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(dir->ei);
 988                 list_del(&dir->list);
 989                 __put_system_dir(dir);
 990         }
 991 }
 992
 993 void event_file_get(struct trace_event_file *file)
 994 {
 995         atomic_inc(&file->ref);
 996 }
 997
 998 void event_file_put(struct trace_event_file *file)
 999 {
1000         if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001                 if (file->flags & EVENT_FILE_FL_FREED)
1002                         kmem_cache_free(file_cachep, file);
1003                 return;
1004         }
1005
1006         if (atomic_dec_and_test(&file->ref)) {
1007                 /* Count should only go to zero when it is freed */
1008                 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009                         return;
1010                 kmem_cache_free(file_cachep, file);
1011         }
1012 }
1013
1014 static void remove_event_file_dir(struct trace_event_file *file)
1015 {
1016         eventfs_remove_dir(file->ei);
1017         list_del(&file->list);
1018         remove_subsystem(file->system);
1019         free_event_filter(file->filter);
1020         file->flags |= EVENT_FILE_FL_FREED;
1021         event_file_put(file);
1022 }
1023
1024 /*
1025  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026  */
1027 static int
1028 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029                               const char *sub, const char *event, int set)
1030 {
1031         struct trace_event_file *file;
1032         struct trace_event_call *call;
1033         const char *name;
1034         int ret = -EINVAL;
1035         int eret = 0;
1036
1037         list_for_each_entry(file, &tr->events, list) {
1038
1039                 call = file->event_call;
1040                 name = trace_event_name(call);
1041
1042                 if (!name || !call->class || !call->class->reg)
1043                         continue;
1044
1045                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046                         continue;
1047
1048                 if (match &&
1049                     strcmp(match, name) != 0 &&
1050                     strcmp(match, call->class->system) != 0)
1051                         continue;
1052
1053                 if (sub && strcmp(sub, call->class->system) != 0)
1054                         continue;
1055
1056                 if (event && strcmp(event, name) != 0)
1057                         continue;
1058
1059                 ret = ftrace_event_enable_disable(file, set);
1060
1061                 /*
1062                  * Save the first error and return that. Some events
1063                  * may still have been enabled, but let the user
1064                  * know that something went wrong.
1065                  */
1066                 if (ret && !eret)
1067                         eret = ret;
1068
1069                 ret = eret;
1070         }
1071
1072         return ret;
1073 }
1074
1075 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076                                   const char *sub, const char *event, int set)
1077 {
1078         int ret;
1079
1080         mutex_lock(&event_mutex);
1081         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082         mutex_unlock(&event_mutex);
1083
1084         return ret;
1085 }
1086
1087 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088 {
1089         char *event = NULL, *sub = NULL, *match;
1090         int ret;
1091
1092         if (!tr)
1093                 return -ENOENT;
1094         /*
1095          * The buf format can be <subsystem>:<event-name>
1096          *  *:<event-name> means any event by that name.
1097          *  :<event-name> is the same.
1098          *
1099          *  <subsystem>:* means all events in that subsystem
1100          *  <subsystem>: means the same.
1101          *
1102          *  <name> (no ':') means all events in a subsystem with
1103          *  the name <name> or any event that matches <name>
1104          */
1105
1106         match = strsep(&buf, ":");
1107         if (buf) {
1108                 sub = match;
1109                 event = buf;
1110                 match = NULL;
1111
1112                 if (!strlen(sub) || strcmp(sub, "*") == 0)
1113                         sub = NULL;
1114                 if (!strlen(event) || strcmp(event, "*") == 0)
1115                         event = NULL;
1116         }
1117
1118         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120         /* Put back the colon to allow this to be called again */
1121         if (buf)
1122                 *(buf - 1) = ':';
1123
1124         return ret;
1125 }
1126
1127 /**
1128  * trace_set_clr_event - enable or disable an event
1129  * @system: system name to match (NULL for any system)
1130  * @event: event name to match (NULL for all events, within system)
1131  * @set: 1 to enable, 0 to disable
1132  *
1133  * This is a way for other parts of the kernel to enable or disable
1134  * event recording.
1135  *
1136  * Returns 0 on success, -EINVAL if the parameters do not match any
1137  * registered events.
1138  */
1139 int trace_set_clr_event(const char *system, const char *event, int set)
1140 {
1141         struct trace_array *tr = top_trace_array();
1142
1143         if (!tr)
1144                 return -ENODEV;
1145
1146         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147 }
1148 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150 /**
1151  * trace_array_set_clr_event - enable or disable an event for a trace array.
1152  * @tr: concerned trace array.
1153  * @system: system name to match (NULL for any system)
1154  * @event: event name to match (NULL for all events, within system)
1155  * @enable: true to enable, false to disable
1156  *
1157  * This is a way for other parts of the kernel to enable or disable
1158  * event recording.
1159  *
1160  * Returns 0 on success, -EINVAL if the parameters do not match any
1161  * registered events.
1162  */
1163 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164                 const char *event, bool enable)
1165 {
1166         int set;
1167
1168         if (!tr)
1169                 return -ENOENT;
1170
1171         set = (enable == true) ? 1 : 0;
1172         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173 }
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE          127
1178
1179 static ssize_t
1180 ftrace_event_write(struct file *file, const char __user *ubuf,
1181                    size_t cnt, loff_t *ppos)
1182 {
1183         struct trace_parser parser;
1184         struct seq_file *m = file->private_data;
1185         struct trace_array *tr = m->private;
1186         ssize_t read, ret;
1187
1188         if (!cnt)
1189                 return 0;
1190
1191         ret = tracing_update_buffers(tr);
1192         if (ret < 0)
1193                 return ret;
1194
1195         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196                 return -ENOMEM;
1197
1198         read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200         if (read >= 0 && trace_parser_loaded((&parser))) {
1201                 int set = 1;
1202
1203                 if (*parser.buffer == '!')
1204                         set = 0;
1205
1206                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207                 if (ret)
1208                         goto out_put;
1209         }
1210
1211         ret = read;
1212
1213  out_put:
1214         trace_parser_put(&parser);
1215
1216         return ret;
1217 }
1218
1219 static void *
1220 t_next(struct seq_file *m, void *v, loff_t *pos)
1221 {
1222         struct trace_event_file *file = v;
1223         struct trace_event_call *call;
1224         struct trace_array *tr = m->private;
1225
1226         (*pos)++;
1227
1228         list_for_each_entry_continue(file, &tr->events, list) {
1229                 call = file->event_call;
1230                 /*
1231                  * The ftrace subsystem is for showing formats only.
1232                  * They can not be enabled or disabled via the event files.
1233                  */
1234                 if (call->class && call->class->reg &&
1235                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236                         return file;
1237         }
1238
1239         return NULL;
1240 }
1241
1242 static void *t_start(struct seq_file *m, loff_t *pos)
1243 {
1244         struct trace_event_file *file;
1245         struct trace_array *tr = m->private;
1246         loff_t l;
1247
1248         mutex_lock(&event_mutex);
1249
1250         file = list_entry(&tr->events, struct trace_event_file, list);
1251         for (l = 0; l <= *pos; ) {
1252                 file = t_next(m, file, &l);
1253                 if (!file)
1254                         break;
1255         }
1256         return file;
1257 }
1258
1259 static void *
1260 s_next(struct seq_file *m, void *v, loff_t *pos)
1261 {
1262         struct trace_event_file *file = v;
1263         struct trace_array *tr = m->private;
1264
1265         (*pos)++;
1266
1267         list_for_each_entry_continue(file, &tr->events, list) {
1268                 if (file->flags & EVENT_FILE_FL_ENABLED)
1269                         return file;
1270         }
1271
1272         return NULL;
1273 }
1274
1275 static void *s_start(struct seq_file *m, loff_t *pos)
1276 {
1277         struct trace_event_file *file;
1278         struct trace_array *tr = m->private;
1279         loff_t l;
1280
1281         mutex_lock(&event_mutex);
1282
1283         file = list_entry(&tr->events, struct trace_event_file, list);
1284         for (l = 0; l <= *pos; ) {
1285                 file = s_next(m, file, &l);
1286                 if (!file)
1287                         break;
1288         }
1289         return file;
1290 }
1291
1292 static int t_show(struct seq_file *m, void *v)
1293 {
1294         struct trace_event_file *file = v;
1295         struct trace_event_call *call = file->event_call;
1296
1297         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298                 seq_printf(m, "%s:", call->class->system);
1299         seq_printf(m, "%s\n", trace_event_name(call));
1300
1301         return 0;
1302 }
1303
1304 static void t_stop(struct seq_file *m, void *p)
1305 {
1306         mutex_unlock(&event_mutex);
1307 }
1308
1309 static void *
1310 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1311 {
1312         struct trace_array *tr = m->private;
1313         struct trace_pid_list *pid_list;
1314
1315         if (type == TRACE_PIDS)
1316                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1317         else
1318                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320         return trace_pid_next(pid_list, v, pos);
1321 }
1322
1323 static void *
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326         return __next(m, v, pos, TRACE_PIDS);
1327 }
1328
1329 static void *
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1331 {
1332         return __next(m, v, pos, TRACE_NO_PIDS);
1333 }
1334
1335 static void *__start(struct seq_file *m, loff_t *pos, int type)
1336         __acquires(RCU)
1337 {
1338         struct trace_pid_list *pid_list;
1339         struct trace_array *tr = m->private;
1340
1341         /*
1342          * Grab the mutex, to keep calls to p_next() having the same
1343          * tr->filtered_pids as p_start() has.
1344          * If we just passed the tr->filtered_pids around, then RCU would
1345          * have been enough, but doing that makes things more complex.
1346          */
1347         mutex_lock(&event_mutex);
1348         rcu_read_lock_sched();
1349
1350         if (type == TRACE_PIDS)
1351                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1352         else
1353                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355         if (!pid_list)
1356                 return NULL;
1357
1358         return trace_pid_start(pid_list, pos);
1359 }
1360
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1362         __acquires(RCU)
1363 {
1364         return __start(m, pos, TRACE_PIDS);
1365 }
1366
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1368         __acquires(RCU)
1369 {
1370         return __start(m, pos, TRACE_NO_PIDS);
1371 }
1372
1373 static void p_stop(struct seq_file *m, void *p)
1374         __releases(RCU)
1375 {
1376         rcu_read_unlock_sched();
1377         mutex_unlock(&event_mutex);
1378 }
1379
1380 static ssize_t
1381 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382                   loff_t *ppos)
1383 {
1384         struct trace_event_file *file;
1385         unsigned long flags;
1386         char buf[4] = "0";
1387
1388         mutex_lock(&event_mutex);
1389         file = event_file_data(filp);
1390         if (likely(file))
1391                 flags = file->flags;
1392         mutex_unlock(&event_mutex);
1393
1394         if (!file || flags & EVENT_FILE_FL_FREED)
1395                 return -ENODEV;
1396
1397         if (flags & EVENT_FILE_FL_ENABLED &&
1398             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399                 strcpy(buf, "1");
1400
1401         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402             flags & EVENT_FILE_FL_SOFT_MODE)
1403                 strcat(buf, "*");
1404
1405         strcat(buf, "\n");
1406
1407         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408 }
1409
1410 static ssize_t
1411 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412                    loff_t *ppos)
1413 {
1414         struct trace_event_file *file;
1415         unsigned long val;
1416         int ret;
1417
1418         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419         if (ret)
1420                 return ret;
1421
1422         switch (val) {
1423         case 0:
1424         case 1:
1425                 ret = -ENODEV;
1426                 mutex_lock(&event_mutex);
1427                 file = event_file_data(filp);
1428                 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429                         ret = tracing_update_buffers(file->tr);
1430                         if (ret < 0) {
1431                                 mutex_unlock(&event_mutex);
1432                                 return ret;
1433                         }
1434                         ret = ftrace_event_enable_disable(file, val);
1435                 }
1436                 mutex_unlock(&event_mutex);
1437                 break;
1438
1439         default:
1440                 return -EINVAL;
1441         }
1442
1443         *ppos += cnt;
1444
1445         return ret ? ret : cnt;
1446 }
1447
1448 static ssize_t
1449 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450                    loff_t *ppos)
1451 {
1452         const char set_to_char[4] = { '?', '0', '1', 'X' };
1453         struct trace_subsystem_dir *dir = filp->private_data;
1454         struct event_subsystem *system = dir->subsystem;
1455         struct trace_event_call *call;
1456         struct trace_event_file *file;
1457         struct trace_array *tr = dir->tr;
1458         char buf[2];
1459         int set = 0;
1460         int ret;
1461
1462         mutex_lock(&event_mutex);
1463         list_for_each_entry(file, &tr->events, list) {
1464                 call = file->event_call;
1465                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466                     !trace_event_name(call) || !call->class || !call->class->reg)
1467                         continue;
1468
1469                 if (system && strcmp(call->class->system, system->name) != 0)
1470                         continue;
1471
1472                 /*
1473                  * We need to find out if all the events are set
1474                  * or if all events or cleared, or if we have
1475                  * a mixture.
1476                  */
1477                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479                 /*
1480                  * If we have a mixture, no need to look further.
1481                  */
1482                 if (set == 3)
1483                         break;
1484         }
1485         mutex_unlock(&event_mutex);
1486
1487         buf[0] = set_to_char[set];
1488         buf[1] = '\n';
1489
1490         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492         return ret;
1493 }
1494
1495 static ssize_t
1496 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497                     loff_t *ppos)
1498 {
1499         struct trace_subsystem_dir *dir = filp->private_data;
1500         struct event_subsystem *system = dir->subsystem;
1501         const char *name = NULL;
1502         unsigned long val;
1503         ssize_t ret;
1504
1505         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506         if (ret)
1507                 return ret;
1508
1509         ret = tracing_update_buffers(dir->tr);
1510         if (ret < 0)
1511                 return ret;
1512
1513         if (val != 0 && val != 1)
1514                 return -EINVAL;
1515
1516         /*
1517          * Opening of "enable" adds a ref count to system,
1518          * so the name is safe to use.
1519          */
1520         if (system)
1521                 name = system->name;
1522
1523         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524         if (ret)
1525                 goto out;
1526
1527         ret = cnt;
1528
1529 out:
1530         *ppos += cnt;
1531
1532         return ret;
1533 }
1534
1535 enum {
1536         FORMAT_HEADER           = 1,
1537         FORMAT_FIELD_SEPERATOR  = 2,
1538         FORMAT_PRINTFMT         = 3,
1539 };
1540
1541 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542 {
1543         struct trace_event_call *call = event_file_data(m->private);
1544         struct list_head *common_head = &ftrace_common_fields;
1545         struct list_head *head = trace_get_fields(call);
1546         struct list_head *node = v;
1547
1548         (*pos)++;
1549
1550         switch ((unsigned long)v) {
1551         case FORMAT_HEADER:
1552                 node = common_head;
1553                 break;
1554
1555         case FORMAT_FIELD_SEPERATOR:
1556                 node = head;
1557                 break;
1558
1559         case FORMAT_PRINTFMT:
1560                 /* all done */
1561                 return NULL;
1562         }
1563
1564         node = node->prev;
1565         if (node == common_head)
1566                 return (void *)FORMAT_FIELD_SEPERATOR;
1567         else if (node == head)
1568                 return (void *)FORMAT_PRINTFMT;
1569         else
1570                 return node;
1571 }
1572
1573 static int f_show(struct seq_file *m, void *v)
1574 {
1575         struct trace_event_call *call = event_file_data(m->private);
1576         struct ftrace_event_field *field;
1577         const char *array_descriptor;
1578
1579         switch ((unsigned long)v) {
1580         case FORMAT_HEADER:
1581                 seq_printf(m, "name: %s\n", trace_event_name(call));
1582                 seq_printf(m, "ID: %d\n", call->event.type);
1583                 seq_puts(m, "format:\n");
1584                 return 0;
1585
1586         case FORMAT_FIELD_SEPERATOR:
1587                 seq_putc(m, '\n');
1588                 return 0;
1589
1590         case FORMAT_PRINTFMT:
1591                 seq_printf(m, "\nprint fmt: %s\n",
1592                            call->print_fmt);
1593                 return 0;
1594         }
1595
1596         field = list_entry(v, struct ftrace_event_field, link);
1597         /*
1598          * Smartly shows the array type(except dynamic array).
1599          * Normal:
1600          *      field:TYPE VAR
1601          * If TYPE := TYPE[LEN], it is shown:
1602          *      field:TYPE VAR[LEN]
1603          */
1604         array_descriptor = strchr(field->type, '[');
1605
1606         if (str_has_prefix(field->type, "__data_loc"))
1607                 array_descriptor = NULL;
1608
1609         if (!array_descriptor)
1610                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611                            field->type, field->name, field->offset,
1612                            field->size, !!field->is_signed);
1613         else if (field->len)
1614                 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615                            (int)(array_descriptor - field->type),
1616                            field->type, field->name,
1617                            field->len, field->offset,
1618                            field->size, !!field->is_signed);
1619         else
1620                 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621                                 (int)(array_descriptor - field->type),
1622                                 field->type, field->name,
1623                                 field->offset, field->size, !!field->is_signed);
1624
1625         return 0;
1626 }
1627
1628 static void *f_start(struct seq_file *m, loff_t *pos)
1629 {
1630         void *p = (void *)FORMAT_HEADER;
1631         loff_t l = 0;
1632
1633         /* ->stop() is called even if ->start() fails */
1634         mutex_lock(&event_mutex);
1635         if (!event_file_data(m->private))
1636                 return ERR_PTR(-ENODEV);
1637
1638         while (l < *pos && p)
1639                 p = f_next(m, p, &l);
1640
1641         return p;
1642 }
1643
1644 static void f_stop(struct seq_file *m, void *p)
1645 {
1646         mutex_unlock(&event_mutex);
1647 }
1648
1649 static const struct seq_operations trace_format_seq_ops = {
1650         .start          = f_start,
1651         .next           = f_next,
1652         .stop           = f_stop,
1653         .show           = f_show,
1654 };
1655
1656 static int trace_format_open(struct inode *inode, struct file *file)
1657 {
1658         struct seq_file *m;
1659         int ret;
1660
1661         /* Do we want to hide event format files on tracefs lockdown? */
1662
1663         ret = seq_open(file, &trace_format_seq_ops);
1664         if (ret < 0)
1665                 return ret;
1666
1667         m = file->private_data;
1668         m->private = file;
1669
1670         return 0;
1671 }
1672
1673 static ssize_t
1674 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1675 {
1676         int id = (long)event_file_data(filp);
1677         char buf[32];
1678         int len;
1679
1680         if (unlikely(!id))
1681                 return -ENODEV;
1682
1683         len = sprintf(buf, "%d\n", id);
1684
1685         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1686 }
1687
1688 static ssize_t
1689 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1690                   loff_t *ppos)
1691 {
1692         struct trace_event_file *file;
1693         struct trace_seq *s;
1694         int r = -ENODEV;
1695
1696         if (*ppos)
1697                 return 0;
1698
1699         s = kmalloc(sizeof(*s), GFP_KERNEL);
1700
1701         if (!s)
1702                 return -ENOMEM;
1703
1704         trace_seq_init(s);
1705
1706         mutex_lock(&event_mutex);
1707         file = event_file_data(filp);
1708         if (file && !(file->flags & EVENT_FILE_FL_FREED))
1709                 print_event_filter(file, s);
1710         mutex_unlock(&event_mutex);
1711
1712         if (file)
1713                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1714                                             s->buffer, trace_seq_used(s));
1715
1716         kfree(s);
1717
1718         return r;
1719 }
1720
1721 static ssize_t
1722 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1723                    loff_t *ppos)
1724 {
1725         struct trace_event_file *file;
1726         char *buf;
1727         int err = -ENODEV;
1728
1729         if (cnt >= PAGE_SIZE)
1730                 return -EINVAL;
1731
1732         buf = memdup_user_nul(ubuf, cnt);
1733         if (IS_ERR(buf))
1734                 return PTR_ERR(buf);
1735
1736         mutex_lock(&event_mutex);
1737         file = event_file_data(filp);
1738         if (file)
1739                 err = apply_event_filter(file, buf);
1740         mutex_unlock(&event_mutex);
1741
1742         kfree(buf);
1743         if (err < 0)
1744                 return err;
1745
1746         *ppos += cnt;
1747
1748         return cnt;
1749 }
1750
1751 static LIST_HEAD(event_subsystems);
1752
1753 static int subsystem_open(struct inode *inode, struct file *filp)
1754 {
1755         struct trace_subsystem_dir *dir = NULL, *iter_dir;
1756         struct trace_array *tr = NULL, *iter_tr;
1757         struct event_subsystem *system = NULL;
1758         int ret;
1759
1760         if (tracing_is_disabled())
1761                 return -ENODEV;
1762
1763         /* Make sure the system still exists */
1764         mutex_lock(&event_mutex);
1765         mutex_lock(&trace_types_lock);
1766         list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1767                 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1768                         if (iter_dir == inode->i_private) {
1769                                 /* Don't open systems with no events */
1770                                 tr = iter_tr;
1771                                 dir = iter_dir;
1772                                 if (dir->nr_events) {
1773                                         __get_system_dir(dir);
1774                                         system = dir->subsystem;
1775                                 }
1776                                 goto exit_loop;
1777                         }
1778                 }
1779         }
1780  exit_loop:
1781         mutex_unlock(&trace_types_lock);
1782         mutex_unlock(&event_mutex);
1783
1784         if (!system)
1785                 return -ENODEV;
1786
1787         /* Still need to increment the ref count of the system */
1788         if (trace_array_get(tr) < 0) {
1789                 put_system(dir);
1790                 return -ENODEV;
1791         }
1792
1793         ret = tracing_open_generic(inode, filp);
1794         if (ret < 0) {
1795                 trace_array_put(tr);
1796                 put_system(dir);
1797         }
1798
1799         return ret;
1800 }
1801
1802 static int system_tr_open(struct inode *inode, struct file *filp)
1803 {
1804         struct trace_subsystem_dir *dir;
1805         struct trace_array *tr = inode->i_private;
1806         int ret;
1807
1808         /* Make a temporary dir that has no system but points to tr */
1809         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1810         if (!dir)
1811                 return -ENOMEM;
1812
1813         ret = tracing_open_generic_tr(inode, filp);
1814         if (ret < 0) {
1815                 kfree(dir);
1816                 return ret;
1817         }
1818         dir->tr = tr;
1819         filp->private_data = dir;
1820
1821         return 0;
1822 }
1823
1824 static int subsystem_release(struct inode *inode, struct file *file)
1825 {
1826         struct trace_subsystem_dir *dir = file->private_data;
1827
1828         trace_array_put(dir->tr);
1829
1830         /*
1831          * If dir->subsystem is NULL, then this is a temporary
1832          * descriptor that was made for a trace_array to enable
1833          * all subsystems.
1834          */
1835         if (dir->subsystem)
1836                 put_system(dir);
1837         else
1838                 kfree(dir);
1839
1840         return 0;
1841 }
1842
1843 static ssize_t
1844 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1845                       loff_t *ppos)
1846 {
1847         struct trace_subsystem_dir *dir = filp->private_data;
1848         struct event_subsystem *system = dir->subsystem;
1849         struct trace_seq *s;
1850         int r;
1851
1852         if (*ppos)
1853                 return 0;
1854
1855         s = kmalloc(sizeof(*s), GFP_KERNEL);
1856         if (!s)
1857                 return -ENOMEM;
1858
1859         trace_seq_init(s);
1860
1861         print_subsystem_event_filter(system, s);
1862         r = simple_read_from_buffer(ubuf, cnt, ppos,
1863                                     s->buffer, trace_seq_used(s));
1864
1865         kfree(s);
1866
1867         return r;
1868 }
1869
1870 static ssize_t
1871 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1872                        loff_t *ppos)
1873 {
1874         struct trace_subsystem_dir *dir = filp->private_data;
1875         char *buf;
1876         int err;
1877
1878         if (cnt >= PAGE_SIZE)
1879                 return -EINVAL;
1880
1881         buf = memdup_user_nul(ubuf, cnt);
1882         if (IS_ERR(buf))
1883                 return PTR_ERR(buf);
1884
1885         err = apply_subsystem_event_filter(dir, buf);
1886         kfree(buf);
1887         if (err < 0)
1888                 return err;
1889
1890         *ppos += cnt;
1891
1892         return cnt;
1893 }
1894
1895 static ssize_t
1896 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1897 {
1898         int (*func)(struct trace_seq *s) = filp->private_data;
1899         struct trace_seq *s;
1900         int r;
1901
1902         if (*ppos)
1903                 return 0;
1904
1905         s = kmalloc(sizeof(*s), GFP_KERNEL);
1906         if (!s)
1907                 return -ENOMEM;
1908
1909         trace_seq_init(s);
1910
1911         func(s);
1912         r = simple_read_from_buffer(ubuf, cnt, ppos,
1913                                     s->buffer, trace_seq_used(s));
1914
1915         kfree(s);
1916
1917         return r;
1918 }
1919
1920 static void ignore_task_cpu(void *data)
1921 {
1922         struct trace_array *tr = data;
1923         struct trace_pid_list *pid_list;
1924         struct trace_pid_list *no_pid_list;
1925
1926         /*
1927          * This function is called by on_each_cpu() while the
1928          * event_mutex is held.
1929          */
1930         pid_list = rcu_dereference_protected(tr->filtered_pids,
1931                                              mutex_is_locked(&event_mutex));
1932         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1933                                              mutex_is_locked(&event_mutex));
1934
1935         this_cpu_write(tr->array_buffer.data->ignore_pid,
1936                        trace_ignore_this_task(pid_list, no_pid_list, current));
1937 }
1938
1939 static void register_pid_events(struct trace_array *tr)
1940 {
1941         /*
1942          * Register a probe that is called before all other probes
1943          * to set ignore_pid if next or prev do not match.
1944          * Register a probe this is called after all other probes
1945          * to only keep ignore_pid set if next pid matches.
1946          */
1947         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1948                                          tr, INT_MAX);
1949         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1950                                          tr, 0);
1951
1952         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1953                                          tr, INT_MAX);
1954         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1955                                          tr, 0);
1956
1957         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1958                                              tr, INT_MAX);
1959         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1960                                              tr, 0);
1961
1962         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1963                                          tr, INT_MAX);
1964         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1965                                          tr, 0);
1966 }
1967
1968 static ssize_t
1969 event_pid_write(struct file *filp, const char __user *ubuf,
1970                 size_t cnt, loff_t *ppos, int type)
1971 {
1972         struct seq_file *m = filp->private_data;
1973         struct trace_array *tr = m->private;
1974         struct trace_pid_list *filtered_pids = NULL;
1975         struct trace_pid_list *other_pids = NULL;
1976         struct trace_pid_list *pid_list;
1977         struct trace_event_file *file;
1978         ssize_t ret;
1979
1980         if (!cnt)
1981                 return 0;
1982
1983         ret = tracing_update_buffers(tr);
1984         if (ret < 0)
1985                 return ret;
1986
1987         mutex_lock(&event_mutex);
1988
1989         if (type == TRACE_PIDS) {
1990                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1991                                                           lockdep_is_held(&event_mutex));
1992                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1993                                                           lockdep_is_held(&event_mutex));
1994         } else {
1995                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1996                                                           lockdep_is_held(&event_mutex));
1997                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1998                                                           lockdep_is_held(&event_mutex));
1999         }
2000
2001         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2002         if (ret < 0)
2003                 goto out;
2004
2005         if (type == TRACE_PIDS)
2006                 rcu_assign_pointer(tr->filtered_pids, pid_list);
2007         else
2008                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2009
2010         list_for_each_entry(file, &tr->events, list) {
2011                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2012         }
2013
2014         if (filtered_pids) {
2015                 tracepoint_synchronize_unregister();
2016                 trace_pid_list_free(filtered_pids);
2017         } else if (pid_list && !other_pids) {
2018                 register_pid_events(tr);
2019         }
2020
2021         /*
2022          * Ignoring of pids is done at task switch. But we have to
2023          * check for those tasks that are currently running.
2024          * Always do this in case a pid was appended or removed.
2025          */
2026         on_each_cpu(ignore_task_cpu, tr, 1);
2027
2028  out:
2029         mutex_unlock(&event_mutex);
2030
2031         if (ret > 0)
2032                 *ppos += ret;
2033
2034         return ret;
2035 }
2036
2037 static ssize_t
2038 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2039                        size_t cnt, loff_t *ppos)
2040 {
2041         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2042 }
2043
2044 static ssize_t
2045 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2046                         size_t cnt, loff_t *ppos)
2047 {
2048         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2049 }
2050
2051 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2052 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2053 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2054 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2055 static int ftrace_event_release(struct inode *inode, struct file *file);
2056
2057 static const struct seq_operations show_event_seq_ops = {
2058         .start = t_start,
2059         .next = t_next,
2060         .show = t_show,
2061         .stop = t_stop,
2062 };
2063
2064 static const struct seq_operations show_set_event_seq_ops = {
2065         .start = s_start,
2066         .next = s_next,
2067         .show = t_show,
2068         .stop = t_stop,
2069 };
2070
2071 static const struct seq_operations show_set_pid_seq_ops = {
2072         .start = p_start,
2073         .next = p_next,
2074         .show = trace_pid_show,
2075         .stop = p_stop,
2076 };
2077
2078 static const struct seq_operations show_set_no_pid_seq_ops = {
2079         .start = np_start,
2080         .next = np_next,
2081         .show = trace_pid_show,
2082         .stop = p_stop,
2083 };
2084
2085 static const struct file_operations ftrace_avail_fops = {
2086         .open = ftrace_event_avail_open,
2087         .read = seq_read,
2088         .llseek = seq_lseek,
2089         .release = seq_release,
2090 };
2091
2092 static const struct file_operations ftrace_set_event_fops = {
2093         .open = ftrace_event_set_open,
2094         .read = seq_read,
2095         .write = ftrace_event_write,
2096         .llseek = seq_lseek,
2097         .release = ftrace_event_release,
2098 };
2099
2100 static const struct file_operations ftrace_set_event_pid_fops = {
2101         .open = ftrace_event_set_pid_open,
2102         .read = seq_read,
2103         .write = ftrace_event_pid_write,
2104         .llseek = seq_lseek,
2105         .release = ftrace_event_release,
2106 };
2107
2108 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2109         .open = ftrace_event_set_npid_open,
2110         .read = seq_read,
2111         .write = ftrace_event_npid_write,
2112         .llseek = seq_lseek,
2113         .release = ftrace_event_release,
2114 };
2115
2116 static const struct file_operations ftrace_enable_fops = {
2117         .open = tracing_open_file_tr,
2118         .read = event_enable_read,
2119         .write = event_enable_write,
2120         .release = tracing_release_file_tr,
2121         .llseek = default_llseek,
2122 };
2123
2124 static const struct file_operations ftrace_event_format_fops = {
2125         .open = trace_format_open,
2126         .read = seq_read,
2127         .llseek = seq_lseek,
2128         .release = seq_release,
2129 };
2130
2131 static const struct file_operations ftrace_event_id_fops = {
2132         .read = event_id_read,
2133         .llseek = default_llseek,
2134 };
2135
2136 static const struct file_operations ftrace_event_filter_fops = {
2137         .open = tracing_open_file_tr,
2138         .read = event_filter_read,
2139         .write = event_filter_write,
2140         .release = tracing_release_file_tr,
2141         .llseek = default_llseek,
2142 };
2143
2144 static const struct file_operations ftrace_subsystem_filter_fops = {
2145         .open = subsystem_open,
2146         .read = subsystem_filter_read,
2147         .write = subsystem_filter_write,
2148         .llseek = default_llseek,
2149         .release = subsystem_release,
2150 };
2151
2152 static const struct file_operations ftrace_system_enable_fops = {
2153         .open = subsystem_open,
2154         .read = system_enable_read,
2155         .write = system_enable_write,
2156         .llseek = default_llseek,
2157         .release = subsystem_release,
2158 };
2159
2160 static const struct file_operations ftrace_tr_enable_fops = {
2161         .open = system_tr_open,
2162         .read = system_enable_read,
2163         .write = system_enable_write,
2164         .llseek = default_llseek,
2165         .release = subsystem_release,
2166 };
2167
2168 static const struct file_operations ftrace_show_header_fops = {
2169         .open = tracing_open_generic,
2170         .read = show_header,
2171         .llseek = default_llseek,
2172 };
2173
2174 static int
2175 ftrace_event_open(struct inode *inode, struct file *file,
2176                   const struct seq_operations *seq_ops)
2177 {
2178         struct seq_file *m;
2179         int ret;
2180
2181         ret = security_locked_down(LOCKDOWN_TRACEFS);
2182         if (ret)
2183                 return ret;
2184
2185         ret = seq_open(file, seq_ops);
2186         if (ret < 0)
2187                 return ret;
2188         m = file->private_data;
2189         /* copy tr over to seq ops */
2190         m->private = inode->i_private;
2191
2192         return ret;
2193 }
2194
2195 static int ftrace_event_release(struct inode *inode, struct file *file)
2196 {
2197         struct trace_array *tr = inode->i_private;
2198
2199         trace_array_put(tr);
2200
2201         return seq_release(inode, file);
2202 }
2203
2204 static int
2205 ftrace_event_avail_open(struct inode *inode, struct file *file)
2206 {
2207         const struct seq_operations *seq_ops = &show_event_seq_ops;
2208
2209         /* Checks for tracefs lockdown */
2210         return ftrace_event_open(inode, file, seq_ops);
2211 }
2212
2213 static int
2214 ftrace_event_set_open(struct inode *inode, struct file *file)
2215 {
2216         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2217         struct trace_array *tr = inode->i_private;
2218         int ret;
2219
2220         ret = tracing_check_open_get_tr(tr);
2221         if (ret)
2222                 return ret;
2223
2224         if ((file->f_mode & FMODE_WRITE) &&
2225             (file->f_flags & O_TRUNC))
2226                 ftrace_clear_events(tr);
2227
2228         ret = ftrace_event_open(inode, file, seq_ops);
2229         if (ret < 0)
2230                 trace_array_put(tr);
2231         return ret;
2232 }
2233
2234 static int
2235 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2236 {
2237         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2238         struct trace_array *tr = inode->i_private;
2239         int ret;
2240
2241         ret = tracing_check_open_get_tr(tr);
2242         if (ret)
2243                 return ret;
2244
2245         if ((file->f_mode & FMODE_WRITE) &&
2246             (file->f_flags & O_TRUNC))
2247                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2248
2249         ret = ftrace_event_open(inode, file, seq_ops);
2250         if (ret < 0)
2251                 trace_array_put(tr);
2252         return ret;
2253 }
2254
2255 static int
2256 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2257 {
2258         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2259         struct trace_array *tr = inode->i_private;
2260         int ret;
2261
2262         ret = tracing_check_open_get_tr(tr);
2263         if (ret)
2264                 return ret;
2265
2266         if ((file->f_mode & FMODE_WRITE) &&
2267             (file->f_flags & O_TRUNC))
2268                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2269
2270         ret = ftrace_event_open(inode, file, seq_ops);
2271         if (ret < 0)
2272                 trace_array_put(tr);
2273         return ret;
2274 }
2275
2276 static struct event_subsystem *
2277 create_new_subsystem(const char *name)
2278 {
2279         struct event_subsystem *system;
2280
2281         /* need to create new entry */
2282         system = kmalloc(sizeof(*system), GFP_KERNEL);
2283         if (!system)
2284                 return NULL;
2285
2286         system->ref_count = 1;
2287
2288         /* Only allocate if dynamic (kprobes and modules) */
2289         system->name = kstrdup_const(name, GFP_KERNEL);
2290         if (!system->name)
2291                 goto out_free;
2292
2293         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2294         if (!system->filter)
2295                 goto out_free;
2296
2297         list_add(&system->list, &event_subsystems);
2298
2299         return system;
2300
2301  out_free:
2302         kfree_const(system->name);
2303         kfree(system);
2304         return NULL;
2305 }
2306
2307 static int system_callback(const char *name, umode_t *mode, void **data,
2308                     const struct file_operations **fops)
2309 {
2310         if (strcmp(name, "filter") == 0)
2311                 *fops = &ftrace_subsystem_filter_fops;
2312
2313         else if (strcmp(name, "enable") == 0)
2314                 *fops = &ftrace_system_enable_fops;
2315
2316         else
2317                 return 0;
2318
2319         *mode = TRACE_MODE_WRITE;
2320         return 1;
2321 }
2322
2323 static struct eventfs_inode *
2324 event_subsystem_dir(struct trace_array *tr, const char *name,
2325                     struct trace_event_file *file, struct eventfs_inode *parent)
2326 {
2327         struct event_subsystem *system, *iter;
2328         struct trace_subsystem_dir *dir;
2329         struct eventfs_inode *ei;
2330         int nr_entries;
2331         static struct eventfs_entry system_entries[] = {
2332                 {
2333                         .name           = "filter",
2334                         .callback       = system_callback,
2335                 },
2336                 {
2337                         .name           = "enable",
2338                         .callback       = system_callback,
2339                 }
2340         };
2341
2342         /* First see if we did not already create this dir */
2343         list_for_each_entry(dir, &tr->systems, list) {
2344                 system = dir->subsystem;
2345                 if (strcmp(system->name, name) == 0) {
2346                         dir->nr_events++;
2347                         file->system = dir;
2348                         return dir->ei;
2349                 }
2350         }
2351
2352         /* Now see if the system itself exists. */
2353         system = NULL;
2354         list_for_each_entry(iter, &event_subsystems, list) {
2355                 if (strcmp(iter->name, name) == 0) {
2356                         system = iter;
2357                         break;
2358                 }
2359         }
2360
2361         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2362         if (!dir)
2363                 goto out_fail;
2364
2365         if (!system) {
2366                 system = create_new_subsystem(name);
2367                 if (!system)
2368                         goto out_free;
2369         } else
2370                 __get_system(system);
2371
2372         /* ftrace only has directories no files */
2373         if (strcmp(name, "ftrace") == 0)
2374                 nr_entries = 0;
2375         else
2376                 nr_entries = ARRAY_SIZE(system_entries);
2377
2378         ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2379         if (IS_ERR(ei)) {
2380                 pr_warn("Failed to create system directory %s\n", name);
2381                 __put_system(system);
2382                 goto out_free;
2383         }
2384
2385         dir->ei = ei;
2386         dir->tr = tr;
2387         dir->ref_count = 1;
2388         dir->nr_events = 1;
2389         dir->subsystem = system;
2390         file->system = dir;
2391
2392         list_add(&dir->list, &tr->systems);
2393
2394         return dir->ei;
2395
2396  out_free:
2397         kfree(dir);
2398  out_fail:
2399         /* Only print this message if failed on memory allocation */
2400         if (!dir || !system)
2401                 pr_warn("No memory to create event subsystem %s\n", name);
2402         return NULL;
2403 }
2404
2405 static int
2406 event_define_fields(struct trace_event_call *call)
2407 {
2408         struct list_head *head;
2409         int ret = 0;
2410
2411         /*
2412          * Other events may have the same class. Only update
2413          * the fields if they are not already defined.
2414          */
2415         head = trace_get_fields(call);
2416         if (list_empty(head)) {
2417                 struct trace_event_fields *field = call->class->fields_array;
2418                 unsigned int offset = sizeof(struct trace_entry);
2419
2420                 for (; field->type; field++) {
2421                         if (field->type == TRACE_FUNCTION_TYPE) {
2422                                 field->define_fields(call);
2423                                 break;
2424                         }
2425
2426                         offset = ALIGN(offset, field->align);
2427                         ret = trace_define_field_ext(call, field->type, field->name,
2428                                                  offset, field->size,
2429                                                  field->is_signed, field->filter_type,
2430                                                  field->len);
2431                         if (WARN_ON_ONCE(ret)) {
2432                                 pr_err("error code is %d\n", ret);
2433                                 break;
2434                         }
2435
2436                         offset += field->size;
2437                 }
2438         }
2439
2440         return ret;
2441 }
2442
2443 static int event_callback(const char *name, umode_t *mode, void **data,
2444                           const struct file_operations **fops)
2445 {
2446         struct trace_event_file *file = *data;
2447         struct trace_event_call *call = file->event_call;
2448
2449         if (strcmp(name, "format") == 0) {
2450                 *mode = TRACE_MODE_READ;
2451                 *fops = &ftrace_event_format_fops;
2452                 *data = call;
2453                 return 1;
2454         }
2455
2456         /*
2457          * Only event directories that can be enabled should have
2458          * triggers or filters, with the exception of the "print"
2459          * event that can have a "trigger" file.
2460          */
2461         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2462                 if (call->class->reg && strcmp(name, "enable") == 0) {
2463                         *mode = TRACE_MODE_WRITE;
2464                         *fops = &ftrace_enable_fops;
2465                         return 1;
2466                 }
2467
2468                 if (strcmp(name, "filter") == 0) {
2469                         *mode = TRACE_MODE_WRITE;
2470                         *fops = &ftrace_event_filter_fops;
2471                         return 1;
2472                 }
2473         }
2474
2475         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2476             strcmp(trace_event_name(call), "print") == 0) {
2477                 if (strcmp(name, "trigger") == 0) {
2478                         *mode = TRACE_MODE_WRITE;
2479                         *fops = &event_trigger_fops;
2480                         return 1;
2481                 }
2482         }
2483
2484 #ifdef CONFIG_PERF_EVENTS
2485         if (call->event.type && call->class->reg &&
2486             strcmp(name, "id") == 0) {
2487                 *mode = TRACE_MODE_READ;
2488                 *data = (void *)(long)call->event.type;
2489                 *fops = &ftrace_event_id_fops;
2490                 return 1;
2491         }
2492 #endif
2493
2494 #ifdef CONFIG_HIST_TRIGGERS
2495         if (strcmp(name, "hist") == 0) {
2496                 *mode = TRACE_MODE_READ;
2497                 *fops = &event_hist_fops;
2498                 return 1;
2499         }
2500 #endif
2501 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2502         if (strcmp(name, "hist_debug") == 0) {
2503                 *mode = TRACE_MODE_READ;
2504                 *fops = &event_hist_debug_fops;
2505                 return 1;
2506         }
2507 #endif
2508 #ifdef CONFIG_TRACE_EVENT_INJECT
2509         if (call->event.type && call->class->reg &&
2510             strcmp(name, "inject") == 0) {
2511                 *mode = 0200;
2512                 *fops = &event_inject_fops;
2513                 return 1;
2514         }
2515 #endif
2516         return 0;
2517 }
2518
2519 static int
2520 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2521 {
2522         struct trace_event_call *call = file->event_call;
2523         struct trace_array *tr = file->tr;
2524         struct eventfs_inode *e_events;
2525         struct eventfs_inode *ei;
2526         const char *name;
2527         int nr_entries;
2528         int ret;
2529         static struct eventfs_entry event_entries[] = {
2530                 {
2531                         .name           = "enable",
2532                         .callback       = event_callback,
2533                 },
2534                 {
2535                         .name           = "filter",
2536                         .callback       = event_callback,
2537                 },
2538                 {
2539                         .name           = "trigger",
2540                         .callback       = event_callback,
2541                 },
2542                 {
2543                         .name           = "format",
2544                         .callback       = event_callback,
2545                 },
2546 #ifdef CONFIG_PERF_EVENTS
2547                 {
2548                         .name           = "id",
2549                         .callback       = event_callback,
2550                 },
2551 #endif
2552 #ifdef CONFIG_HIST_TRIGGERS
2553                 {
2554                         .name           = "hist",
2555                         .callback       = event_callback,
2556                 },
2557 #endif
2558 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2559                 {
2560                         .name           = "hist_debug",
2561                         .callback       = event_callback,
2562                 },
2563 #endif
2564 #ifdef CONFIG_TRACE_EVENT_INJECT
2565                 {
2566                         .name           = "inject",
2567                         .callback       = event_callback,
2568                 },
2569 #endif
2570         };
2571
2572         /*
2573          * If the trace point header did not define TRACE_SYSTEM
2574          * then the system would be called "TRACE_SYSTEM". This should
2575          * never happen.
2576          */
2577         if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2578                 return -ENODEV;
2579
2580         e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2581         if (!e_events)
2582                 return -ENOMEM;
2583
2584         nr_entries = ARRAY_SIZE(event_entries);
2585
2586         name = trace_event_name(call);
2587         ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2588         if (IS_ERR(ei)) {
2589                 pr_warn("Could not create tracefs '%s' directory\n", name);
2590                 return -1;
2591         }
2592
2593         file->ei = ei;
2594
2595         ret = event_define_fields(call);
2596         if (ret < 0) {
2597                 pr_warn("Could not initialize trace point events/%s\n", name);
2598                 return ret;
2599         }
2600
2601         return 0;
2602 }
2603
2604 static void remove_event_from_tracers(struct trace_event_call *call)
2605 {
2606         struct trace_event_file *file;
2607         struct trace_array *tr;
2608
2609         do_for_each_event_file_safe(tr, file) {
2610                 if (file->event_call != call)
2611                         continue;
2612
2613                 remove_event_file_dir(file);
2614                 /*
2615                  * The do_for_each_event_file_safe() is
2616                  * a double loop. After finding the call for this
2617                  * trace_array, we use break to jump to the next
2618                  * trace_array.
2619                  */
2620                 break;
2621         } while_for_each_event_file();
2622 }
2623
2624 static void event_remove(struct trace_event_call *call)
2625 {
2626         struct trace_array *tr;
2627         struct trace_event_file *file;
2628
2629         do_for_each_event_file(tr, file) {
2630                 if (file->event_call != call)
2631                         continue;
2632
2633                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2634                         tr->clear_trace = true;
2635
2636                 ftrace_event_enable_disable(file, 0);
2637                 /*
2638                  * The do_for_each_event_file() is
2639                  * a double loop. After finding the call for this
2640                  * trace_array, we use break to jump to the next
2641                  * trace_array.
2642                  */
2643                 break;
2644         } while_for_each_event_file();
2645
2646         if (call->event.funcs)
2647                 __unregister_trace_event(&call->event);
2648         remove_event_from_tracers(call);
2649         list_del(&call->list);
2650 }
2651
2652 static int event_init(struct trace_event_call *call)
2653 {
2654         int ret = 0;
2655         const char *name;
2656
2657         name = trace_event_name(call);
2658         if (WARN_ON(!name))
2659                 return -EINVAL;
2660
2661         if (call->class->raw_init) {
2662                 ret = call->class->raw_init(call);
2663                 if (ret < 0 && ret != -ENOSYS)
2664                         pr_warn("Could not initialize trace events/%s\n", name);
2665         }
2666
2667         return ret;
2668 }
2669
2670 static int
2671 __register_event(struct trace_event_call *call, struct module *mod)
2672 {
2673         int ret;
2674
2675         ret = event_init(call);
2676         if (ret < 0)
2677                 return ret;
2678
2679         list_add(&call->list, &ftrace_events);
2680         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2681                 atomic_set(&call->refcnt, 0);
2682         else
2683                 call->module = mod;
2684
2685         return 0;
2686 }
2687
2688 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2689 {
2690         int rlen;
2691         int elen;
2692
2693         /* Find the length of the eval value as a string */
2694         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2695         /* Make sure there's enough room to replace the string with the value */
2696         if (len < elen)
2697                 return NULL;
2698
2699         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2700
2701         /* Get the rest of the string of ptr */
2702         rlen = strlen(ptr + len);
2703         memmove(ptr + elen, ptr + len, rlen);
2704         /* Make sure we end the new string */
2705         ptr[elen + rlen] = 0;
2706
2707         return ptr + elen;
2708 }
2709
2710 static void update_event_printk(struct trace_event_call *call,
2711                                 struct trace_eval_map *map)
2712 {
2713         char *ptr;
2714         int quote = 0;
2715         int len = strlen(map->eval_string);
2716
2717         for (ptr = call->print_fmt; *ptr; ptr++) {
2718                 if (*ptr == '\\') {
2719                         ptr++;
2720                         /* paranoid */
2721                         if (!*ptr)
2722                                 break;
2723                         continue;
2724                 }
2725                 if (*ptr == '"') {
2726                         quote ^= 1;
2727                         continue;
2728                 }
2729                 if (quote)
2730                         continue;
2731                 if (isdigit(*ptr)) {
2732                         /* skip numbers */
2733                         do {
2734                                 ptr++;
2735                                 /* Check for alpha chars like ULL */
2736                         } while (isalnum(*ptr));
2737                         if (!*ptr)
2738                                 break;
2739                         /*
2740                          * A number must have some kind of delimiter after
2741                          * it, and we can ignore that too.
2742                          */
2743                         continue;
2744                 }
2745                 if (isalpha(*ptr) || *ptr == '_') {
2746                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2747                             !isalnum(ptr[len]) && ptr[len] != '_') {
2748                                 ptr = eval_replace(ptr, map, len);
2749                                 /* enum/sizeof string smaller than value */
2750                                 if (WARN_ON_ONCE(!ptr))
2751                                         return;
2752                                 /*
2753                                  * No need to decrement here, as eval_replace()
2754                                  * returns the pointer to the character passed
2755                                  * the eval, and two evals can not be placed
2756                                  * back to back without something in between.
2757                                  * We can skip that something in between.
2758                                  */
2759                                 continue;
2760                         }
2761                 skip_more:
2762                         do {
2763                                 ptr++;
2764                         } while (isalnum(*ptr) || *ptr == '_');
2765                         if (!*ptr)
2766                                 break;
2767                         /*
2768                          * If what comes after this variable is a '.' or
2769                          * '->' then we can continue to ignore that string.
2770                          */
2771                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2772                                 ptr += *ptr == '.' ? 1 : 2;
2773                                 if (!*ptr)
2774                                         break;
2775                                 goto skip_more;
2776                         }
2777                         /*
2778                          * Once again, we can skip the delimiter that came
2779                          * after the string.
2780                          */
2781                         continue;
2782                 }
2783         }
2784 }
2785
2786 static void add_str_to_module(struct module *module, char *str)
2787 {
2788         struct module_string *modstr;
2789
2790         modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2791
2792         /*
2793          * If we failed to allocate memory here, then we'll just
2794          * let the str memory leak when the module is removed.
2795          * If this fails to allocate, there's worse problems than
2796          * a leaked string on module removal.
2797          */
2798         if (WARN_ON_ONCE(!modstr))
2799                 return;
2800
2801         modstr->module = module;
2802         modstr->str = str;
2803
2804         list_add(&modstr->next, &module_strings);
2805 }
2806
2807 static void update_event_fields(struct trace_event_call *call,
2808                                 struct trace_eval_map *map)
2809 {
2810         struct ftrace_event_field *field;
2811         struct list_head *head;
2812         char *ptr;
2813         char *str;
2814         int len = strlen(map->eval_string);
2815
2816         /* Dynamic events should never have field maps */
2817         if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2818                 return;
2819
2820         head = trace_get_fields(call);
2821         list_for_each_entry(field, head, link) {
2822                 ptr = strchr(field->type, '[');
2823                 if (!ptr)
2824                         continue;
2825                 ptr++;
2826
2827                 if (!isalpha(*ptr) && *ptr != '_')
2828                         continue;
2829
2830                 if (strncmp(map->eval_string, ptr, len) != 0)
2831                         continue;
2832
2833                 str = kstrdup(field->type, GFP_KERNEL);
2834                 if (WARN_ON_ONCE(!str))
2835                         return;
2836                 ptr = str + (ptr - field->type);
2837                 ptr = eval_replace(ptr, map, len);
2838                 /* enum/sizeof string smaller than value */
2839                 if (WARN_ON_ONCE(!ptr)) {
2840                         kfree(str);
2841                         continue;
2842                 }
2843
2844                 /*
2845                  * If the event is part of a module, then we need to free the string
2846                  * when the module is removed. Otherwise, it will stay allocated
2847                  * until a reboot.
2848                  */
2849                 if (call->module)
2850                         add_str_to_module(call->module, str);
2851
2852                 field->type = str;
2853         }
2854 }
2855
2856 void trace_event_eval_update(struct trace_eval_map **map, int len)
2857 {
2858         struct trace_event_call *call, *p;
2859         const char *last_system = NULL;
2860         bool first = false;
2861         int last_i;
2862         int i;
2863
2864         down_write(&trace_event_sem);
2865         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2866                 /* events are usually grouped together with systems */
2867                 if (!last_system || call->class->system != last_system) {
2868                         first = true;
2869                         last_i = 0;
2870                         last_system = call->class->system;
2871                 }
2872
2873                 /*
2874                  * Since calls are grouped by systems, the likelihood that the
2875                  * next call in the iteration belongs to the same system as the
2876                  * previous call is high. As an optimization, we skip searching
2877                  * for a map[] that matches the call's system if the last call
2878                  * was from the same system. That's what last_i is for. If the
2879                  * call has the same system as the previous call, then last_i
2880                  * will be the index of the first map[] that has a matching
2881                  * system.
2882                  */
2883                 for (i = last_i; i < len; i++) {
2884                         if (call->class->system == map[i]->system) {
2885                                 /* Save the first system if need be */
2886                                 if (first) {
2887                                         last_i = i;
2888                                         first = false;
2889                                 }
2890                                 update_event_printk(call, map[i]);
2891                                 update_event_fields(call, map[i]);
2892                         }
2893                 }
2894                 cond_resched();
2895         }
2896         up_write(&trace_event_sem);
2897 }
2898
2899 static struct trace_event_file *
2900 trace_create_new_event(struct trace_event_call *call,
2901                        struct trace_array *tr)
2902 {
2903         struct trace_pid_list *no_pid_list;
2904         struct trace_pid_list *pid_list;
2905         struct trace_event_file *file;
2906         unsigned int first;
2907
2908         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2909         if (!file)
2910                 return NULL;
2911
2912         pid_list = rcu_dereference_protected(tr->filtered_pids,
2913                                              lockdep_is_held(&event_mutex));
2914         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2915                                              lockdep_is_held(&event_mutex));
2916
2917         if (!trace_pid_list_first(pid_list, &first) ||
2918             !trace_pid_list_first(no_pid_list, &first))
2919                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2920
2921         file->event_call = call;
2922         file->tr = tr;
2923         atomic_set(&file->sm_ref, 0);
2924         atomic_set(&file->tm_ref, 0);
2925         INIT_LIST_HEAD(&file->triggers);
2926         list_add(&file->list, &tr->events);
2927         event_file_get(file);
2928
2929         return file;
2930 }
2931
2932 #define MAX_BOOT_TRIGGERS 32
2933
2934 static struct boot_triggers {
2935         const char              *event;
2936         char                    *trigger;
2937 } bootup_triggers[MAX_BOOT_TRIGGERS];
2938
2939 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2940 static int nr_boot_triggers;
2941
2942 static __init int setup_trace_triggers(char *str)
2943 {
2944         char *trigger;
2945         char *buf;
2946         int i;
2947
2948         strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
2949         trace_set_ring_buffer_expanded(NULL);
2950         disable_tracing_selftest("running event triggers");
2951
2952         buf = bootup_trigger_buf;
2953         for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
2954                 trigger = strsep(&buf, ",");
2955                 if (!trigger)
2956                         break;
2957                 bootup_triggers[i].event = strsep(&trigger, ".");
2958                 bootup_triggers[i].trigger = trigger;
2959                 if (!bootup_triggers[i].trigger)
2960                         break;
2961         }
2962
2963         nr_boot_triggers = i;
2964         return 1;
2965 }
2966 __setup("trace_trigger=", setup_trace_triggers);
2967
2968 /* Add an event to a trace directory */
2969 static int
2970 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2971 {
2972         struct trace_event_file *file;
2973
2974         file = trace_create_new_event(call, tr);
2975         if (!file)
2976                 return -ENOMEM;
2977
2978         if (eventdir_initialized)
2979                 return event_create_dir(tr->event_dir, file);
2980         else
2981                 return event_define_fields(call);
2982 }
2983
2984 static void trace_early_triggers(struct trace_event_file *file, const char *name)
2985 {
2986         int ret;
2987         int i;
2988
2989         for (i = 0; i < nr_boot_triggers; i++) {
2990                 if (strcmp(name, bootup_triggers[i].event))
2991                         continue;
2992                 mutex_lock(&event_mutex);
2993                 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
2994                 mutex_unlock(&event_mutex);
2995                 if (ret)
2996                         pr_err("Failed to register trigger '%s' on event %s\n",
2997                                bootup_triggers[i].trigger,
2998                                bootup_triggers[i].event);
2999         }
3000 }
3001
3002 /*
3003  * Just create a descriptor for early init. A descriptor is required
3004  * for enabling events at boot. We want to enable events before
3005  * the filesystem is initialized.
3006  */
3007 static int
3008 __trace_early_add_new_event(struct trace_event_call *call,
3009                             struct trace_array *tr)
3010 {
3011         struct trace_event_file *file;
3012         int ret;
3013
3014         file = trace_create_new_event(call, tr);
3015         if (!file)
3016                 return -ENOMEM;
3017
3018         ret = event_define_fields(call);
3019         if (ret)
3020                 return ret;
3021
3022         trace_early_triggers(file, trace_event_name(call));
3023
3024         return 0;
3025 }
3026
3027 struct ftrace_module_file_ops;
3028 static void __add_event_to_tracers(struct trace_event_call *call);
3029
3030 /* Add an additional event_call dynamically */
3031 int trace_add_event_call(struct trace_event_call *call)
3032 {
3033         int ret;
3034         lockdep_assert_held(&event_mutex);
3035
3036         mutex_lock(&trace_types_lock);
3037
3038         ret = __register_event(call, NULL);
3039         if (ret >= 0)
3040                 __add_event_to_tracers(call);
3041
3042         mutex_unlock(&trace_types_lock);
3043         return ret;
3044 }
3045 EXPORT_SYMBOL_GPL(trace_add_event_call);
3046
3047 /*
3048  * Must be called under locking of trace_types_lock, event_mutex and
3049  * trace_event_sem.
3050  */
3051 static void __trace_remove_event_call(struct trace_event_call *call)
3052 {
3053         event_remove(call);
3054         trace_destroy_fields(call);
3055         free_event_filter(call->filter);
3056         call->filter = NULL;
3057 }
3058
3059 static int probe_remove_event_call(struct trace_event_call *call)
3060 {
3061         struct trace_array *tr;
3062         struct trace_event_file *file;
3063
3064 #ifdef CONFIG_PERF_EVENTS
3065         if (call->perf_refcount)
3066                 return -EBUSY;
3067 #endif
3068         do_for_each_event_file(tr, file) {
3069                 if (file->event_call != call)
3070                         continue;
3071                 /*
3072                  * We can't rely on ftrace_event_enable_disable(enable => 0)
3073                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3074                  * TRACE_REG_UNREGISTER.
3075                  */
3076                 if (file->flags & EVENT_FILE_FL_ENABLED)
3077                         goto busy;
3078
3079                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3080                         tr->clear_trace = true;
3081                 /*
3082                  * The do_for_each_event_file_safe() is
3083                  * a double loop. After finding the call for this
3084                  * trace_array, we use break to jump to the next
3085                  * trace_array.
3086                  */
3087                 break;
3088         } while_for_each_event_file();
3089
3090         __trace_remove_event_call(call);
3091
3092         return 0;
3093  busy:
3094         /* No need to clear the trace now */
3095         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3096                 tr->clear_trace = false;
3097         }
3098         return -EBUSY;
3099 }
3100
3101 /* Remove an event_call */
3102 int trace_remove_event_call(struct trace_event_call *call)
3103 {
3104         int ret;
3105
3106         lockdep_assert_held(&event_mutex);
3107
3108         mutex_lock(&trace_types_lock);
3109         down_write(&trace_event_sem);
3110         ret = probe_remove_event_call(call);
3111         up_write(&trace_event_sem);
3112         mutex_unlock(&trace_types_lock);
3113
3114         return ret;
3115 }
3116 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3117
3118 #define for_each_event(event, start, end)                       \
3119         for (event = start;                                     \
3120              (unsigned long)event < (unsigned long)end;         \
3121              event++)
3122
3123 #ifdef CONFIG_MODULES
3124
3125 static void trace_module_add_events(struct module *mod)
3126 {
3127         struct trace_event_call **call, **start, **end;
3128
3129         if (!mod->num_trace_events)
3130                 return;
3131
3132         /* Don't add infrastructure for mods without tracepoints */
3133         if (trace_module_has_bad_taint(mod)) {
3134                 pr_err("%s: module has bad taint, not creating trace events\n",
3135                        mod->name);
3136                 return;
3137         }
3138
3139         start = mod->trace_events;
3140         end = mod->trace_events + mod->num_trace_events;
3141
3142         for_each_event(call, start, end) {
3143                 __register_event(*call, mod);
3144                 __add_event_to_tracers(*call);
3145         }
3146 }
3147
3148 static void trace_module_remove_events(struct module *mod)
3149 {
3150         struct trace_event_call *call, *p;
3151         struct module_string *modstr, *m;
3152
3153         down_write(&trace_event_sem);
3154         list_for_each_entry_safe(call, p, &ftrace_events, list) {
3155                 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3156                         continue;
3157                 if (call->module == mod)
3158                         __trace_remove_event_call(call);
3159         }
3160         /* Check for any strings allocade for this module */
3161         list_for_each_entry_safe(modstr, m, &module_strings, next) {
3162                 if (modstr->module != mod)
3163                         continue;
3164                 list_del(&modstr->next);
3165                 kfree(modstr->str);
3166                 kfree(modstr);
3167         }
3168         up_write(&trace_event_sem);
3169
3170         /*
3171          * It is safest to reset the ring buffer if the module being unloaded
3172          * registered any events that were used. The only worry is if
3173          * a new module gets loaded, and takes on the same id as the events
3174          * of this module. When printing out the buffer, traced events left
3175          * over from this module may be passed to the new module events and
3176          * unexpected results may occur.
3177          */
3178         tracing_reset_all_online_cpus_unlocked();
3179 }
3180
3181 static int trace_module_notify(struct notifier_block *self,
3182                                unsigned long val, void *data)
3183 {
3184         struct module *mod = data;
3185
3186         mutex_lock(&event_mutex);
3187         mutex_lock(&trace_types_lock);
3188         switch (val) {
3189         case MODULE_STATE_COMING:
3190                 trace_module_add_events(mod);
3191                 break;
3192         case MODULE_STATE_GOING:
3193                 trace_module_remove_events(mod);
3194                 break;
3195         }
3196         mutex_unlock(&trace_types_lock);
3197         mutex_unlock(&event_mutex);
3198
3199         return NOTIFY_OK;
3200 }
3201
3202 static struct notifier_block trace_module_nb = {
3203         .notifier_call = trace_module_notify,
3204         .priority = 1, /* higher than trace.c module notify */
3205 };
3206 #endif /* CONFIG_MODULES */
3207
3208 /* Create a new event directory structure for a trace directory. */
3209 static void
3210 __trace_add_event_dirs(struct trace_array *tr)
3211 {
3212         struct trace_event_call *call;
3213         int ret;
3214
3215         list_for_each_entry(call, &ftrace_events, list) {
3216                 ret = __trace_add_new_event(call, tr);
3217                 if (ret < 0)
3218                         pr_warn("Could not create directory for event %s\n",
3219                                 trace_event_name(call));
3220         }
3221 }
3222
3223 /* Returns any file that matches the system and event */
3224 struct trace_event_file *
3225 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3226 {
3227         struct trace_event_file *file;
3228         struct trace_event_call *call;
3229         const char *name;
3230
3231         list_for_each_entry(file, &tr->events, list) {
3232
3233                 call = file->event_call;
3234                 name = trace_event_name(call);
3235
3236                 if (!name || !call->class)
3237                         continue;
3238
3239                 if (strcmp(event, name) == 0 &&
3240                     strcmp(system, call->class->system) == 0)
3241                         return file;
3242         }
3243         return NULL;
3244 }
3245
3246 /* Returns valid trace event files that match system and event */
3247 struct trace_event_file *
3248 find_event_file(struct trace_array *tr, const char *system, const char *event)
3249 {
3250         struct trace_event_file *file;
3251
3252         file = __find_event_file(tr, system, event);
3253         if (!file || !file->event_call->class->reg ||
3254             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3255                 return NULL;
3256
3257         return file;
3258 }
3259
3260 /**
3261  * trace_get_event_file - Find and return a trace event file
3262  * @instance: The name of the trace instance containing the event
3263  * @system: The name of the system containing the event
3264  * @event: The name of the event
3265  *
3266  * Return a trace event file given the trace instance name, trace
3267  * system, and trace event name.  If the instance name is NULL, it
3268  * refers to the top-level trace array.
3269  *
3270  * This function will look it up and return it if found, after calling
3271  * trace_array_get() to prevent the instance from going away, and
3272  * increment the event's module refcount to prevent it from being
3273  * removed.
3274  *
3275  * To release the file, call trace_put_event_file(), which will call
3276  * trace_array_put() and decrement the event's module refcount.
3277  *
3278  * Return: The trace event on success, ERR_PTR otherwise.
3279  */
3280 struct trace_event_file *trace_get_event_file(const char *instance,
3281                                               const char *system,
3282                                               const char *event)
3283 {
3284         struct trace_array *tr = top_trace_array();
3285         struct trace_event_file *file = NULL;
3286         int ret = -EINVAL;
3287
3288         if (instance) {
3289                 tr = trace_array_find_get(instance);
3290                 if (!tr)
3291                         return ERR_PTR(-ENOENT);
3292         } else {
3293                 ret = trace_array_get(tr);
3294                 if (ret)
3295                         return ERR_PTR(ret);
3296         }
3297
3298         mutex_lock(&event_mutex);
3299
3300         file = find_event_file(tr, system, event);
3301         if (!file) {
3302                 trace_array_put(tr);
3303                 ret = -EINVAL;
3304                 goto out;
3305         }
3306
3307         /* Don't let event modules unload while in use */
3308         ret = trace_event_try_get_ref(file->event_call);
3309         if (!ret) {
3310                 trace_array_put(tr);
3311                 ret = -EBUSY;
3312                 goto out;
3313         }
3314
3315         ret = 0;
3316  out:
3317         mutex_unlock(&event_mutex);
3318
3319         if (ret)
3320                 file = ERR_PTR(ret);
3321
3322         return file;
3323 }
3324 EXPORT_SYMBOL_GPL(trace_get_event_file);
3325
3326 /**
3327  * trace_put_event_file - Release a file from trace_get_event_file()
3328  * @file: The trace event file
3329  *
3330  * If a file was retrieved using trace_get_event_file(), this should
3331  * be called when it's no longer needed.  It will cancel the previous
3332  * trace_array_get() called by that function, and decrement the
3333  * event's module refcount.
3334  */
3335 void trace_put_event_file(struct trace_event_file *file)
3336 {
3337         mutex_lock(&event_mutex);
3338         trace_event_put_ref(file->event_call);
3339         mutex_unlock(&event_mutex);
3340
3341         trace_array_put(file->tr);
3342 }
3343 EXPORT_SYMBOL_GPL(trace_put_event_file);
3344
3345 #ifdef CONFIG_DYNAMIC_FTRACE
3346
3347 /* Avoid typos */
3348 #define ENABLE_EVENT_STR        "enable_event"
3349 #define DISABLE_EVENT_STR       "disable_event"
3350
3351 struct event_probe_data {
3352         struct trace_event_file *file;
3353         unsigned long                   count;
3354         int                             ref;
3355         bool                            enable;
3356 };
3357
3358 static void update_event_probe(struct event_probe_data *data)
3359 {
3360         if (data->enable)
3361                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3362         else
3363                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3364 }
3365
3366 static void
3367 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3368                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3369                    void *data)
3370 {
3371         struct ftrace_func_mapper *mapper = data;
3372         struct event_probe_data *edata;
3373         void **pdata;
3374
3375         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3376         if (!pdata || !*pdata)
3377                 return;
3378
3379         edata = *pdata;
3380         update_event_probe(edata);
3381 }
3382
3383 static void
3384 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3385                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3386                          void *data)
3387 {
3388         struct ftrace_func_mapper *mapper = data;
3389         struct event_probe_data *edata;
3390         void **pdata;
3391
3392         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3393         if (!pdata || !*pdata)
3394                 return;
3395
3396         edata = *pdata;
3397
3398         if (!edata->count)
3399                 return;
3400
3401         /* Skip if the event is in a state we want to switch to */
3402         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3403                 return;
3404
3405         if (edata->count != -1)
3406                 (edata->count)--;
3407
3408         update_event_probe(edata);
3409 }
3410
3411 static int
3412 event_enable_print(struct seq_file *m, unsigned long ip,
3413                    struct ftrace_probe_ops *ops, void *data)
3414 {
3415         struct ftrace_func_mapper *mapper = data;
3416         struct event_probe_data *edata;
3417         void **pdata;
3418
3419         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3420
3421         if (WARN_ON_ONCE(!pdata || !*pdata))
3422                 return 0;
3423
3424         edata = *pdata;
3425
3426         seq_printf(m, "%ps:", (void *)ip);
3427
3428         seq_printf(m, "%s:%s:%s",
3429                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3430                    edata->file->event_call->class->system,
3431                    trace_event_name(edata->file->event_call));
3432
3433         if (edata->count == -1)
3434                 seq_puts(m, ":unlimited\n");
3435         else
3436                 seq_printf(m, ":count=%ld\n", edata->count);
3437
3438         return 0;
3439 }
3440
3441 static int
3442 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3443                   unsigned long ip, void *init_data, void **data)
3444 {
3445         struct ftrace_func_mapper *mapper = *data;
3446         struct event_probe_data *edata = init_data;
3447         int ret;
3448
3449         if (!mapper) {
3450                 mapper = allocate_ftrace_func_mapper();
3451                 if (!mapper)
3452                         return -ENODEV;
3453                 *data = mapper;
3454         }
3455
3456         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3457         if (ret < 0)
3458                 return ret;
3459
3460         edata->ref++;
3461
3462         return 0;
3463 }
3464
3465 static int free_probe_data(void *data)
3466 {
3467         struct event_probe_data *edata = data;
3468
3469         edata->ref--;
3470         if (!edata->ref) {
3471                 /* Remove the SOFT_MODE flag */
3472                 __ftrace_event_enable_disable(edata->file, 0, 1);
3473                 trace_event_put_ref(edata->file->event_call);
3474                 kfree(edata);
3475         }
3476         return 0;
3477 }
3478
3479 static void
3480 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3481                   unsigned long ip, void *data)
3482 {
3483         struct ftrace_func_mapper *mapper = data;
3484         struct event_probe_data *edata;
3485
3486         if (!ip) {
3487                 if (!mapper)
3488                         return;
3489                 free_ftrace_func_mapper(mapper, free_probe_data);
3490                 return;
3491         }
3492
3493         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3494
3495         if (WARN_ON_ONCE(!edata))
3496                 return;
3497
3498         if (WARN_ON_ONCE(edata->ref <= 0))
3499                 return;
3500
3501         free_probe_data(edata);
3502 }
3503
3504 static struct ftrace_probe_ops event_enable_probe_ops = {
3505         .func                   = event_enable_probe,
3506         .print                  = event_enable_print,
3507         .init                   = event_enable_init,
3508         .free                   = event_enable_free,
3509 };
3510
3511 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3512         .func                   = event_enable_count_probe,
3513         .print                  = event_enable_print,
3514         .init                   = event_enable_init,
3515         .free                   = event_enable_free,
3516 };
3517
3518 static struct ftrace_probe_ops event_disable_probe_ops = {
3519         .func                   = event_enable_probe,
3520         .print                  = event_enable_print,
3521         .init                   = event_enable_init,
3522         .free                   = event_enable_free,
3523 };
3524
3525 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3526         .func                   = event_enable_count_probe,
3527         .print                  = event_enable_print,
3528         .init                   = event_enable_init,
3529         .free                   = event_enable_free,
3530 };
3531
3532 static int
3533 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3534                   char *glob, char *cmd, char *param, int enabled)
3535 {
3536         struct trace_event_file *file;
3537         struct ftrace_probe_ops *ops;
3538         struct event_probe_data *data;
3539         const char *system;
3540         const char *event;
3541         char *number;
3542         bool enable;
3543         int ret;
3544
3545         if (!tr)
3546                 return -ENODEV;
3547
3548         /* hash funcs only work with set_ftrace_filter */
3549         if (!enabled || !param)
3550                 return -EINVAL;
3551
3552         system = strsep(&param, ":");
3553         if (!param)
3554                 return -EINVAL;
3555
3556         event = strsep(&param, ":");
3557
3558         mutex_lock(&event_mutex);
3559
3560         ret = -EINVAL;
3561         file = find_event_file(tr, system, event);
3562         if (!file)
3563                 goto out;
3564
3565         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3566
3567         if (enable)
3568                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3569         else
3570                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3571
3572         if (glob[0] == '!') {
3573                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3574                 goto out;
3575         }
3576
3577         ret = -ENOMEM;
3578
3579         data = kzalloc(sizeof(*data), GFP_KERNEL);
3580         if (!data)
3581                 goto out;
3582
3583         data->enable = enable;
3584         data->count = -1;
3585         data->file = file;
3586
3587         if (!param)
3588                 goto out_reg;
3589
3590         number = strsep(&param, ":");
3591
3592         ret = -EINVAL;
3593         if (!strlen(number))
3594                 goto out_free;
3595
3596         /*
3597          * We use the callback data field (which is a pointer)
3598          * as our counter.
3599          */
3600         ret = kstrtoul(number, 0, &data->count);
3601         if (ret)
3602                 goto out_free;
3603
3604  out_reg:
3605         /* Don't let event modules unload while probe registered */
3606         ret = trace_event_try_get_ref(file->event_call);
3607         if (!ret) {
3608                 ret = -EBUSY;
3609                 goto out_free;
3610         }
3611
3612         ret = __ftrace_event_enable_disable(file, 1, 1);
3613         if (ret < 0)
3614                 goto out_put;
3615
3616         ret = register_ftrace_function_probe(glob, tr, ops, data);
3617         /*
3618          * The above returns on success the # of functions enabled,
3619          * but if it didn't find any functions it returns zero.
3620          * Consider no functions a failure too.
3621          */
3622         if (!ret) {
3623                 ret = -ENOENT;
3624                 goto out_disable;
3625         } else if (ret < 0)
3626                 goto out_disable;
3627         /* Just return zero, not the number of enabled functions */
3628         ret = 0;
3629  out:
3630         mutex_unlock(&event_mutex);
3631         return ret;
3632
3633  out_disable:
3634         __ftrace_event_enable_disable(file, 0, 1);
3635  out_put:
3636         trace_event_put_ref(file->event_call);
3637  out_free:
3638         kfree(data);
3639         goto out;
3640 }
3641
3642 static struct ftrace_func_command event_enable_cmd = {
3643         .name                   = ENABLE_EVENT_STR,
3644         .func                   = event_enable_func,
3645 };
3646
3647 static struct ftrace_func_command event_disable_cmd = {
3648         .name                   = DISABLE_EVENT_STR,
3649         .func                   = event_enable_func,
3650 };
3651
3652 static __init int register_event_cmds(void)
3653 {
3654         int ret;
3655
3656         ret = register_ftrace_command(&event_enable_cmd);
3657         if (WARN_ON(ret < 0))
3658                 return ret;
3659         ret = register_ftrace_command(&event_disable_cmd);
3660         if (WARN_ON(ret < 0))
3661                 unregister_ftrace_command(&event_enable_cmd);
3662         return ret;
3663 }
3664 #else
3665 static inline int register_event_cmds(void) { return 0; }
3666 #endif /* CONFIG_DYNAMIC_FTRACE */
3667
3668 /*
3669  * The top level array and trace arrays created by boot-time tracing
3670  * have already had its trace_event_file descriptors created in order
3671  * to allow for early events to be recorded.
3672  * This function is called after the tracefs has been initialized,
3673  * and we now have to create the files associated to the events.
3674  */
3675 static void __trace_early_add_event_dirs(struct trace_array *tr)
3676 {
3677         struct trace_event_file *file;
3678         int ret;
3679
3680
3681         list_for_each_entry(file, &tr->events, list) {
3682                 ret = event_create_dir(tr->event_dir, file);
3683                 if (ret < 0)
3684                         pr_warn("Could not create directory for event %s\n",
3685                                 trace_event_name(file->event_call));
3686         }
3687 }
3688
3689 /*
3690  * For early boot up, the top trace array and the trace arrays created
3691  * by boot-time tracing require to have a list of events that can be
3692  * enabled. This must be done before the filesystem is set up in order
3693  * to allow events to be traced early.
3694  */
3695 void __trace_early_add_events(struct trace_array *tr)
3696 {
3697         struct trace_event_call *call;
3698         int ret;
3699
3700         list_for_each_entry(call, &ftrace_events, list) {
3701                 /* Early boot up should not have any modules loaded */
3702                 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3703                     WARN_ON_ONCE(call->module))
3704                         continue;
3705
3706                 ret = __trace_early_add_new_event(call, tr);
3707                 if (ret < 0)
3708                         pr_warn("Could not create early event %s\n",
3709                                 trace_event_name(call));
3710         }
3711 }
3712
3713 /* Remove the event directory structure for a trace directory. */
3714 static void
3715 __trace_remove_event_dirs(struct trace_array *tr)
3716 {
3717         struct trace_event_file *file, *next;
3718
3719         list_for_each_entry_safe(file, next, &tr->events, list)
3720                 remove_event_file_dir(file);
3721 }
3722
3723 static void __add_event_to_tracers(struct trace_event_call *call)
3724 {
3725         struct trace_array *tr;
3726
3727         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3728                 __trace_add_new_event(call, tr);
3729 }
3730
3731 extern struct trace_event_call *__start_ftrace_events[];
3732 extern struct trace_event_call *__stop_ftrace_events[];
3733
3734 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3735
3736 static __init int setup_trace_event(char *str)
3737 {
3738         strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3739         trace_set_ring_buffer_expanded(NULL);
3740         disable_tracing_selftest("running event tracing");
3741
3742         return 1;
3743 }
3744 __setup("trace_event=", setup_trace_event);
3745
3746 static int events_callback(const char *name, umode_t *mode, void **data,
3747                            const struct file_operations **fops)
3748 {
3749         if (strcmp(name, "enable") == 0) {
3750                 *mode = TRACE_MODE_WRITE;
3751                 *fops = &ftrace_tr_enable_fops;
3752                 return 1;
3753         }
3754
3755         if (strcmp(name, "header_page") == 0)
3756                 *data = ring_buffer_print_page_header;
3757
3758         else if (strcmp(name, "header_event") == 0)
3759                 *data = ring_buffer_print_entry_header;
3760
3761         else
3762                 return 0;
3763
3764         *mode = TRACE_MODE_READ;
3765         *fops = &ftrace_show_header_fops;
3766         return 1;
3767 }
3768
3769 /* Expects to have event_mutex held when called */
3770 static int
3771 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3772 {
3773         struct eventfs_inode *e_events;
3774         struct dentry *entry;
3775         int nr_entries;
3776         static struct eventfs_entry events_entries[] = {
3777                 {
3778                         .name           = "enable",
3779                         .callback       = events_callback,
3780                 },
3781                 {
3782                         .name           = "header_page",
3783                         .callback       = events_callback,
3784                 },
3785                 {
3786                         .name           = "header_event",
3787                         .callback       = events_callback,
3788                 },
3789         };
3790
3791         entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3792                                   tr, &ftrace_set_event_fops);
3793         if (!entry)
3794                 return -ENOMEM;
3795
3796         nr_entries = ARRAY_SIZE(events_entries);
3797
3798         e_events = eventfs_create_events_dir("events", parent, events_entries,
3799                                              nr_entries, tr);
3800         if (IS_ERR(e_events)) {
3801                 pr_warn("Could not create tracefs 'events' directory\n");
3802                 return -ENOMEM;
3803         }
3804
3805         /* There are not as crucial, just warn if they are not created */
3806
3807         trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3808                           tr, &ftrace_set_event_pid_fops);
3809
3810         trace_create_file("set_event_notrace_pid",
3811                           TRACE_MODE_WRITE, parent, tr,
3812                           &ftrace_set_event_notrace_pid_fops);
3813
3814         tr->event_dir = e_events;
3815
3816         return 0;
3817 }
3818
3819 /**
3820  * event_trace_add_tracer - add a instance of a trace_array to events
3821  * @parent: The parent dentry to place the files/directories for events in
3822  * @tr: The trace array associated with these events
3823  *
3824  * When a new instance is created, it needs to set up its events
3825  * directory, as well as other files associated with events. It also
3826  * creates the event hierarchy in the @parent/events directory.
3827  *
3828  * Returns 0 on success.
3829  *
3830  * Must be called with event_mutex held.
3831  */
3832 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3833 {
3834         int ret;
3835
3836         lockdep_assert_held(&event_mutex);
3837
3838         ret = create_event_toplevel_files(parent, tr);
3839         if (ret)
3840                 goto out;
3841
3842         down_write(&trace_event_sem);
3843         /* If tr already has the event list, it is initialized in early boot. */
3844         if (unlikely(!list_empty(&tr->events)))
3845                 __trace_early_add_event_dirs(tr);
3846         else
3847                 __trace_add_event_dirs(tr);
3848         up_write(&trace_event_sem);
3849
3850  out:
3851         return ret;
3852 }
3853
3854 /*
3855  * The top trace array already had its file descriptors created.
3856  * Now the files themselves need to be created.
3857  */
3858 static __init int
3859 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3860 {
3861         int ret;
3862
3863         mutex_lock(&event_mutex);
3864
3865         ret = create_event_toplevel_files(parent, tr);
3866         if (ret)
3867                 goto out_unlock;
3868
3869         down_write(&trace_event_sem);
3870         __trace_early_add_event_dirs(tr);
3871         up_write(&trace_event_sem);
3872
3873  out_unlock:
3874         mutex_unlock(&event_mutex);
3875
3876         return ret;
3877 }
3878
3879 /* Must be called with event_mutex held */
3880 int event_trace_del_tracer(struct trace_array *tr)
3881 {
3882         lockdep_assert_held(&event_mutex);
3883
3884         /* Disable any event triggers and associated soft-disabled events */
3885         clear_event_triggers(tr);
3886
3887         /* Clear the pid list */
3888         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3889
3890         /* Disable any running events */
3891         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3892
3893         /* Make sure no more events are being executed */
3894         tracepoint_synchronize_unregister();
3895
3896         down_write(&trace_event_sem);
3897         __trace_remove_event_dirs(tr);
3898         eventfs_remove_events_dir(tr->event_dir);
3899         up_write(&trace_event_sem);
3900
3901         tr->event_dir = NULL;
3902
3903         return 0;
3904 }
3905
3906 static __init int event_trace_memsetup(void)
3907 {
3908         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3909         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3910         return 0;
3911 }
3912
3913 __init void
3914 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3915 {
3916         char *token;
3917         int ret;
3918
3919         while (true) {
3920                 token = strsep(&buf, ",");
3921
3922                 if (!token)
3923                         break;
3924
3925                 if (*token) {
3926                         /* Restarting syscalls requires that we stop them first */
3927                         if (disable_first)
3928                                 ftrace_set_clr_event(tr, token, 0);
3929
3930                         ret = ftrace_set_clr_event(tr, token, 1);
3931                         if (ret)
3932                                 pr_warn("Failed to enable trace event: %s\n", token);
3933                 }
3934
3935                 /* Put back the comma to allow this to be called again */
3936                 if (buf)
3937                         *(buf - 1) = ',';
3938         }
3939 }
3940
3941 static __init int event_trace_enable(void)
3942 {
3943         struct trace_array *tr = top_trace_array();
3944         struct trace_event_call **iter, *call;
3945         int ret;
3946
3947         if (!tr)
3948                 return -ENODEV;
3949
3950         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3951
3952                 call = *iter;
3953                 ret = event_init(call);
3954                 if (!ret)
3955                         list_add(&call->list, &ftrace_events);
3956         }
3957
3958         register_trigger_cmds();
3959
3960         /*
3961          * We need the top trace array to have a working set of trace
3962          * points at early init, before the debug files and directories
3963          * are created. Create the file entries now, and attach them
3964          * to the actual file dentries later.
3965          */
3966         __trace_early_add_events(tr);
3967
3968         early_enable_events(tr, bootup_event_buf, false);
3969
3970         trace_printk_start_comm();
3971
3972         register_event_cmds();
3973
3974
3975         return 0;
3976 }
3977
3978 /*
3979  * event_trace_enable() is called from trace_event_init() first to
3980  * initialize events and perhaps start any events that are on the
3981  * command line. Unfortunately, there are some events that will not
3982  * start this early, like the system call tracepoints that need
3983  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3984  * event_trace_enable() is called before pid 1 starts, and this flag
3985  * is never set, making the syscall tracepoint never get reached, but
3986  * the event is enabled regardless (and not doing anything).
3987  */
3988 static __init int event_trace_enable_again(void)
3989 {
3990         struct trace_array *tr;
3991
3992         tr = top_trace_array();
3993         if (!tr)
3994                 return -ENODEV;
3995
3996         early_enable_events(tr, bootup_event_buf, true);
3997
3998         return 0;
3999 }
4000
4001 early_initcall(event_trace_enable_again);
4002
4003 /* Init fields which doesn't related to the tracefs */
4004 static __init int event_trace_init_fields(void)
4005 {
4006         if (trace_define_generic_fields())
4007                 pr_warn("tracing: Failed to allocated generic fields");
4008
4009         if (trace_define_common_fields())
4010                 pr_warn("tracing: Failed to allocate common fields");
4011
4012         return 0;
4013 }
4014
4015 __init int event_trace_init(void)
4016 {
4017         struct trace_array *tr;
4018         int ret;
4019
4020         tr = top_trace_array();
4021         if (!tr)
4022                 return -ENODEV;
4023
4024         trace_create_file("available_events", TRACE_MODE_READ,
4025                           NULL, tr, &ftrace_avail_fops);
4026
4027         ret = early_event_add_tracer(NULL, tr);
4028         if (ret)
4029                 return ret;
4030
4031 #ifdef CONFIG_MODULES
4032         ret = register_module_notifier(&trace_module_nb);
4033         if (ret)
4034                 pr_warn("Failed to register trace events module notifier\n");
4035 #endif
4036
4037         eventdir_initialized = true;
4038
4039         return 0;
4040 }
4041
4042 void __init trace_event_init(void)
4043 {
4044         event_trace_memsetup();
4045         init_ftrace_syscalls();
4046         event_trace_enable();
4047         event_trace_init_fields();
4048 }
4049
4050 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4051
4052 static DEFINE_SPINLOCK(test_spinlock);
4053 static DEFINE_SPINLOCK(test_spinlock_irq);
4054 static DEFINE_MUTEX(test_mutex);
4055
4056 static __init void test_work(struct work_struct *dummy)
4057 {
4058         spin_lock(&test_spinlock);
4059         spin_lock_irq(&test_spinlock_irq);
4060         udelay(1);
4061         spin_unlock_irq(&test_spinlock_irq);
4062         spin_unlock(&test_spinlock);
4063
4064         mutex_lock(&test_mutex);
4065         msleep(1);
4066         mutex_unlock(&test_mutex);
4067 }
4068
4069 static __init int event_test_thread(void *unused)
4070 {
4071         void *test_malloc;
4072
4073         test_malloc = kmalloc(1234, GFP_KERNEL);
4074         if (!test_malloc)
4075                 pr_info("failed to kmalloc\n");
4076
4077         schedule_on_each_cpu(test_work);
4078
4079         kfree(test_malloc);
4080
4081         set_current_state(TASK_INTERRUPTIBLE);
4082         while (!kthread_should_stop()) {
4083                 schedule();
4084                 set_current_state(TASK_INTERRUPTIBLE);
4085         }
4086         __set_current_state(TASK_RUNNING);
4087
4088         return 0;
4089 }
4090
4091 /*
4092  * Do various things that may trigger events.
4093  */
4094 static __init void event_test_stuff(void)
4095 {
4096         struct task_struct *test_thread;
4097
4098         test_thread = kthread_run(event_test_thread, NULL, "test-events");
4099         msleep(1);
4100         kthread_stop(test_thread);
4101 }
4102
4103 /*
4104  * For every trace event defined, we will test each trace point separately,
4105  * and then by groups, and finally all trace points.
4106  */
4107 static __init void event_trace_self_tests(void)
4108 {
4109         struct trace_subsystem_dir *dir;
4110         struct trace_event_file *file;
4111         struct trace_event_call *call;
4112         struct event_subsystem *system;
4113         struct trace_array *tr;
4114         int ret;
4115
4116         tr = top_trace_array();
4117         if (!tr)
4118                 return;
4119
4120         pr_info("Running tests on trace events:\n");
4121
4122         list_for_each_entry(file, &tr->events, list) {
4123
4124                 call = file->event_call;
4125
4126                 /* Only test those that have a probe */
4127                 if (!call->class || !call->class->probe)
4128                         continue;
4129
4130 /*
4131  * Testing syscall events here is pretty useless, but
4132  * we still do it if configured. But this is time consuming.
4133  * What we really need is a user thread to perform the
4134  * syscalls as we test.
4135  */
4136 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4137                 if (call->class->system &&
4138                     strcmp(call->class->system, "syscalls") == 0)
4139                         continue;
4140 #endif
4141
4142                 pr_info("Testing event %s: ", trace_event_name(call));
4143
4144                 /*
4145                  * If an event is already enabled, someone is using
4146                  * it and the self test should not be on.
4147                  */
4148                 if (file->flags & EVENT_FILE_FL_ENABLED) {
4149                         pr_warn("Enabled event during self test!\n");
4150                         WARN_ON_ONCE(1);
4151                         continue;
4152                 }
4153
4154                 ftrace_event_enable_disable(file, 1);
4155                 event_test_stuff();
4156                 ftrace_event_enable_disable(file, 0);
4157
4158                 pr_cont("OK\n");
4159         }
4160
4161         /* Now test at the sub system level */
4162
4163         pr_info("Running tests on trace event systems:\n");
4164
4165         list_for_each_entry(dir, &tr->systems, list) {
4166
4167                 system = dir->subsystem;
4168
4169                 /* the ftrace system is special, skip it */
4170                 if (strcmp(system->name, "ftrace") == 0)
4171                         continue;
4172
4173                 pr_info("Testing event system %s: ", system->name);
4174
4175                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4176                 if (WARN_ON_ONCE(ret)) {
4177                         pr_warn("error enabling system %s\n",
4178                                 system->name);
4179                         continue;
4180                 }
4181
4182                 event_test_stuff();
4183
4184                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4185                 if (WARN_ON_ONCE(ret)) {
4186                         pr_warn("error disabling system %s\n",
4187                                 system->name);
4188                         continue;
4189                 }
4190
4191                 pr_cont("OK\n");
4192         }
4193
4194         /* Test with all events enabled */
4195
4196         pr_info("Running tests on all trace events:\n");
4197         pr_info("Testing all events: ");
4198
4199         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4200         if (WARN_ON_ONCE(ret)) {
4201                 pr_warn("error enabling all events\n");
4202                 return;
4203         }
4204
4205         event_test_stuff();
4206
4207         /* reset sysname */
4208         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4209         if (WARN_ON_ONCE(ret)) {
4210                 pr_warn("error disabling all events\n");
4211                 return;
4212         }
4213
4214         pr_cont("OK\n");
4215 }
4216
4217 #ifdef CONFIG_FUNCTION_TRACER
4218
4219 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4220
4221 static struct trace_event_file event_trace_file __initdata;
4222
4223 static void __init
4224 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4225                           struct ftrace_ops *op, struct ftrace_regs *regs)
4226 {
4227         struct trace_buffer *buffer;
4228         struct ring_buffer_event *event;
4229         struct ftrace_entry *entry;
4230         unsigned int trace_ctx;
4231         long disabled;
4232         int cpu;
4233
4234         trace_ctx = tracing_gen_ctx();
4235         preempt_disable_notrace();
4236         cpu = raw_smp_processor_id();
4237         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4238
4239         if (disabled != 1)
4240                 goto out;
4241
4242         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4243                                                 TRACE_FN, sizeof(*entry),
4244                                                 trace_ctx);
4245         if (!event)
4246                 goto out;
4247         entry   = ring_buffer_event_data(event);
4248         entry->ip                       = ip;
4249         entry->parent_ip                = parent_ip;
4250
4251         event_trigger_unlock_commit(&event_trace_file, buffer, event,
4252                                     entry, trace_ctx);
4253  out:
4254         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4255         preempt_enable_notrace();
4256 }
4257
4258 static struct ftrace_ops trace_ops __initdata  =
4259 {
4260         .func = function_test_events_call,
4261 };
4262
4263 static __init void event_trace_self_test_with_function(void)
4264 {
4265         int ret;
4266
4267         event_trace_file.tr = top_trace_array();
4268         if (WARN_ON(!event_trace_file.tr))
4269                 return;
4270
4271         ret = register_ftrace_function(&trace_ops);
4272         if (WARN_ON(ret < 0)) {
4273                 pr_info("Failed to enable function tracer for event tests\n");
4274                 return;
4275         }
4276         pr_info("Running tests again, along with the function tracer\n");
4277         event_trace_self_tests();
4278         unregister_ftrace_function(&trace_ops);
4279 }
4280 #else
4281 static __init void event_trace_self_test_with_function(void)
4282 {
4283 }
4284 #endif
4285
4286 static __init int event_trace_self_tests_init(void)
4287 {
4288         if (!tracing_selftest_disabled) {
4289                 event_trace_self_tests();
4290                 event_trace_self_test_with_function();
4291         }
4292
4293         return 0;
4294 }
4295
4296 late_initcall(event_trace_self_tests_init);
4297
4298 #endif