kernel/trace/trace_events.c

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