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