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 #ifdef CONFIG_PERF_EVENTS
1674 static ssize_t
1675 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1676 {
1677         int id = (long)event_file_data(filp);
1678         char buf[32];
1679         int len;
1680
1681         if (unlikely(!id))
1682                 return -ENODEV;
1683
1684         len = sprintf(buf, "%d\n", id);
1685
1686         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687 }
1688 #endif
1689
1690 static ssize_t
1691 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1692                   loff_t *ppos)
1693 {
1694         struct trace_event_file *file;
1695         struct trace_seq *s;
1696         int r = -ENODEV;
1697
1698         if (*ppos)
1699                 return 0;
1700
1701         s = kmalloc(sizeof(*s), GFP_KERNEL);
1702
1703         if (!s)
1704                 return -ENOMEM;
1705
1706         trace_seq_init(s);
1707
1708         mutex_lock(&event_mutex);
1709         file = event_file_data(filp);
1710         if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711                 print_event_filter(file, s);
1712         mutex_unlock(&event_mutex);
1713
1714         if (file)
1715                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1716                                             s->buffer, trace_seq_used(s));
1717
1718         kfree(s);
1719
1720         return r;
1721 }
1722
1723 static ssize_t
1724 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1725                    loff_t *ppos)
1726 {
1727         struct trace_event_file *file;
1728         char *buf;
1729         int err = -ENODEV;
1730
1731         if (cnt >= PAGE_SIZE)
1732                 return -EINVAL;
1733
1734         buf = memdup_user_nul(ubuf, cnt);
1735         if (IS_ERR(buf))
1736                 return PTR_ERR(buf);
1737
1738         mutex_lock(&event_mutex);
1739         file = event_file_data(filp);
1740         if (file)
1741                 err = apply_event_filter(file, buf);
1742         mutex_unlock(&event_mutex);
1743
1744         kfree(buf);
1745         if (err < 0)
1746                 return err;
1747
1748         *ppos += cnt;
1749
1750         return cnt;
1751 }
1752
1753 static LIST_HEAD(event_subsystems);
1754
1755 static int subsystem_open(struct inode *inode, struct file *filp)
1756 {
1757         struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758         struct trace_array *tr = NULL, *iter_tr;
1759         struct event_subsystem *system = NULL;
1760         int ret;
1761
1762         if (tracing_is_disabled())
1763                 return -ENODEV;
1764
1765         /* Make sure the system still exists */
1766         mutex_lock(&event_mutex);
1767         mutex_lock(&trace_types_lock);
1768         list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769                 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770                         if (iter_dir == inode->i_private) {
1771                                 /* Don't open systems with no events */
1772                                 tr = iter_tr;
1773                                 dir = iter_dir;
1774                                 if (dir->nr_events) {
1775                                         __get_system_dir(dir);
1776                                         system = dir->subsystem;
1777                                 }
1778                                 goto exit_loop;
1779                         }
1780                 }
1781         }
1782  exit_loop:
1783         mutex_unlock(&trace_types_lock);
1784         mutex_unlock(&event_mutex);
1785
1786         if (!system)
1787                 return -ENODEV;
1788
1789         /* Still need to increment the ref count of the system */
1790         if (trace_array_get(tr) < 0) {
1791                 put_system(dir);
1792                 return -ENODEV;
1793         }
1794
1795         ret = tracing_open_generic(inode, filp);
1796         if (ret < 0) {
1797                 trace_array_put(tr);
1798                 put_system(dir);
1799         }
1800
1801         return ret;
1802 }
1803
1804 static int system_tr_open(struct inode *inode, struct file *filp)
1805 {
1806         struct trace_subsystem_dir *dir;
1807         struct trace_array *tr = inode->i_private;
1808         int ret;
1809
1810         /* Make a temporary dir that has no system but points to tr */
1811         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1812         if (!dir)
1813                 return -ENOMEM;
1814
1815         ret = tracing_open_generic_tr(inode, filp);
1816         if (ret < 0) {
1817                 kfree(dir);
1818                 return ret;
1819         }
1820         dir->tr = tr;
1821         filp->private_data = dir;
1822
1823         return 0;
1824 }
1825
1826 static int subsystem_release(struct inode *inode, struct file *file)
1827 {
1828         struct trace_subsystem_dir *dir = file->private_data;
1829
1830         trace_array_put(dir->tr);
1831
1832         /*
1833          * If dir->subsystem is NULL, then this is a temporary
1834          * descriptor that was made for a trace_array to enable
1835          * all subsystems.
1836          */
1837         if (dir->subsystem)
1838                 put_system(dir);
1839         else
1840                 kfree(dir);
1841
1842         return 0;
1843 }
1844
1845 static ssize_t
1846 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1847                       loff_t *ppos)
1848 {
1849         struct trace_subsystem_dir *dir = filp->private_data;
1850         struct event_subsystem *system = dir->subsystem;
1851         struct trace_seq *s;
1852         int r;
1853
1854         if (*ppos)
1855                 return 0;
1856
1857         s = kmalloc(sizeof(*s), GFP_KERNEL);
1858         if (!s)
1859                 return -ENOMEM;
1860
1861         trace_seq_init(s);
1862
1863         print_subsystem_event_filter(system, s);
1864         r = simple_read_from_buffer(ubuf, cnt, ppos,
1865                                     s->buffer, trace_seq_used(s));
1866
1867         kfree(s);
1868
1869         return r;
1870 }
1871
1872 static ssize_t
1873 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1874                        loff_t *ppos)
1875 {
1876         struct trace_subsystem_dir *dir = filp->private_data;
1877         char *buf;
1878         int err;
1879
1880         if (cnt >= PAGE_SIZE)
1881                 return -EINVAL;
1882
1883         buf = memdup_user_nul(ubuf, cnt);
1884         if (IS_ERR(buf))
1885                 return PTR_ERR(buf);
1886
1887         err = apply_subsystem_event_filter(dir, buf);
1888         kfree(buf);
1889         if (err < 0)
1890                 return err;
1891
1892         *ppos += cnt;
1893
1894         return cnt;
1895 }
1896
1897 static ssize_t
1898 show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1899 {
1900         struct trace_array *tr = filp->private_data;
1901         struct trace_seq *s;
1902         int r;
1903
1904         if (*ppos)
1905                 return 0;
1906
1907         s = kmalloc(sizeof(*s), GFP_KERNEL);
1908         if (!s)
1909                 return -ENOMEM;
1910
1911         trace_seq_init(s);
1912
1913         ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914         r = simple_read_from_buffer(ubuf, cnt, ppos,
1915                                     s->buffer, trace_seq_used(s));
1916
1917         kfree(s);
1918
1919         return r;
1920 }
1921
1922 static ssize_t
1923 show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1924 {
1925         struct trace_seq *s;
1926         int r;
1927
1928         if (*ppos)
1929                 return 0;
1930
1931         s = kmalloc(sizeof(*s), GFP_KERNEL);
1932         if (!s)
1933                 return -ENOMEM;
1934
1935         trace_seq_init(s);
1936
1937         ring_buffer_print_entry_header(s);
1938         r = simple_read_from_buffer(ubuf, cnt, ppos,
1939                                     s->buffer, trace_seq_used(s));
1940
1941         kfree(s);
1942
1943         return r;
1944 }
1945
1946 static void ignore_task_cpu(void *data)
1947 {
1948         struct trace_array *tr = data;
1949         struct trace_pid_list *pid_list;
1950         struct trace_pid_list *no_pid_list;
1951
1952         /*
1953          * This function is called by on_each_cpu() while the
1954          * event_mutex is held.
1955          */
1956         pid_list = rcu_dereference_protected(tr->filtered_pids,
1957                                              mutex_is_locked(&event_mutex));
1958         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959                                              mutex_is_locked(&event_mutex));
1960
1961         this_cpu_write(tr->array_buffer.data->ignore_pid,
1962                        trace_ignore_this_task(pid_list, no_pid_list, current));
1963 }
1964
1965 static void register_pid_events(struct trace_array *tr)
1966 {
1967         /*
1968          * Register a probe that is called before all other probes
1969          * to set ignore_pid if next or prev do not match.
1970          * Register a probe this is called after all other probes
1971          * to only keep ignore_pid set if next pid matches.
1972          */
1973         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1974                                          tr, INT_MAX);
1975         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1976                                          tr, 0);
1977
1978         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1979                                          tr, INT_MAX);
1980         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1981                                          tr, 0);
1982
1983         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1984                                              tr, INT_MAX);
1985         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1986                                              tr, 0);
1987
1988         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1989                                          tr, INT_MAX);
1990         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1991                                          tr, 0);
1992 }
1993
1994 static ssize_t
1995 event_pid_write(struct file *filp, const char __user *ubuf,
1996                 size_t cnt, loff_t *ppos, int type)
1997 {
1998         struct seq_file *m = filp->private_data;
1999         struct trace_array *tr = m->private;
2000         struct trace_pid_list *filtered_pids = NULL;
2001         struct trace_pid_list *other_pids = NULL;
2002         struct trace_pid_list *pid_list;
2003         struct trace_event_file *file;
2004         ssize_t ret;
2005
2006         if (!cnt)
2007                 return 0;
2008
2009         ret = tracing_update_buffers(tr);
2010         if (ret < 0)
2011                 return ret;
2012
2013         mutex_lock(&event_mutex);
2014
2015         if (type == TRACE_PIDS) {
2016                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017                                                           lockdep_is_held(&event_mutex));
2018                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019                                                           lockdep_is_held(&event_mutex));
2020         } else {
2021                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022                                                           lockdep_is_held(&event_mutex));
2023                 other_pids = rcu_dereference_protected(tr->filtered_pids,
2024                                                           lockdep_is_held(&event_mutex));
2025         }
2026
2027         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2028         if (ret < 0)
2029                 goto out;
2030
2031         if (type == TRACE_PIDS)
2032                 rcu_assign_pointer(tr->filtered_pids, pid_list);
2033         else
2034                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2035
2036         list_for_each_entry(file, &tr->events, list) {
2037                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2038         }
2039
2040         if (filtered_pids) {
2041                 tracepoint_synchronize_unregister();
2042                 trace_pid_list_free(filtered_pids);
2043         } else if (pid_list && !other_pids) {
2044                 register_pid_events(tr);
2045         }
2046
2047         /*
2048          * Ignoring of pids is done at task switch. But we have to
2049          * check for those tasks that are currently running.
2050          * Always do this in case a pid was appended or removed.
2051          */
2052         on_each_cpu(ignore_task_cpu, tr, 1);
2053
2054  out:
2055         mutex_unlock(&event_mutex);
2056
2057         if (ret > 0)
2058                 *ppos += ret;
2059
2060         return ret;
2061 }
2062
2063 static ssize_t
2064 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065                        size_t cnt, loff_t *ppos)
2066 {
2067         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2068 }
2069
2070 static ssize_t
2071 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072                         size_t cnt, loff_t *ppos)
2073 {
2074         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2075 }
2076
2077 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081 static int ftrace_event_release(struct inode *inode, struct file *file);
2082
2083 static const struct seq_operations show_event_seq_ops = {
2084         .start = t_start,
2085         .next = t_next,
2086         .show = t_show,
2087         .stop = t_stop,
2088 };
2089
2090 static const struct seq_operations show_set_event_seq_ops = {
2091         .start = s_start,
2092         .next = s_next,
2093         .show = t_show,
2094         .stop = t_stop,
2095 };
2096
2097 static const struct seq_operations show_set_pid_seq_ops = {
2098         .start = p_start,
2099         .next = p_next,
2100         .show = trace_pid_show,
2101         .stop = p_stop,
2102 };
2103
2104 static const struct seq_operations show_set_no_pid_seq_ops = {
2105         .start = np_start,
2106         .next = np_next,
2107         .show = trace_pid_show,
2108         .stop = p_stop,
2109 };
2110
2111 static const struct file_operations ftrace_avail_fops = {
2112         .open = ftrace_event_avail_open,
2113         .read = seq_read,
2114         .llseek = seq_lseek,
2115         .release = seq_release,
2116 };
2117
2118 static const struct file_operations ftrace_set_event_fops = {
2119         .open = ftrace_event_set_open,
2120         .read = seq_read,
2121         .write = ftrace_event_write,
2122         .llseek = seq_lseek,
2123         .release = ftrace_event_release,
2124 };
2125
2126 static const struct file_operations ftrace_set_event_pid_fops = {
2127         .open = ftrace_event_set_pid_open,
2128         .read = seq_read,
2129         .write = ftrace_event_pid_write,
2130         .llseek = seq_lseek,
2131         .release = ftrace_event_release,
2132 };
2133
2134 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135         .open = ftrace_event_set_npid_open,
2136         .read = seq_read,
2137         .write = ftrace_event_npid_write,
2138         .llseek = seq_lseek,
2139         .release = ftrace_event_release,
2140 };
2141
2142 static const struct file_operations ftrace_enable_fops = {
2143         .open = tracing_open_file_tr,
2144         .read = event_enable_read,
2145         .write = event_enable_write,
2146         .release = tracing_release_file_tr,
2147         .llseek = default_llseek,
2148 };
2149
2150 static const struct file_operations ftrace_event_format_fops = {
2151         .open = trace_format_open,
2152         .read = seq_read,
2153         .llseek = seq_lseek,
2154         .release = seq_release,
2155 };
2156
2157 #ifdef CONFIG_PERF_EVENTS
2158 static const struct file_operations ftrace_event_id_fops = {
2159         .read = event_id_read,
2160         .llseek = default_llseek,
2161 };
2162 #endif
2163
2164 static const struct file_operations ftrace_event_filter_fops = {
2165         .open = tracing_open_file_tr,
2166         .read = event_filter_read,
2167         .write = event_filter_write,
2168         .release = tracing_release_file_tr,
2169         .llseek = default_llseek,
2170 };
2171
2172 static const struct file_operations ftrace_subsystem_filter_fops = {
2173         .open = subsystem_open,
2174         .read = subsystem_filter_read,
2175         .write = subsystem_filter_write,
2176         .llseek = default_llseek,
2177         .release = subsystem_release,
2178 };
2179
2180 static const struct file_operations ftrace_system_enable_fops = {
2181         .open = subsystem_open,
2182         .read = system_enable_read,
2183         .write = system_enable_write,
2184         .llseek = default_llseek,
2185         .release = subsystem_release,
2186 };
2187
2188 static const struct file_operations ftrace_tr_enable_fops = {
2189         .open = system_tr_open,
2190         .read = system_enable_read,
2191         .write = system_enable_write,
2192         .llseek = default_llseek,
2193         .release = subsystem_release,
2194 };
2195
2196 static const struct file_operations ftrace_show_header_page_fops = {
2197         .open = tracing_open_generic_tr,
2198         .read = show_header_page_file,
2199         .llseek = default_llseek,
2200         .release = tracing_release_generic_tr,
2201 };
2202
2203 static const struct file_operations ftrace_show_header_event_fops = {
2204         .open = tracing_open_generic_tr,
2205         .read = show_header_event_file,
2206         .llseek = default_llseek,
2207         .release = tracing_release_generic_tr,
2208 };
2209
2210 static int
2211 ftrace_event_open(struct inode *inode, struct file *file,
2212                   const struct seq_operations *seq_ops)
2213 {
2214         struct seq_file *m;
2215         int ret;
2216
2217         ret = security_locked_down(LOCKDOWN_TRACEFS);
2218         if (ret)
2219                 return ret;
2220
2221         ret = seq_open(file, seq_ops);
2222         if (ret < 0)
2223                 return ret;
2224         m = file->private_data;
2225         /* copy tr over to seq ops */
2226         m->private = inode->i_private;
2227
2228         return ret;
2229 }
2230
2231 static int ftrace_event_release(struct inode *inode, struct file *file)
2232 {
2233         struct trace_array *tr = inode->i_private;
2234
2235         trace_array_put(tr);
2236
2237         return seq_release(inode, file);
2238 }
2239
2240 static int
2241 ftrace_event_avail_open(struct inode *inode, struct file *file)
2242 {
2243         const struct seq_operations *seq_ops = &show_event_seq_ops;
2244
2245         /* Checks for tracefs lockdown */
2246         return ftrace_event_open(inode, file, seq_ops);
2247 }
2248
2249 static int
2250 ftrace_event_set_open(struct inode *inode, struct file *file)
2251 {
2252         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253         struct trace_array *tr = inode->i_private;
2254         int ret;
2255
2256         ret = tracing_check_open_get_tr(tr);
2257         if (ret)
2258                 return ret;
2259
2260         if ((file->f_mode & FMODE_WRITE) &&
2261             (file->f_flags & O_TRUNC))
2262                 ftrace_clear_events(tr);
2263
2264         ret = ftrace_event_open(inode, file, seq_ops);
2265         if (ret < 0)
2266                 trace_array_put(tr);
2267         return ret;
2268 }
2269
2270 static int
2271 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2272 {
2273         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274         struct trace_array *tr = inode->i_private;
2275         int ret;
2276
2277         ret = tracing_check_open_get_tr(tr);
2278         if (ret)
2279                 return ret;
2280
2281         if ((file->f_mode & FMODE_WRITE) &&
2282             (file->f_flags & O_TRUNC))
2283                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2284
2285         ret = ftrace_event_open(inode, file, seq_ops);
2286         if (ret < 0)
2287                 trace_array_put(tr);
2288         return ret;
2289 }
2290
2291 static int
2292 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2293 {
2294         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295         struct trace_array *tr = inode->i_private;
2296         int ret;
2297
2298         ret = tracing_check_open_get_tr(tr);
2299         if (ret)
2300                 return ret;
2301
2302         if ((file->f_mode & FMODE_WRITE) &&
2303             (file->f_flags & O_TRUNC))
2304                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2305
2306         ret = ftrace_event_open(inode, file, seq_ops);
2307         if (ret < 0)
2308                 trace_array_put(tr);
2309         return ret;
2310 }
2311
2312 static struct event_subsystem *
2313 create_new_subsystem(const char *name)
2314 {
2315         struct event_subsystem *system;
2316
2317         /* need to create new entry */
2318         system = kmalloc(sizeof(*system), GFP_KERNEL);
2319         if (!system)
2320                 return NULL;
2321
2322         system->ref_count = 1;
2323
2324         /* Only allocate if dynamic (kprobes and modules) */
2325         system->name = kstrdup_const(name, GFP_KERNEL);
2326         if (!system->name)
2327                 goto out_free;
2328
2329         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330         if (!system->filter)
2331                 goto out_free;
2332
2333         list_add(&system->list, &event_subsystems);
2334
2335         return system;
2336
2337  out_free:
2338         kfree_const(system->name);
2339         kfree(system);
2340         return NULL;
2341 }
2342
2343 static int system_callback(const char *name, umode_t *mode, void **data,
2344                     const struct file_operations **fops)
2345 {
2346         if (strcmp(name, "filter") == 0)
2347                 *fops = &ftrace_subsystem_filter_fops;
2348
2349         else if (strcmp(name, "enable") == 0)
2350                 *fops = &ftrace_system_enable_fops;
2351
2352         else
2353                 return 0;
2354
2355         *mode = TRACE_MODE_WRITE;
2356         return 1;
2357 }
2358
2359 static struct eventfs_inode *
2360 event_subsystem_dir(struct trace_array *tr, const char *name,
2361                     struct trace_event_file *file, struct eventfs_inode *parent)
2362 {
2363         struct event_subsystem *system, *iter;
2364         struct trace_subsystem_dir *dir;
2365         struct eventfs_inode *ei;
2366         int nr_entries;
2367         static struct eventfs_entry system_entries[] = {
2368                 {
2369                         .name           = "filter",
2370                         .callback       = system_callback,
2371                 },
2372                 {
2373                         .name           = "enable",
2374                         .callback       = system_callback,
2375                 }
2376         };
2377
2378         /* First see if we did not already create this dir */
2379         list_for_each_entry(dir, &tr->systems, list) {
2380                 system = dir->subsystem;
2381                 if (strcmp(system->name, name) == 0) {
2382                         dir->nr_events++;
2383                         file->system = dir;
2384                         return dir->ei;
2385                 }
2386         }
2387
2388         /* Now see if the system itself exists. */
2389         system = NULL;
2390         list_for_each_entry(iter, &event_subsystems, list) {
2391                 if (strcmp(iter->name, name) == 0) {
2392                         system = iter;
2393                         break;
2394                 }
2395         }
2396
2397         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2398         if (!dir)
2399                 goto out_fail;
2400
2401         if (!system) {
2402                 system = create_new_subsystem(name);
2403                 if (!system)
2404                         goto out_free;
2405         } else
2406                 __get_system(system);
2407
2408         /* ftrace only has directories no files */
2409         if (strcmp(name, "ftrace") == 0)
2410                 nr_entries = 0;
2411         else
2412                 nr_entries = ARRAY_SIZE(system_entries);
2413
2414         ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2415         if (IS_ERR(ei)) {
2416                 pr_warn("Failed to create system directory %s\n", name);
2417                 __put_system(system);
2418                 goto out_free;
2419         }
2420
2421         dir->ei = ei;
2422         dir->tr = tr;
2423         dir->ref_count = 1;
2424         dir->nr_events = 1;
2425         dir->subsystem = system;
2426         file->system = dir;
2427
2428         list_add(&dir->list, &tr->systems);
2429
2430         return dir->ei;
2431
2432  out_free:
2433         kfree(dir);
2434  out_fail:
2435         /* Only print this message if failed on memory allocation */
2436         if (!dir || !system)
2437                 pr_warn("No memory to create event subsystem %s\n", name);
2438         return NULL;
2439 }
2440
2441 static int
2442 event_define_fields(struct trace_event_call *call)
2443 {
2444         struct list_head *head;
2445         int ret = 0;
2446
2447         /*
2448          * Other events may have the same class. Only update
2449          * the fields if they are not already defined.
2450          */
2451         head = trace_get_fields(call);
2452         if (list_empty(head)) {
2453                 struct trace_event_fields *field = call->class->fields_array;
2454                 unsigned int offset = sizeof(struct trace_entry);
2455
2456                 for (; field->type; field++) {
2457                         if (field->type == TRACE_FUNCTION_TYPE) {
2458                                 field->define_fields(call);
2459                                 break;
2460                         }
2461
2462                         offset = ALIGN(offset, field->align);
2463                         ret = trace_define_field_ext(call, field->type, field->name,
2464                                                  offset, field->size,
2465                                                  field->is_signed, field->filter_type,
2466                                                  field->len);
2467                         if (WARN_ON_ONCE(ret)) {
2468                                 pr_err("error code is %d\n", ret);
2469                                 break;
2470                         }
2471
2472                         offset += field->size;
2473                 }
2474         }
2475
2476         return ret;
2477 }
2478
2479 static int event_callback(const char *name, umode_t *mode, void **data,
2480                           const struct file_operations **fops)
2481 {
2482         struct trace_event_file *file = *data;
2483         struct trace_event_call *call = file->event_call;
2484
2485         if (strcmp(name, "format") == 0) {
2486                 *mode = TRACE_MODE_READ;
2487                 *fops = &ftrace_event_format_fops;
2488                 *data = call;
2489                 return 1;
2490         }
2491
2492         /*
2493          * Only event directories that can be enabled should have
2494          * triggers or filters, with the exception of the "print"
2495          * event that can have a "trigger" file.
2496          */
2497         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498                 if (call->class->reg && strcmp(name, "enable") == 0) {
2499                         *mode = TRACE_MODE_WRITE;
2500                         *fops = &ftrace_enable_fops;
2501                         return 1;
2502                 }
2503
2504                 if (strcmp(name, "filter") == 0) {
2505                         *mode = TRACE_MODE_WRITE;
2506                         *fops = &ftrace_event_filter_fops;
2507                         return 1;
2508                 }
2509         }
2510
2511         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512             strcmp(trace_event_name(call), "print") == 0) {
2513                 if (strcmp(name, "trigger") == 0) {
2514                         *mode = TRACE_MODE_WRITE;
2515                         *fops = &event_trigger_fops;
2516                         return 1;
2517                 }
2518         }
2519
2520 #ifdef CONFIG_PERF_EVENTS
2521         if (call->event.type && call->class->reg &&
2522             strcmp(name, "id") == 0) {
2523                 *mode = TRACE_MODE_READ;
2524                 *data = (void *)(long)call->event.type;
2525                 *fops = &ftrace_event_id_fops;
2526                 return 1;
2527         }
2528 #endif
2529
2530 #ifdef CONFIG_HIST_TRIGGERS
2531         if (strcmp(name, "hist") == 0) {
2532                 *mode = TRACE_MODE_READ;
2533                 *fops = &event_hist_fops;
2534                 return 1;
2535         }
2536 #endif
2537 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538         if (strcmp(name, "hist_debug") == 0) {
2539                 *mode = TRACE_MODE_READ;
2540                 *fops = &event_hist_debug_fops;
2541                 return 1;
2542         }
2543 #endif
2544 #ifdef CONFIG_TRACE_EVENT_INJECT
2545         if (call->event.type && call->class->reg &&
2546             strcmp(name, "inject") == 0) {
2547                 *mode = 0200;
2548                 *fops = &event_inject_fops;
2549                 return 1;
2550         }
2551 #endif
2552         return 0;
2553 }
2554
2555 static int
2556 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2557 {
2558         struct trace_event_call *call = file->event_call;
2559         struct trace_array *tr = file->tr;
2560         struct eventfs_inode *e_events;
2561         struct eventfs_inode *ei;
2562         const char *name;
2563         int nr_entries;
2564         int ret;
2565         static struct eventfs_entry event_entries[] = {
2566                 {
2567                         .name           = "enable",
2568                         .callback       = event_callback,
2569                 },
2570                 {
2571                         .name           = "filter",
2572                         .callback       = event_callback,
2573                 },
2574                 {
2575                         .name           = "trigger",
2576                         .callback       = event_callback,
2577                 },
2578                 {
2579                         .name           = "format",
2580                         .callback       = event_callback,
2581                 },
2582 #ifdef CONFIG_PERF_EVENTS
2583                 {
2584                         .name           = "id",
2585                         .callback       = event_callback,
2586                 },
2587 #endif
2588 #ifdef CONFIG_HIST_TRIGGERS
2589                 {
2590                         .name           = "hist",
2591                         .callback       = event_callback,
2592                 },
2593 #endif
2594 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2595                 {
2596                         .name           = "hist_debug",
2597                         .callback       = event_callback,
2598                 },
2599 #endif
2600 #ifdef CONFIG_TRACE_EVENT_INJECT
2601                 {
2602                         .name           = "inject",
2603                         .callback       = event_callback,
2604                 },
2605 #endif
2606         };
2607
2608         /*
2609          * If the trace point header did not define TRACE_SYSTEM
2610          * then the system would be called "TRACE_SYSTEM". This should
2611          * never happen.
2612          */
2613         if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2614                 return -ENODEV;
2615
2616         e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2617         if (!e_events)
2618                 return -ENOMEM;
2619
2620         nr_entries = ARRAY_SIZE(event_entries);
2621
2622         name = trace_event_name(call);
2623         ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2624         if (IS_ERR(ei)) {
2625                 pr_warn("Could not create tracefs '%s' directory\n", name);
2626                 return -1;
2627         }
2628
2629         file->ei = ei;
2630
2631         ret = event_define_fields(call);
2632         if (ret < 0) {
2633                 pr_warn("Could not initialize trace point events/%s\n", name);
2634                 return ret;
2635         }
2636
2637         return 0;
2638 }
2639
2640 static void remove_event_from_tracers(struct trace_event_call *call)
2641 {
2642         struct trace_event_file *file;
2643         struct trace_array *tr;
2644
2645         do_for_each_event_file_safe(tr, file) {
2646                 if (file->event_call != call)
2647                         continue;
2648
2649                 remove_event_file_dir(file);
2650                 /*
2651                  * The do_for_each_event_file_safe() is
2652                  * a double loop. After finding the call for this
2653                  * trace_array, we use break to jump to the next
2654                  * trace_array.
2655                  */
2656                 break;
2657         } while_for_each_event_file();
2658 }
2659
2660 static void event_remove(struct trace_event_call *call)
2661 {
2662         struct trace_array *tr;
2663         struct trace_event_file *file;
2664
2665         do_for_each_event_file(tr, file) {
2666                 if (file->event_call != call)
2667                         continue;
2668
2669                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2670                         tr->clear_trace = true;
2671
2672                 ftrace_event_enable_disable(file, 0);
2673                 /*
2674                  * The do_for_each_event_file() is
2675                  * a double loop. After finding the call for this
2676                  * trace_array, we use break to jump to the next
2677                  * trace_array.
2678                  */
2679                 break;
2680         } while_for_each_event_file();
2681
2682         if (call->event.funcs)
2683                 __unregister_trace_event(&call->event);
2684         remove_event_from_tracers(call);
2685         list_del(&call->list);
2686 }
2687
2688 static int event_init(struct trace_event_call *call)
2689 {
2690         int ret = 0;
2691         const char *name;
2692
2693         name = trace_event_name(call);
2694         if (WARN_ON(!name))
2695                 return -EINVAL;
2696
2697         if (call->class->raw_init) {
2698                 ret = call->class->raw_init(call);
2699                 if (ret < 0 && ret != -ENOSYS)
2700                         pr_warn("Could not initialize trace events/%s\n", name);
2701         }
2702
2703         return ret;
2704 }
2705
2706 static int
2707 __register_event(struct trace_event_call *call, struct module *mod)
2708 {
2709         int ret;
2710
2711         ret = event_init(call);
2712         if (ret < 0)
2713                 return ret;
2714
2715         list_add(&call->list, &ftrace_events);
2716         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2717                 atomic_set(&call->refcnt, 0);
2718         else
2719                 call->module = mod;
2720
2721         return 0;
2722 }
2723
2724 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2725 {
2726         int rlen;
2727         int elen;
2728
2729         /* Find the length of the eval value as a string */
2730         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2731         /* Make sure there's enough room to replace the string with the value */
2732         if (len < elen)
2733                 return NULL;
2734
2735         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2736
2737         /* Get the rest of the string of ptr */
2738         rlen = strlen(ptr + len);
2739         memmove(ptr + elen, ptr + len, rlen);
2740         /* Make sure we end the new string */
2741         ptr[elen + rlen] = 0;
2742
2743         return ptr + elen;
2744 }
2745
2746 static void update_event_printk(struct trace_event_call *call,
2747                                 struct trace_eval_map *map)
2748 {
2749         char *ptr;
2750         int quote = 0;
2751         int len = strlen(map->eval_string);
2752
2753         for (ptr = call->print_fmt; *ptr; ptr++) {
2754                 if (*ptr == '\\') {
2755                         ptr++;
2756                         /* paranoid */
2757                         if (!*ptr)
2758                                 break;
2759                         continue;
2760                 }
2761                 if (*ptr == '"') {
2762                         quote ^= 1;
2763                         continue;
2764                 }
2765                 if (quote)
2766                         continue;
2767                 if (isdigit(*ptr)) {
2768                         /* skip numbers */
2769                         do {
2770                                 ptr++;
2771                                 /* Check for alpha chars like ULL */
2772                         } while (isalnum(*ptr));
2773                         if (!*ptr)
2774                                 break;
2775                         /*
2776                          * A number must have some kind of delimiter after
2777                          * it, and we can ignore that too.
2778                          */
2779                         continue;
2780                 }
2781                 if (isalpha(*ptr) || *ptr == '_') {
2782                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2783                             !isalnum(ptr[len]) && ptr[len] != '_') {
2784                                 ptr = eval_replace(ptr, map, len);
2785                                 /* enum/sizeof string smaller than value */
2786                                 if (WARN_ON_ONCE(!ptr))
2787                                         return;
2788                                 /*
2789                                  * No need to decrement here, as eval_replace()
2790                                  * returns the pointer to the character passed
2791                                  * the eval, and two evals can not be placed
2792                                  * back to back without something in between.
2793                                  * We can skip that something in between.
2794                                  */
2795                                 continue;
2796                         }
2797                 skip_more:
2798                         do {
2799                                 ptr++;
2800                         } while (isalnum(*ptr) || *ptr == '_');
2801                         if (!*ptr)
2802                                 break;
2803                         /*
2804                          * If what comes after this variable is a '.' or
2805                          * '->' then we can continue to ignore that string.
2806                          */
2807                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2808                                 ptr += *ptr == '.' ? 1 : 2;
2809                                 if (!*ptr)
2810                                         break;
2811                                 goto skip_more;
2812                         }
2813                         /*
2814                          * Once again, we can skip the delimiter that came
2815                          * after the string.
2816                          */
2817                         continue;
2818                 }
2819         }
2820 }
2821
2822 static void add_str_to_module(struct module *module, char *str)
2823 {
2824         struct module_string *modstr;
2825
2826         modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2827
2828         /*
2829          * If we failed to allocate memory here, then we'll just
2830          * let the str memory leak when the module is removed.
2831          * If this fails to allocate, there's worse problems than
2832          * a leaked string on module removal.
2833          */
2834         if (WARN_ON_ONCE(!modstr))
2835                 return;
2836
2837         modstr->module = module;
2838         modstr->str = str;
2839
2840         list_add(&modstr->next, &module_strings);
2841 }
2842
2843 static void update_event_fields(struct trace_event_call *call,
2844                                 struct trace_eval_map *map)
2845 {
2846         struct ftrace_event_field *field;
2847         struct list_head *head;
2848         char *ptr;
2849         char *str;
2850         int len = strlen(map->eval_string);
2851
2852         /* Dynamic events should never have field maps */
2853         if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2854                 return;
2855
2856         head = trace_get_fields(call);
2857         list_for_each_entry(field, head, link) {
2858                 ptr = strchr(field->type, '[');
2859                 if (!ptr)
2860                         continue;
2861                 ptr++;
2862
2863                 if (!isalpha(*ptr) && *ptr != '_')
2864                         continue;
2865
2866                 if (strncmp(map->eval_string, ptr, len) != 0)
2867                         continue;
2868
2869                 str = kstrdup(field->type, GFP_KERNEL);
2870                 if (WARN_ON_ONCE(!str))
2871                         return;
2872                 ptr = str + (ptr - field->type);
2873                 ptr = eval_replace(ptr, map, len);
2874                 /* enum/sizeof string smaller than value */
2875                 if (WARN_ON_ONCE(!ptr)) {
2876                         kfree(str);
2877                         continue;
2878                 }
2879
2880                 /*
2881                  * If the event is part of a module, then we need to free the string
2882                  * when the module is removed. Otherwise, it will stay allocated
2883                  * until a reboot.
2884                  */
2885                 if (call->module)
2886                         add_str_to_module(call->module, str);
2887
2888                 field->type = str;
2889         }
2890 }
2891
2892 void trace_event_eval_update(struct trace_eval_map **map, int len)
2893 {
2894         struct trace_event_call *call, *p;
2895         const char *last_system = NULL;
2896         bool first = false;
2897         int last_i;
2898         int i;
2899
2900         down_write(&trace_event_sem);
2901         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2902                 /* events are usually grouped together with systems */
2903                 if (!last_system || call->class->system != last_system) {
2904                         first = true;
2905                         last_i = 0;
2906                         last_system = call->class->system;
2907                 }
2908
2909                 /*
2910                  * Since calls are grouped by systems, the likelihood that the
2911                  * next call in the iteration belongs to the same system as the
2912                  * previous call is high. As an optimization, we skip searching
2913                  * for a map[] that matches the call's system if the last call
2914                  * was from the same system. That's what last_i is for. If the
2915                  * call has the same system as the previous call, then last_i
2916                  * will be the index of the first map[] that has a matching
2917                  * system.
2918                  */
2919                 for (i = last_i; i < len; i++) {
2920                         if (call->class->system == map[i]->system) {
2921                                 /* Save the first system if need be */
2922                                 if (first) {
2923                                         last_i = i;
2924                                         first = false;
2925                                 }
2926                                 update_event_printk(call, map[i]);
2927                                 update_event_fields(call, map[i]);
2928                         }
2929                 }
2930                 cond_resched();
2931         }
2932         up_write(&trace_event_sem);
2933 }
2934
2935 static bool event_in_systems(struct trace_event_call *call,
2936                              const char *systems)
2937 {
2938         const char *system;
2939         const char *p;
2940
2941         if (!systems)
2942                 return true;
2943
2944         system = call->class->system;
2945         p = strstr(systems, system);
2946         if (!p)
2947                 return false;
2948
2949         if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2950                 return false;
2951
2952         p += strlen(system);
2953         return !*p || isspace(*p) || *p == ',';
2954 }
2955
2956 static struct trace_event_file *
2957 trace_create_new_event(struct trace_event_call *call,
2958                        struct trace_array *tr)
2959 {
2960         struct trace_pid_list *no_pid_list;
2961         struct trace_pid_list *pid_list;
2962         struct trace_event_file *file;
2963         unsigned int first;
2964
2965         if (!event_in_systems(call, tr->system_names))
2966                 return NULL;
2967
2968         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2969         if (!file)
2970                 return ERR_PTR(-ENOMEM);
2971
2972         pid_list = rcu_dereference_protected(tr->filtered_pids,
2973                                              lockdep_is_held(&event_mutex));
2974         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2975                                              lockdep_is_held(&event_mutex));
2976
2977         if (!trace_pid_list_first(pid_list, &first) ||
2978             !trace_pid_list_first(no_pid_list, &first))
2979                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2980
2981         file->event_call = call;
2982         file->tr = tr;
2983         atomic_set(&file->sm_ref, 0);
2984         atomic_set(&file->tm_ref, 0);
2985         INIT_LIST_HEAD(&file->triggers);
2986         list_add(&file->list, &tr->events);
2987         event_file_get(file);
2988
2989         return file;
2990 }
2991
2992 #define MAX_BOOT_TRIGGERS 32
2993
2994 static struct boot_triggers {
2995         const char              *event;
2996         char                    *trigger;
2997 } bootup_triggers[MAX_BOOT_TRIGGERS];
2998
2999 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3000 static int nr_boot_triggers;
3001
3002 static __init int setup_trace_triggers(char *str)
3003 {
3004         char *trigger;
3005         char *buf;
3006         int i;
3007
3008         strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3009         trace_set_ring_buffer_expanded(NULL);
3010         disable_tracing_selftest("running event triggers");
3011
3012         buf = bootup_trigger_buf;
3013         for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3014                 trigger = strsep(&buf, ",");
3015                 if (!trigger)
3016                         break;
3017                 bootup_triggers[i].event = strsep(&trigger, ".");
3018                 bootup_triggers[i].trigger = trigger;
3019                 if (!bootup_triggers[i].trigger)
3020                         break;
3021         }
3022
3023         nr_boot_triggers = i;
3024         return 1;
3025 }
3026 __setup("trace_trigger=", setup_trace_triggers);
3027
3028 /* Add an event to a trace directory */
3029 static int
3030 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3031 {
3032         struct trace_event_file *file;
3033
3034         file = trace_create_new_event(call, tr);
3035         /*
3036          * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3037          * allocation, or NULL if the event is not part of the tr->system_names.
3038          * When the event is not part of the tr->system_names, return zero, not
3039          * an error.
3040          */
3041         if (!file)
3042                 return 0;
3043
3044         if (IS_ERR(file))
3045                 return PTR_ERR(file);
3046
3047         if (eventdir_initialized)
3048                 return event_create_dir(tr->event_dir, file);
3049         else
3050                 return event_define_fields(call);
3051 }
3052
3053 static void trace_early_triggers(struct trace_event_file *file, const char *name)
3054 {
3055         int ret;
3056         int i;
3057
3058         for (i = 0; i < nr_boot_triggers; i++) {
3059                 if (strcmp(name, bootup_triggers[i].event))
3060                         continue;
3061                 mutex_lock(&event_mutex);
3062                 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3063                 mutex_unlock(&event_mutex);
3064                 if (ret)
3065                         pr_err("Failed to register trigger '%s' on event %s\n",
3066                                bootup_triggers[i].trigger,
3067                                bootup_triggers[i].event);
3068         }
3069 }
3070
3071 /*
3072  * Just create a descriptor for early init. A descriptor is required
3073  * for enabling events at boot. We want to enable events before
3074  * the filesystem is initialized.
3075  */
3076 static int
3077 __trace_early_add_new_event(struct trace_event_call *call,
3078                             struct trace_array *tr)
3079 {
3080         struct trace_event_file *file;
3081         int ret;
3082
3083         file = trace_create_new_event(call, tr);
3084         /*
3085          * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3086          * allocation, or NULL if the event is not part of the tr->system_names.
3087          * When the event is not part of the tr->system_names, return zero, not
3088          * an error.
3089          */
3090         if (!file)
3091                 return 0;
3092
3093         if (IS_ERR(file))
3094                 return PTR_ERR(file);
3095
3096         ret = event_define_fields(call);
3097         if (ret)
3098                 return ret;
3099
3100         trace_early_triggers(file, trace_event_name(call));
3101
3102         return 0;
3103 }
3104
3105 struct ftrace_module_file_ops;
3106 static void __add_event_to_tracers(struct trace_event_call *call);
3107
3108 /* Add an additional event_call dynamically */
3109 int trace_add_event_call(struct trace_event_call *call)
3110 {
3111         int ret;
3112         lockdep_assert_held(&event_mutex);
3113
3114         mutex_lock(&trace_types_lock);
3115
3116         ret = __register_event(call, NULL);
3117         if (ret >= 0)
3118                 __add_event_to_tracers(call);
3119
3120         mutex_unlock(&trace_types_lock);
3121         return ret;
3122 }
3123 EXPORT_SYMBOL_GPL(trace_add_event_call);
3124
3125 /*
3126  * Must be called under locking of trace_types_lock, event_mutex and
3127  * trace_event_sem.
3128  */
3129 static void __trace_remove_event_call(struct trace_event_call *call)
3130 {
3131         event_remove(call);
3132         trace_destroy_fields(call);
3133         free_event_filter(call->filter);
3134         call->filter = NULL;
3135 }
3136
3137 static int probe_remove_event_call(struct trace_event_call *call)
3138 {
3139         struct trace_array *tr;
3140         struct trace_event_file *file;
3141
3142 #ifdef CONFIG_PERF_EVENTS
3143         if (call->perf_refcount)
3144                 return -EBUSY;
3145 #endif
3146         do_for_each_event_file(tr, file) {
3147                 if (file->event_call != call)
3148                         continue;
3149                 /*
3150                  * We can't rely on ftrace_event_enable_disable(enable => 0)
3151                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3152                  * TRACE_REG_UNREGISTER.
3153                  */
3154                 if (file->flags & EVENT_FILE_FL_ENABLED)
3155                         goto busy;
3156
3157                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3158                         tr->clear_trace = true;
3159                 /*
3160                  * The do_for_each_event_file_safe() is
3161                  * a double loop. After finding the call for this
3162                  * trace_array, we use break to jump to the next
3163                  * trace_array.
3164                  */
3165                 break;
3166         } while_for_each_event_file();
3167
3168         __trace_remove_event_call(call);
3169
3170         return 0;
3171  busy:
3172         /* No need to clear the trace now */
3173         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3174                 tr->clear_trace = false;
3175         }
3176         return -EBUSY;
3177 }
3178
3179 /* Remove an event_call */
3180 int trace_remove_event_call(struct trace_event_call *call)
3181 {
3182         int ret;
3183
3184         lockdep_assert_held(&event_mutex);
3185
3186         mutex_lock(&trace_types_lock);
3187         down_write(&trace_event_sem);
3188         ret = probe_remove_event_call(call);
3189         up_write(&trace_event_sem);
3190         mutex_unlock(&trace_types_lock);
3191
3192         return ret;
3193 }
3194 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3195
3196 #define for_each_event(event, start, end)                       \
3197         for (event = start;                                     \
3198              (unsigned long)event < (unsigned long)end;         \
3199              event++)
3200
3201 #ifdef CONFIG_MODULES
3202
3203 static void trace_module_add_events(struct module *mod)
3204 {
3205         struct trace_event_call **call, **start, **end;
3206
3207         if (!mod->num_trace_events)
3208                 return;
3209
3210         /* Don't add infrastructure for mods without tracepoints */
3211         if (trace_module_has_bad_taint(mod)) {
3212                 pr_err("%s: module has bad taint, not creating trace events\n",
3213                        mod->name);
3214                 return;
3215         }
3216
3217         start = mod->trace_events;
3218         end = mod->trace_events + mod->num_trace_events;
3219
3220         for_each_event(call, start, end) {
3221                 __register_event(*call, mod);
3222                 __add_event_to_tracers(*call);
3223         }
3224 }
3225
3226 static void trace_module_remove_events(struct module *mod)
3227 {
3228         struct trace_event_call *call, *p;
3229         struct module_string *modstr, *m;
3230
3231         down_write(&trace_event_sem);
3232         list_for_each_entry_safe(call, p, &ftrace_events, list) {
3233                 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3234                         continue;
3235                 if (call->module == mod)
3236                         __trace_remove_event_call(call);
3237         }
3238         /* Check for any strings allocade for this module */
3239         list_for_each_entry_safe(modstr, m, &module_strings, next) {
3240                 if (modstr->module != mod)
3241                         continue;
3242                 list_del(&modstr->next);
3243                 kfree(modstr->str);
3244                 kfree(modstr);
3245         }
3246         up_write(&trace_event_sem);
3247
3248         /*
3249          * It is safest to reset the ring buffer if the module being unloaded
3250          * registered any events that were used. The only worry is if
3251          * a new module gets loaded, and takes on the same id as the events
3252          * of this module. When printing out the buffer, traced events left
3253          * over from this module may be passed to the new module events and
3254          * unexpected results may occur.
3255          */
3256         tracing_reset_all_online_cpus_unlocked();
3257 }
3258
3259 static int trace_module_notify(struct notifier_block *self,
3260                                unsigned long val, void *data)
3261 {
3262         struct module *mod = data;
3263
3264         mutex_lock(&event_mutex);
3265         mutex_lock(&trace_types_lock);
3266         switch (val) {
3267         case MODULE_STATE_COMING:
3268                 trace_module_add_events(mod);
3269                 break;
3270         case MODULE_STATE_GOING:
3271                 trace_module_remove_events(mod);
3272                 break;
3273         }
3274         mutex_unlock(&trace_types_lock);
3275         mutex_unlock(&event_mutex);
3276
3277         return NOTIFY_OK;
3278 }
3279
3280 static struct notifier_block trace_module_nb = {
3281         .notifier_call = trace_module_notify,
3282         .priority = 1, /* higher than trace.c module notify */
3283 };
3284 #endif /* CONFIG_MODULES */
3285
3286 /* Create a new event directory structure for a trace directory. */
3287 static void
3288 __trace_add_event_dirs(struct trace_array *tr)
3289 {
3290         struct trace_event_call *call;
3291         int ret;
3292
3293         list_for_each_entry(call, &ftrace_events, list) {
3294                 ret = __trace_add_new_event(call, tr);
3295                 if (ret < 0)
3296                         pr_warn("Could not create directory for event %s\n",
3297                                 trace_event_name(call));
3298         }
3299 }
3300
3301 /* Returns any file that matches the system and event */
3302 struct trace_event_file *
3303 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3304 {
3305         struct trace_event_file *file;
3306         struct trace_event_call *call;
3307         const char *name;
3308
3309         list_for_each_entry(file, &tr->events, list) {
3310
3311                 call = file->event_call;
3312                 name = trace_event_name(call);
3313
3314                 if (!name || !call->class)
3315                         continue;
3316
3317                 if (strcmp(event, name) == 0 &&
3318                     strcmp(system, call->class->system) == 0)
3319                         return file;
3320         }
3321         return NULL;
3322 }
3323
3324 /* Returns valid trace event files that match system and event */
3325 struct trace_event_file *
3326 find_event_file(struct trace_array *tr, const char *system, const char *event)
3327 {
3328         struct trace_event_file *file;
3329
3330         file = __find_event_file(tr, system, event);
3331         if (!file || !file->event_call->class->reg ||
3332             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3333                 return NULL;
3334
3335         return file;
3336 }
3337
3338 /**
3339  * trace_get_event_file - Find and return a trace event file
3340  * @instance: The name of the trace instance containing the event
3341  * @system: The name of the system containing the event
3342  * @event: The name of the event
3343  *
3344  * Return a trace event file given the trace instance name, trace
3345  * system, and trace event name.  If the instance name is NULL, it
3346  * refers to the top-level trace array.
3347  *
3348  * This function will look it up and return it if found, after calling
3349  * trace_array_get() to prevent the instance from going away, and
3350  * increment the event's module refcount to prevent it from being
3351  * removed.
3352  *
3353  * To release the file, call trace_put_event_file(), which will call
3354  * trace_array_put() and decrement the event's module refcount.
3355  *
3356  * Return: The trace event on success, ERR_PTR otherwise.
3357  */
3358 struct trace_event_file *trace_get_event_file(const char *instance,
3359                                               const char *system,
3360                                               const char *event)
3361 {
3362         struct trace_array *tr = top_trace_array();
3363         struct trace_event_file *file = NULL;
3364         int ret = -EINVAL;
3365
3366         if (instance) {
3367                 tr = trace_array_find_get(instance);
3368                 if (!tr)
3369                         return ERR_PTR(-ENOENT);
3370         } else {
3371                 ret = trace_array_get(tr);
3372                 if (ret)
3373                         return ERR_PTR(ret);
3374         }
3375
3376         mutex_lock(&event_mutex);
3377
3378         file = find_event_file(tr, system, event);
3379         if (!file) {
3380                 trace_array_put(tr);
3381                 ret = -EINVAL;
3382                 goto out;
3383         }
3384
3385         /* Don't let event modules unload while in use */
3386         ret = trace_event_try_get_ref(file->event_call);
3387         if (!ret) {
3388                 trace_array_put(tr);
3389                 ret = -EBUSY;
3390                 goto out;
3391         }
3392
3393         ret = 0;
3394  out:
3395         mutex_unlock(&event_mutex);
3396
3397         if (ret)
3398                 file = ERR_PTR(ret);
3399
3400         return file;
3401 }
3402 EXPORT_SYMBOL_GPL(trace_get_event_file);
3403
3404 /**
3405  * trace_put_event_file - Release a file from trace_get_event_file()
3406  * @file: The trace event file
3407  *
3408  * If a file was retrieved using trace_get_event_file(), this should
3409  * be called when it's no longer needed.  It will cancel the previous
3410  * trace_array_get() called by that function, and decrement the
3411  * event's module refcount.
3412  */
3413 void trace_put_event_file(struct trace_event_file *file)
3414 {
3415         mutex_lock(&event_mutex);
3416         trace_event_put_ref(file->event_call);
3417         mutex_unlock(&event_mutex);
3418
3419         trace_array_put(file->tr);
3420 }
3421 EXPORT_SYMBOL_GPL(trace_put_event_file);
3422
3423 #ifdef CONFIG_DYNAMIC_FTRACE
3424
3425 /* Avoid typos */
3426 #define ENABLE_EVENT_STR        "enable_event"
3427 #define DISABLE_EVENT_STR       "disable_event"
3428
3429 struct event_probe_data {
3430         struct trace_event_file *file;
3431         unsigned long                   count;
3432         int                             ref;
3433         bool                            enable;
3434 };
3435
3436 static void update_event_probe(struct event_probe_data *data)
3437 {
3438         if (data->enable)
3439                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3440         else
3441                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3442 }
3443
3444 static void
3445 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3446                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3447                    void *data)
3448 {
3449         struct ftrace_func_mapper *mapper = data;
3450         struct event_probe_data *edata;
3451         void **pdata;
3452
3453         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3454         if (!pdata || !*pdata)
3455                 return;
3456
3457         edata = *pdata;
3458         update_event_probe(edata);
3459 }
3460
3461 static void
3462 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3463                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3464                          void *data)
3465 {
3466         struct ftrace_func_mapper *mapper = data;
3467         struct event_probe_data *edata;
3468         void **pdata;
3469
3470         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3471         if (!pdata || !*pdata)
3472                 return;
3473
3474         edata = *pdata;
3475
3476         if (!edata->count)
3477                 return;
3478
3479         /* Skip if the event is in a state we want to switch to */
3480         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3481                 return;
3482
3483         if (edata->count != -1)
3484                 (edata->count)--;
3485
3486         update_event_probe(edata);
3487 }
3488
3489 static int
3490 event_enable_print(struct seq_file *m, unsigned long ip,
3491                    struct ftrace_probe_ops *ops, void *data)
3492 {
3493         struct ftrace_func_mapper *mapper = data;
3494         struct event_probe_data *edata;
3495         void **pdata;
3496
3497         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3498
3499         if (WARN_ON_ONCE(!pdata || !*pdata))
3500                 return 0;
3501
3502         edata = *pdata;
3503
3504         seq_printf(m, "%ps:", (void *)ip);
3505
3506         seq_printf(m, "%s:%s:%s",
3507                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3508                    edata->file->event_call->class->system,
3509                    trace_event_name(edata->file->event_call));
3510
3511         if (edata->count == -1)
3512                 seq_puts(m, ":unlimited\n");
3513         else
3514                 seq_printf(m, ":count=%ld\n", edata->count);
3515
3516         return 0;
3517 }
3518
3519 static int
3520 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3521                   unsigned long ip, void *init_data, void **data)
3522 {
3523         struct ftrace_func_mapper *mapper = *data;
3524         struct event_probe_data *edata = init_data;
3525         int ret;
3526
3527         if (!mapper) {
3528                 mapper = allocate_ftrace_func_mapper();
3529                 if (!mapper)
3530                         return -ENODEV;
3531                 *data = mapper;
3532         }
3533
3534         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3535         if (ret < 0)
3536                 return ret;
3537
3538         edata->ref++;
3539
3540         return 0;
3541 }
3542
3543 static int free_probe_data(void *data)
3544 {
3545         struct event_probe_data *edata = data;
3546
3547         edata->ref--;
3548         if (!edata->ref) {
3549                 /* Remove the SOFT_MODE flag */
3550                 __ftrace_event_enable_disable(edata->file, 0, 1);
3551                 trace_event_put_ref(edata->file->event_call);
3552                 kfree(edata);
3553         }
3554         return 0;
3555 }
3556
3557 static void
3558 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3559                   unsigned long ip, void *data)
3560 {
3561         struct ftrace_func_mapper *mapper = data;
3562         struct event_probe_data *edata;
3563
3564         if (!ip) {
3565                 if (!mapper)
3566                         return;
3567                 free_ftrace_func_mapper(mapper, free_probe_data);
3568                 return;
3569         }
3570
3571         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3572
3573         if (WARN_ON_ONCE(!edata))
3574                 return;
3575
3576         if (WARN_ON_ONCE(edata->ref <= 0))
3577                 return;
3578
3579         free_probe_data(edata);
3580 }
3581
3582 static struct ftrace_probe_ops event_enable_probe_ops = {
3583         .func                   = event_enable_probe,
3584         .print                  = event_enable_print,
3585         .init                   = event_enable_init,
3586         .free                   = event_enable_free,
3587 };
3588
3589 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3590         .func                   = event_enable_count_probe,
3591         .print                  = event_enable_print,
3592         .init                   = event_enable_init,
3593         .free                   = event_enable_free,
3594 };
3595
3596 static struct ftrace_probe_ops event_disable_probe_ops = {
3597         .func                   = event_enable_probe,
3598         .print                  = event_enable_print,
3599         .init                   = event_enable_init,
3600         .free                   = event_enable_free,
3601 };
3602
3603 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3604         .func                   = event_enable_count_probe,
3605         .print                  = event_enable_print,
3606         .init                   = event_enable_init,
3607         .free                   = event_enable_free,
3608 };
3609
3610 static int
3611 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3612                   char *glob, char *cmd, char *param, int enabled)
3613 {
3614         struct trace_event_file *file;
3615         struct ftrace_probe_ops *ops;
3616         struct event_probe_data *data;
3617         const char *system;
3618         const char *event;
3619         char *number;
3620         bool enable;
3621         int ret;
3622
3623         if (!tr)
3624                 return -ENODEV;
3625
3626         /* hash funcs only work with set_ftrace_filter */
3627         if (!enabled || !param)
3628                 return -EINVAL;
3629
3630         system = strsep(&param, ":");
3631         if (!param)
3632                 return -EINVAL;
3633
3634         event = strsep(&param, ":");
3635
3636         mutex_lock(&event_mutex);
3637
3638         ret = -EINVAL;
3639         file = find_event_file(tr, system, event);
3640         if (!file)
3641                 goto out;
3642
3643         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3644
3645         if (enable)
3646                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3647         else
3648                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3649
3650         if (glob[0] == '!') {
3651                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3652                 goto out;
3653         }
3654
3655         ret = -ENOMEM;
3656
3657         data = kzalloc(sizeof(*data), GFP_KERNEL);
3658         if (!data)
3659                 goto out;
3660
3661         data->enable = enable;
3662         data->count = -1;
3663         data->file = file;
3664
3665         if (!param)
3666                 goto out_reg;
3667
3668         number = strsep(&param, ":");
3669
3670         ret = -EINVAL;
3671         if (!strlen(number))
3672                 goto out_free;
3673
3674         /*
3675          * We use the callback data field (which is a pointer)
3676          * as our counter.
3677          */
3678         ret = kstrtoul(number, 0, &data->count);
3679         if (ret)
3680                 goto out_free;
3681
3682  out_reg:
3683         /* Don't let event modules unload while probe registered */
3684         ret = trace_event_try_get_ref(file->event_call);
3685         if (!ret) {
3686                 ret = -EBUSY;
3687                 goto out_free;
3688         }
3689
3690         ret = __ftrace_event_enable_disable(file, 1, 1);
3691         if (ret < 0)
3692                 goto out_put;
3693
3694         ret = register_ftrace_function_probe(glob, tr, ops, data);
3695         /*
3696          * The above returns on success the # of functions enabled,
3697          * but if it didn't find any functions it returns zero.
3698          * Consider no functions a failure too.
3699          */
3700         if (!ret) {
3701                 ret = -ENOENT;
3702                 goto out_disable;
3703         } else if (ret < 0)
3704                 goto out_disable;
3705         /* Just return zero, not the number of enabled functions */
3706         ret = 0;
3707  out:
3708         mutex_unlock(&event_mutex);
3709         return ret;
3710
3711  out_disable:
3712         __ftrace_event_enable_disable(file, 0, 1);
3713  out_put:
3714         trace_event_put_ref(file->event_call);
3715  out_free:
3716         kfree(data);
3717         goto out;
3718 }
3719
3720 static struct ftrace_func_command event_enable_cmd = {
3721         .name                   = ENABLE_EVENT_STR,
3722         .func                   = event_enable_func,
3723 };
3724
3725 static struct ftrace_func_command event_disable_cmd = {
3726         .name                   = DISABLE_EVENT_STR,
3727         .func                   = event_enable_func,
3728 };
3729
3730 static __init int register_event_cmds(void)
3731 {
3732         int ret;
3733
3734         ret = register_ftrace_command(&event_enable_cmd);
3735         if (WARN_ON(ret < 0))
3736                 return ret;
3737         ret = register_ftrace_command(&event_disable_cmd);
3738         if (WARN_ON(ret < 0))
3739                 unregister_ftrace_command(&event_enable_cmd);
3740         return ret;
3741 }
3742 #else
3743 static inline int register_event_cmds(void) { return 0; }
3744 #endif /* CONFIG_DYNAMIC_FTRACE */
3745
3746 /*
3747  * The top level array and trace arrays created by boot-time tracing
3748  * have already had its trace_event_file descriptors created in order
3749  * to allow for early events to be recorded.
3750  * This function is called after the tracefs has been initialized,
3751  * and we now have to create the files associated to the events.
3752  */
3753 static void __trace_early_add_event_dirs(struct trace_array *tr)
3754 {
3755         struct trace_event_file *file;
3756         int ret;
3757
3758
3759         list_for_each_entry(file, &tr->events, list) {
3760                 ret = event_create_dir(tr->event_dir, file);
3761                 if (ret < 0)
3762                         pr_warn("Could not create directory for event %s\n",
3763                                 trace_event_name(file->event_call));
3764         }
3765 }
3766
3767 /*
3768  * For early boot up, the top trace array and the trace arrays created
3769  * by boot-time tracing require to have a list of events that can be
3770  * enabled. This must be done before the filesystem is set up in order
3771  * to allow events to be traced early.
3772  */
3773 void __trace_early_add_events(struct trace_array *tr)
3774 {
3775         struct trace_event_call *call;
3776         int ret;
3777
3778         list_for_each_entry(call, &ftrace_events, list) {
3779                 /* Early boot up should not have any modules loaded */
3780                 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3781                     WARN_ON_ONCE(call->module))
3782                         continue;
3783
3784                 ret = __trace_early_add_new_event(call, tr);
3785                 if (ret < 0)
3786                         pr_warn("Could not create early event %s\n",
3787                                 trace_event_name(call));
3788         }
3789 }
3790
3791 /* Remove the event directory structure for a trace directory. */
3792 static void
3793 __trace_remove_event_dirs(struct trace_array *tr)
3794 {
3795         struct trace_event_file *file, *next;
3796
3797         list_for_each_entry_safe(file, next, &tr->events, list)
3798                 remove_event_file_dir(file);
3799 }
3800
3801 static void __add_event_to_tracers(struct trace_event_call *call)
3802 {
3803         struct trace_array *tr;
3804
3805         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3806                 __trace_add_new_event(call, tr);
3807 }
3808
3809 extern struct trace_event_call *__start_ftrace_events[];
3810 extern struct trace_event_call *__stop_ftrace_events[];
3811
3812 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3813
3814 static __init int setup_trace_event(char *str)
3815 {
3816         strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3817         trace_set_ring_buffer_expanded(NULL);
3818         disable_tracing_selftest("running event tracing");
3819
3820         return 1;
3821 }
3822 __setup("trace_event=", setup_trace_event);
3823
3824 static int events_callback(const char *name, umode_t *mode, void **data,
3825                            const struct file_operations **fops)
3826 {
3827         if (strcmp(name, "enable") == 0) {
3828                 *mode = TRACE_MODE_WRITE;
3829                 *fops = &ftrace_tr_enable_fops;
3830                 return 1;
3831         }
3832
3833         if (strcmp(name, "header_page") == 0) {
3834                 *mode = TRACE_MODE_READ;
3835                 *fops = &ftrace_show_header_page_fops;
3836
3837         } else if (strcmp(name, "header_event") == 0) {
3838                 *mode = TRACE_MODE_READ;
3839                 *fops = &ftrace_show_header_event_fops;
3840         } else
3841                 return 0;
3842
3843         return 1;
3844 }
3845
3846 /* Expects to have event_mutex held when called */
3847 static int
3848 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3849 {
3850         struct eventfs_inode *e_events;
3851         struct dentry *entry;
3852         int nr_entries;
3853         static struct eventfs_entry events_entries[] = {
3854                 {
3855                         .name           = "enable",
3856                         .callback       = events_callback,
3857                 },
3858                 {
3859                         .name           = "header_page",
3860                         .callback       = events_callback,
3861                 },
3862                 {
3863                         .name           = "header_event",
3864                         .callback       = events_callback,
3865                 },
3866         };
3867
3868         entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3869                                   tr, &ftrace_set_event_fops);
3870         if (!entry)
3871                 return -ENOMEM;
3872
3873         nr_entries = ARRAY_SIZE(events_entries);
3874
3875         e_events = eventfs_create_events_dir("events", parent, events_entries,
3876                                              nr_entries, tr);
3877         if (IS_ERR(e_events)) {
3878                 pr_warn("Could not create tracefs 'events' directory\n");
3879                 return -ENOMEM;
3880         }
3881
3882         /* There are not as crucial, just warn if they are not created */
3883
3884         trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3885                           tr, &ftrace_set_event_pid_fops);
3886
3887         trace_create_file("set_event_notrace_pid",
3888                           TRACE_MODE_WRITE, parent, tr,
3889                           &ftrace_set_event_notrace_pid_fops);
3890
3891         tr->event_dir = e_events;
3892
3893         return 0;
3894 }
3895
3896 /**
3897  * event_trace_add_tracer - add a instance of a trace_array to events
3898  * @parent: The parent dentry to place the files/directories for events in
3899  * @tr: The trace array associated with these events
3900  *
3901  * When a new instance is created, it needs to set up its events
3902  * directory, as well as other files associated with events. It also
3903  * creates the event hierarchy in the @parent/events directory.
3904  *
3905  * Returns 0 on success.
3906  *
3907  * Must be called with event_mutex held.
3908  */
3909 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3910 {
3911         int ret;
3912
3913         lockdep_assert_held(&event_mutex);
3914
3915         ret = create_event_toplevel_files(parent, tr);
3916         if (ret)
3917                 goto out;
3918
3919         down_write(&trace_event_sem);
3920         /* If tr already has the event list, it is initialized in early boot. */
3921         if (unlikely(!list_empty(&tr->events)))
3922                 __trace_early_add_event_dirs(tr);
3923         else
3924                 __trace_add_event_dirs(tr);
3925         up_write(&trace_event_sem);
3926
3927  out:
3928         return ret;
3929 }
3930
3931 /*
3932  * The top trace array already had its file descriptors created.
3933  * Now the files themselves need to be created.
3934  */
3935 static __init int
3936 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3937 {
3938         int ret;
3939
3940         mutex_lock(&event_mutex);
3941
3942         ret = create_event_toplevel_files(parent, tr);
3943         if (ret)
3944                 goto out_unlock;
3945
3946         down_write(&trace_event_sem);
3947         __trace_early_add_event_dirs(tr);
3948         up_write(&trace_event_sem);
3949
3950  out_unlock:
3951         mutex_unlock(&event_mutex);
3952
3953         return ret;
3954 }
3955
3956 /* Must be called with event_mutex held */
3957 int event_trace_del_tracer(struct trace_array *tr)
3958 {
3959         lockdep_assert_held(&event_mutex);
3960
3961         /* Disable any event triggers and associated soft-disabled events */
3962         clear_event_triggers(tr);
3963
3964         /* Clear the pid list */
3965         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3966
3967         /* Disable any running events */
3968         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3969
3970         /* Make sure no more events are being executed */
3971         tracepoint_synchronize_unregister();
3972
3973         down_write(&trace_event_sem);
3974         __trace_remove_event_dirs(tr);
3975         eventfs_remove_events_dir(tr->event_dir);
3976         up_write(&trace_event_sem);
3977
3978         tr->event_dir = NULL;
3979
3980         return 0;
3981 }
3982
3983 static __init int event_trace_memsetup(void)
3984 {
3985         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3986         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3987         return 0;
3988 }
3989
3990 __init void
3991 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3992 {
3993         char *token;
3994         int ret;
3995
3996         while (true) {
3997                 token = strsep(&buf, ",");
3998
3999                 if (!token)
4000                         break;
4001
4002                 if (*token) {
4003                         /* Restarting syscalls requires that we stop them first */
4004                         if (disable_first)
4005                                 ftrace_set_clr_event(tr, token, 0);
4006
4007                         ret = ftrace_set_clr_event(tr, token, 1);
4008                         if (ret)
4009                                 pr_warn("Failed to enable trace event: %s\n", token);
4010                 }
4011
4012                 /* Put back the comma to allow this to be called again */
4013                 if (buf)
4014                         *(buf - 1) = ',';
4015         }
4016 }
4017
4018 static __init int event_trace_enable(void)
4019 {
4020         struct trace_array *tr = top_trace_array();
4021         struct trace_event_call **iter, *call;
4022         int ret;
4023
4024         if (!tr)
4025                 return -ENODEV;
4026
4027         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4028
4029                 call = *iter;
4030                 ret = event_init(call);
4031                 if (!ret)
4032                         list_add(&call->list, &ftrace_events);
4033         }
4034
4035         register_trigger_cmds();
4036
4037         /*
4038          * We need the top trace array to have a working set of trace
4039          * points at early init, before the debug files and directories
4040          * are created. Create the file entries now, and attach them
4041          * to the actual file dentries later.
4042          */
4043         __trace_early_add_events(tr);
4044
4045         early_enable_events(tr, bootup_event_buf, false);
4046
4047         trace_printk_start_comm();
4048
4049         register_event_cmds();
4050
4051
4052         return 0;
4053 }
4054
4055 /*
4056  * event_trace_enable() is called from trace_event_init() first to
4057  * initialize events and perhaps start any events that are on the
4058  * command line. Unfortunately, there are some events that will not
4059  * start this early, like the system call tracepoints that need
4060  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4061  * event_trace_enable() is called before pid 1 starts, and this flag
4062  * is never set, making the syscall tracepoint never get reached, but
4063  * the event is enabled regardless (and not doing anything).
4064  */
4065 static __init int event_trace_enable_again(void)
4066 {
4067         struct trace_array *tr;
4068
4069         tr = top_trace_array();
4070         if (!tr)
4071                 return -ENODEV;
4072
4073         early_enable_events(tr, bootup_event_buf, true);
4074
4075         return 0;
4076 }
4077
4078 early_initcall(event_trace_enable_again);
4079
4080 /* Init fields which doesn't related to the tracefs */
4081 static __init int event_trace_init_fields(void)
4082 {
4083         if (trace_define_generic_fields())
4084                 pr_warn("tracing: Failed to allocated generic fields");
4085
4086         if (trace_define_common_fields())
4087                 pr_warn("tracing: Failed to allocate common fields");
4088
4089         return 0;
4090 }
4091
4092 __init int event_trace_init(void)
4093 {
4094         struct trace_array *tr;
4095         int ret;
4096
4097         tr = top_trace_array();
4098         if (!tr)
4099                 return -ENODEV;
4100
4101         trace_create_file("available_events", TRACE_MODE_READ,
4102                           NULL, tr, &ftrace_avail_fops);
4103
4104         ret = early_event_add_tracer(NULL, tr);
4105         if (ret)
4106                 return ret;
4107
4108 #ifdef CONFIG_MODULES
4109         ret = register_module_notifier(&trace_module_nb);
4110         if (ret)
4111                 pr_warn("Failed to register trace events module notifier\n");
4112 #endif
4113
4114         eventdir_initialized = true;
4115
4116         return 0;
4117 }
4118
4119 void __init trace_event_init(void)
4120 {
4121         event_trace_memsetup();
4122         init_ftrace_syscalls();
4123         event_trace_enable();
4124         event_trace_init_fields();
4125 }
4126
4127 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4128
4129 static DEFINE_SPINLOCK(test_spinlock);
4130 static DEFINE_SPINLOCK(test_spinlock_irq);
4131 static DEFINE_MUTEX(test_mutex);
4132
4133 static __init void test_work(struct work_struct *dummy)
4134 {
4135         spin_lock(&test_spinlock);
4136         spin_lock_irq(&test_spinlock_irq);
4137         udelay(1);
4138         spin_unlock_irq(&test_spinlock_irq);
4139         spin_unlock(&test_spinlock);
4140
4141         mutex_lock(&test_mutex);
4142         msleep(1);
4143         mutex_unlock(&test_mutex);
4144 }
4145
4146 static __init int event_test_thread(void *unused)
4147 {
4148         void *test_malloc;
4149
4150         test_malloc = kmalloc(1234, GFP_KERNEL);
4151         if (!test_malloc)
4152                 pr_info("failed to kmalloc\n");
4153
4154         schedule_on_each_cpu(test_work);
4155
4156         kfree(test_malloc);
4157
4158         set_current_state(TASK_INTERRUPTIBLE);
4159         while (!kthread_should_stop()) {
4160                 schedule();
4161                 set_current_state(TASK_INTERRUPTIBLE);
4162         }
4163         __set_current_state(TASK_RUNNING);
4164
4165         return 0;
4166 }
4167
4168 /*
4169  * Do various things that may trigger events.
4170  */
4171 static __init void event_test_stuff(void)
4172 {
4173         struct task_struct *test_thread;
4174
4175         test_thread = kthread_run(event_test_thread, NULL, "test-events");
4176         msleep(1);
4177         kthread_stop(test_thread);
4178 }
4179
4180 /*
4181  * For every trace event defined, we will test each trace point separately,
4182  * and then by groups, and finally all trace points.
4183  */
4184 static __init void event_trace_self_tests(void)
4185 {
4186         struct trace_subsystem_dir *dir;
4187         struct trace_event_file *file;
4188         struct trace_event_call *call;
4189         struct event_subsystem *system;
4190         struct trace_array *tr;
4191         int ret;
4192
4193         tr = top_trace_array();
4194         if (!tr)
4195                 return;
4196
4197         pr_info("Running tests on trace events:\n");
4198
4199         list_for_each_entry(file, &tr->events, list) {
4200
4201                 call = file->event_call;
4202
4203                 /* Only test those that have a probe */
4204                 if (!call->class || !call->class->probe)
4205                         continue;
4206
4207 /*
4208  * Testing syscall events here is pretty useless, but
4209  * we still do it if configured. But this is time consuming.
4210  * What we really need is a user thread to perform the
4211  * syscalls as we test.
4212  */
4213 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4214                 if (call->class->system &&
4215                     strcmp(call->class->system, "syscalls") == 0)
4216                         continue;
4217 #endif
4218
4219                 pr_info("Testing event %s: ", trace_event_name(call));
4220
4221                 /*
4222                  * If an event is already enabled, someone is using
4223                  * it and the self test should not be on.
4224                  */
4225                 if (file->flags & EVENT_FILE_FL_ENABLED) {
4226                         pr_warn("Enabled event during self test!\n");
4227                         WARN_ON_ONCE(1);
4228                         continue;
4229                 }
4230
4231                 ftrace_event_enable_disable(file, 1);
4232                 event_test_stuff();
4233                 ftrace_event_enable_disable(file, 0);
4234
4235                 pr_cont("OK\n");
4236         }
4237
4238         /* Now test at the sub system level */
4239
4240         pr_info("Running tests on trace event systems:\n");
4241
4242         list_for_each_entry(dir, &tr->systems, list) {
4243
4244                 system = dir->subsystem;
4245
4246                 /* the ftrace system is special, skip it */
4247                 if (strcmp(system->name, "ftrace") == 0)
4248                         continue;
4249
4250                 pr_info("Testing event system %s: ", system->name);
4251
4252                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4253                 if (WARN_ON_ONCE(ret)) {
4254                         pr_warn("error enabling system %s\n",
4255                                 system->name);
4256                         continue;
4257                 }
4258
4259                 event_test_stuff();
4260
4261                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4262                 if (WARN_ON_ONCE(ret)) {
4263                         pr_warn("error disabling system %s\n",
4264                                 system->name);
4265                         continue;
4266                 }
4267
4268                 pr_cont("OK\n");
4269         }
4270
4271         /* Test with all events enabled */
4272
4273         pr_info("Running tests on all trace events:\n");
4274         pr_info("Testing all events: ");
4275
4276         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4277         if (WARN_ON_ONCE(ret)) {
4278                 pr_warn("error enabling all events\n");
4279                 return;
4280         }
4281
4282         event_test_stuff();
4283
4284         /* reset sysname */
4285         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4286         if (WARN_ON_ONCE(ret)) {
4287                 pr_warn("error disabling all events\n");
4288                 return;
4289         }
4290
4291         pr_cont("OK\n");
4292 }
4293
4294 #ifdef CONFIG_FUNCTION_TRACER
4295
4296 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4297
4298 static struct trace_event_file event_trace_file __initdata;
4299
4300 static void __init
4301 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4302                           struct ftrace_ops *op, struct ftrace_regs *regs)
4303 {
4304         struct trace_buffer *buffer;
4305         struct ring_buffer_event *event;
4306         struct ftrace_entry *entry;
4307         unsigned int trace_ctx;
4308         long disabled;
4309         int cpu;
4310
4311         trace_ctx = tracing_gen_ctx();
4312         preempt_disable_notrace();
4313         cpu = raw_smp_processor_id();
4314         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4315
4316         if (disabled != 1)
4317                 goto out;
4318
4319         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4320                                                 TRACE_FN, sizeof(*entry),
4321                                                 trace_ctx);
4322         if (!event)
4323                 goto out;
4324         entry   = ring_buffer_event_data(event);
4325         entry->ip                       = ip;
4326         entry->parent_ip                = parent_ip;
4327
4328         event_trigger_unlock_commit(&event_trace_file, buffer, event,
4329                                     entry, trace_ctx);
4330  out:
4331         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4332         preempt_enable_notrace();
4333 }
4334
4335 static struct ftrace_ops trace_ops __initdata  =
4336 {
4337         .func = function_test_events_call,
4338 };
4339
4340 static __init void event_trace_self_test_with_function(void)
4341 {
4342         int ret;
4343
4344         event_trace_file.tr = top_trace_array();
4345         if (WARN_ON(!event_trace_file.tr))
4346                 return;
4347
4348         ret = register_ftrace_function(&trace_ops);
4349         if (WARN_ON(ret < 0)) {
4350                 pr_info("Failed to enable function tracer for event tests\n");
4351                 return;
4352         }
4353         pr_info("Running tests again, along with the function tracer\n");
4354         event_trace_self_tests();
4355         unregister_ftrace_function(&trace_ops);
4356 }
4357 #else
4358 static __init void event_trace_self_test_with_function(void)
4359 {
4360 }
4361 #endif
4362
4363 static __init int event_trace_self_tests_init(void)
4364 {
4365         if (!tracing_selftest_disabled) {
4366                 event_trace_self_tests();
4367                 event_trace_self_test_with_function();
4368         }
4369
4370         return 0;
4371 }
4372
4373 late_initcall(event_trace_self_tests_init);
4374
4375 #endif