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         mutex_lock(&event_mutex);
 330         do_for_each_event_file(tr, file) {
 331
 332                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 333                         continue;
 334
 335                 if (enable) {
 336                         tracing_start_cmdline_record();
 337                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338                 } else {
 339                         tracing_stop_cmdline_record();
 340                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 341                 }
 342         } while_for_each_event_file();
 343         mutex_unlock(&event_mutex);
 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         mutex_lock(&event_mutex);
 352         do_for_each_event_file(tr, file) {
 353                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 354                         continue;
 355
 356                 if (enable) {
 357                         tracing_start_tgid_record();
 358                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 359                 } else {
 360                         tracing_stop_tgid_record();
 361                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 362                                   &file->flags);
 363                 }
 364         } while_for_each_event_file();
 365         mutex_unlock(&event_mutex);
 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_exit(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_exit(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         /*
 803          * The buf format can be <subsystem>:<event-name>
 804          *  *:<event-name> means any event by that name.
 805          *  :<event-name> is the same.
 806          *
 807          *  <subsystem>:* means all events in that subsystem
 808          *  <subsystem>: means the same.
 809          *
 810          *  <name> (no ':') means all events in a subsystem with
 811          *  the name <name> or any event that matches <name>
 812          */
 813
 814         match = strsep(&buf, ":");
 815         if (buf) {
 816                 sub = match;
 817                 event = buf;
 818                 match = NULL;
 819
 820                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 821                         sub = NULL;
 822                 if (!strlen(event) || strcmp(event, "*") == 0)
 823                         event = NULL;
 824         }
 825
 826         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 827
 828         /* Put back the colon to allow this to be called again */
 829         if (buf)
 830                 *(buf - 1) = ':';
 831
 832         return ret;
 833 }
 834
 835 /**
 836  * trace_set_clr_event - enable or disable an event
 837  * @system: system name to match (NULL for any system)
 838  * @event: event name to match (NULL for all events, within system)
 839  * @set: 1 to enable, 0 to disable
 840  *
 841  * This is a way for other parts of the kernel to enable or disable
 842  * event recording.
 843  *
 844  * Returns 0 on success, -EINVAL if the parameters do not match any
 845  * registered events.
 846  */
 847 int trace_set_clr_event(const char *system, const char *event, int set)
 848 {
 849         struct trace_array *tr = top_trace_array();
 850
 851         if (!tr)
 852                 return -ENODEV;
 853
 854         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 855 }
 856 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 857
 858 /* 128 should be much more than enough */
 859 #define EVENT_BUF_SIZE          127
 860
 861 static ssize_t
 862 ftrace_event_write(struct file *file, const char __user *ubuf,
 863                    size_t cnt, loff_t *ppos)
 864 {
 865         struct trace_parser parser;
 866         struct seq_file *m = file->private_data;
 867         struct trace_array *tr = m->private;
 868         ssize_t read, ret;
 869
 870         if (!cnt)
 871                 return 0;
 872
 873         ret = tracing_update_buffers();
 874         if (ret < 0)
 875                 return ret;
 876
 877         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 878                 return -ENOMEM;
 879
 880         read = trace_get_user(&parser, ubuf, cnt, ppos);
 881
 882         if (read >= 0 && trace_parser_loaded((&parser))) {
 883                 int set = 1;
 884
 885                 if (*parser.buffer == '!')
 886                         set = 0;
 887
 888                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 889                 if (ret)
 890                         goto out_put;
 891         }
 892
 893         ret = read;
 894
 895  out_put:
 896         trace_parser_put(&parser);
 897
 898         return ret;
 899 }
 900
 901 static void *
 902 t_next(struct seq_file *m, void *v, loff_t *pos)
 903 {
 904         struct trace_event_file *file = v;
 905         struct trace_event_call *call;
 906         struct trace_array *tr = m->private;
 907
 908         (*pos)++;
 909
 910         list_for_each_entry_continue(file, &tr->events, list) {
 911                 call = file->event_call;
 912                 /*
 913                  * The ftrace subsystem is for showing formats only.
 914                  * They can not be enabled or disabled via the event files.
 915                  */
 916                 if (call->class && call->class->reg &&
 917                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 918                         return file;
 919         }
 920
 921         return NULL;
 922 }
 923
 924 static void *t_start(struct seq_file *m, loff_t *pos)
 925 {
 926         struct trace_event_file *file;
 927         struct trace_array *tr = m->private;
 928         loff_t l;
 929
 930         mutex_lock(&event_mutex);
 931
 932         file = list_entry(&tr->events, struct trace_event_file, list);
 933         for (l = 0; l <= *pos; ) {
 934                 file = t_next(m, file, &l);
 935                 if (!file)
 936                         break;
 937         }
 938         return file;
 939 }
 940
 941 static void *
 942 s_next(struct seq_file *m, void *v, loff_t *pos)
 943 {
 944         struct trace_event_file *file = v;
 945         struct trace_array *tr = m->private;
 946
 947         (*pos)++;
 948
 949         list_for_each_entry_continue(file, &tr->events, list) {
 950                 if (file->flags & EVENT_FILE_FL_ENABLED)
 951                         return file;
 952         }
 953
 954         return NULL;
 955 }
 956
 957 static void *s_start(struct seq_file *m, loff_t *pos)
 958 {
 959         struct trace_event_file *file;
 960         struct trace_array *tr = m->private;
 961         loff_t l;
 962
 963         mutex_lock(&event_mutex);
 964
 965         file = list_entry(&tr->events, struct trace_event_file, list);
 966         for (l = 0; l <= *pos; ) {
 967                 file = s_next(m, file, &l);
 968                 if (!file)
 969                         break;
 970         }
 971         return file;
 972 }
 973
 974 static int t_show(struct seq_file *m, void *v)
 975 {
 976         struct trace_event_file *file = v;
 977         struct trace_event_call *call = file->event_call;
 978
 979         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 980                 seq_printf(m, "%s:", call->class->system);
 981         seq_printf(m, "%s\n", trace_event_name(call));
 982
 983         return 0;
 984 }
 985
 986 static void t_stop(struct seq_file *m, void *p)
 987 {
 988         mutex_unlock(&event_mutex);
 989 }
 990
 991 static void *
 992 p_next(struct seq_file *m, void *v, loff_t *pos)
 993 {
 994         struct trace_array *tr = m->private;
 995         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 996
 997         return trace_pid_next(pid_list, v, pos);
 998 }
 999
1000 static void *p_start(struct seq_file *m, loff_t *pos)
1001         __acquires(RCU)
1002 {
1003         struct trace_pid_list *pid_list;
1004         struct trace_array *tr = m->private;
1005
1006         /*
1007          * Grab the mutex, to keep calls to p_next() having the same
1008          * tr->filtered_pids as p_start() has.
1009          * If we just passed the tr->filtered_pids around, then RCU would
1010          * have been enough, but doing that makes things more complex.
1011          */
1012         mutex_lock(&event_mutex);
1013         rcu_read_lock_sched();
1014
1015         pid_list = rcu_dereference_sched(tr->filtered_pids);
1016
1017         if (!pid_list)
1018                 return NULL;
1019
1020         return trace_pid_start(pid_list, pos);
1021 }
1022
1023 static void p_stop(struct seq_file *m, void *p)
1024         __releases(RCU)
1025 {
1026         rcu_read_unlock_sched();
1027         mutex_unlock(&event_mutex);
1028 }
1029
1030 static ssize_t
1031 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1032                   loff_t *ppos)
1033 {
1034         struct trace_event_file *file;
1035         unsigned long flags;
1036         char buf[4] = "0";
1037
1038         mutex_lock(&event_mutex);
1039         file = event_file_data(filp);
1040         if (likely(file))
1041                 flags = file->flags;
1042         mutex_unlock(&event_mutex);
1043
1044         if (!file)
1045                 return -ENODEV;
1046
1047         if (flags & EVENT_FILE_FL_ENABLED &&
1048             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1049                 strcpy(buf, "1");
1050
1051         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1052             flags & EVENT_FILE_FL_SOFT_MODE)
1053                 strcat(buf, "*");
1054
1055         strcat(buf, "\n");
1056
1057         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1058 }
1059
1060 static ssize_t
1061 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1062                    loff_t *ppos)
1063 {
1064         struct trace_event_file *file;
1065         unsigned long val;
1066         int ret;
1067
1068         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1069         if (ret)
1070                 return ret;
1071
1072         ret = tracing_update_buffers();
1073         if (ret < 0)
1074                 return ret;
1075
1076         switch (val) {
1077         case 0:
1078         case 1:
1079                 ret = -ENODEV;
1080                 mutex_lock(&event_mutex);
1081                 file = event_file_data(filp);
1082                 if (likely(file))
1083                         ret = ftrace_event_enable_disable(file, val);
1084                 mutex_unlock(&event_mutex);
1085                 break;
1086
1087         default:
1088                 return -EINVAL;
1089         }
1090
1091         *ppos += cnt;
1092
1093         return ret ? ret : cnt;
1094 }
1095
1096 static ssize_t
1097 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1098                    loff_t *ppos)
1099 {
1100         const char set_to_char[4] = { '?', '0', '1', 'X' };
1101         struct trace_subsystem_dir *dir = filp->private_data;
1102         struct event_subsystem *system = dir->subsystem;
1103         struct trace_event_call *call;
1104         struct trace_event_file *file;
1105         struct trace_array *tr = dir->tr;
1106         char buf[2];
1107         int set = 0;
1108         int ret;
1109
1110         mutex_lock(&event_mutex);
1111         list_for_each_entry(file, &tr->events, list) {
1112                 call = file->event_call;
1113                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1114                         continue;
1115
1116                 if (system && strcmp(call->class->system, system->name) != 0)
1117                         continue;
1118
1119                 /*
1120                  * We need to find out if all the events are set
1121                  * or if all events or cleared, or if we have
1122                  * a mixture.
1123                  */
1124                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1125
1126                 /*
1127                  * If we have a mixture, no need to look further.
1128                  */
1129                 if (set == 3)
1130                         break;
1131         }
1132         mutex_unlock(&event_mutex);
1133
1134         buf[0] = set_to_char[set];
1135         buf[1] = '\n';
1136
1137         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1138
1139         return ret;
1140 }
1141
1142 static ssize_t
1143 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1144                     loff_t *ppos)
1145 {
1146         struct trace_subsystem_dir *dir = filp->private_data;
1147         struct event_subsystem *system = dir->subsystem;
1148         const char *name = NULL;
1149         unsigned long val;
1150         ssize_t ret;
1151
1152         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1153         if (ret)
1154                 return ret;
1155
1156         ret = tracing_update_buffers();
1157         if (ret < 0)
1158                 return ret;
1159
1160         if (val != 0 && val != 1)
1161                 return -EINVAL;
1162
1163         /*
1164          * Opening of "enable" adds a ref count to system,
1165          * so the name is safe to use.
1166          */
1167         if (system)
1168                 name = system->name;
1169
1170         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1171         if (ret)
1172                 goto out;
1173
1174         ret = cnt;
1175
1176 out:
1177         *ppos += cnt;
1178
1179         return ret;
1180 }
1181
1182 enum {
1183         FORMAT_HEADER           = 1,
1184         FORMAT_FIELD_SEPERATOR  = 2,
1185         FORMAT_PRINTFMT         = 3,
1186 };
1187
1188 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1189 {
1190         struct trace_event_call *call = event_file_data(m->private);
1191         struct list_head *common_head = &ftrace_common_fields;
1192         struct list_head *head = trace_get_fields(call);
1193         struct list_head *node = v;
1194
1195         (*pos)++;
1196
1197         switch ((unsigned long)v) {
1198         case FORMAT_HEADER:
1199                 node = common_head;
1200                 break;
1201
1202         case FORMAT_FIELD_SEPERATOR:
1203                 node = head;
1204                 break;
1205
1206         case FORMAT_PRINTFMT:
1207                 /* all done */
1208                 return NULL;
1209         }
1210
1211         node = node->prev;
1212         if (node == common_head)
1213                 return (void *)FORMAT_FIELD_SEPERATOR;
1214         else if (node == head)
1215                 return (void *)FORMAT_PRINTFMT;
1216         else
1217                 return node;
1218 }
1219
1220 static int f_show(struct seq_file *m, void *v)
1221 {
1222         struct trace_event_call *call = event_file_data(m->private);
1223         struct ftrace_event_field *field;
1224         const char *array_descriptor;
1225
1226         switch ((unsigned long)v) {
1227         case FORMAT_HEADER:
1228                 seq_printf(m, "name: %s\n", trace_event_name(call));
1229                 seq_printf(m, "ID: %d\n", call->event.type);
1230                 seq_puts(m, "format:\n");
1231                 return 0;
1232
1233         case FORMAT_FIELD_SEPERATOR:
1234                 seq_putc(m, '\n');
1235                 return 0;
1236
1237         case FORMAT_PRINTFMT:
1238                 seq_printf(m, "\nprint fmt: %s\n",
1239                            call->print_fmt);
1240                 return 0;
1241         }
1242
1243         field = list_entry(v, struct ftrace_event_field, link);
1244         /*
1245          * Smartly shows the array type(except dynamic array).
1246          * Normal:
1247          *      field:TYPE VAR
1248          * If TYPE := TYPE[LEN], it is shown:
1249          *      field:TYPE VAR[LEN]
1250          */
1251         array_descriptor = strchr(field->type, '[');
1252
1253         if (!strncmp(field->type, "__data_loc", 10))
1254                 array_descriptor = NULL;
1255
1256         if (!array_descriptor)
1257                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1258                            field->type, field->name, field->offset,
1259                            field->size, !!field->is_signed);
1260         else
1261                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1262                            (int)(array_descriptor - field->type),
1263                            field->type, field->name,
1264                            array_descriptor, field->offset,
1265                            field->size, !!field->is_signed);
1266
1267         return 0;
1268 }
1269
1270 static void *f_start(struct seq_file *m, loff_t *pos)
1271 {
1272         void *p = (void *)FORMAT_HEADER;
1273         loff_t l = 0;
1274
1275         /* ->stop() is called even if ->start() fails */
1276         mutex_lock(&event_mutex);
1277         if (!event_file_data(m->private))
1278                 return ERR_PTR(-ENODEV);
1279
1280         while (l < *pos && p)
1281                 p = f_next(m, p, &l);
1282
1283         return p;
1284 }
1285
1286 static void f_stop(struct seq_file *m, void *p)
1287 {
1288         mutex_unlock(&event_mutex);
1289 }
1290
1291 static const struct seq_operations trace_format_seq_ops = {
1292         .start          = f_start,
1293         .next           = f_next,
1294         .stop           = f_stop,
1295         .show           = f_show,
1296 };
1297
1298 static int trace_format_open(struct inode *inode, struct file *file)
1299 {
1300         struct seq_file *m;
1301         int ret;
1302
1303         ret = seq_open(file, &trace_format_seq_ops);
1304         if (ret < 0)
1305                 return ret;
1306
1307         m = file->private_data;
1308         m->private = file;
1309
1310         return 0;
1311 }
1312
1313 static ssize_t
1314 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1315 {
1316         int id = (long)event_file_data(filp);
1317         char buf[32];
1318         int len;
1319
1320         if (*ppos)
1321                 return 0;
1322
1323         if (unlikely(!id))
1324                 return -ENODEV;
1325
1326         len = sprintf(buf, "%d\n", id);
1327
1328         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1329 }
1330
1331 static ssize_t
1332 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1333                   loff_t *ppos)
1334 {
1335         struct trace_event_file *file;
1336         struct trace_seq *s;
1337         int r = -ENODEV;
1338
1339         if (*ppos)
1340                 return 0;
1341
1342         s = kmalloc(sizeof(*s), GFP_KERNEL);
1343
1344         if (!s)
1345                 return -ENOMEM;
1346
1347         trace_seq_init(s);
1348
1349         mutex_lock(&event_mutex);
1350         file = event_file_data(filp);
1351         if (file)
1352                 print_event_filter(file, s);
1353         mutex_unlock(&event_mutex);
1354
1355         if (file)
1356                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1357                                             s->buffer, trace_seq_used(s));
1358
1359         kfree(s);
1360
1361         return r;
1362 }
1363
1364 static ssize_t
1365 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1366                    loff_t *ppos)
1367 {
1368         struct trace_event_file *file;
1369         char *buf;
1370         int err = -ENODEV;
1371
1372         if (cnt >= PAGE_SIZE)
1373                 return -EINVAL;
1374
1375         buf = memdup_user_nul(ubuf, cnt);
1376         if (IS_ERR(buf))
1377                 return PTR_ERR(buf);
1378
1379         mutex_lock(&event_mutex);
1380         file = event_file_data(filp);
1381         if (file)
1382                 err = apply_event_filter(file, buf);
1383         mutex_unlock(&event_mutex);
1384
1385         kfree(buf);
1386         if (err < 0)
1387                 return err;
1388
1389         *ppos += cnt;
1390
1391         return cnt;
1392 }
1393
1394 static LIST_HEAD(event_subsystems);
1395
1396 static int subsystem_open(struct inode *inode, struct file *filp)
1397 {
1398         struct event_subsystem *system = NULL;
1399         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1400         struct trace_array *tr;
1401         int ret;
1402
1403         if (tracing_is_disabled())
1404                 return -ENODEV;
1405
1406         /* Make sure the system still exists */
1407         mutex_lock(&event_mutex);
1408         mutex_lock(&trace_types_lock);
1409         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1410                 list_for_each_entry(dir, &tr->systems, list) {
1411                         if (dir == inode->i_private) {
1412                                 /* Don't open systems with no events */
1413                                 if (dir->nr_events) {
1414                                         __get_system_dir(dir);
1415                                         system = dir->subsystem;
1416                                 }
1417                                 goto exit_loop;
1418                         }
1419                 }
1420         }
1421  exit_loop:
1422         mutex_unlock(&trace_types_lock);
1423         mutex_unlock(&event_mutex);
1424
1425         if (!system)
1426                 return -ENODEV;
1427
1428         /* Some versions of gcc think dir can be uninitialized here */
1429         WARN_ON(!dir);
1430
1431         /* Still need to increment the ref count of the system */
1432         if (trace_array_get(tr) < 0) {
1433                 put_system(dir);
1434                 return -ENODEV;
1435         }
1436
1437         ret = tracing_open_generic(inode, filp);
1438         if (ret < 0) {
1439                 trace_array_put(tr);
1440                 put_system(dir);
1441         }
1442
1443         return ret;
1444 }
1445
1446 static int system_tr_open(struct inode *inode, struct file *filp)
1447 {
1448         struct trace_subsystem_dir *dir;
1449         struct trace_array *tr = inode->i_private;
1450         int ret;
1451
1452         if (tracing_is_disabled())
1453                 return -ENODEV;
1454
1455         if (trace_array_get(tr) < 0)
1456                 return -ENODEV;
1457
1458         /* Make a temporary dir that has no system but points to tr */
1459         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1460         if (!dir) {
1461                 trace_array_put(tr);
1462                 return -ENOMEM;
1463         }
1464
1465         dir->tr = tr;
1466
1467         ret = tracing_open_generic(inode, filp);
1468         if (ret < 0) {
1469                 trace_array_put(tr);
1470                 kfree(dir);
1471                 return ret;
1472         }
1473
1474         filp->private_data = dir;
1475
1476         return 0;
1477 }
1478
1479 static int subsystem_release(struct inode *inode, struct file *file)
1480 {
1481         struct trace_subsystem_dir *dir = file->private_data;
1482
1483         trace_array_put(dir->tr);
1484
1485         /*
1486          * If dir->subsystem is NULL, then this is a temporary
1487          * descriptor that was made for a trace_array to enable
1488          * all subsystems.
1489          */
1490         if (dir->subsystem)
1491                 put_system(dir);
1492         else
1493                 kfree(dir);
1494
1495         return 0;
1496 }
1497
1498 static ssize_t
1499 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1500                       loff_t *ppos)
1501 {
1502         struct trace_subsystem_dir *dir = filp->private_data;
1503         struct event_subsystem *system = dir->subsystem;
1504         struct trace_seq *s;
1505         int r;
1506
1507         if (*ppos)
1508                 return 0;
1509
1510         s = kmalloc(sizeof(*s), GFP_KERNEL);
1511         if (!s)
1512                 return -ENOMEM;
1513
1514         trace_seq_init(s);
1515
1516         print_subsystem_event_filter(system, s);
1517         r = simple_read_from_buffer(ubuf, cnt, ppos,
1518                                     s->buffer, trace_seq_used(s));
1519
1520         kfree(s);
1521
1522         return r;
1523 }
1524
1525 static ssize_t
1526 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1527                        loff_t *ppos)
1528 {
1529         struct trace_subsystem_dir *dir = filp->private_data;
1530         char *buf;
1531         int err;
1532
1533         if (cnt >= PAGE_SIZE)
1534                 return -EINVAL;
1535
1536         buf = memdup_user_nul(ubuf, cnt);
1537         if (IS_ERR(buf))
1538                 return PTR_ERR(buf);
1539
1540         err = apply_subsystem_event_filter(dir, buf);
1541         kfree(buf);
1542         if (err < 0)
1543                 return err;
1544
1545         *ppos += cnt;
1546
1547         return cnt;
1548 }
1549
1550 static ssize_t
1551 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1552 {
1553         int (*func)(struct trace_seq *s) = filp->private_data;
1554         struct trace_seq *s;
1555         int r;
1556
1557         if (*ppos)
1558                 return 0;
1559
1560         s = kmalloc(sizeof(*s), GFP_KERNEL);
1561         if (!s)
1562                 return -ENOMEM;
1563
1564         trace_seq_init(s);
1565
1566         func(s);
1567         r = simple_read_from_buffer(ubuf, cnt, ppos,
1568                                     s->buffer, trace_seq_used(s));
1569
1570         kfree(s);
1571
1572         return r;
1573 }
1574
1575 static void ignore_task_cpu(void *data)
1576 {
1577         struct trace_array *tr = data;
1578         struct trace_pid_list *pid_list;
1579
1580         /*
1581          * This function is called by on_each_cpu() while the
1582          * event_mutex is held.
1583          */
1584         pid_list = rcu_dereference_protected(tr->filtered_pids,
1585                                              mutex_is_locked(&event_mutex));
1586
1587         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1588                        trace_ignore_this_task(pid_list, current));
1589 }
1590
1591 static ssize_t
1592 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1593                        size_t cnt, loff_t *ppos)
1594 {
1595         struct seq_file *m = filp->private_data;
1596         struct trace_array *tr = m->private;
1597         struct trace_pid_list *filtered_pids = NULL;
1598         struct trace_pid_list *pid_list;
1599         struct trace_event_file *file;
1600         ssize_t ret;
1601
1602         if (!cnt)
1603                 return 0;
1604
1605         ret = tracing_update_buffers();
1606         if (ret < 0)
1607                 return ret;
1608
1609         mutex_lock(&event_mutex);
1610
1611         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1612                                              lockdep_is_held(&event_mutex));
1613
1614         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1615         if (ret < 0)
1616                 goto out;
1617
1618         rcu_assign_pointer(tr->filtered_pids, pid_list);
1619
1620         list_for_each_entry(file, &tr->events, list) {
1621                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1622         }
1623
1624         if (filtered_pids) {
1625                 synchronize_sched();
1626                 trace_free_pid_list(filtered_pids);
1627         } else if (pid_list) {
1628                 /*
1629                  * Register a probe that is called before all other probes
1630                  * to set ignore_pid if next or prev do not match.
1631                  * Register a probe this is called after all other probes
1632                  * to only keep ignore_pid set if next pid matches.
1633                  */
1634                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1635                                                  tr, INT_MAX);
1636                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1637                                                  tr, 0);
1638
1639                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1640                                                  tr, INT_MAX);
1641                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1642                                                  tr, 0);
1643
1644                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1645                                                      tr, INT_MAX);
1646                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1647                                                      tr, 0);
1648
1649                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1650                                                  tr, INT_MAX);
1651                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1652                                                  tr, 0);
1653         }
1654
1655         /*
1656          * Ignoring of pids is done at task switch. But we have to
1657          * check for those tasks that are currently running.
1658          * Always do this in case a pid was appended or removed.
1659          */
1660         on_each_cpu(ignore_task_cpu, tr, 1);
1661
1662  out:
1663         mutex_unlock(&event_mutex);
1664
1665         if (ret > 0)
1666                 *ppos += ret;
1667
1668         return ret;
1669 }
1670
1671 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1672 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_release(struct inode *inode, struct file *file);
1675
1676 static const struct seq_operations show_event_seq_ops = {
1677         .start = t_start,
1678         .next = t_next,
1679         .show = t_show,
1680         .stop = t_stop,
1681 };
1682
1683 static const struct seq_operations show_set_event_seq_ops = {
1684         .start = s_start,
1685         .next = s_next,
1686         .show = t_show,
1687         .stop = t_stop,
1688 };
1689
1690 static const struct seq_operations show_set_pid_seq_ops = {
1691         .start = p_start,
1692         .next = p_next,
1693         .show = trace_pid_show,
1694         .stop = p_stop,
1695 };
1696
1697 static const struct file_operations ftrace_avail_fops = {
1698         .open = ftrace_event_avail_open,
1699         .read = seq_read,
1700         .llseek = seq_lseek,
1701         .release = seq_release,
1702 };
1703
1704 static const struct file_operations ftrace_set_event_fops = {
1705         .open = ftrace_event_set_open,
1706         .read = seq_read,
1707         .write = ftrace_event_write,
1708         .llseek = seq_lseek,
1709         .release = ftrace_event_release,
1710 };
1711
1712 static const struct file_operations ftrace_set_event_pid_fops = {
1713         .open = ftrace_event_set_pid_open,
1714         .read = seq_read,
1715         .write = ftrace_event_pid_write,
1716         .llseek = seq_lseek,
1717         .release = ftrace_event_release,
1718 };
1719
1720 static const struct file_operations ftrace_enable_fops = {
1721         .open = tracing_open_generic,
1722         .read = event_enable_read,
1723         .write = event_enable_write,
1724         .llseek = default_llseek,
1725 };
1726
1727 static const struct file_operations ftrace_event_format_fops = {
1728         .open = trace_format_open,
1729         .read = seq_read,
1730         .llseek = seq_lseek,
1731         .release = seq_release,
1732 };
1733
1734 static const struct file_operations ftrace_event_id_fops = {
1735         .read = event_id_read,
1736         .llseek = default_llseek,
1737 };
1738
1739 static const struct file_operations ftrace_event_filter_fops = {
1740         .open = tracing_open_generic,
1741         .read = event_filter_read,
1742         .write = event_filter_write,
1743         .llseek = default_llseek,
1744 };
1745
1746 static const struct file_operations ftrace_subsystem_filter_fops = {
1747         .open = subsystem_open,
1748         .read = subsystem_filter_read,
1749         .write = subsystem_filter_write,
1750         .llseek = default_llseek,
1751         .release = subsystem_release,
1752 };
1753
1754 static const struct file_operations ftrace_system_enable_fops = {
1755         .open = subsystem_open,
1756         .read = system_enable_read,
1757         .write = system_enable_write,
1758         .llseek = default_llseek,
1759         .release = subsystem_release,
1760 };
1761
1762 static const struct file_operations ftrace_tr_enable_fops = {
1763         .open = system_tr_open,
1764         .read = system_enable_read,
1765         .write = system_enable_write,
1766         .llseek = default_llseek,
1767         .release = subsystem_release,
1768 };
1769
1770 static const struct file_operations ftrace_show_header_fops = {
1771         .open = tracing_open_generic,
1772         .read = show_header,
1773         .llseek = default_llseek,
1774 };
1775
1776 static int
1777 ftrace_event_open(struct inode *inode, struct file *file,
1778                   const struct seq_operations *seq_ops)
1779 {
1780         struct seq_file *m;
1781         int ret;
1782
1783         ret = seq_open(file, seq_ops);
1784         if (ret < 0)
1785                 return ret;
1786         m = file->private_data;
1787         /* copy tr over to seq ops */
1788         m->private = inode->i_private;
1789
1790         return ret;
1791 }
1792
1793 static int ftrace_event_release(struct inode *inode, struct file *file)
1794 {
1795         struct trace_array *tr = inode->i_private;
1796
1797         trace_array_put(tr);
1798
1799         return seq_release(inode, file);
1800 }
1801
1802 static int
1803 ftrace_event_avail_open(struct inode *inode, struct file *file)
1804 {
1805         const struct seq_operations *seq_ops = &show_event_seq_ops;
1806
1807         return ftrace_event_open(inode, file, seq_ops);
1808 }
1809
1810 static int
1811 ftrace_event_set_open(struct inode *inode, struct file *file)
1812 {
1813         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1814         struct trace_array *tr = inode->i_private;
1815         int ret;
1816
1817         if (trace_array_get(tr) < 0)
1818                 return -ENODEV;
1819
1820         if ((file->f_mode & FMODE_WRITE) &&
1821             (file->f_flags & O_TRUNC))
1822                 ftrace_clear_events(tr);
1823
1824         ret = ftrace_event_open(inode, file, seq_ops);
1825         if (ret < 0)
1826                 trace_array_put(tr);
1827         return ret;
1828 }
1829
1830 static int
1831 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1832 {
1833         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1834         struct trace_array *tr = inode->i_private;
1835         int ret;
1836
1837         if (trace_array_get(tr) < 0)
1838                 return -ENODEV;
1839
1840         if ((file->f_mode & FMODE_WRITE) &&
1841             (file->f_flags & O_TRUNC))
1842                 ftrace_clear_event_pids(tr);
1843
1844         ret = ftrace_event_open(inode, file, seq_ops);
1845         if (ret < 0)
1846                 trace_array_put(tr);
1847         return ret;
1848 }
1849
1850 static struct event_subsystem *
1851 create_new_subsystem(const char *name)
1852 {
1853         struct event_subsystem *system;
1854
1855         /* need to create new entry */
1856         system = kmalloc(sizeof(*system), GFP_KERNEL);
1857         if (!system)
1858                 return NULL;
1859
1860         system->ref_count = 1;
1861
1862         /* Only allocate if dynamic (kprobes and modules) */
1863         system->name = kstrdup_const(name, GFP_KERNEL);
1864         if (!system->name)
1865                 goto out_free;
1866
1867         system->filter = NULL;
1868
1869         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1870         if (!system->filter)
1871                 goto out_free;
1872
1873         list_add(&system->list, &event_subsystems);
1874
1875         return system;
1876
1877  out_free:
1878         kfree_const(system->name);
1879         kfree(system);
1880         return NULL;
1881 }
1882
1883 static struct dentry *
1884 event_subsystem_dir(struct trace_array *tr, const char *name,
1885                     struct trace_event_file *file, struct dentry *parent)
1886 {
1887         struct trace_subsystem_dir *dir;
1888         struct event_subsystem *system;
1889         struct dentry *entry;
1890
1891         /* First see if we did not already create this dir */
1892         list_for_each_entry(dir, &tr->systems, list) {
1893                 system = dir->subsystem;
1894                 if (strcmp(system->name, name) == 0) {
1895                         dir->nr_events++;
1896                         file->system = dir;
1897                         return dir->entry;
1898                 }
1899         }
1900
1901         /* Now see if the system itself exists. */
1902         list_for_each_entry(system, &event_subsystems, list) {
1903                 if (strcmp(system->name, name) == 0)
1904                         break;
1905         }
1906         /* Reset system variable when not found */
1907         if (&system->list == &event_subsystems)
1908                 system = NULL;
1909
1910         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1911         if (!dir)
1912                 goto out_fail;
1913
1914         if (!system) {
1915                 system = create_new_subsystem(name);
1916                 if (!system)
1917                         goto out_free;
1918         } else
1919                 __get_system(system);
1920
1921         dir->entry = tracefs_create_dir(name, parent);
1922         if (!dir->entry) {
1923                 pr_warn("Failed to create system directory %s\n", name);
1924                 __put_system(system);
1925                 goto out_free;
1926         }
1927
1928         dir->tr = tr;
1929         dir->ref_count = 1;
1930         dir->nr_events = 1;
1931         dir->subsystem = system;
1932         file->system = dir;
1933
1934         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1935                                     &ftrace_subsystem_filter_fops);
1936         if (!entry) {
1937                 kfree(system->filter);
1938                 system->filter = NULL;
1939                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1940         }
1941
1942         trace_create_file("enable", 0644, dir->entry, dir,
1943                           &ftrace_system_enable_fops);
1944
1945         list_add(&dir->list, &tr->systems);
1946
1947         return dir->entry;
1948
1949  out_free:
1950         kfree(dir);
1951  out_fail:
1952         /* Only print this message if failed on memory allocation */
1953         if (!dir || !system)
1954                 pr_warn("No memory to create event subsystem %s\n", name);
1955         return NULL;
1956 }
1957
1958 static int
1959 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1960 {
1961         struct trace_event_call *call = file->event_call;
1962         struct trace_array *tr = file->tr;
1963         struct list_head *head;
1964         struct dentry *d_events;
1965         const char *name;
1966         int ret;
1967
1968         /*
1969          * If the trace point header did not define TRACE_SYSTEM
1970          * then the system would be called "TRACE_SYSTEM".
1971          */
1972         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1973                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1974                 if (!d_events)
1975                         return -ENOMEM;
1976         } else
1977                 d_events = parent;
1978
1979         name = trace_event_name(call);
1980         file->dir = tracefs_create_dir(name, d_events);
1981         if (!file->dir) {
1982                 pr_warn("Could not create tracefs '%s' directory\n", name);
1983                 return -1;
1984         }
1985
1986         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1987                 trace_create_file("enable", 0644, file->dir, file,
1988                                   &ftrace_enable_fops);
1989
1990 #ifdef CONFIG_PERF_EVENTS
1991         if (call->event.type && call->class->reg)
1992                 trace_create_file("id", 0444, file->dir,
1993                                   (void *)(long)call->event.type,
1994                                   &ftrace_event_id_fops);
1995 #endif
1996
1997         /*
1998          * Other events may have the same class. Only update
1999          * the fields if they are not already defined.
2000          */
2001         head = trace_get_fields(call);
2002         if (list_empty(head)) {
2003                 ret = call->class->define_fields(call);
2004                 if (ret < 0) {
2005                         pr_warn("Could not initialize trace point events/%s\n",
2006                                 name);
2007                         return -1;
2008                 }
2009         }
2010
2011         /*
2012          * Only event directories that can be enabled should have
2013          * triggers or filters.
2014          */
2015         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2016                 trace_create_file("filter", 0644, file->dir, file,
2017                                   &ftrace_event_filter_fops);
2018
2019                 trace_create_file("trigger", 0644, file->dir, file,
2020                                   &event_trigger_fops);
2021         }
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 /* Returns any file that matches the system and event */
2479 struct trace_event_file *
2480 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2481 {
2482         struct trace_event_file *file;
2483         struct trace_event_call *call;
2484         const char *name;
2485
2486         list_for_each_entry(file, &tr->events, list) {
2487
2488                 call = file->event_call;
2489                 name = trace_event_name(call);
2490
2491                 if (!name || !call->class)
2492                         continue;
2493
2494                 if (strcmp(event, name) == 0 &&
2495                     strcmp(system, call->class->system) == 0)
2496                         return file;
2497         }
2498         return NULL;
2499 }
2500
2501 /* Returns valid trace event files that match system and event */
2502 struct trace_event_file *
2503 find_event_file(struct trace_array *tr, const char *system, const char *event)
2504 {
2505         struct trace_event_file *file;
2506
2507         file = __find_event_file(tr, system, event);
2508         if (!file || !file->event_call->class->reg ||
2509             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2510                 return NULL;
2511
2512         return file;
2513 }
2514
2515 #ifdef CONFIG_DYNAMIC_FTRACE
2516
2517 /* Avoid typos */
2518 #define ENABLE_EVENT_STR        "enable_event"
2519 #define DISABLE_EVENT_STR       "disable_event"
2520
2521 struct event_probe_data {
2522         struct trace_event_file *file;
2523         unsigned long                   count;
2524         int                             ref;
2525         bool                            enable;
2526 };
2527
2528 static void update_event_probe(struct event_probe_data *data)
2529 {
2530         if (data->enable)
2531                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2532         else
2533                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2534 }
2535
2536 static void
2537 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2538                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2539                    void *data)
2540 {
2541         struct ftrace_func_mapper *mapper = data;
2542         struct event_probe_data *edata;
2543         void **pdata;
2544
2545         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2546         if (!pdata || !*pdata)
2547                 return;
2548
2549         edata = *pdata;
2550         update_event_probe(edata);
2551 }
2552
2553 static void
2554 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2555                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2556                          void *data)
2557 {
2558         struct ftrace_func_mapper *mapper = data;
2559         struct event_probe_data *edata;
2560         void **pdata;
2561
2562         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2563         if (!pdata || !*pdata)
2564                 return;
2565
2566         edata = *pdata;
2567
2568         if (!edata->count)
2569                 return;
2570
2571         /* Skip if the event is in a state we want to switch to */
2572         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2573                 return;
2574
2575         if (edata->count != -1)
2576                 (edata->count)--;
2577
2578         update_event_probe(edata);
2579 }
2580
2581 static int
2582 event_enable_print(struct seq_file *m, unsigned long ip,
2583                    struct ftrace_probe_ops *ops, void *data)
2584 {
2585         struct ftrace_func_mapper *mapper = data;
2586         struct event_probe_data *edata;
2587         void **pdata;
2588
2589         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2590
2591         if (WARN_ON_ONCE(!pdata || !*pdata))
2592                 return 0;
2593
2594         edata = *pdata;
2595
2596         seq_printf(m, "%ps:", (void *)ip);
2597
2598         seq_printf(m, "%s:%s:%s",
2599                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2600                    edata->file->event_call->class->system,
2601                    trace_event_name(edata->file->event_call));
2602
2603         if (edata->count == -1)
2604                 seq_puts(m, ":unlimited\n");
2605         else
2606                 seq_printf(m, ":count=%ld\n", edata->count);
2607
2608         return 0;
2609 }
2610
2611 static int
2612 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2613                   unsigned long ip, void *init_data, void **data)
2614 {
2615         struct ftrace_func_mapper *mapper = *data;
2616         struct event_probe_data *edata = init_data;
2617         int ret;
2618
2619         if (!mapper) {
2620                 mapper = allocate_ftrace_func_mapper();
2621                 if (!mapper)
2622                         return -ENODEV;
2623                 *data = mapper;
2624         }
2625
2626         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2627         if (ret < 0)
2628                 return ret;
2629
2630         edata->ref++;
2631
2632         return 0;
2633 }
2634
2635 static int free_probe_data(void *data)
2636 {
2637         struct event_probe_data *edata = data;
2638
2639         edata->ref--;
2640         if (!edata->ref) {
2641                 /* Remove the SOFT_MODE flag */
2642                 __ftrace_event_enable_disable(edata->file, 0, 1);
2643                 module_put(edata->file->event_call->mod);
2644                 kfree(edata);
2645         }
2646         return 0;
2647 }
2648
2649 static void
2650 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2651                   unsigned long ip, void *data)
2652 {
2653         struct ftrace_func_mapper *mapper = data;
2654         struct event_probe_data *edata;
2655
2656         if (!ip) {
2657                 if (!mapper)
2658                         return;
2659                 free_ftrace_func_mapper(mapper, free_probe_data);
2660                 return;
2661         }
2662
2663         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2664
2665         if (WARN_ON_ONCE(!edata))
2666                 return;
2667
2668         if (WARN_ON_ONCE(edata->ref <= 0))
2669                 return;
2670
2671         free_probe_data(edata);
2672 }
2673
2674 static struct ftrace_probe_ops event_enable_probe_ops = {
2675         .func                   = event_enable_probe,
2676         .print                  = event_enable_print,
2677         .init                   = event_enable_init,
2678         .free                   = event_enable_free,
2679 };
2680
2681 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2682         .func                   = event_enable_count_probe,
2683         .print                  = event_enable_print,
2684         .init                   = event_enable_init,
2685         .free                   = event_enable_free,
2686 };
2687
2688 static struct ftrace_probe_ops event_disable_probe_ops = {
2689         .func                   = event_enable_probe,
2690         .print                  = event_enable_print,
2691         .init                   = event_enable_init,
2692         .free                   = event_enable_free,
2693 };
2694
2695 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2696         .func                   = event_enable_count_probe,
2697         .print                  = event_enable_print,
2698         .init                   = event_enable_init,
2699         .free                   = event_enable_free,
2700 };
2701
2702 static int
2703 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2704                   char *glob, char *cmd, char *param, int enabled)
2705 {
2706         struct trace_event_file *file;
2707         struct ftrace_probe_ops *ops;
2708         struct event_probe_data *data;
2709         const char *system;
2710         const char *event;
2711         char *number;
2712         bool enable;
2713         int ret;
2714
2715         if (!tr)
2716                 return -ENODEV;
2717
2718         /* hash funcs only work with set_ftrace_filter */
2719         if (!enabled || !param)
2720                 return -EINVAL;
2721
2722         system = strsep(&param, ":");
2723         if (!param)
2724                 return -EINVAL;
2725
2726         event = strsep(&param, ":");
2727
2728         mutex_lock(&event_mutex);
2729
2730         ret = -EINVAL;
2731         file = find_event_file(tr, system, event);
2732         if (!file)
2733                 goto out;
2734
2735         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2736
2737         if (enable)
2738                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2739         else
2740                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2741
2742         if (glob[0] == '!') {
2743                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2744                 goto out;
2745         }
2746
2747         ret = -ENOMEM;
2748
2749         data = kzalloc(sizeof(*data), GFP_KERNEL);
2750         if (!data)
2751                 goto out;
2752
2753         data->enable = enable;
2754         data->count = -1;
2755         data->file = file;
2756
2757         if (!param)
2758                 goto out_reg;
2759
2760         number = strsep(&param, ":");
2761
2762         ret = -EINVAL;
2763         if (!strlen(number))
2764                 goto out_free;
2765
2766         /*
2767          * We use the callback data field (which is a pointer)
2768          * as our counter.
2769          */
2770         ret = kstrtoul(number, 0, &data->count);
2771         if (ret)
2772                 goto out_free;
2773
2774  out_reg:
2775         /* Don't let event modules unload while probe registered */
2776         ret = try_module_get(file->event_call->mod);
2777         if (!ret) {
2778                 ret = -EBUSY;
2779                 goto out_free;
2780         }
2781
2782         ret = __ftrace_event_enable_disable(file, 1, 1);
2783         if (ret < 0)
2784                 goto out_put;
2785
2786         ret = register_ftrace_function_probe(glob, tr, ops, data);
2787         /*
2788          * The above returns on success the # of functions enabled,
2789          * but if it didn't find any functions it returns zero.
2790          * Consider no functions a failure too.
2791          */
2792         if (!ret) {
2793                 ret = -ENOENT;
2794                 goto out_disable;
2795         } else if (ret < 0)
2796                 goto out_disable;
2797         /* Just return zero, not the number of enabled functions */
2798         ret = 0;
2799  out:
2800         mutex_unlock(&event_mutex);
2801         return ret;
2802
2803  out_disable:
2804         __ftrace_event_enable_disable(file, 0, 1);
2805  out_put:
2806         module_put(file->event_call->mod);
2807  out_free:
2808         kfree(data);
2809         goto out;
2810 }
2811
2812 static struct ftrace_func_command event_enable_cmd = {
2813         .name                   = ENABLE_EVENT_STR,
2814         .func                   = event_enable_func,
2815 };
2816
2817 static struct ftrace_func_command event_disable_cmd = {
2818         .name                   = DISABLE_EVENT_STR,
2819         .func                   = event_enable_func,
2820 };
2821
2822 static __init int register_event_cmds(void)
2823 {
2824         int ret;
2825
2826         ret = register_ftrace_command(&event_enable_cmd);
2827         if (WARN_ON(ret < 0))
2828                 return ret;
2829         ret = register_ftrace_command(&event_disable_cmd);
2830         if (WARN_ON(ret < 0))
2831                 unregister_ftrace_command(&event_enable_cmd);
2832         return ret;
2833 }
2834 #else
2835 static inline int register_event_cmds(void) { return 0; }
2836 #endif /* CONFIG_DYNAMIC_FTRACE */
2837
2838 /*
2839  * The top level array has already had its trace_event_file
2840  * descriptors created in order to allow for early events to
2841  * be recorded. This function is called after the tracefs has been
2842  * initialized, and we now have to create the files associated
2843  * to the events.
2844  */
2845 static __init void
2846 __trace_early_add_event_dirs(struct trace_array *tr)
2847 {
2848         struct trace_event_file *file;
2849         int ret;
2850
2851
2852         list_for_each_entry(file, &tr->events, list) {
2853                 ret = event_create_dir(tr->event_dir, file);
2854                 if (ret < 0)
2855                         pr_warn("Could not create directory for event %s\n",
2856                                 trace_event_name(file->event_call));
2857         }
2858 }
2859
2860 /*
2861  * For early boot up, the top trace array requires to have
2862  * a list of events that can be enabled. This must be done before
2863  * the filesystem is set up in order to allow events to be traced
2864  * early.
2865  */
2866 static __init void
2867 __trace_early_add_events(struct trace_array *tr)
2868 {
2869         struct trace_event_call *call;
2870         int ret;
2871
2872         list_for_each_entry(call, &ftrace_events, list) {
2873                 /* Early boot up should not have any modules loaded */
2874                 if (WARN_ON_ONCE(call->mod))
2875                         continue;
2876
2877                 ret = __trace_early_add_new_event(call, tr);
2878                 if (ret < 0)
2879                         pr_warn("Could not create early event %s\n",
2880                                 trace_event_name(call));
2881         }
2882 }
2883
2884 /* Remove the event directory structure for a trace directory. */
2885 static void
2886 __trace_remove_event_dirs(struct trace_array *tr)
2887 {
2888         struct trace_event_file *file, *next;
2889
2890         list_for_each_entry_safe(file, next, &tr->events, list)
2891                 remove_event_file_dir(file);
2892 }
2893
2894 static void __add_event_to_tracers(struct trace_event_call *call)
2895 {
2896         struct trace_array *tr;
2897
2898         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2899                 __trace_add_new_event(call, tr);
2900 }
2901
2902 extern struct trace_event_call *__start_ftrace_events[];
2903 extern struct trace_event_call *__stop_ftrace_events[];
2904
2905 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2906
2907 static __init int setup_trace_event(char *str)
2908 {
2909         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2910         ring_buffer_expanded = true;
2911         tracing_selftest_disabled = true;
2912
2913         return 1;
2914 }
2915 __setup("trace_event=", setup_trace_event);
2916
2917 /* Expects to have event_mutex held when called */
2918 static int
2919 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2920 {
2921         struct dentry *d_events;
2922         struct dentry *entry;
2923
2924         entry = tracefs_create_file("set_event", 0644, parent,
2925                                     tr, &ftrace_set_event_fops);
2926         if (!entry) {
2927                 pr_warn("Could not create tracefs 'set_event' entry\n");
2928                 return -ENOMEM;
2929         }
2930
2931         d_events = tracefs_create_dir("events", parent);
2932         if (!d_events) {
2933                 pr_warn("Could not create tracefs 'events' directory\n");
2934                 return -ENOMEM;
2935         }
2936
2937         entry = trace_create_file("enable", 0644, d_events,
2938                                   tr, &ftrace_tr_enable_fops);
2939         if (!entry) {
2940                 pr_warn("Could not create tracefs 'enable' entry\n");
2941                 return -ENOMEM;
2942         }
2943
2944         /* There are not as crucial, just warn if they are not created */
2945
2946         entry = tracefs_create_file("set_event_pid", 0644, parent,
2947                                     tr, &ftrace_set_event_pid_fops);
2948         if (!entry)
2949                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2950
2951         /* ring buffer internal formats */
2952         entry = trace_create_file("header_page", 0444, d_events,
2953                                   ring_buffer_print_page_header,
2954                                   &ftrace_show_header_fops);
2955         if (!entry)
2956                 pr_warn("Could not create tracefs 'header_page' entry\n");
2957
2958         entry = trace_create_file("header_event", 0444, d_events,
2959                                   ring_buffer_print_entry_header,
2960                                   &ftrace_show_header_fops);
2961         if (!entry)
2962                 pr_warn("Could not create tracefs 'header_event' entry\n");
2963
2964         tr->event_dir = d_events;
2965
2966         return 0;
2967 }
2968
2969 /**
2970  * event_trace_add_tracer - add a instance of a trace_array to events
2971  * @parent: The parent dentry to place the files/directories for events in
2972  * @tr: The trace array associated with these events
2973  *
2974  * When a new instance is created, it needs to set up its events
2975  * directory, as well as other files associated with events. It also
2976  * creates the event hierachry in the @parent/events directory.
2977  *
2978  * Returns 0 on success.
2979  *
2980  * Must be called with event_mutex held.
2981  */
2982 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2983 {
2984         int ret;
2985
2986         lockdep_assert_held(&event_mutex);
2987
2988         ret = create_event_toplevel_files(parent, tr);
2989         if (ret)
2990                 goto out;
2991
2992         down_write(&trace_event_sem);
2993         __trace_add_event_dirs(tr);
2994         up_write(&trace_event_sem);
2995
2996  out:
2997         return ret;
2998 }
2999
3000 /*
3001  * The top trace array already had its file descriptors created.
3002  * Now the files themselves need to be created.
3003  */
3004 static __init int
3005 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3006 {
3007         int ret;
3008
3009         mutex_lock(&event_mutex);
3010
3011         ret = create_event_toplevel_files(parent, tr);
3012         if (ret)
3013                 goto out_unlock;
3014
3015         down_write(&trace_event_sem);
3016         __trace_early_add_event_dirs(tr);
3017         up_write(&trace_event_sem);
3018
3019  out_unlock:
3020         mutex_unlock(&event_mutex);
3021
3022         return ret;
3023 }
3024
3025 /* Must be called with event_mutex held */
3026 int event_trace_del_tracer(struct trace_array *tr)
3027 {
3028         lockdep_assert_held(&event_mutex);
3029
3030         /* Disable any event triggers and associated soft-disabled events */
3031         clear_event_triggers(tr);
3032
3033         /* Clear the pid list */
3034         __ftrace_clear_event_pids(tr);
3035
3036         /* Disable any running events */
3037         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3038
3039         /* Access to events are within rcu_read_lock_sched() */
3040         synchronize_sched();
3041
3042         down_write(&trace_event_sem);
3043         __trace_remove_event_dirs(tr);
3044         tracefs_remove_recursive(tr->event_dir);
3045         up_write(&trace_event_sem);
3046
3047         tr->event_dir = NULL;
3048
3049         return 0;
3050 }
3051
3052 static __init int event_trace_memsetup(void)
3053 {
3054         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3055         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3056         return 0;
3057 }
3058
3059 static __init void
3060 early_enable_events(struct trace_array *tr, bool disable_first)
3061 {
3062         char *buf = bootup_event_buf;
3063         char *token;
3064         int ret;
3065
3066         while (true) {
3067                 token = strsep(&buf, ",");
3068
3069                 if (!token)
3070                         break;
3071
3072                 if (*token) {
3073                         /* Restarting syscalls requires that we stop them first */
3074                         if (disable_first)
3075                                 ftrace_set_clr_event(tr, token, 0);
3076
3077                         ret = ftrace_set_clr_event(tr, token, 1);
3078                         if (ret)
3079                                 pr_warn("Failed to enable trace event: %s\n", token);
3080                 }
3081
3082                 /* Put back the comma to allow this to be called again */
3083                 if (buf)
3084                         *(buf - 1) = ',';
3085         }
3086 }
3087
3088 static __init int event_trace_enable(void)
3089 {
3090         struct trace_array *tr = top_trace_array();
3091         struct trace_event_call **iter, *call;
3092         int ret;
3093
3094         if (!tr)
3095                 return -ENODEV;
3096
3097         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3098
3099                 call = *iter;
3100                 ret = event_init(call);
3101                 if (!ret)
3102                         list_add(&call->list, &ftrace_events);
3103         }
3104
3105         /*
3106          * We need the top trace array to have a working set of trace
3107          * points at early init, before the debug files and directories
3108          * are created. Create the file entries now, and attach them
3109          * to the actual file dentries later.
3110          */
3111         __trace_early_add_events(tr);
3112
3113         early_enable_events(tr, false);
3114
3115         trace_printk_start_comm();
3116
3117         register_event_cmds();
3118
3119         register_trigger_cmds();
3120
3121         return 0;
3122 }
3123
3124 /*
3125  * event_trace_enable() is called from trace_event_init() first to
3126  * initialize events and perhaps start any events that are on the
3127  * command line. Unfortunately, there are some events that will not
3128  * start this early, like the system call tracepoints that need
3129  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3130  * is called before pid 1 starts, and this flag is never set, making
3131  * the syscall tracepoint never get reached, but the event is enabled
3132  * regardless (and not doing anything).
3133  */
3134 static __init int event_trace_enable_again(void)
3135 {
3136         struct trace_array *tr;
3137
3138         tr = top_trace_array();
3139         if (!tr)
3140                 return -ENODEV;
3141
3142         early_enable_events(tr, true);
3143
3144         return 0;
3145 }
3146
3147 early_initcall(event_trace_enable_again);
3148
3149 __init int event_trace_init(void)
3150 {
3151         struct trace_array *tr;
3152         struct dentry *d_tracer;
3153         struct dentry *entry;
3154         int ret;
3155
3156         tr = top_trace_array();
3157         if (!tr)
3158                 return -ENODEV;
3159
3160         d_tracer = tracing_init_dentry();
3161         if (IS_ERR(d_tracer))
3162                 return 0;
3163
3164         entry = tracefs_create_file("available_events", 0444, d_tracer,
3165                                     tr, &ftrace_avail_fops);
3166         if (!entry)
3167                 pr_warn("Could not create tracefs 'available_events' entry\n");
3168
3169         if (trace_define_generic_fields())
3170                 pr_warn("tracing: Failed to allocated generic fields");
3171
3172         if (trace_define_common_fields())
3173                 pr_warn("tracing: Failed to allocate common fields");
3174
3175         ret = early_event_add_tracer(d_tracer, tr);
3176         if (ret)
3177                 return ret;
3178
3179 #ifdef CONFIG_MODULES
3180         ret = register_module_notifier(&trace_module_nb);
3181         if (ret)
3182                 pr_warn("Failed to register trace events module notifier\n");
3183 #endif
3184         return 0;
3185 }
3186
3187 void __init trace_event_init(void)
3188 {
3189         event_trace_memsetup();
3190         init_ftrace_syscalls();
3191         event_trace_enable();
3192 }
3193
3194 #ifdef CONFIG_FTRACE_STARTUP_TEST
3195
3196 static DEFINE_SPINLOCK(test_spinlock);
3197 static DEFINE_SPINLOCK(test_spinlock_irq);
3198 static DEFINE_MUTEX(test_mutex);
3199
3200 static __init void test_work(struct work_struct *dummy)
3201 {
3202         spin_lock(&test_spinlock);
3203         spin_lock_irq(&test_spinlock_irq);
3204         udelay(1);
3205         spin_unlock_irq(&test_spinlock_irq);
3206         spin_unlock(&test_spinlock);
3207
3208         mutex_lock(&test_mutex);
3209         msleep(1);
3210         mutex_unlock(&test_mutex);
3211 }
3212
3213 static __init int event_test_thread(void *unused)
3214 {
3215         void *test_malloc;
3216
3217         test_malloc = kmalloc(1234, GFP_KERNEL);
3218         if (!test_malloc)
3219                 pr_info("failed to kmalloc\n");
3220
3221         schedule_on_each_cpu(test_work);
3222
3223         kfree(test_malloc);
3224
3225         set_current_state(TASK_INTERRUPTIBLE);
3226         while (!kthread_should_stop()) {
3227                 schedule();
3228                 set_current_state(TASK_INTERRUPTIBLE);
3229         }
3230         __set_current_state(TASK_RUNNING);
3231
3232         return 0;
3233 }
3234
3235 /*
3236  * Do various things that may trigger events.
3237  */
3238 static __init void event_test_stuff(void)
3239 {
3240         struct task_struct *test_thread;
3241
3242         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3243         msleep(1);
3244         kthread_stop(test_thread);
3245 }
3246
3247 /*
3248  * For every trace event defined, we will test each trace point separately,
3249  * and then by groups, and finally all trace points.
3250  */
3251 static __init void event_trace_self_tests(void)
3252 {
3253         struct trace_subsystem_dir *dir;
3254         struct trace_event_file *file;
3255         struct trace_event_call *call;
3256         struct event_subsystem *system;
3257         struct trace_array *tr;
3258         int ret;
3259
3260         tr = top_trace_array();
3261         if (!tr)
3262                 return;
3263
3264         pr_info("Running tests on trace events:\n");
3265
3266         list_for_each_entry(file, &tr->events, list) {
3267
3268                 call = file->event_call;
3269
3270                 /* Only test those that have a probe */
3271                 if (!call->class || !call->class->probe)
3272                         continue;
3273
3274 /*
3275  * Testing syscall events here is pretty useless, but
3276  * we still do it if configured. But this is time consuming.
3277  * What we really need is a user thread to perform the
3278  * syscalls as we test.
3279  */
3280 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3281                 if (call->class->system &&
3282                     strcmp(call->class->system, "syscalls") == 0)
3283                         continue;
3284 #endif
3285
3286                 pr_info("Testing event %s: ", trace_event_name(call));
3287
3288                 /*
3289                  * If an event is already enabled, someone is using
3290                  * it and the self test should not be on.
3291                  */
3292                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3293                         pr_warn("Enabled event during self test!\n");
3294                         WARN_ON_ONCE(1);
3295                         continue;
3296                 }
3297
3298                 ftrace_event_enable_disable(file, 1);
3299                 event_test_stuff();
3300                 ftrace_event_enable_disable(file, 0);
3301
3302                 pr_cont("OK\n");
3303         }
3304
3305         /* Now test at the sub system level */
3306
3307         pr_info("Running tests on trace event systems:\n");
3308
3309         list_for_each_entry(dir, &tr->systems, list) {
3310
3311                 system = dir->subsystem;
3312
3313                 /* the ftrace system is special, skip it */
3314                 if (strcmp(system->name, "ftrace") == 0)
3315                         continue;
3316
3317                 pr_info("Testing event system %s: ", system->name);
3318
3319                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3320                 if (WARN_ON_ONCE(ret)) {
3321                         pr_warn("error enabling system %s\n",
3322                                 system->name);
3323                         continue;
3324                 }
3325
3326                 event_test_stuff();
3327
3328                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3329                 if (WARN_ON_ONCE(ret)) {
3330                         pr_warn("error disabling system %s\n",
3331                                 system->name);
3332                         continue;
3333                 }
3334
3335                 pr_cont("OK\n");
3336         }
3337
3338         /* Test with all events enabled */
3339
3340         pr_info("Running tests on all trace events:\n");
3341         pr_info("Testing all events: ");
3342
3343         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3344         if (WARN_ON_ONCE(ret)) {
3345                 pr_warn("error enabling all events\n");
3346                 return;
3347         }
3348
3349         event_test_stuff();
3350
3351         /* reset sysname */
3352         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3353         if (WARN_ON_ONCE(ret)) {
3354                 pr_warn("error disabling all events\n");
3355                 return;
3356         }
3357
3358         pr_cont("OK\n");
3359 }
3360
3361 #ifdef CONFIG_FUNCTION_TRACER
3362
3363 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3364
3365 static struct trace_event_file event_trace_file __initdata;
3366
3367 static void __init
3368 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3369                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3370 {
3371         struct ring_buffer_event *event;
3372         struct ring_buffer *buffer;
3373         struct ftrace_entry *entry;
3374         unsigned long flags;
3375         long disabled;
3376         int cpu;
3377         int pc;
3378
3379         pc = preempt_count();
3380         preempt_disable_notrace();
3381         cpu = raw_smp_processor_id();
3382         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3383
3384         if (disabled != 1)
3385                 goto out;
3386
3387         local_save_flags(flags);
3388
3389         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3390                                                 TRACE_FN, sizeof(*entry),
3391                                                 flags, pc);
3392         if (!event)
3393                 goto out;
3394         entry   = ring_buffer_event_data(event);
3395         entry->ip                       = ip;
3396         entry->parent_ip                = parent_ip;
3397
3398         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3399                                     entry, flags, pc);
3400  out:
3401         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3402         preempt_enable_notrace();
3403 }
3404
3405 static struct ftrace_ops trace_ops __initdata  =
3406 {
3407         .func = function_test_events_call,
3408         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3409 };
3410
3411 static __init void event_trace_self_test_with_function(void)
3412 {
3413         int ret;
3414
3415         event_trace_file.tr = top_trace_array();
3416         if (WARN_ON(!event_trace_file.tr))
3417                 return;
3418
3419         ret = register_ftrace_function(&trace_ops);
3420         if (WARN_ON(ret < 0)) {
3421                 pr_info("Failed to enable function tracer for event tests\n");
3422                 return;
3423         }
3424         pr_info("Running tests again, along with the function tracer\n");
3425         event_trace_self_tests();
3426         unregister_ftrace_function(&trace_ops);
3427 }
3428 #else
3429 static __init void event_trace_self_test_with_function(void)
3430 {
3431 }
3432 #endif
3433
3434 static __init int event_trace_self_tests_init(void)
3435 {
3436         if (!tracing_selftest_disabled) {
3437                 event_trace_self_tests();
3438                 event_trace_self_test_with_function();
3439         }
3440
3441         return 0;
3442 }
3443
3444 late_initcall(event_trace_self_tests_init);
3445
3446 #endif