kernel/trace/trace_events.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * event tracer
   4  *
   5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6  *
   7  *  - Added format output of fields of the trace point.
   8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9  *
  10  */
  11
  12 #define pr_fmt(fmt) fmt
  13
  14 #include <linux/workqueue.h>
  15 #include <linux/security.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/kthread.h>
  18 #include <linux/tracefs.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/module.h>
  21 #include <linux/ctype.h>
  22 #include <linux/sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/delay.h>
  25
  26 #include <trace/events/sched.h>
  27 #include <trace/syscall.h>
  28
  29 #include <asm/setup.h>
  30
  31 #include "trace_output.h"
  32
  33 #undef TRACE_SYSTEM
  34 #define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36 DEFINE_MUTEX(event_mutex);
  37
  38 LIST_HEAD(ftrace_events);
  39 static LIST_HEAD(ftrace_generic_fields);
  40 static LIST_HEAD(ftrace_common_fields);
  41 static bool eventdir_initialized;
  42
  43 static LIST_HEAD(module_strings);
  44
  45 struct module_string {
  46         struct list_head        next;
  47         struct module           *module;
  48         char                    *str;
  49 };
  50
  51 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  52
  53 static struct kmem_cache *field_cachep;
  54 static struct kmem_cache *file_cachep;
  55
  56 static inline int system_refcount(struct event_subsystem *system)
  57 {
  58         return system->ref_count;
  59 }
  60
  61 static int system_refcount_inc(struct event_subsystem *system)
  62 {
  63         return system->ref_count++;
  64 }
  65
  66 static int system_refcount_dec(struct event_subsystem *system)
  67 {
  68         return --system->ref_count;
  69 }
  70
  71 /* Double loops, do not use break, only goto's work */
  72 #define do_for_each_event_file(tr, file)                        \
  73         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  74                 list_for_each_entry(file, &tr->events, list)
  75
  76 #define do_for_each_event_file_safe(tr, file)                   \
  77         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  78                 struct trace_event_file *___n;                          \
  79                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  80
  81 #define while_for_each_event_file()             \
  82         }
  83
  84 static struct ftrace_event_field *
  85 __find_event_field(struct list_head *head, char *name)
  86 {
  87         struct ftrace_event_field *field;
  88
  89         list_for_each_entry(field, head, link) {
  90                 if (!strcmp(field->name, name))
  91                         return field;
  92         }
  93
  94         return NULL;
  95 }
  96
  97 struct ftrace_event_field *
  98 trace_find_event_field(struct trace_event_call *call, char *name)
  99 {
 100         struct ftrace_event_field *field;
 101         struct list_head *head;
 102
 103         head = trace_get_fields(call);
 104         field = __find_event_field(head, name);
 105         if (field)
 106                 return field;
 107
 108         field = __find_event_field(&ftrace_generic_fields, name);
 109         if (field)
 110                 return field;
 111
 112         return __find_event_field(&ftrace_common_fields, name);
 113 }
 114
 115 static int __trace_define_field(struct list_head *head, const char *type,
 116                                 const char *name, int offset, int size,
 117                                 int is_signed, int filter_type, int len)
 118 {
 119         struct ftrace_event_field *field;
 120
 121         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 122         if (!field)
 123                 return -ENOMEM;
 124
 125         field->name = name;
 126         field->type = type;
 127
 128         if (filter_type == FILTER_OTHER)
 129                 field->filter_type = filter_assign_type(type);
 130         else
 131                 field->filter_type = filter_type;
 132
 133         field->offset = offset;
 134         field->size = size;
 135         field->is_signed = is_signed;
 136         field->len = len;
 137
 138         list_add(&field->link, head);
 139
 140         return 0;
 141 }
 142
 143 int trace_define_field(struct trace_event_call *call, const char *type,
 144                        const char *name, int offset, int size, int is_signed,
 145                        int filter_type)
 146 {
 147         struct list_head *head;
 148
 149         if (WARN_ON(!call->class))
 150                 return 0;
 151
 152         head = trace_get_fields(call);
 153         return __trace_define_field(head, type, name, offset, size,
 154                                     is_signed, filter_type, 0);
 155 }
 156 EXPORT_SYMBOL_GPL(trace_define_field);
 157
 158 static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 159                        const char *name, int offset, int size, int is_signed,
 160                        int filter_type, int len)
 161 {
 162         struct list_head *head;
 163
 164         if (WARN_ON(!call->class))
 165                 return 0;
 166
 167         head = trace_get_fields(call);
 168         return __trace_define_field(head, type, name, offset, size,
 169                                     is_signed, filter_type, len);
 170 }
 171
 172 #define __generic_field(type, item, filter_type)                        \
 173         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 174                                    #item, 0, 0, is_signed_type(type),   \
 175                                    filter_type, 0);                     \
 176         if (ret)                                                        \
 177                 return ret;
 178
 179 #define __common_field(type, item)                                      \
 180         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 181                                    "common_" #item,                     \
 182                                    offsetof(typeof(ent), item),         \
 183                                    sizeof(ent.item),                    \
 184                                    is_signed_type(type), FILTER_OTHER, 0);      \
 185         if (ret)                                                        \
 186                 return ret;
 187
 188 static int trace_define_generic_fields(void)
 189 {
 190         int ret;
 191
 192         __generic_field(int, CPU, FILTER_CPU);
 193         __generic_field(int, cpu, FILTER_CPU);
 194         __generic_field(int, common_cpu, FILTER_CPU);
 195         __generic_field(char *, COMM, FILTER_COMM);
 196         __generic_field(char *, comm, FILTER_COMM);
 197         __generic_field(char *, stacktrace, FILTER_STACKTRACE);
 198         __generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 199
 200         return ret;
 201 }
 202
 203 static int trace_define_common_fields(void)
 204 {
 205         int ret;
 206         struct trace_entry ent;
 207
 208         __common_field(unsigned short, type);
 209         __common_field(unsigned char, flags);
 210         /* Holds both preempt_count and migrate_disable */
 211         __common_field(unsigned char, preempt_count);
 212         __common_field(int, pid);
 213
 214         return ret;
 215 }
 216
 217 static void trace_destroy_fields(struct trace_event_call *call)
 218 {
 219         struct ftrace_event_field *field, *next;
 220         struct list_head *head;
 221
 222         head = trace_get_fields(call);
 223         list_for_each_entry_safe(field, next, head, link) {
 224                 list_del(&field->link);
 225                 kmem_cache_free(field_cachep, field);
 226         }
 227 }
 228
 229 /*
 230  * run-time version of trace_event_get_offsets_<call>() that returns the last
 231  * accessible offset of trace fields excluding __dynamic_array bytes
 232  */
 233 int trace_event_get_offsets(struct trace_event_call *call)
 234 {
 235         struct ftrace_event_field *tail;
 236         struct list_head *head;
 237
 238         head = trace_get_fields(call);
 239         /*
 240          * head->next points to the last field with the largest offset,
 241          * since it was added last by trace_define_field()
 242          */
 243         tail = list_first_entry(head, struct ftrace_event_field, link);
 244         return tail->offset + tail->size;
 245 }
 246
 247 /*
 248  * Check if the referenced field is an array and return true,
 249  * as arrays are OK to dereference.
 250  */
 251 static bool test_field(const char *fmt, struct trace_event_call *call)
 252 {
 253         struct trace_event_fields *field = call->class->fields_array;
 254         const char *array_descriptor;
 255         const char *p = fmt;
 256         int len;
 257
 258         if (!(len = str_has_prefix(fmt, "REC->")))
 259                 return false;
 260         fmt += len;
 261         for (p = fmt; *p; p++) {
 262                 if (!isalnum(*p) && *p != '_')
 263                         break;
 264         }
 265         len = p - fmt;
 266
 267         for (; field->type; field++) {
 268                 if (strncmp(field->name, fmt, len) ||
 269                     field->name[len])
 270                         continue;
 271                 array_descriptor = strchr(field->type, '[');
 272                 /* This is an array and is OK to dereference. */
 273                 return array_descriptor != NULL;
 274         }
 275         return false;
 276 }
 277
 278 /*
 279  * Examine the print fmt of the event looking for unsafe dereference
 280  * pointers using %p* that could be recorded in the trace event and
 281  * much later referenced after the pointer was freed. Dereferencing
 282  * pointers are OK, if it is dereferenced into the event itself.
 283  */
 284 static void test_event_printk(struct trace_event_call *call)
 285 {
 286         u64 dereference_flags = 0;
 287         bool first = true;
 288         const char *fmt, *c, *r, *a;
 289         int parens = 0;
 290         char in_quote = 0;
 291         int start_arg = 0;
 292         int arg = 0;
 293         int i;
 294
 295         fmt = call->print_fmt;
 296
 297         if (!fmt)
 298                 return;
 299
 300         for (i = 0; fmt[i]; i++) {
 301                 switch (fmt[i]) {
 302                 case '\\':
 303                         i++;
 304                         if (!fmt[i])
 305                                 return;
 306                         continue;
 307                 case '"':
 308                 case '\'':
 309                         /*
 310                          * The print fmt starts with a string that
 311                          * is processed first to find %p* usage,
 312                          * then after the first string, the print fmt
 313                          * contains arguments that are used to check
 314                          * if the dereferenced %p* usage is safe.
 315                          */
 316                         if (first) {
 317                                 if (fmt[i] == '\'')
 318                                         continue;
 319                                 if (in_quote) {
 320                                         arg = 0;
 321                                         first = false;
 322                                         /*
 323                                          * If there was no %p* uses
 324                                          * the fmt is OK.
 325                                          */
 326                                         if (!dereference_flags)
 327                                                 return;
 328                                 }
 329                         }
 330                         if (in_quote) {
 331                                 if (in_quote == fmt[i])
 332                                         in_quote = 0;
 333                         } else {
 334                                 in_quote = fmt[i];
 335                         }
 336                         continue;
 337                 case '%':
 338                         if (!first || !in_quote)
 339                                 continue;
 340                         i++;
 341                         if (!fmt[i])
 342                                 return;
 343                         switch (fmt[i]) {
 344                         case '%':
 345                                 continue;
 346                         case 'p':
 347                                 /* Find dereferencing fields */
 348                                 switch (fmt[i + 1]) {
 349                                 case 'B': case 'R': case 'r':
 350                                 case 'b': case 'M': case 'm':
 351                                 case 'I': case 'i': case 'E':
 352                                 case 'U': case 'V': case 'N':
 353                                 case 'a': case 'd': case 'D':
 354                                 case 'g': case 't': case 'C':
 355                                 case 'O': case 'f':
 356                                         if (WARN_ONCE(arg == 63,
 357                                                       "Too many args for event: %s",
 358                                                       trace_event_name(call)))
 359                                                 return;
 360                                         dereference_flags |= 1ULL << arg;
 361                                 }
 362                                 break;
 363                         default:
 364                         {
 365                                 bool star = false;
 366                                 int j;
 367
 368                                 /* Increment arg if %*s exists. */
 369                                 for (j = 0; fmt[i + j]; j++) {
 370                                         if (isdigit(fmt[i + j]) ||
 371                                             fmt[i + j] == '.')
 372                                                 continue;
 373                                         if (fmt[i + j] == '*') {
 374                                                 star = true;
 375                                                 continue;
 376                                         }
 377                                         if ((fmt[i + j] == 's') && star)
 378                                                 arg++;
 379                                         break;
 380                                 }
 381                                 break;
 382                         } /* default */
 383
 384                         } /* switch */
 385                         arg++;
 386                         continue;
 387                 case '(':
 388                         if (in_quote)
 389                                 continue;
 390                         parens++;
 391                         continue;
 392                 case ')':
 393                         if (in_quote)
 394                                 continue;
 395                         parens--;
 396                         if (WARN_ONCE(parens < 0,
 397                                       "Paren mismatch for event: %s\narg='%s'\n%*s",
 398                                       trace_event_name(call),
 399                                       fmt + start_arg,
 400                                       (i - start_arg) + 5, "^"))
 401                                 return;
 402                         continue;
 403                 case ',':
 404                         if (in_quote || parens)
 405                                 continue;
 406                         i++;
 407                         while (isspace(fmt[i]))
 408                                 i++;
 409                         start_arg = i;
 410                         if (!(dereference_flags & (1ULL << arg)))
 411                                 goto next_arg;
 412
 413                         /* Find the REC-> in the argument */
 414                         c = strchr(fmt + i, ',');
 415                         r = strstr(fmt + i, "REC->");
 416                         if (r && (!c || r < c)) {
 417                                 /*
 418                                  * Addresses of events on the buffer,
 419                                  * or an array on the buffer is
 420                                  * OK to dereference.
 421                                  * There's ways to fool this, but
 422                                  * this is to catch common mistakes,
 423                                  * not malicious code.
 424                                  */
 425                                 a = strchr(fmt + i, '&');
 426                                 if ((a && (a < r)) || test_field(r, call))
 427                                         dereference_flags &= ~(1ULL << arg);
 428                         } else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
 429                                    (!c || r < c)) {
 430                                 dereference_flags &= ~(1ULL << arg);
 431                         } else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
 432                                    (!c || r < c)) {
 433                                 dereference_flags &= ~(1ULL << arg);
 434                         }
 435
 436                 next_arg:
 437                         i--;
 438                         arg++;
 439                 }
 440         }
 441
 442         /*
 443          * If you triggered the below warning, the trace event reported
 444          * uses an unsafe dereference pointer %p*. As the data stored
 445          * at the trace event time may no longer exist when the trace
 446          * event is printed, dereferencing to the original source is
 447          * unsafe. The source of the dereference must be copied into the
 448          * event itself, and the dereference must access the copy instead.
 449          */
 450         if (WARN_ON_ONCE(dereference_flags)) {
 451                 arg = 1;
 452                 while (!(dereference_flags & 1)) {
 453                         dereference_flags >>= 1;
 454                         arg++;
 455                 }
 456                 pr_warn("event %s has unsafe dereference of argument %d\n",
 457                         trace_event_name(call), arg);
 458                 pr_warn("print_fmt: %s\n", fmt);
 459         }
 460 }
 461
 462 int trace_event_raw_init(struct trace_event_call *call)
 463 {
 464         int id;
 465
 466         id = register_trace_event(&call->event);
 467         if (!id)
 468                 return -ENODEV;
 469
 470         test_event_printk(call);
 471
 472         return 0;
 473 }
 474 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 475
 476 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 477 {
 478         struct trace_array *tr = trace_file->tr;
 479         struct trace_array_cpu *data;
 480         struct trace_pid_list *no_pid_list;
 481         struct trace_pid_list *pid_list;
 482
 483         pid_list = rcu_dereference_raw(tr->filtered_pids);
 484         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 485
 486         if (!pid_list && !no_pid_list)
 487                 return false;
 488
 489         data = this_cpu_ptr(tr->array_buffer.data);
 490
 491         return data->ignore_pid;
 492 }
 493 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 494
 495 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 496                                  struct trace_event_file *trace_file,
 497                                  unsigned long len)
 498 {
 499         struct trace_event_call *event_call = trace_file->event_call;
 500
 501         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 502             trace_event_ignore_this_pid(trace_file))
 503                 return NULL;
 504
 505         /*
 506          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 507          * preemption (adding one to the preempt_count). Since we are
 508          * interested in the preempt_count at the time the tracepoint was
 509          * hit, we need to subtract one to offset the increment.
 510          */
 511         fbuffer->trace_ctx = tracing_gen_ctx_dec();
 512         fbuffer->trace_file = trace_file;
 513
 514         fbuffer->event =
 515                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 516                                                 event_call->event.type, len,
 517                                                 fbuffer->trace_ctx);
 518         if (!fbuffer->event)
 519                 return NULL;
 520
 521         fbuffer->regs = NULL;
 522         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 523         return fbuffer->entry;
 524 }
 525 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 526
 527 int trace_event_reg(struct trace_event_call *call,
 528                     enum trace_reg type, void *data)
 529 {
 530         struct trace_event_file *file = data;
 531
 532         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 533         switch (type) {
 534         case TRACE_REG_REGISTER:
 535                 return tracepoint_probe_register(call->tp,
 536                                                  call->class->probe,
 537                                                  file);
 538         case TRACE_REG_UNREGISTER:
 539                 tracepoint_probe_unregister(call->tp,
 540                                             call->class->probe,
 541                                             file);
 542                 return 0;
 543
 544 #ifdef CONFIG_PERF_EVENTS
 545         case TRACE_REG_PERF_REGISTER:
 546                 return tracepoint_probe_register(call->tp,
 547                                                  call->class->perf_probe,
 548                                                  call);
 549         case TRACE_REG_PERF_UNREGISTER:
 550                 tracepoint_probe_unregister(call->tp,
 551                                             call->class->perf_probe,
 552                                             call);
 553                 return 0;
 554         case TRACE_REG_PERF_OPEN:
 555         case TRACE_REG_PERF_CLOSE:
 556         case TRACE_REG_PERF_ADD:
 557         case TRACE_REG_PERF_DEL:
 558                 return 0;
 559 #endif
 560         }
 561         return 0;
 562 }
 563 EXPORT_SYMBOL_GPL(trace_event_reg);
 564
 565 void trace_event_enable_cmd_record(bool enable)
 566 {
 567         struct trace_event_file *file;
 568         struct trace_array *tr;
 569
 570         lockdep_assert_held(&event_mutex);
 571
 572         do_for_each_event_file(tr, file) {
 573
 574                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 575                         continue;
 576
 577                 if (enable) {
 578                         tracing_start_cmdline_record();
 579                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 580                 } else {
 581                         tracing_stop_cmdline_record();
 582                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 583                 }
 584         } while_for_each_event_file();
 585 }
 586
 587 void trace_event_enable_tgid_record(bool enable)
 588 {
 589         struct trace_event_file *file;
 590         struct trace_array *tr;
 591
 592         lockdep_assert_held(&event_mutex);
 593
 594         do_for_each_event_file(tr, file) {
 595                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 596                         continue;
 597
 598                 if (enable) {
 599                         tracing_start_tgid_record();
 600                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 601                 } else {
 602                         tracing_stop_tgid_record();
 603                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 604                                   &file->flags);
 605                 }
 606         } while_for_each_event_file();
 607 }
 608
 609 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 610                                          int enable, int soft_disable)
 611 {
 612         struct trace_event_call *call = file->event_call;
 613         struct trace_array *tr = file->tr;
 614         int ret = 0;
 615         int disable;
 616
 617         switch (enable) {
 618         case 0:
 619                 /*
 620                  * When soft_disable is set and enable is cleared, the sm_ref
 621                  * reference counter is decremented. If it reaches 0, we want
 622                  * to clear the SOFT_DISABLED flag but leave the event in the
 623                  * state that it was. That is, if the event was enabled and
 624                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 625                  * is set we do not want the event to be enabled before we
 626                  * clear the bit.
 627                  *
 628                  * When soft_disable is not set but the SOFT_MODE flag is,
 629                  * we do nothing. Do not disable the tracepoint, otherwise
 630                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 631                  */
 632                 if (soft_disable) {
 633                         if (atomic_dec_return(&file->sm_ref) > 0)
 634                                 break;
 635                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 636                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 637                         /* Disable use of trace_buffered_event */
 638                         trace_buffered_event_disable();
 639                 } else
 640                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 641
 642                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 643                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 644                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 645                                 tracing_stop_cmdline_record();
 646                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 647                         }
 648
 649                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 650                                 tracing_stop_tgid_record();
 651                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 652                         }
 653
 654                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 655                 }
 656                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 657                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 658                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 659                 else
 660                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 661                 break;
 662         case 1:
 663                 /*
 664                  * When soft_disable is set and enable is set, we want to
 665                  * register the tracepoint for the event, but leave the event
 666                  * as is. That means, if the event was already enabled, we do
 667                  * nothing (but set SOFT_MODE). If the event is disabled, we
 668                  * set SOFT_DISABLED before enabling the event tracepoint, so
 669                  * it still seems to be disabled.
 670                  */
 671                 if (!soft_disable)
 672                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 673                 else {
 674                         if (atomic_inc_return(&file->sm_ref) > 1)
 675                                 break;
 676                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 677                         /* Enable use of trace_buffered_event */
 678                         trace_buffered_event_enable();
 679                 }
 680
 681                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 682                         bool cmd = false, tgid = false;
 683
 684                         /* Keep the event disabled, when going to SOFT_MODE. */
 685                         if (soft_disable)
 686                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 687
 688                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 689                                 cmd = true;
 690                                 tracing_start_cmdline_record();
 691                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 692                         }
 693
 694                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 695                                 tgid = true;
 696                                 tracing_start_tgid_record();
 697                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 698                         }
 699
 700                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 701                         if (ret) {
 702                                 if (cmd)
 703                                         tracing_stop_cmdline_record();
 704                                 if (tgid)
 705                                         tracing_stop_tgid_record();
 706                                 pr_info("event trace: Could not enable event "
 707                                         "%s\n", trace_event_name(call));
 708                                 break;
 709                         }
 710                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 711
 712                         /* WAS_ENABLED gets set but never cleared. */
 713                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 714                 }
 715                 break;
 716         }
 717
 718         return ret;
 719 }
 720
 721 int trace_event_enable_disable(struct trace_event_file *file,
 722                                int enable, int soft_disable)
 723 {
 724         return __ftrace_event_enable_disable(file, enable, soft_disable);
 725 }
 726
 727 static int ftrace_event_enable_disable(struct trace_event_file *file,
 728                                        int enable)
 729 {
 730         return __ftrace_event_enable_disable(file, enable, 0);
 731 }
 732
 733 static void ftrace_clear_events(struct trace_array *tr)
 734 {
 735         struct trace_event_file *file;
 736
 737         mutex_lock(&event_mutex);
 738         list_for_each_entry(file, &tr->events, list) {
 739                 ftrace_event_enable_disable(file, 0);
 740         }
 741         mutex_unlock(&event_mutex);
 742 }
 743
 744 static void
 745 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 746 {
 747         struct trace_pid_list *pid_list;
 748         struct trace_array *tr = data;
 749
 750         pid_list = rcu_dereference_raw(tr->filtered_pids);
 751         trace_filter_add_remove_task(pid_list, NULL, task);
 752
 753         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 754         trace_filter_add_remove_task(pid_list, NULL, task);
 755 }
 756
 757 static void
 758 event_filter_pid_sched_process_fork(void *data,
 759                                     struct task_struct *self,
 760                                     struct task_struct *task)
 761 {
 762         struct trace_pid_list *pid_list;
 763         struct trace_array *tr = data;
 764
 765         pid_list = rcu_dereference_sched(tr->filtered_pids);
 766         trace_filter_add_remove_task(pid_list, self, task);
 767
 768         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 769         trace_filter_add_remove_task(pid_list, self, task);
 770 }
 771
 772 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 773 {
 774         if (enable) {
 775                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 776                                                        tr, INT_MIN);
 777                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 778                                                        tr, INT_MAX);
 779         } else {
 780                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 781                                                     tr);
 782                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 783                                                     tr);
 784         }
 785 }
 786
 787 static void
 788 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 789                                         struct task_struct *prev,
 790                                         struct task_struct *next,
 791                                         unsigned int prev_state)
 792 {
 793         struct trace_array *tr = data;
 794         struct trace_pid_list *no_pid_list;
 795         struct trace_pid_list *pid_list;
 796         bool ret;
 797
 798         pid_list = rcu_dereference_sched(tr->filtered_pids);
 799         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 800
 801         /*
 802          * Sched switch is funny, as we only want to ignore it
 803          * in the notrace case if both prev and next should be ignored.
 804          */
 805         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 806                 trace_ignore_this_task(NULL, no_pid_list, next);
 807
 808         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 809                        (trace_ignore_this_task(pid_list, NULL, prev) &&
 810                         trace_ignore_this_task(pid_list, NULL, next)));
 811 }
 812
 813 static void
 814 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 815                                          struct task_struct *prev,
 816                                          struct task_struct *next,
 817                                          unsigned int prev_state)
 818 {
 819         struct trace_array *tr = data;
 820         struct trace_pid_list *no_pid_list;
 821         struct trace_pid_list *pid_list;
 822
 823         pid_list = rcu_dereference_sched(tr->filtered_pids);
 824         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 825
 826         this_cpu_write(tr->array_buffer.data->ignore_pid,
 827                        trace_ignore_this_task(pid_list, no_pid_list, next));
 828 }
 829
 830 static void
 831 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 832 {
 833         struct trace_array *tr = data;
 834         struct trace_pid_list *no_pid_list;
 835         struct trace_pid_list *pid_list;
 836
 837         /* Nothing to do if we are already tracing */
 838         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 839                 return;
 840
 841         pid_list = rcu_dereference_sched(tr->filtered_pids);
 842         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 843
 844         this_cpu_write(tr->array_buffer.data->ignore_pid,
 845                        trace_ignore_this_task(pid_list, no_pid_list, task));
 846 }
 847
 848 static void
 849 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 850 {
 851         struct trace_array *tr = data;
 852         struct trace_pid_list *no_pid_list;
 853         struct trace_pid_list *pid_list;
 854
 855         /* Nothing to do if we are not tracing */
 856         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 857                 return;
 858
 859         pid_list = rcu_dereference_sched(tr->filtered_pids);
 860         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 861
 862         /* Set tracing if current is enabled */
 863         this_cpu_write(tr->array_buffer.data->ignore_pid,
 864                        trace_ignore_this_task(pid_list, no_pid_list, current));
 865 }
 866
 867 static void unregister_pid_events(struct trace_array *tr)
 868 {
 869         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 870         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 871
 872         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 873         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 874
 875         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 876         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 877
 878         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 879         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 880 }
 881
 882 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 883 {
 884         struct trace_pid_list *pid_list;
 885         struct trace_pid_list *no_pid_list;
 886         struct trace_event_file *file;
 887         int cpu;
 888
 889         pid_list = rcu_dereference_protected(tr->filtered_pids,
 890                                              lockdep_is_held(&event_mutex));
 891         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 892                                              lockdep_is_held(&event_mutex));
 893
 894         /* Make sure there's something to do */
 895         if (!pid_type_enabled(type, pid_list, no_pid_list))
 896                 return;
 897
 898         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 899                 unregister_pid_events(tr);
 900
 901                 list_for_each_entry(file, &tr->events, list) {
 902                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 903                 }
 904
 905                 for_each_possible_cpu(cpu)
 906                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 907         }
 908
 909         if (type & TRACE_PIDS)
 910                 rcu_assign_pointer(tr->filtered_pids, NULL);
 911
 912         if (type & TRACE_NO_PIDS)
 913                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
 914
 915         /* Wait till all users are no longer using pid filtering */
 916         tracepoint_synchronize_unregister();
 917
 918         if ((type & TRACE_PIDS) && pid_list)
 919                 trace_pid_list_free(pid_list);
 920
 921         if ((type & TRACE_NO_PIDS) && no_pid_list)
 922                 trace_pid_list_free(no_pid_list);
 923 }
 924
 925 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 926 {
 927         mutex_lock(&event_mutex);
 928         __ftrace_clear_event_pids(tr, type);
 929         mutex_unlock(&event_mutex);
 930 }
 931
 932 static void __put_system(struct event_subsystem *system)
 933 {
 934         struct event_filter *filter = system->filter;
 935
 936         WARN_ON_ONCE(system_refcount(system) == 0);
 937         if (system_refcount_dec(system))
 938                 return;
 939
 940         list_del(&system->list);
 941
 942         if (filter) {
 943                 kfree(filter->filter_string);
 944                 kfree(filter);
 945         }
 946         kfree_const(system->name);
 947         kfree(system);
 948 }
 949
 950 static void __get_system(struct event_subsystem *system)
 951 {
 952         WARN_ON_ONCE(system_refcount(system) == 0);
 953         system_refcount_inc(system);
 954 }
 955
 956 static void __get_system_dir(struct trace_subsystem_dir *dir)
 957 {
 958         WARN_ON_ONCE(dir->ref_count == 0);
 959         dir->ref_count++;
 960         __get_system(dir->subsystem);
 961 }
 962
 963 static void __put_system_dir(struct trace_subsystem_dir *dir)
 964 {
 965         WARN_ON_ONCE(dir->ref_count == 0);
 966         /* If the subsystem is about to be freed, the dir must be too */
 967         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 968
 969         __put_system(dir->subsystem);
 970         if (!--dir->ref_count)
 971                 kfree(dir);
 972 }
 973
 974 static void put_system(struct trace_subsystem_dir *dir)
 975 {
 976         mutex_lock(&event_mutex);
 977         __put_system_dir(dir);
 978         mutex_unlock(&event_mutex);
 979 }
 980
 981 static void remove_subsystem(struct trace_subsystem_dir *dir)
 982 {
 983         if (!dir)
 984                 return;
 985
 986         if (!--dir->nr_events) {
 987                 eventfs_remove(dir->ef);
 988                 list_del(&dir->list);
 989                 __put_system_dir(dir);
 990         }
 991 }
 992
 993 void event_file_get(struct trace_event_file *file)
 994 {
 995         atomic_inc(&file->ref);
 996 }
 997
 998 void event_file_put(struct trace_event_file *file)
 999 {
1000         if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001                 if (file->flags & EVENT_FILE_FL_FREED)
1002                         kmem_cache_free(file_cachep, file);
1003                 return;
1004         }
1005
1006         if (atomic_dec_and_test(&file->ref)) {
1007                 /* Count should only go to zero when it is freed */
1008                 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009                         return;
1010                 kmem_cache_free(file_cachep, file);
1011         }
1012 }
1013
1014 static void remove_event_file_dir(struct trace_event_file *file)
1015 {
1016         eventfs_remove(file->ef);
1017         list_del(&file->list);
1018         remove_subsystem(file->system);
1019         free_event_filter(file->filter);
1020         file->flags |= EVENT_FILE_FL_FREED;
1021         event_file_put(file);
1022 }
1023
1024 /*
1025  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026  */
1027 static int
1028 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029                               const char *sub, const char *event, int set)
1030 {
1031         struct trace_event_file *file;
1032         struct trace_event_call *call;
1033         const char *name;
1034         int ret = -EINVAL;
1035         int eret = 0;
1036
1037         list_for_each_entry(file, &tr->events, list) {
1038
1039                 call = file->event_call;
1040                 name = trace_event_name(call);
1041
1042                 if (!name || !call->class || !call->class->reg)
1043                         continue;
1044
1045                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046                         continue;
1047
1048                 if (match &&
1049                     strcmp(match, name) != 0 &&
1050                     strcmp(match, call->class->system) != 0)
1051                         continue;
1052
1053                 if (sub && strcmp(sub, call->class->system) != 0)
1054                         continue;
1055
1056                 if (event && strcmp(event, name) != 0)
1057                         continue;
1058
1059                 ret = ftrace_event_enable_disable(file, set);
1060
1061                 /*
1062                  * Save the first error and return that. Some events
1063                  * may still have been enabled, but let the user
1064                  * know that something went wrong.
1065                  */
1066                 if (ret && !eret)
1067                         eret = ret;
1068
1069                 ret = eret;
1070         }
1071
1072         return ret;
1073 }
1074
1075 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076                                   const char *sub, const char *event, int set)
1077 {
1078         int ret;
1079
1080         mutex_lock(&event_mutex);
1081         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082         mutex_unlock(&event_mutex);
1083
1084         return ret;
1085 }
1086
1087 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088 {
1089         char *event = NULL, *sub = NULL, *match;
1090         int ret;
1091
1092         if (!tr)
1093                 return -ENOENT;
1094         /*
1095          * The buf format can be <subsystem>:<event-name>
1096          *  *:<event-name> means any event by that name.
1097          *  :<event-name> is the same.
1098          *
1099          *  <subsystem>:* means all events in that subsystem
1100          *  <subsystem>: means the same.
1101          *
1102          *  <name> (no ':') means all events in a subsystem with
1103          *  the name <name> or any event that matches <name>
1104          */
1105
1106         match = strsep(&buf, ":");
1107         if (buf) {
1108                 sub = match;
1109                 event = buf;
1110                 match = NULL;
1111
1112                 if (!strlen(sub) || strcmp(sub, "*") == 0)
1113                         sub = NULL;
1114                 if (!strlen(event) || strcmp(event, "*") == 0)
1115                         event = NULL;
1116         }
1117
1118         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120         /* Put back the colon to allow this to be called again */
1121         if (buf)
1122                 *(buf - 1) = ':';
1123
1124         return ret;
1125 }
1126
1127 /**
1128  * trace_set_clr_event - enable or disable an event
1129  * @system: system name to match (NULL for any system)
1130  * @event: event name to match (NULL for all events, within system)
1131  * @set: 1 to enable, 0 to disable
1132  *
1133  * This is a way for other parts of the kernel to enable or disable
1134  * event recording.
1135  *
1136  * Returns 0 on success, -EINVAL if the parameters do not match any
1137  * registered events.
1138  */
1139 int trace_set_clr_event(const char *system, const char *event, int set)
1140 {
1141         struct trace_array *tr = top_trace_array();
1142
1143         if (!tr)
1144                 return -ENODEV;
1145
1146         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147 }
1148 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150 /**
1151  * trace_array_set_clr_event - enable or disable an event for a trace array.
1152  * @tr: concerned trace array.
1153  * @system: system name to match (NULL for any system)
1154  * @event: event name to match (NULL for all events, within system)
1155  * @enable: true to enable, false to disable
1156  *
1157  * This is a way for other parts of the kernel to enable or disable
1158  * event recording.
1159  *
1160  * Returns 0 on success, -EINVAL if the parameters do not match any
1161  * registered events.
1162  */
1163 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164                 const char *event, bool enable)
1165 {
1166         int set;
1167
1168         if (!tr)
1169                 return -ENOENT;
1170
1171         set = (enable == true) ? 1 : 0;
1172         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173 }
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE          127
1178
1179 static ssize_t
1180 ftrace_event_write(struct file *file, const char __user *ubuf,
1181                    size_t cnt, loff_t *ppos)
1182 {
1183         struct trace_parser parser;
1184         struct seq_file *m = file->private_data;
1185         struct trace_array *tr = m->private;
1186         ssize_t read, ret;
1187
1188         if (!cnt)
1189                 return 0;
1190
1191         ret = tracing_update_buffers();
1192         if (ret < 0)
1193                 return ret;
1194
1195         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196                 return -ENOMEM;
1197
1198         read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200         if (read >= 0 && trace_parser_loaded((&parser))) {
1201                 int set = 1;
1202
1203                 if (*parser.buffer == '!')
1204                         set = 0;
1205
1206                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207                 if (ret)
1208                         goto out_put;
1209         }
1210
1211         ret = read;
1212
1213  out_put:
1214         trace_parser_put(&parser);
1215
1216         return ret;
1217 }
1218
1219 static void *
1220 t_next(struct seq_file *m, void *v, loff_t *pos)
1221 {
1222         struct trace_event_file *file = v;
1223         struct trace_event_call *call;
1224         struct trace_array *tr = m->private;
1225
1226         (*pos)++;
1227
1228         list_for_each_entry_continue(file, &tr->events, list) {
1229                 call = file->event_call;
1230                 /*
1231                  * The ftrace subsystem is for showing formats only.
1232                  * They can not be enabled or disabled via the event files.
1233                  */
1234                 if (call->class && call->class->reg &&
1235                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236                         return file;
1237         }
1238
1239         return NULL;
1240 }
1241
1242 static void *t_start(struct seq_file *m, loff_t *pos)
1243 {
1244         struct trace_event_file *file;
1245         struct trace_array *tr = m->private;
1246         loff_t l;
1247
1248         mutex_lock(&event_mutex);
1249
1250         file = list_entry(&tr->events, struct trace_event_file, list);
1251         for (l = 0; l <= *pos; ) {
1252                 file = t_next(m, file, &l);
1253                 if (!file)
1254                         break;
1255         }
1256         return file;
1257 }
1258
1259 static void *
1260 s_next(struct seq_file *m, void *v, loff_t *pos)
1261 {
1262         struct trace_event_file *file = v;
1263         struct trace_array *tr = m->private;
1264
1265         (*pos)++;
1266
1267         list_for_each_entry_continue(file, &tr->events, list) {
1268                 if (file->flags & EVENT_FILE_FL_ENABLED)
1269                         return file;
1270         }
1271
1272         return NULL;
1273 }
1274
1275 static void *s_start(struct seq_file *m, loff_t *pos)
1276 {
1277         struct trace_event_file *file;
1278         struct trace_array *tr = m->private;
1279         loff_t l;
1280
1281         mutex_lock(&event_mutex);
1282
1283         file = list_entry(&tr->events, struct trace_event_file, list);
1284         for (l = 0; l <= *pos; ) {
1285                 file = s_next(m, file, &l);
1286                 if (!file)
1287                         break;
1288         }
1289         return file;
1290 }
1291
1292 static int t_show(struct seq_file *m, void *v)
1293 {
1294         struct trace_event_file *file = v;
1295         struct trace_event_call *call = file->event_call;
1296
1297         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298                 seq_printf(m, "%s:", call->class->system);
1299         seq_printf(m, "%s\n", trace_event_name(call));
1300
1301         return 0;
1302 }
1303
1304 static void t_stop(struct seq_file *m, void *p)
1305 {
1306         mutex_unlock(&event_mutex);
1307 }
1308
1309 static void *
1310 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1311 {
1312         struct trace_array *tr = m->private;
1313         struct trace_pid_list *pid_list;
1314
1315         if (type == TRACE_PIDS)
1316                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1317         else
1318                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320         return trace_pid_next(pid_list, v, pos);
1321 }
1322
1323 static void *
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326         return __next(m, v, pos, TRACE_PIDS);
1327 }
1328
1329 static void *
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1331 {
1332         return __next(m, v, pos, TRACE_NO_PIDS);
1333 }
1334
1335 static void *__start(struct seq_file *m, loff_t *pos, int type)
1336         __acquires(RCU)
1337 {
1338         struct trace_pid_list *pid_list;
1339         struct trace_array *tr = m->private;
1340
1341         /*
1342          * Grab the mutex, to keep calls to p_next() having the same
1343          * tr->filtered_pids as p_start() has.
1344          * If we just passed the tr->filtered_pids around, then RCU would
1345          * have been enough, but doing that makes things more complex.
1346          */
1347         mutex_lock(&event_mutex);
1348         rcu_read_lock_sched();
1349
1350         if (type == TRACE_PIDS)
1351                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1352         else
1353                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355         if (!pid_list)
1356                 return NULL;
1357
1358         return trace_pid_start(pid_list, pos);
1359 }
1360
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1362         __acquires(RCU)
1363 {
1364         return __start(m, pos, TRACE_PIDS);
1365 }
1366
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1368         __acquires(RCU)
1369 {
1370         return __start(m, pos, TRACE_NO_PIDS);
1371 }
1372
1373 static void p_stop(struct seq_file *m, void *p)
1374         __releases(RCU)
1375 {
1376         rcu_read_unlock_sched();
1377         mutex_unlock(&event_mutex);
1378 }
1379
1380 static ssize_t
1381 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382                   loff_t *ppos)
1383 {
1384         struct trace_event_file *file;
1385         unsigned long flags;
1386         char buf[4] = "0";
1387
1388         mutex_lock(&event_mutex);
1389         file = event_file_data(filp);
1390         if (likely(file))
1391                 flags = file->flags;
1392         mutex_unlock(&event_mutex);
1393
1394         if (!file || flags & EVENT_FILE_FL_FREED)
1395                 return -ENODEV;
1396
1397         if (flags & EVENT_FILE_FL_ENABLED &&
1398             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399                 strcpy(buf, "1");
1400
1401         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402             flags & EVENT_FILE_FL_SOFT_MODE)
1403                 strcat(buf, "*");
1404
1405         strcat(buf, "\n");
1406
1407         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408 }
1409
1410 static ssize_t
1411 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412                    loff_t *ppos)
1413 {
1414         struct trace_event_file *file;
1415         unsigned long val;
1416         int ret;
1417
1418         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419         if (ret)
1420                 return ret;
1421
1422         ret = tracing_update_buffers();
1423         if (ret < 0)
1424                 return ret;
1425
1426         switch (val) {
1427         case 0:
1428         case 1:
1429                 ret = -ENODEV;
1430                 mutex_lock(&event_mutex);
1431                 file = event_file_data(filp);
1432                 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED)))
1433                         ret = ftrace_event_enable_disable(file, val);
1434                 mutex_unlock(&event_mutex);
1435                 break;
1436
1437         default:
1438                 return -EINVAL;
1439         }
1440
1441         *ppos += cnt;
1442
1443         return ret ? ret : cnt;
1444 }
1445
1446 static ssize_t
1447 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1448                    loff_t *ppos)
1449 {
1450         const char set_to_char[4] = { '?', '0', '1', 'X' };
1451         struct trace_subsystem_dir *dir = filp->private_data;
1452         struct event_subsystem *system = dir->subsystem;
1453         struct trace_event_call *call;
1454         struct trace_event_file *file;
1455         struct trace_array *tr = dir->tr;
1456         char buf[2];
1457         int set = 0;
1458         int ret;
1459
1460         mutex_lock(&event_mutex);
1461         list_for_each_entry(file, &tr->events, list) {
1462                 call = file->event_call;
1463                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1464                     !trace_event_name(call) || !call->class || !call->class->reg)
1465                         continue;
1466
1467                 if (system && strcmp(call->class->system, system->name) != 0)
1468                         continue;
1469
1470                 /*
1471                  * We need to find out if all the events are set
1472                  * or if all events or cleared, or if we have
1473                  * a mixture.
1474                  */
1475                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1476
1477                 /*
1478                  * If we have a mixture, no need to look further.
1479                  */
1480                 if (set == 3)
1481                         break;
1482         }
1483         mutex_unlock(&event_mutex);
1484
1485         buf[0] = set_to_char[set];
1486         buf[1] = '\n';
1487
1488         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1489
1490         return ret;
1491 }
1492
1493 static ssize_t
1494 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1495                     loff_t *ppos)
1496 {
1497         struct trace_subsystem_dir *dir = filp->private_data;
1498         struct event_subsystem *system = dir->subsystem;
1499         const char *name = NULL;
1500         unsigned long val;
1501         ssize_t ret;
1502
1503         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1504         if (ret)
1505                 return ret;
1506
1507         ret = tracing_update_buffers();
1508         if (ret < 0)
1509                 return ret;
1510
1511         if (val != 0 && val != 1)
1512                 return -EINVAL;
1513
1514         /*
1515          * Opening of "enable" adds a ref count to system,
1516          * so the name is safe to use.
1517          */
1518         if (system)
1519                 name = system->name;
1520
1521         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1522         if (ret)
1523                 goto out;
1524
1525         ret = cnt;
1526
1527 out:
1528         *ppos += cnt;
1529
1530         return ret;
1531 }
1532
1533 enum {
1534         FORMAT_HEADER           = 1,
1535         FORMAT_FIELD_SEPERATOR  = 2,
1536         FORMAT_PRINTFMT         = 3,
1537 };
1538
1539 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1540 {
1541         struct trace_event_call *call = event_file_data(m->private);
1542         struct list_head *common_head = &ftrace_common_fields;
1543         struct list_head *head = trace_get_fields(call);
1544         struct list_head *node = v;
1545
1546         (*pos)++;
1547
1548         switch ((unsigned long)v) {
1549         case FORMAT_HEADER:
1550                 node = common_head;
1551                 break;
1552
1553         case FORMAT_FIELD_SEPERATOR:
1554                 node = head;
1555                 break;
1556
1557         case FORMAT_PRINTFMT:
1558                 /* all done */
1559                 return NULL;
1560         }
1561
1562         node = node->prev;
1563         if (node == common_head)
1564                 return (void *)FORMAT_FIELD_SEPERATOR;
1565         else if (node == head)
1566                 return (void *)FORMAT_PRINTFMT;
1567         else
1568                 return node;
1569 }
1570
1571 static int f_show(struct seq_file *m, void *v)
1572 {
1573         struct trace_event_call *call = event_file_data(m->private);
1574         struct ftrace_event_field *field;
1575         const char *array_descriptor;
1576
1577         switch ((unsigned long)v) {
1578         case FORMAT_HEADER:
1579                 seq_printf(m, "name: %s\n", trace_event_name(call));
1580                 seq_printf(m, "ID: %d\n", call->event.type);
1581                 seq_puts(m, "format:\n");
1582                 return 0;
1583
1584         case FORMAT_FIELD_SEPERATOR:
1585                 seq_putc(m, '\n');
1586                 return 0;
1587
1588         case FORMAT_PRINTFMT:
1589                 seq_printf(m, "\nprint fmt: %s\n",
1590                            call->print_fmt);
1591                 return 0;
1592         }
1593
1594         field = list_entry(v, struct ftrace_event_field, link);
1595         /*
1596          * Smartly shows the array type(except dynamic array).
1597          * Normal:
1598          *      field:TYPE VAR
1599          * If TYPE := TYPE[LEN], it is shown:
1600          *      field:TYPE VAR[LEN]
1601          */
1602         array_descriptor = strchr(field->type, '[');
1603
1604         if (str_has_prefix(field->type, "__data_loc"))
1605                 array_descriptor = NULL;
1606
1607         if (!array_descriptor)
1608                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1609                            field->type, field->name, field->offset,
1610                            field->size, !!field->is_signed);
1611         else if (field->len)
1612                 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1613                            (int)(array_descriptor - field->type),
1614                            field->type, field->name,
1615                            field->len, field->offset,
1616                            field->size, !!field->is_signed);
1617         else
1618                 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1619                                 (int)(array_descriptor - field->type),
1620                                 field->type, field->name,
1621                                 field->offset, field->size, !!field->is_signed);
1622
1623         return 0;
1624 }
1625
1626 static void *f_start(struct seq_file *m, loff_t *pos)
1627 {
1628         void *p = (void *)FORMAT_HEADER;
1629         loff_t l = 0;
1630
1631         /* ->stop() is called even if ->start() fails */
1632         mutex_lock(&event_mutex);
1633         if (!event_file_data(m->private))
1634                 return ERR_PTR(-ENODEV);
1635
1636         while (l < *pos && p)
1637                 p = f_next(m, p, &l);
1638
1639         return p;
1640 }
1641
1642 static void f_stop(struct seq_file *m, void *p)
1643 {
1644         mutex_unlock(&event_mutex);
1645 }
1646
1647 static const struct seq_operations trace_format_seq_ops = {
1648         .start          = f_start,
1649         .next           = f_next,
1650         .stop           = f_stop,
1651         .show           = f_show,
1652 };
1653
1654 static int trace_format_open(struct inode *inode, struct file *file)
1655 {
1656         struct seq_file *m;
1657         int ret;
1658
1659         /* Do we want to hide event format files on tracefs lockdown? */
1660
1661         ret = seq_open(file, &trace_format_seq_ops);
1662         if (ret < 0)
1663                 return ret;
1664
1665         m = file->private_data;
1666         m->private = file;
1667
1668         return 0;
1669 }
1670
1671 static ssize_t
1672 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1673 {
1674         int id = (long)event_file_data(filp);
1675         char buf[32];
1676         int len;
1677
1678         if (unlikely(!id))
1679                 return -ENODEV;
1680
1681         len = sprintf(buf, "%d\n", id);
1682
1683         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1684 }
1685
1686 static ssize_t
1687 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1688                   loff_t *ppos)
1689 {
1690         struct trace_event_file *file;
1691         struct trace_seq *s;
1692         int r = -ENODEV;
1693
1694         if (*ppos)
1695                 return 0;
1696
1697         s = kmalloc(sizeof(*s), GFP_KERNEL);
1698
1699         if (!s)
1700                 return -ENOMEM;
1701
1702         trace_seq_init(s);
1703
1704         mutex_lock(&event_mutex);
1705         file = event_file_data(filp);
1706         if (file && !(file->flags & EVENT_FILE_FL_FREED))
1707                 print_event_filter(file, s);
1708         mutex_unlock(&event_mutex);
1709
1710         if (file)
1711                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1712                                             s->buffer, trace_seq_used(s));
1713
1714         kfree(s);
1715
1716         return r;
1717 }
1718
1719 static ssize_t
1720 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1721                    loff_t *ppos)
1722 {
1723         struct trace_event_file *file;
1724         char *buf;
1725         int err = -ENODEV;
1726
1727         if (cnt >= PAGE_SIZE)
1728                 return -EINVAL;
1729
1730         buf = memdup_user_nul(ubuf, cnt);
1731         if (IS_ERR(buf))
1732                 return PTR_ERR(buf);
1733
1734         mutex_lock(&event_mutex);
1735         file = event_file_data(filp);
1736         if (file)
1737                 err = apply_event_filter(file, buf);
1738         mutex_unlock(&event_mutex);
1739
1740         kfree(buf);
1741         if (err < 0)
1742                 return err;
1743
1744         *ppos += cnt;
1745
1746         return cnt;
1747 }
1748
1749 static LIST_HEAD(event_subsystems);
1750
1751 static int subsystem_open(struct inode *inode, struct file *filp)
1752 {
1753         struct trace_subsystem_dir *dir = NULL, *iter_dir;
1754         struct trace_array *tr = NULL, *iter_tr;
1755         struct event_subsystem *system = NULL;
1756         int ret;
1757
1758         if (tracing_is_disabled())
1759                 return -ENODEV;
1760
1761         /* Make sure the system still exists */
1762         mutex_lock(&event_mutex);
1763         mutex_lock(&trace_types_lock);
1764         list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1765                 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1766                         if (iter_dir == inode->i_private) {
1767                                 /* Don't open systems with no events */
1768                                 tr = iter_tr;
1769                                 dir = iter_dir;
1770                                 if (dir->nr_events) {
1771                                         __get_system_dir(dir);
1772                                         system = dir->subsystem;
1773                                 }
1774                                 goto exit_loop;
1775                         }
1776                 }
1777         }
1778  exit_loop:
1779         mutex_unlock(&trace_types_lock);
1780         mutex_unlock(&event_mutex);
1781
1782         if (!system)
1783                 return -ENODEV;
1784
1785         /* Still need to increment the ref count of the system */
1786         if (trace_array_get(tr) < 0) {
1787                 put_system(dir);
1788                 return -ENODEV;
1789         }
1790
1791         ret = tracing_open_generic(inode, filp);
1792         if (ret < 0) {
1793                 trace_array_put(tr);
1794                 put_system(dir);
1795         }
1796
1797         return ret;
1798 }
1799
1800 static int system_tr_open(struct inode *inode, struct file *filp)
1801 {
1802         struct trace_subsystem_dir *dir;
1803         struct trace_array *tr = inode->i_private;
1804         int ret;
1805
1806         /* Make a temporary dir that has no system but points to tr */
1807         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1808         if (!dir)
1809                 return -ENOMEM;
1810
1811         ret = tracing_open_generic_tr(inode, filp);
1812         if (ret < 0) {
1813                 kfree(dir);
1814                 return ret;
1815         }
1816         dir->tr = tr;
1817         filp->private_data = dir;
1818
1819         return 0;
1820 }
1821
1822 static int subsystem_release(struct inode *inode, struct file *file)
1823 {
1824         struct trace_subsystem_dir *dir = file->private_data;
1825
1826         trace_array_put(dir->tr);
1827
1828         /*
1829          * If dir->subsystem is NULL, then this is a temporary
1830          * descriptor that was made for a trace_array to enable
1831          * all subsystems.
1832          */
1833         if (dir->subsystem)
1834                 put_system(dir);
1835         else
1836                 kfree(dir);
1837
1838         return 0;
1839 }
1840
1841 static ssize_t
1842 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1843                       loff_t *ppos)
1844 {
1845         struct trace_subsystem_dir *dir = filp->private_data;
1846         struct event_subsystem *system = dir->subsystem;
1847         struct trace_seq *s;
1848         int r;
1849
1850         if (*ppos)
1851                 return 0;
1852
1853         s = kmalloc(sizeof(*s), GFP_KERNEL);
1854         if (!s)
1855                 return -ENOMEM;
1856
1857         trace_seq_init(s);
1858
1859         print_subsystem_event_filter(system, s);
1860         r = simple_read_from_buffer(ubuf, cnt, ppos,
1861                                     s->buffer, trace_seq_used(s));
1862
1863         kfree(s);
1864
1865         return r;
1866 }
1867
1868 static ssize_t
1869 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1870                        loff_t *ppos)
1871 {
1872         struct trace_subsystem_dir *dir = filp->private_data;
1873         char *buf;
1874         int err;
1875
1876         if (cnt >= PAGE_SIZE)
1877                 return -EINVAL;
1878
1879         buf = memdup_user_nul(ubuf, cnt);
1880         if (IS_ERR(buf))
1881                 return PTR_ERR(buf);
1882
1883         err = apply_subsystem_event_filter(dir, buf);
1884         kfree(buf);
1885         if (err < 0)
1886                 return err;
1887
1888         *ppos += cnt;
1889
1890         return cnt;
1891 }
1892
1893 static ssize_t
1894 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1895 {
1896         int (*func)(struct trace_seq *s) = filp->private_data;
1897         struct trace_seq *s;
1898         int r;
1899
1900         if (*ppos)
1901                 return 0;
1902
1903         s = kmalloc(sizeof(*s), GFP_KERNEL);
1904         if (!s)
1905                 return -ENOMEM;
1906
1907         trace_seq_init(s);
1908
1909         func(s);
1910         r = simple_read_from_buffer(ubuf, cnt, ppos,
1911                                     s->buffer, trace_seq_used(s));
1912
1913         kfree(s);
1914
1915         return r;
1916 }
1917
1918 static void ignore_task_cpu(void *data)
1919 {
1920         struct trace_array *tr = data;
1921         struct trace_pid_list *pid_list;
1922         struct trace_pid_list *no_pid_list;
1923
1924         /*
1925          * This function is called by on_each_cpu() while the
1926          * event_mutex is held.
1927          */
1928         pid_list = rcu_dereference_protected(tr->filtered_pids,
1929                                              mutex_is_locked(&event_mutex));
1930         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1931                                              mutex_is_locked(&event_mutex));
1932
1933         this_cpu_write(tr->array_buffer.data->ignore_pid,
1934                        trace_ignore_this_task(pid_list, no_pid_list, current));
1935 }
1936
1937 static void register_pid_events(struct trace_array *tr)
1938 {
1939         /*
1940          * Register a probe that is called before all other probes
1941          * to set ignore_pid if next or prev do not match.
1942          * Register a probe this is called after all other probes
1943          * to only keep ignore_pid set if next pid matches.
1944          */
1945         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1946                                          tr, INT_MAX);
1947         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1948                                          tr, 0);
1949
1950         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1951                                          tr, INT_MAX);
1952         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1953                                          tr, 0);
1954
1955         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1956                                              tr, INT_MAX);
1957         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1958                                              tr, 0);
1959
1960         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1961                                          tr, INT_MAX);
1962         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1963                                          tr, 0);
1964 }
1965
1966 static ssize_t
1967 event_pid_write(struct file *filp, const char __user *ubuf,
1968                 size_t cnt, loff_t *ppos, int type)
1969 {
1970         struct seq_file *m = filp->private_data;
1971         struct trace_array *tr = m->private;
1972         struct trace_pid_list *filtered_pids = NULL;
1973         struct trace_pid_list *other_pids = NULL;
1974         struct trace_pid_list *pid_list;
1975         struct trace_event_file *file;
1976         ssize_t ret;
1977
1978         if (!cnt)
1979                 return 0;
1980
1981         ret = tracing_update_buffers();
1982         if (ret < 0)
1983                 return ret;
1984
1985         mutex_lock(&event_mutex);
1986
1987         if (type == TRACE_PIDS) {
1988                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1989                                                           lockdep_is_held(&event_mutex));
1990                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1991                                                           lockdep_is_held(&event_mutex));
1992         } else {
1993                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1994                                                           lockdep_is_held(&event_mutex));
1995                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1996                                                           lockdep_is_held(&event_mutex));
1997         }
1998
1999         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2000         if (ret < 0)
2001                 goto out;
2002
2003         if (type == TRACE_PIDS)
2004                 rcu_assign_pointer(tr->filtered_pids, pid_list);
2005         else
2006                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2007
2008         list_for_each_entry(file, &tr->events, list) {
2009                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2010         }
2011
2012         if (filtered_pids) {
2013                 tracepoint_synchronize_unregister();
2014                 trace_pid_list_free(filtered_pids);
2015         } else if (pid_list && !other_pids) {
2016                 register_pid_events(tr);
2017         }
2018
2019         /*
2020          * Ignoring of pids is done at task switch. But we have to
2021          * check for those tasks that are currently running.
2022          * Always do this in case a pid was appended or removed.
2023          */
2024         on_each_cpu(ignore_task_cpu, tr, 1);
2025
2026  out:
2027         mutex_unlock(&event_mutex);
2028
2029         if (ret > 0)
2030                 *ppos += ret;
2031
2032         return ret;
2033 }
2034
2035 static ssize_t
2036 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2037                        size_t cnt, loff_t *ppos)
2038 {
2039         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2040 }
2041
2042 static ssize_t
2043 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2044                         size_t cnt, loff_t *ppos)
2045 {
2046         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2047 }
2048
2049 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2050 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2051 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2052 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2053 static int ftrace_event_release(struct inode *inode, struct file *file);
2054
2055 static const struct seq_operations show_event_seq_ops = {
2056         .start = t_start,
2057         .next = t_next,
2058         .show = t_show,
2059         .stop = t_stop,
2060 };
2061
2062 static const struct seq_operations show_set_event_seq_ops = {
2063         .start = s_start,
2064         .next = s_next,
2065         .show = t_show,
2066         .stop = t_stop,
2067 };
2068
2069 static const struct seq_operations show_set_pid_seq_ops = {
2070         .start = p_start,
2071         .next = p_next,
2072         .show = trace_pid_show,
2073         .stop = p_stop,
2074 };
2075
2076 static const struct seq_operations show_set_no_pid_seq_ops = {
2077         .start = np_start,
2078         .next = np_next,
2079         .show = trace_pid_show,
2080         .stop = p_stop,
2081 };
2082
2083 static const struct file_operations ftrace_avail_fops = {
2084         .open = ftrace_event_avail_open,
2085         .read = seq_read,
2086         .llseek = seq_lseek,
2087         .release = seq_release,
2088 };
2089
2090 static const struct file_operations ftrace_set_event_fops = {
2091         .open = ftrace_event_set_open,
2092         .read = seq_read,
2093         .write = ftrace_event_write,
2094         .llseek = seq_lseek,
2095         .release = ftrace_event_release,
2096 };
2097
2098 static const struct file_operations ftrace_set_event_pid_fops = {
2099         .open = ftrace_event_set_pid_open,
2100         .read = seq_read,
2101         .write = ftrace_event_pid_write,
2102         .llseek = seq_lseek,
2103         .release = ftrace_event_release,
2104 };
2105
2106 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2107         .open = ftrace_event_set_npid_open,
2108         .read = seq_read,
2109         .write = ftrace_event_npid_write,
2110         .llseek = seq_lseek,
2111         .release = ftrace_event_release,
2112 };
2113
2114 static const struct file_operations ftrace_enable_fops = {
2115         .open = tracing_open_file_tr,
2116         .read = event_enable_read,
2117         .write = event_enable_write,
2118         .release = tracing_release_file_tr,
2119         .llseek = default_llseek,
2120 };
2121
2122 static const struct file_operations ftrace_event_format_fops = {
2123         .open = trace_format_open,
2124         .read = seq_read,
2125         .llseek = seq_lseek,
2126         .release = seq_release,
2127 };
2128
2129 static const struct file_operations ftrace_event_id_fops = {
2130         .read = event_id_read,
2131         .llseek = default_llseek,
2132 };
2133
2134 static const struct file_operations ftrace_event_filter_fops = {
2135         .open = tracing_open_file_tr,
2136         .read = event_filter_read,
2137         .write = event_filter_write,
2138         .release = tracing_release_file_tr,
2139         .llseek = default_llseek,
2140 };
2141
2142 static const struct file_operations ftrace_subsystem_filter_fops = {
2143         .open = subsystem_open,
2144         .read = subsystem_filter_read,
2145         .write = subsystem_filter_write,
2146         .llseek = default_llseek,
2147         .release = subsystem_release,
2148 };
2149
2150 static const struct file_operations ftrace_system_enable_fops = {
2151         .open = subsystem_open,
2152         .read = system_enable_read,
2153         .write = system_enable_write,
2154         .llseek = default_llseek,
2155         .release = subsystem_release,
2156 };
2157
2158 static const struct file_operations ftrace_tr_enable_fops = {
2159         .open = system_tr_open,
2160         .read = system_enable_read,
2161         .write = system_enable_write,
2162         .llseek = default_llseek,
2163         .release = subsystem_release,
2164 };
2165
2166 static const struct file_operations ftrace_show_header_fops = {
2167         .open = tracing_open_generic,
2168         .read = show_header,
2169         .llseek = default_llseek,
2170 };
2171
2172 static int
2173 ftrace_event_open(struct inode *inode, struct file *file,
2174                   const struct seq_operations *seq_ops)
2175 {
2176         struct seq_file *m;
2177         int ret;
2178
2179         ret = security_locked_down(LOCKDOWN_TRACEFS);
2180         if (ret)
2181                 return ret;
2182
2183         ret = seq_open(file, seq_ops);
2184         if (ret < 0)
2185                 return ret;
2186         m = file->private_data;
2187         /* copy tr over to seq ops */
2188         m->private = inode->i_private;
2189
2190         return ret;
2191 }
2192
2193 static int ftrace_event_release(struct inode *inode, struct file *file)
2194 {
2195         struct trace_array *tr = inode->i_private;
2196
2197         trace_array_put(tr);
2198
2199         return seq_release(inode, file);
2200 }
2201
2202 static int
2203 ftrace_event_avail_open(struct inode *inode, struct file *file)
2204 {
2205         const struct seq_operations *seq_ops = &show_event_seq_ops;
2206
2207         /* Checks for tracefs lockdown */
2208         return ftrace_event_open(inode, file, seq_ops);
2209 }
2210
2211 static int
2212 ftrace_event_set_open(struct inode *inode, struct file *file)
2213 {
2214         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2215         struct trace_array *tr = inode->i_private;
2216         int ret;
2217
2218         ret = tracing_check_open_get_tr(tr);
2219         if (ret)
2220                 return ret;
2221
2222         if ((file->f_mode & FMODE_WRITE) &&
2223             (file->f_flags & O_TRUNC))
2224                 ftrace_clear_events(tr);
2225
2226         ret = ftrace_event_open(inode, file, seq_ops);
2227         if (ret < 0)
2228                 trace_array_put(tr);
2229         return ret;
2230 }
2231
2232 static int
2233 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2234 {
2235         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2236         struct trace_array *tr = inode->i_private;
2237         int ret;
2238
2239         ret = tracing_check_open_get_tr(tr);
2240         if (ret)
2241                 return ret;
2242
2243         if ((file->f_mode & FMODE_WRITE) &&
2244             (file->f_flags & O_TRUNC))
2245                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2246
2247         ret = ftrace_event_open(inode, file, seq_ops);
2248         if (ret < 0)
2249                 trace_array_put(tr);
2250         return ret;
2251 }
2252
2253 static int
2254 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2255 {
2256         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2257         struct trace_array *tr = inode->i_private;
2258         int ret;
2259
2260         ret = tracing_check_open_get_tr(tr);
2261         if (ret)
2262                 return ret;
2263
2264         if ((file->f_mode & FMODE_WRITE) &&
2265             (file->f_flags & O_TRUNC))
2266                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2267
2268         ret = ftrace_event_open(inode, file, seq_ops);
2269         if (ret < 0)
2270                 trace_array_put(tr);
2271         return ret;
2272 }
2273
2274 static struct event_subsystem *
2275 create_new_subsystem(const char *name)
2276 {
2277         struct event_subsystem *system;
2278
2279         /* need to create new entry */
2280         system = kmalloc(sizeof(*system), GFP_KERNEL);
2281         if (!system)
2282                 return NULL;
2283
2284         system->ref_count = 1;
2285
2286         /* Only allocate if dynamic (kprobes and modules) */
2287         system->name = kstrdup_const(name, GFP_KERNEL);
2288         if (!system->name)
2289                 goto out_free;
2290
2291         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2292         if (!system->filter)
2293                 goto out_free;
2294
2295         list_add(&system->list, &event_subsystems);
2296
2297         return system;
2298
2299  out_free:
2300         kfree_const(system->name);
2301         kfree(system);
2302         return NULL;
2303 }
2304
2305 static struct eventfs_file *
2306 event_subsystem_dir(struct trace_array *tr, const char *name,
2307                     struct trace_event_file *file, struct dentry *parent)
2308 {
2309         struct event_subsystem *system, *iter;
2310         struct trace_subsystem_dir *dir;
2311         struct eventfs_file *ef;
2312         int res;
2313
2314         /* First see if we did not already create this dir */
2315         list_for_each_entry(dir, &tr->systems, list) {
2316                 system = dir->subsystem;
2317                 if (strcmp(system->name, name) == 0) {
2318                         dir->nr_events++;
2319                         file->system = dir;
2320                         return dir->ef;
2321                 }
2322         }
2323
2324         /* Now see if the system itself exists. */
2325         system = NULL;
2326         list_for_each_entry(iter, &event_subsystems, list) {
2327                 if (strcmp(iter->name, name) == 0) {
2328                         system = iter;
2329                         break;
2330                 }
2331         }
2332
2333         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2334         if (!dir)
2335                 goto out_fail;
2336
2337         if (!system) {
2338                 system = create_new_subsystem(name);
2339                 if (!system)
2340                         goto out_free;
2341         } else
2342                 __get_system(system);
2343
2344         ef = eventfs_add_subsystem_dir(name, parent);
2345         if (IS_ERR(ef)) {
2346                 pr_warn("Failed to create system directory %s\n", name);
2347                 __put_system(system);
2348                 goto out_free;
2349         }
2350
2351         dir->ef = ef;
2352         dir->tr = tr;
2353         dir->ref_count = 1;
2354         dir->nr_events = 1;
2355         dir->subsystem = system;
2356         file->system = dir;
2357
2358         /* the ftrace system is special, do not create enable or filter files */
2359         if (strcmp(name, "ftrace") != 0) {
2360
2361                 res = eventfs_add_file("filter", TRACE_MODE_WRITE,
2362                                             dir->ef, dir,
2363                                             &ftrace_subsystem_filter_fops);
2364                 if (res) {
2365                         kfree(system->filter);
2366                         system->filter = NULL;
2367                         pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2368                 }
2369
2370                 eventfs_add_file("enable", TRACE_MODE_WRITE, dir->ef, dir,
2371                                   &ftrace_system_enable_fops);
2372         }
2373
2374         list_add(&dir->list, &tr->systems);
2375
2376         return dir->ef;
2377
2378  out_free:
2379         kfree(dir);
2380  out_fail:
2381         /* Only print this message if failed on memory allocation */
2382         if (!dir || !system)
2383                 pr_warn("No memory to create event subsystem %s\n", name);
2384         return NULL;
2385 }
2386
2387 static int
2388 event_define_fields(struct trace_event_call *call)
2389 {
2390         struct list_head *head;
2391         int ret = 0;
2392
2393         /*
2394          * Other events may have the same class. Only update
2395          * the fields if they are not already defined.
2396          */
2397         head = trace_get_fields(call);
2398         if (list_empty(head)) {
2399                 struct trace_event_fields *field = call->class->fields_array;
2400                 unsigned int offset = sizeof(struct trace_entry);
2401
2402                 for (; field->type; field++) {
2403                         if (field->type == TRACE_FUNCTION_TYPE) {
2404                                 field->define_fields(call);
2405                                 break;
2406                         }
2407
2408                         offset = ALIGN(offset, field->align);
2409                         ret = trace_define_field_ext(call, field->type, field->name,
2410                                                  offset, field->size,
2411                                                  field->is_signed, field->filter_type,
2412                                                  field->len);
2413                         if (WARN_ON_ONCE(ret)) {
2414                                 pr_err("error code is %d\n", ret);
2415                                 break;
2416                         }
2417
2418                         offset += field->size;
2419                 }
2420         }
2421
2422         return ret;
2423 }
2424
2425 static int
2426 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2427 {
2428         struct trace_event_call *call = file->event_call;
2429         struct eventfs_file *ef_subsystem = NULL;
2430         struct trace_array *tr = file->tr;
2431         struct eventfs_file *ef;
2432         const char *name;
2433         int ret;
2434
2435         /*
2436          * If the trace point header did not define TRACE_SYSTEM
2437          * then the system would be called "TRACE_SYSTEM". This should
2438          * never happen.
2439          */
2440         if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2441                 return -ENODEV;
2442
2443         ef_subsystem = event_subsystem_dir(tr, call->class->system, file, parent);
2444         if (!ef_subsystem)
2445                 return -ENOMEM;
2446
2447         name = trace_event_name(call);
2448         ef = eventfs_add_dir(name, ef_subsystem);
2449         if (IS_ERR(ef)) {
2450                 pr_warn("Could not create tracefs '%s' directory\n", name);
2451                 return -1;
2452         }
2453
2454         file->ef = ef;
2455
2456         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2457                 eventfs_add_file("enable", TRACE_MODE_WRITE, file->ef, file,
2458                                   &ftrace_enable_fops);
2459
2460 #ifdef CONFIG_PERF_EVENTS
2461         if (call->event.type && call->class->reg)
2462                 eventfs_add_file("id", TRACE_MODE_READ, file->ef,
2463                                   (void *)(long)call->event.type,
2464                                   &ftrace_event_id_fops);
2465 #endif
2466
2467         ret = event_define_fields(call);
2468         if (ret < 0) {
2469                 pr_warn("Could not initialize trace point events/%s\n", name);
2470                 return ret;
2471         }
2472
2473         /*
2474          * Only event directories that can be enabled should have
2475          * triggers or filters.
2476          */
2477         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2478                 eventfs_add_file("filter", TRACE_MODE_WRITE, file->ef,
2479                                   file, &ftrace_event_filter_fops);
2480
2481                 eventfs_add_file("trigger", TRACE_MODE_WRITE, file->ef,
2482                                   file, &event_trigger_fops);
2483         }
2484
2485 #ifdef CONFIG_HIST_TRIGGERS
2486         eventfs_add_file("hist", TRACE_MODE_READ, file->ef, file,
2487                           &event_hist_fops);
2488 #endif
2489 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2490         eventfs_add_file("hist_debug", TRACE_MODE_READ, file->ef, file,
2491                           &event_hist_debug_fops);
2492 #endif
2493         eventfs_add_file("format", TRACE_MODE_READ, file->ef, call,
2494                           &ftrace_event_format_fops);
2495
2496 #ifdef CONFIG_TRACE_EVENT_INJECT
2497         if (call->event.type && call->class->reg)
2498                 eventfs_add_file("inject", 0200, file->ef, file,
2499                                   &event_inject_fops);
2500 #endif
2501
2502         return 0;
2503 }
2504
2505 static void remove_event_from_tracers(struct trace_event_call *call)
2506 {
2507         struct trace_event_file *file;
2508         struct trace_array *tr;
2509
2510         do_for_each_event_file_safe(tr, file) {
2511                 if (file->event_call != call)
2512                         continue;
2513
2514                 remove_event_file_dir(file);
2515                 /*
2516                  * The do_for_each_event_file_safe() is
2517                  * a double loop. After finding the call for this
2518                  * trace_array, we use break to jump to the next
2519                  * trace_array.
2520                  */
2521                 break;
2522         } while_for_each_event_file();
2523 }
2524
2525 static void event_remove(struct trace_event_call *call)
2526 {
2527         struct trace_array *tr;
2528         struct trace_event_file *file;
2529
2530         do_for_each_event_file(tr, file) {
2531                 if (file->event_call != call)
2532                         continue;
2533
2534                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2535                         tr->clear_trace = true;
2536
2537                 ftrace_event_enable_disable(file, 0);
2538                 /*
2539                  * The do_for_each_event_file() is
2540                  * a double loop. After finding the call for this
2541                  * trace_array, we use break to jump to the next
2542                  * trace_array.
2543                  */
2544                 break;
2545         } while_for_each_event_file();
2546
2547         if (call->event.funcs)
2548                 __unregister_trace_event(&call->event);
2549         remove_event_from_tracers(call);
2550         list_del(&call->list);
2551 }
2552
2553 static int event_init(struct trace_event_call *call)
2554 {
2555         int ret = 0;
2556         const char *name;
2557
2558         name = trace_event_name(call);
2559         if (WARN_ON(!name))
2560                 return -EINVAL;
2561
2562         if (call->class->raw_init) {
2563                 ret = call->class->raw_init(call);
2564                 if (ret < 0 && ret != -ENOSYS)
2565                         pr_warn("Could not initialize trace events/%s\n", name);
2566         }
2567
2568         return ret;
2569 }
2570
2571 static int
2572 __register_event(struct trace_event_call *call, struct module *mod)
2573 {
2574         int ret;
2575
2576         ret = event_init(call);
2577         if (ret < 0)
2578                 return ret;
2579
2580         list_add(&call->list, &ftrace_events);
2581         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2582                 atomic_set(&call->refcnt, 0);
2583         else
2584                 call->module = mod;
2585
2586         return 0;
2587 }
2588
2589 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2590 {
2591         int rlen;
2592         int elen;
2593
2594         /* Find the length of the eval value as a string */
2595         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2596         /* Make sure there's enough room to replace the string with the value */
2597         if (len < elen)
2598                 return NULL;
2599
2600         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2601
2602         /* Get the rest of the string of ptr */
2603         rlen = strlen(ptr + len);
2604         memmove(ptr + elen, ptr + len, rlen);
2605         /* Make sure we end the new string */
2606         ptr[elen + rlen] = 0;
2607
2608         return ptr + elen;
2609 }
2610
2611 static void update_event_printk(struct trace_event_call *call,
2612                                 struct trace_eval_map *map)
2613 {
2614         char *ptr;
2615         int quote = 0;
2616         int len = strlen(map->eval_string);
2617
2618         for (ptr = call->print_fmt; *ptr; ptr++) {
2619                 if (*ptr == '\\') {
2620                         ptr++;
2621                         /* paranoid */
2622                         if (!*ptr)
2623                                 break;
2624                         continue;
2625                 }
2626                 if (*ptr == '"') {
2627                         quote ^= 1;
2628                         continue;
2629                 }
2630                 if (quote)
2631                         continue;
2632                 if (isdigit(*ptr)) {
2633                         /* skip numbers */
2634                         do {
2635                                 ptr++;
2636                                 /* Check for alpha chars like ULL */
2637                         } while (isalnum(*ptr));
2638                         if (!*ptr)
2639                                 break;
2640                         /*
2641                          * A number must have some kind of delimiter after
2642                          * it, and we can ignore that too.
2643                          */
2644                         continue;
2645                 }
2646                 if (isalpha(*ptr) || *ptr == '_') {
2647                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2648                             !isalnum(ptr[len]) && ptr[len] != '_') {
2649                                 ptr = eval_replace(ptr, map, len);
2650                                 /* enum/sizeof string smaller than value */
2651                                 if (WARN_ON_ONCE(!ptr))
2652                                         return;
2653                                 /*
2654                                  * No need to decrement here, as eval_replace()
2655                                  * returns the pointer to the character passed
2656                                  * the eval, and two evals can not be placed
2657                                  * back to back without something in between.
2658                                  * We can skip that something in between.
2659                                  */
2660                                 continue;
2661                         }
2662                 skip_more:
2663                         do {
2664                                 ptr++;
2665                         } while (isalnum(*ptr) || *ptr == '_');
2666                         if (!*ptr)
2667                                 break;
2668                         /*
2669                          * If what comes after this variable is a '.' or
2670                          * '->' then we can continue to ignore that string.
2671                          */
2672                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2673                                 ptr += *ptr == '.' ? 1 : 2;
2674                                 if (!*ptr)
2675                                         break;
2676                                 goto skip_more;
2677                         }
2678                         /*
2679                          * Once again, we can skip the delimiter that came
2680                          * after the string.
2681                          */
2682                         continue;
2683                 }
2684         }
2685 }
2686
2687 static void add_str_to_module(struct module *module, char *str)
2688 {
2689         struct module_string *modstr;
2690
2691         modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2692
2693         /*
2694          * If we failed to allocate memory here, then we'll just
2695          * let the str memory leak when the module is removed.
2696          * If this fails to allocate, there's worse problems than
2697          * a leaked string on module removal.
2698          */
2699         if (WARN_ON_ONCE(!modstr))
2700                 return;
2701
2702         modstr->module = module;
2703         modstr->str = str;
2704
2705         list_add(&modstr->next, &module_strings);
2706 }
2707
2708 static void update_event_fields(struct trace_event_call *call,
2709                                 struct trace_eval_map *map)
2710 {
2711         struct ftrace_event_field *field;
2712         struct list_head *head;
2713         char *ptr;
2714         char *str;
2715         int len = strlen(map->eval_string);
2716
2717         /* Dynamic events should never have field maps */
2718         if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2719                 return;
2720
2721         head = trace_get_fields(call);
2722         list_for_each_entry(field, head, link) {
2723                 ptr = strchr(field->type, '[');
2724                 if (!ptr)
2725                         continue;
2726                 ptr++;
2727
2728                 if (!isalpha(*ptr) && *ptr != '_')
2729                         continue;
2730
2731                 if (strncmp(map->eval_string, ptr, len) != 0)
2732                         continue;
2733
2734                 str = kstrdup(field->type, GFP_KERNEL);
2735                 if (WARN_ON_ONCE(!str))
2736                         return;
2737                 ptr = str + (ptr - field->type);
2738                 ptr = eval_replace(ptr, map, len);
2739                 /* enum/sizeof string smaller than value */
2740                 if (WARN_ON_ONCE(!ptr)) {
2741                         kfree(str);
2742                         continue;
2743                 }
2744
2745                 /*
2746                  * If the event is part of a module, then we need to free the string
2747                  * when the module is removed. Otherwise, it will stay allocated
2748                  * until a reboot.
2749                  */
2750                 if (call->module)
2751                         add_str_to_module(call->module, str);
2752
2753                 field->type = str;
2754         }
2755 }
2756
2757 void trace_event_eval_update(struct trace_eval_map **map, int len)
2758 {
2759         struct trace_event_call *call, *p;
2760         const char *last_system = NULL;
2761         bool first = false;
2762         int last_i;
2763         int i;
2764
2765         down_write(&trace_event_sem);
2766         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2767                 /* events are usually grouped together with systems */
2768                 if (!last_system || call->class->system != last_system) {
2769                         first = true;
2770                         last_i = 0;
2771                         last_system = call->class->system;
2772                 }
2773
2774                 /*
2775                  * Since calls are grouped by systems, the likelihood that the
2776                  * next call in the iteration belongs to the same system as the
2777                  * previous call is high. As an optimization, we skip searching
2778                  * for a map[] that matches the call's system if the last call
2779                  * was from the same system. That's what last_i is for. If the
2780                  * call has the same system as the previous call, then last_i
2781                  * will be the index of the first map[] that has a matching
2782                  * system.
2783                  */
2784                 for (i = last_i; i < len; i++) {
2785                         if (call->class->system == map[i]->system) {
2786                                 /* Save the first system if need be */
2787                                 if (first) {
2788                                         last_i = i;
2789                                         first = false;
2790                                 }
2791                                 update_event_printk(call, map[i]);
2792                                 update_event_fields(call, map[i]);
2793                         }
2794                 }
2795                 cond_resched();
2796         }
2797         up_write(&trace_event_sem);
2798 }
2799
2800 static struct trace_event_file *
2801 trace_create_new_event(struct trace_event_call *call,
2802                        struct trace_array *tr)
2803 {
2804         struct trace_pid_list *no_pid_list;
2805         struct trace_pid_list *pid_list;
2806         struct trace_event_file *file;
2807         unsigned int first;
2808
2809         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2810         if (!file)
2811                 return NULL;
2812
2813         pid_list = rcu_dereference_protected(tr->filtered_pids,
2814                                              lockdep_is_held(&event_mutex));
2815         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2816                                              lockdep_is_held(&event_mutex));
2817
2818         if (!trace_pid_list_first(pid_list, &first) ||
2819             !trace_pid_list_first(no_pid_list, &first))
2820                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2821
2822         file->event_call = call;
2823         file->tr = tr;
2824         atomic_set(&file->sm_ref, 0);
2825         atomic_set(&file->tm_ref, 0);
2826         INIT_LIST_HEAD(&file->triggers);
2827         list_add(&file->list, &tr->events);
2828         event_file_get(file);
2829
2830         return file;
2831 }
2832
2833 #define MAX_BOOT_TRIGGERS 32
2834
2835 static struct boot_triggers {
2836         const char              *event;
2837         char                    *trigger;
2838 } bootup_triggers[MAX_BOOT_TRIGGERS];
2839
2840 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2841 static int nr_boot_triggers;
2842
2843 static __init int setup_trace_triggers(char *str)
2844 {
2845         char *trigger;
2846         char *buf;
2847         int i;
2848
2849         strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
2850         ring_buffer_expanded = true;
2851         disable_tracing_selftest("running event triggers");
2852
2853         buf = bootup_trigger_buf;
2854         for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
2855                 trigger = strsep(&buf, ",");
2856                 if (!trigger)
2857                         break;
2858                 bootup_triggers[i].event = strsep(&trigger, ".");
2859                 bootup_triggers[i].trigger = trigger;
2860                 if (!bootup_triggers[i].trigger)
2861                         break;
2862         }
2863
2864         nr_boot_triggers = i;
2865         return 1;
2866 }
2867 __setup("trace_trigger=", setup_trace_triggers);
2868
2869 /* Add an event to a trace directory */
2870 static int
2871 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2872 {
2873         struct trace_event_file *file;
2874
2875         file = trace_create_new_event(call, tr);
2876         if (!file)
2877                 return -ENOMEM;
2878
2879         if (eventdir_initialized)
2880                 return event_create_dir(tr->event_dir, file);
2881         else
2882                 return event_define_fields(call);
2883 }
2884
2885 static void trace_early_triggers(struct trace_event_file *file, const char *name)
2886 {
2887         int ret;
2888         int i;
2889
2890         for (i = 0; i < nr_boot_triggers; i++) {
2891                 if (strcmp(name, bootup_triggers[i].event))
2892                         continue;
2893                 mutex_lock(&event_mutex);
2894                 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
2895                 mutex_unlock(&event_mutex);
2896                 if (ret)
2897                         pr_err("Failed to register trigger '%s' on event %s\n",
2898                                bootup_triggers[i].trigger,
2899                                bootup_triggers[i].event);
2900         }
2901 }
2902
2903 /*
2904  * Just create a descriptor for early init. A descriptor is required
2905  * for enabling events at boot. We want to enable events before
2906  * the filesystem is initialized.
2907  */
2908 static int
2909 __trace_early_add_new_event(struct trace_event_call *call,
2910                             struct trace_array *tr)
2911 {
2912         struct trace_event_file *file;
2913         int ret;
2914
2915         file = trace_create_new_event(call, tr);
2916         if (!file)
2917                 return -ENOMEM;
2918
2919         ret = event_define_fields(call);
2920         if (ret)
2921                 return ret;
2922
2923         trace_early_triggers(file, trace_event_name(call));
2924
2925         return 0;
2926 }
2927
2928 struct ftrace_module_file_ops;
2929 static void __add_event_to_tracers(struct trace_event_call *call);
2930
2931 /* Add an additional event_call dynamically */
2932 int trace_add_event_call(struct trace_event_call *call)
2933 {
2934         int ret;
2935         lockdep_assert_held(&event_mutex);
2936
2937         mutex_lock(&trace_types_lock);
2938
2939         ret = __register_event(call, NULL);
2940         if (ret >= 0)
2941                 __add_event_to_tracers(call);
2942
2943         mutex_unlock(&trace_types_lock);
2944         return ret;
2945 }
2946 EXPORT_SYMBOL_GPL(trace_add_event_call);
2947
2948 /*
2949  * Must be called under locking of trace_types_lock, event_mutex and
2950  * trace_event_sem.
2951  */
2952 static void __trace_remove_event_call(struct trace_event_call *call)
2953 {
2954         event_remove(call);
2955         trace_destroy_fields(call);
2956         free_event_filter(call->filter);
2957         call->filter = NULL;
2958 }
2959
2960 static int probe_remove_event_call(struct trace_event_call *call)
2961 {
2962         struct trace_array *tr;
2963         struct trace_event_file *file;
2964
2965 #ifdef CONFIG_PERF_EVENTS
2966         if (call->perf_refcount)
2967                 return -EBUSY;
2968 #endif
2969         do_for_each_event_file(tr, file) {
2970                 if (file->event_call != call)
2971                         continue;
2972                 /*
2973                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2974                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2975                  * TRACE_REG_UNREGISTER.
2976                  */
2977                 if (file->flags & EVENT_FILE_FL_ENABLED)
2978                         goto busy;
2979
2980                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2981                         tr->clear_trace = true;
2982                 /*
2983                  * The do_for_each_event_file_safe() is
2984                  * a double loop. After finding the call for this
2985                  * trace_array, we use break to jump to the next
2986                  * trace_array.
2987                  */
2988                 break;
2989         } while_for_each_event_file();
2990
2991         __trace_remove_event_call(call);
2992
2993         return 0;
2994  busy:
2995         /* No need to clear the trace now */
2996         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2997                 tr->clear_trace = false;
2998         }
2999         return -EBUSY;
3000 }
3001
3002 /* Remove an event_call */
3003 int trace_remove_event_call(struct trace_event_call *call)
3004 {
3005         int ret;
3006
3007         lockdep_assert_held(&event_mutex);
3008
3009         mutex_lock(&trace_types_lock);
3010         down_write(&trace_event_sem);
3011         ret = probe_remove_event_call(call);
3012         up_write(&trace_event_sem);
3013         mutex_unlock(&trace_types_lock);
3014
3015         return ret;
3016 }
3017 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3018
3019 #define for_each_event(event, start, end)                       \
3020         for (event = start;                                     \
3021              (unsigned long)event < (unsigned long)end;         \
3022              event++)
3023
3024 #ifdef CONFIG_MODULES
3025
3026 static void trace_module_add_events(struct module *mod)
3027 {
3028         struct trace_event_call **call, **start, **end;
3029
3030         if (!mod->num_trace_events)
3031                 return;
3032
3033         /* Don't add infrastructure for mods without tracepoints */
3034         if (trace_module_has_bad_taint(mod)) {
3035                 pr_err("%s: module has bad taint, not creating trace events\n",
3036                        mod->name);
3037                 return;
3038         }
3039
3040         start = mod->trace_events;
3041         end = mod->trace_events + mod->num_trace_events;
3042
3043         for_each_event(call, start, end) {
3044                 __register_event(*call, mod);
3045                 __add_event_to_tracers(*call);
3046         }
3047 }
3048
3049 static void trace_module_remove_events(struct module *mod)
3050 {
3051         struct trace_event_call *call, *p;
3052         struct module_string *modstr, *m;
3053
3054         down_write(&trace_event_sem);
3055         list_for_each_entry_safe(call, p, &ftrace_events, list) {
3056                 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3057                         continue;
3058                 if (call->module == mod)
3059                         __trace_remove_event_call(call);
3060         }
3061         /* Check for any strings allocade for this module */
3062         list_for_each_entry_safe(modstr, m, &module_strings, next) {
3063                 if (modstr->module != mod)
3064                         continue;
3065                 list_del(&modstr->next);
3066                 kfree(modstr->str);
3067                 kfree(modstr);
3068         }
3069         up_write(&trace_event_sem);
3070
3071         /*
3072          * It is safest to reset the ring buffer if the module being unloaded
3073          * registered any events that were used. The only worry is if
3074          * a new module gets loaded, and takes on the same id as the events
3075          * of this module. When printing out the buffer, traced events left
3076          * over from this module may be passed to the new module events and
3077          * unexpected results may occur.
3078          */
3079         tracing_reset_all_online_cpus_unlocked();
3080 }
3081
3082 static int trace_module_notify(struct notifier_block *self,
3083                                unsigned long val, void *data)
3084 {
3085         struct module *mod = data;
3086
3087         mutex_lock(&event_mutex);
3088         mutex_lock(&trace_types_lock);
3089         switch (val) {
3090         case MODULE_STATE_COMING:
3091                 trace_module_add_events(mod);
3092                 break;
3093         case MODULE_STATE_GOING:
3094                 trace_module_remove_events(mod);
3095                 break;
3096         }
3097         mutex_unlock(&trace_types_lock);
3098         mutex_unlock(&event_mutex);
3099
3100         return NOTIFY_OK;
3101 }
3102
3103 static struct notifier_block trace_module_nb = {
3104         .notifier_call = trace_module_notify,
3105         .priority = 1, /* higher than trace.c module notify */
3106 };
3107 #endif /* CONFIG_MODULES */
3108
3109 /* Create a new event directory structure for a trace directory. */
3110 static void
3111 __trace_add_event_dirs(struct trace_array *tr)
3112 {
3113         struct trace_event_call *call;
3114         int ret;
3115
3116         list_for_each_entry(call, &ftrace_events, list) {
3117                 ret = __trace_add_new_event(call, tr);
3118                 if (ret < 0)
3119                         pr_warn("Could not create directory for event %s\n",
3120                                 trace_event_name(call));
3121         }
3122 }
3123
3124 /* Returns any file that matches the system and event */
3125 struct trace_event_file *
3126 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3127 {
3128         struct trace_event_file *file;
3129         struct trace_event_call *call;
3130         const char *name;
3131
3132         list_for_each_entry(file, &tr->events, list) {
3133
3134                 call = file->event_call;
3135                 name = trace_event_name(call);
3136
3137                 if (!name || !call->class)
3138                         continue;
3139
3140                 if (strcmp(event, name) == 0 &&
3141                     strcmp(system, call->class->system) == 0)
3142                         return file;
3143         }
3144         return NULL;
3145 }
3146
3147 /* Returns valid trace event files that match system and event */
3148 struct trace_event_file *
3149 find_event_file(struct trace_array *tr, const char *system, const char *event)
3150 {
3151         struct trace_event_file *file;
3152
3153         file = __find_event_file(tr, system, event);
3154         if (!file || !file->event_call->class->reg ||
3155             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3156                 return NULL;
3157
3158         return file;
3159 }
3160
3161 /**
3162  * trace_get_event_file - Find and return a trace event file
3163  * @instance: The name of the trace instance containing the event
3164  * @system: The name of the system containing the event
3165  * @event: The name of the event
3166  *
3167  * Return a trace event file given the trace instance name, trace
3168  * system, and trace event name.  If the instance name is NULL, it
3169  * refers to the top-level trace array.
3170  *
3171  * This function will look it up and return it if found, after calling
3172  * trace_array_get() to prevent the instance from going away, and
3173  * increment the event's module refcount to prevent it from being
3174  * removed.
3175  *
3176  * To release the file, call trace_put_event_file(), which will call
3177  * trace_array_put() and decrement the event's module refcount.
3178  *
3179  * Return: The trace event on success, ERR_PTR otherwise.
3180  */
3181 struct trace_event_file *trace_get_event_file(const char *instance,
3182                                               const char *system,
3183                                               const char *event)
3184 {
3185         struct trace_array *tr = top_trace_array();
3186         struct trace_event_file *file = NULL;
3187         int ret = -EINVAL;
3188
3189         if (instance) {
3190                 tr = trace_array_find_get(instance);
3191                 if (!tr)
3192                         return ERR_PTR(-ENOENT);
3193         } else {
3194                 ret = trace_array_get(tr);
3195                 if (ret)
3196                         return ERR_PTR(ret);
3197         }
3198
3199         mutex_lock(&event_mutex);
3200
3201         file = find_event_file(tr, system, event);
3202         if (!file) {
3203                 trace_array_put(tr);
3204                 ret = -EINVAL;
3205                 goto out;
3206         }
3207
3208         /* Don't let event modules unload while in use */
3209         ret = trace_event_try_get_ref(file->event_call);
3210         if (!ret) {
3211                 trace_array_put(tr);
3212                 ret = -EBUSY;
3213                 goto out;
3214         }
3215
3216         ret = 0;
3217  out:
3218         mutex_unlock(&event_mutex);
3219
3220         if (ret)
3221                 file = ERR_PTR(ret);
3222
3223         return file;
3224 }
3225 EXPORT_SYMBOL_GPL(trace_get_event_file);
3226
3227 /**
3228  * trace_put_event_file - Release a file from trace_get_event_file()
3229  * @file: The trace event file
3230  *
3231  * If a file was retrieved using trace_get_event_file(), this should
3232  * be called when it's no longer needed.  It will cancel the previous
3233  * trace_array_get() called by that function, and decrement the
3234  * event's module refcount.
3235  */
3236 void trace_put_event_file(struct trace_event_file *file)
3237 {
3238         mutex_lock(&event_mutex);
3239         trace_event_put_ref(file->event_call);
3240         mutex_unlock(&event_mutex);
3241
3242         trace_array_put(file->tr);
3243 }
3244 EXPORT_SYMBOL_GPL(trace_put_event_file);
3245
3246 #ifdef CONFIG_DYNAMIC_FTRACE
3247
3248 /* Avoid typos */
3249 #define ENABLE_EVENT_STR        "enable_event"
3250 #define DISABLE_EVENT_STR       "disable_event"
3251
3252 struct event_probe_data {
3253         struct trace_event_file *file;
3254         unsigned long                   count;
3255         int                             ref;
3256         bool                            enable;
3257 };
3258
3259 static void update_event_probe(struct event_probe_data *data)
3260 {
3261         if (data->enable)
3262                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3263         else
3264                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3265 }
3266
3267 static void
3268 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3269                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3270                    void *data)
3271 {
3272         struct ftrace_func_mapper *mapper = data;
3273         struct event_probe_data *edata;
3274         void **pdata;
3275
3276         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3277         if (!pdata || !*pdata)
3278                 return;
3279
3280         edata = *pdata;
3281         update_event_probe(edata);
3282 }
3283
3284 static void
3285 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3286                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3287                          void *data)
3288 {
3289         struct ftrace_func_mapper *mapper = data;
3290         struct event_probe_data *edata;
3291         void **pdata;
3292
3293         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3294         if (!pdata || !*pdata)
3295                 return;
3296
3297         edata = *pdata;
3298
3299         if (!edata->count)
3300                 return;
3301
3302         /* Skip if the event is in a state we want to switch to */
3303         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3304                 return;
3305
3306         if (edata->count != -1)
3307                 (edata->count)--;
3308
3309         update_event_probe(edata);
3310 }
3311
3312 static int
3313 event_enable_print(struct seq_file *m, unsigned long ip,
3314                    struct ftrace_probe_ops *ops, void *data)
3315 {
3316         struct ftrace_func_mapper *mapper = data;
3317         struct event_probe_data *edata;
3318         void **pdata;
3319
3320         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3321
3322         if (WARN_ON_ONCE(!pdata || !*pdata))
3323                 return 0;
3324
3325         edata = *pdata;
3326
3327         seq_printf(m, "%ps:", (void *)ip);
3328
3329         seq_printf(m, "%s:%s:%s",
3330                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3331                    edata->file->event_call->class->system,
3332                    trace_event_name(edata->file->event_call));
3333
3334         if (edata->count == -1)
3335                 seq_puts(m, ":unlimited\n");
3336         else
3337                 seq_printf(m, ":count=%ld\n", edata->count);
3338
3339         return 0;
3340 }
3341
3342 static int
3343 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3344                   unsigned long ip, void *init_data, void **data)
3345 {
3346         struct ftrace_func_mapper *mapper = *data;
3347         struct event_probe_data *edata = init_data;
3348         int ret;
3349
3350         if (!mapper) {
3351                 mapper = allocate_ftrace_func_mapper();
3352                 if (!mapper)
3353                         return -ENODEV;
3354                 *data = mapper;
3355         }
3356
3357         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3358         if (ret < 0)
3359                 return ret;
3360
3361         edata->ref++;
3362
3363         return 0;
3364 }
3365
3366 static int free_probe_data(void *data)
3367 {
3368         struct event_probe_data *edata = data;
3369
3370         edata->ref--;
3371         if (!edata->ref) {
3372                 /* Remove the SOFT_MODE flag */
3373                 __ftrace_event_enable_disable(edata->file, 0, 1);
3374                 trace_event_put_ref(edata->file->event_call);
3375                 kfree(edata);
3376         }
3377         return 0;
3378 }
3379
3380 static void
3381 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3382                   unsigned long ip, void *data)
3383 {
3384         struct ftrace_func_mapper *mapper = data;
3385         struct event_probe_data *edata;
3386
3387         if (!ip) {
3388                 if (!mapper)
3389                         return;
3390                 free_ftrace_func_mapper(mapper, free_probe_data);
3391                 return;
3392         }
3393
3394         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3395
3396         if (WARN_ON_ONCE(!edata))
3397                 return;
3398
3399         if (WARN_ON_ONCE(edata->ref <= 0))
3400                 return;
3401
3402         free_probe_data(edata);
3403 }
3404
3405 static struct ftrace_probe_ops event_enable_probe_ops = {
3406         .func                   = event_enable_probe,
3407         .print                  = event_enable_print,
3408         .init                   = event_enable_init,
3409         .free                   = event_enable_free,
3410 };
3411
3412 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3413         .func                   = event_enable_count_probe,
3414         .print                  = event_enable_print,
3415         .init                   = event_enable_init,
3416         .free                   = event_enable_free,
3417 };
3418
3419 static struct ftrace_probe_ops event_disable_probe_ops = {
3420         .func                   = event_enable_probe,
3421         .print                  = event_enable_print,
3422         .init                   = event_enable_init,
3423         .free                   = event_enable_free,
3424 };
3425
3426 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3427         .func                   = event_enable_count_probe,
3428         .print                  = event_enable_print,
3429         .init                   = event_enable_init,
3430         .free                   = event_enable_free,
3431 };
3432
3433 static int
3434 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3435                   char *glob, char *cmd, char *param, int enabled)
3436 {
3437         struct trace_event_file *file;
3438         struct ftrace_probe_ops *ops;
3439         struct event_probe_data *data;
3440         const char *system;
3441         const char *event;
3442         char *number;
3443         bool enable;
3444         int ret;
3445
3446         if (!tr)
3447                 return -ENODEV;
3448
3449         /* hash funcs only work with set_ftrace_filter */
3450         if (!enabled || !param)
3451                 return -EINVAL;
3452
3453         system = strsep(&param, ":");
3454         if (!param)
3455                 return -EINVAL;
3456
3457         event = strsep(&param, ":");
3458
3459         mutex_lock(&event_mutex);
3460
3461         ret = -EINVAL;
3462         file = find_event_file(tr, system, event);
3463         if (!file)
3464                 goto out;
3465
3466         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3467
3468         if (enable)
3469                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3470         else
3471                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3472
3473         if (glob[0] == '!') {
3474                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3475                 goto out;
3476         }
3477
3478         ret = -ENOMEM;
3479
3480         data = kzalloc(sizeof(*data), GFP_KERNEL);
3481         if (!data)
3482                 goto out;
3483
3484         data->enable = enable;
3485         data->count = -1;
3486         data->file = file;
3487
3488         if (!param)
3489                 goto out_reg;
3490
3491         number = strsep(&param, ":");
3492
3493         ret = -EINVAL;
3494         if (!strlen(number))
3495                 goto out_free;
3496
3497         /*
3498          * We use the callback data field (which is a pointer)
3499          * as our counter.
3500          */
3501         ret = kstrtoul(number, 0, &data->count);
3502         if (ret)
3503                 goto out_free;
3504
3505  out_reg:
3506         /* Don't let event modules unload while probe registered */
3507         ret = trace_event_try_get_ref(file->event_call);
3508         if (!ret) {
3509                 ret = -EBUSY;
3510                 goto out_free;
3511         }
3512
3513         ret = __ftrace_event_enable_disable(file, 1, 1);
3514         if (ret < 0)
3515                 goto out_put;
3516
3517         ret = register_ftrace_function_probe(glob, tr, ops, data);
3518         /*
3519          * The above returns on success the # of functions enabled,
3520          * but if it didn't find any functions it returns zero.
3521          * Consider no functions a failure too.
3522          */
3523         if (!ret) {
3524                 ret = -ENOENT;
3525                 goto out_disable;
3526         } else if (ret < 0)
3527                 goto out_disable;
3528         /* Just return zero, not the number of enabled functions */
3529         ret = 0;
3530  out:
3531         mutex_unlock(&event_mutex);
3532         return ret;
3533
3534  out_disable:
3535         __ftrace_event_enable_disable(file, 0, 1);
3536  out_put:
3537         trace_event_put_ref(file->event_call);
3538  out_free:
3539         kfree(data);
3540         goto out;
3541 }
3542
3543 static struct ftrace_func_command event_enable_cmd = {
3544         .name                   = ENABLE_EVENT_STR,
3545         .func                   = event_enable_func,
3546 };
3547
3548 static struct ftrace_func_command event_disable_cmd = {
3549         .name                   = DISABLE_EVENT_STR,
3550         .func                   = event_enable_func,
3551 };
3552
3553 static __init int register_event_cmds(void)
3554 {
3555         int ret;
3556
3557         ret = register_ftrace_command(&event_enable_cmd);
3558         if (WARN_ON(ret < 0))
3559                 return ret;
3560         ret = register_ftrace_command(&event_disable_cmd);
3561         if (WARN_ON(ret < 0))
3562                 unregister_ftrace_command(&event_enable_cmd);
3563         return ret;
3564 }
3565 #else
3566 static inline int register_event_cmds(void) { return 0; }
3567 #endif /* CONFIG_DYNAMIC_FTRACE */
3568
3569 /*
3570  * The top level array and trace arrays created by boot-time tracing
3571  * have already had its trace_event_file descriptors created in order
3572  * to allow for early events to be recorded.
3573  * This function is called after the tracefs has been initialized,
3574  * and we now have to create the files associated to the events.
3575  */
3576 static void __trace_early_add_event_dirs(struct trace_array *tr)
3577 {
3578         struct trace_event_file *file;
3579         int ret;
3580
3581
3582         list_for_each_entry(file, &tr->events, list) {
3583                 ret = event_create_dir(tr->event_dir, file);
3584                 if (ret < 0)
3585                         pr_warn("Could not create directory for event %s\n",
3586                                 trace_event_name(file->event_call));
3587         }
3588 }
3589
3590 /*
3591  * For early boot up, the top trace array and the trace arrays created
3592  * by boot-time tracing require to have a list of events that can be
3593  * enabled. This must be done before the filesystem is set up in order
3594  * to allow events to be traced early.
3595  */
3596 void __trace_early_add_events(struct trace_array *tr)
3597 {
3598         struct trace_event_call *call;
3599         int ret;
3600
3601         list_for_each_entry(call, &ftrace_events, list) {
3602                 /* Early boot up should not have any modules loaded */
3603                 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3604                     WARN_ON_ONCE(call->module))
3605                         continue;
3606
3607                 ret = __trace_early_add_new_event(call, tr);
3608                 if (ret < 0)
3609                         pr_warn("Could not create early event %s\n",
3610                                 trace_event_name(call));
3611         }
3612 }
3613
3614 /* Remove the event directory structure for a trace directory. */
3615 static void
3616 __trace_remove_event_dirs(struct trace_array *tr)
3617 {
3618         struct trace_event_file *file, *next;
3619
3620         list_for_each_entry_safe(file, next, &tr->events, list)
3621                 remove_event_file_dir(file);
3622 }
3623
3624 static void __add_event_to_tracers(struct trace_event_call *call)
3625 {
3626         struct trace_array *tr;
3627
3628         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3629                 __trace_add_new_event(call, tr);
3630 }
3631
3632 extern struct trace_event_call *__start_ftrace_events[];
3633 extern struct trace_event_call *__stop_ftrace_events[];
3634
3635 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3636
3637 static __init int setup_trace_event(char *str)
3638 {
3639         strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3640         ring_buffer_expanded = true;
3641         disable_tracing_selftest("running event tracing");
3642
3643         return 1;
3644 }
3645 __setup("trace_event=", setup_trace_event);
3646
3647 /* Expects to have event_mutex held when called */
3648 static int
3649 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3650 {
3651         struct dentry *d_events;
3652         struct dentry *entry;
3653         int error = 0;
3654
3655         entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3656                                   tr, &ftrace_set_event_fops);
3657         if (!entry)
3658                 return -ENOMEM;
3659
3660         d_events = eventfs_create_events_dir("events", parent);
3661         if (IS_ERR(d_events)) {
3662                 pr_warn("Could not create tracefs 'events' directory\n");
3663                 return -ENOMEM;
3664         }
3665
3666         error = eventfs_add_events_file("enable", TRACE_MODE_WRITE, d_events,
3667                                   tr, &ftrace_tr_enable_fops);
3668         if (error)
3669                 return -ENOMEM;
3670
3671         /* There are not as crucial, just warn if they are not created */
3672
3673         trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3674                           tr, &ftrace_set_event_pid_fops);
3675
3676         trace_create_file("set_event_notrace_pid",
3677                           TRACE_MODE_WRITE, parent, tr,
3678                           &ftrace_set_event_notrace_pid_fops);
3679
3680         /* ring buffer internal formats */
3681         eventfs_add_events_file("header_page", TRACE_MODE_READ, d_events,
3682                                   ring_buffer_print_page_header,
3683                                   &ftrace_show_header_fops);
3684
3685         eventfs_add_events_file("header_event", TRACE_MODE_READ, d_events,
3686                                   ring_buffer_print_entry_header,
3687                                   &ftrace_show_header_fops);
3688
3689         tr->event_dir = d_events;
3690
3691         return 0;
3692 }
3693
3694 /**
3695  * event_trace_add_tracer - add a instance of a trace_array to events
3696  * @parent: The parent dentry to place the files/directories for events in
3697  * @tr: The trace array associated with these events
3698  *
3699  * When a new instance is created, it needs to set up its events
3700  * directory, as well as other files associated with events. It also
3701  * creates the event hierarchy in the @parent/events directory.
3702  *
3703  * Returns 0 on success.
3704  *
3705  * Must be called with event_mutex held.
3706  */
3707 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3708 {
3709         int ret;
3710
3711         lockdep_assert_held(&event_mutex);
3712
3713         ret = create_event_toplevel_files(parent, tr);
3714         if (ret)
3715                 goto out;
3716
3717         down_write(&trace_event_sem);
3718         /* If tr already has the event list, it is initialized in early boot. */
3719         if (unlikely(!list_empty(&tr->events)))
3720                 __trace_early_add_event_dirs(tr);
3721         else
3722                 __trace_add_event_dirs(tr);
3723         up_write(&trace_event_sem);
3724
3725  out:
3726         return ret;
3727 }
3728
3729 /*
3730  * The top trace array already had its file descriptors created.
3731  * Now the files themselves need to be created.
3732  */
3733 static __init int
3734 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3735 {
3736         int ret;
3737
3738         mutex_lock(&event_mutex);
3739
3740         ret = create_event_toplevel_files(parent, tr);
3741         if (ret)
3742                 goto out_unlock;
3743
3744         down_write(&trace_event_sem);
3745         __trace_early_add_event_dirs(tr);
3746         up_write(&trace_event_sem);
3747
3748  out_unlock:
3749         mutex_unlock(&event_mutex);
3750
3751         return ret;
3752 }
3753
3754 /* Must be called with event_mutex held */
3755 int event_trace_del_tracer(struct trace_array *tr)
3756 {
3757         lockdep_assert_held(&event_mutex);
3758
3759         /* Disable any event triggers and associated soft-disabled events */
3760         clear_event_triggers(tr);
3761
3762         /* Clear the pid list */
3763         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3764
3765         /* Disable any running events */
3766         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3767
3768         /* Make sure no more events are being executed */
3769         tracepoint_synchronize_unregister();
3770
3771         down_write(&trace_event_sem);
3772         __trace_remove_event_dirs(tr);
3773         eventfs_remove_events_dir(tr->event_dir);
3774         up_write(&trace_event_sem);
3775
3776         tr->event_dir = NULL;
3777
3778         return 0;
3779 }
3780
3781 static __init int event_trace_memsetup(void)
3782 {
3783         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3784         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3785         return 0;
3786 }
3787
3788 __init void
3789 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3790 {
3791         char *token;
3792         int ret;
3793
3794         while (true) {
3795                 token = strsep(&buf, ",");
3796
3797                 if (!token)
3798                         break;
3799
3800                 if (*token) {
3801                         /* Restarting syscalls requires that we stop them first */
3802                         if (disable_first)
3803                                 ftrace_set_clr_event(tr, token, 0);
3804
3805                         ret = ftrace_set_clr_event(tr, token, 1);
3806                         if (ret)
3807                                 pr_warn("Failed to enable trace event: %s\n", token);
3808                 }
3809
3810                 /* Put back the comma to allow this to be called again */
3811                 if (buf)
3812                         *(buf - 1) = ',';
3813         }
3814 }
3815
3816 static __init int event_trace_enable(void)
3817 {
3818         struct trace_array *tr = top_trace_array();
3819         struct trace_event_call **iter, *call;
3820         int ret;
3821
3822         if (!tr)
3823                 return -ENODEV;
3824
3825         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3826
3827                 call = *iter;
3828                 ret = event_init(call);
3829                 if (!ret)
3830                         list_add(&call->list, &ftrace_events);
3831         }
3832
3833         register_trigger_cmds();
3834
3835         /*
3836          * We need the top trace array to have a working set of trace
3837          * points at early init, before the debug files and directories
3838          * are created. Create the file entries now, and attach them
3839          * to the actual file dentries later.
3840          */
3841         __trace_early_add_events(tr);
3842
3843         early_enable_events(tr, bootup_event_buf, false);
3844
3845         trace_printk_start_comm();
3846
3847         register_event_cmds();
3848
3849
3850         return 0;
3851 }
3852
3853 /*
3854  * event_trace_enable() is called from trace_event_init() first to
3855  * initialize events and perhaps start any events that are on the
3856  * command line. Unfortunately, there are some events that will not
3857  * start this early, like the system call tracepoints that need
3858  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3859  * event_trace_enable() is called before pid 1 starts, and this flag
3860  * is never set, making the syscall tracepoint never get reached, but
3861  * the event is enabled regardless (and not doing anything).
3862  */
3863 static __init int event_trace_enable_again(void)
3864 {
3865         struct trace_array *tr;
3866
3867         tr = top_trace_array();
3868         if (!tr)
3869                 return -ENODEV;
3870
3871         early_enable_events(tr, bootup_event_buf, true);
3872
3873         return 0;
3874 }
3875
3876 early_initcall(event_trace_enable_again);
3877
3878 /* Init fields which doesn't related to the tracefs */
3879 static __init int event_trace_init_fields(void)
3880 {
3881         if (trace_define_generic_fields())
3882                 pr_warn("tracing: Failed to allocated generic fields");
3883
3884         if (trace_define_common_fields())
3885                 pr_warn("tracing: Failed to allocate common fields");
3886
3887         return 0;
3888 }
3889
3890 __init int event_trace_init(void)
3891 {
3892         struct trace_array *tr;
3893         int ret;
3894
3895         tr = top_trace_array();
3896         if (!tr)
3897                 return -ENODEV;
3898
3899         trace_create_file("available_events", TRACE_MODE_READ,
3900                           NULL, tr, &ftrace_avail_fops);
3901
3902         ret = early_event_add_tracer(NULL, tr);
3903         if (ret)
3904                 return ret;
3905
3906 #ifdef CONFIG_MODULES
3907         ret = register_module_notifier(&trace_module_nb);
3908         if (ret)
3909                 pr_warn("Failed to register trace events module notifier\n");
3910 #endif
3911
3912         eventdir_initialized = true;
3913
3914         return 0;
3915 }
3916
3917 void __init trace_event_init(void)
3918 {
3919         event_trace_memsetup();
3920         init_ftrace_syscalls();
3921         event_trace_enable();
3922         event_trace_init_fields();
3923 }
3924
3925 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3926
3927 static DEFINE_SPINLOCK(test_spinlock);
3928 static DEFINE_SPINLOCK(test_spinlock_irq);
3929 static DEFINE_MUTEX(test_mutex);
3930
3931 static __init void test_work(struct work_struct *dummy)
3932 {
3933         spin_lock(&test_spinlock);
3934         spin_lock_irq(&test_spinlock_irq);
3935         udelay(1);
3936         spin_unlock_irq(&test_spinlock_irq);
3937         spin_unlock(&test_spinlock);
3938
3939         mutex_lock(&test_mutex);
3940         msleep(1);
3941         mutex_unlock(&test_mutex);
3942 }
3943
3944 static __init int event_test_thread(void *unused)
3945 {
3946         void *test_malloc;
3947
3948         test_malloc = kmalloc(1234, GFP_KERNEL);
3949         if (!test_malloc)
3950                 pr_info("failed to kmalloc\n");
3951
3952         schedule_on_each_cpu(test_work);
3953
3954         kfree(test_malloc);
3955
3956         set_current_state(TASK_INTERRUPTIBLE);
3957         while (!kthread_should_stop()) {
3958                 schedule();
3959                 set_current_state(TASK_INTERRUPTIBLE);
3960         }
3961         __set_current_state(TASK_RUNNING);
3962
3963         return 0;
3964 }
3965
3966 /*
3967  * Do various things that may trigger events.
3968  */
3969 static __init void event_test_stuff(void)
3970 {
3971         struct task_struct *test_thread;
3972
3973         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3974         msleep(1);
3975         kthread_stop(test_thread);
3976 }
3977
3978 /*
3979  * For every trace event defined, we will test each trace point separately,
3980  * and then by groups, and finally all trace points.
3981  */
3982 static __init void event_trace_self_tests(void)
3983 {
3984         struct trace_subsystem_dir *dir;
3985         struct trace_event_file *file;
3986         struct trace_event_call *call;
3987         struct event_subsystem *system;
3988         struct trace_array *tr;
3989         int ret;
3990
3991         tr = top_trace_array();
3992         if (!tr)
3993                 return;
3994
3995         pr_info("Running tests on trace events:\n");
3996
3997         list_for_each_entry(file, &tr->events, list) {
3998
3999                 call = file->event_call;
4000
4001                 /* Only test those that have a probe */
4002                 if (!call->class || !call->class->probe)
4003                         continue;
4004
4005 /*
4006  * Testing syscall events here is pretty useless, but
4007  * we still do it if configured. But this is time consuming.
4008  * What we really need is a user thread to perform the
4009  * syscalls as we test.
4010  */
4011 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4012                 if (call->class->system &&
4013                     strcmp(call->class->system, "syscalls") == 0)
4014                         continue;
4015 #endif
4016
4017                 pr_info("Testing event %s: ", trace_event_name(call));
4018
4019                 /*
4020                  * If an event is already enabled, someone is using
4021                  * it and the self test should not be on.
4022                  */
4023                 if (file->flags & EVENT_FILE_FL_ENABLED) {
4024                         pr_warn("Enabled event during self test!\n");
4025                         WARN_ON_ONCE(1);
4026                         continue;
4027                 }
4028
4029                 ftrace_event_enable_disable(file, 1);
4030                 event_test_stuff();
4031                 ftrace_event_enable_disable(file, 0);
4032
4033                 pr_cont("OK\n");
4034         }
4035
4036         /* Now test at the sub system level */
4037
4038         pr_info("Running tests on trace event systems:\n");
4039
4040         list_for_each_entry(dir, &tr->systems, list) {
4041
4042                 system = dir->subsystem;
4043
4044                 /* the ftrace system is special, skip it */
4045                 if (strcmp(system->name, "ftrace") == 0)
4046                         continue;
4047
4048                 pr_info("Testing event system %s: ", system->name);
4049
4050                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4051                 if (WARN_ON_ONCE(ret)) {
4052                         pr_warn("error enabling system %s\n",
4053                                 system->name);
4054                         continue;
4055                 }
4056
4057                 event_test_stuff();
4058
4059                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4060                 if (WARN_ON_ONCE(ret)) {
4061                         pr_warn("error disabling system %s\n",
4062                                 system->name);
4063                         continue;
4064                 }
4065
4066                 pr_cont("OK\n");
4067         }
4068
4069         /* Test with all events enabled */
4070
4071         pr_info("Running tests on all trace events:\n");
4072         pr_info("Testing all events: ");
4073
4074         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4075         if (WARN_ON_ONCE(ret)) {
4076                 pr_warn("error enabling all events\n");
4077                 return;
4078         }
4079
4080         event_test_stuff();
4081
4082         /* reset sysname */
4083         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4084         if (WARN_ON_ONCE(ret)) {
4085                 pr_warn("error disabling all events\n");
4086                 return;
4087         }
4088
4089         pr_cont("OK\n");
4090 }
4091
4092 #ifdef CONFIG_FUNCTION_TRACER
4093
4094 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4095
4096 static struct trace_event_file event_trace_file __initdata;
4097
4098 static void __init
4099 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4100                           struct ftrace_ops *op, struct ftrace_regs *regs)
4101 {
4102         struct trace_buffer *buffer;
4103         struct ring_buffer_event *event;
4104         struct ftrace_entry *entry;
4105         unsigned int trace_ctx;
4106         long disabled;
4107         int cpu;
4108
4109         trace_ctx = tracing_gen_ctx();
4110         preempt_disable_notrace();
4111         cpu = raw_smp_processor_id();
4112         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4113
4114         if (disabled != 1)
4115                 goto out;
4116
4117         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4118                                                 TRACE_FN, sizeof(*entry),
4119                                                 trace_ctx);
4120         if (!event)
4121                 goto out;
4122         entry   = ring_buffer_event_data(event);
4123         entry->ip                       = ip;
4124         entry->parent_ip                = parent_ip;
4125
4126         event_trigger_unlock_commit(&event_trace_file, buffer, event,
4127                                     entry, trace_ctx);
4128  out:
4129         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4130         preempt_enable_notrace();
4131 }
4132
4133 static struct ftrace_ops trace_ops __initdata  =
4134 {
4135         .func = function_test_events_call,
4136 };
4137
4138 static __init void event_trace_self_test_with_function(void)
4139 {
4140         int ret;
4141
4142         event_trace_file.tr = top_trace_array();
4143         if (WARN_ON(!event_trace_file.tr))
4144                 return;
4145
4146         ret = register_ftrace_function(&trace_ops);
4147         if (WARN_ON(ret < 0)) {
4148                 pr_info("Failed to enable function tracer for event tests\n");
4149                 return;
4150         }
4151         pr_info("Running tests again, along with the function tracer\n");
4152         event_trace_self_tests();
4153         unregister_ftrace_function(&trace_ops);
4154 }
4155 #else
4156 static __init void event_trace_self_test_with_function(void)
4157 {
4158 }
4159 #endif
4160
4161 static __init int event_trace_self_tests_init(void)
4162 {
4163         if (!tracing_selftest_disabled) {
4164                 event_trace_self_tests();
4165                 event_trace_self_test_with_function();
4166         }
4167
4168         return 0;
4169 }
4170
4171 late_initcall(event_trace_self_tests_init);
4172
4173 #endif