Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * event tracer
   4 *
   5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6 *
   7 *  - Added format output of fields of the trace point.
   8 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9 *
  10 */
  11
  12#define pr_fmt(fmt) fmt
  13
  14#include <linux/workqueue.h>
  15#include <linux/security.h>
  16#include <linux/spinlock.h>
  17#include <linux/kthread.h>
  18#include <linux/tracefs.h>
  19#include <linux/uaccess.h>
  20#include <linux/module.h>
  21#include <linux/ctype.h>
  22#include <linux/sort.h>
  23#include <linux/slab.h>
  24#include <linux/delay.h>
  25
  26#include <trace/events/sched.h>
 
  27
  28#include <asm/setup.h>
  29
  30#include "trace_output.h"
  31
  32#undef TRACE_SYSTEM
  33#define TRACE_SYSTEM "TRACE_SYSTEM"
  34
  35DEFINE_MUTEX(event_mutex);
  36
  37LIST_HEAD(ftrace_events);
  38static LIST_HEAD(ftrace_generic_fields);
  39static LIST_HEAD(ftrace_common_fields);
 
 
 
 
 
 
 
 
 
  40
  41#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  42
  43static struct kmem_cache *field_cachep;
  44static struct kmem_cache *file_cachep;
  45
  46static inline int system_refcount(struct event_subsystem *system)
  47{
  48	return system->ref_count;
  49}
  50
  51static int system_refcount_inc(struct event_subsystem *system)
  52{
  53	return system->ref_count++;
  54}
  55
  56static int system_refcount_dec(struct event_subsystem *system)
  57{
  58	return --system->ref_count;
  59}
  60
  61/* Double loops, do not use break, only goto's work */
  62#define do_for_each_event_file(tr, file)			\
  63	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  64		list_for_each_entry(file, &tr->events, list)
  65
  66#define do_for_each_event_file_safe(tr, file)			\
  67	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  68		struct trace_event_file *___n;				\
  69		list_for_each_entry_safe(file, ___n, &tr->events, list)
  70
  71#define while_for_each_event_file()		\
  72	}
  73
  74static struct ftrace_event_field *
  75__find_event_field(struct list_head *head, char *name)
  76{
  77	struct ftrace_event_field *field;
  78
  79	list_for_each_entry(field, head, link) {
  80		if (!strcmp(field->name, name))
  81			return field;
  82	}
  83
  84	return NULL;
  85}
  86
  87struct ftrace_event_field *
  88trace_find_event_field(struct trace_event_call *call, char *name)
  89{
  90	struct ftrace_event_field *field;
  91	struct list_head *head;
  92
  93	head = trace_get_fields(call);
  94	field = __find_event_field(head, name);
  95	if (field)
  96		return field;
  97
  98	field = __find_event_field(&ftrace_generic_fields, name);
  99	if (field)
 100		return field;
 101
 102	return __find_event_field(&ftrace_common_fields, name);
 103}
 104
 105static int __trace_define_field(struct list_head *head, const char *type,
 106				const char *name, int offset, int size,
 107				int is_signed, int filter_type)
 108{
 109	struct ftrace_event_field *field;
 110
 111	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 112	if (!field)
 113		return -ENOMEM;
 114
 115	field->name = name;
 116	field->type = type;
 117
 118	if (filter_type == FILTER_OTHER)
 119		field->filter_type = filter_assign_type(type);
 120	else
 121		field->filter_type = filter_type;
 122
 123	field->offset = offset;
 124	field->size = size;
 125	field->is_signed = is_signed;
 
 126
 127	list_add(&field->link, head);
 128
 129	return 0;
 130}
 131
 132int trace_define_field(struct trace_event_call *call, const char *type,
 133		       const char *name, int offset, int size, int is_signed,
 134		       int filter_type)
 135{
 136	struct list_head *head;
 137
 138	if (WARN_ON(!call->class))
 139		return 0;
 140
 141	head = trace_get_fields(call);
 142	return __trace_define_field(head, type, name, offset, size,
 143				    is_signed, filter_type);
 144}
 145EXPORT_SYMBOL_GPL(trace_define_field);
 146
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 147#define __generic_field(type, item, filter_type)			\
 148	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 149				   #item, 0, 0, is_signed_type(type),	\
 150				   filter_type);			\
 151	if (ret)							\
 152		return ret;
 153
 154#define __common_field(type, item)					\
 155	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 156				   "common_" #item,			\
 157				   offsetof(typeof(ent), item),		\
 158				   sizeof(ent.item),			\
 159				   is_signed_type(type), FILTER_OTHER);	\
 160	if (ret)							\
 161		return ret;
 162
 163static int trace_define_generic_fields(void)
 164{
 165	int ret;
 166
 167	__generic_field(int, CPU, FILTER_CPU);
 168	__generic_field(int, cpu, FILTER_CPU);
 
 169	__generic_field(char *, COMM, FILTER_COMM);
 170	__generic_field(char *, comm, FILTER_COMM);
 
 
 171
 172	return ret;
 173}
 174
 175static int trace_define_common_fields(void)
 176{
 177	int ret;
 178	struct trace_entry ent;
 179
 180	__common_field(unsigned short, type);
 181	__common_field(unsigned char, flags);
 
 182	__common_field(unsigned char, preempt_count);
 183	__common_field(int, pid);
 184
 185	return ret;
 186}
 187
 188static void trace_destroy_fields(struct trace_event_call *call)
 189{
 190	struct ftrace_event_field *field, *next;
 191	struct list_head *head;
 192
 193	head = trace_get_fields(call);
 194	list_for_each_entry_safe(field, next, head, link) {
 195		list_del(&field->link);
 196		kmem_cache_free(field_cachep, field);
 197	}
 198}
 199
 200/*
 201 * run-time version of trace_event_get_offsets_<call>() that returns the last
 202 * accessible offset of trace fields excluding __dynamic_array bytes
 203 */
 204int trace_event_get_offsets(struct trace_event_call *call)
 205{
 206	struct ftrace_event_field *tail;
 207	struct list_head *head;
 208
 209	head = trace_get_fields(call);
 210	/*
 211	 * head->next points to the last field with the largest offset,
 212	 * since it was added last by trace_define_field()
 213	 */
 214	tail = list_first_entry(head, struct ftrace_event_field, link);
 215	return tail->offset + tail->size;
 216}
 217
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 218int trace_event_raw_init(struct trace_event_call *call)
 219{
 220	int id;
 221
 222	id = register_trace_event(&call->event);
 223	if (!id)
 224		return -ENODEV;
 225
 
 
 226	return 0;
 227}
 228EXPORT_SYMBOL_GPL(trace_event_raw_init);
 229
 230bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 231{
 232	struct trace_array *tr = trace_file->tr;
 233	struct trace_array_cpu *data;
 
 234	struct trace_pid_list *pid_list;
 235
 236	pid_list = rcu_dereference_raw(tr->filtered_pids);
 237	if (!pid_list)
 
 
 238		return false;
 239
 240	data = this_cpu_ptr(tr->trace_buffer.data);
 241
 242	return data->ignore_pid;
 243}
 244EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 245
 246void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 247				 struct trace_event_file *trace_file,
 248				 unsigned long len)
 249{
 250	struct trace_event_call *event_call = trace_file->event_call;
 251
 252	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 253	    trace_event_ignore_this_pid(trace_file))
 254		return NULL;
 255
 256	local_save_flags(fbuffer->flags);
 257	fbuffer->pc = preempt_count();
 258	/*
 259	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 260	 * preemption (adding one to the preempt_count). Since we are
 261	 * interested in the preempt_count at the time the tracepoint was
 262	 * hit, we need to subtract one to offset the increment.
 263	 */
 264	if (IS_ENABLED(CONFIG_PREEMPTION))
 265		fbuffer->pc--;
 266	fbuffer->trace_file = trace_file;
 267
 268	fbuffer->event =
 269		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 270						event_call->event.type, len,
 271						fbuffer->flags, fbuffer->pc);
 272	if (!fbuffer->event)
 273		return NULL;
 274
 
 275	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 276	return fbuffer->entry;
 277}
 278EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 279
 280int trace_event_reg(struct trace_event_call *call,
 281		    enum trace_reg type, void *data)
 282{
 283	struct trace_event_file *file = data;
 284
 285	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 286	switch (type) {
 287	case TRACE_REG_REGISTER:
 288		return tracepoint_probe_register(call->tp,
 289						 call->class->probe,
 290						 file);
 291	case TRACE_REG_UNREGISTER:
 292		tracepoint_probe_unregister(call->tp,
 293					    call->class->probe,
 294					    file);
 295		return 0;
 296
 297#ifdef CONFIG_PERF_EVENTS
 298	case TRACE_REG_PERF_REGISTER:
 299		return tracepoint_probe_register(call->tp,
 300						 call->class->perf_probe,
 301						 call);
 302	case TRACE_REG_PERF_UNREGISTER:
 303		tracepoint_probe_unregister(call->tp,
 304					    call->class->perf_probe,
 305					    call);
 306		return 0;
 307	case TRACE_REG_PERF_OPEN:
 308	case TRACE_REG_PERF_CLOSE:
 309	case TRACE_REG_PERF_ADD:
 310	case TRACE_REG_PERF_DEL:
 311		return 0;
 312#endif
 313	}
 314	return 0;
 315}
 316EXPORT_SYMBOL_GPL(trace_event_reg);
 317
 318void trace_event_enable_cmd_record(bool enable)
 319{
 320	struct trace_event_file *file;
 321	struct trace_array *tr;
 322
 323	mutex_lock(&event_mutex);
 
 324	do_for_each_event_file(tr, file) {
 325
 326		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 327			continue;
 328
 329		if (enable) {
 330			tracing_start_cmdline_record();
 331			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 332		} else {
 333			tracing_stop_cmdline_record();
 334			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 335		}
 336	} while_for_each_event_file();
 337	mutex_unlock(&event_mutex);
 338}
 339
 340void trace_event_enable_tgid_record(bool enable)
 341{
 342	struct trace_event_file *file;
 343	struct trace_array *tr;
 344
 345	mutex_lock(&event_mutex);
 
 346	do_for_each_event_file(tr, file) {
 347		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 348			continue;
 349
 350		if (enable) {
 351			tracing_start_tgid_record();
 352			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 353		} else {
 354			tracing_stop_tgid_record();
 355			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 356				  &file->flags);
 357		}
 358	} while_for_each_event_file();
 359	mutex_unlock(&event_mutex);
 360}
 361
 362static int __ftrace_event_enable_disable(struct trace_event_file *file,
 363					 int enable, int soft_disable)
 364{
 365	struct trace_event_call *call = file->event_call;
 366	struct trace_array *tr = file->tr;
 367	unsigned long file_flags = file->flags;
 368	int ret = 0;
 369	int disable;
 370
 371	switch (enable) {
 372	case 0:
 373		/*
 374		 * When soft_disable is set and enable is cleared, the sm_ref
 375		 * reference counter is decremented. If it reaches 0, we want
 376		 * to clear the SOFT_DISABLED flag but leave the event in the
 377		 * state that it was. That is, if the event was enabled and
 378		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 379		 * is set we do not want the event to be enabled before we
 380		 * clear the bit.
 381		 *
 382		 * When soft_disable is not set but the SOFT_MODE flag is,
 383		 * we do nothing. Do not disable the tracepoint, otherwise
 384		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 385		 */
 386		if (soft_disable) {
 387			if (atomic_dec_return(&file->sm_ref) > 0)
 388				break;
 389			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 390			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 
 
 391		} else
 392			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 393
 394		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 395			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 396			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 397				tracing_stop_cmdline_record();
 398				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 399			}
 400
 401			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 402				tracing_stop_tgid_record();
 403				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 404			}
 405
 406			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 407		}
 408		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 409		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 410			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 411		else
 412			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 413		break;
 414	case 1:
 415		/*
 416		 * When soft_disable is set and enable is set, we want to
 417		 * register the tracepoint for the event, but leave the event
 418		 * as is. That means, if the event was already enabled, we do
 419		 * nothing (but set SOFT_MODE). If the event is disabled, we
 420		 * set SOFT_DISABLED before enabling the event tracepoint, so
 421		 * it still seems to be disabled.
 422		 */
 423		if (!soft_disable)
 424			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 425		else {
 426			if (atomic_inc_return(&file->sm_ref) > 1)
 427				break;
 428			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 
 
 429		}
 430
 431		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 432			bool cmd = false, tgid = false;
 433
 434			/* Keep the event disabled, when going to SOFT_MODE. */
 435			if (soft_disable)
 436				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 437
 438			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 439				cmd = true;
 440				tracing_start_cmdline_record();
 441				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 442			}
 443
 444			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 445				tgid = true;
 446				tracing_start_tgid_record();
 447				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 448			}
 449
 450			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 451			if (ret) {
 452				if (cmd)
 453					tracing_stop_cmdline_record();
 454				if (tgid)
 455					tracing_stop_tgid_record();
 456				pr_info("event trace: Could not enable event "
 457					"%s\n", trace_event_name(call));
 458				break;
 459			}
 460			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 461
 462			/* WAS_ENABLED gets set but never cleared. */
 463			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 464		}
 465		break;
 466	}
 467
 468	/* Enable or disable use of trace_buffered_event */
 469	if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 470	    (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 471		if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 472			trace_buffered_event_enable();
 473		else
 474			trace_buffered_event_disable();
 475	}
 476
 477	return ret;
 478}
 479
 480int trace_event_enable_disable(struct trace_event_file *file,
 481			       int enable, int soft_disable)
 482{
 483	return __ftrace_event_enable_disable(file, enable, soft_disable);
 484}
 485
 486static int ftrace_event_enable_disable(struct trace_event_file *file,
 487				       int enable)
 488{
 489	return __ftrace_event_enable_disable(file, enable, 0);
 490}
 491
 492static void ftrace_clear_events(struct trace_array *tr)
 493{
 494	struct trace_event_file *file;
 495
 496	mutex_lock(&event_mutex);
 497	list_for_each_entry(file, &tr->events, list) {
 498		ftrace_event_enable_disable(file, 0);
 499	}
 500	mutex_unlock(&event_mutex);
 501}
 502
 503static void
 504event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 505{
 506	struct trace_pid_list *pid_list;
 507	struct trace_array *tr = data;
 508
 509	pid_list = rcu_dereference_raw(tr->filtered_pids);
 510	trace_filter_add_remove_task(pid_list, NULL, task);
 
 
 
 511}
 512
 513static void
 514event_filter_pid_sched_process_fork(void *data,
 515				    struct task_struct *self,
 516				    struct task_struct *task)
 517{
 518	struct trace_pid_list *pid_list;
 519	struct trace_array *tr = data;
 520
 521	pid_list = rcu_dereference_sched(tr->filtered_pids);
 522	trace_filter_add_remove_task(pid_list, self, task);
 
 
 
 523}
 524
 525void trace_event_follow_fork(struct trace_array *tr, bool enable)
 526{
 527	if (enable) {
 528		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 529						       tr, INT_MIN);
 530		register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 531						       tr, INT_MAX);
 532	} else {
 533		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 534						    tr);
 535		unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 536						    tr);
 537	}
 538}
 539
 540static void
 541event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 542		    struct task_struct *prev, struct task_struct *next)
 
 
 543{
 544	struct trace_array *tr = data;
 
 545	struct trace_pid_list *pid_list;
 
 546
 547	pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 548
 549	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 550		       trace_ignore_this_task(pid_list, prev) &&
 551		       trace_ignore_this_task(pid_list, next));
 
 
 
 
 
 
 
 552}
 553
 554static void
 555event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 556		    struct task_struct *prev, struct task_struct *next)
 
 
 557{
 558	struct trace_array *tr = data;
 
 559	struct trace_pid_list *pid_list;
 560
 561	pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 562
 563	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 564		       trace_ignore_this_task(pid_list, next));
 565}
 566
 567static void
 568event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 569{
 570	struct trace_array *tr = data;
 
 571	struct trace_pid_list *pid_list;
 572
 573	/* Nothing to do if we are already tracing */
 574	if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 575		return;
 576
 577	pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 578
 579	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 580		       trace_ignore_this_task(pid_list, task));
 581}
 582
 583static void
 584event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 585{
 586	struct trace_array *tr = data;
 
 587	struct trace_pid_list *pid_list;
 588
 589	/* Nothing to do if we are not tracing */
 590	if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 591		return;
 592
 593	pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 594
 595	/* Set tracing if current is enabled */
 596	this_cpu_write(tr->trace_buffer.data->ignore_pid,
 597		       trace_ignore_this_task(pid_list, current));
 598}
 599
 600static void __ftrace_clear_event_pids(struct trace_array *tr)
 601{
 602	struct trace_pid_list *pid_list;
 603	struct trace_event_file *file;
 604	int cpu;
 605
 606	pid_list = rcu_dereference_protected(tr->filtered_pids,
 607					     lockdep_is_held(&event_mutex));
 608	if (!pid_list)
 609		return;
 610
 611	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 612	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 613
 614	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 615	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 616
 617	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 618	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 619
 620	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 621	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 
 622
 623	list_for_each_entry(file, &tr->events, list) {
 624		clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 625	}
 626
 627	for_each_possible_cpu(cpu)
 628		per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 629
 630	rcu_assign_pointer(tr->filtered_pids, NULL);
 
 631
 632	/* Wait till all users are no longer using pid filtering */
 633	tracepoint_synchronize_unregister();
 634
 635	trace_free_pid_list(pid_list);
 
 
 
 
 636}
 637
 638static void ftrace_clear_event_pids(struct trace_array *tr)
 639{
 640	mutex_lock(&event_mutex);
 641	__ftrace_clear_event_pids(tr);
 642	mutex_unlock(&event_mutex);
 643}
 644
 645static void __put_system(struct event_subsystem *system)
 646{
 647	struct event_filter *filter = system->filter;
 648
 649	WARN_ON_ONCE(system_refcount(system) == 0);
 650	if (system_refcount_dec(system))
 651		return;
 652
 653	list_del(&system->list);
 654
 655	if (filter) {
 656		kfree(filter->filter_string);
 657		kfree(filter);
 658	}
 659	kfree_const(system->name);
 660	kfree(system);
 661}
 662
 663static void __get_system(struct event_subsystem *system)
 664{
 665	WARN_ON_ONCE(system_refcount(system) == 0);
 666	system_refcount_inc(system);
 667}
 668
 669static void __get_system_dir(struct trace_subsystem_dir *dir)
 670{
 671	WARN_ON_ONCE(dir->ref_count == 0);
 672	dir->ref_count++;
 673	__get_system(dir->subsystem);
 674}
 675
 676static void __put_system_dir(struct trace_subsystem_dir *dir)
 677{
 678	WARN_ON_ONCE(dir->ref_count == 0);
 679	/* If the subsystem is about to be freed, the dir must be too */
 680	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 681
 682	__put_system(dir->subsystem);
 683	if (!--dir->ref_count)
 684		kfree(dir);
 685}
 686
 687static void put_system(struct trace_subsystem_dir *dir)
 688{
 689	mutex_lock(&event_mutex);
 690	__put_system_dir(dir);
 691	mutex_unlock(&event_mutex);
 692}
 693
 694static void remove_subsystem(struct trace_subsystem_dir *dir)
 695{
 696	if (!dir)
 697		return;
 698
 699	if (!--dir->nr_events) {
 700		tracefs_remove_recursive(dir->entry);
 701		list_del(&dir->list);
 702		__put_system_dir(dir);
 703	}
 704}
 705
 706static void remove_event_file_dir(struct trace_event_file *file)
 707{
 708	struct dentry *dir = file->dir;
 709	struct dentry *child;
 710
 711	if (dir) {
 712		spin_lock(&dir->d_lock);	/* probably unneeded */
 713		list_for_each_entry(child, &dir->d_subdirs, d_child) {
 714			if (d_really_is_positive(child))	/* probably unneeded */
 715				d_inode(child)->i_private = NULL;
 716		}
 717		spin_unlock(&dir->d_lock);
 718
 719		tracefs_remove_recursive(dir);
 
 
 
 
 720	}
 
 721
 
 
 
 722	list_del(&file->list);
 723	remove_subsystem(file->system);
 724	free_event_filter(file->filter);
 725	kmem_cache_free(file_cachep, file);
 
 726}
 727
 728/*
 729 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 730 */
 731static int
 732__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 733			      const char *sub, const char *event, int set)
 734{
 735	struct trace_event_file *file;
 736	struct trace_event_call *call;
 737	const char *name;
 738	int ret = -EINVAL;
 739	int eret = 0;
 740
 741	list_for_each_entry(file, &tr->events, list) {
 742
 743		call = file->event_call;
 744		name = trace_event_name(call);
 745
 746		if (!name || !call->class || !call->class->reg)
 747			continue;
 748
 749		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 750			continue;
 751
 752		if (match &&
 753		    strcmp(match, name) != 0 &&
 754		    strcmp(match, call->class->system) != 0)
 755			continue;
 756
 757		if (sub && strcmp(sub, call->class->system) != 0)
 758			continue;
 759
 760		if (event && strcmp(event, name) != 0)
 761			continue;
 762
 763		ret = ftrace_event_enable_disable(file, set);
 764
 765		/*
 766		 * Save the first error and return that. Some events
 767		 * may still have been enabled, but let the user
 768		 * know that something went wrong.
 769		 */
 770		if (ret && !eret)
 771			eret = ret;
 772
 773		ret = eret;
 774	}
 775
 776	return ret;
 777}
 778
 779static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 780				  const char *sub, const char *event, int set)
 781{
 782	int ret;
 783
 784	mutex_lock(&event_mutex);
 785	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 786	mutex_unlock(&event_mutex);
 787
 788	return ret;
 789}
 790
 791int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 792{
 793	char *event = NULL, *sub = NULL, *match;
 794	int ret;
 795
 
 
 796	/*
 797	 * The buf format can be <subsystem>:<event-name>
 798	 *  *:<event-name> means any event by that name.
 799	 *  :<event-name> is the same.
 800	 *
 801	 *  <subsystem>:* means all events in that subsystem
 802	 *  <subsystem>: means the same.
 803	 *
 804	 *  <name> (no ':') means all events in a subsystem with
 805	 *  the name <name> or any event that matches <name>
 806	 */
 807
 808	match = strsep(&buf, ":");
 809	if (buf) {
 810		sub = match;
 811		event = buf;
 812		match = NULL;
 813
 814		if (!strlen(sub) || strcmp(sub, "*") == 0)
 815			sub = NULL;
 816		if (!strlen(event) || strcmp(event, "*") == 0)
 817			event = NULL;
 818	}
 819
 820	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 821
 822	/* Put back the colon to allow this to be called again */
 823	if (buf)
 824		*(buf - 1) = ':';
 825
 826	return ret;
 827}
 828EXPORT_SYMBOL_GPL(ftrace_set_clr_event);
 829
 830/**
 831 * trace_set_clr_event - enable or disable an event
 832 * @system: system name to match (NULL for any system)
 833 * @event: event name to match (NULL for all events, within system)
 834 * @set: 1 to enable, 0 to disable
 835 *
 836 * This is a way for other parts of the kernel to enable or disable
 837 * event recording.
 838 *
 839 * Returns 0 on success, -EINVAL if the parameters do not match any
 840 * registered events.
 841 */
 842int trace_set_clr_event(const char *system, const char *event, int set)
 843{
 844	struct trace_array *tr = top_trace_array();
 845
 846	if (!tr)
 847		return -ENODEV;
 848
 849	return __ftrace_set_clr_event(tr, NULL, system, event, set);
 850}
 851EXPORT_SYMBOL_GPL(trace_set_clr_event);
 852
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 853/* 128 should be much more than enough */
 854#define EVENT_BUF_SIZE		127
 855
 856static ssize_t
 857ftrace_event_write(struct file *file, const char __user *ubuf,
 858		   size_t cnt, loff_t *ppos)
 859{
 860	struct trace_parser parser;
 861	struct seq_file *m = file->private_data;
 862	struct trace_array *tr = m->private;
 863	ssize_t read, ret;
 864
 865	if (!cnt)
 866		return 0;
 867
 868	ret = tracing_update_buffers();
 869	if (ret < 0)
 870		return ret;
 871
 872	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 873		return -ENOMEM;
 874
 875	read = trace_get_user(&parser, ubuf, cnt, ppos);
 876
 877	if (read >= 0 && trace_parser_loaded((&parser))) {
 878		int set = 1;
 879
 880		if (*parser.buffer == '!')
 881			set = 0;
 882
 883		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 884		if (ret)
 885			goto out_put;
 886	}
 887
 888	ret = read;
 889
 890 out_put:
 891	trace_parser_put(&parser);
 892
 893	return ret;
 894}
 895
 896static void *
 897t_next(struct seq_file *m, void *v, loff_t *pos)
 898{
 899	struct trace_event_file *file = v;
 900	struct trace_event_call *call;
 901	struct trace_array *tr = m->private;
 902
 903	(*pos)++;
 904
 905	list_for_each_entry_continue(file, &tr->events, list) {
 906		call = file->event_call;
 907		/*
 908		 * The ftrace subsystem is for showing formats only.
 909		 * They can not be enabled or disabled via the event files.
 910		 */
 911		if (call->class && call->class->reg &&
 912		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 913			return file;
 914	}
 915
 916	return NULL;
 917}
 918
 919static void *t_start(struct seq_file *m, loff_t *pos)
 920{
 921	struct trace_event_file *file;
 922	struct trace_array *tr = m->private;
 923	loff_t l;
 924
 925	mutex_lock(&event_mutex);
 926
 927	file = list_entry(&tr->events, struct trace_event_file, list);
 928	for (l = 0; l <= *pos; ) {
 929		file = t_next(m, file, &l);
 930		if (!file)
 931			break;
 932	}
 933	return file;
 934}
 935
 936static void *
 937s_next(struct seq_file *m, void *v, loff_t *pos)
 938{
 939	struct trace_event_file *file = v;
 940	struct trace_array *tr = m->private;
 941
 942	(*pos)++;
 943
 944	list_for_each_entry_continue(file, &tr->events, list) {
 945		if (file->flags & EVENT_FILE_FL_ENABLED)
 946			return file;
 947	}
 948
 949	return NULL;
 950}
 951
 952static void *s_start(struct seq_file *m, loff_t *pos)
 953{
 954	struct trace_event_file *file;
 955	struct trace_array *tr = m->private;
 956	loff_t l;
 957
 958	mutex_lock(&event_mutex);
 959
 960	file = list_entry(&tr->events, struct trace_event_file, list);
 961	for (l = 0; l <= *pos; ) {
 962		file = s_next(m, file, &l);
 963		if (!file)
 964			break;
 965	}
 966	return file;
 967}
 968
 969static int t_show(struct seq_file *m, void *v)
 970{
 971	struct trace_event_file *file = v;
 972	struct trace_event_call *call = file->event_call;
 973
 974	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 975		seq_printf(m, "%s:", call->class->system);
 976	seq_printf(m, "%s\n", trace_event_name(call));
 977
 978	return 0;
 979}
 980
 981static void t_stop(struct seq_file *m, void *p)
 982{
 983	mutex_unlock(&event_mutex);
 984}
 985
 986static void *
 987p_next(struct seq_file *m, void *v, loff_t *pos)
 988{
 989	struct trace_array *tr = m->private;
 990	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 
 
 
 
 991
 992	return trace_pid_next(pid_list, v, pos);
 993}
 994
 995static void *p_start(struct seq_file *m, loff_t *pos)
 
 
 
 
 
 
 
 
 
 
 
 
 996	__acquires(RCU)
 997{
 998	struct trace_pid_list *pid_list;
 999	struct trace_array *tr = m->private;
1000
1001	/*
1002	 * Grab the mutex, to keep calls to p_next() having the same
1003	 * tr->filtered_pids as p_start() has.
1004	 * If we just passed the tr->filtered_pids around, then RCU would
1005	 * have been enough, but doing that makes things more complex.
1006	 */
1007	mutex_lock(&event_mutex);
1008	rcu_read_lock_sched();
1009
1010	pid_list = rcu_dereference_sched(tr->filtered_pids);
 
 
 
1011
1012	if (!pid_list)
1013		return NULL;
1014
1015	return trace_pid_start(pid_list, pos);
1016}
1017
 
 
 
 
 
 
 
 
 
 
 
 
1018static void p_stop(struct seq_file *m, void *p)
1019	__releases(RCU)
1020{
1021	rcu_read_unlock_sched();
1022	mutex_unlock(&event_mutex);
1023}
1024
1025static ssize_t
1026event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1027		  loff_t *ppos)
1028{
1029	struct trace_event_file *file;
1030	unsigned long flags;
1031	char buf[4] = "0";
1032
1033	mutex_lock(&event_mutex);
1034	file = event_file_data(filp);
1035	if (likely(file))
1036		flags = file->flags;
1037	mutex_unlock(&event_mutex);
1038
1039	if (!file)
1040		return -ENODEV;
1041
1042	if (flags & EVENT_FILE_FL_ENABLED &&
1043	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1044		strcpy(buf, "1");
1045
1046	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1047	    flags & EVENT_FILE_FL_SOFT_MODE)
1048		strcat(buf, "*");
1049
1050	strcat(buf, "\n");
1051
1052	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1053}
1054
1055static ssize_t
1056event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1057		   loff_t *ppos)
1058{
1059	struct trace_event_file *file;
1060	unsigned long val;
1061	int ret;
1062
1063	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1064	if (ret)
1065		return ret;
1066
1067	ret = tracing_update_buffers();
1068	if (ret < 0)
1069		return ret;
1070
1071	switch (val) {
1072	case 0:
1073	case 1:
1074		ret = -ENODEV;
1075		mutex_lock(&event_mutex);
1076		file = event_file_data(filp);
1077		if (likely(file))
 
 
 
 
 
1078			ret = ftrace_event_enable_disable(file, val);
 
1079		mutex_unlock(&event_mutex);
1080		break;
1081
1082	default:
1083		return -EINVAL;
1084	}
1085
1086	*ppos += cnt;
1087
1088	return ret ? ret : cnt;
1089}
1090
1091static ssize_t
1092system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1093		   loff_t *ppos)
1094{
1095	const char set_to_char[4] = { '?', '0', '1', 'X' };
1096	struct trace_subsystem_dir *dir = filp->private_data;
1097	struct event_subsystem *system = dir->subsystem;
1098	struct trace_event_call *call;
1099	struct trace_event_file *file;
1100	struct trace_array *tr = dir->tr;
1101	char buf[2];
1102	int set = 0;
1103	int ret;
1104
1105	mutex_lock(&event_mutex);
1106	list_for_each_entry(file, &tr->events, list) {
1107		call = file->event_call;
1108		if (!trace_event_name(call) || !call->class || !call->class->reg)
 
1109			continue;
1110
1111		if (system && strcmp(call->class->system, system->name) != 0)
1112			continue;
1113
1114		/*
1115		 * We need to find out if all the events are set
1116		 * or if all events or cleared, or if we have
1117		 * a mixture.
1118		 */
1119		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1120
1121		/*
1122		 * If we have a mixture, no need to look further.
1123		 */
1124		if (set == 3)
1125			break;
1126	}
1127	mutex_unlock(&event_mutex);
1128
1129	buf[0] = set_to_char[set];
1130	buf[1] = '\n';
1131
1132	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1133
1134	return ret;
1135}
1136
1137static ssize_t
1138system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1139		    loff_t *ppos)
1140{
1141	struct trace_subsystem_dir *dir = filp->private_data;
1142	struct event_subsystem *system = dir->subsystem;
1143	const char *name = NULL;
1144	unsigned long val;
1145	ssize_t ret;
1146
1147	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1148	if (ret)
1149		return ret;
1150
1151	ret = tracing_update_buffers();
1152	if (ret < 0)
1153		return ret;
1154
1155	if (val != 0 && val != 1)
1156		return -EINVAL;
1157
1158	/*
1159	 * Opening of "enable" adds a ref count to system,
1160	 * so the name is safe to use.
1161	 */
1162	if (system)
1163		name = system->name;
1164
1165	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1166	if (ret)
1167		goto out;
1168
1169	ret = cnt;
1170
1171out:
1172	*ppos += cnt;
1173
1174	return ret;
1175}
1176
1177enum {
1178	FORMAT_HEADER		= 1,
1179	FORMAT_FIELD_SEPERATOR	= 2,
1180	FORMAT_PRINTFMT		= 3,
1181};
1182
1183static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1184{
1185	struct trace_event_call *call = event_file_data(m->private);
1186	struct list_head *common_head = &ftrace_common_fields;
1187	struct list_head *head = trace_get_fields(call);
1188	struct list_head *node = v;
1189
1190	(*pos)++;
1191
1192	switch ((unsigned long)v) {
1193	case FORMAT_HEADER:
1194		node = common_head;
1195		break;
1196
1197	case FORMAT_FIELD_SEPERATOR:
1198		node = head;
1199		break;
1200
1201	case FORMAT_PRINTFMT:
1202		/* all done */
1203		return NULL;
1204	}
1205
1206	node = node->prev;
1207	if (node == common_head)
1208		return (void *)FORMAT_FIELD_SEPERATOR;
1209	else if (node == head)
1210		return (void *)FORMAT_PRINTFMT;
1211	else
1212		return node;
1213}
1214
1215static int f_show(struct seq_file *m, void *v)
1216{
1217	struct trace_event_call *call = event_file_data(m->private);
1218	struct ftrace_event_field *field;
1219	const char *array_descriptor;
1220
1221	switch ((unsigned long)v) {
1222	case FORMAT_HEADER:
1223		seq_printf(m, "name: %s\n", trace_event_name(call));
1224		seq_printf(m, "ID: %d\n", call->event.type);
1225		seq_puts(m, "format:\n");
1226		return 0;
1227
1228	case FORMAT_FIELD_SEPERATOR:
1229		seq_putc(m, '\n');
1230		return 0;
1231
1232	case FORMAT_PRINTFMT:
1233		seq_printf(m, "\nprint fmt: %s\n",
1234			   call->print_fmt);
1235		return 0;
1236	}
1237
1238	field = list_entry(v, struct ftrace_event_field, link);
1239	/*
1240	 * Smartly shows the array type(except dynamic array).
1241	 * Normal:
1242	 *	field:TYPE VAR
1243	 * If TYPE := TYPE[LEN], it is shown:
1244	 *	field:TYPE VAR[LEN]
1245	 */
1246	array_descriptor = strchr(field->type, '[');
1247
1248	if (str_has_prefix(field->type, "__data_loc"))
1249		array_descriptor = NULL;
1250
1251	if (!array_descriptor)
1252		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1253			   field->type, field->name, field->offset,
1254			   field->size, !!field->is_signed);
1255	else
1256		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1257			   (int)(array_descriptor - field->type),
1258			   field->type, field->name,
1259			   array_descriptor, field->offset,
1260			   field->size, !!field->is_signed);
 
 
 
 
 
1261
1262	return 0;
1263}
1264
1265static void *f_start(struct seq_file *m, loff_t *pos)
1266{
1267	void *p = (void *)FORMAT_HEADER;
1268	loff_t l = 0;
1269
1270	/* ->stop() is called even if ->start() fails */
1271	mutex_lock(&event_mutex);
1272	if (!event_file_data(m->private))
1273		return ERR_PTR(-ENODEV);
1274
1275	while (l < *pos && p)
1276		p = f_next(m, p, &l);
1277
1278	return p;
1279}
1280
1281static void f_stop(struct seq_file *m, void *p)
1282{
1283	mutex_unlock(&event_mutex);
1284}
1285
1286static const struct seq_operations trace_format_seq_ops = {
1287	.start		= f_start,
1288	.next		= f_next,
1289	.stop		= f_stop,
1290	.show		= f_show,
1291};
1292
1293static int trace_format_open(struct inode *inode, struct file *file)
1294{
1295	struct seq_file *m;
1296	int ret;
1297
1298	/* Do we want to hide event format files on tracefs lockdown? */
1299
1300	ret = seq_open(file, &trace_format_seq_ops);
1301	if (ret < 0)
1302		return ret;
1303
1304	m = file->private_data;
1305	m->private = file;
1306
1307	return 0;
1308}
1309
1310static ssize_t
1311event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1312{
1313	int id = (long)event_file_data(filp);
1314	char buf[32];
1315	int len;
1316
1317	if (unlikely(!id))
1318		return -ENODEV;
1319
1320	len = sprintf(buf, "%d\n", id);
1321
1322	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1323}
1324
1325static ssize_t
1326event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1327		  loff_t *ppos)
1328{
1329	struct trace_event_file *file;
1330	struct trace_seq *s;
1331	int r = -ENODEV;
1332
1333	if (*ppos)
1334		return 0;
1335
1336	s = kmalloc(sizeof(*s), GFP_KERNEL);
1337
1338	if (!s)
1339		return -ENOMEM;
1340
1341	trace_seq_init(s);
1342
1343	mutex_lock(&event_mutex);
1344	file = event_file_data(filp);
1345	if (file)
1346		print_event_filter(file, s);
1347	mutex_unlock(&event_mutex);
1348
1349	if (file)
1350		r = simple_read_from_buffer(ubuf, cnt, ppos,
1351					    s->buffer, trace_seq_used(s));
1352
1353	kfree(s);
1354
1355	return r;
1356}
1357
1358static ssize_t
1359event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1360		   loff_t *ppos)
1361{
1362	struct trace_event_file *file;
1363	char *buf;
1364	int err = -ENODEV;
1365
1366	if (cnt >= PAGE_SIZE)
1367		return -EINVAL;
1368
1369	buf = memdup_user_nul(ubuf, cnt);
1370	if (IS_ERR(buf))
1371		return PTR_ERR(buf);
1372
1373	mutex_lock(&event_mutex);
1374	file = event_file_data(filp);
1375	if (file)
1376		err = apply_event_filter(file, buf);
1377	mutex_unlock(&event_mutex);
1378
1379	kfree(buf);
1380	if (err < 0)
1381		return err;
1382
1383	*ppos += cnt;
1384
1385	return cnt;
1386}
1387
1388static LIST_HEAD(event_subsystems);
1389
1390static int subsystem_open(struct inode *inode, struct file *filp)
1391{
 
 
1392	struct event_subsystem *system = NULL;
1393	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1394	struct trace_array *tr;
1395	int ret;
1396
1397	if (tracing_is_disabled())
1398		return -ENODEV;
1399
1400	/* Make sure the system still exists */
1401	mutex_lock(&event_mutex);
1402	mutex_lock(&trace_types_lock);
1403	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1404		list_for_each_entry(dir, &tr->systems, list) {
1405			if (dir == inode->i_private) {
1406				/* Don't open systems with no events */
 
 
1407				if (dir->nr_events) {
1408					__get_system_dir(dir);
1409					system = dir->subsystem;
1410				}
1411				goto exit_loop;
1412			}
1413		}
1414	}
1415 exit_loop:
1416	mutex_unlock(&trace_types_lock);
1417	mutex_unlock(&event_mutex);
1418
1419	if (!system)
1420		return -ENODEV;
1421
1422	/* Some versions of gcc think dir can be uninitialized here */
1423	WARN_ON(!dir);
1424
1425	/* Still need to increment the ref count of the system */
1426	if (trace_array_get(tr) < 0) {
1427		put_system(dir);
1428		return -ENODEV;
1429	}
1430
1431	ret = tracing_open_generic(inode, filp);
1432	if (ret < 0) {
1433		trace_array_put(tr);
1434		put_system(dir);
1435	}
1436
1437	return ret;
1438}
1439
1440static int system_tr_open(struct inode *inode, struct file *filp)
1441{
1442	struct trace_subsystem_dir *dir;
1443	struct trace_array *tr = inode->i_private;
1444	int ret;
1445
1446	/* Make a temporary dir that has no system but points to tr */
1447	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1448	if (!dir)
1449		return -ENOMEM;
1450
1451	ret = tracing_open_generic_tr(inode, filp);
1452	if (ret < 0) {
1453		kfree(dir);
1454		return ret;
1455	}
1456	dir->tr = tr;
1457	filp->private_data = dir;
1458
1459	return 0;
1460}
1461
1462static int subsystem_release(struct inode *inode, struct file *file)
1463{
1464	struct trace_subsystem_dir *dir = file->private_data;
1465
1466	trace_array_put(dir->tr);
1467
1468	/*
1469	 * If dir->subsystem is NULL, then this is a temporary
1470	 * descriptor that was made for a trace_array to enable
1471	 * all subsystems.
1472	 */
1473	if (dir->subsystem)
1474		put_system(dir);
1475	else
1476		kfree(dir);
1477
1478	return 0;
1479}
1480
1481static ssize_t
1482subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1483		      loff_t *ppos)
1484{
1485	struct trace_subsystem_dir *dir = filp->private_data;
1486	struct event_subsystem *system = dir->subsystem;
1487	struct trace_seq *s;
1488	int r;
1489
1490	if (*ppos)
1491		return 0;
1492
1493	s = kmalloc(sizeof(*s), GFP_KERNEL);
1494	if (!s)
1495		return -ENOMEM;
1496
1497	trace_seq_init(s);
1498
1499	print_subsystem_event_filter(system, s);
1500	r = simple_read_from_buffer(ubuf, cnt, ppos,
1501				    s->buffer, trace_seq_used(s));
1502
1503	kfree(s);
1504
1505	return r;
1506}
1507
1508static ssize_t
1509subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1510		       loff_t *ppos)
1511{
1512	struct trace_subsystem_dir *dir = filp->private_data;
1513	char *buf;
1514	int err;
1515
1516	if (cnt >= PAGE_SIZE)
1517		return -EINVAL;
1518
1519	buf = memdup_user_nul(ubuf, cnt);
1520	if (IS_ERR(buf))
1521		return PTR_ERR(buf);
1522
1523	err = apply_subsystem_event_filter(dir, buf);
1524	kfree(buf);
1525	if (err < 0)
1526		return err;
1527
1528	*ppos += cnt;
1529
1530	return cnt;
1531}
1532
1533static ssize_t
1534show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1535{
1536	int (*func)(struct trace_seq *s) = filp->private_data;
1537	struct trace_seq *s;
1538	int r;
1539
1540	if (*ppos)
1541		return 0;
1542
1543	s = kmalloc(sizeof(*s), GFP_KERNEL);
1544	if (!s)
1545		return -ENOMEM;
1546
1547	trace_seq_init(s);
1548
1549	func(s);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1550	r = simple_read_from_buffer(ubuf, cnt, ppos,
1551				    s->buffer, trace_seq_used(s));
1552
1553	kfree(s);
1554
1555	return r;
1556}
1557
1558static void ignore_task_cpu(void *data)
1559{
1560	struct trace_array *tr = data;
1561	struct trace_pid_list *pid_list;
 
1562
1563	/*
1564	 * This function is called by on_each_cpu() while the
1565	 * event_mutex is held.
1566	 */
1567	pid_list = rcu_dereference_protected(tr->filtered_pids,
1568					     mutex_is_locked(&event_mutex));
 
 
 
 
 
 
1569
1570	this_cpu_write(tr->trace_buffer.data->ignore_pid,
1571		       trace_ignore_this_task(pid_list, current));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1572}
1573
1574static ssize_t
1575ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1576		       size_t cnt, loff_t *ppos)
1577{
1578	struct seq_file *m = filp->private_data;
1579	struct trace_array *tr = m->private;
1580	struct trace_pid_list *filtered_pids = NULL;
 
1581	struct trace_pid_list *pid_list;
1582	struct trace_event_file *file;
1583	ssize_t ret;
1584
1585	if (!cnt)
1586		return 0;
1587
1588	ret = tracing_update_buffers();
1589	if (ret < 0)
1590		return ret;
1591
1592	mutex_lock(&event_mutex);
1593
1594	filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1595					     lockdep_is_held(&event_mutex));
 
 
 
 
 
 
 
 
 
1596
1597	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1598	if (ret < 0)
1599		goto out;
1600
1601	rcu_assign_pointer(tr->filtered_pids, pid_list);
 
 
 
1602
1603	list_for_each_entry(file, &tr->events, list) {
1604		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1605	}
1606
1607	if (filtered_pids) {
1608		tracepoint_synchronize_unregister();
1609		trace_free_pid_list(filtered_pids);
1610	} else if (pid_list) {
1611		/*
1612		 * Register a probe that is called before all other probes
1613		 * to set ignore_pid if next or prev do not match.
1614		 * Register a probe this is called after all other probes
1615		 * to only keep ignore_pid set if next pid matches.
1616		 */
1617		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1618						 tr, INT_MAX);
1619		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1620						 tr, 0);
1621
1622		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1623						 tr, INT_MAX);
1624		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1625						 tr, 0);
1626
1627		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1628						     tr, INT_MAX);
1629		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1630						     tr, 0);
1631
1632		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1633						 tr, INT_MAX);
1634		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1635						 tr, 0);
1636	}
1637
1638	/*
1639	 * Ignoring of pids is done at task switch. But we have to
1640	 * check for those tasks that are currently running.
1641	 * Always do this in case a pid was appended or removed.
1642	 */
1643	on_each_cpu(ignore_task_cpu, tr, 1);
1644
1645 out:
1646	mutex_unlock(&event_mutex);
1647
1648	if (ret > 0)
1649		*ppos += ret;
1650
1651	return ret;
1652}
1653
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1654static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1655static int ftrace_event_set_open(struct inode *inode, struct file *file);
1656static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
 
1657static int ftrace_event_release(struct inode *inode, struct file *file);
1658
1659static const struct seq_operations show_event_seq_ops = {
1660	.start = t_start,
1661	.next = t_next,
1662	.show = t_show,
1663	.stop = t_stop,
1664};
1665
1666static const struct seq_operations show_set_event_seq_ops = {
1667	.start = s_start,
1668	.next = s_next,
1669	.show = t_show,
1670	.stop = t_stop,
1671};
1672
1673static const struct seq_operations show_set_pid_seq_ops = {
1674	.start = p_start,
1675	.next = p_next,
1676	.show = trace_pid_show,
1677	.stop = p_stop,
1678};
1679
 
 
 
 
 
 
 
1680static const struct file_operations ftrace_avail_fops = {
1681	.open = ftrace_event_avail_open,
1682	.read = seq_read,
1683	.llseek = seq_lseek,
1684	.release = seq_release,
1685};
1686
1687static const struct file_operations ftrace_set_event_fops = {
1688	.open = ftrace_event_set_open,
1689	.read = seq_read,
1690	.write = ftrace_event_write,
1691	.llseek = seq_lseek,
1692	.release = ftrace_event_release,
1693};
1694
1695static const struct file_operations ftrace_set_event_pid_fops = {
1696	.open = ftrace_event_set_pid_open,
1697	.read = seq_read,
1698	.write = ftrace_event_pid_write,
1699	.llseek = seq_lseek,
1700	.release = ftrace_event_release,
1701};
1702
 
 
 
 
 
 
 
 
1703static const struct file_operations ftrace_enable_fops = {
1704	.open = tracing_open_generic,
1705	.read = event_enable_read,
1706	.write = event_enable_write,
 
1707	.llseek = default_llseek,
1708};
1709
1710static const struct file_operations ftrace_event_format_fops = {
1711	.open = trace_format_open,
1712	.read = seq_read,
1713	.llseek = seq_lseek,
1714	.release = seq_release,
1715};
1716
1717static const struct file_operations ftrace_event_id_fops = {
1718	.read = event_id_read,
1719	.llseek = default_llseek,
1720};
1721
1722static const struct file_operations ftrace_event_filter_fops = {
1723	.open = tracing_open_generic,
1724	.read = event_filter_read,
1725	.write = event_filter_write,
 
1726	.llseek = default_llseek,
1727};
1728
1729static const struct file_operations ftrace_subsystem_filter_fops = {
1730	.open = subsystem_open,
1731	.read = subsystem_filter_read,
1732	.write = subsystem_filter_write,
1733	.llseek = default_llseek,
1734	.release = subsystem_release,
1735};
1736
1737static const struct file_operations ftrace_system_enable_fops = {
1738	.open = subsystem_open,
1739	.read = system_enable_read,
1740	.write = system_enable_write,
1741	.llseek = default_llseek,
1742	.release = subsystem_release,
1743};
1744
1745static const struct file_operations ftrace_tr_enable_fops = {
1746	.open = system_tr_open,
1747	.read = system_enable_read,
1748	.write = system_enable_write,
1749	.llseek = default_llseek,
1750	.release = subsystem_release,
1751};
1752
1753static const struct file_operations ftrace_show_header_fops = {
1754	.open = tracing_open_generic,
1755	.read = show_header,
1756	.llseek = default_llseek,
 
 
 
 
 
 
 
 
1757};
1758
1759static int
1760ftrace_event_open(struct inode *inode, struct file *file,
1761		  const struct seq_operations *seq_ops)
1762{
1763	struct seq_file *m;
1764	int ret;
1765
1766	ret = security_locked_down(LOCKDOWN_TRACEFS);
1767	if (ret)
1768		return ret;
1769
1770	ret = seq_open(file, seq_ops);
1771	if (ret < 0)
1772		return ret;
1773	m = file->private_data;
1774	/* copy tr over to seq ops */
1775	m->private = inode->i_private;
1776
1777	return ret;
1778}
1779
1780static int ftrace_event_release(struct inode *inode, struct file *file)
1781{
1782	struct trace_array *tr = inode->i_private;
1783
1784	trace_array_put(tr);
1785
1786	return seq_release(inode, file);
1787}
1788
1789static int
1790ftrace_event_avail_open(struct inode *inode, struct file *file)
1791{
1792	const struct seq_operations *seq_ops = &show_event_seq_ops;
1793
1794	/* Checks for tracefs lockdown */
1795	return ftrace_event_open(inode, file, seq_ops);
1796}
1797
1798static int
1799ftrace_event_set_open(struct inode *inode, struct file *file)
1800{
1801	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1802	struct trace_array *tr = inode->i_private;
1803	int ret;
1804
1805	ret = tracing_check_open_get_tr(tr);
1806	if (ret)
1807		return ret;
1808
1809	if ((file->f_mode & FMODE_WRITE) &&
1810	    (file->f_flags & O_TRUNC))
1811		ftrace_clear_events(tr);
1812
1813	ret = ftrace_event_open(inode, file, seq_ops);
1814	if (ret < 0)
1815		trace_array_put(tr);
1816	return ret;
1817}
1818
1819static int
1820ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1821{
1822	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1823	struct trace_array *tr = inode->i_private;
1824	int ret;
1825
1826	ret = tracing_check_open_get_tr(tr);
1827	if (ret)
1828		return ret;
1829
1830	if ((file->f_mode & FMODE_WRITE) &&
1831	    (file->f_flags & O_TRUNC))
1832		ftrace_clear_event_pids(tr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1833
1834	ret = ftrace_event_open(inode, file, seq_ops);
1835	if (ret < 0)
1836		trace_array_put(tr);
1837	return ret;
1838}
1839
1840static struct event_subsystem *
1841create_new_subsystem(const char *name)
1842{
1843	struct event_subsystem *system;
1844
1845	/* need to create new entry */
1846	system = kmalloc(sizeof(*system), GFP_KERNEL);
1847	if (!system)
1848		return NULL;
1849
1850	system->ref_count = 1;
1851
1852	/* Only allocate if dynamic (kprobes and modules) */
1853	system->name = kstrdup_const(name, GFP_KERNEL);
1854	if (!system->name)
1855		goto out_free;
1856
1857	system->filter = NULL;
1858
1859	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1860	if (!system->filter)
1861		goto out_free;
1862
1863	list_add(&system->list, &event_subsystems);
1864
1865	return system;
1866
1867 out_free:
1868	kfree_const(system->name);
1869	kfree(system);
1870	return NULL;
1871}
1872
1873static struct dentry *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1874event_subsystem_dir(struct trace_array *tr, const char *name,
1875		    struct trace_event_file *file, struct dentry *parent)
1876{
 
1877	struct trace_subsystem_dir *dir;
1878	struct event_subsystem *system;
1879	struct dentry *entry;
 
 
 
 
 
 
 
 
 
 
1880
1881	/* First see if we did not already create this dir */
1882	list_for_each_entry(dir, &tr->systems, list) {
1883		system = dir->subsystem;
1884		if (strcmp(system->name, name) == 0) {
1885			dir->nr_events++;
1886			file->system = dir;
1887			return dir->entry;
1888		}
1889	}
1890
1891	/* Now see if the system itself exists. */
1892	list_for_each_entry(system, &event_subsystems, list) {
1893		if (strcmp(system->name, name) == 0)
 
 
1894			break;
 
1895	}
1896	/* Reset system variable when not found */
1897	if (&system->list == &event_subsystems)
1898		system = NULL;
1899
1900	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1901	if (!dir)
1902		goto out_fail;
1903
1904	if (!system) {
1905		system = create_new_subsystem(name);
1906		if (!system)
1907			goto out_free;
1908	} else
1909		__get_system(system);
1910
1911	dir->entry = tracefs_create_dir(name, parent);
1912	if (!dir->entry) {
 
 
 
 
 
 
1913		pr_warn("Failed to create system directory %s\n", name);
1914		__put_system(system);
1915		goto out_free;
1916	}
1917
 
1918	dir->tr = tr;
1919	dir->ref_count = 1;
1920	dir->nr_events = 1;
1921	dir->subsystem = system;
1922	file->system = dir;
1923
1924	entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1925				    &ftrace_subsystem_filter_fops);
1926	if (!entry) {
1927		kfree(system->filter);
1928		system->filter = NULL;
1929		pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1930	}
1931
1932	trace_create_file("enable", 0644, dir->entry, dir,
1933			  &ftrace_system_enable_fops);
1934
1935	list_add(&dir->list, &tr->systems);
1936
1937	return dir->entry;
1938
1939 out_free:
1940	kfree(dir);
1941 out_fail:
1942	/* Only print this message if failed on memory allocation */
1943	if (!dir || !system)
1944		pr_warn("No memory to create event subsystem %s\n", name);
1945	return NULL;
1946}
1947
1948static int
1949event_create_dir(struct dentry *parent, struct trace_event_file *file)
1950{
1951	struct trace_event_call *call = file->event_call;
1952	struct trace_array *tr = file->tr;
1953	struct list_head *head;
1954	struct dentry *d_events;
1955	const char *name;
1956	int ret;
1957
1958	/*
1959	 * If the trace point header did not define TRACE_SYSTEM
1960	 * then the system would be called "TRACE_SYSTEM".
1961	 */
1962	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1963		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1964		if (!d_events)
1965			return -ENOMEM;
1966	} else
1967		d_events = parent;
1968
1969	name = trace_event_name(call);
1970	file->dir = tracefs_create_dir(name, d_events);
1971	if (!file->dir) {
1972		pr_warn("Could not create tracefs '%s' directory\n", name);
1973		return -1;
1974	}
1975
1976	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1977		trace_create_file("enable", 0644, file->dir, file,
1978				  &ftrace_enable_fops);
1979
1980#ifdef CONFIG_PERF_EVENTS
1981	if (call->event.type && call->class->reg)
1982		trace_create_file("id", 0444, file->dir,
1983				  (void *)(long)call->event.type,
1984				  &ftrace_event_id_fops);
1985#endif
1986
1987	/*
1988	 * Other events may have the same class. Only update
1989	 * the fields if they are not already defined.
1990	 */
1991	head = trace_get_fields(call);
1992	if (list_empty(head)) {
1993		ret = call->class->define_fields(call);
1994		if (ret < 0) {
1995			pr_warn("Could not initialize trace point events/%s\n",
1996				name);
1997			return -1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1998		}
1999	}
2000
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2001	/*
2002	 * Only event directories that can be enabled should have
2003	 * triggers or filters.
 
2004	 */
2005	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2006		trace_create_file("filter", 0644, file->dir, file,
2007				  &ftrace_event_filter_fops);
 
 
 
2008
2009		trace_create_file("trigger", 0644, file->dir, file,
2010				  &event_trigger_fops);
 
 
 
2011	}
2012
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2013#ifdef CONFIG_HIST_TRIGGERS
2014	trace_create_file("hist", 0444, file->dir, file,
2015			  &event_hist_fops);
 
 
 
2016#endif
2017	trace_create_file("format", 0444, file->dir, call,
2018			  &ftrace_event_format_fops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2019
2020	return 0;
2021}
2022
2023static void remove_event_from_tracers(struct trace_event_call *call)
2024{
2025	struct trace_event_file *file;
2026	struct trace_array *tr;
2027
2028	do_for_each_event_file_safe(tr, file) {
2029		if (file->event_call != call)
2030			continue;
2031
2032		remove_event_file_dir(file);
2033		/*
2034		 * The do_for_each_event_file_safe() is
2035		 * a double loop. After finding the call for this
2036		 * trace_array, we use break to jump to the next
2037		 * trace_array.
2038		 */
2039		break;
2040	} while_for_each_event_file();
2041}
2042
2043static void event_remove(struct trace_event_call *call)
2044{
2045	struct trace_array *tr;
2046	struct trace_event_file *file;
2047
2048	do_for_each_event_file(tr, file) {
2049		if (file->event_call != call)
2050			continue;
2051
2052		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2053			tr->clear_trace = true;
2054
2055		ftrace_event_enable_disable(file, 0);
2056		/*
2057		 * The do_for_each_event_file() is
2058		 * a double loop. After finding the call for this
2059		 * trace_array, we use break to jump to the next
2060		 * trace_array.
2061		 */
2062		break;
2063	} while_for_each_event_file();
2064
2065	if (call->event.funcs)
2066		__unregister_trace_event(&call->event);
2067	remove_event_from_tracers(call);
2068	list_del(&call->list);
2069}
2070
2071static int event_init(struct trace_event_call *call)
2072{
2073	int ret = 0;
2074	const char *name;
2075
2076	name = trace_event_name(call);
2077	if (WARN_ON(!name))
2078		return -EINVAL;
2079
2080	if (call->class->raw_init) {
2081		ret = call->class->raw_init(call);
2082		if (ret < 0 && ret != -ENOSYS)
2083			pr_warn("Could not initialize trace events/%s\n", name);
2084	}
2085
2086	return ret;
2087}
2088
2089static int
2090__register_event(struct trace_event_call *call, struct module *mod)
2091{
2092	int ret;
2093
2094	ret = event_init(call);
2095	if (ret < 0)
2096		return ret;
2097
2098	list_add(&call->list, &ftrace_events);
2099	call->mod = mod;
 
 
 
2100
2101	return 0;
2102}
2103
2104static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2105{
2106	int rlen;
2107	int elen;
2108
2109	/* Find the length of the eval value as a string */
2110	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2111	/* Make sure there's enough room to replace the string with the value */
2112	if (len < elen)
2113		return NULL;
2114
2115	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2116
2117	/* Get the rest of the string of ptr */
2118	rlen = strlen(ptr + len);
2119	memmove(ptr + elen, ptr + len, rlen);
2120	/* Make sure we end the new string */
2121	ptr[elen + rlen] = 0;
2122
2123	return ptr + elen;
2124}
2125
2126static void update_event_printk(struct trace_event_call *call,
2127				struct trace_eval_map *map)
2128{
2129	char *ptr;
2130	int quote = 0;
2131	int len = strlen(map->eval_string);
2132
2133	for (ptr = call->print_fmt; *ptr; ptr++) {
2134		if (*ptr == '\\') {
2135			ptr++;
2136			/* paranoid */
2137			if (!*ptr)
2138				break;
2139			continue;
2140		}
2141		if (*ptr == '"') {
2142			quote ^= 1;
2143			continue;
2144		}
2145		if (quote)
2146			continue;
2147		if (isdigit(*ptr)) {
2148			/* skip numbers */
2149			do {
2150				ptr++;
2151				/* Check for alpha chars like ULL */
2152			} while (isalnum(*ptr));
2153			if (!*ptr)
2154				break;
2155			/*
2156			 * A number must have some kind of delimiter after
2157			 * it, and we can ignore that too.
2158			 */
2159			continue;
2160		}
2161		if (isalpha(*ptr) || *ptr == '_') {
2162			if (strncmp(map->eval_string, ptr, len) == 0 &&
2163			    !isalnum(ptr[len]) && ptr[len] != '_') {
2164				ptr = eval_replace(ptr, map, len);
2165				/* enum/sizeof string smaller than value */
2166				if (WARN_ON_ONCE(!ptr))
2167					return;
2168				/*
2169				 * No need to decrement here, as eval_replace()
2170				 * returns the pointer to the character passed
2171				 * the eval, and two evals can not be placed
2172				 * back to back without something in between.
2173				 * We can skip that something in between.
2174				 */
2175				continue;
2176			}
2177		skip_more:
2178			do {
2179				ptr++;
2180			} while (isalnum(*ptr) || *ptr == '_');
2181			if (!*ptr)
2182				break;
2183			/*
2184			 * If what comes after this variable is a '.' or
2185			 * '->' then we can continue to ignore that string.
2186			 */
2187			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2188				ptr += *ptr == '.' ? 1 : 2;
2189				if (!*ptr)
2190					break;
2191				goto skip_more;
2192			}
2193			/*
2194			 * Once again, we can skip the delimiter that came
2195			 * after the string.
2196			 */
2197			continue;
2198		}
2199	}
2200}
2201
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2202void trace_event_eval_update(struct trace_eval_map **map, int len)
2203{
2204	struct trace_event_call *call, *p;
2205	const char *last_system = NULL;
2206	bool first = false;
2207	int last_i;
2208	int i;
2209
2210	down_write(&trace_event_sem);
2211	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2212		/* events are usually grouped together with systems */
2213		if (!last_system || call->class->system != last_system) {
2214			first = true;
2215			last_i = 0;
2216			last_system = call->class->system;
2217		}
2218
2219		/*
2220		 * Since calls are grouped by systems, the likelyhood that the
2221		 * next call in the iteration belongs to the same system as the
2222		 * previous call is high. As an optimization, we skip seaching
2223		 * for a map[] that matches the call's system if the last call
2224		 * was from the same system. That's what last_i is for. If the
2225		 * call has the same system as the previous call, then last_i
2226		 * will be the index of the first map[] that has a matching
2227		 * system.
2228		 */
2229		for (i = last_i; i < len; i++) {
2230			if (call->class->system == map[i]->system) {
2231				/* Save the first system if need be */
2232				if (first) {
2233					last_i = i;
2234					first = false;
2235				}
2236				update_event_printk(call, map[i]);
 
2237			}
2238		}
 
2239	}
2240	up_write(&trace_event_sem);
2241}
2242
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2243static struct trace_event_file *
2244trace_create_new_event(struct trace_event_call *call,
2245		       struct trace_array *tr)
2246{
 
 
2247	struct trace_event_file *file;
 
 
 
 
2248
2249	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2250	if (!file)
2251		return NULL;
 
 
 
 
 
 
 
 
 
2252
2253	file->event_call = call;
2254	file->tr = tr;
2255	atomic_set(&file->sm_ref, 0);
2256	atomic_set(&file->tm_ref, 0);
2257	INIT_LIST_HEAD(&file->triggers);
2258	list_add(&file->list, &tr->events);
 
2259
2260	return file;
2261}
2262
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2263/* Add an event to a trace directory */
2264static int
2265__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2266{
2267	struct trace_event_file *file;
2268
2269	file = trace_create_new_event(call, tr);
 
 
 
 
 
 
2270	if (!file)
2271		return -ENOMEM;
2272
2273	return event_create_dir(tr->event_dir, file);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2274}
2275
2276/*
2277 * Just create a decriptor for early init. A descriptor is required
2278 * for enabling events at boot. We want to enable events before
2279 * the filesystem is initialized.
2280 */
2281static __init int
2282__trace_early_add_new_event(struct trace_event_call *call,
2283			    struct trace_array *tr)
2284{
2285	struct trace_event_file *file;
 
2286
2287	file = trace_create_new_event(call, tr);
 
 
 
 
 
 
2288	if (!file)
2289		return -ENOMEM;
 
 
 
 
 
 
 
 
 
2290
2291	return 0;
2292}
2293
2294struct ftrace_module_file_ops;
2295static void __add_event_to_tracers(struct trace_event_call *call);
2296
2297/* Add an additional event_call dynamically */
2298int trace_add_event_call(struct trace_event_call *call)
2299{
2300	int ret;
2301	lockdep_assert_held(&event_mutex);
2302
2303	mutex_lock(&trace_types_lock);
2304
2305	ret = __register_event(call, NULL);
2306	if (ret >= 0)
2307		__add_event_to_tracers(call);
2308
2309	mutex_unlock(&trace_types_lock);
2310	return ret;
2311}
 
2312
2313/*
2314 * Must be called under locking of trace_types_lock, event_mutex and
2315 * trace_event_sem.
2316 */
2317static void __trace_remove_event_call(struct trace_event_call *call)
2318{
2319	event_remove(call);
2320	trace_destroy_fields(call);
2321	free_event_filter(call->filter);
2322	call->filter = NULL;
2323}
2324
2325static int probe_remove_event_call(struct trace_event_call *call)
2326{
2327	struct trace_array *tr;
2328	struct trace_event_file *file;
2329
2330#ifdef CONFIG_PERF_EVENTS
2331	if (call->perf_refcount)
2332		return -EBUSY;
2333#endif
2334	do_for_each_event_file(tr, file) {
2335		if (file->event_call != call)
2336			continue;
2337		/*
2338		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2339		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2340		 * TRACE_REG_UNREGISTER.
2341		 */
2342		if (file->flags & EVENT_FILE_FL_ENABLED)
2343			return -EBUSY;
 
 
 
2344		/*
2345		 * The do_for_each_event_file_safe() is
2346		 * a double loop. After finding the call for this
2347		 * trace_array, we use break to jump to the next
2348		 * trace_array.
2349		 */
2350		break;
2351	} while_for_each_event_file();
2352
2353	__trace_remove_event_call(call);
2354
2355	return 0;
 
 
 
 
 
 
2356}
2357
2358/* Remove an event_call */
2359int trace_remove_event_call(struct trace_event_call *call)
2360{
2361	int ret;
2362
2363	lockdep_assert_held(&event_mutex);
2364
2365	mutex_lock(&trace_types_lock);
2366	down_write(&trace_event_sem);
2367	ret = probe_remove_event_call(call);
2368	up_write(&trace_event_sem);
2369	mutex_unlock(&trace_types_lock);
2370
2371	return ret;
2372}
 
2373
2374#define for_each_event(event, start, end)			\
2375	for (event = start;					\
2376	     (unsigned long)event < (unsigned long)end;		\
2377	     event++)
2378
2379#ifdef CONFIG_MODULES
2380
2381static void trace_module_add_events(struct module *mod)
2382{
2383	struct trace_event_call **call, **start, **end;
2384
2385	if (!mod->num_trace_events)
2386		return;
2387
2388	/* Don't add infrastructure for mods without tracepoints */
2389	if (trace_module_has_bad_taint(mod)) {
2390		pr_err("%s: module has bad taint, not creating trace events\n",
2391		       mod->name);
2392		return;
2393	}
2394
2395	start = mod->trace_events;
2396	end = mod->trace_events + mod->num_trace_events;
2397
2398	for_each_event(call, start, end) {
2399		__register_event(*call, mod);
2400		__add_event_to_tracers(*call);
2401	}
2402}
2403
2404static void trace_module_remove_events(struct module *mod)
2405{
2406	struct trace_event_call *call, *p;
 
2407
2408	down_write(&trace_event_sem);
2409	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2410		if (call->mod == mod)
 
 
2411			__trace_remove_event_call(call);
2412	}
 
 
 
 
 
 
 
 
2413	up_write(&trace_event_sem);
2414
2415	/*
2416	 * It is safest to reset the ring buffer if the module being unloaded
2417	 * registered any events that were used. The only worry is if
2418	 * a new module gets loaded, and takes on the same id as the events
2419	 * of this module. When printing out the buffer, traced events left
2420	 * over from this module may be passed to the new module events and
2421	 * unexpected results may occur.
2422	 */
2423	tracing_reset_all_online_cpus();
2424}
2425
2426static int trace_module_notify(struct notifier_block *self,
2427			       unsigned long val, void *data)
2428{
2429	struct module *mod = data;
2430
2431	mutex_lock(&event_mutex);
2432	mutex_lock(&trace_types_lock);
2433	switch (val) {
2434	case MODULE_STATE_COMING:
2435		trace_module_add_events(mod);
2436		break;
2437	case MODULE_STATE_GOING:
2438		trace_module_remove_events(mod);
2439		break;
2440	}
2441	mutex_unlock(&trace_types_lock);
2442	mutex_unlock(&event_mutex);
2443
2444	return 0;
2445}
2446
2447static struct notifier_block trace_module_nb = {
2448	.notifier_call = trace_module_notify,
2449	.priority = 1, /* higher than trace.c module notify */
2450};
2451#endif /* CONFIG_MODULES */
2452
2453/* Create a new event directory structure for a trace directory. */
2454static void
2455__trace_add_event_dirs(struct trace_array *tr)
2456{
2457	struct trace_event_call *call;
2458	int ret;
2459
2460	list_for_each_entry(call, &ftrace_events, list) {
2461		ret = __trace_add_new_event(call, tr);
2462		if (ret < 0)
2463			pr_warn("Could not create directory for event %s\n",
2464				trace_event_name(call));
2465	}
2466}
2467
2468/* Returns any file that matches the system and event */
2469struct trace_event_file *
2470__find_event_file(struct trace_array *tr, const char *system, const char *event)
2471{
2472	struct trace_event_file *file;
2473	struct trace_event_call *call;
2474	const char *name;
2475
2476	list_for_each_entry(file, &tr->events, list) {
2477
2478		call = file->event_call;
2479		name = trace_event_name(call);
2480
2481		if (!name || !call->class)
2482			continue;
2483
2484		if (strcmp(event, name) == 0 &&
2485		    strcmp(system, call->class->system) == 0)
2486			return file;
2487	}
2488	return NULL;
2489}
2490
2491/* Returns valid trace event files that match system and event */
2492struct trace_event_file *
2493find_event_file(struct trace_array *tr, const char *system, const char *event)
2494{
2495	struct trace_event_file *file;
2496
2497	file = __find_event_file(tr, system, event);
2498	if (!file || !file->event_call->class->reg ||
2499	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2500		return NULL;
2501
2502	return file;
2503}
2504
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2505#ifdef CONFIG_DYNAMIC_FTRACE
2506
2507/* Avoid typos */
2508#define ENABLE_EVENT_STR	"enable_event"
2509#define DISABLE_EVENT_STR	"disable_event"
2510
2511struct event_probe_data {
2512	struct trace_event_file	*file;
2513	unsigned long			count;
2514	int				ref;
2515	bool				enable;
2516};
2517
2518static void update_event_probe(struct event_probe_data *data)
2519{
2520	if (data->enable)
2521		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2522	else
2523		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2524}
2525
2526static void
2527event_enable_probe(unsigned long ip, unsigned long parent_ip,
2528		   struct trace_array *tr, struct ftrace_probe_ops *ops,
2529		   void *data)
2530{
2531	struct ftrace_func_mapper *mapper = data;
2532	struct event_probe_data *edata;
2533	void **pdata;
2534
2535	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2536	if (!pdata || !*pdata)
2537		return;
2538
2539	edata = *pdata;
2540	update_event_probe(edata);
2541}
2542
2543static void
2544event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2545			 struct trace_array *tr, struct ftrace_probe_ops *ops,
2546			 void *data)
2547{
2548	struct ftrace_func_mapper *mapper = data;
2549	struct event_probe_data *edata;
2550	void **pdata;
2551
2552	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2553	if (!pdata || !*pdata)
2554		return;
2555
2556	edata = *pdata;
2557
2558	if (!edata->count)
2559		return;
2560
2561	/* Skip if the event is in a state we want to switch to */
2562	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2563		return;
2564
2565	if (edata->count != -1)
2566		(edata->count)--;
2567
2568	update_event_probe(edata);
2569}
2570
2571static int
2572event_enable_print(struct seq_file *m, unsigned long ip,
2573		   struct ftrace_probe_ops *ops, void *data)
2574{
2575	struct ftrace_func_mapper *mapper = data;
2576	struct event_probe_data *edata;
2577	void **pdata;
2578
2579	pdata = ftrace_func_mapper_find_ip(mapper, ip);
2580
2581	if (WARN_ON_ONCE(!pdata || !*pdata))
2582		return 0;
2583
2584	edata = *pdata;
2585
2586	seq_printf(m, "%ps:", (void *)ip);
2587
2588	seq_printf(m, "%s:%s:%s",
2589		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2590		   edata->file->event_call->class->system,
2591		   trace_event_name(edata->file->event_call));
2592
2593	if (edata->count == -1)
2594		seq_puts(m, ":unlimited\n");
2595	else
2596		seq_printf(m, ":count=%ld\n", edata->count);
2597
2598	return 0;
2599}
2600
2601static int
2602event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2603		  unsigned long ip, void *init_data, void **data)
2604{
2605	struct ftrace_func_mapper *mapper = *data;
2606	struct event_probe_data *edata = init_data;
2607	int ret;
2608
2609	if (!mapper) {
2610		mapper = allocate_ftrace_func_mapper();
2611		if (!mapper)
2612			return -ENODEV;
2613		*data = mapper;
2614	}
2615
2616	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2617	if (ret < 0)
2618		return ret;
2619
2620	edata->ref++;
2621
2622	return 0;
2623}
2624
2625static int free_probe_data(void *data)
2626{
2627	struct event_probe_data *edata = data;
2628
2629	edata->ref--;
2630	if (!edata->ref) {
2631		/* Remove the SOFT_MODE flag */
2632		__ftrace_event_enable_disable(edata->file, 0, 1);
2633		module_put(edata->file->event_call->mod);
2634		kfree(edata);
2635	}
2636	return 0;
2637}
2638
2639static void
2640event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2641		  unsigned long ip, void *data)
2642{
2643	struct ftrace_func_mapper *mapper = data;
2644	struct event_probe_data *edata;
2645
2646	if (!ip) {
2647		if (!mapper)
2648			return;
2649		free_ftrace_func_mapper(mapper, free_probe_data);
2650		return;
2651	}
2652
2653	edata = ftrace_func_mapper_remove_ip(mapper, ip);
2654
2655	if (WARN_ON_ONCE(!edata))
2656		return;
2657
2658	if (WARN_ON_ONCE(edata->ref <= 0))
2659		return;
2660
2661	free_probe_data(edata);
2662}
2663
2664static struct ftrace_probe_ops event_enable_probe_ops = {
2665	.func			= event_enable_probe,
2666	.print			= event_enable_print,
2667	.init			= event_enable_init,
2668	.free			= event_enable_free,
2669};
2670
2671static struct ftrace_probe_ops event_enable_count_probe_ops = {
2672	.func			= event_enable_count_probe,
2673	.print			= event_enable_print,
2674	.init			= event_enable_init,
2675	.free			= event_enable_free,
2676};
2677
2678static struct ftrace_probe_ops event_disable_probe_ops = {
2679	.func			= event_enable_probe,
2680	.print			= event_enable_print,
2681	.init			= event_enable_init,
2682	.free			= event_enable_free,
2683};
2684
2685static struct ftrace_probe_ops event_disable_count_probe_ops = {
2686	.func			= event_enable_count_probe,
2687	.print			= event_enable_print,
2688	.init			= event_enable_init,
2689	.free			= event_enable_free,
2690};
2691
2692static int
2693event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2694		  char *glob, char *cmd, char *param, int enabled)
2695{
2696	struct trace_event_file *file;
2697	struct ftrace_probe_ops *ops;
2698	struct event_probe_data *data;
2699	const char *system;
2700	const char *event;
2701	char *number;
2702	bool enable;
2703	int ret;
2704
2705	if (!tr)
2706		return -ENODEV;
2707
2708	/* hash funcs only work with set_ftrace_filter */
2709	if (!enabled || !param)
2710		return -EINVAL;
2711
2712	system = strsep(&param, ":");
2713	if (!param)
2714		return -EINVAL;
2715
2716	event = strsep(&param, ":");
2717
2718	mutex_lock(&event_mutex);
2719
2720	ret = -EINVAL;
2721	file = find_event_file(tr, system, event);
2722	if (!file)
2723		goto out;
2724
2725	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2726
2727	if (enable)
2728		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2729	else
2730		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2731
2732	if (glob[0] == '!') {
2733		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2734		goto out;
2735	}
2736
2737	ret = -ENOMEM;
2738
2739	data = kzalloc(sizeof(*data), GFP_KERNEL);
2740	if (!data)
2741		goto out;
2742
2743	data->enable = enable;
2744	data->count = -1;
2745	data->file = file;
2746
2747	if (!param)
2748		goto out_reg;
2749
2750	number = strsep(&param, ":");
2751
2752	ret = -EINVAL;
2753	if (!strlen(number))
2754		goto out_free;
2755
2756	/*
2757	 * We use the callback data field (which is a pointer)
2758	 * as our counter.
2759	 */
2760	ret = kstrtoul(number, 0, &data->count);
2761	if (ret)
2762		goto out_free;
2763
2764 out_reg:
2765	/* Don't let event modules unload while probe registered */
2766	ret = try_module_get(file->event_call->mod);
2767	if (!ret) {
2768		ret = -EBUSY;
2769		goto out_free;
2770	}
2771
2772	ret = __ftrace_event_enable_disable(file, 1, 1);
2773	if (ret < 0)
2774		goto out_put;
2775
2776	ret = register_ftrace_function_probe(glob, tr, ops, data);
2777	/*
2778	 * The above returns on success the # of functions enabled,
2779	 * but if it didn't find any functions it returns zero.
2780	 * Consider no functions a failure too.
2781	 */
2782	if (!ret) {
2783		ret = -ENOENT;
2784		goto out_disable;
2785	} else if (ret < 0)
2786		goto out_disable;
2787	/* Just return zero, not the number of enabled functions */
2788	ret = 0;
2789 out:
2790	mutex_unlock(&event_mutex);
2791	return ret;
2792
2793 out_disable:
2794	__ftrace_event_enable_disable(file, 0, 1);
2795 out_put:
2796	module_put(file->event_call->mod);
2797 out_free:
2798	kfree(data);
2799	goto out;
2800}
2801
2802static struct ftrace_func_command event_enable_cmd = {
2803	.name			= ENABLE_EVENT_STR,
2804	.func			= event_enable_func,
2805};
2806
2807static struct ftrace_func_command event_disable_cmd = {
2808	.name			= DISABLE_EVENT_STR,
2809	.func			= event_enable_func,
2810};
2811
2812static __init int register_event_cmds(void)
2813{
2814	int ret;
2815
2816	ret = register_ftrace_command(&event_enable_cmd);
2817	if (WARN_ON(ret < 0))
2818		return ret;
2819	ret = register_ftrace_command(&event_disable_cmd);
2820	if (WARN_ON(ret < 0))
2821		unregister_ftrace_command(&event_enable_cmd);
2822	return ret;
2823}
2824#else
2825static inline int register_event_cmds(void) { return 0; }
2826#endif /* CONFIG_DYNAMIC_FTRACE */
2827
2828/*
2829 * The top level array has already had its trace_event_file
2830 * descriptors created in order to allow for early events to
2831 * be recorded. This function is called after the tracefs has been
2832 * initialized, and we now have to create the files associated
2833 * to the events.
2834 */
2835static __init void
2836__trace_early_add_event_dirs(struct trace_array *tr)
2837{
2838	struct trace_event_file *file;
2839	int ret;
2840
2841
2842	list_for_each_entry(file, &tr->events, list) {
2843		ret = event_create_dir(tr->event_dir, file);
2844		if (ret < 0)
2845			pr_warn("Could not create directory for event %s\n",
2846				trace_event_name(file->event_call));
2847	}
2848}
2849
2850/*
2851 * For early boot up, the top trace array requires to have
2852 * a list of events that can be enabled. This must be done before
2853 * the filesystem is set up in order to allow events to be traced
2854 * early.
2855 */
2856static __init void
2857__trace_early_add_events(struct trace_array *tr)
2858{
2859	struct trace_event_call *call;
2860	int ret;
2861
2862	list_for_each_entry(call, &ftrace_events, list) {
2863		/* Early boot up should not have any modules loaded */
2864		if (WARN_ON_ONCE(call->mod))
 
2865			continue;
2866
2867		ret = __trace_early_add_new_event(call, tr);
2868		if (ret < 0)
2869			pr_warn("Could not create early event %s\n",
2870				trace_event_name(call));
2871	}
2872}
2873
2874/* Remove the event directory structure for a trace directory. */
2875static void
2876__trace_remove_event_dirs(struct trace_array *tr)
2877{
2878	struct trace_event_file *file, *next;
2879
2880	list_for_each_entry_safe(file, next, &tr->events, list)
2881		remove_event_file_dir(file);
2882}
2883
2884static void __add_event_to_tracers(struct trace_event_call *call)
2885{
2886	struct trace_array *tr;
2887
2888	list_for_each_entry(tr, &ftrace_trace_arrays, list)
2889		__trace_add_new_event(call, tr);
2890}
2891
2892extern struct trace_event_call *__start_ftrace_events[];
2893extern struct trace_event_call *__stop_ftrace_events[];
2894
2895static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2896
2897static __init int setup_trace_event(char *str)
2898{
2899	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2900	ring_buffer_expanded = true;
2901	tracing_selftest_disabled = true;
2902
2903	return 1;
2904}
2905__setup("trace_event=", setup_trace_event);
2906
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2907/* Expects to have event_mutex held when called */
2908static int
2909create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2910{
2911	struct dentry *d_events;
2912	struct dentry *entry;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2913
2914	entry = tracefs_create_file("set_event", 0644, parent,
2915				    tr, &ftrace_set_event_fops);
2916	if (!entry) {
2917		pr_warn("Could not create tracefs 'set_event' entry\n");
2918		return -ENOMEM;
2919	}
2920
2921	d_events = tracefs_create_dir("events", parent);
2922	if (!d_events) {
2923		pr_warn("Could not create tracefs 'events' directory\n");
2924		return -ENOMEM;
2925	}
2926
2927	entry = trace_create_file("enable", 0644, d_events,
2928				  tr, &ftrace_tr_enable_fops);
2929	if (!entry) {
2930		pr_warn("Could not create tracefs 'enable' entry\n");
2931		return -ENOMEM;
2932	}
2933
2934	/* There are not as crucial, just warn if they are not created */
2935
2936	entry = tracefs_create_file("set_event_pid", 0644, parent,
2937				    tr, &ftrace_set_event_pid_fops);
2938	if (!entry)
2939		pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2940
2941	/* ring buffer internal formats */
2942	entry = trace_create_file("header_page", 0444, d_events,
2943				  ring_buffer_print_page_header,
2944				  &ftrace_show_header_fops);
2945	if (!entry)
2946		pr_warn("Could not create tracefs 'header_page' entry\n");
2947
2948	entry = trace_create_file("header_event", 0444, d_events,
2949				  ring_buffer_print_entry_header,
2950				  &ftrace_show_header_fops);
2951	if (!entry)
2952		pr_warn("Could not create tracefs 'header_event' entry\n");
2953
2954	tr->event_dir = d_events;
2955
2956	return 0;
2957}
2958
2959/**
2960 * event_trace_add_tracer - add a instance of a trace_array to events
2961 * @parent: The parent dentry to place the files/directories for events in
2962 * @tr: The trace array associated with these events
2963 *
2964 * When a new instance is created, it needs to set up its events
2965 * directory, as well as other files associated with events. It also
2966 * creates the event hierachry in the @parent/events directory.
2967 *
2968 * Returns 0 on success.
2969 *
2970 * Must be called with event_mutex held.
2971 */
2972int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2973{
2974	int ret;
2975
2976	lockdep_assert_held(&event_mutex);
2977
2978	ret = create_event_toplevel_files(parent, tr);
2979	if (ret)
2980		goto out;
2981
2982	down_write(&trace_event_sem);
2983	__trace_add_event_dirs(tr);
 
 
 
 
2984	up_write(&trace_event_sem);
2985
2986 out:
2987	return ret;
2988}
2989
2990/*
2991 * The top trace array already had its file descriptors created.
2992 * Now the files themselves need to be created.
2993 */
2994static __init int
2995early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2996{
2997	int ret;
2998
2999	mutex_lock(&event_mutex);
3000
3001	ret = create_event_toplevel_files(parent, tr);
3002	if (ret)
3003		goto out_unlock;
3004
3005	down_write(&trace_event_sem);
3006	__trace_early_add_event_dirs(tr);
3007	up_write(&trace_event_sem);
3008
3009 out_unlock:
3010	mutex_unlock(&event_mutex);
3011
3012	return ret;
3013}
3014
3015/* Must be called with event_mutex held */
3016int event_trace_del_tracer(struct trace_array *tr)
3017{
3018	lockdep_assert_held(&event_mutex);
3019
3020	/* Disable any event triggers and associated soft-disabled events */
3021	clear_event_triggers(tr);
3022
3023	/* Clear the pid list */
3024	__ftrace_clear_event_pids(tr);
3025
3026	/* Disable any running events */
3027	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3028
3029	/* Make sure no more events are being executed */
3030	tracepoint_synchronize_unregister();
3031
3032	down_write(&trace_event_sem);
3033	__trace_remove_event_dirs(tr);
3034	tracefs_remove_recursive(tr->event_dir);
3035	up_write(&trace_event_sem);
3036
3037	tr->event_dir = NULL;
3038
3039	return 0;
3040}
3041
3042static __init int event_trace_memsetup(void)
3043{
3044	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3045	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3046	return 0;
3047}
3048
3049static __init void
3050early_enable_events(struct trace_array *tr, bool disable_first)
3051{
3052	char *buf = bootup_event_buf;
3053	char *token;
3054	int ret;
3055
3056	while (true) {
3057		token = strsep(&buf, ",");
3058
3059		if (!token)
3060			break;
3061
3062		if (*token) {
3063			/* Restarting syscalls requires that we stop them first */
3064			if (disable_first)
3065				ftrace_set_clr_event(tr, token, 0);
3066
3067			ret = ftrace_set_clr_event(tr, token, 1);
3068			if (ret)
3069				pr_warn("Failed to enable trace event: %s\n", token);
3070		}
3071
3072		/* Put back the comma to allow this to be called again */
3073		if (buf)
3074			*(buf - 1) = ',';
3075	}
3076}
3077
3078static __init int event_trace_enable(void)
3079{
3080	struct trace_array *tr = top_trace_array();
3081	struct trace_event_call **iter, *call;
3082	int ret;
3083
3084	if (!tr)
3085		return -ENODEV;
3086
3087	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3088
3089		call = *iter;
3090		ret = event_init(call);
3091		if (!ret)
3092			list_add(&call->list, &ftrace_events);
3093	}
3094
 
 
3095	/*
3096	 * We need the top trace array to have a working set of trace
3097	 * points at early init, before the debug files and directories
3098	 * are created. Create the file entries now, and attach them
3099	 * to the actual file dentries later.
3100	 */
3101	__trace_early_add_events(tr);
3102
3103	early_enable_events(tr, false);
3104
3105	trace_printk_start_comm();
3106
3107	register_event_cmds();
3108
3109	register_trigger_cmds();
3110
3111	return 0;
3112}
3113
3114/*
3115 * event_trace_enable() is called from trace_event_init() first to
3116 * initialize events and perhaps start any events that are on the
3117 * command line. Unfortunately, there are some events that will not
3118 * start this early, like the system call tracepoints that need
3119 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3120 * is called before pid 1 starts, and this flag is never set, making
3121 * the syscall tracepoint never get reached, but the event is enabled
3122 * regardless (and not doing anything).
3123 */
3124static __init int event_trace_enable_again(void)
3125{
3126	struct trace_array *tr;
3127
3128	tr = top_trace_array();
3129	if (!tr)
3130		return -ENODEV;
3131
3132	early_enable_events(tr, true);
3133
3134	return 0;
3135}
3136
3137early_initcall(event_trace_enable_again);
3138
 
 
 
 
 
 
 
 
 
 
 
 
3139__init int event_trace_init(void)
3140{
3141	struct trace_array *tr;
3142	struct dentry *d_tracer;
3143	struct dentry *entry;
3144	int ret;
3145
3146	tr = top_trace_array();
3147	if (!tr)
3148		return -ENODEV;
3149
3150	d_tracer = tracing_init_dentry();
3151	if (IS_ERR(d_tracer))
3152		return 0;
3153
3154	entry = tracefs_create_file("available_events", 0444, d_tracer,
3155				    tr, &ftrace_avail_fops);
3156	if (!entry)
3157		pr_warn("Could not create tracefs 'available_events' entry\n");
3158
3159	if (trace_define_generic_fields())
3160		pr_warn("tracing: Failed to allocated generic fields");
3161
3162	if (trace_define_common_fields())
3163		pr_warn("tracing: Failed to allocate common fields");
3164
3165	ret = early_event_add_tracer(d_tracer, tr);
3166	if (ret)
3167		return ret;
3168
3169#ifdef CONFIG_MODULES
3170	ret = register_module_notifier(&trace_module_nb);
3171	if (ret)
3172		pr_warn("Failed to register trace events module notifier\n");
3173#endif
 
 
 
3174	return 0;
3175}
3176
3177void __init trace_event_init(void)
3178{
3179	event_trace_memsetup();
3180	init_ftrace_syscalls();
3181	event_trace_enable();
 
3182}
3183
3184#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3185
3186static DEFINE_SPINLOCK(test_spinlock);
3187static DEFINE_SPINLOCK(test_spinlock_irq);
3188static DEFINE_MUTEX(test_mutex);
3189
3190static __init void test_work(struct work_struct *dummy)
3191{
3192	spin_lock(&test_spinlock);
3193	spin_lock_irq(&test_spinlock_irq);
3194	udelay(1);
3195	spin_unlock_irq(&test_spinlock_irq);
3196	spin_unlock(&test_spinlock);
3197
3198	mutex_lock(&test_mutex);
3199	msleep(1);
3200	mutex_unlock(&test_mutex);
3201}
3202
3203static __init int event_test_thread(void *unused)
3204{
3205	void *test_malloc;
3206
3207	test_malloc = kmalloc(1234, GFP_KERNEL);
3208	if (!test_malloc)
3209		pr_info("failed to kmalloc\n");
3210
3211	schedule_on_each_cpu(test_work);
3212
3213	kfree(test_malloc);
3214
3215	set_current_state(TASK_INTERRUPTIBLE);
3216	while (!kthread_should_stop()) {
3217		schedule();
3218		set_current_state(TASK_INTERRUPTIBLE);
3219	}
3220	__set_current_state(TASK_RUNNING);
3221
3222	return 0;
3223}
3224
3225/*
3226 * Do various things that may trigger events.
3227 */
3228static __init void event_test_stuff(void)
3229{
3230	struct task_struct *test_thread;
3231
3232	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3233	msleep(1);
3234	kthread_stop(test_thread);
3235}
3236
3237/*
3238 * For every trace event defined, we will test each trace point separately,
3239 * and then by groups, and finally all trace points.
3240 */
3241static __init void event_trace_self_tests(void)
3242{
3243	struct trace_subsystem_dir *dir;
3244	struct trace_event_file *file;
3245	struct trace_event_call *call;
3246	struct event_subsystem *system;
3247	struct trace_array *tr;
3248	int ret;
3249
3250	tr = top_trace_array();
3251	if (!tr)
3252		return;
3253
3254	pr_info("Running tests on trace events:\n");
3255
3256	list_for_each_entry(file, &tr->events, list) {
3257
3258		call = file->event_call;
3259
3260		/* Only test those that have a probe */
3261		if (!call->class || !call->class->probe)
3262			continue;
3263
3264/*
3265 * Testing syscall events here is pretty useless, but
3266 * we still do it if configured. But this is time consuming.
3267 * What we really need is a user thread to perform the
3268 * syscalls as we test.
3269 */
3270#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3271		if (call->class->system &&
3272		    strcmp(call->class->system, "syscalls") == 0)
3273			continue;
3274#endif
3275
3276		pr_info("Testing event %s: ", trace_event_name(call));
3277
3278		/*
3279		 * If an event is already enabled, someone is using
3280		 * it and the self test should not be on.
3281		 */
3282		if (file->flags & EVENT_FILE_FL_ENABLED) {
3283			pr_warn("Enabled event during self test!\n");
3284			WARN_ON_ONCE(1);
3285			continue;
3286		}
3287
3288		ftrace_event_enable_disable(file, 1);
3289		event_test_stuff();
3290		ftrace_event_enable_disable(file, 0);
3291
3292		pr_cont("OK\n");
3293	}
3294
3295	/* Now test at the sub system level */
3296
3297	pr_info("Running tests on trace event systems:\n");
3298
3299	list_for_each_entry(dir, &tr->systems, list) {
3300
3301		system = dir->subsystem;
3302
3303		/* the ftrace system is special, skip it */
3304		if (strcmp(system->name, "ftrace") == 0)
3305			continue;
3306
3307		pr_info("Testing event system %s: ", system->name);
3308
3309		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3310		if (WARN_ON_ONCE(ret)) {
3311			pr_warn("error enabling system %s\n",
3312				system->name);
3313			continue;
3314		}
3315
3316		event_test_stuff();
3317
3318		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3319		if (WARN_ON_ONCE(ret)) {
3320			pr_warn("error disabling system %s\n",
3321				system->name);
3322			continue;
3323		}
3324
3325		pr_cont("OK\n");
3326	}
3327
3328	/* Test with all events enabled */
3329
3330	pr_info("Running tests on all trace events:\n");
3331	pr_info("Testing all events: ");
3332
3333	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3334	if (WARN_ON_ONCE(ret)) {
3335		pr_warn("error enabling all events\n");
3336		return;
3337	}
3338
3339	event_test_stuff();
3340
3341	/* reset sysname */
3342	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3343	if (WARN_ON_ONCE(ret)) {
3344		pr_warn("error disabling all events\n");
3345		return;
3346	}
3347
3348	pr_cont("OK\n");
3349}
3350
3351#ifdef CONFIG_FUNCTION_TRACER
3352
3353static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3354
3355static struct trace_event_file event_trace_file __initdata;
3356
3357static void __init
3358function_test_events_call(unsigned long ip, unsigned long parent_ip,
3359			  struct ftrace_ops *op, struct pt_regs *pt_regs)
3360{
 
3361	struct ring_buffer_event *event;
3362	struct ring_buffer *buffer;
3363	struct ftrace_entry *entry;
3364	unsigned long flags;
3365	long disabled;
3366	int cpu;
3367	int pc;
3368
3369	pc = preempt_count();
3370	preempt_disable_notrace();
3371	cpu = raw_smp_processor_id();
3372	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3373
3374	if (disabled != 1)
3375		goto out;
3376
3377	local_save_flags(flags);
3378
3379	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3380						TRACE_FN, sizeof(*entry),
3381						flags, pc);
3382	if (!event)
3383		goto out;
3384	entry	= ring_buffer_event_data(event);
3385	entry->ip			= ip;
3386	entry->parent_ip		= parent_ip;
3387
3388	event_trigger_unlock_commit(&event_trace_file, buffer, event,
3389				    entry, flags, pc);
3390 out:
3391	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3392	preempt_enable_notrace();
3393}
3394
3395static struct ftrace_ops trace_ops __initdata  =
3396{
3397	.func = function_test_events_call,
3398	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
3399};
3400
3401static __init void event_trace_self_test_with_function(void)
3402{
3403	int ret;
3404
3405	event_trace_file.tr = top_trace_array();
3406	if (WARN_ON(!event_trace_file.tr))
3407		return;
3408
3409	ret = register_ftrace_function(&trace_ops);
3410	if (WARN_ON(ret < 0)) {
3411		pr_info("Failed to enable function tracer for event tests\n");
3412		return;
3413	}
3414	pr_info("Running tests again, along with the function tracer\n");
3415	event_trace_self_tests();
3416	unregister_ftrace_function(&trace_ops);
3417}
3418#else
3419static __init void event_trace_self_test_with_function(void)
3420{
3421}
3422#endif
3423
3424static __init int event_trace_self_tests_init(void)
3425{
3426	if (!tracing_selftest_disabled) {
3427		event_trace_self_tests();
3428		event_trace_self_test_with_function();
3429	}
3430
3431	return 0;
3432}
3433
3434late_initcall(event_trace_self_tests_init);
3435
3436#endif
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * event tracer
   4 *
   5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6 *
   7 *  - Added format output of fields of the trace point.
   8 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9 *
  10 */
  11
  12#define pr_fmt(fmt) fmt
  13
  14#include <linux/workqueue.h>
  15#include <linux/security.h>
  16#include <linux/spinlock.h>
  17#include <linux/kthread.h>
  18#include <linux/tracefs.h>
  19#include <linux/uaccess.h>
  20#include <linux/module.h>
  21#include <linux/ctype.h>
  22#include <linux/sort.h>
  23#include <linux/slab.h>
  24#include <linux/delay.h>
  25
  26#include <trace/events/sched.h>
  27#include <trace/syscall.h>
  28
  29#include <asm/setup.h>
  30
  31#include "trace_output.h"
  32
  33#undef TRACE_SYSTEM
  34#define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36DEFINE_MUTEX(event_mutex);
  37
  38LIST_HEAD(ftrace_events);
  39static LIST_HEAD(ftrace_generic_fields);
  40static LIST_HEAD(ftrace_common_fields);
  41static bool eventdir_initialized;
  42
  43static LIST_HEAD(module_strings);
  44
  45struct module_string {
  46	struct list_head	next;
  47	struct module		*module;
  48	char			*str;
  49};
  50
  51#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  52
  53static struct kmem_cache *field_cachep;
  54static struct kmem_cache *file_cachep;
  55
  56static inline int system_refcount(struct event_subsystem *system)
  57{
  58	return system->ref_count;
  59}
  60
  61static int system_refcount_inc(struct event_subsystem *system)
  62{
  63	return system->ref_count++;
  64}
  65
  66static int system_refcount_dec(struct event_subsystem *system)
  67{
  68	return --system->ref_count;
  69}
  70
  71/* Double loops, do not use break, only goto's work */
  72#define do_for_each_event_file(tr, file)			\
  73	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  74		list_for_each_entry(file, &tr->events, list)
  75
  76#define do_for_each_event_file_safe(tr, file)			\
  77	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
  78		struct trace_event_file *___n;				\
  79		list_for_each_entry_safe(file, ___n, &tr->events, list)
  80
  81#define while_for_each_event_file()		\
  82	}
  83
  84static struct ftrace_event_field *
  85__find_event_field(struct list_head *head, char *name)
  86{
  87	struct ftrace_event_field *field;
  88
  89	list_for_each_entry(field, head, link) {
  90		if (!strcmp(field->name, name))
  91			return field;
  92	}
  93
  94	return NULL;
  95}
  96
  97struct ftrace_event_field *
  98trace_find_event_field(struct trace_event_call *call, char *name)
  99{
 100	struct ftrace_event_field *field;
 101	struct list_head *head;
 102
 103	head = trace_get_fields(call);
 104	field = __find_event_field(head, name);
 105	if (field)
 106		return field;
 107
 108	field = __find_event_field(&ftrace_generic_fields, name);
 109	if (field)
 110		return field;
 111
 112	return __find_event_field(&ftrace_common_fields, name);
 113}
 114
 115static int __trace_define_field(struct list_head *head, const char *type,
 116				const char *name, int offset, int size,
 117				int is_signed, int filter_type, int len)
 118{
 119	struct ftrace_event_field *field;
 120
 121	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 122	if (!field)
 123		return -ENOMEM;
 124
 125	field->name = name;
 126	field->type = type;
 127
 128	if (filter_type == FILTER_OTHER)
 129		field->filter_type = filter_assign_type(type);
 130	else
 131		field->filter_type = filter_type;
 132
 133	field->offset = offset;
 134	field->size = size;
 135	field->is_signed = is_signed;
 136	field->len = len;
 137
 138	list_add(&field->link, head);
 139
 140	return 0;
 141}
 142
 143int trace_define_field(struct trace_event_call *call, const char *type,
 144		       const char *name, int offset, int size, int is_signed,
 145		       int filter_type)
 146{
 147	struct list_head *head;
 148
 149	if (WARN_ON(!call->class))
 150		return 0;
 151
 152	head = trace_get_fields(call);
 153	return __trace_define_field(head, type, name, offset, size,
 154				    is_signed, filter_type, 0);
 155}
 156EXPORT_SYMBOL_GPL(trace_define_field);
 157
 158static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 159		       const char *name, int offset, int size, int is_signed,
 160		       int filter_type, int len)
 161{
 162	struct list_head *head;
 163
 164	if (WARN_ON(!call->class))
 165		return 0;
 166
 167	head = trace_get_fields(call);
 168	return __trace_define_field(head, type, name, offset, size,
 169				    is_signed, filter_type, len);
 170}
 171
 172#define __generic_field(type, item, filter_type)			\
 173	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 174				   #item, 0, 0, is_signed_type(type),	\
 175				   filter_type, 0);			\
 176	if (ret)							\
 177		return ret;
 178
 179#define __common_field(type, item)					\
 180	ret = __trace_define_field(&ftrace_common_fields, #type,	\
 181				   "common_" #item,			\
 182				   offsetof(typeof(ent), item),		\
 183				   sizeof(ent.item),			\
 184				   is_signed_type(type), FILTER_OTHER, 0);	\
 185	if (ret)							\
 186		return ret;
 187
 188static int trace_define_generic_fields(void)
 189{
 190	int ret;
 191
 192	__generic_field(int, CPU, FILTER_CPU);
 193	__generic_field(int, cpu, FILTER_CPU);
 194	__generic_field(int, common_cpu, FILTER_CPU);
 195	__generic_field(char *, COMM, FILTER_COMM);
 196	__generic_field(char *, comm, FILTER_COMM);
 197	__generic_field(char *, stacktrace, FILTER_STACKTRACE);
 198	__generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 199
 200	return ret;
 201}
 202
 203static int trace_define_common_fields(void)
 204{
 205	int ret;
 206	struct trace_entry ent;
 207
 208	__common_field(unsigned short, type);
 209	__common_field(unsigned char, flags);
 210	/* Holds both preempt_count and migrate_disable */
 211	__common_field(unsigned char, preempt_count);
 212	__common_field(int, pid);
 213
 214	return ret;
 215}
 216
 217static void trace_destroy_fields(struct trace_event_call *call)
 218{
 219	struct ftrace_event_field *field, *next;
 220	struct list_head *head;
 221
 222	head = trace_get_fields(call);
 223	list_for_each_entry_safe(field, next, head, link) {
 224		list_del(&field->link);
 225		kmem_cache_free(field_cachep, field);
 226	}
 227}
 228
 229/*
 230 * run-time version of trace_event_get_offsets_<call>() that returns the last
 231 * accessible offset of trace fields excluding __dynamic_array bytes
 232 */
 233int trace_event_get_offsets(struct trace_event_call *call)
 234{
 235	struct ftrace_event_field *tail;
 236	struct list_head *head;
 237
 238	head = trace_get_fields(call);
 239	/*
 240	 * head->next points to the last field with the largest offset,
 241	 * since it was added last by trace_define_field()
 242	 */
 243	tail = list_first_entry(head, struct ftrace_event_field, link);
 244	return tail->offset + tail->size;
 245}
 246
 247/*
 248 * Check if the referenced field is an array and return true,
 249 * as arrays are OK to dereference.
 250 */
 251static bool test_field(const char *fmt, struct trace_event_call *call)
 252{
 253	struct trace_event_fields *field = call->class->fields_array;
 254	const char *array_descriptor;
 255	const char *p = fmt;
 256	int len;
 257
 258	if (!(len = str_has_prefix(fmt, "REC->")))
 259		return false;
 260	fmt += len;
 261	for (p = fmt; *p; p++) {
 262		if (!isalnum(*p) && *p != '_')
 263			break;
 264	}
 265	len = p - fmt;
 266
 267	for (; field->type; field++) {
 268		if (strncmp(field->name, fmt, len) ||
 269		    field->name[len])
 270			continue;
 271		array_descriptor = strchr(field->type, '[');
 272		/* This is an array and is OK to dereference. */
 273		return array_descriptor != NULL;
 274	}
 275	return false;
 276}
 277
 278/*
 279 * Examine the print fmt of the event looking for unsafe dereference
 280 * pointers using %p* that could be recorded in the trace event and
 281 * much later referenced after the pointer was freed. Dereferencing
 282 * pointers are OK, if it is dereferenced into the event itself.
 283 */
 284static void test_event_printk(struct trace_event_call *call)
 285{
 286	u64 dereference_flags = 0;
 287	bool first = true;
 288	const char *fmt, *c, *r, *a;
 289	int parens = 0;
 290	char in_quote = 0;
 291	int start_arg = 0;
 292	int arg = 0;
 293	int i;
 294
 295	fmt = call->print_fmt;
 296
 297	if (!fmt)
 298		return;
 299
 300	for (i = 0; fmt[i]; i++) {
 301		switch (fmt[i]) {
 302		case '\\':
 303			i++;
 304			if (!fmt[i])
 305				return;
 306			continue;
 307		case '"':
 308		case '\'':
 309			/*
 310			 * The print fmt starts with a string that
 311			 * is processed first to find %p* usage,
 312			 * then after the first string, the print fmt
 313			 * contains arguments that are used to check
 314			 * if the dereferenced %p* usage is safe.
 315			 */
 316			if (first) {
 317				if (fmt[i] == '\'')
 318					continue;
 319				if (in_quote) {
 320					arg = 0;
 321					first = false;
 322					/*
 323					 * If there was no %p* uses
 324					 * the fmt is OK.
 325					 */
 326					if (!dereference_flags)
 327						return;
 328				}
 329			}
 330			if (in_quote) {
 331				if (in_quote == fmt[i])
 332					in_quote = 0;
 333			} else {
 334				in_quote = fmt[i];
 335			}
 336			continue;
 337		case '%':
 338			if (!first || !in_quote)
 339				continue;
 340			i++;
 341			if (!fmt[i])
 342				return;
 343			switch (fmt[i]) {
 344			case '%':
 345				continue;
 346			case 'p':
 347				/* Find dereferencing fields */
 348				switch (fmt[i + 1]) {
 349				case 'B': case 'R': case 'r':
 350				case 'b': case 'M': case 'm':
 351				case 'I': case 'i': case 'E':
 352				case 'U': case 'V': case 'N':
 353				case 'a': case 'd': case 'D':
 354				case 'g': case 't': case 'C':
 355				case 'O': case 'f':
 356					if (WARN_ONCE(arg == 63,
 357						      "Too many args for event: %s",
 358						      trace_event_name(call)))
 359						return;
 360					dereference_flags |= 1ULL << arg;
 361				}
 362				break;
 363			default:
 364			{
 365				bool star = false;
 366				int j;
 367
 368				/* Increment arg if %*s exists. */
 369				for (j = 0; fmt[i + j]; j++) {
 370					if (isdigit(fmt[i + j]) ||
 371					    fmt[i + j] == '.')
 372						continue;
 373					if (fmt[i + j] == '*') {
 374						star = true;
 375						continue;
 376					}
 377					if ((fmt[i + j] == 's') && star)
 378						arg++;
 379					break;
 380				}
 381				break;
 382			} /* default */
 383
 384			} /* switch */
 385			arg++;
 386			continue;
 387		case '(':
 388			if (in_quote)
 389				continue;
 390			parens++;
 391			continue;
 392		case ')':
 393			if (in_quote)
 394				continue;
 395			parens--;
 396			if (WARN_ONCE(parens < 0,
 397				      "Paren mismatch for event: %s\narg='%s'\n%*s",
 398				      trace_event_name(call),
 399				      fmt + start_arg,
 400				      (i - start_arg) + 5, "^"))
 401				return;
 402			continue;
 403		case ',':
 404			if (in_quote || parens)
 405				continue;
 406			i++;
 407			while (isspace(fmt[i]))
 408				i++;
 409			start_arg = i;
 410			if (!(dereference_flags & (1ULL << arg)))
 411				goto next_arg;
 412
 413			/* Find the REC-> in the argument */
 414			c = strchr(fmt + i, ',');
 415			r = strstr(fmt + i, "REC->");
 416			if (r && (!c || r < c)) {
 417				/*
 418				 * Addresses of events on the buffer,
 419				 * or an array on the buffer is
 420				 * OK to dereference.
 421				 * There's ways to fool this, but
 422				 * this is to catch common mistakes,
 423				 * not malicious code.
 424				 */
 425				a = strchr(fmt + i, '&');
 426				if ((a && (a < r)) || test_field(r, call))
 427					dereference_flags &= ~(1ULL << arg);
 428			} else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
 429				   (!c || r < c)) {
 430				dereference_flags &= ~(1ULL << arg);
 431			} else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
 432				   (!c || r < c)) {
 433				dereference_flags &= ~(1ULL << arg);
 434			}
 435
 436		next_arg:
 437			i--;
 438			arg++;
 439		}
 440	}
 441
 442	/*
 443	 * If you triggered the below warning, the trace event reported
 444	 * uses an unsafe dereference pointer %p*. As the data stored
 445	 * at the trace event time may no longer exist when the trace
 446	 * event is printed, dereferencing to the original source is
 447	 * unsafe. The source of the dereference must be copied into the
 448	 * event itself, and the dereference must access the copy instead.
 449	 */
 450	if (WARN_ON_ONCE(dereference_flags)) {
 451		arg = 1;
 452		while (!(dereference_flags & 1)) {
 453			dereference_flags >>= 1;
 454			arg++;
 455		}
 456		pr_warn("event %s has unsafe dereference of argument %d\n",
 457			trace_event_name(call), arg);
 458		pr_warn("print_fmt: %s\n", fmt);
 459	}
 460}
 461
 462int trace_event_raw_init(struct trace_event_call *call)
 463{
 464	int id;
 465
 466	id = register_trace_event(&call->event);
 467	if (!id)
 468		return -ENODEV;
 469
 470	test_event_printk(call);
 471
 472	return 0;
 473}
 474EXPORT_SYMBOL_GPL(trace_event_raw_init);
 475
 476bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 477{
 478	struct trace_array *tr = trace_file->tr;
 479	struct trace_array_cpu *data;
 480	struct trace_pid_list *no_pid_list;
 481	struct trace_pid_list *pid_list;
 482
 483	pid_list = rcu_dereference_raw(tr->filtered_pids);
 484	no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 485
 486	if (!pid_list && !no_pid_list)
 487		return false;
 488
 489	data = this_cpu_ptr(tr->array_buffer.data);
 490
 491	return data->ignore_pid;
 492}
 493EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 494
 495void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 496				 struct trace_event_file *trace_file,
 497				 unsigned long len)
 498{
 499	struct trace_event_call *event_call = trace_file->event_call;
 500
 501	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 502	    trace_event_ignore_this_pid(trace_file))
 503		return NULL;
 504
 
 
 505	/*
 506	 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 507	 * preemption (adding one to the preempt_count). Since we are
 508	 * interested in the preempt_count at the time the tracepoint was
 509	 * hit, we need to subtract one to offset the increment.
 510	 */
 511	fbuffer->trace_ctx = tracing_gen_ctx_dec();
 
 512	fbuffer->trace_file = trace_file;
 513
 514	fbuffer->event =
 515		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 516						event_call->event.type, len,
 517						fbuffer->trace_ctx);
 518	if (!fbuffer->event)
 519		return NULL;
 520
 521	fbuffer->regs = NULL;
 522	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 523	return fbuffer->entry;
 524}
 525EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 526
 527int trace_event_reg(struct trace_event_call *call,
 528		    enum trace_reg type, void *data)
 529{
 530	struct trace_event_file *file = data;
 531
 532	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 533	switch (type) {
 534	case TRACE_REG_REGISTER:
 535		return tracepoint_probe_register(call->tp,
 536						 call->class->probe,
 537						 file);
 538	case TRACE_REG_UNREGISTER:
 539		tracepoint_probe_unregister(call->tp,
 540					    call->class->probe,
 541					    file);
 542		return 0;
 543
 544#ifdef CONFIG_PERF_EVENTS
 545	case TRACE_REG_PERF_REGISTER:
 546		return tracepoint_probe_register(call->tp,
 547						 call->class->perf_probe,
 548						 call);
 549	case TRACE_REG_PERF_UNREGISTER:
 550		tracepoint_probe_unregister(call->tp,
 551					    call->class->perf_probe,
 552					    call);
 553		return 0;
 554	case TRACE_REG_PERF_OPEN:
 555	case TRACE_REG_PERF_CLOSE:
 556	case TRACE_REG_PERF_ADD:
 557	case TRACE_REG_PERF_DEL:
 558		return 0;
 559#endif
 560	}
 561	return 0;
 562}
 563EXPORT_SYMBOL_GPL(trace_event_reg);
 564
 565void trace_event_enable_cmd_record(bool enable)
 566{
 567	struct trace_event_file *file;
 568	struct trace_array *tr;
 569
 570	lockdep_assert_held(&event_mutex);
 571
 572	do_for_each_event_file(tr, file) {
 573
 574		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 575			continue;
 576
 577		if (enable) {
 578			tracing_start_cmdline_record();
 579			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 580		} else {
 581			tracing_stop_cmdline_record();
 582			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 583		}
 584	} while_for_each_event_file();
 
 585}
 586
 587void trace_event_enable_tgid_record(bool enable)
 588{
 589	struct trace_event_file *file;
 590	struct trace_array *tr;
 591
 592	lockdep_assert_held(&event_mutex);
 593
 594	do_for_each_event_file(tr, file) {
 595		if (!(file->flags & EVENT_FILE_FL_ENABLED))
 596			continue;
 597
 598		if (enable) {
 599			tracing_start_tgid_record();
 600			set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 601		} else {
 602			tracing_stop_tgid_record();
 603			clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 604				  &file->flags);
 605		}
 606	} while_for_each_event_file();
 
 607}
 608
 609static int __ftrace_event_enable_disable(struct trace_event_file *file,
 610					 int enable, int soft_disable)
 611{
 612	struct trace_event_call *call = file->event_call;
 613	struct trace_array *tr = file->tr;
 
 614	int ret = 0;
 615	int disable;
 616
 617	switch (enable) {
 618	case 0:
 619		/*
 620		 * When soft_disable is set and enable is cleared, the sm_ref
 621		 * reference counter is decremented. If it reaches 0, we want
 622		 * to clear the SOFT_DISABLED flag but leave the event in the
 623		 * state that it was. That is, if the event was enabled and
 624		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 625		 * is set we do not want the event to be enabled before we
 626		 * clear the bit.
 627		 *
 628		 * When soft_disable is not set but the SOFT_MODE flag is,
 629		 * we do nothing. Do not disable the tracepoint, otherwise
 630		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 631		 */
 632		if (soft_disable) {
 633			if (atomic_dec_return(&file->sm_ref) > 0)
 634				break;
 635			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 636			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 637			/* Disable use of trace_buffered_event */
 638			trace_buffered_event_disable();
 639		} else
 640			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 641
 642		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 643			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 644			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 645				tracing_stop_cmdline_record();
 646				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 647			}
 648
 649			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 650				tracing_stop_tgid_record();
 651				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 652			}
 653
 654			call->class->reg(call, TRACE_REG_UNREGISTER, file);
 655		}
 656		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 657		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 658			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 659		else
 660			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 661		break;
 662	case 1:
 663		/*
 664		 * When soft_disable is set and enable is set, we want to
 665		 * register the tracepoint for the event, but leave the event
 666		 * as is. That means, if the event was already enabled, we do
 667		 * nothing (but set SOFT_MODE). If the event is disabled, we
 668		 * set SOFT_DISABLED before enabling the event tracepoint, so
 669		 * it still seems to be disabled.
 670		 */
 671		if (!soft_disable)
 672			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 673		else {
 674			if (atomic_inc_return(&file->sm_ref) > 1)
 675				break;
 676			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 677			/* Enable use of trace_buffered_event */
 678			trace_buffered_event_enable();
 679		}
 680
 681		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 682			bool cmd = false, tgid = false;
 683
 684			/* Keep the event disabled, when going to SOFT_MODE. */
 685			if (soft_disable)
 686				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 687
 688			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 689				cmd = true;
 690				tracing_start_cmdline_record();
 691				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 692			}
 693
 694			if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 695				tgid = true;
 696				tracing_start_tgid_record();
 697				set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 698			}
 699
 700			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 701			if (ret) {
 702				if (cmd)
 703					tracing_stop_cmdline_record();
 704				if (tgid)
 705					tracing_stop_tgid_record();
 706				pr_info("event trace: Could not enable event "
 707					"%s\n", trace_event_name(call));
 708				break;
 709			}
 710			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 711
 712			/* WAS_ENABLED gets set but never cleared. */
 713			set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 714		}
 715		break;
 716	}
 717
 
 
 
 
 
 
 
 
 
 718	return ret;
 719}
 720
 721int trace_event_enable_disable(struct trace_event_file *file,
 722			       int enable, int soft_disable)
 723{
 724	return __ftrace_event_enable_disable(file, enable, soft_disable);
 725}
 726
 727static int ftrace_event_enable_disable(struct trace_event_file *file,
 728				       int enable)
 729{
 730	return __ftrace_event_enable_disable(file, enable, 0);
 731}
 732
 733static void ftrace_clear_events(struct trace_array *tr)
 734{
 735	struct trace_event_file *file;
 736
 737	mutex_lock(&event_mutex);
 738	list_for_each_entry(file, &tr->events, list) {
 739		ftrace_event_enable_disable(file, 0);
 740	}
 741	mutex_unlock(&event_mutex);
 742}
 743
 744static void
 745event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 746{
 747	struct trace_pid_list *pid_list;
 748	struct trace_array *tr = data;
 749
 750	pid_list = rcu_dereference_raw(tr->filtered_pids);
 751	trace_filter_add_remove_task(pid_list, NULL, task);
 752
 753	pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 754	trace_filter_add_remove_task(pid_list, NULL, task);
 755}
 756
 757static void
 758event_filter_pid_sched_process_fork(void *data,
 759				    struct task_struct *self,
 760				    struct task_struct *task)
 761{
 762	struct trace_pid_list *pid_list;
 763	struct trace_array *tr = data;
 764
 765	pid_list = rcu_dereference_sched(tr->filtered_pids);
 766	trace_filter_add_remove_task(pid_list, self, task);
 767
 768	pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 769	trace_filter_add_remove_task(pid_list, self, task);
 770}
 771
 772void trace_event_follow_fork(struct trace_array *tr, bool enable)
 773{
 774	if (enable) {
 775		register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 776						       tr, INT_MIN);
 777		register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 778						       tr, INT_MAX);
 779	} else {
 780		unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 781						    tr);
 782		unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 783						    tr);
 784	}
 785}
 786
 787static void
 788event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 789					struct task_struct *prev,
 790					struct task_struct *next,
 791					unsigned int prev_state)
 792{
 793	struct trace_array *tr = data;
 794	struct trace_pid_list *no_pid_list;
 795	struct trace_pid_list *pid_list;
 796	bool ret;
 797
 798	pid_list = rcu_dereference_sched(tr->filtered_pids);
 799	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 800
 801	/*
 802	 * Sched switch is funny, as we only want to ignore it
 803	 * in the notrace case if both prev and next should be ignored.
 804	 */
 805	ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 806		trace_ignore_this_task(NULL, no_pid_list, next);
 807
 808	this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 809		       (trace_ignore_this_task(pid_list, NULL, prev) &&
 810			trace_ignore_this_task(pid_list, NULL, next)));
 811}
 812
 813static void
 814event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 815					 struct task_struct *prev,
 816					 struct task_struct *next,
 817					 unsigned int prev_state)
 818{
 819	struct trace_array *tr = data;
 820	struct trace_pid_list *no_pid_list;
 821	struct trace_pid_list *pid_list;
 822
 823	pid_list = rcu_dereference_sched(tr->filtered_pids);
 824	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 825
 826	this_cpu_write(tr->array_buffer.data->ignore_pid,
 827		       trace_ignore_this_task(pid_list, no_pid_list, next));
 828}
 829
 830static void
 831event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 832{
 833	struct trace_array *tr = data;
 834	struct trace_pid_list *no_pid_list;
 835	struct trace_pid_list *pid_list;
 836
 837	/* Nothing to do if we are already tracing */
 838	if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 839		return;
 840
 841	pid_list = rcu_dereference_sched(tr->filtered_pids);
 842	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 843
 844	this_cpu_write(tr->array_buffer.data->ignore_pid,
 845		       trace_ignore_this_task(pid_list, no_pid_list, task));
 846}
 847
 848static void
 849event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 850{
 851	struct trace_array *tr = data;
 852	struct trace_pid_list *no_pid_list;
 853	struct trace_pid_list *pid_list;
 854
 855	/* Nothing to do if we are not tracing */
 856	if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 857		return;
 858
 859	pid_list = rcu_dereference_sched(tr->filtered_pids);
 860	no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 861
 862	/* Set tracing if current is enabled */
 863	this_cpu_write(tr->array_buffer.data->ignore_pid,
 864		       trace_ignore_this_task(pid_list, no_pid_list, current));
 865}
 866
 867static void unregister_pid_events(struct trace_array *tr)
 868{
 
 
 
 
 
 
 
 
 
 869	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 870	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 871
 872	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 873	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 874
 875	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 876	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 877
 878	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 879	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 880}
 881
 882static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 883{
 884	struct trace_pid_list *pid_list;
 885	struct trace_pid_list *no_pid_list;
 886	struct trace_event_file *file;
 887	int cpu;
 888
 889	pid_list = rcu_dereference_protected(tr->filtered_pids,
 890					     lockdep_is_held(&event_mutex));
 891	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 892					     lockdep_is_held(&event_mutex));
 893
 894	/* Make sure there's something to do */
 895	if (!pid_type_enabled(type, pid_list, no_pid_list))
 896		return;
 897
 898	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 899		unregister_pid_events(tr);
 900
 901		list_for_each_entry(file, &tr->events, list) {
 902			clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 903		}
 904
 905		for_each_possible_cpu(cpu)
 906			per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 907	}
 908
 909	if (type & TRACE_PIDS)
 910		rcu_assign_pointer(tr->filtered_pids, NULL);
 911
 912	if (type & TRACE_NO_PIDS)
 913		rcu_assign_pointer(tr->filtered_no_pids, NULL);
 914
 915	/* Wait till all users are no longer using pid filtering */
 916	tracepoint_synchronize_unregister();
 917
 918	if ((type & TRACE_PIDS) && pid_list)
 919		trace_pid_list_free(pid_list);
 920
 921	if ((type & TRACE_NO_PIDS) && no_pid_list)
 922		trace_pid_list_free(no_pid_list);
 923}
 924
 925static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 926{
 927	mutex_lock(&event_mutex);
 928	__ftrace_clear_event_pids(tr, type);
 929	mutex_unlock(&event_mutex);
 930}
 931
 932static void __put_system(struct event_subsystem *system)
 933{
 934	struct event_filter *filter = system->filter;
 935
 936	WARN_ON_ONCE(system_refcount(system) == 0);
 937	if (system_refcount_dec(system))
 938		return;
 939
 940	list_del(&system->list);
 941
 942	if (filter) {
 943		kfree(filter->filter_string);
 944		kfree(filter);
 945	}
 946	kfree_const(system->name);
 947	kfree(system);
 948}
 949
 950static void __get_system(struct event_subsystem *system)
 951{
 952	WARN_ON_ONCE(system_refcount(system) == 0);
 953	system_refcount_inc(system);
 954}
 955
 956static void __get_system_dir(struct trace_subsystem_dir *dir)
 957{
 958	WARN_ON_ONCE(dir->ref_count == 0);
 959	dir->ref_count++;
 960	__get_system(dir->subsystem);
 961}
 962
 963static void __put_system_dir(struct trace_subsystem_dir *dir)
 964{
 965	WARN_ON_ONCE(dir->ref_count == 0);
 966	/* If the subsystem is about to be freed, the dir must be too */
 967	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 968
 969	__put_system(dir->subsystem);
 970	if (!--dir->ref_count)
 971		kfree(dir);
 972}
 973
 974static void put_system(struct trace_subsystem_dir *dir)
 975{
 976	mutex_lock(&event_mutex);
 977	__put_system_dir(dir);
 978	mutex_unlock(&event_mutex);
 979}
 980
 981static void remove_subsystem(struct trace_subsystem_dir *dir)
 982{
 983	if (!dir)
 984		return;
 985
 986	if (!--dir->nr_events) {
 987		eventfs_remove_dir(dir->ei);
 988		list_del(&dir->list);
 989		__put_system_dir(dir);
 990	}
 991}
 992
 993void event_file_get(struct trace_event_file *file)
 994{
 995	atomic_inc(&file->ref);
 996}
 997
 998void event_file_put(struct trace_event_file *file)
 999{
1000	if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001		if (file->flags & EVENT_FILE_FL_FREED)
1002			kmem_cache_free(file_cachep, file);
1003		return;
1004	}
1005
1006	if (atomic_dec_and_test(&file->ref)) {
1007		/* Count should only go to zero when it is freed */
1008		if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009			return;
1010		kmem_cache_free(file_cachep, file);
1011	}
1012}
1013
1014static void remove_event_file_dir(struct trace_event_file *file)
1015{
1016	eventfs_remove_dir(file->ei);
1017	list_del(&file->list);
1018	remove_subsystem(file->system);
1019	free_event_filter(file->filter);
1020	file->flags |= EVENT_FILE_FL_FREED;
1021	event_file_put(file);
1022}
1023
1024/*
1025 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026 */
1027static int
1028__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029			      const char *sub, const char *event, int set)
1030{
1031	struct trace_event_file *file;
1032	struct trace_event_call *call;
1033	const char *name;
1034	int ret = -EINVAL;
1035	int eret = 0;
1036
1037	list_for_each_entry(file, &tr->events, list) {
1038
1039		call = file->event_call;
1040		name = trace_event_name(call);
1041
1042		if (!name || !call->class || !call->class->reg)
1043			continue;
1044
1045		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046			continue;
1047
1048		if (match &&
1049		    strcmp(match, name) != 0 &&
1050		    strcmp(match, call->class->system) != 0)
1051			continue;
1052
1053		if (sub && strcmp(sub, call->class->system) != 0)
1054			continue;
1055
1056		if (event && strcmp(event, name) != 0)
1057			continue;
1058
1059		ret = ftrace_event_enable_disable(file, set);
1060
1061		/*
1062		 * Save the first error and return that. Some events
1063		 * may still have been enabled, but let the user
1064		 * know that something went wrong.
1065		 */
1066		if (ret && !eret)
1067			eret = ret;
1068
1069		ret = eret;
1070	}
1071
1072	return ret;
1073}
1074
1075static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076				  const char *sub, const char *event, int set)
1077{
1078	int ret;
1079
1080	mutex_lock(&event_mutex);
1081	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082	mutex_unlock(&event_mutex);
1083
1084	return ret;
1085}
1086
1087int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088{
1089	char *event = NULL, *sub = NULL, *match;
1090	int ret;
1091
1092	if (!tr)
1093		return -ENOENT;
1094	/*
1095	 * The buf format can be <subsystem>:<event-name>
1096	 *  *:<event-name> means any event by that name.
1097	 *  :<event-name> is the same.
1098	 *
1099	 *  <subsystem>:* means all events in that subsystem
1100	 *  <subsystem>: means the same.
1101	 *
1102	 *  <name> (no ':') means all events in a subsystem with
1103	 *  the name <name> or any event that matches <name>
1104	 */
1105
1106	match = strsep(&buf, ":");
1107	if (buf) {
1108		sub = match;
1109		event = buf;
1110		match = NULL;
1111
1112		if (!strlen(sub) || strcmp(sub, "*") == 0)
1113			sub = NULL;
1114		if (!strlen(event) || strcmp(event, "*") == 0)
1115			event = NULL;
1116	}
1117
1118	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120	/* Put back the colon to allow this to be called again */
1121	if (buf)
1122		*(buf - 1) = ':';
1123
1124	return ret;
1125}
 
1126
1127/**
1128 * trace_set_clr_event - enable or disable an event
1129 * @system: system name to match (NULL for any system)
1130 * @event: event name to match (NULL for all events, within system)
1131 * @set: 1 to enable, 0 to disable
1132 *
1133 * This is a way for other parts of the kernel to enable or disable
1134 * event recording.
1135 *
1136 * Returns 0 on success, -EINVAL if the parameters do not match any
1137 * registered events.
1138 */
1139int trace_set_clr_event(const char *system, const char *event, int set)
1140{
1141	struct trace_array *tr = top_trace_array();
1142
1143	if (!tr)
1144		return -ENODEV;
1145
1146	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147}
1148EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150/**
1151 * trace_array_set_clr_event - enable or disable an event for a trace array.
1152 * @tr: concerned trace array.
1153 * @system: system name to match (NULL for any system)
1154 * @event: event name to match (NULL for all events, within system)
1155 * @enable: true to enable, false to disable
1156 *
1157 * This is a way for other parts of the kernel to enable or disable
1158 * event recording.
1159 *
1160 * Returns 0 on success, -EINVAL if the parameters do not match any
1161 * registered events.
1162 */
1163int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164		const char *event, bool enable)
1165{
1166	int set;
1167
1168	if (!tr)
1169		return -ENOENT;
1170
1171	set = (enable == true) ? 1 : 0;
1172	return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173}
1174EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176/* 128 should be much more than enough */
1177#define EVENT_BUF_SIZE		127
1178
1179static ssize_t
1180ftrace_event_write(struct file *file, const char __user *ubuf,
1181		   size_t cnt, loff_t *ppos)
1182{
1183	struct trace_parser parser;
1184	struct seq_file *m = file->private_data;
1185	struct trace_array *tr = m->private;
1186	ssize_t read, ret;
1187
1188	if (!cnt)
1189		return 0;
1190
1191	ret = tracing_update_buffers(tr);
1192	if (ret < 0)
1193		return ret;
1194
1195	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196		return -ENOMEM;
1197
1198	read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200	if (read >= 0 && trace_parser_loaded((&parser))) {
1201		int set = 1;
1202
1203		if (*parser.buffer == '!')
1204			set = 0;
1205
1206		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207		if (ret)
1208			goto out_put;
1209	}
1210
1211	ret = read;
1212
1213 out_put:
1214	trace_parser_put(&parser);
1215
1216	return ret;
1217}
1218
1219static void *
1220t_next(struct seq_file *m, void *v, loff_t *pos)
1221{
1222	struct trace_event_file *file = v;
1223	struct trace_event_call *call;
1224	struct trace_array *tr = m->private;
1225
1226	(*pos)++;
1227
1228	list_for_each_entry_continue(file, &tr->events, list) {
1229		call = file->event_call;
1230		/*
1231		 * The ftrace subsystem is for showing formats only.
1232		 * They can not be enabled or disabled via the event files.
1233		 */
1234		if (call->class && call->class->reg &&
1235		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236			return file;
1237	}
1238
1239	return NULL;
1240}
1241
1242static void *t_start(struct seq_file *m, loff_t *pos)
1243{
1244	struct trace_event_file *file;
1245	struct trace_array *tr = m->private;
1246	loff_t l;
1247
1248	mutex_lock(&event_mutex);
1249
1250	file = list_entry(&tr->events, struct trace_event_file, list);
1251	for (l = 0; l <= *pos; ) {
1252		file = t_next(m, file, &l);
1253		if (!file)
1254			break;
1255	}
1256	return file;
1257}
1258
1259static void *
1260s_next(struct seq_file *m, void *v, loff_t *pos)
1261{
1262	struct trace_event_file *file = v;
1263	struct trace_array *tr = m->private;
1264
1265	(*pos)++;
1266
1267	list_for_each_entry_continue(file, &tr->events, list) {
1268		if (file->flags & EVENT_FILE_FL_ENABLED)
1269			return file;
1270	}
1271
1272	return NULL;
1273}
1274
1275static void *s_start(struct seq_file *m, loff_t *pos)
1276{
1277	struct trace_event_file *file;
1278	struct trace_array *tr = m->private;
1279	loff_t l;
1280
1281	mutex_lock(&event_mutex);
1282
1283	file = list_entry(&tr->events, struct trace_event_file, list);
1284	for (l = 0; l <= *pos; ) {
1285		file = s_next(m, file, &l);
1286		if (!file)
1287			break;
1288	}
1289	return file;
1290}
1291
1292static int t_show(struct seq_file *m, void *v)
1293{
1294	struct trace_event_file *file = v;
1295	struct trace_event_call *call = file->event_call;
1296
1297	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298		seq_printf(m, "%s:", call->class->system);
1299	seq_printf(m, "%s\n", trace_event_name(call));
1300
1301	return 0;
1302}
1303
1304static void t_stop(struct seq_file *m, void *p)
1305{
1306	mutex_unlock(&event_mutex);
1307}
1308
1309static void *
1310__next(struct seq_file *m, void *v, loff_t *pos, int type)
1311{
1312	struct trace_array *tr = m->private;
1313	struct trace_pid_list *pid_list;
1314
1315	if (type == TRACE_PIDS)
1316		pid_list = rcu_dereference_sched(tr->filtered_pids);
1317	else
1318		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320	return trace_pid_next(pid_list, v, pos);
1321}
1322
1323static void *
1324p_next(struct seq_file *m, void *v, loff_t *pos)
1325{
1326	return __next(m, v, pos, TRACE_PIDS);
1327}
1328
1329static void *
1330np_next(struct seq_file *m, void *v, loff_t *pos)
1331{
1332	return __next(m, v, pos, TRACE_NO_PIDS);
1333}
1334
1335static void *__start(struct seq_file *m, loff_t *pos, int type)
1336	__acquires(RCU)
1337{
1338	struct trace_pid_list *pid_list;
1339	struct trace_array *tr = m->private;
1340
1341	/*
1342	 * Grab the mutex, to keep calls to p_next() having the same
1343	 * tr->filtered_pids as p_start() has.
1344	 * If we just passed the tr->filtered_pids around, then RCU would
1345	 * have been enough, but doing that makes things more complex.
1346	 */
1347	mutex_lock(&event_mutex);
1348	rcu_read_lock_sched();
1349
1350	if (type == TRACE_PIDS)
1351		pid_list = rcu_dereference_sched(tr->filtered_pids);
1352	else
1353		pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355	if (!pid_list)
1356		return NULL;
1357
1358	return trace_pid_start(pid_list, pos);
1359}
1360
1361static void *p_start(struct seq_file *m, loff_t *pos)
1362	__acquires(RCU)
1363{
1364	return __start(m, pos, TRACE_PIDS);
1365}
1366
1367static void *np_start(struct seq_file *m, loff_t *pos)
1368	__acquires(RCU)
1369{
1370	return __start(m, pos, TRACE_NO_PIDS);
1371}
1372
1373static void p_stop(struct seq_file *m, void *p)
1374	__releases(RCU)
1375{
1376	rcu_read_unlock_sched();
1377	mutex_unlock(&event_mutex);
1378}
1379
1380static ssize_t
1381event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382		  loff_t *ppos)
1383{
1384	struct trace_event_file *file;
1385	unsigned long flags;
1386	char buf[4] = "0";
1387
1388	mutex_lock(&event_mutex);
1389	file = event_file_data(filp);
1390	if (likely(file))
1391		flags = file->flags;
1392	mutex_unlock(&event_mutex);
1393
1394	if (!file || flags & EVENT_FILE_FL_FREED)
1395		return -ENODEV;
1396
1397	if (flags & EVENT_FILE_FL_ENABLED &&
1398	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399		strcpy(buf, "1");
1400
1401	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402	    flags & EVENT_FILE_FL_SOFT_MODE)
1403		strcat(buf, "*");
1404
1405	strcat(buf, "\n");
1406
1407	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408}
1409
1410static ssize_t
1411event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412		   loff_t *ppos)
1413{
1414	struct trace_event_file *file;
1415	unsigned long val;
1416	int ret;
1417
1418	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419	if (ret)
1420		return ret;
1421
 
 
 
 
1422	switch (val) {
1423	case 0:
1424	case 1:
1425		ret = -ENODEV;
1426		mutex_lock(&event_mutex);
1427		file = event_file_data(filp);
1428		if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429			ret = tracing_update_buffers(file->tr);
1430			if (ret < 0) {
1431				mutex_unlock(&event_mutex);
1432				return ret;
1433			}
1434			ret = ftrace_event_enable_disable(file, val);
1435		}
1436		mutex_unlock(&event_mutex);
1437		break;
1438
1439	default:
1440		return -EINVAL;
1441	}
1442
1443	*ppos += cnt;
1444
1445	return ret ? ret : cnt;
1446}
1447
1448static ssize_t
1449system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450		   loff_t *ppos)
1451{
1452	const char set_to_char[4] = { '?', '0', '1', 'X' };
1453	struct trace_subsystem_dir *dir = filp->private_data;
1454	struct event_subsystem *system = dir->subsystem;
1455	struct trace_event_call *call;
1456	struct trace_event_file *file;
1457	struct trace_array *tr = dir->tr;
1458	char buf[2];
1459	int set = 0;
1460	int ret;
1461
1462	mutex_lock(&event_mutex);
1463	list_for_each_entry(file, &tr->events, list) {
1464		call = file->event_call;
1465		if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466		    !trace_event_name(call) || !call->class || !call->class->reg)
1467			continue;
1468
1469		if (system && strcmp(call->class->system, system->name) != 0)
1470			continue;
1471
1472		/*
1473		 * We need to find out if all the events are set
1474		 * or if all events or cleared, or if we have
1475		 * a mixture.
1476		 */
1477		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479		/*
1480		 * If we have a mixture, no need to look further.
1481		 */
1482		if (set == 3)
1483			break;
1484	}
1485	mutex_unlock(&event_mutex);
1486
1487	buf[0] = set_to_char[set];
1488	buf[1] = '\n';
1489
1490	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492	return ret;
1493}
1494
1495static ssize_t
1496system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497		    loff_t *ppos)
1498{
1499	struct trace_subsystem_dir *dir = filp->private_data;
1500	struct event_subsystem *system = dir->subsystem;
1501	const char *name = NULL;
1502	unsigned long val;
1503	ssize_t ret;
1504
1505	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506	if (ret)
1507		return ret;
1508
1509	ret = tracing_update_buffers(dir->tr);
1510	if (ret < 0)
1511		return ret;
1512
1513	if (val != 0 && val != 1)
1514		return -EINVAL;
1515
1516	/*
1517	 * Opening of "enable" adds a ref count to system,
1518	 * so the name is safe to use.
1519	 */
1520	if (system)
1521		name = system->name;
1522
1523	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524	if (ret)
1525		goto out;
1526
1527	ret = cnt;
1528
1529out:
1530	*ppos += cnt;
1531
1532	return ret;
1533}
1534
1535enum {
1536	FORMAT_HEADER		= 1,
1537	FORMAT_FIELD_SEPERATOR	= 2,
1538	FORMAT_PRINTFMT		= 3,
1539};
1540
1541static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542{
1543	struct trace_event_call *call = event_file_data(m->private);
1544	struct list_head *common_head = &ftrace_common_fields;
1545	struct list_head *head = trace_get_fields(call);
1546	struct list_head *node = v;
1547
1548	(*pos)++;
1549
1550	switch ((unsigned long)v) {
1551	case FORMAT_HEADER:
1552		node = common_head;
1553		break;
1554
1555	case FORMAT_FIELD_SEPERATOR:
1556		node = head;
1557		break;
1558
1559	case FORMAT_PRINTFMT:
1560		/* all done */
1561		return NULL;
1562	}
1563
1564	node = node->prev;
1565	if (node == common_head)
1566		return (void *)FORMAT_FIELD_SEPERATOR;
1567	else if (node == head)
1568		return (void *)FORMAT_PRINTFMT;
1569	else
1570		return node;
1571}
1572
1573static int f_show(struct seq_file *m, void *v)
1574{
1575	struct trace_event_call *call = event_file_data(m->private);
1576	struct ftrace_event_field *field;
1577	const char *array_descriptor;
1578
1579	switch ((unsigned long)v) {
1580	case FORMAT_HEADER:
1581		seq_printf(m, "name: %s\n", trace_event_name(call));
1582		seq_printf(m, "ID: %d\n", call->event.type);
1583		seq_puts(m, "format:\n");
1584		return 0;
1585
1586	case FORMAT_FIELD_SEPERATOR:
1587		seq_putc(m, '\n');
1588		return 0;
1589
1590	case FORMAT_PRINTFMT:
1591		seq_printf(m, "\nprint fmt: %s\n",
1592			   call->print_fmt);
1593		return 0;
1594	}
1595
1596	field = list_entry(v, struct ftrace_event_field, link);
1597	/*
1598	 * Smartly shows the array type(except dynamic array).
1599	 * Normal:
1600	 *	field:TYPE VAR
1601	 * If TYPE := TYPE[LEN], it is shown:
1602	 *	field:TYPE VAR[LEN]
1603	 */
1604	array_descriptor = strchr(field->type, '[');
1605
1606	if (str_has_prefix(field->type, "__data_loc"))
1607		array_descriptor = NULL;
1608
1609	if (!array_descriptor)
1610		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611			   field->type, field->name, field->offset,
1612			   field->size, !!field->is_signed);
1613	else if (field->len)
1614		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615			   (int)(array_descriptor - field->type),
1616			   field->type, field->name,
1617			   field->len, field->offset,
1618			   field->size, !!field->is_signed);
1619	else
1620		seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621				(int)(array_descriptor - field->type),
1622				field->type, field->name,
1623				field->offset, field->size, !!field->is_signed);
1624
1625	return 0;
1626}
1627
1628static void *f_start(struct seq_file *m, loff_t *pos)
1629{
1630	void *p = (void *)FORMAT_HEADER;
1631	loff_t l = 0;
1632
1633	/* ->stop() is called even if ->start() fails */
1634	mutex_lock(&event_mutex);
1635	if (!event_file_data(m->private))
1636		return ERR_PTR(-ENODEV);
1637
1638	while (l < *pos && p)
1639		p = f_next(m, p, &l);
1640
1641	return p;
1642}
1643
1644static void f_stop(struct seq_file *m, void *p)
1645{
1646	mutex_unlock(&event_mutex);
1647}
1648
1649static const struct seq_operations trace_format_seq_ops = {
1650	.start		= f_start,
1651	.next		= f_next,
1652	.stop		= f_stop,
1653	.show		= f_show,
1654};
1655
1656static int trace_format_open(struct inode *inode, struct file *file)
1657{
1658	struct seq_file *m;
1659	int ret;
1660
1661	/* Do we want to hide event format files on tracefs lockdown? */
1662
1663	ret = seq_open(file, &trace_format_seq_ops);
1664	if (ret < 0)
1665		return ret;
1666
1667	m = file->private_data;
1668	m->private = file;
1669
1670	return 0;
1671}
1672
1673static ssize_t
1674event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1675{
1676	int id = (long)event_file_data(filp);
1677	char buf[32];
1678	int len;
1679
1680	if (unlikely(!id))
1681		return -ENODEV;
1682
1683	len = sprintf(buf, "%d\n", id);
1684
1685	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1686}
1687
1688static ssize_t
1689event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1690		  loff_t *ppos)
1691{
1692	struct trace_event_file *file;
1693	struct trace_seq *s;
1694	int r = -ENODEV;
1695
1696	if (*ppos)
1697		return 0;
1698
1699	s = kmalloc(sizeof(*s), GFP_KERNEL);
1700
1701	if (!s)
1702		return -ENOMEM;
1703
1704	trace_seq_init(s);
1705
1706	mutex_lock(&event_mutex);
1707	file = event_file_data(filp);
1708	if (file && !(file->flags & EVENT_FILE_FL_FREED))
1709		print_event_filter(file, s);
1710	mutex_unlock(&event_mutex);
1711
1712	if (file)
1713		r = simple_read_from_buffer(ubuf, cnt, ppos,
1714					    s->buffer, trace_seq_used(s));
1715
1716	kfree(s);
1717
1718	return r;
1719}
1720
1721static ssize_t
1722event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1723		   loff_t *ppos)
1724{
1725	struct trace_event_file *file;
1726	char *buf;
1727	int err = -ENODEV;
1728
1729	if (cnt >= PAGE_SIZE)
1730		return -EINVAL;
1731
1732	buf = memdup_user_nul(ubuf, cnt);
1733	if (IS_ERR(buf))
1734		return PTR_ERR(buf);
1735
1736	mutex_lock(&event_mutex);
1737	file = event_file_data(filp);
1738	if (file)
1739		err = apply_event_filter(file, buf);
1740	mutex_unlock(&event_mutex);
1741
1742	kfree(buf);
1743	if (err < 0)
1744		return err;
1745
1746	*ppos += cnt;
1747
1748	return cnt;
1749}
1750
1751static LIST_HEAD(event_subsystems);
1752
1753static int subsystem_open(struct inode *inode, struct file *filp)
1754{
1755	struct trace_subsystem_dir *dir = NULL, *iter_dir;
1756	struct trace_array *tr = NULL, *iter_tr;
1757	struct event_subsystem *system = NULL;
 
 
1758	int ret;
1759
1760	if (tracing_is_disabled())
1761		return -ENODEV;
1762
1763	/* Make sure the system still exists */
1764	mutex_lock(&event_mutex);
1765	mutex_lock(&trace_types_lock);
1766	list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1767		list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1768			if (iter_dir == inode->i_private) {
1769				/* Don't open systems with no events */
1770				tr = iter_tr;
1771				dir = iter_dir;
1772				if (dir->nr_events) {
1773					__get_system_dir(dir);
1774					system = dir->subsystem;
1775				}
1776				goto exit_loop;
1777			}
1778		}
1779	}
1780 exit_loop:
1781	mutex_unlock(&trace_types_lock);
1782	mutex_unlock(&event_mutex);
1783
1784	if (!system)
1785		return -ENODEV;
1786
 
 
 
1787	/* Still need to increment the ref count of the system */
1788	if (trace_array_get(tr) < 0) {
1789		put_system(dir);
1790		return -ENODEV;
1791	}
1792
1793	ret = tracing_open_generic(inode, filp);
1794	if (ret < 0) {
1795		trace_array_put(tr);
1796		put_system(dir);
1797	}
1798
1799	return ret;
1800}
1801
1802static int system_tr_open(struct inode *inode, struct file *filp)
1803{
1804	struct trace_subsystem_dir *dir;
1805	struct trace_array *tr = inode->i_private;
1806	int ret;
1807
1808	/* Make a temporary dir that has no system but points to tr */
1809	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1810	if (!dir)
1811		return -ENOMEM;
1812
1813	ret = tracing_open_generic_tr(inode, filp);
1814	if (ret < 0) {
1815		kfree(dir);
1816		return ret;
1817	}
1818	dir->tr = tr;
1819	filp->private_data = dir;
1820
1821	return 0;
1822}
1823
1824static int subsystem_release(struct inode *inode, struct file *file)
1825{
1826	struct trace_subsystem_dir *dir = file->private_data;
1827
1828	trace_array_put(dir->tr);
1829
1830	/*
1831	 * If dir->subsystem is NULL, then this is a temporary
1832	 * descriptor that was made for a trace_array to enable
1833	 * all subsystems.
1834	 */
1835	if (dir->subsystem)
1836		put_system(dir);
1837	else
1838		kfree(dir);
1839
1840	return 0;
1841}
1842
1843static ssize_t
1844subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1845		      loff_t *ppos)
1846{
1847	struct trace_subsystem_dir *dir = filp->private_data;
1848	struct event_subsystem *system = dir->subsystem;
1849	struct trace_seq *s;
1850	int r;
1851
1852	if (*ppos)
1853		return 0;
1854
1855	s = kmalloc(sizeof(*s), GFP_KERNEL);
1856	if (!s)
1857		return -ENOMEM;
1858
1859	trace_seq_init(s);
1860
1861	print_subsystem_event_filter(system, s);
1862	r = simple_read_from_buffer(ubuf, cnt, ppos,
1863				    s->buffer, trace_seq_used(s));
1864
1865	kfree(s);
1866
1867	return r;
1868}
1869
1870static ssize_t
1871subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1872		       loff_t *ppos)
1873{
1874	struct trace_subsystem_dir *dir = filp->private_data;
1875	char *buf;
1876	int err;
1877
1878	if (cnt >= PAGE_SIZE)
1879		return -EINVAL;
1880
1881	buf = memdup_user_nul(ubuf, cnt);
1882	if (IS_ERR(buf))
1883		return PTR_ERR(buf);
1884
1885	err = apply_subsystem_event_filter(dir, buf);
1886	kfree(buf);
1887	if (err < 0)
1888		return err;
1889
1890	*ppos += cnt;
1891
1892	return cnt;
1893}
1894
1895static ssize_t
1896show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1897{
1898	struct trace_array *tr = filp->private_data;
1899	struct trace_seq *s;
1900	int r;
1901
1902	if (*ppos)
1903		return 0;
1904
1905	s = kmalloc(sizeof(*s), GFP_KERNEL);
1906	if (!s)
1907		return -ENOMEM;
1908
1909	trace_seq_init(s);
1910
1911	ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1912	r = simple_read_from_buffer(ubuf, cnt, ppos,
1913				    s->buffer, trace_seq_used(s));
1914
1915	kfree(s);
1916
1917	return r;
1918}
1919
1920static ssize_t
1921show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1922{
1923	struct trace_seq *s;
1924	int r;
1925
1926	if (*ppos)
1927		return 0;
1928
1929	s = kmalloc(sizeof(*s), GFP_KERNEL);
1930	if (!s)
1931		return -ENOMEM;
1932
1933	trace_seq_init(s);
1934
1935	ring_buffer_print_entry_header(s);
1936	r = simple_read_from_buffer(ubuf, cnt, ppos,
1937				    s->buffer, trace_seq_used(s));
1938
1939	kfree(s);
1940
1941	return r;
1942}
1943
1944static void ignore_task_cpu(void *data)
1945{
1946	struct trace_array *tr = data;
1947	struct trace_pid_list *pid_list;
1948	struct trace_pid_list *no_pid_list;
1949
1950	/*
1951	 * This function is called by on_each_cpu() while the
1952	 * event_mutex is held.
1953	 */
1954	pid_list = rcu_dereference_protected(tr->filtered_pids,
1955					     mutex_is_locked(&event_mutex));
1956	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1957					     mutex_is_locked(&event_mutex));
1958
1959	this_cpu_write(tr->array_buffer.data->ignore_pid,
1960		       trace_ignore_this_task(pid_list, no_pid_list, current));
1961}
1962
1963static void register_pid_events(struct trace_array *tr)
1964{
1965	/*
1966	 * Register a probe that is called before all other probes
1967	 * to set ignore_pid if next or prev do not match.
1968	 * Register a probe this is called after all other probes
1969	 * to only keep ignore_pid set if next pid matches.
1970	 */
1971	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1972					 tr, INT_MAX);
1973	register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1974					 tr, 0);
1975
1976	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1977					 tr, INT_MAX);
1978	register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1979					 tr, 0);
1980
1981	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1982					     tr, INT_MAX);
1983	register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1984					     tr, 0);
1985
1986	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1987					 tr, INT_MAX);
1988	register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1989					 tr, 0);
1990}
1991
1992static ssize_t
1993event_pid_write(struct file *filp, const char __user *ubuf,
1994		size_t cnt, loff_t *ppos, int type)
1995{
1996	struct seq_file *m = filp->private_data;
1997	struct trace_array *tr = m->private;
1998	struct trace_pid_list *filtered_pids = NULL;
1999	struct trace_pid_list *other_pids = NULL;
2000	struct trace_pid_list *pid_list;
2001	struct trace_event_file *file;
2002	ssize_t ret;
2003
2004	if (!cnt)
2005		return 0;
2006
2007	ret = tracing_update_buffers(tr);
2008	if (ret < 0)
2009		return ret;
2010
2011	mutex_lock(&event_mutex);
2012
2013	if (type == TRACE_PIDS) {
2014		filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2015							  lockdep_is_held(&event_mutex));
2016		other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2017							  lockdep_is_held(&event_mutex));
2018	} else {
2019		filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2020							  lockdep_is_held(&event_mutex));
2021		other_pids = rcu_dereference_protected(tr->filtered_pids,
2022							  lockdep_is_held(&event_mutex));
2023	}
2024
2025	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2026	if (ret < 0)
2027		goto out;
2028
2029	if (type == TRACE_PIDS)
2030		rcu_assign_pointer(tr->filtered_pids, pid_list);
2031	else
2032		rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2033
2034	list_for_each_entry(file, &tr->events, list) {
2035		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2036	}
2037
2038	if (filtered_pids) {
2039		tracepoint_synchronize_unregister();
2040		trace_pid_list_free(filtered_pids);
2041	} else if (pid_list && !other_pids) {
2042		register_pid_events(tr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2043	}
2044
2045	/*
2046	 * Ignoring of pids is done at task switch. But we have to
2047	 * check for those tasks that are currently running.
2048	 * Always do this in case a pid was appended or removed.
2049	 */
2050	on_each_cpu(ignore_task_cpu, tr, 1);
2051
2052 out:
2053	mutex_unlock(&event_mutex);
2054
2055	if (ret > 0)
2056		*ppos += ret;
2057
2058	return ret;
2059}
2060
2061static ssize_t
2062ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2063		       size_t cnt, loff_t *ppos)
2064{
2065	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2066}
2067
2068static ssize_t
2069ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2070			size_t cnt, loff_t *ppos)
2071{
2072	return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2073}
2074
2075static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2076static int ftrace_event_set_open(struct inode *inode, struct file *file);
2077static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2078static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2079static int ftrace_event_release(struct inode *inode, struct file *file);
2080
2081static const struct seq_operations show_event_seq_ops = {
2082	.start = t_start,
2083	.next = t_next,
2084	.show = t_show,
2085	.stop = t_stop,
2086};
2087
2088static const struct seq_operations show_set_event_seq_ops = {
2089	.start = s_start,
2090	.next = s_next,
2091	.show = t_show,
2092	.stop = t_stop,
2093};
2094
2095static const struct seq_operations show_set_pid_seq_ops = {
2096	.start = p_start,
2097	.next = p_next,
2098	.show = trace_pid_show,
2099	.stop = p_stop,
2100};
2101
2102static const struct seq_operations show_set_no_pid_seq_ops = {
2103	.start = np_start,
2104	.next = np_next,
2105	.show = trace_pid_show,
2106	.stop = p_stop,
2107};
2108
2109static const struct file_operations ftrace_avail_fops = {
2110	.open = ftrace_event_avail_open,
2111	.read = seq_read,
2112	.llseek = seq_lseek,
2113	.release = seq_release,
2114};
2115
2116static const struct file_operations ftrace_set_event_fops = {
2117	.open = ftrace_event_set_open,
2118	.read = seq_read,
2119	.write = ftrace_event_write,
2120	.llseek = seq_lseek,
2121	.release = ftrace_event_release,
2122};
2123
2124static const struct file_operations ftrace_set_event_pid_fops = {
2125	.open = ftrace_event_set_pid_open,
2126	.read = seq_read,
2127	.write = ftrace_event_pid_write,
2128	.llseek = seq_lseek,
2129	.release = ftrace_event_release,
2130};
2131
2132static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2133	.open = ftrace_event_set_npid_open,
2134	.read = seq_read,
2135	.write = ftrace_event_npid_write,
2136	.llseek = seq_lseek,
2137	.release = ftrace_event_release,
2138};
2139
2140static const struct file_operations ftrace_enable_fops = {
2141	.open = tracing_open_file_tr,
2142	.read = event_enable_read,
2143	.write = event_enable_write,
2144	.release = tracing_release_file_tr,
2145	.llseek = default_llseek,
2146};
2147
2148static const struct file_operations ftrace_event_format_fops = {
2149	.open = trace_format_open,
2150	.read = seq_read,
2151	.llseek = seq_lseek,
2152	.release = seq_release,
2153};
2154
2155static const struct file_operations ftrace_event_id_fops = {
2156	.read = event_id_read,
2157	.llseek = default_llseek,
2158};
2159
2160static const struct file_operations ftrace_event_filter_fops = {
2161	.open = tracing_open_file_tr,
2162	.read = event_filter_read,
2163	.write = event_filter_write,
2164	.release = tracing_release_file_tr,
2165	.llseek = default_llseek,
2166};
2167
2168static const struct file_operations ftrace_subsystem_filter_fops = {
2169	.open = subsystem_open,
2170	.read = subsystem_filter_read,
2171	.write = subsystem_filter_write,
2172	.llseek = default_llseek,
2173	.release = subsystem_release,
2174};
2175
2176static const struct file_operations ftrace_system_enable_fops = {
2177	.open = subsystem_open,
2178	.read = system_enable_read,
2179	.write = system_enable_write,
2180	.llseek = default_llseek,
2181	.release = subsystem_release,
2182};
2183
2184static const struct file_operations ftrace_tr_enable_fops = {
2185	.open = system_tr_open,
2186	.read = system_enable_read,
2187	.write = system_enable_write,
2188	.llseek = default_llseek,
2189	.release = subsystem_release,
2190};
2191
2192static const struct file_operations ftrace_show_header_page_fops = {
2193	.open = tracing_open_generic_tr,
2194	.read = show_header_page_file,
2195	.llseek = default_llseek,
2196	.release = tracing_release_generic_tr,
2197};
2198
2199static const struct file_operations ftrace_show_header_event_fops = {
2200	.open = tracing_open_generic_tr,
2201	.read = show_header_event_file,
2202	.llseek = default_llseek,
2203	.release = tracing_release_generic_tr,
2204};
2205
2206static int
2207ftrace_event_open(struct inode *inode, struct file *file,
2208		  const struct seq_operations *seq_ops)
2209{
2210	struct seq_file *m;
2211	int ret;
2212
2213	ret = security_locked_down(LOCKDOWN_TRACEFS);
2214	if (ret)
2215		return ret;
2216
2217	ret = seq_open(file, seq_ops);
2218	if (ret < 0)
2219		return ret;
2220	m = file->private_data;
2221	/* copy tr over to seq ops */
2222	m->private = inode->i_private;
2223
2224	return ret;
2225}
2226
2227static int ftrace_event_release(struct inode *inode, struct file *file)
2228{
2229	struct trace_array *tr = inode->i_private;
2230
2231	trace_array_put(tr);
2232
2233	return seq_release(inode, file);
2234}
2235
2236static int
2237ftrace_event_avail_open(struct inode *inode, struct file *file)
2238{
2239	const struct seq_operations *seq_ops = &show_event_seq_ops;
2240
2241	/* Checks for tracefs lockdown */
2242	return ftrace_event_open(inode, file, seq_ops);
2243}
2244
2245static int
2246ftrace_event_set_open(struct inode *inode, struct file *file)
2247{
2248	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2249	struct trace_array *tr = inode->i_private;
2250	int ret;
2251
2252	ret = tracing_check_open_get_tr(tr);
2253	if (ret)
2254		return ret;
2255
2256	if ((file->f_mode & FMODE_WRITE) &&
2257	    (file->f_flags & O_TRUNC))
2258		ftrace_clear_events(tr);
2259
2260	ret = ftrace_event_open(inode, file, seq_ops);
2261	if (ret < 0)
2262		trace_array_put(tr);
2263	return ret;
2264}
2265
2266static int
2267ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2268{
2269	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2270	struct trace_array *tr = inode->i_private;
2271	int ret;
2272
2273	ret = tracing_check_open_get_tr(tr);
2274	if (ret)
2275		return ret;
2276
2277	if ((file->f_mode & FMODE_WRITE) &&
2278	    (file->f_flags & O_TRUNC))
2279		ftrace_clear_event_pids(tr, TRACE_PIDS);
2280
2281	ret = ftrace_event_open(inode, file, seq_ops);
2282	if (ret < 0)
2283		trace_array_put(tr);
2284	return ret;
2285}
2286
2287static int
2288ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2289{
2290	const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2291	struct trace_array *tr = inode->i_private;
2292	int ret;
2293
2294	ret = tracing_check_open_get_tr(tr);
2295	if (ret)
2296		return ret;
2297
2298	if ((file->f_mode & FMODE_WRITE) &&
2299	    (file->f_flags & O_TRUNC))
2300		ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2301
2302	ret = ftrace_event_open(inode, file, seq_ops);
2303	if (ret < 0)
2304		trace_array_put(tr);
2305	return ret;
2306}
2307
2308static struct event_subsystem *
2309create_new_subsystem(const char *name)
2310{
2311	struct event_subsystem *system;
2312
2313	/* need to create new entry */
2314	system = kmalloc(sizeof(*system), GFP_KERNEL);
2315	if (!system)
2316		return NULL;
2317
2318	system->ref_count = 1;
2319
2320	/* Only allocate if dynamic (kprobes and modules) */
2321	system->name = kstrdup_const(name, GFP_KERNEL);
2322	if (!system->name)
2323		goto out_free;
2324
 
 
2325	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2326	if (!system->filter)
2327		goto out_free;
2328
2329	list_add(&system->list, &event_subsystems);
2330
2331	return system;
2332
2333 out_free:
2334	kfree_const(system->name);
2335	kfree(system);
2336	return NULL;
2337}
2338
2339static int system_callback(const char *name, umode_t *mode, void **data,
2340		    const struct file_operations **fops)
2341{
2342	if (strcmp(name, "filter") == 0)
2343		*fops = &ftrace_subsystem_filter_fops;
2344
2345	else if (strcmp(name, "enable") == 0)
2346		*fops = &ftrace_system_enable_fops;
2347
2348	else
2349		return 0;
2350
2351	*mode = TRACE_MODE_WRITE;
2352	return 1;
2353}
2354
2355static struct eventfs_inode *
2356event_subsystem_dir(struct trace_array *tr, const char *name,
2357		    struct trace_event_file *file, struct eventfs_inode *parent)
2358{
2359	struct event_subsystem *system, *iter;
2360	struct trace_subsystem_dir *dir;
2361	struct eventfs_inode *ei;
2362	int nr_entries;
2363	static struct eventfs_entry system_entries[] = {
2364		{
2365			.name		= "filter",
2366			.callback	= system_callback,
2367		},
2368		{
2369			.name		= "enable",
2370			.callback	= system_callback,
2371		}
2372	};
2373
2374	/* First see if we did not already create this dir */
2375	list_for_each_entry(dir, &tr->systems, list) {
2376		system = dir->subsystem;
2377		if (strcmp(system->name, name) == 0) {
2378			dir->nr_events++;
2379			file->system = dir;
2380			return dir->ei;
2381		}
2382	}
2383
2384	/* Now see if the system itself exists. */
2385	system = NULL;
2386	list_for_each_entry(iter, &event_subsystems, list) {
2387		if (strcmp(iter->name, name) == 0) {
2388			system = iter;
2389			break;
2390		}
2391	}
 
 
 
2392
2393	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2394	if (!dir)
2395		goto out_fail;
2396
2397	if (!system) {
2398		system = create_new_subsystem(name);
2399		if (!system)
2400			goto out_free;
2401	} else
2402		__get_system(system);
2403
2404	/* ftrace only has directories no files */
2405	if (strcmp(name, "ftrace") == 0)
2406		nr_entries = 0;
2407	else
2408		nr_entries = ARRAY_SIZE(system_entries);
2409
2410	ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2411	if (IS_ERR(ei)) {
2412		pr_warn("Failed to create system directory %s\n", name);
2413		__put_system(system);
2414		goto out_free;
2415	}
2416
2417	dir->ei = ei;
2418	dir->tr = tr;
2419	dir->ref_count = 1;
2420	dir->nr_events = 1;
2421	dir->subsystem = system;
2422	file->system = dir;
2423
 
 
 
 
 
 
 
 
 
 
 
2424	list_add(&dir->list, &tr->systems);
2425
2426	return dir->ei;
2427
2428 out_free:
2429	kfree(dir);
2430 out_fail:
2431	/* Only print this message if failed on memory allocation */
2432	if (!dir || !system)
2433		pr_warn("No memory to create event subsystem %s\n", name);
2434	return NULL;
2435}
2436
2437static int
2438event_define_fields(struct trace_event_call *call)
2439{
 
 
2440	struct list_head *head;
2441	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2442
2443	/*
2444	 * Other events may have the same class. Only update
2445	 * the fields if they are not already defined.
2446	 */
2447	head = trace_get_fields(call);
2448	if (list_empty(head)) {
2449		struct trace_event_fields *field = call->class->fields_array;
2450		unsigned int offset = sizeof(struct trace_entry);
2451
2452		for (; field->type; field++) {
2453			if (field->type == TRACE_FUNCTION_TYPE) {
2454				field->define_fields(call);
2455				break;
2456			}
2457
2458			offset = ALIGN(offset, field->align);
2459			ret = trace_define_field_ext(call, field->type, field->name,
2460						 offset, field->size,
2461						 field->is_signed, field->filter_type,
2462						 field->len);
2463			if (WARN_ON_ONCE(ret)) {
2464				pr_err("error code is %d\n", ret);
2465				break;
2466			}
2467
2468			offset += field->size;
2469		}
2470	}
2471
2472	return ret;
2473}
2474
2475static int event_callback(const char *name, umode_t *mode, void **data,
2476			  const struct file_operations **fops)
2477{
2478	struct trace_event_file *file = *data;
2479	struct trace_event_call *call = file->event_call;
2480
2481	if (strcmp(name, "format") == 0) {
2482		*mode = TRACE_MODE_READ;
2483		*fops = &ftrace_event_format_fops;
2484		*data = call;
2485		return 1;
2486	}
2487
2488	/*
2489	 * Only event directories that can be enabled should have
2490	 * triggers or filters, with the exception of the "print"
2491	 * event that can have a "trigger" file.
2492	 */
2493	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2494		if (call->class->reg && strcmp(name, "enable") == 0) {
2495			*mode = TRACE_MODE_WRITE;
2496			*fops = &ftrace_enable_fops;
2497			return 1;
2498		}
2499
2500		if (strcmp(name, "filter") == 0) {
2501			*mode = TRACE_MODE_WRITE;
2502			*fops = &ftrace_event_filter_fops;
2503			return 1;
2504		}
2505	}
2506
2507	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2508	    strcmp(trace_event_name(call), "print") == 0) {
2509		if (strcmp(name, "trigger") == 0) {
2510			*mode = TRACE_MODE_WRITE;
2511			*fops = &event_trigger_fops;
2512			return 1;
2513		}
2514	}
2515
2516#ifdef CONFIG_PERF_EVENTS
2517	if (call->event.type && call->class->reg &&
2518	    strcmp(name, "id") == 0) {
2519		*mode = TRACE_MODE_READ;
2520		*data = (void *)(long)call->event.type;
2521		*fops = &ftrace_event_id_fops;
2522		return 1;
2523	}
2524#endif
2525
2526#ifdef CONFIG_HIST_TRIGGERS
2527	if (strcmp(name, "hist") == 0) {
2528		*mode = TRACE_MODE_READ;
2529		*fops = &event_hist_fops;
2530		return 1;
2531	}
2532#endif
2533#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2534	if (strcmp(name, "hist_debug") == 0) {
2535		*mode = TRACE_MODE_READ;
2536		*fops = &event_hist_debug_fops;
2537		return 1;
2538	}
2539#endif
2540#ifdef CONFIG_TRACE_EVENT_INJECT
2541	if (call->event.type && call->class->reg &&
2542	    strcmp(name, "inject") == 0) {
2543		*mode = 0200;
2544		*fops = &event_inject_fops;
2545		return 1;
2546	}
2547#endif
2548	return 0;
2549}
2550
2551static int
2552event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2553{
2554	struct trace_event_call *call = file->event_call;
2555	struct trace_array *tr = file->tr;
2556	struct eventfs_inode *e_events;
2557	struct eventfs_inode *ei;
2558	const char *name;
2559	int nr_entries;
2560	int ret;
2561	static struct eventfs_entry event_entries[] = {
2562		{
2563			.name		= "enable",
2564			.callback	= event_callback,
2565		},
2566		{
2567			.name		= "filter",
2568			.callback	= event_callback,
2569		},
2570		{
2571			.name		= "trigger",
2572			.callback	= event_callback,
2573		},
2574		{
2575			.name		= "format",
2576			.callback	= event_callback,
2577		},
2578#ifdef CONFIG_PERF_EVENTS
2579		{
2580			.name		= "id",
2581			.callback	= event_callback,
2582		},
2583#endif
2584#ifdef CONFIG_HIST_TRIGGERS
2585		{
2586			.name		= "hist",
2587			.callback	= event_callback,
2588		},
2589#endif
2590#ifdef CONFIG_HIST_TRIGGERS_DEBUG
2591		{
2592			.name		= "hist_debug",
2593			.callback	= event_callback,
2594		},
2595#endif
2596#ifdef CONFIG_TRACE_EVENT_INJECT
2597		{
2598			.name		= "inject",
2599			.callback	= event_callback,
2600		},
2601#endif
2602	};
2603
2604	/*
2605	 * If the trace point header did not define TRACE_SYSTEM
2606	 * then the system would be called "TRACE_SYSTEM". This should
2607	 * never happen.
2608	 */
2609	if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2610		return -ENODEV;
2611
2612	e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2613	if (!e_events)
2614		return -ENOMEM;
2615
2616	nr_entries = ARRAY_SIZE(event_entries);
2617
2618	name = trace_event_name(call);
2619	ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2620	if (IS_ERR(ei)) {
2621		pr_warn("Could not create tracefs '%s' directory\n", name);
2622		return -1;
2623	}
2624
2625	file->ei = ei;
2626
2627	ret = event_define_fields(call);
2628	if (ret < 0) {
2629		pr_warn("Could not initialize trace point events/%s\n", name);
2630		return ret;
2631	}
2632
2633	return 0;
2634}
2635
2636static void remove_event_from_tracers(struct trace_event_call *call)
2637{
2638	struct trace_event_file *file;
2639	struct trace_array *tr;
2640
2641	do_for_each_event_file_safe(tr, file) {
2642		if (file->event_call != call)
2643			continue;
2644
2645		remove_event_file_dir(file);
2646		/*
2647		 * The do_for_each_event_file_safe() is
2648		 * a double loop. After finding the call for this
2649		 * trace_array, we use break to jump to the next
2650		 * trace_array.
2651		 */
2652		break;
2653	} while_for_each_event_file();
2654}
2655
2656static void event_remove(struct trace_event_call *call)
2657{
2658	struct trace_array *tr;
2659	struct trace_event_file *file;
2660
2661	do_for_each_event_file(tr, file) {
2662		if (file->event_call != call)
2663			continue;
2664
2665		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2666			tr->clear_trace = true;
2667
2668		ftrace_event_enable_disable(file, 0);
2669		/*
2670		 * The do_for_each_event_file() is
2671		 * a double loop. After finding the call for this
2672		 * trace_array, we use break to jump to the next
2673		 * trace_array.
2674		 */
2675		break;
2676	} while_for_each_event_file();
2677
2678	if (call->event.funcs)
2679		__unregister_trace_event(&call->event);
2680	remove_event_from_tracers(call);
2681	list_del(&call->list);
2682}
2683
2684static int event_init(struct trace_event_call *call)
2685{
2686	int ret = 0;
2687	const char *name;
2688
2689	name = trace_event_name(call);
2690	if (WARN_ON(!name))
2691		return -EINVAL;
2692
2693	if (call->class->raw_init) {
2694		ret = call->class->raw_init(call);
2695		if (ret < 0 && ret != -ENOSYS)
2696			pr_warn("Could not initialize trace events/%s\n", name);
2697	}
2698
2699	return ret;
2700}
2701
2702static int
2703__register_event(struct trace_event_call *call, struct module *mod)
2704{
2705	int ret;
2706
2707	ret = event_init(call);
2708	if (ret < 0)
2709		return ret;
2710
2711	list_add(&call->list, &ftrace_events);
2712	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2713		atomic_set(&call->refcnt, 0);
2714	else
2715		call->module = mod;
2716
2717	return 0;
2718}
2719
2720static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2721{
2722	int rlen;
2723	int elen;
2724
2725	/* Find the length of the eval value as a string */
2726	elen = snprintf(ptr, 0, "%ld", map->eval_value);
2727	/* Make sure there's enough room to replace the string with the value */
2728	if (len < elen)
2729		return NULL;
2730
2731	snprintf(ptr, elen + 1, "%ld", map->eval_value);
2732
2733	/* Get the rest of the string of ptr */
2734	rlen = strlen(ptr + len);
2735	memmove(ptr + elen, ptr + len, rlen);
2736	/* Make sure we end the new string */
2737	ptr[elen + rlen] = 0;
2738
2739	return ptr + elen;
2740}
2741
2742static void update_event_printk(struct trace_event_call *call,
2743				struct trace_eval_map *map)
2744{
2745	char *ptr;
2746	int quote = 0;
2747	int len = strlen(map->eval_string);
2748
2749	for (ptr = call->print_fmt; *ptr; ptr++) {
2750		if (*ptr == '\\') {
2751			ptr++;
2752			/* paranoid */
2753			if (!*ptr)
2754				break;
2755			continue;
2756		}
2757		if (*ptr == '"') {
2758			quote ^= 1;
2759			continue;
2760		}
2761		if (quote)
2762			continue;
2763		if (isdigit(*ptr)) {
2764			/* skip numbers */
2765			do {
2766				ptr++;
2767				/* Check for alpha chars like ULL */
2768			} while (isalnum(*ptr));
2769			if (!*ptr)
2770				break;
2771			/*
2772			 * A number must have some kind of delimiter after
2773			 * it, and we can ignore that too.
2774			 */
2775			continue;
2776		}
2777		if (isalpha(*ptr) || *ptr == '_') {
2778			if (strncmp(map->eval_string, ptr, len) == 0 &&
2779			    !isalnum(ptr[len]) && ptr[len] != '_') {
2780				ptr = eval_replace(ptr, map, len);
2781				/* enum/sizeof string smaller than value */
2782				if (WARN_ON_ONCE(!ptr))
2783					return;
2784				/*
2785				 * No need to decrement here, as eval_replace()
2786				 * returns the pointer to the character passed
2787				 * the eval, and two evals can not be placed
2788				 * back to back without something in between.
2789				 * We can skip that something in between.
2790				 */
2791				continue;
2792			}
2793		skip_more:
2794			do {
2795				ptr++;
2796			} while (isalnum(*ptr) || *ptr == '_');
2797			if (!*ptr)
2798				break;
2799			/*
2800			 * If what comes after this variable is a '.' or
2801			 * '->' then we can continue to ignore that string.
2802			 */
2803			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2804				ptr += *ptr == '.' ? 1 : 2;
2805				if (!*ptr)
2806					break;
2807				goto skip_more;
2808			}
2809			/*
2810			 * Once again, we can skip the delimiter that came
2811			 * after the string.
2812			 */
2813			continue;
2814		}
2815	}
2816}
2817
2818static void add_str_to_module(struct module *module, char *str)
2819{
2820	struct module_string *modstr;
2821
2822	modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2823
2824	/*
2825	 * If we failed to allocate memory here, then we'll just
2826	 * let the str memory leak when the module is removed.
2827	 * If this fails to allocate, there's worse problems than
2828	 * a leaked string on module removal.
2829	 */
2830	if (WARN_ON_ONCE(!modstr))
2831		return;
2832
2833	modstr->module = module;
2834	modstr->str = str;
2835
2836	list_add(&modstr->next, &module_strings);
2837}
2838
2839static void update_event_fields(struct trace_event_call *call,
2840				struct trace_eval_map *map)
2841{
2842	struct ftrace_event_field *field;
2843	struct list_head *head;
2844	char *ptr;
2845	char *str;
2846	int len = strlen(map->eval_string);
2847
2848	/* Dynamic events should never have field maps */
2849	if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2850		return;
2851
2852	head = trace_get_fields(call);
2853	list_for_each_entry(field, head, link) {
2854		ptr = strchr(field->type, '[');
2855		if (!ptr)
2856			continue;
2857		ptr++;
2858
2859		if (!isalpha(*ptr) && *ptr != '_')
2860			continue;
2861
2862		if (strncmp(map->eval_string, ptr, len) != 0)
2863			continue;
2864
2865		str = kstrdup(field->type, GFP_KERNEL);
2866		if (WARN_ON_ONCE(!str))
2867			return;
2868		ptr = str + (ptr - field->type);
2869		ptr = eval_replace(ptr, map, len);
2870		/* enum/sizeof string smaller than value */
2871		if (WARN_ON_ONCE(!ptr)) {
2872			kfree(str);
2873			continue;
2874		}
2875
2876		/*
2877		 * If the event is part of a module, then we need to free the string
2878		 * when the module is removed. Otherwise, it will stay allocated
2879		 * until a reboot.
2880		 */
2881		if (call->module)
2882			add_str_to_module(call->module, str);
2883
2884		field->type = str;
2885	}
2886}
2887
2888void trace_event_eval_update(struct trace_eval_map **map, int len)
2889{
2890	struct trace_event_call *call, *p;
2891	const char *last_system = NULL;
2892	bool first = false;
2893	int last_i;
2894	int i;
2895
2896	down_write(&trace_event_sem);
2897	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2898		/* events are usually grouped together with systems */
2899		if (!last_system || call->class->system != last_system) {
2900			first = true;
2901			last_i = 0;
2902			last_system = call->class->system;
2903		}
2904
2905		/*
2906		 * Since calls are grouped by systems, the likelihood that the
2907		 * next call in the iteration belongs to the same system as the
2908		 * previous call is high. As an optimization, we skip searching
2909		 * for a map[] that matches the call's system if the last call
2910		 * was from the same system. That's what last_i is for. If the
2911		 * call has the same system as the previous call, then last_i
2912		 * will be the index of the first map[] that has a matching
2913		 * system.
2914		 */
2915		for (i = last_i; i < len; i++) {
2916			if (call->class->system == map[i]->system) {
2917				/* Save the first system if need be */
2918				if (first) {
2919					last_i = i;
2920					first = false;
2921				}
2922				update_event_printk(call, map[i]);
2923				update_event_fields(call, map[i]);
2924			}
2925		}
2926		cond_resched();
2927	}
2928	up_write(&trace_event_sem);
2929}
2930
2931static bool event_in_systems(struct trace_event_call *call,
2932			     const char *systems)
2933{
2934	const char *system;
2935	const char *p;
2936
2937	if (!systems)
2938		return true;
2939
2940	system = call->class->system;
2941	p = strstr(systems, system);
2942	if (!p)
2943		return false;
2944
2945	if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2946		return false;
2947
2948	p += strlen(system);
2949	return !*p || isspace(*p) || *p == ',';
2950}
2951
2952static struct trace_event_file *
2953trace_create_new_event(struct trace_event_call *call,
2954		       struct trace_array *tr)
2955{
2956	struct trace_pid_list *no_pid_list;
2957	struct trace_pid_list *pid_list;
2958	struct trace_event_file *file;
2959	unsigned int first;
2960
2961	if (!event_in_systems(call, tr->system_names))
2962		return NULL;
2963
2964	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2965	if (!file)
2966		return ERR_PTR(-ENOMEM);
2967
2968	pid_list = rcu_dereference_protected(tr->filtered_pids,
2969					     lockdep_is_held(&event_mutex));
2970	no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2971					     lockdep_is_held(&event_mutex));
2972
2973	if (!trace_pid_list_first(pid_list, &first) ||
2974	    !trace_pid_list_first(no_pid_list, &first))
2975		file->flags |= EVENT_FILE_FL_PID_FILTER;
2976
2977	file->event_call = call;
2978	file->tr = tr;
2979	atomic_set(&file->sm_ref, 0);
2980	atomic_set(&file->tm_ref, 0);
2981	INIT_LIST_HEAD(&file->triggers);
2982	list_add(&file->list, &tr->events);
2983	event_file_get(file);
2984
2985	return file;
2986}
2987
2988#define MAX_BOOT_TRIGGERS 32
2989
2990static struct boot_triggers {
2991	const char		*event;
2992	char			*trigger;
2993} bootup_triggers[MAX_BOOT_TRIGGERS];
2994
2995static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2996static int nr_boot_triggers;
2997
2998static __init int setup_trace_triggers(char *str)
2999{
3000	char *trigger;
3001	char *buf;
3002	int i;
3003
3004	strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3005	trace_set_ring_buffer_expanded(NULL);
3006	disable_tracing_selftest("running event triggers");
3007
3008	buf = bootup_trigger_buf;
3009	for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3010		trigger = strsep(&buf, ",");
3011		if (!trigger)
3012			break;
3013		bootup_triggers[i].event = strsep(&trigger, ".");
3014		bootup_triggers[i].trigger = trigger;
3015		if (!bootup_triggers[i].trigger)
3016			break;
3017	}
3018
3019	nr_boot_triggers = i;
3020	return 1;
3021}
3022__setup("trace_trigger=", setup_trace_triggers);
3023
3024/* Add an event to a trace directory */
3025static int
3026__trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3027{
3028	struct trace_event_file *file;
3029
3030	file = trace_create_new_event(call, tr);
3031	/*
3032	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3033	 * allocation, or NULL if the event is not part of the tr->system_names.
3034	 * When the event is not part of the tr->system_names, return zero, not
3035	 * an error.
3036	 */
3037	if (!file)
3038		return 0;
3039
3040	if (IS_ERR(file))
3041		return PTR_ERR(file);
3042
3043	if (eventdir_initialized)
3044		return event_create_dir(tr->event_dir, file);
3045	else
3046		return event_define_fields(call);
3047}
3048
3049static void trace_early_triggers(struct trace_event_file *file, const char *name)
3050{
3051	int ret;
3052	int i;
3053
3054	for (i = 0; i < nr_boot_triggers; i++) {
3055		if (strcmp(name, bootup_triggers[i].event))
3056			continue;
3057		mutex_lock(&event_mutex);
3058		ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3059		mutex_unlock(&event_mutex);
3060		if (ret)
3061			pr_err("Failed to register trigger '%s' on event %s\n",
3062			       bootup_triggers[i].trigger,
3063			       bootup_triggers[i].event);
3064	}
3065}
3066
3067/*
3068 * Just create a descriptor for early init. A descriptor is required
3069 * for enabling events at boot. We want to enable events before
3070 * the filesystem is initialized.
3071 */
3072static int
3073__trace_early_add_new_event(struct trace_event_call *call,
3074			    struct trace_array *tr)
3075{
3076	struct trace_event_file *file;
3077	int ret;
3078
3079	file = trace_create_new_event(call, tr);
3080	/*
3081	 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3082	 * allocation, or NULL if the event is not part of the tr->system_names.
3083	 * When the event is not part of the tr->system_names, return zero, not
3084	 * an error.
3085	 */
3086	if (!file)
3087		return 0;
3088
3089	if (IS_ERR(file))
3090		return PTR_ERR(file);
3091
3092	ret = event_define_fields(call);
3093	if (ret)
3094		return ret;
3095
3096	trace_early_triggers(file, trace_event_name(call));
3097
3098	return 0;
3099}
3100
3101struct ftrace_module_file_ops;
3102static void __add_event_to_tracers(struct trace_event_call *call);
3103
3104/* Add an additional event_call dynamically */
3105int trace_add_event_call(struct trace_event_call *call)
3106{
3107	int ret;
3108	lockdep_assert_held(&event_mutex);
3109
3110	mutex_lock(&trace_types_lock);
3111
3112	ret = __register_event(call, NULL);
3113	if (ret >= 0)
3114		__add_event_to_tracers(call);
3115
3116	mutex_unlock(&trace_types_lock);
3117	return ret;
3118}
3119EXPORT_SYMBOL_GPL(trace_add_event_call);
3120
3121/*
3122 * Must be called under locking of trace_types_lock, event_mutex and
3123 * trace_event_sem.
3124 */
3125static void __trace_remove_event_call(struct trace_event_call *call)
3126{
3127	event_remove(call);
3128	trace_destroy_fields(call);
3129	free_event_filter(call->filter);
3130	call->filter = NULL;
3131}
3132
3133static int probe_remove_event_call(struct trace_event_call *call)
3134{
3135	struct trace_array *tr;
3136	struct trace_event_file *file;
3137
3138#ifdef CONFIG_PERF_EVENTS
3139	if (call->perf_refcount)
3140		return -EBUSY;
3141#endif
3142	do_for_each_event_file(tr, file) {
3143		if (file->event_call != call)
3144			continue;
3145		/*
3146		 * We can't rely on ftrace_event_enable_disable(enable => 0)
3147		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3148		 * TRACE_REG_UNREGISTER.
3149		 */
3150		if (file->flags & EVENT_FILE_FL_ENABLED)
3151			goto busy;
3152
3153		if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3154			tr->clear_trace = true;
3155		/*
3156		 * The do_for_each_event_file_safe() is
3157		 * a double loop. After finding the call for this
3158		 * trace_array, we use break to jump to the next
3159		 * trace_array.
3160		 */
3161		break;
3162	} while_for_each_event_file();
3163
3164	__trace_remove_event_call(call);
3165
3166	return 0;
3167 busy:
3168	/* No need to clear the trace now */
3169	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3170		tr->clear_trace = false;
3171	}
3172	return -EBUSY;
3173}
3174
3175/* Remove an event_call */
3176int trace_remove_event_call(struct trace_event_call *call)
3177{
3178	int ret;
3179
3180	lockdep_assert_held(&event_mutex);
3181
3182	mutex_lock(&trace_types_lock);
3183	down_write(&trace_event_sem);
3184	ret = probe_remove_event_call(call);
3185	up_write(&trace_event_sem);
3186	mutex_unlock(&trace_types_lock);
3187
3188	return ret;
3189}
3190EXPORT_SYMBOL_GPL(trace_remove_event_call);
3191
3192#define for_each_event(event, start, end)			\
3193	for (event = start;					\
3194	     (unsigned long)event < (unsigned long)end;		\
3195	     event++)
3196
3197#ifdef CONFIG_MODULES
3198
3199static void trace_module_add_events(struct module *mod)
3200{
3201	struct trace_event_call **call, **start, **end;
3202
3203	if (!mod->num_trace_events)
3204		return;
3205
3206	/* Don't add infrastructure for mods without tracepoints */
3207	if (trace_module_has_bad_taint(mod)) {
3208		pr_err("%s: module has bad taint, not creating trace events\n",
3209		       mod->name);
3210		return;
3211	}
3212
3213	start = mod->trace_events;
3214	end = mod->trace_events + mod->num_trace_events;
3215
3216	for_each_event(call, start, end) {
3217		__register_event(*call, mod);
3218		__add_event_to_tracers(*call);
3219	}
3220}
3221
3222static void trace_module_remove_events(struct module *mod)
3223{
3224	struct trace_event_call *call, *p;
3225	struct module_string *modstr, *m;
3226
3227	down_write(&trace_event_sem);
3228	list_for_each_entry_safe(call, p, &ftrace_events, list) {
3229		if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3230			continue;
3231		if (call->module == mod)
3232			__trace_remove_event_call(call);
3233	}
3234	/* Check for any strings allocade for this module */
3235	list_for_each_entry_safe(modstr, m, &module_strings, next) {
3236		if (modstr->module != mod)
3237			continue;
3238		list_del(&modstr->next);
3239		kfree(modstr->str);
3240		kfree(modstr);
3241	}
3242	up_write(&trace_event_sem);
3243
3244	/*
3245	 * It is safest to reset the ring buffer if the module being unloaded
3246	 * registered any events that were used. The only worry is if
3247	 * a new module gets loaded, and takes on the same id as the events
3248	 * of this module. When printing out the buffer, traced events left
3249	 * over from this module may be passed to the new module events and
3250	 * unexpected results may occur.
3251	 */
3252	tracing_reset_all_online_cpus_unlocked();
3253}
3254
3255static int trace_module_notify(struct notifier_block *self,
3256			       unsigned long val, void *data)
3257{
3258	struct module *mod = data;
3259
3260	mutex_lock(&event_mutex);
3261	mutex_lock(&trace_types_lock);
3262	switch (val) {
3263	case MODULE_STATE_COMING:
3264		trace_module_add_events(mod);
3265		break;
3266	case MODULE_STATE_GOING:
3267		trace_module_remove_events(mod);
3268		break;
3269	}
3270	mutex_unlock(&trace_types_lock);
3271	mutex_unlock(&event_mutex);
3272
3273	return NOTIFY_OK;
3274}
3275
3276static struct notifier_block trace_module_nb = {
3277	.notifier_call = trace_module_notify,
3278	.priority = 1, /* higher than trace.c module notify */
3279};
3280#endif /* CONFIG_MODULES */
3281
3282/* Create a new event directory structure for a trace directory. */
3283static void
3284__trace_add_event_dirs(struct trace_array *tr)
3285{
3286	struct trace_event_call *call;
3287	int ret;
3288
3289	list_for_each_entry(call, &ftrace_events, list) {
3290		ret = __trace_add_new_event(call, tr);
3291		if (ret < 0)
3292			pr_warn("Could not create directory for event %s\n",
3293				trace_event_name(call));
3294	}
3295}
3296
3297/* Returns any file that matches the system and event */
3298struct trace_event_file *
3299__find_event_file(struct trace_array *tr, const char *system, const char *event)
3300{
3301	struct trace_event_file *file;
3302	struct trace_event_call *call;
3303	const char *name;
3304
3305	list_for_each_entry(file, &tr->events, list) {
3306
3307		call = file->event_call;
3308		name = trace_event_name(call);
3309
3310		if (!name || !call->class)
3311			continue;
3312
3313		if (strcmp(event, name) == 0 &&
3314		    strcmp(system, call->class->system) == 0)
3315			return file;
3316	}
3317	return NULL;
3318}
3319
3320/* Returns valid trace event files that match system and event */
3321struct trace_event_file *
3322find_event_file(struct trace_array *tr, const char *system, const char *event)
3323{
3324	struct trace_event_file *file;
3325
3326	file = __find_event_file(tr, system, event);
3327	if (!file || !file->event_call->class->reg ||
3328	    file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3329		return NULL;
3330
3331	return file;
3332}
3333
3334/**
3335 * trace_get_event_file - Find and return a trace event file
3336 * @instance: The name of the trace instance containing the event
3337 * @system: The name of the system containing the event
3338 * @event: The name of the event
3339 *
3340 * Return a trace event file given the trace instance name, trace
3341 * system, and trace event name.  If the instance name is NULL, it
3342 * refers to the top-level trace array.
3343 *
3344 * This function will look it up and return it if found, after calling
3345 * trace_array_get() to prevent the instance from going away, and
3346 * increment the event's module refcount to prevent it from being
3347 * removed.
3348 *
3349 * To release the file, call trace_put_event_file(), which will call
3350 * trace_array_put() and decrement the event's module refcount.
3351 *
3352 * Return: The trace event on success, ERR_PTR otherwise.
3353 */
3354struct trace_event_file *trace_get_event_file(const char *instance,
3355					      const char *system,
3356					      const char *event)
3357{
3358	struct trace_array *tr = top_trace_array();
3359	struct trace_event_file *file = NULL;
3360	int ret = -EINVAL;
3361
3362	if (instance) {
3363		tr = trace_array_find_get(instance);
3364		if (!tr)
3365			return ERR_PTR(-ENOENT);
3366	} else {
3367		ret = trace_array_get(tr);
3368		if (ret)
3369			return ERR_PTR(ret);
3370	}
3371
3372	mutex_lock(&event_mutex);
3373
3374	file = find_event_file(tr, system, event);
3375	if (!file) {
3376		trace_array_put(tr);
3377		ret = -EINVAL;
3378		goto out;
3379	}
3380
3381	/* Don't let event modules unload while in use */
3382	ret = trace_event_try_get_ref(file->event_call);
3383	if (!ret) {
3384		trace_array_put(tr);
3385		ret = -EBUSY;
3386		goto out;
3387	}
3388
3389	ret = 0;
3390 out:
3391	mutex_unlock(&event_mutex);
3392
3393	if (ret)
3394		file = ERR_PTR(ret);
3395
3396	return file;
3397}
3398EXPORT_SYMBOL_GPL(trace_get_event_file);
3399
3400/**
3401 * trace_put_event_file - Release a file from trace_get_event_file()
3402 * @file: The trace event file
3403 *
3404 * If a file was retrieved using trace_get_event_file(), this should
3405 * be called when it's no longer needed.  It will cancel the previous
3406 * trace_array_get() called by that function, and decrement the
3407 * event's module refcount.
3408 */
3409void trace_put_event_file(struct trace_event_file *file)
3410{
3411	mutex_lock(&event_mutex);
3412	trace_event_put_ref(file->event_call);
3413	mutex_unlock(&event_mutex);
3414
3415	trace_array_put(file->tr);
3416}
3417EXPORT_SYMBOL_GPL(trace_put_event_file);
3418
3419#ifdef CONFIG_DYNAMIC_FTRACE
3420
3421/* Avoid typos */
3422#define ENABLE_EVENT_STR	"enable_event"
3423#define DISABLE_EVENT_STR	"disable_event"
3424
3425struct event_probe_data {
3426	struct trace_event_file	*file;
3427	unsigned long			count;
3428	int				ref;
3429	bool				enable;
3430};
3431
3432static void update_event_probe(struct event_probe_data *data)
3433{
3434	if (data->enable)
3435		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3436	else
3437		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3438}
3439
3440static void
3441event_enable_probe(unsigned long ip, unsigned long parent_ip,
3442		   struct trace_array *tr, struct ftrace_probe_ops *ops,
3443		   void *data)
3444{
3445	struct ftrace_func_mapper *mapper = data;
3446	struct event_probe_data *edata;
3447	void **pdata;
3448
3449	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3450	if (!pdata || !*pdata)
3451		return;
3452
3453	edata = *pdata;
3454	update_event_probe(edata);
3455}
3456
3457static void
3458event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3459			 struct trace_array *tr, struct ftrace_probe_ops *ops,
3460			 void *data)
3461{
3462	struct ftrace_func_mapper *mapper = data;
3463	struct event_probe_data *edata;
3464	void **pdata;
3465
3466	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3467	if (!pdata || !*pdata)
3468		return;
3469
3470	edata = *pdata;
3471
3472	if (!edata->count)
3473		return;
3474
3475	/* Skip if the event is in a state we want to switch to */
3476	if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3477		return;
3478
3479	if (edata->count != -1)
3480		(edata->count)--;
3481
3482	update_event_probe(edata);
3483}
3484
3485static int
3486event_enable_print(struct seq_file *m, unsigned long ip,
3487		   struct ftrace_probe_ops *ops, void *data)
3488{
3489	struct ftrace_func_mapper *mapper = data;
3490	struct event_probe_data *edata;
3491	void **pdata;
3492
3493	pdata = ftrace_func_mapper_find_ip(mapper, ip);
3494
3495	if (WARN_ON_ONCE(!pdata || !*pdata))
3496		return 0;
3497
3498	edata = *pdata;
3499
3500	seq_printf(m, "%ps:", (void *)ip);
3501
3502	seq_printf(m, "%s:%s:%s",
3503		   edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3504		   edata->file->event_call->class->system,
3505		   trace_event_name(edata->file->event_call));
3506
3507	if (edata->count == -1)
3508		seq_puts(m, ":unlimited\n");
3509	else
3510		seq_printf(m, ":count=%ld\n", edata->count);
3511
3512	return 0;
3513}
3514
3515static int
3516event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3517		  unsigned long ip, void *init_data, void **data)
3518{
3519	struct ftrace_func_mapper *mapper = *data;
3520	struct event_probe_data *edata = init_data;
3521	int ret;
3522
3523	if (!mapper) {
3524		mapper = allocate_ftrace_func_mapper();
3525		if (!mapper)
3526			return -ENODEV;
3527		*data = mapper;
3528	}
3529
3530	ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3531	if (ret < 0)
3532		return ret;
3533
3534	edata->ref++;
3535
3536	return 0;
3537}
3538
3539static int free_probe_data(void *data)
3540{
3541	struct event_probe_data *edata = data;
3542
3543	edata->ref--;
3544	if (!edata->ref) {
3545		/* Remove the SOFT_MODE flag */
3546		__ftrace_event_enable_disable(edata->file, 0, 1);
3547		trace_event_put_ref(edata->file->event_call);
3548		kfree(edata);
3549	}
3550	return 0;
3551}
3552
3553static void
3554event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3555		  unsigned long ip, void *data)
3556{
3557	struct ftrace_func_mapper *mapper = data;
3558	struct event_probe_data *edata;
3559
3560	if (!ip) {
3561		if (!mapper)
3562			return;
3563		free_ftrace_func_mapper(mapper, free_probe_data);
3564		return;
3565	}
3566
3567	edata = ftrace_func_mapper_remove_ip(mapper, ip);
3568
3569	if (WARN_ON_ONCE(!edata))
3570		return;
3571
3572	if (WARN_ON_ONCE(edata->ref <= 0))
3573		return;
3574
3575	free_probe_data(edata);
3576}
3577
3578static struct ftrace_probe_ops event_enable_probe_ops = {
3579	.func			= event_enable_probe,
3580	.print			= event_enable_print,
3581	.init			= event_enable_init,
3582	.free			= event_enable_free,
3583};
3584
3585static struct ftrace_probe_ops event_enable_count_probe_ops = {
3586	.func			= event_enable_count_probe,
3587	.print			= event_enable_print,
3588	.init			= event_enable_init,
3589	.free			= event_enable_free,
3590};
3591
3592static struct ftrace_probe_ops event_disable_probe_ops = {
3593	.func			= event_enable_probe,
3594	.print			= event_enable_print,
3595	.init			= event_enable_init,
3596	.free			= event_enable_free,
3597};
3598
3599static struct ftrace_probe_ops event_disable_count_probe_ops = {
3600	.func			= event_enable_count_probe,
3601	.print			= event_enable_print,
3602	.init			= event_enable_init,
3603	.free			= event_enable_free,
3604};
3605
3606static int
3607event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3608		  char *glob, char *cmd, char *param, int enabled)
3609{
3610	struct trace_event_file *file;
3611	struct ftrace_probe_ops *ops;
3612	struct event_probe_data *data;
3613	const char *system;
3614	const char *event;
3615	char *number;
3616	bool enable;
3617	int ret;
3618
3619	if (!tr)
3620		return -ENODEV;
3621
3622	/* hash funcs only work with set_ftrace_filter */
3623	if (!enabled || !param)
3624		return -EINVAL;
3625
3626	system = strsep(&param, ":");
3627	if (!param)
3628		return -EINVAL;
3629
3630	event = strsep(&param, ":");
3631
3632	mutex_lock(&event_mutex);
3633
3634	ret = -EINVAL;
3635	file = find_event_file(tr, system, event);
3636	if (!file)
3637		goto out;
3638
3639	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3640
3641	if (enable)
3642		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3643	else
3644		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3645
3646	if (glob[0] == '!') {
3647		ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3648		goto out;
3649	}
3650
3651	ret = -ENOMEM;
3652
3653	data = kzalloc(sizeof(*data), GFP_KERNEL);
3654	if (!data)
3655		goto out;
3656
3657	data->enable = enable;
3658	data->count = -1;
3659	data->file = file;
3660
3661	if (!param)
3662		goto out_reg;
3663
3664	number = strsep(&param, ":");
3665
3666	ret = -EINVAL;
3667	if (!strlen(number))
3668		goto out_free;
3669
3670	/*
3671	 * We use the callback data field (which is a pointer)
3672	 * as our counter.
3673	 */
3674	ret = kstrtoul(number, 0, &data->count);
3675	if (ret)
3676		goto out_free;
3677
3678 out_reg:
3679	/* Don't let event modules unload while probe registered */
3680	ret = trace_event_try_get_ref(file->event_call);
3681	if (!ret) {
3682		ret = -EBUSY;
3683		goto out_free;
3684	}
3685
3686	ret = __ftrace_event_enable_disable(file, 1, 1);
3687	if (ret < 0)
3688		goto out_put;
3689
3690	ret = register_ftrace_function_probe(glob, tr, ops, data);
3691	/*
3692	 * The above returns on success the # of functions enabled,
3693	 * but if it didn't find any functions it returns zero.
3694	 * Consider no functions a failure too.
3695	 */
3696	if (!ret) {
3697		ret = -ENOENT;
3698		goto out_disable;
3699	} else if (ret < 0)
3700		goto out_disable;
3701	/* Just return zero, not the number of enabled functions */
3702	ret = 0;
3703 out:
3704	mutex_unlock(&event_mutex);
3705	return ret;
3706
3707 out_disable:
3708	__ftrace_event_enable_disable(file, 0, 1);
3709 out_put:
3710	trace_event_put_ref(file->event_call);
3711 out_free:
3712	kfree(data);
3713	goto out;
3714}
3715
3716static struct ftrace_func_command event_enable_cmd = {
3717	.name			= ENABLE_EVENT_STR,
3718	.func			= event_enable_func,
3719};
3720
3721static struct ftrace_func_command event_disable_cmd = {
3722	.name			= DISABLE_EVENT_STR,
3723	.func			= event_enable_func,
3724};
3725
3726static __init int register_event_cmds(void)
3727{
3728	int ret;
3729
3730	ret = register_ftrace_command(&event_enable_cmd);
3731	if (WARN_ON(ret < 0))
3732		return ret;
3733	ret = register_ftrace_command(&event_disable_cmd);
3734	if (WARN_ON(ret < 0))
3735		unregister_ftrace_command(&event_enable_cmd);
3736	return ret;
3737}
3738#else
3739static inline int register_event_cmds(void) { return 0; }
3740#endif /* CONFIG_DYNAMIC_FTRACE */
3741
3742/*
3743 * The top level array and trace arrays created by boot-time tracing
3744 * have already had its trace_event_file descriptors created in order
3745 * to allow for early events to be recorded.
3746 * This function is called after the tracefs has been initialized,
3747 * and we now have to create the files associated to the events.
3748 */
3749static void __trace_early_add_event_dirs(struct trace_array *tr)
 
3750{
3751	struct trace_event_file *file;
3752	int ret;
3753
3754
3755	list_for_each_entry(file, &tr->events, list) {
3756		ret = event_create_dir(tr->event_dir, file);
3757		if (ret < 0)
3758			pr_warn("Could not create directory for event %s\n",
3759				trace_event_name(file->event_call));
3760	}
3761}
3762
3763/*
3764 * For early boot up, the top trace array and the trace arrays created
3765 * by boot-time tracing require to have a list of events that can be
3766 * enabled. This must be done before the filesystem is set up in order
3767 * to allow events to be traced early.
3768 */
3769void __trace_early_add_events(struct trace_array *tr)
 
3770{
3771	struct trace_event_call *call;
3772	int ret;
3773
3774	list_for_each_entry(call, &ftrace_events, list) {
3775		/* Early boot up should not have any modules loaded */
3776		if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3777		    WARN_ON_ONCE(call->module))
3778			continue;
3779
3780		ret = __trace_early_add_new_event(call, tr);
3781		if (ret < 0)
3782			pr_warn("Could not create early event %s\n",
3783				trace_event_name(call));
3784	}
3785}
3786
3787/* Remove the event directory structure for a trace directory. */
3788static void
3789__trace_remove_event_dirs(struct trace_array *tr)
3790{
3791	struct trace_event_file *file, *next;
3792
3793	list_for_each_entry_safe(file, next, &tr->events, list)
3794		remove_event_file_dir(file);
3795}
3796
3797static void __add_event_to_tracers(struct trace_event_call *call)
3798{
3799	struct trace_array *tr;
3800
3801	list_for_each_entry(tr, &ftrace_trace_arrays, list)
3802		__trace_add_new_event(call, tr);
3803}
3804
3805extern struct trace_event_call *__start_ftrace_events[];
3806extern struct trace_event_call *__stop_ftrace_events[];
3807
3808static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3809
3810static __init int setup_trace_event(char *str)
3811{
3812	strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3813	trace_set_ring_buffer_expanded(NULL);
3814	disable_tracing_selftest("running event tracing");
3815
3816	return 1;
3817}
3818__setup("trace_event=", setup_trace_event);
3819
3820static int events_callback(const char *name, umode_t *mode, void **data,
3821			   const struct file_operations **fops)
3822{
3823	if (strcmp(name, "enable") == 0) {
3824		*mode = TRACE_MODE_WRITE;
3825		*fops = &ftrace_tr_enable_fops;
3826		return 1;
3827	}
3828
3829	if (strcmp(name, "header_page") == 0) {
3830		*mode = TRACE_MODE_READ;
3831		*fops = &ftrace_show_header_page_fops;
3832
3833	} else if (strcmp(name, "header_event") == 0) {
3834		*mode = TRACE_MODE_READ;
3835		*fops = &ftrace_show_header_event_fops;
3836	} else
3837		return 0;
3838
3839	return 1;
3840}
3841
3842/* Expects to have event_mutex held when called */
3843static int
3844create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3845{
3846	struct eventfs_inode *e_events;
3847	struct dentry *entry;
3848	int nr_entries;
3849	static struct eventfs_entry events_entries[] = {
3850		{
3851			.name		= "enable",
3852			.callback	= events_callback,
3853		},
3854		{
3855			.name		= "header_page",
3856			.callback	= events_callback,
3857		},
3858		{
3859			.name		= "header_event",
3860			.callback	= events_callback,
3861		},
3862	};
3863
3864	entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3865				  tr, &ftrace_set_event_fops);
3866	if (!entry)
 
3867		return -ENOMEM;
 
3868
3869	nr_entries = ARRAY_SIZE(events_entries);
 
 
 
 
3870
3871	e_events = eventfs_create_events_dir("events", parent, events_entries,
3872					     nr_entries, tr);
3873	if (IS_ERR(e_events)) {
3874		pr_warn("Could not create tracefs 'events' directory\n");
3875		return -ENOMEM;
3876	}
3877
3878	/* There are not as crucial, just warn if they are not created */
3879
3880	trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3881			  tr, &ftrace_set_event_pid_fops);
 
 
3882
3883	trace_create_file("set_event_notrace_pid",
3884			  TRACE_MODE_WRITE, parent, tr,
3885			  &ftrace_set_event_notrace_pid_fops);
 
 
 
3886
3887	tr->event_dir = e_events;
 
 
 
 
 
 
3888
3889	return 0;
3890}
3891
3892/**
3893 * event_trace_add_tracer - add a instance of a trace_array to events
3894 * @parent: The parent dentry to place the files/directories for events in
3895 * @tr: The trace array associated with these events
3896 *
3897 * When a new instance is created, it needs to set up its events
3898 * directory, as well as other files associated with events. It also
3899 * creates the event hierarchy in the @parent/events directory.
3900 *
3901 * Returns 0 on success.
3902 *
3903 * Must be called with event_mutex held.
3904 */
3905int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3906{
3907	int ret;
3908
3909	lockdep_assert_held(&event_mutex);
3910
3911	ret = create_event_toplevel_files(parent, tr);
3912	if (ret)
3913		goto out;
3914
3915	down_write(&trace_event_sem);
3916	/* If tr already has the event list, it is initialized in early boot. */
3917	if (unlikely(!list_empty(&tr->events)))
3918		__trace_early_add_event_dirs(tr);
3919	else
3920		__trace_add_event_dirs(tr);
3921	up_write(&trace_event_sem);
3922
3923 out:
3924	return ret;
3925}
3926
3927/*
3928 * The top trace array already had its file descriptors created.
3929 * Now the files themselves need to be created.
3930 */
3931static __init int
3932early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3933{
3934	int ret;
3935
3936	mutex_lock(&event_mutex);
3937
3938	ret = create_event_toplevel_files(parent, tr);
3939	if (ret)
3940		goto out_unlock;
3941
3942	down_write(&trace_event_sem);
3943	__trace_early_add_event_dirs(tr);
3944	up_write(&trace_event_sem);
3945
3946 out_unlock:
3947	mutex_unlock(&event_mutex);
3948
3949	return ret;
3950}
3951
3952/* Must be called with event_mutex held */
3953int event_trace_del_tracer(struct trace_array *tr)
3954{
3955	lockdep_assert_held(&event_mutex);
3956
3957	/* Disable any event triggers and associated soft-disabled events */
3958	clear_event_triggers(tr);
3959
3960	/* Clear the pid list */
3961	__ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3962
3963	/* Disable any running events */
3964	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3965
3966	/* Make sure no more events are being executed */
3967	tracepoint_synchronize_unregister();
3968
3969	down_write(&trace_event_sem);
3970	__trace_remove_event_dirs(tr);
3971	eventfs_remove_events_dir(tr->event_dir);
3972	up_write(&trace_event_sem);
3973
3974	tr->event_dir = NULL;
3975
3976	return 0;
3977}
3978
3979static __init int event_trace_memsetup(void)
3980{
3981	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3982	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3983	return 0;
3984}
3985
3986__init void
3987early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3988{
 
3989	char *token;
3990	int ret;
3991
3992	while (true) {
3993		token = strsep(&buf, ",");
3994
3995		if (!token)
3996			break;
3997
3998		if (*token) {
3999			/* Restarting syscalls requires that we stop them first */
4000			if (disable_first)
4001				ftrace_set_clr_event(tr, token, 0);
4002
4003			ret = ftrace_set_clr_event(tr, token, 1);
4004			if (ret)
4005				pr_warn("Failed to enable trace event: %s\n", token);
4006		}
4007
4008		/* Put back the comma to allow this to be called again */
4009		if (buf)
4010			*(buf - 1) = ',';
4011	}
4012}
4013
4014static __init int event_trace_enable(void)
4015{
4016	struct trace_array *tr = top_trace_array();
4017	struct trace_event_call **iter, *call;
4018	int ret;
4019
4020	if (!tr)
4021		return -ENODEV;
4022
4023	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4024
4025		call = *iter;
4026		ret = event_init(call);
4027		if (!ret)
4028			list_add(&call->list, &ftrace_events);
4029	}
4030
4031	register_trigger_cmds();
4032
4033	/*
4034	 * We need the top trace array to have a working set of trace
4035	 * points at early init, before the debug files and directories
4036	 * are created. Create the file entries now, and attach them
4037	 * to the actual file dentries later.
4038	 */
4039	__trace_early_add_events(tr);
4040
4041	early_enable_events(tr, bootup_event_buf, false);
4042
4043	trace_printk_start_comm();
4044
4045	register_event_cmds();
4046
 
4047
4048	return 0;
4049}
4050
4051/*
4052 * event_trace_enable() is called from trace_event_init() first to
4053 * initialize events and perhaps start any events that are on the
4054 * command line. Unfortunately, there are some events that will not
4055 * start this early, like the system call tracepoints that need
4056 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4057 * event_trace_enable() is called before pid 1 starts, and this flag
4058 * is never set, making the syscall tracepoint never get reached, but
4059 * the event is enabled regardless (and not doing anything).
4060 */
4061static __init int event_trace_enable_again(void)
4062{
4063	struct trace_array *tr;
4064
4065	tr = top_trace_array();
4066	if (!tr)
4067		return -ENODEV;
4068
4069	early_enable_events(tr, bootup_event_buf, true);
4070
4071	return 0;
4072}
4073
4074early_initcall(event_trace_enable_again);
4075
4076/* Init fields which doesn't related to the tracefs */
4077static __init int event_trace_init_fields(void)
4078{
4079	if (trace_define_generic_fields())
4080		pr_warn("tracing: Failed to allocated generic fields");
4081
4082	if (trace_define_common_fields())
4083		pr_warn("tracing: Failed to allocate common fields");
4084
4085	return 0;
4086}
4087
4088__init int event_trace_init(void)
4089{
4090	struct trace_array *tr;
 
 
4091	int ret;
4092
4093	tr = top_trace_array();
4094	if (!tr)
4095		return -ENODEV;
4096
4097	trace_create_file("available_events", TRACE_MODE_READ,
4098			  NULL, tr, &ftrace_avail_fops);
 
4099
4100	ret = early_event_add_tracer(NULL, tr);
 
 
 
 
 
 
 
 
 
 
 
4101	if (ret)
4102		return ret;
4103
4104#ifdef CONFIG_MODULES
4105	ret = register_module_notifier(&trace_module_nb);
4106	if (ret)
4107		pr_warn("Failed to register trace events module notifier\n");
4108#endif
4109
4110	eventdir_initialized = true;
4111
4112	return 0;
4113}
4114
4115void __init trace_event_init(void)
4116{
4117	event_trace_memsetup();
4118	init_ftrace_syscalls();
4119	event_trace_enable();
4120	event_trace_init_fields();
4121}
4122
4123#ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4124
4125static DEFINE_SPINLOCK(test_spinlock);
4126static DEFINE_SPINLOCK(test_spinlock_irq);
4127static DEFINE_MUTEX(test_mutex);
4128
4129static __init void test_work(struct work_struct *dummy)
4130{
4131	spin_lock(&test_spinlock);
4132	spin_lock_irq(&test_spinlock_irq);
4133	udelay(1);
4134	spin_unlock_irq(&test_spinlock_irq);
4135	spin_unlock(&test_spinlock);
4136
4137	mutex_lock(&test_mutex);
4138	msleep(1);
4139	mutex_unlock(&test_mutex);
4140}
4141
4142static __init int event_test_thread(void *unused)
4143{
4144	void *test_malloc;
4145
4146	test_malloc = kmalloc(1234, GFP_KERNEL);
4147	if (!test_malloc)
4148		pr_info("failed to kmalloc\n");
4149
4150	schedule_on_each_cpu(test_work);
4151
4152	kfree(test_malloc);
4153
4154	set_current_state(TASK_INTERRUPTIBLE);
4155	while (!kthread_should_stop()) {
4156		schedule();
4157		set_current_state(TASK_INTERRUPTIBLE);
4158	}
4159	__set_current_state(TASK_RUNNING);
4160
4161	return 0;
4162}
4163
4164/*
4165 * Do various things that may trigger events.
4166 */
4167static __init void event_test_stuff(void)
4168{
4169	struct task_struct *test_thread;
4170
4171	test_thread = kthread_run(event_test_thread, NULL, "test-events");
4172	msleep(1);
4173	kthread_stop(test_thread);
4174}
4175
4176/*
4177 * For every trace event defined, we will test each trace point separately,
4178 * and then by groups, and finally all trace points.
4179 */
4180static __init void event_trace_self_tests(void)
4181{
4182	struct trace_subsystem_dir *dir;
4183	struct trace_event_file *file;
4184	struct trace_event_call *call;
4185	struct event_subsystem *system;
4186	struct trace_array *tr;
4187	int ret;
4188
4189	tr = top_trace_array();
4190	if (!tr)
4191		return;
4192
4193	pr_info("Running tests on trace events:\n");
4194
4195	list_for_each_entry(file, &tr->events, list) {
4196
4197		call = file->event_call;
4198
4199		/* Only test those that have a probe */
4200		if (!call->class || !call->class->probe)
4201			continue;
4202
4203/*
4204 * Testing syscall events here is pretty useless, but
4205 * we still do it if configured. But this is time consuming.
4206 * What we really need is a user thread to perform the
4207 * syscalls as we test.
4208 */
4209#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4210		if (call->class->system &&
4211		    strcmp(call->class->system, "syscalls") == 0)
4212			continue;
4213#endif
4214
4215		pr_info("Testing event %s: ", trace_event_name(call));
4216
4217		/*
4218		 * If an event is already enabled, someone is using
4219		 * it and the self test should not be on.
4220		 */
4221		if (file->flags & EVENT_FILE_FL_ENABLED) {
4222			pr_warn("Enabled event during self test!\n");
4223			WARN_ON_ONCE(1);
4224			continue;
4225		}
4226
4227		ftrace_event_enable_disable(file, 1);
4228		event_test_stuff();
4229		ftrace_event_enable_disable(file, 0);
4230
4231		pr_cont("OK\n");
4232	}
4233
4234	/* Now test at the sub system level */
4235
4236	pr_info("Running tests on trace event systems:\n");
4237
4238	list_for_each_entry(dir, &tr->systems, list) {
4239
4240		system = dir->subsystem;
4241
4242		/* the ftrace system is special, skip it */
4243		if (strcmp(system->name, "ftrace") == 0)
4244			continue;
4245
4246		pr_info("Testing event system %s: ", system->name);
4247
4248		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4249		if (WARN_ON_ONCE(ret)) {
4250			pr_warn("error enabling system %s\n",
4251				system->name);
4252			continue;
4253		}
4254
4255		event_test_stuff();
4256
4257		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4258		if (WARN_ON_ONCE(ret)) {
4259			pr_warn("error disabling system %s\n",
4260				system->name);
4261			continue;
4262		}
4263
4264		pr_cont("OK\n");
4265	}
4266
4267	/* Test with all events enabled */
4268
4269	pr_info("Running tests on all trace events:\n");
4270	pr_info("Testing all events: ");
4271
4272	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4273	if (WARN_ON_ONCE(ret)) {
4274		pr_warn("error enabling all events\n");
4275		return;
4276	}
4277
4278	event_test_stuff();
4279
4280	/* reset sysname */
4281	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4282	if (WARN_ON_ONCE(ret)) {
4283		pr_warn("error disabling all events\n");
4284		return;
4285	}
4286
4287	pr_cont("OK\n");
4288}
4289
4290#ifdef CONFIG_FUNCTION_TRACER
4291
4292static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4293
4294static struct trace_event_file event_trace_file __initdata;
4295
4296static void __init
4297function_test_events_call(unsigned long ip, unsigned long parent_ip,
4298			  struct ftrace_ops *op, struct ftrace_regs *regs)
4299{
4300	struct trace_buffer *buffer;
4301	struct ring_buffer_event *event;
 
4302	struct ftrace_entry *entry;
4303	unsigned int trace_ctx;
4304	long disabled;
4305	int cpu;
 
4306
4307	trace_ctx = tracing_gen_ctx();
4308	preempt_disable_notrace();
4309	cpu = raw_smp_processor_id();
4310	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4311
4312	if (disabled != 1)
4313		goto out;
4314
 
 
4315	event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4316						TRACE_FN, sizeof(*entry),
4317						trace_ctx);
4318	if (!event)
4319		goto out;
4320	entry	= ring_buffer_event_data(event);
4321	entry->ip			= ip;
4322	entry->parent_ip		= parent_ip;
4323
4324	event_trigger_unlock_commit(&event_trace_file, buffer, event,
4325				    entry, trace_ctx);
4326 out:
4327	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4328	preempt_enable_notrace();
4329}
4330
4331static struct ftrace_ops trace_ops __initdata  =
4332{
4333	.func = function_test_events_call,
 
4334};
4335
4336static __init void event_trace_self_test_with_function(void)
4337{
4338	int ret;
4339
4340	event_trace_file.tr = top_trace_array();
4341	if (WARN_ON(!event_trace_file.tr))
4342		return;
4343
4344	ret = register_ftrace_function(&trace_ops);
4345	if (WARN_ON(ret < 0)) {
4346		pr_info("Failed to enable function tracer for event tests\n");
4347		return;
4348	}
4349	pr_info("Running tests again, along with the function tracer\n");
4350	event_trace_self_tests();
4351	unregister_ftrace_function(&trace_ops);
4352}
4353#else
4354static __init void event_trace_self_test_with_function(void)
4355{
4356}
4357#endif
4358
4359static __init int event_trace_self_tests_init(void)
4360{
4361	if (!tracing_selftest_disabled) {
4362		event_trace_self_tests();
4363		event_trace_self_test_with_function();
4364	}
4365
4366	return 0;
4367}
4368
4369late_initcall(event_trace_self_tests_init);
4370
4371#endif