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