kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #define pr_fmt(fmt) fmt
  12
  13 #include <linux/workqueue.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kthread.h>
  16 #include <linux/tracefs.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/bsearch.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/sort.h>
  22 #include <linux/slab.h>
  23 #include <linux/delay.h>
  24
  25 #include <trace/events/sched.h>
  26
  27 #include <asm/setup.h>
  28
  29 #include "trace_output.h"
  30
  31 #undef TRACE_SYSTEM
  32 #define TRACE_SYSTEM "TRACE_SYSTEM"
  33
  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 list_head *
  74 trace_get_fields(struct trace_event_call *event_call)
  75 {
  76         if (!event_call->class->get_fields)
  77                 return &event_call->class->fields;
  78         return event_call->class->get_fields(event_call);
  79 }
  80
  81 static struct ftrace_event_field *
  82 __find_event_field(struct list_head *head, char *name)
  83 {
  84         struct ftrace_event_field *field;
  85
  86         list_for_each_entry(field, head, link) {
  87                 if (!strcmp(field->name, name))
  88                         return field;
  89         }
  90
  91         return NULL;
  92 }
  93
  94 struct ftrace_event_field *
  95 trace_find_event_field(struct trace_event_call *call, char *name)
  96 {
  97         struct ftrace_event_field *field;
  98         struct list_head *head;
  99
 100         head = trace_get_fields(call);
 101         field = __find_event_field(head, name);
 102         if (field)
 103                 return field;
 104
 105         field = __find_event_field(&ftrace_generic_fields, name);
 106         if (field)
 107                 return field;
 108
 109         return __find_event_field(&ftrace_common_fields, name);
 110 }
 111
 112 static int __trace_define_field(struct list_head *head, const char *type,
 113                                 const char *name, int offset, int size,
 114                                 int is_signed, int filter_type)
 115 {
 116         struct ftrace_event_field *field;
 117
 118         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 119         if (!field)
 120                 return -ENOMEM;
 121
 122         field->name = name;
 123         field->type = type;
 124
 125         if (filter_type == FILTER_OTHER)
 126                 field->filter_type = filter_assign_type(type);
 127         else
 128                 field->filter_type = filter_type;
 129
 130         field->offset = offset;
 131         field->size = size;
 132         field->is_signed = is_signed;
 133
 134         list_add(&field->link, head);
 135
 136         return 0;
 137 }
 138
 139 int trace_define_field(struct trace_event_call *call, const char *type,
 140                        const char *name, int offset, int size, int is_signed,
 141                        int filter_type)
 142 {
 143         struct list_head *head;
 144
 145         if (WARN_ON(!call->class))
 146                 return 0;
 147
 148         head = trace_get_fields(call);
 149         return __trace_define_field(head, type, name, offset, size,
 150                                     is_signed, filter_type);
 151 }
 152 EXPORT_SYMBOL_GPL(trace_define_field);
 153
 154 #define __generic_field(type, item, filter_type)                        \
 155         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 156                                    #item, 0, 0, is_signed_type(type),   \
 157                                    filter_type);                        \
 158         if (ret)                                                        \
 159                 return ret;
 160
 161 #define __common_field(type, item)                                      \
 162         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 163                                    "common_" #item,                     \
 164                                    offsetof(typeof(ent), item),         \
 165                                    sizeof(ent.item),                    \
 166                                    is_signed_type(type), FILTER_OTHER); \
 167         if (ret)                                                        \
 168                 return ret;
 169
 170 static int trace_define_generic_fields(void)
 171 {
 172         int ret;
 173
 174         __generic_field(int, CPU, FILTER_CPU);
 175         __generic_field(int, cpu, FILTER_CPU);
 176         __generic_field(char *, COMM, FILTER_COMM);
 177         __generic_field(char *, comm, FILTER_COMM);
 178
 179         return ret;
 180 }
 181
 182 static int trace_define_common_fields(void)
 183 {
 184         int ret;
 185         struct trace_entry ent;
 186
 187         __common_field(unsigned short, type);
 188         __common_field(unsigned char, flags);
 189         __common_field(unsigned char, preempt_count);
 190         __common_field(int, pid);
 191
 192         return ret;
 193 }
 194
 195 static void trace_destroy_fields(struct trace_event_call *call)
 196 {
 197         struct ftrace_event_field *field, *next;
 198         struct list_head *head;
 199
 200         head = trace_get_fields(call);
 201         list_for_each_entry_safe(field, next, head, link) {
 202                 list_del(&field->link);
 203                 kmem_cache_free(field_cachep, field);
 204         }
 205 }
 206
 207 int trace_event_raw_init(struct trace_event_call *call)
 208 {
 209         int id;
 210
 211         id = register_trace_event(&call->event);
 212         if (!id)
 213                 return -ENODEV;
 214
 215         return 0;
 216 }
 217 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 218
 219 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 220 {
 221         struct trace_array *tr = trace_file->tr;
 222         struct trace_array_cpu *data;
 223         struct trace_pid_list *pid_list;
 224
 225         pid_list = rcu_dereference_sched(tr->filtered_pids);
 226         if (!pid_list)
 227                 return false;
 228
 229         data = this_cpu_ptr(tr->trace_buffer.data);
 230
 231         return data->ignore_pid;
 232 }
 233 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 234
 235 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 236                                  struct trace_event_file *trace_file,
 237                                  unsigned long len)
 238 {
 239         struct trace_event_call *event_call = trace_file->event_call;
 240
 241         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 242             trace_event_ignore_this_pid(trace_file))
 243                 return NULL;
 244
 245         local_save_flags(fbuffer->flags);
 246         fbuffer->pc = preempt_count();
 247         fbuffer->trace_file = trace_file;
 248
 249         fbuffer->event =
 250                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 251                                                 event_call->event.type, len,
 252                                                 fbuffer->flags, fbuffer->pc);
 253         if (!fbuffer->event)
 254                 return NULL;
 255
 256         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 257         return fbuffer->entry;
 258 }
 259 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 260
 261 static DEFINE_SPINLOCK(tracepoint_iter_lock);
 262
 263 static void output_printk(struct trace_event_buffer *fbuffer)
 264 {
 265         struct trace_event_call *event_call;
 266         struct trace_event *event;
 267         unsigned long flags;
 268         struct trace_iterator *iter = tracepoint_print_iter;
 269
 270         if (!iter)
 271                 return;
 272
 273         event_call = fbuffer->trace_file->event_call;
 274         if (!event_call || !event_call->event.funcs ||
 275             !event_call->event.funcs->trace)
 276                 return;
 277
 278         event = &fbuffer->trace_file->event_call->event;
 279
 280         spin_lock_irqsave(&tracepoint_iter_lock, flags);
 281         trace_seq_init(&iter->seq);
 282         iter->ent = fbuffer->entry;
 283         event_call->event.funcs->trace(iter, 0, event);
 284         trace_seq_putc(&iter->seq, 0);
 285         printk("%s", iter->seq.buffer);
 286
 287         spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
 288 }
 289
 290 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
 291 {
 292         if (tracepoint_printk)
 293                 output_printk(fbuffer);
 294
 295         event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
 296                                     fbuffer->event, fbuffer->entry,
 297                                     fbuffer->flags, fbuffer->pc);
 298 }
 299 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
 300
 301 int trace_event_reg(struct trace_event_call *call,
 302                     enum trace_reg type, void *data)
 303 {
 304         struct trace_event_file *file = data;
 305
 306         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 307         switch (type) {
 308         case TRACE_REG_REGISTER:
 309                 return tracepoint_probe_register(call->tp,
 310                                                  call->class->probe,
 311                                                  file);
 312         case TRACE_REG_UNREGISTER:
 313                 tracepoint_probe_unregister(call->tp,
 314                                             call->class->probe,
 315                                             file);
 316                 return 0;
 317
 318 #ifdef CONFIG_PERF_EVENTS
 319         case TRACE_REG_PERF_REGISTER:
 320                 return tracepoint_probe_register(call->tp,
 321                                                  call->class->perf_probe,
 322                                                  call);
 323         case TRACE_REG_PERF_UNREGISTER:
 324                 tracepoint_probe_unregister(call->tp,
 325                                             call->class->perf_probe,
 326                                             call);
 327                 return 0;
 328         case TRACE_REG_PERF_OPEN:
 329         case TRACE_REG_PERF_CLOSE:
 330         case TRACE_REG_PERF_ADD:
 331         case TRACE_REG_PERF_DEL:
 332                 return 0;
 333 #endif
 334         }
 335         return 0;
 336 }
 337 EXPORT_SYMBOL_GPL(trace_event_reg);
 338
 339 void trace_event_enable_cmd_record(bool enable)
 340 {
 341         struct trace_event_file *file;
 342         struct trace_array *tr;
 343
 344         mutex_lock(&event_mutex);
 345         do_for_each_event_file(tr, file) {
 346
 347                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 348                         continue;
 349
 350                 if (enable) {
 351                         tracing_start_cmdline_record();
 352                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 353                 } else {
 354                         tracing_stop_cmdline_record();
 355                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 356                 }
 357         } while_for_each_event_file();
 358         mutex_unlock(&event_mutex);
 359 }
 360
 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         int ret = 0;
 367         int disable;
 368
 369         switch (enable) {
 370         case 0:
 371                 /*
 372                  * When soft_disable is set and enable is cleared, the sm_ref
 373                  * reference counter is decremented. If it reaches 0, we want
 374                  * to clear the SOFT_DISABLED flag but leave the event in the
 375                  * state that it was. That is, if the event was enabled and
 376                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 377                  * is set we do not want the event to be enabled before we
 378                  * clear the bit.
 379                  *
 380                  * When soft_disable is not set but the SOFT_MODE flag is,
 381                  * we do nothing. Do not disable the tracepoint, otherwise
 382                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 383                  */
 384                 if (soft_disable) {
 385                         if (atomic_dec_return(&file->sm_ref) > 0)
 386                                 break;
 387                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 388                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 389                 } else
 390                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 391
 392                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 393                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 394                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 395                                 tracing_stop_cmdline_record();
 396                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 397                         }
 398                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 399                 }
 400                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 401                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 402                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 403                 else
 404                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 405                 break;
 406         case 1:
 407                 /*
 408                  * When soft_disable is set and enable is set, we want to
 409                  * register the tracepoint for the event, but leave the event
 410                  * as is. That means, if the event was already enabled, we do
 411                  * nothing (but set SOFT_MODE). If the event is disabled, we
 412                  * set SOFT_DISABLED before enabling the event tracepoint, so
 413                  * it still seems to be disabled.
 414                  */
 415                 if (!soft_disable)
 416                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 417                 else {
 418                         if (atomic_inc_return(&file->sm_ref) > 1)
 419                                 break;
 420                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 421                 }
 422
 423                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 424
 425                         /* Keep the event disabled, when going to SOFT_MODE. */
 426                         if (soft_disable)
 427                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 428
 429                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 430                                 tracing_start_cmdline_record();
 431                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 432                         }
 433                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 434                         if (ret) {
 435                                 tracing_stop_cmdline_record();
 436                                 pr_info("event trace: Could not enable event "
 437                                         "%s\n", trace_event_name(call));
 438                                 break;
 439                         }
 440                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 441
 442                         /* WAS_ENABLED gets set but never cleared. */
 443                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 444                 }
 445                 break;
 446         }
 447
 448         return ret;
 449 }
 450
 451 int trace_event_enable_disable(struct trace_event_file *file,
 452                                int enable, int soft_disable)
 453 {
 454         return __ftrace_event_enable_disable(file, enable, soft_disable);
 455 }
 456
 457 static int ftrace_event_enable_disable(struct trace_event_file *file,
 458                                        int enable)
 459 {
 460         return __ftrace_event_enable_disable(file, enable, 0);
 461 }
 462
 463 static void ftrace_clear_events(struct trace_array *tr)
 464 {
 465         struct trace_event_file *file;
 466
 467         mutex_lock(&event_mutex);
 468         list_for_each_entry(file, &tr->events, list) {
 469                 ftrace_event_enable_disable(file, 0);
 470         }
 471         mutex_unlock(&event_mutex);
 472 }
 473
 474 static int cmp_pid(const void *key, const void *elt)
 475 {
 476         const pid_t *search_pid = key;
 477         const pid_t *pid = elt;
 478
 479         if (*search_pid == *pid)
 480                 return 0;
 481         if (*search_pid < *pid)
 482                 return -1;
 483         return 1;
 484 }
 485
 486 static bool
 487 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
 488 {
 489         pid_t search_pid;
 490         pid_t *pid;
 491
 492         /*
 493          * Return false, because if filtered_pids does not exist,
 494          * all pids are good to trace.
 495          */
 496         if (!filtered_pids)
 497                 return false;
 498
 499         search_pid = task->pid;
 500
 501         pid = bsearch(&search_pid, filtered_pids->pids,
 502                       filtered_pids->nr_pids, sizeof(pid_t),
 503                       cmp_pid);
 504         if (!pid)
 505                 return true;
 506
 507         return false;
 508 }
 509
 510 static void
 511 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 512                     struct task_struct *prev, struct task_struct *next)
 513 {
 514         struct trace_array *tr = data;
 515         struct trace_pid_list *pid_list;
 516
 517         pid_list = rcu_dereference_sched(tr->filtered_pids);
 518
 519         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 520                        check_ignore_pid(pid_list, prev) &&
 521                        check_ignore_pid(pid_list, next));
 522 }
 523
 524 static void
 525 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 526                     struct task_struct *prev, struct task_struct *next)
 527 {
 528         struct trace_array *tr = data;
 529         struct trace_pid_list *pid_list;
 530
 531         pid_list = rcu_dereference_sched(tr->filtered_pids);
 532
 533         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 534                        check_ignore_pid(pid_list, next));
 535 }
 536
 537 static void
 538 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 539 {
 540         struct trace_array *tr = data;
 541         struct trace_pid_list *pid_list;
 542
 543         /* Nothing to do if we are already tracing */
 544         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 545                 return;
 546
 547         pid_list = rcu_dereference_sched(tr->filtered_pids);
 548
 549         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 550                        check_ignore_pid(pid_list, task));
 551 }
 552
 553 static void
 554 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 555 {
 556         struct trace_array *tr = data;
 557         struct trace_pid_list *pid_list;
 558
 559         /* Nothing to do if we are not tracing */
 560         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 561                 return;
 562
 563         pid_list = rcu_dereference_sched(tr->filtered_pids);
 564
 565         /* Set tracing if current is enabled */
 566         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 567                        check_ignore_pid(pid_list, current));
 568 }
 569
 570 static void __ftrace_clear_event_pids(struct trace_array *tr)
 571 {
 572         struct trace_pid_list *pid_list;
 573         struct trace_event_file *file;
 574         int cpu;
 575
 576         pid_list = rcu_dereference_protected(tr->filtered_pids,
 577                                              lockdep_is_held(&event_mutex));
 578         if (!pid_list)
 579                 return;
 580
 581         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 582         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 583
 584         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 585         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 586
 587         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 588         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 589
 590         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 591         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 592
 593         list_for_each_entry(file, &tr->events, list) {
 594                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 595         }
 596
 597         for_each_possible_cpu(cpu)
 598                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 599
 600         rcu_assign_pointer(tr->filtered_pids, NULL);
 601
 602         /* Wait till all users are no longer using pid filtering */
 603         synchronize_sched();
 604
 605         free_pages((unsigned long)pid_list->pids, pid_list->order);
 606         kfree(pid_list);
 607 }
 608
 609 static void ftrace_clear_event_pids(struct trace_array *tr)
 610 {
 611         mutex_lock(&event_mutex);
 612         __ftrace_clear_event_pids(tr);
 613         mutex_unlock(&event_mutex);
 614 }
 615
 616 static void __put_system(struct event_subsystem *system)
 617 {
 618         struct event_filter *filter = system->filter;
 619
 620         WARN_ON_ONCE(system_refcount(system) == 0);
 621         if (system_refcount_dec(system))
 622                 return;
 623
 624         list_del(&system->list);
 625
 626         if (filter) {
 627                 kfree(filter->filter_string);
 628                 kfree(filter);
 629         }
 630         kfree_const(system->name);
 631         kfree(system);
 632 }
 633
 634 static void __get_system(struct event_subsystem *system)
 635 {
 636         WARN_ON_ONCE(system_refcount(system) == 0);
 637         system_refcount_inc(system);
 638 }
 639
 640 static void __get_system_dir(struct trace_subsystem_dir *dir)
 641 {
 642         WARN_ON_ONCE(dir->ref_count == 0);
 643         dir->ref_count++;
 644         __get_system(dir->subsystem);
 645 }
 646
 647 static void __put_system_dir(struct trace_subsystem_dir *dir)
 648 {
 649         WARN_ON_ONCE(dir->ref_count == 0);
 650         /* If the subsystem is about to be freed, the dir must be too */
 651         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 652
 653         __put_system(dir->subsystem);
 654         if (!--dir->ref_count)
 655                 kfree(dir);
 656 }
 657
 658 static void put_system(struct trace_subsystem_dir *dir)
 659 {
 660         mutex_lock(&event_mutex);
 661         __put_system_dir(dir);
 662         mutex_unlock(&event_mutex);
 663 }
 664
 665 static void remove_subsystem(struct trace_subsystem_dir *dir)
 666 {
 667         if (!dir)
 668                 return;
 669
 670         if (!--dir->nr_events) {
 671                 tracefs_remove_recursive(dir->entry);
 672                 list_del(&dir->list);
 673                 __put_system_dir(dir);
 674         }
 675 }
 676
 677 static void remove_event_file_dir(struct trace_event_file *file)
 678 {
 679         struct dentry *dir = file->dir;
 680         struct dentry *child;
 681
 682         if (dir) {
 683                 spin_lock(&dir->d_lock);        /* probably unneeded */
 684                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 685                         if (d_really_is_positive(child))        /* probably unneeded */
 686                                 d_inode(child)->i_private = NULL;
 687                 }
 688                 spin_unlock(&dir->d_lock);
 689
 690                 tracefs_remove_recursive(dir);
 691         }
 692
 693         list_del(&file->list);
 694         remove_subsystem(file->system);
 695         free_event_filter(file->filter);
 696         kmem_cache_free(file_cachep, file);
 697 }
 698
 699 /*
 700  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 701  */
 702 static int
 703 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 704                               const char *sub, const char *event, int set)
 705 {
 706         struct trace_event_file *file;
 707         struct trace_event_call *call;
 708         const char *name;
 709         int ret = -EINVAL;
 710
 711         list_for_each_entry(file, &tr->events, list) {
 712
 713                 call = file->event_call;
 714                 name = trace_event_name(call);
 715
 716                 if (!name || !call->class || !call->class->reg)
 717                         continue;
 718
 719                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 720                         continue;
 721
 722                 if (match &&
 723                     strcmp(match, name) != 0 &&
 724                     strcmp(match, call->class->system) != 0)
 725                         continue;
 726
 727                 if (sub && strcmp(sub, call->class->system) != 0)
 728                         continue;
 729
 730                 if (event && strcmp(event, name) != 0)
 731                         continue;
 732
 733                 ftrace_event_enable_disable(file, set);
 734
 735                 ret = 0;
 736         }
 737
 738         return ret;
 739 }
 740
 741 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 742                                   const char *sub, const char *event, int set)
 743 {
 744         int ret;
 745
 746         mutex_lock(&event_mutex);
 747         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 748         mutex_unlock(&event_mutex);
 749
 750         return ret;
 751 }
 752
 753 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 754 {
 755         char *event = NULL, *sub = NULL, *match;
 756         int ret;
 757
 758         /*
 759          * The buf format can be <subsystem>:<event-name>
 760          *  *:<event-name> means any event by that name.
 761          *  :<event-name> is the same.
 762          *
 763          *  <subsystem>:* means all events in that subsystem
 764          *  <subsystem>: means the same.
 765          *
 766          *  <name> (no ':') means all events in a subsystem with
 767          *  the name <name> or any event that matches <name>
 768          */
 769
 770         match = strsep(&buf, ":");
 771         if (buf) {
 772                 sub = match;
 773                 event = buf;
 774                 match = NULL;
 775
 776                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 777                         sub = NULL;
 778                 if (!strlen(event) || strcmp(event, "*") == 0)
 779                         event = NULL;
 780         }
 781
 782         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 783
 784         /* Put back the colon to allow this to be called again */
 785         if (buf)
 786                 *(buf - 1) = ':';
 787
 788         return ret;
 789 }
 790
 791 /**
 792  * trace_set_clr_event - enable or disable an event
 793  * @system: system name to match (NULL for any system)
 794  * @event: event name to match (NULL for all events, within system)
 795  * @set: 1 to enable, 0 to disable
 796  *
 797  * This is a way for other parts of the kernel to enable or disable
 798  * event recording.
 799  *
 800  * Returns 0 on success, -EINVAL if the parameters do not match any
 801  * registered events.
 802  */
 803 int trace_set_clr_event(const char *system, const char *event, int set)
 804 {
 805         struct trace_array *tr = top_trace_array();
 806
 807         if (!tr)
 808                 return -ENODEV;
 809
 810         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 811 }
 812 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 813
 814 /* 128 should be much more than enough */
 815 #define EVENT_BUF_SIZE          127
 816
 817 static ssize_t
 818 ftrace_event_write(struct file *file, const char __user *ubuf,
 819                    size_t cnt, loff_t *ppos)
 820 {
 821         struct trace_parser parser;
 822         struct seq_file *m = file->private_data;
 823         struct trace_array *tr = m->private;
 824         ssize_t read, ret;
 825
 826         if (!cnt)
 827                 return 0;
 828
 829         ret = tracing_update_buffers();
 830         if (ret < 0)
 831                 return ret;
 832
 833         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 834                 return -ENOMEM;
 835
 836         read = trace_get_user(&parser, ubuf, cnt, ppos);
 837
 838         if (read >= 0 && trace_parser_loaded((&parser))) {
 839                 int set = 1;
 840
 841                 if (*parser.buffer == '!')
 842                         set = 0;
 843
 844                 parser.buffer[parser.idx] = 0;
 845
 846                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 847                 if (ret)
 848                         goto out_put;
 849         }
 850
 851         ret = read;
 852
 853  out_put:
 854         trace_parser_put(&parser);
 855
 856         return ret;
 857 }
 858
 859 static void *
 860 t_next(struct seq_file *m, void *v, loff_t *pos)
 861 {
 862         struct trace_event_file *file = v;
 863         struct trace_event_call *call;
 864         struct trace_array *tr = m->private;
 865
 866         (*pos)++;
 867
 868         list_for_each_entry_continue(file, &tr->events, list) {
 869                 call = file->event_call;
 870                 /*
 871                  * The ftrace subsystem is for showing formats only.
 872                  * They can not be enabled or disabled via the event files.
 873                  */
 874                 if (call->class && call->class->reg &&
 875                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 876                         return file;
 877         }
 878
 879         return NULL;
 880 }
 881
 882 static void *t_start(struct seq_file *m, loff_t *pos)
 883 {
 884         struct trace_event_file *file;
 885         struct trace_array *tr = m->private;
 886         loff_t l;
 887
 888         mutex_lock(&event_mutex);
 889
 890         file = list_entry(&tr->events, struct trace_event_file, list);
 891         for (l = 0; l <= *pos; ) {
 892                 file = t_next(m, file, &l);
 893                 if (!file)
 894                         break;
 895         }
 896         return file;
 897 }
 898
 899 static void *
 900 s_next(struct seq_file *m, void *v, loff_t *pos)
 901 {
 902         struct trace_event_file *file = v;
 903         struct trace_array *tr = m->private;
 904
 905         (*pos)++;
 906
 907         list_for_each_entry_continue(file, &tr->events, list) {
 908                 if (file->flags & EVENT_FILE_FL_ENABLED)
 909                         return file;
 910         }
 911
 912         return NULL;
 913 }
 914
 915 static void *s_start(struct seq_file *m, loff_t *pos)
 916 {
 917         struct trace_event_file *file;
 918         struct trace_array *tr = m->private;
 919         loff_t l;
 920
 921         mutex_lock(&event_mutex);
 922
 923         file = list_entry(&tr->events, struct trace_event_file, list);
 924         for (l = 0; l <= *pos; ) {
 925                 file = s_next(m, file, &l);
 926                 if (!file)
 927                         break;
 928         }
 929         return file;
 930 }
 931
 932 static int t_show(struct seq_file *m, void *v)
 933 {
 934         struct trace_event_file *file = v;
 935         struct trace_event_call *call = file->event_call;
 936
 937         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 938                 seq_printf(m, "%s:", call->class->system);
 939         seq_printf(m, "%s\n", trace_event_name(call));
 940
 941         return 0;
 942 }
 943
 944 static void t_stop(struct seq_file *m, void *p)
 945 {
 946         mutex_unlock(&event_mutex);
 947 }
 948
 949 static void *p_start(struct seq_file *m, loff_t *pos)
 950         __acquires(RCU)
 951 {
 952         struct trace_pid_list *pid_list;
 953         struct trace_array *tr = m->private;
 954
 955         /*
 956          * Grab the mutex, to keep calls to p_next() having the same
 957          * tr->filtered_pids as p_start() has.
 958          * If we just passed the tr->filtered_pids around, then RCU would
 959          * have been enough, but doing that makes things more complex.
 960          */
 961         mutex_lock(&event_mutex);
 962         rcu_read_lock_sched();
 963
 964         pid_list = rcu_dereference_sched(tr->filtered_pids);
 965
 966         if (!pid_list || *pos >= pid_list->nr_pids)
 967                 return NULL;
 968
 969         return (void *)&pid_list->pids[*pos];
 970 }
 971
 972 static void p_stop(struct seq_file *m, void *p)
 973         __releases(RCU)
 974 {
 975         rcu_read_unlock_sched();
 976         mutex_unlock(&event_mutex);
 977 }
 978
 979 static void *
 980 p_next(struct seq_file *m, void *v, loff_t *pos)
 981 {
 982         struct trace_array *tr = m->private;
 983         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 984
 985         (*pos)++;
 986
 987         if (*pos >= pid_list->nr_pids)
 988                 return NULL;
 989
 990         return (void *)&pid_list->pids[*pos];
 991 }
 992
 993 static int p_show(struct seq_file *m, void *v)
 994 {
 995         pid_t *pid = v;
 996
 997         seq_printf(m, "%d\n", *pid);
 998         return 0;
 999 }
1000
1001 static ssize_t
1002 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1003                   loff_t *ppos)
1004 {
1005         struct trace_event_file *file;
1006         unsigned long flags;
1007         char buf[4] = "0";
1008
1009         mutex_lock(&event_mutex);
1010         file = event_file_data(filp);
1011         if (likely(file))
1012                 flags = file->flags;
1013         mutex_unlock(&event_mutex);
1014
1015         if (!file)
1016                 return -ENODEV;
1017
1018         if (flags & EVENT_FILE_FL_ENABLED &&
1019             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1020                 strcpy(buf, "1");
1021
1022         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1023             flags & EVENT_FILE_FL_SOFT_MODE)
1024                 strcat(buf, "*");
1025
1026         strcat(buf, "\n");
1027
1028         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1029 }
1030
1031 static ssize_t
1032 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1033                    loff_t *ppos)
1034 {
1035         struct trace_event_file *file;
1036         unsigned long val;
1037         int ret;
1038
1039         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1040         if (ret)
1041                 return ret;
1042
1043         ret = tracing_update_buffers();
1044         if (ret < 0)
1045                 return ret;
1046
1047         switch (val) {
1048         case 0:
1049         case 1:
1050                 ret = -ENODEV;
1051                 mutex_lock(&event_mutex);
1052                 file = event_file_data(filp);
1053                 if (likely(file))
1054                         ret = ftrace_event_enable_disable(file, val);
1055                 mutex_unlock(&event_mutex);
1056                 break;
1057
1058         default:
1059                 return -EINVAL;
1060         }
1061
1062         *ppos += cnt;
1063
1064         return ret ? ret : cnt;
1065 }
1066
1067 static ssize_t
1068 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1069                    loff_t *ppos)
1070 {
1071         const char set_to_char[4] = { '?', '0', '1', 'X' };
1072         struct trace_subsystem_dir *dir = filp->private_data;
1073         struct event_subsystem *system = dir->subsystem;
1074         struct trace_event_call *call;
1075         struct trace_event_file *file;
1076         struct trace_array *tr = dir->tr;
1077         char buf[2];
1078         int set = 0;
1079         int ret;
1080
1081         mutex_lock(&event_mutex);
1082         list_for_each_entry(file, &tr->events, list) {
1083                 call = file->event_call;
1084                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1085                         continue;
1086
1087                 if (system && strcmp(call->class->system, system->name) != 0)
1088                         continue;
1089
1090                 /*
1091                  * We need to find out if all the events are set
1092                  * or if all events or cleared, or if we have
1093                  * a mixture.
1094                  */
1095                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1096
1097                 /*
1098                  * If we have a mixture, no need to look further.
1099                  */
1100                 if (set == 3)
1101                         break;
1102         }
1103         mutex_unlock(&event_mutex);
1104
1105         buf[0] = set_to_char[set];
1106         buf[1] = '\n';
1107
1108         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1109
1110         return ret;
1111 }
1112
1113 static ssize_t
1114 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1115                     loff_t *ppos)
1116 {
1117         struct trace_subsystem_dir *dir = filp->private_data;
1118         struct event_subsystem *system = dir->subsystem;
1119         const char *name = NULL;
1120         unsigned long val;
1121         ssize_t ret;
1122
1123         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1124         if (ret)
1125                 return ret;
1126
1127         ret = tracing_update_buffers();
1128         if (ret < 0)
1129                 return ret;
1130
1131         if (val != 0 && val != 1)
1132                 return -EINVAL;
1133
1134         /*
1135          * Opening of "enable" adds a ref count to system,
1136          * so the name is safe to use.
1137          */
1138         if (system)
1139                 name = system->name;
1140
1141         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1142         if (ret)
1143                 goto out;
1144
1145         ret = cnt;
1146
1147 out:
1148         *ppos += cnt;
1149
1150         return ret;
1151 }
1152
1153 enum {
1154         FORMAT_HEADER           = 1,
1155         FORMAT_FIELD_SEPERATOR  = 2,
1156         FORMAT_PRINTFMT         = 3,
1157 };
1158
1159 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1160 {
1161         struct trace_event_call *call = event_file_data(m->private);
1162         struct list_head *common_head = &ftrace_common_fields;
1163         struct list_head *head = trace_get_fields(call);
1164         struct list_head *node = v;
1165
1166         (*pos)++;
1167
1168         switch ((unsigned long)v) {
1169         case FORMAT_HEADER:
1170                 node = common_head;
1171                 break;
1172
1173         case FORMAT_FIELD_SEPERATOR:
1174                 node = head;
1175                 break;
1176
1177         case FORMAT_PRINTFMT:
1178                 /* all done */
1179                 return NULL;
1180         }
1181
1182         node = node->prev;
1183         if (node == common_head)
1184                 return (void *)FORMAT_FIELD_SEPERATOR;
1185         else if (node == head)
1186                 return (void *)FORMAT_PRINTFMT;
1187         else
1188                 return node;
1189 }
1190
1191 static int f_show(struct seq_file *m, void *v)
1192 {
1193         struct trace_event_call *call = event_file_data(m->private);
1194         struct ftrace_event_field *field;
1195         const char *array_descriptor;
1196
1197         switch ((unsigned long)v) {
1198         case FORMAT_HEADER:
1199                 seq_printf(m, "name: %s\n", trace_event_name(call));
1200                 seq_printf(m, "ID: %d\n", call->event.type);
1201                 seq_puts(m, "format:\n");
1202                 return 0;
1203
1204         case FORMAT_FIELD_SEPERATOR:
1205                 seq_putc(m, '\n');
1206                 return 0;
1207
1208         case FORMAT_PRINTFMT:
1209                 seq_printf(m, "\nprint fmt: %s\n",
1210                            call->print_fmt);
1211                 return 0;
1212         }
1213
1214         field = list_entry(v, struct ftrace_event_field, link);
1215         /*
1216          * Smartly shows the array type(except dynamic array).
1217          * Normal:
1218          *      field:TYPE VAR
1219          * If TYPE := TYPE[LEN], it is shown:
1220          *      field:TYPE VAR[LEN]
1221          */
1222         array_descriptor = strchr(field->type, '[');
1223
1224         if (!strncmp(field->type, "__data_loc", 10))
1225                 array_descriptor = NULL;
1226
1227         if (!array_descriptor)
1228                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1229                            field->type, field->name, field->offset,
1230                            field->size, !!field->is_signed);
1231         else
1232                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1233                            (int)(array_descriptor - field->type),
1234                            field->type, field->name,
1235                            array_descriptor, field->offset,
1236                            field->size, !!field->is_signed);
1237
1238         return 0;
1239 }
1240
1241 static void *f_start(struct seq_file *m, loff_t *pos)
1242 {
1243         void *p = (void *)FORMAT_HEADER;
1244         loff_t l = 0;
1245
1246         /* ->stop() is called even if ->start() fails */
1247         mutex_lock(&event_mutex);
1248         if (!event_file_data(m->private))
1249                 return ERR_PTR(-ENODEV);
1250
1251         while (l < *pos && p)
1252                 p = f_next(m, p, &l);
1253
1254         return p;
1255 }
1256
1257 static void f_stop(struct seq_file *m, void *p)
1258 {
1259         mutex_unlock(&event_mutex);
1260 }
1261
1262 static const struct seq_operations trace_format_seq_ops = {
1263         .start          = f_start,
1264         .next           = f_next,
1265         .stop           = f_stop,
1266         .show           = f_show,
1267 };
1268
1269 static int trace_format_open(struct inode *inode, struct file *file)
1270 {
1271         struct seq_file *m;
1272         int ret;
1273
1274         ret = seq_open(file, &trace_format_seq_ops);
1275         if (ret < 0)
1276                 return ret;
1277
1278         m = file->private_data;
1279         m->private = file;
1280
1281         return 0;
1282 }
1283
1284 static ssize_t
1285 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1286 {
1287         int id = (long)event_file_data(filp);
1288         char buf[32];
1289         int len;
1290
1291         if (*ppos)
1292                 return 0;
1293
1294         if (unlikely(!id))
1295                 return -ENODEV;
1296
1297         len = sprintf(buf, "%d\n", id);
1298
1299         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1300 }
1301
1302 static ssize_t
1303 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1304                   loff_t *ppos)
1305 {
1306         struct trace_event_file *file;
1307         struct trace_seq *s;
1308         int r = -ENODEV;
1309
1310         if (*ppos)
1311                 return 0;
1312
1313         s = kmalloc(sizeof(*s), GFP_KERNEL);
1314
1315         if (!s)
1316                 return -ENOMEM;
1317
1318         trace_seq_init(s);
1319
1320         mutex_lock(&event_mutex);
1321         file = event_file_data(filp);
1322         if (file)
1323                 print_event_filter(file, s);
1324         mutex_unlock(&event_mutex);
1325
1326         if (file)
1327                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1328                                             s->buffer, trace_seq_used(s));
1329
1330         kfree(s);
1331
1332         return r;
1333 }
1334
1335 static ssize_t
1336 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1337                    loff_t *ppos)
1338 {
1339         struct trace_event_file *file;
1340         char *buf;
1341         int err = -ENODEV;
1342
1343         if (cnt >= PAGE_SIZE)
1344                 return -EINVAL;
1345
1346         buf = memdup_user_nul(ubuf, cnt);
1347         if (IS_ERR(buf))
1348                 return PTR_ERR(buf);
1349
1350         mutex_lock(&event_mutex);
1351         file = event_file_data(filp);
1352         if (file)
1353                 err = apply_event_filter(file, buf);
1354         mutex_unlock(&event_mutex);
1355
1356         kfree(buf);
1357         if (err < 0)
1358                 return err;
1359
1360         *ppos += cnt;
1361
1362         return cnt;
1363 }
1364
1365 static LIST_HEAD(event_subsystems);
1366
1367 static int subsystem_open(struct inode *inode, struct file *filp)
1368 {
1369         struct event_subsystem *system = NULL;
1370         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1371         struct trace_array *tr;
1372         int ret;
1373
1374         if (tracing_is_disabled())
1375                 return -ENODEV;
1376
1377         /* Make sure the system still exists */
1378         mutex_lock(&trace_types_lock);
1379         mutex_lock(&event_mutex);
1380         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1381                 list_for_each_entry(dir, &tr->systems, list) {
1382                         if (dir == inode->i_private) {
1383                                 /* Don't open systems with no events */
1384                                 if (dir->nr_events) {
1385                                         __get_system_dir(dir);
1386                                         system = dir->subsystem;
1387                                 }
1388                                 goto exit_loop;
1389                         }
1390                 }
1391         }
1392  exit_loop:
1393         mutex_unlock(&event_mutex);
1394         mutex_unlock(&trace_types_lock);
1395
1396         if (!system)
1397                 return -ENODEV;
1398
1399         /* Some versions of gcc think dir can be uninitialized here */
1400         WARN_ON(!dir);
1401
1402         /* Still need to increment the ref count of the system */
1403         if (trace_array_get(tr) < 0) {
1404                 put_system(dir);
1405                 return -ENODEV;
1406         }
1407
1408         ret = tracing_open_generic(inode, filp);
1409         if (ret < 0) {
1410                 trace_array_put(tr);
1411                 put_system(dir);
1412         }
1413
1414         return ret;
1415 }
1416
1417 static int system_tr_open(struct inode *inode, struct file *filp)
1418 {
1419         struct trace_subsystem_dir *dir;
1420         struct trace_array *tr = inode->i_private;
1421         int ret;
1422
1423         if (tracing_is_disabled())
1424                 return -ENODEV;
1425
1426         if (trace_array_get(tr) < 0)
1427                 return -ENODEV;
1428
1429         /* Make a temporary dir that has no system but points to tr */
1430         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1431         if (!dir) {
1432                 trace_array_put(tr);
1433                 return -ENOMEM;
1434         }
1435
1436         dir->tr = tr;
1437
1438         ret = tracing_open_generic(inode, filp);
1439         if (ret < 0) {
1440                 trace_array_put(tr);
1441                 kfree(dir);
1442                 return ret;
1443         }
1444
1445         filp->private_data = dir;
1446
1447         return 0;
1448 }
1449
1450 static int subsystem_release(struct inode *inode, struct file *file)
1451 {
1452         struct trace_subsystem_dir *dir = file->private_data;
1453
1454         trace_array_put(dir->tr);
1455
1456         /*
1457          * If dir->subsystem is NULL, then this is a temporary
1458          * descriptor that was made for a trace_array to enable
1459          * all subsystems.
1460          */
1461         if (dir->subsystem)
1462                 put_system(dir);
1463         else
1464                 kfree(dir);
1465
1466         return 0;
1467 }
1468
1469 static ssize_t
1470 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1471                       loff_t *ppos)
1472 {
1473         struct trace_subsystem_dir *dir = filp->private_data;
1474         struct event_subsystem *system = dir->subsystem;
1475         struct trace_seq *s;
1476         int r;
1477
1478         if (*ppos)
1479                 return 0;
1480
1481         s = kmalloc(sizeof(*s), GFP_KERNEL);
1482         if (!s)
1483                 return -ENOMEM;
1484
1485         trace_seq_init(s);
1486
1487         print_subsystem_event_filter(system, s);
1488         r = simple_read_from_buffer(ubuf, cnt, ppos,
1489                                     s->buffer, trace_seq_used(s));
1490
1491         kfree(s);
1492
1493         return r;
1494 }
1495
1496 static ssize_t
1497 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1498                        loff_t *ppos)
1499 {
1500         struct trace_subsystem_dir *dir = filp->private_data;
1501         char *buf;
1502         int err;
1503
1504         if (cnt >= PAGE_SIZE)
1505                 return -EINVAL;
1506
1507         buf = memdup_user_nul(ubuf, cnt);
1508         if (IS_ERR(buf))
1509                 return PTR_ERR(buf);
1510
1511         err = apply_subsystem_event_filter(dir, buf);
1512         kfree(buf);
1513         if (err < 0)
1514                 return err;
1515
1516         *ppos += cnt;
1517
1518         return cnt;
1519 }
1520
1521 static ssize_t
1522 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1523 {
1524         int (*func)(struct trace_seq *s) = filp->private_data;
1525         struct trace_seq *s;
1526         int r;
1527
1528         if (*ppos)
1529                 return 0;
1530
1531         s = kmalloc(sizeof(*s), GFP_KERNEL);
1532         if (!s)
1533                 return -ENOMEM;
1534
1535         trace_seq_init(s);
1536
1537         func(s);
1538         r = simple_read_from_buffer(ubuf, cnt, ppos,
1539                                     s->buffer, trace_seq_used(s));
1540
1541         kfree(s);
1542
1543         return r;
1544 }
1545
1546 static int max_pids(struct trace_pid_list *pid_list)
1547 {
1548         return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1549 }
1550
1551 static void ignore_task_cpu(void *data)
1552 {
1553         struct trace_array *tr = data;
1554         struct trace_pid_list *pid_list;
1555
1556         /*
1557          * This function is called by on_each_cpu() while the
1558          * event_mutex is held.
1559          */
1560         pid_list = rcu_dereference_protected(tr->filtered_pids,
1561                                              mutex_is_locked(&event_mutex));
1562
1563         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1564                        check_ignore_pid(pid_list, current));
1565 }
1566
1567 static ssize_t
1568 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1569                        size_t cnt, loff_t *ppos)
1570 {
1571         struct seq_file *m = filp->private_data;
1572         struct trace_array *tr = m->private;
1573         struct trace_pid_list *filtered_pids = NULL;
1574         struct trace_pid_list *pid_list = NULL;
1575         struct trace_event_file *file;
1576         struct trace_parser parser;
1577         unsigned long val;
1578         loff_t this_pos;
1579         ssize_t read = 0;
1580         ssize_t ret = 0;
1581         pid_t pid;
1582         int i;
1583
1584         if (!cnt)
1585                 return 0;
1586
1587         ret = tracing_update_buffers();
1588         if (ret < 0)
1589                 return ret;
1590
1591         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1592                 return -ENOMEM;
1593
1594         mutex_lock(&event_mutex);
1595         /*
1596          * Load as many pids into the array before doing a
1597          * swap from the tr->filtered_pids to the new list.
1598          */
1599         while (cnt > 0) {
1600
1601                 this_pos = 0;
1602
1603                 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1604                 if (ret < 0 || !trace_parser_loaded(&parser))
1605                         break;
1606
1607                 read += ret;
1608                 ubuf += ret;
1609                 cnt -= ret;
1610
1611                 parser.buffer[parser.idx] = 0;
1612
1613                 ret = -EINVAL;
1614                 if (kstrtoul(parser.buffer, 0, &val))
1615                         break;
1616                 if (val > INT_MAX)
1617                         break;
1618
1619                 pid = (pid_t)val;
1620
1621                 ret = -ENOMEM;
1622                 if (!pid_list) {
1623                         pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1624                         if (!pid_list)
1625                                 break;
1626
1627                         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1628                                                         lockdep_is_held(&event_mutex));
1629                         if (filtered_pids)
1630                                 pid_list->order = filtered_pids->order;
1631                         else
1632                                 pid_list->order = 0;
1633
1634                         pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1635                                                                   pid_list->order);
1636                         if (!pid_list->pids)
1637                                 break;
1638
1639                         if (filtered_pids) {
1640                                 pid_list->nr_pids = filtered_pids->nr_pids;
1641                                 memcpy(pid_list->pids, filtered_pids->pids,
1642                                        pid_list->nr_pids * sizeof(pid_t));
1643                         } else
1644                                 pid_list->nr_pids = 0;
1645                 }
1646
1647                 if (pid_list->nr_pids >= max_pids(pid_list)) {
1648                         pid_t *pid_page;
1649
1650                         pid_page = (void *)__get_free_pages(GFP_KERNEL,
1651                                                             pid_list->order + 1);
1652                         if (!pid_page)
1653                                 break;
1654                         memcpy(pid_page, pid_list->pids,
1655                                pid_list->nr_pids * sizeof(pid_t));
1656                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1657
1658                         pid_list->order++;
1659                         pid_list->pids = pid_page;
1660                 }
1661
1662                 pid_list->pids[pid_list->nr_pids++] = pid;
1663                 trace_parser_clear(&parser);
1664                 ret = 0;
1665         }
1666         trace_parser_put(&parser);
1667
1668         if (ret < 0) {
1669                 if (pid_list)
1670                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1671                 kfree(pid_list);
1672                 mutex_unlock(&event_mutex);
1673                 return ret;
1674         }
1675
1676         if (!pid_list) {
1677                 mutex_unlock(&event_mutex);
1678                 return ret;
1679         }
1680
1681         sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1682
1683         /* Remove duplicates */
1684         for (i = 1; i < pid_list->nr_pids; i++) {
1685                 int start = i;
1686
1687                 while (i < pid_list->nr_pids &&
1688                        pid_list->pids[i - 1] == pid_list->pids[i])
1689                         i++;
1690
1691                 if (start != i) {
1692                         if (i < pid_list->nr_pids) {
1693                                 memmove(&pid_list->pids[start], &pid_list->pids[i],
1694                                         (pid_list->nr_pids - i) * sizeof(pid_t));
1695                                 pid_list->nr_pids -= i - start;
1696                                 i = start;
1697                         } else
1698                                 pid_list->nr_pids = start;
1699                 }
1700         }
1701
1702         rcu_assign_pointer(tr->filtered_pids, pid_list);
1703
1704         list_for_each_entry(file, &tr->events, list) {
1705                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1706         }
1707
1708         if (filtered_pids) {
1709                 synchronize_sched();
1710
1711                 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1712                 kfree(filtered_pids);
1713         } else {
1714                 /*
1715                  * Register a probe that is called before all other probes
1716                  * to set ignore_pid if next or prev do not match.
1717                  * Register a probe this is called after all other probes
1718                  * to only keep ignore_pid set if next pid matches.
1719                  */
1720                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1721                                                  tr, INT_MAX);
1722                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1723                                                  tr, 0);
1724
1725                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1726                                                  tr, INT_MAX);
1727                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1728                                                  tr, 0);
1729
1730                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1731                                                      tr, INT_MAX);
1732                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1733                                                      tr, 0);
1734
1735                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1736                                                  tr, INT_MAX);
1737                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1738                                                  tr, 0);
1739         }
1740
1741         /*
1742          * Ignoring of pids is done at task switch. But we have to
1743          * check for those tasks that are currently running.
1744          * Always do this in case a pid was appended or removed.
1745          */
1746         on_each_cpu(ignore_task_cpu, tr, 1);
1747
1748         mutex_unlock(&event_mutex);
1749
1750         ret = read;
1751         *ppos += read;
1752
1753         return ret;
1754 }
1755
1756 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1757 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1758 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1759 static int ftrace_event_release(struct inode *inode, struct file *file);
1760
1761 static const struct seq_operations show_event_seq_ops = {
1762         .start = t_start,
1763         .next = t_next,
1764         .show = t_show,
1765         .stop = t_stop,
1766 };
1767
1768 static const struct seq_operations show_set_event_seq_ops = {
1769         .start = s_start,
1770         .next = s_next,
1771         .show = t_show,
1772         .stop = t_stop,
1773 };
1774
1775 static const struct seq_operations show_set_pid_seq_ops = {
1776         .start = p_start,
1777         .next = p_next,
1778         .show = p_show,
1779         .stop = p_stop,
1780 };
1781
1782 static const struct file_operations ftrace_avail_fops = {
1783         .open = ftrace_event_avail_open,
1784         .read = seq_read,
1785         .llseek = seq_lseek,
1786         .release = seq_release,
1787 };
1788
1789 static const struct file_operations ftrace_set_event_fops = {
1790         .open = ftrace_event_set_open,
1791         .read = seq_read,
1792         .write = ftrace_event_write,
1793         .llseek = seq_lseek,
1794         .release = ftrace_event_release,
1795 };
1796
1797 static const struct file_operations ftrace_set_event_pid_fops = {
1798         .open = ftrace_event_set_pid_open,
1799         .read = seq_read,
1800         .write = ftrace_event_pid_write,
1801         .llseek = seq_lseek,
1802         .release = ftrace_event_release,
1803 };
1804
1805 static const struct file_operations ftrace_enable_fops = {
1806         .open = tracing_open_generic,
1807         .read = event_enable_read,
1808         .write = event_enable_write,
1809         .llseek = default_llseek,
1810 };
1811
1812 static const struct file_operations ftrace_event_format_fops = {
1813         .open = trace_format_open,
1814         .read = seq_read,
1815         .llseek = seq_lseek,
1816         .release = seq_release,
1817 };
1818
1819 static const struct file_operations ftrace_event_id_fops = {
1820         .read = event_id_read,
1821         .llseek = default_llseek,
1822 };
1823
1824 static const struct file_operations ftrace_event_filter_fops = {
1825         .open = tracing_open_generic,
1826         .read = event_filter_read,
1827         .write = event_filter_write,
1828         .llseek = default_llseek,
1829 };
1830
1831 static const struct file_operations ftrace_subsystem_filter_fops = {
1832         .open = subsystem_open,
1833         .read = subsystem_filter_read,
1834         .write = subsystem_filter_write,
1835         .llseek = default_llseek,
1836         .release = subsystem_release,
1837 };
1838
1839 static const struct file_operations ftrace_system_enable_fops = {
1840         .open = subsystem_open,
1841         .read = system_enable_read,
1842         .write = system_enable_write,
1843         .llseek = default_llseek,
1844         .release = subsystem_release,
1845 };
1846
1847 static const struct file_operations ftrace_tr_enable_fops = {
1848         .open = system_tr_open,
1849         .read = system_enable_read,
1850         .write = system_enable_write,
1851         .llseek = default_llseek,
1852         .release = subsystem_release,
1853 };
1854
1855 static const struct file_operations ftrace_show_header_fops = {
1856         .open = tracing_open_generic,
1857         .read = show_header,
1858         .llseek = default_llseek,
1859 };
1860
1861 static int
1862 ftrace_event_open(struct inode *inode, struct file *file,
1863                   const struct seq_operations *seq_ops)
1864 {
1865         struct seq_file *m;
1866         int ret;
1867
1868         ret = seq_open(file, seq_ops);
1869         if (ret < 0)
1870                 return ret;
1871         m = file->private_data;
1872         /* copy tr over to seq ops */
1873         m->private = inode->i_private;
1874
1875         return ret;
1876 }
1877
1878 static int ftrace_event_release(struct inode *inode, struct file *file)
1879 {
1880         struct trace_array *tr = inode->i_private;
1881
1882         trace_array_put(tr);
1883
1884         return seq_release(inode, file);
1885 }
1886
1887 static int
1888 ftrace_event_avail_open(struct inode *inode, struct file *file)
1889 {
1890         const struct seq_operations *seq_ops = &show_event_seq_ops;
1891
1892         return ftrace_event_open(inode, file, seq_ops);
1893 }
1894
1895 static int
1896 ftrace_event_set_open(struct inode *inode, struct file *file)
1897 {
1898         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1899         struct trace_array *tr = inode->i_private;
1900         int ret;
1901
1902         if (trace_array_get(tr) < 0)
1903                 return -ENODEV;
1904
1905         if ((file->f_mode & FMODE_WRITE) &&
1906             (file->f_flags & O_TRUNC))
1907                 ftrace_clear_events(tr);
1908
1909         ret = ftrace_event_open(inode, file, seq_ops);
1910         if (ret < 0)
1911                 trace_array_put(tr);
1912         return ret;
1913 }
1914
1915 static int
1916 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1917 {
1918         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1919         struct trace_array *tr = inode->i_private;
1920         int ret;
1921
1922         if (trace_array_get(tr) < 0)
1923                 return -ENODEV;
1924
1925         if ((file->f_mode & FMODE_WRITE) &&
1926             (file->f_flags & O_TRUNC))
1927                 ftrace_clear_event_pids(tr);
1928
1929         ret = ftrace_event_open(inode, file, seq_ops);
1930         if (ret < 0)
1931                 trace_array_put(tr);
1932         return ret;
1933 }
1934
1935 static struct event_subsystem *
1936 create_new_subsystem(const char *name)
1937 {
1938         struct event_subsystem *system;
1939
1940         /* need to create new entry */
1941         system = kmalloc(sizeof(*system), GFP_KERNEL);
1942         if (!system)
1943                 return NULL;
1944
1945         system->ref_count = 1;
1946
1947         /* Only allocate if dynamic (kprobes and modules) */
1948         system->name = kstrdup_const(name, GFP_KERNEL);
1949         if (!system->name)
1950                 goto out_free;
1951
1952         system->filter = NULL;
1953
1954         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1955         if (!system->filter)
1956                 goto out_free;
1957
1958         list_add(&system->list, &event_subsystems);
1959
1960         return system;
1961
1962  out_free:
1963         kfree_const(system->name);
1964         kfree(system);
1965         return NULL;
1966 }
1967
1968 static struct dentry *
1969 event_subsystem_dir(struct trace_array *tr, const char *name,
1970                     struct trace_event_file *file, struct dentry *parent)
1971 {
1972         struct trace_subsystem_dir *dir;
1973         struct event_subsystem *system;
1974         struct dentry *entry;
1975
1976         /* First see if we did not already create this dir */
1977         list_for_each_entry(dir, &tr->systems, list) {
1978                 system = dir->subsystem;
1979                 if (strcmp(system->name, name) == 0) {
1980                         dir->nr_events++;
1981                         file->system = dir;
1982                         return dir->entry;
1983                 }
1984         }
1985
1986         /* Now see if the system itself exists. */
1987         list_for_each_entry(system, &event_subsystems, list) {
1988                 if (strcmp(system->name, name) == 0)
1989                         break;
1990         }
1991         /* Reset system variable when not found */
1992         if (&system->list == &event_subsystems)
1993                 system = NULL;
1994
1995         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1996         if (!dir)
1997                 goto out_fail;
1998
1999         if (!system) {
2000                 system = create_new_subsystem(name);
2001                 if (!system)
2002                         goto out_free;
2003         } else
2004                 __get_system(system);
2005
2006         dir->entry = tracefs_create_dir(name, parent);
2007         if (!dir->entry) {
2008                 pr_warn("Failed to create system directory %s\n", name);
2009                 __put_system(system);
2010                 goto out_free;
2011         }
2012
2013         dir->tr = tr;
2014         dir->ref_count = 1;
2015         dir->nr_events = 1;
2016         dir->subsystem = system;
2017         file->system = dir;
2018
2019         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2020                                     &ftrace_subsystem_filter_fops);
2021         if (!entry) {
2022                 kfree(system->filter);
2023                 system->filter = NULL;
2024                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2025         }
2026
2027         trace_create_file("enable", 0644, dir->entry, dir,
2028                           &ftrace_system_enable_fops);
2029
2030         list_add(&dir->list, &tr->systems);
2031
2032         return dir->entry;
2033
2034  out_free:
2035         kfree(dir);
2036  out_fail:
2037         /* Only print this message if failed on memory allocation */
2038         if (!dir || !system)
2039                 pr_warn("No memory to create event subsystem %s\n", name);
2040         return NULL;
2041 }
2042
2043 static int
2044 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2045 {
2046         struct trace_event_call *call = file->event_call;
2047         struct trace_array *tr = file->tr;
2048         struct list_head *head;
2049         struct dentry *d_events;
2050         const char *name;
2051         int ret;
2052
2053         /*
2054          * If the trace point header did not define TRACE_SYSTEM
2055          * then the system would be called "TRACE_SYSTEM".
2056          */
2057         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2058                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2059                 if (!d_events)
2060                         return -ENOMEM;
2061         } else
2062                 d_events = parent;
2063
2064         name = trace_event_name(call);
2065         file->dir = tracefs_create_dir(name, d_events);
2066         if (!file->dir) {
2067                 pr_warn("Could not create tracefs '%s' directory\n", name);
2068                 return -1;
2069         }
2070
2071         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2072                 trace_create_file("enable", 0644, file->dir, file,
2073                                   &ftrace_enable_fops);
2074
2075 #ifdef CONFIG_PERF_EVENTS
2076         if (call->event.type && call->class->reg)
2077                 trace_create_file("id", 0444, file->dir,
2078                                   (void *)(long)call->event.type,
2079                                   &ftrace_event_id_fops);
2080 #endif
2081
2082         /*
2083          * Other events may have the same class. Only update
2084          * the fields if they are not already defined.
2085          */
2086         head = trace_get_fields(call);
2087         if (list_empty(head)) {
2088                 ret = call->class->define_fields(call);
2089                 if (ret < 0) {
2090                         pr_warn("Could not initialize trace point events/%s\n",
2091                                 name);
2092                         return -1;
2093                 }
2094         }
2095         trace_create_file("filter", 0644, file->dir, file,
2096                           &ftrace_event_filter_fops);
2097
2098         /*
2099          * Only event directories that can be enabled should have
2100          * triggers.
2101          */
2102         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2103                 trace_create_file("trigger", 0644, file->dir, file,
2104                                   &event_trigger_fops);
2105
2106         trace_create_file("format", 0444, file->dir, call,
2107                           &ftrace_event_format_fops);
2108
2109         return 0;
2110 }
2111
2112 static void remove_event_from_tracers(struct trace_event_call *call)
2113 {
2114         struct trace_event_file *file;
2115         struct trace_array *tr;
2116
2117         do_for_each_event_file_safe(tr, file) {
2118                 if (file->event_call != call)
2119                         continue;
2120
2121                 remove_event_file_dir(file);
2122                 /*
2123                  * The do_for_each_event_file_safe() is
2124                  * a double loop. After finding the call for this
2125                  * trace_array, we use break to jump to the next
2126                  * trace_array.
2127                  */
2128                 break;
2129         } while_for_each_event_file();
2130 }
2131
2132 static void event_remove(struct trace_event_call *call)
2133 {
2134         struct trace_array *tr;
2135         struct trace_event_file *file;
2136
2137         do_for_each_event_file(tr, file) {
2138                 if (file->event_call != call)
2139                         continue;
2140                 ftrace_event_enable_disable(file, 0);
2141                 /*
2142                  * The do_for_each_event_file() is
2143                  * a double loop. After finding the call for this
2144                  * trace_array, we use break to jump to the next
2145                  * trace_array.
2146                  */
2147                 break;
2148         } while_for_each_event_file();
2149
2150         if (call->event.funcs)
2151                 __unregister_trace_event(&call->event);
2152         remove_event_from_tracers(call);
2153         list_del(&call->list);
2154 }
2155
2156 static int event_init(struct trace_event_call *call)
2157 {
2158         int ret = 0;
2159         const char *name;
2160
2161         name = trace_event_name(call);
2162         if (WARN_ON(!name))
2163                 return -EINVAL;
2164
2165         if (call->class->raw_init) {
2166                 ret = call->class->raw_init(call);
2167                 if (ret < 0 && ret != -ENOSYS)
2168                         pr_warn("Could not initialize trace events/%s\n", name);
2169         }
2170
2171         return ret;
2172 }
2173
2174 static int
2175 __register_event(struct trace_event_call *call, struct module *mod)
2176 {
2177         int ret;
2178
2179         ret = event_init(call);
2180         if (ret < 0)
2181                 return ret;
2182
2183         list_add(&call->list, &ftrace_events);
2184         call->mod = mod;
2185
2186         return 0;
2187 }
2188
2189 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2190 {
2191         int rlen;
2192         int elen;
2193
2194         /* Find the length of the enum value as a string */
2195         elen = snprintf(ptr, 0, "%ld", map->enum_value);
2196         /* Make sure there's enough room to replace the string with the value */
2197         if (len < elen)
2198                 return NULL;
2199
2200         snprintf(ptr, elen + 1, "%ld", map->enum_value);
2201
2202         /* Get the rest of the string of ptr */
2203         rlen = strlen(ptr + len);
2204         memmove(ptr + elen, ptr + len, rlen);
2205         /* Make sure we end the new string */
2206         ptr[elen + rlen] = 0;
2207
2208         return ptr + elen;
2209 }
2210
2211 static void update_event_printk(struct trace_event_call *call,
2212                                 struct trace_enum_map *map)
2213 {
2214         char *ptr;
2215         int quote = 0;
2216         int len = strlen(map->enum_string);
2217
2218         for (ptr = call->print_fmt; *ptr; ptr++) {
2219                 if (*ptr == '\\') {
2220                         ptr++;
2221                         /* paranoid */
2222                         if (!*ptr)
2223                                 break;
2224                         continue;
2225                 }
2226                 if (*ptr == '"') {
2227                         quote ^= 1;
2228                         continue;
2229                 }
2230                 if (quote)
2231                         continue;
2232                 if (isdigit(*ptr)) {
2233                         /* skip numbers */
2234                         do {
2235                                 ptr++;
2236                                 /* Check for alpha chars like ULL */
2237                         } while (isalnum(*ptr));
2238                         if (!*ptr)
2239                                 break;
2240                         /*
2241                          * A number must have some kind of delimiter after
2242                          * it, and we can ignore that too.
2243                          */
2244                         continue;
2245                 }
2246                 if (isalpha(*ptr) || *ptr == '_') {
2247                         if (strncmp(map->enum_string, ptr, len) == 0 &&
2248                             !isalnum(ptr[len]) && ptr[len] != '_') {
2249                                 ptr = enum_replace(ptr, map, len);
2250                                 /* Hmm, enum string smaller than value */
2251                                 if (WARN_ON_ONCE(!ptr))
2252                                         return;
2253                                 /*
2254                                  * No need to decrement here, as enum_replace()
2255                                  * returns the pointer to the character passed
2256                                  * the enum, and two enums can not be placed
2257                                  * back to back without something in between.
2258                                  * We can skip that something in between.
2259                                  */
2260                                 continue;
2261                         }
2262                 skip_more:
2263                         do {
2264                                 ptr++;
2265                         } while (isalnum(*ptr) || *ptr == '_');
2266                         if (!*ptr)
2267                                 break;
2268                         /*
2269                          * If what comes after this variable is a '.' or
2270                          * '->' then we can continue to ignore that string.
2271                          */
2272                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2273                                 ptr += *ptr == '.' ? 1 : 2;
2274                                 if (!*ptr)
2275                                         break;
2276                                 goto skip_more;
2277                         }
2278                         /*
2279                          * Once again, we can skip the delimiter that came
2280                          * after the string.
2281                          */
2282                         continue;
2283                 }
2284         }
2285 }
2286
2287 void trace_event_enum_update(struct trace_enum_map **map, int len)
2288 {
2289         struct trace_event_call *call, *p;
2290         const char *last_system = NULL;
2291         int last_i;
2292         int i;
2293
2294         down_write(&trace_event_sem);
2295         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2296                 /* events are usually grouped together with systems */
2297                 if (!last_system || call->class->system != last_system) {
2298                         last_i = 0;
2299                         last_system = call->class->system;
2300                 }
2301
2302                 for (i = last_i; i < len; i++) {
2303                         if (call->class->system == map[i]->system) {
2304                                 /* Save the first system if need be */
2305                                 if (!last_i)
2306                                         last_i = i;
2307                                 update_event_printk(call, map[i]);
2308                         }
2309                 }
2310         }
2311         up_write(&trace_event_sem);
2312 }
2313
2314 static struct trace_event_file *
2315 trace_create_new_event(struct trace_event_call *call,
2316                        struct trace_array *tr)
2317 {
2318         struct trace_event_file *file;
2319
2320         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2321         if (!file)
2322                 return NULL;
2323
2324         file->event_call = call;
2325         file->tr = tr;
2326         atomic_set(&file->sm_ref, 0);
2327         atomic_set(&file->tm_ref, 0);
2328         INIT_LIST_HEAD(&file->triggers);
2329         list_add(&file->list, &tr->events);
2330
2331         return file;
2332 }
2333
2334 /* Add an event to a trace directory */
2335 static int
2336 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2337 {
2338         struct trace_event_file *file;
2339
2340         file = trace_create_new_event(call, tr);
2341         if (!file)
2342                 return -ENOMEM;
2343
2344         return event_create_dir(tr->event_dir, file);
2345 }
2346
2347 /*
2348  * Just create a decriptor for early init. A descriptor is required
2349  * for enabling events at boot. We want to enable events before
2350  * the filesystem is initialized.
2351  */
2352 static __init int
2353 __trace_early_add_new_event(struct trace_event_call *call,
2354                             struct trace_array *tr)
2355 {
2356         struct trace_event_file *file;
2357
2358         file = trace_create_new_event(call, tr);
2359         if (!file)
2360                 return -ENOMEM;
2361
2362         return 0;
2363 }
2364
2365 struct ftrace_module_file_ops;
2366 static void __add_event_to_tracers(struct trace_event_call *call);
2367
2368 /* Add an additional event_call dynamically */
2369 int trace_add_event_call(struct trace_event_call *call)
2370 {
2371         int ret;
2372         mutex_lock(&trace_types_lock);
2373         mutex_lock(&event_mutex);
2374
2375         ret = __register_event(call, NULL);
2376         if (ret >= 0)
2377                 __add_event_to_tracers(call);
2378
2379         mutex_unlock(&event_mutex);
2380         mutex_unlock(&trace_types_lock);
2381         return ret;
2382 }
2383
2384 /*
2385  * Must be called under locking of trace_types_lock, event_mutex and
2386  * trace_event_sem.
2387  */
2388 static void __trace_remove_event_call(struct trace_event_call *call)
2389 {
2390         event_remove(call);
2391         trace_destroy_fields(call);
2392         free_event_filter(call->filter);
2393         call->filter = NULL;
2394 }
2395
2396 static int probe_remove_event_call(struct trace_event_call *call)
2397 {
2398         struct trace_array *tr;
2399         struct trace_event_file *file;
2400
2401 #ifdef CONFIG_PERF_EVENTS
2402         if (call->perf_refcount)
2403                 return -EBUSY;
2404 #endif
2405         do_for_each_event_file(tr, file) {
2406                 if (file->event_call != call)
2407                         continue;
2408                 /*
2409                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2410                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2411                  * TRACE_REG_UNREGISTER.
2412                  */
2413                 if (file->flags & EVENT_FILE_FL_ENABLED)
2414                         return -EBUSY;
2415                 /*
2416                  * The do_for_each_event_file_safe() is
2417                  * a double loop. After finding the call for this
2418                  * trace_array, we use break to jump to the next
2419                  * trace_array.
2420                  */
2421                 break;
2422         } while_for_each_event_file();
2423
2424         __trace_remove_event_call(call);
2425
2426         return 0;
2427 }
2428
2429 /* Remove an event_call */
2430 int trace_remove_event_call(struct trace_event_call *call)
2431 {
2432         int ret;
2433
2434         mutex_lock(&trace_types_lock);
2435         mutex_lock(&event_mutex);
2436         down_write(&trace_event_sem);
2437         ret = probe_remove_event_call(call);
2438         up_write(&trace_event_sem);
2439         mutex_unlock(&event_mutex);
2440         mutex_unlock(&trace_types_lock);
2441
2442         return ret;
2443 }
2444
2445 #define for_each_event(event, start, end)                       \
2446         for (event = start;                                     \
2447              (unsigned long)event < (unsigned long)end;         \
2448              event++)
2449
2450 #ifdef CONFIG_MODULES
2451
2452 static void trace_module_add_events(struct module *mod)
2453 {
2454         struct trace_event_call **call, **start, **end;
2455
2456         if (!mod->num_trace_events)
2457                 return;
2458
2459         /* Don't add infrastructure for mods without tracepoints */
2460         if (trace_module_has_bad_taint(mod)) {
2461                 pr_err("%s: module has bad taint, not creating trace events\n",
2462                        mod->name);
2463                 return;
2464         }
2465
2466         start = mod->trace_events;
2467         end = mod->trace_events + mod->num_trace_events;
2468
2469         for_each_event(call, start, end) {
2470                 __register_event(*call, mod);
2471                 __add_event_to_tracers(*call);
2472         }
2473 }
2474
2475 static void trace_module_remove_events(struct module *mod)
2476 {
2477         struct trace_event_call *call, *p;
2478         bool clear_trace = false;
2479
2480         down_write(&trace_event_sem);
2481         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2482                 if (call->mod == mod) {
2483                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2484                                 clear_trace = true;
2485                         __trace_remove_event_call(call);
2486                 }
2487         }
2488         up_write(&trace_event_sem);
2489
2490         /*
2491          * It is safest to reset the ring buffer if the module being unloaded
2492          * registered any events that were used. The only worry is if
2493          * a new module gets loaded, and takes on the same id as the events
2494          * of this module. When printing out the buffer, traced events left
2495          * over from this module may be passed to the new module events and
2496          * unexpected results may occur.
2497          */
2498         if (clear_trace)
2499                 tracing_reset_all_online_cpus();
2500 }
2501
2502 static int trace_module_notify(struct notifier_block *self,
2503                                unsigned long val, void *data)
2504 {
2505         struct module *mod = data;
2506
2507         mutex_lock(&trace_types_lock);
2508         mutex_lock(&event_mutex);
2509         switch (val) {
2510         case MODULE_STATE_COMING:
2511                 trace_module_add_events(mod);
2512                 break;
2513         case MODULE_STATE_GOING:
2514                 trace_module_remove_events(mod);
2515                 break;
2516         }
2517         mutex_unlock(&event_mutex);
2518         mutex_unlock(&trace_types_lock);
2519
2520         return 0;
2521 }
2522
2523 static struct notifier_block trace_module_nb = {
2524         .notifier_call = trace_module_notify,
2525         .priority = 1, /* higher than trace.c module notify */
2526 };
2527 #endif /* CONFIG_MODULES */
2528
2529 /* Create a new event directory structure for a trace directory. */
2530 static void
2531 __trace_add_event_dirs(struct trace_array *tr)
2532 {
2533         struct trace_event_call *call;
2534         int ret;
2535
2536         list_for_each_entry(call, &ftrace_events, list) {
2537                 ret = __trace_add_new_event(call, tr);
2538                 if (ret < 0)
2539                         pr_warn("Could not create directory for event %s\n",
2540                                 trace_event_name(call));
2541         }
2542 }
2543
2544 struct trace_event_file *
2545 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2546 {
2547         struct trace_event_file *file;
2548         struct trace_event_call *call;
2549         const char *name;
2550
2551         list_for_each_entry(file, &tr->events, list) {
2552
2553                 call = file->event_call;
2554                 name = trace_event_name(call);
2555
2556                 if (!name || !call->class || !call->class->reg)
2557                         continue;
2558
2559                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2560                         continue;
2561
2562                 if (strcmp(event, name) == 0 &&
2563                     strcmp(system, call->class->system) == 0)
2564                         return file;
2565         }
2566         return NULL;
2567 }
2568
2569 #ifdef CONFIG_DYNAMIC_FTRACE
2570
2571 /* Avoid typos */
2572 #define ENABLE_EVENT_STR        "enable_event"
2573 #define DISABLE_EVENT_STR       "disable_event"
2574
2575 struct event_probe_data {
2576         struct trace_event_file *file;
2577         unsigned long                   count;
2578         int                             ref;
2579         bool                            enable;
2580 };
2581
2582 static void
2583 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2584 {
2585         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2586         struct event_probe_data *data = *pdata;
2587
2588         if (!data)
2589                 return;
2590
2591         if (data->enable)
2592                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2593         else
2594                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2595 }
2596
2597 static void
2598 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2599 {
2600         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2601         struct event_probe_data *data = *pdata;
2602
2603         if (!data)
2604                 return;
2605
2606         if (!data->count)
2607                 return;
2608
2609         /* Skip if the event is in a state we want to switch to */
2610         if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2611                 return;
2612
2613         if (data->count != -1)
2614                 (data->count)--;
2615
2616         event_enable_probe(ip, parent_ip, _data);
2617 }
2618
2619 static int
2620 event_enable_print(struct seq_file *m, unsigned long ip,
2621                       struct ftrace_probe_ops *ops, void *_data)
2622 {
2623         struct event_probe_data *data = _data;
2624
2625         seq_printf(m, "%ps:", (void *)ip);
2626
2627         seq_printf(m, "%s:%s:%s",
2628                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2629                    data->file->event_call->class->system,
2630                    trace_event_name(data->file->event_call));
2631
2632         if (data->count == -1)
2633                 seq_puts(m, ":unlimited\n");
2634         else
2635                 seq_printf(m, ":count=%ld\n", data->count);
2636
2637         return 0;
2638 }
2639
2640 static int
2641 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2642                   void **_data)
2643 {
2644         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2645         struct event_probe_data *data = *pdata;
2646
2647         data->ref++;
2648         return 0;
2649 }
2650
2651 static void
2652 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2653                   void **_data)
2654 {
2655         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2656         struct event_probe_data *data = *pdata;
2657
2658         if (WARN_ON_ONCE(data->ref <= 0))
2659                 return;
2660
2661         data->ref--;
2662         if (!data->ref) {
2663                 /* Remove the SOFT_MODE flag */
2664                 __ftrace_event_enable_disable(data->file, 0, 1);
2665                 module_put(data->file->event_call->mod);
2666                 kfree(data);
2667         }
2668         *pdata = NULL;
2669 }
2670
2671 static struct ftrace_probe_ops event_enable_probe_ops = {
2672         .func                   = event_enable_probe,
2673         .print                  = event_enable_print,
2674         .init                   = event_enable_init,
2675         .free                   = event_enable_free,
2676 };
2677
2678 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2679         .func                   = event_enable_count_probe,
2680         .print                  = event_enable_print,
2681         .init                   = event_enable_init,
2682         .free                   = event_enable_free,
2683 };
2684
2685 static struct ftrace_probe_ops event_disable_probe_ops = {
2686         .func                   = event_enable_probe,
2687         .print                  = event_enable_print,
2688         .init                   = event_enable_init,
2689         .free                   = event_enable_free,
2690 };
2691
2692 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2693         .func                   = event_enable_count_probe,
2694         .print                  = event_enable_print,
2695         .init                   = event_enable_init,
2696         .free                   = event_enable_free,
2697 };
2698
2699 static int
2700 event_enable_func(struct ftrace_hash *hash,
2701                   char *glob, char *cmd, char *param, int enabled)
2702 {
2703         struct trace_array *tr = top_trace_array();
2704         struct trace_event_file *file;
2705         struct ftrace_probe_ops *ops;
2706         struct event_probe_data *data;
2707         const char *system;
2708         const char *event;
2709         char *number;
2710         bool enable;
2711         int ret;
2712
2713         if (!tr)
2714                 return -ENODEV;
2715
2716         /* hash funcs only work with set_ftrace_filter */
2717         if (!enabled || !param)
2718                 return -EINVAL;
2719
2720         system = strsep(&param, ":");
2721         if (!param)
2722                 return -EINVAL;
2723
2724         event = strsep(&param, ":");
2725
2726         mutex_lock(&event_mutex);
2727
2728         ret = -EINVAL;
2729         file = find_event_file(tr, system, event);
2730         if (!file)
2731                 goto out;
2732
2733         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2734
2735         if (enable)
2736                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2737         else
2738                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2739
2740         if (glob[0] == '!') {
2741                 unregister_ftrace_function_probe_func(glob+1, ops);
2742                 ret = 0;
2743                 goto out;
2744         }
2745
2746         ret = -ENOMEM;
2747         data = kzalloc(sizeof(*data), GFP_KERNEL);
2748         if (!data)
2749                 goto out;
2750
2751         data->enable = enable;
2752         data->count = -1;
2753         data->file = file;
2754
2755         if (!param)
2756                 goto out_reg;
2757
2758         number = strsep(&param, ":");
2759
2760         ret = -EINVAL;
2761         if (!strlen(number))
2762                 goto out_free;
2763
2764         /*
2765          * We use the callback data field (which is a pointer)
2766          * as our counter.
2767          */
2768         ret = kstrtoul(number, 0, &data->count);
2769         if (ret)
2770                 goto out_free;
2771
2772  out_reg:
2773         /* Don't let event modules unload while probe registered */
2774         ret = try_module_get(file->event_call->mod);
2775         if (!ret) {
2776                 ret = -EBUSY;
2777                 goto out_free;
2778         }
2779
2780         ret = __ftrace_event_enable_disable(file, 1, 1);
2781         if (ret < 0)
2782                 goto out_put;
2783         ret = register_ftrace_function_probe(glob, ops, data);
2784         /*
2785          * The above returns on success the # of functions enabled,
2786          * but if it didn't find any functions it returns zero.
2787          * Consider no functions a failure too.
2788          */
2789         if (!ret) {
2790                 ret = -ENOENT;
2791                 goto out_disable;
2792         } else if (ret < 0)
2793                 goto out_disable;
2794         /* Just return zero, not the number of enabled functions */
2795         ret = 0;
2796  out:
2797         mutex_unlock(&event_mutex);
2798         return ret;
2799
2800  out_disable:
2801         __ftrace_event_enable_disable(file, 0, 1);
2802  out_put:
2803         module_put(file->event_call->mod);
2804  out_free:
2805         kfree(data);
2806         goto out;
2807 }
2808
2809 static struct ftrace_func_command event_enable_cmd = {
2810         .name                   = ENABLE_EVENT_STR,
2811         .func                   = event_enable_func,
2812 };
2813
2814 static struct ftrace_func_command event_disable_cmd = {
2815         .name                   = DISABLE_EVENT_STR,
2816         .func                   = event_enable_func,
2817 };
2818
2819 static __init int register_event_cmds(void)
2820 {
2821         int ret;
2822
2823         ret = register_ftrace_command(&event_enable_cmd);
2824         if (WARN_ON(ret < 0))
2825                 return ret;
2826         ret = register_ftrace_command(&event_disable_cmd);
2827         if (WARN_ON(ret < 0))
2828                 unregister_ftrace_command(&event_enable_cmd);
2829         return ret;
2830 }
2831 #else
2832 static inline int register_event_cmds(void) { return 0; }
2833 #endif /* CONFIG_DYNAMIC_FTRACE */
2834
2835 /*
2836  * The top level array has already had its trace_event_file
2837  * descriptors created in order to allow for early events to
2838  * be recorded. This function is called after the tracefs has been
2839  * initialized, and we now have to create the files associated
2840  * to the events.
2841  */
2842 static __init void
2843 __trace_early_add_event_dirs(struct trace_array *tr)
2844 {
2845         struct trace_event_file *file;
2846         int ret;
2847
2848
2849         list_for_each_entry(file, &tr->events, list) {
2850                 ret = event_create_dir(tr->event_dir, file);
2851                 if (ret < 0)
2852                         pr_warn("Could not create directory for event %s\n",
2853                                 trace_event_name(file->event_call));
2854         }
2855 }
2856
2857 /*
2858  * For early boot up, the top trace array requires to have
2859  * a list of events that can be enabled. This must be done before
2860  * the filesystem is set up in order to allow events to be traced
2861  * early.
2862  */
2863 static __init void
2864 __trace_early_add_events(struct trace_array *tr)
2865 {
2866         struct trace_event_call *call;
2867         int ret;
2868
2869         list_for_each_entry(call, &ftrace_events, list) {
2870                 /* Early boot up should not have any modules loaded */
2871                 if (WARN_ON_ONCE(call->mod))
2872                         continue;
2873
2874                 ret = __trace_early_add_new_event(call, tr);
2875                 if (ret < 0)
2876                         pr_warn("Could not create early event %s\n",
2877                                 trace_event_name(call));
2878         }
2879 }
2880
2881 /* Remove the event directory structure for a trace directory. */
2882 static void
2883 __trace_remove_event_dirs(struct trace_array *tr)
2884 {
2885         struct trace_event_file *file, *next;
2886
2887         list_for_each_entry_safe(file, next, &tr->events, list)
2888                 remove_event_file_dir(file);
2889 }
2890
2891 static void __add_event_to_tracers(struct trace_event_call *call)
2892 {
2893         struct trace_array *tr;
2894
2895         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2896                 __trace_add_new_event(call, tr);
2897 }
2898
2899 extern struct trace_event_call *__start_ftrace_events[];
2900 extern struct trace_event_call *__stop_ftrace_events[];
2901
2902 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2903
2904 static __init int setup_trace_event(char *str)
2905 {
2906         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2907         ring_buffer_expanded = true;
2908         tracing_selftest_disabled = true;
2909
2910         return 1;
2911 }
2912 __setup("trace_event=", setup_trace_event);
2913
2914 /* Expects to have event_mutex held when called */
2915 static int
2916 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2917 {
2918         struct dentry *d_events;
2919         struct dentry *entry;
2920
2921         entry = tracefs_create_file("set_event", 0644, parent,
2922                                     tr, &ftrace_set_event_fops);
2923         if (!entry) {
2924                 pr_warn("Could not create tracefs 'set_event' entry\n");
2925                 return -ENOMEM;
2926         }
2927
2928         d_events = tracefs_create_dir("events", parent);
2929         if (!d_events) {
2930                 pr_warn("Could not create tracefs 'events' directory\n");
2931                 return -ENOMEM;
2932         }
2933
2934         entry = tracefs_create_file("set_event_pid", 0644, parent,
2935                                     tr, &ftrace_set_event_pid_fops);
2936
2937         /* ring buffer internal formats */
2938         trace_create_file("header_page", 0444, d_events,
2939                           ring_buffer_print_page_header,
2940                           &ftrace_show_header_fops);
2941
2942         trace_create_file("header_event", 0444, d_events,
2943                           ring_buffer_print_entry_header,
2944                           &ftrace_show_header_fops);
2945
2946         trace_create_file("enable", 0644, d_events,
2947                           tr, &ftrace_tr_enable_fops);
2948
2949         tr->event_dir = d_events;
2950
2951         return 0;
2952 }
2953
2954 /**
2955  * event_trace_add_tracer - add a instance of a trace_array to events
2956  * @parent: The parent dentry to place the files/directories for events in
2957  * @tr: The trace array associated with these events
2958  *
2959  * When a new instance is created, it needs to set up its events
2960  * directory, as well as other files associated with events. It also
2961  * creates the event hierachry in the @parent/events directory.
2962  *
2963  * Returns 0 on success.
2964  */
2965 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2966 {
2967         int ret;
2968
2969         mutex_lock(&event_mutex);
2970
2971         ret = create_event_toplevel_files(parent, tr);
2972         if (ret)
2973                 goto out_unlock;
2974
2975         down_write(&trace_event_sem);
2976         __trace_add_event_dirs(tr);
2977         up_write(&trace_event_sem);
2978
2979  out_unlock:
2980         mutex_unlock(&event_mutex);
2981
2982         return ret;
2983 }
2984
2985 /*
2986  * The top trace array already had its file descriptors created.
2987  * Now the files themselves need to be created.
2988  */
2989 static __init int
2990 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2991 {
2992         int ret;
2993
2994         mutex_lock(&event_mutex);
2995
2996         ret = create_event_toplevel_files(parent, tr);
2997         if (ret)
2998                 goto out_unlock;
2999
3000         down_write(&trace_event_sem);
3001         __trace_early_add_event_dirs(tr);
3002         up_write(&trace_event_sem);
3003
3004  out_unlock:
3005         mutex_unlock(&event_mutex);
3006
3007         return ret;
3008 }
3009
3010 int event_trace_del_tracer(struct trace_array *tr)
3011 {
3012         mutex_lock(&event_mutex);
3013
3014         /* Disable any event triggers and associated soft-disabled events */
3015         clear_event_triggers(tr);
3016
3017         /* Clear the pid list */
3018         __ftrace_clear_event_pids(tr);
3019
3020         /* Disable any running events */
3021         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3022
3023         /* Access to events are within rcu_read_lock_sched() */
3024         synchronize_sched();
3025
3026         down_write(&trace_event_sem);
3027         __trace_remove_event_dirs(tr);
3028         tracefs_remove_recursive(tr->event_dir);
3029         up_write(&trace_event_sem);
3030
3031         tr->event_dir = NULL;
3032
3033         mutex_unlock(&event_mutex);
3034
3035         return 0;
3036 }
3037
3038 static __init int event_trace_memsetup(void)
3039 {
3040         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3041         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3042         return 0;
3043 }
3044
3045 static __init void
3046 early_enable_events(struct trace_array *tr, bool disable_first)
3047 {
3048         char *buf = bootup_event_buf;
3049         char *token;
3050         int ret;
3051
3052         while (true) {
3053                 token = strsep(&buf, ",");
3054
3055                 if (!token)
3056                         break;
3057
3058                 if (*token) {
3059                         /* Restarting syscalls requires that we stop them first */
3060                         if (disable_first)
3061                                 ftrace_set_clr_event(tr, token, 0);
3062
3063                         ret = ftrace_set_clr_event(tr, token, 1);
3064                         if (ret)
3065                                 pr_warn("Failed to enable trace event: %s\n", token);
3066                 }
3067
3068                 /* Put back the comma to allow this to be called again */
3069                 if (buf)
3070                         *(buf - 1) = ',';
3071         }
3072 }
3073
3074 static __init int event_trace_enable(void)
3075 {
3076         struct trace_array *tr = top_trace_array();
3077         struct trace_event_call **iter, *call;
3078         int ret;
3079
3080         if (!tr)
3081                 return -ENODEV;
3082
3083         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3084
3085                 call = *iter;
3086                 ret = event_init(call);
3087                 if (!ret)
3088                         list_add(&call->list, &ftrace_events);
3089         }
3090
3091         /*
3092          * We need the top trace array to have a working set of trace
3093          * points at early init, before the debug files and directories
3094          * are created. Create the file entries now, and attach them
3095          * to the actual file dentries later.
3096          */
3097         __trace_early_add_events(tr);
3098
3099         early_enable_events(tr, false);
3100
3101         trace_printk_start_comm();
3102
3103         register_event_cmds();
3104
3105         register_trigger_cmds();
3106
3107         return 0;
3108 }
3109
3110 /*
3111  * event_trace_enable() is called from trace_event_init() first to
3112  * initialize events and perhaps start any events that are on the
3113  * command line. Unfortunately, there are some events that will not
3114  * start this early, like the system call tracepoints that need
3115  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3116  * is called before pid 1 starts, and this flag is never set, making
3117  * the syscall tracepoint never get reached, but the event is enabled
3118  * regardless (and not doing anything).
3119  */
3120 static __init int event_trace_enable_again(void)
3121 {
3122         struct trace_array *tr;
3123
3124         tr = top_trace_array();
3125         if (!tr)
3126                 return -ENODEV;
3127
3128         early_enable_events(tr, true);
3129
3130         return 0;
3131 }
3132
3133 early_initcall(event_trace_enable_again);
3134
3135 static __init int event_trace_init(void)
3136 {
3137         struct trace_array *tr;
3138         struct dentry *d_tracer;
3139         struct dentry *entry;
3140         int ret;
3141
3142         tr = top_trace_array();
3143         if (!tr)
3144                 return -ENODEV;
3145
3146         d_tracer = tracing_init_dentry();
3147         if (IS_ERR(d_tracer))
3148                 return 0;
3149
3150         entry = tracefs_create_file("available_events", 0444, d_tracer,
3151                                     tr, &ftrace_avail_fops);
3152         if (!entry)
3153                 pr_warn("Could not create tracefs 'available_events' entry\n");
3154
3155         if (trace_define_generic_fields())
3156                 pr_warn("tracing: Failed to allocated generic fields");
3157
3158         if (trace_define_common_fields())
3159                 pr_warn("tracing: Failed to allocate common fields");
3160
3161         ret = early_event_add_tracer(d_tracer, tr);
3162         if (ret)
3163                 return ret;
3164
3165 #ifdef CONFIG_MODULES
3166         ret = register_module_notifier(&trace_module_nb);
3167         if (ret)
3168                 pr_warn("Failed to register trace events module notifier\n");
3169 #endif
3170         return 0;
3171 }
3172
3173 void __init trace_event_init(void)
3174 {
3175         event_trace_memsetup();
3176         init_ftrace_syscalls();
3177         event_trace_enable();
3178 }
3179
3180 fs_initcall(event_trace_init);
3181
3182 #ifdef CONFIG_FTRACE_STARTUP_TEST
3183
3184 static DEFINE_SPINLOCK(test_spinlock);
3185 static DEFINE_SPINLOCK(test_spinlock_irq);
3186 static DEFINE_MUTEX(test_mutex);
3187
3188 static __init void test_work(struct work_struct *dummy)
3189 {
3190         spin_lock(&test_spinlock);
3191         spin_lock_irq(&test_spinlock_irq);
3192         udelay(1);
3193         spin_unlock_irq(&test_spinlock_irq);
3194         spin_unlock(&test_spinlock);
3195
3196         mutex_lock(&test_mutex);
3197         msleep(1);
3198         mutex_unlock(&test_mutex);
3199 }
3200
3201 static __init int event_test_thread(void *unused)
3202 {
3203         void *test_malloc;
3204
3205         test_malloc = kmalloc(1234, GFP_KERNEL);
3206         if (!test_malloc)
3207                 pr_info("failed to kmalloc\n");
3208
3209         schedule_on_each_cpu(test_work);
3210
3211         kfree(test_malloc);
3212
3213         set_current_state(TASK_INTERRUPTIBLE);
3214         while (!kthread_should_stop()) {
3215                 schedule();
3216                 set_current_state(TASK_INTERRUPTIBLE);
3217         }
3218         __set_current_state(TASK_RUNNING);
3219
3220         return 0;
3221 }
3222
3223 /*
3224  * Do various things that may trigger events.
3225  */
3226 static __init void event_test_stuff(void)
3227 {
3228         struct task_struct *test_thread;
3229
3230         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3231         msleep(1);
3232         kthread_stop(test_thread);
3233 }
3234
3235 /*
3236  * For every trace event defined, we will test each trace point separately,
3237  * and then by groups, and finally all trace points.
3238  */
3239 static __init void event_trace_self_tests(void)
3240 {
3241         struct trace_subsystem_dir *dir;
3242         struct trace_event_file *file;
3243         struct trace_event_call *call;
3244         struct event_subsystem *system;
3245         struct trace_array *tr;
3246         int ret;
3247
3248         tr = top_trace_array();
3249         if (!tr)
3250                 return;
3251
3252         pr_info("Running tests on trace events:\n");
3253
3254         list_for_each_entry(file, &tr->events, list) {
3255
3256                 call = file->event_call;
3257
3258                 /* Only test those that have a probe */
3259                 if (!call->class || !call->class->probe)
3260                         continue;
3261
3262 /*
3263  * Testing syscall events here is pretty useless, but
3264  * we still do it if configured. But this is time consuming.
3265  * What we really need is a user thread to perform the
3266  * syscalls as we test.
3267  */
3268 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3269                 if (call->class->system &&
3270                     strcmp(call->class->system, "syscalls") == 0)
3271                         continue;
3272 #endif
3273
3274                 pr_info("Testing event %s: ", trace_event_name(call));
3275
3276                 /*
3277                  * If an event is already enabled, someone is using
3278                  * it and the self test should not be on.
3279                  */
3280                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3281                         pr_warn("Enabled event during self test!\n");
3282                         WARN_ON_ONCE(1);
3283                         continue;
3284                 }
3285
3286                 ftrace_event_enable_disable(file, 1);
3287                 event_test_stuff();
3288                 ftrace_event_enable_disable(file, 0);
3289
3290                 pr_cont("OK\n");
3291         }
3292
3293         /* Now test at the sub system level */
3294
3295         pr_info("Running tests on trace event systems:\n");
3296
3297         list_for_each_entry(dir, &tr->systems, list) {
3298
3299                 system = dir->subsystem;
3300
3301                 /* the ftrace system is special, skip it */
3302                 if (strcmp(system->name, "ftrace") == 0)
3303                         continue;
3304
3305                 pr_info("Testing event system %s: ", system->name);
3306
3307                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3308                 if (WARN_ON_ONCE(ret)) {
3309                         pr_warn("error enabling system %s\n",
3310                                 system->name);
3311                         continue;
3312                 }
3313
3314                 event_test_stuff();
3315
3316                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3317                 if (WARN_ON_ONCE(ret)) {
3318                         pr_warn("error disabling system %s\n",
3319                                 system->name);
3320                         continue;
3321                 }
3322
3323                 pr_cont("OK\n");
3324         }
3325
3326         /* Test with all events enabled */
3327
3328         pr_info("Running tests on all trace events:\n");
3329         pr_info("Testing all events: ");
3330
3331         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3332         if (WARN_ON_ONCE(ret)) {
3333                 pr_warn("error enabling all events\n");
3334                 return;
3335         }
3336
3337         event_test_stuff();
3338
3339         /* reset sysname */
3340         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3341         if (WARN_ON_ONCE(ret)) {
3342                 pr_warn("error disabling all events\n");
3343                 return;
3344         }
3345
3346         pr_cont("OK\n");
3347 }
3348
3349 #ifdef CONFIG_FUNCTION_TRACER
3350
3351 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3352
3353 static struct trace_array *event_tr;
3354
3355 static void __init
3356 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3357                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3358 {
3359         struct ring_buffer_event *event;
3360         struct ring_buffer *buffer;
3361         struct ftrace_entry *entry;
3362         unsigned long flags;
3363         long disabled;
3364         int cpu;
3365         int pc;
3366
3367         pc = preempt_count();
3368         preempt_disable_notrace();
3369         cpu = raw_smp_processor_id();
3370         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3371
3372         if (disabled != 1)
3373                 goto out;
3374
3375         local_save_flags(flags);
3376
3377         event = trace_current_buffer_lock_reserve(&buffer,
3378                                                   TRACE_FN, sizeof(*entry),
3379                                                   flags, pc);
3380         if (!event)
3381                 goto out;
3382         entry   = ring_buffer_event_data(event);
3383         entry->ip                       = ip;
3384         entry->parent_ip                = parent_ip;
3385
3386         trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3387
3388  out:
3389         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3390         preempt_enable_notrace();
3391 }
3392
3393 static struct ftrace_ops trace_ops __initdata  =
3394 {
3395         .func = function_test_events_call,
3396         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3397 };
3398
3399 static __init void event_trace_self_test_with_function(void)
3400 {
3401         int ret;
3402         event_tr = top_trace_array();
3403         if (WARN_ON(!event_tr))
3404                 return;
3405         ret = register_ftrace_function(&trace_ops);
3406         if (WARN_ON(ret < 0)) {
3407                 pr_info("Failed to enable function tracer for event tests\n");
3408                 return;
3409         }
3410         pr_info("Running tests again, along with the function tracer\n");
3411         event_trace_self_tests();
3412         unregister_ftrace_function(&trace_ops);
3413 }
3414 #else
3415 static __init void event_trace_self_test_with_function(void)
3416 {
3417 }
3418 #endif
3419
3420 static __init int event_trace_self_tests_init(void)
3421 {
3422         if (!tracing_selftest_disabled) {
3423                 event_trace_self_tests();
3424                 event_trace_self_test_with_function();
3425         }
3426
3427         return 0;
3428 }
3429
3430 late_initcall(event_trace_self_tests_init);
3431
3432 #endif