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