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
v6.13.7
   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, const 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				int need_test)
 119{
 120	struct ftrace_event_field *field;
 121
 122	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 123	if (!field)
 124		return -ENOMEM;
 125
 126	field->name = name;
 127	field->type = type;
 128
 129	if (filter_type == FILTER_OTHER)
 130		field->filter_type = filter_assign_type(type);
 131	else
 132		field->filter_type = filter_type;
 133
 134	field->offset = offset;
 135	field->size = size;
 136	field->is_signed = is_signed;
 137	field->needs_test = need_test;
 138	field->len = len;
 139
 140	list_add(&field->link, head);
 141
 142	return 0;
 143}
 144
 145int trace_define_field(struct trace_event_call *call, const char *type,
 146		       const char *name, int offset, int size, int is_signed,
 147		       int filter_type)
 148{
 149	struct list_head *head;
 150
 151	if (WARN_ON(!call->class))
 152		return 0;
 153
 154	head = trace_get_fields(call);
 155	return __trace_define_field(head, type, name, offset, size,
 156				    is_signed, filter_type, 0, 0);
 157}
 158EXPORT_SYMBOL_GPL(trace_define_field);
 159
 160static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 161		       const char *name, int offset, int size, int is_signed,
 162		       int filter_type, int len, int need_test)
 163{
 164	struct list_head *head;
 165
 166	if (WARN_ON(!call->class))
 167		return 0;
 168
 169	head = trace_get_fields(call);
 170	return __trace_define_field(head, type, name, offset, size,
 171				    is_signed, filter_type, len, need_test);
 172}
 173
 174#define __generic_field(type, item, filter_type)			\
 175	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 176				   #item, 0, 0, is_signed_type(type),	\
 177				   filter_type, 0, 0);			\
 178	if (ret)							\
 179		return ret;
 180
 181#define __common_field(type, item)					\
 182	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 183				   "common_" #item,			\
 184				   offsetof(typeof(ent), item),		\
 185				   sizeof(ent.item),			\
 186				   is_signed_type(type), FILTER_OTHER,	\
 187				   0, 0);				\
 188	if (ret)							\
 189		return ret;
 190
 191static int trace_define_generic_fields(void)
 192{
 193	int ret;
 194
 195	__generic_field(int, CPU, FILTER_CPU);
 196	__generic_field(int, cpu, FILTER_CPU);
 197	__generic_field(int, common_cpu, FILTER_CPU);
 198	__generic_field(char *, COMM, FILTER_COMM);
 199	__generic_field(char *, comm, FILTER_COMM);
 200	__generic_field(char *, stacktrace, FILTER_STACKTRACE);
 201	__generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 202
 203	return ret;
 204}
 205
 206static int trace_define_common_fields(void)
 207{
 208	int ret;
 209	struct trace_entry ent;
 210
 211	__common_field(unsigned short, type);
 212	__common_field(unsigned char, flags);
 213	/* Holds both preempt_count and migrate_disable */
 214	__common_field(unsigned char, preempt_count);
 215	__common_field(int, pid);
 216
 217	return ret;
 218}
 219
 220static void trace_destroy_fields(struct trace_event_call *call)
 221{
 222	struct ftrace_event_field *field, *next;
 223	struct list_head *head;
 224
 225	head = trace_get_fields(call);
 226	list_for_each_entry_safe(field, next, head, link) {
 227		list_del(&field->link);
 228		kmem_cache_free(field_cachep, field);
 229	}
 230}
 231
 232/*
 233 * run-time version of trace_event_get_offsets_<call>() that returns the last
 234 * accessible offset of trace fields excluding __dynamic_array bytes
 235 */
 236int trace_event_get_offsets(struct trace_event_call *call)
 237{
 238	struct ftrace_event_field *tail;
 239	struct list_head *head;
 240
 241	head = trace_get_fields(call);
 242	/*
 243	 * head->next points to the last field with the largest offset,
 244	 * since it was added last by trace_define_field()
 245	 */
 246	tail = list_first_entry(head, struct ftrace_event_field, link);
 247	return tail->offset + tail->size;
 248}
 249
 250
 251static struct trace_event_fields *find_event_field(const char *fmt,
 252						   struct trace_event_call *call)
 253{
 254	struct trace_event_fields *field = call->class->fields_array;
 255	const char *p = fmt;
 256	int len;
 257
 258	if (!(len = str_has_prefix(fmt, "REC->")))
 259		return NULL;
 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) || field->name[len])
 269			continue;
 270
 271		return field;
 272	}
 273	return NULL;
 274}
 275
 276/*
 277 * Check if the referenced field is an array and return true,
 278 * as arrays are OK to dereference.
 279 */
 280static bool test_field(const char *fmt, struct trace_event_call *call)
 281{
 282	struct trace_event_fields *field;
 283
 284	field = find_event_field(fmt, call);
 285	if (!field)
 286		return false;
 287
 288	/* This is an array and is OK to dereference. */
 289	return strchr(field->type, '[') != NULL;
 290}
 291
 292/* Look for a string within an argument */
 293static bool find_print_string(const char *arg, const char *str, const char *end)
 294{
 295	const char *r;
 296
 297	r = strstr(arg, str);
 298	return r && r < end;
 299}
 300
 301/* Return true if the argument pointer is safe */
 302static bool process_pointer(const char *fmt, int len, struct trace_event_call *call)
 303{
 304	const char *r, *e, *a;
 305
 306	e = fmt + len;
 307
 308	/* Find the REC-> in the argument */
 309	r = strstr(fmt, "REC->");
 310	if (r && r < e) {
 311		/*
 312		 * Addresses of events on the buffer, or an array on the buffer is
 313		 * OK to dereference. There's ways to fool this, but
 314		 * this is to catch common mistakes, not malicious code.
 315		 */
 316		a = strchr(fmt, '&');
 317		if ((a && (a < r)) || test_field(r, call))
 318			return true;
 319	} else if (find_print_string(fmt, "__get_dynamic_array(", e)) {
 320		return true;
 321	} else if (find_print_string(fmt, "__get_rel_dynamic_array(", e)) {
 322		return true;
 323	} else if (find_print_string(fmt, "__get_dynamic_array_len(", e)) {
 324		return true;
 325	} else if (find_print_string(fmt, "__get_rel_dynamic_array_len(", e)) {
 326		return true;
 327	} else if (find_print_string(fmt, "__get_sockaddr(", e)) {
 328		return true;
 329	} else if (find_print_string(fmt, "__get_rel_sockaddr(", e)) {
 330		return true;
 331	}
 332	return false;
 333}
 334
 335/* Return true if the string is safe */
 336static bool process_string(const char *fmt, int len, struct trace_event_call *call)
 337{
 338	struct trace_event_fields *field;
 339	const char *r, *e, *s;
 340
 341	e = fmt + len;
 342
 343	/*
 344	 * There are several helper functions that return strings.
 345	 * If the argument contains a function, then assume its field is valid.
 346	 * It is considered that the argument has a function if it has:
 347	 *   alphanumeric or '_' before a parenthesis.
 348	 */
 349	s = fmt;
 350	do {
 351		r = strstr(s, "(");
 352		if (!r || r >= e)
 353			break;
 354		for (int i = 1; r - i >= s; i++) {
 355			char ch = *(r - i);
 356			if (isspace(ch))
 357				continue;
 358			if (isalnum(ch) || ch == '_')
 359				return true;
 360			/* Anything else, this isn't a function */
 361			break;
 362		}
 363		/* A function could be wrapped in parethesis, try the next one */
 364		s = r + 1;
 365	} while (s < e);
 366
 367	/*
 368	 * Check for arrays. If the argument has: foo[REC->val]
 369	 * then it is very likely that foo is an array of strings
 370	 * that are safe to use.
 371	 */
 372	r = strstr(s, "[");
 373	if (r && r < e) {
 374		r = strstr(r, "REC->");
 375		if (r && r < e)
 376			return true;
 377	}
 378
 379	/*
 380	 * If there's any strings in the argument consider this arg OK as it
 381	 * could be: REC->field ? "foo" : "bar" and we don't want to get into
 382	 * verifying that logic here.
 383	 */
 384	if (find_print_string(fmt, "\"", e))
 385		return true;
 386
 387	/* Dereferenced strings are also valid like any other pointer */
 388	if (process_pointer(fmt, len, call))
 389		return true;
 390
 391	/* Make sure the field is found */
 392	field = find_event_field(fmt, call);
 393	if (!field)
 394		return false;
 395
 396	/* Test this field's string before printing the event */
 397	call->flags |= TRACE_EVENT_FL_TEST_STR;
 398	field->needs_test = 1;
 399
 400	return true;
 401}
 402
 403/*
 404 * Examine the print fmt of the event looking for unsafe dereference
 405 * pointers using %p* that could be recorded in the trace event and
 406 * much later referenced after the pointer was freed. Dereferencing
 407 * pointers are OK, if it is dereferenced into the event itself.
 408 */
 409static void test_event_printk(struct trace_event_call *call)
 410{
 411	u64 dereference_flags = 0;
 412	u64 string_flags = 0;
 413	bool first = true;
 414	const char *fmt;
 415	int parens = 0;
 416	char in_quote = 0;
 417	int start_arg = 0;
 418	int arg = 0;
 419	int i, e;
 420
 421	fmt = call->print_fmt;
 422
 423	if (!fmt)
 424		return;
 425
 426	for (i = 0; fmt[i]; i++) {
 427		switch (fmt[i]) {
 428		case '\\':
 429			i++;
 430			if (!fmt[i])
 431				return;
 432			continue;
 433		case '"':
 434		case '\'':
 435			/*
 436			 * The print fmt starts with a string that
 437			 * is processed first to find %p* usage,
 438			 * then after the first string, the print fmt
 439			 * contains arguments that are used to check
 440			 * if the dereferenced %p* usage is safe.
 441			 */
 442			if (first) {
 443				if (fmt[i] == '\'')
 444					continue;
 445				if (in_quote) {
 446					arg = 0;
 447					first = false;
 448					/*
 449					 * If there was no %p* uses
 450					 * the fmt is OK.
 451					 */
 452					if (!dereference_flags)
 453						return;
 454				}
 455			}
 456			if (in_quote) {
 457				if (in_quote == fmt[i])
 458					in_quote = 0;
 459			} else {
 460				in_quote = fmt[i];
 461			}
 462			continue;
 463		case '%':
 464			if (!first || !in_quote)
 465				continue;
 466			i++;
 467			if (!fmt[i])
 468				return;
 469			switch (fmt[i]) {
 470			case '%':
 471				continue;
 472			case 'p':
 473				/* Find dereferencing fields */
 474				switch (fmt[i + 1]) {
 475				case 'B': case 'R': case 'r':
 476				case 'b': case 'M': case 'm':
 477				case 'I': case 'i': case 'E':
 478				case 'U': case 'V': case 'N':
 479				case 'a': case 'd': case 'D':
 480				case 'g': case 't': case 'C':
 481				case 'O': case 'f':
 482					if (WARN_ONCE(arg == 63,
 483						      "Too many args for event: %s",
 484						      trace_event_name(call)))
 485						return;
 486					dereference_flags |= 1ULL << arg;
 487				}
 488				break;
 489			default:
 490			{
 491				bool star = false;
 492				int j;
 493
 494				/* Increment arg if %*s exists. */
 495				for (j = 0; fmt[i + j]; j++) {
 496					if (isdigit(fmt[i + j]) ||
 497					    fmt[i + j] == '.')
 498						continue;
 499					if (fmt[i + j] == '*') {
 500						star = true;
 501						continue;
 502					}
 503					if ((fmt[i + j] == 's')) {
 504						if (star)
 505							arg++;
 506						if (WARN_ONCE(arg == 63,
 507							      "Too many args for event: %s",
 508							      trace_event_name(call)))
 509							return;
 510						dereference_flags |= 1ULL << arg;
 511						string_flags |= 1ULL << arg;
 512					}
 513					break;
 514				}
 515				break;
 516			} /* default */
 517
 518			} /* switch */
 519			arg++;
 520			continue;
 521		case '(':
 522			if (in_quote)
 523				continue;
 524			parens++;
 525			continue;
 526		case ')':
 527			if (in_quote)
 528				continue;
 529			parens--;
 530			if (WARN_ONCE(parens < 0,
 531				      "Paren mismatch for event: %s\narg='%s'\n%*s",
 532				      trace_event_name(call),
 533				      fmt + start_arg,
 534				      (i - start_arg) + 5, "^"))
 535				return;
 536			continue;
 537		case ',':
 538			if (in_quote || parens)
 539				continue;
 540			e = i;
 541			i++;
 542			while (isspace(fmt[i]))
 543				i++;
 544
 545			/*
 546			 * If start_arg is zero, then this is the start of the
 547			 * first argument. The processing of the argument happens
 548			 * when the end of the argument is found, as it needs to
 549			 * handle paranthesis and such.
 550			 */
 551			if (!start_arg) {
 552				start_arg = i;
 553				/* Balance out the i++ in the for loop */
 554				i--;
 555				continue;
 556			}
 557
 558			if (dereference_flags & (1ULL << arg)) {
 559				if (string_flags & (1ULL << arg)) {
 560					if (process_string(fmt + start_arg, e - start_arg, call))
 561						dereference_flags &= ~(1ULL << arg);
 562				} else if (process_pointer(fmt + start_arg, e - start_arg, call))
 563					dereference_flags &= ~(1ULL << arg);
 564			}
 565
 566			start_arg = i;
 567			arg++;
 568			/* Balance out the i++ in the for loop */
 569			i--;
 570		}
 571	}
 572
 573	if (dereference_flags & (1ULL << arg)) {
 574		if (string_flags & (1ULL << arg)) {
 575			if (process_string(fmt + start_arg, i - start_arg, call))
 576				dereference_flags &= ~(1ULL << arg);
 577		} else if (process_pointer(fmt + start_arg, i - start_arg, call))
 578			dereference_flags &= ~(1ULL << arg);
 579	}
 580
 581	/*
 582	 * If you triggered the below warning, the trace event reported
 583	 * uses an unsafe dereference pointer %p*. As the data stored
 584	 * at the trace event time may no longer exist when the trace
 585	 * event is printed, dereferencing to the original source is
 586	 * unsafe. The source of the dereference must be copied into the
 587	 * event itself, and the dereference must access the copy instead.
 588	 */
 589	if (WARN_ON_ONCE(dereference_flags)) {
 590		arg = 1;
 591		while (!(dereference_flags & 1)) {
 592			dereference_flags >>= 1;
 593			arg++;
 594		}
 595		pr_warn("event %s has unsafe dereference of argument %d\n",
 596			trace_event_name(call), arg);
 597		pr_warn("print_fmt: %s\n", fmt);
 598	}
 599}
 600
 601int trace_event_raw_init(struct trace_event_call *call)
 602{
 603	int id;
 604
 605	id = register_trace_event(&call->event);
 606	if (!id)
 607		return -ENODEV;
 608
 609	test_event_printk(call);
 610
 611	return 0;
 612}
 613EXPORT_SYMBOL_GPL(trace_event_raw_init);
 614
 615bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 616{
 617	struct trace_array *tr = trace_file->tr;
 618	struct trace_array_cpu *data;
 619	struct trace_pid_list *no_pid_list;
 620	struct trace_pid_list *pid_list;
 621
 622	pid_list = rcu_dereference_raw(tr->filtered_pids);
 623	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 624
 625	if (!pid_list && !no_pid_list)
 626		return false;
 627
 628	data = this_cpu_ptr(tr->array_buffer.data);
 629
 630	return data->ignore_pid;
 631}
 632EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 633
 634void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 635				 struct trace_event_file *trace_file,
 636				 unsigned long len)
 637{
 638	struct trace_event_call *event_call = trace_file->event_call;
 639
 640	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 641	    trace_event_ignore_this_pid(trace_file))
 642		return NULL;
 643
 644	/*
 645	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 646	 * preemption (adding one to the preempt_count). Since we are
 647	 * interested in the preempt_count at the time the tracepoint was
 648	 * hit, we need to subtract one to offset the increment.
 649	 */
 650	fbuffer->trace_ctx = tracing_gen_ctx_dec();
 651	fbuffer->trace_file = trace_file;
 652
 653	fbuffer->event =
 654		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 655						event_call->event.type, len,
 656						fbuffer->trace_ctx);
 657	if (!fbuffer->event)
 658		return NULL;
 659
 660	fbuffer->regs = NULL;
 661	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 662	return fbuffer->entry;
 663}
 664EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 665
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 666int trace_event_reg(struct trace_event_call *call,
 667		    enum trace_reg type, void *data)
 668{
 669	struct trace_event_file *file = data;
 670
 671	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 672	switch (type) {
 673	case TRACE_REG_REGISTER:
 674		return tracepoint_probe_register(call->tp,
 675						 call->class->probe,
 676						 file);
 677	case TRACE_REG_UNREGISTER:
 678		tracepoint_probe_unregister(call->tp,
 679					    call->class->probe,
 680					    file);
 681		return 0;
 682
 683#ifdef CONFIG_PERF_EVENTS
 684	case TRACE_REG_PERF_REGISTER:
 685		return tracepoint_probe_register(call->tp,
 686						 call->class->perf_probe,
 687						 call);
 688	case TRACE_REG_PERF_UNREGISTER:
 689		tracepoint_probe_unregister(call->tp,
 690					    call->class->perf_probe,
 691					    call);
 692		return 0;
 693	case TRACE_REG_PERF_OPEN:
 694	case TRACE_REG_PERF_CLOSE:
 695	case TRACE_REG_PERF_ADD:
 696	case TRACE_REG_PERF_DEL:
 697		return 0;
 698#endif
 699	}
 700	return 0;
 701}
 702EXPORT_SYMBOL_GPL(trace_event_reg);
 703
 704void trace_event_enable_cmd_record(bool enable)
 705{
 706	struct trace_event_file *file;
 707	struct trace_array *tr;
 708
 709	lockdep_assert_held(&event_mutex);
 710
 711	do_for_each_event_file(tr, file) {
 712
 713		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 714			continue;
 715
 716		if (enable) {
 717			tracing_start_cmdline_record();
 718			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 719		} else {
 720			tracing_stop_cmdline_record();
 721			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 722		}
 723	} while_for_each_event_file();
 724}
 725
 726void trace_event_enable_tgid_record(bool enable)
 727{
 728	struct trace_event_file *file;
 729	struct trace_array *tr;
 730
 731	lockdep_assert_held(&event_mutex);
 732
 733	do_for_each_event_file(tr, file) {
 734		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 735			continue;
 736
 737		if (enable) {
 738			tracing_start_tgid_record();
 739			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 740		} else {
 741			tracing_stop_tgid_record();
 742			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 743				  &file->flags);
 744		}
 745	} while_for_each_event_file();
 746}
 747
 748static int __ftrace_event_enable_disable(struct trace_event_file *file,
 749					 int enable, int soft_disable)
 750{
 751	struct trace_event_call *call = file->event_call;
 752	struct trace_array *tr = file->tr;
 753	int ret = 0;
 754	int disable;
 755
 756	switch (enable) {
 757	case 0:
 758		/*
 759		 * When soft_disable is set and enable is cleared, the sm_ref
 760		 * reference counter is decremented. If it reaches 0, we want
 761		 * to clear the SOFT_DISABLED flag but leave the event in the
 762		 * state that it was. That is, if the event was enabled and
 763		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 764		 * is set we do not want the event to be enabled before we
 765		 * clear the bit.
 766		 *
 767		 * When soft_disable is not set but the SOFT_MODE flag is,
 768		 * we do nothing. Do not disable the tracepoint, otherwise
 769		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 770		 */
 771		if (soft_disable) {
 772			if (atomic_dec_return(&file->sm_ref) > 0)
 773				break;
 774			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 775			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 776			/* Disable use of trace_buffered_event */
 777			trace_buffered_event_disable();
 778		} else
 779			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 780
 781		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 782			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 783			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 784				tracing_stop_cmdline_record();
 785				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 786			}
 787
 788			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 789				tracing_stop_tgid_record();
 790				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 791			}
 792
 793			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 794		}
 795		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 796		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 797			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 798		else
 799			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 800		break;
 801	case 1:
 802		/*
 803		 * When soft_disable is set and enable is set, we want to
 804		 * register the tracepoint for the event, but leave the event
 805		 * as is. That means, if the event was already enabled, we do
 806		 * nothing (but set SOFT_MODE). If the event is disabled, we
 807		 * set SOFT_DISABLED before enabling the event tracepoint, so
 808		 * it still seems to be disabled.
 809		 */
 810		if (!soft_disable)
 811			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 812		else {
 813			if (atomic_inc_return(&file->sm_ref) > 1)
 814				break;
 815			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 816			/* Enable use of trace_buffered_event */
 817			trace_buffered_event_enable();
 818		}
 819
 820		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 821			bool cmd = false, tgid = false;
 822
 823			/* Keep the event disabled, when going to SOFT_MODE. */
 824			if (soft_disable)
 825				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 826
 827			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 828				cmd = true;
 829				tracing_start_cmdline_record();
 830				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 831			}
 832
 833			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 834				tgid = true;
 835				tracing_start_tgid_record();
 836				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 837			}
 838
 839			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 840			if (ret) {
 841				if (cmd)
 842					tracing_stop_cmdline_record();
 843				if (tgid)
 844					tracing_stop_tgid_record();
 845				pr_info("event trace: Could not enable event "
 846					"%s\n", trace_event_name(call));
 847				break;
 848			}
 849			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 850
 851			/* WAS_ENABLED gets set but never cleared. */
 852			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 853		}
 854		break;
 855	}
 856
 857	return ret;
 858}
 859
 860int trace_event_enable_disable(struct trace_event_file *file,
 861			       int enable, int soft_disable)
 862{
 863	return __ftrace_event_enable_disable(file, enable, soft_disable);
 864}
 865
 866static int ftrace_event_enable_disable(struct trace_event_file *file,
 867				       int enable)
 868{
 869	return __ftrace_event_enable_disable(file, enable, 0);
 870}
 871
 872static void ftrace_clear_events(struct trace_array *tr)
 873{
 874	struct trace_event_file *file;
 875
 876	mutex_lock(&event_mutex);
 877	list_for_each_entry(file, &tr->events, list) {
 878		ftrace_event_enable_disable(file, 0);
 879	}
 880	mutex_unlock(&event_mutex);
 881}
 882
 883static void
 884event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 885{
 886	struct trace_pid_list *pid_list;
 887	struct trace_array *tr = data;
 888
 889	pid_list = rcu_dereference_raw(tr->filtered_pids);
 890	trace_filter_add_remove_task(pid_list, NULL, task);
 891
 892	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 893	trace_filter_add_remove_task(pid_list, NULL, task);
 894}
 895
 896static void
 897event_filter_pid_sched_process_fork(void *data,
 898				    struct task_struct *self,
 899				    struct task_struct *task)
 900{
 901	struct trace_pid_list *pid_list;
 902	struct trace_array *tr = data;
 
 
 
 
 
 
 
 903
 904	pid_list = rcu_dereference_sched(tr->filtered_pids);
 905	trace_filter_add_remove_task(pid_list, self, task);
 906
 907	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 908	trace_filter_add_remove_task(pid_list, self, task);
 909}
 
 
 910
 911void trace_event_follow_fork(struct trace_array *tr, bool enable)
 912{
 913	if (enable) {
 914		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 915						       tr, INT_MIN);
 916		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 917						       tr, INT_MAX);
 918	} else {
 919		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 920						    tr);
 921		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 922						    tr);
 923	}
 924}
 925
 926static void
 927event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 928					struct task_struct *prev,
 929					struct task_struct *next,
 930					unsigned int prev_state)
 931{
 932	struct trace_array *tr = data;
 933	struct trace_pid_list *no_pid_list;
 934	struct trace_pid_list *pid_list;
 935	bool ret;
 936
 937	pid_list = rcu_dereference_sched(tr->filtered_pids);
 938	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 939
 940	/*
 941	 * Sched switch is funny, as we only want to ignore it
 942	 * in the notrace case if both prev and next should be ignored.
 943	 */
 944	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 945		trace_ignore_this_task(NULL, no_pid_list, next);
 946
 947	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 948		       (trace_ignore_this_task(pid_list, NULL, prev) &&
 949			trace_ignore_this_task(pid_list, NULL, next)));
 950}
 951
 952static void
 953event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 954					 struct task_struct *prev,
 955					 struct task_struct *next,
 956					 unsigned int prev_state)
 957{
 958	struct trace_array *tr = data;
 959	struct trace_pid_list *no_pid_list;
 960	struct trace_pid_list *pid_list;
 961
 962	pid_list = rcu_dereference_sched(tr->filtered_pids);
 963	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 964
 965	this_cpu_write(tr->array_buffer.data->ignore_pid,
 966		       trace_ignore_this_task(pid_list, no_pid_list, next));
 967}
 968
 969static void
 970event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 971{
 972	struct trace_array *tr = data;
 973	struct trace_pid_list *no_pid_list;
 974	struct trace_pid_list *pid_list;
 975
 976	/* Nothing to do if we are already tracing */
 977	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 978		return;
 979
 980	pid_list = rcu_dereference_sched(tr->filtered_pids);
 981	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 982
 983	this_cpu_write(tr->array_buffer.data->ignore_pid,
 984		       trace_ignore_this_task(pid_list, no_pid_list, task));
 985}
 986
 987static void
 988event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 989{
 990	struct trace_array *tr = data;
 991	struct trace_pid_list *no_pid_list;
 992	struct trace_pid_list *pid_list;
 993
 994	/* Nothing to do if we are not tracing */
 995	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 996		return;
 997
 998	pid_list = rcu_dereference_sched(tr->filtered_pids);
 999	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1000
1001	/* Set tracing if current is enabled */
1002	this_cpu_write(tr->array_buffer.data->ignore_pid,
1003		       trace_ignore_this_task(pid_list, no_pid_list, current));
1004}
1005
1006static void unregister_pid_events(struct trace_array *tr)
1007{
 
 
 
 
 
 
 
 
 
1008	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
1009	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
1010
1011	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
1012	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
1013
1014	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
1015	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
1016
1017	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
1018	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
1019}
1020
1021static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
1022{
1023	struct trace_pid_list *pid_list;
1024	struct trace_pid_list *no_pid_list;
1025	struct trace_event_file *file;
1026	int cpu;
1027
1028	pid_list = rcu_dereference_protected(tr->filtered_pids,
1029					     lockdep_is_held(&event_mutex));
1030	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1031					     lockdep_is_held(&event_mutex));
1032
1033	/* Make sure there's something to do */
1034	if (!pid_type_enabled(type, pid_list, no_pid_list))
1035		return;
1036
1037	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
1038		unregister_pid_events(tr);
1039
1040		list_for_each_entry(file, &tr->events, list) {
1041			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1042		}
1043
1044		for_each_possible_cpu(cpu)
1045			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
1046	}
1047
1048	if (type & TRACE_PIDS)
1049		rcu_assign_pointer(tr->filtered_pids, NULL);
1050
1051	if (type & TRACE_NO_PIDS)
1052		rcu_assign_pointer(tr->filtered_no_pids, NULL);
1053
1054	/* Wait till all users are no longer using pid filtering */
1055	tracepoint_synchronize_unregister();
1056
1057	if ((type & TRACE_PIDS) && pid_list)
1058		trace_pid_list_free(pid_list);
1059
1060	if ((type & TRACE_NO_PIDS) && no_pid_list)
1061		trace_pid_list_free(no_pid_list);
1062}
1063
1064static void ftrace_clear_event_pids(struct trace_array *tr, int type)
1065{
1066	mutex_lock(&event_mutex);
1067	__ftrace_clear_event_pids(tr, type);
1068	mutex_unlock(&event_mutex);
1069}
1070
1071static void __put_system(struct event_subsystem *system)
1072{
1073	struct event_filter *filter = system->filter;
1074
1075	WARN_ON_ONCE(system_refcount(system) == 0);
1076	if (system_refcount_dec(system))
1077		return;
1078
1079	list_del(&system->list);
1080
1081	if (filter) {
1082		kfree(filter->filter_string);
1083		kfree(filter);
1084	}
1085	kfree_const(system->name);
1086	kfree(system);
1087}
1088
1089static void __get_system(struct event_subsystem *system)
1090{
1091	WARN_ON_ONCE(system_refcount(system) == 0);
1092	system_refcount_inc(system);
1093}
1094
1095static void __get_system_dir(struct trace_subsystem_dir *dir)
1096{
1097	WARN_ON_ONCE(dir->ref_count == 0);
1098	dir->ref_count++;
1099	__get_system(dir->subsystem);
1100}
1101
1102static void __put_system_dir(struct trace_subsystem_dir *dir)
1103{
1104	WARN_ON_ONCE(dir->ref_count == 0);
1105	/* If the subsystem is about to be freed, the dir must be too */
1106	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
1107
1108	__put_system(dir->subsystem);
1109	if (!--dir->ref_count)
1110		kfree(dir);
1111}
1112
1113static void put_system(struct trace_subsystem_dir *dir)
1114{
1115	mutex_lock(&event_mutex);
1116	__put_system_dir(dir);
1117	mutex_unlock(&event_mutex);
1118}
1119
1120static void remove_subsystem(struct trace_subsystem_dir *dir)
1121{
1122	if (!dir)
1123		return;
1124
1125	if (!--dir->nr_events) {
1126		eventfs_remove_dir(dir->ei);
1127		list_del(&dir->list);
1128		__put_system_dir(dir);
1129	}
1130}
1131
1132void event_file_get(struct trace_event_file *file)
1133{
1134	refcount_inc(&file->ref);
1135}
1136
1137void event_file_put(struct trace_event_file *file)
1138{
1139	if (WARN_ON_ONCE(!refcount_read(&file->ref))) {
1140		if (file->flags & EVENT_FILE_FL_FREED)
1141			kmem_cache_free(file_cachep, file);
1142		return;
1143	}
1144
1145	if (refcount_dec_and_test(&file->ref)) {
1146		/* Count should only go to zero when it is freed */
1147		if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1148			return;
1149		kmem_cache_free(file_cachep, file);
1150	}
1151}
1152
1153static void remove_event_file_dir(struct trace_event_file *file)
1154{
1155	eventfs_remove_dir(file->ei);
1156	list_del(&file->list);
1157	remove_subsystem(file->system);
1158	free_event_filter(file->filter);
1159	file->flags |= EVENT_FILE_FL_FREED;
1160	event_file_put(file);
1161}
1162
1163/*
1164 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1165 */
1166static int
1167__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1168			      const char *sub, const char *event, int set)
1169{
1170	struct trace_event_file *file;
1171	struct trace_event_call *call;
1172	const char *name;
1173	int ret = -EINVAL;
1174	int eret = 0;
1175
1176	list_for_each_entry(file, &tr->events, list) {
1177
1178		call = file->event_call;
1179		name = trace_event_name(call);
1180
1181		if (!name || !call->class || !call->class->reg)
1182			continue;
1183
1184		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1185			continue;
1186
1187		if (match &&
1188		    strcmp(match, name) != 0 &&
1189		    strcmp(match, call->class->system) != 0)
1190			continue;
1191
1192		if (sub && strcmp(sub, call->class->system) != 0)
1193			continue;
1194
1195		if (event && strcmp(event, name) != 0)
1196			continue;
1197
1198		ret = ftrace_event_enable_disable(file, set);
1199
1200		/*
1201		 * Save the first error and return that. Some events
1202		 * may still have been enabled, but let the user
1203		 * know that something went wrong.
1204		 */
1205		if (ret && !eret)
1206			eret = ret;
1207
1208		ret = eret;
1209	}
1210
1211	return ret;
1212}
1213
1214static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1215				  const char *sub, const char *event, int set)
1216{
1217	int ret;
1218
1219	mutex_lock(&event_mutex);
1220	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1221	mutex_unlock(&event_mutex);
1222
1223	return ret;
1224}
1225
1226int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1227{
1228	char *event = NULL, *sub = NULL, *match;
1229	int ret;
1230
1231	if (!tr)
1232		return -ENOENT;
1233	/*
1234	 * The buf format can be <subsystem>:<event-name>
1235	 *  *:<event-name> means any event by that name.
1236	 *  :<event-name> is the same.
1237	 *
1238	 *  <subsystem>:* means all events in that subsystem
1239	 *  <subsystem>: means the same.
1240	 *
1241	 *  <name> (no ':') means all events in a subsystem with
1242	 *  the name <name> or any event that matches <name>
1243	 */
1244
1245	match = strsep(&buf, ":");
1246	if (buf) {
1247		sub = match;
1248		event = buf;
1249		match = NULL;
1250
1251		if (!strlen(sub) || strcmp(sub, "*") == 0)
1252			sub = NULL;
1253		if (!strlen(event) || strcmp(event, "*") == 0)
1254			event = NULL;
1255	}
1256
1257	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1258
1259	/* Put back the colon to allow this to be called again */
1260	if (buf)
1261		*(buf - 1) = ':';
1262
1263	return ret;
1264}
1265
1266/**
1267 * trace_set_clr_event - enable or disable an event
1268 * @system: system name to match (NULL for any system)
1269 * @event: event name to match (NULL for all events, within system)
1270 * @set: 1 to enable, 0 to disable
1271 *
1272 * This is a way for other parts of the kernel to enable or disable
1273 * event recording.
1274 *
1275 * Returns 0 on success, -EINVAL if the parameters do not match any
1276 * registered events.
1277 */
1278int trace_set_clr_event(const char *system, const char *event, int set)
1279{
1280	struct trace_array *tr = top_trace_array();
1281
1282	if (!tr)
1283		return -ENODEV;
1284
1285	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1286}
1287EXPORT_SYMBOL_GPL(trace_set_clr_event);
1288
1289/**
1290 * trace_array_set_clr_event - enable or disable an event for a trace array.
1291 * @tr: concerned trace array.
1292 * @system: system name to match (NULL for any system)
1293 * @event: event name to match (NULL for all events, within system)
1294 * @enable: true to enable, false to disable
1295 *
1296 * This is a way for other parts of the kernel to enable or disable
1297 * event recording.
1298 *
1299 * Returns 0 on success, -EINVAL if the parameters do not match any
1300 * registered events.
1301 */
1302int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1303		const char *event, bool enable)
1304{
1305	int set;
1306
1307	if (!tr)
1308		return -ENOENT;
1309
1310	set = (enable == true) ? 1 : 0;
1311	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1312}
1313EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1314
1315/* 128 should be much more than enough */
1316#define EVENT_BUF_SIZE		127
1317
1318static ssize_t
1319ftrace_event_write(struct file *file, const char __user *ubuf,
1320		   size_t cnt, loff_t *ppos)
1321{
1322	struct trace_parser parser;
1323	struct seq_file *m = file->private_data;
1324	struct trace_array *tr = m->private;
1325	ssize_t read, ret;
1326
1327	if (!cnt)
1328		return 0;
1329
1330	ret = tracing_update_buffers(tr);
1331	if (ret < 0)
1332		return ret;
1333
1334	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1335		return -ENOMEM;
1336
1337	read = trace_get_user(&parser, ubuf, cnt, ppos);
1338
1339	if (read >= 0 && trace_parser_loaded((&parser))) {
1340		int set = 1;
1341
1342		if (*parser.buffer == '!')
1343			set = 0;
1344
 
 
1345		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1346		if (ret)
1347			goto out_put;
1348	}
1349
1350	ret = read;
1351
1352 out_put:
1353	trace_parser_put(&parser);
1354
1355	return ret;
1356}
1357
1358static void *
1359t_next(struct seq_file *m, void *v, loff_t *pos)
1360{
1361	struct trace_event_file *file = v;
1362	struct trace_event_call *call;
1363	struct trace_array *tr = m->private;
1364
1365	(*pos)++;
1366
1367	list_for_each_entry_continue(file, &tr->events, list) {
1368		call = file->event_call;
1369		/*
1370		 * The ftrace subsystem is for showing formats only.
1371		 * They can not be enabled or disabled via the event files.
1372		 */
1373		if (call->class && call->class->reg &&
1374		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1375			return file;
1376	}
1377
1378	return NULL;
1379}
1380
1381static void *t_start(struct seq_file *m, loff_t *pos)
1382{
1383	struct trace_event_file *file;
1384	struct trace_array *tr = m->private;
1385	loff_t l;
1386
1387	mutex_lock(&event_mutex);
1388
1389	file = list_entry(&tr->events, struct trace_event_file, list);
1390	for (l = 0; l <= *pos; ) {
1391		file = t_next(m, file, &l);
1392		if (!file)
1393			break;
1394	}
1395	return file;
1396}
1397
1398static void *
1399s_next(struct seq_file *m, void *v, loff_t *pos)
1400{
1401	struct trace_event_file *file = v;
1402	struct trace_array *tr = m->private;
1403
1404	(*pos)++;
1405
1406	list_for_each_entry_continue(file, &tr->events, list) {
1407		if (file->flags & EVENT_FILE_FL_ENABLED)
1408			return file;
1409	}
1410
1411	return NULL;
1412}
1413
1414static void *s_start(struct seq_file *m, loff_t *pos)
1415{
1416	struct trace_event_file *file;
1417	struct trace_array *tr = m->private;
1418	loff_t l;
1419
1420	mutex_lock(&event_mutex);
1421
1422	file = list_entry(&tr->events, struct trace_event_file, list);
1423	for (l = 0; l <= *pos; ) {
1424		file = s_next(m, file, &l);
1425		if (!file)
1426			break;
1427	}
1428	return file;
1429}
1430
1431static int t_show(struct seq_file *m, void *v)
1432{
1433	struct trace_event_file *file = v;
1434	struct trace_event_call *call = file->event_call;
1435
1436	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1437		seq_printf(m, "%s:", call->class->system);
1438	seq_printf(m, "%s\n", trace_event_name(call));
1439
1440	return 0;
1441}
1442
1443static void t_stop(struct seq_file *m, void *p)
1444{
1445	mutex_unlock(&event_mutex);
1446}
1447
1448static void *
1449__next(struct seq_file *m, void *v, loff_t *pos, int type)
1450{
1451	struct trace_array *tr = m->private;
1452	struct trace_pid_list *pid_list;
1453
1454	if (type == TRACE_PIDS)
1455		pid_list = rcu_dereference_sched(tr->filtered_pids);
1456	else
1457		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1458
1459	return trace_pid_next(pid_list, v, pos);
1460}
1461
1462static void *
1463p_next(struct seq_file *m, void *v, loff_t *pos)
1464{
1465	return __next(m, v, pos, TRACE_PIDS);
1466}
1467
1468static void *
1469np_next(struct seq_file *m, void *v, loff_t *pos)
1470{
1471	return __next(m, v, pos, TRACE_NO_PIDS);
1472}
1473
1474static void *__start(struct seq_file *m, loff_t *pos, int type)
1475	__acquires(RCU)
1476{
1477	struct trace_pid_list *pid_list;
1478	struct trace_array *tr = m->private;
1479
1480	/*
1481	 * Grab the mutex, to keep calls to p_next() having the same
1482	 * tr->filtered_pids as p_start() has.
1483	 * If we just passed the tr->filtered_pids around, then RCU would
1484	 * have been enough, but doing that makes things more complex.
1485	 */
1486	mutex_lock(&event_mutex);
1487	rcu_read_lock_sched();
1488
1489	if (type == TRACE_PIDS)
1490		pid_list = rcu_dereference_sched(tr->filtered_pids);
1491	else
1492		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1493
1494	if (!pid_list)
1495		return NULL;
1496
1497	return trace_pid_start(pid_list, pos);
1498}
1499
1500static void *p_start(struct seq_file *m, loff_t *pos)
1501	__acquires(RCU)
1502{
1503	return __start(m, pos, TRACE_PIDS);
 
1504}
1505
1506static void *np_start(struct seq_file *m, loff_t *pos)
1507	__acquires(RCU)
1508{
1509	return __start(m, pos, TRACE_NO_PIDS);
 
 
 
 
 
 
 
 
1510}
1511
1512static void p_stop(struct seq_file *m, void *p)
1513	__releases(RCU)
1514{
1515	rcu_read_unlock_sched();
1516	mutex_unlock(&event_mutex);
 
 
1517}
1518
1519static ssize_t
1520event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1521		  loff_t *ppos)
1522{
1523	struct trace_event_file *file;
1524	unsigned long flags;
1525	char buf[4] = "0";
1526
1527	mutex_lock(&event_mutex);
1528	file = event_file_file(filp);
1529	if (likely(file))
1530		flags = file->flags;
1531	mutex_unlock(&event_mutex);
1532
1533	if (!file)
1534		return -ENODEV;
1535
1536	if (flags & EVENT_FILE_FL_ENABLED &&
1537	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1538		strcpy(buf, "1");
1539
1540	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1541	    flags & EVENT_FILE_FL_SOFT_MODE)
1542		strcat(buf, "*");
1543
1544	strcat(buf, "\n");
1545
1546	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1547}
1548
1549static ssize_t
1550event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1551		   loff_t *ppos)
1552{
1553	struct trace_event_file *file;
1554	unsigned long val;
1555	int ret;
1556
1557	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1558	if (ret)
1559		return ret;
1560
 
 
 
 
1561	switch (val) {
1562	case 0:
1563	case 1:
1564		ret = -ENODEV;
1565		mutex_lock(&event_mutex);
1566		file = event_file_file(filp);
1567		if (likely(file)) {
1568			ret = tracing_update_buffers(file->tr);
1569			if (ret < 0) {
1570				mutex_unlock(&event_mutex);
1571				return ret;
1572			}
1573			ret = ftrace_event_enable_disable(file, val);
1574		}
1575		mutex_unlock(&event_mutex);
1576		break;
1577
1578	default:
1579		return -EINVAL;
1580	}
1581
1582	*ppos += cnt;
1583
1584	return ret ? ret : cnt;
1585}
1586
1587static ssize_t
1588system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1589		   loff_t *ppos)
1590{
1591	const char set_to_char[4] = { '?', '0', '1', 'X' };
1592	struct trace_subsystem_dir *dir = filp->private_data;
1593	struct event_subsystem *system = dir->subsystem;
1594	struct trace_event_call *call;
1595	struct trace_event_file *file;
1596	struct trace_array *tr = dir->tr;
1597	char buf[2];
1598	int set = 0;
1599	int ret;
1600
1601	mutex_lock(&event_mutex);
1602	list_for_each_entry(file, &tr->events, list) {
1603		call = file->event_call;
1604		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1605		    !trace_event_name(call) || !call->class || !call->class->reg)
1606			continue;
1607
1608		if (system && strcmp(call->class->system, system->name) != 0)
1609			continue;
1610
1611		/*
1612		 * We need to find out if all the events are set
1613		 * or if all events or cleared, or if we have
1614		 * a mixture.
1615		 */
1616		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1617
1618		/*
1619		 * If we have a mixture, no need to look further.
1620		 */
1621		if (set == 3)
1622			break;
1623	}
1624	mutex_unlock(&event_mutex);
1625
1626	buf[0] = set_to_char[set];
1627	buf[1] = '\n';
1628
1629	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1630
1631	return ret;
1632}
1633
1634static ssize_t
1635system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1636		    loff_t *ppos)
1637{
1638	struct trace_subsystem_dir *dir = filp->private_data;
1639	struct event_subsystem *system = dir->subsystem;
1640	const char *name = NULL;
1641	unsigned long val;
1642	ssize_t ret;
1643
1644	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1645	if (ret)
1646		return ret;
1647
1648	ret = tracing_update_buffers(dir->tr);
1649	if (ret < 0)
1650		return ret;
1651
1652	if (val != 0 && val != 1)
1653		return -EINVAL;
1654
1655	/*
1656	 * Opening of "enable" adds a ref count to system,
1657	 * so the name is safe to use.
1658	 */
1659	if (system)
1660		name = system->name;
1661
1662	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1663	if (ret)
1664		goto out;
1665
1666	ret = cnt;
1667
1668out:
1669	*ppos += cnt;
1670
1671	return ret;
1672}
1673
1674enum {
1675	FORMAT_HEADER		= 1,
1676	FORMAT_FIELD_SEPERATOR	= 2,
1677	FORMAT_PRINTFMT		= 3,
1678};
1679
1680static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1681{
1682	struct trace_event_file *file = event_file_data(m->private);
1683	struct trace_event_call *call = file->event_call;
1684	struct list_head *common_head = &ftrace_common_fields;
1685	struct list_head *head = trace_get_fields(call);
1686	struct list_head *node = v;
1687
1688	(*pos)++;
1689
1690	switch ((unsigned long)v) {
1691	case FORMAT_HEADER:
1692		node = common_head;
1693		break;
1694
1695	case FORMAT_FIELD_SEPERATOR:
1696		node = head;
1697		break;
1698
1699	case FORMAT_PRINTFMT:
1700		/* all done */
1701		return NULL;
1702	}
1703
1704	node = node->prev;
1705	if (node == common_head)
1706		return (void *)FORMAT_FIELD_SEPERATOR;
1707	else if (node == head)
1708		return (void *)FORMAT_PRINTFMT;
1709	else
1710		return node;
1711}
1712
1713static int f_show(struct seq_file *m, void *v)
1714{
1715	struct trace_event_file *file = event_file_data(m->private);
1716	struct trace_event_call *call = file->event_call;
1717	struct ftrace_event_field *field;
1718	const char *array_descriptor;
1719
1720	switch ((unsigned long)v) {
1721	case FORMAT_HEADER:
1722		seq_printf(m, "name: %s\n", trace_event_name(call));
1723		seq_printf(m, "ID: %d\n", call->event.type);
1724		seq_puts(m, "format:\n");
1725		return 0;
1726
1727	case FORMAT_FIELD_SEPERATOR:
1728		seq_putc(m, '\n');
1729		return 0;
1730
1731	case FORMAT_PRINTFMT:
1732		seq_printf(m, "\nprint fmt: %s\n",
1733			   call->print_fmt);
1734		return 0;
1735	}
1736
1737	field = list_entry(v, struct ftrace_event_field, link);
1738	/*
1739	 * Smartly shows the array type(except dynamic array).
1740	 * Normal:
1741	 *	field:TYPE VAR
1742	 * If TYPE := TYPE[LEN], it is shown:
1743	 *	field:TYPE VAR[LEN]
1744	 */
1745	array_descriptor = strchr(field->type, '[');
1746
1747	if (str_has_prefix(field->type, "__data_loc"))
1748		array_descriptor = NULL;
1749
1750	if (!array_descriptor)
1751		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1752			   field->type, field->name, field->offset,
1753			   field->size, !!field->is_signed);
1754	else if (field->len)
1755		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1756			   (int)(array_descriptor - field->type),
1757			   field->type, field->name,
1758			   field->len, field->offset,
1759			   field->size, !!field->is_signed);
1760	else
1761		seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1762				(int)(array_descriptor - field->type),
1763				field->type, field->name,
1764				field->offset, field->size, !!field->is_signed);
1765
1766	return 0;
1767}
1768
1769static void *f_start(struct seq_file *m, loff_t *pos)
1770{
1771	struct trace_event_file *file;
1772	void *p = (void *)FORMAT_HEADER;
1773	loff_t l = 0;
1774
1775	/* ->stop() is called even if ->start() fails */
1776	mutex_lock(&event_mutex);
1777	file = event_file_file(m->private);
1778	if (!file)
1779		return ERR_PTR(-ENODEV);
1780
1781	while (l < *pos && p)
1782		p = f_next(m, p, &l);
1783
1784	return p;
1785}
1786
1787static void f_stop(struct seq_file *m, void *p)
1788{
1789	mutex_unlock(&event_mutex);
1790}
1791
1792static const struct seq_operations trace_format_seq_ops = {
1793	.start		= f_start,
1794	.next		= f_next,
1795	.stop		= f_stop,
1796	.show		= f_show,
1797};
1798
1799static int trace_format_open(struct inode *inode, struct file *file)
1800{
1801	struct seq_file *m;
1802	int ret;
1803
1804	/* Do we want to hide event format files on tracefs lockdown? */
1805
1806	ret = seq_open(file, &trace_format_seq_ops);
1807	if (ret < 0)
1808		return ret;
1809
1810	m = file->private_data;
1811	m->private = file;
1812
1813	return 0;
1814}
1815
1816#ifdef CONFIG_PERF_EVENTS
1817static ssize_t
1818event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1819{
1820	int id = (long)event_file_data(filp);
1821	char buf[32];
1822	int len;
1823
 
 
 
1824	if (unlikely(!id))
1825		return -ENODEV;
1826
1827	len = sprintf(buf, "%d\n", id);
1828
1829	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1830}
1831#endif
1832
1833static ssize_t
1834event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1835		  loff_t *ppos)
1836{
1837	struct trace_event_file *file;
1838	struct trace_seq *s;
1839	int r = -ENODEV;
1840
1841	if (*ppos)
1842		return 0;
1843
1844	s = kmalloc(sizeof(*s), GFP_KERNEL);
1845
1846	if (!s)
1847		return -ENOMEM;
1848
1849	trace_seq_init(s);
1850
1851	mutex_lock(&event_mutex);
1852	file = event_file_file(filp);
1853	if (file)
1854		print_event_filter(file, s);
1855	mutex_unlock(&event_mutex);
1856
1857	if (file)
1858		r = simple_read_from_buffer(ubuf, cnt, ppos,
1859					    s->buffer, trace_seq_used(s));
1860
1861	kfree(s);
1862
1863	return r;
1864}
1865
1866static ssize_t
1867event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1868		   loff_t *ppos)
1869{
1870	struct trace_event_file *file;
1871	char *buf;
1872	int err = -ENODEV;
1873
1874	if (cnt >= PAGE_SIZE)
1875		return -EINVAL;
1876
1877	buf = memdup_user_nul(ubuf, cnt);
1878	if (IS_ERR(buf))
1879		return PTR_ERR(buf);
1880
1881	mutex_lock(&event_mutex);
1882	file = event_file_file(filp);
1883	if (file) {
1884		if (file->flags & EVENT_FILE_FL_FREED)
1885			err = -ENODEV;
1886		else
1887			err = apply_event_filter(file, buf);
1888	}
1889	mutex_unlock(&event_mutex);
1890
1891	kfree(buf);
1892	if (err < 0)
1893		return err;
1894
1895	*ppos += cnt;
1896
1897	return cnt;
1898}
1899
1900static LIST_HEAD(event_subsystems);
1901
1902static int subsystem_open(struct inode *inode, struct file *filp)
1903{
1904	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1905	struct trace_array *tr = NULL, *iter_tr;
1906	struct event_subsystem *system = NULL;
 
 
1907	int ret;
1908
1909	if (tracing_is_disabled())
1910		return -ENODEV;
1911
1912	/* Make sure the system still exists */
 
1913	mutex_lock(&event_mutex);
1914	mutex_lock(&trace_types_lock);
1915	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1916		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1917			if (iter_dir == inode->i_private) {
1918				/* Don't open systems with no events */
1919				tr = iter_tr;
1920				dir = iter_dir;
1921				if (dir->nr_events) {
1922					__get_system_dir(dir);
1923					system = dir->subsystem;
1924				}
1925				goto exit_loop;
1926			}
1927		}
1928	}
1929 exit_loop:
 
1930	mutex_unlock(&trace_types_lock);
1931	mutex_unlock(&event_mutex);
1932
1933	if (!system)
1934		return -ENODEV;
1935
 
 
 
1936	/* Still need to increment the ref count of the system */
1937	if (trace_array_get(tr) < 0) {
1938		put_system(dir);
1939		return -ENODEV;
1940	}
1941
1942	ret = tracing_open_generic(inode, filp);
1943	if (ret < 0) {
1944		trace_array_put(tr);
1945		put_system(dir);
1946	}
1947
1948	return ret;
1949}
1950
1951static int system_tr_open(struct inode *inode, struct file *filp)
1952{
1953	struct trace_subsystem_dir *dir;
1954	struct trace_array *tr = inode->i_private;
1955	int ret;
1956
 
 
 
 
 
 
1957	/* Make a temporary dir that has no system but points to tr */
1958	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1959	if (!dir)
 
1960		return -ENOMEM;
 
 
 
1961
1962	ret = tracing_open_generic_tr(inode, filp);
1963	if (ret < 0) {
 
1964		kfree(dir);
1965		return ret;
1966	}
1967	dir->tr = tr;
1968	filp->private_data = dir;
1969
1970	return 0;
1971}
1972
1973static int subsystem_release(struct inode *inode, struct file *file)
1974{
1975	struct trace_subsystem_dir *dir = file->private_data;
1976
1977	trace_array_put(dir->tr);
1978
1979	/*
1980	 * If dir->subsystem is NULL, then this is a temporary
1981	 * descriptor that was made for a trace_array to enable
1982	 * all subsystems.
1983	 */
1984	if (dir->subsystem)
1985		put_system(dir);
1986	else
1987		kfree(dir);
1988
1989	return 0;
1990}
1991
1992static ssize_t
1993subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1994		      loff_t *ppos)
1995{
1996	struct trace_subsystem_dir *dir = filp->private_data;
1997	struct event_subsystem *system = dir->subsystem;
1998	struct trace_seq *s;
1999	int r;
2000
2001	if (*ppos)
2002		return 0;
2003
2004	s = kmalloc(sizeof(*s), GFP_KERNEL);
2005	if (!s)
2006		return -ENOMEM;
2007
2008	trace_seq_init(s);
2009
2010	print_subsystem_event_filter(system, s);
2011	r = simple_read_from_buffer(ubuf, cnt, ppos,
2012				    s->buffer, trace_seq_used(s));
2013
2014	kfree(s);
2015
2016	return r;
2017}
2018
2019static ssize_t
2020subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
2021		       loff_t *ppos)
2022{
2023	struct trace_subsystem_dir *dir = filp->private_data;
2024	char *buf;
2025	int err;
2026
2027	if (cnt >= PAGE_SIZE)
2028		return -EINVAL;
2029
2030	buf = memdup_user_nul(ubuf, cnt);
2031	if (IS_ERR(buf))
2032		return PTR_ERR(buf);
2033
2034	err = apply_subsystem_event_filter(dir, buf);
2035	kfree(buf);
2036	if (err < 0)
2037		return err;
2038
2039	*ppos += cnt;
2040
2041	return cnt;
2042}
2043
2044static ssize_t
2045show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
2046{
2047	struct trace_array *tr = filp->private_data;
2048	struct trace_seq *s;
2049	int r;
2050
2051	if (*ppos)
2052		return 0;
2053
2054	s = kmalloc(sizeof(*s), GFP_KERNEL);
2055	if (!s)
2056		return -ENOMEM;
2057
2058	trace_seq_init(s);
2059
2060	ring_buffer_print_page_header(tr->array_buffer.buffer, s);
2061	r = simple_read_from_buffer(ubuf, cnt, ppos,
2062				    s->buffer, trace_seq_used(s));
2063
2064	kfree(s);
2065
2066	return r;
2067}
2068
2069static ssize_t
2070show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
2071{
2072	struct trace_seq *s;
2073	int r;
2074
2075	if (*ppos)
2076		return 0;
2077
2078	s = kmalloc(sizeof(*s), GFP_KERNEL);
2079	if (!s)
2080		return -ENOMEM;
2081
2082	trace_seq_init(s);
2083
2084	ring_buffer_print_entry_header(s);
2085	r = simple_read_from_buffer(ubuf, cnt, ppos,
2086				    s->buffer, trace_seq_used(s));
2087
2088	kfree(s);
2089
2090	return r;
2091}
2092
2093static void ignore_task_cpu(void *data)
2094{
2095	struct trace_array *tr = data;
2096	struct trace_pid_list *pid_list;
2097	struct trace_pid_list *no_pid_list;
2098
2099	/*
2100	 * This function is called by on_each_cpu() while the
2101	 * event_mutex is held.
2102	 */
2103	pid_list = rcu_dereference_protected(tr->filtered_pids,
2104					     mutex_is_locked(&event_mutex));
2105	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2106					     mutex_is_locked(&event_mutex));
2107
2108	this_cpu_write(tr->array_buffer.data->ignore_pid,
2109		       trace_ignore_this_task(pid_list, no_pid_list, current));
2110}
2111
2112static void register_pid_events(struct trace_array *tr)
2113{
2114	/*
2115	 * Register a probe that is called before all other probes
2116	 * to set ignore_pid if next or prev do not match.
2117	 * Register a probe this is called after all other probes
2118	 * to only keep ignore_pid set if next pid matches.
2119	 */
2120	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
2121					 tr, INT_MAX);
2122	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
2123					 tr, 0);
2124
2125	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
2126					 tr, INT_MAX);
2127	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
2128					 tr, 0);
2129
2130	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
2131					     tr, INT_MAX);
2132	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
2133					     tr, 0);
2134
2135	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
2136					 tr, INT_MAX);
2137	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
2138					 tr, 0);
2139}
2140
2141static ssize_t
2142event_pid_write(struct file *filp, const char __user *ubuf,
2143		size_t cnt, loff_t *ppos, int type)
2144{
2145	struct seq_file *m = filp->private_data;
2146	struct trace_array *tr = m->private;
2147	struct trace_pid_list *filtered_pids = NULL;
2148	struct trace_pid_list *other_pids = NULL;
2149	struct trace_pid_list *pid_list;
2150	struct trace_event_file *file;
2151	ssize_t ret;
 
 
 
 
 
 
2152
2153	if (!cnt)
2154		return 0;
2155
2156	ret = tracing_update_buffers(tr);
2157	if (ret < 0)
2158		return ret;
2159
 
 
 
2160	mutex_lock(&event_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2161
2162	if (type == TRACE_PIDS) {
2163		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2164							  lockdep_is_held(&event_mutex));
2165		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2166							  lockdep_is_held(&event_mutex));
2167	} else {
2168		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2169							  lockdep_is_held(&event_mutex));
2170		other_pids = rcu_dereference_protected(tr->filtered_pids,
2171							  lockdep_is_held(&event_mutex));
 
 
 
 
 
 
 
2172	}
2173
2174	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2175	if (ret < 0)
2176		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2177
2178	if (type == TRACE_PIDS)
2179		rcu_assign_pointer(tr->filtered_pids, pid_list);
2180	else
2181		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2182
2183	list_for_each_entry(file, &tr->events, list) {
2184		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2185	}
2186
2187	if (filtered_pids) {
2188		tracepoint_synchronize_unregister();
2189		trace_pid_list_free(filtered_pids);
2190	} else if (pid_list && !other_pids) {
2191		register_pid_events(tr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2192	}
2193
2194	/*
2195	 * Ignoring of pids is done at task switch. But we have to
2196	 * check for those tasks that are currently running.
2197	 * Always do this in case a pid was appended or removed.
2198	 */
2199	on_each_cpu(ignore_task_cpu, tr, 1);
2200
2201 out:
2202	mutex_unlock(&event_mutex);
2203
2204	if (ret > 0)
2205		*ppos += ret;
2206
2207	return ret;
2208}
2209
2210static ssize_t
2211ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2212		       size_t cnt, loff_t *ppos)
2213{
2214	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2215}
2216
2217static ssize_t
2218ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2219			size_t cnt, loff_t *ppos)
2220{
2221	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2222}
2223
2224static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2225static int ftrace_event_set_open(struct inode *inode, struct file *file);
2226static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2227static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2228static int ftrace_event_release(struct inode *inode, struct file *file);
2229
2230static const struct seq_operations show_event_seq_ops = {
2231	.start = t_start,
2232	.next = t_next,
2233	.show = t_show,
2234	.stop = t_stop,
2235};
2236
2237static const struct seq_operations show_set_event_seq_ops = {
2238	.start = s_start,
2239	.next = s_next,
2240	.show = t_show,
2241	.stop = t_stop,
2242};
2243
2244static const struct seq_operations show_set_pid_seq_ops = {
2245	.start = p_start,
2246	.next = p_next,
2247	.show = trace_pid_show,
2248	.stop = p_stop,
2249};
2250
2251static const struct seq_operations show_set_no_pid_seq_ops = {
2252	.start = np_start,
2253	.next = np_next,
2254	.show = trace_pid_show,
2255	.stop = p_stop,
2256};
2257
2258static const struct file_operations ftrace_avail_fops = {
2259	.open = ftrace_event_avail_open,
2260	.read = seq_read,
2261	.llseek = seq_lseek,
2262	.release = seq_release,
2263};
2264
2265static const struct file_operations ftrace_set_event_fops = {
2266	.open = ftrace_event_set_open,
2267	.read = seq_read,
2268	.write = ftrace_event_write,
2269	.llseek = seq_lseek,
2270	.release = ftrace_event_release,
2271};
2272
2273static const struct file_operations ftrace_set_event_pid_fops = {
2274	.open = ftrace_event_set_pid_open,
2275	.read = seq_read,
2276	.write = ftrace_event_pid_write,
2277	.llseek = seq_lseek,
2278	.release = ftrace_event_release,
2279};
2280
2281static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2282	.open = ftrace_event_set_npid_open,
2283	.read = seq_read,
2284	.write = ftrace_event_npid_write,
2285	.llseek = seq_lseek,
2286	.release = ftrace_event_release,
2287};
2288
2289static const struct file_operations ftrace_enable_fops = {
2290	.open = tracing_open_file_tr,
2291	.read = event_enable_read,
2292	.write = event_enable_write,
2293	.release = tracing_release_file_tr,
2294	.llseek = default_llseek,
2295};
2296
2297static const struct file_operations ftrace_event_format_fops = {
2298	.open = trace_format_open,
2299	.read = seq_read,
2300	.llseek = seq_lseek,
2301	.release = seq_release,
2302};
2303
2304#ifdef CONFIG_PERF_EVENTS
2305static const struct file_operations ftrace_event_id_fops = {
2306	.read = event_id_read,
2307	.llseek = default_llseek,
2308};
2309#endif
2310
2311static const struct file_operations ftrace_event_filter_fops = {
2312	.open = tracing_open_file_tr,
2313	.read = event_filter_read,
2314	.write = event_filter_write,
2315	.release = tracing_release_file_tr,
2316	.llseek = default_llseek,
2317};
2318
2319static const struct file_operations ftrace_subsystem_filter_fops = {
2320	.open = subsystem_open,
2321	.read = subsystem_filter_read,
2322	.write = subsystem_filter_write,
2323	.llseek = default_llseek,
2324	.release = subsystem_release,
2325};
2326
2327static const struct file_operations ftrace_system_enable_fops = {
2328	.open = subsystem_open,
2329	.read = system_enable_read,
2330	.write = system_enable_write,
2331	.llseek = default_llseek,
2332	.release = subsystem_release,
2333};
2334
2335static const struct file_operations ftrace_tr_enable_fops = {
2336	.open = system_tr_open,
2337	.read = system_enable_read,
2338	.write = system_enable_write,
2339	.llseek = default_llseek,
2340	.release = subsystem_release,
2341};
2342
2343static const struct file_operations ftrace_show_header_page_fops = {
2344	.open = tracing_open_generic_tr,
2345	.read = show_header_page_file,
2346	.llseek = default_llseek,
2347	.release = tracing_release_generic_tr,
2348};
2349
2350static const struct file_operations ftrace_show_header_event_fops = {
2351	.open = tracing_open_generic_tr,
2352	.read = show_header_event_file,
2353	.llseek = default_llseek,
2354	.release = tracing_release_generic_tr,
2355};
2356
2357static int
2358ftrace_event_open(struct inode *inode, struct file *file,
2359		  const struct seq_operations *seq_ops)
2360{
2361	struct seq_file *m;
2362	int ret;
2363
2364	ret = security_locked_down(LOCKDOWN_TRACEFS);
2365	if (ret)
2366		return ret;
2367
2368	ret = seq_open(file, seq_ops);
2369	if (ret < 0)
2370		return ret;
2371	m = file->private_data;
2372	/* copy tr over to seq ops */
2373	m->private = inode->i_private;
2374
2375	return ret;
2376}
2377
2378static int ftrace_event_release(struct inode *inode, struct file *file)
2379{
2380	struct trace_array *tr = inode->i_private;
2381
2382	trace_array_put(tr);
2383
2384	return seq_release(inode, file);
2385}
2386
2387static int
2388ftrace_event_avail_open(struct inode *inode, struct file *file)
2389{
2390	const struct seq_operations *seq_ops = &show_event_seq_ops;
2391
2392	/* Checks for tracefs lockdown */
2393	return ftrace_event_open(inode, file, seq_ops);
2394}
2395
2396static int
2397ftrace_event_set_open(struct inode *inode, struct file *file)
2398{
2399	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2400	struct trace_array *tr = inode->i_private;
2401	int ret;
2402
2403	ret = tracing_check_open_get_tr(tr);
2404	if (ret)
2405		return ret;
2406
2407	if ((file->f_mode & FMODE_WRITE) &&
2408	    (file->f_flags & O_TRUNC))
2409		ftrace_clear_events(tr);
2410
2411	ret = ftrace_event_open(inode, file, seq_ops);
2412	if (ret < 0)
2413		trace_array_put(tr);
2414	return ret;
2415}
2416
2417static int
2418ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2419{
2420	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2421	struct trace_array *tr = inode->i_private;
2422	int ret;
2423
2424	ret = tracing_check_open_get_tr(tr);
2425	if (ret)
2426		return ret;
2427
2428	if ((file->f_mode & FMODE_WRITE) &&
2429	    (file->f_flags & O_TRUNC))
2430		ftrace_clear_event_pids(tr, TRACE_PIDS);
2431
2432	ret = ftrace_event_open(inode, file, seq_ops);
2433	if (ret < 0)
2434		trace_array_put(tr);
2435	return ret;
2436}
2437
2438static int
2439ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2440{
2441	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2442	struct trace_array *tr = inode->i_private;
2443	int ret;
2444
2445	ret = tracing_check_open_get_tr(tr);
2446	if (ret)
2447		return ret;
2448
2449	if ((file->f_mode & FMODE_WRITE) &&
2450	    (file->f_flags & O_TRUNC))
2451		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2452
2453	ret = ftrace_event_open(inode, file, seq_ops);
2454	if (ret < 0)
2455		trace_array_put(tr);
2456	return ret;
2457}
2458
2459static struct event_subsystem *
2460create_new_subsystem(const char *name)
2461{
2462	struct event_subsystem *system;
2463
2464	/* need to create new entry */
2465	system = kmalloc(sizeof(*system), GFP_KERNEL);
2466	if (!system)
2467		return NULL;
2468
2469	system->ref_count = 1;
2470
2471	/* Only allocate if dynamic (kprobes and modules) */
2472	system->name = kstrdup_const(name, GFP_KERNEL);
2473	if (!system->name)
2474		goto out_free;
2475
 
 
2476	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2477	if (!system->filter)
2478		goto out_free;
2479
2480	list_add(&system->list, &event_subsystems);
2481
2482	return system;
2483
2484 out_free:
2485	kfree_const(system->name);
2486	kfree(system);
2487	return NULL;
2488}
2489
2490static int system_callback(const char *name, umode_t *mode, void **data,
2491		    const struct file_operations **fops)
2492{
2493	if (strcmp(name, "filter") == 0)
2494		*fops = &ftrace_subsystem_filter_fops;
2495
2496	else if (strcmp(name, "enable") == 0)
2497		*fops = &ftrace_system_enable_fops;
2498
2499	else
2500		return 0;
2501
2502	*mode = TRACE_MODE_WRITE;
2503	return 1;
2504}
2505
2506static struct eventfs_inode *
2507event_subsystem_dir(struct trace_array *tr, const char *name,
2508		    struct trace_event_file *file, struct eventfs_inode *parent)
2509{
2510	struct event_subsystem *system, *iter;
2511	struct trace_subsystem_dir *dir;
2512	struct eventfs_inode *ei;
2513	int nr_entries;
2514	static struct eventfs_entry system_entries[] = {
2515		{
2516			.name		= "filter",
2517			.callback	= system_callback,
2518		},
2519		{
2520			.name		= "enable",
2521			.callback	= system_callback,
2522		}
2523	};
2524
2525	/* First see if we did not already create this dir */
2526	list_for_each_entry(dir, &tr->systems, list) {
2527		system = dir->subsystem;
2528		if (strcmp(system->name, name) == 0) {
2529			dir->nr_events++;
2530			file->system = dir;
2531			return dir->ei;
2532		}
2533	}
2534
2535	/* Now see if the system itself exists. */
2536	system = NULL;
2537	list_for_each_entry(iter, &event_subsystems, list) {
2538		if (strcmp(iter->name, name) == 0) {
2539			system = iter;
2540			break;
2541		}
2542	}
 
 
 
2543
2544	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2545	if (!dir)
2546		goto out_fail;
2547
2548	if (!system) {
2549		system = create_new_subsystem(name);
2550		if (!system)
2551			goto out_free;
2552	} else
2553		__get_system(system);
2554
2555	/* ftrace only has directories no files */
2556	if (strcmp(name, "ftrace") == 0)
2557		nr_entries = 0;
2558	else
2559		nr_entries = ARRAY_SIZE(system_entries);
2560
2561	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2562	if (IS_ERR(ei)) {
2563		pr_warn("Failed to create system directory %s\n", name);
2564		__put_system(system);
2565		goto out_free;
2566	}
2567
2568	dir->ei = ei;
2569	dir->tr = tr;
2570	dir->ref_count = 1;
2571	dir->nr_events = 1;
2572	dir->subsystem = system;
2573	file->system = dir;
2574
 
 
 
 
 
 
 
 
 
 
 
2575	list_add(&dir->list, &tr->systems);
2576
2577	return dir->ei;
2578
2579 out_free:
2580	kfree(dir);
2581 out_fail:
2582	/* Only print this message if failed on memory allocation */
2583	if (!dir || !system)
2584		pr_warn("No memory to create event subsystem %s\n", name);
2585	return NULL;
2586}
2587
2588static int
2589event_define_fields(struct trace_event_call *call)
2590{
 
 
2591	struct list_head *head;
2592	int ret = 0;
 
 
2593
2594	/*
2595	 * Other events may have the same class. Only update
2596	 * the fields if they are not already defined.
2597	 */
2598	head = trace_get_fields(call);
2599	if (list_empty(head)) {
2600		struct trace_event_fields *field = call->class->fields_array;
2601		unsigned int offset = sizeof(struct trace_entry);
 
 
2602
2603		for (; field->type; field++) {
2604			if (field->type == TRACE_FUNCTION_TYPE) {
2605				field->define_fields(call);
2606				break;
2607			}
2608
2609			offset = ALIGN(offset, field->align);
2610			ret = trace_define_field_ext(call, field->type, field->name,
2611						 offset, field->size,
2612						 field->is_signed, field->filter_type,
2613						 field->len, field->needs_test);
2614			if (WARN_ON_ONCE(ret)) {
2615				pr_err("error code is %d\n", ret);
2616				break;
2617			}
2618
2619			offset += field->size;
2620		}
2621	}
2622
2623	return ret;
2624}
 
2625
2626static int event_callback(const char *name, umode_t *mode, void **data,
2627			  const struct file_operations **fops)
2628{
2629	struct trace_event_file *file = *data;
2630	struct trace_event_call *call = file->event_call;
2631
2632	if (strcmp(name, "format") == 0) {
2633		*mode = TRACE_MODE_READ;
2634		*fops = &ftrace_event_format_fops;
2635		return 1;
2636	}
2637
2638	/*
2639	 * Only event directories that can be enabled should have
2640	 * triggers or filters, with the exception of the "print"
2641	 * event that can have a "trigger" file.
2642	 */
2643	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2644		if (call->class->reg && strcmp(name, "enable") == 0) {
2645			*mode = TRACE_MODE_WRITE;
2646			*fops = &ftrace_enable_fops;
2647			return 1;
2648		}
2649
2650		if (strcmp(name, "filter") == 0) {
2651			*mode = TRACE_MODE_WRITE;
2652			*fops = &ftrace_event_filter_fops;
2653			return 1;
2654		}
2655	}
2656
2657	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2658	    strcmp(trace_event_name(call), "print") == 0) {
2659		if (strcmp(name, "trigger") == 0) {
2660			*mode = TRACE_MODE_WRITE;
2661			*fops = &event_trigger_fops;
2662			return 1;
2663		}
2664	}
2665
2666#ifdef CONFIG_PERF_EVENTS
2667	if (call->event.type && call->class->reg &&
2668	    strcmp(name, "id") == 0) {
2669		*mode = TRACE_MODE_READ;
2670		*data = (void *)(long)call->event.type;
2671		*fops = &ftrace_event_id_fops;
2672		return 1;
2673	}
2674#endif
2675
2676#ifdef CONFIG_HIST_TRIGGERS
2677	if (strcmp(name, "hist") == 0) {
2678		*mode = TRACE_MODE_READ;
2679		*fops = &event_hist_fops;
2680		return 1;
2681	}
2682#endif
2683#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2684	if (strcmp(name, "hist_debug") == 0) {
2685		*mode = TRACE_MODE_READ;
2686		*fops = &event_hist_debug_fops;
2687		return 1;
2688	}
2689#endif
2690#ifdef CONFIG_TRACE_EVENT_INJECT
2691	if (call->event.type && call->class->reg &&
2692	    strcmp(name, "inject") == 0) {
2693		*mode = 0200;
2694		*fops = &event_inject_fops;
2695		return 1;
2696	}
2697#endif
2698	return 0;
2699}
2700
2701/* The file is incremented on creation and freeing the enable file decrements it */
2702static void event_release(const char *name, void *data)
2703{
2704	struct trace_event_file *file = data;
2705
2706	event_file_put(file);
2707}
2708
2709static int
2710event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2711{
2712	struct trace_event_call *call = file->event_call;
2713	struct trace_array *tr = file->tr;
2714	struct eventfs_inode *e_events;
2715	struct eventfs_inode *ei;
2716	const char *name;
2717	int nr_entries;
2718	int ret;
2719	static struct eventfs_entry event_entries[] = {
2720		{
2721			.name		= "enable",
2722			.callback	= event_callback,
2723			.release	= event_release,
2724		},
2725		{
2726			.name		= "filter",
2727			.callback	= event_callback,
2728		},
2729		{
2730			.name		= "trigger",
2731			.callback	= event_callback,
2732		},
2733		{
2734			.name		= "format",
2735			.callback	= event_callback,
2736		},
2737#ifdef CONFIG_PERF_EVENTS
2738		{
2739			.name		= "id",
2740			.callback	= event_callback,
2741		},
2742#endif
2743#ifdef CONFIG_HIST_TRIGGERS
2744		{
2745			.name		= "hist",
2746			.callback	= event_callback,
2747		},
2748#endif
2749#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2750		{
2751			.name		= "hist_debug",
2752			.callback	= event_callback,
2753		},
2754#endif
2755#ifdef CONFIG_TRACE_EVENT_INJECT
2756		{
2757			.name		= "inject",
2758			.callback	= event_callback,
2759		},
2760#endif
2761	};
2762
2763	/*
2764	 * If the trace point header did not define TRACE_SYSTEM
2765	 * then the system would be called "TRACE_SYSTEM". This should
2766	 * never happen.
2767	 */
2768	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2769		return -ENODEV;
2770
2771	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2772	if (!e_events)
2773		return -ENOMEM;
2774
2775	nr_entries = ARRAY_SIZE(event_entries);
2776
2777	name = trace_event_name(call);
2778	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2779	if (IS_ERR(ei)) {
2780		pr_warn("Could not create tracefs '%s' directory\n", name);
2781		return -1;
2782	}
2783
2784	file->ei = ei;
2785
2786	ret = event_define_fields(call);
2787	if (ret < 0) {
2788		pr_warn("Could not initialize trace point events/%s\n", name);
2789		return ret;
2790	}
2791
2792	/* Gets decremented on freeing of the "enable" file */
2793	event_file_get(file);
2794
2795	return 0;
2796}
2797
2798static void remove_event_from_tracers(struct trace_event_call *call)
2799{
2800	struct trace_event_file *file;
2801	struct trace_array *tr;
2802
2803	do_for_each_event_file_safe(tr, file) {
2804		if (file->event_call != call)
2805			continue;
2806
2807		remove_event_file_dir(file);
2808		/*
2809		 * The do_for_each_event_file_safe() is
2810		 * a double loop. After finding the call for this
2811		 * trace_array, we use break to jump to the next
2812		 * trace_array.
2813		 */
2814		break;
2815	} while_for_each_event_file();
2816}
2817
2818static void event_remove(struct trace_event_call *call)
2819{
2820	struct trace_array *tr;
2821	struct trace_event_file *file;
2822
2823	do_for_each_event_file(tr, file) {
2824		if (file->event_call != call)
2825			continue;
2826
2827		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2828			tr->clear_trace = true;
2829
2830		ftrace_event_enable_disable(file, 0);
2831		/*
2832		 * The do_for_each_event_file() is
2833		 * a double loop. After finding the call for this
2834		 * trace_array, we use break to jump to the next
2835		 * trace_array.
2836		 */
2837		break;
2838	} while_for_each_event_file();
2839
2840	if (call->event.funcs)
2841		__unregister_trace_event(&call->event);
2842	remove_event_from_tracers(call);
2843	list_del(&call->list);
2844}
2845
2846static int event_init(struct trace_event_call *call)
2847{
2848	int ret = 0;
2849	const char *name;
2850
2851	name = trace_event_name(call);
2852	if (WARN_ON(!name))
2853		return -EINVAL;
2854
2855	if (call->class->raw_init) {
2856		ret = call->class->raw_init(call);
2857		if (ret < 0 && ret != -ENOSYS)
2858			pr_warn("Could not initialize trace events/%s\n", name);
2859	}
2860
2861	return ret;
2862}
2863
2864static int
2865__register_event(struct trace_event_call *call, struct module *mod)
2866{
2867	int ret;
2868
2869	ret = event_init(call);
2870	if (ret < 0)
2871		return ret;
2872
2873	list_add(&call->list, &ftrace_events);
2874	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2875		atomic_set(&call->refcnt, 0);
2876	else
2877		call->module = mod;
2878
2879	return 0;
2880}
2881
2882static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2883{
2884	int rlen;
2885	int elen;
2886
2887	/* Find the length of the eval value as a string */
2888	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2889	/* Make sure there's enough room to replace the string with the value */
2890	if (len < elen)
2891		return NULL;
2892
2893	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2894
2895	/* Get the rest of the string of ptr */
2896	rlen = strlen(ptr + len);
2897	memmove(ptr + elen, ptr + len, rlen);
2898	/* Make sure we end the new string */
2899	ptr[elen + rlen] = 0;
2900
2901	return ptr + elen;
2902}
2903
2904static void update_event_printk(struct trace_event_call *call,
2905				struct trace_eval_map *map)
2906{
2907	char *ptr;
2908	int quote = 0;
2909	int len = strlen(map->eval_string);
2910
2911	for (ptr = call->print_fmt; *ptr; ptr++) {
2912		if (*ptr == '\\') {
2913			ptr++;
2914			/* paranoid */
2915			if (!*ptr)
2916				break;
2917			continue;
2918		}
2919		if (*ptr == '"') {
2920			quote ^= 1;
2921			continue;
2922		}
2923		if (quote)
2924			continue;
2925		if (isdigit(*ptr)) {
2926			/* skip numbers */
2927			do {
2928				ptr++;
2929				/* Check for alpha chars like ULL */
2930			} while (isalnum(*ptr));
2931			if (!*ptr)
2932				break;
2933			/*
2934			 * A number must have some kind of delimiter after
2935			 * it, and we can ignore that too.
2936			 */
2937			continue;
2938		}
2939		if (isalpha(*ptr) || *ptr == '_') {
2940			if (strncmp(map->eval_string, ptr, len) == 0 &&
2941			    !isalnum(ptr[len]) && ptr[len] != '_') {
2942				ptr = eval_replace(ptr, map, len);
2943				/* enum/sizeof string smaller than value */
2944				if (WARN_ON_ONCE(!ptr))
2945					return;
2946				/*
2947				 * No need to decrement here, as eval_replace()
2948				 * returns the pointer to the character passed
2949				 * the eval, and two evals can not be placed
2950				 * back to back without something in between.
2951				 * We can skip that something in between.
2952				 */
2953				continue;
2954			}
2955		skip_more:
2956			do {
2957				ptr++;
2958			} while (isalnum(*ptr) || *ptr == '_');
2959			if (!*ptr)
2960				break;
2961			/*
2962			 * If what comes after this variable is a '.' or
2963			 * '->' then we can continue to ignore that string.
2964			 */
2965			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2966				ptr += *ptr == '.' ? 1 : 2;
2967				if (!*ptr)
2968					break;
2969				goto skip_more;
2970			}
2971			/*
2972			 * Once again, we can skip the delimiter that came
2973			 * after the string.
2974			 */
2975			continue;
2976		}
2977	}
2978}
2979
2980static void add_str_to_module(struct module *module, char *str)
2981{
2982	struct module_string *modstr;
2983
2984	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2985
2986	/*
2987	 * If we failed to allocate memory here, then we'll just
2988	 * let the str memory leak when the module is removed.
2989	 * If this fails to allocate, there's worse problems than
2990	 * a leaked string on module removal.
2991	 */
2992	if (WARN_ON_ONCE(!modstr))
2993		return;
2994
2995	modstr->module = module;
2996	modstr->str = str;
2997
2998	list_add(&modstr->next, &module_strings);
2999}
3000
3001static void update_event_fields(struct trace_event_call *call,
3002				struct trace_eval_map *map)
3003{
3004	struct ftrace_event_field *field;
3005	struct list_head *head;
3006	char *ptr;
3007	char *str;
3008	int len = strlen(map->eval_string);
3009
3010	/* Dynamic events should never have field maps */
3011	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
3012		return;
3013
3014	head = trace_get_fields(call);
3015	list_for_each_entry(field, head, link) {
3016		ptr = strchr(field->type, '[');
3017		if (!ptr)
3018			continue;
3019		ptr++;
3020
3021		if (!isalpha(*ptr) && *ptr != '_')
3022			continue;
3023
3024		if (strncmp(map->eval_string, ptr, len) != 0)
3025			continue;
3026
3027		str = kstrdup(field->type, GFP_KERNEL);
3028		if (WARN_ON_ONCE(!str))
3029			return;
3030		ptr = str + (ptr - field->type);
3031		ptr = eval_replace(ptr, map, len);
3032		/* enum/sizeof string smaller than value */
3033		if (WARN_ON_ONCE(!ptr)) {
3034			kfree(str);
3035			continue;
3036		}
3037
3038		/*
3039		 * If the event is part of a module, then we need to free the string
3040		 * when the module is removed. Otherwise, it will stay allocated
3041		 * until a reboot.
3042		 */
3043		if (call->module)
3044			add_str_to_module(call->module, str);
3045
3046		field->type = str;
3047	}
3048}
3049
3050void trace_event_eval_update(struct trace_eval_map **map, int len)
3051{
3052	struct trace_event_call *call, *p;
3053	const char *last_system = NULL;
3054	bool first = false;
3055	int last_i;
3056	int i;
3057
3058	down_write(&trace_event_sem);
3059	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3060		/* events are usually grouped together with systems */
3061		if (!last_system || call->class->system != last_system) {
3062			first = true;
3063			last_i = 0;
3064			last_system = call->class->system;
3065		}
3066
3067		/*
3068		 * Since calls are grouped by systems, the likelihood that the
3069		 * next call in the iteration belongs to the same system as the
3070		 * previous call is high. As an optimization, we skip searching
3071		 * for a map[] that matches the call's system if the last call
3072		 * was from the same system. That's what last_i is for. If the
3073		 * call has the same system as the previous call, then last_i
3074		 * will be the index of the first map[] that has a matching
3075		 * system.
3076		 */
3077		for (i = last_i; i < len; i++) {
3078			if (call->class->system == map[i]->system) {
3079				/* Save the first system if need be */
3080				if (first) {
3081					last_i = i;
3082					first = false;
3083				}
3084				update_event_printk(call, map[i]);
3085				update_event_fields(call, map[i]);
3086			}
3087		}
3088		cond_resched();
3089	}
3090	up_write(&trace_event_sem);
3091}
3092
3093static bool event_in_systems(struct trace_event_call *call,
3094			     const char *systems)
3095{
3096	const char *system;
3097	const char *p;
3098
3099	if (!systems)
3100		return true;
3101
3102	system = call->class->system;
3103	p = strstr(systems, system);
3104	if (!p)
3105		return false;
3106
3107	if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
3108		return false;
3109
3110	p += strlen(system);
3111	return !*p || isspace(*p) || *p == ',';
3112}
3113
3114static struct trace_event_file *
3115trace_create_new_event(struct trace_event_call *call,
3116		       struct trace_array *tr)
3117{
3118	struct trace_pid_list *no_pid_list;
3119	struct trace_pid_list *pid_list;
3120	struct trace_event_file *file;
3121	unsigned int first;
3122
3123	if (!event_in_systems(call, tr->system_names))
3124		return NULL;
3125
3126	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
3127	if (!file)
3128		return ERR_PTR(-ENOMEM);
3129
3130	pid_list = rcu_dereference_protected(tr->filtered_pids,
3131					     lockdep_is_held(&event_mutex));
3132	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
3133					     lockdep_is_held(&event_mutex));
3134
3135	if (!trace_pid_list_first(pid_list, &first) ||
3136	    !trace_pid_list_first(no_pid_list, &first))
3137		file->flags |= EVENT_FILE_FL_PID_FILTER;
3138
3139	file->event_call = call;
3140	file->tr = tr;
3141	atomic_set(&file->sm_ref, 0);
3142	atomic_set(&file->tm_ref, 0);
3143	INIT_LIST_HEAD(&file->triggers);
3144	list_add(&file->list, &tr->events);
3145	refcount_set(&file->ref, 1);
3146
3147	return file;
3148}
3149
3150#define MAX_BOOT_TRIGGERS 32
3151
3152static struct boot_triggers {
3153	const char		*event;
3154	char			*trigger;
3155} bootup_triggers[MAX_BOOT_TRIGGERS];
3156
3157static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3158static int nr_boot_triggers;
3159
3160static __init int setup_trace_triggers(char *str)
3161{
3162	char *trigger;
3163	char *buf;
3164	int i;
3165
3166	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3167	trace_set_ring_buffer_expanded(NULL);
3168	disable_tracing_selftest("running event triggers");
3169
3170	buf = bootup_trigger_buf;
3171	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3172		trigger = strsep(&buf, ",");
3173		if (!trigger)
3174			break;
3175		bootup_triggers[i].event = strsep(&trigger, ".");
3176		bootup_triggers[i].trigger = trigger;
3177		if (!bootup_triggers[i].trigger)
3178			break;
3179	}
3180
3181	nr_boot_triggers = i;
3182	return 1;
3183}
3184__setup("trace_trigger=", setup_trace_triggers);
3185
3186/* Add an event to a trace directory */
3187static int
3188__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3189{
3190	struct trace_event_file *file;
3191
3192	file = trace_create_new_event(call, tr);
3193	/*
3194	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3195	 * allocation, or NULL if the event is not part of the tr->system_names.
3196	 * When the event is not part of the tr->system_names, return zero, not
3197	 * an error.
3198	 */
3199	if (!file)
3200		return 0;
3201
3202	if (IS_ERR(file))
3203		return PTR_ERR(file);
3204
3205	if (eventdir_initialized)
3206		return event_create_dir(tr->event_dir, file);
3207	else
3208		return event_define_fields(call);
3209}
3210
3211static void trace_early_triggers(struct trace_event_file *file, const char *name)
3212{
3213	int ret;
3214	int i;
3215
3216	for (i = 0; i < nr_boot_triggers; i++) {
3217		if (strcmp(name, bootup_triggers[i].event))
3218			continue;
3219		mutex_lock(&event_mutex);
3220		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3221		mutex_unlock(&event_mutex);
3222		if (ret)
3223			pr_err("Failed to register trigger '%s' on event %s\n",
3224			       bootup_triggers[i].trigger,
3225			       bootup_triggers[i].event);
3226	}
3227}
3228
3229/*
3230 * Just create a descriptor for early init. A descriptor is required
3231 * for enabling events at boot. We want to enable events before
3232 * the filesystem is initialized.
3233 */
3234static int
3235__trace_early_add_new_event(struct trace_event_call *call,
3236			    struct trace_array *tr)
3237{
3238	struct trace_event_file *file;
3239	int ret;
3240
3241	file = trace_create_new_event(call, tr);
3242	/*
3243	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3244	 * allocation, or NULL if the event is not part of the tr->system_names.
3245	 * When the event is not part of the tr->system_names, return zero, not
3246	 * an error.
3247	 */
3248	if (!file)
3249		return 0;
3250
3251	if (IS_ERR(file))
3252		return PTR_ERR(file);
3253
3254	ret = event_define_fields(call);
3255	if (ret)
3256		return ret;
3257
3258	trace_early_triggers(file, trace_event_name(call));
3259
3260	return 0;
3261}
3262
3263struct ftrace_module_file_ops;
3264static void __add_event_to_tracers(struct trace_event_call *call);
3265
3266/* Add an additional event_call dynamically */
3267int trace_add_event_call(struct trace_event_call *call)
3268{
3269	int ret;
3270	lockdep_assert_held(&event_mutex);
3271
3272	mutex_lock(&trace_types_lock);
 
3273
3274	ret = __register_event(call, NULL);
3275	if (ret >= 0)
3276		__add_event_to_tracers(call);
3277
 
3278	mutex_unlock(&trace_types_lock);
3279	return ret;
3280}
3281EXPORT_SYMBOL_GPL(trace_add_event_call);
3282
3283/*
3284 * Must be called under locking of trace_types_lock, event_mutex and
3285 * trace_event_sem.
3286 */
3287static void __trace_remove_event_call(struct trace_event_call *call)
3288{
3289	event_remove(call);
3290	trace_destroy_fields(call);
 
 
3291}
3292
3293static int probe_remove_event_call(struct trace_event_call *call)
3294{
3295	struct trace_array *tr;
3296	struct trace_event_file *file;
3297
3298#ifdef CONFIG_PERF_EVENTS
3299	if (call->perf_refcount)
3300		return -EBUSY;
3301#endif
3302	do_for_each_event_file(tr, file) {
3303		if (file->event_call != call)
3304			continue;
3305		/*
3306		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3307		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3308		 * TRACE_REG_UNREGISTER.
3309		 */
3310		if (file->flags & EVENT_FILE_FL_ENABLED)
3311			goto busy;
3312
3313		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3314			tr->clear_trace = true;
3315		/*
3316		 * The do_for_each_event_file_safe() is
3317		 * a double loop. After finding the call for this
3318		 * trace_array, we use break to jump to the next
3319		 * trace_array.
3320		 */
3321		break;
3322	} while_for_each_event_file();
3323
3324	__trace_remove_event_call(call);
3325
3326	return 0;
3327 busy:
3328	/* No need to clear the trace now */
3329	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3330		tr->clear_trace = false;
3331	}
3332	return -EBUSY;
3333}
3334
3335/* Remove an event_call */
3336int trace_remove_event_call(struct trace_event_call *call)
3337{
3338	int ret;
3339
3340	lockdep_assert_held(&event_mutex);
3341
3342	mutex_lock(&trace_types_lock);
 
3343	down_write(&trace_event_sem);
3344	ret = probe_remove_event_call(call);
3345	up_write(&trace_event_sem);
 
3346	mutex_unlock(&trace_types_lock);
3347
3348	return ret;
3349}
3350EXPORT_SYMBOL_GPL(trace_remove_event_call);
3351
3352#define for_each_event(event, start, end)			\
3353	for (event = start;					\
3354	     (unsigned long)event < (unsigned long)end;		\
3355	     event++)
3356
3357#ifdef CONFIG_MODULES
3358
3359static void trace_module_add_events(struct module *mod)
3360{
3361	struct trace_event_call **call, **start, **end;
3362
3363	if (!mod->num_trace_events)
3364		return;
3365
3366	/* Don't add infrastructure for mods without tracepoints */
3367	if (trace_module_has_bad_taint(mod)) {
3368		pr_err("%s: module has bad taint, not creating trace events\n",
3369		       mod->name);
3370		return;
3371	}
3372
3373	start = mod->trace_events;
3374	end = mod->trace_events + mod->num_trace_events;
3375
3376	for_each_event(call, start, end) {
3377		__register_event(*call, mod);
3378		__add_event_to_tracers(*call);
3379	}
3380}
3381
3382static void trace_module_remove_events(struct module *mod)
3383{
3384	struct trace_event_call *call, *p;
3385	struct module_string *modstr, *m;
3386
3387	down_write(&trace_event_sem);
3388	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3389		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3390			continue;
3391		if (call->module == mod)
3392			__trace_remove_event_call(call);
3393	}
3394	/* Check for any strings allocade for this module */
3395	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3396		if (modstr->module != mod)
3397			continue;
3398		list_del(&modstr->next);
3399		kfree(modstr->str);
3400		kfree(modstr);
3401	}
3402	up_write(&trace_event_sem);
3403
3404	/*
3405	 * It is safest to reset the ring buffer if the module being unloaded
3406	 * registered any events that were used. The only worry is if
3407	 * a new module gets loaded, and takes on the same id as the events
3408	 * of this module. When printing out the buffer, traced events left
3409	 * over from this module may be passed to the new module events and
3410	 * unexpected results may occur.
3411	 */
3412	tracing_reset_all_online_cpus_unlocked();
 
3413}
3414
3415static int trace_module_notify(struct notifier_block *self,
3416			       unsigned long val, void *data)
3417{
3418	struct module *mod = data;
3419
 
3420	mutex_lock(&event_mutex);
3421	mutex_lock(&trace_types_lock);
3422	switch (val) {
3423	case MODULE_STATE_COMING:
3424		trace_module_add_events(mod);
3425		break;
3426	case MODULE_STATE_GOING:
3427		trace_module_remove_events(mod);
3428		break;
3429	}
 
3430	mutex_unlock(&trace_types_lock);
3431	mutex_unlock(&event_mutex);
3432
3433	return NOTIFY_OK;
3434}
3435
3436static struct notifier_block trace_module_nb = {
3437	.notifier_call = trace_module_notify,
3438	.priority = 1, /* higher than trace.c module notify */
3439};
3440#endif /* CONFIG_MODULES */
3441
3442/* Create a new event directory structure for a trace directory. */
3443static void
3444__trace_add_event_dirs(struct trace_array *tr)
3445{
3446	struct trace_event_call *call;
3447	int ret;
3448
3449	list_for_each_entry(call, &ftrace_events, list) {
3450		ret = __trace_add_new_event(call, tr);
3451		if (ret < 0)
3452			pr_warn("Could not create directory for event %s\n",
3453				trace_event_name(call));
3454	}
3455}
3456
3457/* Returns any file that matches the system and event */
3458struct trace_event_file *
3459__find_event_file(struct trace_array *tr, const char *system, const char *event)
3460{
3461	struct trace_event_file *file;
3462	struct trace_event_call *call;
3463	const char *name;
3464
3465	list_for_each_entry(file, &tr->events, list) {
3466
3467		call = file->event_call;
3468		name = trace_event_name(call);
3469
3470		if (!name || !call->class)
 
 
 
3471			continue;
3472
3473		if (strcmp(event, name) == 0 &&
3474		    strcmp(system, call->class->system) == 0)
3475			return file;
3476	}
3477	return NULL;
3478}
3479
3480/* Returns valid trace event files that match system and event */
3481struct trace_event_file *
3482find_event_file(struct trace_array *tr, const char *system, const char *event)
3483{
3484	struct trace_event_file *file;
3485
3486	file = __find_event_file(tr, system, event);
3487	if (!file || !file->event_call->class->reg ||
3488	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3489		return NULL;
3490
3491	return file;
3492}
3493
3494/**
3495 * trace_get_event_file - Find and return a trace event file
3496 * @instance: The name of the trace instance containing the event
3497 * @system: The name of the system containing the event
3498 * @event: The name of the event
3499 *
3500 * Return a trace event file given the trace instance name, trace
3501 * system, and trace event name.  If the instance name is NULL, it
3502 * refers to the top-level trace array.
3503 *
3504 * This function will look it up and return it if found, after calling
3505 * trace_array_get() to prevent the instance from going away, and
3506 * increment the event's module refcount to prevent it from being
3507 * removed.
3508 *
3509 * To release the file, call trace_put_event_file(), which will call
3510 * trace_array_put() and decrement the event's module refcount.
3511 *
3512 * Return: The trace event on success, ERR_PTR otherwise.
3513 */
3514struct trace_event_file *trace_get_event_file(const char *instance,
3515					      const char *system,
3516					      const char *event)
3517{
3518	struct trace_array *tr = top_trace_array();
3519	struct trace_event_file *file = NULL;
3520	int ret = -EINVAL;
3521
3522	if (instance) {
3523		tr = trace_array_find_get(instance);
3524		if (!tr)
3525			return ERR_PTR(-ENOENT);
3526	} else {
3527		ret = trace_array_get(tr);
3528		if (ret)
3529			return ERR_PTR(ret);
3530	}
3531
3532	mutex_lock(&event_mutex);
3533
3534	file = find_event_file(tr, system, event);
3535	if (!file) {
3536		trace_array_put(tr);
3537		ret = -EINVAL;
3538		goto out;
3539	}
3540
3541	/* Don't let event modules unload while in use */
3542	ret = trace_event_try_get_ref(file->event_call);
3543	if (!ret) {
3544		trace_array_put(tr);
3545		ret = -EBUSY;
3546		goto out;
3547	}
3548
3549	ret = 0;
3550 out:
3551	mutex_unlock(&event_mutex);
3552
3553	if (ret)
3554		file = ERR_PTR(ret);
3555
3556	return file;
3557}
3558EXPORT_SYMBOL_GPL(trace_get_event_file);
3559
3560/**
3561 * trace_put_event_file - Release a file from trace_get_event_file()
3562 * @file: The trace event file
3563 *
3564 * If a file was retrieved using trace_get_event_file(), this should
3565 * be called when it's no longer needed.  It will cancel the previous
3566 * trace_array_get() called by that function, and decrement the
3567 * event's module refcount.
3568 */
3569void trace_put_event_file(struct trace_event_file *file)
3570{
3571	mutex_lock(&event_mutex);
3572	trace_event_put_ref(file->event_call);
3573	mutex_unlock(&event_mutex);
3574
3575	trace_array_put(file->tr);
3576}
3577EXPORT_SYMBOL_GPL(trace_put_event_file);
3578
3579#ifdef CONFIG_DYNAMIC_FTRACE
3580
3581/* Avoid typos */
3582#define ENABLE_EVENT_STR	"enable_event"
3583#define DISABLE_EVENT_STR	"disable_event"
3584
3585struct event_probe_data {
3586	struct trace_event_file	*file;
3587	unsigned long			count;
3588	int				ref;
3589	bool				enable;
3590};
3591
3592static void update_event_probe(struct event_probe_data *data)
 
3593{
 
 
 
 
 
 
3594	if (data->enable)
3595		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3596	else
3597		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3598}
3599
3600static void
3601event_enable_probe(unsigned long ip, unsigned long parent_ip,
3602		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3603		   void *data)
3604{
3605	struct ftrace_func_mapper *mapper = data;
3606	struct event_probe_data *edata;
3607	void **pdata;
3608
3609	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3610	if (!pdata || !*pdata)
3611		return;
3612
3613	edata = *pdata;
3614	update_event_probe(edata);
3615}
3616
3617static void
3618event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3619			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3620			 void *data)
3621{
3622	struct ftrace_func_mapper *mapper = data;
3623	struct event_probe_data *edata;
3624	void **pdata;
3625
3626	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3627	if (!pdata || !*pdata)
3628		return;
3629
3630	edata = *pdata;
3631
3632	if (!edata->count)
3633		return;
3634
3635	/* Skip if the event is in a state we want to switch to */
3636	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3637		return;
3638
3639	if (edata->count != -1)
3640		(edata->count)--;
3641
3642	update_event_probe(edata);
3643}
3644
3645static int
3646event_enable_print(struct seq_file *m, unsigned long ip,
3647		   struct ftrace_probe_ops *ops, void *data)
3648{
3649	struct ftrace_func_mapper *mapper = data;
3650	struct event_probe_data *edata;
3651	void **pdata;
3652
3653	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3654
3655	if (WARN_ON_ONCE(!pdata || !*pdata))
3656		return 0;
3657
3658	edata = *pdata;
3659
3660	seq_printf(m, "%ps:", (void *)ip);
3661
3662	seq_printf(m, "%s:%s:%s",
3663		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3664		   edata->file->event_call->class->system,
3665		   trace_event_name(edata->file->event_call));
3666
3667	if (edata->count == -1)
3668		seq_puts(m, ":unlimited\n");
3669	else
3670		seq_printf(m, ":count=%ld\n", edata->count);
3671
3672	return 0;
3673}
3674
3675static int
3676event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3677		  unsigned long ip, void *init_data, void **data)
3678{
3679	struct ftrace_func_mapper *mapper = *data;
3680	struct event_probe_data *edata = init_data;
3681	int ret;
3682
3683	if (!mapper) {
3684		mapper = allocate_ftrace_func_mapper();
3685		if (!mapper)
3686			return -ENODEV;
3687		*data = mapper;
3688	}
3689
3690	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3691	if (ret < 0)
3692		return ret;
3693
3694	edata->ref++;
3695
3696	return 0;
3697}
3698
3699static int free_probe_data(void *data)
3700{
3701	struct event_probe_data *edata = data;
3702
3703	edata->ref--;
3704	if (!edata->ref) {
3705		/* Remove the SOFT_MODE flag */
3706		__ftrace_event_enable_disable(edata->file, 0, 1);
3707		trace_event_put_ref(edata->file->event_call);
3708		kfree(edata);
3709	}
3710	return 0;
3711}
3712
3713static void
3714event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3715		  unsigned long ip, void *data)
3716{
3717	struct ftrace_func_mapper *mapper = data;
3718	struct event_probe_data *edata;
3719
3720	if (!ip) {
3721		if (!mapper)
3722			return;
3723		free_ftrace_func_mapper(mapper, free_probe_data);
3724		return;
 
 
 
 
 
 
 
3725	}
3726
3727	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3728
3729	if (WARN_ON_ONCE(!edata))
3730		return;
3731
3732	if (WARN_ON_ONCE(edata->ref <= 0))
3733		return;
3734
3735	free_probe_data(edata);
3736}
3737
3738static struct ftrace_probe_ops event_enable_probe_ops = {
3739	.func			= event_enable_probe,
3740	.print			= event_enable_print,
3741	.init			= event_enable_init,
3742	.free			= event_enable_free,
3743};
3744
3745static struct ftrace_probe_ops event_enable_count_probe_ops = {
3746	.func			= event_enable_count_probe,
3747	.print			= event_enable_print,
3748	.init			= event_enable_init,
3749	.free			= event_enable_free,
3750};
3751
3752static struct ftrace_probe_ops event_disable_probe_ops = {
3753	.func			= event_enable_probe,
3754	.print			= event_enable_print,
3755	.init			= event_enable_init,
3756	.free			= event_enable_free,
3757};
3758
3759static struct ftrace_probe_ops event_disable_count_probe_ops = {
3760	.func			= event_enable_count_probe,
3761	.print			= event_enable_print,
3762	.init			= event_enable_init,
3763	.free			= event_enable_free,
3764};
3765
3766static int
3767event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3768		  char *glob, char *cmd, char *param, int enabled)
3769{
 
3770	struct trace_event_file *file;
3771	struct ftrace_probe_ops *ops;
3772	struct event_probe_data *data;
3773	const char *system;
3774	const char *event;
3775	char *number;
3776	bool enable;
3777	int ret;
3778
3779	if (!tr)
3780		return -ENODEV;
3781
3782	/* hash funcs only work with set_ftrace_filter */
3783	if (!enabled || !param)
3784		return -EINVAL;
3785
3786	system = strsep(&param, ":");
3787	if (!param)
3788		return -EINVAL;
3789
3790	event = strsep(&param, ":");
3791
3792	mutex_lock(&event_mutex);
3793
3794	ret = -EINVAL;
3795	file = find_event_file(tr, system, event);
3796	if (!file)
3797		goto out;
3798
3799	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3800
3801	if (enable)
3802		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3803	else
3804		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3805
3806	if (glob[0] == '!') {
3807		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
 
3808		goto out;
3809	}
3810
3811	ret = -ENOMEM;
3812
3813	data = kzalloc(sizeof(*data), GFP_KERNEL);
3814	if (!data)
3815		goto out;
3816
3817	data->enable = enable;
3818	data->count = -1;
3819	data->file = file;
3820
3821	if (!param)
3822		goto out_reg;
3823
3824	number = strsep(&param, ":");
3825
3826	ret = -EINVAL;
3827	if (!strlen(number))
3828		goto out_free;
3829
3830	/*
3831	 * We use the callback data field (which is a pointer)
3832	 * as our counter.
3833	 */
3834	ret = kstrtoul(number, 0, &data->count);
3835	if (ret)
3836		goto out_free;
3837
3838 out_reg:
3839	/* Don't let event modules unload while probe registered */
3840	ret = trace_event_try_get_ref(file->event_call);
3841	if (!ret) {
3842		ret = -EBUSY;
3843		goto out_free;
3844	}
3845
3846	ret = __ftrace_event_enable_disable(file, 1, 1);
3847	if (ret < 0)
3848		goto out_put;
3849
3850	ret = register_ftrace_function_probe(glob, tr, ops, data);
3851	/*
3852	 * The above returns on success the # of functions enabled,
3853	 * but if it didn't find any functions it returns zero.
3854	 * Consider no functions a failure too.
3855	 */
3856	if (!ret) {
3857		ret = -ENOENT;
3858		goto out_disable;
3859	} else if (ret < 0)
3860		goto out_disable;
3861	/* Just return zero, not the number of enabled functions */
3862	ret = 0;
3863 out:
3864	mutex_unlock(&event_mutex);
3865	return ret;
3866
3867 out_disable:
3868	__ftrace_event_enable_disable(file, 0, 1);
3869 out_put:
3870	trace_event_put_ref(file->event_call);
3871 out_free:
3872	kfree(data);
3873	goto out;
3874}
3875
3876static struct ftrace_func_command event_enable_cmd = {
3877	.name			= ENABLE_EVENT_STR,
3878	.func			= event_enable_func,
3879};
3880
3881static struct ftrace_func_command event_disable_cmd = {
3882	.name			= DISABLE_EVENT_STR,
3883	.func			= event_enable_func,
3884};
3885
3886static __init int register_event_cmds(void)
3887{
3888	int ret;
3889
3890	ret = register_ftrace_command(&event_enable_cmd);
3891	if (WARN_ON(ret < 0))
3892		return ret;
3893	ret = register_ftrace_command(&event_disable_cmd);
3894	if (WARN_ON(ret < 0))
3895		unregister_ftrace_command(&event_enable_cmd);
3896	return ret;
3897}
3898#else
3899static inline int register_event_cmds(void) { return 0; }
3900#endif /* CONFIG_DYNAMIC_FTRACE */
3901
3902/*
3903 * The top level array and trace arrays created by boot-time tracing
3904 * have already had its trace_event_file descriptors created in order
3905 * to allow for early events to be recorded.
3906 * This function is called after the tracefs has been initialized,
3907 * and we now have to create the files associated to the events.
3908 */
3909static void __trace_early_add_event_dirs(struct trace_array *tr)
 
3910{
3911	struct trace_event_file *file;
3912	int ret;
3913
3914
3915	list_for_each_entry(file, &tr->events, list) {
3916		ret = event_create_dir(tr->event_dir, file);
3917		if (ret < 0)
3918			pr_warn("Could not create directory for event %s\n",
3919				trace_event_name(file->event_call));
3920	}
3921}
3922
3923/*
3924 * For early boot up, the top trace array and the trace arrays created
3925 * by boot-time tracing require to have a list of events that can be
3926 * enabled. This must be done before the filesystem is set up in order
3927 * to allow events to be traced early.
3928 */
3929void __trace_early_add_events(struct trace_array *tr)
 
3930{
3931	struct trace_event_call *call;
3932	int ret;
3933
3934	list_for_each_entry(call, &ftrace_events, list) {
3935		/* Early boot up should not have any modules loaded */
3936		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3937		    WARN_ON_ONCE(call->module))
3938			continue;
3939
3940		ret = __trace_early_add_new_event(call, tr);
3941		if (ret < 0)
3942			pr_warn("Could not create early event %s\n",
3943				trace_event_name(call));
3944	}
3945}
3946
3947/* Remove the event directory structure for a trace directory. */
3948static void
3949__trace_remove_event_dirs(struct trace_array *tr)
3950{
3951	struct trace_event_file *file, *next;
3952
3953	list_for_each_entry_safe(file, next, &tr->events, list)
3954		remove_event_file_dir(file);
3955}
3956
3957static void __add_event_to_tracers(struct trace_event_call *call)
3958{
3959	struct trace_array *tr;
3960
3961	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3962		__trace_add_new_event(call, tr);
3963}
3964
3965extern struct trace_event_call *__start_ftrace_events[];
3966extern struct trace_event_call *__stop_ftrace_events[];
3967
3968static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3969
3970static __init int setup_trace_event(char *str)
3971{
3972	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3973	trace_set_ring_buffer_expanded(NULL);
3974	disable_tracing_selftest("running event tracing");
3975
3976	return 1;
3977}
3978__setup("trace_event=", setup_trace_event);
3979
3980static int events_callback(const char *name, umode_t *mode, void **data,
3981			   const struct file_operations **fops)
3982{
3983	if (strcmp(name, "enable") == 0) {
3984		*mode = TRACE_MODE_WRITE;
3985		*fops = &ftrace_tr_enable_fops;
3986		return 1;
3987	}
3988
3989	if (strcmp(name, "header_page") == 0) {
3990		*mode = TRACE_MODE_READ;
3991		*fops = &ftrace_show_header_page_fops;
3992
3993	} else if (strcmp(name, "header_event") == 0) {
3994		*mode = TRACE_MODE_READ;
3995		*fops = &ftrace_show_header_event_fops;
3996	} else
3997		return 0;
3998
3999	return 1;
4000}
4001
4002/* Expects to have event_mutex held when called */
4003static int
4004create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
4005{
4006	struct eventfs_inode *e_events;
4007	struct dentry *entry;
4008	int nr_entries;
4009	static struct eventfs_entry events_entries[] = {
4010		{
4011			.name		= "enable",
4012			.callback	= events_callback,
4013		},
4014		{
4015			.name		= "header_page",
4016			.callback	= events_callback,
4017		},
4018		{
4019			.name		= "header_event",
4020			.callback	= events_callback,
4021		},
4022	};
4023
4024	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
4025				  tr, &ftrace_set_event_fops);
4026	if (!entry)
 
4027		return -ENOMEM;
 
4028
4029	nr_entries = ARRAY_SIZE(events_entries);
4030
4031	e_events = eventfs_create_events_dir("events", parent, events_entries,
4032					     nr_entries, tr);
4033	if (IS_ERR(e_events)) {
4034		pr_warn("Could not create tracefs 'events' directory\n");
4035		return -ENOMEM;
4036	}
4037
4038	/* There are not as crucial, just warn if they are not created */
 
4039
4040	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
4041			  tr, &ftrace_set_event_pid_fops);
 
 
 
 
 
 
4042
4043	trace_create_file("set_event_notrace_pid",
4044			  TRACE_MODE_WRITE, parent, tr,
4045			  &ftrace_set_event_notrace_pid_fops);
4046
4047	tr->event_dir = e_events;
4048
4049	return 0;
4050}
4051
4052/**
4053 * event_trace_add_tracer - add a instance of a trace_array to events
4054 * @parent: The parent dentry to place the files/directories for events in
4055 * @tr: The trace array associated with these events
4056 *
4057 * When a new instance is created, it needs to set up its events
4058 * directory, as well as other files associated with events. It also
4059 * creates the event hierarchy in the @parent/events directory.
4060 *
4061 * Returns 0 on success.
4062 *
4063 * Must be called with event_mutex held.
4064 */
4065int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
4066{
4067	int ret;
4068
4069	lockdep_assert_held(&event_mutex);
4070
4071	ret = create_event_toplevel_files(parent, tr);
4072	if (ret)
4073		goto out;
4074
4075	down_write(&trace_event_sem);
4076	/* If tr already has the event list, it is initialized in early boot. */
4077	if (unlikely(!list_empty(&tr->events)))
4078		__trace_early_add_event_dirs(tr);
4079	else
4080		__trace_add_event_dirs(tr);
4081	up_write(&trace_event_sem);
4082
4083 out:
 
 
4084	return ret;
4085}
4086
4087/*
4088 * The top trace array already had its file descriptors created.
4089 * Now the files themselves need to be created.
4090 */
4091static __init int
4092early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
4093{
4094	int ret;
4095
4096	mutex_lock(&event_mutex);
4097
4098	ret = create_event_toplevel_files(parent, tr);
4099	if (ret)
4100		goto out_unlock;
4101
4102	down_write(&trace_event_sem);
4103	__trace_early_add_event_dirs(tr);
4104	up_write(&trace_event_sem);
4105
4106 out_unlock:
4107	mutex_unlock(&event_mutex);
4108
4109	return ret;
4110}
4111
4112/* Must be called with event_mutex held */
4113int event_trace_del_tracer(struct trace_array *tr)
4114{
4115	lockdep_assert_held(&event_mutex);
4116
4117	/* Disable any event triggers and associated soft-disabled events */
4118	clear_event_triggers(tr);
4119
4120	/* Clear the pid list */
4121	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
4122
4123	/* Disable any running events */
4124	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
4125
4126	/* Make sure no more events are being executed */
4127	tracepoint_synchronize_unregister();
4128
4129	down_write(&trace_event_sem);
4130	__trace_remove_event_dirs(tr);
4131	eventfs_remove_events_dir(tr->event_dir);
4132	up_write(&trace_event_sem);
4133
4134	tr->event_dir = NULL;
4135
 
 
4136	return 0;
4137}
4138
4139static __init int event_trace_memsetup(void)
4140{
4141	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
4142	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
4143	return 0;
4144}
4145
4146__init void
4147early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4148{
 
4149	char *token;
4150	int ret;
4151
4152	while (true) {
4153		token = strsep(&buf, ",");
4154
4155		if (!token)
4156			break;
4157
4158		if (*token) {
4159			/* Restarting syscalls requires that we stop them first */
4160			if (disable_first)
4161				ftrace_set_clr_event(tr, token, 0);
4162
4163			ret = ftrace_set_clr_event(tr, token, 1);
4164			if (ret)
4165				pr_warn("Failed to enable trace event: %s\n", token);
4166		}
4167
4168		/* Put back the comma to allow this to be called again */
4169		if (buf)
4170			*(buf - 1) = ',';
4171	}
4172}
4173
4174static __init int event_trace_enable(void)
4175{
4176	struct trace_array *tr = top_trace_array();
4177	struct trace_event_call **iter, *call;
4178	int ret;
4179
4180	if (!tr)
4181		return -ENODEV;
4182
4183	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4184
4185		call = *iter;
4186		ret = event_init(call);
4187		if (!ret)
4188			list_add(&call->list, &ftrace_events);
4189	}
4190
4191	register_trigger_cmds();
4192
4193	/*
4194	 * We need the top trace array to have a working set of trace
4195	 * points at early init, before the debug files and directories
4196	 * are created. Create the file entries now, and attach them
4197	 * to the actual file dentries later.
4198	 */
4199	__trace_early_add_events(tr);
4200
4201	early_enable_events(tr, bootup_event_buf, false);
4202
4203	trace_printk_start_comm();
4204
4205	register_event_cmds();
4206
 
4207
4208	return 0;
4209}
4210
4211/*
4212 * event_trace_enable() is called from trace_event_init() first to
4213 * initialize events and perhaps start any events that are on the
4214 * command line. Unfortunately, there are some events that will not
4215 * start this early, like the system call tracepoints that need
4216 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4217 * event_trace_enable() is called before pid 1 starts, and this flag
4218 * is never set, making the syscall tracepoint never get reached, but
4219 * the event is enabled regardless (and not doing anything).
4220 */
4221static __init int event_trace_enable_again(void)
4222{
4223	struct trace_array *tr;
4224
4225	tr = top_trace_array();
4226	if (!tr)
4227		return -ENODEV;
4228
4229	early_enable_events(tr, bootup_event_buf, true);
4230
4231	return 0;
4232}
4233
4234early_initcall(event_trace_enable_again);
4235
4236/* Init fields which doesn't related to the tracefs */
4237static __init int event_trace_init_fields(void)
4238{
4239	if (trace_define_generic_fields())
4240		pr_warn("tracing: Failed to allocated generic fields");
4241
4242	if (trace_define_common_fields())
4243		pr_warn("tracing: Failed to allocate common fields");
4244
4245	return 0;
4246}
4247
4248__init int event_trace_init(void)
4249{
4250	struct trace_array *tr;
 
 
4251	int ret;
4252
4253	tr = top_trace_array();
4254	if (!tr)
4255		return -ENODEV;
4256
4257	trace_create_file("available_events", TRACE_MODE_READ,
4258			  NULL, tr, &ftrace_avail_fops);
 
4259
4260	ret = early_event_add_tracer(NULL, tr);
 
 
 
 
 
 
 
 
 
 
 
4261	if (ret)
4262		return ret;
4263
4264#ifdef CONFIG_MODULES
4265	ret = register_module_notifier(&trace_module_nb);
4266	if (ret)
4267		pr_warn("Failed to register trace events module notifier\n");
4268#endif
4269
4270	eventdir_initialized = true;
4271
4272	return 0;
4273}
4274
4275void __init trace_event_init(void)
4276{
4277	event_trace_memsetup();
4278	init_ftrace_syscalls();
4279	event_trace_enable();
4280	event_trace_init_fields();
4281}
4282
4283#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
 
 
4284
4285static DEFINE_SPINLOCK(test_spinlock);
4286static DEFINE_SPINLOCK(test_spinlock_irq);
4287static DEFINE_MUTEX(test_mutex);
4288
4289static __init void test_work(struct work_struct *dummy)
4290{
4291	spin_lock(&test_spinlock);
4292	spin_lock_irq(&test_spinlock_irq);
4293	udelay(1);
4294	spin_unlock_irq(&test_spinlock_irq);
4295	spin_unlock(&test_spinlock);
4296
4297	mutex_lock(&test_mutex);
4298	msleep(1);
4299	mutex_unlock(&test_mutex);
4300}
4301
4302static __init int event_test_thread(void *unused)
4303{
4304	void *test_malloc;
4305
4306	test_malloc = kmalloc(1234, GFP_KERNEL);
4307	if (!test_malloc)
4308		pr_info("failed to kmalloc\n");
4309
4310	schedule_on_each_cpu(test_work);
4311
4312	kfree(test_malloc);
4313
4314	set_current_state(TASK_INTERRUPTIBLE);
4315	while (!kthread_should_stop()) {
4316		schedule();
4317		set_current_state(TASK_INTERRUPTIBLE);
4318	}
4319	__set_current_state(TASK_RUNNING);
4320
4321	return 0;
4322}
4323
4324/*
4325 * Do various things that may trigger events.
4326 */
4327static __init void event_test_stuff(void)
4328{
4329	struct task_struct *test_thread;
4330
4331	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4332	msleep(1);
4333	kthread_stop(test_thread);
4334}
4335
4336/*
4337 * For every trace event defined, we will test each trace point separately,
4338 * and then by groups, and finally all trace points.
4339 */
4340static __init void event_trace_self_tests(void)
4341{
4342	struct trace_subsystem_dir *dir;
4343	struct trace_event_file *file;
4344	struct trace_event_call *call;
4345	struct event_subsystem *system;
4346	struct trace_array *tr;
4347	int ret;
4348
4349	tr = top_trace_array();
4350	if (!tr)
4351		return;
4352
4353	pr_info("Running tests on trace events:\n");
4354
4355	list_for_each_entry(file, &tr->events, list) {
4356
4357		call = file->event_call;
4358
4359		/* Only test those that have a probe */
4360		if (!call->class || !call->class->probe)
4361			continue;
4362
4363/*
4364 * Testing syscall events here is pretty useless, but
4365 * we still do it if configured. But this is time consuming.
4366 * What we really need is a user thread to perform the
4367 * syscalls as we test.
4368 */
4369#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4370		if (call->class->system &&
4371		    strcmp(call->class->system, "syscalls") == 0)
4372			continue;
4373#endif
4374
4375		pr_info("Testing event %s: ", trace_event_name(call));
4376
4377		/*
4378		 * If an event is already enabled, someone is using
4379		 * it and the self test should not be on.
4380		 */
4381		if (file->flags & EVENT_FILE_FL_ENABLED) {
4382			pr_warn("Enabled event during self test!\n");
4383			WARN_ON_ONCE(1);
4384			continue;
4385		}
4386
4387		ftrace_event_enable_disable(file, 1);
4388		event_test_stuff();
4389		ftrace_event_enable_disable(file, 0);
4390
4391		pr_cont("OK\n");
4392	}
4393
4394	/* Now test at the sub system level */
4395
4396	pr_info("Running tests on trace event systems:\n");
4397
4398	list_for_each_entry(dir, &tr->systems, list) {
4399
4400		system = dir->subsystem;
4401
4402		/* the ftrace system is special, skip it */
4403		if (strcmp(system->name, "ftrace") == 0)
4404			continue;
4405
4406		pr_info("Testing event system %s: ", system->name);
4407
4408		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4409		if (WARN_ON_ONCE(ret)) {
4410			pr_warn("error enabling system %s\n",
4411				system->name);
4412			continue;
4413		}
4414
4415		event_test_stuff();
4416
4417		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4418		if (WARN_ON_ONCE(ret)) {
4419			pr_warn("error disabling system %s\n",
4420				system->name);
4421			continue;
4422		}
4423
4424		pr_cont("OK\n");
4425	}
4426
4427	/* Test with all events enabled */
4428
4429	pr_info("Running tests on all trace events:\n");
4430	pr_info("Testing all events: ");
4431
4432	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4433	if (WARN_ON_ONCE(ret)) {
4434		pr_warn("error enabling all events\n");
4435		return;
4436	}
4437
4438	event_test_stuff();
4439
4440	/* reset sysname */
4441	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4442	if (WARN_ON_ONCE(ret)) {
4443		pr_warn("error disabling all events\n");
4444		return;
4445	}
4446
4447	pr_cont("OK\n");
4448}
4449
4450#ifdef CONFIG_FUNCTION_TRACER
4451
4452static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4453
4454static struct trace_event_file event_trace_file __initdata;
4455
4456static void __init
4457function_test_events_call(unsigned long ip, unsigned long parent_ip,
4458			  struct ftrace_ops *op, struct ftrace_regs *regs)
4459{
4460	struct trace_buffer *buffer;
4461	struct ring_buffer_event *event;
 
4462	struct ftrace_entry *entry;
4463	unsigned int trace_ctx;
4464	long disabled;
4465	int cpu;
 
4466
4467	trace_ctx = tracing_gen_ctx();
4468	preempt_disable_notrace();
4469	cpu = raw_smp_processor_id();
4470	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4471
4472	if (disabled != 1)
4473		goto out;
4474
4475	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4476						TRACE_FN, sizeof(*entry),
4477						trace_ctx);
 
 
4478	if (!event)
4479		goto out;
4480	entry	= ring_buffer_event_data(event);
4481	entry->ip			= ip;
4482	entry->parent_ip		= parent_ip;
4483
4484	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4485				    entry, trace_ctx);
4486 out:
4487	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4488	preempt_enable_notrace();
4489}
4490
4491static struct ftrace_ops trace_ops __initdata  =
4492{
4493	.func = function_test_events_call,
 
4494};
4495
4496static __init void event_trace_self_test_with_function(void)
4497{
4498	int ret;
4499
4500	event_trace_file.tr = top_trace_array();
4501	if (WARN_ON(!event_trace_file.tr))
4502		return;
4503
4504	ret = register_ftrace_function(&trace_ops);
4505	if (WARN_ON(ret < 0)) {
4506		pr_info("Failed to enable function tracer for event tests\n");
4507		return;
4508	}
4509	pr_info("Running tests again, along with the function tracer\n");
4510	event_trace_self_tests();
4511	unregister_ftrace_function(&trace_ops);
4512}
4513#else
4514static __init void event_trace_self_test_with_function(void)
4515{
4516}
4517#endif
4518
4519static __init int event_trace_self_tests_init(void)
4520{
4521	if (!tracing_selftest_disabled) {
4522		event_trace_self_tests();
4523		event_trace_self_test_with_function();
4524	}
4525
4526	return 0;
4527}
4528
4529late_initcall(event_trace_self_tests_init);
4530
4531#endif