kernel/trace/trace_events.c

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