{
 677	WARN_ON_ONCE(dir->ref_count == 0);
 678	dir->ref_count++;
 679	__get_system(dir->subsystem);
 680}
 681
 682static void __put_system_dir(struct trace_subsystem_dir *dir)
 683{
 684	WARN_ON_ONCE(dir->ref_count == 0);
 685	/* If the subsystem is about to be freed, the dir must be too */
 686	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 687
 688	__put_system(dir->subsystem);
 689	if (!--dir->ref_count)
 690		kfree(dir);
 691}
 692
 693static void put_system(struct trace_subsystem_dir *dir)
 694{
 695	mutex_lock(&event_mutex);
 696	__put_system_dir(dir);
 697	mutex_unlock(&event_mutex);
 698}
 699
 700static void remove_subsystem(struct trace_subsystem_dir *dir)
 701{
 702	if (!dir)
 703		return;
 704
 705	if (!--dir->nr_events) {
 706		tracefs_remove_recursive(dir->entry);
 707		list_del(&dir->list);
 708		__put_system_dir(dir);
 709	}
 710}
 711
 712static void remove_event_file_dir(struct trace_event_file *file)
 713{
 714	struct dentry *dir = file->dir;
 715	struct dentry *child;
 716
 717	if (dir) {
 718		spin_lock(&dir->d_lock);	/* probably unneeded */
 719		list_for_each_entry(child, &dir->d_subdirs, d_child) {
 720			if (d_really_is_positive(child))	/* probably unneeded */
 721				d_inode(child)->i_private = NULL;
 722		}
 723		spin_unlock(&dir->d_lock);
 724
 725		tracefs_remove_recursive(dir);
 726	}
 727
 728	list_del(&file->list);
 729	remove_subsystem(file->system);
 730	free_event_filter(file->filter);
 731	kmem_cache_free(file_cachep, file);
 732}
 733
 734/*
 735 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 736 */
 737static int
 738__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 739			      const char *sub, const char *event, int set)
 740{
 741	struct trace_event_file *file;
 742	struct trace_event_call *call;
 743	const char *name;
 744	int ret = -EINVAL;
 745	int eret = 0;
 746
 747	list_for_each_entry(file, &tr->events, list) {
 748
 749		call = file->event_call;
 750		name = trace_event_name(call);
 751
 752		if (!name || !call->class || !call->class->reg)
 753			continue;
 754
 755		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 756			continue;
 757
 758		if (match &&
 759		    strcmp(match, name) != 0 &&
 760		    strcmp(match, call->class->system) != 0)
 761			continue;
 762
 763		if (sub && strcmp(sub, call->class->system) != 0)
 764			continue;
 765
 766		if (event && strcmp(event, name) != 0)
 767			continue;
 768
 769		ret = ftrace_event_enable_disable(file, set);
 770
 771		/*
 772		 * Save the first error and return that. Some events
 773		 * may still have been enabled, but let the user
 774		 * know that something went wrong.
 775		 */
 776		if (ret && !eret)
 777			eret = ret;
 778
 779		ret = eret;
 780	}
 781
 782	return ret;
 783}
 784
 785static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 786				  const char *sub, const char *event, int set)
 787{
 788	int ret;
 789
 790	mutex_lock(&event_mutex);
 791	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 792	mutex_unlock(&event_mutex);
 793
 794	return ret;
 795}
 796
 797static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 798{
 799	char *event = NULL, *sub = NULL, *match;
 800	int ret;
 801
 802	/*
 803	 * The buf format can be <subsystem>:<event-name>
 804	 *  *:<event-name> means any event by that name.
 805	 *  :<event-name> is the same.
 806	 *
 807	 *  <subsystem>:* means all events in that subsystem
 808	 *  <subsystem>: means the same.
 809	 *
 810	 *  <name> (no ':') means all events in a subsystem with
 811	 *  the name <name> or any event that matches <name>
 812	 */
 813
 814	match = strsep(&buf, ":");
 815	if (buf) {
 816		sub = match;
 817		event = buf;
 818		match = NULL;
 819
 820		if (!strlen(sub) || strcmp(sub, "*") == 0)
 821			sub = NULL;
 822		if (!strlen(event) || strcmp(event, "*") == 0)
 823			event = NULL;
 824	}
 825
 826	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 827
 828	/* Put back the colon to allow this to be called again */
 829	if (buf)
 830		*(buf - 1) = ':';
 831
 832	return ret;
 833}
 834
 835/**
 836 * trace_set_clr_event - enable or disable an event
 837 * @system: system name to match (NULL for any system)
 838 * @event: event name to match (NULL for all events, within system)
 839 * @set: 1 to enable, 0 to disable
 840 *
 841 * This is a way for other parts of the kernel to enable or disable
 842 * event recording.
 843 *
 844 * Returns 0 on success, -EINVAL if the parameters do not match any
 845 * registered events.
 846 */
 847int trace_set_clr_event(const char *system, const char *event, int set)
 848{
 849	struct trace_array *tr = top_trace_array();
 850
 851	if (!tr)
 852		return -ENODEV;
 853
 854	return __ftrace_set_clr_event(tr, NULL, system, event, set);
 855}
 856EXPORT_SYMBOL_GPL(trace_set_clr_event);
 857
 858/* 128 should be much more than enough */
 859#define EVENT_BUF_SIZE		127
 860
 861static ssize_t
 862ftrace_event_write(struct file *file, const char __user *ubuf,
 863		   size_t cnt, loff_t *ppos)
 864{
 865	struct trace_parser parser;
 866	struct seq_file *m = file->private_data;
 867	struct trace_array *tr = m->private;
 868	ssize_t read, ret;
 869
 870	if (!cnt)
 871		return 0;
 872
 873	ret = tracing_update_buffers();
 874	if (ret < 0)
 875		return ret;
 876
 877	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 878		return -ENOMEM;
 879
 880	read = trace_get_user(&parser, ubuf, cnt, ppos);
 881
 882	if (read >= 0 && trace_parser_loaded((&parser))) {
 883		int set = 1;
 884
 885		if (*parser.buffer == '!')
 886			set = 0;
 887
 
 
 888		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 889		if (ret)
 890			goto out_put;
 891	}
 892
 893	ret = read;
 894
 895 out_put:
 896	trace_parser_put(&parser);
 897
 898	return ret;
 899}
 900
 901static void *
 902t_next(struct seq_file *m, void *v, loff_t *pos)
 903{
 904	struct trace_event_file *file = v;
 905	struct trace_event_call *call;
 906	struct trace_array *tr = m->private;
 907
 908	(*pos)++;
 909
 910	list_for_each_entry_continue(file, &tr->events, list) {
 911		call = file->event_call;
 912		/*
 913		 * The ftrace subsystem is for showing formats only.
 914		 * They can not be enabled or disabled via the event files.
 915		 */
 916		if (call->class && call->class->reg &&
 917		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 918			return file;
 919	}
 920
 921	return NULL;
 922}
 923
 924static void *t_start(struct seq_file *m, loff_t *pos)
 925{
 926	struct trace_event_file *file;
 927	struct trace_array *tr = m->private;
 928	loff_t l;
 929
 930	mutex_lock(&event_mutex);
 931
 932	file = list_entry(&tr->events, struct trace_event_file, list);
 933	for (l = 0; l <= *pos; ) {
 934		file = t_next(m, file, &l);
 935		if (!file)
 936			break;
 937	}
 938	return file;
 939}
 940
 941static void *
 942s_next(struct seq_file *m, void *v, loff_t *pos)
 943{
 944	struct trace_event_file *file = v;
 945	struct trace_array *tr = m->private;
 946
 947	(*pos)++;
 948
 949	list_for_each_entry_continue(file, &tr->events, list) {
 950		if (file->flags & EVENT_FILE_FL_ENABLED)
 951			return file;
 952	}
 953
 954	return NULL;
 955}
 956
 957static void *s_start(struct seq_file *m, loff_t *pos)
 958{
 959	struct trace_event_file *file;
 960	struct trace_array *tr = m->private;
 961	loff_t l;
 962
 963	mutex_lock(&event_mutex);
 964
 965	file = list_entry(&tr->events, struct trace_event_file, list);
 966	for (l = 0; l <= *pos; ) {
 967		file = s_next(m, file, &l);
 968		if (!file)
 969			break;
 970	}
 971	return file;
 972}
 973
 974static int t_show(struct seq_file *m, void *v)
 975{
 976	struct trace_event_file *file = v;
 977	struct trace_event_call *call = file->event_call;
 978
 979	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 980		seq_printf(m, "%s:", call->class->system);
 981	seq_printf(m, "%s\n", trace_event_name(call));
 982
 983	return 0;
 984}
 985
 986static void t_stop(struct seq_file *m, void *p)
 987{
 988	mutex_unlock(&event_mutex);
 989}
 990
 991static void *
 992p_next(struct seq_file *m, void *v, loff_t *pos)
 993{
 994	struct trace_array *tr = m->private;
 995	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 996
 997	return trace_pid_next(pid_list, v, pos);
 998}
 999
1000static void *p_start(struct seq_file *m, loff_t *pos)
1001	__acquires(RCU)
1002{
1003	struct trace_pid_list *pid_list;
1004	struct trace_array *tr = m->private;
1005
1006	/*
1007	 * Grab the mutex, to keep calls to p_next() having the same
1008	 * tr->filtered_pids as p_start() has.
1009	 * If we just passed the tr->filtered_pids around, then RCU would
1010	 * have been enough, but doing that makes things more complex.
1011	 */
1012	mutex_lock(&event_mutex);
1013	rcu_read_lock_sched();
1014
1015	pid_list = rcu_dereference_sched(tr->filtered_pids);
1016
1017	if (!pid_list)
1018		return NULL;
1019
1020	return trace_pid_start(pid_list, pos);
1021}
1022
1023static void p_stop(struct seq_file *m, void *p)
1024	__releases(RCU)
1025{
1026	rcu_read_unlock_sched();
1027	mutex_unlock(&event_mutex);
1028}
1029
1030static ssize_t
1031event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1032		  loff_t *ppos)
1033{
1034	struct trace_event_file *file;
1035	unsigned long flags;
1036	char buf[4] = "0";
1037
1038	mutex_lock(&event_mutex);
1039	file = event_file_data(filp);
1040	if (likely(file))
1041		flags = file->flags;
1042	mutex_unlock(&event_mutex);
1043
1044	if (!file)
1045		return -ENODEV;
1046
1047	if (flags & EVENT_FILE_FL_ENABLED &&
1048	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1049		strcpy(buf, "1");
1050
1051	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1052	    flags & EVENT_FILE_FL_SOFT_MODE)
1053		strcat(buf, "*");
1054
1055	strcat(buf, "\n");
1056
1057	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1058}
1059
1060static ssize_t
1061event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1062		   loff_t *ppos)
1063{
1064	struct trace_event_file *file;
1065	unsigned long val;
1066	int ret;
1067
1068	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1069	if (ret)
1070		return ret;
1071
1072	ret = tracing_update_buffers();
1073	if (ret < 0)
1074		return ret;
1075
1076	switch (val) {
1077	case 0:
1078	case 1:
1079		ret = -ENODEV;
1080		mutex_lock(&event_mutex);
1081		file = event_file_data(filp);
1082		if (likely(file))
1083			ret = ftrace_event_enable_disable(file, val);
1084		mutex_unlock(&event_mutex);
1085		break;
1086
1087	default:
1088		return -EINVAL;
1089	}
1090
1091	*ppos += cnt;
1092
1093	return ret ? ret : cnt;
1094}
1095
1096static ssize_t
1097system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1098		   loff_t *ppos)
1099{
1100	const char set_to_char[4] = { '?', '0', '1', 'X' };
1101	struct trace_subsystem_dir *dir = filp->private_data;
1102	struct event_subsystem *system = dir->subsystem;
1103	struct trace_event_call *call;
1104	struct trace_event_file *file;
1105	struct trace_array *tr = dir->tr;
1106	char buf[2];
1107	int set = 0;
1108	int ret;
1109
1110	mutex_lock(&event_mutex);
1111	list_for_each_entry(file, &tr->events, list) {
1112		call = file->event_call;
1113		if (!trace_event_name(call) || !call->class || !call->class->reg)
1114			continue;
1115
1116		if (system && strcmp(call->class->system, system->name) != 0)
1117			continue;
1118
1119		/*
1120		 * We need to find out if all the events are set
1121		 * or if all events or cleared, or if we have
1122		 * a mixture.
1123		 */
1124		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1125
1126		/*
1127		 * If we have a mixture, no need to look further.
1128		 */
1129		if (set == 3)
1130			break;
1131	}
1132	mutex_unlock(&event_mutex);
1133
1134	buf[0] = set_to_char[set];
1135	buf[1] = '\n';
1136
1137	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1138
1139	return ret;
1140}
1141
1142static ssize_t
1143system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1144		    loff_t *ppos)
1145{
1146	struct trace_subsystem_dir *dir = filp->private_data;
1147	struct event_subsystem *system = dir->subsystem;
1148	const char *name = NULL;
1149	unsigned long val;
1150	ssize_t ret;
1151
1152	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1153	if (ret)
1154		return ret;
1155
1156	ret = tracing_update_buffers();
1157	if (ret < 0)
1158		return ret;
1159
1160	if (val != 0 && val != 1)
1161		return -EINVAL;
1162
1163	/*
1164	 * Opening of "enable" adds a ref count to system,
1165	 * so the name is safe to use.
1166	 */
1167	if (system)
1168		name = system->name;
1169
1170	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1171	if (ret)
1172		goto out;
1173
1174	ret = cnt;
1175
1176out:
1177	*ppos += cnt;
1178
1179	return ret;
1180}
1181
1182enum {
1183	FORMAT_HEADER		= 1,
1184	FORMAT_FIELD_SEPERATOR	= 2,
1185	FORMAT_PRINTFMT		= 3,
1186};
1187
1188static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1189{
1190	struct trace_event_call *call = event_file_data(m->private);
1191	struct list_head *common_head = &ftrace_common_fields;
1192	struct list_head *head = trace_get_fields(call);
1193	struct list_head *node = v;
1194
1195	(*pos)++;
1196
1197	switch ((unsigned long)v) {
1198	case FORMAT_HEADER:
1199		node = common_head;
1200		break;
1201
1202	case FORMAT_FIELD_SEPERATOR:
1203		node = head;
1204		break;
1205
1206	case FORMAT_PRINTFMT:
1207		/* all done */
1208		return NULL;
1209	}
1210
1211	node = node->prev;
1212	if (node == common_head)
1213		return (void *)FORMAT_FIELD_SEPERATOR;
1214	else if (node == head)
1215		return (void *)FORMAT_PRINTFMT;
1216	else
1217		return node;
1218}
1219
1220static int f_show(struct seq_file *m, void *v)
1221{
1222	struct trace_event_call *call = event_file_data(m->private);
1223	struct ftrace_event_field *field;
1224	const char *array_descriptor;
1225
1226	switch ((unsigned long)v) {
1227	case FORMAT_HEADER:
1228		seq_printf(m, "name: %s\n", trace_event_name(call));
1229		seq_printf(m, "ID: %d\n", call->event.type);
1230		seq_puts(m, "format:\n");
1231		return 0;
1232
1233	case FORMAT_FIELD_SEPERATOR:
1234		seq_putc(m, '\n');
1235		return 0;
1236
1237	case FORMAT_PRINTFMT:
1238		seq_printf(m, "\nprint fmt: %s\n",
1239			   call->print_fmt);
1240		return 0;
1241	}
1242
1243	field = list_entry(v, struct ftrace_event_field, link);
1244	/*
1245	 * Smartly shows the array type(except dynamic array).
1246	 * Normal:
1247	 *	field:TYPE VAR
1248	 * If TYPE := TYPE[LEN], it is shown:
1249	 *	field:TYPE VAR[LEN]
1250	 */
1251	array_descriptor = strchr(field->type, '[');
1252
1253	if (!strncmp(field->type, "__data_loc", 10))
1254		array_descriptor = NULL;
1255
1256	if (!array_descriptor)
1257		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1258			   field->type, field->name, field->offset,
1259			   field->size, !!field->is_signed);
1260	else
1261		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1262			   (int)(array_descriptor - field->type),
1263			   field->type, field->name,
1264			   array_descriptor, field->offset,
1265			   field->size, !!field->is_signed);
1266
1267	return 0;
1268}
1269
1270static void *f_start(struct seq_file *m, loff_t *pos)
1271{
1272	void *p = (void *)FORMAT_HEADER;
1273	loff_t l = 0;
1274
1275	/* ->stop() is called even if ->start() fails */
1276	mutex_lock(&event_mutex);
1277	if (!event_file_data(m->private))
1278		return ERR_PTR(-ENODEV);
1279
1280	while (l < *pos && p)
1281		p = f_next(m, p, &l);
1282
1283	return p;
1284}
1285
1286static void f_stop(struct seq_file *m, void *p)
1287{
1288	mutex_unlock(&event_mutex);
1289}
1290
1291static const struct seq_operations trace_format_seq_ops = {
1292	.start		= f_start,
1293	.next		= f_next,
1294	.stop		= f_stop,
1295	.show		= f_show,
1296};
1297
1298static int trace_format_open(struct inode *inode, struct file *file)
1299{
1300	struct seq_file *m;
1301	int ret;
1302
1303	ret = seq_open(file, &trace_format_seq_ops);
1304	if (ret < 0)
1305		return ret;
1306
1307	m = file->private_data;
1308	m->private = file;
1309
1310	return 0;
1311}
1312
1313static ssize_t
1314event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1315{
1316	int id = (long)event_file_data(filp);
1317	char buf[32];
1318	int len;
1319
1320	if (*ppos)
1321		return 0;
1322
1323	if (unlikely(!id))
1324		return -ENODEV;
1325
1326	len = sprintf(buf, "%d\n", id);
1327
1328	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1329}
1330
1331static ssize_t
1332event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1333		  loff_t *ppos)
1334{
1335	struct trace_event_file *file;
1336	struct trace_seq *s;
1337	int r = -ENODEV;
1338
1339	if (*ppos)
1340		return 0;
1341
1342	s = kmalloc(sizeof(*s), GFP_KERNEL);
1343
1344	if (!s)
1345		return -ENOMEM;
1346
1347	trace_seq_init(s);
1348
1349	mutex_lock(&event_mutex);
1350	file = event_file_data(filp);
1351	if (file)
1352		print_event_filter(file, s);
1353	mutex_unlock(&event_mutex);
1354
1355	if (file)
1356		r = simple_read_from_buffer(ubuf, cnt, ppos,
1357					    s->buffer, trace_seq_used(s));
1358
1359	kfree(s);
1360
1361	return r;
1362}
1363
1364static ssize_t
1365event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1366		   loff_t *ppos)
1367{
1368	struct trace_event_file *file;
1369	char *buf;
1370	int err = -ENODEV;
1371
1372	if (cnt >= PAGE_SIZE)
1373		return -EINVAL;
1374
1375	buf = memdup_user_nul(ubuf, cnt);
1376	if (IS_ERR(buf))
1377		return PTR_ERR(buf);
1378
1379	mutex_lock(&event_mutex);
1380	file = event_file_data(filp);
1381	if (file)
1382		err = apply_event_filter(file, buf);
1383	mutex_unlock(&event_mutex);
1384
1385	kfree(buf);
1386	if (err < 0)
1387		return err;
1388
1389	*ppos += cnt;
1390
1391	return cnt;
1392}
1393
1394static LIST_HEAD(event_subsystems);
1395
1396static int subsystem_open(struct inode *inode, struct file *filp)
1397{
1398	struct event_subsystem *system = NULL;
1399	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1400	struct trace_array *tr;
1401	int ret;
1402
1403	if (tracing_is_disabled())
1404		return -ENODEV;
1405
1406	/* Make sure the system still exists */
1407	mutex_lock(&event_mutex);
1408	mutex_lock(&trace_types_lock);
 
1409	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1410		list_for_each_entry(dir, &tr->systems, list) {
1411			if (dir == inode->i_private) {
1412				/* Don't open systems with no events */
1413				if (dir->nr_events) {
1414					__get_system_dir(dir);
1415					system = dir->subsystem;
1416				}
1417				goto exit_loop;
1418			}
1419		}
1420	}
1421 exit_loop:
1422	mutex_unlock(&trace_types_lock);
1423	mutex_unlock(&event_mutex);
 
1424
1425	if (!system)
1426		return -ENODEV;
1427
1428	/* Some versions of gcc think dir can be uninitialized here */
1429	WARN_ON(!dir);
1430
1431	/* Still need to increment the ref count of the system */
1432	if (trace_array_get(tr) < 0) {
1433		put_system(dir);
1434		return -ENODEV;
1435	}
1436
1437	ret = tracing_open_generic(inode, filp);
1438	if (ret < 0) {
1439		trace_array_put(tr);
1440		put_system(dir);
1441	}
1442
1443	return ret;
1444}
1445
1446static int system_tr_open(struct inode *inode, struct file *filp)
1447{
1448	struct trace_subsystem_dir *dir;
1449	struct trace_array *tr = inode->i_private;
1450	int ret;
1451
1452	if (tracing_is_disabled())
1453		return -ENODEV;
1454
1455	if (trace_array_get(tr) < 0)
1456		return -ENODEV;
1457
1458	/* Make a temporary dir that has no system but points to tr */
1459	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1460	if (!dir) {
1461		trace_array_put(tr);
1462		return -ENOMEM;
1463	}
1464
1465	dir->tr = tr;
1466
1467	ret = tracing_open_generic(inode, filp);
1468	if (ret < 0) {
1469		trace_array_put(tr);
1470		kfree(dir);
1471		return ret;
1472	}
1473
1474	filp->private_data = dir;
1475
1476	return 0;
1477}
1478
1479static int subsystem_release(struct inode *inode, struct file *file)
1480{
1481	struct trace_subsystem_dir *dir = file->private_data;
1482
1483	trace_array_put(dir->tr);
1484
1485	/*
1486	 * If dir->subsystem is NULL, then this is a temporary
1487	 * descriptor that was made for a trace_array to enable
1488	 * all subsystems.
1489	 */
1490	if (dir->subsystem)
1491		put_system(dir);
1492	else
1493		kfree(dir);
1494
1495	return 0;
1496}
1497
1498static ssize_t
1499subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1500		      loff_t *ppos)
1501{
1502	struct trace_subsystem_dir *dir = filp->private_data;
1503	struct event_subsystem *system = dir->subsystem;
1504	struct trace_seq *s;
1505	int r;
1506
1507	if (*ppos)
1508		return 0;
1509
1510	s = kmalloc(sizeof(*s), GFP_KERNEL);
1511	if (!s)
1512		return -ENOMEM;
1513
1514	trace_seq_init(s);
1515
1516	print_subsystem_event_filter(system, s);
1517	r = simple_read_from_buffer(ubuf, cnt, ppos,
1518				    s->buffer, trace_seq_used(s));
1519
1520	kfree(s);
1521
1522	return r;
1523}
1524
1525static ssize_t
1526subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1527		       loff_t *ppos)
1528{
1529	struct trace_subsystem_dir *dir = filp->private_data;
1530	char *buf;
1531	int err;
1532
1533	if (cnt >= PAGE_SIZE)
1534		return -EINVAL;
1535
1536	buf = memdup_user_nul(ubuf, cnt);
1537	if (IS_ERR(buf))
1538		return PTR_ERR(buf);
1539
1540	err = apply_subsystem_event_filter(dir, buf);
1541	kfree(buf);
1542	if (err < 0)
1543		return err;
1544
1545	*ppos += cnt;
1546
1547	return cnt;
1548}
1549
1550static ssize_t
1551show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1552{
1553	int (*func)(struct trace_seq *s) = filp->private_data;
1554	struct trace_seq *s;
1555	int r;
1556
1557	if (*ppos)
1558		return 0;
1559
1560	s = kmalloc(sizeof(*s), GFP_KERNEL);
1561	if (!s)
1562		return -ENOMEM;
1563
1564	trace_seq_init(s);
1565
1566	func(s);
1567	r = simple_read_from_buffer(ubuf, cnt, ppos,
1568				    s->buffer, trace_seq_used(s));
1569
1570	kfree(s);
1571
1572	return r;
1573}
1574
1575static void ignore_task_cpu(void *data)
1576{
1577	struct trace_array *tr = data;
1578	struct trace_pid_list *pid_list;
1579
1580	/*
1581	 * This function is called by on_each_cpu() while the
1582	 * event_mutex is held.
1583	 */
1584	pid_list = rcu_dereference_protected(tr->filtered_pids,
1585					     mutex_is_locked(&event_mutex));
1586
1587	this_cpu_write(tr->trace_buffer.data->ignore_pid,
1588		       trace_ignore_this_task(pid_list, current));
1589}
1590
1591static ssize_t
1592ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1593		       size_t cnt, loff_t *ppos)
1594{
1595	struct seq_file *m = filp->private_data;
1596	struct trace_array *tr = m->private;
1597	struct trace_pid_list *filtered_pids = NULL;
1598	struct trace_pid_list *pid_list;
1599	struct trace_event_file *file;
1600	ssize_t ret;
1601
1602	if (!cnt)
1603		return 0;
1604
1605	ret = tracing_update_buffers();
1606	if (ret < 0)
1607		return ret;
1608
1609	mutex_lock(&event_mutex);
1610
1611	filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1612					     lockdep_is_held(&event_mutex));
1613
1614	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1615	if (ret < 0)
1616		goto out;
1617
1618	rcu_assign_pointer(tr->filtered_pids, pid_list);
1619
1620	list_for_each_entry(file, &tr->events, list) {
1621		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1622	}
1623
1624	if (filtered_pids) {
1625		synchronize_sched();
1626		trace_free_pid_list(filtered_pids);
1627	} else if (pid_list) {
1628		/*
1629		 * Register a probe that is called before all other probes
1630		 * to set ignore_pid if next or prev do not match.
1631		 * Register a probe this is called after all other probes
1632		 * to only keep ignore_pid set if next pid matches.
1633		 */
1634		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1635						 tr, INT_MAX);
1636		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1637						 tr, 0);
1638
1639		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1640						 tr, INT_MAX);
1641		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1642						 tr, 0);
1643
1644		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1645						     tr, INT_MAX);
1646		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1647						     tr, 0);
1648
1649		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1650						 tr, INT_MAX);
1651		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1652						 tr, 0);
1653	}
1654
1655	/*
1656	 * Ignoring of pids is done at task switch. But we have to
1657	 * check for those tasks that are currently running.
1658	 * Always do this in case a pid was appended or removed.
1659	 */
1660	on_each_cpu(ignore_task_cpu, tr, 1);
1661
1662 out:
1663	mutex_unlock(&event_mutex);
1664
1665	if (ret > 0)
1666		*ppos += ret;
1667
1668	return ret;
1669}
1670
1671static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1672static int ftrace_event_set_open(struct inode *inode, struct file *file);
1673static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1674static int ftrace_event_release(struct inode *inode, struct file *file);
1675
1676static const struct seq_operations show_event_seq_ops = {
1677	.start = t_start,
1678	.next = t_next,
1679	.show = t_show,
1680	.stop = t_stop,
1681};
1682
1683static const struct seq_operations show_set_event_seq_ops = {
1684	.start = s_start,
1685	.next = s_next,
1686	.show = t_show,
1687	.stop = t_stop,
1688};
1689
1690static const struct seq_operations show_set_pid_seq_ops = {
1691	.start = p_start,
1692	.next = p_next,
1693	.show = trace_pid_show,
1694	.stop = p_stop,
1695};
1696
1697static const struct file_operations ftrace_avail_fops = {
1698	.open = ftrace_event_avail_open,
1699	.read = seq_read,
1700	.llseek = seq_lseek,
1701	.release = seq_release,
1702};
1703
1704static const struct file_operations ftrace_set_event_fops = {
1705	.open = ftrace_event_set_open,
1706	.read = seq_read,
1707	.write = ftrace_event_write,
1708	.llseek = seq_lseek,
1709	.release = ftrace_event_release,
1710};
1711
1712static const struct file_operations ftrace_set_event_pid_fops = {
1713	.open = ftrace_event_set_pid_open,
1714	.read = seq_read,
1715	.write = ftrace_event_pid_write,
1716	.llseek = seq_lseek,
1717	.release = ftrace_event_release,
1718};
1719
1720static const struct file_operations ftrace_enable_fops = {
1721	.open = tracing_open_generic,
1722	.read = event_enable_read,
1723	.write = event_enable_write,
1724	.llseek = default_llseek,
1725};
1726
1727static const struct file_operations ftrace_event_format_fops = {
1728	.open = trace_format_open,
1729	.read = seq_read,
1730	.llseek = seq_lseek,
1731	.release = seq_release,
1732};
1733
1734static const struct file_operations ftrace_event_id_fops = {
1735	.read = event_id_read,
1736	.llseek = default_llseek,
1737};
1738
1739static const struct file_operations ftrace_event_filter_fops = {
1740	.open = tracing_open_generic,
1741	.read = event_filter_read,
1742	.write = event_filter_write,
1743	.llseek = default_llseek,
1744};
1745
1746static const struct file_operations ftrace_subsystem_filter_fops = {
1747	.open = subsystem_open,
1748	.read = subsystem_filter_read,
1749	.write = subsystem_filter_write,
1750	.llseek = default_llseek,
1751	.release = subsystem_release,
1752};
1753
1754static const struct file_operations ftrace_system_enable_fops = {
1755	.open = subsystem_open,
1756	.read = system_enable_read,
1757	.write = system_enable_write,
1758	.llseek = default_llseek,
1759	.release = subsystem_release,
1760};
1761
1762static const struct file_operations ftrace_tr_enable_fops = {
1763	.open = system_tr_open,
1764	.read = system_enable_read,
1765	.write = system_enable_write,
1766	.llseek = default_llseek,
1767	.release = subsystem_release,
1768};
1769
1770static const struct file_operations ftrace_show_header_fops = {
1771	.open = tracing_open_generic,
1772	.read = show_header,
1773	.llseek = default_llseek,
1774};
1775
1776static int
1777ftrace_event_open(struct inode *inode, struct file *file,
1778		  const struct seq_operations *seq_ops)
1779{
1780	struct seq_file *m;
1781	int ret;
1782
1783	ret = seq_open(file, seq_ops);
1784	if (ret < 0)
1785		return ret;
1786	m = file->private_data;
1787	/* copy tr over to seq ops */
1788	m->private = inode->i_private;
1789
1790	return ret;
1791}
1792
1793static int ftrace_event_release(struct inode *inode, struct file *file)
1794{
1795	struct trace_array *tr = inode->i_private;
1796
1797	trace_array_put(tr);
1798
1799	return seq_release(inode, file);
1800}
1801
1802static int
1803ftrace_event_avail_open(struct inode *inode, struct file *file)
1804{
1805	const struct seq_operations *seq_ops = &show_event_seq_ops;
1806
1807	return ftrace_event_open(inode, file, seq_ops);
1808}
1809
1810static int
1811ftrace_event_set_open(struct inode *inode, struct file *file)
1812{
1813	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1814	struct trace_array *tr = inode->i_private;
1815	int ret;
1816
1817	if (trace_array_get(tr) < 0)
1818		return -ENODEV;
1819
1820	if ((file->f_mode & FMODE_WRITE) &&
1821	    (file->f_flags & O_TRUNC))
1822		ftrace_clear_events(tr);
1823
1824	ret = ftrace_event_open(inode, file, seq_ops);
1825	if (ret < 0)
1826		trace_array_put(tr);
1827	return ret;
1828}
1829
1830static int
1831ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1832{
1833	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1834	struct trace_array *tr = inode->i_private;
1835	int ret;
1836
1837	if (trace_array_get(tr) < 0)
1838		return -ENODEV;
1839
1840	if ((file->f_mode & FMODE_WRITE) &&
1841	    (file->f_flags & O_TRUNC))
1842		ftrace_clear_event_pids(tr);
1843
1844	ret = ftrace_event_open(inode, file, seq_ops);
1845	if (ret < 0)
1846		trace_array_put(tr);
1847	return ret;
1848}
1849
1850static struct event_subsystem *
1851create_new_subsystem(const char *name)
1852{
1853	struct event_subsystem *system;
1854
1855	/* need to create new entry */
1856	system = kmalloc(sizeof(*system), GFP_KERNEL);
1857	if (!system)
1858		return NULL;
1859
1860	system->ref_count = 1;
1861
1862	/* Only allocate if dynamic (kprobes and modules) */
1863	system->name = kstrdup_const(name, GFP_KERNEL);
1864	if (!system->name)
1865		goto out_free;
1866
1867	system->filter = NULL;
1868
1869	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1870	if (!system->filter)
1871		goto out_free;
1872
1873	list_add(&system->list, &event_subsystems);
1874
1875	return system;
1876
1877 out_free:
1878	kfree_const(system->name);
1879	kfree(system);
1880	return NULL;
1881}
1882
1883static struct dentry *
1884event_subsystem_dir(struct trace_array *tr, const char *name,
1885		    struct trace_event_file *file, struct dentry *parent)
1886{
1887	struct trace_subsystem_dir *dir;
1888	struct event_subsystem *system;
1889	struct dentry *entry;
1890
1891	/* First see if we did not already create this dir */
1892	list_for_each_entry(dir, &tr->systems, list) {
1893		system = dir->subsystem;
1894		if (strcmp(system->name, name) == 0) {
1895			dir->nr_events++;
1896			file->system = dir;
1897			return dir->entry;
1898		}
1899	}
1900
1901	/* Now see if the system itself exists. */
1902	list_for_each_entry(system, &event_subsystems, list) {
1903		if (strcmp(system->name, name) == 0)
1904			break;
1905	}
1906	/* Reset system variable when not found */
1907	if (&system->list == &event_subsystems)
1908		system = NULL;
1909
1910	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1911	if (!dir)
1912		goto out_fail;
1913
1914	if (!system) {
1915		system = create_new_subsystem(name);
1916		if (!system)
1917			goto out_free;
1918	} else
1919		__get_system(system);
1920
1921	dir->entry = tracefs_create_dir(name, parent);
1922	if (!dir->entry) {
1923		pr_warn("Failed to create system directory %s\n", name);
1924		__put_system(system);
1925		goto out_free;
1926	}
1927
1928	dir->tr = tr;
1929	dir->ref_count = 1;
1930	dir->nr_events = 1;
1931	dir->subsystem = system;
1932	file->system = dir;
1933
1934	entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1935				    &ftrace_subsystem_filter_fops);
1936	if (!entry) {
1937		kfree(system->filter);
1938		system->filter = NULL;
1939		pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1940	}
1941
1942	trace_create_file("enable", 0644, dir->entry, dir,
1943			  &ftrace_system_enable_fops);
1944
1945	list_add(&dir->list, &tr->systems);
1946
1947	return dir->entry;
1948
1949 out_free:
1950	kfree(dir);
1951 out_fail:
1952	/* Only print this message if failed on memory allocation */
1953	if (!dir || !system)
1954		pr_warn("No memory to create event subsystem %s\n", name);
1955	return NULL;
1956}
1957
1958static int
1959event_create_dir(struct dentry *parent, struct trace_event_file *file)
1960{
1961	struct trace_event_call *call = file->event_call;
1962	struct trace_array *tr = file->tr;
1963	struct list_head *head;
1964	struct dentry *d_events;
1965	const char *name;
1966	int ret;
1967
1968	/*
1969	 * If the trace point header did not define TRACE_SYSTEM
1970	 * then the system would be called "TRACE_SYSTEM".
1971	 */
1972	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1973		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1974		if (!d_events)
1975			return -ENOMEM;
1976	} else
1977		d_events = parent;
1978
1979	name = trace_event_name(call);
1980	file->dir = tracefs_create_dir(name, d_events);
1981	if (!file->dir) {
1982		pr_warn("Could not create tracefs '%s' directory\n", name);
1983		return -1;
1984	}
1985
1986	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1987		trace_create_file("enable", 0644, file->dir, file,
1988				  &ftrace_enable_fops);
1989
1990#ifdef CONFIG_PERF_EVENTS
1991	if (call->event.type && call->class->reg)
1992		trace_create_file("id", 0444, file->dir,
1993				  (void *)(long)call->event.type,
1994				  &ftrace_event_id_fops);
1995#endif
1996
1997	/*
1998	 * Other events may have the same class. Only update
1999	 * the fields if they are not already defined.
2000	 */
2001	head = trace_get_fields(call);
2002	if (list_empty(head)) {
2003		ret = call->class->define_fields(call);
2004		if (ret < 0) {
2005			pr_warn("Could not initialize trace point events/%s\n",
2006				name);
2007			return -1;
2008		}
2009	}
2010	trace_create_file("filter", 0644, file->dir, file,
2011			  &ftrace_event_filter_fops);
2012
2013	/*
2014	 * Only event directories that can be enabled should have
2015	 * triggers.
2016	 */
2017	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2018		trace_create_file("trigger", 0644, file->dir, file,
2019				  &event_trigger_fops);
2020
2021#ifdef CONFIG_HIST_TRIGGERS
2022	trace_create_file("hist", 0444, file->dir, file,
2023			  &event_hist_fops);
2024#endif
2025	trace_create_file("format", 0444, file->dir, call,
2026			  &ftrace_event_format_fops);
2027
2028	return 0;
2029}
2030
2031static void remove_event_from_tracers(struct trace_event_call *call)
2032{
2033	struct trace_event_file *file;
2034	struct trace_array *tr;
2035
2036	do_for_each_event_file_safe(tr, file) {
2037		if (file->event_call != call)
2038			continue;
2039
2040		remove_event_file_dir(file);
2041		/*
2042		 * The do_for_each_event_file_safe() is
2043		 * a double loop. After finding the call for this
2044		 * trace_array, we use break to jump to the next
2045		 * trace_array.
2046		 */
2047		break;
2048	} while_for_each_event_file();
2049}
2050
2051static void event_remove(struct trace_event_call *call)
2052{
2053	struct trace_array *tr;
2054	struct trace_event_file *file;
2055
2056	do_for_each_event_file(tr, file) {
2057		if (file->event_call != call)
2058			continue;
2059
2060		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2061			tr->clear_trace = true;
2062
2063		ftrace_event_enable_disable(file, 0);
2064		/*
2065		 * The do_for_each_event_file() is
2066		 * a double loop. After finding the call for this
2067		 * trace_array, we use break to jump to the next
2068		 * trace_array.
2069		 */
2070		break;
2071	} while_for_each_event_file();
2072
2073	if (call->event.funcs)
2074		__unregister_trace_event(&call->event);
2075	remove_event_from_tracers(call);
2076	list_del(&call->list);
2077}
2078
2079static int event_init(struct trace_event_call *call)
2080{
2081	int ret = 0;
2082	const char *name;
2083
2084	name = trace_event_name(call);
2085	if (WARN_ON(!name))
2086		return -EINVAL;
2087
2088	if (call->class->raw_init) {
2089		ret = call->class->raw_init(call);
2090		if (ret < 0 && ret != -ENOSYS)
2091			pr_warn("Could not initialize trace events/%s\n", name);
2092	}
2093
2094	return ret;
2095}
2096
2097static int
2098__register_event(struct trace_event_call *call, struct module *mod)
2099{
2100	int ret;
2101
2102	ret = event_init(call);
2103	if (ret < 0)
2104		return ret;
2105
2106	list_add(&call->list, &ftrace_events);
2107	call->mod = mod;
2108
2109	return 0;
2110}
2111
2112static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2113{
2114	int rlen;
2115	int elen;
2116
2117	/* Find the length of the eval value as a string */
2118	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2119	/* Make sure there's enough room to replace the string with the value */
2120	if (len < elen)
2121		return NULL;
2122
2123	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2124
2125	/* Get the rest of the string of ptr */
2126	rlen = strlen(ptr + len);
2127	memmove(ptr + elen, ptr + len, rlen);
2128	/* Make sure we end the new string */
2129	ptr[elen + rlen] = 0;
2130
2131	return ptr + elen;
2132}
2133
2134static void update_event_printk(struct trace_event_call *call,
2135				struct trace_eval_map *map)
2136{
2137	char *ptr;
2138	int quote = 0;
2139	int len = strlen(map->eval_string);
2140
2141	for (ptr = call->print_fmt; *ptr; ptr++) {
2142		if (*ptr == '\\') {
2143			ptr++;
2144			/* paranoid */
2145			if (!*ptr)
2146				break;
2147			continue;
2148		}
2149		if (*ptr == '"') {
2150			quote ^= 1;
2151			continue;
2152		}
2153		if (quote)
2154			continue;
2155		if (isdigit(*ptr)) {
2156			/* skip numbers */
2157			do {
2158				ptr++;
2159				/* Check for alpha chars like ULL */
2160			} while (isalnum(*ptr));
2161			if (!*ptr)
2162				break;
2163			/*
2164			 * A number must have some kind of delimiter after
2165			 * it, and we can ignore that too.
2166			 */
2167			continue;
2168		}
2169		if (isalpha(*ptr) || *ptr == '_') {
2170			if (strncmp(map->eval_string, ptr, len) == 0 &&
2171			    !isalnum(ptr[len]) && ptr[len] != '_') {
2172				ptr = eval_replace(ptr, map, len);
2173				/* enum/sizeof string smaller than value */
2174				if (WARN_ON_ONCE(!ptr))
2175					return;
2176				/*
2177				 * No need to decrement here, as eval_replace()
2178				 * returns the pointer to the character passed
2179				 * the eval, and two evals can not be placed
2180				 * back to back without something in between.
2181				 * We can skip that something in between.
2182				 */
2183				continue;
2184			}
2185		skip_more:
2186			do {
2187				ptr++;
2188			} while (isalnum(*ptr) || *ptr == '_');
2189			if (!*ptr)
2190				break;
2191			/*
2192			 * If what comes after this variable is a '.' or
2193			 * '->' then we can continue to ignore that string.
2194			 */
2195			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2196				ptr += *ptr == '.' ? 1 : 2;
2197				if (!*ptr)
2198					break;
2199				goto skip_more;
2200			}
2201			/*
2202			 * Once again, we can skip the delimiter that came
2203			 * after the string.
2204			 */
2205			continue;
2206		}
2207	}
2208}
2209
2210void trace_event_eval_update(struct trace_eval_map **map, int len)
2211{
2212	struct trace_event_call *call, *p;
2213	const char *last_system = NULL;
2214	bool first = false;
2215	int last_i;
2216	int i;
2217
2218	down_write(&trace_event_sem);
2219	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2220		/* events are usually grouped together with systems */
2221		if (!last_system || call->class->system != last_system) {
2222			first = true;
2223			last_i = 0;
2224			last_system = call->class->system;
2225		}
2226
2227		/*
2228		 * Since calls are grouped by systems, the likelyhood that the
2229		 * next call in the iteration belongs to the same system as the
2230		 * previous call is high. As an optimization, we skip seaching
2231		 * for a map[] that matches the call's system if the last call
2232		 * was from the same system. That's what last_i is for. If the
2233		 * call has the same system as the previous call, then last_i
2234		 * will be the index of the first map[] that has a matching
2235		 * system.
2236		 */
2237		for (i = last_i; i < len; i++) {
2238			if (call->class->system == map[i]->system) {
2239				/* Save the first system if need be */
2240				if (first) {
2241					last_i = i;
2242					first = false;
2243				}
2244				update_event_printk(call, map[i]);
2245			}
2246		}
2247	}
2248	up_write(&trace_event_sem);
2249}
2250
2251static struct trace_event_file *
2252trace_create_new_event(struct trace_event_call *call,
2253		       struct trace_array *tr)
2254{
2255	struct trace_event_file *file;
2256
2257	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2258	if (!file)
2259		return NULL;
2260
2261	file->event_call = call;
2262	file->tr = tr;
2263	atomic_set(&file->sm_ref, 0);
2264	atomic_set(&file->tm_ref, 0);
2265	INIT_LIST_HEAD(&file->triggers);
2266	list_add(&file->list, &tr->events);
2267
2268	return file;
2269}
2270
2271/* Add an event to a trace directory */
2272static int
2273__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2274{
2275	struct trace_event_file *file;
2276
2277	file = trace_create_new_event(call, tr);
2278	if (!file)
2279		return -ENOMEM;
2280
2281	return event_create_dir(tr->event_dir, file);
2282}
2283
2284/*
2285 * Just create a decriptor for early init. A descriptor is required
2286 * for enabling events at boot. We want to enable events before
2287 * the filesystem is initialized.
2288 */
2289static __init int
2290__trace_early_add_new_event(struct trace_event_call *call,
2291			    struct trace_array *tr)
2292{
2293	struct trace_event_file *file;
2294
2295	file = trace_create_new_event(call, tr);
2296	if (!file)
2297		return -ENOMEM;
2298
2299	return 0;
2300}
2301
2302struct ftrace_module_file_ops;
2303static void __add_event_to_tracers(struct trace_event_call *call);
2304
2305/* Add an additional event_call dynamically */
2306int trace_add_event_call(struct trace_event_call *call)
2307{
2308	int ret;
2309	mutex_lock(&event_mutex);
2310	mutex_lock(&trace_types_lock);
 
2311
2312	ret = __register_event(call, NULL);
2313	if (ret >= 0)
2314		__add_event_to_tracers(call);
2315
2316	mutex_unlock(&trace_types_lock);
2317	mutex_unlock(&event_mutex);
 
2318	return ret;
2319}
2320
2321/*
2322 * Must be called under locking of trace_types_lock, event_mutex and
2323 * trace_event_sem.
2324 */
2325static void __trace_remove_event_call(struct trace_event_call *call)
2326{
2327	event_remove(call);
2328	trace_destroy_fields(call);
2329	free_event_filter(call->filter);
2330	call->filter = NULL;
2331}
2332
2333static int probe_remove_event_call(struct trace_event_call *call)
2334{
2335	struct trace_array *tr;
2336	struct trace_event_file *file;
2337
2338#ifdef CONFIG_PERF_EVENTS
2339	if (call->perf_refcount)
2340		return -EBUSY;
2341#endif
2342	do_for_each_event_file(tr, file) {
2343		if (file->event_call != call)
2344			continue;
2345		/*
2346		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2347		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2348		 * TRACE_REG_UNREGISTER.
2349		 */
2350		if (file->flags & EVENT_FILE_FL_ENABLED)
2351			return -EBUSY;
2352		/*
2353		 * The do_for_each_event_file_safe() is
2354		 * a double loop. After finding the call for this
2355		 * trace_array, we use break to jump to the next
2356		 * trace_array.
2357		 */
2358		break;
2359	} while_for_each_event_file();
2360
2361	__trace_remove_event_call(call);
2362
2363	return 0;
2364}
2365
2366/* Remove an event_call */
2367int trace_remove_event_call(struct trace_event_call *call)
2368{
2369	int ret;
2370
2371	mutex_lock(&event_mutex);
2372	mutex_lock(&trace_types_lock);
 
2373	down_write(&trace_event_sem);
2374	ret = probe_remove_event_call(call);
2375	up_write(&trace_event_sem);
2376	mutex_unlock(&trace_types_lock);
2377	mutex_unlock(&event_mutex);
 
2378
2379	return ret;
2380}
2381
2382#define for_each_event(event, start, end)			\
2383	for (event = start;					\
2384	     (unsigned long)event < (unsigned long)end;		\
2385	     event++)
2386
2387#ifdef CONFIG_MODULES
2388
2389static void trace_module_add_events(struct module *mod)
2390{
2391	struct trace_event_call **call, **start, **end;
2392
2393	if (!mod->num_trace_events)
2394		return;
2395
2396	/* Don't add infrastructure for mods without tracepoints */
2397	if (trace_module_has_bad_taint(mod)) {
2398		pr_err("%s: module has bad taint, not creating trace events\n",
2399		       mod->name);
2400		return;
2401	}
2402
2403	start = mod->trace_events;
2404	end = mod->trace_events + mod->num_trace_events;
2405
2406	for_each_event(call, start, end) {
2407		__register_event(*call, mod);
2408		__add_event_to_tracers(*call);
2409	}
2410}
2411
2412static void trace_module_remove_events(struct module *mod)
2413{
2414	struct trace_event_call *call, *p;
 
2415
2416	down_write(&trace_event_sem);
2417	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2418		if (call->mod == mod)
 
 
2419			__trace_remove_event_call(call);
 
2420	}
2421	up_write(&trace_event_sem);
2422
2423	/*
2424	 * It is safest to reset the ring buffer if the module being unloaded
2425	 * registered any events that were used. The only worry is if
2426	 * a new module gets loaded, and takes on the same id as the events
2427	 * of this module. When printing out the buffer, traced events left
2428	 * over from this module may be passed to the new module events and
2429	 * unexpected results may occur.
2430	 */
2431	tracing_reset_all_online_cpus();
 
2432}
2433
2434static int trace_module_notify(struct notifier_block *self,
2435			       unsigned long val, void *data)
2436{
2437	struct module *mod = data;
2438
2439	mutex_lock(&event_mutex);
2440	mutex_lock(&trace_types_lock);
 
2441	switch (val) {
2442	case MODULE_STATE_COMING:
2443		trace_module_add_events(mod);
2444		break;
2445	case MODULE_STATE_GOING:
2446		trace_module_remove_events(mod);
2447		break;
2448	}
2449	mutex_unlock(&trace_types_lock);
2450	mutex_unlock(&event_mutex);
 
2451
2452	return 0;
2453}
2454
2455static struct notifier_block trace_module_nb = {
2456	.notifier_call = trace_module_notify,
2457	.priority = 1, /* higher than trace.c module notify */
2458};
2459#endif /* CONFIG_MODULES */
2460
2461/* Create a new event directory structure for a trace directory. */
2462static void
2463__trace_add_event_dirs(struct trace_array *tr)
2464{
2465	struct trace_event_call *call;
2466	int ret;
2467
2468	list_for_each_entry(call, &ftrace_events, list) {
2469		ret = __trace_add_new_event(call, tr);
2470		if (ret < 0)
2471			pr_warn("Could not create directory for event %s\n",
2472				trace_event_name(call));
2473	}
2474}
2475
2476struct trace_event_file *
2477find_event_file(struct trace_array *tr, const char *system,  const char *event)
2478{
2479	struct trace_event_file *file;
2480	struct trace_event_call *call;
2481	const char *name;
2482
2483	list_for_each_entry(file, &tr->events, list) {
2484
2485		call = file->event_call;
2486		name = trace_event_name(call);
2487
2488		if (!name || !call->class || !call->class->reg)
2489			continue;
2490
2491		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2492			continue;
2493
2494		if (strcmp(event, name) == 0 &&
2495		    strcmp(system, call->class->system) == 0)
2496			return file;
2497	}
2498	return NULL;
2499}
2500
2501#ifdef CONFIG_DYNAMIC_FTRACE
2502
2503/* Avoid typos */
2504#define ENABLE_EVENT_STR	"enable_event"
2505#define DISABLE_EVENT_STR	"disable_event"
2506
2507struct event_probe_data {
2508	struct trace_event_file	*file;
2509	unsigned long			count;
2510	int				ref;
2511	bool				enable;
2512};
2513
2514static void update_event_probe(struct event_probe_data *data)
 
2515{
 
 
 
 
 
 
2516	if (data->enable)
2517		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2518	else
2519		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2520}
2521
2522static void
2523event_enable_probe(unsigned long ip, unsigned long parent_ip,
2524		   struct trace_array *tr, struct ftrace_probe_ops *ops,
2525		   void *data)
2526{
2527	struct ftrace_func_mapper *mapper = data;
2528	struct event_probe_data *edata;
2529	void **pdata;
2530
2531	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2532	if (!pdata || !*pdata)
2533		return;
2534
2535	edata = *pdata;
2536	update_event_probe(edata);
2537}
2538
2539static void
2540event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2541			 struct trace_array *tr, struct ftrace_probe_ops *ops,
2542			 void *data)
2543{
2544	struct ftrace_func_mapper *mapper = data;
2545	struct event_probe_data *edata;
2546	void **pdata;
2547
2548	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2549	if (!pdata || !*pdata)
2550		return;
2551
2552	edata = *pdata;
2553
2554	if (!edata->count)
2555		return;
2556
2557	/* Skip if the event is in a state we want to switch to */
2558	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2559		return;
2560
2561	if (edata->count != -1)
2562		(edata->count)--;
2563
2564	update_event_probe(edata);
2565}
2566
2567static int
2568event_enable_print(struct seq_file *m, unsigned long ip,
2569		   struct ftrace_probe_ops *ops, void *data)
2570{
2571	struct ftrace_func_mapper *mapper = data;
2572	struct event_probe_data *edata;
2573	void **pdata;
2574
2575	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2576
2577	if (WARN_ON_ONCE(!pdata || !*pdata))
2578		return 0;
2579
2580	edata = *pdata;
2581
2582	seq_printf(m, "%ps:", (void *)ip);
2583
2584	seq_printf(m, "%s:%s:%s",
2585		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2586		   edata->file->event_call->class->system,
2587		   trace_event_name(edata->file->event_call));
2588
2589	if (edata->count == -1)
2590		seq_puts(m, ":unlimited\n");
2591	else
2592		seq_printf(m, ":count=%ld\n", edata->count);
2593
2594	return 0;
2595}
2596
2597static int
2598event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2599		  unsigned long ip, void *init_data, void **data)
2600{
2601	struct ftrace_func_mapper *mapper = *data;
2602	struct event_probe_data *edata = init_data;
2603	int ret;
2604
2605	if (!mapper) {
2606		mapper = allocate_ftrace_func_mapper();
2607		if (!mapper)
2608			return -ENODEV;
2609		*data = mapper;
2610	}
2611
2612	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2613	if (ret < 0)
2614		return ret;
2615
2616	edata->ref++;
2617
2618	return 0;
2619}
2620
2621static int free_probe_data(void *data)
2622{
2623	struct event_probe_data *edata = data;
 
2624
2625	edata->ref--;
2626	if (!edata->ref) {
2627		/* Remove the SOFT_MODE flag */
2628		__ftrace_event_enable_disable(edata->file, 0, 1);
2629		module_put(edata->file->event_call->mod);
2630		kfree(edata);
2631	}
2632	return 0;
2633}
2634
2635static void
2636event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2637		  unsigned long ip, void *data)
2638{
2639	struct ftrace_func_mapper *mapper = data;
2640	struct event_probe_data *edata;
2641
2642	if (!ip) {
2643		if (!mapper)
2644			return;
2645		free_ftrace_func_mapper(mapper, free_probe_data);
2646		return;
2647	}
2648
2649	edata = ftrace_func_mapper_remove_ip(mapper, ip);
2650
2651	if (WARN_ON_ONCE(!edata))
2652		return;
2653
2654	if (WARN_ON_ONCE(edata->ref <= 0))
2655		return;
2656
2657	free_probe_data(edata);
 
 
 
 
2658}
2659
2660static struct ftrace_probe_ops event_enable_probe_ops = {
2661	.func			= event_enable_probe,
2662	.print			= event_enable_print,
2663	.init			= event_enable_init,
2664	.free			= event_enable_free,
2665};
2666
2667static struct ftrace_probe_ops event_enable_count_probe_ops = {
2668	.func			= event_enable_count_probe,
2669	.print			= event_enable_print,
2670	.init			= event_enable_init,
2671	.free			= event_enable_free,
2672};
2673
2674static struct ftrace_probe_ops event_disable_probe_ops = {
2675	.func			= event_enable_probe,
2676	.print			= event_enable_print,
2677	.init			= event_enable_init,
2678	.free			= event_enable_free,
2679};
2680
2681static struct ftrace_probe_ops event_disable_count_probe_ops = {
2682	.func			= event_enable_count_probe,
2683	.print			= event_enable_print,
2684	.init			= event_enable_init,
2685	.free			= event_enable_free,
2686};
2687
2688static int
2689event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2690		  char *glob, char *cmd, char *param, int enabled)
2691{
 
2692	struct trace_event_file *file;
2693	struct ftrace_probe_ops *ops;
2694	struct event_probe_data *data;
2695	const char *system;
2696	const char *event;
2697	char *number;
2698	bool enable;
2699	int ret;
2700
2701	if (!tr)
2702		return -ENODEV;
2703
2704	/* hash funcs only work with set_ftrace_filter */
2705	if (!enabled || !param)
2706		return -EINVAL;
2707
2708	system = strsep(&param, ":");
2709	if (!param)
2710		return -EINVAL;
2711
2712	event = strsep(&param, ":");
2713
2714	mutex_lock(&event_mutex);
2715
2716	ret = -EINVAL;
2717	file = find_event_file(tr, system, event);
2718	if (!file)
2719		goto out;
2720
2721	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2722
2723	if (enable)
2724		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2725	else
2726		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2727
2728	if (glob[0] == '!') {
2729		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
 
2730		goto out;
2731	}
2732
2733	ret = -ENOMEM;
2734
2735	data = kzalloc(sizeof(*data), GFP_KERNEL);
2736	if (!data)
2737		goto out;
2738
2739	data->enable = enable;
2740	data->count = -1;
2741	data->file = file;
2742
2743	if (!param)
2744		goto out_reg;
2745
2746	number = strsep(&param, ":");
2747
2748	ret = -EINVAL;
2749	if (!strlen(number))
2750		goto out_free;
2751
2752	/*
2753	 * We use the callback data field (which is a pointer)
2754	 * as our counter.
2755	 */
2756	ret = kstrtoul(number, 0, &data->count);
2757	if (ret)
2758		goto out_free;
2759
2760 out_reg:
2761	/* Don't let event modules unload while probe registered */
2762	ret = try_module_get(file->event_call->mod);
2763	if (!ret) {
2764		ret = -EBUSY;
2765		goto out_free;
2766	}
2767
2768	ret = __ftrace_event_enable_disable(file, 1, 1);
2769	if (ret < 0)
2770		goto out_put;
2771
2772	ret = register_ftrace_function_probe(glob, tr, ops, data);
2773	/*
2774	 * The above returns on success the # of functions enabled,
2775	 * but if it didn't find any functions it returns zero.
2776	 * Consider no functions a failure too.
2777	 */
2778	if (!ret) {
2779		ret = -ENOENT;
2780		goto out_disable;
2781	} else if (ret < 0)
2782		goto out_disable;
2783	/* Just return zero, not the number of enabled functions */
2784	ret = 0;
2785 out:
2786	mutex_unlock(&event_mutex);
2787	return ret;
2788
2789 out_disable:
2790	__ftrace_event_enable_disable(file, 0, 1);
2791 out_put:
2792	module_put(file->event_call->mod);
2793 out_free:
2794	kfree(data);
2795	goto out;
2796}
2797
2798static struct ftrace_func_command event_enable_cmd = {
2799	.name			= ENABLE_EVENT_STR,
2800	.func			= event_enable_func,
2801};
2802
2803static struct ftrace_func_command event_disable_cmd = {
2804	.name			= DISABLE_EVENT_STR,
2805	.func			= event_enable_func,
2806};
2807
2808static __init int register_event_cmds(void)
2809{
2810	int ret;
2811
2812	ret = register_ftrace_command(&event_enable_cmd);
2813	if (WARN_ON(ret < 0))
2814		return ret;
2815	ret = register_ftrace_command(&event_disable_cmd);
2816	if (WARN_ON(ret < 0))
2817		unregister_ftrace_command(&event_enable_cmd);
2818	return ret;
2819}
2820#else
2821static inline int register_event_cmds(void) { return 0; }
2822#endif /* CONFIG_DYNAMIC_FTRACE */
2823
2824/*
2825 * The top level array has already had its trace_event_file
2826 * descriptors created in order to allow for early events to
2827 * be recorded. This function is called after the tracefs has been
2828 * initialized, and we now have to create the files associated
2829 * to the events.
2830 */
2831static __init void
2832__trace_early_add_event_dirs(struct trace_array *tr)
2833{
2834	struct trace_event_file *file;
2835	int ret;
2836
2837
2838	list_for_each_entry(file, &tr->events, list) {
2839		ret = event_create_dir(tr->event_dir, file);
2840		if (ret < 0)
2841			pr_warn("Could not create directory for event %s\n",
2842				trace_event_name(file->event_call));
2843	}
2844}
2845
2846/*
2847 * For early boot up, the top trace array requires to have
2848 * a list of events that can be enabled. This must be done before
2849 * the filesystem is set up in order to allow events to be traced
2850 * early.
2851 */
2852static __init void
2853__trace_early_add_events(struct trace_array *tr)
2854{
2855	struct trace_event_call *call;
2856	int ret;
2857
2858	list_for_each_entry(call, &ftrace_events, list) {
2859		/* Early boot up should not have any modules loaded */
2860		if (WARN_ON_ONCE(call->mod))
2861			continue;
2862
2863		ret = __trace_early_add_new_event(call, tr);
2864		if (ret < 0)
2865			pr_warn("Could not create early event %s\n",
2866				trace_event_name(call));
2867	}
2868}
2869
2870/* Remove the event directory structure for a trace directory. */
2871static void
2872__trace_remove_event_dirs(struct trace_array *tr)
2873{
2874	struct trace_event_file *file, *next;
2875
2876	list_for_each_entry_safe(file, next, &tr->events, list)
2877		remove_event_file_dir(file);
2878}
2879
2880static void __add_event_to_tracers(struct trace_event_call *call)
2881{
2882	struct trace_array *tr;
2883
2884	list_for_each_entry(tr, &ftrace_trace_arrays, list)
2885		__trace_add_new_event(call, tr);
2886}
2887
2888extern struct trace_event_call *__start_ftrace_events[];
2889extern struct trace_event_call *__stop_ftrace_events[];
2890
2891static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2892
2893static __init int setup_trace_event(char *str)
2894{
2895	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2896	ring_buffer_expanded = true;
2897	tracing_selftest_disabled = true;
2898
2899	return 1;
2900}
2901__setup("trace_event=", setup_trace_event);
2902
2903/* Expects to have event_mutex held when called */
2904static int
2905create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2906{
2907	struct dentry *d_events;
2908	struct dentry *entry;
2909
2910	entry = tracefs_create_file("set_event", 0644, parent,
2911				    tr, &ftrace_set_event_fops);
2912	if (!entry) {
2913		pr_warn("Could not create tracefs 'set_event' entry\n");
2914		return -ENOMEM;
2915	}
2916
2917	d_events = tracefs_create_dir("events", parent);
2918	if (!d_events) {
2919		pr_warn("Could not create tracefs 'events' directory\n");
2920		return -ENOMEM;
2921	}
2922
2923	entry = trace_create_file("enable", 0644, d_events,
2924				  tr, &ftrace_tr_enable_fops);
2925	if (!entry) {
2926		pr_warn("Could not create tracefs 'enable' entry\n");
2927		return -ENOMEM;
2928	}
2929
2930	/* There are not as crucial, just warn if they are not created */
2931
2932	entry = tracefs_create_file("set_event_pid", 0644, parent,
2933				    tr, &ftrace_set_event_pid_fops);
2934	if (!entry)
2935		pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2936
2937	/* ring buffer internal formats */
2938	entry = trace_create_file("header_page", 0444, d_events,
2939				  ring_buffer_print_page_header,
2940				  &ftrace_show_header_fops);
2941	if (!entry)
2942		pr_warn("Could not create tracefs 'header_page' entry\n");
2943
2944	entry = trace_create_file("header_event", 0444, d_events,
2945				  ring_buffer_print_entry_header,
2946				  &ftrace_show_header_fops);
2947	if (!entry)
2948		pr_warn("Could not create tracefs 'header_event' entry\n");
2949
2950	tr->event_dir = d_events;
2951
2952	return 0;
2953}
2954
2955/**
2956 * event_trace_add_tracer - add a instance of a trace_array to events
2957 * @parent: The parent dentry to place the files/directories for events in
2958 * @tr: The trace array associated with these events
2959 *
2960 * When a new instance is created, it needs to set up its events
2961 * directory, as well as other files associated with events. It also
2962 * creates the event hierachry in the @parent/events directory.
2963 *
2964 * Returns 0 on success.
2965 *
2966 * Must be called with event_mutex held.
2967 */
2968int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2969{
2970	int ret;
2971
2972	lockdep_assert_held(&event_mutex);
2973
2974	ret = create_event_toplevel_files(parent, tr);
2975	if (ret)
2976		goto out;
2977
2978	down_write(&trace_event_sem);
2979	__trace_add_event_dirs(tr);
2980	up_write(&trace_event_sem);
2981
2982 out:
 
 
2983	return ret;
2984}
2985
2986/*
2987 * The top trace array already had its file descriptors created.
2988 * Now the files themselves need to be created.
2989 */
2990static __init int
2991early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2992{
2993	int ret;
2994
2995	mutex_lock(&event_mutex);
2996
2997	ret = create_event_toplevel_files(parent, tr);
2998	if (ret)
2999		goto out_unlock;
3000
3001	down_write(&trace_event_sem);
3002	__trace_early_add_event_dirs(tr);
3003	up_write(&trace_event_sem);
3004
3005 out_unlock:
3006	mutex_unlock(&event_mutex);
3007
3008	return ret;
3009}
3010
3011/* Must be called with event_mutex held */
3012int event_trace_del_tracer(struct trace_array *tr)
3013{
3014	lockdep_assert_held(&event_mutex);
3015
3016	/* Disable any event triggers and associated soft-disabled events */
3017	clear_event_triggers(tr);
3018
3019	/* Clear the pid list */
3020	__ftrace_clear_event_pids(tr);
3021
3022	/* Disable any running events */
3023	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3024
3025	/* Access to events are within rcu_read_lock_sched() */
3026	synchronize_sched();
3027
3028	down_write(&trace_event_sem);
3029	__trace_remove_event_dirs(tr);
3030	tracefs_remove_recursive(tr->event_dir);
3031	up_write(&trace_event_sem);
3032
3033	tr->event_dir = NULL;
 
 
3034
3035	return 0;
3036}
3037
3038static __init int event_trace_memsetup(void)
3039{
3040	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3041	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3042	return 0;
3043}
3044
3045static __init void
3046early_enable_events(struct trace_array *tr, bool disable_first)
3047{
3048	char *buf = bootup_event_buf;
3049	char *token;
3050	int ret;
3051
3052	while (true) {
3053		token = strsep(&buf, ",");
3054
3055		if (!token)
3056			break;
3057
3058		if (*token) {
3059			/* Restarting syscalls requires that we stop them first */
3060			if (disable_first)
3061				ftrace_set_clr_event(tr, token, 0);
3062
3063			ret = ftrace_set_clr_event(tr, token, 1);
3064			if (ret)
3065				pr_warn("Failed to enable trace event: %s\n", token);
3066		}
3067
3068		/* Put back the comma to allow this to be called again */
3069		if (buf)
3070			*(buf - 1) = ',';
3071	}
3072}
3073
3074static __init int event_trace_enable(void)
3075{
3076	struct trace_array *tr = top_trace_array();
3077	struct trace_event_call **iter, *call;
3078	int ret;
3079
3080	if (!tr)
3081		return -ENODEV;
3082
3083	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3084
3085		call = *iter;
3086		ret = event_init(call);
3087		if (!ret)
3088			list_add(&call->list, &ftrace_events);
3089	}
3090
3091	/*
3092	 * We need the top trace array to have a working set of trace
3093	 * points at early init, before the debug files and directories
3094	 * are created. Create the file entries now, and attach them
3095	 * to the actual file dentries later.
3096	 */
3097	__trace_early_add_events(tr);
3098
3099	early_enable_events(tr, false);
3100
3101	trace_printk_start_comm();
3102
3103	register_event_cmds();
3104
3105	register_trigger_cmds();
3106
3107	return 0;
3108}
3109
3110/*
3111 * event_trace_enable() is called from trace_event_init() first to
3112 * initialize events and perhaps start any events that are on the
3113 * command line. Unfortunately, there are some events that will not
3114 * start this early, like the system call tracepoints that need
3115 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3116 * is called before pid 1 starts, and this flag is never set, making
3117 * the syscall tracepoint never get reached, but the event is enabled
3118 * regardless (and not doing anything).
3119 */
3120static __init int event_trace_enable_again(void)
3121{
3122	struct trace_array *tr;
3123
3124	tr = top_trace_array();
3125	if (!tr)
3126		return -ENODEV;
3127
3128	early_enable_events(tr, true);
3129
3130	return 0;
3131}
3132
3133early_initcall(event_trace_enable_again);
3134
3135static __init int event_trace_init(void)
3136{
3137	struct trace_array *tr;
3138	struct dentry *d_tracer;
3139	struct dentry *entry;
3140	int ret;
3141
3142	tr = top_trace_array();
3143	if (!tr)
3144		return -ENODEV;
3145
3146	d_tracer = tracing_init_dentry();
3147	if (IS_ERR(d_tracer))
3148		return 0;
3149
3150	entry = tracefs_create_file("available_events", 0444, d_tracer,
3151				    tr, &ftrace_avail_fops);
3152	if (!entry)
3153		pr_warn("Could not create tracefs 'available_events' entry\n");
3154
3155	if (trace_define_generic_fields())
3156		pr_warn("tracing: Failed to allocated generic fields");
3157
3158	if (trace_define_common_fields())
3159		pr_warn("tracing: Failed to allocate common fields");
3160
3161	ret = early_event_add_tracer(d_tracer, tr);
3162	if (ret)
3163		return ret;
3164
3165#ifdef CONFIG_MODULES
3166	ret = register_module_notifier(&trace_module_nb);
3167	if (ret)
3168		pr_warn("Failed to register trace events module notifier\n");
3169#endif
3170	return 0;
3171}
3172
3173void __init trace_event_init(void)
3174{
3175	event_trace_memsetup();
3176	init_ftrace_syscalls();
3177	event_trace_enable();
3178}
3179
3180fs_initcall(event_trace_init);
3181
3182#ifdef CONFIG_FTRACE_STARTUP_TEST
3183
3184static DEFINE_SPINLOCK(test_spinlock);
3185static DEFINE_SPINLOCK(test_spinlock_irq);
3186static DEFINE_MUTEX(test_mutex);
3187
3188static __init void test_work(struct work_struct *dummy)
3189{
3190	spin_lock(&test_spinlock);
3191	spin_lock_irq(&test_spinlock_irq);
3192	udelay(1);
3193	spin_unlock_irq(&test_spinlock_irq);
3194	spin_unlock(&test_spinlock);
3195
3196	mutex_lock(&test_mutex);
3197	msleep(1);
3198	mutex_unlock(&test_mutex);
3199}
3200
3201static __init int event_test_thread(void *unused)
3202{
3203	void *test_malloc;
3204
3205	test_malloc = kmalloc(1234, GFP_KERNEL);
3206	if (!test_malloc)
3207		pr_info("failed to kmalloc\n");
3208
3209	schedule_on_each_cpu(test_work);
3210
3211	kfree(test_malloc);
3212
3213	set_current_state(TASK_INTERRUPTIBLE);
3214	while (!kthread_should_stop()) {
3215		schedule();
3216		set_current_state(TASK_INTERRUPTIBLE);
3217	}
3218	__set_current_state(TASK_RUNNING);
3219
3220	return 0;
3221}
3222
3223/*
3224 * Do various things that may trigger events.
3225 */
3226static __init void event_test_stuff(void)
3227{
3228	struct task_struct *test_thread;
3229
3230	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3231	msleep(1);
3232	kthread_stop(test_thread);
3233}
3234
3235/*
3236 * For every trace event defined, we will test each trace point separately,
3237 * and then by groups, and finally all trace points.
3238 */
3239static __init void event_trace_self_tests(void)
3240{
3241	struct trace_subsystem_dir *dir;
3242	struct trace_event_file *file;
3243	struct trace_event_call *call;
3244	struct event_subsystem *system;
3245	struct trace_array *tr;
3246	int ret;
3247
3248	tr = top_trace_array();
3249	if (!tr)
3250		return;
3251
3252	pr_info("Running tests on trace events:\n");
3253
3254	list_for_each_entry(file, &tr->events, list) {
3255
3256		call = file->event_call;
3257
3258		/* Only test those that have a probe */
3259		if (!call->class || !call->class->probe)
3260			continue;
3261
3262/*
3263 * Testing syscall events here is pretty useless, but
3264 * we still do it if configured. But this is time consuming.
3265 * What we really need is a user thread to perform the
3266 * syscalls as we test.
3267 */
3268#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3269		if (call->class->system &&
3270		    strcmp(call->class->system, "syscalls") == 0)
3271			continue;
3272#endif
3273
3274		pr_info("Testing event %s: ", trace_event_name(call));
3275
3276		/*
3277		 * If an event is already enabled, someone is using
3278		 * it and the self test should not be on.
3279		 */
3280		if (file->flags & EVENT_FILE_FL_ENABLED) {
3281			pr_warn("Enabled event during self test!\n");
3282			WARN_ON_ONCE(1);
3283			continue;
3284		}
3285
3286		ftrace_event_enable_disable(file, 1);
3287		event_test_stuff();
3288		ftrace_event_enable_disable(file, 0);
3289
3290		pr_cont("OK\n");
3291	}
3292
3293	/* Now test at the sub system level */
3294
3295	pr_info("Running tests on trace event systems:\n");
3296
3297	list_for_each_entry(dir, &tr->systems, list) {
3298
3299		system = dir->subsystem;
3300
3301		/* the ftrace system is special, skip it */
3302		if (strcmp(system->name, "ftrace") == 0)
3303			continue;
3304
3305		pr_info("Testing event system %s: ", system->name);
3306
3307		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3308		if (WARN_ON_ONCE(ret)) {
3309			pr_warn("error enabling system %s\n",
3310				system->name);
3311			continue;
3312		}
3313
3314		event_test_stuff();
3315
3316		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3317		if (WARN_ON_ONCE(ret)) {
3318			pr_warn("error disabling system %s\n",
3319				system->name);
3320			continue;
3321		}
3322
3323		pr_cont("OK\n");
3324	}
3325
3326	/* Test with all events enabled */
3327
3328	pr_info("Running tests on all trace events:\n");
3329	pr_info("Testing all events: ");
3330
3331	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3332	if (WARN_ON_ONCE(ret)) {
3333		pr_warn("error enabling all events\n");
3334		return;
3335	}
3336
3337	event_test_stuff();
3338
3339	/* reset sysname */
3340	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3341	if (WARN_ON_ONCE(ret)) {
3342		pr_warn("error disabling all events\n");
3343		return;
3344	}
3345
3346	pr_cont("OK\n");
3347}
3348
3349#ifdef CONFIG_FUNCTION_TRACER
3350
3351static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3352
3353static struct trace_event_file event_trace_file __initdata;
3354
3355static void __init
3356function_test_events_call(unsigned long ip, unsigned long parent_ip,
3357			  struct ftrace_ops *op, struct pt_regs *pt_regs)
3358{
3359	struct ring_buffer_event *event;
3360	struct ring_buffer *buffer;
3361	struct ftrace_entry *entry;
3362	unsigned long flags;
3363	long disabled;
3364	int cpu;
3365	int pc;
3366
3367	pc = preempt_count();
3368	preempt_disable_notrace();
3369	cpu = raw_smp_processor_id();
3370	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3371
3372	if (disabled != 1)
3373		goto out;
3374
3375	local_save_flags(flags);
3376
3377	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3378						TRACE_FN, sizeof(*entry),
3379						flags, pc);
3380	if (!event)
3381		goto out;
3382	entry	= ring_buffer_event_data(event);
3383	entry->ip			= ip;
3384	entry->parent_ip		= parent_ip;
3385
3386	event_trigger_unlock_commit(&event_trace_file, buffer, event,
3387				    entry, flags, pc);
3388 out:
3389	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3390	preempt_enable_notrace();
3391}
3392
3393static struct ftrace_ops trace_ops __initdata  =
3394{
3395	.func = function_test_events_call,
3396	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
3397};
3398
3399static __init void event_trace_self_test_with_function(void)
3400{
3401	int ret;
3402
3403	event_trace_file.tr = top_trace_array();
3404	if (WARN_ON(!event_trace_file.tr))
3405		return;
3406
3407	ret = register_ftrace_function(&trace_ops);
3408	if (WARN_ON(ret < 0)) {
3409		pr_info("Failed to enable function tracer for event tests\n");
3410		return;
3411	}
3412	pr_info("Running tests again, along with the function tracer\n");
3413	event_trace_self_tests();
3414	unregister_ftrace_function(&trace_ops);
3415}
3416#else
3417static __init void event_trace_self_test_with_function(void)
3418{
3419}
3420#endif
3421
3422static __init int event_trace_self_tests_init(void)
3423{
3424	if (!tracing_selftest_disabled) {
3425		event_trace_self_tests();
3426		event_trace_self_test_with_function();
3427	}
3428
3429	return 0;
3430}
3431
3432late_initcall(event_trace_self_tests_init);
3433
3434#endif
v4.10.11
   1/*
   2 * event tracer
   3 *
   4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5 *
   6 *  - Added format output of fields of the trace point.
   7 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8 *
   9 */
  10
  11#define pr_fmt(fmt) fmt
  12
  13#include <linux/workqueue.h>
  14#include <linux/spinlock.h>
  15#include <linux/kthread.h>
  16#include <linux/tracefs.h>
  17#include <linux/uaccess.h>
  18#include <linux/module.h>
  19#include <linux/ctype.h>
  20#include <linux/sort.h>
  21#include <linux/slab.h>
  22#include <linux/delay.h>
  23
  24#include <trace/events/sched.h>
  25
  26#include <asm/setup.h>
  27
  28#include "trace_output.h"
  29
  30#undef TRACE_SYSTEM
  31#define TRACE_SYSTEM "TRACE_SYSTEM"
  32
  33DEFINE_MUTEX(event_mutex);
  34
  35LIST_HEAD(ftrace_events);
  36static LIST_HEAD(ftrace_generic_fields);
  37static LIST_HEAD(ftrace_common_fields);
  38
  39#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41static struct kmem_cache *field_cachep;
  42static struct kmem_cache *file_cachep;
  43
  44static inline int system_refcount(struct event_subsystem *system)
  45{
  46	return system->ref_count;
  47}
  48
  49static int system_refcount_inc(struct event_subsystem *system)
  50{
  51	return system->ref_count++;
  52}
  53
  54static int system_refcount_dec(struct event_subsystem *system)
  55{
  56	return --system->ref_count;
  57}
  58
  59/* Double loops, do not use break, only goto's work */
  60#define do_for_each_event_file(tr, file)			\
  61	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  62		list_for_each_entry(file, &tr->events, list)
  63
  64#define do_for_each_event_file_safe(tr, file)			\
  65	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  66		struct trace_event_file *___n;				\
  67		list_for_each_entry_safe(file, ___n, &tr->events, list)
  68
  69#define while_for_each_event_file()		\
  70	}
  71
  72static struct list_head *
  73trace_get_fields(struct trace_event_call *event_call)
  74{
  75	if (!event_call->class->get_fields)
  76		return &event_call->class->fields;
  77	return event_call->class->get_fields(event_call);
  78}
  79
  80static struct ftrace_event_field *
  81__find_event_field(struct list_head *head, char *name)
  82{
  83	struct ftrace_event_field *field;
  84
  85	list_for_each_entry(field, head, link) {
  86		if (!strcmp(field->name, name))
  87			return field;
  88	}
  89
  90	return NULL;
  91}
  92
  93struct ftrace_event_field *
  94trace_find_event_field(struct trace_event_call *call, char *name)
  95{
  96	struct ftrace_event_field *field;
  97	struct list_head *head;
  98
  99	head = trace_get_fields(call);
 100	field = __find_event_field(head, name);
 101	if (field)
 102		return field;
 103
 104	field = __find_event_field(&ftrace_generic_fields, name);
 105	if (field)
 106		return field;
 107
 108	return __find_event_field(&ftrace_common_fields, name);
 109}
 110
 111static int __trace_define_field(struct list_head *head, const char *type,
 112				const char *name, int offset, int size,
 113				int is_signed, int filter_type)
 114{
 115	struct ftrace_event_field *field;
 116
 117	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 118	if (!field)
 119		return -ENOMEM;
 120
 121	field->name = name;
 122	field->type = type;
 123
 124	if (filter_type == FILTER_OTHER)
 125		field->filter_type = filter_assign_type(type);
 126	else
 127		field->filter_type = filter_type;
 128
 129	field->offset = offset;
 130	field->size = size;
 131	field->is_signed = is_signed;
 132
 133	list_add(&field->link, head);
 134
 135	return 0;
 136}
 137
 138int trace_define_field(struct trace_event_call *call, const char *type,
 139		       const char *name, int offset, int size, int is_signed,
 140		       int filter_type)
 141{
 142	struct list_head *head;
 143
 144	if (WARN_ON(!call->class))
 145		return 0;
 146
 147	head = trace_get_fields(call);
 148	return __trace_define_field(head, type, name, offset, size,
 149				    is_signed, filter_type);
 150}
 151EXPORT_SYMBOL_GPL(trace_define_field);
 152
 153#define __generic_field(type, item, filter_type)			\
 154	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 155				   #item, 0, 0, is_signed_type(type),	\
 156				   filter_type);			\
 157	if (ret)							\
 158		return ret;
 159
 160#define __common_field(type, item)					\
 161	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 162				   "common_" #item,			\
 163				   offsetof(typeof(ent), item),		\
 164				   sizeof(ent.item),			\
 165				   is_signed_type(type), FILTER_OTHER);	\
 166	if (ret)							\
 167		return ret;
 168
 169static int trace_define_generic_fields(void)
 170{
 171	int ret;
 172
 173	__generic_field(int, CPU, FILTER_CPU);
 174	__generic_field(int, cpu, FILTER_CPU);
 175	__generic_field(char *, COMM, FILTER_COMM);
 176	__generic_field(char *, comm, FILTER_COMM);
 177
 178	return ret;
 179}
 180
 181static int trace_define_common_fields(void)
 182{
 183	int ret;
 184	struct trace_entry ent;
 185
 186	__common_field(unsigned short, type);
 187	__common_field(unsigned char, flags);
 188	__common_field(unsigned char, preempt_count);
 189	__common_field(int, pid);
 190
 191	return ret;
 192}
 193
 194static void trace_destroy_fields(struct trace_event_call *call)
 195{
 196	struct ftrace_event_field *field, *next;
 197	struct list_head *head;
 198
 199	head = trace_get_fields(call);
 200	list_for_each_entry_safe(field, next, head, link) {
 201		list_del(&field->link);
 202		kmem_cache_free(field_cachep, field);
 203	}
 204}
 205
 206/*
 207 * run-time version of trace_event_get_offsets_<call>() that returns the last
 208 * accessible offset of trace fields excluding __dynamic_array bytes
 209 */
 210int trace_event_get_offsets(struct trace_event_call *call)
 211{
 212	struct ftrace_event_field *tail;
 213	struct list_head *head;
 214
 215	head = trace_get_fields(call);
 216	/*
 217	 * head->next points to the last field with the largest offset,
 218	 * since it was added last by trace_define_field()
 219	 */
 220	tail = list_first_entry(head, struct ftrace_event_field, link);
 221	return tail->offset + tail->size;
 222}
 223
 224int trace_event_raw_init(struct trace_event_call *call)
 225{
 226	int id;
 227
 228	id = register_trace_event(&call->event);
 229	if (!id)
 230		return -ENODEV;
 231
 232	return 0;
 233}
 234EXPORT_SYMBOL_GPL(trace_event_raw_init);
 235
 236bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 237{
 238	struct trace_array *tr = trace_file->tr;
 239	struct trace_array_cpu *data;
 240	struct trace_pid_list *pid_list;
 241
 242	pid_list = rcu_dereference_sched(tr->filtered_pids);
 243	if (!pid_list)
 244		return false;
 245
 246	data = this_cpu_ptr(tr->trace_buffer.data);
 247
 248	return data->ignore_pid;
 249}
 250EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 251
 252void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 253				 struct trace_event_file *trace_file,
 254				 unsigned long len)
 255{
 256	struct trace_event_call *event_call = trace_file->event_call;
 257
 258	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 259	    trace_event_ignore_this_pid(trace_file))
 260		return NULL;
 261
 262	local_save_flags(fbuffer->flags);
 263	fbuffer->pc = preempt_count();
 264	/*
 265	 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
 266	 * preemption (adding one to the preempt_count). Since we are
 267	 * interested in the preempt_count at the time the tracepoint was
 268	 * hit, we need to subtract one to offset the increment.
 269	 */
 270	if (IS_ENABLED(CONFIG_PREEMPT))
 271		fbuffer->pc--;
 272	fbuffer->trace_file = trace_file;
 273
 274	fbuffer->event =
 275		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 276						event_call->event.type, len,
 277						fbuffer->flags, fbuffer->pc);
 278	if (!fbuffer->event)
 279		return NULL;
 280
 281	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 282	return fbuffer->entry;
 283}
 284EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 285
 286int trace_event_reg(struct trace_event_call *call,
 287		    enum trace_reg type, void *data)
 288{
 289	struct trace_event_file *file = data;
 290
 291	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 292	switch (type) {
 293	case TRACE_REG_REGISTER:
 294		return tracepoint_probe_register(call->tp,
 295						 call->class->probe,
 296						 file);
 297	case TRACE_REG_UNREGISTER:
 298		tracepoint_probe_unregister(call->tp,
 299					    call->class->probe,
 300					    file);
 301		return 0;
 302
 303#ifdef CONFIG_PERF_EVENTS
 304	case TRACE_REG_PERF_REGISTER:
 305		return tracepoint_probe_register(call->tp,
 306						 call->class->perf_probe,
 307						 call);
 308	case TRACE_REG_PERF_UNREGISTER:
 309		tracepoint_probe_unregister(call->tp,
 310					    call->class->perf_probe,
 311					    call);
 312		return 0;
 313	case TRACE_REG_PERF_OPEN:
 314	case TRACE_REG_PERF_CLOSE:
 315	case TRACE_REG_PERF_ADD:
 316	case TRACE_REG_PERF_DEL:
 317		return 0;
 318#endif
 319	}
 320	return 0;
 321}
 322EXPORT_SYMBOL_GPL(trace_event_reg);
 323
 324void trace_event_enable_cmd_record(bool enable)
 325{
 326	struct trace_event_file *file;
 327	struct trace_array *tr;
 328
 329	mutex_lock(&event_mutex);
 330	do_for_each_event_file(tr, file) {
 331
 332		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 333			continue;
 334
 335		if (enable) {
 336			tracing_start_cmdline_record();
 337			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338		} else {
 339			tracing_stop_cmdline_record();
 340			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 341		}
 342	} while_for_each_event_file();
 343	mutex_unlock(&event_mutex);
 344}
 345
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 346static int __ftrace_event_enable_disable(struct trace_event_file *file,
 347					 int enable, int soft_disable)
 348{
 349	struct trace_event_call *call = file->event_call;
 350	struct trace_array *tr = file->tr;
 351	unsigned long file_flags = file->flags;
 352	int ret = 0;
 353	int disable;
 354
 355	switch (enable) {
 356	case 0:
 357		/*
 358		 * When soft_disable is set and enable is cleared, the sm_ref
 359		 * reference counter is decremented. If it reaches 0, we want
 360		 * to clear the SOFT_DISABLED flag but leave the event in the
 361		 * state that it was. That is, if the event was enabled and
 362		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 363		 * is set we do not want the event to be enabled before we
 364		 * clear the bit.
 365		 *
 366		 * When soft_disable is not set but the SOFT_MODE flag is,
 367		 * we do nothing. Do not disable the tracepoint, otherwise
 368		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 369		 */
 370		if (soft_disable) {
 371			if (atomic_dec_return(&file->sm_ref) > 0)
 372				break;
 373			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 374			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 375		} else
 376			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 377
 378		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 379			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 380			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 381				tracing_stop_cmdline_record();
 382				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 383			}
 
 
 
 
 
 
 384			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 385		}
 386		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 387		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 388			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 389		else
 390			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 391		break;
 392	case 1:
 393		/*
 394		 * When soft_disable is set and enable is set, we want to
 395		 * register the tracepoint for the event, but leave the event
 396		 * as is. That means, if the event was already enabled, we do
 397		 * nothing (but set SOFT_MODE). If the event is disabled, we
 398		 * set SOFT_DISABLED before enabling the event tracepoint, so
 399		 * it still seems to be disabled.
 400		 */
 401		if (!soft_disable)
 402			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 403		else {
 404			if (atomic_inc_return(&file->sm_ref) > 1)
 405				break;
 406			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 407		}
 408
 409		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 
 410
 411			/* Keep the event disabled, when going to SOFT_MODE. */
 412			if (soft_disable)
 413				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 414
 415			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 
 416				tracing_start_cmdline_record();
 417				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 418			}
 
 
 
 
 
 
 
 419			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 420			if (ret) {
 421				tracing_stop_cmdline_record();
 
 
 
 422				pr_info("event trace: Could not enable event "
 423					"%s\n", trace_event_name(call));
 424				break;
 425			}
 426			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 427
 428			/* WAS_ENABLED gets set but never cleared. */
 429			call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 430		}
 431		break;
 432	}
 433
 434	/* Enable or disable use of trace_buffered_event */
 435	if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 436	    (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 437		if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 438			trace_buffered_event_enable();
 439		else
 440			trace_buffered_event_disable();
 441	}
 442
 443	return ret;
 444}
 445
 446int trace_event_enable_disable(struct trace_event_file *file,
 447			       int enable, int soft_disable)
 448{
 449	return __ftrace_event_enable_disable(file, enable, soft_disable);
 450}
 451
 452static int ftrace_event_enable_disable(struct trace_event_file *file,
 453				       int enable)
 454{
 455	return __ftrace_event_enable_disable(file, enable, 0);
 456}
 457
 458static void ftrace_clear_events(struct trace_array *tr)
 459{
 460	struct trace_event_file *file;
 461
 462	mutex_lock(&event_mutex);
 463	list_for_each_entry(file, &tr->events, list) {
 464		ftrace_event_enable_disable(file, 0);
 465	}
 466	mutex_unlock(&event_mutex);
 467}
 468
 469static void
 470event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 471{
 472	struct trace_pid_list *pid_list;
 473	struct trace_array *tr = data;
 474
 475	pid_list = rcu_dereference_sched(tr->filtered_pids);
 476	trace_filter_add_remove_task(pid_list, NULL, task);
 477}
 478
 479static void
 480event_filter_pid_sched_process_fork(void *data,
 481				    struct task_struct *self,
 482				    struct task_struct *task)
 483{
 484	struct trace_pid_list *pid_list;
 485	struct trace_array *tr = data;
 486
 487	pid_list = rcu_dereference_sched(tr->filtered_pids);
 488	trace_filter_add_remove_task(pid_list, self, task);
 489}
 490
 491void trace_event_follow_fork(struct trace_array *tr, bool enable)
 492{
 493	if (enable) {
 494		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 495						       tr, INT_MIN);
 496		register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 497						       tr, INT_MAX);
 498	} else {
 499		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 500						    tr);
 501		unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 502						    tr);
 503	}
 504}
 505
 506static void
 507event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 508		    struct task_struct *prev, struct task_struct *next)
 509{
 510	struct trace_array *tr = data;
 511	struct trace_pid_list *pid_list;
 512
 513	pid_list = rcu_dereference_sched(tr->filtered_pids);
 514
 515	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 516		       trace_ignore_this_task(pid_list, prev) &&
 517		       trace_ignore_this_task(pid_list, next));
 518}
 519
 520static void
 521event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 522		    struct task_struct *prev, struct task_struct *next)
 523{
 524	struct trace_array *tr = data;
 525	struct trace_pid_list *pid_list;
 526
 527	pid_list = rcu_dereference_sched(tr->filtered_pids);
 528
 529	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 530		       trace_ignore_this_task(pid_list, next));
 531}
 532
 533static void
 534event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 535{
 536	struct trace_array *tr = data;
 537	struct trace_pid_list *pid_list;
 538
 539	/* Nothing to do if we are already tracing */
 540	if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 541		return;
 542
 543	pid_list = rcu_dereference_sched(tr->filtered_pids);
 544
 545	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 546		       trace_ignore_this_task(pid_list, task));
 547}
 548
 549static void
 550event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 551{
 552	struct trace_array *tr = data;
 553	struct trace_pid_list *pid_list;
 554
 555	/* Nothing to do if we are not tracing */
 556	if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 557		return;
 558
 559	pid_list = rcu_dereference_sched(tr->filtered_pids);
 560
 561	/* Set tracing if current is enabled */
 562	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 563		       trace_ignore_this_task(pid_list, current));
 564}
 565
 566static void __ftrace_clear_event_pids(struct trace_array *tr)
 567{
 568	struct trace_pid_list *pid_list;
 569	struct trace_event_file *file;
 570	int cpu;
 571
 572	pid_list = rcu_dereference_protected(tr->filtered_pids,
 573					     lockdep_is_held(&event_mutex));
 574	if (!pid_list)
 575		return;
 576
 577	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 578	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 579
 580	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 581	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 582
 583	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 584	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 585
 586	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 587	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 588
 589	list_for_each_entry(file, &tr->events, list) {
 590		clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 591	}
 592
 593	for_each_possible_cpu(cpu)
 594		per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 595
 596	rcu_assign_pointer(tr->filtered_pids, NULL);
 597
 598	/* Wait till all users are no longer using pid filtering */
 599	synchronize_sched();
 600
 601	trace_free_pid_list(pid_list);
 602}
 603
 604static void ftrace_clear_event_pids(struct trace_array *tr)
 605{
 606	mutex_lock(&event_mutex);
 607	__ftrace_clear_event_pids(tr);
 608	mutex_unlock(&event_mutex);
 609}
 610
 611static void __put_system(struct event_subsystem *system)
 612{
 613	struct event_filter *filter = system->filter;
 614
 615	WARN_ON_ONCE(system_refcount(system) == 0);
 616	if (system_refcount_dec(system))
 617		return;
 618
 619	list_del(&system->list);
 620
 621	if (filter) {
 622		kfree(filter->filter_string);
 623		kfree(filter);
 624	}
 625	kfree_const(system->name);
 626	kfree(system);
 627}
 628
 629static void __get_system(struct event_subsystem *system)
 630{
 631	WARN_ON_ONCE(system_refcount(system) == 0);
 632	system_refcount_inc(system);
 633}
 634
 635static void __get_system_dir(struct trace_subsystem_dir *dir)
 636{
 637	WARN_ON_ONCE(dir->ref_count == 0);
 638	dir->ref_count++;
 639	__get_system(dir->subsystem);
 640}
 641
 642static void __put_system_dir(struct trace_subsystem_dir *dir)
 643{
 644	WARN_ON_ONCE(dir->ref_count == 0);
 645	/* If the subsystem is about to be freed, the dir must be too */
 646	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 647
 648	__put_system(dir->subsystem);
 649	if (!--dir->ref_count)
 650		kfree(dir);
 651}
 652
 653static void put_system(struct trace_subsystem_dir *dir)
 654{
 655	mutex_lock(&event_mutex);
 656	__put_system_dir(dir);
 657	mutex_unlock(&event_mutex);
 658}
 659
 660static void remove_subsystem(struct trace_subsystem_dir *dir)
 661{
 662	if (!dir)
 663		return;
 664
 665	if (!--dir->nr_events) {
 666		tracefs_remove_recursive(dir->entry);
 667		list_del(&dir->list);
 668		__put_system_dir(dir);
 669	}
 670}
 671
 672static void remove_event_file_dir(struct trace_event_file *file)
 673{
 674	struct dentry *dir = file->dir;
 675	struct dentry *child;
 676
 677	if (dir) {
 678		spin_lock(&dir->d_lock);	/* probably unneeded */
 679		list_for_each_entry(child, &dir->d_subdirs, d_child) {
 680			if (d_really_is_positive(child))	/* probably unneeded */
 681				d_inode(child)->i_private = NULL;
 682		}
 683		spin_unlock(&dir->d_lock);
 684
 685		tracefs_remove_recursive(dir);
 686	}
 687
 688	list_del(&file->list);
 689	remove_subsystem(file->system);
 690	free_event_filter(file->filter);
 691	kmem_cache_free(file_cachep, file);
 692}
 693
 694/*
 695 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 696 */
 697static int
 698__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 699			      const char *sub, const char *event, int set)
 700{
 701	struct trace_event_file *file;
 702	struct trace_event_call *call;
 703	const char *name;
 704	int ret = -EINVAL;
 705	int eret = 0;
 706
 707	list_for_each_entry(file, &tr->events, list) {
 708
 709		call = file->event_call;
 710		name = trace_event_name(call);
 711
 712		if (!name || !call->class || !call->class->reg)
 713			continue;
 714
 715		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 716			continue;
 717
 718		if (match &&
 719		    strcmp(match, name) != 0 &&
 720		    strcmp(match, call->class->system) != 0)
 721			continue;
 722
 723		if (sub && strcmp(sub, call->class->system) != 0)
 724			continue;
 725
 726		if (event && strcmp(event, name) != 0)
 727			continue;
 728
 729		ret = ftrace_event_enable_disable(file, set);
 730
 731		/*
 732		 * Save the first error and return that. Some events
 733		 * may still have been enabled, but let the user
 734		 * know that something went wrong.
 735		 */
 736		if (ret && !eret)
 737			eret = ret;
 738
 739		ret = eret;
 740	}
 741
 742	return ret;
 743}
 744
 745static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 746				  const char *sub, const char *event, int set)
 747{
 748	int ret;
 749
 750	mutex_lock(&event_mutex);
 751	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 752	mutex_unlock(&event_mutex);
 753
 754	return ret;
 755}
 756
 757static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 758{
 759	char *event = NULL, *sub = NULL, *match;
 760	int ret;
 761
 762	/*
 763	 * The buf format can be <subsystem>:<event-name>
 764	 *  *:<event-name> means any event by that name.
 765	 *  :<event-name> is the same.
 766	 *
 767	 *  <subsystem>:* means all events in that subsystem
 768	 *  <subsystem>: means the same.
 769	 *
 770	 *  <name> (no ':') means all events in a subsystem with
 771	 *  the name <name> or any event that matches <name>
 772	 */
 773
 774	match = strsep(&buf, ":");
 775	if (buf) {
 776		sub = match;
 777		event = buf;
 778		match = NULL;
 779
 780		if (!strlen(sub) || strcmp(sub, "*") == 0)
 781			sub = NULL;
 782		if (!strlen(event) || strcmp(event, "*") == 0)
 783			event = NULL;
 784	}
 785
 786	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 787
 788	/* Put back the colon to allow this to be called again */
 789	if (buf)
 790		*(buf - 1) = ':';
 791
 792	return ret;
 793}
 794
 795/**
 796 * trace_set_clr_event - enable or disable an event
 797 * @system: system name to match (NULL for any system)
 798 * @event: event name to match (NULL for all events, within system)
 799 * @set: 1 to enable, 0 to disable
 800 *
 801 * This is a way for other parts of the kernel to enable or disable
 802 * event recording.
 803 *
 804 * Returns 0 on success, -EINVAL if the parameters do not match any
 805 * registered events.
 806 */
 807int trace_set_clr_event(const char *system, const char *event, int set)
 808{
 809	struct trace_array *tr = top_trace_array();
 810
 811	if (!tr)
 812		return -ENODEV;
 813
 814	return __ftrace_set_clr_event(tr, NULL, system, event, set);
 815}
 816EXPORT_SYMBOL_GPL(trace_set_clr_event);
 817
 818/* 128 should be much more than enough */
 819#define EVENT_BUF_SIZE		127
 820
 821static ssize_t
 822ftrace_event_write(struct file *file, const char __user *ubuf,
 823		   size_t cnt, loff_t *ppos)
 824{
 825	struct trace_parser parser;
 826	struct seq_file *m = file->private_data;
 827	struct trace_array *tr = m->private;
 828	ssize_t read, ret;
 829
 830	if (!cnt)
 831		return 0;
 832
 833	ret = tracing_update_buffers();
 834	if (ret < 0)
 835		return ret;
 836
 837	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 838		return -ENOMEM;
 839
 840	read = trace_get_user(&parser, ubuf, cnt, ppos);
 841
 842	if (read >= 0 && trace_parser_loaded((&parser))) {
 843		int set = 1;
 844
 845		if (*parser.buffer == '!')
 846			set = 0;
 847
 848		parser.buffer[parser.idx] = 0;
 849
 850		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 851		if (ret)
 852			goto out_put;
 853	}
 854
 855	ret = read;
 856
 857 out_put:
 858	trace_parser_put(&parser);
 859
 860	return ret;
 861}
 862
 863static void *
 864t_next(struct seq_file *m, void *v, loff_t *pos)
 865{
 866	struct trace_event_file *file = v;
 867	struct trace_event_call *call;
 868	struct trace_array *tr = m->private;
 869
 870	(*pos)++;
 871
 872	list_for_each_entry_continue(file, &tr->events, list) {
 873		call = file->event_call;
 874		/*
 875		 * The ftrace subsystem is for showing formats only.
 876		 * They can not be enabled or disabled via the event files.
 877		 */
 878		if (call->class && call->class->reg &&
 879		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 880			return file;
 881	}
 882
 883	return NULL;
 884}
 885
 886static void *t_start(struct seq_file *m, loff_t *pos)
 887{
 888	struct trace_event_file *file;
 889	struct trace_array *tr = m->private;
 890	loff_t l;
 891
 892	mutex_lock(&event_mutex);
 893
 894	file = list_entry(&tr->events, struct trace_event_file, list);
 895	for (l = 0; l <= *pos; ) {
 896		file = t_next(m, file, &l);
 897		if (!file)
 898			break;
 899	}
 900	return file;
 901}
 902
 903static void *
 904s_next(struct seq_file *m, void *v, loff_t *pos)
 905{
 906	struct trace_event_file *file = v;
 907	struct trace_array *tr = m->private;
 908
 909	(*pos)++;
 910
 911	list_for_each_entry_continue(file, &tr->events, list) {
 912		if (file->flags & EVENT_FILE_FL_ENABLED)
 913			return file;
 914	}
 915
 916	return NULL;
 917}
 918
 919static void *s_start(struct seq_file *m, loff_t *pos)
 920{
 921	struct trace_event_file *file;
 922	struct trace_array *tr = m->private;
 923	loff_t l;
 924
 925	mutex_lock(&event_mutex);
 926
 927	file = list_entry(&tr->events, struct trace_event_file, list);
 928	for (l = 0; l <= *pos; ) {
 929		file = s_next(m, file, &l);
 930		if (!file)
 931			break;
 932	}
 933	return file;
 934}
 935
 936static int t_show(struct seq_file *m, void *v)
 937{
 938	struct trace_event_file *file = v;
 939	struct trace_event_call *call = file->event_call;
 940
 941	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 942		seq_printf(m, "%s:", call->class->system);
 943	seq_printf(m, "%s\n", trace_event_name(call));
 944
 945	return 0;
 946}
 947
 948static void t_stop(struct seq_file *m, void *p)
 949{
 950	mutex_unlock(&event_mutex);
 951}
 952
 953static void *
 954p_next(struct seq_file *m, void *v, loff_t *pos)
 955{
 956	struct trace_array *tr = m->private;
 957	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 958
 959	return trace_pid_next(pid_list, v, pos);
 960}
 961
 962static void *p_start(struct seq_file *m, loff_t *pos)
 963	__acquires(RCU)
 964{
 965	struct trace_pid_list *pid_list;
 966	struct trace_array *tr = m->private;
 967
 968	/*
 969	 * Grab the mutex, to keep calls to p_next() having the same
 970	 * tr->filtered_pids as p_start() has.
 971	 * If we just passed the tr->filtered_pids around, then RCU would
 972	 * have been enough, but doing that makes things more complex.
 973	 */
 974	mutex_lock(&event_mutex);
 975	rcu_read_lock_sched();
 976
 977	pid_list = rcu_dereference_sched(tr->filtered_pids);
 978
 979	if (!pid_list)
 980		return NULL;
 981
 982	return trace_pid_start(pid_list, pos);
 983}
 984
 985static void p_stop(struct seq_file *m, void *p)
 986	__releases(RCU)
 987{
 988	rcu_read_unlock_sched();
 989	mutex_unlock(&event_mutex);
 990}
 991
 992static ssize_t
 993event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 994		  loff_t *ppos)
 995{
 996	struct trace_event_file *file;
 997	unsigned long flags;
 998	char buf[4] = "0";
 999
1000	mutex_lock(&event_mutex);
1001	file = event_file_data(filp);
1002	if (likely(file))
1003		flags = file->flags;
1004	mutex_unlock(&event_mutex);
1005
1006	if (!file)
1007		return -ENODEV;
1008
1009	if (flags & EVENT_FILE_FL_ENABLED &&
1010	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1011		strcpy(buf, "1");
1012
1013	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1014	    flags & EVENT_FILE_FL_SOFT_MODE)
1015		strcat(buf, "*");
1016
1017	strcat(buf, "\n");
1018
1019	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1020}
1021
1022static ssize_t
1023event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1024		   loff_t *ppos)
1025{
1026	struct trace_event_file *file;
1027	unsigned long val;
1028	int ret;
1029
1030	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1031	if (ret)
1032		return ret;
1033
1034	ret = tracing_update_buffers();
1035	if (ret < 0)
1036		return ret;
1037
1038	switch (val) {
1039	case 0:
1040	case 1:
1041		ret = -ENODEV;
1042		mutex_lock(&event_mutex);
1043		file = event_file_data(filp);
1044		if (likely(file))
1045			ret = ftrace_event_enable_disable(file, val);
1046		mutex_unlock(&event_mutex);
1047		break;
1048
1049	default:
1050		return -EINVAL;
1051	}
1052
1053	*ppos += cnt;
1054
1055	return ret ? ret : cnt;
1056}
1057
1058static ssize_t
1059system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1060		   loff_t *ppos)
1061{
1062	const char set_to_char[4] = { '?', '0', '1', 'X' };
1063	struct trace_subsystem_dir *dir = filp->private_data;
1064	struct event_subsystem *system = dir->subsystem;
1065	struct trace_event_call *call;
1066	struct trace_event_file *file;
1067	struct trace_array *tr = dir->tr;
1068	char buf[2];
1069	int set = 0;
1070	int ret;
1071
1072	mutex_lock(&event_mutex);
1073	list_for_each_entry(file, &tr->events, list) {
1074		call = file->event_call;
1075		if (!trace_event_name(call) || !call->class || !call->class->reg)
1076			continue;
1077
1078		if (system && strcmp(call->class->system, system->name) != 0)
1079			continue;
1080
1081		/*
1082		 * We need to find out if all the events are set
1083		 * or if all events or cleared, or if we have
1084		 * a mixture.
1085		 */
1086		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1087
1088		/*
1089		 * If we have a mixture, no need to look further.
1090		 */
1091		if (set == 3)
1092			break;
1093	}
1094	mutex_unlock(&event_mutex);
1095
1096	buf[0] = set_to_char[set];
1097	buf[1] = '\n';
1098
1099	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1100
1101	return ret;
1102}
1103
1104static ssize_t
1105system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1106		    loff_t *ppos)
1107{
1108	struct trace_subsystem_dir *dir = filp->private_data;
1109	struct event_subsystem *system = dir->subsystem;
1110	const char *name = NULL;
1111	unsigned long val;
1112	ssize_t ret;
1113
1114	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1115	if (ret)
1116		return ret;
1117
1118	ret = tracing_update_buffers();
1119	if (ret < 0)
1120		return ret;
1121
1122	if (val != 0 && val != 1)
1123		return -EINVAL;
1124
1125	/*
1126	 * Opening of "enable" adds a ref count to system,
1127	 * so the name is safe to use.
1128	 */
1129	if (system)
1130		name = system->name;
1131
1132	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1133	if (ret)
1134		goto out;
1135
1136	ret = cnt;
1137
1138out:
1139	*ppos += cnt;
1140
1141	return ret;
1142}
1143
1144enum {
1145	FORMAT_HEADER		= 1,
1146	FORMAT_FIELD_SEPERATOR	= 2,
1147	FORMAT_PRINTFMT		= 3,
1148};
1149
1150static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1151{
1152	struct trace_event_call *call = event_file_data(m->private);
1153	struct list_head *common_head = &ftrace_common_fields;
1154	struct list_head *head = trace_get_fields(call);
1155	struct list_head *node = v;
1156
1157	(*pos)++;
1158
1159	switch ((unsigned long)v) {
1160	case FORMAT_HEADER:
1161		node = common_head;
1162		break;
1163
1164	case FORMAT_FIELD_SEPERATOR:
1165		node = head;
1166		break;
1167
1168	case FORMAT_PRINTFMT:
1169		/* all done */
1170		return NULL;
1171	}
1172
1173	node = node->prev;
1174	if (node == common_head)
1175		return (void *)FORMAT_FIELD_SEPERATOR;
1176	else if (node == head)
1177		return (void *)FORMAT_PRINTFMT;
1178	else
1179		return node;
1180}
1181
1182static int f_show(struct seq_file *m, void *v)
1183{
1184	struct trace_event_call *call = event_file_data(m->private);
1185	struct ftrace_event_field *field;
1186	const char *array_descriptor;
1187
1188	switch ((unsigned long)v) {
1189	case FORMAT_HEADER:
1190		seq_printf(m, "name: %s\n", trace_event_name(call));
1191		seq_printf(m, "ID: %d\n", call->event.type);
1192		seq_puts(m, "format:\n");
1193		return 0;
1194
1195	case FORMAT_FIELD_SEPERATOR:
1196		seq_putc(m, '\n');
1197		return 0;
1198
1199	case FORMAT_PRINTFMT:
1200		seq_printf(m, "\nprint fmt: %s\n",
1201			   call->print_fmt);
1202		return 0;
1203	}
1204
1205	field = list_entry(v, struct ftrace_event_field, link);
1206	/*
1207	 * Smartly shows the array type(except dynamic array).
1208	 * Normal:
1209	 *	field:TYPE VAR
1210	 * If TYPE := TYPE[LEN], it is shown:
1211	 *	field:TYPE VAR[LEN]
1212	 */
1213	array_descriptor = strchr(field->type, '[');
1214
1215	if (!strncmp(field->type, "__data_loc", 10))
1216		array_descriptor = NULL;
1217
1218	if (!array_descriptor)
1219		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1220			   field->type, field->name, field->offset,
1221			   field->size, !!field->is_signed);
1222	else
1223		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1224			   (int)(array_descriptor - field->type),
1225			   field->type, field->name,
1226			   array_descriptor, field->offset,
1227			   field->size, !!field->is_signed);
1228
1229	return 0;
1230}
1231
1232static void *f_start(struct seq_file *m, loff_t *pos)
1233{
1234	void *p = (void *)FORMAT_HEADER;
1235	loff_t l = 0;
1236
1237	/* ->stop() is called even if ->start() fails */
1238	mutex_lock(&event_mutex);
1239	if (!event_file_data(m->private))
1240		return ERR_PTR(-ENODEV);
1241
1242	while (l < *pos && p)
1243		p = f_next(m, p, &l);
1244
1245	return p;
1246}
1247
1248static void f_stop(struct seq_file *m, void *p)
1249{
1250	mutex_unlock(&event_mutex);
1251}
1252
1253static const struct seq_operations trace_format_seq_ops = {
1254	.start		= f_start,
1255	.next		= f_next,
1256	.stop		= f_stop,
1257	.show		= f_show,
1258};
1259
1260static int trace_format_open(struct inode *inode, struct file *file)
1261{
1262	struct seq_file *m;
1263	int ret;
1264
1265	ret = seq_open(file, &trace_format_seq_ops);
1266	if (ret < 0)
1267		return ret;
1268
1269	m = file->private_data;
1270	m->private = file;
1271
1272	return 0;
1273}
1274
1275static ssize_t
1276event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1277{
1278	int id = (long)event_file_data(filp);
1279	char buf[32];
1280	int len;
1281
1282	if (*ppos)
1283		return 0;
1284
1285	if (unlikely(!id))
1286		return -ENODEV;
1287
1288	len = sprintf(buf, "%d\n", id);
1289
1290	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1291}
1292
1293static ssize_t
1294event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1295		  loff_t *ppos)
1296{
1297	struct trace_event_file *file;
1298	struct trace_seq *s;
1299	int r = -ENODEV;
1300
1301	if (*ppos)
1302		return 0;
1303
1304	s = kmalloc(sizeof(*s), GFP_KERNEL);
1305
1306	if (!s)
1307		return -ENOMEM;
1308
1309	trace_seq_init(s);
1310
1311	mutex_lock(&event_mutex);
1312	file = event_file_data(filp);
1313	if (file)
1314		print_event_filter(file, s);
1315	mutex_unlock(&event_mutex);
1316
1317	if (file)
1318		r = simple_read_from_buffer(ubuf, cnt, ppos,
1319					    s->buffer, trace_seq_used(s));
1320
1321	kfree(s);
1322
1323	return r;
1324}
1325
1326static ssize_t
1327event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1328		   loff_t *ppos)
1329{
1330	struct trace_event_file *file;
1331	char *buf;
1332	int err = -ENODEV;
1333
1334	if (cnt >= PAGE_SIZE)
1335		return -EINVAL;
1336
1337	buf = memdup_user_nul(ubuf, cnt);
1338	if (IS_ERR(buf))
1339		return PTR_ERR(buf);
1340
1341	mutex_lock(&event_mutex);
1342	file = event_file_data(filp);
1343	if (file)
1344		err = apply_event_filter(file, buf);
1345	mutex_unlock(&event_mutex);
1346
1347	kfree(buf);
1348	if (err < 0)
1349		return err;
1350
1351	*ppos += cnt;
1352
1353	return cnt;
1354}
1355
1356static LIST_HEAD(event_subsystems);
1357
1358static int subsystem_open(struct inode *inode, struct file *filp)
1359{
1360	struct event_subsystem *system = NULL;
1361	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1362	struct trace_array *tr;
1363	int ret;
1364
1365	if (tracing_is_disabled())
1366		return -ENODEV;
1367
1368	/* Make sure the system still exists */
 
1369	mutex_lock(&trace_types_lock);
1370	mutex_lock(&event_mutex);
1371	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1372		list_for_each_entry(dir, &tr->systems, list) {
1373			if (dir == inode->i_private) {
1374				/* Don't open systems with no events */
1375				if (dir->nr_events) {
1376					__get_system_dir(dir);
1377					system = dir->subsystem;
1378				}
1379				goto exit_loop;
1380			}
1381		}
1382	}
1383 exit_loop:
 
1384	mutex_unlock(&event_mutex);
1385	mutex_unlock(&trace_types_lock);
1386
1387	if (!system)
1388		return -ENODEV;
1389
1390	/* Some versions of gcc think dir can be uninitialized here */
1391	WARN_ON(!dir);
1392
1393	/* Still need to increment the ref count of the system */
1394	if (trace_array_get(tr) < 0) {
1395		put_system(dir);
1396		return -ENODEV;
1397	}
1398
1399	ret = tracing_open_generic(inode, filp);
1400	if (ret < 0) {
1401		trace_array_put(tr);
1402		put_system(dir);
1403	}
1404
1405	return ret;
1406}
1407
1408static int system_tr_open(struct inode *inode, struct file *filp)
1409{
1410	struct trace_subsystem_dir *dir;
1411	struct trace_array *tr = inode->i_private;
1412	int ret;
1413
1414	if (tracing_is_disabled())
1415		return -ENODEV;
1416
1417	if (trace_array_get(tr) < 0)
1418		return -ENODEV;
1419
1420	/* Make a temporary dir that has no system but points to tr */
1421	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1422	if (!dir) {
1423		trace_array_put(tr);
1424		return -ENOMEM;
1425	}
1426
1427	dir->tr = tr;
1428
1429	ret = tracing_open_generic(inode, filp);
1430	if (ret < 0) {
1431		trace_array_put(tr);
1432		kfree(dir);
1433		return ret;
1434	}
1435
1436	filp->private_data = dir;
1437
1438	return 0;
1439}
1440
1441static int subsystem_release(struct inode *inode, struct file *file)
1442{
1443	struct trace_subsystem_dir *dir = file->private_data;
1444
1445	trace_array_put(dir->tr);
1446
1447	/*
1448	 * If dir->subsystem is NULL, then this is a temporary
1449	 * descriptor that was made for a trace_array to enable
1450	 * all subsystems.
1451	 */
1452	if (dir->subsystem)
1453		put_system(dir);
1454	else
1455		kfree(dir);
1456
1457	return 0;
1458}
1459
1460static ssize_t
1461subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1462		      loff_t *ppos)
1463{
1464	struct trace_subsystem_dir *dir = filp->private_data;
1465	struct event_subsystem *system = dir->subsystem;
1466	struct trace_seq *s;
1467	int r;
1468
1469	if (*ppos)
1470		return 0;
1471
1472	s = kmalloc(sizeof(*s), GFP_KERNEL);
1473	if (!s)
1474		return -ENOMEM;
1475
1476	trace_seq_init(s);
1477
1478	print_subsystem_event_filter(system, s);
1479	r = simple_read_from_buffer(ubuf, cnt, ppos,
1480				    s->buffer, trace_seq_used(s));
1481
1482	kfree(s);
1483
1484	return r;
1485}
1486
1487static ssize_t
1488subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1489		       loff_t *ppos)
1490{
1491	struct trace_subsystem_dir *dir = filp->private_data;
1492	char *buf;
1493	int err;
1494
1495	if (cnt >= PAGE_SIZE)
1496		return -EINVAL;
1497
1498	buf = memdup_user_nul(ubuf, cnt);
1499	if (IS_ERR(buf))
1500		return PTR_ERR(buf);
1501
1502	err = apply_subsystem_event_filter(dir, buf);
1503	kfree(buf);
1504	if (err < 0)
1505		return err;
1506
1507	*ppos += cnt;
1508
1509	return cnt;
1510}
1511
1512static ssize_t
1513show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1514{
1515	int (*func)(struct trace_seq *s) = filp->private_data;
1516	struct trace_seq *s;
1517	int r;
1518
1519	if (*ppos)
1520		return 0;
1521
1522	s = kmalloc(sizeof(*s), GFP_KERNEL);
1523	if (!s)
1524		return -ENOMEM;
1525
1526	trace_seq_init(s);
1527
1528	func(s);
1529	r = simple_read_from_buffer(ubuf, cnt, ppos,
1530				    s->buffer, trace_seq_used(s));
1531
1532	kfree(s);
1533
1534	return r;
1535}
1536
1537static void ignore_task_cpu(void *data)
1538{
1539	struct trace_array *tr = data;
1540	struct trace_pid_list *pid_list;
1541
1542	/*
1543	 * This function is called by on_each_cpu() while the
1544	 * event_mutex is held.
1545	 */
1546	pid_list = rcu_dereference_protected(tr->filtered_pids,
1547					     mutex_is_locked(&event_mutex));
1548
1549	this_cpu_write(tr->trace_buffer.data->ignore_pid,
1550		       trace_ignore_this_task(pid_list, current));
1551}
1552
1553static ssize_t
1554ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1555		       size_t cnt, loff_t *ppos)
1556{
1557	struct seq_file *m = filp->private_data;
1558	struct trace_array *tr = m->private;
1559	struct trace_pid_list *filtered_pids = NULL;
1560	struct trace_pid_list *pid_list;
1561	struct trace_event_file *file;
1562	ssize_t ret;
1563
1564	if (!cnt)
1565		return 0;
1566
1567	ret = tracing_update_buffers();
1568	if (ret < 0)
1569		return ret;
1570
1571	mutex_lock(&event_mutex);
1572
1573	filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1574					     lockdep_is_held(&event_mutex));
1575
1576	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1577	if (ret < 0)
1578		goto out;
1579
1580	rcu_assign_pointer(tr->filtered_pids, pid_list);
1581
1582	list_for_each_entry(file, &tr->events, list) {
1583		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1584	}
1585
1586	if (filtered_pids) {
1587		synchronize_sched();
1588		trace_free_pid_list(filtered_pids);
1589	} else if (pid_list) {
1590		/*
1591		 * Register a probe that is called before all other probes
1592		 * to set ignore_pid if next or prev do not match.
1593		 * Register a probe this is called after all other probes
1594		 * to only keep ignore_pid set if next pid matches.
1595		 */
1596		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1597						 tr, INT_MAX);
1598		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1599						 tr, 0);
1600
1601		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1602						 tr, INT_MAX);
1603		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1604						 tr, 0);
1605
1606		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1607						     tr, INT_MAX);
1608		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1609						     tr, 0);
1610
1611		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1612						 tr, INT_MAX);
1613		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1614						 tr, 0);
1615	}
1616
1617	/*
1618	 * Ignoring of pids is done at task switch. But we have to
1619	 * check for those tasks that are currently running.
1620	 * Always do this in case a pid was appended or removed.
1621	 */
1622	on_each_cpu(ignore_task_cpu, tr, 1);
1623
1624 out:
1625	mutex_unlock(&event_mutex);
1626
1627	if (ret > 0)
1628		*ppos += ret;
1629
1630	return ret;
1631}
1632
1633static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1634static int ftrace_event_set_open(struct inode *inode, struct file *file);
1635static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1636static int ftrace_event_release(struct inode *inode, struct file *file);
1637
1638static const struct seq_operations show_event_seq_ops = {
1639	.start = t_start,
1640	.next = t_next,
1641	.show = t_show,
1642	.stop = t_stop,
1643};
1644
1645static const struct seq_operations show_set_event_seq_ops = {
1646	.start = s_start,
1647	.next = s_next,
1648	.show = t_show,
1649	.stop = t_stop,
1650};
1651
1652static const struct seq_operations show_set_pid_seq_ops = {
1653	.start = p_start,
1654	.next = p_next,
1655	.show = trace_pid_show,
1656	.stop = p_stop,
1657};
1658
1659static const struct file_operations ftrace_avail_fops = {
1660	.open = ftrace_event_avail_open,
1661	.read = seq_read,
1662	.llseek = seq_lseek,
1663	.release = seq_release,
1664};
1665
1666static const struct file_operations ftrace_set_event_fops = {
1667	.open = ftrace_event_set_open,
1668	.read = seq_read,
1669	.write = ftrace_event_write,
1670	.llseek = seq_lseek,
1671	.release = ftrace_event_release,
1672};
1673
1674static const struct file_operations ftrace_set_event_pid_fops = {
1675	.open = ftrace_event_set_pid_open,
1676	.read = seq_read,
1677	.write = ftrace_event_pid_write,
1678	.llseek = seq_lseek,
1679	.release = ftrace_event_release,
1680};
1681
1682static const struct file_operations ftrace_enable_fops = {
1683	.open = tracing_open_generic,
1684	.read = event_enable_read,
1685	.write = event_enable_write,
1686	.llseek = default_llseek,
1687};
1688
1689static const struct file_operations ftrace_event_format_fops = {
1690	.open = trace_format_open,
1691	.read = seq_read,
1692	.llseek = seq_lseek,
1693	.release = seq_release,
1694};
1695
1696static const struct file_operations ftrace_event_id_fops = {
1697	.read = event_id_read,
1698	.llseek = default_llseek,
1699};
1700
1701static const struct file_operations ftrace_event_filter_fops = {
1702	.open = tracing_open_generic,
1703	.read = event_filter_read,
1704	.write = event_filter_write,
1705	.llseek = default_llseek,
1706};
1707
1708static const struct file_operations ftrace_subsystem_filter_fops = {
1709	.open = subsystem_open,
1710	.read = subsystem_filter_read,
1711	.write = subsystem_filter_write,
1712	.llseek = default_llseek,
1713	.release = subsystem_release,
1714};
1715
1716static const struct file_operations ftrace_system_enable_fops = {
1717	.open = subsystem_open,
1718	.read = system_enable_read,
1719	.write = system_enable_write,
1720	.llseek = default_llseek,
1721	.release = subsystem_release,
1722};
1723
1724static const struct file_operations ftrace_tr_enable_fops = {
1725	.open = system_tr_open,
1726	.read = system_enable_read,
1727	.write = system_enable_write,
1728	.llseek = default_llseek,
1729	.release = subsystem_release,
1730};
1731
1732static const struct file_operations ftrace_show_header_fops = {
1733	.open = tracing_open_generic,
1734	.read = show_header,
1735	.llseek = default_llseek,
1736};
1737
1738static int
1739ftrace_event_open(struct inode *inode, struct file *file,
1740		  const struct seq_operations *seq_ops)
1741{
1742	struct seq_file *m;
1743	int ret;
1744
1745	ret = seq_open(file, seq_ops);
1746	if (ret < 0)
1747		return ret;
1748	m = file->private_data;
1749	/* copy tr over to seq ops */
1750	m->private = inode->i_private;
1751
1752	return ret;
1753}
1754
1755static int ftrace_event_release(struct inode *inode, struct file *file)
1756{
1757	struct trace_array *tr = inode->i_private;
1758
1759	trace_array_put(tr);
1760
1761	return seq_release(inode, file);
1762}
1763
1764static int
1765ftrace_event_avail_open(struct inode *inode, struct file *file)
1766{
1767	const struct seq_operations *seq_ops = &show_event_seq_ops;
1768
1769	return ftrace_event_open(inode, file, seq_ops);
1770}
1771
1772static int
1773ftrace_event_set_open(struct inode *inode, struct file *file)
1774{
1775	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1776	struct trace_array *tr = inode->i_private;
1777	int ret;
1778
1779	if (trace_array_get(tr) < 0)
1780		return -ENODEV;
1781
1782	if ((file->f_mode & FMODE_WRITE) &&
1783	    (file->f_flags & O_TRUNC))
1784		ftrace_clear_events(tr);
1785
1786	ret = ftrace_event_open(inode, file, seq_ops);
1787	if (ret < 0)
1788		trace_array_put(tr);
1789	return ret;
1790}
1791
1792static int
1793ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1794{
1795	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1796	struct trace_array *tr = inode->i_private;
1797	int ret;
1798
1799	if (trace_array_get(tr) < 0)
1800		return -ENODEV;
1801
1802	if ((file->f_mode & FMODE_WRITE) &&
1803	    (file->f_flags & O_TRUNC))
1804		ftrace_clear_event_pids(tr);
1805
1806	ret = ftrace_event_open(inode, file, seq_ops);
1807	if (ret < 0)
1808		trace_array_put(tr);
1809	return ret;
1810}
1811
1812static struct event_subsystem *
1813create_new_subsystem(const char *name)
1814{
1815	struct event_subsystem *system;
1816
1817	/* need to create new entry */
1818	system = kmalloc(sizeof(*system), GFP_KERNEL);
1819	if (!system)
1820		return NULL;
1821
1822	system->ref_count = 1;
1823
1824	/* Only allocate if dynamic (kprobes and modules) */
1825	system->name = kstrdup_const(name, GFP_KERNEL);
1826	if (!system->name)
1827		goto out_free;
1828
1829	system->filter = NULL;
1830
1831	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1832	if (!system->filter)
1833		goto out_free;
1834
1835	list_add(&system->list, &event_subsystems);
1836
1837	return system;
1838
1839 out_free:
1840	kfree_const(system->name);
1841	kfree(system);
1842	return NULL;
1843}
1844
1845static struct dentry *
1846event_subsystem_dir(struct trace_array *tr, const char *name,
1847		    struct trace_event_file *file, struct dentry *parent)
1848{
1849	struct trace_subsystem_dir *dir;
1850	struct event_subsystem *system;
1851	struct dentry *entry;
1852
1853	/* First see if we did not already create this dir */
1854	list_for_each_entry(dir, &tr->systems, list) {
1855		system = dir->subsystem;
1856		if (strcmp(system->name, name) == 0) {
1857			dir->nr_events++;
1858			file->system = dir;
1859			return dir->entry;
1860		}
1861	}
1862
1863	/* Now see if the system itself exists. */
1864	list_for_each_entry(system, &event_subsystems, list) {
1865		if (strcmp(system->name, name) == 0)
1866			break;
1867	}
1868	/* Reset system variable when not found */
1869	if (&system->list == &event_subsystems)
1870		system = NULL;
1871
1872	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1873	if (!dir)
1874		goto out_fail;
1875
1876	if (!system) {
1877		system = create_new_subsystem(name);
1878		if (!system)
1879			goto out_free;
1880	} else
1881		__get_system(system);
1882
1883	dir->entry = tracefs_create_dir(name, parent);
1884	if (!dir->entry) {
1885		pr_warn("Failed to create system directory %s\n", name);
1886		__put_system(system);
1887		goto out_free;
1888	}
1889
1890	dir->tr = tr;
1891	dir->ref_count = 1;
1892	dir->nr_events = 1;
1893	dir->subsystem = system;
1894	file->system = dir;
1895
1896	entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1897				    &ftrace_subsystem_filter_fops);
1898	if (!entry) {
1899		kfree(system->filter);
1900		system->filter = NULL;
1901		pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1902	}
1903
1904	trace_create_file("enable", 0644, dir->entry, dir,
1905			  &ftrace_system_enable_fops);
1906
1907	list_add(&dir->list, &tr->systems);
1908
1909	return dir->entry;
1910
1911 out_free:
1912	kfree(dir);
1913 out_fail:
1914	/* Only print this message if failed on memory allocation */
1915	if (!dir || !system)
1916		pr_warn("No memory to create event subsystem %s\n", name);
1917	return NULL;
1918}
1919
1920static int
1921event_create_dir(struct dentry *parent, struct trace_event_file *file)
1922{
1923	struct trace_event_call *call = file->event_call;
1924	struct trace_array *tr = file->tr;
1925	struct list_head *head;
1926	struct dentry *d_events;
1927	const char *name;
1928	int ret;
1929
1930	/*
1931	 * If the trace point header did not define TRACE_SYSTEM
1932	 * then the system would be called "TRACE_SYSTEM".
1933	 */
1934	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1935		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1936		if (!d_events)
1937			return -ENOMEM;
1938	} else
1939		d_events = parent;
1940
1941	name = trace_event_name(call);
1942	file->dir = tracefs_create_dir(name, d_events);
1943	if (!file->dir) {
1944		pr_warn("Could not create tracefs '%s' directory\n", name);
1945		return -1;
1946	}
1947
1948	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1949		trace_create_file("enable", 0644, file->dir, file,
1950				  &ftrace_enable_fops);
1951
1952#ifdef CONFIG_PERF_EVENTS
1953	if (call->event.type && call->class->reg)
1954		trace_create_file("id", 0444, file->dir,
1955				  (void *)(long)call->event.type,
1956				  &ftrace_event_id_fops);
1957#endif
1958
1959	/*
1960	 * Other events may have the same class. Only update
1961	 * the fields if they are not already defined.
1962	 */
1963	head = trace_get_fields(call);
1964	if (list_empty(head)) {
1965		ret = call->class->define_fields(call);
1966		if (ret < 0) {
1967			pr_warn("Could not initialize trace point events/%s\n",
1968				name);
1969			return -1;
1970		}
1971	}
1972	trace_create_file("filter", 0644, file->dir, file,
1973			  &ftrace_event_filter_fops);
1974
1975	/*
1976	 * Only event directories that can be enabled should have
1977	 * triggers.
1978	 */
1979	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1980		trace_create_file("trigger", 0644, file->dir, file,
1981				  &event_trigger_fops);
1982
1983#ifdef CONFIG_HIST_TRIGGERS
1984	trace_create_file("hist", 0444, file->dir, file,
1985			  &event_hist_fops);
1986#endif
1987	trace_create_file("format", 0444, file->dir, call,
1988			  &ftrace_event_format_fops);
1989
1990	return 0;
1991}
1992
1993static void remove_event_from_tracers(struct trace_event_call *call)
1994{
1995	struct trace_event_file *file;
1996	struct trace_array *tr;
1997
1998	do_for_each_event_file_safe(tr, file) {
1999		if (file->event_call != call)
2000			continue;
2001
2002		remove_event_file_dir(file);
2003		/*
2004		 * The do_for_each_event_file_safe() is
2005		 * a double loop. After finding the call for this
2006		 * trace_array, we use break to jump to the next
2007		 * trace_array.
2008		 */
2009		break;
2010	} while_for_each_event_file();
2011}
2012
2013static void event_remove(struct trace_event_call *call)
2014{
2015	struct trace_array *tr;
2016	struct trace_event_file *file;
2017
2018	do_for_each_event_file(tr, file) {
2019		if (file->event_call != call)
2020			continue;
 
 
 
 
2021		ftrace_event_enable_disable(file, 0);
2022		/*
2023		 * The do_for_each_event_file() is
2024		 * a double loop. After finding the call for this
2025		 * trace_array, we use break to jump to the next
2026		 * trace_array.
2027		 */
2028		break;
2029	} while_for_each_event_file();
2030
2031	if (call->event.funcs)
2032		__unregister_trace_event(&call->event);
2033	remove_event_from_tracers(call);
2034	list_del(&call->list);
2035}
2036
2037static int event_init(struct trace_event_call *call)
2038{
2039	int ret = 0;
2040	const char *name;
2041
2042	name = trace_event_name(call);
2043	if (WARN_ON(!name))
2044		return -EINVAL;
2045
2046	if (call->class->raw_init) {
2047		ret = call->class->raw_init(call);
2048		if (ret < 0 && ret != -ENOSYS)
2049			pr_warn("Could not initialize trace events/%s\n", name);
2050	}
2051
2052	return ret;
2053}
2054
2055static int
2056__register_event(struct trace_event_call *call, struct module *mod)
2057{
2058	int ret;
2059
2060	ret = event_init(call);
2061	if (ret < 0)
2062		return ret;
2063
2064	list_add(&call->list, &ftrace_events);
2065	call->mod = mod;
2066
2067	return 0;
2068}
2069
2070static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2071{
2072	int rlen;
2073	int elen;
2074
2075	/* Find the length of the enum value as a string */
2076	elen = snprintf(ptr, 0, "%ld", map->enum_value);
2077	/* Make sure there's enough room to replace the string with the value */
2078	if (len < elen)
2079		return NULL;
2080
2081	snprintf(ptr, elen + 1, "%ld", map->enum_value);
2082
2083	/* Get the rest of the string of ptr */
2084	rlen = strlen(ptr + len);
2085	memmove(ptr + elen, ptr + len, rlen);
2086	/* Make sure we end the new string */
2087	ptr[elen + rlen] = 0;
2088
2089	return ptr + elen;
2090}
2091
2092static void update_event_printk(struct trace_event_call *call,
2093				struct trace_enum_map *map)
2094{
2095	char *ptr;
2096	int quote = 0;
2097	int len = strlen(map->enum_string);
2098
2099	for (ptr = call->print_fmt; *ptr; ptr++) {
2100		if (*ptr == '\\') {
2101			ptr++;
2102			/* paranoid */
2103			if (!*ptr)
2104				break;
2105			continue;
2106		}
2107		if (*ptr == '"') {
2108			quote ^= 1;
2109			continue;
2110		}
2111		if (quote)
2112			continue;
2113		if (isdigit(*ptr)) {
2114			/* skip numbers */
2115			do {
2116				ptr++;
2117				/* Check for alpha chars like ULL */
2118			} while (isalnum(*ptr));
2119			if (!*ptr)
2120				break;
2121			/*
2122			 * A number must have some kind of delimiter after
2123			 * it, and we can ignore that too.
2124			 */
2125			continue;
2126		}
2127		if (isalpha(*ptr) || *ptr == '_') {
2128			if (strncmp(map->enum_string, ptr, len) == 0 &&
2129			    !isalnum(ptr[len]) && ptr[len] != '_') {
2130				ptr = enum_replace(ptr, map, len);
2131				/* Hmm, enum string smaller than value */
2132				if (WARN_ON_ONCE(!ptr))
2133					return;
2134				/*
2135				 * No need to decrement here, as enum_replace()
2136				 * returns the pointer to the character passed
2137				 * the enum, and two enums can not be placed
2138				 * back to back without something in between.
2139				 * We can skip that something in between.
2140				 */
2141				continue;
2142			}
2143		skip_more:
2144			do {
2145				ptr++;
2146			} while (isalnum(*ptr) || *ptr == '_');
2147			if (!*ptr)
2148				break;
2149			/*
2150			 * If what comes after this variable is a '.' or
2151			 * '->' then we can continue to ignore that string.
2152			 */
2153			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2154				ptr += *ptr == '.' ? 1 : 2;
2155				if (!*ptr)
2156					break;
2157				goto skip_more;
2158			}
2159			/*
2160			 * Once again, we can skip the delimiter that came
2161			 * after the string.
2162			 */
2163			continue;
2164		}
2165	}
2166}
2167
2168void trace_event_enum_update(struct trace_enum_map **map, int len)
2169{
2170	struct trace_event_call *call, *p;
2171	const char *last_system = NULL;
 
2172	int last_i;
2173	int i;
2174
2175	down_write(&trace_event_sem);
2176	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2177		/* events are usually grouped together with systems */
2178		if (!last_system || call->class->system != last_system) {
 
2179			last_i = 0;
2180			last_system = call->class->system;
2181		}
2182
 
 
 
 
 
 
 
 
 
 
2183		for (i = last_i; i < len; i++) {
2184			if (call->class->system == map[i]->system) {
2185				/* Save the first system if need be */
2186				if (!last_i)
2187					last_i = i;
 
 
2188				update_event_printk(call, map[i]);
2189			}
2190		}
2191	}
2192	up_write(&trace_event_sem);
2193}
2194
2195static struct trace_event_file *
2196trace_create_new_event(struct trace_event_call *call,
2197		       struct trace_array *tr)
2198{
2199	struct trace_event_file *file;
2200
2201	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2202	if (!file)
2203		return NULL;
2204
2205	file->event_call = call;
2206	file->tr = tr;
2207	atomic_set(&file->sm_ref, 0);
2208	atomic_set(&file->tm_ref, 0);
2209	INIT_LIST_HEAD(&file->triggers);
2210	list_add(&file->list, &tr->events);
2211
2212	return file;
2213}
2214
2215/* Add an event to a trace directory */
2216static int
2217__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2218{
2219	struct trace_event_file *file;
2220
2221	file = trace_create_new_event(call, tr);
2222	if (!file)
2223		return -ENOMEM;
2224
2225	return event_create_dir(tr->event_dir, file);
2226}
2227
2228/*
2229 * Just create a decriptor for early init. A descriptor is required
2230 * for enabling events at boot. We want to enable events before
2231 * the filesystem is initialized.
2232 */
2233static __init int
2234__trace_early_add_new_event(struct trace_event_call *call,
2235			    struct trace_array *tr)
2236{
2237	struct trace_event_file *file;
2238
2239	file = trace_create_new_event(call, tr);
2240	if (!file)
2241		return -ENOMEM;
2242
2243	return 0;
2244}
2245
2246struct ftrace_module_file_ops;
2247static void __add_event_to_tracers(struct trace_event_call *call);
2248
2249/* Add an additional event_call dynamically */
2250int trace_add_event_call(struct trace_event_call *call)
2251{
2252	int ret;
 
2253	mutex_lock(&trace_types_lock);
2254	mutex_lock(&event_mutex);
2255
2256	ret = __register_event(call, NULL);
2257	if (ret >= 0)
2258		__add_event_to_tracers(call);
2259
 
2260	mutex_unlock(&event_mutex);
2261	mutex_unlock(&trace_types_lock);
2262	return ret;
2263}
2264
2265/*
2266 * Must be called under locking of trace_types_lock, event_mutex and
2267 * trace_event_sem.
2268 */
2269static void __trace_remove_event_call(struct trace_event_call *call)
2270{
2271	event_remove(call);
2272	trace_destroy_fields(call);
2273	free_event_filter(call->filter);
2274	call->filter = NULL;
2275}
2276
2277static int probe_remove_event_call(struct trace_event_call *call)
2278{
2279	struct trace_array *tr;
2280	struct trace_event_file *file;
2281
2282#ifdef CONFIG_PERF_EVENTS
2283	if (call->perf_refcount)
2284		return -EBUSY;
2285#endif
2286	do_for_each_event_file(tr, file) {
2287		if (file->event_call != call)
2288			continue;
2289		/*
2290		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2291		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2292		 * TRACE_REG_UNREGISTER.
2293		 */
2294		if (file->flags & EVENT_FILE_FL_ENABLED)
2295			return -EBUSY;
2296		/*
2297		 * The do_for_each_event_file_safe() is
2298		 * a double loop. After finding the call for this
2299		 * trace_array, we use break to jump to the next
2300		 * trace_array.
2301		 */
2302		break;
2303	} while_for_each_event_file();
2304
2305	__trace_remove_event_call(call);
2306
2307	return 0;
2308}
2309
2310/* Remove an event_call */
2311int trace_remove_event_call(struct trace_event_call *call)
2312{
2313	int ret;
2314
 
2315	mutex_lock(&trace_types_lock);
2316	mutex_lock(&event_mutex);
2317	down_write(&trace_event_sem);
2318	ret = probe_remove_event_call(call);
2319	up_write(&trace_event_sem);
 
2320	mutex_unlock(&event_mutex);
2321	mutex_unlock(&trace_types_lock);
2322
2323	return ret;
2324}
2325
2326#define for_each_event(event, start, end)			\
2327	for (event = start;					\
2328	     (unsigned long)event < (unsigned long)end;		\
2329	     event++)
2330
2331#ifdef CONFIG_MODULES
2332
2333static void trace_module_add_events(struct module *mod)
2334{
2335	struct trace_event_call **call, **start, **end;
2336
2337	if (!mod->num_trace_events)
2338		return;
2339
2340	/* Don't add infrastructure for mods without tracepoints */
2341	if (trace_module_has_bad_taint(mod)) {
2342		pr_err("%s: module has bad taint, not creating trace events\n",
2343		       mod->name);
2344		return;
2345	}
2346
2347	start = mod->trace_events;
2348	end = mod->trace_events + mod->num_trace_events;
2349
2350	for_each_event(call, start, end) {
2351		__register_event(*call, mod);
2352		__add_event_to_tracers(*call);
2353	}
2354}
2355
2356static void trace_module_remove_events(struct module *mod)
2357{
2358	struct trace_event_call *call, *p;
2359	bool clear_trace = false;
2360
2361	down_write(&trace_event_sem);
2362	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2363		if (call->mod == mod) {
2364			if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2365				clear_trace = true;
2366			__trace_remove_event_call(call);
2367		}
2368	}
2369	up_write(&trace_event_sem);
2370
2371	/*
2372	 * It is safest to reset the ring buffer if the module being unloaded
2373	 * registered any events that were used. The only worry is if
2374	 * a new module gets loaded, and takes on the same id as the events
2375	 * of this module. When printing out the buffer, traced events left
2376	 * over from this module may be passed to the new module events and
2377	 * unexpected results may occur.
2378	 */
2379	if (clear_trace)
2380		tracing_reset_all_online_cpus();
2381}
2382
2383static int trace_module_notify(struct notifier_block *self,
2384			       unsigned long val, void *data)
2385{
2386	struct module *mod = data;
2387
 
2388	mutex_lock(&trace_types_lock);
2389	mutex_lock(&event_mutex);
2390	switch (val) {
2391	case MODULE_STATE_COMING:
2392		trace_module_add_events(mod);
2393		break;
2394	case MODULE_STATE_GOING:
2395		trace_module_remove_events(mod);
2396		break;
2397	}
 
2398	mutex_unlock(&event_mutex);
2399	mutex_unlock(&trace_types_lock);
2400
2401	return 0;
2402}
2403
2404static struct notifier_block trace_module_nb = {
2405	.notifier_call = trace_module_notify,
2406	.priority = 1, /* higher than trace.c module notify */
2407};
2408#endif /* CONFIG_MODULES */
2409
2410/* Create a new event directory structure for a trace directory. */
2411static void
2412__trace_add_event_dirs(struct trace_array *tr)
2413{
2414	struct trace_event_call *call;
2415	int ret;
2416
2417	list_for_each_entry(call, &ftrace_events, list) {
2418		ret = __trace_add_new_event(call, tr);
2419		if (ret < 0)
2420			pr_warn("Could not create directory for event %s\n",
2421				trace_event_name(call));
2422	}
2423}
2424
2425struct trace_event_file *
2426find_event_file(struct trace_array *tr, const char *system,  const char *event)
2427{
2428	struct trace_event_file *file;
2429	struct trace_event_call *call;
2430	const char *name;
2431
2432	list_for_each_entry(file, &tr->events, list) {
2433
2434		call = file->event_call;
2435		name = trace_event_name(call);
2436
2437		if (!name || !call->class || !call->class->reg)
2438			continue;
2439
2440		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2441			continue;
2442
2443		if (strcmp(event, name) == 0 &&
2444		    strcmp(system, call->class->system) == 0)
2445			return file;
2446	}
2447	return NULL;
2448}
2449
2450#ifdef CONFIG_DYNAMIC_FTRACE
2451
2452/* Avoid typos */
2453#define ENABLE_EVENT_STR	"enable_event"
2454#define DISABLE_EVENT_STR	"disable_event"
2455
2456struct event_probe_data {
2457	struct trace_event_file	*file;
2458	unsigned long			count;
2459	int				ref;
2460	bool				enable;
2461};
2462
2463static void
2464event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2465{
2466	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2467	struct event_probe_data *data = *pdata;
2468
2469	if (!data)
2470		return;
2471
2472	if (data->enable)
2473		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2474	else
2475		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2476}
2477
2478static void
2479event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2480{
2481	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2482	struct event_probe_data *data = *pdata;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2483
2484	if (!data)
 
2485		return;
2486
2487	if (!data->count)
 
 
2488		return;
2489
2490	/* Skip if the event is in a state we want to switch to */
2491	if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2492		return;
2493
2494	if (data->count != -1)
2495		(data->count)--;
2496
2497	event_enable_probe(ip, parent_ip, _data);
2498}
2499
2500static int
2501event_enable_print(struct seq_file *m, unsigned long ip,
2502		      struct ftrace_probe_ops *ops, void *_data)
2503{
2504	struct event_probe_data *data = _data;
 
 
 
 
 
 
 
 
 
2505
2506	seq_printf(m, "%ps:", (void *)ip);
2507
2508	seq_printf(m, "%s:%s:%s",
2509		   data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2510		   data->file->event_call->class->system,
2511		   trace_event_name(data->file->event_call));
2512
2513	if (data->count == -1)
2514		seq_puts(m, ":unlimited\n");
2515	else
2516		seq_printf(m, ":count=%ld\n", data->count);
2517
2518	return 0;
2519}
2520
2521static int
2522event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2523		  void **_data)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2524{
2525	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2526	struct event_probe_data *data = *pdata;
2527
2528	data->ref++;
 
 
 
 
 
 
2529	return 0;
2530}
2531
2532static void
2533event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2534		  void **_data)
2535{
2536	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2537	struct event_probe_data *data = *pdata;
 
 
 
 
 
 
 
 
 
2538
2539	if (WARN_ON_ONCE(data->ref <= 0))
2540		return;
2541
2542	data->ref--;
2543	if (!data->ref) {
2544		/* Remove the SOFT_MODE flag */
2545		__ftrace_event_enable_disable(data->file, 0, 1);
2546		module_put(data->file->event_call->mod);
2547		kfree(data);
2548	}
2549	*pdata = NULL;
2550}
2551
2552static struct ftrace_probe_ops event_enable_probe_ops = {
2553	.func			= event_enable_probe,
2554	.print			= event_enable_print,
2555	.init			= event_enable_init,
2556	.free			= event_enable_free,
2557};
2558
2559static struct ftrace_probe_ops event_enable_count_probe_ops = {
2560	.func			= event_enable_count_probe,
2561	.print			= event_enable_print,
2562	.init			= event_enable_init,
2563	.free			= event_enable_free,
2564};
2565
2566static struct ftrace_probe_ops event_disable_probe_ops = {
2567	.func			= event_enable_probe,
2568	.print			= event_enable_print,
2569	.init			= event_enable_init,
2570	.free			= event_enable_free,
2571};
2572
2573static struct ftrace_probe_ops event_disable_count_probe_ops = {
2574	.func			= event_enable_count_probe,
2575	.print			= event_enable_print,
2576	.init			= event_enable_init,
2577	.free			= event_enable_free,
2578};
2579
2580static int
2581event_enable_func(struct ftrace_hash *hash,
2582		  char *glob, char *cmd, char *param, int enabled)
2583{
2584	struct trace_array *tr = top_trace_array();
2585	struct trace_event_file *file;
2586	struct ftrace_probe_ops *ops;
2587	struct event_probe_data *data;
2588	const char *system;
2589	const char *event;
2590	char *number;
2591	bool enable;
2592	int ret;
2593
2594	if (!tr)
2595		return -ENODEV;
2596
2597	/* hash funcs only work with set_ftrace_filter */
2598	if (!enabled || !param)
2599		return -EINVAL;
2600
2601	system = strsep(&param, ":");
2602	if (!param)
2603		return -EINVAL;
2604
2605	event = strsep(&param, ":");
2606
2607	mutex_lock(&event_mutex);
2608
2609	ret = -EINVAL;
2610	file = find_event_file(tr, system, event);
2611	if (!file)
2612		goto out;
2613
2614	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2615
2616	if (enable)
2617		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2618	else
2619		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2620
2621	if (glob[0] == '!') {
2622		unregister_ftrace_function_probe_func(glob+1, ops);
2623		ret = 0;
2624		goto out;
2625	}
2626
2627	ret = -ENOMEM;
 
2628	data = kzalloc(sizeof(*data), GFP_KERNEL);
2629	if (!data)
2630		goto out;
2631
2632	data->enable = enable;
2633	data->count = -1;
2634	data->file = file;
2635
2636	if (!param)
2637		goto out_reg;
2638
2639	number = strsep(&param, ":");
2640
2641	ret = -EINVAL;
2642	if (!strlen(number))
2643		goto out_free;
2644
2645	/*
2646	 * We use the callback data field (which is a pointer)
2647	 * as our counter.
2648	 */
2649	ret = kstrtoul(number, 0, &data->count);
2650	if (ret)
2651		goto out_free;
2652
2653 out_reg:
2654	/* Don't let event modules unload while probe registered */
2655	ret = try_module_get(file->event_call->mod);
2656	if (!ret) {
2657		ret = -EBUSY;
2658		goto out_free;
2659	}
2660
2661	ret = __ftrace_event_enable_disable(file, 1, 1);
2662	if (ret < 0)
2663		goto out_put;
2664	ret = register_ftrace_function_probe(glob, ops, data);
 
2665	/*
2666	 * The above returns on success the # of functions enabled,
2667	 * but if it didn't find any functions it returns zero.
2668	 * Consider no functions a failure too.
2669	 */
2670	if (!ret) {
2671		ret = -ENOENT;
2672		goto out_disable;
2673	} else if (ret < 0)
2674		goto out_disable;
2675	/* Just return zero, not the number of enabled functions */
2676	ret = 0;
2677 out:
2678	mutex_unlock(&event_mutex);
2679	return ret;
2680
2681 out_disable:
2682	__ftrace_event_enable_disable(file, 0, 1);
2683 out_put:
2684	module_put(file->event_call->mod);
2685 out_free:
2686	kfree(data);
2687	goto out;
2688}
2689
2690static struct ftrace_func_command event_enable_cmd = {
2691	.name			= ENABLE_EVENT_STR,
2692	.func			= event_enable_func,
2693};
2694
2695static struct ftrace_func_command event_disable_cmd = {
2696	.name			= DISABLE_EVENT_STR,
2697	.func			= event_enable_func,
2698};
2699
2700static __init int register_event_cmds(void)
2701{
2702	int ret;
2703
2704	ret = register_ftrace_command(&event_enable_cmd);
2705	if (WARN_ON(ret < 0))
2706		return ret;
2707	ret = register_ftrace_command(&event_disable_cmd);
2708	if (WARN_ON(ret < 0))
2709		unregister_ftrace_command(&event_enable_cmd);
2710	return ret;
2711}
2712#else
2713static inline int register_event_cmds(void) { return 0; }
2714#endif /* CONFIG_DYNAMIC_FTRACE */
2715
2716/*
2717 * The top level array has already had its trace_event_file
2718 * descriptors created in order to allow for early events to
2719 * be recorded. This function is called after the tracefs has been
2720 * initialized, and we now have to create the files associated
2721 * to the events.
2722 */
2723static __init void
2724__trace_early_add_event_dirs(struct trace_array *tr)
2725{
2726	struct trace_event_file *file;
2727	int ret;
2728
2729
2730	list_for_each_entry(file, &tr->events, list) {
2731		ret = event_create_dir(tr->event_dir, file);
2732		if (ret < 0)
2733			pr_warn("Could not create directory for event %s\n",
2734				trace_event_name(file->event_call));
2735	}
2736}
2737
2738/*
2739 * For early boot up, the top trace array requires to have
2740 * a list of events that can be enabled. This must be done before
2741 * the filesystem is set up in order to allow events to be traced
2742 * early.
2743 */
2744static __init void
2745__trace_early_add_events(struct trace_array *tr)
2746{
2747	struct trace_event_call *call;
2748	int ret;
2749
2750	list_for_each_entry(call, &ftrace_events, list) {
2751		/* Early boot up should not have any modules loaded */
2752		if (WARN_ON_ONCE(call->mod))
2753			continue;
2754
2755		ret = __trace_early_add_new_event(call, tr);
2756		if (ret < 0)
2757			pr_warn("Could not create early event %s\n",
2758				trace_event_name(call));
2759	}
2760}
2761
2762/* Remove the event directory structure for a trace directory. */
2763static void
2764__trace_remove_event_dirs(struct trace_array *tr)
2765{
2766	struct trace_event_file *file, *next;
2767
2768	list_for_each_entry_safe(file, next, &tr->events, list)
2769		remove_event_file_dir(file);
2770}
2771
2772static void __add_event_to_tracers(struct trace_event_call *call)
2773{
2774	struct trace_array *tr;
2775
2776	list_for_each_entry(tr, &ftrace_trace_arrays, list)
2777		__trace_add_new_event(call, tr);
2778}
2779
2780extern struct trace_event_call *__start_ftrace_events[];
2781extern struct trace_event_call *__stop_ftrace_events[];
2782
2783static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2784
2785static __init int setup_trace_event(char *str)
2786{
2787	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2788	ring_buffer_expanded = true;
2789	tracing_selftest_disabled = true;
2790
2791	return 1;
2792}
2793__setup("trace_event=", setup_trace_event);
2794
2795/* Expects to have event_mutex held when called */
2796static int
2797create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2798{
2799	struct dentry *d_events;
2800	struct dentry *entry;
2801
2802	entry = tracefs_create_file("set_event", 0644, parent,
2803				    tr, &ftrace_set_event_fops);
2804	if (!entry) {
2805		pr_warn("Could not create tracefs 'set_event' entry\n");
2806		return -ENOMEM;
2807	}
2808
2809	d_events = tracefs_create_dir("events", parent);
2810	if (!d_events) {
2811		pr_warn("Could not create tracefs 'events' directory\n");
2812		return -ENOMEM;
2813	}
2814
2815	entry = trace_create_file("enable", 0644, d_events,
2816				  tr, &ftrace_tr_enable_fops);
2817	if (!entry) {
2818		pr_warn("Could not create tracefs 'enable' entry\n");
2819		return -ENOMEM;
2820	}
2821
2822	/* There are not as crucial, just warn if they are not created */
2823
2824	entry = tracefs_create_file("set_event_pid", 0644, parent,
2825				    tr, &ftrace_set_event_pid_fops);
2826	if (!entry)
2827		pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2828
2829	/* ring buffer internal formats */
2830	entry = trace_create_file("header_page", 0444, d_events,
2831				  ring_buffer_print_page_header,
2832				  &ftrace_show_header_fops);
2833	if (!entry)
2834		pr_warn("Could not create tracefs 'header_page' entry\n");
2835
2836	entry = trace_create_file("header_event", 0444, d_events,
2837				  ring_buffer_print_entry_header,
2838				  &ftrace_show_header_fops);
2839	if (!entry)
2840		pr_warn("Could not create tracefs 'header_event' entry\n");
2841
2842	tr->event_dir = d_events;
2843
2844	return 0;
2845}
2846
2847/**
2848 * event_trace_add_tracer - add a instance of a trace_array to events
2849 * @parent: The parent dentry to place the files/directories for events in
2850 * @tr: The trace array associated with these events
2851 *
2852 * When a new instance is created, it needs to set up its events
2853 * directory, as well as other files associated with events. It also
2854 * creates the event hierachry in the @parent/events directory.
2855 *
2856 * Returns 0 on success.
 
 
2857 */
2858int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2859{
2860	int ret;
2861
2862	mutex_lock(&event_mutex);
2863
2864	ret = create_event_toplevel_files(parent, tr);
2865	if (ret)
2866		goto out_unlock;
2867
2868	down_write(&trace_event_sem);
2869	__trace_add_event_dirs(tr);
2870	up_write(&trace_event_sem);
2871
2872 out_unlock:
2873	mutex_unlock(&event_mutex);
2874
2875	return ret;
2876}
2877
2878/*
2879 * The top trace array already had its file descriptors created.
2880 * Now the files themselves need to be created.
2881 */
2882static __init int
2883early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2884{
2885	int ret;
2886
2887	mutex_lock(&event_mutex);
2888
2889	ret = create_event_toplevel_files(parent, tr);
2890	if (ret)
2891		goto out_unlock;
2892
2893	down_write(&trace_event_sem);
2894	__trace_early_add_event_dirs(tr);
2895	up_write(&trace_event_sem);
2896
2897 out_unlock:
2898	mutex_unlock(&event_mutex);
2899
2900	return ret;
2901}
2902
 
2903int event_trace_del_tracer(struct trace_array *tr)
2904{
2905	mutex_lock(&event_mutex);
2906
2907	/* Disable any event triggers and associated soft-disabled events */
2908	clear_event_triggers(tr);
2909
2910	/* Clear the pid list */
2911	__ftrace_clear_event_pids(tr);
2912
2913	/* Disable any running events */
2914	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2915
2916	/* Access to events are within rcu_read_lock_sched() */
2917	synchronize_sched();
2918
2919	down_write(&trace_event_sem);
2920	__trace_remove_event_dirs(tr);
2921	tracefs_remove_recursive(tr->event_dir);
2922	up_write(&trace_event_sem);
2923
2924	tr->event_dir = NULL;
2925
2926	mutex_unlock(&event_mutex);
2927
2928	return 0;
2929}
2930
2931static __init int event_trace_memsetup(void)
2932{
2933	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2934	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
2935	return 0;
2936}
2937
2938static __init void
2939early_enable_events(struct trace_array *tr, bool disable_first)
2940{
2941	char *buf = bootup_event_buf;
2942	char *token;
2943	int ret;
2944
2945	while (true) {
2946		token = strsep(&buf, ",");
2947
2948		if (!token)
2949			break;
2950
2951		if (*token) {
2952			/* Restarting syscalls requires that we stop them first */
2953			if (disable_first)
2954				ftrace_set_clr_event(tr, token, 0);
2955
2956			ret = ftrace_set_clr_event(tr, token, 1);
2957			if (ret)
2958				pr_warn("Failed to enable trace event: %s\n", token);
2959		}
2960
2961		/* Put back the comma to allow this to be called again */
2962		if (buf)
2963			*(buf - 1) = ',';
2964	}
2965}
2966
2967static __init int event_trace_enable(void)
2968{
2969	struct trace_array *tr = top_trace_array();
2970	struct trace_event_call **iter, *call;
2971	int ret;
2972
2973	if (!tr)
2974		return -ENODEV;
2975
2976	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2977
2978		call = *iter;
2979		ret = event_init(call);
2980		if (!ret)
2981			list_add(&call->list, &ftrace_events);
2982	}
2983
2984	/*
2985	 * We need the top trace array to have a working set of trace
2986	 * points at early init, before the debug files and directories
2987	 * are created. Create the file entries now, and attach them
2988	 * to the actual file dentries later.
2989	 */
2990	__trace_early_add_events(tr);
2991
2992	early_enable_events(tr, false);
2993
2994	trace_printk_start_comm();
2995
2996	register_event_cmds();
2997
2998	register_trigger_cmds();
2999
3000	return 0;
3001}
3002
3003/*
3004 * event_trace_enable() is called from trace_event_init() first to
3005 * initialize events and perhaps start any events that are on the
3006 * command line. Unfortunately, there are some events that will not
3007 * start this early, like the system call tracepoints that need
3008 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3009 * is called before pid 1 starts, and this flag is never set, making
3010 * the syscall tracepoint never get reached, but the event is enabled
3011 * regardless (and not doing anything).
3012 */
3013static __init int event_trace_enable_again(void)
3014{
3015	struct trace_array *tr;
3016
3017	tr = top_trace_array();
3018	if (!tr)
3019		return -ENODEV;
3020
3021	early_enable_events(tr, true);
3022
3023	return 0;
3024}
3025
3026early_initcall(event_trace_enable_again);
3027
3028static __init int event_trace_init(void)
3029{
3030	struct trace_array *tr;
3031	struct dentry *d_tracer;
3032	struct dentry *entry;
3033	int ret;
3034
3035	tr = top_trace_array();
3036	if (!tr)
3037		return -ENODEV;
3038
3039	d_tracer = tracing_init_dentry();
3040	if (IS_ERR(d_tracer))
3041		return 0;
3042
3043	entry = tracefs_create_file("available_events", 0444, d_tracer,
3044				    tr, &ftrace_avail_fops);
3045	if (!entry)
3046		pr_warn("Could not create tracefs 'available_events' entry\n");
3047
3048	if (trace_define_generic_fields())
3049		pr_warn("tracing: Failed to allocated generic fields");
3050
3051	if (trace_define_common_fields())
3052		pr_warn("tracing: Failed to allocate common fields");
3053
3054	ret = early_event_add_tracer(d_tracer, tr);
3055	if (ret)
3056		return ret;
3057
3058#ifdef CONFIG_MODULES
3059	ret = register_module_notifier(&trace_module_nb);
3060	if (ret)
3061		pr_warn("Failed to register trace events module notifier\n");
3062#endif
3063	return 0;
3064}
3065
3066void __init trace_event_init(void)
3067{
3068	event_trace_memsetup();
3069	init_ftrace_syscalls();
3070	event_trace_enable();
3071}
3072
3073fs_initcall(event_trace_init);
3074
3075#ifdef CONFIG_FTRACE_STARTUP_TEST
3076
3077static DEFINE_SPINLOCK(test_spinlock);
3078static DEFINE_SPINLOCK(test_spinlock_irq);
3079static DEFINE_MUTEX(test_mutex);
3080
3081static __init void test_work(struct work_struct *dummy)
3082{
3083	spin_lock(&test_spinlock);
3084	spin_lock_irq(&test_spinlock_irq);
3085	udelay(1);
3086	spin_unlock_irq(&test_spinlock_irq);
3087	spin_unlock(&test_spinlock);
3088
3089	mutex_lock(&test_mutex);
3090	msleep(1);
3091	mutex_unlock(&test_mutex);
3092}
3093
3094static __init int event_test_thread(void *unused)
3095{
3096	void *test_malloc;
3097
3098	test_malloc = kmalloc(1234, GFP_KERNEL);
3099	if (!test_malloc)
3100		pr_info("failed to kmalloc\n");
3101
3102	schedule_on_each_cpu(test_work);
3103
3104	kfree(test_malloc);
3105
3106	set_current_state(TASK_INTERRUPTIBLE);
3107	while (!kthread_should_stop()) {
3108		schedule();
3109		set_current_state(TASK_INTERRUPTIBLE);
3110	}
3111	__set_current_state(TASK_RUNNING);
3112
3113	return 0;
3114}
3115
3116/*
3117 * Do various things that may trigger events.
3118 */
3119static __init void event_test_stuff(void)
3120{
3121	struct task_struct *test_thread;
3122
3123	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3124	msleep(1);
3125	kthread_stop(test_thread);
3126}
3127
3128/*
3129 * For every trace event defined, we will test each trace point separately,
3130 * and then by groups, and finally all trace points.
3131 */
3132static __init void event_trace_self_tests(void)
3133{
3134	struct trace_subsystem_dir *dir;
3135	struct trace_event_file *file;
3136	struct trace_event_call *call;
3137	struct event_subsystem *system;
3138	struct trace_array *tr;
3139	int ret;
3140
3141	tr = top_trace_array();
3142	if (!tr)
3143		return;
3144
3145	pr_info("Running tests on trace events:\n");
3146
3147	list_for_each_entry(file, &tr->events, list) {
3148
3149		call = file->event_call;
3150
3151		/* Only test those that have a probe */
3152		if (!call->class || !call->class->probe)
3153			continue;
3154
3155/*
3156 * Testing syscall events here is pretty useless, but
3157 * we still do it if configured. But this is time consuming.
3158 * What we really need is a user thread to perform the
3159 * syscalls as we test.
3160 */
3161#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3162		if (call->class->system &&
3163		    strcmp(call->class->system, "syscalls") == 0)
3164			continue;
3165#endif
3166
3167		pr_info("Testing event %s: ", trace_event_name(call));
3168
3169		/*
3170		 * If an event is already enabled, someone is using
3171		 * it and the self test should not be on.
3172		 */
3173		if (file->flags & EVENT_FILE_FL_ENABLED) {
3174			pr_warn("Enabled event during self test!\n");
3175			WARN_ON_ONCE(1);
3176			continue;
3177		}
3178
3179		ftrace_event_enable_disable(file, 1);
3180		event_test_stuff();
3181		ftrace_event_enable_disable(file, 0);
3182
3183		pr_cont("OK\n");
3184	}
3185
3186	/* Now test at the sub system level */
3187
3188	pr_info("Running tests on trace event systems:\n");
3189
3190	list_for_each_entry(dir, &tr->systems, list) {
3191
3192		system = dir->subsystem;
3193
3194		/* the ftrace system is special, skip it */
3195		if (strcmp(system->name, "ftrace") == 0)
3196			continue;
3197
3198		pr_info("Testing event system %s: ", system->name);
3199
3200		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3201		if (WARN_ON_ONCE(ret)) {
3202			pr_warn("error enabling system %s\n",
3203				system->name);
3204			continue;
3205		}
3206
3207		event_test_stuff();
3208
3209		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3210		if (WARN_ON_ONCE(ret)) {
3211			pr_warn("error disabling system %s\n",
3212				system->name);
3213			continue;
3214		}
3215
3216		pr_cont("OK\n");
3217	}
3218
3219	/* Test with all events enabled */
3220
3221	pr_info("Running tests on all trace events:\n");
3222	pr_info("Testing all events: ");
3223
3224	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3225	if (WARN_ON_ONCE(ret)) {
3226		pr_warn("error enabling all events\n");
3227		return;
3228	}
3229
3230	event_test_stuff();
3231
3232	/* reset sysname */
3233	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3234	if (WARN_ON_ONCE(ret)) {
3235		pr_warn("error disabling all events\n");
3236		return;
3237	}
3238
3239	pr_cont("OK\n");
3240}
3241
3242#ifdef CONFIG_FUNCTION_TRACER
3243
3244static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3245
3246static struct trace_event_file event_trace_file __initdata;
3247
3248static void __init
3249function_test_events_call(unsigned long ip, unsigned long parent_ip,
3250			  struct ftrace_ops *op, struct pt_regs *pt_regs)
3251{
3252	struct ring_buffer_event *event;
3253	struct ring_buffer *buffer;
3254	struct ftrace_entry *entry;
3255	unsigned long flags;
3256	long disabled;
3257	int cpu;
3258	int pc;
3259
3260	pc = preempt_count();
3261	preempt_disable_notrace();
3262	cpu = raw_smp_processor_id();
3263	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3264
3265	if (disabled != 1)
3266		goto out;
3267
3268	local_save_flags(flags);
3269
3270	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3271						TRACE_FN, sizeof(*entry),
3272						flags, pc);
3273	if (!event)
3274		goto out;
3275	entry	= ring_buffer_event_data(event);
3276	entry->ip			= ip;
3277	entry->parent_ip		= parent_ip;
3278
3279	event_trigger_unlock_commit(&event_trace_file, buffer, event,
3280				    entry, flags, pc);
3281 out:
3282	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3283	preempt_enable_notrace();
3284}
3285
3286static struct ftrace_ops trace_ops __initdata  =
3287{
3288	.func = function_test_events_call,
3289	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
3290};
3291
3292static __init void event_trace_self_test_with_function(void)
3293{
3294	int ret;
3295
3296	event_trace_file.tr = top_trace_array();
3297	if (WARN_ON(!event_trace_file.tr))
3298		return;
3299
3300	ret = register_ftrace_function(&trace_ops);
3301	if (WARN_ON(ret < 0)) {
3302		pr_info("Failed to enable function tracer for event tests\n");
3303		return;
3304	}
3305	pr_info("Running tests again, along with the function tracer\n");
3306	event_trace_self_tests();
3307	unregister_ftrace_function(&trace_ops);
3308}
3309#else
3310static __init void event_trace_self_test_with_function(void)
3311{
3312}
3313#endif
3314
3315static __init int event_trace_self_tests_init(void)
3316{
3317	if (!tracing_selftest_disabled) {
3318		event_trace_self_tests();
3319		event_trace_self_test_with_function();
3320	}
3321
3322	return 0;
3323}
3324
3325late_initcall(event_trace_self_tests_init);
3326
3327#endif