kernel/trace/trace_events.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * event tracer
   4  *
   5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6  *
   7  *  - Added format output of fields of the trace point.
   8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9  *
  10  */
  11
  12 #define pr_fmt(fmt) fmt
  13
  14 #include <linux/workqueue.h>
  15 #include <linux/security.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/kthread.h>
  18 #include <linux/tracefs.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/module.h>
  21 #include <linux/ctype.h>
  22 #include <linux/sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/delay.h>
  25
  26 #include <trace/events/sched.h>
  27 #include <trace/syscall.h>
  28
  29 #include <asm/setup.h>
  30
  31 #include "trace_output.h"
  32
  33 #undef TRACE_SYSTEM
  34 #define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36 DEFINE_MUTEX(event_mutex);
  37
  38 LIST_HEAD(ftrace_events);
  39 static LIST_HEAD(ftrace_generic_fields);
  40 static LIST_HEAD(ftrace_common_fields);
  41 static bool eventdir_initialized;
  42
  43 static LIST_HEAD(module_strings);
  44
  45 struct module_string {
  46         struct list_head        next;
  47         struct module           *module;
  48         char                    *str;
  49 };
  50
  51 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  52
  53 static struct kmem_cache *field_cachep;
  54 static struct kmem_cache *file_cachep;
  55
  56 static inline int system_refcount(struct event_subsystem *system)
  57 {
  58         return system->ref_count;
  59 }
  60
  61 static int system_refcount_inc(struct event_subsystem *system)
  62 {
  63         return system->ref_count++;
  64 }
  65
  66 static int system_refcount_dec(struct event_subsystem *system)
  67 {
  68         return --system->ref_count;
  69 }
  70
  71 /* Double loops, do not use break, only goto's work */
  72 #define do_for_each_event_file(tr, file)                        \
  73         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  74                 list_for_each_entry(file, &tr->events, list)
  75
  76 #define do_for_each_event_file_safe(tr, file)                   \
  77         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  78                 struct trace_event_file *___n;                          \
  79                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  80
  81 #define while_for_each_event_file()             \
  82         }
  83
  84 static struct ftrace_event_field *
  85 __find_event_field(struct list_head *head, char *name)
  86 {
  87         struct ftrace_event_field *field;
  88
  89         list_for_each_entry(field, head, link) {
  90                 if (!strcmp(field->name, name))
  91                         return field;
  92         }
  93
  94         return NULL;
  95 }
  96
  97 struct ftrace_event_field *
  98 trace_find_event_field(struct trace_event_call *call, char *name)
  99 {
 100         struct ftrace_event_field *field;
 101         struct list_head *head;
 102
 103         head = trace_get_fields(call);
 104         field = __find_event_field(head, name);
 105         if (field)
 106                 return field;
 107
 108         field = __find_event_field(&ftrace_generic_fields, name);
 109         if (field)
 110                 return field;
 111
 112         return __find_event_field(&ftrace_common_fields, name);
 113 }
 114
 115 static int __trace_define_field(struct list_head *head, const char *type,
 116                                 const char *name, int offset, int size,
 117                                 int is_signed, int filter_type, int len)
 118 {
 119         struct ftrace_event_field *field;
 120
 121         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 122         if (!field)
 123                 return -ENOMEM;
 124
 125         field->name = name;
 126         field->type = type;
 127
 128         if (filter_type == FILTER_OTHER)
 129                 field->filter_type = filter_assign_type(type);
 130         else
 131                 field->filter_type = filter_type;
 132
 133         field->offset = offset;
 134         field->size = size;
 135         field->is_signed = is_signed;
 136         field->len = len;
 137
 138         list_add(&field->link, head);
 139
 140         return 0;
 141 }
 142
 143 int trace_define_field(struct trace_event_call *call, const char *type,
 144                        const char *name, int offset, int size, int is_signed,
 145                        int filter_type)
 146 {
 147         struct list_head *head;
 148
 149         if (WARN_ON(!call->class))
 150                 return 0;
 151
 152         head = trace_get_fields(call);
 153         return __trace_define_field(head, type, name, offset, size,
 154                                     is_signed, filter_type, 0);
 155 }
 156 EXPORT_SYMBOL_GPL(trace_define_field);
 157
 158 static int trace_define_field_ext(struct trace_event_call *call, const char *type,
 159                        const char *name, int offset, int size, int is_signed,
 160                        int filter_type, int len)
 161 {
 162         struct list_head *head;
 163
 164         if (WARN_ON(!call->class))
 165                 return 0;
 166
 167         head = trace_get_fields(call);
 168         return __trace_define_field(head, type, name, offset, size,
 169                                     is_signed, filter_type, len);
 170 }
 171
 172 #define __generic_field(type, item, filter_type)                        \
 173         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 174                                    #item, 0, 0, is_signed_type(type),   \
 175                                    filter_type, 0);                     \
 176         if (ret)                                                        \
 177                 return ret;
 178
 179 #define __common_field(type, item)                                      \
 180         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 181                                    "common_" #item,                     \
 182                                    offsetof(typeof(ent), item),         \
 183                                    sizeof(ent.item),                    \
 184                                    is_signed_type(type), FILTER_OTHER, 0);      \
 185         if (ret)                                                        \
 186                 return ret;
 187
 188 static int trace_define_generic_fields(void)
 189 {
 190         int ret;
 191
 192         __generic_field(int, CPU, FILTER_CPU);
 193         __generic_field(int, cpu, FILTER_CPU);
 194         __generic_field(int, common_cpu, FILTER_CPU);
 195         __generic_field(char *, COMM, FILTER_COMM);
 196         __generic_field(char *, comm, FILTER_COMM);
 197         __generic_field(char *, stacktrace, FILTER_STACKTRACE);
 198         __generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
 199
 200         return ret;
 201 }
 202
 203 static int trace_define_common_fields(void)
 204 {
 205         int ret;
 206         struct trace_entry ent;
 207
 208         __common_field(unsigned short, type);
 209         __common_field(unsigned char, flags);
 210         /* Holds both preempt_count and migrate_disable */
 211         __common_field(unsigned char, preempt_count);
 212         __common_field(int, pid);
 213
 214         return ret;
 215 }
 216
 217 static void trace_destroy_fields(struct trace_event_call *call)
 218 {
 219         struct ftrace_event_field *field, *next;
 220         struct list_head *head;
 221
 222         head = trace_get_fields(call);
 223         list_for_each_entry_safe(field, next, head, link) {
 224                 list_del(&field->link);
 225                 kmem_cache_free(field_cachep, field);
 226         }
 227 }
 228
 229 /*
 230  * run-time version of trace_event_get_offsets_<call>() that returns the last
 231  * accessible offset of trace fields excluding __dynamic_array bytes
 232  */
 233 int trace_event_get_offsets(struct trace_event_call *call)
 234 {
 235         struct ftrace_event_field *tail;
 236         struct list_head *head;
 237
 238         head = trace_get_fields(call);
 239         /*
 240          * head->next points to the last field with the largest offset,
 241          * since it was added last by trace_define_field()
 242          */
 243         tail = list_first_entry(head, struct ftrace_event_field, link);
 244         return tail->offset + tail->size;
 245 }
 246
 247 /*
 248  * Check if the referenced field is an array and return true,
 249  * as arrays are OK to dereference.
 250  */
 251 static bool test_field(const char *fmt, struct trace_event_call *call)
 252 {
 253         struct trace_event_fields *field = call->class->fields_array;
 254         const char *array_descriptor;
 255         const char *p = fmt;
 256         int len;
 257
 258         if (!(len = str_has_prefix(fmt, "REC->")))
 259                 return false;
 260         fmt += len;
 261         for (p = fmt; *p; p++) {
 262                 if (!isalnum(*p) && *p != '_')
 263                         break;
 264         }
 265         len = p - fmt;
 266
 267         for (; field->type; field++) {
 268                 if (strncmp(field->name, fmt, len) ||
 269                     field->name[len])
 270                         continue;
 271                 array_descriptor = strchr(field->type, '[');
 272                 /* This is an array and is OK to dereference. */
 273                 return array_descriptor != NULL;
 274         }
 275         return false;
 276 }
 277
 278 /*
 279  * Examine the print fmt of the event looking for unsafe dereference
 280  * pointers using %p* that could be recorded in the trace event and
 281  * much later referenced after the pointer was freed. Dereferencing
 282  * pointers are OK, if it is dereferenced into the event itself.
 283  */
 284 static void test_event_printk(struct trace_event_call *call)
 285 {
 286         u64 dereference_flags = 0;
 287         bool first = true;
 288         const char *fmt, *c, *r, *a;
 289         int parens = 0;
 290         char in_quote = 0;
 291         int start_arg = 0;
 292         int arg = 0;
 293         int i;
 294
 295         fmt = call->print_fmt;
 296
 297         if (!fmt)
 298                 return;
 299
 300         for (i = 0; fmt[i]; i++) {
 301                 switch (fmt[i]) {
 302                 case '\\':
 303                         i++;
 304                         if (!fmt[i])
 305                                 return;
 306                         continue;
 307                 case '"':
 308                 case '\'':
 309                         /*
 310                          * The print fmt starts with a string that
 311                          * is processed first to find %p* usage,
 312                          * then after the first string, the print fmt
 313                          * contains arguments that are used to check
 314                          * if the dereferenced %p* usage is safe.
 315                          */
 316                         if (first) {
 317                                 if (fmt[i] == '\'')
 318                                         continue;
 319                                 if (in_quote) {
 320                                         arg = 0;
 321                                         first = false;
 322                                         /*
 323                                          * If there was no %p* uses
 324                                          * the fmt is OK.
 325                                          */
 326                                         if (!dereference_flags)
 327                                                 return;
 328                                 }
 329                         }
 330                         if (in_quote) {
 331                                 if (in_quote == fmt[i])
 332                                         in_quote = 0;
 333                         } else {
 334                                 in_quote = fmt[i];
 335                         }
 336                         continue;
 337                 case '%':
 338                         if (!first || !in_quote)
 339                                 continue;
 340                         i++;
 341                         if (!fmt[i])
 342                                 return;
 343                         switch (fmt[i]) {
 344                         case '%':
 345                                 continue;
 346                         case 'p':
 347                                 /* Find dereferencing fields */
 348                                 switch (fmt[i + 1]) {
 349                                 case 'B': case 'R': case 'r':
 350                                 case 'b': case 'M': case 'm':
 351                                 case 'I': case 'i': case 'E':
 352                                 case 'U': case 'V': case 'N':
 353                                 case 'a': case 'd': case 'D':
 354                                 case 'g': case 't': case 'C':
 355                                 case 'O': case 'f':
 356                                         if (WARN_ONCE(arg == 63,
 357                                                       "Too many args for event: %s",
 358                                                       trace_event_name(call)))
 359                                                 return;
 360                                         dereference_flags |= 1ULL << arg;
 361                                 }
 362                                 break;
 363                         default:
 364                         {
 365                                 bool star = false;
 366                                 int j;
 367
 368                                 /* Increment arg if %*s exists. */
 369                                 for (j = 0; fmt[i + j]; j++) {
 370                                         if (isdigit(fmt[i + j]) ||
 371                                             fmt[i + j] == '.')
 372                                                 continue;
 373                                         if (fmt[i + j] == '*') {
 374                                                 star = true;
 375                                                 continue;
 376                                         }
 377                                         if ((fmt[i + j] == 's') && star)
 378                                                 arg++;
 379                                         break;
 380                                 }
 381                                 break;
 382                         } /* default */
 383
 384                         } /* switch */
 385                         arg++;
 386                         continue;
 387                 case '(':
 388                         if (in_quote)
 389                                 continue;
 390                         parens++;
 391                         continue;
 392                 case ')':
 393                         if (in_quote)
 394                                 continue;
 395                         parens--;
 396                         if (WARN_ONCE(parens < 0,
 397                                       "Paren mismatch for event: %s\narg='%s'\n%*s",
 398                                       trace_event_name(call),
 399                                       fmt + start_arg,
 400                                       (i - start_arg) + 5, "^"))
 401                                 return;
 402                         continue;
 403                 case ',':
 404                         if (in_quote || parens)
 405                                 continue;
 406                         i++;
 407                         while (isspace(fmt[i]))
 408                                 i++;
 409                         start_arg = i;
 410                         if (!(dereference_flags & (1ULL << arg)))
 411                                 goto next_arg;
 412
 413                         /* Find the REC-> in the argument */
 414                         c = strchr(fmt + i, ',');
 415                         r = strstr(fmt + i, "REC->");
 416                         if (r && (!c || r < c)) {
 417                                 /*
 418                                  * Addresses of events on the buffer,
 419                                  * or an array on the buffer is
 420                                  * OK to dereference.
 421                                  * There's ways to fool this, but
 422                                  * this is to catch common mistakes,
 423                                  * not malicious code.
 424                                  */
 425                                 a = strchr(fmt + i, '&');
 426                                 if ((a && (a < r)) || test_field(r, call))
 427                                         dereference_flags &= ~(1ULL << arg);
 428                         } else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
 429                                    (!c || r < c)) {
 430                                 dereference_flags &= ~(1ULL << arg);
 431                         } else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
 432                                    (!c || r < c)) {
 433                                 dereference_flags &= ~(1ULL << arg);
 434                         }
 435
 436                 next_arg:
 437                         i--;
 438                         arg++;
 439                 }
 440         }
 441
 442         /*
 443          * If you triggered the below warning, the trace event reported
 444          * uses an unsafe dereference pointer %p*. As the data stored
 445          * at the trace event time may no longer exist when the trace
 446          * event is printed, dereferencing to the original source is
 447          * unsafe. The source of the dereference must be copied into the
 448          * event itself, and the dereference must access the copy instead.
 449          */
 450         if (WARN_ON_ONCE(dereference_flags)) {
 451                 arg = 1;
 452                 while (!(dereference_flags & 1)) {
 453                         dereference_flags >>= 1;
 454                         arg++;
 455                 }
 456                 pr_warn("event %s has unsafe dereference of argument %d\n",
 457                         trace_event_name(call), arg);
 458                 pr_warn("print_fmt: %s\n", fmt);
 459         }
 460 }
 461
 462 int trace_event_raw_init(struct trace_event_call *call)
 463 {
 464         int id;
 465
 466         id = register_trace_event(&call->event);
 467         if (!id)
 468                 return -ENODEV;
 469
 470         test_event_printk(call);
 471
 472         return 0;
 473 }
 474 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 475
 476 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 477 {
 478         struct trace_array *tr = trace_file->tr;
 479         struct trace_array_cpu *data;
 480         struct trace_pid_list *no_pid_list;
 481         struct trace_pid_list *pid_list;
 482
 483         pid_list = rcu_dereference_raw(tr->filtered_pids);
 484         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 485
 486         if (!pid_list && !no_pid_list)
 487                 return false;
 488
 489         data = this_cpu_ptr(tr->array_buffer.data);
 490
 491         return data->ignore_pid;
 492 }
 493 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 494
 495 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 496                                  struct trace_event_file *trace_file,
 497                                  unsigned long len)
 498 {
 499         struct trace_event_call *event_call = trace_file->event_call;
 500
 501         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 502             trace_event_ignore_this_pid(trace_file))
 503                 return NULL;
 504
 505         /*
 506          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 507          * preemption (adding one to the preempt_count). Since we are
 508          * interested in the preempt_count at the time the tracepoint was
 509          * hit, we need to subtract one to offset the increment.
 510          */
 511         fbuffer->trace_ctx = tracing_gen_ctx_dec();
 512         fbuffer->trace_file = trace_file;
 513
 514         fbuffer->event =
 515                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 516                                                 event_call->event.type, len,
 517                                                 fbuffer->trace_ctx);
 518         if (!fbuffer->event)
 519                 return NULL;
 520
 521         fbuffer->regs = NULL;
 522         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 523         return fbuffer->entry;
 524 }
 525 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 526
 527 int trace_event_reg(struct trace_event_call *call,
 528                     enum trace_reg type, void *data)
 529 {
 530         struct trace_event_file *file = data;
 531
 532         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 533         switch (type) {
 534         case TRACE_REG_REGISTER:
 535                 return tracepoint_probe_register(call->tp,
 536                                                  call->class->probe,
 537                                                  file);
 538         case TRACE_REG_UNREGISTER:
 539                 tracepoint_probe_unregister(call->tp,
 540                                             call->class->probe,
 541                                             file);
 542                 return 0;
 543
 544 #ifdef CONFIG_PERF_EVENTS
 545         case TRACE_REG_PERF_REGISTER:
 546                 return tracepoint_probe_register(call->tp,
 547                                                  call->class->perf_probe,
 548                                                  call);
 549         case TRACE_REG_PERF_UNREGISTER:
 550                 tracepoint_probe_unregister(call->tp,
 551                                             call->class->perf_probe,
 552                                             call);
 553                 return 0;
 554         case TRACE_REG_PERF_OPEN:
 555         case TRACE_REG_PERF_CLOSE:
 556         case TRACE_REG_PERF_ADD:
 557         case TRACE_REG_PERF_DEL:
 558                 return 0;
 559 #endif
 560         }
 561         return 0;
 562 }
 563 EXPORT_SYMBOL_GPL(trace_event_reg);
 564
 565 void trace_event_enable_cmd_record(bool enable)
 566 {
 567         struct trace_event_file *file;
 568         struct trace_array *tr;
 569
 570         lockdep_assert_held(&event_mutex);
 571
 572         do_for_each_event_file(tr, file) {
 573
 574                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 575                         continue;
 576
 577                 if (enable) {
 578                         tracing_start_cmdline_record();
 579                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 580                 } else {
 581                         tracing_stop_cmdline_record();
 582                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 583                 }
 584         } while_for_each_event_file();
 585 }
 586
 587 void trace_event_enable_tgid_record(bool enable)
 588 {
 589         struct trace_event_file *file;
 590         struct trace_array *tr;
 591
 592         lockdep_assert_held(&event_mutex);
 593
 594         do_for_each_event_file(tr, file) {
 595                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 596                         continue;
 597
 598                 if (enable) {
 599                         tracing_start_tgid_record();
 600                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 601                 } else {
 602                         tracing_stop_tgid_record();
 603                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 604                                   &file->flags);
 605                 }
 606         } while_for_each_event_file();
 607 }
 608
 609 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 610                                          int enable, int soft_disable)
 611 {
 612         struct trace_event_call *call = file->event_call;
 613         struct trace_array *tr = file->tr;
 614         int ret = 0;
 615         int disable;
 616
 617         switch (enable) {
 618         case 0:
 619                 /*
 620                  * When soft_disable is set and enable is cleared, the sm_ref
 621                  * reference counter is decremented. If it reaches 0, we want
 622                  * to clear the SOFT_DISABLED flag but leave the event in the
 623                  * state that it was. That is, if the event was enabled and
 624                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 625                  * is set we do not want the event to be enabled before we
 626                  * clear the bit.
 627                  *
 628                  * When soft_disable is not set but the SOFT_MODE flag is,
 629                  * we do nothing. Do not disable the tracepoint, otherwise
 630                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 631                  */
 632                 if (soft_disable) {
 633                         if (atomic_dec_return(&file->sm_ref) > 0)
 634                                 break;
 635                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 636                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 637                         /* Disable use of trace_buffered_event */
 638                         trace_buffered_event_disable();
 639                 } else
 640                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 641
 642                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 643                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 644                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 645                                 tracing_stop_cmdline_record();
 646                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 647                         }
 648
 649                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 650                                 tracing_stop_tgid_record();
 651                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 652                         }
 653
 654                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 655                 }
 656                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 657                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 658                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 659                 else
 660                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 661                 break;
 662         case 1:
 663                 /*
 664                  * When soft_disable is set and enable is set, we want to
 665                  * register the tracepoint for the event, but leave the event
 666                  * as is. That means, if the event was already enabled, we do
 667                  * nothing (but set SOFT_MODE). If the event is disabled, we
 668                  * set SOFT_DISABLED before enabling the event tracepoint, so
 669                  * it still seems to be disabled.
 670                  */
 671                 if (!soft_disable)
 672                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 673                 else {
 674                         if (atomic_inc_return(&file->sm_ref) > 1)
 675                                 break;
 676                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 677                         /* Enable use of trace_buffered_event */
 678                         trace_buffered_event_enable();
 679                 }
 680
 681                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 682                         bool cmd = false, tgid = false;
 683
 684                         /* Keep the event disabled, when going to SOFT_MODE. */
 685                         if (soft_disable)
 686                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 687
 688                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 689                                 cmd = true;
 690                                 tracing_start_cmdline_record();
 691                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 692                         }
 693
 694                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 695                                 tgid = true;
 696                                 tracing_start_tgid_record();
 697                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 698                         }
 699
 700                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 701                         if (ret) {
 702                                 if (cmd)
 703                                         tracing_stop_cmdline_record();
 704                                 if (tgid)
 705                                         tracing_stop_tgid_record();
 706                                 pr_info("event trace: Could not enable event "
 707                                         "%s\n", trace_event_name(call));
 708                                 break;
 709                         }
 710                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 711
 712                         /* WAS_ENABLED gets set but never cleared. */
 713                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 714                 }
 715                 break;
 716         }
 717
 718         return ret;
 719 }
 720
 721 int trace_event_enable_disable(struct trace_event_file *file,
 722                                int enable, int soft_disable)
 723 {
 724         return __ftrace_event_enable_disable(file, enable, soft_disable);
 725 }
 726
 727 static int ftrace_event_enable_disable(struct trace_event_file *file,
 728                                        int enable)
 729 {
 730         return __ftrace_event_enable_disable(file, enable, 0);
 731 }
 732
 733 static void ftrace_clear_events(struct trace_array *tr)
 734 {
 735         struct trace_event_file *file;
 736
 737         mutex_lock(&event_mutex);
 738         list_for_each_entry(file, &tr->events, list) {
 739                 ftrace_event_enable_disable(file, 0);
 740         }
 741         mutex_unlock(&event_mutex);
 742 }
 743
 744 static void
 745 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 746 {
 747         struct trace_pid_list *pid_list;
 748         struct trace_array *tr = data;
 749
 750         pid_list = rcu_dereference_raw(tr->filtered_pids);
 751         trace_filter_add_remove_task(pid_list, NULL, task);
 752
 753         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 754         trace_filter_add_remove_task(pid_list, NULL, task);
 755 }
 756
 757 static void
 758 event_filter_pid_sched_process_fork(void *data,
 759                                     struct task_struct *self,
 760                                     struct task_struct *task)
 761 {
 762         struct trace_pid_list *pid_list;
 763         struct trace_array *tr = data;
 764
 765         pid_list = rcu_dereference_sched(tr->filtered_pids);
 766         trace_filter_add_remove_task(pid_list, self, task);
 767
 768         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 769         trace_filter_add_remove_task(pid_list, self, task);
 770 }
 771
 772 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 773 {
 774         if (enable) {
 775                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 776                                                        tr, INT_MIN);
 777                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 778                                                        tr, INT_MAX);
 779         } else {
 780                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 781                                                     tr);
 782                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 783                                                     tr);
 784         }
 785 }
 786
 787 static void
 788 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 789                                         struct task_struct *prev,
 790                                         struct task_struct *next,
 791                                         unsigned int prev_state)
 792 {
 793         struct trace_array *tr = data;
 794         struct trace_pid_list *no_pid_list;
 795         struct trace_pid_list *pid_list;
 796         bool ret;
 797
 798         pid_list = rcu_dereference_sched(tr->filtered_pids);
 799         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 800
 801         /*
 802          * Sched switch is funny, as we only want to ignore it
 803          * in the notrace case if both prev and next should be ignored.
 804          */
 805         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 806                 trace_ignore_this_task(NULL, no_pid_list, next);
 807
 808         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 809                        (trace_ignore_this_task(pid_list, NULL, prev) &&
 810                         trace_ignore_this_task(pid_list, NULL, next)));
 811 }
 812
 813 static void
 814 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 815                                          struct task_struct *prev,
 816                                          struct task_struct *next,
 817                                          unsigned int prev_state)
 818 {
 819         struct trace_array *tr = data;
 820         struct trace_pid_list *no_pid_list;
 821         struct trace_pid_list *pid_list;
 822
 823         pid_list = rcu_dereference_sched(tr->filtered_pids);
 824         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 825
 826         this_cpu_write(tr->array_buffer.data->ignore_pid,
 827                        trace_ignore_this_task(pid_list, no_pid_list, next));
 828 }
 829
 830 static void
 831 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 832 {
 833         struct trace_array *tr = data;
 834         struct trace_pid_list *no_pid_list;
 835         struct trace_pid_list *pid_list;
 836
 837         /* Nothing to do if we are already tracing */
 838         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 839                 return;
 840
 841         pid_list = rcu_dereference_sched(tr->filtered_pids);
 842         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 843
 844         this_cpu_write(tr->array_buffer.data->ignore_pid,
 845                        trace_ignore_this_task(pid_list, no_pid_list, task));
 846 }
 847
 848 static void
 849 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 850 {
 851         struct trace_array *tr = data;
 852         struct trace_pid_list *no_pid_list;
 853         struct trace_pid_list *pid_list;
 854
 855         /* Nothing to do if we are not tracing */
 856         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 857                 return;
 858
 859         pid_list = rcu_dereference_sched(tr->filtered_pids);
 860         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 861
 862         /* Set tracing if current is enabled */
 863         this_cpu_write(tr->array_buffer.data->ignore_pid,
 864                        trace_ignore_this_task(pid_list, no_pid_list, current));
 865 }
 866
 867 static void unregister_pid_events(struct trace_array *tr)
 868 {
 869         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 870         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 871
 872         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 873         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 874
 875         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 876         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 877
 878         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 879         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 880 }
 881
 882 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 883 {
 884         struct trace_pid_list *pid_list;
 885         struct trace_pid_list *no_pid_list;
 886         struct trace_event_file *file;
 887         int cpu;
 888
 889         pid_list = rcu_dereference_protected(tr->filtered_pids,
 890                                              lockdep_is_held(&event_mutex));
 891         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 892                                              lockdep_is_held(&event_mutex));
 893
 894         /* Make sure there's something to do */
 895         if (!pid_type_enabled(type, pid_list, no_pid_list))
 896                 return;
 897
 898         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 899                 unregister_pid_events(tr);
 900
 901                 list_for_each_entry(file, &tr->events, list) {
 902                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 903                 }
 904
 905                 for_each_possible_cpu(cpu)
 906                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 907         }
 908
 909         if (type & TRACE_PIDS)
 910                 rcu_assign_pointer(tr->filtered_pids, NULL);
 911
 912         if (type & TRACE_NO_PIDS)
 913                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
 914
 915         /* Wait till all users are no longer using pid filtering */
 916         tracepoint_synchronize_unregister();
 917
 918         if ((type & TRACE_PIDS) && pid_list)
 919                 trace_pid_list_free(pid_list);
 920
 921         if ((type & TRACE_NO_PIDS) && no_pid_list)
 922                 trace_pid_list_free(no_pid_list);
 923 }
 924
 925 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 926 {
 927         mutex_lock(&event_mutex);
 928         __ftrace_clear_event_pids(tr, type);
 929         mutex_unlock(&event_mutex);
 930 }
 931
 932 static void __put_system(struct event_subsystem *system)
 933 {
 934         struct event_filter *filter = system->filter;
 935
 936         WARN_ON_ONCE(system_refcount(system) == 0);
 937         if (system_refcount_dec(system))
 938                 return;
 939
 940         list_del(&system->list);
 941
 942         if (filter) {
 943                 kfree(filter->filter_string);
 944                 kfree(filter);
 945         }
 946         kfree_const(system->name);
 947         kfree(system);
 948 }
 949
 950 static void __get_system(struct event_subsystem *system)
 951 {
 952         WARN_ON_ONCE(system_refcount(system) == 0);
 953         system_refcount_inc(system);
 954 }
 955
 956 static void __get_system_dir(struct trace_subsystem_dir *dir)
 957 {
 958         WARN_ON_ONCE(dir->ref_count == 0);
 959         dir->ref_count++;
 960         __get_system(dir->subsystem);
 961 }
 962
 963 static void __put_system_dir(struct trace_subsystem_dir *dir)
 964 {
 965         WARN_ON_ONCE(dir->ref_count == 0);
 966         /* If the subsystem is about to be freed, the dir must be too */
 967         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 968
 969         __put_system(dir->subsystem);
 970         if (!--dir->ref_count)
 971                 kfree(dir);
 972 }
 973
 974 static void put_system(struct trace_subsystem_dir *dir)
 975 {
 976         mutex_lock(&event_mutex);
 977         __put_system_dir(dir);
 978         mutex_unlock(&event_mutex);
 979 }
 980
 981 static void remove_subsystem(struct trace_subsystem_dir *dir)
 982 {
 983         if (!dir)
 984                 return;
 985
 986         if (!--dir->nr_events) {
 987                 eventfs_remove_dir(dir->ei);
 988                 list_del(&dir->list);
 989                 __put_system_dir(dir);
 990         }
 991 }
 992
 993 void event_file_get(struct trace_event_file *file)
 994 {
 995         atomic_inc(&file->ref);
 996 }
 997
 998 void event_file_put(struct trace_event_file *file)
 999 {
1000         if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001                 if (file->flags & EVENT_FILE_FL_FREED)
1002                         kmem_cache_free(file_cachep, file);
1003                 return;
1004         }
1005
1006         if (atomic_dec_and_test(&file->ref)) {
1007                 /* Count should only go to zero when it is freed */
1008                 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1009                         return;
1010                 kmem_cache_free(file_cachep, file);
1011         }
1012 }
1013
1014 static void remove_event_file_dir(struct trace_event_file *file)
1015 {
1016         eventfs_remove_dir(file->ei);
1017         list_del(&file->list);
1018         remove_subsystem(file->system);
1019         free_event_filter(file->filter);
1020         file->flags |= EVENT_FILE_FL_FREED;
1021         event_file_put(file);
1022 }
1023
1024 /*
1025  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1026  */
1027 static int
1028 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029                               const char *sub, const char *event, int set)
1030 {
1031         struct trace_event_file *file;
1032         struct trace_event_call *call;
1033         const char *name;
1034         int ret = -EINVAL;
1035         int eret = 0;
1036
1037         list_for_each_entry(file, &tr->events, list) {
1038
1039                 call = file->event_call;
1040                 name = trace_event_name(call);
1041
1042                 if (!name || !call->class || !call->class->reg)
1043                         continue;
1044
1045                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1046                         continue;
1047
1048                 if (match &&
1049                     strcmp(match, name) != 0 &&
1050                     strcmp(match, call->class->system) != 0)
1051                         continue;
1052
1053                 if (sub && strcmp(sub, call->class->system) != 0)
1054                         continue;
1055
1056                 if (event && strcmp(event, name) != 0)
1057                         continue;
1058
1059                 ret = ftrace_event_enable_disable(file, set);
1060
1061                 /*
1062                  * Save the first error and return that. Some events
1063                  * may still have been enabled, but let the user
1064                  * know that something went wrong.
1065                  */
1066                 if (ret && !eret)
1067                         eret = ret;
1068
1069                 ret = eret;
1070         }
1071
1072         return ret;
1073 }
1074
1075 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076                                   const char *sub, const char *event, int set)
1077 {
1078         int ret;
1079
1080         mutex_lock(&event_mutex);
1081         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082         mutex_unlock(&event_mutex);
1083
1084         return ret;
1085 }
1086
1087 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1088 {
1089         char *event = NULL, *sub = NULL, *match;
1090         int ret;
1091
1092         if (!tr)
1093                 return -ENOENT;
1094         /*
1095          * The buf format can be <subsystem>:<event-name>
1096          *  *:<event-name> means any event by that name.
1097          *  :<event-name> is the same.
1098          *
1099          *  <subsystem>:* means all events in that subsystem
1100          *  <subsystem>: means the same.
1101          *
1102          *  <name> (no ':') means all events in a subsystem with
1103          *  the name <name> or any event that matches <name>
1104          */
1105
1106         match = strsep(&buf, ":");
1107         if (buf) {
1108                 sub = match;
1109                 event = buf;
1110                 match = NULL;
1111
1112                 if (!strlen(sub) || strcmp(sub, "*") == 0)
1113                         sub = NULL;
1114                 if (!strlen(event) || strcmp(event, "*") == 0)
1115                         event = NULL;
1116         }
1117
1118         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1119
1120         /* Put back the colon to allow this to be called again */
1121         if (buf)
1122                 *(buf - 1) = ':';
1123
1124         return ret;
1125 }
1126
1127 /**
1128  * trace_set_clr_event - enable or disable an event
1129  * @system: system name to match (NULL for any system)
1130  * @event: event name to match (NULL for all events, within system)
1131  * @set: 1 to enable, 0 to disable
1132  *
1133  * This is a way for other parts of the kernel to enable or disable
1134  * event recording.
1135  *
1136  * Returns 0 on success, -EINVAL if the parameters do not match any
1137  * registered events.
1138  */
1139 int trace_set_clr_event(const char *system, const char *event, int set)
1140 {
1141         struct trace_array *tr = top_trace_array();
1142
1143         if (!tr)
1144                 return -ENODEV;
1145
1146         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1147 }
1148 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1149
1150 /**
1151  * trace_array_set_clr_event - enable or disable an event for a trace array.
1152  * @tr: concerned trace array.
1153  * @system: system name to match (NULL for any system)
1154  * @event: event name to match (NULL for all events, within system)
1155  * @enable: true to enable, false to disable
1156  *
1157  * This is a way for other parts of the kernel to enable or disable
1158  * event recording.
1159  *
1160  * Returns 0 on success, -EINVAL if the parameters do not match any
1161  * registered events.
1162  */
1163 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164                 const char *event, bool enable)
1165 {
1166         int set;
1167
1168         if (!tr)
1169                 return -ENOENT;
1170
1171         set = (enable == true) ? 1 : 0;
1172         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1173 }
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1175
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE          127
1178
1179 static ssize_t
1180 ftrace_event_write(struct file *file, const char __user *ubuf,
1181                    size_t cnt, loff_t *ppos)
1182 {
1183         struct trace_parser parser;
1184         struct seq_file *m = file->private_data;
1185         struct trace_array *tr = m->private;
1186         ssize_t read, ret;
1187
1188         if (!cnt)
1189                 return 0;
1190
1191         ret = tracing_update_buffers(tr);
1192         if (ret < 0)
1193                 return ret;
1194
1195         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1196                 return -ENOMEM;
1197
1198         read = trace_get_user(&parser, ubuf, cnt, ppos);
1199
1200         if (read >= 0 && trace_parser_loaded((&parser))) {
1201                 int set = 1;
1202
1203                 if (*parser.buffer == '!')
1204                         set = 0;
1205
1206                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1207                 if (ret)
1208                         goto out_put;
1209         }
1210
1211         ret = read;
1212
1213  out_put:
1214         trace_parser_put(&parser);
1215
1216         return ret;
1217 }
1218
1219 static void *
1220 t_next(struct seq_file *m, void *v, loff_t *pos)
1221 {
1222         struct trace_event_file *file = v;
1223         struct trace_event_call *call;
1224         struct trace_array *tr = m->private;
1225
1226         (*pos)++;
1227
1228         list_for_each_entry_continue(file, &tr->events, list) {
1229                 call = file->event_call;
1230                 /*
1231                  * The ftrace subsystem is for showing formats only.
1232                  * They can not be enabled or disabled via the event files.
1233                  */
1234                 if (call->class && call->class->reg &&
1235                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1236                         return file;
1237         }
1238
1239         return NULL;
1240 }
1241
1242 static void *t_start(struct seq_file *m, loff_t *pos)
1243 {
1244         struct trace_event_file *file;
1245         struct trace_array *tr = m->private;
1246         loff_t l;
1247
1248         mutex_lock(&event_mutex);
1249
1250         file = list_entry(&tr->events, struct trace_event_file, list);
1251         for (l = 0; l <= *pos; ) {
1252                 file = t_next(m, file, &l);
1253                 if (!file)
1254                         break;
1255         }
1256         return file;
1257 }
1258
1259 static void *
1260 s_next(struct seq_file *m, void *v, loff_t *pos)
1261 {
1262         struct trace_event_file *file = v;
1263         struct trace_array *tr = m->private;
1264
1265         (*pos)++;
1266
1267         list_for_each_entry_continue(file, &tr->events, list) {
1268                 if (file->flags & EVENT_FILE_FL_ENABLED)
1269                         return file;
1270         }
1271
1272         return NULL;
1273 }
1274
1275 static void *s_start(struct seq_file *m, loff_t *pos)
1276 {
1277         struct trace_event_file *file;
1278         struct trace_array *tr = m->private;
1279         loff_t l;
1280
1281         mutex_lock(&event_mutex);
1282
1283         file = list_entry(&tr->events, struct trace_event_file, list);
1284         for (l = 0; l <= *pos; ) {
1285                 file = s_next(m, file, &l);
1286                 if (!file)
1287                         break;
1288         }
1289         return file;
1290 }
1291
1292 static int t_show(struct seq_file *m, void *v)
1293 {
1294         struct trace_event_file *file = v;
1295         struct trace_event_call *call = file->event_call;
1296
1297         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298                 seq_printf(m, "%s:", call->class->system);
1299         seq_printf(m, "%s\n", trace_event_name(call));
1300
1301         return 0;
1302 }
1303
1304 static void t_stop(struct seq_file *m, void *p)
1305 {
1306         mutex_unlock(&event_mutex);
1307 }
1308
1309 static void *
1310 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1311 {
1312         struct trace_array *tr = m->private;
1313         struct trace_pid_list *pid_list;
1314
1315         if (type == TRACE_PIDS)
1316                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1317         else
1318                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1319
1320         return trace_pid_next(pid_list, v, pos);
1321 }
1322
1323 static void *
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326         return __next(m, v, pos, TRACE_PIDS);
1327 }
1328
1329 static void *
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1331 {
1332         return __next(m, v, pos, TRACE_NO_PIDS);
1333 }
1334
1335 static void *__start(struct seq_file *m, loff_t *pos, int type)
1336         __acquires(RCU)
1337 {
1338         struct trace_pid_list *pid_list;
1339         struct trace_array *tr = m->private;
1340
1341         /*
1342          * Grab the mutex, to keep calls to p_next() having the same
1343          * tr->filtered_pids as p_start() has.
1344          * If we just passed the tr->filtered_pids around, then RCU would
1345          * have been enough, but doing that makes things more complex.
1346          */
1347         mutex_lock(&event_mutex);
1348         rcu_read_lock_sched();
1349
1350         if (type == TRACE_PIDS)
1351                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1352         else
1353                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1354
1355         if (!pid_list)
1356                 return NULL;
1357
1358         return trace_pid_start(pid_list, pos);
1359 }
1360
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1362         __acquires(RCU)
1363 {
1364         return __start(m, pos, TRACE_PIDS);
1365 }
1366
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1368         __acquires(RCU)
1369 {
1370         return __start(m, pos, TRACE_NO_PIDS);
1371 }
1372
1373 static void p_stop(struct seq_file *m, void *p)
1374         __releases(RCU)
1375 {
1376         rcu_read_unlock_sched();
1377         mutex_unlock(&event_mutex);
1378 }
1379
1380 static ssize_t
1381 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1382                   loff_t *ppos)
1383 {
1384         struct trace_event_file *file;
1385         unsigned long flags;
1386         char buf[4] = "0";
1387
1388         mutex_lock(&event_mutex);
1389         file = event_file_data(filp);
1390         if (likely(file))
1391                 flags = file->flags;
1392         mutex_unlock(&event_mutex);
1393
1394         if (!file || flags & EVENT_FILE_FL_FREED)
1395                 return -ENODEV;
1396
1397         if (flags & EVENT_FILE_FL_ENABLED &&
1398             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1399                 strcpy(buf, "1");
1400
1401         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402             flags & EVENT_FILE_FL_SOFT_MODE)
1403                 strcat(buf, "*");
1404
1405         strcat(buf, "\n");
1406
1407         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1408 }
1409
1410 static ssize_t
1411 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1412                    loff_t *ppos)
1413 {
1414         struct trace_event_file *file;
1415         unsigned long val;
1416         int ret;
1417
1418         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1419         if (ret)
1420                 return ret;
1421
1422         switch (val) {
1423         case 0:
1424         case 1:
1425                 ret = -ENODEV;
1426                 mutex_lock(&event_mutex);
1427                 file = event_file_data(filp);
1428                 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429                         ret = tracing_update_buffers(file->tr);
1430                         if (ret < 0) {
1431                                 mutex_unlock(&event_mutex);
1432                                 return ret;
1433                         }
1434                         ret = ftrace_event_enable_disable(file, val);
1435                 }
1436                 mutex_unlock(&event_mutex);
1437                 break;
1438
1439         default:
1440                 return -EINVAL;
1441         }
1442
1443         *ppos += cnt;
1444
1445         return ret ? ret : cnt;
1446 }
1447
1448 static ssize_t
1449 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450                    loff_t *ppos)
1451 {
1452         const char set_to_char[4] = { '?', '0', '1', 'X' };
1453         struct trace_subsystem_dir *dir = filp->private_data;
1454         struct event_subsystem *system = dir->subsystem;
1455         struct trace_event_call *call;
1456         struct trace_event_file *file;
1457         struct trace_array *tr = dir->tr;
1458         char buf[2];
1459         int set = 0;
1460         int ret;
1461
1462         mutex_lock(&event_mutex);
1463         list_for_each_entry(file, &tr->events, list) {
1464                 call = file->event_call;
1465                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466                     !trace_event_name(call) || !call->class || !call->class->reg)
1467                         continue;
1468
1469                 if (system && strcmp(call->class->system, system->name) != 0)
1470                         continue;
1471
1472                 /*
1473                  * We need to find out if all the events are set
1474                  * or if all events or cleared, or if we have
1475                  * a mixture.
1476                  */
1477                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478
1479                 /*
1480                  * If we have a mixture, no need to look further.
1481                  */
1482                 if (set == 3)
1483                         break;
1484         }
1485         mutex_unlock(&event_mutex);
1486
1487         buf[0] = set_to_char[set];
1488         buf[1] = '\n';
1489
1490         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1491
1492         return ret;
1493 }
1494
1495 static ssize_t
1496 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497                     loff_t *ppos)
1498 {
1499         struct trace_subsystem_dir *dir = filp->private_data;
1500         struct event_subsystem *system = dir->subsystem;
1501         const char *name = NULL;
1502         unsigned long val;
1503         ssize_t ret;
1504
1505         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1506         if (ret)
1507                 return ret;
1508
1509         ret = tracing_update_buffers(dir->tr);
1510         if (ret < 0)
1511                 return ret;
1512
1513         if (val != 0 && val != 1)
1514                 return -EINVAL;
1515
1516         /*
1517          * Opening of "enable" adds a ref count to system,
1518          * so the name is safe to use.
1519          */
1520         if (system)
1521                 name = system->name;
1522
1523         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1524         if (ret)
1525                 goto out;
1526
1527         ret = cnt;
1528
1529 out:
1530         *ppos += cnt;
1531
1532         return ret;
1533 }
1534
1535 enum {
1536         FORMAT_HEADER           = 1,
1537         FORMAT_FIELD_SEPERATOR  = 2,
1538         FORMAT_PRINTFMT         = 3,
1539 };
1540
1541 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1542 {
1543         struct trace_event_call *call = event_file_data(m->private);
1544         struct list_head *common_head = &ftrace_common_fields;
1545         struct list_head *head = trace_get_fields(call);
1546         struct list_head *node = v;
1547
1548         (*pos)++;
1549
1550         switch ((unsigned long)v) {
1551         case FORMAT_HEADER:
1552                 node = common_head;
1553                 break;
1554
1555         case FORMAT_FIELD_SEPERATOR:
1556                 node = head;
1557                 break;
1558
1559         case FORMAT_PRINTFMT:
1560                 /* all done */
1561                 return NULL;
1562         }
1563
1564         node = node->prev;
1565         if (node == common_head)
1566                 return (void *)FORMAT_FIELD_SEPERATOR;
1567         else if (node == head)
1568                 return (void *)FORMAT_PRINTFMT;
1569         else
1570                 return node;
1571 }
1572
1573 static int f_show(struct seq_file *m, void *v)
1574 {
1575         struct trace_event_call *call = event_file_data(m->private);
1576         struct ftrace_event_field *field;
1577         const char *array_descriptor;
1578
1579         switch ((unsigned long)v) {
1580         case FORMAT_HEADER:
1581                 seq_printf(m, "name: %s\n", trace_event_name(call));
1582                 seq_printf(m, "ID: %d\n", call->event.type);
1583                 seq_puts(m, "format:\n");
1584                 return 0;
1585
1586         case FORMAT_FIELD_SEPERATOR:
1587                 seq_putc(m, '\n');
1588                 return 0;
1589
1590         case FORMAT_PRINTFMT:
1591                 seq_printf(m, "\nprint fmt: %s\n",
1592                            call->print_fmt);
1593                 return 0;
1594         }
1595
1596         field = list_entry(v, struct ftrace_event_field, link);
1597         /*
1598          * Smartly shows the array type(except dynamic array).
1599          * Normal:
1600          *      field:TYPE VAR
1601          * If TYPE := TYPE[LEN], it is shown:
1602          *      field:TYPE VAR[LEN]
1603          */
1604         array_descriptor = strchr(field->type, '[');
1605
1606         if (str_has_prefix(field->type, "__data_loc"))
1607                 array_descriptor = NULL;
1608
1609         if (!array_descriptor)
1610                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611                            field->type, field->name, field->offset,
1612                            field->size, !!field->is_signed);
1613         else if (field->len)
1614                 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615                            (int)(array_descriptor - field->type),
1616                            field->type, field->name,
1617                            field->len, field->offset,
1618                            field->size, !!field->is_signed);
1619         else
1620                 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621                                 (int)(array_descriptor - field->type),
1622                                 field->type, field->name,
1623                                 field->offset, field->size, !!field->is_signed);
1624
1625         return 0;
1626 }
1627
1628 static void *f_start(struct seq_file *m, loff_t *pos)
1629 {
1630         void *p = (void *)FORMAT_HEADER;
1631         loff_t l = 0;
1632
1633         /* ->stop() is called even if ->start() fails */
1634         mutex_lock(&event_mutex);
1635         if (!event_file_data(m->private))
1636                 return ERR_PTR(-ENODEV);
1637
1638         while (l < *pos && p)
1639                 p = f_next(m, p, &l);
1640
1641         return p;
1642 }
1643
1644 static void f_stop(struct seq_file *m, void *p)
1645 {
1646         mutex_unlock(&event_mutex);
1647 }
1648
1649 static const struct seq_operations trace_format_seq_ops = {
1650         .start          = f_start,
1651         .next           = f_next,
1652         .stop           = f_stop,
1653         .show           = f_show,
1654 };
1655
1656 static int trace_format_open(struct inode *inode, struct file *file)
1657 {
1658         struct seq_file *m;
1659         int ret;
1660
1661         /* Do we want to hide event format files on tracefs lockdown? */
1662
1663         ret = seq_open(file, &trace_format_seq_ops);
1664         if (ret < 0)
1665                 return ret;
1666
1667         m = file->private_data;
1668         m->private = file;
1669
1670         return 0;
1671 }
1672
1673 #ifdef CONFIG_PERF_EVENTS
1674 static ssize_t
1675 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1676 {
1677         int id = (long)event_file_data(filp);
1678         char buf[32];
1679         int len;
1680
1681         if (unlikely(!id))
1682                 return -ENODEV;
1683
1684         len = sprintf(buf, "%d\n", id);
1685
1686         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687 }
1688 #endif
1689
1690 static ssize_t
1691 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1692                   loff_t *ppos)
1693 {
1694         struct trace_event_file *file;
1695         struct trace_seq *s;
1696         int r = -ENODEV;
1697
1698         if (*ppos)
1699                 return 0;
1700
1701         s = kmalloc(sizeof(*s), GFP_KERNEL);
1702
1703         if (!s)
1704                 return -ENOMEM;
1705
1706         trace_seq_init(s);
1707
1708         mutex_lock(&event_mutex);
1709         file = event_file_data(filp);
1710         if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711                 print_event_filter(file, s);
1712         mutex_unlock(&event_mutex);
1713
1714         if (file)
1715                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1716                                             s->buffer, trace_seq_used(s));
1717
1718         kfree(s);
1719
1720         return r;
1721 }
1722
1723 static ssize_t
1724 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1725                    loff_t *ppos)
1726 {
1727         struct trace_event_file *file;
1728         char *buf;
1729         int err = -ENODEV;
1730
1731         if (cnt >= PAGE_SIZE)
1732                 return -EINVAL;
1733
1734         buf = memdup_user_nul(ubuf, cnt);
1735         if (IS_ERR(buf))
1736                 return PTR_ERR(buf);
1737
1738         mutex_lock(&event_mutex);
1739         file = event_file_data(filp);
1740         if (file)
1741                 err = apply_event_filter(file, buf);
1742         mutex_unlock(&event_mutex);
1743
1744         kfree(buf);
1745         if (err < 0)
1746                 return err;
1747
1748         *ppos += cnt;
1749
1750         return cnt;
1751 }
1752
1753 static LIST_HEAD(event_subsystems);
1754
1755 static int subsystem_open(struct inode *inode, struct file *filp)
1756 {
1757         struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758         struct trace_array *tr = NULL, *iter_tr;
1759         struct event_subsystem *system = NULL;
1760         int ret;
1761
1762         if (tracing_is_disabled())
1763                 return -ENODEV;
1764
1765         /* Make sure the system still exists */
1766         mutex_lock(&event_mutex);
1767         mutex_lock(&trace_types_lock);
1768         list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769                 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770                         if (iter_dir == inode->i_private) {
1771                                 /* Don't open systems with no events */
1772                                 tr = iter_tr;
1773                                 dir = iter_dir;
1774                                 if (dir->nr_events) {
1775                                         __get_system_dir(dir);
1776                                         system = dir->subsystem;
1777                                 }
1778                                 goto exit_loop;
1779                         }
1780                 }
1781         }
1782  exit_loop:
1783         mutex_unlock(&trace_types_lock);
1784         mutex_unlock(&event_mutex);
1785
1786         if (!system)
1787                 return -ENODEV;
1788
1789         /* Still need to increment the ref count of the system */
1790         if (trace_array_get(tr) < 0) {
1791                 put_system(dir);
1792                 return -ENODEV;
1793         }
1794
1795         ret = tracing_open_generic(inode, filp);
1796         if (ret < 0) {
1797                 trace_array_put(tr);
1798                 put_system(dir);
1799         }
1800
1801         return ret;
1802 }
1803
1804 static int system_tr_open(struct inode *inode, struct file *filp)
1805 {
1806         struct trace_subsystem_dir *dir;
1807         struct trace_array *tr = inode->i_private;
1808         int ret;
1809
1810         /* Make a temporary dir that has no system but points to tr */
1811         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1812         if (!dir)
1813                 return -ENOMEM;
1814
1815         ret = tracing_open_generic_tr(inode, filp);
1816         if (ret < 0) {
1817                 kfree(dir);
1818                 return ret;
1819         }
1820         dir->tr = tr;
1821         filp->private_data = dir;
1822
1823         return 0;
1824 }
1825
1826 static int subsystem_release(struct inode *inode, struct file *file)
1827 {
1828         struct trace_subsystem_dir *dir = file->private_data;
1829
1830         trace_array_put(dir->tr);
1831
1832         /*
1833          * If dir->subsystem is NULL, then this is a temporary
1834          * descriptor that was made for a trace_array to enable
1835          * all subsystems.
1836          */
1837         if (dir->subsystem)
1838                 put_system(dir);
1839         else
1840                 kfree(dir);
1841
1842         return 0;
1843 }
1844
1845 static ssize_t
1846 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1847                       loff_t *ppos)
1848 {
1849         struct trace_subsystem_dir *dir = filp->private_data;
1850         struct event_subsystem *system = dir->subsystem;
1851         struct trace_seq *s;
1852         int r;
1853
1854         if (*ppos)
1855                 return 0;
1856
1857         s = kmalloc(sizeof(*s), GFP_KERNEL);
1858         if (!s)
1859                 return -ENOMEM;
1860
1861         trace_seq_init(s);
1862
1863         print_subsystem_event_filter(system, s);
1864         r = simple_read_from_buffer(ubuf, cnt, ppos,
1865                                     s->buffer, trace_seq_used(s));
1866
1867         kfree(s);
1868
1869         return r;
1870 }
1871
1872 static ssize_t
1873 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1874                        loff_t *ppos)
1875 {
1876         struct trace_subsystem_dir *dir = filp->private_data;
1877         char *buf;
1878         int err;
1879
1880         if (cnt >= PAGE_SIZE)
1881                 return -EINVAL;
1882
1883         buf = memdup_user_nul(ubuf, cnt);
1884         if (IS_ERR(buf))
1885                 return PTR_ERR(buf);
1886
1887         err = apply_subsystem_event_filter(dir, buf);
1888         kfree(buf);
1889         if (err < 0)
1890                 return err;
1891
1892         *ppos += cnt;
1893
1894         return cnt;
1895 }
1896
1897 static ssize_t
1898 show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1899 {
1900         struct trace_array *tr = filp->private_data;
1901         struct trace_seq *s;
1902         int r;
1903
1904         if (*ppos)
1905                 return 0;
1906
1907         s = kmalloc(sizeof(*s), GFP_KERNEL);
1908         if (!s)
1909                 return -ENOMEM;
1910
1911         trace_seq_init(s);
1912
1913         ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914         r = simple_read_from_buffer(ubuf, cnt, ppos,
1915                                     s->buffer, trace_seq_used(s));
1916
1917         kfree(s);
1918
1919         return r;
1920 }
1921
1922 static ssize_t
1923 show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1924 {
1925         struct trace_seq *s;
1926         int r;
1927
1928         if (*ppos)
1929                 return 0;
1930
1931         s = kmalloc(sizeof(*s), GFP_KERNEL);
1932         if (!s)
1933                 return -ENOMEM;
1934
1935         trace_seq_init(s);
1936
1937         ring_buffer_print_entry_header(s);
1938         r = simple_read_from_buffer(ubuf, cnt, ppos,
1939                                     s->buffer, trace_seq_used(s));
1940
1941         kfree(s);
1942
1943         return r;
1944 }
1945
1946 static void ignore_task_cpu(void *data)
1947 {
1948         struct trace_array *tr = data;
1949         struct trace_pid_list *pid_list;
1950         struct trace_pid_list *no_pid_list;
1951
1952         /*
1953          * This function is called by on_each_cpu() while the
1954          * event_mutex is held.
1955          */
1956         pid_list = rcu_dereference_protected(tr->filtered_pids,
1957                                              mutex_is_locked(&event_mutex));
1958         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959                                              mutex_is_locked(&event_mutex));
1960
1961         this_cpu_write(tr->array_buffer.data->ignore_pid,
1962                        trace_ignore_this_task(pid_list, no_pid_list, current));
1963 }
1964
1965 static void register_pid_events(struct trace_array *tr)
1966 {
1967         /*
1968          * Register a probe that is called before all other probes
1969          * to set ignore_pid if next or prev do not match.
1970          * Register a probe this is called after all other probes
1971          * to only keep ignore_pid set if next pid matches.
1972          */
1973         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1974                                          tr, INT_MAX);
1975         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1976                                          tr, 0);
1977
1978         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1979                                          tr, INT_MAX);
1980         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1981                                          tr, 0);
1982
1983         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1984                                              tr, INT_MAX);
1985         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1986                                              tr, 0);
1987
1988         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1989                                          tr, INT_MAX);
1990         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1991                                          tr, 0);
1992 }
1993
1994 static ssize_t
1995 event_pid_write(struct file *filp, const char __user *ubuf,
1996                 size_t cnt, loff_t *ppos, int type)
1997 {
1998         struct seq_file *m = filp->private_data;
1999         struct trace_array *tr = m->private;
2000         struct trace_pid_list *filtered_pids = NULL;
2001         struct trace_pid_list *other_pids = NULL;
2002         struct trace_pid_list *pid_list;
2003         struct trace_event_file *file;
2004         ssize_t ret;
2005
2006         if (!cnt)
2007                 return 0;
2008
2009         ret = tracing_update_buffers(tr);
2010         if (ret < 0)
2011                 return ret;
2012
2013         mutex_lock(&event_mutex);
2014
2015         if (type == TRACE_PIDS) {
2016                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017                                                           lockdep_is_held(&event_mutex));
2018                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019                                                           lockdep_is_held(&event_mutex));
2020         } else {
2021                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022                                                           lockdep_is_held(&event_mutex));
2023                 other_pids = rcu_dereference_protected(tr->filtered_pids,
2024                                                           lockdep_is_held(&event_mutex));
2025         }
2026
2027         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2028         if (ret < 0)
2029                 goto out;
2030
2031         if (type == TRACE_PIDS)
2032                 rcu_assign_pointer(tr->filtered_pids, pid_list);
2033         else
2034                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2035
2036         list_for_each_entry(file, &tr->events, list) {
2037                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2038         }
2039
2040         if (filtered_pids) {
2041                 tracepoint_synchronize_unregister();
2042                 trace_pid_list_free(filtered_pids);
2043         } else if (pid_list && !other_pids) {
2044                 register_pid_events(tr);
2045         }
2046
2047         /*
2048          * Ignoring of pids is done at task switch. But we have to
2049          * check for those tasks that are currently running.
2050          * Always do this in case a pid was appended or removed.
2051          */
2052         on_each_cpu(ignore_task_cpu, tr, 1);
2053
2054  out:
2055         mutex_unlock(&event_mutex);
2056
2057         if (ret > 0)
2058                 *ppos += ret;
2059
2060         return ret;
2061 }
2062
2063 static ssize_t
2064 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065                        size_t cnt, loff_t *ppos)
2066 {
2067         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2068 }
2069
2070 static ssize_t
2071 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072                         size_t cnt, loff_t *ppos)
2073 {
2074         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2075 }
2076
2077 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081 static int ftrace_event_release(struct inode *inode, struct file *file);
2082
2083 static const struct seq_operations show_event_seq_ops = {
2084         .start = t_start,
2085         .next = t_next,
2086         .show = t_show,
2087         .stop = t_stop,
2088 };
2089
2090 static const struct seq_operations show_set_event_seq_ops = {
2091         .start = s_start,
2092         .next = s_next,
2093         .show = t_show,
2094         .stop = t_stop,
2095 };
2096
2097 static const struct seq_operations show_set_pid_seq_ops = {
2098         .start = p_start,
2099         .next = p_next,
2100         .show = trace_pid_show,
2101         .stop = p_stop,
2102 };
2103
2104 static const struct seq_operations show_set_no_pid_seq_ops = {
2105         .start = np_start,
2106         .next = np_next,
2107         .show = trace_pid_show,
2108         .stop = p_stop,
2109 };
2110
2111 static const struct file_operations ftrace_avail_fops = {
2112         .open = ftrace_event_avail_open,
2113         .read = seq_read,
2114         .llseek = seq_lseek,
2115         .release = seq_release,
2116 };
2117
2118 static const struct file_operations ftrace_set_event_fops = {
2119         .open = ftrace_event_set_open,
2120         .read = seq_read,
2121         .write = ftrace_event_write,
2122         .llseek = seq_lseek,
2123         .release = ftrace_event_release,
2124 };
2125
2126 static const struct file_operations ftrace_set_event_pid_fops = {
2127         .open = ftrace_event_set_pid_open,
2128         .read = seq_read,
2129         .write = ftrace_event_pid_write,
2130         .llseek = seq_lseek,
2131         .release = ftrace_event_release,
2132 };
2133
2134 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135         .open = ftrace_event_set_npid_open,
2136         .read = seq_read,
2137         .write = ftrace_event_npid_write,
2138         .llseek = seq_lseek,
2139         .release = ftrace_event_release,
2140 };
2141
2142 static const struct file_operations ftrace_enable_fops = {
2143         .open = tracing_open_file_tr,
2144         .read = event_enable_read,
2145         .write = event_enable_write,
2146         .release = tracing_release_file_tr,
2147         .llseek = default_llseek,
2148 };
2149
2150 static const struct file_operations ftrace_event_format_fops = {
2151         .open = trace_format_open,
2152         .read = seq_read,
2153         .llseek = seq_lseek,
2154         .release = seq_release,
2155 };
2156
2157 #ifdef CONFIG_PERF_EVENTS
2158 static const struct file_operations ftrace_event_id_fops = {
2159         .read = event_id_read,
2160         .llseek = default_llseek,
2161 };
2162 #endif
2163
2164 static const struct file_operations ftrace_event_filter_fops = {
2165         .open = tracing_open_file_tr,
2166         .read = event_filter_read,
2167         .write = event_filter_write,
2168         .release = tracing_release_file_tr,
2169         .llseek = default_llseek,
2170 };
2171
2172 static const struct file_operations ftrace_subsystem_filter_fops = {
2173         .open = subsystem_open,
2174         .read = subsystem_filter_read,
2175         .write = subsystem_filter_write,
2176         .llseek = default_llseek,
2177         .release = subsystem_release,
2178 };
2179
2180 static const struct file_operations ftrace_system_enable_fops = {
2181         .open = subsystem_open,
2182         .read = system_enable_read,
2183         .write = system_enable_write,
2184         .llseek = default_llseek,
2185         .release = subsystem_release,
2186 };
2187
2188 static const struct file_operations ftrace_tr_enable_fops = {
2189         .open = system_tr_open,
2190         .read = system_enable_read,
2191         .write = system_enable_write,
2192         .llseek = default_llseek,
2193         .release = subsystem_release,
2194 };
2195
2196 static const struct file_operations ftrace_show_header_page_fops = {
2197         .open = tracing_open_generic_tr,
2198         .read = show_header_page_file,
2199         .llseek = default_llseek,
2200         .release = tracing_release_generic_tr,
2201 };
2202
2203 static const struct file_operations ftrace_show_header_event_fops = {
2204         .open = tracing_open_generic_tr,
2205         .read = show_header_event_file,
2206         .llseek = default_llseek,
2207         .release = tracing_release_generic_tr,
2208 };
2209
2210 static int
2211 ftrace_event_open(struct inode *inode, struct file *file,
2212                   const struct seq_operations *seq_ops)
2213 {
2214         struct seq_file *m;
2215         int ret;
2216
2217         ret = security_locked_down(LOCKDOWN_TRACEFS);
2218         if (ret)
2219                 return ret;
2220
2221         ret = seq_open(file, seq_ops);
2222         if (ret < 0)
2223                 return ret;
2224         m = file->private_data;
2225         /* copy tr over to seq ops */
2226         m->private = inode->i_private;
2227
2228         return ret;
2229 }
2230
2231 static int ftrace_event_release(struct inode *inode, struct file *file)
2232 {
2233         struct trace_array *tr = inode->i_private;
2234
2235         trace_array_put(tr);
2236
2237         return seq_release(inode, file);
2238 }
2239
2240 static int
2241 ftrace_event_avail_open(struct inode *inode, struct file *file)
2242 {
2243         const struct seq_operations *seq_ops = &show_event_seq_ops;
2244
2245         /* Checks for tracefs lockdown */
2246         return ftrace_event_open(inode, file, seq_ops);
2247 }
2248
2249 static int
2250 ftrace_event_set_open(struct inode *inode, struct file *file)
2251 {
2252         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253         struct trace_array *tr = inode->i_private;
2254         int ret;
2255
2256         ret = tracing_check_open_get_tr(tr);
2257         if (ret)
2258                 return ret;
2259
2260         if ((file->f_mode & FMODE_WRITE) &&
2261             (file->f_flags & O_TRUNC))
2262                 ftrace_clear_events(tr);
2263
2264         ret = ftrace_event_open(inode, file, seq_ops);
2265         if (ret < 0)
2266                 trace_array_put(tr);
2267         return ret;
2268 }
2269
2270 static int
2271 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2272 {
2273         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274         struct trace_array *tr = inode->i_private;
2275         int ret;
2276
2277         ret = tracing_check_open_get_tr(tr);
2278         if (ret)
2279                 return ret;
2280
2281         if ((file->f_mode & FMODE_WRITE) &&
2282             (file->f_flags & O_TRUNC))
2283                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2284
2285         ret = ftrace_event_open(inode, file, seq_ops);
2286         if (ret < 0)
2287                 trace_array_put(tr);
2288         return ret;
2289 }
2290
2291 static int
2292 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2293 {
2294         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295         struct trace_array *tr = inode->i_private;
2296         int ret;
2297
2298         ret = tracing_check_open_get_tr(tr);
2299         if (ret)
2300                 return ret;
2301
2302         if ((file->f_mode & FMODE_WRITE) &&
2303             (file->f_flags & O_TRUNC))
2304                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2305
2306         ret = ftrace_event_open(inode, file, seq_ops);
2307         if (ret < 0)
2308                 trace_array_put(tr);
2309         return ret;
2310 }
2311
2312 static struct event_subsystem *
2313 create_new_subsystem(const char *name)
2314 {
2315         struct event_subsystem *system;
2316
2317         /* need to create new entry */
2318         system = kmalloc(sizeof(*system), GFP_KERNEL);
2319         if (!system)
2320                 return NULL;
2321
2322         system->ref_count = 1;
2323
2324         /* Only allocate if dynamic (kprobes and modules) */
2325         system->name = kstrdup_const(name, GFP_KERNEL);
2326         if (!system->name)
2327                 goto out_free;
2328
2329         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330         if (!system->filter)
2331                 goto out_free;
2332
2333         list_add(&system->list, &event_subsystems);
2334
2335         return system;
2336
2337  out_free:
2338         kfree_const(system->name);
2339         kfree(system);
2340         return NULL;
2341 }
2342
2343 static int system_callback(const char *name, umode_t *mode, void **data,
2344                     const struct file_operations **fops)
2345 {
2346         if (strcmp(name, "filter") == 0)
2347                 *fops = &ftrace_subsystem_filter_fops;
2348
2349         else if (strcmp(name, "enable") == 0)
2350                 *fops = &ftrace_system_enable_fops;
2351
2352         else
2353                 return 0;
2354
2355         *mode = TRACE_MODE_WRITE;
2356         return 1;
2357 }
2358
2359 static struct eventfs_inode *
2360 event_subsystem_dir(struct trace_array *tr, const char *name,
2361                     struct trace_event_file *file, struct eventfs_inode *parent)
2362 {
2363         struct event_subsystem *system, *iter;
2364         struct trace_subsystem_dir *dir;
2365         struct eventfs_inode *ei;
2366         int nr_entries;
2367         static struct eventfs_entry system_entries[] = {
2368                 {
2369                         .name           = "filter",
2370                         .callback       = system_callback,
2371                 },
2372                 {
2373                         .name           = "enable",
2374                         .callback       = system_callback,
2375                 }
2376         };
2377
2378         /* First see if we did not already create this dir */
2379         list_for_each_entry(dir, &tr->systems, list) {
2380                 system = dir->subsystem;
2381                 if (strcmp(system->name, name) == 0) {
2382                         dir->nr_events++;
2383                         file->system = dir;
2384                         return dir->ei;
2385                 }
2386         }
2387
2388         /* Now see if the system itself exists. */
2389         system = NULL;
2390         list_for_each_entry(iter, &event_subsystems, list) {
2391                 if (strcmp(iter->name, name) == 0) {
2392                         system = iter;
2393                         break;
2394                 }
2395         }
2396
2397         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2398         if (!dir)
2399                 goto out_fail;
2400
2401         if (!system) {
2402                 system = create_new_subsystem(name);
2403                 if (!system)
2404                         goto out_free;
2405         } else
2406                 __get_system(system);
2407
2408         /* ftrace only has directories no files */
2409         if (strcmp(name, "ftrace") == 0)
2410                 nr_entries = 0;
2411         else
2412                 nr_entries = ARRAY_SIZE(system_entries);
2413
2414         ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2415         if (IS_ERR(ei)) {
2416                 pr_warn("Failed to create system directory %s\n", name);
2417                 __put_system(system);
2418                 goto out_free;
2419         }
2420
2421         dir->ei = ei;
2422         dir->tr = tr;
2423         dir->ref_count = 1;
2424         dir->nr_events = 1;
2425         dir->subsystem = system;
2426         file->system = dir;
2427
2428         list_add(&dir->list, &tr->systems);
2429
2430         return dir->ei;
2431
2432  out_free:
2433         kfree(dir);
2434  out_fail:
2435         /* Only print this message if failed on memory allocation */
2436         if (!dir || !system)
2437                 pr_warn("No memory to create event subsystem %s\n", name);
2438         return NULL;
2439 }
2440
2441 static int
2442 event_define_fields(struct trace_event_call *call)
2443 {
2444         struct list_head *head;
2445         int ret = 0;
2446
2447         /*
2448          * Other events may have the same class. Only update
2449          * the fields if they are not already defined.
2450          */
2451         head = trace_get_fields(call);
2452         if (list_empty(head)) {
2453                 struct trace_event_fields *field = call->class->fields_array;
2454                 unsigned int offset = sizeof(struct trace_entry);
2455
2456                 for (; field->type; field++) {
2457                         if (field->type == TRACE_FUNCTION_TYPE) {
2458                                 field->define_fields(call);
2459                                 break;
2460                         }
2461
2462                         offset = ALIGN(offset, field->align);
2463                         ret = trace_define_field_ext(call, field->type, field->name,
2464                                                  offset, field->size,
2465                                                  field->is_signed, field->filter_type,
2466                                                  field->len);
2467                         if (WARN_ON_ONCE(ret)) {
2468                                 pr_err("error code is %d\n", ret);
2469                                 break;
2470                         }
2471
2472                         offset += field->size;
2473                 }
2474         }
2475
2476         return ret;
2477 }
2478
2479 static int event_callback(const char *name, umode_t *mode, void **data,
2480                           const struct file_operations **fops)
2481 {
2482         struct trace_event_file *file = *data;
2483         struct trace_event_call *call = file->event_call;
2484
2485         if (strcmp(name, "format") == 0) {
2486                 *mode = TRACE_MODE_READ;
2487                 *fops = &ftrace_event_format_fops;
2488                 *data = call;
2489                 return 1;
2490         }
2491
2492         /*
2493          * Only event directories that can be enabled should have
2494          * triggers or filters, with the exception of the "print"
2495          * event that can have a "trigger" file.
2496          */
2497         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498                 if (call->class->reg && strcmp(name, "enable") == 0) {
2499                         *mode = TRACE_MODE_WRITE;
2500                         *fops = &ftrace_enable_fops;
2501                         return 1;
2502                 }
2503
2504                 if (strcmp(name, "filter") == 0) {
2505                         *mode = TRACE_MODE_WRITE;
2506                         *fops = &ftrace_event_filter_fops;
2507                         return 1;
2508                 }
2509         }
2510
2511         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512             strcmp(trace_event_name(call), "print") == 0) {
2513                 if (strcmp(name, "trigger") == 0) {
2514                         *mode = TRACE_MODE_WRITE;
2515                         *fops = &event_trigger_fops;
2516                         return 1;
2517                 }
2518         }
2519
2520 #ifdef CONFIG_PERF_EVENTS
2521         if (call->event.type && call->class->reg &&
2522             strcmp(name, "id") == 0) {
2523                 *mode = TRACE_MODE_READ;
2524                 *data = (void *)(long)call->event.type;
2525                 *fops = &ftrace_event_id_fops;
2526                 return 1;
2527         }
2528 #endif
2529
2530 #ifdef CONFIG_HIST_TRIGGERS
2531         if (strcmp(name, "hist") == 0) {
2532                 *mode = TRACE_MODE_READ;
2533                 *fops = &event_hist_fops;
2534                 return 1;
2535         }
2536 #endif
2537 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538         if (strcmp(name, "hist_debug") == 0) {
2539                 *mode = TRACE_MODE_READ;
2540                 *fops = &event_hist_debug_fops;
2541                 return 1;
2542         }
2543 #endif
2544 #ifdef CONFIG_TRACE_EVENT_INJECT
2545         if (call->event.type && call->class->reg &&
2546             strcmp(name, "inject") == 0) {
2547                 *mode = 0200;
2548                 *fops = &event_inject_fops;
2549                 return 1;
2550         }
2551 #endif
2552         return 0;
2553 }
2554
2555 /* The file is incremented on creation and freeing the enable file decrements it */
2556 static void event_release(const char *name, void *data)
2557 {
2558         struct trace_event_file *file = data;
2559
2560         event_file_put(file);
2561 }
2562
2563 static int
2564 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2565 {
2566         struct trace_event_call *call = file->event_call;
2567         struct trace_array *tr = file->tr;
2568         struct eventfs_inode *e_events;
2569         struct eventfs_inode *ei;
2570         const char *name;
2571         int nr_entries;
2572         int ret;
2573         static struct eventfs_entry event_entries[] = {
2574                 {
2575                         .name           = "enable",
2576                         .callback       = event_callback,
2577                         .release        = event_release,
2578                 },
2579                 {
2580                         .name           = "filter",
2581                         .callback       = event_callback,
2582                 },
2583                 {
2584                         .name           = "trigger",
2585                         .callback       = event_callback,
2586                 },
2587                 {
2588                         .name           = "format",
2589                         .callback       = event_callback,
2590                 },
2591 #ifdef CONFIG_PERF_EVENTS
2592                 {
2593                         .name           = "id",
2594                         .callback       = event_callback,
2595                 },
2596 #endif
2597 #ifdef CONFIG_HIST_TRIGGERS
2598                 {
2599                         .name           = "hist",
2600                         .callback       = event_callback,
2601                 },
2602 #endif
2603 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2604                 {
2605                         .name           = "hist_debug",
2606                         .callback       = event_callback,
2607                 },
2608 #endif
2609 #ifdef CONFIG_TRACE_EVENT_INJECT
2610                 {
2611                         .name           = "inject",
2612                         .callback       = event_callback,
2613                 },
2614 #endif
2615         };
2616
2617         /*
2618          * If the trace point header did not define TRACE_SYSTEM
2619          * then the system would be called "TRACE_SYSTEM". This should
2620          * never happen.
2621          */
2622         if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2623                 return -ENODEV;
2624
2625         e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2626         if (!e_events)
2627                 return -ENOMEM;
2628
2629         nr_entries = ARRAY_SIZE(event_entries);
2630
2631         name = trace_event_name(call);
2632         ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2633         if (IS_ERR(ei)) {
2634                 pr_warn("Could not create tracefs '%s' directory\n", name);
2635                 return -1;
2636         }
2637
2638         file->ei = ei;
2639
2640         ret = event_define_fields(call);
2641         if (ret < 0) {
2642                 pr_warn("Could not initialize trace point events/%s\n", name);
2643                 return ret;
2644         }
2645
2646         /* Gets decremented on freeing of the "enable" file */
2647         event_file_get(file);
2648
2649         return 0;
2650 }
2651
2652 static void remove_event_from_tracers(struct trace_event_call *call)
2653 {
2654         struct trace_event_file *file;
2655         struct trace_array *tr;
2656
2657         do_for_each_event_file_safe(tr, file) {
2658                 if (file->event_call != call)
2659                         continue;
2660
2661                 remove_event_file_dir(file);
2662                 /*
2663                  * The do_for_each_event_file_safe() is
2664                  * a double loop. After finding the call for this
2665                  * trace_array, we use break to jump to the next
2666                  * trace_array.
2667                  */
2668                 break;
2669         } while_for_each_event_file();
2670 }
2671
2672 static void event_remove(struct trace_event_call *call)
2673 {
2674         struct trace_array *tr;
2675         struct trace_event_file *file;
2676
2677         do_for_each_event_file(tr, file) {
2678                 if (file->event_call != call)
2679                         continue;
2680
2681                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2682                         tr->clear_trace = true;
2683
2684                 ftrace_event_enable_disable(file, 0);
2685                 /*
2686                  * The do_for_each_event_file() is
2687                  * a double loop. After finding the call for this
2688                  * trace_array, we use break to jump to the next
2689                  * trace_array.
2690                  */
2691                 break;
2692         } while_for_each_event_file();
2693
2694         if (call->event.funcs)
2695                 __unregister_trace_event(&call->event);
2696         remove_event_from_tracers(call);
2697         list_del(&call->list);
2698 }
2699
2700 static int event_init(struct trace_event_call *call)
2701 {
2702         int ret = 0;
2703         const char *name;
2704
2705         name = trace_event_name(call);
2706         if (WARN_ON(!name))
2707                 return -EINVAL;
2708
2709         if (call->class->raw_init) {
2710                 ret = call->class->raw_init(call);
2711                 if (ret < 0 && ret != -ENOSYS)
2712                         pr_warn("Could not initialize trace events/%s\n", name);
2713         }
2714
2715         return ret;
2716 }
2717
2718 static int
2719 __register_event(struct trace_event_call *call, struct module *mod)
2720 {
2721         int ret;
2722
2723         ret = event_init(call);
2724         if (ret < 0)
2725                 return ret;
2726
2727         list_add(&call->list, &ftrace_events);
2728         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2729                 atomic_set(&call->refcnt, 0);
2730         else
2731                 call->module = mod;
2732
2733         return 0;
2734 }
2735
2736 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2737 {
2738         int rlen;
2739         int elen;
2740
2741         /* Find the length of the eval value as a string */
2742         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2743         /* Make sure there's enough room to replace the string with the value */
2744         if (len < elen)
2745                 return NULL;
2746
2747         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2748
2749         /* Get the rest of the string of ptr */
2750         rlen = strlen(ptr + len);
2751         memmove(ptr + elen, ptr + len, rlen);
2752         /* Make sure we end the new string */
2753         ptr[elen + rlen] = 0;
2754
2755         return ptr + elen;
2756 }
2757
2758 static void update_event_printk(struct trace_event_call *call,
2759                                 struct trace_eval_map *map)
2760 {
2761         char *ptr;
2762         int quote = 0;
2763         int len = strlen(map->eval_string);
2764
2765         for (ptr = call->print_fmt; *ptr; ptr++) {
2766                 if (*ptr == '\\') {
2767                         ptr++;
2768                         /* paranoid */
2769                         if (!*ptr)
2770                                 break;
2771                         continue;
2772                 }
2773                 if (*ptr == '"') {
2774                         quote ^= 1;
2775                         continue;
2776                 }
2777                 if (quote)
2778                         continue;
2779                 if (isdigit(*ptr)) {
2780                         /* skip numbers */
2781                         do {
2782                                 ptr++;
2783                                 /* Check for alpha chars like ULL */
2784                         } while (isalnum(*ptr));
2785                         if (!*ptr)
2786                                 break;
2787                         /*
2788                          * A number must have some kind of delimiter after
2789                          * it, and we can ignore that too.
2790                          */
2791                         continue;
2792                 }
2793                 if (isalpha(*ptr) || *ptr == '_') {
2794                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2795                             !isalnum(ptr[len]) && ptr[len] != '_') {
2796                                 ptr = eval_replace(ptr, map, len);
2797                                 /* enum/sizeof string smaller than value */
2798                                 if (WARN_ON_ONCE(!ptr))
2799                                         return;
2800                                 /*
2801                                  * No need to decrement here, as eval_replace()
2802                                  * returns the pointer to the character passed
2803                                  * the eval, and two evals can not be placed
2804                                  * back to back without something in between.
2805                                  * We can skip that something in between.
2806                                  */
2807                                 continue;
2808                         }
2809                 skip_more:
2810                         do {
2811                                 ptr++;
2812                         } while (isalnum(*ptr) || *ptr == '_');
2813                         if (!*ptr)
2814                                 break;
2815                         /*
2816                          * If what comes after this variable is a '.' or
2817                          * '->' then we can continue to ignore that string.
2818                          */
2819                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2820                                 ptr += *ptr == '.' ? 1 : 2;
2821                                 if (!*ptr)
2822                                         break;
2823                                 goto skip_more;
2824                         }
2825                         /*
2826                          * Once again, we can skip the delimiter that came
2827                          * after the string.
2828                          */
2829                         continue;
2830                 }
2831         }
2832 }
2833
2834 static void add_str_to_module(struct module *module, char *str)
2835 {
2836         struct module_string *modstr;
2837
2838         modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2839
2840         /*
2841          * If we failed to allocate memory here, then we'll just
2842          * let the str memory leak when the module is removed.
2843          * If this fails to allocate, there's worse problems than
2844          * a leaked string on module removal.
2845          */
2846         if (WARN_ON_ONCE(!modstr))
2847                 return;
2848
2849         modstr->module = module;
2850         modstr->str = str;
2851
2852         list_add(&modstr->next, &module_strings);
2853 }
2854
2855 static void update_event_fields(struct trace_event_call *call,
2856                                 struct trace_eval_map *map)
2857 {
2858         struct ftrace_event_field *field;
2859         struct list_head *head;
2860         char *ptr;
2861         char *str;
2862         int len = strlen(map->eval_string);
2863
2864         /* Dynamic events should never have field maps */
2865         if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2866                 return;
2867
2868         head = trace_get_fields(call);
2869         list_for_each_entry(field, head, link) {
2870                 ptr = strchr(field->type, '[');
2871                 if (!ptr)
2872                         continue;
2873                 ptr++;
2874
2875                 if (!isalpha(*ptr) && *ptr != '_')
2876                         continue;
2877
2878                 if (strncmp(map->eval_string, ptr, len) != 0)
2879                         continue;
2880
2881                 str = kstrdup(field->type, GFP_KERNEL);
2882                 if (WARN_ON_ONCE(!str))
2883                         return;
2884                 ptr = str + (ptr - field->type);
2885                 ptr = eval_replace(ptr, map, len);
2886                 /* enum/sizeof string smaller than value */
2887                 if (WARN_ON_ONCE(!ptr)) {
2888                         kfree(str);
2889                         continue;
2890                 }
2891
2892                 /*
2893                  * If the event is part of a module, then we need to free the string
2894                  * when the module is removed. Otherwise, it will stay allocated
2895                  * until a reboot.
2896                  */
2897                 if (call->module)
2898                         add_str_to_module(call->module, str);
2899
2900                 field->type = str;
2901         }
2902 }
2903
2904 void trace_event_eval_update(struct trace_eval_map **map, int len)
2905 {
2906         struct trace_event_call *call, *p;
2907         const char *last_system = NULL;
2908         bool first = false;
2909         int last_i;
2910         int i;
2911
2912         down_write(&trace_event_sem);
2913         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2914                 /* events are usually grouped together with systems */
2915                 if (!last_system || call->class->system != last_system) {
2916                         first = true;
2917                         last_i = 0;
2918                         last_system = call->class->system;
2919                 }
2920
2921                 /*
2922                  * Since calls are grouped by systems, the likelihood that the
2923                  * next call in the iteration belongs to the same system as the
2924                  * previous call is high. As an optimization, we skip searching
2925                  * for a map[] that matches the call's system if the last call
2926                  * was from the same system. That's what last_i is for. If the
2927                  * call has the same system as the previous call, then last_i
2928                  * will be the index of the first map[] that has a matching
2929                  * system.
2930                  */
2931                 for (i = last_i; i < len; i++) {
2932                         if (call->class->system == map[i]->system) {
2933                                 /* Save the first system if need be */
2934                                 if (first) {
2935                                         last_i = i;
2936                                         first = false;
2937                                 }
2938                                 update_event_printk(call, map[i]);
2939                                 update_event_fields(call, map[i]);
2940                         }
2941                 }
2942                 cond_resched();
2943         }
2944         up_write(&trace_event_sem);
2945 }
2946
2947 static bool event_in_systems(struct trace_event_call *call,
2948                              const char *systems)
2949 {
2950         const char *system;
2951         const char *p;
2952
2953         if (!systems)
2954                 return true;
2955
2956         system = call->class->system;
2957         p = strstr(systems, system);
2958         if (!p)
2959                 return false;
2960
2961         if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2962                 return false;
2963
2964         p += strlen(system);
2965         return !*p || isspace(*p) || *p == ',';
2966 }
2967
2968 static struct trace_event_file *
2969 trace_create_new_event(struct trace_event_call *call,
2970                        struct trace_array *tr)
2971 {
2972         struct trace_pid_list *no_pid_list;
2973         struct trace_pid_list *pid_list;
2974         struct trace_event_file *file;
2975         unsigned int first;
2976
2977         if (!event_in_systems(call, tr->system_names))
2978                 return NULL;
2979
2980         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2981         if (!file)
2982                 return ERR_PTR(-ENOMEM);
2983
2984         pid_list = rcu_dereference_protected(tr->filtered_pids,
2985                                              lockdep_is_held(&event_mutex));
2986         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2987                                              lockdep_is_held(&event_mutex));
2988
2989         if (!trace_pid_list_first(pid_list, &first) ||
2990             !trace_pid_list_first(no_pid_list, &first))
2991                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2992
2993         file->event_call = call;
2994         file->tr = tr;
2995         atomic_set(&file->sm_ref, 0);
2996         atomic_set(&file->tm_ref, 0);
2997         INIT_LIST_HEAD(&file->triggers);
2998         list_add(&file->list, &tr->events);
2999         event_file_get(file);
3000
3001         return file;
3002 }
3003
3004 #define MAX_BOOT_TRIGGERS 32
3005
3006 static struct boot_triggers {
3007         const char              *event;
3008         char                    *trigger;
3009 } bootup_triggers[MAX_BOOT_TRIGGERS];
3010
3011 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3012 static int nr_boot_triggers;
3013
3014 static __init int setup_trace_triggers(char *str)
3015 {
3016         char *trigger;
3017         char *buf;
3018         int i;
3019
3020         strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3021         trace_set_ring_buffer_expanded(NULL);
3022         disable_tracing_selftest("running event triggers");
3023
3024         buf = bootup_trigger_buf;
3025         for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3026                 trigger = strsep(&buf, ",");
3027                 if (!trigger)
3028                         break;
3029                 bootup_triggers[i].event = strsep(&trigger, ".");
3030                 bootup_triggers[i].trigger = trigger;
3031                 if (!bootup_triggers[i].trigger)
3032                         break;
3033         }
3034
3035         nr_boot_triggers = i;
3036         return 1;
3037 }
3038 __setup("trace_trigger=", setup_trace_triggers);
3039
3040 /* Add an event to a trace directory */
3041 static int
3042 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3043 {
3044         struct trace_event_file *file;
3045
3046         file = trace_create_new_event(call, tr);
3047         /*
3048          * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3049          * allocation, or NULL if the event is not part of the tr->system_names.
3050          * When the event is not part of the tr->system_names, return zero, not
3051          * an error.
3052          */
3053         if (!file)
3054                 return 0;
3055
3056         if (IS_ERR(file))
3057                 return PTR_ERR(file);
3058
3059         if (eventdir_initialized)
3060                 return event_create_dir(tr->event_dir, file);
3061         else
3062                 return event_define_fields(call);
3063 }
3064
3065 static void trace_early_triggers(struct trace_event_file *file, const char *name)
3066 {
3067         int ret;
3068         int i;
3069
3070         for (i = 0; i < nr_boot_triggers; i++) {
3071                 if (strcmp(name, bootup_triggers[i].event))
3072                         continue;
3073                 mutex_lock(&event_mutex);
3074                 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3075                 mutex_unlock(&event_mutex);
3076                 if (ret)
3077                         pr_err("Failed to register trigger '%s' on event %s\n",
3078                                bootup_triggers[i].trigger,
3079                                bootup_triggers[i].event);
3080         }
3081 }
3082
3083 /*
3084  * Just create a descriptor for early init. A descriptor is required
3085  * for enabling events at boot. We want to enable events before
3086  * the filesystem is initialized.
3087  */
3088 static int
3089 __trace_early_add_new_event(struct trace_event_call *call,
3090                             struct trace_array *tr)
3091 {
3092         struct trace_event_file *file;
3093         int ret;
3094
3095         file = trace_create_new_event(call, tr);
3096         /*
3097          * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3098          * allocation, or NULL if the event is not part of the tr->system_names.
3099          * When the event is not part of the tr->system_names, return zero, not
3100          * an error.
3101          */
3102         if (!file)
3103                 return 0;
3104
3105         if (IS_ERR(file))
3106                 return PTR_ERR(file);
3107
3108         ret = event_define_fields(call);
3109         if (ret)
3110                 return ret;
3111
3112         trace_early_triggers(file, trace_event_name(call));
3113
3114         return 0;
3115 }
3116
3117 struct ftrace_module_file_ops;
3118 static void __add_event_to_tracers(struct trace_event_call *call);
3119
3120 /* Add an additional event_call dynamically */
3121 int trace_add_event_call(struct trace_event_call *call)
3122 {
3123         int ret;
3124         lockdep_assert_held(&event_mutex);
3125
3126         mutex_lock(&trace_types_lock);
3127
3128         ret = __register_event(call, NULL);
3129         if (ret >= 0)
3130                 __add_event_to_tracers(call);
3131
3132         mutex_unlock(&trace_types_lock);
3133         return ret;
3134 }
3135 EXPORT_SYMBOL_GPL(trace_add_event_call);
3136
3137 /*
3138  * Must be called under locking of trace_types_lock, event_mutex and
3139  * trace_event_sem.
3140  */
3141 static void __trace_remove_event_call(struct trace_event_call *call)
3142 {
3143         event_remove(call);
3144         trace_destroy_fields(call);
3145         free_event_filter(call->filter);
3146         call->filter = NULL;
3147 }
3148
3149 static int probe_remove_event_call(struct trace_event_call *call)
3150 {
3151         struct trace_array *tr;
3152         struct trace_event_file *file;
3153
3154 #ifdef CONFIG_PERF_EVENTS
3155         if (call->perf_refcount)
3156                 return -EBUSY;
3157 #endif
3158         do_for_each_event_file(tr, file) {
3159                 if (file->event_call != call)
3160                         continue;
3161                 /*
3162                  * We can't rely on ftrace_event_enable_disable(enable => 0)
3163                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3164                  * TRACE_REG_UNREGISTER.
3165                  */
3166                 if (file->flags & EVENT_FILE_FL_ENABLED)
3167                         goto busy;
3168
3169                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3170                         tr->clear_trace = true;
3171                 /*
3172                  * The do_for_each_event_file_safe() is
3173                  * a double loop. After finding the call for this
3174                  * trace_array, we use break to jump to the next
3175                  * trace_array.
3176                  */
3177                 break;
3178         } while_for_each_event_file();
3179
3180         __trace_remove_event_call(call);
3181
3182         return 0;
3183  busy:
3184         /* No need to clear the trace now */
3185         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3186                 tr->clear_trace = false;
3187         }
3188         return -EBUSY;
3189 }
3190
3191 /* Remove an event_call */
3192 int trace_remove_event_call(struct trace_event_call *call)
3193 {
3194         int ret;
3195
3196         lockdep_assert_held(&event_mutex);
3197
3198         mutex_lock(&trace_types_lock);
3199         down_write(&trace_event_sem);
3200         ret = probe_remove_event_call(call);
3201         up_write(&trace_event_sem);
3202         mutex_unlock(&trace_types_lock);
3203
3204         return ret;
3205 }
3206 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3207
3208 #define for_each_event(event, start, end)                       \
3209         for (event = start;                                     \
3210              (unsigned long)event < (unsigned long)end;         \
3211              event++)
3212
3213 #ifdef CONFIG_MODULES
3214
3215 static void trace_module_add_events(struct module *mod)
3216 {
3217         struct trace_event_call **call, **start, **end;
3218
3219         if (!mod->num_trace_events)
3220                 return;
3221
3222         /* Don't add infrastructure for mods without tracepoints */
3223         if (trace_module_has_bad_taint(mod)) {
3224                 pr_err("%s: module has bad taint, not creating trace events\n",
3225                        mod->name);
3226                 return;
3227         }
3228
3229         start = mod->trace_events;
3230         end = mod->trace_events + mod->num_trace_events;
3231
3232         for_each_event(call, start, end) {
3233                 __register_event(*call, mod);
3234                 __add_event_to_tracers(*call);
3235         }
3236 }
3237
3238 static void trace_module_remove_events(struct module *mod)
3239 {
3240         struct trace_event_call *call, *p;
3241         struct module_string *modstr, *m;
3242
3243         down_write(&trace_event_sem);
3244         list_for_each_entry_safe(call, p, &ftrace_events, list) {
3245                 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3246                         continue;
3247                 if (call->module == mod)
3248                         __trace_remove_event_call(call);
3249         }
3250         /* Check for any strings allocade for this module */
3251         list_for_each_entry_safe(modstr, m, &module_strings, next) {
3252                 if (modstr->module != mod)
3253                         continue;
3254                 list_del(&modstr->next);
3255                 kfree(modstr->str);
3256                 kfree(modstr);
3257         }
3258         up_write(&trace_event_sem);
3259
3260         /*
3261          * It is safest to reset the ring buffer if the module being unloaded
3262          * registered any events that were used. The only worry is if
3263          * a new module gets loaded, and takes on the same id as the events
3264          * of this module. When printing out the buffer, traced events left
3265          * over from this module may be passed to the new module events and
3266          * unexpected results may occur.
3267          */
3268         tracing_reset_all_online_cpus_unlocked();
3269 }
3270
3271 static int trace_module_notify(struct notifier_block *self,
3272                                unsigned long val, void *data)
3273 {
3274         struct module *mod = data;
3275
3276         mutex_lock(&event_mutex);
3277         mutex_lock(&trace_types_lock);
3278         switch (val) {
3279         case MODULE_STATE_COMING:
3280                 trace_module_add_events(mod);
3281                 break;
3282         case MODULE_STATE_GOING:
3283                 trace_module_remove_events(mod);
3284                 break;
3285         }
3286         mutex_unlock(&trace_types_lock);
3287         mutex_unlock(&event_mutex);
3288
3289         return NOTIFY_OK;
3290 }
3291
3292 static struct notifier_block trace_module_nb = {
3293         .notifier_call = trace_module_notify,
3294         .priority = 1, /* higher than trace.c module notify */
3295 };
3296 #endif /* CONFIG_MODULES */
3297
3298 /* Create a new event directory structure for a trace directory. */
3299 static void
3300 __trace_add_event_dirs(struct trace_array *tr)
3301 {
3302         struct trace_event_call *call;
3303         int ret;
3304
3305         list_for_each_entry(call, &ftrace_events, list) {
3306                 ret = __trace_add_new_event(call, tr);
3307                 if (ret < 0)
3308                         pr_warn("Could not create directory for event %s\n",
3309                                 trace_event_name(call));
3310         }
3311 }
3312
3313 /* Returns any file that matches the system and event */
3314 struct trace_event_file *
3315 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3316 {
3317         struct trace_event_file *file;
3318         struct trace_event_call *call;
3319         const char *name;
3320
3321         list_for_each_entry(file, &tr->events, list) {
3322
3323                 call = file->event_call;
3324                 name = trace_event_name(call);
3325
3326                 if (!name || !call->class)
3327                         continue;
3328
3329                 if (strcmp(event, name) == 0 &&
3330                     strcmp(system, call->class->system) == 0)
3331                         return file;
3332         }
3333         return NULL;
3334 }
3335
3336 /* Returns valid trace event files that match system and event */
3337 struct trace_event_file *
3338 find_event_file(struct trace_array *tr, const char *system, const char *event)
3339 {
3340         struct trace_event_file *file;
3341
3342         file = __find_event_file(tr, system, event);
3343         if (!file || !file->event_call->class->reg ||
3344             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3345                 return NULL;
3346
3347         return file;
3348 }
3349
3350 /**
3351  * trace_get_event_file - Find and return a trace event file
3352  * @instance: The name of the trace instance containing the event
3353  * @system: The name of the system containing the event
3354  * @event: The name of the event
3355  *
3356  * Return a trace event file given the trace instance name, trace
3357  * system, and trace event name.  If the instance name is NULL, it
3358  * refers to the top-level trace array.
3359  *
3360  * This function will look it up and return it if found, after calling
3361  * trace_array_get() to prevent the instance from going away, and
3362  * increment the event's module refcount to prevent it from being
3363  * removed.
3364  *
3365  * To release the file, call trace_put_event_file(), which will call
3366  * trace_array_put() and decrement the event's module refcount.
3367  *
3368  * Return: The trace event on success, ERR_PTR otherwise.
3369  */
3370 struct trace_event_file *trace_get_event_file(const char *instance,
3371                                               const char *system,
3372                                               const char *event)
3373 {
3374         struct trace_array *tr = top_trace_array();
3375         struct trace_event_file *file = NULL;
3376         int ret = -EINVAL;
3377
3378         if (instance) {
3379                 tr = trace_array_find_get(instance);
3380                 if (!tr)
3381                         return ERR_PTR(-ENOENT);
3382         } else {
3383                 ret = trace_array_get(tr);
3384                 if (ret)
3385                         return ERR_PTR(ret);
3386         }
3387
3388         mutex_lock(&event_mutex);
3389
3390         file = find_event_file(tr, system, event);
3391         if (!file) {
3392                 trace_array_put(tr);
3393                 ret = -EINVAL;
3394                 goto out;
3395         }
3396
3397         /* Don't let event modules unload while in use */
3398         ret = trace_event_try_get_ref(file->event_call);
3399         if (!ret) {
3400                 trace_array_put(tr);
3401                 ret = -EBUSY;
3402                 goto out;
3403         }
3404
3405         ret = 0;
3406  out:
3407         mutex_unlock(&event_mutex);
3408
3409         if (ret)
3410                 file = ERR_PTR(ret);
3411
3412         return file;
3413 }
3414 EXPORT_SYMBOL_GPL(trace_get_event_file);
3415
3416 /**
3417  * trace_put_event_file - Release a file from trace_get_event_file()
3418  * @file: The trace event file
3419  *
3420  * If a file was retrieved using trace_get_event_file(), this should
3421  * be called when it's no longer needed.  It will cancel the previous
3422  * trace_array_get() called by that function, and decrement the
3423  * event's module refcount.
3424  */
3425 void trace_put_event_file(struct trace_event_file *file)
3426 {
3427         mutex_lock(&event_mutex);
3428         trace_event_put_ref(file->event_call);
3429         mutex_unlock(&event_mutex);
3430
3431         trace_array_put(file->tr);
3432 }
3433 EXPORT_SYMBOL_GPL(trace_put_event_file);
3434
3435 #ifdef CONFIG_DYNAMIC_FTRACE
3436
3437 /* Avoid typos */
3438 #define ENABLE_EVENT_STR        "enable_event"
3439 #define DISABLE_EVENT_STR       "disable_event"
3440
3441 struct event_probe_data {
3442         struct trace_event_file *file;
3443         unsigned long                   count;
3444         int                             ref;
3445         bool                            enable;
3446 };
3447
3448 static void update_event_probe(struct event_probe_data *data)
3449 {
3450         if (data->enable)
3451                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3452         else
3453                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3454 }
3455
3456 static void
3457 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3458                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3459                    void *data)
3460 {
3461         struct ftrace_func_mapper *mapper = data;
3462         struct event_probe_data *edata;
3463         void **pdata;
3464
3465         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3466         if (!pdata || !*pdata)
3467                 return;
3468
3469         edata = *pdata;
3470         update_event_probe(edata);
3471 }
3472
3473 static void
3474 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3475                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3476                          void *data)
3477 {
3478         struct ftrace_func_mapper *mapper = data;
3479         struct event_probe_data *edata;
3480         void **pdata;
3481
3482         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3483         if (!pdata || !*pdata)
3484                 return;
3485
3486         edata = *pdata;
3487
3488         if (!edata->count)
3489                 return;
3490
3491         /* Skip if the event is in a state we want to switch to */
3492         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3493                 return;
3494
3495         if (edata->count != -1)
3496                 (edata->count)--;
3497
3498         update_event_probe(edata);
3499 }
3500
3501 static int
3502 event_enable_print(struct seq_file *m, unsigned long ip,
3503                    struct ftrace_probe_ops *ops, void *data)
3504 {
3505         struct ftrace_func_mapper *mapper = data;
3506         struct event_probe_data *edata;
3507         void **pdata;
3508
3509         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3510
3511         if (WARN_ON_ONCE(!pdata || !*pdata))
3512                 return 0;
3513
3514         edata = *pdata;
3515
3516         seq_printf(m, "%ps:", (void *)ip);
3517
3518         seq_printf(m, "%s:%s:%s",
3519                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3520                    edata->file->event_call->class->system,
3521                    trace_event_name(edata->file->event_call));
3522
3523         if (edata->count == -1)
3524                 seq_puts(m, ":unlimited\n");
3525         else
3526                 seq_printf(m, ":count=%ld\n", edata->count);
3527
3528         return 0;
3529 }
3530
3531 static int
3532 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3533                   unsigned long ip, void *init_data, void **data)
3534 {
3535         struct ftrace_func_mapper *mapper = *data;
3536         struct event_probe_data *edata = init_data;
3537         int ret;
3538
3539         if (!mapper) {
3540                 mapper = allocate_ftrace_func_mapper();
3541                 if (!mapper)
3542                         return -ENODEV;
3543                 *data = mapper;
3544         }
3545
3546         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3547         if (ret < 0)
3548                 return ret;
3549
3550         edata->ref++;
3551
3552         return 0;
3553 }
3554
3555 static int free_probe_data(void *data)
3556 {
3557         struct event_probe_data *edata = data;
3558
3559         edata->ref--;
3560         if (!edata->ref) {
3561                 /* Remove the SOFT_MODE flag */
3562                 __ftrace_event_enable_disable(edata->file, 0, 1);
3563                 trace_event_put_ref(edata->file->event_call);
3564                 kfree(edata);
3565         }
3566         return 0;
3567 }
3568
3569 static void
3570 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3571                   unsigned long ip, void *data)
3572 {
3573         struct ftrace_func_mapper *mapper = data;
3574         struct event_probe_data *edata;
3575
3576         if (!ip) {
3577                 if (!mapper)
3578                         return;
3579                 free_ftrace_func_mapper(mapper, free_probe_data);
3580                 return;
3581         }
3582
3583         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3584
3585         if (WARN_ON_ONCE(!edata))
3586                 return;
3587
3588         if (WARN_ON_ONCE(edata->ref <= 0))
3589                 return;
3590
3591         free_probe_data(edata);
3592 }
3593
3594 static struct ftrace_probe_ops event_enable_probe_ops = {
3595         .func                   = event_enable_probe,
3596         .print                  = event_enable_print,
3597         .init                   = event_enable_init,
3598         .free                   = event_enable_free,
3599 };
3600
3601 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3602         .func                   = event_enable_count_probe,
3603         .print                  = event_enable_print,
3604         .init                   = event_enable_init,
3605         .free                   = event_enable_free,
3606 };
3607
3608 static struct ftrace_probe_ops event_disable_probe_ops = {
3609         .func                   = event_enable_probe,
3610         .print                  = event_enable_print,
3611         .init                   = event_enable_init,
3612         .free                   = event_enable_free,
3613 };
3614
3615 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3616         .func                   = event_enable_count_probe,
3617         .print                  = event_enable_print,
3618         .init                   = event_enable_init,
3619         .free                   = event_enable_free,
3620 };
3621
3622 static int
3623 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3624                   char *glob, char *cmd, char *param, int enabled)
3625 {
3626         struct trace_event_file *file;
3627         struct ftrace_probe_ops *ops;
3628         struct event_probe_data *data;
3629         const char *system;
3630         const char *event;
3631         char *number;
3632         bool enable;
3633         int ret;
3634
3635         if (!tr)
3636                 return -ENODEV;
3637
3638         /* hash funcs only work with set_ftrace_filter */
3639         if (!enabled || !param)
3640                 return -EINVAL;
3641
3642         system = strsep(&param, ":");
3643         if (!param)
3644                 return -EINVAL;
3645
3646         event = strsep(&param, ":");
3647
3648         mutex_lock(&event_mutex);
3649
3650         ret = -EINVAL;
3651         file = find_event_file(tr, system, event);
3652         if (!file)
3653                 goto out;
3654
3655         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3656
3657         if (enable)
3658                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3659         else
3660                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3661
3662         if (glob[0] == '!') {
3663                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3664                 goto out;
3665         }
3666
3667         ret = -ENOMEM;
3668
3669         data = kzalloc(sizeof(*data), GFP_KERNEL);
3670         if (!data)
3671                 goto out;
3672
3673         data->enable = enable;
3674         data->count = -1;
3675         data->file = file;
3676
3677         if (!param)
3678                 goto out_reg;
3679
3680         number = strsep(&param, ":");
3681
3682         ret = -EINVAL;
3683         if (!strlen(number))
3684                 goto out_free;
3685
3686         /*
3687          * We use the callback data field (which is a pointer)
3688          * as our counter.
3689          */
3690         ret = kstrtoul(number, 0, &data->count);
3691         if (ret)
3692                 goto out_free;
3693
3694  out_reg:
3695         /* Don't let event modules unload while probe registered */
3696         ret = trace_event_try_get_ref(file->event_call);
3697         if (!ret) {
3698                 ret = -EBUSY;
3699                 goto out_free;
3700         }
3701
3702         ret = __ftrace_event_enable_disable(file, 1, 1);
3703         if (ret < 0)
3704                 goto out_put;
3705
3706         ret = register_ftrace_function_probe(glob, tr, ops, data);
3707         /*
3708          * The above returns on success the # of functions enabled,
3709          * but if it didn't find any functions it returns zero.
3710          * Consider no functions a failure too.
3711          */
3712         if (!ret) {
3713                 ret = -ENOENT;
3714                 goto out_disable;
3715         } else if (ret < 0)
3716                 goto out_disable;
3717         /* Just return zero, not the number of enabled functions */
3718         ret = 0;
3719  out:
3720         mutex_unlock(&event_mutex);
3721         return ret;
3722
3723  out_disable:
3724         __ftrace_event_enable_disable(file, 0, 1);
3725  out_put:
3726         trace_event_put_ref(file->event_call);
3727  out_free:
3728         kfree(data);
3729         goto out;
3730 }
3731
3732 static struct ftrace_func_command event_enable_cmd = {
3733         .name                   = ENABLE_EVENT_STR,
3734         .func                   = event_enable_func,
3735 };
3736
3737 static struct ftrace_func_command event_disable_cmd = {
3738         .name                   = DISABLE_EVENT_STR,
3739         .func                   = event_enable_func,
3740 };
3741
3742 static __init int register_event_cmds(void)
3743 {
3744         int ret;
3745
3746         ret = register_ftrace_command(&event_enable_cmd);
3747         if (WARN_ON(ret < 0))
3748                 return ret;
3749         ret = register_ftrace_command(&event_disable_cmd);
3750         if (WARN_ON(ret < 0))
3751                 unregister_ftrace_command(&event_enable_cmd);
3752         return ret;
3753 }
3754 #else
3755 static inline int register_event_cmds(void) { return 0; }
3756 #endif /* CONFIG_DYNAMIC_FTRACE */
3757
3758 /*
3759  * The top level array and trace arrays created by boot-time tracing
3760  * have already had its trace_event_file descriptors created in order
3761  * to allow for early events to be recorded.
3762  * This function is called after the tracefs has been initialized,
3763  * and we now have to create the files associated to the events.
3764  */
3765 static void __trace_early_add_event_dirs(struct trace_array *tr)
3766 {
3767         struct trace_event_file *file;
3768         int ret;
3769
3770
3771         list_for_each_entry(file, &tr->events, list) {
3772                 ret = event_create_dir(tr->event_dir, file);
3773                 if (ret < 0)
3774                         pr_warn("Could not create directory for event %s\n",
3775                                 trace_event_name(file->event_call));
3776         }
3777 }
3778
3779 /*
3780  * For early boot up, the top trace array and the trace arrays created
3781  * by boot-time tracing require to have a list of events that can be
3782  * enabled. This must be done before the filesystem is set up in order
3783  * to allow events to be traced early.
3784  */
3785 void __trace_early_add_events(struct trace_array *tr)
3786 {
3787         struct trace_event_call *call;
3788         int ret;
3789
3790         list_for_each_entry(call, &ftrace_events, list) {
3791                 /* Early boot up should not have any modules loaded */
3792                 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3793                     WARN_ON_ONCE(call->module))
3794                         continue;
3795
3796                 ret = __trace_early_add_new_event(call, tr);
3797                 if (ret < 0)
3798                         pr_warn("Could not create early event %s\n",
3799                                 trace_event_name(call));
3800         }
3801 }
3802
3803 /* Remove the event directory structure for a trace directory. */
3804 static void
3805 __trace_remove_event_dirs(struct trace_array *tr)
3806 {
3807         struct trace_event_file *file, *next;
3808
3809         list_for_each_entry_safe(file, next, &tr->events, list)
3810                 remove_event_file_dir(file);
3811 }
3812
3813 static void __add_event_to_tracers(struct trace_event_call *call)
3814 {
3815         struct trace_array *tr;
3816
3817         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3818                 __trace_add_new_event(call, tr);
3819 }
3820
3821 extern struct trace_event_call *__start_ftrace_events[];
3822 extern struct trace_event_call *__stop_ftrace_events[];
3823
3824 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3825
3826 static __init int setup_trace_event(char *str)
3827 {
3828         strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3829         trace_set_ring_buffer_expanded(NULL);
3830         disable_tracing_selftest("running event tracing");
3831
3832         return 1;
3833 }
3834 __setup("trace_event=", setup_trace_event);
3835
3836 static int events_callback(const char *name, umode_t *mode, void **data,
3837                            const struct file_operations **fops)
3838 {
3839         if (strcmp(name, "enable") == 0) {
3840                 *mode = TRACE_MODE_WRITE;
3841                 *fops = &ftrace_tr_enable_fops;
3842                 return 1;
3843         }
3844
3845         if (strcmp(name, "header_page") == 0) {
3846                 *mode = TRACE_MODE_READ;
3847                 *fops = &ftrace_show_header_page_fops;
3848
3849         } else if (strcmp(name, "header_event") == 0) {
3850                 *mode = TRACE_MODE_READ;
3851                 *fops = &ftrace_show_header_event_fops;
3852         } else
3853                 return 0;
3854
3855         return 1;
3856 }
3857
3858 /* Expects to have event_mutex held when called */
3859 static int
3860 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3861 {
3862         struct eventfs_inode *e_events;
3863         struct dentry *entry;
3864         int nr_entries;
3865         static struct eventfs_entry events_entries[] = {
3866                 {
3867                         .name           = "enable",
3868                         .callback       = events_callback,
3869                 },
3870                 {
3871                         .name           = "header_page",
3872                         .callback       = events_callback,
3873                 },
3874                 {
3875                         .name           = "header_event",
3876                         .callback       = events_callback,
3877                 },
3878         };
3879
3880         entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3881                                   tr, &ftrace_set_event_fops);
3882         if (!entry)
3883                 return -ENOMEM;
3884
3885         nr_entries = ARRAY_SIZE(events_entries);
3886
3887         e_events = eventfs_create_events_dir("events", parent, events_entries,
3888                                              nr_entries, tr);
3889         if (IS_ERR(e_events)) {
3890                 pr_warn("Could not create tracefs 'events' directory\n");
3891                 return -ENOMEM;
3892         }
3893
3894         /* There are not as crucial, just warn if they are not created */
3895
3896         trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3897                           tr, &ftrace_set_event_pid_fops);
3898
3899         trace_create_file("set_event_notrace_pid",
3900                           TRACE_MODE_WRITE, parent, tr,
3901                           &ftrace_set_event_notrace_pid_fops);
3902
3903         tr->event_dir = e_events;
3904
3905         return 0;
3906 }
3907
3908 /**
3909  * event_trace_add_tracer - add a instance of a trace_array to events
3910  * @parent: The parent dentry to place the files/directories for events in
3911  * @tr: The trace array associated with these events
3912  *
3913  * When a new instance is created, it needs to set up its events
3914  * directory, as well as other files associated with events. It also
3915  * creates the event hierarchy in the @parent/events directory.
3916  *
3917  * Returns 0 on success.
3918  *
3919  * Must be called with event_mutex held.
3920  */
3921 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3922 {
3923         int ret;
3924
3925         lockdep_assert_held(&event_mutex);
3926
3927         ret = create_event_toplevel_files(parent, tr);
3928         if (ret)
3929                 goto out;
3930
3931         down_write(&trace_event_sem);
3932         /* If tr already has the event list, it is initialized in early boot. */
3933         if (unlikely(!list_empty(&tr->events)))
3934                 __trace_early_add_event_dirs(tr);
3935         else
3936                 __trace_add_event_dirs(tr);
3937         up_write(&trace_event_sem);
3938
3939  out:
3940         return ret;
3941 }
3942
3943 /*
3944  * The top trace array already had its file descriptors created.
3945  * Now the files themselves need to be created.
3946  */
3947 static __init int
3948 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3949 {
3950         int ret;
3951
3952         mutex_lock(&event_mutex);
3953
3954         ret = create_event_toplevel_files(parent, tr);
3955         if (ret)
3956                 goto out_unlock;
3957
3958         down_write(&trace_event_sem);
3959         __trace_early_add_event_dirs(tr);
3960         up_write(&trace_event_sem);
3961
3962  out_unlock:
3963         mutex_unlock(&event_mutex);
3964
3965         return ret;
3966 }
3967
3968 /* Must be called with event_mutex held */
3969 int event_trace_del_tracer(struct trace_array *tr)
3970 {
3971         lockdep_assert_held(&event_mutex);
3972
3973         /* Disable any event triggers and associated soft-disabled events */
3974         clear_event_triggers(tr);
3975
3976         /* Clear the pid list */
3977         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3978
3979         /* Disable any running events */
3980         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3981
3982         /* Make sure no more events are being executed */
3983         tracepoint_synchronize_unregister();
3984
3985         down_write(&trace_event_sem);
3986         __trace_remove_event_dirs(tr);
3987         eventfs_remove_events_dir(tr->event_dir);
3988         up_write(&trace_event_sem);
3989
3990         tr->event_dir = NULL;
3991
3992         return 0;
3993 }
3994
3995 static __init int event_trace_memsetup(void)
3996 {
3997         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3998         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3999         return 0;
4000 }
4001
4002 __init void
4003 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4004 {
4005         char *token;
4006         int ret;
4007
4008         while (true) {
4009                 token = strsep(&buf, ",");
4010
4011                 if (!token)
4012                         break;
4013
4014                 if (*token) {
4015                         /* Restarting syscalls requires that we stop them first */
4016                         if (disable_first)
4017                                 ftrace_set_clr_event(tr, token, 0);
4018
4019                         ret = ftrace_set_clr_event(tr, token, 1);
4020                         if (ret)
4021                                 pr_warn("Failed to enable trace event: %s\n", token);
4022                 }
4023
4024                 /* Put back the comma to allow this to be called again */
4025                 if (buf)
4026                         *(buf - 1) = ',';
4027         }
4028 }
4029
4030 static __init int event_trace_enable(void)
4031 {
4032         struct trace_array *tr = top_trace_array();
4033         struct trace_event_call **iter, *call;
4034         int ret;
4035
4036         if (!tr)
4037                 return -ENODEV;
4038
4039         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4040
4041                 call = *iter;
4042                 ret = event_init(call);
4043                 if (!ret)
4044                         list_add(&call->list, &ftrace_events);
4045         }
4046
4047         register_trigger_cmds();
4048
4049         /*
4050          * We need the top trace array to have a working set of trace
4051          * points at early init, before the debug files and directories
4052          * are created. Create the file entries now, and attach them
4053          * to the actual file dentries later.
4054          */
4055         __trace_early_add_events(tr);
4056
4057         early_enable_events(tr, bootup_event_buf, false);
4058
4059         trace_printk_start_comm();
4060
4061         register_event_cmds();
4062
4063
4064         return 0;
4065 }
4066
4067 /*
4068  * event_trace_enable() is called from trace_event_init() first to
4069  * initialize events and perhaps start any events that are on the
4070  * command line. Unfortunately, there are some events that will not
4071  * start this early, like the system call tracepoints that need
4072  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4073  * event_trace_enable() is called before pid 1 starts, and this flag
4074  * is never set, making the syscall tracepoint never get reached, but
4075  * the event is enabled regardless (and not doing anything).
4076  */
4077 static __init int event_trace_enable_again(void)
4078 {
4079         struct trace_array *tr;
4080
4081         tr = top_trace_array();
4082         if (!tr)
4083                 return -ENODEV;
4084
4085         early_enable_events(tr, bootup_event_buf, true);
4086
4087         return 0;
4088 }
4089
4090 early_initcall(event_trace_enable_again);
4091
4092 /* Init fields which doesn't related to the tracefs */
4093 static __init int event_trace_init_fields(void)
4094 {
4095         if (trace_define_generic_fields())
4096                 pr_warn("tracing: Failed to allocated generic fields");
4097
4098         if (trace_define_common_fields())
4099                 pr_warn("tracing: Failed to allocate common fields");
4100
4101         return 0;
4102 }
4103
4104 __init int event_trace_init(void)
4105 {
4106         struct trace_array *tr;
4107         int ret;
4108
4109         tr = top_trace_array();
4110         if (!tr)
4111                 return -ENODEV;
4112
4113         trace_create_file("available_events", TRACE_MODE_READ,
4114                           NULL, tr, &ftrace_avail_fops);
4115
4116         ret = early_event_add_tracer(NULL, tr);
4117         if (ret)
4118                 return ret;
4119
4120 #ifdef CONFIG_MODULES
4121         ret = register_module_notifier(&trace_module_nb);
4122         if (ret)
4123                 pr_warn("Failed to register trace events module notifier\n");
4124 #endif
4125
4126         eventdir_initialized = true;
4127
4128         return 0;
4129 }
4130
4131 void __init trace_event_init(void)
4132 {
4133         event_trace_memsetup();
4134         init_ftrace_syscalls();
4135         event_trace_enable();
4136         event_trace_init_fields();
4137 }
4138
4139 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4140
4141 static DEFINE_SPINLOCK(test_spinlock);
4142 static DEFINE_SPINLOCK(test_spinlock_irq);
4143 static DEFINE_MUTEX(test_mutex);
4144
4145 static __init void test_work(struct work_struct *dummy)
4146 {
4147         spin_lock(&test_spinlock);
4148         spin_lock_irq(&test_spinlock_irq);
4149         udelay(1);
4150         spin_unlock_irq(&test_spinlock_irq);
4151         spin_unlock(&test_spinlock);
4152
4153         mutex_lock(&test_mutex);
4154         msleep(1);
4155         mutex_unlock(&test_mutex);
4156 }
4157
4158 static __init int event_test_thread(void *unused)
4159 {
4160         void *test_malloc;
4161
4162         test_malloc = kmalloc(1234, GFP_KERNEL);
4163         if (!test_malloc)
4164                 pr_info("failed to kmalloc\n");
4165
4166         schedule_on_each_cpu(test_work);
4167
4168         kfree(test_malloc);
4169
4170         set_current_state(TASK_INTERRUPTIBLE);
4171         while (!kthread_should_stop()) {
4172                 schedule();
4173                 set_current_state(TASK_INTERRUPTIBLE);
4174         }
4175         __set_current_state(TASK_RUNNING);
4176
4177         return 0;
4178 }
4179
4180 /*
4181  * Do various things that may trigger events.
4182  */
4183 static __init void event_test_stuff(void)
4184 {
4185         struct task_struct *test_thread;
4186
4187         test_thread = kthread_run(event_test_thread, NULL, "test-events");
4188         msleep(1);
4189         kthread_stop(test_thread);
4190 }
4191
4192 /*
4193  * For every trace event defined, we will test each trace point separately,
4194  * and then by groups, and finally all trace points.
4195  */
4196 static __init void event_trace_self_tests(void)
4197 {
4198         struct trace_subsystem_dir *dir;
4199         struct trace_event_file *file;
4200         struct trace_event_call *call;
4201         struct event_subsystem *system;
4202         struct trace_array *tr;
4203         int ret;
4204
4205         tr = top_trace_array();
4206         if (!tr)
4207                 return;
4208
4209         pr_info("Running tests on trace events:\n");
4210
4211         list_for_each_entry(file, &tr->events, list) {
4212
4213                 call = file->event_call;
4214
4215                 /* Only test those that have a probe */
4216                 if (!call->class || !call->class->probe)
4217                         continue;
4218
4219 /*
4220  * Testing syscall events here is pretty useless, but
4221  * we still do it if configured. But this is time consuming.
4222  * What we really need is a user thread to perform the
4223  * syscalls as we test.
4224  */
4225 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4226                 if (call->class->system &&
4227                     strcmp(call->class->system, "syscalls") == 0)
4228                         continue;
4229 #endif
4230
4231                 pr_info("Testing event %s: ", trace_event_name(call));
4232
4233                 /*
4234                  * If an event is already enabled, someone is using
4235                  * it and the self test should not be on.
4236                  */
4237                 if (file->flags & EVENT_FILE_FL_ENABLED) {
4238                         pr_warn("Enabled event during self test!\n");
4239                         WARN_ON_ONCE(1);
4240                         continue;
4241                 }
4242
4243                 ftrace_event_enable_disable(file, 1);
4244                 event_test_stuff();
4245                 ftrace_event_enable_disable(file, 0);
4246
4247                 pr_cont("OK\n");
4248         }
4249
4250         /* Now test at the sub system level */
4251
4252         pr_info("Running tests on trace event systems:\n");
4253
4254         list_for_each_entry(dir, &tr->systems, list) {
4255
4256                 system = dir->subsystem;
4257
4258                 /* the ftrace system is special, skip it */
4259                 if (strcmp(system->name, "ftrace") == 0)
4260                         continue;
4261
4262                 pr_info("Testing event system %s: ", system->name);
4263
4264                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4265                 if (WARN_ON_ONCE(ret)) {
4266                         pr_warn("error enabling system %s\n",
4267                                 system->name);
4268                         continue;
4269                 }
4270
4271                 event_test_stuff();
4272
4273                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4274                 if (WARN_ON_ONCE(ret)) {
4275                         pr_warn("error disabling system %s\n",
4276                                 system->name);
4277                         continue;
4278                 }
4279
4280                 pr_cont("OK\n");
4281         }
4282
4283         /* Test with all events enabled */
4284
4285         pr_info("Running tests on all trace events:\n");
4286         pr_info("Testing all events: ");
4287
4288         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4289         if (WARN_ON_ONCE(ret)) {
4290                 pr_warn("error enabling all events\n");
4291                 return;
4292         }
4293
4294         event_test_stuff();
4295
4296         /* reset sysname */
4297         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4298         if (WARN_ON_ONCE(ret)) {
4299                 pr_warn("error disabling all events\n");
4300                 return;
4301         }
4302
4303         pr_cont("OK\n");
4304 }
4305
4306 #ifdef CONFIG_FUNCTION_TRACER
4307
4308 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4309
4310 static struct trace_event_file event_trace_file __initdata;
4311
4312 static void __init
4313 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4314                           struct ftrace_ops *op, struct ftrace_regs *regs)
4315 {
4316         struct trace_buffer *buffer;
4317         struct ring_buffer_event *event;
4318         struct ftrace_entry *entry;
4319         unsigned int trace_ctx;
4320         long disabled;
4321         int cpu;
4322
4323         trace_ctx = tracing_gen_ctx();
4324         preempt_disable_notrace();
4325         cpu = raw_smp_processor_id();
4326         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4327
4328         if (disabled != 1)
4329                 goto out;
4330
4331         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4332                                                 TRACE_FN, sizeof(*entry),
4333                                                 trace_ctx);
4334         if (!event)
4335                 goto out;
4336         entry   = ring_buffer_event_data(event);
4337         entry->ip                       = ip;
4338         entry->parent_ip                = parent_ip;
4339
4340         event_trigger_unlock_commit(&event_trace_file, buffer, event,
4341                                     entry, trace_ctx);
4342  out:
4343         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4344         preempt_enable_notrace();
4345 }
4346
4347 static struct ftrace_ops trace_ops __initdata  =
4348 {
4349         .func = function_test_events_call,
4350 };
4351
4352 static __init void event_trace_self_test_with_function(void)
4353 {
4354         int ret;
4355
4356         event_trace_file.tr = top_trace_array();
4357         if (WARN_ON(!event_trace_file.tr))
4358                 return;
4359
4360         ret = register_ftrace_function(&trace_ops);
4361         if (WARN_ON(ret < 0)) {
4362                 pr_info("Failed to enable function tracer for event tests\n");
4363                 return;
4364         }
4365         pr_info("Running tests again, along with the function tracer\n");
4366         event_trace_self_tests();
4367         unregister_ftrace_function(&trace_ops);
4368 }
4369 #else
4370 static __init void event_trace_self_test_with_function(void)
4371 {
4372 }
4373 #endif
4374
4375 static __init int event_trace_self_tests_init(void)
4376 {
4377         if (!tracing_selftest_disabled) {
4378                 event_trace_self_tests();
4379                 event_trace_self_test_with_function();
4380         }
4381
4382         return 0;
4383 }
4384
4385 late_initcall(event_trace_self_tests_init);
4386
4387 #endif