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
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * event tracer
   4 *
   5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6 *
   7 *  - Added format output of fields of the trace point.
   8 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9 *
  10 */
  11
  12#define pr_fmt(fmt) fmt
  13
  14#include <linux/workqueue.h>
  15#include <linux/security.h>
  16#include <linux/spinlock.h>
  17#include <linux/kthread.h>
  18#include <linux/tracefs.h>
  19#include <linux/uaccess.h>
 
  20#include <linux/module.h>
  21#include <linux/ctype.h>
  22#include <linux/sort.h>
  23#include <linux/slab.h>
  24#include <linux/delay.h>
  25
  26#include <trace/events/sched.h>
  27#include <trace/syscall.h>
  28
  29#include <asm/setup.h>
  30
  31#include "trace_output.h"
  32
  33#undef TRACE_SYSTEM
  34#define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36DEFINE_MUTEX(event_mutex);
  37
  38LIST_HEAD(ftrace_events);
  39static LIST_HEAD(ftrace_generic_fields);
  40static LIST_HEAD(ftrace_common_fields);
  41static bool eventdir_initialized;
  42
  43static LIST_HEAD(module_strings);
  44
  45struct module_string {
  46	struct list_head	next;
  47	struct module		*module;
  48	char			*str;
  49};
  50
  51#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  52
  53static struct kmem_cache *field_cachep;
  54static struct kmem_cache *file_cachep;
  55
  56static inline int system_refcount(struct event_subsystem *system)
  57{
  58	return system->ref_count;
  59}
  60
  61static int system_refcount_inc(struct event_subsystem *system)
  62{
  63	return system->ref_count++;
  64}
  65
  66static int system_refcount_dec(struct event_subsystem *system)
  67{
  68	return --system->ref_count;
  69}
  70
  71/* Double loops, do not use break, only goto's work */
  72#define do_for_each_event_file(tr, file)			\
  73	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  74		list_for_each_entry(file, &tr->events, list)
  75
  76#define do_for_each_event_file_safe(tr, file)			\
  77	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  78		struct trace_event_file *___n;				\
  79		list_for_each_entry_safe(file, ___n, &tr->events, list)
  80
  81#define while_for_each_event_file()		\
  82	}
  83
 
 
 
 
 
 
 
 
  84static struct ftrace_event_field *
  85__find_event_field(struct list_head *head, char *name)
  86{
  87	struct ftrace_event_field *field;
  88
  89	list_for_each_entry(field, head, link) {
  90		if (!strcmp(field->name, name))
  91			return field;
  92	}
  93
  94	return NULL;
  95}
  96
  97struct ftrace_event_field *
  98trace_find_event_field(struct trace_event_call *call, char *name)
  99{
 100	struct ftrace_event_field *field;
 101	struct list_head *head;
 102
 103	head = trace_get_fields(call);
 104	field = __find_event_field(head, name);
 105	if (field)
 106		return field;
 107
 108	field = __find_event_field(&ftrace_generic_fields, name);
 109	if (field)
 110		return field;
 111
 112	return __find_event_field(&ftrace_common_fields, name);
 113}
 114
 115static int __trace_define_field(struct list_head *head, const char *type,
 116				const char *name, int offset, int size,
 117				int is_signed, int filter_type, int len)
 118{
 119	struct ftrace_event_field *field;
 120
 121	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 122	if (!field)
 123		return -ENOMEM;
 124
 125	field->name = name;
 126	field->type = type;
 127
 128	if (filter_type == FILTER_OTHER)
 129		field->filter_type = filter_assign_type(type);
 130	else
 131		field->filter_type = filter_type;
 132
 133	field->offset = offset;
 134	field->size = size;
 135	field->is_signed = is_signed;
 136	field->len = len;
 137
 138	list_add(&field->link, head);
 139
 140	return 0;
 141}
 142
 143int trace_define_field(struct trace_event_call *call, const char *type,
 144		       const char *name, int offset, int size, int is_signed,
 145		       int filter_type)
 146{
 147	struct list_head *head;
 148
 149	if (WARN_ON(!call->class))
 150		return 0;
 151
 152	head = trace_get_fields(call);
 153	return __trace_define_field(head, type, name, offset, size,
 154				    is_signed, filter_type, 0);
 155}
 156EXPORT_SYMBOL_GPL(trace_define_field);
 157
 158static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 159		       const char *name, int offset, int size, int is_signed,
 160		       int filter_type, int len)
 161{
 162	struct list_head *head;
 163
 164	if (WARN_ON(!call->class))
 165		return 0;
 166
 167	head = trace_get_fields(call);
 168	return __trace_define_field(head, type, name, offset, size,
 169				    is_signed, filter_type, len);
 170}
 171
 172#define __generic_field(type, item, filter_type)			\
 173	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 174				   #item, 0, 0, is_signed_type(type),	\
 175				   filter_type, 0);			\
 176	if (ret)							\
 177		return ret;
 178
 179#define __common_field(type, item)					\
 180	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 181				   "common_" #item,			\
 182				   offsetof(typeof(ent), item),		\
 183				   sizeof(ent.item),			\
 184				   is_signed_type(type), FILTER_OTHER, 0);	\
 185	if (ret)							\
 186		return ret;
 187
 188static int trace_define_generic_fields(void)
 189{
 190	int ret;
 191
 192	__generic_field(int, CPU, FILTER_CPU);
 193	__generic_field(int, cpu, FILTER_CPU);
 194	__generic_field(int, common_cpu, FILTER_CPU);
 195	__generic_field(char *, COMM, FILTER_COMM);
 196	__generic_field(char *, comm, FILTER_COMM);
 197	__generic_field(char *, stacktrace, FILTER_STACKTRACE);
 198	__generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 199
 200	return ret;
 201}
 202
 203static int trace_define_common_fields(void)
 204{
 205	int ret;
 206	struct trace_entry ent;
 207
 208	__common_field(unsigned short, type);
 209	__common_field(unsigned char, flags);
 210	/* Holds both preempt_count and migrate_disable */
 211	__common_field(unsigned char, preempt_count);
 212	__common_field(int, pid);
 213
 214	return ret;
 215}
 216
 217static void trace_destroy_fields(struct trace_event_call *call)
 218{
 219	struct ftrace_event_field *field, *next;
 220	struct list_head *head;
 221
 222	head = trace_get_fields(call);
 223	list_for_each_entry_safe(field, next, head, link) {
 224		list_del(&field->link);
 225		kmem_cache_free(field_cachep, field);
 226	}
 227}
 228
 229/*
 230 * run-time version of trace_event_get_offsets_<call>() that returns the last
 231 * accessible offset of trace fields excluding __dynamic_array bytes
 232 */
 233int trace_event_get_offsets(struct trace_event_call *call)
 234{
 235	struct ftrace_event_field *tail;
 236	struct list_head *head;
 237
 238	head = trace_get_fields(call);
 239	/*
 240	 * head->next points to the last field with the largest offset,
 241	 * since it was added last by trace_define_field()
 242	 */
 243	tail = list_first_entry(head, struct ftrace_event_field, link);
 244	return tail->offset + tail->size;
 245}
 246
 247/*
 248 * Check if the referenced field is an array and return true,
 249 * as arrays are OK to dereference.
 250 */
 251static bool test_field(const char *fmt, struct trace_event_call *call)
 252{
 253	struct trace_event_fields *field = call->class->fields_array;
 254	const char *array_descriptor;
 255	const char *p = fmt;
 256	int len;
 257
 258	if (!(len = str_has_prefix(fmt, "REC->")))
 259		return false;
 260	fmt += len;
 261	for (p = fmt; *p; p++) {
 262		if (!isalnum(*p) && *p != '_')
 263			break;
 264	}
 265	len = p - fmt;
 266
 267	for (; field->type; field++) {
 268		if (strncmp(field->name, fmt, len) ||
 269		    field->name[len])
 270			continue;
 271		array_descriptor = strchr(field->type, '[');
 272		/* This is an array and is OK to dereference. */
 273		return array_descriptor != NULL;
 274	}
 275	return false;
 276}
 277
 278/*
 279 * Examine the print fmt of the event looking for unsafe dereference
 280 * pointers using %p* that could be recorded in the trace event and
 281 * much later referenced after the pointer was freed. Dereferencing
 282 * pointers are OK, if it is dereferenced into the event itself.
 283 */
 284static void test_event_printk(struct trace_event_call *call)
 285{
 286	u64 dereference_flags = 0;
 287	bool first = true;
 288	const char *fmt, *c, *r, *a;
 289	int parens = 0;
 290	char in_quote = 0;
 291	int start_arg = 0;
 292	int arg = 0;
 293	int i;
 294
 295	fmt = call->print_fmt;
 296
 297	if (!fmt)
 298		return;
 299
 300	for (i = 0; fmt[i]; i++) {
 301		switch (fmt[i]) {
 302		case '\\':
 303			i++;
 304			if (!fmt[i])
 305				return;
 306			continue;
 307		case '"':
 308		case '\'':
 309			/*
 310			 * The print fmt starts with a string that
 311			 * is processed first to find %p* usage,
 312			 * then after the first string, the print fmt
 313			 * contains arguments that are used to check
 314			 * if the dereferenced %p* usage is safe.
 315			 */
 316			if (first) {
 317				if (fmt[i] == '\'')
 318					continue;
 319				if (in_quote) {
 320					arg = 0;
 321					first = false;
 322					/*
 323					 * If there was no %p* uses
 324					 * the fmt is OK.
 325					 */
 326					if (!dereference_flags)
 327						return;
 328				}
 329			}
 330			if (in_quote) {
 331				if (in_quote == fmt[i])
 332					in_quote = 0;
 333			} else {
 334				in_quote = fmt[i];
 335			}
 336			continue;
 337		case '%':
 338			if (!first || !in_quote)
 339				continue;
 340			i++;
 341			if (!fmt[i])
 342				return;
 343			switch (fmt[i]) {
 344			case '%':
 345				continue;
 346			case 'p':
 347				/* Find dereferencing fields */
 348				switch (fmt[i + 1]) {
 349				case 'B': case 'R': case 'r':
 350				case 'b': case 'M': case 'm':
 351				case 'I': case 'i': case 'E':
 352				case 'U': case 'V': case 'N':
 353				case 'a': case 'd': case 'D':
 354				case 'g': case 't': case 'C':
 355				case 'O': case 'f':
 356					if (WARN_ONCE(arg == 63,
 357						      "Too many args for event: %s",
 358						      trace_event_name(call)))
 359						return;
 360					dereference_flags |= 1ULL << arg;
 361				}
 362				break;
 363			default:
 364			{
 365				bool star = false;
 366				int j;
 367
 368				/* Increment arg if %*s exists. */
 369				for (j = 0; fmt[i + j]; j++) {
 370					if (isdigit(fmt[i + j]) ||
 371					    fmt[i + j] == '.')
 372						continue;
 373					if (fmt[i + j] == '*') {
 374						star = true;
 375						continue;
 376					}
 377					if ((fmt[i + j] == 's') && star)
 378						arg++;
 379					break;
 380				}
 381				break;
 382			} /* default */
 383
 384			} /* switch */
 385			arg++;
 386			continue;
 387		case '(':
 388			if (in_quote)
 389				continue;
 390			parens++;
 391			continue;
 392		case ')':
 393			if (in_quote)
 394				continue;
 395			parens--;
 396			if (WARN_ONCE(parens < 0,
 397				      "Paren mismatch for event: %s\narg='%s'\n%*s",
 398				      trace_event_name(call),
 399				      fmt + start_arg,
 400				      (i - start_arg) + 5, "^"))
 401				return;
 402			continue;
 403		case ',':
 404			if (in_quote || parens)
 405				continue;
 406			i++;
 407			while (isspace(fmt[i]))
 408				i++;
 409			start_arg = i;
 410			if (!(dereference_flags & (1ULL << arg)))
 411				goto next_arg;
 412
 413			/* Find the REC-> in the argument */
 414			c = strchr(fmt + i, ',');
 415			r = strstr(fmt + i, "REC->");
 416			if (r && (!c || r < c)) {
 417				/*
 418				 * Addresses of events on the buffer,
 419				 * or an array on the buffer is
 420				 * OK to dereference.
 421				 * There's ways to fool this, but
 422				 * this is to catch common mistakes,
 423				 * not malicious code.
 424				 */
 425				a = strchr(fmt + i, '&');
 426				if ((a && (a < r)) || test_field(r, call))
 427					dereference_flags &= ~(1ULL << arg);
 428			} else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
 429				   (!c || r < c)) {
 430				dereference_flags &= ~(1ULL << arg);
 431			} else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
 432				   (!c || r < c)) {
 433				dereference_flags &= ~(1ULL << arg);
 434			}
 435
 436		next_arg:
 437			i--;
 438			arg++;
 439		}
 440	}
 441
 442	/*
 443	 * If you triggered the below warning, the trace event reported
 444	 * uses an unsafe dereference pointer %p*. As the data stored
 445	 * at the trace event time may no longer exist when the trace
 446	 * event is printed, dereferencing to the original source is
 447	 * unsafe. The source of the dereference must be copied into the
 448	 * event itself, and the dereference must access the copy instead.
 449	 */
 450	if (WARN_ON_ONCE(dereference_flags)) {
 451		arg = 1;
 452		while (!(dereference_flags & 1)) {
 453			dereference_flags >>= 1;
 454			arg++;
 455		}
 456		pr_warn("event %s has unsafe dereference of argument %d\n",
 457			trace_event_name(call), arg);
 458		pr_warn("print_fmt: %s\n", fmt);
 459	}
 460}
 461
 462int trace_event_raw_init(struct trace_event_call *call)
 463{
 464	int id;
 465
 466	id = register_trace_event(&call->event);
 467	if (!id)
 468		return -ENODEV;
 469
 470	test_event_printk(call);
 471
 472	return 0;
 473}
 474EXPORT_SYMBOL_GPL(trace_event_raw_init);
 475
 476bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 477{
 478	struct trace_array *tr = trace_file->tr;
 479	struct trace_array_cpu *data;
 480	struct trace_pid_list *no_pid_list;
 481	struct trace_pid_list *pid_list;
 482
 483	pid_list = rcu_dereference_raw(tr->filtered_pids);
 484	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 485
 486	if (!pid_list && !no_pid_list)
 487		return false;
 488
 489	data = this_cpu_ptr(tr->array_buffer.data);
 490
 491	return data->ignore_pid;
 492}
 493EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 494
 495void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 496				 struct trace_event_file *trace_file,
 497				 unsigned long len)
 498{
 499	struct trace_event_call *event_call = trace_file->event_call;
 500
 501	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 502	    trace_event_ignore_this_pid(trace_file))
 503		return NULL;
 504
 505	/*
 506	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 507	 * preemption (adding one to the preempt_count). Since we are
 508	 * interested in the preempt_count at the time the tracepoint was
 509	 * hit, we need to subtract one to offset the increment.
 510	 */
 511	fbuffer->trace_ctx = tracing_gen_ctx_dec();
 512	fbuffer->trace_file = trace_file;
 513
 514	fbuffer->event =
 515		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 516						event_call->event.type, len,
 517						fbuffer->trace_ctx);
 518	if (!fbuffer->event)
 519		return NULL;
 520
 521	fbuffer->regs = NULL;
 522	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 523	return fbuffer->entry;
 524}
 525EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 526
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 527int trace_event_reg(struct trace_event_call *call,
 528		    enum trace_reg type, void *data)
 529{
 530	struct trace_event_file *file = data;
 531
 532	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 533	switch (type) {
 534	case TRACE_REG_REGISTER:
 535		return tracepoint_probe_register(call->tp,
 536						 call->class->probe,
 537						 file);
 538	case TRACE_REG_UNREGISTER:
 539		tracepoint_probe_unregister(call->tp,
 540					    call->class->probe,
 541					    file);
 542		return 0;
 543
 544#ifdef CONFIG_PERF_EVENTS
 545	case TRACE_REG_PERF_REGISTER:
 546		return tracepoint_probe_register(call->tp,
 547						 call->class->perf_probe,
 548						 call);
 549	case TRACE_REG_PERF_UNREGISTER:
 550		tracepoint_probe_unregister(call->tp,
 551					    call->class->perf_probe,
 552					    call);
 553		return 0;
 554	case TRACE_REG_PERF_OPEN:
 555	case TRACE_REG_PERF_CLOSE:
 556	case TRACE_REG_PERF_ADD:
 557	case TRACE_REG_PERF_DEL:
 558		return 0;
 559#endif
 560	}
 561	return 0;
 562}
 563EXPORT_SYMBOL_GPL(trace_event_reg);
 564
 565void trace_event_enable_cmd_record(bool enable)
 566{
 567	struct trace_event_file *file;
 568	struct trace_array *tr;
 569
 570	lockdep_assert_held(&event_mutex);
 571
 572	do_for_each_event_file(tr, file) {
 573
 574		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 575			continue;
 576
 577		if (enable) {
 578			tracing_start_cmdline_record();
 579			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 580		} else {
 581			tracing_stop_cmdline_record();
 582			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 583		}
 584	} while_for_each_event_file();
 585}
 586
 587void trace_event_enable_tgid_record(bool enable)
 588{
 589	struct trace_event_file *file;
 590	struct trace_array *tr;
 591
 592	lockdep_assert_held(&event_mutex);
 593
 594	do_for_each_event_file(tr, file) {
 595		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 596			continue;
 597
 598		if (enable) {
 599			tracing_start_tgid_record();
 600			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 601		} else {
 602			tracing_stop_tgid_record();
 603			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 604				  &file->flags);
 605		}
 606	} while_for_each_event_file();
 607}
 608
 609static int __ftrace_event_enable_disable(struct trace_event_file *file,
 610					 int enable, int soft_disable)
 611{
 612	struct trace_event_call *call = file->event_call;
 613	struct trace_array *tr = file->tr;
 614	int ret = 0;
 615	int disable;
 616
 617	switch (enable) {
 618	case 0:
 619		/*
 620		 * When soft_disable is set and enable is cleared, the sm_ref
 621		 * reference counter is decremented. If it reaches 0, we want
 622		 * to clear the SOFT_DISABLED flag but leave the event in the
 623		 * state that it was. That is, if the event was enabled and
 624		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 625		 * is set we do not want the event to be enabled before we
 626		 * clear the bit.
 627		 *
 628		 * When soft_disable is not set but the SOFT_MODE flag is,
 629		 * we do nothing. Do not disable the tracepoint, otherwise
 630		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 631		 */
 632		if (soft_disable) {
 633			if (atomic_dec_return(&file->sm_ref) > 0)
 634				break;
 635			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 636			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 637			/* Disable use of trace_buffered_event */
 638			trace_buffered_event_disable();
 639		} else
 640			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 641
 642		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 643			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 644			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 645				tracing_stop_cmdline_record();
 646				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 647			}
 648
 649			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 650				tracing_stop_tgid_record();
 651				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 652			}
 653
 654			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 655		}
 656		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 657		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 658			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 659		else
 660			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 661		break;
 662	case 1:
 663		/*
 664		 * When soft_disable is set and enable is set, we want to
 665		 * register the tracepoint for the event, but leave the event
 666		 * as is. That means, if the event was already enabled, we do
 667		 * nothing (but set SOFT_MODE). If the event is disabled, we
 668		 * set SOFT_DISABLED before enabling the event tracepoint, so
 669		 * it still seems to be disabled.
 670		 */
 671		if (!soft_disable)
 672			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 673		else {
 674			if (atomic_inc_return(&file->sm_ref) > 1)
 675				break;
 676			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 677			/* Enable use of trace_buffered_event */
 678			trace_buffered_event_enable();
 679		}
 680
 681		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 682			bool cmd = false, tgid = false;
 683
 684			/* Keep the event disabled, when going to SOFT_MODE. */
 685			if (soft_disable)
 686				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 687
 688			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 689				cmd = true;
 690				tracing_start_cmdline_record();
 691				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 692			}
 693
 694			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 695				tgid = true;
 696				tracing_start_tgid_record();
 697				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 698			}
 699
 700			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 701			if (ret) {
 702				if (cmd)
 703					tracing_stop_cmdline_record();
 704				if (tgid)
 705					tracing_stop_tgid_record();
 706				pr_info("event trace: Could not enable event "
 707					"%s\n", trace_event_name(call));
 708				break;
 709			}
 710			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 711
 712			/* WAS_ENABLED gets set but never cleared. */
 713			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 714		}
 715		break;
 716	}
 717
 718	return ret;
 719}
 720
 721int trace_event_enable_disable(struct trace_event_file *file,
 722			       int enable, int soft_disable)
 723{
 724	return __ftrace_event_enable_disable(file, enable, soft_disable);
 725}
 726
 727static int ftrace_event_enable_disable(struct trace_event_file *file,
 728				       int enable)
 729{
 730	return __ftrace_event_enable_disable(file, enable, 0);
 731}
 732
 733static void ftrace_clear_events(struct trace_array *tr)
 734{
 735	struct trace_event_file *file;
 736
 737	mutex_lock(&event_mutex);
 738	list_for_each_entry(file, &tr->events, list) {
 739		ftrace_event_enable_disable(file, 0);
 740	}
 741	mutex_unlock(&event_mutex);
 742}
 743
 744static void
 745event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 746{
 747	struct trace_pid_list *pid_list;
 748	struct trace_array *tr = data;
 749
 750	pid_list = rcu_dereference_raw(tr->filtered_pids);
 751	trace_filter_add_remove_task(pid_list, NULL, task);
 752
 753	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 754	trace_filter_add_remove_task(pid_list, NULL, task);
 755}
 756
 757static void
 758event_filter_pid_sched_process_fork(void *data,
 759				    struct task_struct *self,
 760				    struct task_struct *task)
 761{
 762	struct trace_pid_list *pid_list;
 763	struct trace_array *tr = data;
 
 
 
 
 
 
 
 764
 765	pid_list = rcu_dereference_sched(tr->filtered_pids);
 766	trace_filter_add_remove_task(pid_list, self, task);
 767
 768	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 769	trace_filter_add_remove_task(pid_list, self, task);
 770}
 
 
 771
 772void trace_event_follow_fork(struct trace_array *tr, bool enable)
 773{
 774	if (enable) {
 775		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 776						       tr, INT_MIN);
 777		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 778						       tr, INT_MAX);
 779	} else {
 780		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 781						    tr);
 782		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 783						    tr);
 784	}
 785}
 786
 787static void
 788event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 789					struct task_struct *prev,
 790					struct task_struct *next,
 791					unsigned int prev_state)
 792{
 793	struct trace_array *tr = data;
 794	struct trace_pid_list *no_pid_list;
 795	struct trace_pid_list *pid_list;
 796	bool ret;
 797
 798	pid_list = rcu_dereference_sched(tr->filtered_pids);
 799	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 800
 801	/*
 802	 * Sched switch is funny, as we only want to ignore it
 803	 * in the notrace case if both prev and next should be ignored.
 804	 */
 805	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 806		trace_ignore_this_task(NULL, no_pid_list, next);
 807
 808	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 809		       (trace_ignore_this_task(pid_list, NULL, prev) &&
 810			trace_ignore_this_task(pid_list, NULL, next)));
 811}
 812
 813static void
 814event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 815					 struct task_struct *prev,
 816					 struct task_struct *next,
 817					 unsigned int prev_state)
 818{
 819	struct trace_array *tr = data;
 820	struct trace_pid_list *no_pid_list;
 821	struct trace_pid_list *pid_list;
 822
 823	pid_list = rcu_dereference_sched(tr->filtered_pids);
 824	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 825
 826	this_cpu_write(tr->array_buffer.data->ignore_pid,
 827		       trace_ignore_this_task(pid_list, no_pid_list, next));
 828}
 829
 830static void
 831event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 832{
 833	struct trace_array *tr = data;
 834	struct trace_pid_list *no_pid_list;
 835	struct trace_pid_list *pid_list;
 836
 837	/* Nothing to do if we are already tracing */
 838	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 839		return;
 840
 841	pid_list = rcu_dereference_sched(tr->filtered_pids);
 842	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 843
 844	this_cpu_write(tr->array_buffer.data->ignore_pid,
 845		       trace_ignore_this_task(pid_list, no_pid_list, task));
 846}
 847
 848static void
 849event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 850{
 851	struct trace_array *tr = data;
 852	struct trace_pid_list *no_pid_list;
 853	struct trace_pid_list *pid_list;
 854
 855	/* Nothing to do if we are not tracing */
 856	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 857		return;
 858
 859	pid_list = rcu_dereference_sched(tr->filtered_pids);
 860	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 861
 862	/* Set tracing if current is enabled */
 863	this_cpu_write(tr->array_buffer.data->ignore_pid,
 864		       trace_ignore_this_task(pid_list, no_pid_list, current));
 865}
 866
 867static void unregister_pid_events(struct trace_array *tr)
 868{
 
 
 
 
 
 
 
 
 
 869	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 870	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 871
 872	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 873	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 874
 875	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 876	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 877
 878	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 879	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 880}
 881
 882static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 883{
 884	struct trace_pid_list *pid_list;
 885	struct trace_pid_list *no_pid_list;
 886	struct trace_event_file *file;
 887	int cpu;
 888
 889	pid_list = rcu_dereference_protected(tr->filtered_pids,
 890					     lockdep_is_held(&event_mutex));
 891	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 892					     lockdep_is_held(&event_mutex));
 893
 894	/* Make sure there's something to do */
 895	if (!pid_type_enabled(type, pid_list, no_pid_list))
 896		return;
 897
 898	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 899		unregister_pid_events(tr);
 900
 901		list_for_each_entry(file, &tr->events, list) {
 902			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 903		}
 904
 905		for_each_possible_cpu(cpu)
 906			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 907	}
 908
 909	if (type & TRACE_PIDS)
 910		rcu_assign_pointer(tr->filtered_pids, NULL);
 911
 912	if (type & TRACE_NO_PIDS)
 913		rcu_assign_pointer(tr->filtered_no_pids, NULL);
 914
 915	/* Wait till all users are no longer using pid filtering */
 916	tracepoint_synchronize_unregister();
 917
 918	if ((type & TRACE_PIDS) && pid_list)
 919		trace_pid_list_free(pid_list);
 920
 921	if ((type & TRACE_NO_PIDS) && no_pid_list)
 922		trace_pid_list_free(no_pid_list);
 923}
 924
 925static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 926{
 927	mutex_lock(&event_mutex);
 928	__ftrace_clear_event_pids(tr, type);
 929	mutex_unlock(&event_mutex);
 930}
 931
 932static void __put_system(struct event_subsystem *system)
 933{
 934	struct event_filter *filter = system->filter;
 935
 936	WARN_ON_ONCE(system_refcount(system) == 0);
 937	if (system_refcount_dec(system))
 938		return;
 939
 940	list_del(&system->list);
 941
 942	if (filter) {
 943		kfree(filter->filter_string);
 944		kfree(filter);
 945	}
 946	kfree_const(system->name);
 947	kfree(system);
 948}
 949
 950static void __get_system(struct event_subsystem *system)
 951{
 952	WARN_ON_ONCE(system_refcount(system) == 0);
 953	system_refcount_inc(system);
 954}
 955
 956static void __get_system_dir(struct trace_subsystem_dir *dir)
 957{
 958	WARN_ON_ONCE(dir->ref_count == 0);
 959	dir->ref_count++;
 960	__get_system(dir->subsystem);
 961}
 962
 963static void __put_system_dir(struct trace_subsystem_dir *dir)
 964{
 965	WARN_ON_ONCE(dir->ref_count == 0);
 966	/* If the subsystem is about to be freed, the dir must be too */
 967	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 968
 969	__put_system(dir->subsystem);
 970	if (!--dir->ref_count)
 971		kfree(dir);
 972}
 973
 974static void put_system(struct trace_subsystem_dir *dir)
 975{
 976	mutex_lock(&event_mutex);
 977	__put_system_dir(dir);
 978	mutex_unlock(&event_mutex);
 979}
 980
 981static void remove_subsystem(struct trace_subsystem_dir *dir)
 982{
 983	if (!dir)
 984		return;
 985
 986	if (!--dir->nr_events) {
 987		eventfs_remove_dir(dir->ei);
 988		list_del(&dir->list);
 989		__put_system_dir(dir);
 990	}
 991}
 992
 993void event_file_get(struct trace_event_file *file)
 994{
 995	atomic_inc(&file->ref);
 996}
 997
 998void event_file_put(struct trace_event_file *file)
 999{
1000	if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001		if (file->flags & EVENT_FILE_FL_FREED)
1002			kmem_cache_free(file_cachep, file);
1003		return;
1004	}
1005
1006	if (atomic_dec_and_test(&file->ref)) {
1007		/* Count should only go to zero when it is freed */
1008		if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009			return;
1010		kmem_cache_free(file_cachep, file);
1011	}
1012}
1013
1014static void remove_event_file_dir(struct trace_event_file *file)
1015{
1016	eventfs_remove_dir(file->ei);
1017	list_del(&file->list);
1018	remove_subsystem(file->system);
1019	free_event_filter(file->filter);
1020	file->flags |= EVENT_FILE_FL_FREED;
1021	event_file_put(file);
1022}
1023
1024/*
1025 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026 */
1027static int
1028__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029			      const char *sub, const char *event, int set)
1030{
1031	struct trace_event_file *file;
1032	struct trace_event_call *call;
1033	const char *name;
1034	int ret = -EINVAL;
1035	int eret = 0;
1036
1037	list_for_each_entry(file, &tr->events, list) {
1038
1039		call = file->event_call;
1040		name = trace_event_name(call);
1041
1042		if (!name || !call->class || !call->class->reg)
1043			continue;
1044
1045		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046			continue;
1047
1048		if (match &&
1049		    strcmp(match, name) != 0 &&
1050		    strcmp(match, call->class->system) != 0)
1051			continue;
1052
1053		if (sub && strcmp(sub, call->class->system) != 0)
1054			continue;
1055
1056		if (event && strcmp(event, name) != 0)
1057			continue;
1058
1059		ret = ftrace_event_enable_disable(file, set);
1060
1061		/*
1062		 * Save the first error and return that. Some events
1063		 * may still have been enabled, but let the user
1064		 * know that something went wrong.
1065		 */
1066		if (ret && !eret)
1067			eret = ret;
1068
1069		ret = eret;
1070	}
1071
1072	return ret;
1073}
1074
1075static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076				  const char *sub, const char *event, int set)
1077{
1078	int ret;
1079
1080	mutex_lock(&event_mutex);
1081	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082	mutex_unlock(&event_mutex);
1083
1084	return ret;
1085}
1086
1087int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088{
1089	char *event = NULL, *sub = NULL, *match;
1090	int ret;
1091
1092	if (!tr)
1093		return -ENOENT;
1094	/*
1095	 * The buf format can be <subsystem>:<event-name>
1096	 *  *:<event-name> means any event by that name.
1097	 *  :<event-name> is the same.
1098	 *
1099	 *  <subsystem>:* means all events in that subsystem
1100	 *  <subsystem>: means the same.
1101	 *
1102	 *  <name> (no ':') means all events in a subsystem with
1103	 *  the name <name> or any event that matches <name>
1104	 */
1105
1106	match = strsep(&buf, ":");
1107	if (buf) {
1108		sub = match;
1109		event = buf;
1110		match = NULL;
1111
1112		if (!strlen(sub) || strcmp(sub, "*") == 0)
1113			sub = NULL;
1114		if (!strlen(event) || strcmp(event, "*") == 0)
1115			event = NULL;
1116	}
1117
1118	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120	/* Put back the colon to allow this to be called again */
1121	if (buf)
1122		*(buf - 1) = ':';
1123
1124	return ret;
1125}
1126
1127/**
1128 * trace_set_clr_event - enable or disable an event
1129 * @system: system name to match (NULL for any system)
1130 * @event: event name to match (NULL for all events, within system)
1131 * @set: 1 to enable, 0 to disable
1132 *
1133 * This is a way for other parts of the kernel to enable or disable
1134 * event recording.
1135 *
1136 * Returns 0 on success, -EINVAL if the parameters do not match any
1137 * registered events.
1138 */
1139int trace_set_clr_event(const char *system, const char *event, int set)
1140{
1141	struct trace_array *tr = top_trace_array();
1142
1143	if (!tr)
1144		return -ENODEV;
1145
1146	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147}
1148EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150/**
1151 * trace_array_set_clr_event - enable or disable an event for a trace array.
1152 * @tr: concerned trace array.
1153 * @system: system name to match (NULL for any system)
1154 * @event: event name to match (NULL for all events, within system)
1155 * @enable: true to enable, false to disable
1156 *
1157 * This is a way for other parts of the kernel to enable or disable
1158 * event recording.
1159 *
1160 * Returns 0 on success, -EINVAL if the parameters do not match any
1161 * registered events.
1162 */
1163int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164		const char *event, bool enable)
1165{
1166	int set;
1167
1168	if (!tr)
1169		return -ENOENT;
1170
1171	set = (enable == true) ? 1 : 0;
1172	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173}
1174EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176/* 128 should be much more than enough */
1177#define EVENT_BUF_SIZE		127
1178
1179static ssize_t
1180ftrace_event_write(struct file *file, const char __user *ubuf,
1181		   size_t cnt, loff_t *ppos)
1182{
1183	struct trace_parser parser;
1184	struct seq_file *m = file->private_data;
1185	struct trace_array *tr = m->private;
1186	ssize_t read, ret;
1187
1188	if (!cnt)
1189		return 0;
1190
1191	ret = tracing_update_buffers(tr);
1192	if (ret < 0)
1193		return ret;
1194
1195	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196		return -ENOMEM;
1197
1198	read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200	if (read >= 0 && trace_parser_loaded((&parser))) {
1201		int set = 1;
1202
1203		if (*parser.buffer == '!')
1204			set = 0;
1205
 
 
1206		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207		if (ret)
1208			goto out_put;
1209	}
1210
1211	ret = read;
1212
1213 out_put:
1214	trace_parser_put(&parser);
1215
1216	return ret;
1217}
1218
1219static void *
1220t_next(struct seq_file *m, void *v, loff_t *pos)
1221{
1222	struct trace_event_file *file = v;
1223	struct trace_event_call *call;
1224	struct trace_array *tr = m->private;
1225
1226	(*pos)++;
1227
1228	list_for_each_entry_continue(file, &tr->events, list) {
1229		call = file->event_call;
1230		/*
1231		 * The ftrace subsystem is for showing formats only.
1232		 * They can not be enabled or disabled via the event files.
1233		 */
1234		if (call->class && call->class->reg &&
1235		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236			return file;
1237	}
1238
1239	return NULL;
1240}
1241
1242static void *t_start(struct seq_file *m, loff_t *pos)
1243{
1244	struct trace_event_file *file;
1245	struct trace_array *tr = m->private;
1246	loff_t l;
1247
1248	mutex_lock(&event_mutex);
1249
1250	file = list_entry(&tr->events, struct trace_event_file, list);
1251	for (l = 0; l <= *pos; ) {
1252		file = t_next(m, file, &l);
1253		if (!file)
1254			break;
1255	}
1256	return file;
1257}
1258
1259static void *
1260s_next(struct seq_file *m, void *v, loff_t *pos)
1261{
1262	struct trace_event_file *file = v;
1263	struct trace_array *tr = m->private;
1264
1265	(*pos)++;
1266
1267	list_for_each_entry_continue(file, &tr->events, list) {
1268		if (file->flags & EVENT_FILE_FL_ENABLED)
1269			return file;
1270	}
1271
1272	return NULL;
1273}
1274
1275static void *s_start(struct seq_file *m, loff_t *pos)
1276{
1277	struct trace_event_file *file;
1278	struct trace_array *tr = m->private;
1279	loff_t l;
1280
1281	mutex_lock(&event_mutex);
1282
1283	file = list_entry(&tr->events, struct trace_event_file, list);
1284	for (l = 0; l <= *pos; ) {
1285		file = s_next(m, file, &l);
1286		if (!file)
1287			break;
1288	}
1289	return file;
1290}
1291
1292static int t_show(struct seq_file *m, void *v)
1293{
1294	struct trace_event_file *file = v;
1295	struct trace_event_call *call = file->event_call;
1296
1297	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298		seq_printf(m, "%s:", call->class->system);
1299	seq_printf(m, "%s\n", trace_event_name(call));
1300
1301	return 0;
1302}
1303
1304static void t_stop(struct seq_file *m, void *p)
1305{
1306	mutex_unlock(&event_mutex);
1307}
1308
1309static void *
1310__next(struct seq_file *m, void *v, loff_t *pos, int type)
1311{
1312	struct trace_array *tr = m->private;
1313	struct trace_pid_list *pid_list;
1314
1315	if (type == TRACE_PIDS)
1316		pid_list = rcu_dereference_sched(tr->filtered_pids);
1317	else
1318		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320	return trace_pid_next(pid_list, v, pos);
1321}
1322
1323static void *
1324p_next(struct seq_file *m, void *v, loff_t *pos)
1325{
1326	return __next(m, v, pos, TRACE_PIDS);
1327}
1328
1329static void *
1330np_next(struct seq_file *m, void *v, loff_t *pos)
1331{
1332	return __next(m, v, pos, TRACE_NO_PIDS);
1333}
1334
1335static void *__start(struct seq_file *m, loff_t *pos, int type)
1336	__acquires(RCU)
1337{
1338	struct trace_pid_list *pid_list;
1339	struct trace_array *tr = m->private;
1340
1341	/*
1342	 * Grab the mutex, to keep calls to p_next() having the same
1343	 * tr->filtered_pids as p_start() has.
1344	 * If we just passed the tr->filtered_pids around, then RCU would
1345	 * have been enough, but doing that makes things more complex.
1346	 */
1347	mutex_lock(&event_mutex);
1348	rcu_read_lock_sched();
1349
1350	if (type == TRACE_PIDS)
1351		pid_list = rcu_dereference_sched(tr->filtered_pids);
1352	else
1353		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355	if (!pid_list)
1356		return NULL;
1357
1358	return trace_pid_start(pid_list, pos);
1359}
1360
1361static void *p_start(struct seq_file *m, loff_t *pos)
1362	__acquires(RCU)
1363{
1364	return __start(m, pos, TRACE_PIDS);
 
1365}
1366
1367static void *np_start(struct seq_file *m, loff_t *pos)
1368	__acquires(RCU)
1369{
1370	return __start(m, pos, TRACE_NO_PIDS);
 
 
 
 
 
 
 
 
1371}
1372
1373static void p_stop(struct seq_file *m, void *p)
1374	__releases(RCU)
1375{
1376	rcu_read_unlock_sched();
1377	mutex_unlock(&event_mutex);
 
 
1378}
1379
1380static ssize_t
1381event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382		  loff_t *ppos)
1383{
1384	struct trace_event_file *file;
1385	unsigned long flags;
1386	char buf[4] = "0";
1387
1388	mutex_lock(&event_mutex);
1389	file = event_file_data(filp);
1390	if (likely(file))
1391		flags = file->flags;
1392	mutex_unlock(&event_mutex);
1393
1394	if (!file || flags & EVENT_FILE_FL_FREED)
1395		return -ENODEV;
1396
1397	if (flags & EVENT_FILE_FL_ENABLED &&
1398	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399		strcpy(buf, "1");
1400
1401	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402	    flags & EVENT_FILE_FL_SOFT_MODE)
1403		strcat(buf, "*");
1404
1405	strcat(buf, "\n");
1406
1407	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408}
1409
1410static ssize_t
1411event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412		   loff_t *ppos)
1413{
1414	struct trace_event_file *file;
1415	unsigned long val;
1416	int ret;
1417
1418	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419	if (ret)
1420		return ret;
1421
 
 
 
 
1422	switch (val) {
1423	case 0:
1424	case 1:
1425		ret = -ENODEV;
1426		mutex_lock(&event_mutex);
1427		file = event_file_data(filp);
1428		if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429			ret = tracing_update_buffers(file->tr);
1430			if (ret < 0) {
1431				mutex_unlock(&event_mutex);
1432				return ret;
1433			}
1434			ret = ftrace_event_enable_disable(file, val);
1435		}
1436		mutex_unlock(&event_mutex);
1437		break;
1438
1439	default:
1440		return -EINVAL;
1441	}
1442
1443	*ppos += cnt;
1444
1445	return ret ? ret : cnt;
1446}
1447
1448static ssize_t
1449system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450		   loff_t *ppos)
1451{
1452	const char set_to_char[4] = { '?', '0', '1', 'X' };
1453	struct trace_subsystem_dir *dir = filp->private_data;
1454	struct event_subsystem *system = dir->subsystem;
1455	struct trace_event_call *call;
1456	struct trace_event_file *file;
1457	struct trace_array *tr = dir->tr;
1458	char buf[2];
1459	int set = 0;
1460	int ret;
1461
1462	mutex_lock(&event_mutex);
1463	list_for_each_entry(file, &tr->events, list) {
1464		call = file->event_call;
1465		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466		    !trace_event_name(call) || !call->class || !call->class->reg)
1467			continue;
1468
1469		if (system && strcmp(call->class->system, system->name) != 0)
1470			continue;
1471
1472		/*
1473		 * We need to find out if all the events are set
1474		 * or if all events or cleared, or if we have
1475		 * a mixture.
1476		 */
1477		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479		/*
1480		 * If we have a mixture, no need to look further.
1481		 */
1482		if (set == 3)
1483			break;
1484	}
1485	mutex_unlock(&event_mutex);
1486
1487	buf[0] = set_to_char[set];
1488	buf[1] = '\n';
1489
1490	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492	return ret;
1493}
1494
1495static ssize_t
1496system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497		    loff_t *ppos)
1498{
1499	struct trace_subsystem_dir *dir = filp->private_data;
1500	struct event_subsystem *system = dir->subsystem;
1501	const char *name = NULL;
1502	unsigned long val;
1503	ssize_t ret;
1504
1505	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506	if (ret)
1507		return ret;
1508
1509	ret = tracing_update_buffers(dir->tr);
1510	if (ret < 0)
1511		return ret;
1512
1513	if (val != 0 && val != 1)
1514		return -EINVAL;
1515
1516	/*
1517	 * Opening of "enable" adds a ref count to system,
1518	 * so the name is safe to use.
1519	 */
1520	if (system)
1521		name = system->name;
1522
1523	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524	if (ret)
1525		goto out;
1526
1527	ret = cnt;
1528
1529out:
1530	*ppos += cnt;
1531
1532	return ret;
1533}
1534
1535enum {
1536	FORMAT_HEADER		= 1,
1537	FORMAT_FIELD_SEPERATOR	= 2,
1538	FORMAT_PRINTFMT		= 3,
1539};
1540
1541static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542{
1543	struct trace_event_call *call = event_file_data(m->private);
1544	struct list_head *common_head = &ftrace_common_fields;
1545	struct list_head *head = trace_get_fields(call);
1546	struct list_head *node = v;
1547
1548	(*pos)++;
1549
1550	switch ((unsigned long)v) {
1551	case FORMAT_HEADER:
1552		node = common_head;
1553		break;
1554
1555	case FORMAT_FIELD_SEPERATOR:
1556		node = head;
1557		break;
1558
1559	case FORMAT_PRINTFMT:
1560		/* all done */
1561		return NULL;
1562	}
1563
1564	node = node->prev;
1565	if (node == common_head)
1566		return (void *)FORMAT_FIELD_SEPERATOR;
1567	else if (node == head)
1568		return (void *)FORMAT_PRINTFMT;
1569	else
1570		return node;
1571}
1572
1573static int f_show(struct seq_file *m, void *v)
1574{
1575	struct trace_event_call *call = event_file_data(m->private);
1576	struct ftrace_event_field *field;
1577	const char *array_descriptor;
1578
1579	switch ((unsigned long)v) {
1580	case FORMAT_HEADER:
1581		seq_printf(m, "name: %s\n", trace_event_name(call));
1582		seq_printf(m, "ID: %d\n", call->event.type);
1583		seq_puts(m, "format:\n");
1584		return 0;
1585
1586	case FORMAT_FIELD_SEPERATOR:
1587		seq_putc(m, '\n');
1588		return 0;
1589
1590	case FORMAT_PRINTFMT:
1591		seq_printf(m, "\nprint fmt: %s\n",
1592			   call->print_fmt);
1593		return 0;
1594	}
1595
1596	field = list_entry(v, struct ftrace_event_field, link);
1597	/*
1598	 * Smartly shows the array type(except dynamic array).
1599	 * Normal:
1600	 *	field:TYPE VAR
1601	 * If TYPE := TYPE[LEN], it is shown:
1602	 *	field:TYPE VAR[LEN]
1603	 */
1604	array_descriptor = strchr(field->type, '[');
1605
1606	if (str_has_prefix(field->type, "__data_loc"))
1607		array_descriptor = NULL;
1608
1609	if (!array_descriptor)
1610		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611			   field->type, field->name, field->offset,
1612			   field->size, !!field->is_signed);
1613	else if (field->len)
1614		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615			   (int)(array_descriptor - field->type),
1616			   field->type, field->name,
1617			   field->len, field->offset,
1618			   field->size, !!field->is_signed);
1619	else
1620		seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621				(int)(array_descriptor - field->type),
1622				field->type, field->name,
1623				field->offset, field->size, !!field->is_signed);
1624
1625	return 0;
1626}
1627
1628static void *f_start(struct seq_file *m, loff_t *pos)
1629{
1630	void *p = (void *)FORMAT_HEADER;
1631	loff_t l = 0;
1632
1633	/* ->stop() is called even if ->start() fails */
1634	mutex_lock(&event_mutex);
1635	if (!event_file_data(m->private))
1636		return ERR_PTR(-ENODEV);
1637
1638	while (l < *pos && p)
1639		p = f_next(m, p, &l);
1640
1641	return p;
1642}
1643
1644static void f_stop(struct seq_file *m, void *p)
1645{
1646	mutex_unlock(&event_mutex);
1647}
1648
1649static const struct seq_operations trace_format_seq_ops = {
1650	.start		= f_start,
1651	.next		= f_next,
1652	.stop		= f_stop,
1653	.show		= f_show,
1654};
1655
1656static int trace_format_open(struct inode *inode, struct file *file)
1657{
1658	struct seq_file *m;
1659	int ret;
1660
1661	/* Do we want to hide event format files on tracefs lockdown? */
1662
1663	ret = seq_open(file, &trace_format_seq_ops);
1664	if (ret < 0)
1665		return ret;
1666
1667	m = file->private_data;
1668	m->private = file;
1669
1670	return 0;
1671}
1672
1673#ifdef CONFIG_PERF_EVENTS
1674static ssize_t
1675event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1676{
1677	int id = (long)event_file_data(filp);
1678	char buf[32];
1679	int len;
1680
 
 
 
1681	if (unlikely(!id))
1682		return -ENODEV;
1683
1684	len = sprintf(buf, "%d\n", id);
1685
1686	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687}
1688#endif
1689
1690static ssize_t
1691event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1692		  loff_t *ppos)
1693{
1694	struct trace_event_file *file;
1695	struct trace_seq *s;
1696	int r = -ENODEV;
1697
1698	if (*ppos)
1699		return 0;
1700
1701	s = kmalloc(sizeof(*s), GFP_KERNEL);
1702
1703	if (!s)
1704		return -ENOMEM;
1705
1706	trace_seq_init(s);
1707
1708	mutex_lock(&event_mutex);
1709	file = event_file_data(filp);
1710	if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711		print_event_filter(file, s);
1712	mutex_unlock(&event_mutex);
1713
1714	if (file)
1715		r = simple_read_from_buffer(ubuf, cnt, ppos,
1716					    s->buffer, trace_seq_used(s));
1717
1718	kfree(s);
1719
1720	return r;
1721}
1722
1723static ssize_t
1724event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1725		   loff_t *ppos)
1726{
1727	struct trace_event_file *file;
1728	char *buf;
1729	int err = -ENODEV;
1730
1731	if (cnt >= PAGE_SIZE)
1732		return -EINVAL;
1733
1734	buf = memdup_user_nul(ubuf, cnt);
1735	if (IS_ERR(buf))
1736		return PTR_ERR(buf);
1737
1738	mutex_lock(&event_mutex);
1739	file = event_file_data(filp);
1740	if (file)
1741		err = apply_event_filter(file, buf);
1742	mutex_unlock(&event_mutex);
1743
1744	kfree(buf);
1745	if (err < 0)
1746		return err;
1747
1748	*ppos += cnt;
1749
1750	return cnt;
1751}
1752
1753static LIST_HEAD(event_subsystems);
1754
1755static int subsystem_open(struct inode *inode, struct file *filp)
1756{
1757	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758	struct trace_array *tr = NULL, *iter_tr;
1759	struct event_subsystem *system = NULL;
 
 
1760	int ret;
1761
1762	if (tracing_is_disabled())
1763		return -ENODEV;
1764
1765	/* Make sure the system still exists */
 
1766	mutex_lock(&event_mutex);
1767	mutex_lock(&trace_types_lock);
1768	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770			if (iter_dir == inode->i_private) {
1771				/* Don't open systems with no events */
1772				tr = iter_tr;
1773				dir = iter_dir;
1774				if (dir->nr_events) {
1775					__get_system_dir(dir);
1776					system = dir->subsystem;
1777				}
1778				goto exit_loop;
1779			}
1780		}
1781	}
1782 exit_loop:
 
1783	mutex_unlock(&trace_types_lock);
1784	mutex_unlock(&event_mutex);
1785
1786	if (!system)
1787		return -ENODEV;
1788
 
 
 
1789	/* Still need to increment the ref count of the system */
1790	if (trace_array_get(tr) < 0) {
1791		put_system(dir);
1792		return -ENODEV;
1793	}
1794
1795	ret = tracing_open_generic(inode, filp);
1796	if (ret < 0) {
1797		trace_array_put(tr);
1798		put_system(dir);
1799	}
1800
1801	return ret;
1802}
1803
1804static int system_tr_open(struct inode *inode, struct file *filp)
1805{
1806	struct trace_subsystem_dir *dir;
1807	struct trace_array *tr = inode->i_private;
1808	int ret;
1809
 
 
 
 
 
 
1810	/* Make a temporary dir that has no system but points to tr */
1811	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1812	if (!dir)
 
1813		return -ENOMEM;
 
1814
1815	ret = tracing_open_generic_tr(inode, filp);
 
 
1816	if (ret < 0) {
 
1817		kfree(dir);
1818		return ret;
1819	}
1820	dir->tr = tr;
1821	filp->private_data = dir;
1822
1823	return 0;
1824}
1825
1826static int subsystem_release(struct inode *inode, struct file *file)
1827{
1828	struct trace_subsystem_dir *dir = file->private_data;
1829
1830	trace_array_put(dir->tr);
1831
1832	/*
1833	 * If dir->subsystem is NULL, then this is a temporary
1834	 * descriptor that was made for a trace_array to enable
1835	 * all subsystems.
1836	 */
1837	if (dir->subsystem)
1838		put_system(dir);
1839	else
1840		kfree(dir);
1841
1842	return 0;
1843}
1844
1845static ssize_t
1846subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1847		      loff_t *ppos)
1848{
1849	struct trace_subsystem_dir *dir = filp->private_data;
1850	struct event_subsystem *system = dir->subsystem;
1851	struct trace_seq *s;
1852	int r;
1853
1854	if (*ppos)
1855		return 0;
1856
1857	s = kmalloc(sizeof(*s), GFP_KERNEL);
1858	if (!s)
1859		return -ENOMEM;
1860
1861	trace_seq_init(s);
1862
1863	print_subsystem_event_filter(system, s);
1864	r = simple_read_from_buffer(ubuf, cnt, ppos,
1865				    s->buffer, trace_seq_used(s));
1866
1867	kfree(s);
1868
1869	return r;
1870}
1871
1872static ssize_t
1873subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1874		       loff_t *ppos)
1875{
1876	struct trace_subsystem_dir *dir = filp->private_data;
1877	char *buf;
1878	int err;
1879
1880	if (cnt >= PAGE_SIZE)
1881		return -EINVAL;
1882
1883	buf = memdup_user_nul(ubuf, cnt);
1884	if (IS_ERR(buf))
1885		return PTR_ERR(buf);
1886
1887	err = apply_subsystem_event_filter(dir, buf);
1888	kfree(buf);
1889	if (err < 0)
1890		return err;
1891
1892	*ppos += cnt;
1893
1894	return cnt;
1895}
1896
1897static ssize_t
1898show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1899{
1900	struct trace_array *tr = filp->private_data;
1901	struct trace_seq *s;
1902	int r;
1903
1904	if (*ppos)
1905		return 0;
1906
1907	s = kmalloc(sizeof(*s), GFP_KERNEL);
1908	if (!s)
1909		return -ENOMEM;
1910
1911	trace_seq_init(s);
1912
1913	ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914	r = simple_read_from_buffer(ubuf, cnt, ppos,
1915				    s->buffer, trace_seq_used(s));
1916
1917	kfree(s);
1918
1919	return r;
1920}
1921
1922static ssize_t
1923show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1924{
1925	struct trace_seq *s;
1926	int r;
1927
1928	if (*ppos)
1929		return 0;
1930
1931	s = kmalloc(sizeof(*s), GFP_KERNEL);
1932	if (!s)
1933		return -ENOMEM;
1934
1935	trace_seq_init(s);
1936
1937	ring_buffer_print_entry_header(s);
1938	r = simple_read_from_buffer(ubuf, cnt, ppos,
1939				    s->buffer, trace_seq_used(s));
1940
1941	kfree(s);
1942
1943	return r;
1944}
1945
1946static void ignore_task_cpu(void *data)
1947{
1948	struct trace_array *tr = data;
1949	struct trace_pid_list *pid_list;
1950	struct trace_pid_list *no_pid_list;
1951
1952	/*
1953	 * This function is called by on_each_cpu() while the
1954	 * event_mutex is held.
1955	 */
1956	pid_list = rcu_dereference_protected(tr->filtered_pids,
1957					     mutex_is_locked(&event_mutex));
1958	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959					     mutex_is_locked(&event_mutex));
1960
1961	this_cpu_write(tr->array_buffer.data->ignore_pid,
1962		       trace_ignore_this_task(pid_list, no_pid_list, current));
1963}
1964
1965static void register_pid_events(struct trace_array *tr)
1966{
1967	/*
1968	 * Register a probe that is called before all other probes
1969	 * to set ignore_pid if next or prev do not match.
1970	 * Register a probe this is called after all other probes
1971	 * to only keep ignore_pid set if next pid matches.
1972	 */
1973	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1974					 tr, INT_MAX);
1975	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1976					 tr, 0);
1977
1978	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1979					 tr, INT_MAX);
1980	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1981					 tr, 0);
1982
1983	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1984					     tr, INT_MAX);
1985	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1986					     tr, 0);
1987
1988	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1989					 tr, INT_MAX);
1990	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1991					 tr, 0);
1992}
1993
1994static ssize_t
1995event_pid_write(struct file *filp, const char __user *ubuf,
1996		size_t cnt, loff_t *ppos, int type)
1997{
1998	struct seq_file *m = filp->private_data;
1999	struct trace_array *tr = m->private;
2000	struct trace_pid_list *filtered_pids = NULL;
2001	struct trace_pid_list *other_pids = NULL;
2002	struct trace_pid_list *pid_list;
2003	struct trace_event_file *file;
2004	ssize_t ret;
 
 
 
 
 
 
2005
2006	if (!cnt)
2007		return 0;
2008
2009	ret = tracing_update_buffers(tr);
2010	if (ret < 0)
2011		return ret;
2012
 
 
 
2013	mutex_lock(&event_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2014
2015	if (type == TRACE_PIDS) {
2016		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017							  lockdep_is_held(&event_mutex));
2018		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019							  lockdep_is_held(&event_mutex));
2020	} else {
2021		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022							  lockdep_is_held(&event_mutex));
2023		other_pids = rcu_dereference_protected(tr->filtered_pids,
2024							  lockdep_is_held(&event_mutex));
 
2025	}
2026
2027	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2028	if (ret < 0)
2029		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2030
2031	if (type == TRACE_PIDS)
2032		rcu_assign_pointer(tr->filtered_pids, pid_list);
2033	else
2034		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2035
2036	list_for_each_entry(file, &tr->events, list) {
2037		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2038	}
2039
2040	if (filtered_pids) {
2041		tracepoint_synchronize_unregister();
2042		trace_pid_list_free(filtered_pids);
2043	} else if (pid_list && !other_pids) {
2044		register_pid_events(tr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2045	}
2046
2047	/*
2048	 * Ignoring of pids is done at task switch. But we have to
2049	 * check for those tasks that are currently running.
2050	 * Always do this in case a pid was appended or removed.
2051	 */
2052	on_each_cpu(ignore_task_cpu, tr, 1);
2053
2054 out:
2055	mutex_unlock(&event_mutex);
2056
2057	if (ret > 0)
2058		*ppos += ret;
2059
2060	return ret;
2061}
2062
2063static ssize_t
2064ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065		       size_t cnt, loff_t *ppos)
2066{
2067	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2068}
2069
2070static ssize_t
2071ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072			size_t cnt, loff_t *ppos)
2073{
2074	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2075}
2076
2077static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081static int ftrace_event_release(struct inode *inode, struct file *file);
2082
2083static const struct seq_operations show_event_seq_ops = {
2084	.start = t_start,
2085	.next = t_next,
2086	.show = t_show,
2087	.stop = t_stop,
2088};
2089
2090static const struct seq_operations show_set_event_seq_ops = {
2091	.start = s_start,
2092	.next = s_next,
2093	.show = t_show,
2094	.stop = t_stop,
2095};
2096
2097static const struct seq_operations show_set_pid_seq_ops = {
2098	.start = p_start,
2099	.next = p_next,
2100	.show = trace_pid_show,
2101	.stop = p_stop,
2102};
2103
2104static const struct seq_operations show_set_no_pid_seq_ops = {
2105	.start = np_start,
2106	.next = np_next,
2107	.show = trace_pid_show,
2108	.stop = p_stop,
2109};
2110
2111static const struct file_operations ftrace_avail_fops = {
2112	.open = ftrace_event_avail_open,
2113	.read = seq_read,
2114	.llseek = seq_lseek,
2115	.release = seq_release,
2116};
2117
2118static const struct file_operations ftrace_set_event_fops = {
2119	.open = ftrace_event_set_open,
2120	.read = seq_read,
2121	.write = ftrace_event_write,
2122	.llseek = seq_lseek,
2123	.release = ftrace_event_release,
2124};
2125
2126static const struct file_operations ftrace_set_event_pid_fops = {
2127	.open = ftrace_event_set_pid_open,
2128	.read = seq_read,
2129	.write = ftrace_event_pid_write,
2130	.llseek = seq_lseek,
2131	.release = ftrace_event_release,
2132};
2133
2134static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135	.open = ftrace_event_set_npid_open,
2136	.read = seq_read,
2137	.write = ftrace_event_npid_write,
2138	.llseek = seq_lseek,
2139	.release = ftrace_event_release,
2140};
2141
2142static const struct file_operations ftrace_enable_fops = {
2143	.open = tracing_open_file_tr,
2144	.read = event_enable_read,
2145	.write = event_enable_write,
2146	.release = tracing_release_file_tr,
2147	.llseek = default_llseek,
2148};
2149
2150static const struct file_operations ftrace_event_format_fops = {
2151	.open = trace_format_open,
2152	.read = seq_read,
2153	.llseek = seq_lseek,
2154	.release = seq_release,
2155};
2156
2157#ifdef CONFIG_PERF_EVENTS
2158static const struct file_operations ftrace_event_id_fops = {
2159	.read = event_id_read,
2160	.llseek = default_llseek,
2161};
2162#endif
2163
2164static const struct file_operations ftrace_event_filter_fops = {
2165	.open = tracing_open_file_tr,
2166	.read = event_filter_read,
2167	.write = event_filter_write,
2168	.release = tracing_release_file_tr,
2169	.llseek = default_llseek,
2170};
2171
2172static const struct file_operations ftrace_subsystem_filter_fops = {
2173	.open = subsystem_open,
2174	.read = subsystem_filter_read,
2175	.write = subsystem_filter_write,
2176	.llseek = default_llseek,
2177	.release = subsystem_release,
2178};
2179
2180static const struct file_operations ftrace_system_enable_fops = {
2181	.open = subsystem_open,
2182	.read = system_enable_read,
2183	.write = system_enable_write,
2184	.llseek = default_llseek,
2185	.release = subsystem_release,
2186};
2187
2188static const struct file_operations ftrace_tr_enable_fops = {
2189	.open = system_tr_open,
2190	.read = system_enable_read,
2191	.write = system_enable_write,
2192	.llseek = default_llseek,
2193	.release = subsystem_release,
2194};
2195
2196static const struct file_operations ftrace_show_header_page_fops = {
2197	.open = tracing_open_generic_tr,
2198	.read = show_header_page_file,
2199	.llseek = default_llseek,
2200	.release = tracing_release_generic_tr,
2201};
2202
2203static const struct file_operations ftrace_show_header_event_fops = {
2204	.open = tracing_open_generic_tr,
2205	.read = show_header_event_file,
2206	.llseek = default_llseek,
2207	.release = tracing_release_generic_tr,
2208};
2209
2210static int
2211ftrace_event_open(struct inode *inode, struct file *file,
2212		  const struct seq_operations *seq_ops)
2213{
2214	struct seq_file *m;
2215	int ret;
2216
2217	ret = security_locked_down(LOCKDOWN_TRACEFS);
2218	if (ret)
2219		return ret;
2220
2221	ret = seq_open(file, seq_ops);
2222	if (ret < 0)
2223		return ret;
2224	m = file->private_data;
2225	/* copy tr over to seq ops */
2226	m->private = inode->i_private;
2227
2228	return ret;
2229}
2230
2231static int ftrace_event_release(struct inode *inode, struct file *file)
2232{
2233	struct trace_array *tr = inode->i_private;
2234
2235	trace_array_put(tr);
2236
2237	return seq_release(inode, file);
2238}
2239
2240static int
2241ftrace_event_avail_open(struct inode *inode, struct file *file)
2242{
2243	const struct seq_operations *seq_ops = &show_event_seq_ops;
2244
2245	/* Checks for tracefs lockdown */
2246	return ftrace_event_open(inode, file, seq_ops);
2247}
2248
2249static int
2250ftrace_event_set_open(struct inode *inode, struct file *file)
2251{
2252	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253	struct trace_array *tr = inode->i_private;
2254	int ret;
2255
2256	ret = tracing_check_open_get_tr(tr);
2257	if (ret)
2258		return ret;
2259
2260	if ((file->f_mode & FMODE_WRITE) &&
2261	    (file->f_flags & O_TRUNC))
2262		ftrace_clear_events(tr);
2263
2264	ret = ftrace_event_open(inode, file, seq_ops);
2265	if (ret < 0)
2266		trace_array_put(tr);
2267	return ret;
2268}
2269
2270static int
2271ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2272{
2273	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274	struct trace_array *tr = inode->i_private;
2275	int ret;
2276
2277	ret = tracing_check_open_get_tr(tr);
2278	if (ret)
2279		return ret;
2280
2281	if ((file->f_mode & FMODE_WRITE) &&
2282	    (file->f_flags & O_TRUNC))
2283		ftrace_clear_event_pids(tr, TRACE_PIDS);
2284
2285	ret = ftrace_event_open(inode, file, seq_ops);
2286	if (ret < 0)
2287		trace_array_put(tr);
2288	return ret;
2289}
2290
2291static int
2292ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2293{
2294	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295	struct trace_array *tr = inode->i_private;
2296	int ret;
2297
2298	ret = tracing_check_open_get_tr(tr);
2299	if (ret)
2300		return ret;
2301
2302	if ((file->f_mode & FMODE_WRITE) &&
2303	    (file->f_flags & O_TRUNC))
2304		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2305
2306	ret = ftrace_event_open(inode, file, seq_ops);
2307	if (ret < 0)
2308		trace_array_put(tr);
2309	return ret;
2310}
2311
2312static struct event_subsystem *
2313create_new_subsystem(const char *name)
2314{
2315	struct event_subsystem *system;
2316
2317	/* need to create new entry */
2318	system = kmalloc(sizeof(*system), GFP_KERNEL);
2319	if (!system)
2320		return NULL;
2321
2322	system->ref_count = 1;
2323
2324	/* Only allocate if dynamic (kprobes and modules) */
2325	system->name = kstrdup_const(name, GFP_KERNEL);
2326	if (!system->name)
2327		goto out_free;
2328
 
 
2329	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330	if (!system->filter)
2331		goto out_free;
2332
2333	list_add(&system->list, &event_subsystems);
2334
2335	return system;
2336
2337 out_free:
2338	kfree_const(system->name);
2339	kfree(system);
2340	return NULL;
2341}
2342
2343static int system_callback(const char *name, umode_t *mode, void **data,
2344		    const struct file_operations **fops)
2345{
2346	if (strcmp(name, "filter") == 0)
2347		*fops = &ftrace_subsystem_filter_fops;
2348
2349	else if (strcmp(name, "enable") == 0)
2350		*fops = &ftrace_system_enable_fops;
2351
2352	else
2353		return 0;
2354
2355	*mode = TRACE_MODE_WRITE;
2356	return 1;
2357}
2358
2359static struct eventfs_inode *
2360event_subsystem_dir(struct trace_array *tr, const char *name,
2361		    struct trace_event_file *file, struct eventfs_inode *parent)
2362{
2363	struct event_subsystem *system, *iter;
2364	struct trace_subsystem_dir *dir;
2365	struct eventfs_inode *ei;
2366	int nr_entries;
2367	static struct eventfs_entry system_entries[] = {
2368		{
2369			.name		= "filter",
2370			.callback	= system_callback,
2371		},
2372		{
2373			.name		= "enable",
2374			.callback	= system_callback,
2375		}
2376	};
2377
2378	/* First see if we did not already create this dir */
2379	list_for_each_entry(dir, &tr->systems, list) {
2380		system = dir->subsystem;
2381		if (strcmp(system->name, name) == 0) {
2382			dir->nr_events++;
2383			file->system = dir;
2384			return dir->ei;
2385		}
2386	}
2387
2388	/* Now see if the system itself exists. */
2389	system = NULL;
2390	list_for_each_entry(iter, &event_subsystems, list) {
2391		if (strcmp(iter->name, name) == 0) {
2392			system = iter;
2393			break;
2394		}
2395	}
 
 
 
2396
2397	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2398	if (!dir)
2399		goto out_fail;
2400
2401	if (!system) {
2402		system = create_new_subsystem(name);
2403		if (!system)
2404			goto out_free;
2405	} else
2406		__get_system(system);
2407
2408	/* ftrace only has directories no files */
2409	if (strcmp(name, "ftrace") == 0)
2410		nr_entries = 0;
2411	else
2412		nr_entries = ARRAY_SIZE(system_entries);
2413
2414	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2415	if (IS_ERR(ei)) {
2416		pr_warn("Failed to create system directory %s\n", name);
2417		__put_system(system);
2418		goto out_free;
2419	}
2420
2421	dir->ei = ei;
2422	dir->tr = tr;
2423	dir->ref_count = 1;
2424	dir->nr_events = 1;
2425	dir->subsystem = system;
2426	file->system = dir;
2427
 
 
 
 
 
 
 
 
 
 
 
2428	list_add(&dir->list, &tr->systems);
2429
2430	return dir->ei;
2431
2432 out_free:
2433	kfree(dir);
2434 out_fail:
2435	/* Only print this message if failed on memory allocation */
2436	if (!dir || !system)
2437		pr_warn("No memory to create event subsystem %s\n", name);
2438	return NULL;
2439}
2440
2441static int
2442event_define_fields(struct trace_event_call *call)
2443{
 
 
2444	struct list_head *head;
2445	int ret = 0;
 
 
2446
2447	/*
2448	 * Other events may have the same class. Only update
2449	 * the fields if they are not already defined.
2450	 */
2451	head = trace_get_fields(call);
2452	if (list_empty(head)) {
2453		struct trace_event_fields *field = call->class->fields_array;
2454		unsigned int offset = sizeof(struct trace_entry);
 
 
2455
2456		for (; field->type; field++) {
2457			if (field->type == TRACE_FUNCTION_TYPE) {
2458				field->define_fields(call);
2459				break;
2460			}
2461
2462			offset = ALIGN(offset, field->align);
2463			ret = trace_define_field_ext(call, field->type, field->name,
2464						 offset, field->size,
2465						 field->is_signed, field->filter_type,
2466						 field->len);
2467			if (WARN_ON_ONCE(ret)) {
2468				pr_err("error code is %d\n", ret);
2469				break;
2470			}
2471
2472			offset += field->size;
2473		}
2474	}
2475
2476	return ret;
2477}
 
2478
2479static int event_callback(const char *name, umode_t *mode, void **data,
2480			  const struct file_operations **fops)
2481{
2482	struct trace_event_file *file = *data;
2483	struct trace_event_call *call = file->event_call;
2484
2485	if (strcmp(name, "format") == 0) {
2486		*mode = TRACE_MODE_READ;
2487		*fops = &ftrace_event_format_fops;
2488		*data = call;
2489		return 1;
2490	}
2491
2492	/*
2493	 * Only event directories that can be enabled should have
2494	 * triggers or filters, with the exception of the "print"
2495	 * event that can have a "trigger" file.
2496	 */
2497	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498		if (call->class->reg && strcmp(name, "enable") == 0) {
2499			*mode = TRACE_MODE_WRITE;
2500			*fops = &ftrace_enable_fops;
2501			return 1;
 
 
2502		}
2503
2504		if (strcmp(name, "filter") == 0) {
2505			*mode = TRACE_MODE_WRITE;
2506			*fops = &ftrace_event_filter_fops;
2507			return 1;
2508		}
2509	}
2510
2511	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512	    strcmp(trace_event_name(call), "print") == 0) {
2513		if (strcmp(name, "trigger") == 0) {
2514			*mode = TRACE_MODE_WRITE;
2515			*fops = &event_trigger_fops;
2516			return 1;
2517		}
2518	}
2519
2520#ifdef CONFIG_PERF_EVENTS
2521	if (call->event.type && call->class->reg &&
2522	    strcmp(name, "id") == 0) {
2523		*mode = TRACE_MODE_READ;
2524		*data = (void *)(long)call->event.type;
2525		*fops = &ftrace_event_id_fops;
2526		return 1;
2527	}
2528#endif
2529
2530#ifdef CONFIG_HIST_TRIGGERS
2531	if (strcmp(name, "hist") == 0) {
2532		*mode = TRACE_MODE_READ;
2533		*fops = &event_hist_fops;
2534		return 1;
2535	}
2536#endif
2537#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538	if (strcmp(name, "hist_debug") == 0) {
2539		*mode = TRACE_MODE_READ;
2540		*fops = &event_hist_debug_fops;
2541		return 1;
2542	}
2543#endif
2544#ifdef CONFIG_TRACE_EVENT_INJECT
2545	if (call->event.type && call->class->reg &&
2546	    strcmp(name, "inject") == 0) {
2547		*mode = 0200;
2548		*fops = &event_inject_fops;
2549		return 1;
2550	}
2551#endif
2552	return 0;
2553}
2554
2555/* The file is incremented on creation and freeing the enable file decrements it */
2556static void event_release(const char *name, void *data)
2557{
2558	struct trace_event_file *file = data;
2559
2560	event_file_put(file);
2561}
2562
2563static int
2564event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2565{
2566	struct trace_event_call *call = file->event_call;
2567	struct trace_array *tr = file->tr;
2568	struct eventfs_inode *e_events;
2569	struct eventfs_inode *ei;
2570	const char *name;
2571	int nr_entries;
2572	int ret;
2573	static struct eventfs_entry event_entries[] = {
2574		{
2575			.name		= "enable",
2576			.callback	= event_callback,
2577			.release	= event_release,
2578		},
2579		{
2580			.name		= "filter",
2581			.callback	= event_callback,
2582		},
2583		{
2584			.name		= "trigger",
2585			.callback	= event_callback,
2586		},
2587		{
2588			.name		= "format",
2589			.callback	= event_callback,
2590		},
2591#ifdef CONFIG_PERF_EVENTS
2592		{
2593			.name		= "id",
2594			.callback	= event_callback,
2595		},
2596#endif
2597#ifdef CONFIG_HIST_TRIGGERS
2598		{
2599			.name		= "hist",
2600			.callback	= event_callback,
2601		},
2602#endif
2603#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2604		{
2605			.name		= "hist_debug",
2606			.callback	= event_callback,
2607		},
2608#endif
2609#ifdef CONFIG_TRACE_EVENT_INJECT
2610		{
2611			.name		= "inject",
2612			.callback	= event_callback,
2613		},
2614#endif
2615	};
2616
2617	/*
2618	 * If the trace point header did not define TRACE_SYSTEM
2619	 * then the system would be called "TRACE_SYSTEM". This should
2620	 * never happen.
2621	 */
2622	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2623		return -ENODEV;
2624
2625	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2626	if (!e_events)
2627		return -ENOMEM;
2628
2629	nr_entries = ARRAY_SIZE(event_entries);
2630
2631	name = trace_event_name(call);
2632	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2633	if (IS_ERR(ei)) {
2634		pr_warn("Could not create tracefs '%s' directory\n", name);
2635		return -1;
2636	}
2637
2638	file->ei = ei;
2639
2640	ret = event_define_fields(call);
2641	if (ret < 0) {
2642		pr_warn("Could not initialize trace point events/%s\n", name);
2643		return ret;
2644	}
2645
2646	/* Gets decremented on freeing of the "enable" file */
2647	event_file_get(file);
2648
2649	return 0;
2650}
2651
2652static void remove_event_from_tracers(struct trace_event_call *call)
2653{
2654	struct trace_event_file *file;
2655	struct trace_array *tr;
2656
2657	do_for_each_event_file_safe(tr, file) {
2658		if (file->event_call != call)
2659			continue;
2660
2661		remove_event_file_dir(file);
2662		/*
2663		 * The do_for_each_event_file_safe() is
2664		 * a double loop. After finding the call for this
2665		 * trace_array, we use break to jump to the next
2666		 * trace_array.
2667		 */
2668		break;
2669	} while_for_each_event_file();
2670}
2671
2672static void event_remove(struct trace_event_call *call)
2673{
2674	struct trace_array *tr;
2675	struct trace_event_file *file;
2676
2677	do_for_each_event_file(tr, file) {
2678		if (file->event_call != call)
2679			continue;
2680
2681		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2682			tr->clear_trace = true;
2683
2684		ftrace_event_enable_disable(file, 0);
2685		/*
2686		 * The do_for_each_event_file() is
2687		 * a double loop. After finding the call for this
2688		 * trace_array, we use break to jump to the next
2689		 * trace_array.
2690		 */
2691		break;
2692	} while_for_each_event_file();
2693
2694	if (call->event.funcs)
2695		__unregister_trace_event(&call->event);
2696	remove_event_from_tracers(call);
2697	list_del(&call->list);
2698}
2699
2700static int event_init(struct trace_event_call *call)
2701{
2702	int ret = 0;
2703	const char *name;
2704
2705	name = trace_event_name(call);
2706	if (WARN_ON(!name))
2707		return -EINVAL;
2708
2709	if (call->class->raw_init) {
2710		ret = call->class->raw_init(call);
2711		if (ret < 0 && ret != -ENOSYS)
2712			pr_warn("Could not initialize trace events/%s\n", name);
2713	}
2714
2715	return ret;
2716}
2717
2718static int
2719__register_event(struct trace_event_call *call, struct module *mod)
2720{
2721	int ret;
2722
2723	ret = event_init(call);
2724	if (ret < 0)
2725		return ret;
2726
2727	list_add(&call->list, &ftrace_events);
2728	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2729		atomic_set(&call->refcnt, 0);
2730	else
2731		call->module = mod;
2732
2733	return 0;
2734}
2735
2736static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2737{
2738	int rlen;
2739	int elen;
2740
2741	/* Find the length of the eval value as a string */
2742	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2743	/* Make sure there's enough room to replace the string with the value */
2744	if (len < elen)
2745		return NULL;
2746
2747	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2748
2749	/* Get the rest of the string of ptr */
2750	rlen = strlen(ptr + len);
2751	memmove(ptr + elen, ptr + len, rlen);
2752	/* Make sure we end the new string */
2753	ptr[elen + rlen] = 0;
2754
2755	return ptr + elen;
2756}
2757
2758static void update_event_printk(struct trace_event_call *call,
2759				struct trace_eval_map *map)
2760{
2761	char *ptr;
2762	int quote = 0;
2763	int len = strlen(map->eval_string);
2764
2765	for (ptr = call->print_fmt; *ptr; ptr++) {
2766		if (*ptr == '\\') {
2767			ptr++;
2768			/* paranoid */
2769			if (!*ptr)
2770				break;
2771			continue;
2772		}
2773		if (*ptr == '"') {
2774			quote ^= 1;
2775			continue;
2776		}
2777		if (quote)
2778			continue;
2779		if (isdigit(*ptr)) {
2780			/* skip numbers */
2781			do {
2782				ptr++;
2783				/* Check for alpha chars like ULL */
2784			} while (isalnum(*ptr));
2785			if (!*ptr)
2786				break;
2787			/*
2788			 * A number must have some kind of delimiter after
2789			 * it, and we can ignore that too.
2790			 */
2791			continue;
2792		}
2793		if (isalpha(*ptr) || *ptr == '_') {
2794			if (strncmp(map->eval_string, ptr, len) == 0 &&
2795			    !isalnum(ptr[len]) && ptr[len] != '_') {
2796				ptr = eval_replace(ptr, map, len);
2797				/* enum/sizeof string smaller than value */
2798				if (WARN_ON_ONCE(!ptr))
2799					return;
2800				/*
2801				 * No need to decrement here, as eval_replace()
2802				 * returns the pointer to the character passed
2803				 * the eval, and two evals can not be placed
2804				 * back to back without something in between.
2805				 * We can skip that something in between.
2806				 */
2807				continue;
2808			}
2809		skip_more:
2810			do {
2811				ptr++;
2812			} while (isalnum(*ptr) || *ptr == '_');
2813			if (!*ptr)
2814				break;
2815			/*
2816			 * If what comes after this variable is a '.' or
2817			 * '->' then we can continue to ignore that string.
2818			 */
2819			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2820				ptr += *ptr == '.' ? 1 : 2;
2821				if (!*ptr)
2822					break;
2823				goto skip_more;
2824			}
2825			/*
2826			 * Once again, we can skip the delimiter that came
2827			 * after the string.
2828			 */
2829			continue;
2830		}
2831	}
2832}
2833
2834static void add_str_to_module(struct module *module, char *str)
2835{
2836	struct module_string *modstr;
2837
2838	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2839
2840	/*
2841	 * If we failed to allocate memory here, then we'll just
2842	 * let the str memory leak when the module is removed.
2843	 * If this fails to allocate, there's worse problems than
2844	 * a leaked string on module removal.
2845	 */
2846	if (WARN_ON_ONCE(!modstr))
2847		return;
2848
2849	modstr->module = module;
2850	modstr->str = str;
2851
2852	list_add(&modstr->next, &module_strings);
2853}
2854
2855static void update_event_fields(struct trace_event_call *call,
2856				struct trace_eval_map *map)
2857{
2858	struct ftrace_event_field *field;
2859	struct list_head *head;
2860	char *ptr;
2861	char *str;
2862	int len = strlen(map->eval_string);
2863
2864	/* Dynamic events should never have field maps */
2865	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2866		return;
2867
2868	head = trace_get_fields(call);
2869	list_for_each_entry(field, head, link) {
2870		ptr = strchr(field->type, '[');
2871		if (!ptr)
2872			continue;
2873		ptr++;
2874
2875		if (!isalpha(*ptr) && *ptr != '_')
2876			continue;
2877
2878		if (strncmp(map->eval_string, ptr, len) != 0)
2879			continue;
2880
2881		str = kstrdup(field->type, GFP_KERNEL);
2882		if (WARN_ON_ONCE(!str))
2883			return;
2884		ptr = str + (ptr - field->type);
2885		ptr = eval_replace(ptr, map, len);
2886		/* enum/sizeof string smaller than value */
2887		if (WARN_ON_ONCE(!ptr)) {
2888			kfree(str);
2889			continue;
2890		}
2891
2892		/*
2893		 * If the event is part of a module, then we need to free the string
2894		 * when the module is removed. Otherwise, it will stay allocated
2895		 * until a reboot.
2896		 */
2897		if (call->module)
2898			add_str_to_module(call->module, str);
2899
2900		field->type = str;
2901	}
2902}
2903
2904void trace_event_eval_update(struct trace_eval_map **map, int len)
2905{
2906	struct trace_event_call *call, *p;
2907	const char *last_system = NULL;
2908	bool first = false;
2909	int last_i;
2910	int i;
2911
2912	down_write(&trace_event_sem);
2913	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2914		/* events are usually grouped together with systems */
2915		if (!last_system || call->class->system != last_system) {
2916			first = true;
2917			last_i = 0;
2918			last_system = call->class->system;
2919		}
2920
2921		/*
2922		 * Since calls are grouped by systems, the likelihood that the
2923		 * next call in the iteration belongs to the same system as the
2924		 * previous call is high. As an optimization, we skip searching
2925		 * for a map[] that matches the call's system if the last call
2926		 * was from the same system. That's what last_i is for. If the
2927		 * call has the same system as the previous call, then last_i
2928		 * will be the index of the first map[] that has a matching
2929		 * system.
2930		 */
2931		for (i = last_i; i < len; i++) {
2932			if (call->class->system == map[i]->system) {
2933				/* Save the first system if need be */
2934				if (first) {
2935					last_i = i;
2936					first = false;
2937				}
2938				update_event_printk(call, map[i]);
2939				update_event_fields(call, map[i]);
2940			}
2941		}
2942		cond_resched();
2943	}
2944	up_write(&trace_event_sem);
2945}
2946
2947static bool event_in_systems(struct trace_event_call *call,
2948			     const char *systems)
2949{
2950	const char *system;
2951	const char *p;
2952
2953	if (!systems)
2954		return true;
2955
2956	system = call->class->system;
2957	p = strstr(systems, system);
2958	if (!p)
2959		return false;
2960
2961	if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2962		return false;
2963
2964	p += strlen(system);
2965	return !*p || isspace(*p) || *p == ',';
2966}
2967
2968static struct trace_event_file *
2969trace_create_new_event(struct trace_event_call *call,
2970		       struct trace_array *tr)
2971{
2972	struct trace_pid_list *no_pid_list;
2973	struct trace_pid_list *pid_list;
2974	struct trace_event_file *file;
2975	unsigned int first;
2976
2977	if (!event_in_systems(call, tr->system_names))
2978		return NULL;
2979
2980	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2981	if (!file)
2982		return ERR_PTR(-ENOMEM);
2983
2984	pid_list = rcu_dereference_protected(tr->filtered_pids,
2985					     lockdep_is_held(&event_mutex));
2986	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2987					     lockdep_is_held(&event_mutex));
2988
2989	if (!trace_pid_list_first(pid_list, &first) ||
2990	    !trace_pid_list_first(no_pid_list, &first))
2991		file->flags |= EVENT_FILE_FL_PID_FILTER;
2992
2993	file->event_call = call;
2994	file->tr = tr;
2995	atomic_set(&file->sm_ref, 0);
2996	atomic_set(&file->tm_ref, 0);
2997	INIT_LIST_HEAD(&file->triggers);
2998	list_add(&file->list, &tr->events);
2999	event_file_get(file);
3000
3001	return file;
3002}
3003
3004#define MAX_BOOT_TRIGGERS 32
3005
3006static struct boot_triggers {
3007	const char		*event;
3008	char			*trigger;
3009} bootup_triggers[MAX_BOOT_TRIGGERS];
3010
3011static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3012static int nr_boot_triggers;
3013
3014static __init int setup_trace_triggers(char *str)
3015{
3016	char *trigger;
3017	char *buf;
3018	int i;
3019
3020	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3021	trace_set_ring_buffer_expanded(NULL);
3022	disable_tracing_selftest("running event triggers");
3023
3024	buf = bootup_trigger_buf;
3025	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3026		trigger = strsep(&buf, ",");
3027		if (!trigger)
3028			break;
3029		bootup_triggers[i].event = strsep(&trigger, ".");
3030		bootup_triggers[i].trigger = trigger;
3031		if (!bootup_triggers[i].trigger)
3032			break;
3033	}
3034
3035	nr_boot_triggers = i;
3036	return 1;
3037}
3038__setup("trace_trigger=", setup_trace_triggers);
3039
3040/* Add an event to a trace directory */
3041static int
3042__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3043{
3044	struct trace_event_file *file;
3045
3046	file = trace_create_new_event(call, tr);
3047	/*
3048	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3049	 * allocation, or NULL if the event is not part of the tr->system_names.
3050	 * When the event is not part of the tr->system_names, return zero, not
3051	 * an error.
3052	 */
3053	if (!file)
3054		return 0;
3055
3056	if (IS_ERR(file))
3057		return PTR_ERR(file);
3058
3059	if (eventdir_initialized)
3060		return event_create_dir(tr->event_dir, file);
3061	else
3062		return event_define_fields(call);
3063}
3064
3065static void trace_early_triggers(struct trace_event_file *file, const char *name)
3066{
3067	int ret;
3068	int i;
3069
3070	for (i = 0; i < nr_boot_triggers; i++) {
3071		if (strcmp(name, bootup_triggers[i].event))
3072			continue;
3073		mutex_lock(&event_mutex);
3074		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3075		mutex_unlock(&event_mutex);
3076		if (ret)
3077			pr_err("Failed to register trigger '%s' on event %s\n",
3078			       bootup_triggers[i].trigger,
3079			       bootup_triggers[i].event);
3080	}
3081}
3082
3083/*
3084 * Just create a descriptor for early init. A descriptor is required
3085 * for enabling events at boot. We want to enable events before
3086 * the filesystem is initialized.
3087 */
3088static int
3089__trace_early_add_new_event(struct trace_event_call *call,
3090			    struct trace_array *tr)
3091{
3092	struct trace_event_file *file;
3093	int ret;
3094
3095	file = trace_create_new_event(call, tr);
3096	/*
3097	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3098	 * allocation, or NULL if the event is not part of the tr->system_names.
3099	 * When the event is not part of the tr->system_names, return zero, not
3100	 * an error.
3101	 */
3102	if (!file)
3103		return 0;
3104
3105	if (IS_ERR(file))
3106		return PTR_ERR(file);
3107
3108	ret = event_define_fields(call);
3109	if (ret)
3110		return ret;
3111
3112	trace_early_triggers(file, trace_event_name(call));
3113
3114	return 0;
3115}
3116
3117struct ftrace_module_file_ops;
3118static void __add_event_to_tracers(struct trace_event_call *call);
3119
3120/* Add an additional event_call dynamically */
3121int trace_add_event_call(struct trace_event_call *call)
3122{
3123	int ret;
3124	lockdep_assert_held(&event_mutex);
3125
3126	mutex_lock(&trace_types_lock);
 
3127
3128	ret = __register_event(call, NULL);
3129	if (ret >= 0)
3130		__add_event_to_tracers(call);
3131
 
3132	mutex_unlock(&trace_types_lock);
3133	return ret;
3134}
3135EXPORT_SYMBOL_GPL(trace_add_event_call);
3136
3137/*
3138 * Must be called under locking of trace_types_lock, event_mutex and
3139 * trace_event_sem.
3140 */
3141static void __trace_remove_event_call(struct trace_event_call *call)
3142{
3143	event_remove(call);
3144	trace_destroy_fields(call);
3145	free_event_filter(call->filter);
3146	call->filter = NULL;
3147}
3148
3149static int probe_remove_event_call(struct trace_event_call *call)
3150{
3151	struct trace_array *tr;
3152	struct trace_event_file *file;
3153
3154#ifdef CONFIG_PERF_EVENTS
3155	if (call->perf_refcount)
3156		return -EBUSY;
3157#endif
3158	do_for_each_event_file(tr, file) {
3159		if (file->event_call != call)
3160			continue;
3161		/*
3162		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3163		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3164		 * TRACE_REG_UNREGISTER.
3165		 */
3166		if (file->flags & EVENT_FILE_FL_ENABLED)
3167			goto busy;
3168
3169		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3170			tr->clear_trace = true;
3171		/*
3172		 * The do_for_each_event_file_safe() is
3173		 * a double loop. After finding the call for this
3174		 * trace_array, we use break to jump to the next
3175		 * trace_array.
3176		 */
3177		break;
3178	} while_for_each_event_file();
3179
3180	__trace_remove_event_call(call);
3181
3182	return 0;
3183 busy:
3184	/* No need to clear the trace now */
3185	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3186		tr->clear_trace = false;
3187	}
3188	return -EBUSY;
3189}
3190
3191/* Remove an event_call */
3192int trace_remove_event_call(struct trace_event_call *call)
3193{
3194	int ret;
3195
3196	lockdep_assert_held(&event_mutex);
3197
3198	mutex_lock(&trace_types_lock);
 
3199	down_write(&trace_event_sem);
3200	ret = probe_remove_event_call(call);
3201	up_write(&trace_event_sem);
 
3202	mutex_unlock(&trace_types_lock);
3203
3204	return ret;
3205}
3206EXPORT_SYMBOL_GPL(trace_remove_event_call);
3207
3208#define for_each_event(event, start, end)			\
3209	for (event = start;					\
3210	     (unsigned long)event < (unsigned long)end;		\
3211	     event++)
3212
3213#ifdef CONFIG_MODULES
3214
3215static void trace_module_add_events(struct module *mod)
3216{
3217	struct trace_event_call **call, **start, **end;
3218
3219	if (!mod->num_trace_events)
3220		return;
3221
3222	/* Don't add infrastructure for mods without tracepoints */
3223	if (trace_module_has_bad_taint(mod)) {
3224		pr_err("%s: module has bad taint, not creating trace events\n",
3225		       mod->name);
3226		return;
3227	}
3228
3229	start = mod->trace_events;
3230	end = mod->trace_events + mod->num_trace_events;
3231
3232	for_each_event(call, start, end) {
3233		__register_event(*call, mod);
3234		__add_event_to_tracers(*call);
3235	}
3236}
3237
3238static void trace_module_remove_events(struct module *mod)
3239{
3240	struct trace_event_call *call, *p;
3241	struct module_string *modstr, *m;
3242
3243	down_write(&trace_event_sem);
3244	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3245		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3246			continue;
3247		if (call->module == mod)
3248			__trace_remove_event_call(call);
3249	}
3250	/* Check for any strings allocade for this module */
3251	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3252		if (modstr->module != mod)
3253			continue;
3254		list_del(&modstr->next);
3255		kfree(modstr->str);
3256		kfree(modstr);
3257	}
3258	up_write(&trace_event_sem);
3259
3260	/*
3261	 * It is safest to reset the ring buffer if the module being unloaded
3262	 * registered any events that were used. The only worry is if
3263	 * a new module gets loaded, and takes on the same id as the events
3264	 * of this module. When printing out the buffer, traced events left
3265	 * over from this module may be passed to the new module events and
3266	 * unexpected results may occur.
3267	 */
3268	tracing_reset_all_online_cpus_unlocked();
 
3269}
3270
3271static int trace_module_notify(struct notifier_block *self,
3272			       unsigned long val, void *data)
3273{
3274	struct module *mod = data;
3275
 
3276	mutex_lock(&event_mutex);
3277	mutex_lock(&trace_types_lock);
3278	switch (val) {
3279	case MODULE_STATE_COMING:
3280		trace_module_add_events(mod);
3281		break;
3282	case MODULE_STATE_GOING:
3283		trace_module_remove_events(mod);
3284		break;
3285	}
 
3286	mutex_unlock(&trace_types_lock);
3287	mutex_unlock(&event_mutex);
3288
3289	return NOTIFY_OK;
3290}
3291
3292static struct notifier_block trace_module_nb = {
3293	.notifier_call = trace_module_notify,
3294	.priority = 1, /* higher than trace.c module notify */
3295};
3296#endif /* CONFIG_MODULES */
3297
3298/* Create a new event directory structure for a trace directory. */
3299static void
3300__trace_add_event_dirs(struct trace_array *tr)
3301{
3302	struct trace_event_call *call;
3303	int ret;
3304
3305	list_for_each_entry(call, &ftrace_events, list) {
3306		ret = __trace_add_new_event(call, tr);
3307		if (ret < 0)
3308			pr_warn("Could not create directory for event %s\n",
3309				trace_event_name(call));
3310	}
3311}
3312
3313/* Returns any file that matches the system and event */
3314struct trace_event_file *
3315__find_event_file(struct trace_array *tr, const char *system, const char *event)
3316{
3317	struct trace_event_file *file;
3318	struct trace_event_call *call;
3319	const char *name;
3320
3321	list_for_each_entry(file, &tr->events, list) {
3322
3323		call = file->event_call;
3324		name = trace_event_name(call);
3325
3326		if (!name || !call->class)
 
 
 
3327			continue;
3328
3329		if (strcmp(event, name) == 0 &&
3330		    strcmp(system, call->class->system) == 0)
3331			return file;
3332	}
3333	return NULL;
3334}
3335
3336/* Returns valid trace event files that match system and event */
3337struct trace_event_file *
3338find_event_file(struct trace_array *tr, const char *system, const char *event)
3339{
3340	struct trace_event_file *file;
3341
3342	file = __find_event_file(tr, system, event);
3343	if (!file || !file->event_call->class->reg ||
3344	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3345		return NULL;
3346
3347	return file;
3348}
3349
3350/**
3351 * trace_get_event_file - Find and return a trace event file
3352 * @instance: The name of the trace instance containing the event
3353 * @system: The name of the system containing the event
3354 * @event: The name of the event
3355 *
3356 * Return a trace event file given the trace instance name, trace
3357 * system, and trace event name.  If the instance name is NULL, it
3358 * refers to the top-level trace array.
3359 *
3360 * This function will look it up and return it if found, after calling
3361 * trace_array_get() to prevent the instance from going away, and
3362 * increment the event's module refcount to prevent it from being
3363 * removed.
3364 *
3365 * To release the file, call trace_put_event_file(), which will call
3366 * trace_array_put() and decrement the event's module refcount.
3367 *
3368 * Return: The trace event on success, ERR_PTR otherwise.
3369 */
3370struct trace_event_file *trace_get_event_file(const char *instance,
3371					      const char *system,
3372					      const char *event)
3373{
3374	struct trace_array *tr = top_trace_array();
3375	struct trace_event_file *file = NULL;
3376	int ret = -EINVAL;
3377
3378	if (instance) {
3379		tr = trace_array_find_get(instance);
3380		if (!tr)
3381			return ERR_PTR(-ENOENT);
3382	} else {
3383		ret = trace_array_get(tr);
3384		if (ret)
3385			return ERR_PTR(ret);
3386	}
3387
3388	mutex_lock(&event_mutex);
3389
3390	file = find_event_file(tr, system, event);
3391	if (!file) {
3392		trace_array_put(tr);
3393		ret = -EINVAL;
3394		goto out;
3395	}
3396
3397	/* Don't let event modules unload while in use */
3398	ret = trace_event_try_get_ref(file->event_call);
3399	if (!ret) {
3400		trace_array_put(tr);
3401		ret = -EBUSY;
3402		goto out;
3403	}
3404
3405	ret = 0;
3406 out:
3407	mutex_unlock(&event_mutex);
3408
3409	if (ret)
3410		file = ERR_PTR(ret);
3411
3412	return file;
3413}
3414EXPORT_SYMBOL_GPL(trace_get_event_file);
3415
3416/**
3417 * trace_put_event_file - Release a file from trace_get_event_file()
3418 * @file: The trace event file
3419 *
3420 * If a file was retrieved using trace_get_event_file(), this should
3421 * be called when it's no longer needed.  It will cancel the previous
3422 * trace_array_get() called by that function, and decrement the
3423 * event's module refcount.
3424 */
3425void trace_put_event_file(struct trace_event_file *file)
3426{
3427	mutex_lock(&event_mutex);
3428	trace_event_put_ref(file->event_call);
3429	mutex_unlock(&event_mutex);
3430
3431	trace_array_put(file->tr);
3432}
3433EXPORT_SYMBOL_GPL(trace_put_event_file);
3434
3435#ifdef CONFIG_DYNAMIC_FTRACE
3436
3437/* Avoid typos */
3438#define ENABLE_EVENT_STR	"enable_event"
3439#define DISABLE_EVENT_STR	"disable_event"
3440
3441struct event_probe_data {
3442	struct trace_event_file	*file;
3443	unsigned long			count;
3444	int				ref;
3445	bool				enable;
3446};
3447
3448static void update_event_probe(struct event_probe_data *data)
 
3449{
 
 
 
 
 
 
3450	if (data->enable)
3451		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3452	else
3453		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3454}
3455
3456static void
3457event_enable_probe(unsigned long ip, unsigned long parent_ip,
3458		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3459		   void *data)
3460{
3461	struct ftrace_func_mapper *mapper = data;
3462	struct event_probe_data *edata;
3463	void **pdata;
3464
3465	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3466	if (!pdata || !*pdata)
3467		return;
3468
3469	edata = *pdata;
3470	update_event_probe(edata);
3471}
3472
3473static void
3474event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3475			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3476			 void *data)
3477{
3478	struct ftrace_func_mapper *mapper = data;
3479	struct event_probe_data *edata;
3480	void **pdata;
3481
3482	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3483	if (!pdata || !*pdata)
3484		return;
3485
3486	edata = *pdata;
3487
3488	if (!edata->count)
3489		return;
3490
3491	/* Skip if the event is in a state we want to switch to */
3492	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3493		return;
3494
3495	if (edata->count != -1)
3496		(edata->count)--;
3497
3498	update_event_probe(edata);
3499}
3500
3501static int
3502event_enable_print(struct seq_file *m, unsigned long ip,
3503		   struct ftrace_probe_ops *ops, void *data)
3504{
3505	struct ftrace_func_mapper *mapper = data;
3506	struct event_probe_data *edata;
3507	void **pdata;
3508
3509	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3510
3511	if (WARN_ON_ONCE(!pdata || !*pdata))
3512		return 0;
3513
3514	edata = *pdata;
3515
3516	seq_printf(m, "%ps:", (void *)ip);
3517
3518	seq_printf(m, "%s:%s:%s",
3519		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3520		   edata->file->event_call->class->system,
3521		   trace_event_name(edata->file->event_call));
3522
3523	if (edata->count == -1)
3524		seq_puts(m, ":unlimited\n");
3525	else
3526		seq_printf(m, ":count=%ld\n", edata->count);
3527
3528	return 0;
3529}
3530
3531static int
3532event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3533		  unsigned long ip, void *init_data, void **data)
3534{
3535	struct ftrace_func_mapper *mapper = *data;
3536	struct event_probe_data *edata = init_data;
3537	int ret;
3538
3539	if (!mapper) {
3540		mapper = allocate_ftrace_func_mapper();
3541		if (!mapper)
3542			return -ENODEV;
3543		*data = mapper;
3544	}
3545
3546	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3547	if (ret < 0)
3548		return ret;
3549
3550	edata->ref++;
3551
3552	return 0;
3553}
3554
3555static int free_probe_data(void *data)
3556{
3557	struct event_probe_data *edata = data;
 
3558
3559	edata->ref--;
3560	if (!edata->ref) {
3561		/* Remove the SOFT_MODE flag */
3562		__ftrace_event_enable_disable(edata->file, 0, 1);
3563		trace_event_put_ref(edata->file->event_call);
3564		kfree(edata);
3565	}
3566	return 0;
3567}
3568
3569static void
3570event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3571		  unsigned long ip, void *data)
3572{
3573	struct ftrace_func_mapper *mapper = data;
3574	struct event_probe_data *edata;
3575
3576	if (!ip) {
3577		if (!mapper)
3578			return;
3579		free_ftrace_func_mapper(mapper, free_probe_data);
3580		return;
 
 
 
 
 
 
 
3581	}
3582
3583	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3584
3585	if (WARN_ON_ONCE(!edata))
3586		return;
3587
3588	if (WARN_ON_ONCE(edata->ref <= 0))
3589		return;
3590
3591	free_probe_data(edata);
3592}
3593
3594static struct ftrace_probe_ops event_enable_probe_ops = {
3595	.func			= event_enable_probe,
3596	.print			= event_enable_print,
3597	.init			= event_enable_init,
3598	.free			= event_enable_free,
3599};
3600
3601static struct ftrace_probe_ops event_enable_count_probe_ops = {
3602	.func			= event_enable_count_probe,
3603	.print			= event_enable_print,
3604	.init			= event_enable_init,
3605	.free			= event_enable_free,
3606};
3607
3608static struct ftrace_probe_ops event_disable_probe_ops = {
3609	.func			= event_enable_probe,
3610	.print			= event_enable_print,
3611	.init			= event_enable_init,
3612	.free			= event_enable_free,
3613};
3614
3615static struct ftrace_probe_ops event_disable_count_probe_ops = {
3616	.func			= event_enable_count_probe,
3617	.print			= event_enable_print,
3618	.init			= event_enable_init,
3619	.free			= event_enable_free,
3620};
3621
3622static int
3623event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3624		  char *glob, char *cmd, char *param, int enabled)
3625{
 
3626	struct trace_event_file *file;
3627	struct ftrace_probe_ops *ops;
3628	struct event_probe_data *data;
3629	const char *system;
3630	const char *event;
3631	char *number;
3632	bool enable;
3633	int ret;
3634
3635	if (!tr)
3636		return -ENODEV;
3637
3638	/* hash funcs only work with set_ftrace_filter */
3639	if (!enabled || !param)
3640		return -EINVAL;
3641
3642	system = strsep(&param, ":");
3643	if (!param)
3644		return -EINVAL;
3645
3646	event = strsep(&param, ":");
3647
3648	mutex_lock(&event_mutex);
3649
3650	ret = -EINVAL;
3651	file = find_event_file(tr, system, event);
3652	if (!file)
3653		goto out;
3654
3655	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3656
3657	if (enable)
3658		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3659	else
3660		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3661
3662	if (glob[0] == '!') {
3663		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
 
3664		goto out;
3665	}
3666
3667	ret = -ENOMEM;
3668
3669	data = kzalloc(sizeof(*data), GFP_KERNEL);
3670	if (!data)
3671		goto out;
3672
3673	data->enable = enable;
3674	data->count = -1;
3675	data->file = file;
3676
3677	if (!param)
3678		goto out_reg;
3679
3680	number = strsep(&param, ":");
3681
3682	ret = -EINVAL;
3683	if (!strlen(number))
3684		goto out_free;
3685
3686	/*
3687	 * We use the callback data field (which is a pointer)
3688	 * as our counter.
3689	 */
3690	ret = kstrtoul(number, 0, &data->count);
3691	if (ret)
3692		goto out_free;
3693
3694 out_reg:
3695	/* Don't let event modules unload while probe registered */
3696	ret = trace_event_try_get_ref(file->event_call);
3697	if (!ret) {
3698		ret = -EBUSY;
3699		goto out_free;
3700	}
3701
3702	ret = __ftrace_event_enable_disable(file, 1, 1);
3703	if (ret < 0)
3704		goto out_put;
3705
3706	ret = register_ftrace_function_probe(glob, tr, ops, data);
3707	/*
3708	 * The above returns on success the # of functions enabled,
3709	 * but if it didn't find any functions it returns zero.
3710	 * Consider no functions a failure too.
3711	 */
3712	if (!ret) {
3713		ret = -ENOENT;
3714		goto out_disable;
3715	} else if (ret < 0)
3716		goto out_disable;
3717	/* Just return zero, not the number of enabled functions */
3718	ret = 0;
3719 out:
3720	mutex_unlock(&event_mutex);
3721	return ret;
3722
3723 out_disable:
3724	__ftrace_event_enable_disable(file, 0, 1);
3725 out_put:
3726	trace_event_put_ref(file->event_call);
3727 out_free:
3728	kfree(data);
3729	goto out;
3730}
3731
3732static struct ftrace_func_command event_enable_cmd = {
3733	.name			= ENABLE_EVENT_STR,
3734	.func			= event_enable_func,
3735};
3736
3737static struct ftrace_func_command event_disable_cmd = {
3738	.name			= DISABLE_EVENT_STR,
3739	.func			= event_enable_func,
3740};
3741
3742static __init int register_event_cmds(void)
3743{
3744	int ret;
3745
3746	ret = register_ftrace_command(&event_enable_cmd);
3747	if (WARN_ON(ret < 0))
3748		return ret;
3749	ret = register_ftrace_command(&event_disable_cmd);
3750	if (WARN_ON(ret < 0))
3751		unregister_ftrace_command(&event_enable_cmd);
3752	return ret;
3753}
3754#else
3755static inline int register_event_cmds(void) { return 0; }
3756#endif /* CONFIG_DYNAMIC_FTRACE */
3757
3758/*
3759 * The top level array and trace arrays created by boot-time tracing
3760 * have already had its trace_event_file descriptors created in order
3761 * to allow for early events to be recorded.
3762 * This function is called after the tracefs has been initialized,
3763 * and we now have to create the files associated to the events.
3764 */
3765static void __trace_early_add_event_dirs(struct trace_array *tr)
 
3766{
3767	struct trace_event_file *file;
3768	int ret;
3769
3770
3771	list_for_each_entry(file, &tr->events, list) {
3772		ret = event_create_dir(tr->event_dir, file);
3773		if (ret < 0)
3774			pr_warn("Could not create directory for event %s\n",
3775				trace_event_name(file->event_call));
3776	}
3777}
3778
3779/*
3780 * For early boot up, the top trace array and the trace arrays created
3781 * by boot-time tracing require to have a list of events that can be
3782 * enabled. This must be done before the filesystem is set up in order
3783 * to allow events to be traced early.
3784 */
3785void __trace_early_add_events(struct trace_array *tr)
 
3786{
3787	struct trace_event_call *call;
3788	int ret;
3789
3790	list_for_each_entry(call, &ftrace_events, list) {
3791		/* Early boot up should not have any modules loaded */
3792		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3793		    WARN_ON_ONCE(call->module))
3794			continue;
3795
3796		ret = __trace_early_add_new_event(call, tr);
3797		if (ret < 0)
3798			pr_warn("Could not create early event %s\n",
3799				trace_event_name(call));
3800	}
3801}
3802
3803/* Remove the event directory structure for a trace directory. */
3804static void
3805__trace_remove_event_dirs(struct trace_array *tr)
3806{
3807	struct trace_event_file *file, *next;
3808
3809	list_for_each_entry_safe(file, next, &tr->events, list)
3810		remove_event_file_dir(file);
3811}
3812
3813static void __add_event_to_tracers(struct trace_event_call *call)
3814{
3815	struct trace_array *tr;
3816
3817	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3818		__trace_add_new_event(call, tr);
3819}
3820
3821extern struct trace_event_call *__start_ftrace_events[];
3822extern struct trace_event_call *__stop_ftrace_events[];
3823
3824static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3825
3826static __init int setup_trace_event(char *str)
3827{
3828	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3829	trace_set_ring_buffer_expanded(NULL);
3830	disable_tracing_selftest("running event tracing");
3831
3832	return 1;
3833}
3834__setup("trace_event=", setup_trace_event);
3835
3836static int events_callback(const char *name, umode_t *mode, void **data,
3837			   const struct file_operations **fops)
3838{
3839	if (strcmp(name, "enable") == 0) {
3840		*mode = TRACE_MODE_WRITE;
3841		*fops = &ftrace_tr_enable_fops;
3842		return 1;
3843	}
3844
3845	if (strcmp(name, "header_page") == 0) {
3846		*mode = TRACE_MODE_READ;
3847		*fops = &ftrace_show_header_page_fops;
3848
3849	} else if (strcmp(name, "header_event") == 0) {
3850		*mode = TRACE_MODE_READ;
3851		*fops = &ftrace_show_header_event_fops;
3852	} else
3853		return 0;
3854
3855	return 1;
3856}
3857
3858/* Expects to have event_mutex held when called */
3859static int
3860create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3861{
3862	struct eventfs_inode *e_events;
3863	struct dentry *entry;
3864	int nr_entries;
3865	static struct eventfs_entry events_entries[] = {
3866		{
3867			.name		= "enable",
3868			.callback	= events_callback,
3869		},
3870		{
3871			.name		= "header_page",
3872			.callback	= events_callback,
3873		},
3874		{
3875			.name		= "header_event",
3876			.callback	= events_callback,
3877		},
3878	};
3879
3880	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3881				  tr, &ftrace_set_event_fops);
3882	if (!entry)
 
3883		return -ENOMEM;
 
3884
3885	nr_entries = ARRAY_SIZE(events_entries);
3886
3887	e_events = eventfs_create_events_dir("events", parent, events_entries,
3888					     nr_entries, tr);
3889	if (IS_ERR(e_events)) {
3890		pr_warn("Could not create tracefs 'events' directory\n");
3891		return -ENOMEM;
3892	}
3893
3894	/* There are not as crucial, just warn if they are not created */
 
3895
3896	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3897			  tr, &ftrace_set_event_pid_fops);
 
 
 
 
 
 
3898
3899	trace_create_file("set_event_notrace_pid",
3900			  TRACE_MODE_WRITE, parent, tr,
3901			  &ftrace_set_event_notrace_pid_fops);
3902
3903	tr->event_dir = e_events;
3904
3905	return 0;
3906}
3907
3908/**
3909 * event_trace_add_tracer - add a instance of a trace_array to events
3910 * @parent: The parent dentry to place the files/directories for events in
3911 * @tr: The trace array associated with these events
3912 *
3913 * When a new instance is created, it needs to set up its events
3914 * directory, as well as other files associated with events. It also
3915 * creates the event hierarchy in the @parent/events directory.
3916 *
3917 * Returns 0 on success.
3918 *
3919 * Must be called with event_mutex held.
3920 */
3921int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3922{
3923	int ret;
3924
3925	lockdep_assert_held(&event_mutex);
3926
3927	ret = create_event_toplevel_files(parent, tr);
3928	if (ret)
3929		goto out;
3930
3931	down_write(&trace_event_sem);
3932	/* If tr already has the event list, it is initialized in early boot. */
3933	if (unlikely(!list_empty(&tr->events)))
3934		__trace_early_add_event_dirs(tr);
3935	else
3936		__trace_add_event_dirs(tr);
3937	up_write(&trace_event_sem);
3938
3939 out:
 
 
3940	return ret;
3941}
3942
3943/*
3944 * The top trace array already had its file descriptors created.
3945 * Now the files themselves need to be created.
3946 */
3947static __init int
3948early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3949{
3950	int ret;
3951
3952	mutex_lock(&event_mutex);
3953
3954	ret = create_event_toplevel_files(parent, tr);
3955	if (ret)
3956		goto out_unlock;
3957
3958	down_write(&trace_event_sem);
3959	__trace_early_add_event_dirs(tr);
3960	up_write(&trace_event_sem);
3961
3962 out_unlock:
3963	mutex_unlock(&event_mutex);
3964
3965	return ret;
3966}
3967
3968/* Must be called with event_mutex held */
3969int event_trace_del_tracer(struct trace_array *tr)
3970{
3971	lockdep_assert_held(&event_mutex);
3972
3973	/* Disable any event triggers and associated soft-disabled events */
3974	clear_event_triggers(tr);
3975
3976	/* Clear the pid list */
3977	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3978
3979	/* Disable any running events */
3980	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3981
3982	/* Make sure no more events are being executed */
3983	tracepoint_synchronize_unregister();
3984
3985	down_write(&trace_event_sem);
3986	__trace_remove_event_dirs(tr);
3987	eventfs_remove_events_dir(tr->event_dir);
3988	up_write(&trace_event_sem);
3989
3990	tr->event_dir = NULL;
3991
 
 
3992	return 0;
3993}
3994
3995static __init int event_trace_memsetup(void)
3996{
3997	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3998	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3999	return 0;
4000}
4001
4002__init void
4003early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4004{
 
4005	char *token;
4006	int ret;
4007
4008	while (true) {
4009		token = strsep(&buf, ",");
4010
4011		if (!token)
4012			break;
4013
4014		if (*token) {
4015			/* Restarting syscalls requires that we stop them first */
4016			if (disable_first)
4017				ftrace_set_clr_event(tr, token, 0);
4018
4019			ret = ftrace_set_clr_event(tr, token, 1);
4020			if (ret)
4021				pr_warn("Failed to enable trace event: %s\n", token);
4022		}
4023
4024		/* Put back the comma to allow this to be called again */
4025		if (buf)
4026			*(buf - 1) = ',';
4027	}
4028}
4029
4030static __init int event_trace_enable(void)
4031{
4032	struct trace_array *tr = top_trace_array();
4033	struct trace_event_call **iter, *call;
4034	int ret;
4035
4036	if (!tr)
4037		return -ENODEV;
4038
4039	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4040
4041		call = *iter;
4042		ret = event_init(call);
4043		if (!ret)
4044			list_add(&call->list, &ftrace_events);
4045	}
4046
4047	register_trigger_cmds();
4048
4049	/*
4050	 * We need the top trace array to have a working set of trace
4051	 * points at early init, before the debug files and directories
4052	 * are created. Create the file entries now, and attach them
4053	 * to the actual file dentries later.
4054	 */
4055	__trace_early_add_events(tr);
4056
4057	early_enable_events(tr, bootup_event_buf, false);
4058
4059	trace_printk_start_comm();
4060
4061	register_event_cmds();
4062
 
4063
4064	return 0;
4065}
4066
4067/*
4068 * event_trace_enable() is called from trace_event_init() first to
4069 * initialize events and perhaps start any events that are on the
4070 * command line. Unfortunately, there are some events that will not
4071 * start this early, like the system call tracepoints that need
4072 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4073 * event_trace_enable() is called before pid 1 starts, and this flag
4074 * is never set, making the syscall tracepoint never get reached, but
4075 * the event is enabled regardless (and not doing anything).
4076 */
4077static __init int event_trace_enable_again(void)
4078{
4079	struct trace_array *tr;
4080
4081	tr = top_trace_array();
4082	if (!tr)
4083		return -ENODEV;
4084
4085	early_enable_events(tr, bootup_event_buf, true);
4086
4087	return 0;
4088}
4089
4090early_initcall(event_trace_enable_again);
4091
4092/* Init fields which doesn't related to the tracefs */
4093static __init int event_trace_init_fields(void)
4094{
4095	if (trace_define_generic_fields())
4096		pr_warn("tracing: Failed to allocated generic fields");
4097
4098	if (trace_define_common_fields())
4099		pr_warn("tracing: Failed to allocate common fields");
4100
4101	return 0;
4102}
4103
4104__init int event_trace_init(void)
4105{
4106	struct trace_array *tr;
 
 
4107	int ret;
4108
4109	tr = top_trace_array();
4110	if (!tr)
4111		return -ENODEV;
4112
4113	trace_create_file("available_events", TRACE_MODE_READ,
4114			  NULL, tr, &ftrace_avail_fops);
 
 
 
 
 
 
4115
4116	ret = early_event_add_tracer(NULL, tr);
 
 
 
 
 
 
4117	if (ret)
4118		return ret;
4119
4120#ifdef CONFIG_MODULES
4121	ret = register_module_notifier(&trace_module_nb);
4122	if (ret)
4123		pr_warn("Failed to register trace events module notifier\n");
4124#endif
4125
4126	eventdir_initialized = true;
4127
4128	return 0;
4129}
4130
4131void __init trace_event_init(void)
4132{
4133	event_trace_memsetup();
4134	init_ftrace_syscalls();
4135	event_trace_enable();
4136	event_trace_init_fields();
4137}
4138
4139#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
 
 
4140
4141static DEFINE_SPINLOCK(test_spinlock);
4142static DEFINE_SPINLOCK(test_spinlock_irq);
4143static DEFINE_MUTEX(test_mutex);
4144
4145static __init void test_work(struct work_struct *dummy)
4146{
4147	spin_lock(&test_spinlock);
4148	spin_lock_irq(&test_spinlock_irq);
4149	udelay(1);
4150	spin_unlock_irq(&test_spinlock_irq);
4151	spin_unlock(&test_spinlock);
4152
4153	mutex_lock(&test_mutex);
4154	msleep(1);
4155	mutex_unlock(&test_mutex);
4156}
4157
4158static __init int event_test_thread(void *unused)
4159{
4160	void *test_malloc;
4161
4162	test_malloc = kmalloc(1234, GFP_KERNEL);
4163	if (!test_malloc)
4164		pr_info("failed to kmalloc\n");
4165
4166	schedule_on_each_cpu(test_work);
4167
4168	kfree(test_malloc);
4169
4170	set_current_state(TASK_INTERRUPTIBLE);
4171	while (!kthread_should_stop()) {
4172		schedule();
4173		set_current_state(TASK_INTERRUPTIBLE);
4174	}
4175	__set_current_state(TASK_RUNNING);
4176
4177	return 0;
4178}
4179
4180/*
4181 * Do various things that may trigger events.
4182 */
4183static __init void event_test_stuff(void)
4184{
4185	struct task_struct *test_thread;
4186
4187	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4188	msleep(1);
4189	kthread_stop(test_thread);
4190}
4191
4192/*
4193 * For every trace event defined, we will test each trace point separately,
4194 * and then by groups, and finally all trace points.
4195 */
4196static __init void event_trace_self_tests(void)
4197{
4198	struct trace_subsystem_dir *dir;
4199	struct trace_event_file *file;
4200	struct trace_event_call *call;
4201	struct event_subsystem *system;
4202	struct trace_array *tr;
4203	int ret;
4204
4205	tr = top_trace_array();
4206	if (!tr)
4207		return;
4208
4209	pr_info("Running tests on trace events:\n");
4210
4211	list_for_each_entry(file, &tr->events, list) {
4212
4213		call = file->event_call;
4214
4215		/* Only test those that have a probe */
4216		if (!call->class || !call->class->probe)
4217			continue;
4218
4219/*
4220 * Testing syscall events here is pretty useless, but
4221 * we still do it if configured. But this is time consuming.
4222 * What we really need is a user thread to perform the
4223 * syscalls as we test.
4224 */
4225#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4226		if (call->class->system &&
4227		    strcmp(call->class->system, "syscalls") == 0)
4228			continue;
4229#endif
4230
4231		pr_info("Testing event %s: ", trace_event_name(call));
4232
4233		/*
4234		 * If an event is already enabled, someone is using
4235		 * it and the self test should not be on.
4236		 */
4237		if (file->flags & EVENT_FILE_FL_ENABLED) {
4238			pr_warn("Enabled event during self test!\n");
4239			WARN_ON_ONCE(1);
4240			continue;
4241		}
4242
4243		ftrace_event_enable_disable(file, 1);
4244		event_test_stuff();
4245		ftrace_event_enable_disable(file, 0);
4246
4247		pr_cont("OK\n");
4248	}
4249
4250	/* Now test at the sub system level */
4251
4252	pr_info("Running tests on trace event systems:\n");
4253
4254	list_for_each_entry(dir, &tr->systems, list) {
4255
4256		system = dir->subsystem;
4257
4258		/* the ftrace system is special, skip it */
4259		if (strcmp(system->name, "ftrace") == 0)
4260			continue;
4261
4262		pr_info("Testing event system %s: ", system->name);
4263
4264		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4265		if (WARN_ON_ONCE(ret)) {
4266			pr_warn("error enabling system %s\n",
4267				system->name);
4268			continue;
4269		}
4270
4271		event_test_stuff();
4272
4273		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4274		if (WARN_ON_ONCE(ret)) {
4275			pr_warn("error disabling system %s\n",
4276				system->name);
4277			continue;
4278		}
4279
4280		pr_cont("OK\n");
4281	}
4282
4283	/* Test with all events enabled */
4284
4285	pr_info("Running tests on all trace events:\n");
4286	pr_info("Testing all events: ");
4287
4288	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4289	if (WARN_ON_ONCE(ret)) {
4290		pr_warn("error enabling all events\n");
4291		return;
4292	}
4293
4294	event_test_stuff();
4295
4296	/* reset sysname */
4297	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4298	if (WARN_ON_ONCE(ret)) {
4299		pr_warn("error disabling all events\n");
4300		return;
4301	}
4302
4303	pr_cont("OK\n");
4304}
4305
4306#ifdef CONFIG_FUNCTION_TRACER
4307
4308static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4309
4310static struct trace_event_file event_trace_file __initdata;
4311
4312static void __init
4313function_test_events_call(unsigned long ip, unsigned long parent_ip,
4314			  struct ftrace_ops *op, struct ftrace_regs *regs)
4315{
4316	struct trace_buffer *buffer;
4317	struct ring_buffer_event *event;
 
4318	struct ftrace_entry *entry;
4319	unsigned int trace_ctx;
4320	long disabled;
4321	int cpu;
 
4322
4323	trace_ctx = tracing_gen_ctx();
4324	preempt_disable_notrace();
4325	cpu = raw_smp_processor_id();
4326	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4327
4328	if (disabled != 1)
4329		goto out;
4330
4331	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4332						TRACE_FN, sizeof(*entry),
4333						trace_ctx);
 
 
4334	if (!event)
4335		goto out;
4336	entry	= ring_buffer_event_data(event);
4337	entry->ip			= ip;
4338	entry->parent_ip		= parent_ip;
4339
4340	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4341				    entry, trace_ctx);
4342 out:
4343	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4344	preempt_enable_notrace();
4345}
4346
4347static struct ftrace_ops trace_ops __initdata  =
4348{
4349	.func = function_test_events_call,
 
4350};
4351
4352static __init void event_trace_self_test_with_function(void)
4353{
4354	int ret;
4355
4356	event_trace_file.tr = top_trace_array();
4357	if (WARN_ON(!event_trace_file.tr))
4358		return;
4359
4360	ret = register_ftrace_function(&trace_ops);
4361	if (WARN_ON(ret < 0)) {
4362		pr_info("Failed to enable function tracer for event tests\n");
4363		return;
4364	}
4365	pr_info("Running tests again, along with the function tracer\n");
4366	event_trace_self_tests();
4367	unregister_ftrace_function(&trace_ops);
4368}
4369#else
4370static __init void event_trace_self_test_with_function(void)
4371{
4372}
4373#endif
4374
4375static __init int event_trace_self_tests_init(void)
4376{
4377	if (!tracing_selftest_disabled) {
4378		event_trace_self_tests();
4379		event_trace_self_test_with_function();
4380	}
4381
4382	return 0;
4383}
4384
4385late_initcall(event_trace_self_tests_init);
4386
4387#endif