kernel/trace/trace_events.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * event tracer
   4  *
   5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6  *
   7  *  - Added format output of fields of the trace point.
   8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9  *
  10  */
  11
  12 #define pr_fmt(fmt) fmt
  13
  14 #include <linux/workqueue.h>
  15 #include <linux/security.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/kthread.h>
  18 #include <linux/tracefs.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/module.h>
  21 #include <linux/ctype.h>
  22 #include <linux/sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/delay.h>
  25
  26 #include <trace/events/sched.h>
  27 #include <trace/syscall.h>
  28
  29 #include <asm/setup.h>
  30
  31 #include "trace_output.h"
  32
  33 #undef TRACE_SYSTEM
  34 #define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36 DEFINE_MUTEX(event_mutex);
  37
  38 LIST_HEAD(ftrace_events);
  39 static LIST_HEAD(ftrace_generic_fields);
  40 static LIST_HEAD(ftrace_common_fields);
  41
  42 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  43
  44 static struct kmem_cache *field_cachep;
  45 static struct kmem_cache *file_cachep;
  46
  47 static inline int system_refcount(struct event_subsystem *system)
  48 {
  49         return system->ref_count;
  50 }
  51
  52 static int system_refcount_inc(struct event_subsystem *system)
  53 {
  54         return system->ref_count++;
  55 }
  56
  57 static int system_refcount_dec(struct event_subsystem *system)
  58 {
  59         return --system->ref_count;
  60 }
  61
  62 /* Double loops, do not use break, only goto's work */
  63 #define do_for_each_event_file(tr, file)                        \
  64         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  65                 list_for_each_entry(file, &tr->events, list)
  66
  67 #define do_for_each_event_file_safe(tr, file)                   \
  68         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  69                 struct trace_event_file *___n;                          \
  70                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  71
  72 #define while_for_each_event_file()             \
  73         }
  74
  75 static struct ftrace_event_field *
  76 __find_event_field(struct list_head *head, char *name)
  77 {
  78         struct ftrace_event_field *field;
  79
  80         list_for_each_entry(field, head, link) {
  81                 if (!strcmp(field->name, name))
  82                         return field;
  83         }
  84
  85         return NULL;
  86 }
  87
  88 struct ftrace_event_field *
  89 trace_find_event_field(struct trace_event_call *call, char *name)
  90 {
  91         struct ftrace_event_field *field;
  92         struct list_head *head;
  93
  94         head = trace_get_fields(call);
  95         field = __find_event_field(head, name);
  96         if (field)
  97                 return field;
  98
  99         field = __find_event_field(&ftrace_generic_fields, name);
 100         if (field)
 101                 return field;
 102
 103         return __find_event_field(&ftrace_common_fields, name);
 104 }
 105
 106 static int __trace_define_field(struct list_head *head, const char *type,
 107                                 const char *name, int offset, int size,
 108                                 int is_signed, int filter_type)
 109 {
 110         struct ftrace_event_field *field;
 111
 112         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 113         if (!field)
 114                 return -ENOMEM;
 115
 116         field->name = name;
 117         field->type = type;
 118
 119         if (filter_type == FILTER_OTHER)
 120                 field->filter_type = filter_assign_type(type);
 121         else
 122                 field->filter_type = filter_type;
 123
 124         field->offset = offset;
 125         field->size = size;
 126         field->is_signed = is_signed;
 127
 128         list_add(&field->link, head);
 129
 130         return 0;
 131 }
 132
 133 int trace_define_field(struct trace_event_call *call, const char *type,
 134                        const char *name, int offset, int size, int is_signed,
 135                        int filter_type)
 136 {
 137         struct list_head *head;
 138
 139         if (WARN_ON(!call->class))
 140                 return 0;
 141
 142         head = trace_get_fields(call);
 143         return __trace_define_field(head, type, name, offset, size,
 144                                     is_signed, filter_type);
 145 }
 146 EXPORT_SYMBOL_GPL(trace_define_field);
 147
 148 #define __generic_field(type, item, filter_type)                        \
 149         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 150                                    #item, 0, 0, is_signed_type(type),   \
 151                                    filter_type);                        \
 152         if (ret)                                                        \
 153                 return ret;
 154
 155 #define __common_field(type, item)                                      \
 156         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 157                                    "common_" #item,                     \
 158                                    offsetof(typeof(ent), item),         \
 159                                    sizeof(ent.item),                    \
 160                                    is_signed_type(type), FILTER_OTHER); \
 161         if (ret)                                                        \
 162                 return ret;
 163
 164 static int trace_define_generic_fields(void)
 165 {
 166         int ret;
 167
 168         __generic_field(int, CPU, FILTER_CPU);
 169         __generic_field(int, cpu, FILTER_CPU);
 170         __generic_field(char *, COMM, FILTER_COMM);
 171         __generic_field(char *, comm, FILTER_COMM);
 172
 173         return ret;
 174 }
 175
 176 static int trace_define_common_fields(void)
 177 {
 178         int ret;
 179         struct trace_entry ent;
 180
 181         __common_field(unsigned short, type);
 182         __common_field(unsigned char, flags);
 183         __common_field(unsigned char, preempt_count);
 184         __common_field(int, pid);
 185
 186         return ret;
 187 }
 188
 189 static void trace_destroy_fields(struct trace_event_call *call)
 190 {
 191         struct ftrace_event_field *field, *next;
 192         struct list_head *head;
 193
 194         head = trace_get_fields(call);
 195         list_for_each_entry_safe(field, next, head, link) {
 196                 list_del(&field->link);
 197                 kmem_cache_free(field_cachep, field);
 198         }
 199 }
 200
 201 /*
 202  * run-time version of trace_event_get_offsets_<call>() that returns the last
 203  * accessible offset of trace fields excluding __dynamic_array bytes
 204  */
 205 int trace_event_get_offsets(struct trace_event_call *call)
 206 {
 207         struct ftrace_event_field *tail;
 208         struct list_head *head;
 209
 210         head = trace_get_fields(call);
 211         /*
 212          * head->next points to the last field with the largest offset,
 213          * since it was added last by trace_define_field()
 214          */
 215         tail = list_first_entry(head, struct ftrace_event_field, link);
 216         return tail->offset + tail->size;
 217 }
 218
 219 int trace_event_raw_init(struct trace_event_call *call)
 220 {
 221         int id;
 222
 223         id = register_trace_event(&call->event);
 224         if (!id)
 225                 return -ENODEV;
 226
 227         return 0;
 228 }
 229 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 230
 231 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 232 {
 233         struct trace_array *tr = trace_file->tr;
 234         struct trace_array_cpu *data;
 235         struct trace_pid_list *pid_list;
 236
 237         pid_list = rcu_dereference_raw(tr->filtered_pids);
 238         if (!pid_list)
 239                 return false;
 240
 241         data = this_cpu_ptr(tr->array_buffer.data);
 242
 243         return data->ignore_pid;
 244 }
 245 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 246
 247 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 248                                  struct trace_event_file *trace_file,
 249                                  unsigned long len)
 250 {
 251         struct trace_event_call *event_call = trace_file->event_call;
 252
 253         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 254             trace_event_ignore_this_pid(trace_file))
 255                 return NULL;
 256
 257         local_save_flags(fbuffer->flags);
 258         fbuffer->pc = preempt_count();
 259         /*
 260          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 261          * preemption (adding one to the preempt_count). Since we are
 262          * interested in the preempt_count at the time the tracepoint was
 263          * hit, we need to subtract one to offset the increment.
 264          */
 265         if (IS_ENABLED(CONFIG_PREEMPTION))
 266                 fbuffer->pc--;
 267         fbuffer->trace_file = trace_file;
 268
 269         fbuffer->event =
 270                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 271                                                 event_call->event.type, len,
 272                                                 fbuffer->flags, fbuffer->pc);
 273         if (!fbuffer->event)
 274                 return NULL;
 275
 276         fbuffer->regs = NULL;
 277         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 278         return fbuffer->entry;
 279 }
 280 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 281
 282 int trace_event_reg(struct trace_event_call *call,
 283                     enum trace_reg type, void *data)
 284 {
 285         struct trace_event_file *file = data;
 286
 287         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 288         switch (type) {
 289         case TRACE_REG_REGISTER:
 290                 return tracepoint_probe_register(call->tp,
 291                                                  call->class->probe,
 292                                                  file);
 293         case TRACE_REG_UNREGISTER:
 294                 tracepoint_probe_unregister(call->tp,
 295                                             call->class->probe,
 296                                             file);
 297                 return 0;
 298
 299 #ifdef CONFIG_PERF_EVENTS
 300         case TRACE_REG_PERF_REGISTER:
 301                 return tracepoint_probe_register(call->tp,
 302                                                  call->class->perf_probe,
 303                                                  call);
 304         case TRACE_REG_PERF_UNREGISTER:
 305                 tracepoint_probe_unregister(call->tp,
 306                                             call->class->perf_probe,
 307                                             call);
 308                 return 0;
 309         case TRACE_REG_PERF_OPEN:
 310         case TRACE_REG_PERF_CLOSE:
 311         case TRACE_REG_PERF_ADD:
 312         case TRACE_REG_PERF_DEL:
 313                 return 0;
 314 #endif
 315         }
 316         return 0;
 317 }
 318 EXPORT_SYMBOL_GPL(trace_event_reg);
 319
 320 void trace_event_enable_cmd_record(bool enable)
 321 {
 322         struct trace_event_file *file;
 323         struct trace_array *tr;
 324
 325         lockdep_assert_held(&event_mutex);
 326
 327         do_for_each_event_file(tr, file) {
 328
 329                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 330                         continue;
 331
 332                 if (enable) {
 333                         tracing_start_cmdline_record();
 334                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 335                 } else {
 336                         tracing_stop_cmdline_record();
 337                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338                 }
 339         } while_for_each_event_file();
 340 }
 341
 342 void trace_event_enable_tgid_record(bool enable)
 343 {
 344         struct trace_event_file *file;
 345         struct trace_array *tr;
 346
 347         lockdep_assert_held(&event_mutex);
 348
 349         do_for_each_event_file(tr, file) {
 350                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 351                         continue;
 352
 353                 if (enable) {
 354                         tracing_start_tgid_record();
 355                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 356                 } else {
 357                         tracing_stop_tgid_record();
 358                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 359                                   &file->flags);
 360                 }
 361         } while_for_each_event_file();
 362 }
 363
 364 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 365                                          int enable, int soft_disable)
 366 {
 367         struct trace_event_call *call = file->event_call;
 368         struct trace_array *tr = file->tr;
 369         unsigned long file_flags = file->flags;
 370         int ret = 0;
 371         int disable;
 372
 373         switch (enable) {
 374         case 0:
 375                 /*
 376                  * When soft_disable is set and enable is cleared, the sm_ref
 377                  * reference counter is decremented. If it reaches 0, we want
 378                  * to clear the SOFT_DISABLED flag but leave the event in the
 379                  * state that it was. That is, if the event was enabled and
 380                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 381                  * is set we do not want the event to be enabled before we
 382                  * clear the bit.
 383                  *
 384                  * When soft_disable is not set but the SOFT_MODE flag is,
 385                  * we do nothing. Do not disable the tracepoint, otherwise
 386                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 387                  */
 388                 if (soft_disable) {
 389                         if (atomic_dec_return(&file->sm_ref) > 0)
 390                                 break;
 391                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 392                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 393                 } else
 394                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 395
 396                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 397                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 398                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 399                                 tracing_stop_cmdline_record();
 400                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 401                         }
 402
 403                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 404                                 tracing_stop_tgid_record();
 405                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 406                         }
 407
 408                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 409                 }
 410                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 411                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 412                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 413                 else
 414                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 415                 break;
 416         case 1:
 417                 /*
 418                  * When soft_disable is set and enable is set, we want to
 419                  * register the tracepoint for the event, but leave the event
 420                  * as is. That means, if the event was already enabled, we do
 421                  * nothing (but set SOFT_MODE). If the event is disabled, we
 422                  * set SOFT_DISABLED before enabling the event tracepoint, so
 423                  * it still seems to be disabled.
 424                  */
 425                 if (!soft_disable)
 426                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 427                 else {
 428                         if (atomic_inc_return(&file->sm_ref) > 1)
 429                                 break;
 430                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 431                 }
 432
 433                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 434                         bool cmd = false, tgid = false;
 435
 436                         /* Keep the event disabled, when going to SOFT_MODE. */
 437                         if (soft_disable)
 438                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 439
 440                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 441                                 cmd = true;
 442                                 tracing_start_cmdline_record();
 443                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 444                         }
 445
 446                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 447                                 tgid = true;
 448                                 tracing_start_tgid_record();
 449                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 450                         }
 451
 452                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 453                         if (ret) {
 454                                 if (cmd)
 455                                         tracing_stop_cmdline_record();
 456                                 if (tgid)
 457                                         tracing_stop_tgid_record();
 458                                 pr_info("event trace: Could not enable event "
 459                                         "%s\n", trace_event_name(call));
 460                                 break;
 461                         }
 462                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 463
 464                         /* WAS_ENABLED gets set but never cleared. */
 465                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 466                 }
 467                 break;
 468         }
 469
 470         /* Enable or disable use of trace_buffered_event */
 471         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 472             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 473                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 474                         trace_buffered_event_enable();
 475                 else
 476                         trace_buffered_event_disable();
 477         }
 478
 479         return ret;
 480 }
 481
 482 int trace_event_enable_disable(struct trace_event_file *file,
 483                                int enable, int soft_disable)
 484 {
 485         return __ftrace_event_enable_disable(file, enable, soft_disable);
 486 }
 487
 488 static int ftrace_event_enable_disable(struct trace_event_file *file,
 489                                        int enable)
 490 {
 491         return __ftrace_event_enable_disable(file, enable, 0);
 492 }
 493
 494 static void ftrace_clear_events(struct trace_array *tr)
 495 {
 496         struct trace_event_file *file;
 497
 498         mutex_lock(&event_mutex);
 499         list_for_each_entry(file, &tr->events, list) {
 500                 ftrace_event_enable_disable(file, 0);
 501         }
 502         mutex_unlock(&event_mutex);
 503 }
 504
 505 static void
 506 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 507 {
 508         struct trace_pid_list *pid_list;
 509         struct trace_array *tr = data;
 510
 511         pid_list = rcu_dereference_raw(tr->filtered_pids);
 512         trace_filter_add_remove_task(pid_list, NULL, task);
 513 }
 514
 515 static void
 516 event_filter_pid_sched_process_fork(void *data,
 517                                     struct task_struct *self,
 518                                     struct task_struct *task)
 519 {
 520         struct trace_pid_list *pid_list;
 521         struct trace_array *tr = data;
 522
 523         pid_list = rcu_dereference_sched(tr->filtered_pids);
 524         trace_filter_add_remove_task(pid_list, self, task);
 525 }
 526
 527 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 528 {
 529         if (enable) {
 530                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 531                                                        tr, INT_MIN);
 532                 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 533                                                        tr, INT_MAX);
 534         } else {
 535                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 536                                                     tr);
 537                 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 538                                                     tr);
 539         }
 540 }
 541
 542 static void
 543 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 544                     struct task_struct *prev, struct task_struct *next)
 545 {
 546         struct trace_array *tr = data;
 547         struct trace_pid_list *pid_list;
 548
 549         pid_list = rcu_dereference_sched(tr->filtered_pids);
 550
 551         this_cpu_write(tr->array_buffer.data->ignore_pid,
 552                        trace_ignore_this_task(pid_list, prev) &&
 553                        trace_ignore_this_task(pid_list, next));
 554 }
 555
 556 static void
 557 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 558                     struct task_struct *prev, struct task_struct *next)
 559 {
 560         struct trace_array *tr = data;
 561         struct trace_pid_list *pid_list;
 562
 563         pid_list = rcu_dereference_sched(tr->filtered_pids);
 564
 565         this_cpu_write(tr->array_buffer.data->ignore_pid,
 566                        trace_ignore_this_task(pid_list, next));
 567 }
 568
 569 static void
 570 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 571 {
 572         struct trace_array *tr = data;
 573         struct trace_pid_list *pid_list;
 574
 575         /* Nothing to do if we are already tracing */
 576         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 577                 return;
 578
 579         pid_list = rcu_dereference_sched(tr->filtered_pids);
 580
 581         this_cpu_write(tr->array_buffer.data->ignore_pid,
 582                        trace_ignore_this_task(pid_list, task));
 583 }
 584
 585 static void
 586 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 587 {
 588         struct trace_array *tr = data;
 589         struct trace_pid_list *pid_list;
 590
 591         /* Nothing to do if we are not tracing */
 592         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 593                 return;
 594
 595         pid_list = rcu_dereference_sched(tr->filtered_pids);
 596
 597         /* Set tracing if current is enabled */
 598         this_cpu_write(tr->array_buffer.data->ignore_pid,
 599                        trace_ignore_this_task(pid_list, current));
 600 }
 601
 602 static void __ftrace_clear_event_pids(struct trace_array *tr)
 603 {
 604         struct trace_pid_list *pid_list;
 605         struct trace_event_file *file;
 606         int cpu;
 607
 608         pid_list = rcu_dereference_protected(tr->filtered_pids,
 609                                              lockdep_is_held(&event_mutex));
 610         if (!pid_list)
 611                 return;
 612
 613         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 614         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 615
 616         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 617         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 618
 619         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 620         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 621
 622         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 623         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 624
 625         list_for_each_entry(file, &tr->events, list) {
 626                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 627         }
 628
 629         for_each_possible_cpu(cpu)
 630                 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 631
 632         rcu_assign_pointer(tr->filtered_pids, NULL);
 633
 634         /* Wait till all users are no longer using pid filtering */
 635         tracepoint_synchronize_unregister();
 636
 637         trace_free_pid_list(pid_list);
 638 }
 639
 640 static void ftrace_clear_event_pids(struct trace_array *tr)
 641 {
 642         mutex_lock(&event_mutex);
 643         __ftrace_clear_event_pids(tr);
 644         mutex_unlock(&event_mutex);
 645 }
 646
 647 static void __put_system(struct event_subsystem *system)
 648 {
 649         struct event_filter *filter = system->filter;
 650
 651         WARN_ON_ONCE(system_refcount(system) == 0);
 652         if (system_refcount_dec(system))
 653                 return;
 654
 655         list_del(&system->list);
 656
 657         if (filter) {
 658                 kfree(filter->filter_string);
 659                 kfree(filter);
 660         }
 661         kfree_const(system->name);
 662         kfree(system);
 663 }
 664
 665 static void __get_system(struct event_subsystem *system)
 666 {
 667         WARN_ON_ONCE(system_refcount(system) == 0);
 668         system_refcount_inc(system);
 669 }
 670
 671 static void __get_system_dir(struct trace_subsystem_dir *dir)
 672 {
 673         WARN_ON_ONCE(dir->ref_count == 0);
 674         dir->ref_count++;
 675         __get_system(dir->subsystem);
 676 }
 677
 678 static void __put_system_dir(struct trace_subsystem_dir *dir)
 679 {
 680         WARN_ON_ONCE(dir->ref_count == 0);
 681         /* If the subsystem is about to be freed, the dir must be too */
 682         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 683
 684         __put_system(dir->subsystem);
 685         if (!--dir->ref_count)
 686                 kfree(dir);
 687 }
 688
 689 static void put_system(struct trace_subsystem_dir *dir)
 690 {
 691         mutex_lock(&event_mutex);
 692         __put_system_dir(dir);
 693         mutex_unlock(&event_mutex);
 694 }
 695
 696 static void remove_subsystem(struct trace_subsystem_dir *dir)
 697 {
 698         if (!dir)
 699                 return;
 700
 701         if (!--dir->nr_events) {
 702                 tracefs_remove(dir->entry);
 703                 list_del(&dir->list);
 704                 __put_system_dir(dir);
 705         }
 706 }
 707
 708 static void remove_event_file_dir(struct trace_event_file *file)
 709 {
 710         struct dentry *dir = file->dir;
 711         struct dentry *child;
 712
 713         if (dir) {
 714                 spin_lock(&dir->d_lock);        /* probably unneeded */
 715                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 716                         if (d_really_is_positive(child))        /* probably unneeded */
 717                                 d_inode(child)->i_private = NULL;
 718                 }
 719                 spin_unlock(&dir->d_lock);
 720
 721                 tracefs_remove(dir);
 722         }
 723
 724         list_del(&file->list);
 725         remove_subsystem(file->system);
 726         free_event_filter(file->filter);
 727         kmem_cache_free(file_cachep, file);
 728 }
 729
 730 /*
 731  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 732  */
 733 static int
 734 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 735                               const char *sub, const char *event, int set)
 736 {
 737         struct trace_event_file *file;
 738         struct trace_event_call *call;
 739         const char *name;
 740         int ret = -EINVAL;
 741         int eret = 0;
 742
 743         list_for_each_entry(file, &tr->events, list) {
 744
 745                 call = file->event_call;
 746                 name = trace_event_name(call);
 747
 748                 if (!name || !call->class || !call->class->reg)
 749                         continue;
 750
 751                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 752                         continue;
 753
 754                 if (match &&
 755                     strcmp(match, name) != 0 &&
 756                     strcmp(match, call->class->system) != 0)
 757                         continue;
 758
 759                 if (sub && strcmp(sub, call->class->system) != 0)
 760                         continue;
 761
 762                 if (event && strcmp(event, name) != 0)
 763                         continue;
 764
 765                 ret = ftrace_event_enable_disable(file, set);
 766
 767                 /*
 768                  * Save the first error and return that. Some events
 769                  * may still have been enabled, but let the user
 770                  * know that something went wrong.
 771                  */
 772                 if (ret && !eret)
 773                         eret = ret;
 774
 775                 ret = eret;
 776         }
 777
 778         return ret;
 779 }
 780
 781 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 782                                   const char *sub, const char *event, int set)
 783 {
 784         int ret;
 785
 786         mutex_lock(&event_mutex);
 787         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 788         mutex_unlock(&event_mutex);
 789
 790         return ret;
 791 }
 792
 793 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 794 {
 795         char *event = NULL, *sub = NULL, *match;
 796         int ret;
 797
 798         if (!tr)
 799                 return -ENOENT;
 800         /*
 801          * The buf format can be <subsystem>:<event-name>
 802          *  *:<event-name> means any event by that name.
 803          *  :<event-name> is the same.
 804          *
 805          *  <subsystem>:* means all events in that subsystem
 806          *  <subsystem>: means the same.
 807          *
 808          *  <name> (no ':') means all events in a subsystem with
 809          *  the name <name> or any event that matches <name>
 810          */
 811
 812         match = strsep(&buf, ":");
 813         if (buf) {
 814                 sub = match;
 815                 event = buf;
 816                 match = NULL;
 817
 818                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 819                         sub = NULL;
 820                 if (!strlen(event) || strcmp(event, "*") == 0)
 821                         event = NULL;
 822         }
 823
 824         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 825
 826         /* Put back the colon to allow this to be called again */
 827         if (buf)
 828                 *(buf - 1) = ':';
 829
 830         return ret;
 831 }
 832
 833 /**
 834  * trace_set_clr_event - enable or disable an event
 835  * @system: system name to match (NULL for any system)
 836  * @event: event name to match (NULL for all events, within system)
 837  * @set: 1 to enable, 0 to disable
 838  *
 839  * This is a way for other parts of the kernel to enable or disable
 840  * event recording.
 841  *
 842  * Returns 0 on success, -EINVAL if the parameters do not match any
 843  * registered events.
 844  */
 845 int trace_set_clr_event(const char *system, const char *event, int set)
 846 {
 847         struct trace_array *tr = top_trace_array();
 848
 849         if (!tr)
 850                 return -ENODEV;
 851
 852         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 853 }
 854 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 855
 856 /**
 857  * trace_array_set_clr_event - enable or disable an event for a trace array.
 858  * @tr: concerned trace array.
 859  * @system: system name to match (NULL for any system)
 860  * @event: event name to match (NULL for all events, within system)
 861  * @enable: true to enable, false to disable
 862  *
 863  * This is a way for other parts of the kernel to enable or disable
 864  * event recording.
 865  *
 866  * Returns 0 on success, -EINVAL if the parameters do not match any
 867  * registered events.
 868  */
 869 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
 870                 const char *event, bool enable)
 871 {
 872         int set;
 873
 874         if (!tr)
 875                 return -ENOENT;
 876
 877         set = (enable == true) ? 1 : 0;
 878         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 879 }
 880 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
 881
 882 /* 128 should be much more than enough */
 883 #define EVENT_BUF_SIZE          127
 884
 885 static ssize_t
 886 ftrace_event_write(struct file *file, const char __user *ubuf,
 887                    size_t cnt, loff_t *ppos)
 888 {
 889         struct trace_parser parser;
 890         struct seq_file *m = file->private_data;
 891         struct trace_array *tr = m->private;
 892         ssize_t read, ret;
 893
 894         if (!cnt)
 895                 return 0;
 896
 897         ret = tracing_update_buffers();
 898         if (ret < 0)
 899                 return ret;
 900
 901         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 902                 return -ENOMEM;
 903
 904         read = trace_get_user(&parser, ubuf, cnt, ppos);
 905
 906         if (read >= 0 && trace_parser_loaded((&parser))) {
 907                 int set = 1;
 908
 909                 if (*parser.buffer == '!')
 910                         set = 0;
 911
 912                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 913                 if (ret)
 914                         goto out_put;
 915         }
 916
 917         ret = read;
 918
 919  out_put:
 920         trace_parser_put(&parser);
 921
 922         return ret;
 923 }
 924
 925 static void *
 926 t_next(struct seq_file *m, void *v, loff_t *pos)
 927 {
 928         struct trace_event_file *file = v;
 929         struct trace_event_call *call;
 930         struct trace_array *tr = m->private;
 931
 932         (*pos)++;
 933
 934         list_for_each_entry_continue(file, &tr->events, list) {
 935                 call = file->event_call;
 936                 /*
 937                  * The ftrace subsystem is for showing formats only.
 938                  * They can not be enabled or disabled via the event files.
 939                  */
 940                 if (call->class && call->class->reg &&
 941                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 942                         return file;
 943         }
 944
 945         return NULL;
 946 }
 947
 948 static void *t_start(struct seq_file *m, loff_t *pos)
 949 {
 950         struct trace_event_file *file;
 951         struct trace_array *tr = m->private;
 952         loff_t l;
 953
 954         mutex_lock(&event_mutex);
 955
 956         file = list_entry(&tr->events, struct trace_event_file, list);
 957         for (l = 0; l <= *pos; ) {
 958                 file = t_next(m, file, &l);
 959                 if (!file)
 960                         break;
 961         }
 962         return file;
 963 }
 964
 965 static void *
 966 s_next(struct seq_file *m, void *v, loff_t *pos)
 967 {
 968         struct trace_event_file *file = v;
 969         struct trace_array *tr = m->private;
 970
 971         (*pos)++;
 972
 973         list_for_each_entry_continue(file, &tr->events, list) {
 974                 if (file->flags & EVENT_FILE_FL_ENABLED)
 975                         return file;
 976         }
 977
 978         return NULL;
 979 }
 980
 981 static void *s_start(struct seq_file *m, loff_t *pos)
 982 {
 983         struct trace_event_file *file;
 984         struct trace_array *tr = m->private;
 985         loff_t l;
 986
 987         mutex_lock(&event_mutex);
 988
 989         file = list_entry(&tr->events, struct trace_event_file, list);
 990         for (l = 0; l <= *pos; ) {
 991                 file = s_next(m, file, &l);
 992                 if (!file)
 993                         break;
 994         }
 995         return file;
 996 }
 997
 998 static int t_show(struct seq_file *m, void *v)
 999 {
1000         struct trace_event_file *file = v;
1001         struct trace_event_call *call = file->event_call;
1002
1003         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1004                 seq_printf(m, "%s:", call->class->system);
1005         seq_printf(m, "%s\n", trace_event_name(call));
1006
1007         return 0;
1008 }
1009
1010 static void t_stop(struct seq_file *m, void *p)
1011 {
1012         mutex_unlock(&event_mutex);
1013 }
1014
1015 static void *
1016 p_next(struct seq_file *m, void *v, loff_t *pos)
1017 {
1018         struct trace_array *tr = m->private;
1019         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
1020
1021         return trace_pid_next(pid_list, v, pos);
1022 }
1023
1024 static void *p_start(struct seq_file *m, loff_t *pos)
1025         __acquires(RCU)
1026 {
1027         struct trace_pid_list *pid_list;
1028         struct trace_array *tr = m->private;
1029
1030         /*
1031          * Grab the mutex, to keep calls to p_next() having the same
1032          * tr->filtered_pids as p_start() has.
1033          * If we just passed the tr->filtered_pids around, then RCU would
1034          * have been enough, but doing that makes things more complex.
1035          */
1036         mutex_lock(&event_mutex);
1037         rcu_read_lock_sched();
1038
1039         pid_list = rcu_dereference_sched(tr->filtered_pids);
1040
1041         if (!pid_list)
1042                 return NULL;
1043
1044         return trace_pid_start(pid_list, pos);
1045 }
1046
1047 static void p_stop(struct seq_file *m, void *p)
1048         __releases(RCU)
1049 {
1050         rcu_read_unlock_sched();
1051         mutex_unlock(&event_mutex);
1052 }
1053
1054 static ssize_t
1055 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1056                   loff_t *ppos)
1057 {
1058         struct trace_event_file *file;
1059         unsigned long flags;
1060         char buf[4] = "0";
1061
1062         mutex_lock(&event_mutex);
1063         file = event_file_data(filp);
1064         if (likely(file))
1065                 flags = file->flags;
1066         mutex_unlock(&event_mutex);
1067
1068         if (!file)
1069                 return -ENODEV;
1070
1071         if (flags & EVENT_FILE_FL_ENABLED &&
1072             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1073                 strcpy(buf, "1");
1074
1075         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1076             flags & EVENT_FILE_FL_SOFT_MODE)
1077                 strcat(buf, "*");
1078
1079         strcat(buf, "\n");
1080
1081         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1082 }
1083
1084 static ssize_t
1085 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1086                    loff_t *ppos)
1087 {
1088         struct trace_event_file *file;
1089         unsigned long val;
1090         int ret;
1091
1092         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1093         if (ret)
1094                 return ret;
1095
1096         ret = tracing_update_buffers();
1097         if (ret < 0)
1098                 return ret;
1099
1100         switch (val) {
1101         case 0:
1102         case 1:
1103                 ret = -ENODEV;
1104                 mutex_lock(&event_mutex);
1105                 file = event_file_data(filp);
1106                 if (likely(file))
1107                         ret = ftrace_event_enable_disable(file, val);
1108                 mutex_unlock(&event_mutex);
1109                 break;
1110
1111         default:
1112                 return -EINVAL;
1113         }
1114
1115         *ppos += cnt;
1116
1117         return ret ? ret : cnt;
1118 }
1119
1120 static ssize_t
1121 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1122                    loff_t *ppos)
1123 {
1124         const char set_to_char[4] = { '?', '0', '1', 'X' };
1125         struct trace_subsystem_dir *dir = filp->private_data;
1126         struct event_subsystem *system = dir->subsystem;
1127         struct trace_event_call *call;
1128         struct trace_event_file *file;
1129         struct trace_array *tr = dir->tr;
1130         char buf[2];
1131         int set = 0;
1132         int ret;
1133
1134         mutex_lock(&event_mutex);
1135         list_for_each_entry(file, &tr->events, list) {
1136                 call = file->event_call;
1137                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1138                         continue;
1139
1140                 if (system && strcmp(call->class->system, system->name) != 0)
1141                         continue;
1142
1143                 /*
1144                  * We need to find out if all the events are set
1145                  * or if all events or cleared, or if we have
1146                  * a mixture.
1147                  */
1148                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1149
1150                 /*
1151                  * If we have a mixture, no need to look further.
1152                  */
1153                 if (set == 3)
1154                         break;
1155         }
1156         mutex_unlock(&event_mutex);
1157
1158         buf[0] = set_to_char[set];
1159         buf[1] = '\n';
1160
1161         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1162
1163         return ret;
1164 }
1165
1166 static ssize_t
1167 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1168                     loff_t *ppos)
1169 {
1170         struct trace_subsystem_dir *dir = filp->private_data;
1171         struct event_subsystem *system = dir->subsystem;
1172         const char *name = NULL;
1173         unsigned long val;
1174         ssize_t ret;
1175
1176         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1177         if (ret)
1178                 return ret;
1179
1180         ret = tracing_update_buffers();
1181         if (ret < 0)
1182                 return ret;
1183
1184         if (val != 0 && val != 1)
1185                 return -EINVAL;
1186
1187         /*
1188          * Opening of "enable" adds a ref count to system,
1189          * so the name is safe to use.
1190          */
1191         if (system)
1192                 name = system->name;
1193
1194         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1195         if (ret)
1196                 goto out;
1197
1198         ret = cnt;
1199
1200 out:
1201         *ppos += cnt;
1202
1203         return ret;
1204 }
1205
1206 enum {
1207         FORMAT_HEADER           = 1,
1208         FORMAT_FIELD_SEPERATOR  = 2,
1209         FORMAT_PRINTFMT         = 3,
1210 };
1211
1212 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1213 {
1214         struct trace_event_call *call = event_file_data(m->private);
1215         struct list_head *common_head = &ftrace_common_fields;
1216         struct list_head *head = trace_get_fields(call);
1217         struct list_head *node = v;
1218
1219         (*pos)++;
1220
1221         switch ((unsigned long)v) {
1222         case FORMAT_HEADER:
1223                 node = common_head;
1224                 break;
1225
1226         case FORMAT_FIELD_SEPERATOR:
1227                 node = head;
1228                 break;
1229
1230         case FORMAT_PRINTFMT:
1231                 /* all done */
1232                 return NULL;
1233         }
1234
1235         node = node->prev;
1236         if (node == common_head)
1237                 return (void *)FORMAT_FIELD_SEPERATOR;
1238         else if (node == head)
1239                 return (void *)FORMAT_PRINTFMT;
1240         else
1241                 return node;
1242 }
1243
1244 static int f_show(struct seq_file *m, void *v)
1245 {
1246         struct trace_event_call *call = event_file_data(m->private);
1247         struct ftrace_event_field *field;
1248         const char *array_descriptor;
1249
1250         switch ((unsigned long)v) {
1251         case FORMAT_HEADER:
1252                 seq_printf(m, "name: %s\n", trace_event_name(call));
1253                 seq_printf(m, "ID: %d\n", call->event.type);
1254                 seq_puts(m, "format:\n");
1255                 return 0;
1256
1257         case FORMAT_FIELD_SEPERATOR:
1258                 seq_putc(m, '\n');
1259                 return 0;
1260
1261         case FORMAT_PRINTFMT:
1262                 seq_printf(m, "\nprint fmt: %s\n",
1263                            call->print_fmt);
1264                 return 0;
1265         }
1266
1267         field = list_entry(v, struct ftrace_event_field, link);
1268         /*
1269          * Smartly shows the array type(except dynamic array).
1270          * Normal:
1271          *      field:TYPE VAR
1272          * If TYPE := TYPE[LEN], it is shown:
1273          *      field:TYPE VAR[LEN]
1274          */
1275         array_descriptor = strchr(field->type, '[');
1276
1277         if (str_has_prefix(field->type, "__data_loc"))
1278                 array_descriptor = NULL;
1279
1280         if (!array_descriptor)
1281                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1282                            field->type, field->name, field->offset,
1283                            field->size, !!field->is_signed);
1284         else
1285                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1286                            (int)(array_descriptor - field->type),
1287                            field->type, field->name,
1288                            array_descriptor, field->offset,
1289                            field->size, !!field->is_signed);
1290
1291         return 0;
1292 }
1293
1294 static void *f_start(struct seq_file *m, loff_t *pos)
1295 {
1296         void *p = (void *)FORMAT_HEADER;
1297         loff_t l = 0;
1298
1299         /* ->stop() is called even if ->start() fails */
1300         mutex_lock(&event_mutex);
1301         if (!event_file_data(m->private))
1302                 return ERR_PTR(-ENODEV);
1303
1304         while (l < *pos && p)
1305                 p = f_next(m, p, &l);
1306
1307         return p;
1308 }
1309
1310 static void f_stop(struct seq_file *m, void *p)
1311 {
1312         mutex_unlock(&event_mutex);
1313 }
1314
1315 static const struct seq_operations trace_format_seq_ops = {
1316         .start          = f_start,
1317         .next           = f_next,
1318         .stop           = f_stop,
1319         .show           = f_show,
1320 };
1321
1322 static int trace_format_open(struct inode *inode, struct file *file)
1323 {
1324         struct seq_file *m;
1325         int ret;
1326
1327         /* Do we want to hide event format files on tracefs lockdown? */
1328
1329         ret = seq_open(file, &trace_format_seq_ops);
1330         if (ret < 0)
1331                 return ret;
1332
1333         m = file->private_data;
1334         m->private = file;
1335
1336         return 0;
1337 }
1338
1339 static ssize_t
1340 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1341 {
1342         int id = (long)event_file_data(filp);
1343         char buf[32];
1344         int len;
1345
1346         if (unlikely(!id))
1347                 return -ENODEV;
1348
1349         len = sprintf(buf, "%d\n", id);
1350
1351         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1352 }
1353
1354 static ssize_t
1355 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1356                   loff_t *ppos)
1357 {
1358         struct trace_event_file *file;
1359         struct trace_seq *s;
1360         int r = -ENODEV;
1361
1362         if (*ppos)
1363                 return 0;
1364
1365         s = kmalloc(sizeof(*s), GFP_KERNEL);
1366
1367         if (!s)
1368                 return -ENOMEM;
1369
1370         trace_seq_init(s);
1371
1372         mutex_lock(&event_mutex);
1373         file = event_file_data(filp);
1374         if (file)
1375                 print_event_filter(file, s);
1376         mutex_unlock(&event_mutex);
1377
1378         if (file)
1379                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1380                                             s->buffer, trace_seq_used(s));
1381
1382         kfree(s);
1383
1384         return r;
1385 }
1386
1387 static ssize_t
1388 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1389                    loff_t *ppos)
1390 {
1391         struct trace_event_file *file;
1392         char *buf;
1393         int err = -ENODEV;
1394
1395         if (cnt >= PAGE_SIZE)
1396                 return -EINVAL;
1397
1398         buf = memdup_user_nul(ubuf, cnt);
1399         if (IS_ERR(buf))
1400                 return PTR_ERR(buf);
1401
1402         mutex_lock(&event_mutex);
1403         file = event_file_data(filp);
1404         if (file)
1405                 err = apply_event_filter(file, buf);
1406         mutex_unlock(&event_mutex);
1407
1408         kfree(buf);
1409         if (err < 0)
1410                 return err;
1411
1412         *ppos += cnt;
1413
1414         return cnt;
1415 }
1416
1417 static LIST_HEAD(event_subsystems);
1418
1419 static int subsystem_open(struct inode *inode, struct file *filp)
1420 {
1421         struct event_subsystem *system = NULL;
1422         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1423         struct trace_array *tr;
1424         int ret;
1425
1426         if (tracing_is_disabled())
1427                 return -ENODEV;
1428
1429         /* Make sure the system still exists */
1430         mutex_lock(&event_mutex);
1431         mutex_lock(&trace_types_lock);
1432         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1433                 list_for_each_entry(dir, &tr->systems, list) {
1434                         if (dir == inode->i_private) {
1435                                 /* Don't open systems with no events */
1436                                 if (dir->nr_events) {
1437                                         __get_system_dir(dir);
1438                                         system = dir->subsystem;
1439                                 }
1440                                 goto exit_loop;
1441                         }
1442                 }
1443         }
1444  exit_loop:
1445         mutex_unlock(&trace_types_lock);
1446         mutex_unlock(&event_mutex);
1447
1448         if (!system)
1449                 return -ENODEV;
1450
1451         /* Some versions of gcc think dir can be uninitialized here */
1452         WARN_ON(!dir);
1453
1454         /* Still need to increment the ref count of the system */
1455         if (trace_array_get(tr) < 0) {
1456                 put_system(dir);
1457                 return -ENODEV;
1458         }
1459
1460         ret = tracing_open_generic(inode, filp);
1461         if (ret < 0) {
1462                 trace_array_put(tr);
1463                 put_system(dir);
1464         }
1465
1466         return ret;
1467 }
1468
1469 static int system_tr_open(struct inode *inode, struct file *filp)
1470 {
1471         struct trace_subsystem_dir *dir;
1472         struct trace_array *tr = inode->i_private;
1473         int ret;
1474
1475         /* Make a temporary dir that has no system but points to tr */
1476         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1477         if (!dir)
1478                 return -ENOMEM;
1479
1480         ret = tracing_open_generic_tr(inode, filp);
1481         if (ret < 0) {
1482                 kfree(dir);
1483                 return ret;
1484         }
1485         dir->tr = tr;
1486         filp->private_data = dir;
1487
1488         return 0;
1489 }
1490
1491 static int subsystem_release(struct inode *inode, struct file *file)
1492 {
1493         struct trace_subsystem_dir *dir = file->private_data;
1494
1495         trace_array_put(dir->tr);
1496
1497         /*
1498          * If dir->subsystem is NULL, then this is a temporary
1499          * descriptor that was made for a trace_array to enable
1500          * all subsystems.
1501          */
1502         if (dir->subsystem)
1503                 put_system(dir);
1504         else
1505                 kfree(dir);
1506
1507         return 0;
1508 }
1509
1510 static ssize_t
1511 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1512                       loff_t *ppos)
1513 {
1514         struct trace_subsystem_dir *dir = filp->private_data;
1515         struct event_subsystem *system = dir->subsystem;
1516         struct trace_seq *s;
1517         int r;
1518
1519         if (*ppos)
1520                 return 0;
1521
1522         s = kmalloc(sizeof(*s), GFP_KERNEL);
1523         if (!s)
1524                 return -ENOMEM;
1525
1526         trace_seq_init(s);
1527
1528         print_subsystem_event_filter(system, s);
1529         r = simple_read_from_buffer(ubuf, cnt, ppos,
1530                                     s->buffer, trace_seq_used(s));
1531
1532         kfree(s);
1533
1534         return r;
1535 }
1536
1537 static ssize_t
1538 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1539                        loff_t *ppos)
1540 {
1541         struct trace_subsystem_dir *dir = filp->private_data;
1542         char *buf;
1543         int err;
1544
1545         if (cnt >= PAGE_SIZE)
1546                 return -EINVAL;
1547
1548         buf = memdup_user_nul(ubuf, cnt);
1549         if (IS_ERR(buf))
1550                 return PTR_ERR(buf);
1551
1552         err = apply_subsystem_event_filter(dir, buf);
1553         kfree(buf);
1554         if (err < 0)
1555                 return err;
1556
1557         *ppos += cnt;
1558
1559         return cnt;
1560 }
1561
1562 static ssize_t
1563 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1564 {
1565         int (*func)(struct trace_seq *s) = filp->private_data;
1566         struct trace_seq *s;
1567         int r;
1568
1569         if (*ppos)
1570                 return 0;
1571
1572         s = kmalloc(sizeof(*s), GFP_KERNEL);
1573         if (!s)
1574                 return -ENOMEM;
1575
1576         trace_seq_init(s);
1577
1578         func(s);
1579         r = simple_read_from_buffer(ubuf, cnt, ppos,
1580                                     s->buffer, trace_seq_used(s));
1581
1582         kfree(s);
1583
1584         return r;
1585 }
1586
1587 static void ignore_task_cpu(void *data)
1588 {
1589         struct trace_array *tr = data;
1590         struct trace_pid_list *pid_list;
1591
1592         /*
1593          * This function is called by on_each_cpu() while the
1594          * event_mutex is held.
1595          */
1596         pid_list = rcu_dereference_protected(tr->filtered_pids,
1597                                              mutex_is_locked(&event_mutex));
1598
1599         this_cpu_write(tr->array_buffer.data->ignore_pid,
1600                        trace_ignore_this_task(pid_list, current));
1601 }
1602
1603 static ssize_t
1604 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1605                        size_t cnt, loff_t *ppos)
1606 {
1607         struct seq_file *m = filp->private_data;
1608         struct trace_array *tr = m->private;
1609         struct trace_pid_list *filtered_pids = NULL;
1610         struct trace_pid_list *pid_list;
1611         struct trace_event_file *file;
1612         ssize_t ret;
1613
1614         if (!cnt)
1615                 return 0;
1616
1617         ret = tracing_update_buffers();
1618         if (ret < 0)
1619                 return ret;
1620
1621         mutex_lock(&event_mutex);
1622
1623         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1624                                              lockdep_is_held(&event_mutex));
1625
1626         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1627         if (ret < 0)
1628                 goto out;
1629
1630         rcu_assign_pointer(tr->filtered_pids, pid_list);
1631
1632         list_for_each_entry(file, &tr->events, list) {
1633                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1634         }
1635
1636         if (filtered_pids) {
1637                 tracepoint_synchronize_unregister();
1638                 trace_free_pid_list(filtered_pids);
1639         } else if (pid_list) {
1640                 /*
1641                  * Register a probe that is called before all other probes
1642                  * to set ignore_pid if next or prev do not match.
1643                  * Register a probe this is called after all other probes
1644                  * to only keep ignore_pid set if next pid matches.
1645                  */
1646                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1647                                                  tr, INT_MAX);
1648                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1649                                                  tr, 0);
1650
1651                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1652                                                  tr, INT_MAX);
1653                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1654                                                  tr, 0);
1655
1656                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1657                                                      tr, INT_MAX);
1658                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1659                                                      tr, 0);
1660
1661                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1662                                                  tr, INT_MAX);
1663                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1664                                                  tr, 0);
1665         }
1666
1667         /*
1668          * Ignoring of pids is done at task switch. But we have to
1669          * check for those tasks that are currently running.
1670          * Always do this in case a pid was appended or removed.
1671          */
1672         on_each_cpu(ignore_task_cpu, tr, 1);
1673
1674  out:
1675         mutex_unlock(&event_mutex);
1676
1677         if (ret > 0)
1678                 *ppos += ret;
1679
1680         return ret;
1681 }
1682
1683 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1684 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1685 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1686 static int ftrace_event_release(struct inode *inode, struct file *file);
1687
1688 static const struct seq_operations show_event_seq_ops = {
1689         .start = t_start,
1690         .next = t_next,
1691         .show = t_show,
1692         .stop = t_stop,
1693 };
1694
1695 static const struct seq_operations show_set_event_seq_ops = {
1696         .start = s_start,
1697         .next = s_next,
1698         .show = t_show,
1699         .stop = t_stop,
1700 };
1701
1702 static const struct seq_operations show_set_pid_seq_ops = {
1703         .start = p_start,
1704         .next = p_next,
1705         .show = trace_pid_show,
1706         .stop = p_stop,
1707 };
1708
1709 static const struct file_operations ftrace_avail_fops = {
1710         .open = ftrace_event_avail_open,
1711         .read = seq_read,
1712         .llseek = seq_lseek,
1713         .release = seq_release,
1714 };
1715
1716 static const struct file_operations ftrace_set_event_fops = {
1717         .open = ftrace_event_set_open,
1718         .read = seq_read,
1719         .write = ftrace_event_write,
1720         .llseek = seq_lseek,
1721         .release = ftrace_event_release,
1722 };
1723
1724 static const struct file_operations ftrace_set_event_pid_fops = {
1725         .open = ftrace_event_set_pid_open,
1726         .read = seq_read,
1727         .write = ftrace_event_pid_write,
1728         .llseek = seq_lseek,
1729         .release = ftrace_event_release,
1730 };
1731
1732 static const struct file_operations ftrace_enable_fops = {
1733         .open = tracing_open_generic,
1734         .read = event_enable_read,
1735         .write = event_enable_write,
1736         .llseek = default_llseek,
1737 };
1738
1739 static const struct file_operations ftrace_event_format_fops = {
1740         .open = trace_format_open,
1741         .read = seq_read,
1742         .llseek = seq_lseek,
1743         .release = seq_release,
1744 };
1745
1746 static const struct file_operations ftrace_event_id_fops = {
1747         .read = event_id_read,
1748         .llseek = default_llseek,
1749 };
1750
1751 static const struct file_operations ftrace_event_filter_fops = {
1752         .open = tracing_open_generic,
1753         .read = event_filter_read,
1754         .write = event_filter_write,
1755         .llseek = default_llseek,
1756 };
1757
1758 static const struct file_operations ftrace_subsystem_filter_fops = {
1759         .open = subsystem_open,
1760         .read = subsystem_filter_read,
1761         .write = subsystem_filter_write,
1762         .llseek = default_llseek,
1763         .release = subsystem_release,
1764 };
1765
1766 static const struct file_operations ftrace_system_enable_fops = {
1767         .open = subsystem_open,
1768         .read = system_enable_read,
1769         .write = system_enable_write,
1770         .llseek = default_llseek,
1771         .release = subsystem_release,
1772 };
1773
1774 static const struct file_operations ftrace_tr_enable_fops = {
1775         .open = system_tr_open,
1776         .read = system_enable_read,
1777         .write = system_enable_write,
1778         .llseek = default_llseek,
1779         .release = subsystem_release,
1780 };
1781
1782 static const struct file_operations ftrace_show_header_fops = {
1783         .open = tracing_open_generic,
1784         .read = show_header,
1785         .llseek = default_llseek,
1786 };
1787
1788 static int
1789 ftrace_event_open(struct inode *inode, struct file *file,
1790                   const struct seq_operations *seq_ops)
1791 {
1792         struct seq_file *m;
1793         int ret;
1794
1795         ret = security_locked_down(LOCKDOWN_TRACEFS);
1796         if (ret)
1797                 return ret;
1798
1799         ret = seq_open(file, seq_ops);
1800         if (ret < 0)
1801                 return ret;
1802         m = file->private_data;
1803         /* copy tr over to seq ops */
1804         m->private = inode->i_private;
1805
1806         return ret;
1807 }
1808
1809 static int ftrace_event_release(struct inode *inode, struct file *file)
1810 {
1811         struct trace_array *tr = inode->i_private;
1812
1813         trace_array_put(tr);
1814
1815         return seq_release(inode, file);
1816 }
1817
1818 static int
1819 ftrace_event_avail_open(struct inode *inode, struct file *file)
1820 {
1821         const struct seq_operations *seq_ops = &show_event_seq_ops;
1822
1823         /* Checks for tracefs lockdown */
1824         return ftrace_event_open(inode, file, seq_ops);
1825 }
1826
1827 static int
1828 ftrace_event_set_open(struct inode *inode, struct file *file)
1829 {
1830         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1831         struct trace_array *tr = inode->i_private;
1832         int ret;
1833
1834         ret = tracing_check_open_get_tr(tr);
1835         if (ret)
1836                 return ret;
1837
1838         if ((file->f_mode & FMODE_WRITE) &&
1839             (file->f_flags & O_TRUNC))
1840                 ftrace_clear_events(tr);
1841
1842         ret = ftrace_event_open(inode, file, seq_ops);
1843         if (ret < 0)
1844                 trace_array_put(tr);
1845         return ret;
1846 }
1847
1848 static int
1849 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1850 {
1851         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1852         struct trace_array *tr = inode->i_private;
1853         int ret;
1854
1855         ret = tracing_check_open_get_tr(tr);
1856         if (ret)
1857                 return ret;
1858
1859         if ((file->f_mode & FMODE_WRITE) &&
1860             (file->f_flags & O_TRUNC))
1861                 ftrace_clear_event_pids(tr);
1862
1863         ret = ftrace_event_open(inode, file, seq_ops);
1864         if (ret < 0)
1865                 trace_array_put(tr);
1866         return ret;
1867 }
1868
1869 static struct event_subsystem *
1870 create_new_subsystem(const char *name)
1871 {
1872         struct event_subsystem *system;
1873
1874         /* need to create new entry */
1875         system = kmalloc(sizeof(*system), GFP_KERNEL);
1876         if (!system)
1877                 return NULL;
1878
1879         system->ref_count = 1;
1880
1881         /* Only allocate if dynamic (kprobes and modules) */
1882         system->name = kstrdup_const(name, GFP_KERNEL);
1883         if (!system->name)
1884                 goto out_free;
1885
1886         system->filter = NULL;
1887
1888         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1889         if (!system->filter)
1890                 goto out_free;
1891
1892         list_add(&system->list, &event_subsystems);
1893
1894         return system;
1895
1896  out_free:
1897         kfree_const(system->name);
1898         kfree(system);
1899         return NULL;
1900 }
1901
1902 static struct dentry *
1903 event_subsystem_dir(struct trace_array *tr, const char *name,
1904                     struct trace_event_file *file, struct dentry *parent)
1905 {
1906         struct trace_subsystem_dir *dir;
1907         struct event_subsystem *system;
1908         struct dentry *entry;
1909
1910         /* First see if we did not already create this dir */
1911         list_for_each_entry(dir, &tr->systems, list) {
1912                 system = dir->subsystem;
1913                 if (strcmp(system->name, name) == 0) {
1914                         dir->nr_events++;
1915                         file->system = dir;
1916                         return dir->entry;
1917                 }
1918         }
1919
1920         /* Now see if the system itself exists. */
1921         list_for_each_entry(system, &event_subsystems, list) {
1922                 if (strcmp(system->name, name) == 0)
1923                         break;
1924         }
1925         /* Reset system variable when not found */
1926         if (&system->list == &event_subsystems)
1927                 system = NULL;
1928
1929         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1930         if (!dir)
1931                 goto out_fail;
1932
1933         if (!system) {
1934                 system = create_new_subsystem(name);
1935                 if (!system)
1936                         goto out_free;
1937         } else
1938                 __get_system(system);
1939
1940         dir->entry = tracefs_create_dir(name, parent);
1941         if (!dir->entry) {
1942                 pr_warn("Failed to create system directory %s\n", name);
1943                 __put_system(system);
1944                 goto out_free;
1945         }
1946
1947         dir->tr = tr;
1948         dir->ref_count = 1;
1949         dir->nr_events = 1;
1950         dir->subsystem = system;
1951         file->system = dir;
1952
1953         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1954                                     &ftrace_subsystem_filter_fops);
1955         if (!entry) {
1956                 kfree(system->filter);
1957                 system->filter = NULL;
1958                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1959         }
1960
1961         trace_create_file("enable", 0644, dir->entry, dir,
1962                           &ftrace_system_enable_fops);
1963
1964         list_add(&dir->list, &tr->systems);
1965
1966         return dir->entry;
1967
1968  out_free:
1969         kfree(dir);
1970  out_fail:
1971         /* Only print this message if failed on memory allocation */
1972         if (!dir || !system)
1973                 pr_warn("No memory to create event subsystem %s\n", name);
1974         return NULL;
1975 }
1976
1977 static int
1978 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1979 {
1980         struct trace_event_call *call = file->event_call;
1981         struct trace_array *tr = file->tr;
1982         struct list_head *head;
1983         struct dentry *d_events;
1984         const char *name;
1985         int ret;
1986
1987         /*
1988          * If the trace point header did not define TRACE_SYSTEM
1989          * then the system would be called "TRACE_SYSTEM".
1990          */
1991         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1992                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1993                 if (!d_events)
1994                         return -ENOMEM;
1995         } else
1996                 d_events = parent;
1997
1998         name = trace_event_name(call);
1999         file->dir = tracefs_create_dir(name, d_events);
2000         if (!file->dir) {
2001                 pr_warn("Could not create tracefs '%s' directory\n", name);
2002                 return -1;
2003         }
2004
2005         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2006                 trace_create_file("enable", 0644, file->dir, file,
2007                                   &ftrace_enable_fops);
2008
2009 #ifdef CONFIG_PERF_EVENTS
2010         if (call->event.type && call->class->reg)
2011                 trace_create_file("id", 0444, file->dir,
2012                                   (void *)(long)call->event.type,
2013                                   &ftrace_event_id_fops);
2014 #endif
2015
2016         /*
2017          * Other events may have the same class. Only update
2018          * the fields if they are not already defined.
2019          */
2020         head = trace_get_fields(call);
2021         if (list_empty(head)) {
2022                 struct trace_event_fields *field = call->class->fields_array;
2023                 unsigned int offset = sizeof(struct trace_entry);
2024
2025                 for (; field->type; field++) {
2026                         if (field->type == TRACE_FUNCTION_TYPE) {
2027                                 ret = field->define_fields(call);
2028                                 break;
2029                         }
2030
2031                         offset = ALIGN(offset, field->align);
2032                         ret = trace_define_field(call, field->type, field->name,
2033                                                  offset, field->size,
2034                                                  field->is_signed, field->filter_type);
2035                         if (ret)
2036                                 break;
2037
2038                         offset += field->size;
2039                 }
2040                 if (ret < 0) {
2041                         pr_warn("Could not initialize trace point events/%s\n",
2042                                 name);
2043                         return -1;
2044                 }
2045         }
2046
2047         /*
2048          * Only event directories that can be enabled should have
2049          * triggers or filters.
2050          */
2051         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2052                 trace_create_file("filter", 0644, file->dir, file,
2053                                   &ftrace_event_filter_fops);
2054
2055                 trace_create_file("trigger", 0644, file->dir, file,
2056                                   &event_trigger_fops);
2057         }
2058
2059 #ifdef CONFIG_HIST_TRIGGERS
2060         trace_create_file("hist", 0444, file->dir, file,
2061                           &event_hist_fops);
2062 #endif
2063         trace_create_file("format", 0444, file->dir, call,
2064                           &ftrace_event_format_fops);
2065
2066 #ifdef CONFIG_TRACE_EVENT_INJECT
2067         if (call->event.type && call->class->reg)
2068                 trace_create_file("inject", 0200, file->dir, file,
2069                                   &event_inject_fops);
2070 #endif
2071
2072         return 0;
2073 }
2074
2075 static void remove_event_from_tracers(struct trace_event_call *call)
2076 {
2077         struct trace_event_file *file;
2078         struct trace_array *tr;
2079
2080         do_for_each_event_file_safe(tr, file) {
2081                 if (file->event_call != call)
2082                         continue;
2083
2084                 remove_event_file_dir(file);
2085                 /*
2086                  * The do_for_each_event_file_safe() is
2087                  * a double loop. After finding the call for this
2088                  * trace_array, we use break to jump to the next
2089                  * trace_array.
2090                  */
2091                 break;
2092         } while_for_each_event_file();
2093 }
2094
2095 static void event_remove(struct trace_event_call *call)
2096 {
2097         struct trace_array *tr;
2098         struct trace_event_file *file;
2099
2100         do_for_each_event_file(tr, file) {
2101                 if (file->event_call != call)
2102                         continue;
2103
2104                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2105                         tr->clear_trace = true;
2106
2107                 ftrace_event_enable_disable(file, 0);
2108                 /*
2109                  * The do_for_each_event_file() is
2110                  * a double loop. After finding the call for this
2111                  * trace_array, we use break to jump to the next
2112                  * trace_array.
2113                  */
2114                 break;
2115         } while_for_each_event_file();
2116
2117         if (call->event.funcs)
2118                 __unregister_trace_event(&call->event);
2119         remove_event_from_tracers(call);
2120         list_del(&call->list);
2121 }
2122
2123 static int event_init(struct trace_event_call *call)
2124 {
2125         int ret = 0;
2126         const char *name;
2127
2128         name = trace_event_name(call);
2129         if (WARN_ON(!name))
2130                 return -EINVAL;
2131
2132         if (call->class->raw_init) {
2133                 ret = call->class->raw_init(call);
2134                 if (ret < 0 && ret != -ENOSYS)
2135                         pr_warn("Could not initialize trace events/%s\n", name);
2136         }
2137
2138         return ret;
2139 }
2140
2141 static int
2142 __register_event(struct trace_event_call *call, struct module *mod)
2143 {
2144         int ret;
2145
2146         ret = event_init(call);
2147         if (ret < 0)
2148                 return ret;
2149
2150         list_add(&call->list, &ftrace_events);
2151         call->mod = mod;
2152
2153         return 0;
2154 }
2155
2156 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2157 {
2158         int rlen;
2159         int elen;
2160
2161         /* Find the length of the eval value as a string */
2162         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2163         /* Make sure there's enough room to replace the string with the value */
2164         if (len < elen)
2165                 return NULL;
2166
2167         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2168
2169         /* Get the rest of the string of ptr */
2170         rlen = strlen(ptr + len);
2171         memmove(ptr + elen, ptr + len, rlen);
2172         /* Make sure we end the new string */
2173         ptr[elen + rlen] = 0;
2174
2175         return ptr + elen;
2176 }
2177
2178 static void update_event_printk(struct trace_event_call *call,
2179                                 struct trace_eval_map *map)
2180 {
2181         char *ptr;
2182         int quote = 0;
2183         int len = strlen(map->eval_string);
2184
2185         for (ptr = call->print_fmt; *ptr; ptr++) {
2186                 if (*ptr == '\\') {
2187                         ptr++;
2188                         /* paranoid */
2189                         if (!*ptr)
2190                                 break;
2191                         continue;
2192                 }
2193                 if (*ptr == '"') {
2194                         quote ^= 1;
2195                         continue;
2196                 }
2197                 if (quote)
2198                         continue;
2199                 if (isdigit(*ptr)) {
2200                         /* skip numbers */
2201                         do {
2202                                 ptr++;
2203                                 /* Check for alpha chars like ULL */
2204                         } while (isalnum(*ptr));
2205                         if (!*ptr)
2206                                 break;
2207                         /*
2208                          * A number must have some kind of delimiter after
2209                          * it, and we can ignore that too.
2210                          */
2211                         continue;
2212                 }
2213                 if (isalpha(*ptr) || *ptr == '_') {
2214                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2215                             !isalnum(ptr[len]) && ptr[len] != '_') {
2216                                 ptr = eval_replace(ptr, map, len);
2217                                 /* enum/sizeof string smaller than value */
2218                                 if (WARN_ON_ONCE(!ptr))
2219                                         return;
2220                                 /*
2221                                  * No need to decrement here, as eval_replace()
2222                                  * returns the pointer to the character passed
2223                                  * the eval, and two evals can not be placed
2224                                  * back to back without something in between.
2225                                  * We can skip that something in between.
2226                                  */
2227                                 continue;
2228                         }
2229                 skip_more:
2230                         do {
2231                                 ptr++;
2232                         } while (isalnum(*ptr) || *ptr == '_');
2233                         if (!*ptr)
2234                                 break;
2235                         /*
2236                          * If what comes after this variable is a '.' or
2237                          * '->' then we can continue to ignore that string.
2238                          */
2239                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2240                                 ptr += *ptr == '.' ? 1 : 2;
2241                                 if (!*ptr)
2242                                         break;
2243                                 goto skip_more;
2244                         }
2245                         /*
2246                          * Once again, we can skip the delimiter that came
2247                          * after the string.
2248                          */
2249                         continue;
2250                 }
2251         }
2252 }
2253
2254 void trace_event_eval_update(struct trace_eval_map **map, int len)
2255 {
2256         struct trace_event_call *call, *p;
2257         const char *last_system = NULL;
2258         bool first = false;
2259         int last_i;
2260         int i;
2261
2262         down_write(&trace_event_sem);
2263         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2264                 /* events are usually grouped together with systems */
2265                 if (!last_system || call->class->system != last_system) {
2266                         first = true;
2267                         last_i = 0;
2268                         last_system = call->class->system;
2269                 }
2270
2271                 /*
2272                  * Since calls are grouped by systems, the likelyhood that the
2273                  * next call in the iteration belongs to the same system as the
2274                  * previous call is high. As an optimization, we skip seaching
2275                  * for a map[] that matches the call's system if the last call
2276                  * was from the same system. That's what last_i is for. If the
2277                  * call has the same system as the previous call, then last_i
2278                  * will be the index of the first map[] that has a matching
2279                  * system.
2280                  */
2281                 for (i = last_i; i < len; i++) {
2282                         if (call->class->system == map[i]->system) {
2283                                 /* Save the first system if need be */
2284                                 if (first) {
2285                                         last_i = i;
2286                                         first = false;
2287                                 }
2288                                 update_event_printk(call, map[i]);
2289                         }
2290                 }
2291         }
2292         up_write(&trace_event_sem);
2293 }
2294
2295 static struct trace_event_file *
2296 trace_create_new_event(struct trace_event_call *call,
2297                        struct trace_array *tr)
2298 {
2299         struct trace_event_file *file;
2300
2301         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2302         if (!file)
2303                 return NULL;
2304
2305         file->event_call = call;
2306         file->tr = tr;
2307         atomic_set(&file->sm_ref, 0);
2308         atomic_set(&file->tm_ref, 0);
2309         INIT_LIST_HEAD(&file->triggers);
2310         list_add(&file->list, &tr->events);
2311
2312         return file;
2313 }
2314
2315 /* Add an event to a trace directory */
2316 static int
2317 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2318 {
2319         struct trace_event_file *file;
2320
2321         file = trace_create_new_event(call, tr);
2322         if (!file)
2323                 return -ENOMEM;
2324
2325         return event_create_dir(tr->event_dir, file);
2326 }
2327
2328 /*
2329  * Just create a decriptor for early init. A descriptor is required
2330  * for enabling events at boot. We want to enable events before
2331  * the filesystem is initialized.
2332  */
2333 static __init int
2334 __trace_early_add_new_event(struct trace_event_call *call,
2335                             struct trace_array *tr)
2336 {
2337         struct trace_event_file *file;
2338
2339         file = trace_create_new_event(call, tr);
2340         if (!file)
2341                 return -ENOMEM;
2342
2343         return 0;
2344 }
2345
2346 struct ftrace_module_file_ops;
2347 static void __add_event_to_tracers(struct trace_event_call *call);
2348
2349 /* Add an additional event_call dynamically */
2350 int trace_add_event_call(struct trace_event_call *call)
2351 {
2352         int ret;
2353         lockdep_assert_held(&event_mutex);
2354
2355         mutex_lock(&trace_types_lock);
2356
2357         ret = __register_event(call, NULL);
2358         if (ret >= 0)
2359                 __add_event_to_tracers(call);
2360
2361         mutex_unlock(&trace_types_lock);
2362         return ret;
2363 }
2364
2365 /*
2366  * Must be called under locking of trace_types_lock, event_mutex and
2367  * trace_event_sem.
2368  */
2369 static void __trace_remove_event_call(struct trace_event_call *call)
2370 {
2371         event_remove(call);
2372         trace_destroy_fields(call);
2373         free_event_filter(call->filter);
2374         call->filter = NULL;
2375 }
2376
2377 static int probe_remove_event_call(struct trace_event_call *call)
2378 {
2379         struct trace_array *tr;
2380         struct trace_event_file *file;
2381
2382 #ifdef CONFIG_PERF_EVENTS
2383         if (call->perf_refcount)
2384                 return -EBUSY;
2385 #endif
2386         do_for_each_event_file(tr, file) {
2387                 if (file->event_call != call)
2388                         continue;
2389                 /*
2390                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2391                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2392                  * TRACE_REG_UNREGISTER.
2393                  */
2394                 if (file->flags & EVENT_FILE_FL_ENABLED)
2395                         return -EBUSY;
2396                 /*
2397                  * The do_for_each_event_file_safe() is
2398                  * a double loop. After finding the call for this
2399                  * trace_array, we use break to jump to the next
2400                  * trace_array.
2401                  */
2402                 break;
2403         } while_for_each_event_file();
2404
2405         __trace_remove_event_call(call);
2406
2407         return 0;
2408 }
2409
2410 /* Remove an event_call */
2411 int trace_remove_event_call(struct trace_event_call *call)
2412 {
2413         int ret;
2414
2415         lockdep_assert_held(&event_mutex);
2416
2417         mutex_lock(&trace_types_lock);
2418         down_write(&trace_event_sem);
2419         ret = probe_remove_event_call(call);
2420         up_write(&trace_event_sem);
2421         mutex_unlock(&trace_types_lock);
2422
2423         return ret;
2424 }
2425
2426 #define for_each_event(event, start, end)                       \
2427         for (event = start;                                     \
2428              (unsigned long)event < (unsigned long)end;         \
2429              event++)
2430
2431 #ifdef CONFIG_MODULES
2432
2433 static void trace_module_add_events(struct module *mod)
2434 {
2435         struct trace_event_call **call, **start, **end;
2436
2437         if (!mod->num_trace_events)
2438                 return;
2439
2440         /* Don't add infrastructure for mods without tracepoints */
2441         if (trace_module_has_bad_taint(mod)) {
2442                 pr_err("%s: module has bad taint, not creating trace events\n",
2443                        mod->name);
2444                 return;
2445         }
2446
2447         start = mod->trace_events;
2448         end = mod->trace_events + mod->num_trace_events;
2449
2450         for_each_event(call, start, end) {
2451                 __register_event(*call, mod);
2452                 __add_event_to_tracers(*call);
2453         }
2454 }
2455
2456 static void trace_module_remove_events(struct module *mod)
2457 {
2458         struct trace_event_call *call, *p;
2459
2460         down_write(&trace_event_sem);
2461         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2462                 if (call->mod == mod)
2463                         __trace_remove_event_call(call);
2464         }
2465         up_write(&trace_event_sem);
2466
2467         /*
2468          * It is safest to reset the ring buffer if the module being unloaded
2469          * registered any events that were used. The only worry is if
2470          * a new module gets loaded, and takes on the same id as the events
2471          * of this module. When printing out the buffer, traced events left
2472          * over from this module may be passed to the new module events and
2473          * unexpected results may occur.
2474          */
2475         tracing_reset_all_online_cpus();
2476 }
2477
2478 static int trace_module_notify(struct notifier_block *self,
2479                                unsigned long val, void *data)
2480 {
2481         struct module *mod = data;
2482
2483         mutex_lock(&event_mutex);
2484         mutex_lock(&trace_types_lock);
2485         switch (val) {
2486         case MODULE_STATE_COMING:
2487                 trace_module_add_events(mod);
2488                 break;
2489         case MODULE_STATE_GOING:
2490                 trace_module_remove_events(mod);
2491                 break;
2492         }
2493         mutex_unlock(&trace_types_lock);
2494         mutex_unlock(&event_mutex);
2495
2496         return 0;
2497 }
2498
2499 static struct notifier_block trace_module_nb = {
2500         .notifier_call = trace_module_notify,
2501         .priority = 1, /* higher than trace.c module notify */
2502 };
2503 #endif /* CONFIG_MODULES */
2504
2505 /* Create a new event directory structure for a trace directory. */
2506 static void
2507 __trace_add_event_dirs(struct trace_array *tr)
2508 {
2509         struct trace_event_call *call;
2510         int ret;
2511
2512         list_for_each_entry(call, &ftrace_events, list) {
2513                 ret = __trace_add_new_event(call, tr);
2514                 if (ret < 0)
2515                         pr_warn("Could not create directory for event %s\n",
2516                                 trace_event_name(call));
2517         }
2518 }
2519
2520 /* Returns any file that matches the system and event */
2521 struct trace_event_file *
2522 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2523 {
2524         struct trace_event_file *file;
2525         struct trace_event_call *call;
2526         const char *name;
2527
2528         list_for_each_entry(file, &tr->events, list) {
2529
2530                 call = file->event_call;
2531                 name = trace_event_name(call);
2532
2533                 if (!name || !call->class)
2534                         continue;
2535
2536                 if (strcmp(event, name) == 0 &&
2537                     strcmp(system, call->class->system) == 0)
2538                         return file;
2539         }
2540         return NULL;
2541 }
2542
2543 /* Returns valid trace event files that match system and event */
2544 struct trace_event_file *
2545 find_event_file(struct trace_array *tr, const char *system, const char *event)
2546 {
2547         struct trace_event_file *file;
2548
2549         file = __find_event_file(tr, system, event);
2550         if (!file || !file->event_call->class->reg ||
2551             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2552                 return NULL;
2553
2554         return file;
2555 }
2556
2557 /**
2558  * trace_get_event_file - Find and return a trace event file
2559  * @instance: The name of the trace instance containing the event
2560  * @system: The name of the system containing the event
2561  * @event: The name of the event
2562  *
2563  * Return a trace event file given the trace instance name, trace
2564  * system, and trace event name.  If the instance name is NULL, it
2565  * refers to the top-level trace array.
2566  *
2567  * This function will look it up and return it if found, after calling
2568  * trace_array_get() to prevent the instance from going away, and
2569  * increment the event's module refcount to prevent it from being
2570  * removed.
2571  *
2572  * To release the file, call trace_put_event_file(), which will call
2573  * trace_array_put() and decrement the event's module refcount.
2574  *
2575  * Return: The trace event on success, ERR_PTR otherwise.
2576  */
2577 struct trace_event_file *trace_get_event_file(const char *instance,
2578                                               const char *system,
2579                                               const char *event)
2580 {
2581         struct trace_array *tr = top_trace_array();
2582         struct trace_event_file *file = NULL;
2583         int ret = -EINVAL;
2584
2585         if (instance) {
2586                 tr = trace_array_find_get(instance);
2587                 if (!tr)
2588                         return ERR_PTR(-ENOENT);
2589         } else {
2590                 ret = trace_array_get(tr);
2591                 if (ret)
2592                         return ERR_PTR(ret);
2593         }
2594
2595         mutex_lock(&event_mutex);
2596
2597         file = find_event_file(tr, system, event);
2598         if (!file) {
2599                 trace_array_put(tr);
2600                 ret = -EINVAL;
2601                 goto out;
2602         }
2603
2604         /* Don't let event modules unload while in use */
2605         ret = try_module_get(file->event_call->mod);
2606         if (!ret) {
2607                 trace_array_put(tr);
2608                 ret = -EBUSY;
2609                 goto out;
2610         }
2611
2612         ret = 0;
2613  out:
2614         mutex_unlock(&event_mutex);
2615
2616         if (ret)
2617                 file = ERR_PTR(ret);
2618
2619         return file;
2620 }
2621 EXPORT_SYMBOL_GPL(trace_get_event_file);
2622
2623 /**
2624  * trace_put_event_file - Release a file from trace_get_event_file()
2625  * @file: The trace event file
2626  *
2627  * If a file was retrieved using trace_get_event_file(), this should
2628  * be called when it's no longer needed.  It will cancel the previous
2629  * trace_array_get() called by that function, and decrement the
2630  * event's module refcount.
2631  */
2632 void trace_put_event_file(struct trace_event_file *file)
2633 {
2634         mutex_lock(&event_mutex);
2635         module_put(file->event_call->mod);
2636         mutex_unlock(&event_mutex);
2637
2638         trace_array_put(file->tr);
2639 }
2640 EXPORT_SYMBOL_GPL(trace_put_event_file);
2641
2642 #ifdef CONFIG_DYNAMIC_FTRACE
2643
2644 /* Avoid typos */
2645 #define ENABLE_EVENT_STR        "enable_event"
2646 #define DISABLE_EVENT_STR       "disable_event"
2647
2648 struct event_probe_data {
2649         struct trace_event_file *file;
2650         unsigned long                   count;
2651         int                             ref;
2652         bool                            enable;
2653 };
2654
2655 static void update_event_probe(struct event_probe_data *data)
2656 {
2657         if (data->enable)
2658                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2659         else
2660                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2661 }
2662
2663 static void
2664 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2665                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2666                    void *data)
2667 {
2668         struct ftrace_func_mapper *mapper = data;
2669         struct event_probe_data *edata;
2670         void **pdata;
2671
2672         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2673         if (!pdata || !*pdata)
2674                 return;
2675
2676         edata = *pdata;
2677         update_event_probe(edata);
2678 }
2679
2680 static void
2681 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2682                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2683                          void *data)
2684 {
2685         struct ftrace_func_mapper *mapper = data;
2686         struct event_probe_data *edata;
2687         void **pdata;
2688
2689         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2690         if (!pdata || !*pdata)
2691                 return;
2692
2693         edata = *pdata;
2694
2695         if (!edata->count)
2696                 return;
2697
2698         /* Skip if the event is in a state we want to switch to */
2699         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2700                 return;
2701
2702         if (edata->count != -1)
2703                 (edata->count)--;
2704
2705         update_event_probe(edata);
2706 }
2707
2708 static int
2709 event_enable_print(struct seq_file *m, unsigned long ip,
2710                    struct ftrace_probe_ops *ops, void *data)
2711 {
2712         struct ftrace_func_mapper *mapper = data;
2713         struct event_probe_data *edata;
2714         void **pdata;
2715
2716         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2717
2718         if (WARN_ON_ONCE(!pdata || !*pdata))
2719                 return 0;
2720
2721         edata = *pdata;
2722
2723         seq_printf(m, "%ps:", (void *)ip);
2724
2725         seq_printf(m, "%s:%s:%s",
2726                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2727                    edata->file->event_call->class->system,
2728                    trace_event_name(edata->file->event_call));
2729
2730         if (edata->count == -1)
2731                 seq_puts(m, ":unlimited\n");
2732         else
2733                 seq_printf(m, ":count=%ld\n", edata->count);
2734
2735         return 0;
2736 }
2737
2738 static int
2739 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2740                   unsigned long ip, void *init_data, void **data)
2741 {
2742         struct ftrace_func_mapper *mapper = *data;
2743         struct event_probe_data *edata = init_data;
2744         int ret;
2745
2746         if (!mapper) {
2747                 mapper = allocate_ftrace_func_mapper();
2748                 if (!mapper)
2749                         return -ENODEV;
2750                 *data = mapper;
2751         }
2752
2753         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2754         if (ret < 0)
2755                 return ret;
2756
2757         edata->ref++;
2758
2759         return 0;
2760 }
2761
2762 static int free_probe_data(void *data)
2763 {
2764         struct event_probe_data *edata = data;
2765
2766         edata->ref--;
2767         if (!edata->ref) {
2768                 /* Remove the SOFT_MODE flag */
2769                 __ftrace_event_enable_disable(edata->file, 0, 1);
2770                 module_put(edata->file->event_call->mod);
2771                 kfree(edata);
2772         }
2773         return 0;
2774 }
2775
2776 static void
2777 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2778                   unsigned long ip, void *data)
2779 {
2780         struct ftrace_func_mapper *mapper = data;
2781         struct event_probe_data *edata;
2782
2783         if (!ip) {
2784                 if (!mapper)
2785                         return;
2786                 free_ftrace_func_mapper(mapper, free_probe_data);
2787                 return;
2788         }
2789
2790         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2791
2792         if (WARN_ON_ONCE(!edata))
2793                 return;
2794
2795         if (WARN_ON_ONCE(edata->ref <= 0))
2796                 return;
2797
2798         free_probe_data(edata);
2799 }
2800
2801 static struct ftrace_probe_ops event_enable_probe_ops = {
2802         .func                   = event_enable_probe,
2803         .print                  = event_enable_print,
2804         .init                   = event_enable_init,
2805         .free                   = event_enable_free,
2806 };
2807
2808 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2809         .func                   = event_enable_count_probe,
2810         .print                  = event_enable_print,
2811         .init                   = event_enable_init,
2812         .free                   = event_enable_free,
2813 };
2814
2815 static struct ftrace_probe_ops event_disable_probe_ops = {
2816         .func                   = event_enable_probe,
2817         .print                  = event_enable_print,
2818         .init                   = event_enable_init,
2819         .free                   = event_enable_free,
2820 };
2821
2822 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2823         .func                   = event_enable_count_probe,
2824         .print                  = event_enable_print,
2825         .init                   = event_enable_init,
2826         .free                   = event_enable_free,
2827 };
2828
2829 static int
2830 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2831                   char *glob, char *cmd, char *param, int enabled)
2832 {
2833         struct trace_event_file *file;
2834         struct ftrace_probe_ops *ops;
2835         struct event_probe_data *data;
2836         const char *system;
2837         const char *event;
2838         char *number;
2839         bool enable;
2840         int ret;
2841
2842         if (!tr)
2843                 return -ENODEV;
2844
2845         /* hash funcs only work with set_ftrace_filter */
2846         if (!enabled || !param)
2847                 return -EINVAL;
2848
2849         system = strsep(&param, ":");
2850         if (!param)
2851                 return -EINVAL;
2852
2853         event = strsep(&param, ":");
2854
2855         mutex_lock(&event_mutex);
2856
2857         ret = -EINVAL;
2858         file = find_event_file(tr, system, event);
2859         if (!file)
2860                 goto out;
2861
2862         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2863
2864         if (enable)
2865                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2866         else
2867                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2868
2869         if (glob[0] == '!') {
2870                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2871                 goto out;
2872         }
2873
2874         ret = -ENOMEM;
2875
2876         data = kzalloc(sizeof(*data), GFP_KERNEL);
2877         if (!data)
2878                 goto out;
2879
2880         data->enable = enable;
2881         data->count = -1;
2882         data->file = file;
2883
2884         if (!param)
2885                 goto out_reg;
2886
2887         number = strsep(&param, ":");
2888
2889         ret = -EINVAL;
2890         if (!strlen(number))
2891                 goto out_free;
2892
2893         /*
2894          * We use the callback data field (which is a pointer)
2895          * as our counter.
2896          */
2897         ret = kstrtoul(number, 0, &data->count);
2898         if (ret)
2899                 goto out_free;
2900
2901  out_reg:
2902         /* Don't let event modules unload while probe registered */
2903         ret = try_module_get(file->event_call->mod);
2904         if (!ret) {
2905                 ret = -EBUSY;
2906                 goto out_free;
2907         }
2908
2909         ret = __ftrace_event_enable_disable(file, 1, 1);
2910         if (ret < 0)
2911                 goto out_put;
2912
2913         ret = register_ftrace_function_probe(glob, tr, ops, data);
2914         /*
2915          * The above returns on success the # of functions enabled,
2916          * but if it didn't find any functions it returns zero.
2917          * Consider no functions a failure too.
2918          */
2919         if (!ret) {
2920                 ret = -ENOENT;
2921                 goto out_disable;
2922         } else if (ret < 0)
2923                 goto out_disable;
2924         /* Just return zero, not the number of enabled functions */
2925         ret = 0;
2926  out:
2927         mutex_unlock(&event_mutex);
2928         return ret;
2929
2930  out_disable:
2931         __ftrace_event_enable_disable(file, 0, 1);
2932  out_put:
2933         module_put(file->event_call->mod);
2934  out_free:
2935         kfree(data);
2936         goto out;
2937 }
2938
2939 static struct ftrace_func_command event_enable_cmd = {
2940         .name                   = ENABLE_EVENT_STR,
2941         .func                   = event_enable_func,
2942 };
2943
2944 static struct ftrace_func_command event_disable_cmd = {
2945         .name                   = DISABLE_EVENT_STR,
2946         .func                   = event_enable_func,
2947 };
2948
2949 static __init int register_event_cmds(void)
2950 {
2951         int ret;
2952
2953         ret = register_ftrace_command(&event_enable_cmd);
2954         if (WARN_ON(ret < 0))
2955                 return ret;
2956         ret = register_ftrace_command(&event_disable_cmd);
2957         if (WARN_ON(ret < 0))
2958                 unregister_ftrace_command(&event_enable_cmd);
2959         return ret;
2960 }
2961 #else
2962 static inline int register_event_cmds(void) { return 0; }
2963 #endif /* CONFIG_DYNAMIC_FTRACE */
2964
2965 /*
2966  * The top level array has already had its trace_event_file
2967  * descriptors created in order to allow for early events to
2968  * be recorded. This function is called after the tracefs has been
2969  * initialized, and we now have to create the files associated
2970  * to the events.
2971  */
2972 static __init void
2973 __trace_early_add_event_dirs(struct trace_array *tr)
2974 {
2975         struct trace_event_file *file;
2976         int ret;
2977
2978
2979         list_for_each_entry(file, &tr->events, list) {
2980                 ret = event_create_dir(tr->event_dir, file);
2981                 if (ret < 0)
2982                         pr_warn("Could not create directory for event %s\n",
2983                                 trace_event_name(file->event_call));
2984         }
2985 }
2986
2987 /*
2988  * For early boot up, the top trace array requires to have
2989  * a list of events that can be enabled. This must be done before
2990  * the filesystem is set up in order to allow events to be traced
2991  * early.
2992  */
2993 static __init void
2994 __trace_early_add_events(struct trace_array *tr)
2995 {
2996         struct trace_event_call *call;
2997         int ret;
2998
2999         list_for_each_entry(call, &ftrace_events, list) {
3000                 /* Early boot up should not have any modules loaded */
3001                 if (WARN_ON_ONCE(call->mod))
3002                         continue;
3003
3004                 ret = __trace_early_add_new_event(call, tr);
3005                 if (ret < 0)
3006                         pr_warn("Could not create early event %s\n",
3007                                 trace_event_name(call));
3008         }
3009 }
3010
3011 /* Remove the event directory structure for a trace directory. */
3012 static void
3013 __trace_remove_event_dirs(struct trace_array *tr)
3014 {
3015         struct trace_event_file *file, *next;
3016
3017         list_for_each_entry_safe(file, next, &tr->events, list)
3018                 remove_event_file_dir(file);
3019 }
3020
3021 static void __add_event_to_tracers(struct trace_event_call *call)
3022 {
3023         struct trace_array *tr;
3024
3025         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3026                 __trace_add_new_event(call, tr);
3027 }
3028
3029 extern struct trace_event_call *__start_ftrace_events[];
3030 extern struct trace_event_call *__stop_ftrace_events[];
3031
3032 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3033
3034 static __init int setup_trace_event(char *str)
3035 {
3036         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3037         ring_buffer_expanded = true;
3038         tracing_selftest_disabled = true;
3039
3040         return 1;
3041 }
3042 __setup("trace_event=", setup_trace_event);
3043
3044 /* Expects to have event_mutex held when called */
3045 static int
3046 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3047 {
3048         struct dentry *d_events;
3049         struct dentry *entry;
3050
3051         entry = tracefs_create_file("set_event", 0644, parent,
3052                                     tr, &ftrace_set_event_fops);
3053         if (!entry) {
3054                 pr_warn("Could not create tracefs 'set_event' entry\n");
3055                 return -ENOMEM;
3056         }
3057
3058         d_events = tracefs_create_dir("events", parent);
3059         if (!d_events) {
3060                 pr_warn("Could not create tracefs 'events' directory\n");
3061                 return -ENOMEM;
3062         }
3063
3064         entry = trace_create_file("enable", 0644, d_events,
3065                                   tr, &ftrace_tr_enable_fops);
3066         if (!entry) {
3067                 pr_warn("Could not create tracefs 'enable' entry\n");
3068                 return -ENOMEM;
3069         }
3070
3071         /* There are not as crucial, just warn if they are not created */
3072
3073         entry = tracefs_create_file("set_event_pid", 0644, parent,
3074                                     tr, &ftrace_set_event_pid_fops);
3075         if (!entry)
3076                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3077
3078         /* ring buffer internal formats */
3079         entry = trace_create_file("header_page", 0444, d_events,
3080                                   ring_buffer_print_page_header,
3081                                   &ftrace_show_header_fops);
3082         if (!entry)
3083                 pr_warn("Could not create tracefs 'header_page' entry\n");
3084
3085         entry = trace_create_file("header_event", 0444, d_events,
3086                                   ring_buffer_print_entry_header,
3087                                   &ftrace_show_header_fops);
3088         if (!entry)
3089                 pr_warn("Could not create tracefs 'header_event' entry\n");
3090
3091         tr->event_dir = d_events;
3092
3093         return 0;
3094 }
3095
3096 /**
3097  * event_trace_add_tracer - add a instance of a trace_array to events
3098  * @parent: The parent dentry to place the files/directories for events in
3099  * @tr: The trace array associated with these events
3100  *
3101  * When a new instance is created, it needs to set up its events
3102  * directory, as well as other files associated with events. It also
3103  * creates the event hierachry in the @parent/events directory.
3104  *
3105  * Returns 0 on success.
3106  *
3107  * Must be called with event_mutex held.
3108  */
3109 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3110 {
3111         int ret;
3112
3113         lockdep_assert_held(&event_mutex);
3114
3115         ret = create_event_toplevel_files(parent, tr);
3116         if (ret)
3117                 goto out;
3118
3119         down_write(&trace_event_sem);
3120         __trace_add_event_dirs(tr);
3121         up_write(&trace_event_sem);
3122
3123  out:
3124         return ret;
3125 }
3126
3127 /*
3128  * The top trace array already had its file descriptors created.
3129  * Now the files themselves need to be created.
3130  */
3131 static __init int
3132 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3133 {
3134         int ret;
3135
3136         mutex_lock(&event_mutex);
3137
3138         ret = create_event_toplevel_files(parent, tr);
3139         if (ret)
3140                 goto out_unlock;
3141
3142         down_write(&trace_event_sem);
3143         __trace_early_add_event_dirs(tr);
3144         up_write(&trace_event_sem);
3145
3146  out_unlock:
3147         mutex_unlock(&event_mutex);
3148
3149         return ret;
3150 }
3151
3152 /* Must be called with event_mutex held */
3153 int event_trace_del_tracer(struct trace_array *tr)
3154 {
3155         lockdep_assert_held(&event_mutex);
3156
3157         /* Disable any event triggers and associated soft-disabled events */
3158         clear_event_triggers(tr);
3159
3160         /* Clear the pid list */
3161         __ftrace_clear_event_pids(tr);
3162
3163         /* Disable any running events */
3164         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3165
3166         /* Make sure no more events are being executed */
3167         tracepoint_synchronize_unregister();
3168
3169         down_write(&trace_event_sem);
3170         __trace_remove_event_dirs(tr);
3171         tracefs_remove(tr->event_dir);
3172         up_write(&trace_event_sem);
3173
3174         tr->event_dir = NULL;
3175
3176         return 0;
3177 }
3178
3179 static __init int event_trace_memsetup(void)
3180 {
3181         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3182         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3183         return 0;
3184 }
3185
3186 static __init void
3187 early_enable_events(struct trace_array *tr, bool disable_first)
3188 {
3189         char *buf = bootup_event_buf;
3190         char *token;
3191         int ret;
3192
3193         while (true) {
3194                 token = strsep(&buf, ",");
3195
3196                 if (!token)
3197                         break;
3198
3199                 if (*token) {
3200                         /* Restarting syscalls requires that we stop them first */
3201                         if (disable_first)
3202                                 ftrace_set_clr_event(tr, token, 0);
3203
3204                         ret = ftrace_set_clr_event(tr, token, 1);
3205                         if (ret)
3206                                 pr_warn("Failed to enable trace event: %s\n", token);
3207                 }
3208
3209                 /* Put back the comma to allow this to be called again */
3210                 if (buf)
3211                         *(buf - 1) = ',';
3212         }
3213 }
3214
3215 static __init int event_trace_enable(void)
3216 {
3217         struct trace_array *tr = top_trace_array();
3218         struct trace_event_call **iter, *call;
3219         int ret;
3220
3221         if (!tr)
3222                 return -ENODEV;
3223
3224         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3225
3226                 call = *iter;
3227                 ret = event_init(call);
3228                 if (!ret)
3229                         list_add(&call->list, &ftrace_events);
3230         }
3231
3232         /*
3233          * We need the top trace array to have a working set of trace
3234          * points at early init, before the debug files and directories
3235          * are created. Create the file entries now, and attach them
3236          * to the actual file dentries later.
3237          */
3238         __trace_early_add_events(tr);
3239
3240         early_enable_events(tr, false);
3241
3242         trace_printk_start_comm();
3243
3244         register_event_cmds();
3245
3246         register_trigger_cmds();
3247
3248         return 0;
3249 }
3250
3251 /*
3252  * event_trace_enable() is called from trace_event_init() first to
3253  * initialize events and perhaps start any events that are on the
3254  * command line. Unfortunately, there are some events that will not
3255  * start this early, like the system call tracepoints that need
3256  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3257  * is called before pid 1 starts, and this flag is never set, making
3258  * the syscall tracepoint never get reached, but the event is enabled
3259  * regardless (and not doing anything).
3260  */
3261 static __init int event_trace_enable_again(void)
3262 {
3263         struct trace_array *tr;
3264
3265         tr = top_trace_array();
3266         if (!tr)
3267                 return -ENODEV;
3268
3269         early_enable_events(tr, true);
3270
3271         return 0;
3272 }
3273
3274 early_initcall(event_trace_enable_again);
3275
3276 __init int event_trace_init(void)
3277 {
3278         struct trace_array *tr;
3279         struct dentry *d_tracer;
3280         struct dentry *entry;
3281         int ret;
3282
3283         tr = top_trace_array();
3284         if (!tr)
3285                 return -ENODEV;
3286
3287         d_tracer = tracing_init_dentry();
3288         if (IS_ERR(d_tracer))
3289                 return 0;
3290
3291         entry = tracefs_create_file("available_events", 0444, d_tracer,
3292                                     tr, &ftrace_avail_fops);
3293         if (!entry)
3294                 pr_warn("Could not create tracefs 'available_events' entry\n");
3295
3296         if (trace_define_generic_fields())
3297                 pr_warn("tracing: Failed to allocated generic fields");
3298
3299         if (trace_define_common_fields())
3300                 pr_warn("tracing: Failed to allocate common fields");
3301
3302         ret = early_event_add_tracer(d_tracer, tr);
3303         if (ret)
3304                 return ret;
3305
3306 #ifdef CONFIG_MODULES
3307         ret = register_module_notifier(&trace_module_nb);
3308         if (ret)
3309                 pr_warn("Failed to register trace events module notifier\n");
3310 #endif
3311         return 0;
3312 }
3313
3314 void __init trace_event_init(void)
3315 {
3316         event_trace_memsetup();
3317         init_ftrace_syscalls();
3318         event_trace_enable();
3319 }
3320
3321 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3322
3323 static DEFINE_SPINLOCK(test_spinlock);
3324 static DEFINE_SPINLOCK(test_spinlock_irq);
3325 static DEFINE_MUTEX(test_mutex);
3326
3327 static __init void test_work(struct work_struct *dummy)
3328 {
3329         spin_lock(&test_spinlock);
3330         spin_lock_irq(&test_spinlock_irq);
3331         udelay(1);
3332         spin_unlock_irq(&test_spinlock_irq);
3333         spin_unlock(&test_spinlock);
3334
3335         mutex_lock(&test_mutex);
3336         msleep(1);
3337         mutex_unlock(&test_mutex);
3338 }
3339
3340 static __init int event_test_thread(void *unused)
3341 {
3342         void *test_malloc;
3343
3344         test_malloc = kmalloc(1234, GFP_KERNEL);
3345         if (!test_malloc)
3346                 pr_info("failed to kmalloc\n");
3347
3348         schedule_on_each_cpu(test_work);
3349
3350         kfree(test_malloc);
3351
3352         set_current_state(TASK_INTERRUPTIBLE);
3353         while (!kthread_should_stop()) {
3354                 schedule();
3355                 set_current_state(TASK_INTERRUPTIBLE);
3356         }
3357         __set_current_state(TASK_RUNNING);
3358
3359         return 0;
3360 }
3361
3362 /*
3363  * Do various things that may trigger events.
3364  */
3365 static __init void event_test_stuff(void)
3366 {
3367         struct task_struct *test_thread;
3368
3369         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3370         msleep(1);
3371         kthread_stop(test_thread);
3372 }
3373
3374 /*
3375  * For every trace event defined, we will test each trace point separately,
3376  * and then by groups, and finally all trace points.
3377  */
3378 static __init void event_trace_self_tests(void)
3379 {
3380         struct trace_subsystem_dir *dir;
3381         struct trace_event_file *file;
3382         struct trace_event_call *call;
3383         struct event_subsystem *system;
3384         struct trace_array *tr;
3385         int ret;
3386
3387         tr = top_trace_array();
3388         if (!tr)
3389                 return;
3390
3391         pr_info("Running tests on trace events:\n");
3392
3393         list_for_each_entry(file, &tr->events, list) {
3394
3395                 call = file->event_call;
3396
3397                 /* Only test those that have a probe */
3398                 if (!call->class || !call->class->probe)
3399                         continue;
3400
3401 /*
3402  * Testing syscall events here is pretty useless, but
3403  * we still do it if configured. But this is time consuming.
3404  * What we really need is a user thread to perform the
3405  * syscalls as we test.
3406  */
3407 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3408                 if (call->class->system &&
3409                     strcmp(call->class->system, "syscalls") == 0)
3410                         continue;
3411 #endif
3412
3413                 pr_info("Testing event %s: ", trace_event_name(call));
3414
3415                 /*
3416                  * If an event is already enabled, someone is using
3417                  * it and the self test should not be on.
3418                  */
3419                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3420                         pr_warn("Enabled event during self test!\n");
3421                         WARN_ON_ONCE(1);
3422                         continue;
3423                 }
3424
3425                 ftrace_event_enable_disable(file, 1);
3426                 event_test_stuff();
3427                 ftrace_event_enable_disable(file, 0);
3428
3429                 pr_cont("OK\n");
3430         }
3431
3432         /* Now test at the sub system level */
3433
3434         pr_info("Running tests on trace event systems:\n");
3435
3436         list_for_each_entry(dir, &tr->systems, list) {
3437
3438                 system = dir->subsystem;
3439
3440                 /* the ftrace system is special, skip it */
3441                 if (strcmp(system->name, "ftrace") == 0)
3442                         continue;
3443
3444                 pr_info("Testing event system %s: ", system->name);
3445
3446                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3447                 if (WARN_ON_ONCE(ret)) {
3448                         pr_warn("error enabling system %s\n",
3449                                 system->name);
3450                         continue;
3451                 }
3452
3453                 event_test_stuff();
3454
3455                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3456                 if (WARN_ON_ONCE(ret)) {
3457                         pr_warn("error disabling system %s\n",
3458                                 system->name);
3459                         continue;
3460                 }
3461
3462                 pr_cont("OK\n");
3463         }
3464
3465         /* Test with all events enabled */
3466
3467         pr_info("Running tests on all trace events:\n");
3468         pr_info("Testing all events: ");
3469
3470         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3471         if (WARN_ON_ONCE(ret)) {
3472                 pr_warn("error enabling all events\n");
3473                 return;
3474         }
3475
3476         event_test_stuff();
3477
3478         /* reset sysname */
3479         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3480         if (WARN_ON_ONCE(ret)) {
3481                 pr_warn("error disabling all events\n");
3482                 return;
3483         }
3484
3485         pr_cont("OK\n");
3486 }
3487
3488 #ifdef CONFIG_FUNCTION_TRACER
3489
3490 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3491
3492 static struct trace_event_file event_trace_file __initdata;
3493
3494 static void __init
3495 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3496                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3497 {
3498         struct trace_buffer *buffer;
3499         struct ring_buffer_event *event;
3500         struct ftrace_entry *entry;
3501         unsigned long flags;
3502         long disabled;
3503         int cpu;
3504         int pc;
3505
3506         pc = preempt_count();
3507         preempt_disable_notrace();
3508         cpu = raw_smp_processor_id();
3509         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3510
3511         if (disabled != 1)
3512                 goto out;
3513
3514         local_save_flags(flags);
3515
3516         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3517                                                 TRACE_FN, sizeof(*entry),
3518                                                 flags, pc);
3519         if (!event)
3520                 goto out;
3521         entry   = ring_buffer_event_data(event);
3522         entry->ip                       = ip;
3523         entry->parent_ip                = parent_ip;
3524
3525         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3526                                     entry, flags, pc);
3527  out:
3528         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3529         preempt_enable_notrace();
3530 }
3531
3532 static struct ftrace_ops trace_ops __initdata  =
3533 {
3534         .func = function_test_events_call,
3535         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3536 };
3537
3538 static __init void event_trace_self_test_with_function(void)
3539 {
3540         int ret;
3541
3542         event_trace_file.tr = top_trace_array();
3543         if (WARN_ON(!event_trace_file.tr))
3544                 return;
3545
3546         ret = register_ftrace_function(&trace_ops);
3547         if (WARN_ON(ret < 0)) {
3548                 pr_info("Failed to enable function tracer for event tests\n");
3549                 return;
3550         }
3551         pr_info("Running tests again, along with the function tracer\n");
3552         event_trace_self_tests();
3553         unregister_ftrace_function(&trace_ops);
3554 }
3555 #else
3556 static __init void event_trace_self_test_with_function(void)
3557 {
3558 }
3559 #endif
3560
3561 static __init int event_trace_self_tests_init(void)
3562 {
3563         if (!tracing_selftest_disabled) {
3564                 event_trace_self_tests();
3565                 event_trace_self_test_with_function();
3566         }
3567
3568         return 0;
3569 }
3570
3571 late_initcall(event_trace_self_tests_init);
3572
3573 #endif