kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #define pr_fmt(fmt) fmt
  12
  13 #include <linux/workqueue.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kthread.h>
  16 #include <linux/tracefs.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/bsearch.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/sort.h>
  22 #include <linux/slab.h>
  23 #include <linux/delay.h>
  24
  25 #include <trace/events/sched.h>
  26
  27 #include <asm/setup.h>
  28
  29 #include "trace_output.h"
  30
  31 #undef TRACE_SYSTEM
  32 #define TRACE_SYSTEM "TRACE_SYSTEM"
  33
  34 DEFINE_MUTEX(event_mutex);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_generic_fields);
  38 static LIST_HEAD(ftrace_common_fields);
  39
  40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  41
  42 static struct kmem_cache *field_cachep;
  43 static struct kmem_cache *file_cachep;
  44
  45 static inline int system_refcount(struct event_subsystem *system)
  46 {
  47         return system->ref_count;
  48 }
  49
  50 static int system_refcount_inc(struct event_subsystem *system)
  51 {
  52         return system->ref_count++;
  53 }
  54
  55 static int system_refcount_dec(struct event_subsystem *system)
  56 {
  57         return --system->ref_count;
  58 }
  59
  60 /* Double loops, do not use break, only goto's work */
  61 #define do_for_each_event_file(tr, file)                        \
  62         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  63                 list_for_each_entry(file, &tr->events, list)
  64
  65 #define do_for_each_event_file_safe(tr, file)                   \
  66         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  67                 struct trace_event_file *___n;                          \
  68                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  69
  70 #define while_for_each_event_file()             \
  71         }
  72
  73 static struct list_head *
  74 trace_get_fields(struct trace_event_call *event_call)
  75 {
  76         if (!event_call->class->get_fields)
  77                 return &event_call->class->fields;
  78         return event_call->class->get_fields(event_call);
  79 }
  80
  81 static struct ftrace_event_field *
  82 __find_event_field(struct list_head *head, char *name)
  83 {
  84         struct ftrace_event_field *field;
  85
  86         list_for_each_entry(field, head, link) {
  87                 if (!strcmp(field->name, name))
  88                         return field;
  89         }
  90
  91         return NULL;
  92 }
  93
  94 struct ftrace_event_field *
  95 trace_find_event_field(struct trace_event_call *call, char *name)
  96 {
  97         struct ftrace_event_field *field;
  98         struct list_head *head;
  99
 100         head = trace_get_fields(call);
 101         field = __find_event_field(head, name);
 102         if (field)
 103                 return field;
 104
 105         field = __find_event_field(&ftrace_generic_fields, name);
 106         if (field)
 107                 return field;
 108
 109         return __find_event_field(&ftrace_common_fields, name);
 110 }
 111
 112 static int __trace_define_field(struct list_head *head, const char *type,
 113                                 const char *name, int offset, int size,
 114                                 int is_signed, int filter_type)
 115 {
 116         struct ftrace_event_field *field;
 117
 118         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 119         if (!field)
 120                 return -ENOMEM;
 121
 122         field->name = name;
 123         field->type = type;
 124
 125         if (filter_type == FILTER_OTHER)
 126                 field->filter_type = filter_assign_type(type);
 127         else
 128                 field->filter_type = filter_type;
 129
 130         field->offset = offset;
 131         field->size = size;
 132         field->is_signed = is_signed;
 133
 134         list_add(&field->link, head);
 135
 136         return 0;
 137 }
 138
 139 int trace_define_field(struct trace_event_call *call, const char *type,
 140                        const char *name, int offset, int size, int is_signed,
 141                        int filter_type)
 142 {
 143         struct list_head *head;
 144
 145         if (WARN_ON(!call->class))
 146                 return 0;
 147
 148         head = trace_get_fields(call);
 149         return __trace_define_field(head, type, name, offset, size,
 150                                     is_signed, filter_type);
 151 }
 152 EXPORT_SYMBOL_GPL(trace_define_field);
 153
 154 #define __generic_field(type, item, filter_type)                        \
 155         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 156                                    #item, 0, 0, is_signed_type(type),   \
 157                                    filter_type);                        \
 158         if (ret)                                                        \
 159                 return ret;
 160
 161 #define __common_field(type, item)                                      \
 162         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 163                                    "common_" #item,                     \
 164                                    offsetof(typeof(ent), item),         \
 165                                    sizeof(ent.item),                    \
 166                                    is_signed_type(type), FILTER_OTHER); \
 167         if (ret)                                                        \
 168                 return ret;
 169
 170 static int trace_define_generic_fields(void)
 171 {
 172         int ret;
 173
 174         __generic_field(int, CPU, FILTER_CPU);
 175         __generic_field(int, cpu, FILTER_CPU);
 176         __generic_field(char *, COMM, FILTER_COMM);
 177         __generic_field(char *, comm, FILTER_COMM);
 178
 179         return ret;
 180 }
 181
 182 static int trace_define_common_fields(void)
 183 {
 184         int ret;
 185         struct trace_entry ent;
 186
 187         __common_field(unsigned short, type);
 188         __common_field(unsigned char, flags);
 189         __common_field(unsigned char, preempt_count);
 190         __common_field(int, pid);
 191
 192         return ret;
 193 }
 194
 195 static void trace_destroy_fields(struct trace_event_call *call)
 196 {
 197         struct ftrace_event_field *field, *next;
 198         struct list_head *head;
 199
 200         head = trace_get_fields(call);
 201         list_for_each_entry_safe(field, next, head, link) {
 202                 list_del(&field->link);
 203                 kmem_cache_free(field_cachep, field);
 204         }
 205 }
 206
 207 int trace_event_raw_init(struct trace_event_call *call)
 208 {
 209         int id;
 210
 211         id = register_trace_event(&call->event);
 212         if (!id)
 213                 return -ENODEV;
 214
 215         return 0;
 216 }
 217 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 218
 219 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 220 {
 221         struct trace_array *tr = trace_file->tr;
 222         struct trace_array_cpu *data;
 223         struct trace_pid_list *pid_list;
 224
 225         pid_list = rcu_dereference_sched(tr->filtered_pids);
 226         if (!pid_list)
 227                 return false;
 228
 229         data = this_cpu_ptr(tr->trace_buffer.data);
 230
 231         return data->ignore_pid;
 232 }
 233 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 234
 235 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 236                                  struct trace_event_file *trace_file,
 237                                  unsigned long len)
 238 {
 239         struct trace_event_call *event_call = trace_file->event_call;
 240
 241         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 242             trace_event_ignore_this_pid(trace_file))
 243                 return NULL;
 244
 245         local_save_flags(fbuffer->flags);
 246         fbuffer->pc = preempt_count();
 247         fbuffer->trace_file = trace_file;
 248
 249         fbuffer->event =
 250                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 251                                                 event_call->event.type, len,
 252                                                 fbuffer->flags, fbuffer->pc);
 253         if (!fbuffer->event)
 254                 return NULL;
 255
 256         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 257         return fbuffer->entry;
 258 }
 259 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 260
 261 static DEFINE_SPINLOCK(tracepoint_iter_lock);
 262
 263 static void output_printk(struct trace_event_buffer *fbuffer)
 264 {
 265         struct trace_event_call *event_call;
 266         struct trace_event *event;
 267         unsigned long flags;
 268         struct trace_iterator *iter = tracepoint_print_iter;
 269
 270         if (!iter)
 271                 return;
 272
 273         event_call = fbuffer->trace_file->event_call;
 274         if (!event_call || !event_call->event.funcs ||
 275             !event_call->event.funcs->trace)
 276                 return;
 277
 278         event = &fbuffer->trace_file->event_call->event;
 279
 280         spin_lock_irqsave(&tracepoint_iter_lock, flags);
 281         trace_seq_init(&iter->seq);
 282         iter->ent = fbuffer->entry;
 283         event_call->event.funcs->trace(iter, 0, event);
 284         trace_seq_putc(&iter->seq, 0);
 285         printk("%s", iter->seq.buffer);
 286
 287         spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
 288 }
 289
 290 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
 291 {
 292         if (tracepoint_printk)
 293                 output_printk(fbuffer);
 294
 295         event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
 296                                     fbuffer->event, fbuffer->entry,
 297                                     fbuffer->flags, fbuffer->pc);
 298 }
 299 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
 300
 301 int trace_event_reg(struct trace_event_call *call,
 302                     enum trace_reg type, void *data)
 303 {
 304         struct trace_event_file *file = data;
 305
 306         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 307         switch (type) {
 308         case TRACE_REG_REGISTER:
 309                 return tracepoint_probe_register(call->tp,
 310                                                  call->class->probe,
 311                                                  file);
 312         case TRACE_REG_UNREGISTER:
 313                 tracepoint_probe_unregister(call->tp,
 314                                             call->class->probe,
 315                                             file);
 316                 return 0;
 317
 318 #ifdef CONFIG_PERF_EVENTS
 319         case TRACE_REG_PERF_REGISTER:
 320                 return tracepoint_probe_register(call->tp,
 321                                                  call->class->perf_probe,
 322                                                  call);
 323         case TRACE_REG_PERF_UNREGISTER:
 324                 tracepoint_probe_unregister(call->tp,
 325                                             call->class->perf_probe,
 326                                             call);
 327                 return 0;
 328         case TRACE_REG_PERF_OPEN:
 329         case TRACE_REG_PERF_CLOSE:
 330         case TRACE_REG_PERF_ADD:
 331         case TRACE_REG_PERF_DEL:
 332                 return 0;
 333 #endif
 334         }
 335         return 0;
 336 }
 337 EXPORT_SYMBOL_GPL(trace_event_reg);
 338
 339 void trace_event_enable_cmd_record(bool enable)
 340 {
 341         struct trace_event_file *file;
 342         struct trace_array *tr;
 343
 344         mutex_lock(&event_mutex);
 345         do_for_each_event_file(tr, file) {
 346
 347                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 348                         continue;
 349
 350                 if (enable) {
 351                         tracing_start_cmdline_record();
 352                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 353                 } else {
 354                         tracing_stop_cmdline_record();
 355                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 356                 }
 357         } while_for_each_event_file();
 358         mutex_unlock(&event_mutex);
 359 }
 360
 361 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 362                                          int enable, int soft_disable)
 363 {
 364         struct trace_event_call *call = file->event_call;
 365         struct trace_array *tr = file->tr;
 366         int ret = 0;
 367         int disable;
 368
 369         switch (enable) {
 370         case 0:
 371                 /*
 372                  * When soft_disable is set and enable is cleared, the sm_ref
 373                  * reference counter is decremented. If it reaches 0, we want
 374                  * to clear the SOFT_DISABLED flag but leave the event in the
 375                  * state that it was. That is, if the event was enabled and
 376                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 377                  * is set we do not want the event to be enabled before we
 378                  * clear the bit.
 379                  *
 380                  * When soft_disable is not set but the SOFT_MODE flag is,
 381                  * we do nothing. Do not disable the tracepoint, otherwise
 382                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 383                  */
 384                 if (soft_disable) {
 385                         if (atomic_dec_return(&file->sm_ref) > 0)
 386                                 break;
 387                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 388                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 389                 } else
 390                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 391
 392                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 393                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 394                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 395                                 tracing_stop_cmdline_record();
 396                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 397                         }
 398                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 399                 }
 400                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 401                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 402                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 403                 else
 404                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 405                 break;
 406         case 1:
 407                 /*
 408                  * When soft_disable is set and enable is set, we want to
 409                  * register the tracepoint for the event, but leave the event
 410                  * as is. That means, if the event was already enabled, we do
 411                  * nothing (but set SOFT_MODE). If the event is disabled, we
 412                  * set SOFT_DISABLED before enabling the event tracepoint, so
 413                  * it still seems to be disabled.
 414                  */
 415                 if (!soft_disable)
 416                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 417                 else {
 418                         if (atomic_inc_return(&file->sm_ref) > 1)
 419                                 break;
 420                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 421                 }
 422
 423                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 424
 425                         /* Keep the event disabled, when going to SOFT_MODE. */
 426                         if (soft_disable)
 427                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 428
 429                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 430                                 tracing_start_cmdline_record();
 431                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 432                         }
 433                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 434                         if (ret) {
 435                                 tracing_stop_cmdline_record();
 436                                 pr_info("event trace: Could not enable event "
 437                                         "%s\n", trace_event_name(call));
 438                                 break;
 439                         }
 440                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 441
 442                         /* WAS_ENABLED gets set but never cleared. */
 443                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 444                 }
 445                 break;
 446         }
 447
 448         return ret;
 449 }
 450
 451 int trace_event_enable_disable(struct trace_event_file *file,
 452                                int enable, int soft_disable)
 453 {
 454         return __ftrace_event_enable_disable(file, enable, soft_disable);
 455 }
 456
 457 static int ftrace_event_enable_disable(struct trace_event_file *file,
 458                                        int enable)
 459 {
 460         return __ftrace_event_enable_disable(file, enable, 0);
 461 }
 462
 463 static void ftrace_clear_events(struct trace_array *tr)
 464 {
 465         struct trace_event_file *file;
 466
 467         mutex_lock(&event_mutex);
 468         list_for_each_entry(file, &tr->events, list) {
 469                 ftrace_event_enable_disable(file, 0);
 470         }
 471         mutex_unlock(&event_mutex);
 472 }
 473
 474 static int cmp_pid(const void *key, const void *elt)
 475 {
 476         const pid_t *search_pid = key;
 477         const pid_t *pid = elt;
 478
 479         if (*search_pid == *pid)
 480                 return 0;
 481         if (*search_pid < *pid)
 482                 return -1;
 483         return 1;
 484 }
 485
 486 static bool
 487 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
 488 {
 489         pid_t search_pid;
 490         pid_t *pid;
 491
 492         /*
 493          * Return false, because if filtered_pids does not exist,
 494          * all pids are good to trace.
 495          */
 496         if (!filtered_pids)
 497                 return false;
 498
 499         search_pid = task->pid;
 500
 501         pid = bsearch(&search_pid, filtered_pids->pids,
 502                       filtered_pids->nr_pids, sizeof(pid_t),
 503                       cmp_pid);
 504         if (!pid)
 505                 return true;
 506
 507         return false;
 508 }
 509
 510 static void
 511 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 512                     struct task_struct *prev, struct task_struct *next)
 513 {
 514         struct trace_array *tr = data;
 515         struct trace_pid_list *pid_list;
 516
 517         pid_list = rcu_dereference_sched(tr->filtered_pids);
 518
 519         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 520                        check_ignore_pid(pid_list, prev) &&
 521                        check_ignore_pid(pid_list, next));
 522 }
 523
 524 static void
 525 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 526                     struct task_struct *prev, struct task_struct *next)
 527 {
 528         struct trace_array *tr = data;
 529         struct trace_pid_list *pid_list;
 530
 531         pid_list = rcu_dereference_sched(tr->filtered_pids);
 532
 533         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 534                        check_ignore_pid(pid_list, next));
 535 }
 536
 537 static void
 538 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 539 {
 540         struct trace_array *tr = data;
 541         struct trace_pid_list *pid_list;
 542
 543         /* Nothing to do if we are already tracing */
 544         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 545                 return;
 546
 547         pid_list = rcu_dereference_sched(tr->filtered_pids);
 548
 549         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 550                        check_ignore_pid(pid_list, task));
 551 }
 552
 553 static void
 554 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 555 {
 556         struct trace_array *tr = data;
 557         struct trace_pid_list *pid_list;
 558
 559         /* Nothing to do if we are not tracing */
 560         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 561                 return;
 562
 563         pid_list = rcu_dereference_sched(tr->filtered_pids);
 564
 565         /* Set tracing if current is enabled */
 566         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 567                        check_ignore_pid(pid_list, current));
 568 }
 569
 570 static void __ftrace_clear_event_pids(struct trace_array *tr)
 571 {
 572         struct trace_pid_list *pid_list;
 573         struct trace_event_file *file;
 574         int cpu;
 575
 576         pid_list = rcu_dereference_protected(tr->filtered_pids,
 577                                              lockdep_is_held(&event_mutex));
 578         if (!pid_list)
 579                 return;
 580
 581         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 582         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 583
 584         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 585         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 586
 587         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 588         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 589
 590         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 591         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 592
 593         list_for_each_entry(file, &tr->events, list) {
 594                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 595         }
 596
 597         for_each_possible_cpu(cpu)
 598                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 599
 600         rcu_assign_pointer(tr->filtered_pids, NULL);
 601
 602         /* Wait till all users are no longer using pid filtering */
 603         synchronize_sched();
 604
 605         free_pages((unsigned long)pid_list->pids, pid_list->order);
 606         kfree(pid_list);
 607 }
 608
 609 static void ftrace_clear_event_pids(struct trace_array *tr)
 610 {
 611         mutex_lock(&event_mutex);
 612         __ftrace_clear_event_pids(tr);
 613         mutex_unlock(&event_mutex);
 614 }
 615
 616 static void __put_system(struct event_subsystem *system)
 617 {
 618         struct event_filter *filter = system->filter;
 619
 620         WARN_ON_ONCE(system_refcount(system) == 0);
 621         if (system_refcount_dec(system))
 622                 return;
 623
 624         list_del(&system->list);
 625
 626         if (filter) {
 627                 kfree(filter->filter_string);
 628                 kfree(filter);
 629         }
 630         kfree_const(system->name);
 631         kfree(system);
 632 }
 633
 634 static void __get_system(struct event_subsystem *system)
 635 {
 636         WARN_ON_ONCE(system_refcount(system) == 0);
 637         system_refcount_inc(system);
 638 }
 639
 640 static void __get_system_dir(struct trace_subsystem_dir *dir)
 641 {
 642         WARN_ON_ONCE(dir->ref_count == 0);
 643         dir->ref_count++;
 644         __get_system(dir->subsystem);
 645 }
 646
 647 static void __put_system_dir(struct trace_subsystem_dir *dir)
 648 {
 649         WARN_ON_ONCE(dir->ref_count == 0);
 650         /* If the subsystem is about to be freed, the dir must be too */
 651         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 652
 653         __put_system(dir->subsystem);
 654         if (!--dir->ref_count)
 655                 kfree(dir);
 656 }
 657
 658 static void put_system(struct trace_subsystem_dir *dir)
 659 {
 660         mutex_lock(&event_mutex);
 661         __put_system_dir(dir);
 662         mutex_unlock(&event_mutex);
 663 }
 664
 665 static void remove_subsystem(struct trace_subsystem_dir *dir)
 666 {
 667         if (!dir)
 668                 return;
 669
 670         if (!--dir->nr_events) {
 671                 tracefs_remove_recursive(dir->entry);
 672                 list_del(&dir->list);
 673                 __put_system_dir(dir);
 674         }
 675 }
 676
 677 static void remove_event_file_dir(struct trace_event_file *file)
 678 {
 679         struct dentry *dir = file->dir;
 680         struct dentry *child;
 681
 682         if (dir) {
 683                 spin_lock(&dir->d_lock);        /* probably unneeded */
 684                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 685                         if (d_really_is_positive(child))        /* probably unneeded */
 686                                 d_inode(child)->i_private = NULL;
 687                 }
 688                 spin_unlock(&dir->d_lock);
 689
 690                 tracefs_remove_recursive(dir);
 691         }
 692
 693         list_del(&file->list);
 694         remove_subsystem(file->system);
 695         free_event_filter(file->filter);
 696         kmem_cache_free(file_cachep, file);
 697 }
 698
 699 /*
 700  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 701  */
 702 static int
 703 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 704                               const char *sub, const char *event, int set)
 705 {
 706         struct trace_event_file *file;
 707         struct trace_event_call *call;
 708         const char *name;
 709         int ret = -EINVAL;
 710
 711         list_for_each_entry(file, &tr->events, list) {
 712
 713                 call = file->event_call;
 714                 name = trace_event_name(call);
 715
 716                 if (!name || !call->class || !call->class->reg)
 717                         continue;
 718
 719                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 720                         continue;
 721
 722                 if (match &&
 723                     strcmp(match, name) != 0 &&
 724                     strcmp(match, call->class->system) != 0)
 725                         continue;
 726
 727                 if (sub && strcmp(sub, call->class->system) != 0)
 728                         continue;
 729
 730                 if (event && strcmp(event, name) != 0)
 731                         continue;
 732
 733                 ftrace_event_enable_disable(file, set);
 734
 735                 ret = 0;
 736         }
 737
 738         return ret;
 739 }
 740
 741 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 742                                   const char *sub, const char *event, int set)
 743 {
 744         int ret;
 745
 746         mutex_lock(&event_mutex);
 747         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 748         mutex_unlock(&event_mutex);
 749
 750         return ret;
 751 }
 752
 753 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 754 {
 755         char *event = NULL, *sub = NULL, *match;
 756         int ret;
 757
 758         if (!tr)
 759                 return -ENOENT;
 760         /*
 761          * The buf format can be <subsystem>:<event-name>
 762          *  *:<event-name> means any event by that name.
 763          *  :<event-name> is the same.
 764          *
 765          *  <subsystem>:* means all events in that subsystem
 766          *  <subsystem>: means the same.
 767          *
 768          *  <name> (no ':') means all events in a subsystem with
 769          *  the name <name> or any event that matches <name>
 770          */
 771
 772         match = strsep(&buf, ":");
 773         if (buf) {
 774                 sub = match;
 775                 event = buf;
 776                 match = NULL;
 777
 778                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 779                         sub = NULL;
 780                 if (!strlen(event) || strcmp(event, "*") == 0)
 781                         event = NULL;
 782         }
 783
 784         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 785
 786         /* Put back the colon to allow this to be called again */
 787         if (buf)
 788                 *(buf - 1) = ':';
 789
 790         return ret;
 791 }
 792
 793 /**
 794  * trace_set_clr_event - enable or disable an event
 795  * @system: system name to match (NULL for any system)
 796  * @event: event name to match (NULL for all events, within system)
 797  * @set: 1 to enable, 0 to disable
 798  *
 799  * This is a way for other parts of the kernel to enable or disable
 800  * event recording.
 801  *
 802  * Returns 0 on success, -EINVAL if the parameters do not match any
 803  * registered events.
 804  */
 805 int trace_set_clr_event(const char *system, const char *event, int set)
 806 {
 807         struct trace_array *tr = top_trace_array();
 808
 809         if (!tr)
 810                 return -ENODEV;
 811
 812         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 813 }
 814 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 815
 816 /* 128 should be much more than enough */
 817 #define EVENT_BUF_SIZE          127
 818
 819 static ssize_t
 820 ftrace_event_write(struct file *file, const char __user *ubuf,
 821                    size_t cnt, loff_t *ppos)
 822 {
 823         struct trace_parser parser;
 824         struct seq_file *m = file->private_data;
 825         struct trace_array *tr = m->private;
 826         ssize_t read, ret;
 827
 828         if (!cnt)
 829                 return 0;
 830
 831         ret = tracing_update_buffers();
 832         if (ret < 0)
 833                 return ret;
 834
 835         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 836                 return -ENOMEM;
 837
 838         read = trace_get_user(&parser, ubuf, cnt, ppos);
 839
 840         if (read >= 0 && trace_parser_loaded((&parser))) {
 841                 int set = 1;
 842
 843                 if (*parser.buffer == '!')
 844                         set = 0;
 845
 846                 parser.buffer[parser.idx] = 0;
 847
 848                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 849                 if (ret)
 850                         goto out_put;
 851         }
 852
 853         ret = read;
 854
 855  out_put:
 856         trace_parser_put(&parser);
 857
 858         return ret;
 859 }
 860
 861 static void *
 862 t_next(struct seq_file *m, void *v, loff_t *pos)
 863 {
 864         struct trace_event_file *file = v;
 865         struct trace_event_call *call;
 866         struct trace_array *tr = m->private;
 867
 868         (*pos)++;
 869
 870         list_for_each_entry_continue(file, &tr->events, list) {
 871                 call = file->event_call;
 872                 /*
 873                  * The ftrace subsystem is for showing formats only.
 874                  * They can not be enabled or disabled via the event files.
 875                  */
 876                 if (call->class && call->class->reg &&
 877                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 878                         return file;
 879         }
 880
 881         return NULL;
 882 }
 883
 884 static void *t_start(struct seq_file *m, loff_t *pos)
 885 {
 886         struct trace_event_file *file;
 887         struct trace_array *tr = m->private;
 888         loff_t l;
 889
 890         mutex_lock(&event_mutex);
 891
 892         file = list_entry(&tr->events, struct trace_event_file, list);
 893         for (l = 0; l <= *pos; ) {
 894                 file = t_next(m, file, &l);
 895                 if (!file)
 896                         break;
 897         }
 898         return file;
 899 }
 900
 901 static void *
 902 s_next(struct seq_file *m, void *v, loff_t *pos)
 903 {
 904         struct trace_event_file *file = v;
 905         struct trace_array *tr = m->private;
 906
 907         (*pos)++;
 908
 909         list_for_each_entry_continue(file, &tr->events, list) {
 910                 if (file->flags & EVENT_FILE_FL_ENABLED)
 911                         return file;
 912         }
 913
 914         return NULL;
 915 }
 916
 917 static void *s_start(struct seq_file *m, loff_t *pos)
 918 {
 919         struct trace_event_file *file;
 920         struct trace_array *tr = m->private;
 921         loff_t l;
 922
 923         mutex_lock(&event_mutex);
 924
 925         file = list_entry(&tr->events, struct trace_event_file, list);
 926         for (l = 0; l <= *pos; ) {
 927                 file = s_next(m, file, &l);
 928                 if (!file)
 929                         break;
 930         }
 931         return file;
 932 }
 933
 934 static int t_show(struct seq_file *m, void *v)
 935 {
 936         struct trace_event_file *file = v;
 937         struct trace_event_call *call = file->event_call;
 938
 939         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 940                 seq_printf(m, "%s:", call->class->system);
 941         seq_printf(m, "%s\n", trace_event_name(call));
 942
 943         return 0;
 944 }
 945
 946 static void t_stop(struct seq_file *m, void *p)
 947 {
 948         mutex_unlock(&event_mutex);
 949 }
 950
 951 static void *p_start(struct seq_file *m, loff_t *pos)
 952         __acquires(RCU)
 953 {
 954         struct trace_pid_list *pid_list;
 955         struct trace_array *tr = m->private;
 956
 957         /*
 958          * Grab the mutex, to keep calls to p_next() having the same
 959          * tr->filtered_pids as p_start() has.
 960          * If we just passed the tr->filtered_pids around, then RCU would
 961          * have been enough, but doing that makes things more complex.
 962          */
 963         mutex_lock(&event_mutex);
 964         rcu_read_lock_sched();
 965
 966         pid_list = rcu_dereference_sched(tr->filtered_pids);
 967
 968         if (!pid_list || *pos >= pid_list->nr_pids)
 969                 return NULL;
 970
 971         return (void *)&pid_list->pids[*pos];
 972 }
 973
 974 static void p_stop(struct seq_file *m, void *p)
 975         __releases(RCU)
 976 {
 977         rcu_read_unlock_sched();
 978         mutex_unlock(&event_mutex);
 979 }
 980
 981 static void *
 982 p_next(struct seq_file *m, void *v, loff_t *pos)
 983 {
 984         struct trace_array *tr = m->private;
 985         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 986
 987         (*pos)++;
 988
 989         if (*pos >= pid_list->nr_pids)
 990                 return NULL;
 991
 992         return (void *)&pid_list->pids[*pos];
 993 }
 994
 995 static int p_show(struct seq_file *m, void *v)
 996 {
 997         pid_t *pid = v;
 998
 999         seq_printf(m, "%d\n", *pid);
1000         return 0;
1001 }
1002
1003 static ssize_t
1004 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1005                   loff_t *ppos)
1006 {
1007         struct trace_event_file *file;
1008         unsigned long flags;
1009         char buf[4] = "0";
1010
1011         mutex_lock(&event_mutex);
1012         file = event_file_data(filp);
1013         if (likely(file))
1014                 flags = file->flags;
1015         mutex_unlock(&event_mutex);
1016
1017         if (!file)
1018                 return -ENODEV;
1019
1020         if (flags & EVENT_FILE_FL_ENABLED &&
1021             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1022                 strcpy(buf, "1");
1023
1024         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1025             flags & EVENT_FILE_FL_SOFT_MODE)
1026                 strcat(buf, "*");
1027
1028         strcat(buf, "\n");
1029
1030         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1031 }
1032
1033 static ssize_t
1034 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1035                    loff_t *ppos)
1036 {
1037         struct trace_event_file *file;
1038         unsigned long val;
1039         int ret;
1040
1041         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1042         if (ret)
1043                 return ret;
1044
1045         ret = tracing_update_buffers();
1046         if (ret < 0)
1047                 return ret;
1048
1049         switch (val) {
1050         case 0:
1051         case 1:
1052                 ret = -ENODEV;
1053                 mutex_lock(&event_mutex);
1054                 file = event_file_data(filp);
1055                 if (likely(file))
1056                         ret = ftrace_event_enable_disable(file, val);
1057                 mutex_unlock(&event_mutex);
1058                 break;
1059
1060         default:
1061                 return -EINVAL;
1062         }
1063
1064         *ppos += cnt;
1065
1066         return ret ? ret : cnt;
1067 }
1068
1069 static ssize_t
1070 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1071                    loff_t *ppos)
1072 {
1073         const char set_to_char[4] = { '?', '0', '1', 'X' };
1074         struct trace_subsystem_dir *dir = filp->private_data;
1075         struct event_subsystem *system = dir->subsystem;
1076         struct trace_event_call *call;
1077         struct trace_event_file *file;
1078         struct trace_array *tr = dir->tr;
1079         char buf[2];
1080         int set = 0;
1081         int ret;
1082
1083         mutex_lock(&event_mutex);
1084         list_for_each_entry(file, &tr->events, list) {
1085                 call = file->event_call;
1086                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1087                     !trace_event_name(call) || !call->class || !call->class->reg)
1088                         continue;
1089
1090                 if (system && strcmp(call->class->system, system->name) != 0)
1091                         continue;
1092
1093                 /*
1094                  * We need to find out if all the events are set
1095                  * or if all events or cleared, or if we have
1096                  * a mixture.
1097                  */
1098                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1099
1100                 /*
1101                  * If we have a mixture, no need to look further.
1102                  */
1103                 if (set == 3)
1104                         break;
1105         }
1106         mutex_unlock(&event_mutex);
1107
1108         buf[0] = set_to_char[set];
1109         buf[1] = '\n';
1110
1111         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1112
1113         return ret;
1114 }
1115
1116 static ssize_t
1117 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1118                     loff_t *ppos)
1119 {
1120         struct trace_subsystem_dir *dir = filp->private_data;
1121         struct event_subsystem *system = dir->subsystem;
1122         const char *name = NULL;
1123         unsigned long val;
1124         ssize_t ret;
1125
1126         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1127         if (ret)
1128                 return ret;
1129
1130         ret = tracing_update_buffers();
1131         if (ret < 0)
1132                 return ret;
1133
1134         if (val != 0 && val != 1)
1135                 return -EINVAL;
1136
1137         /*
1138          * Opening of "enable" adds a ref count to system,
1139          * so the name is safe to use.
1140          */
1141         if (system)
1142                 name = system->name;
1143
1144         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1145         if (ret)
1146                 goto out;
1147
1148         ret = cnt;
1149
1150 out:
1151         *ppos += cnt;
1152
1153         return ret;
1154 }
1155
1156 enum {
1157         FORMAT_HEADER           = 1,
1158         FORMAT_FIELD_SEPERATOR  = 2,
1159         FORMAT_PRINTFMT         = 3,
1160 };
1161
1162 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1163 {
1164         struct trace_event_call *call = event_file_data(m->private);
1165         struct list_head *common_head = &ftrace_common_fields;
1166         struct list_head *head = trace_get_fields(call);
1167         struct list_head *node = v;
1168
1169         (*pos)++;
1170
1171         switch ((unsigned long)v) {
1172         case FORMAT_HEADER:
1173                 node = common_head;
1174                 break;
1175
1176         case FORMAT_FIELD_SEPERATOR:
1177                 node = head;
1178                 break;
1179
1180         case FORMAT_PRINTFMT:
1181                 /* all done */
1182                 return NULL;
1183         }
1184
1185         node = node->prev;
1186         if (node == common_head)
1187                 return (void *)FORMAT_FIELD_SEPERATOR;
1188         else if (node == head)
1189                 return (void *)FORMAT_PRINTFMT;
1190         else
1191                 return node;
1192 }
1193
1194 static int f_show(struct seq_file *m, void *v)
1195 {
1196         struct trace_event_call *call = event_file_data(m->private);
1197         struct ftrace_event_field *field;
1198         const char *array_descriptor;
1199
1200         switch ((unsigned long)v) {
1201         case FORMAT_HEADER:
1202                 seq_printf(m, "name: %s\n", trace_event_name(call));
1203                 seq_printf(m, "ID: %d\n", call->event.type);
1204                 seq_puts(m, "format:\n");
1205                 return 0;
1206
1207         case FORMAT_FIELD_SEPERATOR:
1208                 seq_putc(m, '\n');
1209                 return 0;
1210
1211         case FORMAT_PRINTFMT:
1212                 seq_printf(m, "\nprint fmt: %s\n",
1213                            call->print_fmt);
1214                 return 0;
1215         }
1216
1217         field = list_entry(v, struct ftrace_event_field, link);
1218         /*
1219          * Smartly shows the array type(except dynamic array).
1220          * Normal:
1221          *      field:TYPE VAR
1222          * If TYPE := TYPE[LEN], it is shown:
1223          *      field:TYPE VAR[LEN]
1224          */
1225         array_descriptor = strchr(field->type, '[');
1226
1227         if (!strncmp(field->type, "__data_loc", 10))
1228                 array_descriptor = NULL;
1229
1230         if (!array_descriptor)
1231                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1232                            field->type, field->name, field->offset,
1233                            field->size, !!field->is_signed);
1234         else
1235                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1236                            (int)(array_descriptor - field->type),
1237                            field->type, field->name,
1238                            array_descriptor, field->offset,
1239                            field->size, !!field->is_signed);
1240
1241         return 0;
1242 }
1243
1244 static void *f_start(struct seq_file *m, loff_t *pos)
1245 {
1246         void *p = (void *)FORMAT_HEADER;
1247         loff_t l = 0;
1248
1249         /* ->stop() is called even if ->start() fails */
1250         mutex_lock(&event_mutex);
1251         if (!event_file_data(m->private))
1252                 return ERR_PTR(-ENODEV);
1253
1254         while (l < *pos && p)
1255                 p = f_next(m, p, &l);
1256
1257         return p;
1258 }
1259
1260 static void f_stop(struct seq_file *m, void *p)
1261 {
1262         mutex_unlock(&event_mutex);
1263 }
1264
1265 static const struct seq_operations trace_format_seq_ops = {
1266         .start          = f_start,
1267         .next           = f_next,
1268         .stop           = f_stop,
1269         .show           = f_show,
1270 };
1271
1272 static int trace_format_open(struct inode *inode, struct file *file)
1273 {
1274         struct seq_file *m;
1275         int ret;
1276
1277         ret = seq_open(file, &trace_format_seq_ops);
1278         if (ret < 0)
1279                 return ret;
1280
1281         m = file->private_data;
1282         m->private = file;
1283
1284         return 0;
1285 }
1286
1287 static ssize_t
1288 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1289 {
1290         int id = (long)event_file_data(filp);
1291         char buf[32];
1292         int len;
1293
1294         if (unlikely(!id))
1295                 return -ENODEV;
1296
1297         len = sprintf(buf, "%d\n", id);
1298
1299         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1300 }
1301
1302 static ssize_t
1303 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1304                   loff_t *ppos)
1305 {
1306         struct trace_event_file *file;
1307         struct trace_seq *s;
1308         int r = -ENODEV;
1309
1310         if (*ppos)
1311                 return 0;
1312
1313         s = kmalloc(sizeof(*s), GFP_KERNEL);
1314
1315         if (!s)
1316                 return -ENOMEM;
1317
1318         trace_seq_init(s);
1319
1320         mutex_lock(&event_mutex);
1321         file = event_file_data(filp);
1322         if (file)
1323                 print_event_filter(file, s);
1324         mutex_unlock(&event_mutex);
1325
1326         if (file)
1327                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1328                                             s->buffer, trace_seq_used(s));
1329
1330         kfree(s);
1331
1332         return r;
1333 }
1334
1335 static ssize_t
1336 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1337                    loff_t *ppos)
1338 {
1339         struct trace_event_file *file;
1340         char *buf;
1341         int err = -ENODEV;
1342
1343         if (cnt >= PAGE_SIZE)
1344                 return -EINVAL;
1345
1346         buf = (char *)__get_free_page(GFP_TEMPORARY);
1347         if (!buf)
1348                 return -ENOMEM;
1349
1350         if (copy_from_user(buf, ubuf, cnt)) {
1351                 free_page((unsigned long) buf);
1352                 return -EFAULT;
1353         }
1354         buf[cnt] = '\0';
1355
1356         mutex_lock(&event_mutex);
1357         file = event_file_data(filp);
1358         if (file)
1359                 err = apply_event_filter(file, buf);
1360         mutex_unlock(&event_mutex);
1361
1362         free_page((unsigned long) buf);
1363         if (err < 0)
1364                 return err;
1365
1366         *ppos += cnt;
1367
1368         return cnt;
1369 }
1370
1371 static LIST_HEAD(event_subsystems);
1372
1373 static int subsystem_open(struct inode *inode, struct file *filp)
1374 {
1375         struct event_subsystem *system = NULL;
1376         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1377         struct trace_array *tr;
1378         int ret;
1379
1380         if (tracing_is_disabled())
1381                 return -ENODEV;
1382
1383         /* Make sure the system still exists */
1384         mutex_lock(&trace_types_lock);
1385         mutex_lock(&event_mutex);
1386         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1387                 list_for_each_entry(dir, &tr->systems, list) {
1388                         if (dir == inode->i_private) {
1389                                 /* Don't open systems with no events */
1390                                 if (dir->nr_events) {
1391                                         __get_system_dir(dir);
1392                                         system = dir->subsystem;
1393                                 }
1394                                 goto exit_loop;
1395                         }
1396                 }
1397         }
1398  exit_loop:
1399         mutex_unlock(&event_mutex);
1400         mutex_unlock(&trace_types_lock);
1401
1402         if (!system)
1403                 return -ENODEV;
1404
1405         /* Some versions of gcc think dir can be uninitialized here */
1406         WARN_ON(!dir);
1407
1408         /* Still need to increment the ref count of the system */
1409         if (trace_array_get(tr) < 0) {
1410                 put_system(dir);
1411                 return -ENODEV;
1412         }
1413
1414         ret = tracing_open_generic(inode, filp);
1415         if (ret < 0) {
1416                 trace_array_put(tr);
1417                 put_system(dir);
1418         }
1419
1420         return ret;
1421 }
1422
1423 static int system_tr_open(struct inode *inode, struct file *filp)
1424 {
1425         struct trace_subsystem_dir *dir;
1426         struct trace_array *tr = inode->i_private;
1427         int ret;
1428
1429         if (tracing_is_disabled())
1430                 return -ENODEV;
1431
1432         if (trace_array_get(tr) < 0)
1433                 return -ENODEV;
1434
1435         /* Make a temporary dir that has no system but points to tr */
1436         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1437         if (!dir) {
1438                 trace_array_put(tr);
1439                 return -ENOMEM;
1440         }
1441
1442         dir->tr = tr;
1443
1444         ret = tracing_open_generic(inode, filp);
1445         if (ret < 0) {
1446                 trace_array_put(tr);
1447                 kfree(dir);
1448                 return ret;
1449         }
1450
1451         filp->private_data = dir;
1452
1453         return 0;
1454 }
1455
1456 static int subsystem_release(struct inode *inode, struct file *file)
1457 {
1458         struct trace_subsystem_dir *dir = file->private_data;
1459
1460         trace_array_put(dir->tr);
1461
1462         /*
1463          * If dir->subsystem is NULL, then this is a temporary
1464          * descriptor that was made for a trace_array to enable
1465          * all subsystems.
1466          */
1467         if (dir->subsystem)
1468                 put_system(dir);
1469         else
1470                 kfree(dir);
1471
1472         return 0;
1473 }
1474
1475 static ssize_t
1476 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1477                       loff_t *ppos)
1478 {
1479         struct trace_subsystem_dir *dir = filp->private_data;
1480         struct event_subsystem *system = dir->subsystem;
1481         struct trace_seq *s;
1482         int r;
1483
1484         if (*ppos)
1485                 return 0;
1486
1487         s = kmalloc(sizeof(*s), GFP_KERNEL);
1488         if (!s)
1489                 return -ENOMEM;
1490
1491         trace_seq_init(s);
1492
1493         print_subsystem_event_filter(system, s);
1494         r = simple_read_from_buffer(ubuf, cnt, ppos,
1495                                     s->buffer, trace_seq_used(s));
1496
1497         kfree(s);
1498
1499         return r;
1500 }
1501
1502 static ssize_t
1503 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1504                        loff_t *ppos)
1505 {
1506         struct trace_subsystem_dir *dir = filp->private_data;
1507         char *buf;
1508         int err;
1509
1510         if (cnt >= PAGE_SIZE)
1511                 return -EINVAL;
1512
1513         buf = (char *)__get_free_page(GFP_TEMPORARY);
1514         if (!buf)
1515                 return -ENOMEM;
1516
1517         if (copy_from_user(buf, ubuf, cnt)) {
1518                 free_page((unsigned long) buf);
1519                 return -EFAULT;
1520         }
1521         buf[cnt] = '\0';
1522
1523         err = apply_subsystem_event_filter(dir, buf);
1524         free_page((unsigned long) buf);
1525         if (err < 0)
1526                 return err;
1527
1528         *ppos += cnt;
1529
1530         return cnt;
1531 }
1532
1533 static ssize_t
1534 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1535 {
1536         int (*func)(struct trace_seq *s) = filp->private_data;
1537         struct trace_seq *s;
1538         int r;
1539
1540         if (*ppos)
1541                 return 0;
1542
1543         s = kmalloc(sizeof(*s), GFP_KERNEL);
1544         if (!s)
1545                 return -ENOMEM;
1546
1547         trace_seq_init(s);
1548
1549         func(s);
1550         r = simple_read_from_buffer(ubuf, cnt, ppos,
1551                                     s->buffer, trace_seq_used(s));
1552
1553         kfree(s);
1554
1555         return r;
1556 }
1557
1558 static int max_pids(struct trace_pid_list *pid_list)
1559 {
1560         return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1561 }
1562
1563 static void ignore_task_cpu(void *data)
1564 {
1565         struct trace_array *tr = data;
1566         struct trace_pid_list *pid_list;
1567
1568         /*
1569          * This function is called by on_each_cpu() while the
1570          * event_mutex is held.
1571          */
1572         pid_list = rcu_dereference_protected(tr->filtered_pids,
1573                                              mutex_is_locked(&event_mutex));
1574
1575         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1576                        check_ignore_pid(pid_list, current));
1577 }
1578
1579 static ssize_t
1580 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1581                        size_t cnt, loff_t *ppos)
1582 {
1583         struct seq_file *m = filp->private_data;
1584         struct trace_array *tr = m->private;
1585         struct trace_pid_list *filtered_pids = NULL;
1586         struct trace_pid_list *pid_list = NULL;
1587         struct trace_event_file *file;
1588         struct trace_parser parser;
1589         unsigned long val;
1590         loff_t this_pos;
1591         ssize_t read = 0;
1592         ssize_t ret = 0;
1593         pid_t pid;
1594         int i;
1595
1596         if (!cnt)
1597                 return 0;
1598
1599         ret = tracing_update_buffers();
1600         if (ret < 0)
1601                 return ret;
1602
1603         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1604                 return -ENOMEM;
1605
1606         mutex_lock(&event_mutex);
1607         /*
1608          * Load as many pids into the array before doing a
1609          * swap from the tr->filtered_pids to the new list.
1610          */
1611         while (cnt > 0) {
1612
1613                 this_pos = 0;
1614
1615                 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1616                 if (ret < 0 || !trace_parser_loaded(&parser))
1617                         break;
1618
1619                 read += ret;
1620                 ubuf += ret;
1621                 cnt -= ret;
1622
1623                 parser.buffer[parser.idx] = 0;
1624
1625                 ret = -EINVAL;
1626                 if (kstrtoul(parser.buffer, 0, &val))
1627                         break;
1628                 if (val > INT_MAX)
1629                         break;
1630
1631                 pid = (pid_t)val;
1632
1633                 ret = -ENOMEM;
1634                 if (!pid_list) {
1635                         pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1636                         if (!pid_list)
1637                                 break;
1638
1639                         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1640                                                         lockdep_is_held(&event_mutex));
1641                         if (filtered_pids)
1642                                 pid_list->order = filtered_pids->order;
1643                         else
1644                                 pid_list->order = 0;
1645
1646                         pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1647                                                                   pid_list->order);
1648                         if (!pid_list->pids)
1649                                 break;
1650
1651                         if (filtered_pids) {
1652                                 pid_list->nr_pids = filtered_pids->nr_pids;
1653                                 memcpy(pid_list->pids, filtered_pids->pids,
1654                                        pid_list->nr_pids * sizeof(pid_t));
1655                         } else
1656                                 pid_list->nr_pids = 0;
1657                 }
1658
1659                 if (pid_list->nr_pids >= max_pids(pid_list)) {
1660                         pid_t *pid_page;
1661
1662                         pid_page = (void *)__get_free_pages(GFP_KERNEL,
1663                                                             pid_list->order + 1);
1664                         if (!pid_page)
1665                                 break;
1666                         memcpy(pid_page, pid_list->pids,
1667                                pid_list->nr_pids * sizeof(pid_t));
1668                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1669
1670                         pid_list->order++;
1671                         pid_list->pids = pid_page;
1672                 }
1673
1674                 pid_list->pids[pid_list->nr_pids++] = pid;
1675                 trace_parser_clear(&parser);
1676                 ret = 0;
1677         }
1678         trace_parser_put(&parser);
1679
1680         if (ret < 0) {
1681                 if (pid_list)
1682                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1683                 kfree(pid_list);
1684                 mutex_unlock(&event_mutex);
1685                 return ret;
1686         }
1687
1688         if (!pid_list) {
1689                 mutex_unlock(&event_mutex);
1690                 return ret;
1691         }
1692
1693         sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1694
1695         /* Remove duplicates */
1696         for (i = 1; i < pid_list->nr_pids; i++) {
1697                 int start = i;
1698
1699                 while (i < pid_list->nr_pids &&
1700                        pid_list->pids[i - 1] == pid_list->pids[i])
1701                         i++;
1702
1703                 if (start != i) {
1704                         if (i < pid_list->nr_pids) {
1705                                 memmove(&pid_list->pids[start], &pid_list->pids[i],
1706                                         (pid_list->nr_pids - i) * sizeof(pid_t));
1707                                 pid_list->nr_pids -= i - start;
1708                                 i = start;
1709                         } else
1710                                 pid_list->nr_pids = start;
1711                 }
1712         }
1713
1714         rcu_assign_pointer(tr->filtered_pids, pid_list);
1715
1716         list_for_each_entry(file, &tr->events, list) {
1717                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1718         }
1719
1720         if (filtered_pids) {
1721                 synchronize_sched();
1722
1723                 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1724                 kfree(filtered_pids);
1725         } else {
1726                 /*
1727                  * Register a probe that is called before all other probes
1728                  * to set ignore_pid if next or prev do not match.
1729                  * Register a probe this is called after all other probes
1730                  * to only keep ignore_pid set if next pid matches.
1731                  */
1732                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1733                                                  tr, INT_MAX);
1734                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1735                                                  tr, 0);
1736
1737                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1738                                                  tr, INT_MAX);
1739                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1740                                                  tr, 0);
1741
1742                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1743                                                      tr, INT_MAX);
1744                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1745                                                      tr, 0);
1746
1747                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1748                                                  tr, INT_MAX);
1749                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1750                                                  tr, 0);
1751         }
1752
1753         /*
1754          * Ignoring of pids is done at task switch. But we have to
1755          * check for those tasks that are currently running.
1756          * Always do this in case a pid was appended or removed.
1757          */
1758         on_each_cpu(ignore_task_cpu, tr, 1);
1759
1760         mutex_unlock(&event_mutex);
1761
1762         ret = read;
1763         *ppos += read;
1764
1765         return ret;
1766 }
1767
1768 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1769 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1770 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1771 static int ftrace_event_release(struct inode *inode, struct file *file);
1772
1773 static const struct seq_operations show_event_seq_ops = {
1774         .start = t_start,
1775         .next = t_next,
1776         .show = t_show,
1777         .stop = t_stop,
1778 };
1779
1780 static const struct seq_operations show_set_event_seq_ops = {
1781         .start = s_start,
1782         .next = s_next,
1783         .show = t_show,
1784         .stop = t_stop,
1785 };
1786
1787 static const struct seq_operations show_set_pid_seq_ops = {
1788         .start = p_start,
1789         .next = p_next,
1790         .show = p_show,
1791         .stop = p_stop,
1792 };
1793
1794 static const struct file_operations ftrace_avail_fops = {
1795         .open = ftrace_event_avail_open,
1796         .read = seq_read,
1797         .llseek = seq_lseek,
1798         .release = seq_release,
1799 };
1800
1801 static const struct file_operations ftrace_set_event_fops = {
1802         .open = ftrace_event_set_open,
1803         .read = seq_read,
1804         .write = ftrace_event_write,
1805         .llseek = seq_lseek,
1806         .release = ftrace_event_release,
1807 };
1808
1809 static const struct file_operations ftrace_set_event_pid_fops = {
1810         .open = ftrace_event_set_pid_open,
1811         .read = seq_read,
1812         .write = ftrace_event_pid_write,
1813         .llseek = seq_lseek,
1814         .release = ftrace_event_release,
1815 };
1816
1817 static const struct file_operations ftrace_enable_fops = {
1818         .open = tracing_open_generic,
1819         .read = event_enable_read,
1820         .write = event_enable_write,
1821         .llseek = default_llseek,
1822 };
1823
1824 static const struct file_operations ftrace_event_format_fops = {
1825         .open = trace_format_open,
1826         .read = seq_read,
1827         .llseek = seq_lseek,
1828         .release = seq_release,
1829 };
1830
1831 static const struct file_operations ftrace_event_id_fops = {
1832         .read = event_id_read,
1833         .llseek = default_llseek,
1834 };
1835
1836 static const struct file_operations ftrace_event_filter_fops = {
1837         .open = tracing_open_generic,
1838         .read = event_filter_read,
1839         .write = event_filter_write,
1840         .llseek = default_llseek,
1841 };
1842
1843 static const struct file_operations ftrace_subsystem_filter_fops = {
1844         .open = subsystem_open,
1845         .read = subsystem_filter_read,
1846         .write = subsystem_filter_write,
1847         .llseek = default_llseek,
1848         .release = subsystem_release,
1849 };
1850
1851 static const struct file_operations ftrace_system_enable_fops = {
1852         .open = subsystem_open,
1853         .read = system_enable_read,
1854         .write = system_enable_write,
1855         .llseek = default_llseek,
1856         .release = subsystem_release,
1857 };
1858
1859 static const struct file_operations ftrace_tr_enable_fops = {
1860         .open = system_tr_open,
1861         .read = system_enable_read,
1862         .write = system_enable_write,
1863         .llseek = default_llseek,
1864         .release = subsystem_release,
1865 };
1866
1867 static const struct file_operations ftrace_show_header_fops = {
1868         .open = tracing_open_generic,
1869         .read = show_header,
1870         .llseek = default_llseek,
1871 };
1872
1873 static int
1874 ftrace_event_open(struct inode *inode, struct file *file,
1875                   const struct seq_operations *seq_ops)
1876 {
1877         struct seq_file *m;
1878         int ret;
1879
1880         ret = seq_open(file, seq_ops);
1881         if (ret < 0)
1882                 return ret;
1883         m = file->private_data;
1884         /* copy tr over to seq ops */
1885         m->private = inode->i_private;
1886
1887         return ret;
1888 }
1889
1890 static int ftrace_event_release(struct inode *inode, struct file *file)
1891 {
1892         struct trace_array *tr = inode->i_private;
1893
1894         trace_array_put(tr);
1895
1896         return seq_release(inode, file);
1897 }
1898
1899 static int
1900 ftrace_event_avail_open(struct inode *inode, struct file *file)
1901 {
1902         const struct seq_operations *seq_ops = &show_event_seq_ops;
1903
1904         return ftrace_event_open(inode, file, seq_ops);
1905 }
1906
1907 static int
1908 ftrace_event_set_open(struct inode *inode, struct file *file)
1909 {
1910         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1911         struct trace_array *tr = inode->i_private;
1912         int ret;
1913
1914         if (trace_array_get(tr) < 0)
1915                 return -ENODEV;
1916
1917         if ((file->f_mode & FMODE_WRITE) &&
1918             (file->f_flags & O_TRUNC))
1919                 ftrace_clear_events(tr);
1920
1921         ret = ftrace_event_open(inode, file, seq_ops);
1922         if (ret < 0)
1923                 trace_array_put(tr);
1924         return ret;
1925 }
1926
1927 static int
1928 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1929 {
1930         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1931         struct trace_array *tr = inode->i_private;
1932         int ret;
1933
1934         if (trace_array_get(tr) < 0)
1935                 return -ENODEV;
1936
1937         if ((file->f_mode & FMODE_WRITE) &&
1938             (file->f_flags & O_TRUNC))
1939                 ftrace_clear_event_pids(tr);
1940
1941         ret = ftrace_event_open(inode, file, seq_ops);
1942         if (ret < 0)
1943                 trace_array_put(tr);
1944         return ret;
1945 }
1946
1947 static struct event_subsystem *
1948 create_new_subsystem(const char *name)
1949 {
1950         struct event_subsystem *system;
1951
1952         /* need to create new entry */
1953         system = kmalloc(sizeof(*system), GFP_KERNEL);
1954         if (!system)
1955                 return NULL;
1956
1957         system->ref_count = 1;
1958
1959         /* Only allocate if dynamic (kprobes and modules) */
1960         system->name = kstrdup_const(name, GFP_KERNEL);
1961         if (!system->name)
1962                 goto out_free;
1963
1964         system->filter = NULL;
1965
1966         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1967         if (!system->filter)
1968                 goto out_free;
1969
1970         list_add(&system->list, &event_subsystems);
1971
1972         return system;
1973
1974  out_free:
1975         kfree_const(system->name);
1976         kfree(system);
1977         return NULL;
1978 }
1979
1980 static struct dentry *
1981 event_subsystem_dir(struct trace_array *tr, const char *name,
1982                     struct trace_event_file *file, struct dentry *parent)
1983 {
1984         struct trace_subsystem_dir *dir;
1985         struct event_subsystem *system;
1986         struct dentry *entry;
1987
1988         /* First see if we did not already create this dir */
1989         list_for_each_entry(dir, &tr->systems, list) {
1990                 system = dir->subsystem;
1991                 if (strcmp(system->name, name) == 0) {
1992                         dir->nr_events++;
1993                         file->system = dir;
1994                         return dir->entry;
1995                 }
1996         }
1997
1998         /* Now see if the system itself exists. */
1999         list_for_each_entry(system, &event_subsystems, list) {
2000                 if (strcmp(system->name, name) == 0)
2001                         break;
2002         }
2003         /* Reset system variable when not found */
2004         if (&system->list == &event_subsystems)
2005                 system = NULL;
2006
2007         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2008         if (!dir)
2009                 goto out_fail;
2010
2011         if (!system) {
2012                 system = create_new_subsystem(name);
2013                 if (!system)
2014                         goto out_free;
2015         } else
2016                 __get_system(system);
2017
2018         dir->entry = tracefs_create_dir(name, parent);
2019         if (!dir->entry) {
2020                 pr_warn("Failed to create system directory %s\n", name);
2021                 __put_system(system);
2022                 goto out_free;
2023         }
2024
2025         dir->tr = tr;
2026         dir->ref_count = 1;
2027         dir->nr_events = 1;
2028         dir->subsystem = system;
2029         file->system = dir;
2030
2031         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2032                                     &ftrace_subsystem_filter_fops);
2033         if (!entry) {
2034                 kfree(system->filter);
2035                 system->filter = NULL;
2036                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2037         }
2038
2039         trace_create_file("enable", 0644, dir->entry, dir,
2040                           &ftrace_system_enable_fops);
2041
2042         list_add(&dir->list, &tr->systems);
2043
2044         return dir->entry;
2045
2046  out_free:
2047         kfree(dir);
2048  out_fail:
2049         /* Only print this message if failed on memory allocation */
2050         if (!dir || !system)
2051                 pr_warn("No memory to create event subsystem %s\n", name);
2052         return NULL;
2053 }
2054
2055 static int
2056 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2057 {
2058         struct trace_event_call *call = file->event_call;
2059         struct trace_array *tr = file->tr;
2060         struct list_head *head;
2061         struct dentry *d_events;
2062         const char *name;
2063         int ret;
2064
2065         /*
2066          * If the trace point header did not define TRACE_SYSTEM
2067          * then the system would be called "TRACE_SYSTEM".
2068          */
2069         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2070                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2071                 if (!d_events)
2072                         return -ENOMEM;
2073         } else
2074                 d_events = parent;
2075
2076         name = trace_event_name(call);
2077         file->dir = tracefs_create_dir(name, d_events);
2078         if (!file->dir) {
2079                 pr_warn("Could not create tracefs '%s' directory\n", name);
2080                 return -1;
2081         }
2082
2083         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2084                 trace_create_file("enable", 0644, file->dir, file,
2085                                   &ftrace_enable_fops);
2086
2087 #ifdef CONFIG_PERF_EVENTS
2088         if (call->event.type && call->class->reg)
2089                 trace_create_file("id", 0444, file->dir,
2090                                   (void *)(long)call->event.type,
2091                                   &ftrace_event_id_fops);
2092 #endif
2093
2094         /*
2095          * Other events may have the same class. Only update
2096          * the fields if they are not already defined.
2097          */
2098         head = trace_get_fields(call);
2099         if (list_empty(head)) {
2100                 ret = call->class->define_fields(call);
2101                 if (ret < 0) {
2102                         pr_warn("Could not initialize trace point events/%s\n",
2103                                 name);
2104                         return -1;
2105                 }
2106         }
2107         trace_create_file("filter", 0644, file->dir, file,
2108                           &ftrace_event_filter_fops);
2109
2110         /*
2111          * Only event directories that can be enabled should have
2112          * triggers.
2113          */
2114         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2115                 trace_create_file("trigger", 0644, file->dir, file,
2116                                   &event_trigger_fops);
2117
2118         trace_create_file("format", 0444, file->dir, call,
2119                           &ftrace_event_format_fops);
2120
2121         return 0;
2122 }
2123
2124 static void remove_event_from_tracers(struct trace_event_call *call)
2125 {
2126         struct trace_event_file *file;
2127         struct trace_array *tr;
2128
2129         do_for_each_event_file_safe(tr, file) {
2130                 if (file->event_call != call)
2131                         continue;
2132
2133                 remove_event_file_dir(file);
2134                 /*
2135                  * The do_for_each_event_file_safe() is
2136                  * a double loop. After finding the call for this
2137                  * trace_array, we use break to jump to the next
2138                  * trace_array.
2139                  */
2140                 break;
2141         } while_for_each_event_file();
2142 }
2143
2144 static void event_remove(struct trace_event_call *call)
2145 {
2146         struct trace_array *tr;
2147         struct trace_event_file *file;
2148
2149         do_for_each_event_file(tr, file) {
2150                 if (file->event_call != call)
2151                         continue;
2152                 ftrace_event_enable_disable(file, 0);
2153                 /*
2154                  * The do_for_each_event_file() is
2155                  * a double loop. After finding the call for this
2156                  * trace_array, we use break to jump to the next
2157                  * trace_array.
2158                  */
2159                 break;
2160         } while_for_each_event_file();
2161
2162         if (call->event.funcs)
2163                 __unregister_trace_event(&call->event);
2164         remove_event_from_tracers(call);
2165         list_del(&call->list);
2166 }
2167
2168 static int event_init(struct trace_event_call *call)
2169 {
2170         int ret = 0;
2171         const char *name;
2172
2173         name = trace_event_name(call);
2174         if (WARN_ON(!name))
2175                 return -EINVAL;
2176
2177         if (call->class->raw_init) {
2178                 ret = call->class->raw_init(call);
2179                 if (ret < 0 && ret != -ENOSYS)
2180                         pr_warn("Could not initialize trace events/%s\n", name);
2181         }
2182
2183         return ret;
2184 }
2185
2186 static int
2187 __register_event(struct trace_event_call *call, struct module *mod)
2188 {
2189         int ret;
2190
2191         ret = event_init(call);
2192         if (ret < 0)
2193                 return ret;
2194
2195         list_add(&call->list, &ftrace_events);
2196         call->mod = mod;
2197
2198         return 0;
2199 }
2200
2201 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2202 {
2203         int rlen;
2204         int elen;
2205
2206         /* Find the length of the enum value as a string */
2207         elen = snprintf(ptr, 0, "%ld", map->enum_value);
2208         /* Make sure there's enough room to replace the string with the value */
2209         if (len < elen)
2210                 return NULL;
2211
2212         snprintf(ptr, elen + 1, "%ld", map->enum_value);
2213
2214         /* Get the rest of the string of ptr */
2215         rlen = strlen(ptr + len);
2216         memmove(ptr + elen, ptr + len, rlen);
2217         /* Make sure we end the new string */
2218         ptr[elen + rlen] = 0;
2219
2220         return ptr + elen;
2221 }
2222
2223 static void update_event_printk(struct trace_event_call *call,
2224                                 struct trace_enum_map *map)
2225 {
2226         char *ptr;
2227         int quote = 0;
2228         int len = strlen(map->enum_string);
2229
2230         for (ptr = call->print_fmt; *ptr; ptr++) {
2231                 if (*ptr == '\\') {
2232                         ptr++;
2233                         /* paranoid */
2234                         if (!*ptr)
2235                                 break;
2236                         continue;
2237                 }
2238                 if (*ptr == '"') {
2239                         quote ^= 1;
2240                         continue;
2241                 }
2242                 if (quote)
2243                         continue;
2244                 if (isdigit(*ptr)) {
2245                         /* skip numbers */
2246                         do {
2247                                 ptr++;
2248                                 /* Check for alpha chars like ULL */
2249                         } while (isalnum(*ptr));
2250                         if (!*ptr)
2251                                 break;
2252                         /*
2253                          * A number must have some kind of delimiter after
2254                          * it, and we can ignore that too.
2255                          */
2256                         continue;
2257                 }
2258                 if (isalpha(*ptr) || *ptr == '_') {
2259                         if (strncmp(map->enum_string, ptr, len) == 0 &&
2260                             !isalnum(ptr[len]) && ptr[len] != '_') {
2261                                 ptr = enum_replace(ptr, map, len);
2262                                 /* Hmm, enum string smaller than value */
2263                                 if (WARN_ON_ONCE(!ptr))
2264                                         return;
2265                                 /*
2266                                  * No need to decrement here, as enum_replace()
2267                                  * returns the pointer to the character passed
2268                                  * the enum, and two enums can not be placed
2269                                  * back to back without something in between.
2270                                  * We can skip that something in between.
2271                                  */
2272                                 continue;
2273                         }
2274                 skip_more:
2275                         do {
2276                                 ptr++;
2277                         } while (isalnum(*ptr) || *ptr == '_');
2278                         if (!*ptr)
2279                                 break;
2280                         /*
2281                          * If what comes after this variable is a '.' or
2282                          * '->' then we can continue to ignore that string.
2283                          */
2284                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2285                                 ptr += *ptr == '.' ? 1 : 2;
2286                                 if (!*ptr)
2287                                         break;
2288                                 goto skip_more;
2289                         }
2290                         /*
2291                          * Once again, we can skip the delimiter that came
2292                          * after the string.
2293                          */
2294                         continue;
2295                 }
2296         }
2297 }
2298
2299 void trace_event_enum_update(struct trace_enum_map **map, int len)
2300 {
2301         struct trace_event_call *call, *p;
2302         const char *last_system = NULL;
2303         bool first = false;
2304         int last_i;
2305         int i;
2306
2307         down_write(&trace_event_sem);
2308         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2309                 /* events are usually grouped together with systems */
2310                 if (!last_system || call->class->system != last_system) {
2311                         first = true;
2312                         last_i = 0;
2313                         last_system = call->class->system;
2314                 }
2315
2316                 /*
2317                  * Since calls are grouped by systems, the likelyhood that the
2318                  * next call in the iteration belongs to the same system as the
2319                  * previous call is high. As an optimization, we skip seaching
2320                  * for a map[] that matches the call's system if the last call
2321                  * was from the same system. That's what last_i is for. If the
2322                  * call has the same system as the previous call, then last_i
2323                  * will be the index of the first map[] that has a matching
2324                  * system.
2325                  */
2326                 for (i = last_i; i < len; i++) {
2327                         if (call->class->system == map[i]->system) {
2328                                 /* Save the first system if need be */
2329                                 if (first) {
2330                                         last_i = i;
2331                                         first = false;
2332                                 }
2333                                 update_event_printk(call, map[i]);
2334                         }
2335                 }
2336         }
2337         up_write(&trace_event_sem);
2338 }
2339
2340 static struct trace_event_file *
2341 trace_create_new_event(struct trace_event_call *call,
2342                        struct trace_array *tr)
2343 {
2344         struct trace_pid_list *pid_list;
2345         struct trace_event_file *file;
2346
2347         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2348         if (!file)
2349                 return NULL;
2350
2351         pid_list = rcu_dereference_protected(tr->filtered_pids,
2352                                              lockdep_is_held(&event_mutex));
2353
2354         if (pid_list)
2355                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2356
2357         file->event_call = call;
2358         file->tr = tr;
2359         atomic_set(&file->sm_ref, 0);
2360         atomic_set(&file->tm_ref, 0);
2361         INIT_LIST_HEAD(&file->triggers);
2362         list_add(&file->list, &tr->events);
2363
2364         return file;
2365 }
2366
2367 /* Add an event to a trace directory */
2368 static int
2369 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2370 {
2371         struct trace_event_file *file;
2372
2373         file = trace_create_new_event(call, tr);
2374         if (!file)
2375                 return -ENOMEM;
2376
2377         return event_create_dir(tr->event_dir, file);
2378 }
2379
2380 /*
2381  * Just create a decriptor for early init. A descriptor is required
2382  * for enabling events at boot. We want to enable events before
2383  * the filesystem is initialized.
2384  */
2385 static __init int
2386 __trace_early_add_new_event(struct trace_event_call *call,
2387                             struct trace_array *tr)
2388 {
2389         struct trace_event_file *file;
2390
2391         file = trace_create_new_event(call, tr);
2392         if (!file)
2393                 return -ENOMEM;
2394
2395         return 0;
2396 }
2397
2398 struct ftrace_module_file_ops;
2399 static void __add_event_to_tracers(struct trace_event_call *call);
2400
2401 /* Add an additional event_call dynamically */
2402 int trace_add_event_call(struct trace_event_call *call)
2403 {
2404         int ret;
2405         mutex_lock(&trace_types_lock);
2406         mutex_lock(&event_mutex);
2407
2408         ret = __register_event(call, NULL);
2409         if (ret >= 0)
2410                 __add_event_to_tracers(call);
2411
2412         mutex_unlock(&event_mutex);
2413         mutex_unlock(&trace_types_lock);
2414         return ret;
2415 }
2416
2417 /*
2418  * Must be called under locking of trace_types_lock, event_mutex and
2419  * trace_event_sem.
2420  */
2421 static void __trace_remove_event_call(struct trace_event_call *call)
2422 {
2423         event_remove(call);
2424         trace_destroy_fields(call);
2425         free_event_filter(call->filter);
2426         call->filter = NULL;
2427 }
2428
2429 static int probe_remove_event_call(struct trace_event_call *call)
2430 {
2431         struct trace_array *tr;
2432         struct trace_event_file *file;
2433
2434 #ifdef CONFIG_PERF_EVENTS
2435         if (call->perf_refcount)
2436                 return -EBUSY;
2437 #endif
2438         do_for_each_event_file(tr, file) {
2439                 if (file->event_call != call)
2440                         continue;
2441                 /*
2442                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2443                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2444                  * TRACE_REG_UNREGISTER.
2445                  */
2446                 if (file->flags & EVENT_FILE_FL_ENABLED)
2447                         return -EBUSY;
2448                 /*
2449                  * The do_for_each_event_file_safe() is
2450                  * a double loop. After finding the call for this
2451                  * trace_array, we use break to jump to the next
2452                  * trace_array.
2453                  */
2454                 break;
2455         } while_for_each_event_file();
2456
2457         __trace_remove_event_call(call);
2458
2459         return 0;
2460 }
2461
2462 /* Remove an event_call */
2463 int trace_remove_event_call(struct trace_event_call *call)
2464 {
2465         int ret;
2466
2467         mutex_lock(&trace_types_lock);
2468         mutex_lock(&event_mutex);
2469         down_write(&trace_event_sem);
2470         ret = probe_remove_event_call(call);
2471         up_write(&trace_event_sem);
2472         mutex_unlock(&event_mutex);
2473         mutex_unlock(&trace_types_lock);
2474
2475         return ret;
2476 }
2477
2478 #define for_each_event(event, start, end)                       \
2479         for (event = start;                                     \
2480              (unsigned long)event < (unsigned long)end;         \
2481              event++)
2482
2483 #ifdef CONFIG_MODULES
2484
2485 static void trace_module_add_events(struct module *mod)
2486 {
2487         struct trace_event_call **call, **start, **end;
2488
2489         if (!mod->num_trace_events)
2490                 return;
2491
2492         /* Don't add infrastructure for mods without tracepoints */
2493         if (trace_module_has_bad_taint(mod)) {
2494                 pr_err("%s: module has bad taint, not creating trace events\n",
2495                        mod->name);
2496                 return;
2497         }
2498
2499         start = mod->trace_events;
2500         end = mod->trace_events + mod->num_trace_events;
2501
2502         for_each_event(call, start, end) {
2503                 __register_event(*call, mod);
2504                 __add_event_to_tracers(*call);
2505         }
2506 }
2507
2508 static void trace_module_remove_events(struct module *mod)
2509 {
2510         struct trace_event_call *call, *p;
2511         bool clear_trace = false;
2512
2513         down_write(&trace_event_sem);
2514         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2515                 if (call->mod == mod) {
2516                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2517                                 clear_trace = true;
2518                         __trace_remove_event_call(call);
2519                 }
2520         }
2521         up_write(&trace_event_sem);
2522
2523         /*
2524          * It is safest to reset the ring buffer if the module being unloaded
2525          * registered any events that were used. The only worry is if
2526          * a new module gets loaded, and takes on the same id as the events
2527          * of this module. When printing out the buffer, traced events left
2528          * over from this module may be passed to the new module events and
2529          * unexpected results may occur.
2530          */
2531         if (clear_trace)
2532                 tracing_reset_all_online_cpus();
2533 }
2534
2535 static int trace_module_notify(struct notifier_block *self,
2536                                unsigned long val, void *data)
2537 {
2538         struct module *mod = data;
2539
2540         mutex_lock(&trace_types_lock);
2541         mutex_lock(&event_mutex);
2542         switch (val) {
2543         case MODULE_STATE_COMING:
2544                 trace_module_add_events(mod);
2545                 break;
2546         case MODULE_STATE_GOING:
2547                 trace_module_remove_events(mod);
2548                 break;
2549         }
2550         mutex_unlock(&event_mutex);
2551         mutex_unlock(&trace_types_lock);
2552
2553         return 0;
2554 }
2555
2556 static struct notifier_block trace_module_nb = {
2557         .notifier_call = trace_module_notify,
2558         .priority = 1, /* higher than trace.c module notify */
2559 };
2560 #endif /* CONFIG_MODULES */
2561
2562 /* Create a new event directory structure for a trace directory. */
2563 static void
2564 __trace_add_event_dirs(struct trace_array *tr)
2565 {
2566         struct trace_event_call *call;
2567         int ret;
2568
2569         list_for_each_entry(call, &ftrace_events, list) {
2570                 ret = __trace_add_new_event(call, tr);
2571                 if (ret < 0)
2572                         pr_warn("Could not create directory for event %s\n",
2573                                 trace_event_name(call));
2574         }
2575 }
2576
2577 struct trace_event_file *
2578 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2579 {
2580         struct trace_event_file *file;
2581         struct trace_event_call *call;
2582         const char *name;
2583
2584         list_for_each_entry(file, &tr->events, list) {
2585
2586                 call = file->event_call;
2587                 name = trace_event_name(call);
2588
2589                 if (!name || !call->class || !call->class->reg)
2590                         continue;
2591
2592                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2593                         continue;
2594
2595                 if (strcmp(event, name) == 0 &&
2596                     strcmp(system, call->class->system) == 0)
2597                         return file;
2598         }
2599         return NULL;
2600 }
2601
2602 #ifdef CONFIG_DYNAMIC_FTRACE
2603
2604 /* Avoid typos */
2605 #define ENABLE_EVENT_STR        "enable_event"
2606 #define DISABLE_EVENT_STR       "disable_event"
2607
2608 struct event_probe_data {
2609         struct trace_event_file *file;
2610         unsigned long                   count;
2611         int                             ref;
2612         bool                            enable;
2613 };
2614
2615 static void
2616 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2617 {
2618         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2619         struct event_probe_data *data = *pdata;
2620
2621         if (!data)
2622                 return;
2623
2624         if (data->enable)
2625                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2626         else
2627                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2628 }
2629
2630 static void
2631 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2632 {
2633         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2634         struct event_probe_data *data = *pdata;
2635
2636         if (!data)
2637                 return;
2638
2639         if (!data->count)
2640                 return;
2641
2642         /* Skip if the event is in a state we want to switch to */
2643         if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2644                 return;
2645
2646         if (data->count != -1)
2647                 (data->count)--;
2648
2649         event_enable_probe(ip, parent_ip, _data);
2650 }
2651
2652 static int
2653 event_enable_print(struct seq_file *m, unsigned long ip,
2654                       struct ftrace_probe_ops *ops, void *_data)
2655 {
2656         struct event_probe_data *data = _data;
2657
2658         seq_printf(m, "%ps:", (void *)ip);
2659
2660         seq_printf(m, "%s:%s:%s",
2661                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2662                    data->file->event_call->class->system,
2663                    trace_event_name(data->file->event_call));
2664
2665         if (data->count == -1)
2666                 seq_puts(m, ":unlimited\n");
2667         else
2668                 seq_printf(m, ":count=%ld\n", data->count);
2669
2670         return 0;
2671 }
2672
2673 static int
2674 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2675                   void **_data)
2676 {
2677         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2678         struct event_probe_data *data = *pdata;
2679
2680         data->ref++;
2681         return 0;
2682 }
2683
2684 static void
2685 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2686                   void **_data)
2687 {
2688         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2689         struct event_probe_data *data = *pdata;
2690
2691         if (WARN_ON_ONCE(data->ref <= 0))
2692                 return;
2693
2694         data->ref--;
2695         if (!data->ref) {
2696                 /* Remove the SOFT_MODE flag */
2697                 __ftrace_event_enable_disable(data->file, 0, 1);
2698                 module_put(data->file->event_call->mod);
2699                 kfree(data);
2700         }
2701         *pdata = NULL;
2702 }
2703
2704 static struct ftrace_probe_ops event_enable_probe_ops = {
2705         .func                   = event_enable_probe,
2706         .print                  = event_enable_print,
2707         .init                   = event_enable_init,
2708         .free                   = event_enable_free,
2709 };
2710
2711 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2712         .func                   = event_enable_count_probe,
2713         .print                  = event_enable_print,
2714         .init                   = event_enable_init,
2715         .free                   = event_enable_free,
2716 };
2717
2718 static struct ftrace_probe_ops event_disable_probe_ops = {
2719         .func                   = event_enable_probe,
2720         .print                  = event_enable_print,
2721         .init                   = event_enable_init,
2722         .free                   = event_enable_free,
2723 };
2724
2725 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2726         .func                   = event_enable_count_probe,
2727         .print                  = event_enable_print,
2728         .init                   = event_enable_init,
2729         .free                   = event_enable_free,
2730 };
2731
2732 static int
2733 event_enable_func(struct ftrace_hash *hash,
2734                   char *glob, char *cmd, char *param, int enabled)
2735 {
2736         struct trace_array *tr = top_trace_array();
2737         struct trace_event_file *file;
2738         struct ftrace_probe_ops *ops;
2739         struct event_probe_data *data;
2740         const char *system;
2741         const char *event;
2742         char *number;
2743         bool enable;
2744         int ret;
2745
2746         if (!tr)
2747                 return -ENODEV;
2748
2749         /* hash funcs only work with set_ftrace_filter */
2750         if (!enabled || !param)
2751                 return -EINVAL;
2752
2753         system = strsep(&param, ":");
2754         if (!param)
2755                 return -EINVAL;
2756
2757         event = strsep(&param, ":");
2758
2759         mutex_lock(&event_mutex);
2760
2761         ret = -EINVAL;
2762         file = find_event_file(tr, system, event);
2763         if (!file)
2764                 goto out;
2765
2766         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2767
2768         if (enable)
2769                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2770         else
2771                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2772
2773         if (glob[0] == '!') {
2774                 unregister_ftrace_function_probe_func(glob+1, ops);
2775                 ret = 0;
2776                 goto out;
2777         }
2778
2779         ret = -ENOMEM;
2780         data = kzalloc(sizeof(*data), GFP_KERNEL);
2781         if (!data)
2782                 goto out;
2783
2784         data->enable = enable;
2785         data->count = -1;
2786         data->file = file;
2787
2788         if (!param)
2789                 goto out_reg;
2790
2791         number = strsep(&param, ":");
2792
2793         ret = -EINVAL;
2794         if (!strlen(number))
2795                 goto out_free;
2796
2797         /*
2798          * We use the callback data field (which is a pointer)
2799          * as our counter.
2800          */
2801         ret = kstrtoul(number, 0, &data->count);
2802         if (ret)
2803                 goto out_free;
2804
2805  out_reg:
2806         /* Don't let event modules unload while probe registered */
2807         ret = try_module_get(file->event_call->mod);
2808         if (!ret) {
2809                 ret = -EBUSY;
2810                 goto out_free;
2811         }
2812
2813         ret = __ftrace_event_enable_disable(file, 1, 1);
2814         if (ret < 0)
2815                 goto out_put;
2816         ret = register_ftrace_function_probe(glob, ops, data);
2817         /*
2818          * The above returns on success the # of functions enabled,
2819          * but if it didn't find any functions it returns zero.
2820          * Consider no functions a failure too.
2821          */
2822         if (!ret) {
2823                 ret = -ENOENT;
2824                 goto out_disable;
2825         } else if (ret < 0)
2826                 goto out_disable;
2827         /* Just return zero, not the number of enabled functions */
2828         ret = 0;
2829  out:
2830         mutex_unlock(&event_mutex);
2831         return ret;
2832
2833  out_disable:
2834         __ftrace_event_enable_disable(file, 0, 1);
2835  out_put:
2836         module_put(file->event_call->mod);
2837  out_free:
2838         kfree(data);
2839         goto out;
2840 }
2841
2842 static struct ftrace_func_command event_enable_cmd = {
2843         .name                   = ENABLE_EVENT_STR,
2844         .func                   = event_enable_func,
2845 };
2846
2847 static struct ftrace_func_command event_disable_cmd = {
2848         .name                   = DISABLE_EVENT_STR,
2849         .func                   = event_enable_func,
2850 };
2851
2852 static __init int register_event_cmds(void)
2853 {
2854         int ret;
2855
2856         ret = register_ftrace_command(&event_enable_cmd);
2857         if (WARN_ON(ret < 0))
2858                 return ret;
2859         ret = register_ftrace_command(&event_disable_cmd);
2860         if (WARN_ON(ret < 0))
2861                 unregister_ftrace_command(&event_enable_cmd);
2862         return ret;
2863 }
2864 #else
2865 static inline int register_event_cmds(void) { return 0; }
2866 #endif /* CONFIG_DYNAMIC_FTRACE */
2867
2868 /*
2869  * The top level array has already had its trace_event_file
2870  * descriptors created in order to allow for early events to
2871  * be recorded. This function is called after the tracefs has been
2872  * initialized, and we now have to create the files associated
2873  * to the events.
2874  */
2875 static __init void
2876 __trace_early_add_event_dirs(struct trace_array *tr)
2877 {
2878         struct trace_event_file *file;
2879         int ret;
2880
2881
2882         list_for_each_entry(file, &tr->events, list) {
2883                 ret = event_create_dir(tr->event_dir, file);
2884                 if (ret < 0)
2885                         pr_warn("Could not create directory for event %s\n",
2886                                 trace_event_name(file->event_call));
2887         }
2888 }
2889
2890 /*
2891  * For early boot up, the top trace array requires to have
2892  * a list of events that can be enabled. This must be done before
2893  * the filesystem is set up in order to allow events to be traced
2894  * early.
2895  */
2896 static __init void
2897 __trace_early_add_events(struct trace_array *tr)
2898 {
2899         struct trace_event_call *call;
2900         int ret;
2901
2902         list_for_each_entry(call, &ftrace_events, list) {
2903                 /* Early boot up should not have any modules loaded */
2904                 if (WARN_ON_ONCE(call->mod))
2905                         continue;
2906
2907                 ret = __trace_early_add_new_event(call, tr);
2908                 if (ret < 0)
2909                         pr_warn("Could not create early event %s\n",
2910                                 trace_event_name(call));
2911         }
2912 }
2913
2914 /* Remove the event directory structure for a trace directory. */
2915 static void
2916 __trace_remove_event_dirs(struct trace_array *tr)
2917 {
2918         struct trace_event_file *file, *next;
2919
2920         list_for_each_entry_safe(file, next, &tr->events, list)
2921                 remove_event_file_dir(file);
2922 }
2923
2924 static void __add_event_to_tracers(struct trace_event_call *call)
2925 {
2926         struct trace_array *tr;
2927
2928         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2929                 __trace_add_new_event(call, tr);
2930 }
2931
2932 extern struct trace_event_call *__start_ftrace_events[];
2933 extern struct trace_event_call *__stop_ftrace_events[];
2934
2935 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2936
2937 static __init int setup_trace_event(char *str)
2938 {
2939         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2940         ring_buffer_expanded = true;
2941         tracing_selftest_disabled = true;
2942
2943         return 1;
2944 }
2945 __setup("trace_event=", setup_trace_event);
2946
2947 /* Expects to have event_mutex held when called */
2948 static int
2949 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2950 {
2951         struct dentry *d_events;
2952         struct dentry *entry;
2953
2954         entry = tracefs_create_file("set_event", 0644, parent,
2955                                     tr, &ftrace_set_event_fops);
2956         if (!entry) {
2957                 pr_warn("Could not create tracefs 'set_event' entry\n");
2958                 return -ENOMEM;
2959         }
2960
2961         d_events = tracefs_create_dir("events", parent);
2962         if (!d_events) {
2963                 pr_warn("Could not create tracefs 'events' directory\n");
2964                 return -ENOMEM;
2965         }
2966
2967         entry = tracefs_create_file("set_event_pid", 0644, parent,
2968                                     tr, &ftrace_set_event_pid_fops);
2969
2970         /* ring buffer internal formats */
2971         trace_create_file("header_page", 0444, d_events,
2972                           ring_buffer_print_page_header,
2973                           &ftrace_show_header_fops);
2974
2975         trace_create_file("header_event", 0444, d_events,
2976                           ring_buffer_print_entry_header,
2977                           &ftrace_show_header_fops);
2978
2979         trace_create_file("enable", 0644, d_events,
2980                           tr, &ftrace_tr_enable_fops);
2981
2982         tr->event_dir = d_events;
2983
2984         return 0;
2985 }
2986
2987 /**
2988  * event_trace_add_tracer - add a instance of a trace_array to events
2989  * @parent: The parent dentry to place the files/directories for events in
2990  * @tr: The trace array associated with these events
2991  *
2992  * When a new instance is created, it needs to set up its events
2993  * directory, as well as other files associated with events. It also
2994  * creates the event hierachry in the @parent/events directory.
2995  *
2996  * Returns 0 on success.
2997  */
2998 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2999 {
3000         int ret;
3001
3002         mutex_lock(&event_mutex);
3003
3004         ret = create_event_toplevel_files(parent, tr);
3005         if (ret)
3006                 goto out_unlock;
3007
3008         down_write(&trace_event_sem);
3009         __trace_add_event_dirs(tr);
3010         up_write(&trace_event_sem);
3011
3012  out_unlock:
3013         mutex_unlock(&event_mutex);
3014
3015         return ret;
3016 }
3017
3018 /*
3019  * The top trace array already had its file descriptors created.
3020  * Now the files themselves need to be created.
3021  */
3022 static __init int
3023 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3024 {
3025         int ret;
3026
3027         mutex_lock(&event_mutex);
3028
3029         ret = create_event_toplevel_files(parent, tr);
3030         if (ret)
3031                 goto out_unlock;
3032
3033         down_write(&trace_event_sem);
3034         __trace_early_add_event_dirs(tr);
3035         up_write(&trace_event_sem);
3036
3037  out_unlock:
3038         mutex_unlock(&event_mutex);
3039
3040         return ret;
3041 }
3042
3043 int event_trace_del_tracer(struct trace_array *tr)
3044 {
3045         mutex_lock(&event_mutex);
3046
3047         /* Disable any event triggers and associated soft-disabled events */
3048         clear_event_triggers(tr);
3049
3050         /* Clear the pid list */
3051         __ftrace_clear_event_pids(tr);
3052
3053         /* Disable any running events */
3054         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3055
3056         /* Access to events are within rcu_read_lock_sched() */
3057         synchronize_sched();
3058
3059         down_write(&trace_event_sem);
3060         __trace_remove_event_dirs(tr);
3061         tracefs_remove_recursive(tr->event_dir);
3062         up_write(&trace_event_sem);
3063
3064         tr->event_dir = NULL;
3065
3066         mutex_unlock(&event_mutex);
3067
3068         return 0;
3069 }
3070
3071 static __init int event_trace_memsetup(void)
3072 {
3073         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3074         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3075         return 0;
3076 }
3077
3078 static __init void
3079 early_enable_events(struct trace_array *tr, bool disable_first)
3080 {
3081         char *buf = bootup_event_buf;
3082         char *token;
3083         int ret;
3084
3085         while (true) {
3086                 token = strsep(&buf, ",");
3087
3088                 if (!token)
3089                         break;
3090
3091                 if (*token) {
3092                         /* Restarting syscalls requires that we stop them first */
3093                         if (disable_first)
3094                                 ftrace_set_clr_event(tr, token, 0);
3095
3096                         ret = ftrace_set_clr_event(tr, token, 1);
3097                         if (ret)
3098                                 pr_warn("Failed to enable trace event: %s\n", token);
3099                 }
3100
3101                 /* Put back the comma to allow this to be called again */
3102                 if (buf)
3103                         *(buf - 1) = ',';
3104         }
3105 }
3106
3107 static __init int event_trace_enable(void)
3108 {
3109         struct trace_array *tr = top_trace_array();
3110         struct trace_event_call **iter, *call;
3111         int ret;
3112
3113         if (!tr)
3114                 return -ENODEV;
3115
3116         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3117
3118                 call = *iter;
3119                 ret = event_init(call);
3120                 if (!ret)
3121                         list_add(&call->list, &ftrace_events);
3122         }
3123
3124         /*
3125          * We need the top trace array to have a working set of trace
3126          * points at early init, before the debug files and directories
3127          * are created. Create the file entries now, and attach them
3128          * to the actual file dentries later.
3129          */
3130         __trace_early_add_events(tr);
3131
3132         early_enable_events(tr, false);
3133
3134         trace_printk_start_comm();
3135
3136         register_event_cmds();
3137
3138         register_trigger_cmds();
3139
3140         return 0;
3141 }
3142
3143 /*
3144  * event_trace_enable() is called from trace_event_init() first to
3145  * initialize events and perhaps start any events that are on the
3146  * command line. Unfortunately, there are some events that will not
3147  * start this early, like the system call tracepoints that need
3148  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3149  * is called before pid 1 starts, and this flag is never set, making
3150  * the syscall tracepoint never get reached, but the event is enabled
3151  * regardless (and not doing anything).
3152  */
3153 static __init int event_trace_enable_again(void)
3154 {
3155         struct trace_array *tr;
3156
3157         tr = top_trace_array();
3158         if (!tr)
3159                 return -ENODEV;
3160
3161         early_enable_events(tr, true);
3162
3163         return 0;
3164 }
3165
3166 early_initcall(event_trace_enable_again);
3167
3168 static __init int event_trace_init(void)
3169 {
3170         struct trace_array *tr;
3171         struct dentry *d_tracer;
3172         struct dentry *entry;
3173         int ret;
3174
3175         tr = top_trace_array();
3176         if (!tr)
3177                 return -ENODEV;
3178
3179         d_tracer = tracing_init_dentry();
3180         if (IS_ERR(d_tracer))
3181                 return 0;
3182
3183         entry = tracefs_create_file("available_events", 0444, d_tracer,
3184                                     tr, &ftrace_avail_fops);
3185         if (!entry)
3186                 pr_warn("Could not create tracefs 'available_events' entry\n");
3187
3188         if (trace_define_generic_fields())
3189                 pr_warn("tracing: Failed to allocated generic fields");
3190
3191         if (trace_define_common_fields())
3192                 pr_warn("tracing: Failed to allocate common fields");
3193
3194         ret = early_event_add_tracer(d_tracer, tr);
3195         if (ret)
3196                 return ret;
3197
3198 #ifdef CONFIG_MODULES
3199         ret = register_module_notifier(&trace_module_nb);
3200         if (ret)
3201                 pr_warn("Failed to register trace events module notifier\n");
3202 #endif
3203         return 0;
3204 }
3205
3206 void __init trace_event_init(void)
3207 {
3208         event_trace_memsetup();
3209         init_ftrace_syscalls();
3210         event_trace_enable();
3211 }
3212
3213 fs_initcall(event_trace_init);
3214
3215 #ifdef CONFIG_FTRACE_STARTUP_TEST
3216
3217 static DEFINE_SPINLOCK(test_spinlock);
3218 static DEFINE_SPINLOCK(test_spinlock_irq);
3219 static DEFINE_MUTEX(test_mutex);
3220
3221 static __init void test_work(struct work_struct *dummy)
3222 {
3223         spin_lock(&test_spinlock);
3224         spin_lock_irq(&test_spinlock_irq);
3225         udelay(1);
3226         spin_unlock_irq(&test_spinlock_irq);
3227         spin_unlock(&test_spinlock);
3228
3229         mutex_lock(&test_mutex);
3230         msleep(1);
3231         mutex_unlock(&test_mutex);
3232 }
3233
3234 static __init int event_test_thread(void *unused)
3235 {
3236         void *test_malloc;
3237
3238         test_malloc = kmalloc(1234, GFP_KERNEL);
3239         if (!test_malloc)
3240                 pr_info("failed to kmalloc\n");
3241
3242         schedule_on_each_cpu(test_work);
3243
3244         kfree(test_malloc);
3245
3246         set_current_state(TASK_INTERRUPTIBLE);
3247         while (!kthread_should_stop()) {
3248                 schedule();
3249                 set_current_state(TASK_INTERRUPTIBLE);
3250         }
3251         __set_current_state(TASK_RUNNING);
3252
3253         return 0;
3254 }
3255
3256 /*
3257  * Do various things that may trigger events.
3258  */
3259 static __init void event_test_stuff(void)
3260 {
3261         struct task_struct *test_thread;
3262
3263         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3264         msleep(1);
3265         kthread_stop(test_thread);
3266 }
3267
3268 /*
3269  * For every trace event defined, we will test each trace point separately,
3270  * and then by groups, and finally all trace points.
3271  */
3272 static __init void event_trace_self_tests(void)
3273 {
3274         struct trace_subsystem_dir *dir;
3275         struct trace_event_file *file;
3276         struct trace_event_call *call;
3277         struct event_subsystem *system;
3278         struct trace_array *tr;
3279         int ret;
3280
3281         tr = top_trace_array();
3282         if (!tr)
3283                 return;
3284
3285         pr_info("Running tests on trace events:\n");
3286
3287         list_for_each_entry(file, &tr->events, list) {
3288
3289                 call = file->event_call;
3290
3291                 /* Only test those that have a probe */
3292                 if (!call->class || !call->class->probe)
3293                         continue;
3294
3295 /*
3296  * Testing syscall events here is pretty useless, but
3297  * we still do it if configured. But this is time consuming.
3298  * What we really need is a user thread to perform the
3299  * syscalls as we test.
3300  */
3301 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3302                 if (call->class->system &&
3303                     strcmp(call->class->system, "syscalls") == 0)
3304                         continue;
3305 #endif
3306
3307                 pr_info("Testing event %s: ", trace_event_name(call));
3308
3309                 /*
3310                  * If an event is already enabled, someone is using
3311                  * it and the self test should not be on.
3312                  */
3313                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3314                         pr_warn("Enabled event during self test!\n");
3315                         WARN_ON_ONCE(1);
3316                         continue;
3317                 }
3318
3319                 ftrace_event_enable_disable(file, 1);
3320                 event_test_stuff();
3321                 ftrace_event_enable_disable(file, 0);
3322
3323                 pr_cont("OK\n");
3324         }
3325
3326         /* Now test at the sub system level */
3327
3328         pr_info("Running tests on trace event systems:\n");
3329
3330         list_for_each_entry(dir, &tr->systems, list) {
3331
3332                 system = dir->subsystem;
3333
3334                 /* the ftrace system is special, skip it */
3335                 if (strcmp(system->name, "ftrace") == 0)
3336                         continue;
3337
3338                 pr_info("Testing event system %s: ", system->name);
3339
3340                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3341                 if (WARN_ON_ONCE(ret)) {
3342                         pr_warn("error enabling system %s\n",
3343                                 system->name);
3344                         continue;
3345                 }
3346
3347                 event_test_stuff();
3348
3349                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3350                 if (WARN_ON_ONCE(ret)) {
3351                         pr_warn("error disabling system %s\n",
3352                                 system->name);
3353                         continue;
3354                 }
3355
3356                 pr_cont("OK\n");
3357         }
3358
3359         /* Test with all events enabled */
3360
3361         pr_info("Running tests on all trace events:\n");
3362         pr_info("Testing all events: ");
3363
3364         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3365         if (WARN_ON_ONCE(ret)) {
3366                 pr_warn("error enabling all events\n");
3367                 return;
3368         }
3369
3370         event_test_stuff();
3371
3372         /* reset sysname */
3373         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3374         if (WARN_ON_ONCE(ret)) {
3375                 pr_warn("error disabling all events\n");
3376                 return;
3377         }
3378
3379         pr_cont("OK\n");
3380 }
3381
3382 #ifdef CONFIG_FUNCTION_TRACER
3383
3384 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3385
3386 static struct trace_array *event_tr;
3387
3388 static void __init
3389 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3390                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3391 {
3392         struct ring_buffer_event *event;
3393         struct ring_buffer *buffer;
3394         struct ftrace_entry *entry;
3395         unsigned long flags;
3396         long disabled;
3397         int cpu;
3398         int pc;
3399
3400         pc = preempt_count();
3401         preempt_disable_notrace();
3402         cpu = raw_smp_processor_id();
3403         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3404
3405         if (disabled != 1)
3406                 goto out;
3407
3408         local_save_flags(flags);
3409
3410         event = trace_current_buffer_lock_reserve(&buffer,
3411                                                   TRACE_FN, sizeof(*entry),
3412                                                   flags, pc);
3413         if (!event)
3414                 goto out;
3415         entry   = ring_buffer_event_data(event);
3416         entry->ip                       = ip;
3417         entry->parent_ip                = parent_ip;
3418
3419         trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3420
3421  out:
3422         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3423         preempt_enable_notrace();
3424 }
3425
3426 static struct ftrace_ops trace_ops __initdata  =
3427 {
3428         .func = function_test_events_call,
3429         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3430 };
3431
3432 static __init void event_trace_self_test_with_function(void)
3433 {
3434         int ret;
3435         event_tr = top_trace_array();
3436         if (WARN_ON(!event_tr))
3437                 return;
3438         ret = register_ftrace_function(&trace_ops);
3439         if (WARN_ON(ret < 0)) {
3440                 pr_info("Failed to enable function tracer for event tests\n");
3441                 return;
3442         }
3443         pr_info("Running tests again, along with the function tracer\n");
3444         event_trace_self_tests();
3445         unregister_ftrace_function(&trace_ops);
3446 }
3447 #else
3448 static __init void event_trace_self_test_with_function(void)
3449 {
3450 }
3451 #endif
3452
3453 static __init int event_trace_self_tests_init(void)
3454 {
3455         if (!tracing_selftest_disabled) {
3456                 event_trace_self_tests();
3457                 event_trace_self_test_with_function();
3458         }
3459
3460         return 0;
3461 }
3462
3463 late_initcall(event_trace_self_tests_init);
3464
3465 #endif