Linux Audio

Check our new training course

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