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