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                 tracefs_remove(dir->entry);
 988                 list_del(&dir->list);
 989                 __put_system_dir(dir);
 990         }
 991 }
 992
 993 static void remove_event_file_dir(struct trace_event_file *file)
 994 {
 995         struct dentry *dir = file->dir;
 996         struct dentry *child;
 997
 998         if (dir) {
 999                 spin_lock(&dir->d_lock);        /* probably unneeded */
1000                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
1001                         if (d_really_is_positive(child))        /* probably unneeded */
1002                                 d_inode(child)->i_private = NULL;
1003                 }
1004                 spin_unlock(&dir->d_lock);
1005
1006                 tracefs_remove(dir);
1007         }
1008
1009         list_del(&file->list);
1010         remove_subsystem(file->system);
1011         free_event_filter(file->filter);
1012         kmem_cache_free(file_cachep, file);
1013 }
1014
1015 /*
1016  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1017  */
1018 static int
1019 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1020                               const char *sub, const char *event, int set)
1021 {
1022         struct trace_event_file *file;
1023         struct trace_event_call *call;
1024         const char *name;
1025         int ret = -EINVAL;
1026         int eret = 0;
1027
1028         list_for_each_entry(file, &tr->events, list) {
1029
1030                 call = file->event_call;
1031                 name = trace_event_name(call);
1032
1033                 if (!name || !call->class || !call->class->reg)
1034                         continue;
1035
1036                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1037                         continue;
1038
1039                 if (match &&
1040                     strcmp(match, name) != 0 &&
1041                     strcmp(match, call->class->system) != 0)
1042                         continue;
1043
1044                 if (sub && strcmp(sub, call->class->system) != 0)
1045                         continue;
1046
1047                 if (event && strcmp(event, name) != 0)
1048                         continue;
1049
1050                 ret = ftrace_event_enable_disable(file, set);
1051
1052                 /*
1053                  * Save the first error and return that. Some events
1054                  * may still have been enabled, but let the user
1055                  * know that something went wrong.
1056                  */
1057                 if (ret && !eret)
1058                         eret = ret;
1059
1060                 ret = eret;
1061         }
1062
1063         return ret;
1064 }
1065
1066 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1067                                   const char *sub, const char *event, int set)
1068 {
1069         int ret;
1070
1071         mutex_lock(&event_mutex);
1072         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1073         mutex_unlock(&event_mutex);
1074
1075         return ret;
1076 }
1077
1078 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1079 {
1080         char *event = NULL, *sub = NULL, *match;
1081         int ret;
1082
1083         if (!tr)
1084                 return -ENOENT;
1085         /*
1086          * The buf format can be <subsystem>:<event-name>
1087          *  *:<event-name> means any event by that name.
1088          *  :<event-name> is the same.
1089          *
1090          *  <subsystem>:* means all events in that subsystem
1091          *  <subsystem>: means the same.
1092          *
1093          *  <name> (no ':') means all events in a subsystem with
1094          *  the name <name> or any event that matches <name>
1095          */
1096
1097         match = strsep(&buf, ":");
1098         if (buf) {
1099                 sub = match;
1100                 event = buf;
1101                 match = NULL;
1102
1103                 if (!strlen(sub) || strcmp(sub, "*") == 0)
1104                         sub = NULL;
1105                 if (!strlen(event) || strcmp(event, "*") == 0)
1106                         event = NULL;
1107         }
1108
1109         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1110
1111         /* Put back the colon to allow this to be called again */
1112         if (buf)
1113                 *(buf - 1) = ':';
1114
1115         return ret;
1116 }
1117
1118 /**
1119  * trace_set_clr_event - enable or disable an event
1120  * @system: system name to match (NULL for any system)
1121  * @event: event name to match (NULL for all events, within system)
1122  * @set: 1 to enable, 0 to disable
1123  *
1124  * This is a way for other parts of the kernel to enable or disable
1125  * event recording.
1126  *
1127  * Returns 0 on success, -EINVAL if the parameters do not match any
1128  * registered events.
1129  */
1130 int trace_set_clr_event(const char *system, const char *event, int set)
1131 {
1132         struct trace_array *tr = top_trace_array();
1133
1134         if (!tr)
1135                 return -ENODEV;
1136
1137         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1138 }
1139 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1140
1141 /**
1142  * trace_array_set_clr_event - enable or disable an event for a trace array.
1143  * @tr: concerned trace array.
1144  * @system: system name to match (NULL for any system)
1145  * @event: event name to match (NULL for all events, within system)
1146  * @enable: true to enable, false to disable
1147  *
1148  * This is a way for other parts of the kernel to enable or disable
1149  * event recording.
1150  *
1151  * Returns 0 on success, -EINVAL if the parameters do not match any
1152  * registered events.
1153  */
1154 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1155                 const char *event, bool enable)
1156 {
1157         int set;
1158
1159         if (!tr)
1160                 return -ENOENT;
1161
1162         set = (enable == true) ? 1 : 0;
1163         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1164 }
1165 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1166
1167 /* 128 should be much more than enough */
1168 #define EVENT_BUF_SIZE          127
1169
1170 static ssize_t
1171 ftrace_event_write(struct file *file, const char __user *ubuf,
1172                    size_t cnt, loff_t *ppos)
1173 {
1174         struct trace_parser parser;
1175         struct seq_file *m = file->private_data;
1176         struct trace_array *tr = m->private;
1177         ssize_t read, ret;
1178
1179         if (!cnt)
1180                 return 0;
1181
1182         ret = tracing_update_buffers();
1183         if (ret < 0)
1184                 return ret;
1185
1186         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1187                 return -ENOMEM;
1188
1189         read = trace_get_user(&parser, ubuf, cnt, ppos);
1190
1191         if (read >= 0 && trace_parser_loaded((&parser))) {
1192                 int set = 1;
1193
1194                 if (*parser.buffer == '!')
1195                         set = 0;
1196
1197                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1198                 if (ret)
1199                         goto out_put;
1200         }
1201
1202         ret = read;
1203
1204  out_put:
1205         trace_parser_put(&parser);
1206
1207         return ret;
1208 }
1209
1210 static void *
1211 t_next(struct seq_file *m, void *v, loff_t *pos)
1212 {
1213         struct trace_event_file *file = v;
1214         struct trace_event_call *call;
1215         struct trace_array *tr = m->private;
1216
1217         (*pos)++;
1218
1219         list_for_each_entry_continue(file, &tr->events, list) {
1220                 call = file->event_call;
1221                 /*
1222                  * The ftrace subsystem is for showing formats only.
1223                  * They can not be enabled or disabled via the event files.
1224                  */
1225                 if (call->class && call->class->reg &&
1226                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1227                         return file;
1228         }
1229
1230         return NULL;
1231 }
1232
1233 static void *t_start(struct seq_file *m, loff_t *pos)
1234 {
1235         struct trace_event_file *file;
1236         struct trace_array *tr = m->private;
1237         loff_t l;
1238
1239         mutex_lock(&event_mutex);
1240
1241         file = list_entry(&tr->events, struct trace_event_file, list);
1242         for (l = 0; l <= *pos; ) {
1243                 file = t_next(m, file, &l);
1244                 if (!file)
1245                         break;
1246         }
1247         return file;
1248 }
1249
1250 static void *
1251 s_next(struct seq_file *m, void *v, loff_t *pos)
1252 {
1253         struct trace_event_file *file = v;
1254         struct trace_array *tr = m->private;
1255
1256         (*pos)++;
1257
1258         list_for_each_entry_continue(file, &tr->events, list) {
1259                 if (file->flags & EVENT_FILE_FL_ENABLED)
1260                         return file;
1261         }
1262
1263         return NULL;
1264 }
1265
1266 static void *s_start(struct seq_file *m, loff_t *pos)
1267 {
1268         struct trace_event_file *file;
1269         struct trace_array *tr = m->private;
1270         loff_t l;
1271
1272         mutex_lock(&event_mutex);
1273
1274         file = list_entry(&tr->events, struct trace_event_file, list);
1275         for (l = 0; l <= *pos; ) {
1276                 file = s_next(m, file, &l);
1277                 if (!file)
1278                         break;
1279         }
1280         return file;
1281 }
1282
1283 static int t_show(struct seq_file *m, void *v)
1284 {
1285         struct trace_event_file *file = v;
1286         struct trace_event_call *call = file->event_call;
1287
1288         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1289                 seq_printf(m, "%s:", call->class->system);
1290         seq_printf(m, "%s\n", trace_event_name(call));
1291
1292         return 0;
1293 }
1294
1295 static void t_stop(struct seq_file *m, void *p)
1296 {
1297         mutex_unlock(&event_mutex);
1298 }
1299
1300 static void *
1301 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1302 {
1303         struct trace_array *tr = m->private;
1304         struct trace_pid_list *pid_list;
1305
1306         if (type == TRACE_PIDS)
1307                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1308         else
1309                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1310
1311         return trace_pid_next(pid_list, v, pos);
1312 }
1313
1314 static void *
1315 p_next(struct seq_file *m, void *v, loff_t *pos)
1316 {
1317         return __next(m, v, pos, TRACE_PIDS);
1318 }
1319
1320 static void *
1321 np_next(struct seq_file *m, void *v, loff_t *pos)
1322 {
1323         return __next(m, v, pos, TRACE_NO_PIDS);
1324 }
1325
1326 static void *__start(struct seq_file *m, loff_t *pos, int type)
1327         __acquires(RCU)
1328 {
1329         struct trace_pid_list *pid_list;
1330         struct trace_array *tr = m->private;
1331
1332         /*
1333          * Grab the mutex, to keep calls to p_next() having the same
1334          * tr->filtered_pids as p_start() has.
1335          * If we just passed the tr->filtered_pids around, then RCU would
1336          * have been enough, but doing that makes things more complex.
1337          */
1338         mutex_lock(&event_mutex);
1339         rcu_read_lock_sched();
1340
1341         if (type == TRACE_PIDS)
1342                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1343         else
1344                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1345
1346         if (!pid_list)
1347                 return NULL;
1348
1349         return trace_pid_start(pid_list, pos);
1350 }
1351
1352 static void *p_start(struct seq_file *m, loff_t *pos)
1353         __acquires(RCU)
1354 {
1355         return __start(m, pos, TRACE_PIDS);
1356 }
1357
1358 static void *np_start(struct seq_file *m, loff_t *pos)
1359         __acquires(RCU)
1360 {
1361         return __start(m, pos, TRACE_NO_PIDS);
1362 }
1363
1364 static void p_stop(struct seq_file *m, void *p)
1365         __releases(RCU)
1366 {
1367         rcu_read_unlock_sched();
1368         mutex_unlock(&event_mutex);
1369 }
1370
1371 static ssize_t
1372 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1373                   loff_t *ppos)
1374 {
1375         struct trace_event_file *file;
1376         unsigned long flags;
1377         char buf[4] = "0";
1378
1379         mutex_lock(&event_mutex);
1380         file = event_file_data(filp);
1381         if (likely(file))
1382                 flags = file->flags;
1383         mutex_unlock(&event_mutex);
1384
1385         if (!file)
1386                 return -ENODEV;
1387
1388         if (flags & EVENT_FILE_FL_ENABLED &&
1389             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1390                 strcpy(buf, "1");
1391
1392         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1393             flags & EVENT_FILE_FL_SOFT_MODE)
1394                 strcat(buf, "*");
1395
1396         strcat(buf, "\n");
1397
1398         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1399 }
1400
1401 static ssize_t
1402 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1403                    loff_t *ppos)
1404 {
1405         struct trace_event_file *file;
1406         unsigned long val;
1407         int ret;
1408
1409         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1410         if (ret)
1411                 return ret;
1412
1413         ret = tracing_update_buffers();
1414         if (ret < 0)
1415                 return ret;
1416
1417         switch (val) {
1418         case 0:
1419         case 1:
1420                 ret = -ENODEV;
1421                 mutex_lock(&event_mutex);
1422                 file = event_file_data(filp);
1423                 if (likely(file))
1424                         ret = ftrace_event_enable_disable(file, val);
1425                 mutex_unlock(&event_mutex);
1426                 break;
1427
1428         default:
1429                 return -EINVAL;
1430         }
1431
1432         *ppos += cnt;
1433
1434         return ret ? ret : cnt;
1435 }
1436
1437 static ssize_t
1438 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1439                    loff_t *ppos)
1440 {
1441         const char set_to_char[4] = { '?', '0', '1', 'X' };
1442         struct trace_subsystem_dir *dir = filp->private_data;
1443         struct event_subsystem *system = dir->subsystem;
1444         struct trace_event_call *call;
1445         struct trace_event_file *file;
1446         struct trace_array *tr = dir->tr;
1447         char buf[2];
1448         int set = 0;
1449         int ret;
1450
1451         mutex_lock(&event_mutex);
1452         list_for_each_entry(file, &tr->events, list) {
1453                 call = file->event_call;
1454                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1455                     !trace_event_name(call) || !call->class || !call->class->reg)
1456                         continue;
1457
1458                 if (system && strcmp(call->class->system, system->name) != 0)
1459                         continue;
1460
1461                 /*
1462                  * We need to find out if all the events are set
1463                  * or if all events or cleared, or if we have
1464                  * a mixture.
1465                  */
1466                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1467
1468                 /*
1469                  * If we have a mixture, no need to look further.
1470                  */
1471                 if (set == 3)
1472                         break;
1473         }
1474         mutex_unlock(&event_mutex);
1475
1476         buf[0] = set_to_char[set];
1477         buf[1] = '\n';
1478
1479         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1480
1481         return ret;
1482 }
1483
1484 static ssize_t
1485 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1486                     loff_t *ppos)
1487 {
1488         struct trace_subsystem_dir *dir = filp->private_data;
1489         struct event_subsystem *system = dir->subsystem;
1490         const char *name = NULL;
1491         unsigned long val;
1492         ssize_t ret;
1493
1494         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1495         if (ret)
1496                 return ret;
1497
1498         ret = tracing_update_buffers();
1499         if (ret < 0)
1500                 return ret;
1501
1502         if (val != 0 && val != 1)
1503                 return -EINVAL;
1504
1505         /*
1506          * Opening of "enable" adds a ref count to system,
1507          * so the name is safe to use.
1508          */
1509         if (system)
1510                 name = system->name;
1511
1512         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1513         if (ret)
1514                 goto out;
1515
1516         ret = cnt;
1517
1518 out:
1519         *ppos += cnt;
1520
1521         return ret;
1522 }
1523
1524 enum {
1525         FORMAT_HEADER           = 1,
1526         FORMAT_FIELD_SEPERATOR  = 2,
1527         FORMAT_PRINTFMT         = 3,
1528 };
1529
1530 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1531 {
1532         struct trace_event_call *call = event_file_data(m->private);
1533         struct list_head *common_head = &ftrace_common_fields;
1534         struct list_head *head = trace_get_fields(call);
1535         struct list_head *node = v;
1536
1537         (*pos)++;
1538
1539         switch ((unsigned long)v) {
1540         case FORMAT_HEADER:
1541                 node = common_head;
1542                 break;
1543
1544         case FORMAT_FIELD_SEPERATOR:
1545                 node = head;
1546                 break;
1547
1548         case FORMAT_PRINTFMT:
1549                 /* all done */
1550                 return NULL;
1551         }
1552
1553         node = node->prev;
1554         if (node == common_head)
1555                 return (void *)FORMAT_FIELD_SEPERATOR;
1556         else if (node == head)
1557                 return (void *)FORMAT_PRINTFMT;
1558         else
1559                 return node;
1560 }
1561
1562 static int f_show(struct seq_file *m, void *v)
1563 {
1564         struct trace_event_call *call = event_file_data(m->private);
1565         struct ftrace_event_field *field;
1566         const char *array_descriptor;
1567
1568         switch ((unsigned long)v) {
1569         case FORMAT_HEADER:
1570                 seq_printf(m, "name: %s\n", trace_event_name(call));
1571                 seq_printf(m, "ID: %d\n", call->event.type);
1572                 seq_puts(m, "format:\n");
1573                 return 0;
1574
1575         case FORMAT_FIELD_SEPERATOR:
1576                 seq_putc(m, '\n');
1577                 return 0;
1578
1579         case FORMAT_PRINTFMT:
1580                 seq_printf(m, "\nprint fmt: %s\n",
1581                            call->print_fmt);
1582                 return 0;
1583         }
1584
1585         field = list_entry(v, struct ftrace_event_field, link);
1586         /*
1587          * Smartly shows the array type(except dynamic array).
1588          * Normal:
1589          *      field:TYPE VAR
1590          * If TYPE := TYPE[LEN], it is shown:
1591          *      field:TYPE VAR[LEN]
1592          */
1593         array_descriptor = strchr(field->type, '[');
1594
1595         if (str_has_prefix(field->type, "__data_loc"))
1596                 array_descriptor = NULL;
1597
1598         if (!array_descriptor)
1599                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1600                            field->type, field->name, field->offset,
1601                            field->size, !!field->is_signed);
1602         else if (field->len)
1603                 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1604                            (int)(array_descriptor - field->type),
1605                            field->type, field->name,
1606                            field->len, field->offset,
1607                            field->size, !!field->is_signed);
1608         else
1609                 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1610                                 (int)(array_descriptor - field->type),
1611                                 field->type, field->name,
1612                                 field->offset, field->size, !!field->is_signed);
1613
1614         return 0;
1615 }
1616
1617 static void *f_start(struct seq_file *m, loff_t *pos)
1618 {
1619         void *p = (void *)FORMAT_HEADER;
1620         loff_t l = 0;
1621
1622         /* ->stop() is called even if ->start() fails */
1623         mutex_lock(&event_mutex);
1624         if (!event_file_data(m->private))
1625                 return ERR_PTR(-ENODEV);
1626
1627         while (l < *pos && p)
1628                 p = f_next(m, p, &l);
1629
1630         return p;
1631 }
1632
1633 static void f_stop(struct seq_file *m, void *p)
1634 {
1635         mutex_unlock(&event_mutex);
1636 }
1637
1638 static const struct seq_operations trace_format_seq_ops = {
1639         .start          = f_start,
1640         .next           = f_next,
1641         .stop           = f_stop,
1642         .show           = f_show,
1643 };
1644
1645 static int trace_format_open(struct inode *inode, struct file *file)
1646 {
1647         struct seq_file *m;
1648         int ret;
1649
1650         /* Do we want to hide event format files on tracefs lockdown? */
1651
1652         ret = seq_open(file, &trace_format_seq_ops);
1653         if (ret < 0)
1654                 return ret;
1655
1656         m = file->private_data;
1657         m->private = file;
1658
1659         return 0;
1660 }
1661
1662 static ssize_t
1663 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1664 {
1665         int id = (long)event_file_data(filp);
1666         char buf[32];
1667         int len;
1668
1669         if (unlikely(!id))
1670                 return -ENODEV;
1671
1672         len = sprintf(buf, "%d\n", id);
1673
1674         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1675 }
1676
1677 static ssize_t
1678 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1679                   loff_t *ppos)
1680 {
1681         struct trace_event_file *file;
1682         struct trace_seq *s;
1683         int r = -ENODEV;
1684
1685         if (*ppos)
1686                 return 0;
1687
1688         s = kmalloc(sizeof(*s), GFP_KERNEL);
1689
1690         if (!s)
1691                 return -ENOMEM;
1692
1693         trace_seq_init(s);
1694
1695         mutex_lock(&event_mutex);
1696         file = event_file_data(filp);
1697         if (file)
1698                 print_event_filter(file, s);
1699         mutex_unlock(&event_mutex);
1700
1701         if (file)
1702                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1703                                             s->buffer, trace_seq_used(s));
1704
1705         kfree(s);
1706
1707         return r;
1708 }
1709
1710 static ssize_t
1711 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1712                    loff_t *ppos)
1713 {
1714         struct trace_event_file *file;
1715         char *buf;
1716         int err = -ENODEV;
1717
1718         if (cnt >= PAGE_SIZE)
1719                 return -EINVAL;
1720
1721         buf = memdup_user_nul(ubuf, cnt);
1722         if (IS_ERR(buf))
1723                 return PTR_ERR(buf);
1724
1725         mutex_lock(&event_mutex);
1726         file = event_file_data(filp);
1727         if (file)
1728                 err = apply_event_filter(file, buf);
1729         mutex_unlock(&event_mutex);
1730
1731         kfree(buf);
1732         if (err < 0)
1733                 return err;
1734
1735         *ppos += cnt;
1736
1737         return cnt;
1738 }
1739
1740 static LIST_HEAD(event_subsystems);
1741
1742 static int subsystem_open(struct inode *inode, struct file *filp)
1743 {
1744         struct trace_subsystem_dir *dir = NULL, *iter_dir;
1745         struct trace_array *tr = NULL, *iter_tr;
1746         struct event_subsystem *system = NULL;
1747         int ret;
1748
1749         if (tracing_is_disabled())
1750                 return -ENODEV;
1751
1752         /* Make sure the system still exists */
1753         mutex_lock(&event_mutex);
1754         mutex_lock(&trace_types_lock);
1755         list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1756                 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1757                         if (iter_dir == inode->i_private) {
1758                                 /* Don't open systems with no events */
1759                                 tr = iter_tr;
1760                                 dir = iter_dir;
1761                                 if (dir->nr_events) {
1762                                         __get_system_dir(dir);
1763                                         system = dir->subsystem;
1764                                 }
1765                                 goto exit_loop;
1766                         }
1767                 }
1768         }
1769  exit_loop:
1770         mutex_unlock(&trace_types_lock);
1771         mutex_unlock(&event_mutex);
1772
1773         if (!system)
1774                 return -ENODEV;
1775
1776         /* Still need to increment the ref count of the system */
1777         if (trace_array_get(tr) < 0) {
1778                 put_system(dir);
1779                 return -ENODEV;
1780         }
1781
1782         ret = tracing_open_generic(inode, filp);
1783         if (ret < 0) {
1784                 trace_array_put(tr);
1785                 put_system(dir);
1786         }
1787
1788         return ret;
1789 }
1790
1791 static int system_tr_open(struct inode *inode, struct file *filp)
1792 {
1793         struct trace_subsystem_dir *dir;
1794         struct trace_array *tr = inode->i_private;
1795         int ret;
1796
1797         /* Make a temporary dir that has no system but points to tr */
1798         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1799         if (!dir)
1800                 return -ENOMEM;
1801
1802         ret = tracing_open_generic_tr(inode, filp);
1803         if (ret < 0) {
1804                 kfree(dir);
1805                 return ret;
1806         }
1807         dir->tr = tr;
1808         filp->private_data = dir;
1809
1810         return 0;
1811 }
1812
1813 static int subsystem_release(struct inode *inode, struct file *file)
1814 {
1815         struct trace_subsystem_dir *dir = file->private_data;
1816
1817         trace_array_put(dir->tr);
1818
1819         /*
1820          * If dir->subsystem is NULL, then this is a temporary
1821          * descriptor that was made for a trace_array to enable
1822          * all subsystems.
1823          */
1824         if (dir->subsystem)
1825                 put_system(dir);
1826         else
1827                 kfree(dir);
1828
1829         return 0;
1830 }
1831
1832 static ssize_t
1833 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1834                       loff_t *ppos)
1835 {
1836         struct trace_subsystem_dir *dir = filp->private_data;
1837         struct event_subsystem *system = dir->subsystem;
1838         struct trace_seq *s;
1839         int r;
1840
1841         if (*ppos)
1842                 return 0;
1843
1844         s = kmalloc(sizeof(*s), GFP_KERNEL);
1845         if (!s)
1846                 return -ENOMEM;
1847
1848         trace_seq_init(s);
1849
1850         print_subsystem_event_filter(system, s);
1851         r = simple_read_from_buffer(ubuf, cnt, ppos,
1852                                     s->buffer, trace_seq_used(s));
1853
1854         kfree(s);
1855
1856         return r;
1857 }
1858
1859 static ssize_t
1860 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1861                        loff_t *ppos)
1862 {
1863         struct trace_subsystem_dir *dir = filp->private_data;
1864         char *buf;
1865         int err;
1866
1867         if (cnt >= PAGE_SIZE)
1868                 return -EINVAL;
1869
1870         buf = memdup_user_nul(ubuf, cnt);
1871         if (IS_ERR(buf))
1872                 return PTR_ERR(buf);
1873
1874         err = apply_subsystem_event_filter(dir, buf);
1875         kfree(buf);
1876         if (err < 0)
1877                 return err;
1878
1879         *ppos += cnt;
1880
1881         return cnt;
1882 }
1883
1884 static ssize_t
1885 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1886 {
1887         int (*func)(struct trace_seq *s) = filp->private_data;
1888         struct trace_seq *s;
1889         int r;
1890
1891         if (*ppos)
1892                 return 0;
1893
1894         s = kmalloc(sizeof(*s), GFP_KERNEL);
1895         if (!s)
1896                 return -ENOMEM;
1897
1898         trace_seq_init(s);
1899
1900         func(s);
1901         r = simple_read_from_buffer(ubuf, cnt, ppos,
1902                                     s->buffer, trace_seq_used(s));
1903
1904         kfree(s);
1905
1906         return r;
1907 }
1908
1909 static void ignore_task_cpu(void *data)
1910 {
1911         struct trace_array *tr = data;
1912         struct trace_pid_list *pid_list;
1913         struct trace_pid_list *no_pid_list;
1914
1915         /*
1916          * This function is called by on_each_cpu() while the
1917          * event_mutex is held.
1918          */
1919         pid_list = rcu_dereference_protected(tr->filtered_pids,
1920                                              mutex_is_locked(&event_mutex));
1921         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1922                                              mutex_is_locked(&event_mutex));
1923
1924         this_cpu_write(tr->array_buffer.data->ignore_pid,
1925                        trace_ignore_this_task(pid_list, no_pid_list, current));
1926 }
1927
1928 static void register_pid_events(struct trace_array *tr)
1929 {
1930         /*
1931          * Register a probe that is called before all other probes
1932          * to set ignore_pid if next or prev do not match.
1933          * Register a probe this is called after all other probes
1934          * to only keep ignore_pid set if next pid matches.
1935          */
1936         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1937                                          tr, INT_MAX);
1938         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1939                                          tr, 0);
1940
1941         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1942                                          tr, INT_MAX);
1943         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1944                                          tr, 0);
1945
1946         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1947                                              tr, INT_MAX);
1948         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1949                                              tr, 0);
1950
1951         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1952                                          tr, INT_MAX);
1953         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1954                                          tr, 0);
1955 }
1956
1957 static ssize_t
1958 event_pid_write(struct file *filp, const char __user *ubuf,
1959                 size_t cnt, loff_t *ppos, int type)
1960 {
1961         struct seq_file *m = filp->private_data;
1962         struct trace_array *tr = m->private;
1963         struct trace_pid_list *filtered_pids = NULL;
1964         struct trace_pid_list *other_pids = NULL;
1965         struct trace_pid_list *pid_list;
1966         struct trace_event_file *file;
1967         ssize_t ret;
1968
1969         if (!cnt)
1970                 return 0;
1971
1972         ret = tracing_update_buffers();
1973         if (ret < 0)
1974                 return ret;
1975
1976         mutex_lock(&event_mutex);
1977
1978         if (type == TRACE_PIDS) {
1979                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1980                                                           lockdep_is_held(&event_mutex));
1981                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1982                                                           lockdep_is_held(&event_mutex));
1983         } else {
1984                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1985                                                           lockdep_is_held(&event_mutex));
1986                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1987                                                           lockdep_is_held(&event_mutex));
1988         }
1989
1990         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1991         if (ret < 0)
1992                 goto out;
1993
1994         if (type == TRACE_PIDS)
1995                 rcu_assign_pointer(tr->filtered_pids, pid_list);
1996         else
1997                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1998
1999         list_for_each_entry(file, &tr->events, list) {
2000                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2001         }
2002
2003         if (filtered_pids) {
2004                 tracepoint_synchronize_unregister();
2005                 trace_pid_list_free(filtered_pids);
2006         } else if (pid_list && !other_pids) {
2007                 register_pid_events(tr);
2008         }
2009
2010         /*
2011          * Ignoring of pids is done at task switch. But we have to
2012          * check for those tasks that are currently running.
2013          * Always do this in case a pid was appended or removed.
2014          */
2015         on_each_cpu(ignore_task_cpu, tr, 1);
2016
2017  out:
2018         mutex_unlock(&event_mutex);
2019
2020         if (ret > 0)
2021                 *ppos += ret;
2022
2023         return ret;
2024 }
2025
2026 static ssize_t
2027 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2028                        size_t cnt, loff_t *ppos)
2029 {
2030         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2031 }
2032
2033 static ssize_t
2034 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2035                         size_t cnt, loff_t *ppos)
2036 {
2037         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2038 }
2039
2040 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2041 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2042 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2043 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2044 static int ftrace_event_release(struct inode *inode, struct file *file);
2045
2046 static const struct seq_operations show_event_seq_ops = {
2047         .start = t_start,
2048         .next = t_next,
2049         .show = t_show,
2050         .stop = t_stop,
2051 };
2052
2053 static const struct seq_operations show_set_event_seq_ops = {
2054         .start = s_start,
2055         .next = s_next,
2056         .show = t_show,
2057         .stop = t_stop,
2058 };
2059
2060 static const struct seq_operations show_set_pid_seq_ops = {
2061         .start = p_start,
2062         .next = p_next,
2063         .show = trace_pid_show,
2064         .stop = p_stop,
2065 };
2066
2067 static const struct seq_operations show_set_no_pid_seq_ops = {
2068         .start = np_start,
2069         .next = np_next,
2070         .show = trace_pid_show,
2071         .stop = p_stop,
2072 };
2073
2074 static const struct file_operations ftrace_avail_fops = {
2075         .open = ftrace_event_avail_open,
2076         .read = seq_read,
2077         .llseek = seq_lseek,
2078         .release = seq_release,
2079 };
2080
2081 static const struct file_operations ftrace_set_event_fops = {
2082         .open = ftrace_event_set_open,
2083         .read = seq_read,
2084         .write = ftrace_event_write,
2085         .llseek = seq_lseek,
2086         .release = ftrace_event_release,
2087 };
2088
2089 static const struct file_operations ftrace_set_event_pid_fops = {
2090         .open = ftrace_event_set_pid_open,
2091         .read = seq_read,
2092         .write = ftrace_event_pid_write,
2093         .llseek = seq_lseek,
2094         .release = ftrace_event_release,
2095 };
2096
2097 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2098         .open = ftrace_event_set_npid_open,
2099         .read = seq_read,
2100         .write = ftrace_event_npid_write,
2101         .llseek = seq_lseek,
2102         .release = ftrace_event_release,
2103 };
2104
2105 static const struct file_operations ftrace_enable_fops = {
2106         .open = tracing_open_generic,
2107         .read = event_enable_read,
2108         .write = event_enable_write,
2109         .llseek = default_llseek,
2110 };
2111
2112 static const struct file_operations ftrace_event_format_fops = {
2113         .open = trace_format_open,
2114         .read = seq_read,
2115         .llseek = seq_lseek,
2116         .release = seq_release,
2117 };
2118
2119 static const struct file_operations ftrace_event_id_fops = {
2120         .read = event_id_read,
2121         .llseek = default_llseek,
2122 };
2123
2124 static const struct file_operations ftrace_event_filter_fops = {
2125         .open = tracing_open_generic,
2126         .read = event_filter_read,
2127         .write = event_filter_write,
2128         .llseek = default_llseek,
2129 };
2130
2131 static const struct file_operations ftrace_subsystem_filter_fops = {
2132         .open = subsystem_open,
2133         .read = subsystem_filter_read,
2134         .write = subsystem_filter_write,
2135         .llseek = default_llseek,
2136         .release = subsystem_release,
2137 };
2138
2139 static const struct file_operations ftrace_system_enable_fops = {
2140         .open = subsystem_open,
2141         .read = system_enable_read,
2142         .write = system_enable_write,
2143         .llseek = default_llseek,
2144         .release = subsystem_release,
2145 };
2146
2147 static const struct file_operations ftrace_tr_enable_fops = {
2148         .open = system_tr_open,
2149         .read = system_enable_read,
2150         .write = system_enable_write,
2151         .llseek = default_llseek,
2152         .release = subsystem_release,
2153 };
2154
2155 static const struct file_operations ftrace_show_header_fops = {
2156         .open = tracing_open_generic,
2157         .read = show_header,
2158         .llseek = default_llseek,
2159 };
2160
2161 static int
2162 ftrace_event_open(struct inode *inode, struct file *file,
2163                   const struct seq_operations *seq_ops)
2164 {
2165         struct seq_file *m;
2166         int ret;
2167
2168         ret = security_locked_down(LOCKDOWN_TRACEFS);
2169         if (ret)
2170                 return ret;
2171
2172         ret = seq_open(file, seq_ops);
2173         if (ret < 0)
2174                 return ret;
2175         m = file->private_data;
2176         /* copy tr over to seq ops */
2177         m->private = inode->i_private;
2178
2179         return ret;
2180 }
2181
2182 static int ftrace_event_release(struct inode *inode, struct file *file)
2183 {
2184         struct trace_array *tr = inode->i_private;
2185
2186         trace_array_put(tr);
2187
2188         return seq_release(inode, file);
2189 }
2190
2191 static int
2192 ftrace_event_avail_open(struct inode *inode, struct file *file)
2193 {
2194         const struct seq_operations *seq_ops = &show_event_seq_ops;
2195
2196         /* Checks for tracefs lockdown */
2197         return ftrace_event_open(inode, file, seq_ops);
2198 }
2199
2200 static int
2201 ftrace_event_set_open(struct inode *inode, struct file *file)
2202 {
2203         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2204         struct trace_array *tr = inode->i_private;
2205         int ret;
2206
2207         ret = tracing_check_open_get_tr(tr);
2208         if (ret)
2209                 return ret;
2210
2211         if ((file->f_mode & FMODE_WRITE) &&
2212             (file->f_flags & O_TRUNC))
2213                 ftrace_clear_events(tr);
2214
2215         ret = ftrace_event_open(inode, file, seq_ops);
2216         if (ret < 0)
2217                 trace_array_put(tr);
2218         return ret;
2219 }
2220
2221 static int
2222 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2223 {
2224         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2225         struct trace_array *tr = inode->i_private;
2226         int ret;
2227
2228         ret = tracing_check_open_get_tr(tr);
2229         if (ret)
2230                 return ret;
2231
2232         if ((file->f_mode & FMODE_WRITE) &&
2233             (file->f_flags & O_TRUNC))
2234                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2235
2236         ret = ftrace_event_open(inode, file, seq_ops);
2237         if (ret < 0)
2238                 trace_array_put(tr);
2239         return ret;
2240 }
2241
2242 static int
2243 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2244 {
2245         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2246         struct trace_array *tr = inode->i_private;
2247         int ret;
2248
2249         ret = tracing_check_open_get_tr(tr);
2250         if (ret)
2251                 return ret;
2252
2253         if ((file->f_mode & FMODE_WRITE) &&
2254             (file->f_flags & O_TRUNC))
2255                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2256
2257         ret = ftrace_event_open(inode, file, seq_ops);
2258         if (ret < 0)
2259                 trace_array_put(tr);
2260         return ret;
2261 }
2262
2263 static struct event_subsystem *
2264 create_new_subsystem(const char *name)
2265 {
2266         struct event_subsystem *system;
2267
2268         /* need to create new entry */
2269         system = kmalloc(sizeof(*system), GFP_KERNEL);
2270         if (!system)
2271                 return NULL;
2272
2273         system->ref_count = 1;
2274
2275         /* Only allocate if dynamic (kprobes and modules) */
2276         system->name = kstrdup_const(name, GFP_KERNEL);
2277         if (!system->name)
2278                 goto out_free;
2279
2280         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2281         if (!system->filter)
2282                 goto out_free;
2283
2284         list_add(&system->list, &event_subsystems);
2285
2286         return system;
2287
2288  out_free:
2289         kfree_const(system->name);
2290         kfree(system);
2291         return NULL;
2292 }
2293
2294 static struct dentry *
2295 event_subsystem_dir(struct trace_array *tr, const char *name,
2296                     struct trace_event_file *file, struct dentry *parent)
2297 {
2298         struct event_subsystem *system, *iter;
2299         struct trace_subsystem_dir *dir;
2300         struct dentry *entry;
2301
2302         /* First see if we did not already create this dir */
2303         list_for_each_entry(dir, &tr->systems, list) {
2304                 system = dir->subsystem;
2305                 if (strcmp(system->name, name) == 0) {
2306                         dir->nr_events++;
2307                         file->system = dir;
2308                         return dir->entry;
2309                 }
2310         }
2311
2312         /* Now see if the system itself exists. */
2313         system = NULL;
2314         list_for_each_entry(iter, &event_subsystems, list) {
2315                 if (strcmp(iter->name, name) == 0) {
2316                         system = iter;
2317                         break;
2318                 }
2319         }
2320
2321         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2322         if (!dir)
2323                 goto out_fail;
2324
2325         if (!system) {
2326                 system = create_new_subsystem(name);
2327                 if (!system)
2328                         goto out_free;
2329         } else
2330                 __get_system(system);
2331
2332         dir->entry = tracefs_create_dir(name, parent);
2333         if (!dir->entry) {
2334                 pr_warn("Failed to create system directory %s\n", name);
2335                 __put_system(system);
2336                 goto out_free;
2337         }
2338
2339         dir->tr = tr;
2340         dir->ref_count = 1;
2341         dir->nr_events = 1;
2342         dir->subsystem = system;
2343         file->system = dir;
2344
2345         /* the ftrace system is special, do not create enable or filter files */
2346         if (strcmp(name, "ftrace") != 0) {
2347
2348                 entry = tracefs_create_file("filter", TRACE_MODE_WRITE,
2349                                             dir->entry, dir,
2350                                             &ftrace_subsystem_filter_fops);
2351                 if (!entry) {
2352                         kfree(system->filter);
2353                         system->filter = NULL;
2354                         pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2355                 }
2356
2357                 trace_create_file("enable", TRACE_MODE_WRITE, dir->entry, dir,
2358                                   &ftrace_system_enable_fops);
2359         }
2360
2361         list_add(&dir->list, &tr->systems);
2362
2363         return dir->entry;
2364
2365  out_free:
2366         kfree(dir);
2367  out_fail:
2368         /* Only print this message if failed on memory allocation */
2369         if (!dir || !system)
2370                 pr_warn("No memory to create event subsystem %s\n", name);
2371         return NULL;
2372 }
2373
2374 static int
2375 event_define_fields(struct trace_event_call *call)
2376 {
2377         struct list_head *head;
2378         int ret = 0;
2379
2380         /*
2381          * Other events may have the same class. Only update
2382          * the fields if they are not already defined.
2383          */
2384         head = trace_get_fields(call);
2385         if (list_empty(head)) {
2386                 struct trace_event_fields *field = call->class->fields_array;
2387                 unsigned int offset = sizeof(struct trace_entry);
2388
2389                 for (; field->type; field++) {
2390                         if (field->type == TRACE_FUNCTION_TYPE) {
2391                                 field->define_fields(call);
2392                                 break;
2393                         }
2394
2395                         offset = ALIGN(offset, field->align);
2396                         ret = trace_define_field_ext(call, field->type, field->name,
2397                                                  offset, field->size,
2398                                                  field->is_signed, field->filter_type,
2399                                                  field->len);
2400                         if (WARN_ON_ONCE(ret)) {
2401                                 pr_err("error code is %d\n", ret);
2402                                 break;
2403                         }
2404
2405                         offset += field->size;
2406                 }
2407         }
2408
2409         return ret;
2410 }
2411
2412 static int
2413 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2414 {
2415         struct trace_event_call *call = file->event_call;
2416         struct trace_array *tr = file->tr;
2417         struct dentry *d_events;
2418         const char *name;
2419         int ret;
2420
2421         /*
2422          * If the trace point header did not define TRACE_SYSTEM
2423          * then the system would be called "TRACE_SYSTEM". This should
2424          * never happen.
2425          */
2426         if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2427                 return -ENODEV;
2428
2429         d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2430         if (!d_events)
2431                 return -ENOMEM;
2432
2433         name = trace_event_name(call);
2434         file->dir = tracefs_create_dir(name, d_events);
2435         if (!file->dir) {
2436                 pr_warn("Could not create tracefs '%s' directory\n", name);
2437                 return -1;
2438         }
2439
2440         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2441                 trace_create_file("enable", TRACE_MODE_WRITE, file->dir, file,
2442                                   &ftrace_enable_fops);
2443
2444 #ifdef CONFIG_PERF_EVENTS
2445         if (call->event.type && call->class->reg)
2446                 trace_create_file("id", TRACE_MODE_READ, file->dir,
2447                                   (void *)(long)call->event.type,
2448                                   &ftrace_event_id_fops);
2449 #endif
2450
2451         ret = event_define_fields(call);
2452         if (ret < 0) {
2453                 pr_warn("Could not initialize trace point events/%s\n", name);
2454                 return ret;
2455         }
2456
2457         /*
2458          * Only event directories that can be enabled should have
2459          * triggers or filters.
2460          */
2461         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2462                 trace_create_file("filter", TRACE_MODE_WRITE, file->dir,
2463                                   file, &ftrace_event_filter_fops);
2464
2465                 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
2466                                   file, &event_trigger_fops);
2467         }
2468
2469 #ifdef CONFIG_HIST_TRIGGERS
2470         trace_create_file("hist", TRACE_MODE_READ, file->dir, file,
2471                           &event_hist_fops);
2472 #endif
2473 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2474         trace_create_file("hist_debug", TRACE_MODE_READ, file->dir, file,
2475                           &event_hist_debug_fops);
2476 #endif
2477         trace_create_file("format", TRACE_MODE_READ, file->dir, call,
2478                           &ftrace_event_format_fops);
2479
2480 #ifdef CONFIG_TRACE_EVENT_INJECT
2481         if (call->event.type && call->class->reg)
2482                 trace_create_file("inject", 0200, file->dir, file,
2483                                   &event_inject_fops);
2484 #endif
2485
2486         return 0;
2487 }
2488
2489 static void remove_event_from_tracers(struct trace_event_call *call)
2490 {
2491         struct trace_event_file *file;
2492         struct trace_array *tr;
2493
2494         do_for_each_event_file_safe(tr, file) {
2495                 if (file->event_call != call)
2496                         continue;
2497
2498                 remove_event_file_dir(file);
2499                 /*
2500                  * The do_for_each_event_file_safe() is
2501                  * a double loop. After finding the call for this
2502                  * trace_array, we use break to jump to the next
2503                  * trace_array.
2504                  */
2505                 break;
2506         } while_for_each_event_file();
2507 }
2508
2509 static void event_remove(struct trace_event_call *call)
2510 {
2511         struct trace_array *tr;
2512         struct trace_event_file *file;
2513
2514         do_for_each_event_file(tr, file) {
2515                 if (file->event_call != call)
2516                         continue;
2517
2518                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2519                         tr->clear_trace = true;
2520
2521                 ftrace_event_enable_disable(file, 0);
2522                 /*
2523                  * The do_for_each_event_file() is
2524                  * a double loop. After finding the call for this
2525                  * trace_array, we use break to jump to the next
2526                  * trace_array.
2527                  */
2528                 break;
2529         } while_for_each_event_file();
2530
2531         if (call->event.funcs)
2532                 __unregister_trace_event(&call->event);
2533         remove_event_from_tracers(call);
2534         list_del(&call->list);
2535 }
2536
2537 static int event_init(struct trace_event_call *call)
2538 {
2539         int ret = 0;
2540         const char *name;
2541
2542         name = trace_event_name(call);
2543         if (WARN_ON(!name))
2544                 return -EINVAL;
2545
2546         if (call->class->raw_init) {
2547                 ret = call->class->raw_init(call);
2548                 if (ret < 0 && ret != -ENOSYS)
2549                         pr_warn("Could not initialize trace events/%s\n", name);
2550         }
2551
2552         return ret;
2553 }
2554
2555 static int
2556 __register_event(struct trace_event_call *call, struct module *mod)
2557 {
2558         int ret;
2559
2560         ret = event_init(call);
2561         if (ret < 0)
2562                 return ret;
2563
2564         list_add(&call->list, &ftrace_events);
2565         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2566                 atomic_set(&call->refcnt, 0);
2567         else
2568                 call->module = mod;
2569
2570         return 0;
2571 }
2572
2573 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2574 {
2575         int rlen;
2576         int elen;
2577
2578         /* Find the length of the eval value as a string */
2579         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2580         /* Make sure there's enough room to replace the string with the value */
2581         if (len < elen)
2582                 return NULL;
2583
2584         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2585
2586         /* Get the rest of the string of ptr */
2587         rlen = strlen(ptr + len);
2588         memmove(ptr + elen, ptr + len, rlen);
2589         /* Make sure we end the new string */
2590         ptr[elen + rlen] = 0;
2591
2592         return ptr + elen;
2593 }
2594
2595 static void update_event_printk(struct trace_event_call *call,
2596                                 struct trace_eval_map *map)
2597 {
2598         char *ptr;
2599         int quote = 0;
2600         int len = strlen(map->eval_string);
2601
2602         for (ptr = call->print_fmt; *ptr; ptr++) {
2603                 if (*ptr == '\\') {
2604                         ptr++;
2605                         /* paranoid */
2606                         if (!*ptr)
2607                                 break;
2608                         continue;
2609                 }
2610                 if (*ptr == '"') {
2611                         quote ^= 1;
2612                         continue;
2613                 }
2614                 if (quote)
2615                         continue;
2616                 if (isdigit(*ptr)) {
2617                         /* skip numbers */
2618                         do {
2619                                 ptr++;
2620                                 /* Check for alpha chars like ULL */
2621                         } while (isalnum(*ptr));
2622                         if (!*ptr)
2623                                 break;
2624                         /*
2625                          * A number must have some kind of delimiter after
2626                          * it, and we can ignore that too.
2627                          */
2628                         continue;
2629                 }
2630                 if (isalpha(*ptr) || *ptr == '_') {
2631                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2632                             !isalnum(ptr[len]) && ptr[len] != '_') {
2633                                 ptr = eval_replace(ptr, map, len);
2634                                 /* enum/sizeof string smaller than value */
2635                                 if (WARN_ON_ONCE(!ptr))
2636                                         return;
2637                                 /*
2638                                  * No need to decrement here, as eval_replace()
2639                                  * returns the pointer to the character passed
2640                                  * the eval, and two evals can not be placed
2641                                  * back to back without something in between.
2642                                  * We can skip that something in between.
2643                                  */
2644                                 continue;
2645                         }
2646                 skip_more:
2647                         do {
2648                                 ptr++;
2649                         } while (isalnum(*ptr) || *ptr == '_');
2650                         if (!*ptr)
2651                                 break;
2652                         /*
2653                          * If what comes after this variable is a '.' or
2654                          * '->' then we can continue to ignore that string.
2655                          */
2656                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2657                                 ptr += *ptr == '.' ? 1 : 2;
2658                                 if (!*ptr)
2659                                         break;
2660                                 goto skip_more;
2661                         }
2662                         /*
2663                          * Once again, we can skip the delimiter that came
2664                          * after the string.
2665                          */
2666                         continue;
2667                 }
2668         }
2669 }
2670
2671 static void add_str_to_module(struct module *module, char *str)
2672 {
2673         struct module_string *modstr;
2674
2675         modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2676
2677         /*
2678          * If we failed to allocate memory here, then we'll just
2679          * let the str memory leak when the module is removed.
2680          * If this fails to allocate, there's worse problems than
2681          * a leaked string on module removal.
2682          */
2683         if (WARN_ON_ONCE(!modstr))
2684                 return;
2685
2686         modstr->module = module;
2687         modstr->str = str;
2688
2689         list_add(&modstr->next, &module_strings);
2690 }
2691
2692 static void update_event_fields(struct trace_event_call *call,
2693                                 struct trace_eval_map *map)
2694 {
2695         struct ftrace_event_field *field;
2696         struct list_head *head;
2697         char *ptr;
2698         char *str;
2699         int len = strlen(map->eval_string);
2700
2701         /* Dynamic events should never have field maps */
2702         if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2703                 return;
2704
2705         head = trace_get_fields(call);
2706         list_for_each_entry(field, head, link) {
2707                 ptr = strchr(field->type, '[');
2708                 if (!ptr)
2709                         continue;
2710                 ptr++;
2711
2712                 if (!isalpha(*ptr) && *ptr != '_')
2713                         continue;
2714
2715                 if (strncmp(map->eval_string, ptr, len) != 0)
2716                         continue;
2717
2718                 str = kstrdup(field->type, GFP_KERNEL);
2719                 if (WARN_ON_ONCE(!str))
2720                         return;
2721                 ptr = str + (ptr - field->type);
2722                 ptr = eval_replace(ptr, map, len);
2723                 /* enum/sizeof string smaller than value */
2724                 if (WARN_ON_ONCE(!ptr)) {
2725                         kfree(str);
2726                         continue;
2727                 }
2728
2729                 /*
2730                  * If the event is part of a module, then we need to free the string
2731                  * when the module is removed. Otherwise, it will stay allocated
2732                  * until a reboot.
2733                  */
2734                 if (call->module)
2735                         add_str_to_module(call->module, str);
2736
2737                 field->type = str;
2738         }
2739 }
2740
2741 void trace_event_eval_update(struct trace_eval_map **map, int len)
2742 {
2743         struct trace_event_call *call, *p;
2744         const char *last_system = NULL;
2745         bool first = false;
2746         int last_i;
2747         int i;
2748
2749         down_write(&trace_event_sem);
2750         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2751                 /* events are usually grouped together with systems */
2752                 if (!last_system || call->class->system != last_system) {
2753                         first = true;
2754                         last_i = 0;
2755                         last_system = call->class->system;
2756                 }
2757
2758                 /*
2759                  * Since calls are grouped by systems, the likelihood that the
2760                  * next call in the iteration belongs to the same system as the
2761                  * previous call is high. As an optimization, we skip searching
2762                  * for a map[] that matches the call's system if the last call
2763                  * was from the same system. That's what last_i is for. If the
2764                  * call has the same system as the previous call, then last_i
2765                  * will be the index of the first map[] that has a matching
2766                  * system.
2767                  */
2768                 for (i = last_i; i < len; i++) {
2769                         if (call->class->system == map[i]->system) {
2770                                 /* Save the first system if need be */
2771                                 if (first) {
2772                                         last_i = i;
2773                                         first = false;
2774                                 }
2775                                 update_event_printk(call, map[i]);
2776                                 update_event_fields(call, map[i]);
2777                         }
2778                 }
2779         }
2780         up_write(&trace_event_sem);
2781 }
2782
2783 static struct trace_event_file *
2784 trace_create_new_event(struct trace_event_call *call,
2785                        struct trace_array *tr)
2786 {
2787         struct trace_pid_list *no_pid_list;
2788         struct trace_pid_list *pid_list;
2789         struct trace_event_file *file;
2790         unsigned int first;
2791
2792         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2793         if (!file)
2794                 return NULL;
2795
2796         pid_list = rcu_dereference_protected(tr->filtered_pids,
2797                                              lockdep_is_held(&event_mutex));
2798         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2799                                              lockdep_is_held(&event_mutex));
2800
2801         if (!trace_pid_list_first(pid_list, &first) ||
2802             !trace_pid_list_first(no_pid_list, &first))
2803                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2804
2805         file->event_call = call;
2806         file->tr = tr;
2807         atomic_set(&file->sm_ref, 0);
2808         atomic_set(&file->tm_ref, 0);
2809         INIT_LIST_HEAD(&file->triggers);
2810         list_add(&file->list, &tr->events);
2811
2812         return file;
2813 }
2814
2815 #define MAX_BOOT_TRIGGERS 32
2816
2817 static struct boot_triggers {
2818         const char              *event;
2819         char                    *trigger;
2820 } bootup_triggers[MAX_BOOT_TRIGGERS];
2821
2822 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2823 static int nr_boot_triggers;
2824
2825 static __init int setup_trace_triggers(char *str)
2826 {
2827         char *trigger;
2828         char *buf;
2829         int i;
2830
2831         strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
2832         ring_buffer_expanded = true;
2833         disable_tracing_selftest("running event triggers");
2834
2835         buf = bootup_trigger_buf;
2836         for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
2837                 trigger = strsep(&buf, ",");
2838                 if (!trigger)
2839                         break;
2840                 bootup_triggers[i].event = strsep(&trigger, ".");
2841                 bootup_triggers[i].trigger = trigger;
2842                 if (!bootup_triggers[i].trigger)
2843                         break;
2844         }
2845
2846         nr_boot_triggers = i;
2847         return 1;
2848 }
2849 __setup("trace_trigger=", setup_trace_triggers);
2850
2851 /* Add an event to a trace directory */
2852 static int
2853 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2854 {
2855         struct trace_event_file *file;
2856
2857         file = trace_create_new_event(call, tr);
2858         if (!file)
2859                 return -ENOMEM;
2860
2861         if (eventdir_initialized)
2862                 return event_create_dir(tr->event_dir, file);
2863         else
2864                 return event_define_fields(call);
2865 }
2866
2867 static void trace_early_triggers(struct trace_event_file *file, const char *name)
2868 {
2869         int ret;
2870         int i;
2871
2872         for (i = 0; i < nr_boot_triggers; i++) {
2873                 if (strcmp(name, bootup_triggers[i].event))
2874                         continue;
2875                 mutex_lock(&event_mutex);
2876                 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
2877                 mutex_unlock(&event_mutex);
2878                 if (ret)
2879                         pr_err("Failed to register trigger '%s' on event %s\n",
2880                                bootup_triggers[i].trigger,
2881                                bootup_triggers[i].event);
2882         }
2883 }
2884
2885 /*
2886  * Just create a descriptor for early init. A descriptor is required
2887  * for enabling events at boot. We want to enable events before
2888  * the filesystem is initialized.
2889  */
2890 static int
2891 __trace_early_add_new_event(struct trace_event_call *call,
2892                             struct trace_array *tr)
2893 {
2894         struct trace_event_file *file;
2895         int ret;
2896
2897         file = trace_create_new_event(call, tr);
2898         if (!file)
2899                 return -ENOMEM;
2900
2901         ret = event_define_fields(call);
2902         if (ret)
2903                 return ret;
2904
2905         trace_early_triggers(file, trace_event_name(call));
2906
2907         return 0;
2908 }
2909
2910 struct ftrace_module_file_ops;
2911 static void __add_event_to_tracers(struct trace_event_call *call);
2912
2913 /* Add an additional event_call dynamically */
2914 int trace_add_event_call(struct trace_event_call *call)
2915 {
2916         int ret;
2917         lockdep_assert_held(&event_mutex);
2918
2919         mutex_lock(&trace_types_lock);
2920
2921         ret = __register_event(call, NULL);
2922         if (ret >= 0)
2923                 __add_event_to_tracers(call);
2924
2925         mutex_unlock(&trace_types_lock);
2926         return ret;
2927 }
2928 EXPORT_SYMBOL_GPL(trace_add_event_call);
2929
2930 /*
2931  * Must be called under locking of trace_types_lock, event_mutex and
2932  * trace_event_sem.
2933  */
2934 static void __trace_remove_event_call(struct trace_event_call *call)
2935 {
2936         event_remove(call);
2937         trace_destroy_fields(call);
2938         free_event_filter(call->filter);
2939         call->filter = NULL;
2940 }
2941
2942 static int probe_remove_event_call(struct trace_event_call *call)
2943 {
2944         struct trace_array *tr;
2945         struct trace_event_file *file;
2946
2947 #ifdef CONFIG_PERF_EVENTS
2948         if (call->perf_refcount)
2949                 return -EBUSY;
2950 #endif
2951         do_for_each_event_file(tr, file) {
2952                 if (file->event_call != call)
2953                         continue;
2954                 /*
2955                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2956                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2957                  * TRACE_REG_UNREGISTER.
2958                  */
2959                 if (file->flags & EVENT_FILE_FL_ENABLED)
2960                         goto busy;
2961
2962                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2963                         tr->clear_trace = true;
2964                 /*
2965                  * The do_for_each_event_file_safe() is
2966                  * a double loop. After finding the call for this
2967                  * trace_array, we use break to jump to the next
2968                  * trace_array.
2969                  */
2970                 break;
2971         } while_for_each_event_file();
2972
2973         __trace_remove_event_call(call);
2974
2975         return 0;
2976  busy:
2977         /* No need to clear the trace now */
2978         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2979                 tr->clear_trace = false;
2980         }
2981         return -EBUSY;
2982 }
2983
2984 /* Remove an event_call */
2985 int trace_remove_event_call(struct trace_event_call *call)
2986 {
2987         int ret;
2988
2989         lockdep_assert_held(&event_mutex);
2990
2991         mutex_lock(&trace_types_lock);
2992         down_write(&trace_event_sem);
2993         ret = probe_remove_event_call(call);
2994         up_write(&trace_event_sem);
2995         mutex_unlock(&trace_types_lock);
2996
2997         return ret;
2998 }
2999 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3000
3001 #define for_each_event(event, start, end)                       \
3002         for (event = start;                                     \
3003              (unsigned long)event < (unsigned long)end;         \
3004              event++)
3005
3006 #ifdef CONFIG_MODULES
3007
3008 static void trace_module_add_events(struct module *mod)
3009 {
3010         struct trace_event_call **call, **start, **end;
3011
3012         if (!mod->num_trace_events)
3013                 return;
3014
3015         /* Don't add infrastructure for mods without tracepoints */
3016         if (trace_module_has_bad_taint(mod)) {
3017                 pr_err("%s: module has bad taint, not creating trace events\n",
3018                        mod->name);
3019                 return;
3020         }
3021
3022         start = mod->trace_events;
3023         end = mod->trace_events + mod->num_trace_events;
3024
3025         for_each_event(call, start, end) {
3026                 __register_event(*call, mod);
3027                 __add_event_to_tracers(*call);
3028         }
3029 }
3030
3031 static void trace_module_remove_events(struct module *mod)
3032 {
3033         struct trace_event_call *call, *p;
3034         struct module_string *modstr, *m;
3035
3036         down_write(&trace_event_sem);
3037         list_for_each_entry_safe(call, p, &ftrace_events, list) {
3038                 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3039                         continue;
3040                 if (call->module == mod)
3041                         __trace_remove_event_call(call);
3042         }
3043         /* Check for any strings allocade for this module */
3044         list_for_each_entry_safe(modstr, m, &module_strings, next) {
3045                 if (modstr->module != mod)
3046                         continue;
3047                 list_del(&modstr->next);
3048                 kfree(modstr->str);
3049                 kfree(modstr);
3050         }
3051         up_write(&trace_event_sem);
3052
3053         /*
3054          * It is safest to reset the ring buffer if the module being unloaded
3055          * registered any events that were used. The only worry is if
3056          * a new module gets loaded, and takes on the same id as the events
3057          * of this module. When printing out the buffer, traced events left
3058          * over from this module may be passed to the new module events and
3059          * unexpected results may occur.
3060          */
3061         tracing_reset_all_online_cpus_unlocked();
3062 }
3063
3064 static int trace_module_notify(struct notifier_block *self,
3065                                unsigned long val, void *data)
3066 {
3067         struct module *mod = data;
3068
3069         mutex_lock(&event_mutex);
3070         mutex_lock(&trace_types_lock);
3071         switch (val) {
3072         case MODULE_STATE_COMING:
3073                 trace_module_add_events(mod);
3074                 break;
3075         case MODULE_STATE_GOING:
3076                 trace_module_remove_events(mod);
3077                 break;
3078         }
3079         mutex_unlock(&trace_types_lock);
3080         mutex_unlock(&event_mutex);
3081
3082         return NOTIFY_OK;
3083 }
3084
3085 static struct notifier_block trace_module_nb = {
3086         .notifier_call = trace_module_notify,
3087         .priority = 1, /* higher than trace.c module notify */
3088 };
3089 #endif /* CONFIG_MODULES */
3090
3091 /* Create a new event directory structure for a trace directory. */
3092 static void
3093 __trace_add_event_dirs(struct trace_array *tr)
3094 {
3095         struct trace_event_call *call;
3096         int ret;
3097
3098         list_for_each_entry(call, &ftrace_events, list) {
3099                 ret = __trace_add_new_event(call, tr);
3100                 if (ret < 0)
3101                         pr_warn("Could not create directory for event %s\n",
3102                                 trace_event_name(call));
3103         }
3104 }
3105
3106 /* Returns any file that matches the system and event */
3107 struct trace_event_file *
3108 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3109 {
3110         struct trace_event_file *file;
3111         struct trace_event_call *call;
3112         const char *name;
3113
3114         list_for_each_entry(file, &tr->events, list) {
3115
3116                 call = file->event_call;
3117                 name = trace_event_name(call);
3118
3119                 if (!name || !call->class)
3120                         continue;
3121
3122                 if (strcmp(event, name) == 0 &&
3123                     strcmp(system, call->class->system) == 0)
3124                         return file;
3125         }
3126         return NULL;
3127 }
3128
3129 /* Returns valid trace event files that match system and event */
3130 struct trace_event_file *
3131 find_event_file(struct trace_array *tr, const char *system, const char *event)
3132 {
3133         struct trace_event_file *file;
3134
3135         file = __find_event_file(tr, system, event);
3136         if (!file || !file->event_call->class->reg ||
3137             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3138                 return NULL;
3139
3140         return file;
3141 }
3142
3143 /**
3144  * trace_get_event_file - Find and return a trace event file
3145  * @instance: The name of the trace instance containing the event
3146  * @system: The name of the system containing the event
3147  * @event: The name of the event
3148  *
3149  * Return a trace event file given the trace instance name, trace
3150  * system, and trace event name.  If the instance name is NULL, it
3151  * refers to the top-level trace array.
3152  *
3153  * This function will look it up and return it if found, after calling
3154  * trace_array_get() to prevent the instance from going away, and
3155  * increment the event's module refcount to prevent it from being
3156  * removed.
3157  *
3158  * To release the file, call trace_put_event_file(), which will call
3159  * trace_array_put() and decrement the event's module refcount.
3160  *
3161  * Return: The trace event on success, ERR_PTR otherwise.
3162  */
3163 struct trace_event_file *trace_get_event_file(const char *instance,
3164                                               const char *system,
3165                                               const char *event)
3166 {
3167         struct trace_array *tr = top_trace_array();
3168         struct trace_event_file *file = NULL;
3169         int ret = -EINVAL;
3170
3171         if (instance) {
3172                 tr = trace_array_find_get(instance);
3173                 if (!tr)
3174                         return ERR_PTR(-ENOENT);
3175         } else {
3176                 ret = trace_array_get(tr);
3177                 if (ret)
3178                         return ERR_PTR(ret);
3179         }
3180
3181         mutex_lock(&event_mutex);
3182
3183         file = find_event_file(tr, system, event);
3184         if (!file) {
3185                 trace_array_put(tr);
3186                 ret = -EINVAL;
3187                 goto out;
3188         }
3189
3190         /* Don't let event modules unload while in use */
3191         ret = trace_event_try_get_ref(file->event_call);
3192         if (!ret) {
3193                 trace_array_put(tr);
3194                 ret = -EBUSY;
3195                 goto out;
3196         }
3197
3198         ret = 0;
3199  out:
3200         mutex_unlock(&event_mutex);
3201
3202         if (ret)
3203                 file = ERR_PTR(ret);
3204
3205         return file;
3206 }
3207 EXPORT_SYMBOL_GPL(trace_get_event_file);
3208
3209 /**
3210  * trace_put_event_file - Release a file from trace_get_event_file()
3211  * @file: The trace event file
3212  *
3213  * If a file was retrieved using trace_get_event_file(), this should
3214  * be called when it's no longer needed.  It will cancel the previous
3215  * trace_array_get() called by that function, and decrement the
3216  * event's module refcount.
3217  */
3218 void trace_put_event_file(struct trace_event_file *file)
3219 {
3220         mutex_lock(&event_mutex);
3221         trace_event_put_ref(file->event_call);
3222         mutex_unlock(&event_mutex);
3223
3224         trace_array_put(file->tr);
3225 }
3226 EXPORT_SYMBOL_GPL(trace_put_event_file);
3227
3228 #ifdef CONFIG_DYNAMIC_FTRACE
3229
3230 /* Avoid typos */
3231 #define ENABLE_EVENT_STR        "enable_event"
3232 #define DISABLE_EVENT_STR       "disable_event"
3233
3234 struct event_probe_data {
3235         struct trace_event_file *file;
3236         unsigned long                   count;
3237         int                             ref;
3238         bool                            enable;
3239 };
3240
3241 static void update_event_probe(struct event_probe_data *data)
3242 {
3243         if (data->enable)
3244                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3245         else
3246                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3247 }
3248
3249 static void
3250 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3251                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3252                    void *data)
3253 {
3254         struct ftrace_func_mapper *mapper = data;
3255         struct event_probe_data *edata;
3256         void **pdata;
3257
3258         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3259         if (!pdata || !*pdata)
3260                 return;
3261
3262         edata = *pdata;
3263         update_event_probe(edata);
3264 }
3265
3266 static void
3267 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3268                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3269                          void *data)
3270 {
3271         struct ftrace_func_mapper *mapper = data;
3272         struct event_probe_data *edata;
3273         void **pdata;
3274
3275         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3276         if (!pdata || !*pdata)
3277                 return;
3278
3279         edata = *pdata;
3280
3281         if (!edata->count)
3282                 return;
3283
3284         /* Skip if the event is in a state we want to switch to */
3285         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3286                 return;
3287
3288         if (edata->count != -1)
3289                 (edata->count)--;
3290
3291         update_event_probe(edata);
3292 }
3293
3294 static int
3295 event_enable_print(struct seq_file *m, unsigned long ip,
3296                    struct ftrace_probe_ops *ops, void *data)
3297 {
3298         struct ftrace_func_mapper *mapper = data;
3299         struct event_probe_data *edata;
3300         void **pdata;
3301
3302         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3303
3304         if (WARN_ON_ONCE(!pdata || !*pdata))
3305                 return 0;
3306
3307         edata = *pdata;
3308
3309         seq_printf(m, "%ps:", (void *)ip);
3310
3311         seq_printf(m, "%s:%s:%s",
3312                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3313                    edata->file->event_call->class->system,
3314                    trace_event_name(edata->file->event_call));
3315
3316         if (edata->count == -1)
3317                 seq_puts(m, ":unlimited\n");
3318         else
3319                 seq_printf(m, ":count=%ld\n", edata->count);
3320
3321         return 0;
3322 }
3323
3324 static int
3325 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3326                   unsigned long ip, void *init_data, void **data)
3327 {
3328         struct ftrace_func_mapper *mapper = *data;
3329         struct event_probe_data *edata = init_data;
3330         int ret;
3331
3332         if (!mapper) {
3333                 mapper = allocate_ftrace_func_mapper();
3334                 if (!mapper)
3335                         return -ENODEV;
3336                 *data = mapper;
3337         }
3338
3339         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3340         if (ret < 0)
3341                 return ret;
3342
3343         edata->ref++;
3344
3345         return 0;
3346 }
3347
3348 static int free_probe_data(void *data)
3349 {
3350         struct event_probe_data *edata = data;
3351
3352         edata->ref--;
3353         if (!edata->ref) {
3354                 /* Remove the SOFT_MODE flag */
3355                 __ftrace_event_enable_disable(edata->file, 0, 1);
3356                 trace_event_put_ref(edata->file->event_call);
3357                 kfree(edata);
3358         }
3359         return 0;
3360 }
3361
3362 static void
3363 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3364                   unsigned long ip, void *data)
3365 {
3366         struct ftrace_func_mapper *mapper = data;
3367         struct event_probe_data *edata;
3368
3369         if (!ip) {
3370                 if (!mapper)
3371                         return;
3372                 free_ftrace_func_mapper(mapper, free_probe_data);
3373                 return;
3374         }
3375
3376         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3377
3378         if (WARN_ON_ONCE(!edata))
3379                 return;
3380
3381         if (WARN_ON_ONCE(edata->ref <= 0))
3382                 return;
3383
3384         free_probe_data(edata);
3385 }
3386
3387 static struct ftrace_probe_ops event_enable_probe_ops = {
3388         .func                   = event_enable_probe,
3389         .print                  = event_enable_print,
3390         .init                   = event_enable_init,
3391         .free                   = event_enable_free,
3392 };
3393
3394 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3395         .func                   = event_enable_count_probe,
3396         .print                  = event_enable_print,
3397         .init                   = event_enable_init,
3398         .free                   = event_enable_free,
3399 };
3400
3401 static struct ftrace_probe_ops event_disable_probe_ops = {
3402         .func                   = event_enable_probe,
3403         .print                  = event_enable_print,
3404         .init                   = event_enable_init,
3405         .free                   = event_enable_free,
3406 };
3407
3408 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3409         .func                   = event_enable_count_probe,
3410         .print                  = event_enable_print,
3411         .init                   = event_enable_init,
3412         .free                   = event_enable_free,
3413 };
3414
3415 static int
3416 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3417                   char *glob, char *cmd, char *param, int enabled)
3418 {
3419         struct trace_event_file *file;
3420         struct ftrace_probe_ops *ops;
3421         struct event_probe_data *data;
3422         const char *system;
3423         const char *event;
3424         char *number;
3425         bool enable;
3426         int ret;
3427
3428         if (!tr)
3429                 return -ENODEV;
3430
3431         /* hash funcs only work with set_ftrace_filter */
3432         if (!enabled || !param)
3433                 return -EINVAL;
3434
3435         system = strsep(&param, ":");
3436         if (!param)
3437                 return -EINVAL;
3438
3439         event = strsep(&param, ":");
3440
3441         mutex_lock(&event_mutex);
3442
3443         ret = -EINVAL;
3444         file = find_event_file(tr, system, event);
3445         if (!file)
3446                 goto out;
3447
3448         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3449
3450         if (enable)
3451                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3452         else
3453                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3454
3455         if (glob[0] == '!') {
3456                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3457                 goto out;
3458         }
3459
3460         ret = -ENOMEM;
3461
3462         data = kzalloc(sizeof(*data), GFP_KERNEL);
3463         if (!data)
3464                 goto out;
3465
3466         data->enable = enable;
3467         data->count = -1;
3468         data->file = file;
3469
3470         if (!param)
3471                 goto out_reg;
3472
3473         number = strsep(&param, ":");
3474
3475         ret = -EINVAL;
3476         if (!strlen(number))
3477                 goto out_free;
3478
3479         /*
3480          * We use the callback data field (which is a pointer)
3481          * as our counter.
3482          */
3483         ret = kstrtoul(number, 0, &data->count);
3484         if (ret)
3485                 goto out_free;
3486
3487  out_reg:
3488         /* Don't let event modules unload while probe registered */
3489         ret = trace_event_try_get_ref(file->event_call);
3490         if (!ret) {
3491                 ret = -EBUSY;
3492                 goto out_free;
3493         }
3494
3495         ret = __ftrace_event_enable_disable(file, 1, 1);
3496         if (ret < 0)
3497                 goto out_put;
3498
3499         ret = register_ftrace_function_probe(glob, tr, ops, data);
3500         /*
3501          * The above returns on success the # of functions enabled,
3502          * but if it didn't find any functions it returns zero.
3503          * Consider no functions a failure too.
3504          */
3505         if (!ret) {
3506                 ret = -ENOENT;
3507                 goto out_disable;
3508         } else if (ret < 0)
3509                 goto out_disable;
3510         /* Just return zero, not the number of enabled functions */
3511         ret = 0;
3512  out:
3513         mutex_unlock(&event_mutex);
3514         return ret;
3515
3516  out_disable:
3517         __ftrace_event_enable_disable(file, 0, 1);
3518  out_put:
3519         trace_event_put_ref(file->event_call);
3520  out_free:
3521         kfree(data);
3522         goto out;
3523 }
3524
3525 static struct ftrace_func_command event_enable_cmd = {
3526         .name                   = ENABLE_EVENT_STR,
3527         .func                   = event_enable_func,
3528 };
3529
3530 static struct ftrace_func_command event_disable_cmd = {
3531         .name                   = DISABLE_EVENT_STR,
3532         .func                   = event_enable_func,
3533 };
3534
3535 static __init int register_event_cmds(void)
3536 {
3537         int ret;
3538
3539         ret = register_ftrace_command(&event_enable_cmd);
3540         if (WARN_ON(ret < 0))
3541                 return ret;
3542         ret = register_ftrace_command(&event_disable_cmd);
3543         if (WARN_ON(ret < 0))
3544                 unregister_ftrace_command(&event_enable_cmd);
3545         return ret;
3546 }
3547 #else
3548 static inline int register_event_cmds(void) { return 0; }
3549 #endif /* CONFIG_DYNAMIC_FTRACE */
3550
3551 /*
3552  * The top level array and trace arrays created by boot-time tracing
3553  * have already had its trace_event_file descriptors created in order
3554  * to allow for early events to be recorded.
3555  * This function is called after the tracefs has been initialized,
3556  * and we now have to create the files associated to the events.
3557  */
3558 static void __trace_early_add_event_dirs(struct trace_array *tr)
3559 {
3560         struct trace_event_file *file;
3561         int ret;
3562
3563
3564         list_for_each_entry(file, &tr->events, list) {
3565                 ret = event_create_dir(tr->event_dir, file);
3566                 if (ret < 0)
3567                         pr_warn("Could not create directory for event %s\n",
3568                                 trace_event_name(file->event_call));
3569         }
3570 }
3571
3572 /*
3573  * For early boot up, the top trace array and the trace arrays created
3574  * by boot-time tracing require to have a list of events that can be
3575  * enabled. This must be done before the filesystem is set up in order
3576  * to allow events to be traced early.
3577  */
3578 void __trace_early_add_events(struct trace_array *tr)
3579 {
3580         struct trace_event_call *call;
3581         int ret;
3582
3583         list_for_each_entry(call, &ftrace_events, list) {
3584                 /* Early boot up should not have any modules loaded */
3585                 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3586                     WARN_ON_ONCE(call->module))
3587                         continue;
3588
3589                 ret = __trace_early_add_new_event(call, tr);
3590                 if (ret < 0)
3591                         pr_warn("Could not create early event %s\n",
3592                                 trace_event_name(call));
3593         }
3594 }
3595
3596 /* Remove the event directory structure for a trace directory. */
3597 static void
3598 __trace_remove_event_dirs(struct trace_array *tr)
3599 {
3600         struct trace_event_file *file, *next;
3601
3602         list_for_each_entry_safe(file, next, &tr->events, list)
3603                 remove_event_file_dir(file);
3604 }
3605
3606 static void __add_event_to_tracers(struct trace_event_call *call)
3607 {
3608         struct trace_array *tr;
3609
3610         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3611                 __trace_add_new_event(call, tr);
3612 }
3613
3614 extern struct trace_event_call *__start_ftrace_events[];
3615 extern struct trace_event_call *__stop_ftrace_events[];
3616
3617 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3618
3619 static __init int setup_trace_event(char *str)
3620 {
3621         strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3622         ring_buffer_expanded = true;
3623         disable_tracing_selftest("running event tracing");
3624
3625         return 1;
3626 }
3627 __setup("trace_event=", setup_trace_event);
3628
3629 /* Expects to have event_mutex held when called */
3630 static int
3631 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3632 {
3633         struct dentry *d_events;
3634         struct dentry *entry;
3635
3636         entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3637                                   tr, &ftrace_set_event_fops);
3638         if (!entry)
3639                 return -ENOMEM;
3640
3641         d_events = tracefs_create_dir("events", parent);
3642         if (!d_events) {
3643                 pr_warn("Could not create tracefs 'events' directory\n");
3644                 return -ENOMEM;
3645         }
3646
3647         entry = trace_create_file("enable", TRACE_MODE_WRITE, d_events,
3648                                   tr, &ftrace_tr_enable_fops);
3649         if (!entry)
3650                 return -ENOMEM;
3651
3652         /* There are not as crucial, just warn if they are not created */
3653
3654         trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3655                           tr, &ftrace_set_event_pid_fops);
3656
3657         trace_create_file("set_event_notrace_pid",
3658                           TRACE_MODE_WRITE, parent, tr,
3659                           &ftrace_set_event_notrace_pid_fops);
3660
3661         /* ring buffer internal formats */
3662         trace_create_file("header_page", TRACE_MODE_READ, d_events,
3663                                   ring_buffer_print_page_header,
3664                                   &ftrace_show_header_fops);
3665
3666         trace_create_file("header_event", TRACE_MODE_READ, d_events,
3667                                   ring_buffer_print_entry_header,
3668                                   &ftrace_show_header_fops);
3669
3670         tr->event_dir = d_events;
3671
3672         return 0;
3673 }
3674
3675 /**
3676  * event_trace_add_tracer - add a instance of a trace_array to events
3677  * @parent: The parent dentry to place the files/directories for events in
3678  * @tr: The trace array associated with these events
3679  *
3680  * When a new instance is created, it needs to set up its events
3681  * directory, as well as other files associated with events. It also
3682  * creates the event hierarchy in the @parent/events directory.
3683  *
3684  * Returns 0 on success.
3685  *
3686  * Must be called with event_mutex held.
3687  */
3688 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3689 {
3690         int ret;
3691
3692         lockdep_assert_held(&event_mutex);
3693
3694         ret = create_event_toplevel_files(parent, tr);
3695         if (ret)
3696                 goto out;
3697
3698         down_write(&trace_event_sem);
3699         /* If tr already has the event list, it is initialized in early boot. */
3700         if (unlikely(!list_empty(&tr->events)))
3701                 __trace_early_add_event_dirs(tr);
3702         else
3703                 __trace_add_event_dirs(tr);
3704         up_write(&trace_event_sem);
3705
3706  out:
3707         return ret;
3708 }
3709
3710 /*
3711  * The top trace array already had its file descriptors created.
3712  * Now the files themselves need to be created.
3713  */
3714 static __init int
3715 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3716 {
3717         int ret;
3718
3719         mutex_lock(&event_mutex);
3720
3721         ret = create_event_toplevel_files(parent, tr);
3722         if (ret)
3723                 goto out_unlock;
3724
3725         down_write(&trace_event_sem);
3726         __trace_early_add_event_dirs(tr);
3727         up_write(&trace_event_sem);
3728
3729  out_unlock:
3730         mutex_unlock(&event_mutex);
3731
3732         return ret;
3733 }
3734
3735 /* Must be called with event_mutex held */
3736 int event_trace_del_tracer(struct trace_array *tr)
3737 {
3738         lockdep_assert_held(&event_mutex);
3739
3740         /* Disable any event triggers and associated soft-disabled events */
3741         clear_event_triggers(tr);
3742
3743         /* Clear the pid list */
3744         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3745
3746         /* Disable any running events */
3747         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3748
3749         /* Make sure no more events are being executed */
3750         tracepoint_synchronize_unregister();
3751
3752         down_write(&trace_event_sem);
3753         __trace_remove_event_dirs(tr);
3754         tracefs_remove(tr->event_dir);
3755         up_write(&trace_event_sem);
3756
3757         tr->event_dir = NULL;
3758
3759         return 0;
3760 }
3761
3762 static __init int event_trace_memsetup(void)
3763 {
3764         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3765         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3766         return 0;
3767 }
3768
3769 __init void
3770 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3771 {
3772         char *token;
3773         int ret;
3774
3775         while (true) {
3776                 token = strsep(&buf, ",");
3777
3778                 if (!token)
3779                         break;
3780
3781                 if (*token) {
3782                         /* Restarting syscalls requires that we stop them first */
3783                         if (disable_first)
3784                                 ftrace_set_clr_event(tr, token, 0);
3785
3786                         ret = ftrace_set_clr_event(tr, token, 1);
3787                         if (ret)
3788                                 pr_warn("Failed to enable trace event: %s\n", token);
3789                 }
3790
3791                 /* Put back the comma to allow this to be called again */
3792                 if (buf)
3793                         *(buf - 1) = ',';
3794         }
3795 }
3796
3797 static __init int event_trace_enable(void)
3798 {
3799         struct trace_array *tr = top_trace_array();
3800         struct trace_event_call **iter, *call;
3801         int ret;
3802
3803         if (!tr)
3804                 return -ENODEV;
3805
3806         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3807
3808                 call = *iter;
3809                 ret = event_init(call);
3810                 if (!ret)
3811                         list_add(&call->list, &ftrace_events);
3812         }
3813
3814         register_trigger_cmds();
3815
3816         /*
3817          * We need the top trace array to have a working set of trace
3818          * points at early init, before the debug files and directories
3819          * are created. Create the file entries now, and attach them
3820          * to the actual file dentries later.
3821          */
3822         __trace_early_add_events(tr);
3823
3824         early_enable_events(tr, bootup_event_buf, false);
3825
3826         trace_printk_start_comm();
3827
3828         register_event_cmds();
3829
3830
3831         return 0;
3832 }
3833
3834 /*
3835  * event_trace_enable() is called from trace_event_init() first to
3836  * initialize events and perhaps start any events that are on the
3837  * command line. Unfortunately, there are some events that will not
3838  * start this early, like the system call tracepoints that need
3839  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3840  * event_trace_enable() is called before pid 1 starts, and this flag
3841  * is never set, making the syscall tracepoint never get reached, but
3842  * the event is enabled regardless (and not doing anything).
3843  */
3844 static __init int event_trace_enable_again(void)
3845 {
3846         struct trace_array *tr;
3847
3848         tr = top_trace_array();
3849         if (!tr)
3850                 return -ENODEV;
3851
3852         early_enable_events(tr, bootup_event_buf, true);
3853
3854         return 0;
3855 }
3856
3857 early_initcall(event_trace_enable_again);
3858
3859 /* Init fields which doesn't related to the tracefs */
3860 static __init int event_trace_init_fields(void)
3861 {
3862         if (trace_define_generic_fields())
3863                 pr_warn("tracing: Failed to allocated generic fields");
3864
3865         if (trace_define_common_fields())
3866                 pr_warn("tracing: Failed to allocate common fields");
3867
3868         return 0;
3869 }
3870
3871 __init int event_trace_init(void)
3872 {
3873         struct trace_array *tr;
3874         int ret;
3875
3876         tr = top_trace_array();
3877         if (!tr)
3878                 return -ENODEV;
3879
3880         trace_create_file("available_events", TRACE_MODE_READ,
3881                           NULL, tr, &ftrace_avail_fops);
3882
3883         ret = early_event_add_tracer(NULL, tr);
3884         if (ret)
3885                 return ret;
3886
3887 #ifdef CONFIG_MODULES
3888         ret = register_module_notifier(&trace_module_nb);
3889         if (ret)
3890                 pr_warn("Failed to register trace events module notifier\n");
3891 #endif
3892
3893         eventdir_initialized = true;
3894
3895         return 0;
3896 }
3897
3898 void __init trace_event_init(void)
3899 {
3900         event_trace_memsetup();
3901         init_ftrace_syscalls();
3902         event_trace_enable();
3903         event_trace_init_fields();
3904 }
3905
3906 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3907
3908 static DEFINE_SPINLOCK(test_spinlock);
3909 static DEFINE_SPINLOCK(test_spinlock_irq);
3910 static DEFINE_MUTEX(test_mutex);
3911
3912 static __init void test_work(struct work_struct *dummy)
3913 {
3914         spin_lock(&test_spinlock);
3915         spin_lock_irq(&test_spinlock_irq);
3916         udelay(1);
3917         spin_unlock_irq(&test_spinlock_irq);
3918         spin_unlock(&test_spinlock);
3919
3920         mutex_lock(&test_mutex);
3921         msleep(1);
3922         mutex_unlock(&test_mutex);
3923 }
3924
3925 static __init int event_test_thread(void *unused)
3926 {
3927         void *test_malloc;
3928
3929         test_malloc = kmalloc(1234, GFP_KERNEL);
3930         if (!test_malloc)
3931                 pr_info("failed to kmalloc\n");
3932
3933         schedule_on_each_cpu(test_work);
3934
3935         kfree(test_malloc);
3936
3937         set_current_state(TASK_INTERRUPTIBLE);
3938         while (!kthread_should_stop()) {
3939                 schedule();
3940                 set_current_state(TASK_INTERRUPTIBLE);
3941         }
3942         __set_current_state(TASK_RUNNING);
3943
3944         return 0;
3945 }
3946
3947 /*
3948  * Do various things that may trigger events.
3949  */
3950 static __init void event_test_stuff(void)
3951 {
3952         struct task_struct *test_thread;
3953
3954         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3955         msleep(1);
3956         kthread_stop(test_thread);
3957 }
3958
3959 /*
3960  * For every trace event defined, we will test each trace point separately,
3961  * and then by groups, and finally all trace points.
3962  */
3963 static __init void event_trace_self_tests(void)
3964 {
3965         struct trace_subsystem_dir *dir;
3966         struct trace_event_file *file;
3967         struct trace_event_call *call;
3968         struct event_subsystem *system;
3969         struct trace_array *tr;
3970         int ret;
3971
3972         tr = top_trace_array();
3973         if (!tr)
3974                 return;
3975
3976         pr_info("Running tests on trace events:\n");
3977
3978         list_for_each_entry(file, &tr->events, list) {
3979
3980                 call = file->event_call;
3981
3982                 /* Only test those that have a probe */
3983                 if (!call->class || !call->class->probe)
3984                         continue;
3985
3986 /*
3987  * Testing syscall events here is pretty useless, but
3988  * we still do it if configured. But this is time consuming.
3989  * What we really need is a user thread to perform the
3990  * syscalls as we test.
3991  */
3992 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3993                 if (call->class->system &&
3994                     strcmp(call->class->system, "syscalls") == 0)
3995                         continue;
3996 #endif
3997
3998                 pr_info("Testing event %s: ", trace_event_name(call));
3999
4000                 /*
4001                  * If an event is already enabled, someone is using
4002                  * it and the self test should not be on.
4003                  */
4004                 if (file->flags & EVENT_FILE_FL_ENABLED) {
4005                         pr_warn("Enabled event during self test!\n");
4006                         WARN_ON_ONCE(1);
4007                         continue;
4008                 }
4009
4010                 ftrace_event_enable_disable(file, 1);
4011                 event_test_stuff();
4012                 ftrace_event_enable_disable(file, 0);
4013
4014                 pr_cont("OK\n");
4015         }
4016
4017         /* Now test at the sub system level */
4018
4019         pr_info("Running tests on trace event systems:\n");
4020
4021         list_for_each_entry(dir, &tr->systems, list) {
4022
4023                 system = dir->subsystem;
4024
4025                 /* the ftrace system is special, skip it */
4026                 if (strcmp(system->name, "ftrace") == 0)
4027                         continue;
4028
4029                 pr_info("Testing event system %s: ", system->name);
4030
4031                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4032                 if (WARN_ON_ONCE(ret)) {
4033                         pr_warn("error enabling system %s\n",
4034                                 system->name);
4035                         continue;
4036                 }
4037
4038                 event_test_stuff();
4039
4040                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4041                 if (WARN_ON_ONCE(ret)) {
4042                         pr_warn("error disabling system %s\n",
4043                                 system->name);
4044                         continue;
4045                 }
4046
4047                 pr_cont("OK\n");
4048         }
4049
4050         /* Test with all events enabled */
4051
4052         pr_info("Running tests on all trace events:\n");
4053         pr_info("Testing all events: ");
4054
4055         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4056         if (WARN_ON_ONCE(ret)) {
4057                 pr_warn("error enabling all events\n");
4058                 return;
4059         }
4060
4061         event_test_stuff();
4062
4063         /* reset sysname */
4064         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4065         if (WARN_ON_ONCE(ret)) {
4066                 pr_warn("error disabling all events\n");
4067                 return;
4068         }
4069
4070         pr_cont("OK\n");
4071 }
4072
4073 #ifdef CONFIG_FUNCTION_TRACER
4074
4075 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4076
4077 static struct trace_event_file event_trace_file __initdata;
4078
4079 static void __init
4080 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4081                           struct ftrace_ops *op, struct ftrace_regs *regs)
4082 {
4083         struct trace_buffer *buffer;
4084         struct ring_buffer_event *event;
4085         struct ftrace_entry *entry;
4086         unsigned int trace_ctx;
4087         long disabled;
4088         int cpu;
4089
4090         trace_ctx = tracing_gen_ctx();
4091         preempt_disable_notrace();
4092         cpu = raw_smp_processor_id();
4093         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4094
4095         if (disabled != 1)
4096                 goto out;
4097
4098         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4099                                                 TRACE_FN, sizeof(*entry),
4100                                                 trace_ctx);
4101         if (!event)
4102                 goto out;
4103         entry   = ring_buffer_event_data(event);
4104         entry->ip                       = ip;
4105         entry->parent_ip                = parent_ip;
4106
4107         event_trigger_unlock_commit(&event_trace_file, buffer, event,
4108                                     entry, trace_ctx);
4109  out:
4110         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4111         preempt_enable_notrace();
4112 }
4113
4114 static struct ftrace_ops trace_ops __initdata  =
4115 {
4116         .func = function_test_events_call,
4117 };
4118
4119 static __init void event_trace_self_test_with_function(void)
4120 {
4121         int ret;
4122
4123         event_trace_file.tr = top_trace_array();
4124         if (WARN_ON(!event_trace_file.tr))
4125                 return;
4126
4127         ret = register_ftrace_function(&trace_ops);
4128         if (WARN_ON(ret < 0)) {
4129                 pr_info("Failed to enable function tracer for event tests\n");
4130                 return;
4131         }
4132         pr_info("Running tests again, along with the function tracer\n");
4133         event_trace_self_tests();
4134         unregister_ftrace_function(&trace_ops);
4135 }
4136 #else
4137 static __init void event_trace_self_test_with_function(void)
4138 {
4139 }
4140 #endif
4141
4142 static __init int event_trace_self_tests_init(void)
4143 {
4144         if (!tracing_selftest_disabled) {
4145                 event_trace_self_tests();
4146                 event_trace_self_test_with_function();
4147         }
4148
4149         return 0;
4150 }
4151
4152 late_initcall(event_trace_self_tests_init);
4153
4154 #endif