Linux Audio

Check our new training course

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