kernel/trace/trace_events.c

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