Linux Audio

Check our new training course

Loading...
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
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * event tracer
   4 *
   5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6 *
   7 *  - Added format output of fields of the trace point.
   8 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9 *
  10 */
  11
  12#define pr_fmt(fmt) fmt
  13
  14#include <linux/workqueue.h>
  15#include <linux/security.h>
  16#include <linux/spinlock.h>
  17#include <linux/kthread.h>
  18#include <linux/tracefs.h>
  19#include <linux/uaccess.h>
  20#include <linux/module.h>
  21#include <linux/ctype.h>
  22#include <linux/sort.h>
  23#include <linux/slab.h>
  24#include <linux/delay.h>
  25
  26#include <trace/events/sched.h>
  27#include <trace/syscall.h>
  28
  29#include <asm/setup.h>
  30
  31#include "trace_output.h"
  32
  33#undef TRACE_SYSTEM
  34#define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36DEFINE_MUTEX(event_mutex);
  37
  38LIST_HEAD(ftrace_events);
  39static LIST_HEAD(ftrace_generic_fields);
  40static LIST_HEAD(ftrace_common_fields);
  41static bool eventdir_initialized;
  42
  43#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  44
  45static struct kmem_cache *field_cachep;
  46static struct kmem_cache *file_cachep;
  47
  48static inline int system_refcount(struct event_subsystem *system)
  49{
  50	return system->ref_count;
  51}
  52
  53static int system_refcount_inc(struct event_subsystem *system)
  54{
  55	return system->ref_count++;
  56}
  57
  58static int system_refcount_dec(struct event_subsystem *system)
  59{
  60	return --system->ref_count;
  61}
  62
  63/* Double loops, do not use break, only goto's work */
  64#define do_for_each_event_file(tr, file)			\
  65	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  66		list_for_each_entry(file, &tr->events, list)
  67
  68#define do_for_each_event_file_safe(tr, file)			\
  69	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  70		struct trace_event_file *___n;				\
  71		list_for_each_entry_safe(file, ___n, &tr->events, list)
  72
  73#define while_for_each_event_file()		\
  74	}
  75
 
 
 
 
 
 
 
 
  76static struct ftrace_event_field *
  77__find_event_field(struct list_head *head, char *name)
  78{
  79	struct ftrace_event_field *field;
  80
  81	list_for_each_entry(field, head, link) {
  82		if (!strcmp(field->name, name))
  83			return field;
  84	}
  85
  86	return NULL;
  87}
  88
  89struct ftrace_event_field *
  90trace_find_event_field(struct trace_event_call *call, char *name)
  91{
  92	struct ftrace_event_field *field;
  93	struct list_head *head;
  94
  95	head = trace_get_fields(call);
  96	field = __find_event_field(head, name);
  97	if (field)
  98		return field;
  99
 100	field = __find_event_field(&ftrace_generic_fields, name);
 101	if (field)
 102		return field;
 103
 104	return __find_event_field(&ftrace_common_fields, name);
 105}
 106
 107static int __trace_define_field(struct list_head *head, const char *type,
 108				const char *name, int offset, int size,
 109				int is_signed, int filter_type)
 110{
 111	struct ftrace_event_field *field;
 112
 113	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 114	if (!field)
 115		return -ENOMEM;
 116
 117	field->name = name;
 118	field->type = type;
 119
 120	if (filter_type == FILTER_OTHER)
 121		field->filter_type = filter_assign_type(type);
 122	else
 123		field->filter_type = filter_type;
 124
 125	field->offset = offset;
 126	field->size = size;
 127	field->is_signed = is_signed;
 128
 129	list_add(&field->link, head);
 130
 131	return 0;
 132}
 133
 134int trace_define_field(struct trace_event_call *call, const char *type,
 135		       const char *name, int offset, int size, int is_signed,
 136		       int filter_type)
 137{
 138	struct list_head *head;
 139
 140	if (WARN_ON(!call->class))
 141		return 0;
 142
 143	head = trace_get_fields(call);
 144	return __trace_define_field(head, type, name, offset, size,
 145				    is_signed, filter_type);
 146}
 147EXPORT_SYMBOL_GPL(trace_define_field);
 148
 149#define __generic_field(type, item, filter_type)			\
 150	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 151				   #item, 0, 0, is_signed_type(type),	\
 152				   filter_type);			\
 153	if (ret)							\
 154		return ret;
 155
 156#define __common_field(type, item)					\
 157	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 158				   "common_" #item,			\
 159				   offsetof(typeof(ent), item),		\
 160				   sizeof(ent.item),			\
 161				   is_signed_type(type), FILTER_OTHER);	\
 162	if (ret)							\
 163		return ret;
 164
 165static int trace_define_generic_fields(void)
 166{
 167	int ret;
 168
 169	__generic_field(int, CPU, FILTER_CPU);
 170	__generic_field(int, cpu, FILTER_CPU);
 171	__generic_field(char *, COMM, FILTER_COMM);
 172	__generic_field(char *, comm, FILTER_COMM);
 173
 174	return ret;
 175}
 176
 177static int trace_define_common_fields(void)
 178{
 179	int ret;
 180	struct trace_entry ent;
 181
 182	__common_field(unsigned short, type);
 183	__common_field(unsigned char, flags);
 184	__common_field(unsigned char, preempt_count);
 185	__common_field(int, pid);
 186
 187	return ret;
 188}
 189
 190static void trace_destroy_fields(struct trace_event_call *call)
 191{
 192	struct ftrace_event_field *field, *next;
 193	struct list_head *head;
 194
 195	head = trace_get_fields(call);
 196	list_for_each_entry_safe(field, next, head, link) {
 197		list_del(&field->link);
 198		kmem_cache_free(field_cachep, field);
 199	}
 200}
 201
 202/*
 203 * run-time version of trace_event_get_offsets_<call>() that returns the last
 204 * accessible offset of trace fields excluding __dynamic_array bytes
 205 */
 206int trace_event_get_offsets(struct trace_event_call *call)
 207{
 208	struct ftrace_event_field *tail;
 209	struct list_head *head;
 210
 211	head = trace_get_fields(call);
 212	/*
 213	 * head->next points to the last field with the largest offset,
 214	 * since it was added last by trace_define_field()
 215	 */
 216	tail = list_first_entry(head, struct ftrace_event_field, link);
 217	return tail->offset + tail->size;
 218}
 219
 220/*
 221 * Check if the referenced field is an array and return true,
 222 * as arrays are OK to dereference.
 223 */
 224static bool test_field(const char *fmt, struct trace_event_call *call)
 225{
 226	struct trace_event_fields *field = call->class->fields_array;
 227	const char *array_descriptor;
 228	const char *p = fmt;
 229	int len;
 230
 231	if (!(len = str_has_prefix(fmt, "REC->")))
 232		return false;
 233	fmt += len;
 234	for (p = fmt; *p; p++) {
 235		if (!isalnum(*p) && *p != '_')
 236			break;
 237	}
 238	len = p - fmt;
 239
 240	for (; field->type; field++) {
 241		if (strncmp(field->name, fmt, len) ||
 242		    field->name[len])
 243			continue;
 244		array_descriptor = strchr(field->type, '[');
 245		/* This is an array and is OK to dereference. */
 246		return array_descriptor != NULL;
 247	}
 248	return false;
 249}
 250
 251/*
 252 * Examine the print fmt of the event looking for unsafe dereference
 253 * pointers using %p* that could be recorded in the trace event and
 254 * much later referenced after the pointer was freed. Dereferencing
 255 * pointers are OK, if it is dereferenced into the event itself.
 256 */
 257static void test_event_printk(struct trace_event_call *call)
 258{
 259	u64 dereference_flags = 0;
 260	bool first = true;
 261	const char *fmt, *c, *r, *a;
 262	int parens = 0;
 263	char in_quote = 0;
 264	int start_arg = 0;
 265	int arg = 0;
 266	int i;
 267
 268	fmt = call->print_fmt;
 269
 270	if (!fmt)
 271		return;
 272
 273	for (i = 0; fmt[i]; i++) {
 274		switch (fmt[i]) {
 275		case '\\':
 276			i++;
 277			if (!fmt[i])
 278				return;
 279			continue;
 280		case '"':
 281		case '\'':
 282			/*
 283			 * The print fmt starts with a string that
 284			 * is processed first to find %p* usage,
 285			 * then after the first string, the print fmt
 286			 * contains arguments that are used to check
 287			 * if the dereferenced %p* usage is safe.
 288			 */
 289			if (first) {
 290				if (fmt[i] == '\'')
 291					continue;
 292				if (in_quote) {
 293					arg = 0;
 294					first = false;
 295					/*
 296					 * If there was no %p* uses
 297					 * the fmt is OK.
 298					 */
 299					if (!dereference_flags)
 300						return;
 301				}
 302			}
 303			if (in_quote) {
 304				if (in_quote == fmt[i])
 305					in_quote = 0;
 306			} else {
 307				in_quote = fmt[i];
 308			}
 309			continue;
 310		case '%':
 311			if (!first || !in_quote)
 312				continue;
 313			i++;
 314			if (!fmt[i])
 315				return;
 316			switch (fmt[i]) {
 317			case '%':
 318				continue;
 319			case 'p':
 320				/* Find dereferencing fields */
 321				switch (fmt[i + 1]) {
 322				case 'B': case 'R': case 'r':
 323				case 'b': case 'M': case 'm':
 324				case 'I': case 'i': case 'E':
 325				case 'U': case 'V': case 'N':
 326				case 'a': case 'd': case 'D':
 327				case 'g': case 't': case 'C':
 328				case 'O': case 'f':
 329					if (WARN_ONCE(arg == 63,
 330						      "Too many args for event: %s",
 331						      trace_event_name(call)))
 332						return;
 333					dereference_flags |= 1ULL << arg;
 334				}
 335				break;
 336			default:
 337			{
 338				bool star = false;
 339				int j;
 340
 341				/* Increment arg if %*s exists. */
 342				for (j = 0; fmt[i + j]; j++) {
 343					if (isdigit(fmt[i + j]) ||
 344					    fmt[i + j] == '.')
 345						continue;
 346					if (fmt[i + j] == '*') {
 347						star = true;
 348						continue;
 349					}
 350					if ((fmt[i + j] == 's') && star)
 351						arg++;
 352					break;
 353				}
 354				break;
 355			} /* default */
 356
 357			} /* switch */
 358			arg++;
 359			continue;
 360		case '(':
 361			if (in_quote)
 362				continue;
 363			parens++;
 364			continue;
 365		case ')':
 366			if (in_quote)
 367				continue;
 368			parens--;
 369			if (WARN_ONCE(parens < 0,
 370				      "Paren mismatch for event: %s\narg='%s'\n%*s",
 371				      trace_event_name(call),
 372				      fmt + start_arg,
 373				      (i - start_arg) + 5, "^"))
 374				return;
 375			continue;
 376		case ',':
 377			if (in_quote || parens)
 378				continue;
 379			i++;
 380			while (isspace(fmt[i]))
 381				i++;
 382			start_arg = i;
 383			if (!(dereference_flags & (1ULL << arg)))
 384				goto next_arg;
 385
 386			/* Find the REC-> in the argument */
 387			c = strchr(fmt + i, ',');
 388			r = strstr(fmt + i, "REC->");
 389			if (r && (!c || r < c)) {
 390				/*
 391				 * Addresses of events on the buffer,
 392				 * or an array on the buffer is
 393				 * OK to dereference.
 394				 * There's ways to fool this, but
 395				 * this is to catch common mistakes,
 396				 * not malicious code.
 397				 */
 398				a = strchr(fmt + i, '&');
 399				if ((a && (a < r)) || test_field(r, call))
 400					dereference_flags &= ~(1ULL << arg);
 401			}
 402		next_arg:
 403			i--;
 404			arg++;
 405		}
 406	}
 407
 408	/*
 409	 * If you triggered the below warning, the trace event reported
 410	 * uses an unsafe dereference pointer %p*. As the data stored
 411	 * at the trace event time may no longer exist when the trace
 412	 * event is printed, dereferencing to the original source is
 413	 * unsafe. The source of the dereference must be copied into the
 414	 * event itself, and the dereference must access the copy instead.
 415	 */
 416	if (WARN_ON_ONCE(dereference_flags)) {
 417		arg = 1;
 418		while (!(dereference_flags & 1)) {
 419			dereference_flags >>= 1;
 420			arg++;
 421		}
 422		pr_warn("event %s has unsafe dereference of argument %d\n",
 423			trace_event_name(call), arg);
 424		pr_warn("print_fmt: %s\n", fmt);
 425	}
 426}
 427
 428int trace_event_raw_init(struct trace_event_call *call)
 429{
 430	int id;
 431
 432	id = register_trace_event(&call->event);
 433	if (!id)
 434		return -ENODEV;
 435
 436	test_event_printk(call);
 437
 438	return 0;
 439}
 440EXPORT_SYMBOL_GPL(trace_event_raw_init);
 441
 442bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 443{
 444	struct trace_array *tr = trace_file->tr;
 445	struct trace_array_cpu *data;
 446	struct trace_pid_list *no_pid_list;
 447	struct trace_pid_list *pid_list;
 448
 449	pid_list = rcu_dereference_raw(tr->filtered_pids);
 450	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 451
 452	if (!pid_list && !no_pid_list)
 453		return false;
 454
 455	data = this_cpu_ptr(tr->array_buffer.data);
 456
 457	return data->ignore_pid;
 458}
 459EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 460
 461void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 462				 struct trace_event_file *trace_file,
 463				 unsigned long len)
 464{
 465	struct trace_event_call *event_call = trace_file->event_call;
 466
 467	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 468	    trace_event_ignore_this_pid(trace_file))
 469		return NULL;
 470
 
 
 471	/*
 472	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 473	 * preemption (adding one to the preempt_count). Since we are
 474	 * interested in the preempt_count at the time the tracepoint was
 475	 * hit, we need to subtract one to offset the increment.
 476	 */
 477	fbuffer->trace_ctx = tracing_gen_ctx_dec();
 
 478	fbuffer->trace_file = trace_file;
 479
 480	fbuffer->event =
 481		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 482						event_call->event.type, len,
 483						fbuffer->trace_ctx);
 484	if (!fbuffer->event)
 485		return NULL;
 486
 487	fbuffer->regs = NULL;
 488	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 489	return fbuffer->entry;
 490}
 491EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 492
 493int trace_event_reg(struct trace_event_call *call,
 494		    enum trace_reg type, void *data)
 495{
 496	struct trace_event_file *file = data;
 497
 498	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 499	switch (type) {
 500	case TRACE_REG_REGISTER:
 501		return tracepoint_probe_register(call->tp,
 502						 call->class->probe,
 503						 file);
 504	case TRACE_REG_UNREGISTER:
 505		tracepoint_probe_unregister(call->tp,
 506					    call->class->probe,
 507					    file);
 508		return 0;
 509
 510#ifdef CONFIG_PERF_EVENTS
 511	case TRACE_REG_PERF_REGISTER:
 512		return tracepoint_probe_register(call->tp,
 513						 call->class->perf_probe,
 514						 call);
 515	case TRACE_REG_PERF_UNREGISTER:
 516		tracepoint_probe_unregister(call->tp,
 517					    call->class->perf_probe,
 518					    call);
 519		return 0;
 520	case TRACE_REG_PERF_OPEN:
 521	case TRACE_REG_PERF_CLOSE:
 522	case TRACE_REG_PERF_ADD:
 523	case TRACE_REG_PERF_DEL:
 524		return 0;
 525#endif
 526	}
 527	return 0;
 528}
 529EXPORT_SYMBOL_GPL(trace_event_reg);
 530
 531void trace_event_enable_cmd_record(bool enable)
 532{
 533	struct trace_event_file *file;
 534	struct trace_array *tr;
 535
 536	lockdep_assert_held(&event_mutex);
 537
 538	do_for_each_event_file(tr, file) {
 539
 540		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 541			continue;
 542
 543		if (enable) {
 544			tracing_start_cmdline_record();
 545			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 546		} else {
 547			tracing_stop_cmdline_record();
 548			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 549		}
 550	} while_for_each_event_file();
 551}
 552
 553void trace_event_enable_tgid_record(bool enable)
 554{
 555	struct trace_event_file *file;
 556	struct trace_array *tr;
 557
 558	lockdep_assert_held(&event_mutex);
 559
 560	do_for_each_event_file(tr, file) {
 561		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 562			continue;
 563
 564		if (enable) {
 565			tracing_start_tgid_record();
 566			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 567		} else {
 568			tracing_stop_tgid_record();
 569			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 570				  &file->flags);
 571		}
 572	} while_for_each_event_file();
 573}
 574
 575static int __ftrace_event_enable_disable(struct trace_event_file *file,
 576					 int enable, int soft_disable)
 577{
 578	struct trace_event_call *call = file->event_call;
 579	struct trace_array *tr = file->tr;
 580	unsigned long file_flags = file->flags;
 581	int ret = 0;
 582	int disable;
 583
 584	switch (enable) {
 585	case 0:
 586		/*
 587		 * When soft_disable is set and enable is cleared, the sm_ref
 588		 * reference counter is decremented. If it reaches 0, we want
 589		 * to clear the SOFT_DISABLED flag but leave the event in the
 590		 * state that it was. That is, if the event was enabled and
 591		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 592		 * is set we do not want the event to be enabled before we
 593		 * clear the bit.
 594		 *
 595		 * When soft_disable is not set but the SOFT_MODE flag is,
 596		 * we do nothing. Do not disable the tracepoint, otherwise
 597		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 598		 */
 599		if (soft_disable) {
 600			if (atomic_dec_return(&file->sm_ref) > 0)
 601				break;
 602			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 603			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 604		} else
 605			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 606
 607		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 608			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 609			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 610				tracing_stop_cmdline_record();
 611				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 612			}
 613
 614			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 615				tracing_stop_tgid_record();
 616				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 617			}
 618
 619			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 620		}
 621		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 622		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 623			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 624		else
 625			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 626		break;
 627	case 1:
 628		/*
 629		 * When soft_disable is set and enable is set, we want to
 630		 * register the tracepoint for the event, but leave the event
 631		 * as is. That means, if the event was already enabled, we do
 632		 * nothing (but set SOFT_MODE). If the event is disabled, we
 633		 * set SOFT_DISABLED before enabling the event tracepoint, so
 634		 * it still seems to be disabled.
 635		 */
 636		if (!soft_disable)
 637			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 638		else {
 639			if (atomic_inc_return(&file->sm_ref) > 1)
 640				break;
 641			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 642		}
 643
 644		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 645			bool cmd = false, tgid = false;
 646
 647			/* Keep the event disabled, when going to SOFT_MODE. */
 648			if (soft_disable)
 649				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 650
 651			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 652				cmd = true;
 653				tracing_start_cmdline_record();
 654				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 655			}
 656
 657			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 658				tgid = true;
 659				tracing_start_tgid_record();
 660				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 661			}
 662
 663			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 664			if (ret) {
 665				if (cmd)
 666					tracing_stop_cmdline_record();
 667				if (tgid)
 668					tracing_stop_tgid_record();
 669				pr_info("event trace: Could not enable event "
 670					"%s\n", trace_event_name(call));
 671				break;
 672			}
 673			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 674
 675			/* WAS_ENABLED gets set but never cleared. */
 676			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 677		}
 678		break;
 679	}
 680
 681	/* Enable or disable use of trace_buffered_event */
 682	if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 683	    (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 684		if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 685			trace_buffered_event_enable();
 686		else
 687			trace_buffered_event_disable();
 688	}
 689
 690	return ret;
 691}
 692
 693int trace_event_enable_disable(struct trace_event_file *file,
 694			       int enable, int soft_disable)
 695{
 696	return __ftrace_event_enable_disable(file, enable, soft_disable);
 697}
 698
 699static int ftrace_event_enable_disable(struct trace_event_file *file,
 700				       int enable)
 701{
 702	return __ftrace_event_enable_disable(file, enable, 0);
 703}
 704
 705static void ftrace_clear_events(struct trace_array *tr)
 706{
 707	struct trace_event_file *file;
 708
 709	mutex_lock(&event_mutex);
 710	list_for_each_entry(file, &tr->events, list) {
 711		ftrace_event_enable_disable(file, 0);
 712	}
 713	mutex_unlock(&event_mutex);
 714}
 715
 716static void
 717event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 718{
 719	struct trace_pid_list *pid_list;
 720	struct trace_array *tr = data;
 721
 722	pid_list = rcu_dereference_raw(tr->filtered_pids);
 723	trace_filter_add_remove_task(pid_list, NULL, task);
 724
 725	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 726	trace_filter_add_remove_task(pid_list, NULL, task);
 727}
 728
 729static void
 730event_filter_pid_sched_process_fork(void *data,
 731				    struct task_struct *self,
 732				    struct task_struct *task)
 733{
 734	struct trace_pid_list *pid_list;
 735	struct trace_array *tr = data;
 736
 737	pid_list = rcu_dereference_sched(tr->filtered_pids);
 738	trace_filter_add_remove_task(pid_list, self, task);
 739
 740	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 741	trace_filter_add_remove_task(pid_list, self, task);
 742}
 743
 744void trace_event_follow_fork(struct trace_array *tr, bool enable)
 745{
 746	if (enable) {
 747		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 748						       tr, INT_MIN);
 749		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 750						       tr, INT_MAX);
 751	} else {
 752		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 753						    tr);
 754		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 755						    tr);
 756	}
 757}
 758
 759static void
 760event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 761		    struct task_struct *prev, struct task_struct *next)
 762{
 763	struct trace_array *tr = data;
 764	struct trace_pid_list *no_pid_list;
 765	struct trace_pid_list *pid_list;
 766	bool ret;
 767
 768	pid_list = rcu_dereference_sched(tr->filtered_pids);
 769	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 770
 771	/*
 772	 * Sched switch is funny, as we only want to ignore it
 773	 * in the notrace case if both prev and next should be ignored.
 774	 */
 775	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 776		trace_ignore_this_task(NULL, no_pid_list, next);
 777
 778	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 779		       (trace_ignore_this_task(pid_list, NULL, prev) &&
 780			trace_ignore_this_task(pid_list, NULL, next)));
 781}
 782
 783static void
 784event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 785		    struct task_struct *prev, struct task_struct *next)
 786{
 787	struct trace_array *tr = data;
 788	struct trace_pid_list *no_pid_list;
 789	struct trace_pid_list *pid_list;
 790
 791	pid_list = rcu_dereference_sched(tr->filtered_pids);
 792	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 793
 794	this_cpu_write(tr->array_buffer.data->ignore_pid,
 795		       trace_ignore_this_task(pid_list, no_pid_list, next));
 796}
 797
 798static void
 799event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 800{
 801	struct trace_array *tr = data;
 802	struct trace_pid_list *no_pid_list;
 803	struct trace_pid_list *pid_list;
 804
 805	/* Nothing to do if we are already tracing */
 806	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 807		return;
 808
 809	pid_list = rcu_dereference_sched(tr->filtered_pids);
 810	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 811
 812	this_cpu_write(tr->array_buffer.data->ignore_pid,
 813		       trace_ignore_this_task(pid_list, no_pid_list, task));
 814}
 815
 816static void
 817event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 818{
 819	struct trace_array *tr = data;
 820	struct trace_pid_list *no_pid_list;
 821	struct trace_pid_list *pid_list;
 822
 823	/* Nothing to do if we are not tracing */
 824	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 825		return;
 826
 827	pid_list = rcu_dereference_sched(tr->filtered_pids);
 828	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 829
 830	/* Set tracing if current is enabled */
 831	this_cpu_write(tr->array_buffer.data->ignore_pid,
 832		       trace_ignore_this_task(pid_list, no_pid_list, current));
 833}
 834
 835static void unregister_pid_events(struct trace_array *tr)
 836{
 
 
 
 
 
 
 
 
 
 837	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 838	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 839
 840	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 841	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 842
 843	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 844	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 845
 846	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 847	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 848}
 849
 850static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 851{
 852	struct trace_pid_list *pid_list;
 853	struct trace_pid_list *no_pid_list;
 854	struct trace_event_file *file;
 855	int cpu;
 856
 857	pid_list = rcu_dereference_protected(tr->filtered_pids,
 858					     lockdep_is_held(&event_mutex));
 859	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 860					     lockdep_is_held(&event_mutex));
 861
 862	/* Make sure there's something to do */
 863	if (!pid_type_enabled(type, pid_list, no_pid_list))
 864		return;
 865
 866	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 867		unregister_pid_events(tr);
 868
 869		list_for_each_entry(file, &tr->events, list) {
 870			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 871		}
 872
 873		for_each_possible_cpu(cpu)
 874			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 875	}
 876
 877	if (type & TRACE_PIDS)
 878		rcu_assign_pointer(tr->filtered_pids, NULL);
 879
 880	if (type & TRACE_NO_PIDS)
 881		rcu_assign_pointer(tr->filtered_no_pids, NULL);
 882
 883	/* Wait till all users are no longer using pid filtering */
 884	tracepoint_synchronize_unregister();
 885
 886	if ((type & TRACE_PIDS) && pid_list)
 887		trace_free_pid_list(pid_list);
 888
 889	if ((type & TRACE_NO_PIDS) && no_pid_list)
 890		trace_free_pid_list(no_pid_list);
 891}
 892
 893static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 894{
 895	mutex_lock(&event_mutex);
 896	__ftrace_clear_event_pids(tr, type);
 897	mutex_unlock(&event_mutex);
 898}
 899
 900static void __put_system(struct event_subsystem *system)
 901{
 902	struct event_filter *filter = system->filter;
 903
 904	WARN_ON_ONCE(system_refcount(system) == 0);
 905	if (system_refcount_dec(system))
 906		return;
 907
 908	list_del(&system->list);
 909
 910	if (filter) {
 911		kfree(filter->filter_string);
 912		kfree(filter);
 913	}
 914	kfree_const(system->name);
 915	kfree(system);
 916}
 917
 918static void __get_system(struct event_subsystem *system)
 919{
 920	WARN_ON_ONCE(system_refcount(system) == 0);
 921	system_refcount_inc(system);
 922}
 923
 924static void __get_system_dir(struct trace_subsystem_dir *dir)
 925{
 926	WARN_ON_ONCE(dir->ref_count == 0);
 927	dir->ref_count++;
 928	__get_system(dir->subsystem);
 929}
 930
 931static void __put_system_dir(struct trace_subsystem_dir *dir)
 932{
 933	WARN_ON_ONCE(dir->ref_count == 0);
 934	/* If the subsystem is about to be freed, the dir must be too */
 935	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 936
 937	__put_system(dir->subsystem);
 938	if (!--dir->ref_count)
 939		kfree(dir);
 940}
 941
 942static void put_system(struct trace_subsystem_dir *dir)
 943{
 944	mutex_lock(&event_mutex);
 945	__put_system_dir(dir);
 946	mutex_unlock(&event_mutex);
 947}
 948
 949static void remove_subsystem(struct trace_subsystem_dir *dir)
 950{
 951	if (!dir)
 952		return;
 953
 954	if (!--dir->nr_events) {
 955		tracefs_remove(dir->entry);
 956		list_del(&dir->list);
 957		__put_system_dir(dir);
 958	}
 959}
 960
 961static void remove_event_file_dir(struct trace_event_file *file)
 962{
 963	struct dentry *dir = file->dir;
 964	struct dentry *child;
 965
 966	if (dir) {
 967		spin_lock(&dir->d_lock);	/* probably unneeded */
 968		list_for_each_entry(child, &dir->d_subdirs, d_child) {
 969			if (d_really_is_positive(child))	/* probably unneeded */
 970				d_inode(child)->i_private = NULL;
 971		}
 972		spin_unlock(&dir->d_lock);
 973
 974		tracefs_remove(dir);
 975	}
 976
 977	list_del(&file->list);
 978	remove_subsystem(file->system);
 979	free_event_filter(file->filter);
 980	kmem_cache_free(file_cachep, file);
 981}
 982
 983/*
 984 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 985 */
 986static int
 987__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 988			      const char *sub, const char *event, int set)
 989{
 990	struct trace_event_file *file;
 991	struct trace_event_call *call;
 992	const char *name;
 993	int ret = -EINVAL;
 994	int eret = 0;
 995
 996	list_for_each_entry(file, &tr->events, list) {
 997
 998		call = file->event_call;
 999		name = trace_event_name(call);
1000
1001		if (!name || !call->class || !call->class->reg)
1002			continue;
1003
1004		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1005			continue;
1006
1007		if (match &&
1008		    strcmp(match, name) != 0 &&
1009		    strcmp(match, call->class->system) != 0)
1010			continue;
1011
1012		if (sub && strcmp(sub, call->class->system) != 0)
1013			continue;
1014
1015		if (event && strcmp(event, name) != 0)
1016			continue;
1017
1018		ret = ftrace_event_enable_disable(file, set);
1019
1020		/*
1021		 * Save the first error and return that. Some events
1022		 * may still have been enabled, but let the user
1023		 * know that something went wrong.
1024		 */
1025		if (ret && !eret)
1026			eret = ret;
1027
1028		ret = eret;
1029	}
1030
1031	return ret;
1032}
1033
1034static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1035				  const char *sub, const char *event, int set)
1036{
1037	int ret;
1038
1039	mutex_lock(&event_mutex);
1040	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1041	mutex_unlock(&event_mutex);
1042
1043	return ret;
1044}
1045
1046int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1047{
1048	char *event = NULL, *sub = NULL, *match;
1049	int ret;
1050
1051	if (!tr)
1052		return -ENOENT;
1053	/*
1054	 * The buf format can be <subsystem>:<event-name>
1055	 *  *:<event-name> means any event by that name.
1056	 *  :<event-name> is the same.
1057	 *
1058	 *  <subsystem>:* means all events in that subsystem
1059	 *  <subsystem>: means the same.
1060	 *
1061	 *  <name> (no ':') means all events in a subsystem with
1062	 *  the name <name> or any event that matches <name>
1063	 */
1064
1065	match = strsep(&buf, ":");
1066	if (buf) {
1067		sub = match;
1068		event = buf;
1069		match = NULL;
1070
1071		if (!strlen(sub) || strcmp(sub, "*") == 0)
1072			sub = NULL;
1073		if (!strlen(event) || strcmp(event, "*") == 0)
1074			event = NULL;
1075	}
1076
1077	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1078
1079	/* Put back the colon to allow this to be called again */
1080	if (buf)
1081		*(buf - 1) = ':';
1082
1083	return ret;
1084}
1085
1086/**
1087 * trace_set_clr_event - enable or disable an event
1088 * @system: system name to match (NULL for any system)
1089 * @event: event name to match (NULL for all events, within system)
1090 * @set: 1 to enable, 0 to disable
1091 *
1092 * This is a way for other parts of the kernel to enable or disable
1093 * event recording.
1094 *
1095 * Returns 0 on success, -EINVAL if the parameters do not match any
1096 * registered events.
1097 */
1098int trace_set_clr_event(const char *system, const char *event, int set)
1099{
1100	struct trace_array *tr = top_trace_array();
1101
1102	if (!tr)
1103		return -ENODEV;
1104
1105	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1106}
1107EXPORT_SYMBOL_GPL(trace_set_clr_event);
1108
1109/**
1110 * trace_array_set_clr_event - enable or disable an event for a trace array.
1111 * @tr: concerned trace array.
1112 * @system: system name to match (NULL for any system)
1113 * @event: event name to match (NULL for all events, within system)
1114 * @enable: true to enable, false to disable
1115 *
1116 * This is a way for other parts of the kernel to enable or disable
1117 * event recording.
1118 *
1119 * Returns 0 on success, -EINVAL if the parameters do not match any
1120 * registered events.
1121 */
1122int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1123		const char *event, bool enable)
1124{
1125	int set;
1126
1127	if (!tr)
1128		return -ENOENT;
1129
1130	set = (enable == true) ? 1 : 0;
1131	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1132}
1133EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1134
1135/* 128 should be much more than enough */
1136#define EVENT_BUF_SIZE		127
1137
1138static ssize_t
1139ftrace_event_write(struct file *file, const char __user *ubuf,
1140		   size_t cnt, loff_t *ppos)
1141{
1142	struct trace_parser parser;
1143	struct seq_file *m = file->private_data;
1144	struct trace_array *tr = m->private;
1145	ssize_t read, ret;
1146
1147	if (!cnt)
1148		return 0;
1149
1150	ret = tracing_update_buffers();
1151	if (ret < 0)
1152		return ret;
1153
1154	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1155		return -ENOMEM;
1156
1157	read = trace_get_user(&parser, ubuf, cnt, ppos);
1158
1159	if (read >= 0 && trace_parser_loaded((&parser))) {
1160		int set = 1;
1161
1162		if (*parser.buffer == '!')
1163			set = 0;
1164
 
 
1165		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1166		if (ret)
1167			goto out_put;
1168	}
1169
1170	ret = read;
1171
1172 out_put:
1173	trace_parser_put(&parser);
1174
1175	return ret;
1176}
1177
1178static void *
1179t_next(struct seq_file *m, void *v, loff_t *pos)
1180{
1181	struct trace_event_file *file = v;
1182	struct trace_event_call *call;
1183	struct trace_array *tr = m->private;
1184
1185	(*pos)++;
1186
1187	list_for_each_entry_continue(file, &tr->events, list) {
1188		call = file->event_call;
1189		/*
1190		 * The ftrace subsystem is for showing formats only.
1191		 * They can not be enabled or disabled via the event files.
1192		 */
1193		if (call->class && call->class->reg &&
1194		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1195			return file;
1196	}
1197
1198	return NULL;
1199}
1200
1201static void *t_start(struct seq_file *m, loff_t *pos)
1202{
1203	struct trace_event_file *file;
1204	struct trace_array *tr = m->private;
1205	loff_t l;
1206
1207	mutex_lock(&event_mutex);
1208
1209	file = list_entry(&tr->events, struct trace_event_file, list);
1210	for (l = 0; l <= *pos; ) {
1211		file = t_next(m, file, &l);
1212		if (!file)
1213			break;
1214	}
1215	return file;
1216}
1217
1218static void *
1219s_next(struct seq_file *m, void *v, loff_t *pos)
1220{
1221	struct trace_event_file *file = v;
1222	struct trace_array *tr = m->private;
1223
1224	(*pos)++;
1225
1226	list_for_each_entry_continue(file, &tr->events, list) {
1227		if (file->flags & EVENT_FILE_FL_ENABLED)
1228			return file;
1229	}
1230
1231	return NULL;
1232}
1233
1234static void *s_start(struct seq_file *m, loff_t *pos)
1235{
1236	struct trace_event_file *file;
1237	struct trace_array *tr = m->private;
1238	loff_t l;
1239
1240	mutex_lock(&event_mutex);
1241
1242	file = list_entry(&tr->events, struct trace_event_file, list);
1243	for (l = 0; l <= *pos; ) {
1244		file = s_next(m, file, &l);
1245		if (!file)
1246			break;
1247	}
1248	return file;
1249}
1250
1251static int t_show(struct seq_file *m, void *v)
1252{
1253	struct trace_event_file *file = v;
1254	struct trace_event_call *call = file->event_call;
1255
1256	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1257		seq_printf(m, "%s:", call->class->system);
1258	seq_printf(m, "%s\n", trace_event_name(call));
1259
1260	return 0;
1261}
1262
1263static void t_stop(struct seq_file *m, void *p)
1264{
1265	mutex_unlock(&event_mutex);
1266}
1267
1268static void *
1269__next(struct seq_file *m, void *v, loff_t *pos, int type)
1270{
1271	struct trace_array *tr = m->private;
1272	struct trace_pid_list *pid_list;
1273
1274	if (type == TRACE_PIDS)
1275		pid_list = rcu_dereference_sched(tr->filtered_pids);
1276	else
1277		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1278
1279	return trace_pid_next(pid_list, v, pos);
1280}
1281
1282static void *
1283p_next(struct seq_file *m, void *v, loff_t *pos)
1284{
1285	return __next(m, v, pos, TRACE_PIDS);
1286}
1287
1288static void *
1289np_next(struct seq_file *m, void *v, loff_t *pos)
1290{
1291	return __next(m, v, pos, TRACE_NO_PIDS);
1292}
1293
1294static void *__start(struct seq_file *m, loff_t *pos, int type)
1295	__acquires(RCU)
1296{
1297	struct trace_pid_list *pid_list;
1298	struct trace_array *tr = m->private;
1299
1300	/*
1301	 * Grab the mutex, to keep calls to p_next() having the same
1302	 * tr->filtered_pids as p_start() has.
1303	 * If we just passed the tr->filtered_pids around, then RCU would
1304	 * have been enough, but doing that makes things more complex.
1305	 */
1306	mutex_lock(&event_mutex);
1307	rcu_read_lock_sched();
1308
1309	if (type == TRACE_PIDS)
1310		pid_list = rcu_dereference_sched(tr->filtered_pids);
1311	else
1312		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1313
1314	if (!pid_list)
1315		return NULL;
1316
1317	return trace_pid_start(pid_list, pos);
1318}
1319
1320static void *p_start(struct seq_file *m, loff_t *pos)
1321	__acquires(RCU)
1322{
1323	return __start(m, pos, TRACE_PIDS);
1324}
1325
1326static void *np_start(struct seq_file *m, loff_t *pos)
1327	__acquires(RCU)
1328{
1329	return __start(m, pos, TRACE_NO_PIDS);
1330}
1331
1332static void p_stop(struct seq_file *m, void *p)
1333	__releases(RCU)
1334{
1335	rcu_read_unlock_sched();
1336	mutex_unlock(&event_mutex);
1337}
1338
1339static ssize_t
1340event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1341		  loff_t *ppos)
1342{
1343	struct trace_event_file *file;
1344	unsigned long flags;
1345	char buf[4] = "0";
1346
1347	mutex_lock(&event_mutex);
1348	file = event_file_data(filp);
1349	if (likely(file))
1350		flags = file->flags;
1351	mutex_unlock(&event_mutex);
1352
1353	if (!file)
1354		return -ENODEV;
1355
1356	if (flags & EVENT_FILE_FL_ENABLED &&
1357	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1358		strcpy(buf, "1");
1359
1360	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1361	    flags & EVENT_FILE_FL_SOFT_MODE)
1362		strcat(buf, "*");
1363
1364	strcat(buf, "\n");
1365
1366	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1367}
1368
1369static ssize_t
1370event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1371		   loff_t *ppos)
1372{
1373	struct trace_event_file *file;
1374	unsigned long val;
1375	int ret;
1376
1377	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1378	if (ret)
1379		return ret;
1380
1381	ret = tracing_update_buffers();
1382	if (ret < 0)
1383		return ret;
1384
1385	switch (val) {
1386	case 0:
1387	case 1:
1388		ret = -ENODEV;
1389		mutex_lock(&event_mutex);
1390		file = event_file_data(filp);
1391		if (likely(file))
1392			ret = ftrace_event_enable_disable(file, val);
1393		mutex_unlock(&event_mutex);
1394		break;
1395
1396	default:
1397		return -EINVAL;
1398	}
1399
1400	*ppos += cnt;
1401
1402	return ret ? ret : cnt;
1403}
1404
1405static ssize_t
1406system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1407		   loff_t *ppos)
1408{
1409	const char set_to_char[4] = { '?', '0', '1', 'X' };
1410	struct trace_subsystem_dir *dir = filp->private_data;
1411	struct event_subsystem *system = dir->subsystem;
1412	struct trace_event_call *call;
1413	struct trace_event_file *file;
1414	struct trace_array *tr = dir->tr;
1415	char buf[2];
1416	int set = 0;
1417	int ret;
1418
1419	mutex_lock(&event_mutex);
1420	list_for_each_entry(file, &tr->events, list) {
1421		call = file->event_call;
1422		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1423		    !trace_event_name(call) || !call->class || !call->class->reg)
1424			continue;
1425
1426		if (system && strcmp(call->class->system, system->name) != 0)
1427			continue;
1428
1429		/*
1430		 * We need to find out if all the events are set
1431		 * or if all events or cleared, or if we have
1432		 * a mixture.
1433		 */
1434		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1435
1436		/*
1437		 * If we have a mixture, no need to look further.
1438		 */
1439		if (set == 3)
1440			break;
1441	}
1442	mutex_unlock(&event_mutex);
1443
1444	buf[0] = set_to_char[set];
1445	buf[1] = '\n';
1446
1447	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1448
1449	return ret;
1450}
1451
1452static ssize_t
1453system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1454		    loff_t *ppos)
1455{
1456	struct trace_subsystem_dir *dir = filp->private_data;
1457	struct event_subsystem *system = dir->subsystem;
1458	const char *name = NULL;
1459	unsigned long val;
1460	ssize_t ret;
1461
1462	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1463	if (ret)
1464		return ret;
1465
1466	ret = tracing_update_buffers();
1467	if (ret < 0)
1468		return ret;
1469
1470	if (val != 0 && val != 1)
1471		return -EINVAL;
1472
1473	/*
1474	 * Opening of "enable" adds a ref count to system,
1475	 * so the name is safe to use.
1476	 */
1477	if (system)
1478		name = system->name;
1479
1480	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1481	if (ret)
1482		goto out;
1483
1484	ret = cnt;
1485
1486out:
1487	*ppos += cnt;
1488
1489	return ret;
1490}
1491
1492enum {
1493	FORMAT_HEADER		= 1,
1494	FORMAT_FIELD_SEPERATOR	= 2,
1495	FORMAT_PRINTFMT		= 3,
1496};
1497
1498static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1499{
1500	struct trace_event_call *call = event_file_data(m->private);
1501	struct list_head *common_head = &ftrace_common_fields;
1502	struct list_head *head = trace_get_fields(call);
1503	struct list_head *node = v;
1504
1505	(*pos)++;
1506
1507	switch ((unsigned long)v) {
1508	case FORMAT_HEADER:
1509		node = common_head;
1510		break;
1511
1512	case FORMAT_FIELD_SEPERATOR:
1513		node = head;
1514		break;
1515
1516	case FORMAT_PRINTFMT:
1517		/* all done */
1518		return NULL;
1519	}
1520
1521	node = node->prev;
1522	if (node == common_head)
1523		return (void *)FORMAT_FIELD_SEPERATOR;
1524	else if (node == head)
1525		return (void *)FORMAT_PRINTFMT;
1526	else
1527		return node;
1528}
1529
1530static int f_show(struct seq_file *m, void *v)
1531{
1532	struct trace_event_call *call = event_file_data(m->private);
1533	struct ftrace_event_field *field;
1534	const char *array_descriptor;
1535
1536	switch ((unsigned long)v) {
1537	case FORMAT_HEADER:
1538		seq_printf(m, "name: %s\n", trace_event_name(call));
1539		seq_printf(m, "ID: %d\n", call->event.type);
1540		seq_puts(m, "format:\n");
1541		return 0;
1542
1543	case FORMAT_FIELD_SEPERATOR:
1544		seq_putc(m, '\n');
1545		return 0;
1546
1547	case FORMAT_PRINTFMT:
1548		seq_printf(m, "\nprint fmt: %s\n",
1549			   call->print_fmt);
1550		return 0;
1551	}
1552
1553	field = list_entry(v, struct ftrace_event_field, link);
1554	/*
1555	 * Smartly shows the array type(except dynamic array).
1556	 * Normal:
1557	 *	field:TYPE VAR
1558	 * If TYPE := TYPE[LEN], it is shown:
1559	 *	field:TYPE VAR[LEN]
1560	 */
1561	array_descriptor = strchr(field->type, '[');
1562
1563	if (str_has_prefix(field->type, "__data_loc"))
1564		array_descriptor = NULL;
1565
1566	if (!array_descriptor)
1567		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1568			   field->type, field->name, field->offset,
1569			   field->size, !!field->is_signed);
1570	else
1571		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1572			   (int)(array_descriptor - field->type),
1573			   field->type, field->name,
1574			   array_descriptor, field->offset,
1575			   field->size, !!field->is_signed);
1576
1577	return 0;
1578}
1579
1580static void *f_start(struct seq_file *m, loff_t *pos)
1581{
1582	void *p = (void *)FORMAT_HEADER;
1583	loff_t l = 0;
1584
1585	/* ->stop() is called even if ->start() fails */
1586	mutex_lock(&event_mutex);
1587	if (!event_file_data(m->private))
1588		return ERR_PTR(-ENODEV);
1589
1590	while (l < *pos && p)
1591		p = f_next(m, p, &l);
1592
1593	return p;
1594}
1595
1596static void f_stop(struct seq_file *m, void *p)
1597{
1598	mutex_unlock(&event_mutex);
1599}
1600
1601static const struct seq_operations trace_format_seq_ops = {
1602	.start		= f_start,
1603	.next		= f_next,
1604	.stop		= f_stop,
1605	.show		= f_show,
1606};
1607
1608static int trace_format_open(struct inode *inode, struct file *file)
1609{
1610	struct seq_file *m;
1611	int ret;
1612
1613	/* Do we want to hide event format files on tracefs lockdown? */
1614
1615	ret = seq_open(file, &trace_format_seq_ops);
1616	if (ret < 0)
1617		return ret;
1618
1619	m = file->private_data;
1620	m->private = file;
1621
1622	return 0;
1623}
1624
1625static ssize_t
1626event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1627{
1628	int id = (long)event_file_data(filp);
1629	char buf[32];
1630	int len;
1631
 
 
 
1632	if (unlikely(!id))
1633		return -ENODEV;
1634
1635	len = sprintf(buf, "%d\n", id);
1636
1637	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1638}
1639
1640static ssize_t
1641event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1642		  loff_t *ppos)
1643{
1644	struct trace_event_file *file;
1645	struct trace_seq *s;
1646	int r = -ENODEV;
1647
1648	if (*ppos)
1649		return 0;
1650
1651	s = kmalloc(sizeof(*s), GFP_KERNEL);
1652
1653	if (!s)
1654		return -ENOMEM;
1655
1656	trace_seq_init(s);
1657
1658	mutex_lock(&event_mutex);
1659	file = event_file_data(filp);
1660	if (file)
1661		print_event_filter(file, s);
1662	mutex_unlock(&event_mutex);
1663
1664	if (file)
1665		r = simple_read_from_buffer(ubuf, cnt, ppos,
1666					    s->buffer, trace_seq_used(s));
1667
1668	kfree(s);
1669
1670	return r;
1671}
1672
1673static ssize_t
1674event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1675		   loff_t *ppos)
1676{
1677	struct trace_event_file *file;
1678	char *buf;
1679	int err = -ENODEV;
1680
1681	if (cnt >= PAGE_SIZE)
1682		return -EINVAL;
1683
1684	buf = memdup_user_nul(ubuf, cnt);
1685	if (IS_ERR(buf))
1686		return PTR_ERR(buf);
1687
1688	mutex_lock(&event_mutex);
1689	file = event_file_data(filp);
1690	if (file)
1691		err = apply_event_filter(file, buf);
1692	mutex_unlock(&event_mutex);
1693
1694	kfree(buf);
1695	if (err < 0)
1696		return err;
1697
1698	*ppos += cnt;
1699
1700	return cnt;
1701}
1702
1703static LIST_HEAD(event_subsystems);
1704
1705static int subsystem_open(struct inode *inode, struct file *filp)
1706{
1707	struct event_subsystem *system = NULL;
1708	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1709	struct trace_array *tr;
1710	int ret;
1711
1712	if (tracing_is_disabled())
1713		return -ENODEV;
1714
1715	/* Make sure the system still exists */
 
1716	mutex_lock(&event_mutex);
1717	mutex_lock(&trace_types_lock);
1718	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1719		list_for_each_entry(dir, &tr->systems, list) {
1720			if (dir == inode->i_private) {
1721				/* Don't open systems with no events */
1722				if (dir->nr_events) {
1723					__get_system_dir(dir);
1724					system = dir->subsystem;
1725				}
1726				goto exit_loop;
1727			}
1728		}
1729	}
1730 exit_loop:
 
1731	mutex_unlock(&trace_types_lock);
1732	mutex_unlock(&event_mutex);
1733
1734	if (!system)
1735		return -ENODEV;
1736
1737	/* Some versions of gcc think dir can be uninitialized here */
1738	WARN_ON(!dir);
1739
1740	/* Still need to increment the ref count of the system */
1741	if (trace_array_get(tr) < 0) {
1742		put_system(dir);
1743		return -ENODEV;
1744	}
1745
1746	ret = tracing_open_generic(inode, filp);
1747	if (ret < 0) {
1748		trace_array_put(tr);
1749		put_system(dir);
1750	}
1751
1752	return ret;
1753}
1754
1755static int system_tr_open(struct inode *inode, struct file *filp)
1756{
1757	struct trace_subsystem_dir *dir;
1758	struct trace_array *tr = inode->i_private;
1759	int ret;
1760
 
 
 
 
 
 
1761	/* Make a temporary dir that has no system but points to tr */
1762	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1763	if (!dir)
 
1764		return -ENOMEM;
 
1765
1766	ret = tracing_open_generic_tr(inode, filp);
 
 
1767	if (ret < 0) {
 
1768		kfree(dir);
1769		return ret;
1770	}
1771	dir->tr = tr;
1772	filp->private_data = dir;
1773
1774	return 0;
1775}
1776
1777static int subsystem_release(struct inode *inode, struct file *file)
1778{
1779	struct trace_subsystem_dir *dir = file->private_data;
1780
1781	trace_array_put(dir->tr);
1782
1783	/*
1784	 * If dir->subsystem is NULL, then this is a temporary
1785	 * descriptor that was made for a trace_array to enable
1786	 * all subsystems.
1787	 */
1788	if (dir->subsystem)
1789		put_system(dir);
1790	else
1791		kfree(dir);
1792
1793	return 0;
1794}
1795
1796static ssize_t
1797subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1798		      loff_t *ppos)
1799{
1800	struct trace_subsystem_dir *dir = filp->private_data;
1801	struct event_subsystem *system = dir->subsystem;
1802	struct trace_seq *s;
1803	int r;
1804
1805	if (*ppos)
1806		return 0;
1807
1808	s = kmalloc(sizeof(*s), GFP_KERNEL);
1809	if (!s)
1810		return -ENOMEM;
1811
1812	trace_seq_init(s);
1813
1814	print_subsystem_event_filter(system, s);
1815	r = simple_read_from_buffer(ubuf, cnt, ppos,
1816				    s->buffer, trace_seq_used(s));
1817
1818	kfree(s);
1819
1820	return r;
1821}
1822
1823static ssize_t
1824subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1825		       loff_t *ppos)
1826{
1827	struct trace_subsystem_dir *dir = filp->private_data;
1828	char *buf;
1829	int err;
1830
1831	if (cnt >= PAGE_SIZE)
1832		return -EINVAL;
1833
1834	buf = memdup_user_nul(ubuf, cnt);
1835	if (IS_ERR(buf))
1836		return PTR_ERR(buf);
1837
1838	err = apply_subsystem_event_filter(dir, buf);
1839	kfree(buf);
1840	if (err < 0)
1841		return err;
1842
1843	*ppos += cnt;
1844
1845	return cnt;
1846}
1847
1848static ssize_t
1849show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1850{
1851	int (*func)(struct trace_seq *s) = filp->private_data;
1852	struct trace_seq *s;
1853	int r;
1854
1855	if (*ppos)
1856		return 0;
1857
1858	s = kmalloc(sizeof(*s), GFP_KERNEL);
1859	if (!s)
1860		return -ENOMEM;
1861
1862	trace_seq_init(s);
1863
1864	func(s);
1865	r = simple_read_from_buffer(ubuf, cnt, ppos,
1866				    s->buffer, trace_seq_used(s));
1867
1868	kfree(s);
1869
1870	return r;
1871}
1872
1873static void ignore_task_cpu(void *data)
1874{
1875	struct trace_array *tr = data;
1876	struct trace_pid_list *pid_list;
1877	struct trace_pid_list *no_pid_list;
1878
1879	/*
1880	 * This function is called by on_each_cpu() while the
1881	 * event_mutex is held.
1882	 */
1883	pid_list = rcu_dereference_protected(tr->filtered_pids,
1884					     mutex_is_locked(&event_mutex));
1885	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1886					     mutex_is_locked(&event_mutex));
1887
1888	this_cpu_write(tr->array_buffer.data->ignore_pid,
1889		       trace_ignore_this_task(pid_list, no_pid_list, current));
1890}
1891
1892static void register_pid_events(struct trace_array *tr)
1893{
1894	/*
1895	 * Register a probe that is called before all other probes
1896	 * to set ignore_pid if next or prev do not match.
1897	 * Register a probe this is called after all other probes
1898	 * to only keep ignore_pid set if next pid matches.
1899	 */
1900	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1901					 tr, INT_MAX);
1902	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1903					 tr, 0);
1904
1905	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1906					 tr, INT_MAX);
1907	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1908					 tr, 0);
1909
1910	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1911					     tr, INT_MAX);
1912	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1913					     tr, 0);
1914
1915	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1916					 tr, INT_MAX);
1917	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1918					 tr, 0);
1919}
1920
1921static ssize_t
1922event_pid_write(struct file *filp, const char __user *ubuf,
1923		size_t cnt, loff_t *ppos, int type)
1924{
1925	struct seq_file *m = filp->private_data;
1926	struct trace_array *tr = m->private;
1927	struct trace_pid_list *filtered_pids = NULL;
1928	struct trace_pid_list *other_pids = NULL;
1929	struct trace_pid_list *pid_list;
1930	struct trace_event_file *file;
1931	ssize_t ret;
1932
1933	if (!cnt)
1934		return 0;
1935
1936	ret = tracing_update_buffers();
1937	if (ret < 0)
1938		return ret;
1939
1940	mutex_lock(&event_mutex);
1941
1942	if (type == TRACE_PIDS) {
1943		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1944							  lockdep_is_held(&event_mutex));
1945		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1946							  lockdep_is_held(&event_mutex));
1947	} else {
1948		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1949							  lockdep_is_held(&event_mutex));
1950		other_pids = rcu_dereference_protected(tr->filtered_pids,
1951							  lockdep_is_held(&event_mutex));
1952	}
1953
1954	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1955	if (ret < 0)
1956		goto out;
1957
1958	if (type == TRACE_PIDS)
1959		rcu_assign_pointer(tr->filtered_pids, pid_list);
1960	else
1961		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1962
1963	list_for_each_entry(file, &tr->events, list) {
1964		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1965	}
1966
1967	if (filtered_pids) {
1968		tracepoint_synchronize_unregister();
1969		trace_free_pid_list(filtered_pids);
1970	} else if (pid_list && !other_pids) {
1971		register_pid_events(tr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1972	}
1973
1974	/*
1975	 * Ignoring of pids is done at task switch. But we have to
1976	 * check for those tasks that are currently running.
1977	 * Always do this in case a pid was appended or removed.
1978	 */
1979	on_each_cpu(ignore_task_cpu, tr, 1);
1980
1981 out:
1982	mutex_unlock(&event_mutex);
1983
1984	if (ret > 0)
1985		*ppos += ret;
1986
1987	return ret;
1988}
1989
1990static ssize_t
1991ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1992		       size_t cnt, loff_t *ppos)
1993{
1994	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1995}
1996
1997static ssize_t
1998ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1999			size_t cnt, loff_t *ppos)
2000{
2001	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2002}
2003
2004static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2005static int ftrace_event_set_open(struct inode *inode, struct file *file);
2006static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2007static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2008static int ftrace_event_release(struct inode *inode, struct file *file);
2009
2010static const struct seq_operations show_event_seq_ops = {
2011	.start = t_start,
2012	.next = t_next,
2013	.show = t_show,
2014	.stop = t_stop,
2015};
2016
2017static const struct seq_operations show_set_event_seq_ops = {
2018	.start = s_start,
2019	.next = s_next,
2020	.show = t_show,
2021	.stop = t_stop,
2022};
2023
2024static const struct seq_operations show_set_pid_seq_ops = {
2025	.start = p_start,
2026	.next = p_next,
2027	.show = trace_pid_show,
2028	.stop = p_stop,
2029};
2030
2031static const struct seq_operations show_set_no_pid_seq_ops = {
2032	.start = np_start,
2033	.next = np_next,
2034	.show = trace_pid_show,
2035	.stop = p_stop,
2036};
2037
2038static const struct file_operations ftrace_avail_fops = {
2039	.open = ftrace_event_avail_open,
2040	.read = seq_read,
2041	.llseek = seq_lseek,
2042	.release = seq_release,
2043};
2044
2045static const struct file_operations ftrace_set_event_fops = {
2046	.open = ftrace_event_set_open,
2047	.read = seq_read,
2048	.write = ftrace_event_write,
2049	.llseek = seq_lseek,
2050	.release = ftrace_event_release,
2051};
2052
2053static const struct file_operations ftrace_set_event_pid_fops = {
2054	.open = ftrace_event_set_pid_open,
2055	.read = seq_read,
2056	.write = ftrace_event_pid_write,
2057	.llseek = seq_lseek,
2058	.release = ftrace_event_release,
2059};
2060
2061static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2062	.open = ftrace_event_set_npid_open,
2063	.read = seq_read,
2064	.write = ftrace_event_npid_write,
2065	.llseek = seq_lseek,
2066	.release = ftrace_event_release,
2067};
2068
2069static const struct file_operations ftrace_enable_fops = {
2070	.open = tracing_open_generic,
2071	.read = event_enable_read,
2072	.write = event_enable_write,
2073	.llseek = default_llseek,
2074};
2075
2076static const struct file_operations ftrace_event_format_fops = {
2077	.open = trace_format_open,
2078	.read = seq_read,
2079	.llseek = seq_lseek,
2080	.release = seq_release,
2081};
2082
2083static const struct file_operations ftrace_event_id_fops = {
2084	.read = event_id_read,
2085	.llseek = default_llseek,
2086};
2087
2088static const struct file_operations ftrace_event_filter_fops = {
2089	.open = tracing_open_generic,
2090	.read = event_filter_read,
2091	.write = event_filter_write,
2092	.llseek = default_llseek,
2093};
2094
2095static const struct file_operations ftrace_subsystem_filter_fops = {
2096	.open = subsystem_open,
2097	.read = subsystem_filter_read,
2098	.write = subsystem_filter_write,
2099	.llseek = default_llseek,
2100	.release = subsystem_release,
2101};
2102
2103static const struct file_operations ftrace_system_enable_fops = {
2104	.open = subsystem_open,
2105	.read = system_enable_read,
2106	.write = system_enable_write,
2107	.llseek = default_llseek,
2108	.release = subsystem_release,
2109};
2110
2111static const struct file_operations ftrace_tr_enable_fops = {
2112	.open = system_tr_open,
2113	.read = system_enable_read,
2114	.write = system_enable_write,
2115	.llseek = default_llseek,
2116	.release = subsystem_release,
2117};
2118
2119static const struct file_operations ftrace_show_header_fops = {
2120	.open = tracing_open_generic,
2121	.read = show_header,
2122	.llseek = default_llseek,
2123};
2124
2125static int
2126ftrace_event_open(struct inode *inode, struct file *file,
2127		  const struct seq_operations *seq_ops)
2128{
2129	struct seq_file *m;
2130	int ret;
2131
2132	ret = security_locked_down(LOCKDOWN_TRACEFS);
2133	if (ret)
2134		return ret;
2135
2136	ret = seq_open(file, seq_ops);
2137	if (ret < 0)
2138		return ret;
2139	m = file->private_data;
2140	/* copy tr over to seq ops */
2141	m->private = inode->i_private;
2142
2143	return ret;
2144}
2145
2146static int ftrace_event_release(struct inode *inode, struct file *file)
2147{
2148	struct trace_array *tr = inode->i_private;
2149
2150	trace_array_put(tr);
2151
2152	return seq_release(inode, file);
2153}
2154
2155static int
2156ftrace_event_avail_open(struct inode *inode, struct file *file)
2157{
2158	const struct seq_operations *seq_ops = &show_event_seq_ops;
2159
2160	/* Checks for tracefs lockdown */
2161	return ftrace_event_open(inode, file, seq_ops);
2162}
2163
2164static int
2165ftrace_event_set_open(struct inode *inode, struct file *file)
2166{
2167	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2168	struct trace_array *tr = inode->i_private;
2169	int ret;
2170
2171	ret = tracing_check_open_get_tr(tr);
2172	if (ret)
2173		return ret;
2174
2175	if ((file->f_mode & FMODE_WRITE) &&
2176	    (file->f_flags & O_TRUNC))
2177		ftrace_clear_events(tr);
2178
2179	ret = ftrace_event_open(inode, file, seq_ops);
2180	if (ret < 0)
2181		trace_array_put(tr);
2182	return ret;
2183}
2184
2185static int
2186ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2187{
2188	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2189	struct trace_array *tr = inode->i_private;
2190	int ret;
2191
2192	ret = tracing_check_open_get_tr(tr);
2193	if (ret)
2194		return ret;
2195
2196	if ((file->f_mode & FMODE_WRITE) &&
2197	    (file->f_flags & O_TRUNC))
2198		ftrace_clear_event_pids(tr, TRACE_PIDS);
2199
2200	ret = ftrace_event_open(inode, file, seq_ops);
2201	if (ret < 0)
2202		trace_array_put(tr);
2203	return ret;
2204}
2205
2206static int
2207ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2208{
2209	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2210	struct trace_array *tr = inode->i_private;
2211	int ret;
2212
2213	ret = tracing_check_open_get_tr(tr);
2214	if (ret)
2215		return ret;
2216
2217	if ((file->f_mode & FMODE_WRITE) &&
2218	    (file->f_flags & O_TRUNC))
2219		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2220
2221	ret = ftrace_event_open(inode, file, seq_ops);
2222	if (ret < 0)
2223		trace_array_put(tr);
2224	return ret;
2225}
2226
2227static struct event_subsystem *
2228create_new_subsystem(const char *name)
2229{
2230	struct event_subsystem *system;
2231
2232	/* need to create new entry */
2233	system = kmalloc(sizeof(*system), GFP_KERNEL);
2234	if (!system)
2235		return NULL;
2236
2237	system->ref_count = 1;
2238
2239	/* Only allocate if dynamic (kprobes and modules) */
2240	system->name = kstrdup_const(name, GFP_KERNEL);
2241	if (!system->name)
2242		goto out_free;
2243
2244	system->filter = NULL;
2245
2246	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2247	if (!system->filter)
2248		goto out_free;
2249
2250	list_add(&system->list, &event_subsystems);
2251
2252	return system;
2253
2254 out_free:
2255	kfree_const(system->name);
2256	kfree(system);
2257	return NULL;
2258}
2259
2260static struct dentry *
2261event_subsystem_dir(struct trace_array *tr, const char *name,
2262		    struct trace_event_file *file, struct dentry *parent)
2263{
2264	struct trace_subsystem_dir *dir;
2265	struct event_subsystem *system;
2266	struct dentry *entry;
2267
2268	/* First see if we did not already create this dir */
2269	list_for_each_entry(dir, &tr->systems, list) {
2270		system = dir->subsystem;
2271		if (strcmp(system->name, name) == 0) {
2272			dir->nr_events++;
2273			file->system = dir;
2274			return dir->entry;
2275		}
2276	}
2277
2278	/* Now see if the system itself exists. */
2279	list_for_each_entry(system, &event_subsystems, list) {
2280		if (strcmp(system->name, name) == 0)
2281			break;
2282	}
2283	/* Reset system variable when not found */
2284	if (&system->list == &event_subsystems)
2285		system = NULL;
2286
2287	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2288	if (!dir)
2289		goto out_fail;
2290
2291	if (!system) {
2292		system = create_new_subsystem(name);
2293		if (!system)
2294			goto out_free;
2295	} else
2296		__get_system(system);
2297
2298	dir->entry = tracefs_create_dir(name, parent);
2299	if (!dir->entry) {
2300		pr_warn("Failed to create system directory %s\n", name);
2301		__put_system(system);
2302		goto out_free;
2303	}
2304
2305	dir->tr = tr;
2306	dir->ref_count = 1;
2307	dir->nr_events = 1;
2308	dir->subsystem = system;
2309	file->system = dir;
2310
2311	/* the ftrace system is special, do not create enable or filter files */
2312	if (strcmp(name, "ftrace") != 0) {
2313
2314		entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2315					    &ftrace_subsystem_filter_fops);
2316		if (!entry) {
2317			kfree(system->filter);
2318			system->filter = NULL;
2319			pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2320		}
2321
2322		trace_create_file("enable", 0644, dir->entry, dir,
2323				  &ftrace_system_enable_fops);
2324	}
2325
2326	list_add(&dir->list, &tr->systems);
2327
2328	return dir->entry;
2329
2330 out_free:
2331	kfree(dir);
2332 out_fail:
2333	/* Only print this message if failed on memory allocation */
2334	if (!dir || !system)
2335		pr_warn("No memory to create event subsystem %s\n", name);
2336	return NULL;
2337}
2338
2339static int
2340event_define_fields(struct trace_event_call *call)
2341{
2342	struct list_head *head;
2343	int ret = 0;
2344
2345	/*
2346	 * Other events may have the same class. Only update
2347	 * the fields if they are not already defined.
2348	 */
2349	head = trace_get_fields(call);
2350	if (list_empty(head)) {
2351		struct trace_event_fields *field = call->class->fields_array;
2352		unsigned int offset = sizeof(struct trace_entry);
2353
2354		for (; field->type; field++) {
2355			if (field->type == TRACE_FUNCTION_TYPE) {
2356				field->define_fields(call);
2357				break;
2358			}
2359
2360			offset = ALIGN(offset, field->align);
2361			ret = trace_define_field(call, field->type, field->name,
2362						 offset, field->size,
2363						 field->is_signed, field->filter_type);
2364			if (WARN_ON_ONCE(ret)) {
2365				pr_err("error code is %d\n", ret);
2366				break;
2367			}
2368
2369			offset += field->size;
2370		}
2371	}
2372
2373	return ret;
2374}
2375
2376static int
2377event_create_dir(struct dentry *parent, struct trace_event_file *file)
2378{
2379	struct trace_event_call *call = file->event_call;
2380	struct trace_array *tr = file->tr;
 
2381	struct dentry *d_events;
2382	const char *name;
2383	int ret;
2384
2385	/*
2386	 * If the trace point header did not define TRACE_SYSTEM
2387	 * then the system would be called "TRACE_SYSTEM".
2388	 */
2389	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2390		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2391		if (!d_events)
2392			return -ENOMEM;
2393	} else
2394		d_events = parent;
2395
2396	name = trace_event_name(call);
2397	file->dir = tracefs_create_dir(name, d_events);
2398	if (!file->dir) {
2399		pr_warn("Could not create tracefs '%s' directory\n", name);
2400		return -1;
2401	}
2402
2403	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2404		trace_create_file("enable", 0644, file->dir, file,
2405				  &ftrace_enable_fops);
2406
2407#ifdef CONFIG_PERF_EVENTS
2408	if (call->event.type && call->class->reg)
2409		trace_create_file("id", 0444, file->dir,
2410				  (void *)(long)call->event.type,
2411				  &ftrace_event_id_fops);
2412#endif
2413
2414	ret = event_define_fields(call);
2415	if (ret < 0) {
2416		pr_warn("Could not initialize trace point events/%s\n", name);
2417		return ret;
 
 
 
 
 
 
 
 
2418	}
 
 
2419
2420	/*
2421	 * Only event directories that can be enabled should have
2422	 * triggers or filters.
2423	 */
2424	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2425		trace_create_file("filter", 0644, file->dir, file,
2426				  &ftrace_event_filter_fops);
2427
2428		trace_create_file("trigger", 0644, file->dir, file,
2429				  &event_trigger_fops);
2430	}
2431
2432#ifdef CONFIG_HIST_TRIGGERS
2433	trace_create_file("hist", 0444, file->dir, file,
2434			  &event_hist_fops);
2435#endif
2436#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2437	trace_create_file("hist_debug", 0444, file->dir, file,
2438			  &event_hist_debug_fops);
2439#endif
2440	trace_create_file("format", 0444, file->dir, call,
2441			  &ftrace_event_format_fops);
2442
2443#ifdef CONFIG_TRACE_EVENT_INJECT
2444	if (call->event.type && call->class->reg)
2445		trace_create_file("inject", 0200, file->dir, file,
2446				  &event_inject_fops);
2447#endif
2448
2449	return 0;
2450}
2451
2452static void remove_event_from_tracers(struct trace_event_call *call)
2453{
2454	struct trace_event_file *file;
2455	struct trace_array *tr;
2456
2457	do_for_each_event_file_safe(tr, file) {
2458		if (file->event_call != call)
2459			continue;
2460
2461		remove_event_file_dir(file);
2462		/*
2463		 * The do_for_each_event_file_safe() is
2464		 * a double loop. After finding the call for this
2465		 * trace_array, we use break to jump to the next
2466		 * trace_array.
2467		 */
2468		break;
2469	} while_for_each_event_file();
2470}
2471
2472static void event_remove(struct trace_event_call *call)
2473{
2474	struct trace_array *tr;
2475	struct trace_event_file *file;
2476
2477	do_for_each_event_file(tr, file) {
2478		if (file->event_call != call)
2479			continue;
2480
2481		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2482			tr->clear_trace = true;
2483
2484		ftrace_event_enable_disable(file, 0);
2485		/*
2486		 * The do_for_each_event_file() is
2487		 * a double loop. After finding the call for this
2488		 * trace_array, we use break to jump to the next
2489		 * trace_array.
2490		 */
2491		break;
2492	} while_for_each_event_file();
2493
2494	if (call->event.funcs)
2495		__unregister_trace_event(&call->event);
2496	remove_event_from_tracers(call);
2497	list_del(&call->list);
2498}
2499
2500static int event_init(struct trace_event_call *call)
2501{
2502	int ret = 0;
2503	const char *name;
2504
2505	name = trace_event_name(call);
2506	if (WARN_ON(!name))
2507		return -EINVAL;
2508
2509	if (call->class->raw_init) {
2510		ret = call->class->raw_init(call);
2511		if (ret < 0 && ret != -ENOSYS)
2512			pr_warn("Could not initialize trace events/%s\n", name);
2513	}
2514
2515	return ret;
2516}
2517
2518static int
2519__register_event(struct trace_event_call *call, struct module *mod)
2520{
2521	int ret;
2522
2523	ret = event_init(call);
2524	if (ret < 0)
2525		return ret;
2526
2527	list_add(&call->list, &ftrace_events);
2528	call->mod = mod;
2529
2530	return 0;
2531}
2532
2533static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2534{
2535	int rlen;
2536	int elen;
2537
2538	/* Find the length of the eval value as a string */
2539	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2540	/* Make sure there's enough room to replace the string with the value */
2541	if (len < elen)
2542		return NULL;
2543
2544	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2545
2546	/* Get the rest of the string of ptr */
2547	rlen = strlen(ptr + len);
2548	memmove(ptr + elen, ptr + len, rlen);
2549	/* Make sure we end the new string */
2550	ptr[elen + rlen] = 0;
2551
2552	return ptr + elen;
2553}
2554
2555static void update_event_printk(struct trace_event_call *call,
2556				struct trace_eval_map *map)
2557{
2558	char *ptr;
2559	int quote = 0;
2560	int len = strlen(map->eval_string);
2561
2562	for (ptr = call->print_fmt; *ptr; ptr++) {
2563		if (*ptr == '\\') {
2564			ptr++;
2565			/* paranoid */
2566			if (!*ptr)
2567				break;
2568			continue;
2569		}
2570		if (*ptr == '"') {
2571			quote ^= 1;
2572			continue;
2573		}
2574		if (quote)
2575			continue;
2576		if (isdigit(*ptr)) {
2577			/* skip numbers */
2578			do {
2579				ptr++;
2580				/* Check for alpha chars like ULL */
2581			} while (isalnum(*ptr));
2582			if (!*ptr)
2583				break;
2584			/*
2585			 * A number must have some kind of delimiter after
2586			 * it, and we can ignore that too.
2587			 */
2588			continue;
2589		}
2590		if (isalpha(*ptr) || *ptr == '_') {
2591			if (strncmp(map->eval_string, ptr, len) == 0 &&
2592			    !isalnum(ptr[len]) && ptr[len] != '_') {
2593				ptr = eval_replace(ptr, map, len);
2594				/* enum/sizeof string smaller than value */
2595				if (WARN_ON_ONCE(!ptr))
2596					return;
2597				/*
2598				 * No need to decrement here, as eval_replace()
2599				 * returns the pointer to the character passed
2600				 * the eval, and two evals can not be placed
2601				 * back to back without something in between.
2602				 * We can skip that something in between.
2603				 */
2604				continue;
2605			}
2606		skip_more:
2607			do {
2608				ptr++;
2609			} while (isalnum(*ptr) || *ptr == '_');
2610			if (!*ptr)
2611				break;
2612			/*
2613			 * If what comes after this variable is a '.' or
2614			 * '->' then we can continue to ignore that string.
2615			 */
2616			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2617				ptr += *ptr == '.' ? 1 : 2;
2618				if (!*ptr)
2619					break;
2620				goto skip_more;
2621			}
2622			/*
2623			 * Once again, we can skip the delimiter that came
2624			 * after the string.
2625			 */
2626			continue;
2627		}
2628	}
2629}
2630
2631void trace_event_eval_update(struct trace_eval_map **map, int len)
2632{
2633	struct trace_event_call *call, *p;
2634	const char *last_system = NULL;
2635	bool first = false;
2636	int last_i;
2637	int i;
2638
2639	down_write(&trace_event_sem);
2640	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2641		/* events are usually grouped together with systems */
2642		if (!last_system || call->class->system != last_system) {
2643			first = true;
2644			last_i = 0;
2645			last_system = call->class->system;
2646		}
2647
2648		/*
2649		 * Since calls are grouped by systems, the likelihood that the
2650		 * next call in the iteration belongs to the same system as the
2651		 * previous call is high. As an optimization, we skip searching
2652		 * for a map[] that matches the call's system if the last call
2653		 * was from the same system. That's what last_i is for. If the
2654		 * call has the same system as the previous call, then last_i
2655		 * will be the index of the first map[] that has a matching
2656		 * system.
2657		 */
2658		for (i = last_i; i < len; i++) {
2659			if (call->class->system == map[i]->system) {
2660				/* Save the first system if need be */
2661				if (first) {
2662					last_i = i;
2663					first = false;
2664				}
2665				update_event_printk(call, map[i]);
2666			}
2667		}
2668	}
2669	up_write(&trace_event_sem);
2670}
2671
2672static struct trace_event_file *
2673trace_create_new_event(struct trace_event_call *call,
2674		       struct trace_array *tr)
2675{
2676	struct trace_event_file *file;
2677
2678	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2679	if (!file)
2680		return NULL;
2681
2682	file->event_call = call;
2683	file->tr = tr;
2684	atomic_set(&file->sm_ref, 0);
2685	atomic_set(&file->tm_ref, 0);
2686	INIT_LIST_HEAD(&file->triggers);
2687	list_add(&file->list, &tr->events);
2688
2689	return file;
2690}
2691
2692/* Add an event to a trace directory */
2693static int
2694__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2695{
2696	struct trace_event_file *file;
2697
2698	file = trace_create_new_event(call, tr);
2699	if (!file)
2700		return -ENOMEM;
2701
2702	if (eventdir_initialized)
2703		return event_create_dir(tr->event_dir, file);
2704	else
2705		return event_define_fields(call);
2706}
2707
2708/*
2709 * Just create a descriptor for early init. A descriptor is required
2710 * for enabling events at boot. We want to enable events before
2711 * the filesystem is initialized.
2712 */
2713static int
2714__trace_early_add_new_event(struct trace_event_call *call,
2715			    struct trace_array *tr)
2716{
2717	struct trace_event_file *file;
2718
2719	file = trace_create_new_event(call, tr);
2720	if (!file)
2721		return -ENOMEM;
2722
2723	return event_define_fields(call);
2724}
2725
2726struct ftrace_module_file_ops;
2727static void __add_event_to_tracers(struct trace_event_call *call);
2728
2729/* Add an additional event_call dynamically */
2730int trace_add_event_call(struct trace_event_call *call)
2731{
2732	int ret;
2733	lockdep_assert_held(&event_mutex);
2734
2735	mutex_lock(&trace_types_lock);
 
2736
2737	ret = __register_event(call, NULL);
2738	if (ret >= 0)
2739		__add_event_to_tracers(call);
2740
 
2741	mutex_unlock(&trace_types_lock);
2742	return ret;
2743}
2744
2745/*
2746 * Must be called under locking of trace_types_lock, event_mutex and
2747 * trace_event_sem.
2748 */
2749static void __trace_remove_event_call(struct trace_event_call *call)
2750{
2751	event_remove(call);
2752	trace_destroy_fields(call);
2753	free_event_filter(call->filter);
2754	call->filter = NULL;
2755}
2756
2757static int probe_remove_event_call(struct trace_event_call *call)
2758{
2759	struct trace_array *tr;
2760	struct trace_event_file *file;
2761
2762#ifdef CONFIG_PERF_EVENTS
2763	if (call->perf_refcount)
2764		return -EBUSY;
2765#endif
2766	do_for_each_event_file(tr, file) {
2767		if (file->event_call != call)
2768			continue;
2769		/*
2770		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2771		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2772		 * TRACE_REG_UNREGISTER.
2773		 */
2774		if (file->flags & EVENT_FILE_FL_ENABLED)
2775			return -EBUSY;
2776		/*
2777		 * The do_for_each_event_file_safe() is
2778		 * a double loop. After finding the call for this
2779		 * trace_array, we use break to jump to the next
2780		 * trace_array.
2781		 */
2782		break;
2783	} while_for_each_event_file();
2784
2785	__trace_remove_event_call(call);
2786
2787	return 0;
2788}
2789
2790/* Remove an event_call */
2791int trace_remove_event_call(struct trace_event_call *call)
2792{
2793	int ret;
2794
2795	lockdep_assert_held(&event_mutex);
2796
2797	mutex_lock(&trace_types_lock);
 
2798	down_write(&trace_event_sem);
2799	ret = probe_remove_event_call(call);
2800	up_write(&trace_event_sem);
 
2801	mutex_unlock(&trace_types_lock);
2802
2803	return ret;
2804}
2805
2806#define for_each_event(event, start, end)			\
2807	for (event = start;					\
2808	     (unsigned long)event < (unsigned long)end;		\
2809	     event++)
2810
2811#ifdef CONFIG_MODULES
2812
2813static void trace_module_add_events(struct module *mod)
2814{
2815	struct trace_event_call **call, **start, **end;
2816
2817	if (!mod->num_trace_events)
2818		return;
2819
2820	/* Don't add infrastructure for mods without tracepoints */
2821	if (trace_module_has_bad_taint(mod)) {
2822		pr_err("%s: module has bad taint, not creating trace events\n",
2823		       mod->name);
2824		return;
2825	}
2826
2827	start = mod->trace_events;
2828	end = mod->trace_events + mod->num_trace_events;
2829
2830	for_each_event(call, start, end) {
2831		__register_event(*call, mod);
2832		__add_event_to_tracers(*call);
2833	}
2834}
2835
2836static void trace_module_remove_events(struct module *mod)
2837{
2838	struct trace_event_call *call, *p;
 
2839
2840	down_write(&trace_event_sem);
2841	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2842		if (call->mod == mod)
 
 
2843			__trace_remove_event_call(call);
 
2844	}
2845	up_write(&trace_event_sem);
2846
2847	/*
2848	 * It is safest to reset the ring buffer if the module being unloaded
2849	 * registered any events that were used. The only worry is if
2850	 * a new module gets loaded, and takes on the same id as the events
2851	 * of this module. When printing out the buffer, traced events left
2852	 * over from this module may be passed to the new module events and
2853	 * unexpected results may occur.
2854	 */
2855	tracing_reset_all_online_cpus();
 
2856}
2857
2858static int trace_module_notify(struct notifier_block *self,
2859			       unsigned long val, void *data)
2860{
2861	struct module *mod = data;
2862
 
2863	mutex_lock(&event_mutex);
2864	mutex_lock(&trace_types_lock);
2865	switch (val) {
2866	case MODULE_STATE_COMING:
2867		trace_module_add_events(mod);
2868		break;
2869	case MODULE_STATE_GOING:
2870		trace_module_remove_events(mod);
2871		break;
2872	}
 
2873	mutex_unlock(&trace_types_lock);
2874	mutex_unlock(&event_mutex);
2875
2876	return NOTIFY_OK;
2877}
2878
2879static struct notifier_block trace_module_nb = {
2880	.notifier_call = trace_module_notify,
2881	.priority = 1, /* higher than trace.c module notify */
2882};
2883#endif /* CONFIG_MODULES */
2884
2885/* Create a new event directory structure for a trace directory. */
2886static void
2887__trace_add_event_dirs(struct trace_array *tr)
2888{
2889	struct trace_event_call *call;
2890	int ret;
2891
2892	list_for_each_entry(call, &ftrace_events, list) {
2893		ret = __trace_add_new_event(call, tr);
2894		if (ret < 0)
2895			pr_warn("Could not create directory for event %s\n",
2896				trace_event_name(call));
2897	}
2898}
2899
2900/* Returns any file that matches the system and event */
2901struct trace_event_file *
2902__find_event_file(struct trace_array *tr, const char *system, const char *event)
2903{
2904	struct trace_event_file *file;
2905	struct trace_event_call *call;
2906	const char *name;
2907
2908	list_for_each_entry(file, &tr->events, list) {
2909
2910		call = file->event_call;
2911		name = trace_event_name(call);
2912
2913		if (!name || !call->class)
 
 
 
2914			continue;
2915
2916		if (strcmp(event, name) == 0 &&
2917		    strcmp(system, call->class->system) == 0)
2918			return file;
2919	}
2920	return NULL;
2921}
2922
2923/* Returns valid trace event files that match system and event */
2924struct trace_event_file *
2925find_event_file(struct trace_array *tr, const char *system, const char *event)
2926{
2927	struct trace_event_file *file;
2928
2929	file = __find_event_file(tr, system, event);
2930	if (!file || !file->event_call->class->reg ||
2931	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2932		return NULL;
2933
2934	return file;
2935}
2936
2937/**
2938 * trace_get_event_file - Find and return a trace event file
2939 * @instance: The name of the trace instance containing the event
2940 * @system: The name of the system containing the event
2941 * @event: The name of the event
2942 *
2943 * Return a trace event file given the trace instance name, trace
2944 * system, and trace event name.  If the instance name is NULL, it
2945 * refers to the top-level trace array.
2946 *
2947 * This function will look it up and return it if found, after calling
2948 * trace_array_get() to prevent the instance from going away, and
2949 * increment the event's module refcount to prevent it from being
2950 * removed.
2951 *
2952 * To release the file, call trace_put_event_file(), which will call
2953 * trace_array_put() and decrement the event's module refcount.
2954 *
2955 * Return: The trace event on success, ERR_PTR otherwise.
2956 */
2957struct trace_event_file *trace_get_event_file(const char *instance,
2958					      const char *system,
2959					      const char *event)
2960{
2961	struct trace_array *tr = top_trace_array();
2962	struct trace_event_file *file = NULL;
2963	int ret = -EINVAL;
2964
2965	if (instance) {
2966		tr = trace_array_find_get(instance);
2967		if (!tr)
2968			return ERR_PTR(-ENOENT);
2969	} else {
2970		ret = trace_array_get(tr);
2971		if (ret)
2972			return ERR_PTR(ret);
2973	}
2974
2975	mutex_lock(&event_mutex);
2976
2977	file = find_event_file(tr, system, event);
2978	if (!file) {
2979		trace_array_put(tr);
2980		ret = -EINVAL;
2981		goto out;
2982	}
2983
2984	/* Don't let event modules unload while in use */
2985	ret = try_module_get(file->event_call->mod);
2986	if (!ret) {
2987		trace_array_put(tr);
2988		ret = -EBUSY;
2989		goto out;
2990	}
2991
2992	ret = 0;
2993 out:
2994	mutex_unlock(&event_mutex);
2995
2996	if (ret)
2997		file = ERR_PTR(ret);
2998
2999	return file;
3000}
3001EXPORT_SYMBOL_GPL(trace_get_event_file);
3002
3003/**
3004 * trace_put_event_file - Release a file from trace_get_event_file()
3005 * @file: The trace event file
3006 *
3007 * If a file was retrieved using trace_get_event_file(), this should
3008 * be called when it's no longer needed.  It will cancel the previous
3009 * trace_array_get() called by that function, and decrement the
3010 * event's module refcount.
3011 */
3012void trace_put_event_file(struct trace_event_file *file)
3013{
3014	mutex_lock(&event_mutex);
3015	module_put(file->event_call->mod);
3016	mutex_unlock(&event_mutex);
3017
3018	trace_array_put(file->tr);
3019}
3020EXPORT_SYMBOL_GPL(trace_put_event_file);
3021
3022#ifdef CONFIG_DYNAMIC_FTRACE
3023
3024/* Avoid typos */
3025#define ENABLE_EVENT_STR	"enable_event"
3026#define DISABLE_EVENT_STR	"disable_event"
3027
3028struct event_probe_data {
3029	struct trace_event_file	*file;
3030	unsigned long			count;
3031	int				ref;
3032	bool				enable;
3033};
3034
3035static void update_event_probe(struct event_probe_data *data)
 
3036{
 
 
 
 
 
 
3037	if (data->enable)
3038		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3039	else
3040		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3041}
3042
3043static void
3044event_enable_probe(unsigned long ip, unsigned long parent_ip,
3045		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3046		   void *data)
3047{
3048	struct ftrace_func_mapper *mapper = data;
3049	struct event_probe_data *edata;
3050	void **pdata;
3051
3052	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3053	if (!pdata || !*pdata)
3054		return;
3055
3056	edata = *pdata;
3057	update_event_probe(edata);
3058}
3059
3060static void
3061event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3062			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3063			 void *data)
3064{
3065	struct ftrace_func_mapper *mapper = data;
3066	struct event_probe_data *edata;
3067	void **pdata;
3068
3069	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3070	if (!pdata || !*pdata)
3071		return;
3072
3073	edata = *pdata;
3074
3075	if (!edata->count)
3076		return;
3077
3078	/* Skip if the event is in a state we want to switch to */
3079	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3080		return;
3081
3082	if (edata->count != -1)
3083		(edata->count)--;
3084
3085	update_event_probe(edata);
3086}
3087
3088static int
3089event_enable_print(struct seq_file *m, unsigned long ip,
3090		   struct ftrace_probe_ops *ops, void *data)
3091{
3092	struct ftrace_func_mapper *mapper = data;
3093	struct event_probe_data *edata;
3094	void **pdata;
3095
3096	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3097
3098	if (WARN_ON_ONCE(!pdata || !*pdata))
3099		return 0;
3100
3101	edata = *pdata;
3102
3103	seq_printf(m, "%ps:", (void *)ip);
3104
3105	seq_printf(m, "%s:%s:%s",
3106		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3107		   edata->file->event_call->class->system,
3108		   trace_event_name(edata->file->event_call));
3109
3110	if (edata->count == -1)
3111		seq_puts(m, ":unlimited\n");
3112	else
3113		seq_printf(m, ":count=%ld\n", edata->count);
3114
3115	return 0;
3116}
3117
3118static int
3119event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3120		  unsigned long ip, void *init_data, void **data)
3121{
3122	struct ftrace_func_mapper *mapper = *data;
3123	struct event_probe_data *edata = init_data;
3124	int ret;
3125
3126	if (!mapper) {
3127		mapper = allocate_ftrace_func_mapper();
3128		if (!mapper)
3129			return -ENODEV;
3130		*data = mapper;
3131	}
3132
3133	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3134	if (ret < 0)
3135		return ret;
3136
3137	edata->ref++;
3138
3139	return 0;
3140}
3141
3142static int free_probe_data(void *data)
3143{
3144	struct event_probe_data *edata = data;
3145
3146	edata->ref--;
3147	if (!edata->ref) {
3148		/* Remove the SOFT_MODE flag */
3149		__ftrace_event_enable_disable(edata->file, 0, 1);
3150		module_put(edata->file->event_call->mod);
3151		kfree(edata);
3152	}
3153	return 0;
3154}
3155
3156static void
3157event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3158		  unsigned long ip, void *data)
3159{
3160	struct ftrace_func_mapper *mapper = data;
3161	struct event_probe_data *edata;
3162
3163	if (!ip) {
3164		if (!mapper)
3165			return;
3166		free_ftrace_func_mapper(mapper, free_probe_data);
3167		return;
 
 
 
 
 
 
 
3168	}
3169
3170	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3171
3172	if (WARN_ON_ONCE(!edata))
3173		return;
3174
3175	if (WARN_ON_ONCE(edata->ref <= 0))
3176		return;
3177
3178	free_probe_data(edata);
3179}
3180
3181static struct ftrace_probe_ops event_enable_probe_ops = {
3182	.func			= event_enable_probe,
3183	.print			= event_enable_print,
3184	.init			= event_enable_init,
3185	.free			= event_enable_free,
3186};
3187
3188static struct ftrace_probe_ops event_enable_count_probe_ops = {
3189	.func			= event_enable_count_probe,
3190	.print			= event_enable_print,
3191	.init			= event_enable_init,
3192	.free			= event_enable_free,
3193};
3194
3195static struct ftrace_probe_ops event_disable_probe_ops = {
3196	.func			= event_enable_probe,
3197	.print			= event_enable_print,
3198	.init			= event_enable_init,
3199	.free			= event_enable_free,
3200};
3201
3202static struct ftrace_probe_ops event_disable_count_probe_ops = {
3203	.func			= event_enable_count_probe,
3204	.print			= event_enable_print,
3205	.init			= event_enable_init,
3206	.free			= event_enable_free,
3207};
3208
3209static int
3210event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3211		  char *glob, char *cmd, char *param, int enabled)
3212{
 
3213	struct trace_event_file *file;
3214	struct ftrace_probe_ops *ops;
3215	struct event_probe_data *data;
3216	const char *system;
3217	const char *event;
3218	char *number;
3219	bool enable;
3220	int ret;
3221
3222	if (!tr)
3223		return -ENODEV;
3224
3225	/* hash funcs only work with set_ftrace_filter */
3226	if (!enabled || !param)
3227		return -EINVAL;
3228
3229	system = strsep(&param, ":");
3230	if (!param)
3231		return -EINVAL;
3232
3233	event = strsep(&param, ":");
3234
3235	mutex_lock(&event_mutex);
3236
3237	ret = -EINVAL;
3238	file = find_event_file(tr, system, event);
3239	if (!file)
3240		goto out;
3241
3242	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3243
3244	if (enable)
3245		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3246	else
3247		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3248
3249	if (glob[0] == '!') {
3250		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
 
3251		goto out;
3252	}
3253
3254	ret = -ENOMEM;
3255
3256	data = kzalloc(sizeof(*data), GFP_KERNEL);
3257	if (!data)
3258		goto out;
3259
3260	data->enable = enable;
3261	data->count = -1;
3262	data->file = file;
3263
3264	if (!param)
3265		goto out_reg;
3266
3267	number = strsep(&param, ":");
3268
3269	ret = -EINVAL;
3270	if (!strlen(number))
3271		goto out_free;
3272
3273	/*
3274	 * We use the callback data field (which is a pointer)
3275	 * as our counter.
3276	 */
3277	ret = kstrtoul(number, 0, &data->count);
3278	if (ret)
3279		goto out_free;
3280
3281 out_reg:
3282	/* Don't let event modules unload while probe registered */
3283	ret = try_module_get(file->event_call->mod);
3284	if (!ret) {
3285		ret = -EBUSY;
3286		goto out_free;
3287	}
3288
3289	ret = __ftrace_event_enable_disable(file, 1, 1);
3290	if (ret < 0)
3291		goto out_put;
3292
3293	ret = register_ftrace_function_probe(glob, tr, ops, data);
3294	/*
3295	 * The above returns on success the # of functions enabled,
3296	 * but if it didn't find any functions it returns zero.
3297	 * Consider no functions a failure too.
3298	 */
3299	if (!ret) {
3300		ret = -ENOENT;
3301		goto out_disable;
3302	} else if (ret < 0)
3303		goto out_disable;
3304	/* Just return zero, not the number of enabled functions */
3305	ret = 0;
3306 out:
3307	mutex_unlock(&event_mutex);
3308	return ret;
3309
3310 out_disable:
3311	__ftrace_event_enable_disable(file, 0, 1);
3312 out_put:
3313	module_put(file->event_call->mod);
3314 out_free:
3315	kfree(data);
3316	goto out;
3317}
3318
3319static struct ftrace_func_command event_enable_cmd = {
3320	.name			= ENABLE_EVENT_STR,
3321	.func			= event_enable_func,
3322};
3323
3324static struct ftrace_func_command event_disable_cmd = {
3325	.name			= DISABLE_EVENT_STR,
3326	.func			= event_enable_func,
3327};
3328
3329static __init int register_event_cmds(void)
3330{
3331	int ret;
3332
3333	ret = register_ftrace_command(&event_enable_cmd);
3334	if (WARN_ON(ret < 0))
3335		return ret;
3336	ret = register_ftrace_command(&event_disable_cmd);
3337	if (WARN_ON(ret < 0))
3338		unregister_ftrace_command(&event_enable_cmd);
3339	return ret;
3340}
3341#else
3342static inline int register_event_cmds(void) { return 0; }
3343#endif /* CONFIG_DYNAMIC_FTRACE */
3344
3345/*
3346 * The top level array and trace arrays created by boot-time tracing
3347 * have already had its trace_event_file descriptors created in order
3348 * to allow for early events to be recorded.
3349 * This function is called after the tracefs has been initialized,
3350 * and we now have to create the files associated to the events.
3351 */
3352static void __trace_early_add_event_dirs(struct trace_array *tr)
 
3353{
3354	struct trace_event_file *file;
3355	int ret;
3356
3357
3358	list_for_each_entry(file, &tr->events, list) {
3359		ret = event_create_dir(tr->event_dir, file);
3360		if (ret < 0)
3361			pr_warn("Could not create directory for event %s\n",
3362				trace_event_name(file->event_call));
3363	}
3364}
3365
3366/*
3367 * For early boot up, the top trace array and the trace arrays created
3368 * by boot-time tracing require to have a list of events that can be
3369 * enabled. This must be done before the filesystem is set up in order
3370 * to allow events to be traced early.
3371 */
3372void __trace_early_add_events(struct trace_array *tr)
 
3373{
3374	struct trace_event_call *call;
3375	int ret;
3376
3377	list_for_each_entry(call, &ftrace_events, list) {
3378		/* Early boot up should not have any modules loaded */
3379		if (WARN_ON_ONCE(call->mod))
3380			continue;
3381
3382		ret = __trace_early_add_new_event(call, tr);
3383		if (ret < 0)
3384			pr_warn("Could not create early event %s\n",
3385				trace_event_name(call));
3386	}
3387}
3388
3389/* Remove the event directory structure for a trace directory. */
3390static void
3391__trace_remove_event_dirs(struct trace_array *tr)
3392{
3393	struct trace_event_file *file, *next;
3394
3395	list_for_each_entry_safe(file, next, &tr->events, list)
3396		remove_event_file_dir(file);
3397}
3398
3399static void __add_event_to_tracers(struct trace_event_call *call)
3400{
3401	struct trace_array *tr;
3402
3403	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3404		__trace_add_new_event(call, tr);
3405}
3406
3407extern struct trace_event_call *__start_ftrace_events[];
3408extern struct trace_event_call *__stop_ftrace_events[];
3409
3410static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3411
3412static __init int setup_trace_event(char *str)
3413{
3414	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3415	ring_buffer_expanded = true;
3416	disable_tracing_selftest("running event tracing");
3417
3418	return 1;
3419}
3420__setup("trace_event=", setup_trace_event);
3421
3422/* Expects to have event_mutex held when called */
3423static int
3424create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3425{
3426	struct dentry *d_events;
3427	struct dentry *entry;
3428
3429	entry = tracefs_create_file("set_event", 0644, parent,
3430				    tr, &ftrace_set_event_fops);
3431	if (!entry) {
3432		pr_warn("Could not create tracefs 'set_event' entry\n");
3433		return -ENOMEM;
3434	}
3435
3436	d_events = tracefs_create_dir("events", parent);
3437	if (!d_events) {
3438		pr_warn("Could not create tracefs 'events' directory\n");
3439		return -ENOMEM;
3440	}
3441
3442	entry = trace_create_file("enable", 0644, d_events,
3443				  tr, &ftrace_tr_enable_fops);
3444	if (!entry) {
3445		pr_warn("Could not create tracefs 'enable' entry\n");
3446		return -ENOMEM;
3447	}
3448
3449	/* There are not as crucial, just warn if they are not created */
3450
3451	entry = tracefs_create_file("set_event_pid", 0644, parent,
3452				    tr, &ftrace_set_event_pid_fops);
3453	if (!entry)
3454		pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3455
3456	entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3457				    tr, &ftrace_set_event_notrace_pid_fops);
3458	if (!entry)
3459		pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3460
3461	/* ring buffer internal formats */
3462	entry = trace_create_file("header_page", 0444, d_events,
3463				  ring_buffer_print_page_header,
3464				  &ftrace_show_header_fops);
3465	if (!entry)
3466		pr_warn("Could not create tracefs 'header_page' entry\n");
3467
3468	entry = trace_create_file("header_event", 0444, d_events,
3469				  ring_buffer_print_entry_header,
3470				  &ftrace_show_header_fops);
3471	if (!entry)
3472		pr_warn("Could not create tracefs 'header_event' entry\n");
3473
3474	tr->event_dir = d_events;
3475
3476	return 0;
3477}
3478
3479/**
3480 * event_trace_add_tracer - add a instance of a trace_array to events
3481 * @parent: The parent dentry to place the files/directories for events in
3482 * @tr: The trace array associated with these events
3483 *
3484 * When a new instance is created, it needs to set up its events
3485 * directory, as well as other files associated with events. It also
3486 * creates the event hierarchy in the @parent/events directory.
3487 *
3488 * Returns 0 on success.
3489 *
3490 * Must be called with event_mutex held.
3491 */
3492int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3493{
3494	int ret;
3495
3496	lockdep_assert_held(&event_mutex);
3497
3498	ret = create_event_toplevel_files(parent, tr);
3499	if (ret)
3500		goto out;
3501
3502	down_write(&trace_event_sem);
3503	/* If tr already has the event list, it is initialized in early boot. */
3504	if (unlikely(!list_empty(&tr->events)))
3505		__trace_early_add_event_dirs(tr);
3506	else
3507		__trace_add_event_dirs(tr);
3508	up_write(&trace_event_sem);
3509
3510 out:
 
 
3511	return ret;
3512}
3513
3514/*
3515 * The top trace array already had its file descriptors created.
3516 * Now the files themselves need to be created.
3517 */
3518static __init int
3519early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3520{
3521	int ret;
3522
3523	mutex_lock(&event_mutex);
3524
3525	ret = create_event_toplevel_files(parent, tr);
3526	if (ret)
3527		goto out_unlock;
3528
3529	down_write(&trace_event_sem);
3530	__trace_early_add_event_dirs(tr);
3531	up_write(&trace_event_sem);
3532
3533 out_unlock:
3534	mutex_unlock(&event_mutex);
3535
3536	return ret;
3537}
3538
3539/* Must be called with event_mutex held */
3540int event_trace_del_tracer(struct trace_array *tr)
3541{
3542	lockdep_assert_held(&event_mutex);
3543
3544	/* Disable any event triggers and associated soft-disabled events */
3545	clear_event_triggers(tr);
3546
3547	/* Clear the pid list */
3548	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3549
3550	/* Disable any running events */
3551	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3552
3553	/* Make sure no more events are being executed */
3554	tracepoint_synchronize_unregister();
3555
3556	down_write(&trace_event_sem);
3557	__trace_remove_event_dirs(tr);
3558	tracefs_remove(tr->event_dir);
3559	up_write(&trace_event_sem);
3560
3561	tr->event_dir = NULL;
3562
 
 
3563	return 0;
3564}
3565
3566static __init int event_trace_memsetup(void)
3567{
3568	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3569	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3570	return 0;
3571}
3572
3573static __init void
3574early_enable_events(struct trace_array *tr, bool disable_first)
3575{
3576	char *buf = bootup_event_buf;
3577	char *token;
3578	int ret;
3579
3580	while (true) {
3581		token = strsep(&buf, ",");
3582
3583		if (!token)
3584			break;
3585
3586		if (*token) {
3587			/* Restarting syscalls requires that we stop them first */
3588			if (disable_first)
3589				ftrace_set_clr_event(tr, token, 0);
3590
3591			ret = ftrace_set_clr_event(tr, token, 1);
3592			if (ret)
3593				pr_warn("Failed to enable trace event: %s\n", token);
3594		}
3595
3596		/* Put back the comma to allow this to be called again */
3597		if (buf)
3598			*(buf - 1) = ',';
3599	}
3600}
3601
3602static __init int event_trace_enable(void)
3603{
3604	struct trace_array *tr = top_trace_array();
3605	struct trace_event_call **iter, *call;
3606	int ret;
3607
3608	if (!tr)
3609		return -ENODEV;
3610
3611	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3612
3613		call = *iter;
3614		ret = event_init(call);
3615		if (!ret)
3616			list_add(&call->list, &ftrace_events);
3617	}
3618
3619	/*
3620	 * We need the top trace array to have a working set of trace
3621	 * points at early init, before the debug files and directories
3622	 * are created. Create the file entries now, and attach them
3623	 * to the actual file dentries later.
3624	 */
3625	__trace_early_add_events(tr);
3626
3627	early_enable_events(tr, false);
3628
3629	trace_printk_start_comm();
3630
3631	register_event_cmds();
3632
3633	register_trigger_cmds();
3634
3635	return 0;
3636}
3637
3638/*
3639 * event_trace_enable() is called from trace_event_init() first to
3640 * initialize events and perhaps start any events that are on the
3641 * command line. Unfortunately, there are some events that will not
3642 * start this early, like the system call tracepoints that need
3643 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3644 * event_trace_enable() is called before pid 1 starts, and this flag
3645 * is never set, making the syscall tracepoint never get reached, but
3646 * the event is enabled regardless (and not doing anything).
3647 */
3648static __init int event_trace_enable_again(void)
3649{
3650	struct trace_array *tr;
3651
3652	tr = top_trace_array();
3653	if (!tr)
3654		return -ENODEV;
3655
3656	early_enable_events(tr, true);
3657
3658	return 0;
3659}
3660
3661early_initcall(event_trace_enable_again);
3662
3663/* Init fields which doesn't related to the tracefs */
3664static __init int event_trace_init_fields(void)
3665{
3666	if (trace_define_generic_fields())
3667		pr_warn("tracing: Failed to allocated generic fields");
3668
3669	if (trace_define_common_fields())
3670		pr_warn("tracing: Failed to allocate common fields");
3671
3672	return 0;
3673}
3674
3675__init int event_trace_init(void)
3676{
3677	struct trace_array *tr;
 
3678	struct dentry *entry;
3679	int ret;
3680
3681	tr = top_trace_array();
3682	if (!tr)
3683		return -ENODEV;
3684
3685	entry = tracefs_create_file("available_events", 0444, NULL,
 
 
 
 
3686				    tr, &ftrace_avail_fops);
3687	if (!entry)
3688		pr_warn("Could not create tracefs 'available_events' entry\n");
3689
3690	ret = early_event_add_tracer(NULL, tr);
 
 
 
 
 
 
3691	if (ret)
3692		return ret;
3693
3694#ifdef CONFIG_MODULES
3695	ret = register_module_notifier(&trace_module_nb);
3696	if (ret)
3697		pr_warn("Failed to register trace events module notifier\n");
3698#endif
3699
3700	eventdir_initialized = true;
3701
3702	return 0;
3703}
3704
3705void __init trace_event_init(void)
3706{
3707	event_trace_memsetup();
3708	init_ftrace_syscalls();
3709	event_trace_enable();
3710	event_trace_init_fields();
3711}
3712
3713#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
 
 
3714
3715static DEFINE_SPINLOCK(test_spinlock);
3716static DEFINE_SPINLOCK(test_spinlock_irq);
3717static DEFINE_MUTEX(test_mutex);
3718
3719static __init void test_work(struct work_struct *dummy)
3720{
3721	spin_lock(&test_spinlock);
3722	spin_lock_irq(&test_spinlock_irq);
3723	udelay(1);
3724	spin_unlock_irq(&test_spinlock_irq);
3725	spin_unlock(&test_spinlock);
3726
3727	mutex_lock(&test_mutex);
3728	msleep(1);
3729	mutex_unlock(&test_mutex);
3730}
3731
3732static __init int event_test_thread(void *unused)
3733{
3734	void *test_malloc;
3735
3736	test_malloc = kmalloc(1234, GFP_KERNEL);
3737	if (!test_malloc)
3738		pr_info("failed to kmalloc\n");
3739
3740	schedule_on_each_cpu(test_work);
3741
3742	kfree(test_malloc);
3743
3744	set_current_state(TASK_INTERRUPTIBLE);
3745	while (!kthread_should_stop()) {
3746		schedule();
3747		set_current_state(TASK_INTERRUPTIBLE);
3748	}
3749	__set_current_state(TASK_RUNNING);
3750
3751	return 0;
3752}
3753
3754/*
3755 * Do various things that may trigger events.
3756 */
3757static __init void event_test_stuff(void)
3758{
3759	struct task_struct *test_thread;
3760
3761	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3762	msleep(1);
3763	kthread_stop(test_thread);
3764}
3765
3766/*
3767 * For every trace event defined, we will test each trace point separately,
3768 * and then by groups, and finally all trace points.
3769 */
3770static __init void event_trace_self_tests(void)
3771{
3772	struct trace_subsystem_dir *dir;
3773	struct trace_event_file *file;
3774	struct trace_event_call *call;
3775	struct event_subsystem *system;
3776	struct trace_array *tr;
3777	int ret;
3778
3779	tr = top_trace_array();
3780	if (!tr)
3781		return;
3782
3783	pr_info("Running tests on trace events:\n");
3784
3785	list_for_each_entry(file, &tr->events, list) {
3786
3787		call = file->event_call;
3788
3789		/* Only test those that have a probe */
3790		if (!call->class || !call->class->probe)
3791			continue;
3792
3793/*
3794 * Testing syscall events here is pretty useless, but
3795 * we still do it if configured. But this is time consuming.
3796 * What we really need is a user thread to perform the
3797 * syscalls as we test.
3798 */
3799#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3800		if (call->class->system &&
3801		    strcmp(call->class->system, "syscalls") == 0)
3802			continue;
3803#endif
3804
3805		pr_info("Testing event %s: ", trace_event_name(call));
3806
3807		/*
3808		 * If an event is already enabled, someone is using
3809		 * it and the self test should not be on.
3810		 */
3811		if (file->flags & EVENT_FILE_FL_ENABLED) {
3812			pr_warn("Enabled event during self test!\n");
3813			WARN_ON_ONCE(1);
3814			continue;
3815		}
3816
3817		ftrace_event_enable_disable(file, 1);
3818		event_test_stuff();
3819		ftrace_event_enable_disable(file, 0);
3820
3821		pr_cont("OK\n");
3822	}
3823
3824	/* Now test at the sub system level */
3825
3826	pr_info("Running tests on trace event systems:\n");
3827
3828	list_for_each_entry(dir, &tr->systems, list) {
3829
3830		system = dir->subsystem;
3831
3832		/* the ftrace system is special, skip it */
3833		if (strcmp(system->name, "ftrace") == 0)
3834			continue;
3835
3836		pr_info("Testing event system %s: ", system->name);
3837
3838		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3839		if (WARN_ON_ONCE(ret)) {
3840			pr_warn("error enabling system %s\n",
3841				system->name);
3842			continue;
3843		}
3844
3845		event_test_stuff();
3846
3847		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3848		if (WARN_ON_ONCE(ret)) {
3849			pr_warn("error disabling system %s\n",
3850				system->name);
3851			continue;
3852		}
3853
3854		pr_cont("OK\n");
3855	}
3856
3857	/* Test with all events enabled */
3858
3859	pr_info("Running tests on all trace events:\n");
3860	pr_info("Testing all events: ");
3861
3862	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3863	if (WARN_ON_ONCE(ret)) {
3864		pr_warn("error enabling all events\n");
3865		return;
3866	}
3867
3868	event_test_stuff();
3869
3870	/* reset sysname */
3871	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3872	if (WARN_ON_ONCE(ret)) {
3873		pr_warn("error disabling all events\n");
3874		return;
3875	}
3876
3877	pr_cont("OK\n");
3878}
3879
3880#ifdef CONFIG_FUNCTION_TRACER
3881
3882static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3883
3884static struct trace_event_file event_trace_file __initdata;
3885
3886static void __init
3887function_test_events_call(unsigned long ip, unsigned long parent_ip,
3888			  struct ftrace_ops *op, struct ftrace_regs *regs)
3889{
3890	struct trace_buffer *buffer;
3891	struct ring_buffer_event *event;
 
3892	struct ftrace_entry *entry;
3893	unsigned int trace_ctx;
3894	long disabled;
3895	int cpu;
 
3896
3897	trace_ctx = tracing_gen_ctx();
3898	preempt_disable_notrace();
3899	cpu = raw_smp_processor_id();
3900	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3901
3902	if (disabled != 1)
3903		goto out;
3904
 
 
3905	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3906						TRACE_FN, sizeof(*entry),
3907						trace_ctx);
3908	if (!event)
3909		goto out;
3910	entry	= ring_buffer_event_data(event);
3911	entry->ip			= ip;
3912	entry->parent_ip		= parent_ip;
3913
3914	event_trigger_unlock_commit(&event_trace_file, buffer, event,
3915				    entry, trace_ctx);
3916 out:
3917	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3918	preempt_enable_notrace();
3919}
3920
3921static struct ftrace_ops trace_ops __initdata  =
3922{
3923	.func = function_test_events_call,
 
3924};
3925
3926static __init void event_trace_self_test_with_function(void)
3927{
3928	int ret;
3929
3930	event_trace_file.tr = top_trace_array();
3931	if (WARN_ON(!event_trace_file.tr))
3932		return;
3933
3934	ret = register_ftrace_function(&trace_ops);
3935	if (WARN_ON(ret < 0)) {
3936		pr_info("Failed to enable function tracer for event tests\n");
3937		return;
3938	}
3939	pr_info("Running tests again, along with the function tracer\n");
3940	event_trace_self_tests();
3941	unregister_ftrace_function(&trace_ops);
3942}
3943#else
3944static __init void event_trace_self_test_with_function(void)
3945{
3946}
3947#endif
3948
3949static __init int event_trace_self_tests_init(void)
3950{
3951	if (!tracing_selftest_disabled) {
3952		event_trace_self_tests();
3953		event_trace_self_test_with_function();
3954	}
3955
3956	return 0;
3957}
3958
3959late_initcall(event_trace_self_tests_init);
3960
3961#endif