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 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  44
  45 static struct kmem_cache *field_cachep;
  46 static struct kmem_cache *file_cachep;
  47
  48 static inline int system_refcount(struct event_subsystem *system)
  49 {
  50         return system->ref_count;
  51 }
  52
  53 static int system_refcount_inc(struct event_subsystem *system)
  54 {
  55         return system->ref_count++;
  56 }
  57
  58 static int system_refcount_dec(struct event_subsystem *system)
  59 {
  60         return --system->ref_count;
  61 }
  62
  63 /* Double loops, do not use break, only goto's work */
  64 #define do_for_each_event_file(tr, file)                        \
  65         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  66                 list_for_each_entry(file, &tr->events, list)
  67
  68 #define do_for_each_event_file_safe(tr, file)                   \
  69         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  70                 struct trace_event_file *___n;                          \
  71                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  72
  73 #define while_for_each_event_file()             \
  74         }
  75
  76 static struct ftrace_event_field *
  77 __find_event_field(struct list_head *head, char *name)
  78 {
  79         struct ftrace_event_field *field;
  80
  81         list_for_each_entry(field, head, link) {
  82                 if (!strcmp(field->name, name))
  83                         return field;
  84         }
  85
  86         return NULL;
  87 }
  88
  89 struct ftrace_event_field *
  90 trace_find_event_field(struct trace_event_call *call, char *name)
  91 {
  92         struct ftrace_event_field *field;
  93         struct list_head *head;
  94
  95         head = trace_get_fields(call);
  96         field = __find_event_field(head, name);
  97         if (field)
  98                 return field;
  99
 100         field = __find_event_field(&ftrace_generic_fields, name);
 101         if (field)
 102                 return field;
 103
 104         return __find_event_field(&ftrace_common_fields, name);
 105 }
 106
 107 static int __trace_define_field(struct list_head *head, const char *type,
 108                                 const char *name, int offset, int size,
 109                                 int is_signed, int filter_type)
 110 {
 111         struct ftrace_event_field *field;
 112
 113         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 114         if (!field)
 115                 return -ENOMEM;
 116
 117         field->name = name;
 118         field->type = type;
 119
 120         if (filter_type == FILTER_OTHER)
 121                 field->filter_type = filter_assign_type(type);
 122         else
 123                 field->filter_type = filter_type;
 124
 125         field->offset = offset;
 126         field->size = size;
 127         field->is_signed = is_signed;
 128
 129         list_add(&field->link, head);
 130
 131         return 0;
 132 }
 133
 134 int trace_define_field(struct trace_event_call *call, const char *type,
 135                        const char *name, int offset, int size, int is_signed,
 136                        int filter_type)
 137 {
 138         struct list_head *head;
 139
 140         if (WARN_ON(!call->class))
 141                 return 0;
 142
 143         head = trace_get_fields(call);
 144         return __trace_define_field(head, type, name, offset, size,
 145                                     is_signed, filter_type);
 146 }
 147 EXPORT_SYMBOL_GPL(trace_define_field);
 148
 149 #define __generic_field(type, item, filter_type)                        \
 150         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 151                                    #item, 0, 0, is_signed_type(type),   \
 152                                    filter_type);                        \
 153         if (ret)                                                        \
 154                 return ret;
 155
 156 #define __common_field(type, item)                                      \
 157         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 158                                    "common_" #item,                     \
 159                                    offsetof(typeof(ent), item),         \
 160                                    sizeof(ent.item),                    \
 161                                    is_signed_type(type), FILTER_OTHER); \
 162         if (ret)                                                        \
 163                 return ret;
 164
 165 static int trace_define_generic_fields(void)
 166 {
 167         int ret;
 168
 169         __generic_field(int, CPU, FILTER_CPU);
 170         __generic_field(int, cpu, FILTER_CPU);
 171         __generic_field(char *, COMM, FILTER_COMM);
 172         __generic_field(char *, comm, FILTER_COMM);
 173
 174         return ret;
 175 }
 176
 177 static int trace_define_common_fields(void)
 178 {
 179         int ret;
 180         struct trace_entry ent;
 181
 182         __common_field(unsigned short, type);
 183         __common_field(unsigned char, flags);
 184         __common_field(unsigned char, preempt_count);
 185         __common_field(int, pid);
 186
 187         return ret;
 188 }
 189
 190 static void trace_destroy_fields(struct trace_event_call *call)
 191 {
 192         struct ftrace_event_field *field, *next;
 193         struct list_head *head;
 194
 195         head = trace_get_fields(call);
 196         list_for_each_entry_safe(field, next, head, link) {
 197                 list_del(&field->link);
 198                 kmem_cache_free(field_cachep, field);
 199         }
 200 }
 201
 202 /*
 203  * run-time version of trace_event_get_offsets_<call>() that returns the last
 204  * accessible offset of trace fields excluding __dynamic_array bytes
 205  */
 206 int trace_event_get_offsets(struct trace_event_call *call)
 207 {
 208         struct ftrace_event_field *tail;
 209         struct list_head *head;
 210
 211         head = trace_get_fields(call);
 212         /*
 213          * head->next points to the last field with the largest offset,
 214          * since it was added last by trace_define_field()
 215          */
 216         tail = list_first_entry(head, struct ftrace_event_field, link);
 217         return tail->offset + tail->size;
 218 }
 219
 220 /*
 221  * Check if the referenced field is an array and return true,
 222  * as arrays are OK to dereference.
 223  */
 224 static bool test_field(const char *fmt, struct trace_event_call *call)
 225 {
 226         struct trace_event_fields *field = call->class->fields_array;
 227         const char *array_descriptor;
 228         const char *p = fmt;
 229         int len;
 230
 231         if (!(len = str_has_prefix(fmt, "REC->")))
 232                 return false;
 233         fmt += len;
 234         for (p = fmt; *p; p++) {
 235                 if (!isalnum(*p) && *p != '_')
 236                         break;
 237         }
 238         len = p - fmt;
 239
 240         for (; field->type; field++) {
 241                 if (strncmp(field->name, fmt, len) ||
 242                     field->name[len])
 243                         continue;
 244                 array_descriptor = strchr(field->type, '[');
 245                 /* This is an array and is OK to dereference. */
 246                 return array_descriptor != NULL;
 247         }
 248         return false;
 249 }
 250
 251 /*
 252  * Examine the print fmt of the event looking for unsafe dereference
 253  * pointers using %p* that could be recorded in the trace event and
 254  * much later referenced after the pointer was freed. Dereferencing
 255  * pointers are OK, if it is dereferenced into the event itself.
 256  */
 257 static void test_event_printk(struct trace_event_call *call)
 258 {
 259         u64 dereference_flags = 0;
 260         bool first = true;
 261         const char *fmt, *c, *r, *a;
 262         int parens = 0;
 263         char in_quote = 0;
 264         int start_arg = 0;
 265         int arg = 0;
 266         int i;
 267
 268         fmt = call->print_fmt;
 269
 270         if (!fmt)
 271                 return;
 272
 273         for (i = 0; fmt[i]; i++) {
 274                 switch (fmt[i]) {
 275                 case '\\':
 276                         i++;
 277                         if (!fmt[i])
 278                                 return;
 279                         continue;
 280                 case '"':
 281                 case '\'':
 282                         /*
 283                          * The print fmt starts with a string that
 284                          * is processed first to find %p* usage,
 285                          * then after the first string, the print fmt
 286                          * contains arguments that are used to check
 287                          * if the dereferenced %p* usage is safe.
 288                          */
 289                         if (first) {
 290                                 if (fmt[i] == '\'')
 291                                         continue;
 292                                 if (in_quote) {
 293                                         arg = 0;
 294                                         first = false;
 295                                         /*
 296                                          * If there was no %p* uses
 297                                          * the fmt is OK.
 298                                          */
 299                                         if (!dereference_flags)
 300                                                 return;
 301                                 }
 302                         }
 303                         if (in_quote) {
 304                                 if (in_quote == fmt[i])
 305                                         in_quote = 0;
 306                         } else {
 307                                 in_quote = fmt[i];
 308                         }
 309                         continue;
 310                 case '%':
 311                         if (!first || !in_quote)
 312                                 continue;
 313                         i++;
 314                         if (!fmt[i])
 315                                 return;
 316                         switch (fmt[i]) {
 317                         case '%':
 318                                 continue;
 319                         case 'p':
 320                                 /* Find dereferencing fields */
 321                                 switch (fmt[i + 1]) {
 322                                 case 'B': case 'R': case 'r':
 323                                 case 'b': case 'M': case 'm':
 324                                 case 'I': case 'i': case 'E':
 325                                 case 'U': case 'V': case 'N':
 326                                 case 'a': case 'd': case 'D':
 327                                 case 'g': case 't': case 'C':
 328                                 case 'O': case 'f':
 329                                         if (WARN_ONCE(arg == 63,
 330                                                       "Too many args for event: %s",
 331                                                       trace_event_name(call)))
 332                                                 return;
 333                                         dereference_flags |= 1ULL << arg;
 334                                 }
 335                                 break;
 336                         default:
 337                         {
 338                                 bool star = false;
 339                                 int j;
 340
 341                                 /* Increment arg if %*s exists. */
 342                                 for (j = 0; fmt[i + j]; j++) {
 343                                         if (isdigit(fmt[i + j]) ||
 344                                             fmt[i + j] == '.')
 345                                                 continue;
 346                                         if (fmt[i + j] == '*') {
 347                                                 star = true;
 348                                                 continue;
 349                                         }
 350                                         if ((fmt[i + j] == 's') && star)
 351                                                 arg++;
 352                                         break;
 353                                 }
 354                                 break;
 355                         } /* default */
 356
 357                         } /* switch */
 358                         arg++;
 359                         continue;
 360                 case '(':
 361                         if (in_quote)
 362                                 continue;
 363                         parens++;
 364                         continue;
 365                 case ')':
 366                         if (in_quote)
 367                                 continue;
 368                         parens--;
 369                         if (WARN_ONCE(parens < 0,
 370                                       "Paren mismatch for event: %s\narg='%s'\n%*s",
 371                                       trace_event_name(call),
 372                                       fmt + start_arg,
 373                                       (i - start_arg) + 5, "^"))
 374                                 return;
 375                         continue;
 376                 case ',':
 377                         if (in_quote || parens)
 378                                 continue;
 379                         i++;
 380                         while (isspace(fmt[i]))
 381                                 i++;
 382                         start_arg = i;
 383                         if (!(dereference_flags & (1ULL << arg)))
 384                                 goto next_arg;
 385
 386                         /* Find the REC-> in the argument */
 387                         c = strchr(fmt + i, ',');
 388                         r = strstr(fmt + i, "REC->");
 389                         if (r && (!c || r < c)) {
 390                                 /*
 391                                  * Addresses of events on the buffer,
 392                                  * or an array on the buffer is
 393                                  * OK to dereference.
 394                                  * There's ways to fool this, but
 395                                  * this is to catch common mistakes,
 396                                  * not malicious code.
 397                                  */
 398                                 a = strchr(fmt + i, '&');
 399                                 if ((a && (a < r)) || test_field(r, call))
 400                                         dereference_flags &= ~(1ULL << arg);
 401                         }
 402                 next_arg:
 403                         i--;
 404                         arg++;
 405                 }
 406         }
 407
 408         /*
 409          * If you triggered the below warning, the trace event reported
 410          * uses an unsafe dereference pointer %p*. As the data stored
 411          * at the trace event time may no longer exist when the trace
 412          * event is printed, dereferencing to the original source is
 413          * unsafe. The source of the dereference must be copied into the
 414          * event itself, and the dereference must access the copy instead.
 415          */
 416         if (WARN_ON_ONCE(dereference_flags)) {
 417                 arg = 1;
 418                 while (!(dereference_flags & 1)) {
 419                         dereference_flags >>= 1;
 420                         arg++;
 421                 }
 422                 pr_warn("event %s has unsafe dereference of argument %d\n",
 423                         trace_event_name(call), arg);
 424                 pr_warn("print_fmt: %s\n", fmt);
 425         }
 426 }
 427
 428 int trace_event_raw_init(struct trace_event_call *call)
 429 {
 430         int id;
 431
 432         id = register_trace_event(&call->event);
 433         if (!id)
 434                 return -ENODEV;
 435
 436         test_event_printk(call);
 437
 438         return 0;
 439 }
 440 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 441
 442 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 443 {
 444         struct trace_array *tr = trace_file->tr;
 445         struct trace_array_cpu *data;
 446         struct trace_pid_list *no_pid_list;
 447         struct trace_pid_list *pid_list;
 448
 449         pid_list = rcu_dereference_raw(tr->filtered_pids);
 450         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 451
 452         if (!pid_list && !no_pid_list)
 453                 return false;
 454
 455         data = this_cpu_ptr(tr->array_buffer.data);
 456
 457         return data->ignore_pid;
 458 }
 459 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 460
 461 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 462                                  struct trace_event_file *trace_file,
 463                                  unsigned long len)
 464 {
 465         struct trace_event_call *event_call = trace_file->event_call;
 466
 467         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 468             trace_event_ignore_this_pid(trace_file))
 469                 return NULL;
 470
 471         /*
 472          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 473          * preemption (adding one to the preempt_count). Since we are
 474          * interested in the preempt_count at the time the tracepoint was
 475          * hit, we need to subtract one to offset the increment.
 476          */
 477         fbuffer->trace_ctx = tracing_gen_ctx_dec();
 478         fbuffer->trace_file = trace_file;
 479
 480         fbuffer->event =
 481                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 482                                                 event_call->event.type, len,
 483                                                 fbuffer->trace_ctx);
 484         if (!fbuffer->event)
 485                 return NULL;
 486
 487         fbuffer->regs = NULL;
 488         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 489         return fbuffer->entry;
 490 }
 491 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 492
 493 int trace_event_reg(struct trace_event_call *call,
 494                     enum trace_reg type, void *data)
 495 {
 496         struct trace_event_file *file = data;
 497
 498         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 499         switch (type) {
 500         case TRACE_REG_REGISTER:
 501                 return tracepoint_probe_register(call->tp,
 502                                                  call->class->probe,
 503                                                  file);
 504         case TRACE_REG_UNREGISTER:
 505                 tracepoint_probe_unregister(call->tp,
 506                                             call->class->probe,
 507                                             file);
 508                 return 0;
 509
 510 #ifdef CONFIG_PERF_EVENTS
 511         case TRACE_REG_PERF_REGISTER:
 512                 return tracepoint_probe_register(call->tp,
 513                                                  call->class->perf_probe,
 514                                                  call);
 515         case TRACE_REG_PERF_UNREGISTER:
 516                 tracepoint_probe_unregister(call->tp,
 517                                             call->class->perf_probe,
 518                                             call);
 519                 return 0;
 520         case TRACE_REG_PERF_OPEN:
 521         case TRACE_REG_PERF_CLOSE:
 522         case TRACE_REG_PERF_ADD:
 523         case TRACE_REG_PERF_DEL:
 524                 return 0;
 525 #endif
 526         }
 527         return 0;
 528 }
 529 EXPORT_SYMBOL_GPL(trace_event_reg);
 530
 531 void trace_event_enable_cmd_record(bool enable)
 532 {
 533         struct trace_event_file *file;
 534         struct trace_array *tr;
 535
 536         lockdep_assert_held(&event_mutex);
 537
 538         do_for_each_event_file(tr, file) {
 539
 540                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 541                         continue;
 542
 543                 if (enable) {
 544                         tracing_start_cmdline_record();
 545                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 546                 } else {
 547                         tracing_stop_cmdline_record();
 548                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 549                 }
 550         } while_for_each_event_file();
 551 }
 552
 553 void trace_event_enable_tgid_record(bool enable)
 554 {
 555         struct trace_event_file *file;
 556         struct trace_array *tr;
 557
 558         lockdep_assert_held(&event_mutex);
 559
 560         do_for_each_event_file(tr, file) {
 561                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 562                         continue;
 563
 564                 if (enable) {
 565                         tracing_start_tgid_record();
 566                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 567                 } else {
 568                         tracing_stop_tgid_record();
 569                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 570                                   &file->flags);
 571                 }
 572         } while_for_each_event_file();
 573 }
 574
 575 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 576                                          int enable, int soft_disable)
 577 {
 578         struct trace_event_call *call = file->event_call;
 579         struct trace_array *tr = file->tr;
 580         unsigned long file_flags = file->flags;
 581         int ret = 0;
 582         int disable;
 583
 584         switch (enable) {
 585         case 0:
 586                 /*
 587                  * When soft_disable is set and enable is cleared, the sm_ref
 588                  * reference counter is decremented. If it reaches 0, we want
 589                  * to clear the SOFT_DISABLED flag but leave the event in the
 590                  * state that it was. That is, if the event was enabled and
 591                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 592                  * is set we do not want the event to be enabled before we
 593                  * clear the bit.
 594                  *
 595                  * When soft_disable is not set but the SOFT_MODE flag is,
 596                  * we do nothing. Do not disable the tracepoint, otherwise
 597                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 598                  */
 599                 if (soft_disable) {
 600                         if (atomic_dec_return(&file->sm_ref) > 0)
 601                                 break;
 602                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 603                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 604                 } else
 605                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 606
 607                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 608                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 609                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 610                                 tracing_stop_cmdline_record();
 611                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 612                         }
 613
 614                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 615                                 tracing_stop_tgid_record();
 616                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 617                         }
 618
 619                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 620                 }
 621                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 622                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 623                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 624                 else
 625                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 626                 break;
 627         case 1:
 628                 /*
 629                  * When soft_disable is set and enable is set, we want to
 630                  * register the tracepoint for the event, but leave the event
 631                  * as is. That means, if the event was already enabled, we do
 632                  * nothing (but set SOFT_MODE). If the event is disabled, we
 633                  * set SOFT_DISABLED before enabling the event tracepoint, so
 634                  * it still seems to be disabled.
 635                  */
 636                 if (!soft_disable)
 637                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 638                 else {
 639                         if (atomic_inc_return(&file->sm_ref) > 1)
 640                                 break;
 641                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 642                 }
 643
 644                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 645                         bool cmd = false, tgid = false;
 646
 647                         /* Keep the event disabled, when going to SOFT_MODE. */
 648                         if (soft_disable)
 649                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 650
 651                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 652                                 cmd = true;
 653                                 tracing_start_cmdline_record();
 654                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 655                         }
 656
 657                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 658                                 tgid = true;
 659                                 tracing_start_tgid_record();
 660                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 661                         }
 662
 663                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 664                         if (ret) {
 665                                 if (cmd)
 666                                         tracing_stop_cmdline_record();
 667                                 if (tgid)
 668                                         tracing_stop_tgid_record();
 669                                 pr_info("event trace: Could not enable event "
 670                                         "%s\n", trace_event_name(call));
 671                                 break;
 672                         }
 673                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 674
 675                         /* WAS_ENABLED gets set but never cleared. */
 676                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 677                 }
 678                 break;
 679         }
 680
 681         /* Enable or disable use of trace_buffered_event */
 682         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 683             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 684                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 685                         trace_buffered_event_enable();
 686                 else
 687                         trace_buffered_event_disable();
 688         }
 689
 690         return ret;
 691 }
 692
 693 int trace_event_enable_disable(struct trace_event_file *file,
 694                                int enable, int soft_disable)
 695 {
 696         return __ftrace_event_enable_disable(file, enable, soft_disable);
 697 }
 698
 699 static int ftrace_event_enable_disable(struct trace_event_file *file,
 700                                        int enable)
 701 {
 702         return __ftrace_event_enable_disable(file, enable, 0);
 703 }
 704
 705 static void ftrace_clear_events(struct trace_array *tr)
 706 {
 707         struct trace_event_file *file;
 708
 709         mutex_lock(&event_mutex);
 710         list_for_each_entry(file, &tr->events, list) {
 711                 ftrace_event_enable_disable(file, 0);
 712         }
 713         mutex_unlock(&event_mutex);
 714 }
 715
 716 static void
 717 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 718 {
 719         struct trace_pid_list *pid_list;
 720         struct trace_array *tr = data;
 721
 722         pid_list = rcu_dereference_raw(tr->filtered_pids);
 723         trace_filter_add_remove_task(pid_list, NULL, task);
 724
 725         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 726         trace_filter_add_remove_task(pid_list, NULL, task);
 727 }
 728
 729 static void
 730 event_filter_pid_sched_process_fork(void *data,
 731                                     struct task_struct *self,
 732                                     struct task_struct *task)
 733 {
 734         struct trace_pid_list *pid_list;
 735         struct trace_array *tr = data;
 736
 737         pid_list = rcu_dereference_sched(tr->filtered_pids);
 738         trace_filter_add_remove_task(pid_list, self, task);
 739
 740         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 741         trace_filter_add_remove_task(pid_list, self, task);
 742 }
 743
 744 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 745 {
 746         if (enable) {
 747                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 748                                                        tr, INT_MIN);
 749                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 750                                                        tr, INT_MAX);
 751         } else {
 752                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 753                                                     tr);
 754                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 755                                                     tr);
 756         }
 757 }
 758
 759 static void
 760 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 761                     struct task_struct *prev, struct task_struct *next)
 762 {
 763         struct trace_array *tr = data;
 764         struct trace_pid_list *no_pid_list;
 765         struct trace_pid_list *pid_list;
 766         bool ret;
 767
 768         pid_list = rcu_dereference_sched(tr->filtered_pids);
 769         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 770
 771         /*
 772          * Sched switch is funny, as we only want to ignore it
 773          * in the notrace case if both prev and next should be ignored.
 774          */
 775         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 776                 trace_ignore_this_task(NULL, no_pid_list, next);
 777
 778         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 779                        (trace_ignore_this_task(pid_list, NULL, prev) &&
 780                         trace_ignore_this_task(pid_list, NULL, next)));
 781 }
 782
 783 static void
 784 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 785                     struct task_struct *prev, struct task_struct *next)
 786 {
 787         struct trace_array *tr = data;
 788         struct trace_pid_list *no_pid_list;
 789         struct trace_pid_list *pid_list;
 790
 791         pid_list = rcu_dereference_sched(tr->filtered_pids);
 792         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 793
 794         this_cpu_write(tr->array_buffer.data->ignore_pid,
 795                        trace_ignore_this_task(pid_list, no_pid_list, next));
 796 }
 797
 798 static void
 799 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 800 {
 801         struct trace_array *tr = data;
 802         struct trace_pid_list *no_pid_list;
 803         struct trace_pid_list *pid_list;
 804
 805         /* Nothing to do if we are already tracing */
 806         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 807                 return;
 808
 809         pid_list = rcu_dereference_sched(tr->filtered_pids);
 810         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 811
 812         this_cpu_write(tr->array_buffer.data->ignore_pid,
 813                        trace_ignore_this_task(pid_list, no_pid_list, task));
 814 }
 815
 816 static void
 817 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 818 {
 819         struct trace_array *tr = data;
 820         struct trace_pid_list *no_pid_list;
 821         struct trace_pid_list *pid_list;
 822
 823         /* Nothing to do if we are not tracing */
 824         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 825                 return;
 826
 827         pid_list = rcu_dereference_sched(tr->filtered_pids);
 828         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 829
 830         /* Set tracing if current is enabled */
 831         this_cpu_write(tr->array_buffer.data->ignore_pid,
 832                        trace_ignore_this_task(pid_list, no_pid_list, current));
 833 }
 834
 835 static void unregister_pid_events(struct trace_array *tr)
 836 {
 837         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 838         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 839
 840         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 841         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 842
 843         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 844         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 845
 846         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 847         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 848 }
 849
 850 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 851 {
 852         struct trace_pid_list *pid_list;
 853         struct trace_pid_list *no_pid_list;
 854         struct trace_event_file *file;
 855         int cpu;
 856
 857         pid_list = rcu_dereference_protected(tr->filtered_pids,
 858                                              lockdep_is_held(&event_mutex));
 859         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 860                                              lockdep_is_held(&event_mutex));
 861
 862         /* Make sure there's something to do */
 863         if (!pid_type_enabled(type, pid_list, no_pid_list))
 864                 return;
 865
 866         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 867                 unregister_pid_events(tr);
 868
 869                 list_for_each_entry(file, &tr->events, list) {
 870                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 871                 }
 872
 873                 for_each_possible_cpu(cpu)
 874                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 875         }
 876
 877         if (type & TRACE_PIDS)
 878                 rcu_assign_pointer(tr->filtered_pids, NULL);
 879
 880         if (type & TRACE_NO_PIDS)
 881                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
 882
 883         /* Wait till all users are no longer using pid filtering */
 884         tracepoint_synchronize_unregister();
 885
 886         if ((type & TRACE_PIDS) && pid_list)
 887                 trace_free_pid_list(pid_list);
 888
 889         if ((type & TRACE_NO_PIDS) && no_pid_list)
 890                 trace_free_pid_list(no_pid_list);
 891 }
 892
 893 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 894 {
 895         mutex_lock(&event_mutex);
 896         __ftrace_clear_event_pids(tr, type);
 897         mutex_unlock(&event_mutex);
 898 }
 899
 900 static void __put_system(struct event_subsystem *system)
 901 {
 902         struct event_filter *filter = system->filter;
 903
 904         WARN_ON_ONCE(system_refcount(system) == 0);
 905         if (system_refcount_dec(system))
 906                 return;
 907
 908         list_del(&system->list);
 909
 910         if (filter) {
 911                 kfree(filter->filter_string);
 912                 kfree(filter);
 913         }
 914         kfree_const(system->name);
 915         kfree(system);
 916 }
 917
 918 static void __get_system(struct event_subsystem *system)
 919 {
 920         WARN_ON_ONCE(system_refcount(system) == 0);
 921         system_refcount_inc(system);
 922 }
 923
 924 static void __get_system_dir(struct trace_subsystem_dir *dir)
 925 {
 926         WARN_ON_ONCE(dir->ref_count == 0);
 927         dir->ref_count++;
 928         __get_system(dir->subsystem);
 929 }
 930
 931 static void __put_system_dir(struct trace_subsystem_dir *dir)
 932 {
 933         WARN_ON_ONCE(dir->ref_count == 0);
 934         /* If the subsystem is about to be freed, the dir must be too */
 935         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 936
 937         __put_system(dir->subsystem);
 938         if (!--dir->ref_count)
 939                 kfree(dir);
 940 }
 941
 942 static void put_system(struct trace_subsystem_dir *dir)
 943 {
 944         mutex_lock(&event_mutex);
 945         __put_system_dir(dir);
 946         mutex_unlock(&event_mutex);
 947 }
 948
 949 static void remove_subsystem(struct trace_subsystem_dir *dir)
 950 {
 951         if (!dir)
 952                 return;
 953
 954         if (!--dir->nr_events) {
 955                 tracefs_remove(dir->entry);
 956                 list_del(&dir->list);
 957                 __put_system_dir(dir);
 958         }
 959 }
 960
 961 static void remove_event_file_dir(struct trace_event_file *file)
 962 {
 963         struct dentry *dir = file->dir;
 964         struct dentry *child;
 965
 966         if (dir) {
 967                 spin_lock(&dir->d_lock);        /* probably unneeded */
 968                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 969                         if (d_really_is_positive(child))        /* probably unneeded */
 970                                 d_inode(child)->i_private = NULL;
 971                 }
 972                 spin_unlock(&dir->d_lock);
 973
 974                 tracefs_remove(dir);
 975         }
 976
 977         list_del(&file->list);
 978         remove_subsystem(file->system);
 979         free_event_filter(file->filter);
 980         kmem_cache_free(file_cachep, file);
 981 }
 982
 983 /*
 984  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 985  */
 986 static int
 987 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 988                               const char *sub, const char *event, int set)
 989 {
 990         struct trace_event_file *file;
 991         struct trace_event_call *call;
 992         const char *name;
 993         int ret = -EINVAL;
 994         int eret = 0;
 995
 996         list_for_each_entry(file, &tr->events, list) {
 997
 998                 call = file->event_call;
 999                 name = trace_event_name(call);
1000
1001                 if (!name || !call->class || !call->class->reg)
1002                         continue;
1003
1004                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1005                         continue;
1006
1007                 if (match &&
1008                     strcmp(match, name) != 0 &&
1009                     strcmp(match, call->class->system) != 0)
1010                         continue;
1011
1012                 if (sub && strcmp(sub, call->class->system) != 0)
1013                         continue;
1014
1015                 if (event && strcmp(event, name) != 0)
1016                         continue;
1017
1018                 ret = ftrace_event_enable_disable(file, set);
1019
1020                 /*
1021                  * Save the first error and return that. Some events
1022                  * may still have been enabled, but let the user
1023                  * know that something went wrong.
1024                  */
1025                 if (ret && !eret)
1026                         eret = ret;
1027
1028                 ret = eret;
1029         }
1030
1031         return ret;
1032 }
1033
1034 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1035                                   const char *sub, const char *event, int set)
1036 {
1037         int ret;
1038
1039         mutex_lock(&event_mutex);
1040         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1041         mutex_unlock(&event_mutex);
1042
1043         return ret;
1044 }
1045
1046 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1047 {
1048         char *event = NULL, *sub = NULL, *match;
1049         int ret;
1050
1051         if (!tr)
1052                 return -ENOENT;
1053         /*
1054          * The buf format can be <subsystem>:<event-name>
1055          *  *:<event-name> means any event by that name.
1056          *  :<event-name> is the same.
1057          *
1058          *  <subsystem>:* means all events in that subsystem
1059          *  <subsystem>: means the same.
1060          *
1061          *  <name> (no ':') means all events in a subsystem with
1062          *  the name <name> or any event that matches <name>
1063          */
1064
1065         match = strsep(&buf, ":");
1066         if (buf) {
1067                 sub = match;
1068                 event = buf;
1069                 match = NULL;
1070
1071                 if (!strlen(sub) || strcmp(sub, "*") == 0)
1072                         sub = NULL;
1073                 if (!strlen(event) || strcmp(event, "*") == 0)
1074                         event = NULL;
1075         }
1076
1077         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1078
1079         /* Put back the colon to allow this to be called again */
1080         if (buf)
1081                 *(buf - 1) = ':';
1082
1083         return ret;
1084 }
1085
1086 /**
1087  * trace_set_clr_event - enable or disable an event
1088  * @system: system name to match (NULL for any system)
1089  * @event: event name to match (NULL for all events, within system)
1090  * @set: 1 to enable, 0 to disable
1091  *
1092  * This is a way for other parts of the kernel to enable or disable
1093  * event recording.
1094  *
1095  * Returns 0 on success, -EINVAL if the parameters do not match any
1096  * registered events.
1097  */
1098 int trace_set_clr_event(const char *system, const char *event, int set)
1099 {
1100         struct trace_array *tr = top_trace_array();
1101
1102         if (!tr)
1103                 return -ENODEV;
1104
1105         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1106 }
1107 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1108
1109 /**
1110  * trace_array_set_clr_event - enable or disable an event for a trace array.
1111  * @tr: concerned trace array.
1112  * @system: system name to match (NULL for any system)
1113  * @event: event name to match (NULL for all events, within system)
1114  * @enable: true to enable, false to disable
1115  *
1116  * This is a way for other parts of the kernel to enable or disable
1117  * event recording.
1118  *
1119  * Returns 0 on success, -EINVAL if the parameters do not match any
1120  * registered events.
1121  */
1122 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1123                 const char *event, bool enable)
1124 {
1125         int set;
1126
1127         if (!tr)
1128                 return -ENOENT;
1129
1130         set = (enable == true) ? 1 : 0;
1131         return __ftrace_set_clr_event(tr, NULL, system, event, set);
1132 }
1133 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1134
1135 /* 128 should be much more than enough */
1136 #define EVENT_BUF_SIZE          127
1137
1138 static ssize_t
1139 ftrace_event_write(struct file *file, const char __user *ubuf,
1140                    size_t cnt, loff_t *ppos)
1141 {
1142         struct trace_parser parser;
1143         struct seq_file *m = file->private_data;
1144         struct trace_array *tr = m->private;
1145         ssize_t read, ret;
1146
1147         if (!cnt)
1148                 return 0;
1149
1150         ret = tracing_update_buffers();
1151         if (ret < 0)
1152                 return ret;
1153
1154         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1155                 return -ENOMEM;
1156
1157         read = trace_get_user(&parser, ubuf, cnt, ppos);
1158
1159         if (read >= 0 && trace_parser_loaded((&parser))) {
1160                 int set = 1;
1161
1162                 if (*parser.buffer == '!')
1163                         set = 0;
1164
1165                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1166                 if (ret)
1167                         goto out_put;
1168         }
1169
1170         ret = read;
1171
1172  out_put:
1173         trace_parser_put(&parser);
1174
1175         return ret;
1176 }
1177
1178 static void *
1179 t_next(struct seq_file *m, void *v, loff_t *pos)
1180 {
1181         struct trace_event_file *file = v;
1182         struct trace_event_call *call;
1183         struct trace_array *tr = m->private;
1184
1185         (*pos)++;
1186
1187         list_for_each_entry_continue(file, &tr->events, list) {
1188                 call = file->event_call;
1189                 /*
1190                  * The ftrace subsystem is for showing formats only.
1191                  * They can not be enabled or disabled via the event files.
1192                  */
1193                 if (call->class && call->class->reg &&
1194                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1195                         return file;
1196         }
1197
1198         return NULL;
1199 }
1200
1201 static void *t_start(struct seq_file *m, loff_t *pos)
1202 {
1203         struct trace_event_file *file;
1204         struct trace_array *tr = m->private;
1205         loff_t l;
1206
1207         mutex_lock(&event_mutex);
1208
1209         file = list_entry(&tr->events, struct trace_event_file, list);
1210         for (l = 0; l <= *pos; ) {
1211                 file = t_next(m, file, &l);
1212                 if (!file)
1213                         break;
1214         }
1215         return file;
1216 }
1217
1218 static void *
1219 s_next(struct seq_file *m, void *v, loff_t *pos)
1220 {
1221         struct trace_event_file *file = v;
1222         struct trace_array *tr = m->private;
1223
1224         (*pos)++;
1225
1226         list_for_each_entry_continue(file, &tr->events, list) {
1227                 if (file->flags & EVENT_FILE_FL_ENABLED)
1228                         return file;
1229         }
1230
1231         return NULL;
1232 }
1233
1234 static void *s_start(struct seq_file *m, loff_t *pos)
1235 {
1236         struct trace_event_file *file;
1237         struct trace_array *tr = m->private;
1238         loff_t l;
1239
1240         mutex_lock(&event_mutex);
1241
1242         file = list_entry(&tr->events, struct trace_event_file, list);
1243         for (l = 0; l <= *pos; ) {
1244                 file = s_next(m, file, &l);
1245                 if (!file)
1246                         break;
1247         }
1248         return file;
1249 }
1250
1251 static int t_show(struct seq_file *m, void *v)
1252 {
1253         struct trace_event_file *file = v;
1254         struct trace_event_call *call = file->event_call;
1255
1256         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1257                 seq_printf(m, "%s:", call->class->system);
1258         seq_printf(m, "%s\n", trace_event_name(call));
1259
1260         return 0;
1261 }
1262
1263 static void t_stop(struct seq_file *m, void *p)
1264 {
1265         mutex_unlock(&event_mutex);
1266 }
1267
1268 static void *
1269 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1270 {
1271         struct trace_array *tr = m->private;
1272         struct trace_pid_list *pid_list;
1273
1274         if (type == TRACE_PIDS)
1275                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1276         else
1277                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1278
1279         return trace_pid_next(pid_list, v, pos);
1280 }
1281
1282 static void *
1283 p_next(struct seq_file *m, void *v, loff_t *pos)
1284 {
1285         return __next(m, v, pos, TRACE_PIDS);
1286 }
1287
1288 static void *
1289 np_next(struct seq_file *m, void *v, loff_t *pos)
1290 {
1291         return __next(m, v, pos, TRACE_NO_PIDS);
1292 }
1293
1294 static void *__start(struct seq_file *m, loff_t *pos, int type)
1295         __acquires(RCU)
1296 {
1297         struct trace_pid_list *pid_list;
1298         struct trace_array *tr = m->private;
1299
1300         /*
1301          * Grab the mutex, to keep calls to p_next() having the same
1302          * tr->filtered_pids as p_start() has.
1303          * If we just passed the tr->filtered_pids around, then RCU would
1304          * have been enough, but doing that makes things more complex.
1305          */
1306         mutex_lock(&event_mutex);
1307         rcu_read_lock_sched();
1308
1309         if (type == TRACE_PIDS)
1310                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1311         else
1312                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1313
1314         if (!pid_list)
1315                 return NULL;
1316
1317         return trace_pid_start(pid_list, pos);
1318 }
1319
1320 static void *p_start(struct seq_file *m, loff_t *pos)
1321         __acquires(RCU)
1322 {
1323         return __start(m, pos, TRACE_PIDS);
1324 }
1325
1326 static void *np_start(struct seq_file *m, loff_t *pos)
1327         __acquires(RCU)
1328 {
1329         return __start(m, pos, TRACE_NO_PIDS);
1330 }
1331
1332 static void p_stop(struct seq_file *m, void *p)
1333         __releases(RCU)
1334 {
1335         rcu_read_unlock_sched();
1336         mutex_unlock(&event_mutex);
1337 }
1338
1339 static ssize_t
1340 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1341                   loff_t *ppos)
1342 {
1343         struct trace_event_file *file;
1344         unsigned long flags;
1345         char buf[4] = "0";
1346
1347         mutex_lock(&event_mutex);
1348         file = event_file_data(filp);
1349         if (likely(file))
1350                 flags = file->flags;
1351         mutex_unlock(&event_mutex);
1352
1353         if (!file)
1354                 return -ENODEV;
1355
1356         if (flags & EVENT_FILE_FL_ENABLED &&
1357             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1358                 strcpy(buf, "1");
1359
1360         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1361             flags & EVENT_FILE_FL_SOFT_MODE)
1362                 strcat(buf, "*");
1363
1364         strcat(buf, "\n");
1365
1366         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1367 }
1368
1369 static ssize_t
1370 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1371                    loff_t *ppos)
1372 {
1373         struct trace_event_file *file;
1374         unsigned long val;
1375         int ret;
1376
1377         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1378         if (ret)
1379                 return ret;
1380
1381         ret = tracing_update_buffers();
1382         if (ret < 0)
1383                 return ret;
1384
1385         switch (val) {
1386         case 0:
1387         case 1:
1388                 ret = -ENODEV;
1389                 mutex_lock(&event_mutex);
1390                 file = event_file_data(filp);
1391                 if (likely(file))
1392                         ret = ftrace_event_enable_disable(file, val);
1393                 mutex_unlock(&event_mutex);
1394                 break;
1395
1396         default:
1397                 return -EINVAL;
1398         }
1399
1400         *ppos += cnt;
1401
1402         return ret ? ret : cnt;
1403 }
1404
1405 static ssize_t
1406 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1407                    loff_t *ppos)
1408 {
1409         const char set_to_char[4] = { '?', '0', '1', 'X' };
1410         struct trace_subsystem_dir *dir = filp->private_data;
1411         struct event_subsystem *system = dir->subsystem;
1412         struct trace_event_call *call;
1413         struct trace_event_file *file;
1414         struct trace_array *tr = dir->tr;
1415         char buf[2];
1416         int set = 0;
1417         int ret;
1418
1419         mutex_lock(&event_mutex);
1420         list_for_each_entry(file, &tr->events, list) {
1421                 call = file->event_call;
1422                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1423                     !trace_event_name(call) || !call->class || !call->class->reg)
1424                         continue;
1425
1426                 if (system && strcmp(call->class->system, system->name) != 0)
1427                         continue;
1428
1429                 /*
1430                  * We need to find out if all the events are set
1431                  * or if all events or cleared, or if we have
1432                  * a mixture.
1433                  */
1434                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1435
1436                 /*
1437                  * If we have a mixture, no need to look further.
1438                  */
1439                 if (set == 3)
1440                         break;
1441         }
1442         mutex_unlock(&event_mutex);
1443
1444         buf[0] = set_to_char[set];
1445         buf[1] = '\n';
1446
1447         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1448
1449         return ret;
1450 }
1451
1452 static ssize_t
1453 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1454                     loff_t *ppos)
1455 {
1456         struct trace_subsystem_dir *dir = filp->private_data;
1457         struct event_subsystem *system = dir->subsystem;
1458         const char *name = NULL;
1459         unsigned long val;
1460         ssize_t ret;
1461
1462         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1463         if (ret)
1464                 return ret;
1465
1466         ret = tracing_update_buffers();
1467         if (ret < 0)
1468                 return ret;
1469
1470         if (val != 0 && val != 1)
1471                 return -EINVAL;
1472
1473         /*
1474          * Opening of "enable" adds a ref count to system,
1475          * so the name is safe to use.
1476          */
1477         if (system)
1478                 name = system->name;
1479
1480         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1481         if (ret)
1482                 goto out;
1483
1484         ret = cnt;
1485
1486 out:
1487         *ppos += cnt;
1488
1489         return ret;
1490 }
1491
1492 enum {
1493         FORMAT_HEADER           = 1,
1494         FORMAT_FIELD_SEPERATOR  = 2,
1495         FORMAT_PRINTFMT         = 3,
1496 };
1497
1498 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1499 {
1500         struct trace_event_call *call = event_file_data(m->private);
1501         struct list_head *common_head = &ftrace_common_fields;
1502         struct list_head *head = trace_get_fields(call);
1503         struct list_head *node = v;
1504
1505         (*pos)++;
1506
1507         switch ((unsigned long)v) {
1508         case FORMAT_HEADER:
1509                 node = common_head;
1510                 break;
1511
1512         case FORMAT_FIELD_SEPERATOR:
1513                 node = head;
1514                 break;
1515
1516         case FORMAT_PRINTFMT:
1517                 /* all done */
1518                 return NULL;
1519         }
1520
1521         node = node->prev;
1522         if (node == common_head)
1523                 return (void *)FORMAT_FIELD_SEPERATOR;
1524         else if (node == head)
1525                 return (void *)FORMAT_PRINTFMT;
1526         else
1527                 return node;
1528 }
1529
1530 static int f_show(struct seq_file *m, void *v)
1531 {
1532         struct trace_event_call *call = event_file_data(m->private);
1533         struct ftrace_event_field *field;
1534         const char *array_descriptor;
1535
1536         switch ((unsigned long)v) {
1537         case FORMAT_HEADER:
1538                 seq_printf(m, "name: %s\n", trace_event_name(call));
1539                 seq_printf(m, "ID: %d\n", call->event.type);
1540                 seq_puts(m, "format:\n");
1541                 return 0;
1542
1543         case FORMAT_FIELD_SEPERATOR:
1544                 seq_putc(m, '\n');
1545                 return 0;
1546
1547         case FORMAT_PRINTFMT:
1548                 seq_printf(m, "\nprint fmt: %s\n",
1549                            call->print_fmt);
1550                 return 0;
1551         }
1552
1553         field = list_entry(v, struct ftrace_event_field, link);
1554         /*
1555          * Smartly shows the array type(except dynamic array).
1556          * Normal:
1557          *      field:TYPE VAR
1558          * If TYPE := TYPE[LEN], it is shown:
1559          *      field:TYPE VAR[LEN]
1560          */
1561         array_descriptor = strchr(field->type, '[');
1562
1563         if (str_has_prefix(field->type, "__data_loc"))
1564                 array_descriptor = NULL;
1565
1566         if (!array_descriptor)
1567                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1568                            field->type, field->name, field->offset,
1569                            field->size, !!field->is_signed);
1570         else
1571                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1572                            (int)(array_descriptor - field->type),
1573                            field->type, field->name,
1574                            array_descriptor, field->offset,
1575                            field->size, !!field->is_signed);
1576
1577         return 0;
1578 }
1579
1580 static void *f_start(struct seq_file *m, loff_t *pos)
1581 {
1582         void *p = (void *)FORMAT_HEADER;
1583         loff_t l = 0;
1584
1585         /* ->stop() is called even if ->start() fails */
1586         mutex_lock(&event_mutex);
1587         if (!event_file_data(m->private))
1588                 return ERR_PTR(-ENODEV);
1589
1590         while (l < *pos && p)
1591                 p = f_next(m, p, &l);
1592
1593         return p;
1594 }
1595
1596 static void f_stop(struct seq_file *m, void *p)
1597 {
1598         mutex_unlock(&event_mutex);
1599 }
1600
1601 static const struct seq_operations trace_format_seq_ops = {
1602         .start          = f_start,
1603         .next           = f_next,
1604         .stop           = f_stop,
1605         .show           = f_show,
1606 };
1607
1608 static int trace_format_open(struct inode *inode, struct file *file)
1609 {
1610         struct seq_file *m;
1611         int ret;
1612
1613         /* Do we want to hide event format files on tracefs lockdown? */
1614
1615         ret = seq_open(file, &trace_format_seq_ops);
1616         if (ret < 0)
1617                 return ret;
1618
1619         m = file->private_data;
1620         m->private = file;
1621
1622         return 0;
1623 }
1624
1625 static ssize_t
1626 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1627 {
1628         int id = (long)event_file_data(filp);
1629         char buf[32];
1630         int len;
1631
1632         if (unlikely(!id))
1633                 return -ENODEV;
1634
1635         len = sprintf(buf, "%d\n", id);
1636
1637         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1638 }
1639
1640 static ssize_t
1641 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1642                   loff_t *ppos)
1643 {
1644         struct trace_event_file *file;
1645         struct trace_seq *s;
1646         int r = -ENODEV;
1647
1648         if (*ppos)
1649                 return 0;
1650
1651         s = kmalloc(sizeof(*s), GFP_KERNEL);
1652
1653         if (!s)
1654                 return -ENOMEM;
1655
1656         trace_seq_init(s);
1657
1658         mutex_lock(&event_mutex);
1659         file = event_file_data(filp);
1660         if (file)
1661                 print_event_filter(file, s);
1662         mutex_unlock(&event_mutex);
1663
1664         if (file)
1665                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1666                                             s->buffer, trace_seq_used(s));
1667
1668         kfree(s);
1669
1670         return r;
1671 }
1672
1673 static ssize_t
1674 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1675                    loff_t *ppos)
1676 {
1677         struct trace_event_file *file;
1678         char *buf;
1679         int err = -ENODEV;
1680
1681         if (cnt >= PAGE_SIZE)
1682                 return -EINVAL;
1683
1684         buf = memdup_user_nul(ubuf, cnt);
1685         if (IS_ERR(buf))
1686                 return PTR_ERR(buf);
1687
1688         mutex_lock(&event_mutex);
1689         file = event_file_data(filp);
1690         if (file)
1691                 err = apply_event_filter(file, buf);
1692         mutex_unlock(&event_mutex);
1693
1694         kfree(buf);
1695         if (err < 0)
1696                 return err;
1697
1698         *ppos += cnt;
1699
1700         return cnt;
1701 }
1702
1703 static LIST_HEAD(event_subsystems);
1704
1705 static int subsystem_open(struct inode *inode, struct file *filp)
1706 {
1707         struct event_subsystem *system = NULL;
1708         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1709         struct trace_array *tr;
1710         int ret;
1711
1712         if (tracing_is_disabled())
1713                 return -ENODEV;
1714
1715         /* Make sure the system still exists */
1716         mutex_lock(&event_mutex);
1717         mutex_lock(&trace_types_lock);
1718         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1719                 list_for_each_entry(dir, &tr->systems, list) {
1720                         if (dir == inode->i_private) {
1721                                 /* Don't open systems with no events */
1722                                 if (dir->nr_events) {
1723                                         __get_system_dir(dir);
1724                                         system = dir->subsystem;
1725                                 }
1726                                 goto exit_loop;
1727                         }
1728                 }
1729         }
1730  exit_loop:
1731         mutex_unlock(&trace_types_lock);
1732         mutex_unlock(&event_mutex);
1733
1734         if (!system)
1735                 return -ENODEV;
1736
1737         /* Some versions of gcc think dir can be uninitialized here */
1738         WARN_ON(!dir);
1739
1740         /* Still need to increment the ref count of the system */
1741         if (trace_array_get(tr) < 0) {
1742                 put_system(dir);
1743                 return -ENODEV;
1744         }
1745
1746         ret = tracing_open_generic(inode, filp);
1747         if (ret < 0) {
1748                 trace_array_put(tr);
1749                 put_system(dir);
1750         }
1751
1752         return ret;
1753 }
1754
1755 static int system_tr_open(struct inode *inode, struct file *filp)
1756 {
1757         struct trace_subsystem_dir *dir;
1758         struct trace_array *tr = inode->i_private;
1759         int ret;
1760
1761         /* Make a temporary dir that has no system but points to tr */
1762         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1763         if (!dir)
1764                 return -ENOMEM;
1765
1766         ret = tracing_open_generic_tr(inode, filp);
1767         if (ret < 0) {
1768                 kfree(dir);
1769                 return ret;
1770         }
1771         dir->tr = tr;
1772         filp->private_data = dir;
1773
1774         return 0;
1775 }
1776
1777 static int subsystem_release(struct inode *inode, struct file *file)
1778 {
1779         struct trace_subsystem_dir *dir = file->private_data;
1780
1781         trace_array_put(dir->tr);
1782
1783         /*
1784          * If dir->subsystem is NULL, then this is a temporary
1785          * descriptor that was made for a trace_array to enable
1786          * all subsystems.
1787          */
1788         if (dir->subsystem)
1789                 put_system(dir);
1790         else
1791                 kfree(dir);
1792
1793         return 0;
1794 }
1795
1796 static ssize_t
1797 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1798                       loff_t *ppos)
1799 {
1800         struct trace_subsystem_dir *dir = filp->private_data;
1801         struct event_subsystem *system = dir->subsystem;
1802         struct trace_seq *s;
1803         int r;
1804
1805         if (*ppos)
1806                 return 0;
1807
1808         s = kmalloc(sizeof(*s), GFP_KERNEL);
1809         if (!s)
1810                 return -ENOMEM;
1811
1812         trace_seq_init(s);
1813
1814         print_subsystem_event_filter(system, s);
1815         r = simple_read_from_buffer(ubuf, cnt, ppos,
1816                                     s->buffer, trace_seq_used(s));
1817
1818         kfree(s);
1819
1820         return r;
1821 }
1822
1823 static ssize_t
1824 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1825                        loff_t *ppos)
1826 {
1827         struct trace_subsystem_dir *dir = filp->private_data;
1828         char *buf;
1829         int err;
1830
1831         if (cnt >= PAGE_SIZE)
1832                 return -EINVAL;
1833
1834         buf = memdup_user_nul(ubuf, cnt);
1835         if (IS_ERR(buf))
1836                 return PTR_ERR(buf);
1837
1838         err = apply_subsystem_event_filter(dir, buf);
1839         kfree(buf);
1840         if (err < 0)
1841                 return err;
1842
1843         *ppos += cnt;
1844
1845         return cnt;
1846 }
1847
1848 static ssize_t
1849 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1850 {
1851         int (*func)(struct trace_seq *s) = filp->private_data;
1852         struct trace_seq *s;
1853         int r;
1854
1855         if (*ppos)
1856                 return 0;
1857
1858         s = kmalloc(sizeof(*s), GFP_KERNEL);
1859         if (!s)
1860                 return -ENOMEM;
1861
1862         trace_seq_init(s);
1863
1864         func(s);
1865         r = simple_read_from_buffer(ubuf, cnt, ppos,
1866                                     s->buffer, trace_seq_used(s));
1867
1868         kfree(s);
1869
1870         return r;
1871 }
1872
1873 static void ignore_task_cpu(void *data)
1874 {
1875         struct trace_array *tr = data;
1876         struct trace_pid_list *pid_list;
1877         struct trace_pid_list *no_pid_list;
1878
1879         /*
1880          * This function is called by on_each_cpu() while the
1881          * event_mutex is held.
1882          */
1883         pid_list = rcu_dereference_protected(tr->filtered_pids,
1884                                              mutex_is_locked(&event_mutex));
1885         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1886                                              mutex_is_locked(&event_mutex));
1887
1888         this_cpu_write(tr->array_buffer.data->ignore_pid,
1889                        trace_ignore_this_task(pid_list, no_pid_list, current));
1890 }
1891
1892 static void register_pid_events(struct trace_array *tr)
1893 {
1894         /*
1895          * Register a probe that is called before all other probes
1896          * to set ignore_pid if next or prev do not match.
1897          * Register a probe this is called after all other probes
1898          * to only keep ignore_pid set if next pid matches.
1899          */
1900         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1901                                          tr, INT_MAX);
1902         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1903                                          tr, 0);
1904
1905         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1906                                          tr, INT_MAX);
1907         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1908                                          tr, 0);
1909
1910         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1911                                              tr, INT_MAX);
1912         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1913                                              tr, 0);
1914
1915         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1916                                          tr, INT_MAX);
1917         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1918                                          tr, 0);
1919 }
1920
1921 static ssize_t
1922 event_pid_write(struct file *filp, const char __user *ubuf,
1923                 size_t cnt, loff_t *ppos, int type)
1924 {
1925         struct seq_file *m = filp->private_data;
1926         struct trace_array *tr = m->private;
1927         struct trace_pid_list *filtered_pids = NULL;
1928         struct trace_pid_list *other_pids = NULL;
1929         struct trace_pid_list *pid_list;
1930         struct trace_event_file *file;
1931         ssize_t ret;
1932
1933         if (!cnt)
1934                 return 0;
1935
1936         ret = tracing_update_buffers();
1937         if (ret < 0)
1938                 return ret;
1939
1940         mutex_lock(&event_mutex);
1941
1942         if (type == TRACE_PIDS) {
1943                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1944                                                           lockdep_is_held(&event_mutex));
1945                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1946                                                           lockdep_is_held(&event_mutex));
1947         } else {
1948                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1949                                                           lockdep_is_held(&event_mutex));
1950                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1951                                                           lockdep_is_held(&event_mutex));
1952         }
1953
1954         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1955         if (ret < 0)
1956                 goto out;
1957
1958         if (type == TRACE_PIDS)
1959                 rcu_assign_pointer(tr->filtered_pids, pid_list);
1960         else
1961                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1962
1963         list_for_each_entry(file, &tr->events, list) {
1964                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1965         }
1966
1967         if (filtered_pids) {
1968                 tracepoint_synchronize_unregister();
1969                 trace_free_pid_list(filtered_pids);
1970         } else if (pid_list && !other_pids) {
1971                 register_pid_events(tr);
1972         }
1973
1974         /*
1975          * Ignoring of pids is done at task switch. But we have to
1976          * check for those tasks that are currently running.
1977          * Always do this in case a pid was appended or removed.
1978          */
1979         on_each_cpu(ignore_task_cpu, tr, 1);
1980
1981  out:
1982         mutex_unlock(&event_mutex);
1983
1984         if (ret > 0)
1985                 *ppos += ret;
1986
1987         return ret;
1988 }
1989
1990 static ssize_t
1991 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1992                        size_t cnt, loff_t *ppos)
1993 {
1994         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1995 }
1996
1997 static ssize_t
1998 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1999                         size_t cnt, loff_t *ppos)
2000 {
2001         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2002 }
2003
2004 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2005 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2006 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2007 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2008 static int ftrace_event_release(struct inode *inode, struct file *file);
2009
2010 static const struct seq_operations show_event_seq_ops = {
2011         .start = t_start,
2012         .next = t_next,
2013         .show = t_show,
2014         .stop = t_stop,
2015 };
2016
2017 static const struct seq_operations show_set_event_seq_ops = {
2018         .start = s_start,
2019         .next = s_next,
2020         .show = t_show,
2021         .stop = t_stop,
2022 };
2023
2024 static const struct seq_operations show_set_pid_seq_ops = {
2025         .start = p_start,
2026         .next = p_next,
2027         .show = trace_pid_show,
2028         .stop = p_stop,
2029 };
2030
2031 static const struct seq_operations show_set_no_pid_seq_ops = {
2032         .start = np_start,
2033         .next = np_next,
2034         .show = trace_pid_show,
2035         .stop = p_stop,
2036 };
2037
2038 static const struct file_operations ftrace_avail_fops = {
2039         .open = ftrace_event_avail_open,
2040         .read = seq_read,
2041         .llseek = seq_lseek,
2042         .release = seq_release,
2043 };
2044
2045 static const struct file_operations ftrace_set_event_fops = {
2046         .open = ftrace_event_set_open,
2047         .read = seq_read,
2048         .write = ftrace_event_write,
2049         .llseek = seq_lseek,
2050         .release = ftrace_event_release,
2051 };
2052
2053 static const struct file_operations ftrace_set_event_pid_fops = {
2054         .open = ftrace_event_set_pid_open,
2055         .read = seq_read,
2056         .write = ftrace_event_pid_write,
2057         .llseek = seq_lseek,
2058         .release = ftrace_event_release,
2059 };
2060
2061 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2062         .open = ftrace_event_set_npid_open,
2063         .read = seq_read,
2064         .write = ftrace_event_npid_write,
2065         .llseek = seq_lseek,
2066         .release = ftrace_event_release,
2067 };
2068
2069 static const struct file_operations ftrace_enable_fops = {
2070         .open = tracing_open_generic,
2071         .read = event_enable_read,
2072         .write = event_enable_write,
2073         .llseek = default_llseek,
2074 };
2075
2076 static const struct file_operations ftrace_event_format_fops = {
2077         .open = trace_format_open,
2078         .read = seq_read,
2079         .llseek = seq_lseek,
2080         .release = seq_release,
2081 };
2082
2083 static const struct file_operations ftrace_event_id_fops = {
2084         .read = event_id_read,
2085         .llseek = default_llseek,
2086 };
2087
2088 static const struct file_operations ftrace_event_filter_fops = {
2089         .open = tracing_open_generic,
2090         .read = event_filter_read,
2091         .write = event_filter_write,
2092         .llseek = default_llseek,
2093 };
2094
2095 static const struct file_operations ftrace_subsystem_filter_fops = {
2096         .open = subsystem_open,
2097         .read = subsystem_filter_read,
2098         .write = subsystem_filter_write,
2099         .llseek = default_llseek,
2100         .release = subsystem_release,
2101 };
2102
2103 static const struct file_operations ftrace_system_enable_fops = {
2104         .open = subsystem_open,
2105         .read = system_enable_read,
2106         .write = system_enable_write,
2107         .llseek = default_llseek,
2108         .release = subsystem_release,
2109 };
2110
2111 static const struct file_operations ftrace_tr_enable_fops = {
2112         .open = system_tr_open,
2113         .read = system_enable_read,
2114         .write = system_enable_write,
2115         .llseek = default_llseek,
2116         .release = subsystem_release,
2117 };
2118
2119 static const struct file_operations ftrace_show_header_fops = {
2120         .open = tracing_open_generic,
2121         .read = show_header,
2122         .llseek = default_llseek,
2123 };
2124
2125 static int
2126 ftrace_event_open(struct inode *inode, struct file *file,
2127                   const struct seq_operations *seq_ops)
2128 {
2129         struct seq_file *m;
2130         int ret;
2131
2132         ret = security_locked_down(LOCKDOWN_TRACEFS);
2133         if (ret)
2134                 return ret;
2135
2136         ret = seq_open(file, seq_ops);
2137         if (ret < 0)
2138                 return ret;
2139         m = file->private_data;
2140         /* copy tr over to seq ops */
2141         m->private = inode->i_private;
2142
2143         return ret;
2144 }
2145
2146 static int ftrace_event_release(struct inode *inode, struct file *file)
2147 {
2148         struct trace_array *tr = inode->i_private;
2149
2150         trace_array_put(tr);
2151
2152         return seq_release(inode, file);
2153 }
2154
2155 static int
2156 ftrace_event_avail_open(struct inode *inode, struct file *file)
2157 {
2158         const struct seq_operations *seq_ops = &show_event_seq_ops;
2159
2160         /* Checks for tracefs lockdown */
2161         return ftrace_event_open(inode, file, seq_ops);
2162 }
2163
2164 static int
2165 ftrace_event_set_open(struct inode *inode, struct file *file)
2166 {
2167         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2168         struct trace_array *tr = inode->i_private;
2169         int ret;
2170
2171         ret = tracing_check_open_get_tr(tr);
2172         if (ret)
2173                 return ret;
2174
2175         if ((file->f_mode & FMODE_WRITE) &&
2176             (file->f_flags & O_TRUNC))
2177                 ftrace_clear_events(tr);
2178
2179         ret = ftrace_event_open(inode, file, seq_ops);
2180         if (ret < 0)
2181                 trace_array_put(tr);
2182         return ret;
2183 }
2184
2185 static int
2186 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2187 {
2188         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2189         struct trace_array *tr = inode->i_private;
2190         int ret;
2191
2192         ret = tracing_check_open_get_tr(tr);
2193         if (ret)
2194                 return ret;
2195
2196         if ((file->f_mode & FMODE_WRITE) &&
2197             (file->f_flags & O_TRUNC))
2198                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2199
2200         ret = ftrace_event_open(inode, file, seq_ops);
2201         if (ret < 0)
2202                 trace_array_put(tr);
2203         return ret;
2204 }
2205
2206 static int
2207 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2208 {
2209         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2210         struct trace_array *tr = inode->i_private;
2211         int ret;
2212
2213         ret = tracing_check_open_get_tr(tr);
2214         if (ret)
2215                 return ret;
2216
2217         if ((file->f_mode & FMODE_WRITE) &&
2218             (file->f_flags & O_TRUNC))
2219                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2220
2221         ret = ftrace_event_open(inode, file, seq_ops);
2222         if (ret < 0)
2223                 trace_array_put(tr);
2224         return ret;
2225 }
2226
2227 static struct event_subsystem *
2228 create_new_subsystem(const char *name)
2229 {
2230         struct event_subsystem *system;
2231
2232         /* need to create new entry */
2233         system = kmalloc(sizeof(*system), GFP_KERNEL);
2234         if (!system)
2235                 return NULL;
2236
2237         system->ref_count = 1;
2238
2239         /* Only allocate if dynamic (kprobes and modules) */
2240         system->name = kstrdup_const(name, GFP_KERNEL);
2241         if (!system->name)
2242                 goto out_free;
2243
2244         system->filter = NULL;
2245
2246         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2247         if (!system->filter)
2248                 goto out_free;
2249
2250         list_add(&system->list, &event_subsystems);
2251
2252         return system;
2253
2254  out_free:
2255         kfree_const(system->name);
2256         kfree(system);
2257         return NULL;
2258 }
2259
2260 static struct dentry *
2261 event_subsystem_dir(struct trace_array *tr, const char *name,
2262                     struct trace_event_file *file, struct dentry *parent)
2263 {
2264         struct trace_subsystem_dir *dir;
2265         struct event_subsystem *system;
2266         struct dentry *entry;
2267
2268         /* First see if we did not already create this dir */
2269         list_for_each_entry(dir, &tr->systems, list) {
2270                 system = dir->subsystem;
2271                 if (strcmp(system->name, name) == 0) {
2272                         dir->nr_events++;
2273                         file->system = dir;
2274                         return dir->entry;
2275                 }
2276         }
2277
2278         /* Now see if the system itself exists. */
2279         list_for_each_entry(system, &event_subsystems, list) {
2280                 if (strcmp(system->name, name) == 0)
2281                         break;
2282         }
2283         /* Reset system variable when not found */
2284         if (&system->list == &event_subsystems)
2285                 system = NULL;
2286
2287         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2288         if (!dir)
2289                 goto out_fail;
2290
2291         if (!system) {
2292                 system = create_new_subsystem(name);
2293                 if (!system)
2294                         goto out_free;
2295         } else
2296                 __get_system(system);
2297
2298         dir->entry = tracefs_create_dir(name, parent);
2299         if (!dir->entry) {
2300                 pr_warn("Failed to create system directory %s\n", name);
2301                 __put_system(system);
2302                 goto out_free;
2303         }
2304
2305         dir->tr = tr;
2306         dir->ref_count = 1;
2307         dir->nr_events = 1;
2308         dir->subsystem = system;
2309         file->system = dir;
2310
2311         /* the ftrace system is special, do not create enable or filter files */
2312         if (strcmp(name, "ftrace") != 0) {
2313
2314                 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2315                                             &ftrace_subsystem_filter_fops);
2316                 if (!entry) {
2317                         kfree(system->filter);
2318                         system->filter = NULL;
2319                         pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2320                 }
2321
2322                 trace_create_file("enable", 0644, dir->entry, dir,
2323                                   &ftrace_system_enable_fops);
2324         }
2325
2326         list_add(&dir->list, &tr->systems);
2327
2328         return dir->entry;
2329
2330  out_free:
2331         kfree(dir);
2332  out_fail:
2333         /* Only print this message if failed on memory allocation */
2334         if (!dir || !system)
2335                 pr_warn("No memory to create event subsystem %s\n", name);
2336         return NULL;
2337 }
2338
2339 static int
2340 event_define_fields(struct trace_event_call *call)
2341 {
2342         struct list_head *head;
2343         int ret = 0;
2344
2345         /*
2346          * Other events may have the same class. Only update
2347          * the fields if they are not already defined.
2348          */
2349         head = trace_get_fields(call);
2350         if (list_empty(head)) {
2351                 struct trace_event_fields *field = call->class->fields_array;
2352                 unsigned int offset = sizeof(struct trace_entry);
2353
2354                 for (; field->type; field++) {
2355                         if (field->type == TRACE_FUNCTION_TYPE) {
2356                                 field->define_fields(call);
2357                                 break;
2358                         }
2359
2360                         offset = ALIGN(offset, field->align);
2361                         ret = trace_define_field(call, field->type, field->name,
2362                                                  offset, field->size,
2363                                                  field->is_signed, field->filter_type);
2364                         if (WARN_ON_ONCE(ret)) {
2365                                 pr_err("error code is %d\n", ret);
2366                                 break;
2367                         }
2368
2369                         offset += field->size;
2370                 }
2371         }
2372
2373         return ret;
2374 }
2375
2376 static int
2377 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2378 {
2379         struct trace_event_call *call = file->event_call;
2380         struct trace_array *tr = file->tr;
2381         struct dentry *d_events;
2382         const char *name;
2383         int ret;
2384
2385         /*
2386          * If the trace point header did not define TRACE_SYSTEM
2387          * then the system would be called "TRACE_SYSTEM".
2388          */
2389         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2390                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2391                 if (!d_events)
2392                         return -ENOMEM;
2393         } else
2394                 d_events = parent;
2395
2396         name = trace_event_name(call);
2397         file->dir = tracefs_create_dir(name, d_events);
2398         if (!file->dir) {
2399                 pr_warn("Could not create tracefs '%s' directory\n", name);
2400                 return -1;
2401         }
2402
2403         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2404                 trace_create_file("enable", 0644, file->dir, file,
2405                                   &ftrace_enable_fops);
2406
2407 #ifdef CONFIG_PERF_EVENTS
2408         if (call->event.type && call->class->reg)
2409                 trace_create_file("id", 0444, file->dir,
2410                                   (void *)(long)call->event.type,
2411                                   &ftrace_event_id_fops);
2412 #endif
2413
2414         ret = event_define_fields(call);
2415         if (ret < 0) {
2416                 pr_warn("Could not initialize trace point events/%s\n", name);
2417                 return ret;
2418         }
2419
2420         /*
2421          * Only event directories that can be enabled should have
2422          * triggers or filters.
2423          */
2424         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2425                 trace_create_file("filter", 0644, file->dir, file,
2426                                   &ftrace_event_filter_fops);
2427
2428                 trace_create_file("trigger", 0644, file->dir, file,
2429                                   &event_trigger_fops);
2430         }
2431
2432 #ifdef CONFIG_HIST_TRIGGERS
2433         trace_create_file("hist", 0444, file->dir, file,
2434                           &event_hist_fops);
2435 #endif
2436 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2437         trace_create_file("hist_debug", 0444, file->dir, file,
2438                           &event_hist_debug_fops);
2439 #endif
2440         trace_create_file("format", 0444, file->dir, call,
2441                           &ftrace_event_format_fops);
2442
2443 #ifdef CONFIG_TRACE_EVENT_INJECT
2444         if (call->event.type && call->class->reg)
2445                 trace_create_file("inject", 0200, file->dir, file,
2446                                   &event_inject_fops);
2447 #endif
2448
2449         return 0;
2450 }
2451
2452 static void remove_event_from_tracers(struct trace_event_call *call)
2453 {
2454         struct trace_event_file *file;
2455         struct trace_array *tr;
2456
2457         do_for_each_event_file_safe(tr, file) {
2458                 if (file->event_call != call)
2459                         continue;
2460
2461                 remove_event_file_dir(file);
2462                 /*
2463                  * The do_for_each_event_file_safe() is
2464                  * a double loop. After finding the call for this
2465                  * trace_array, we use break to jump to the next
2466                  * trace_array.
2467                  */
2468                 break;
2469         } while_for_each_event_file();
2470 }
2471
2472 static void event_remove(struct trace_event_call *call)
2473 {
2474         struct trace_array *tr;
2475         struct trace_event_file *file;
2476
2477         do_for_each_event_file(tr, file) {
2478                 if (file->event_call != call)
2479                         continue;
2480
2481                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2482                         tr->clear_trace = true;
2483
2484                 ftrace_event_enable_disable(file, 0);
2485                 /*
2486                  * The do_for_each_event_file() is
2487                  * a double loop. After finding the call for this
2488                  * trace_array, we use break to jump to the next
2489                  * trace_array.
2490                  */
2491                 break;
2492         } while_for_each_event_file();
2493
2494         if (call->event.funcs)
2495                 __unregister_trace_event(&call->event);
2496         remove_event_from_tracers(call);
2497         list_del(&call->list);
2498 }
2499
2500 static int event_init(struct trace_event_call *call)
2501 {
2502         int ret = 0;
2503         const char *name;
2504
2505         name = trace_event_name(call);
2506         if (WARN_ON(!name))
2507                 return -EINVAL;
2508
2509         if (call->class->raw_init) {
2510                 ret = call->class->raw_init(call);
2511                 if (ret < 0 && ret != -ENOSYS)
2512                         pr_warn("Could not initialize trace events/%s\n", name);
2513         }
2514
2515         return ret;
2516 }
2517
2518 static int
2519 __register_event(struct trace_event_call *call, struct module *mod)
2520 {
2521         int ret;
2522
2523         ret = event_init(call);
2524         if (ret < 0)
2525                 return ret;
2526
2527         list_add(&call->list, &ftrace_events);
2528         call->mod = mod;
2529
2530         return 0;
2531 }
2532
2533 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2534 {
2535         int rlen;
2536         int elen;
2537
2538         /* Find the length of the eval value as a string */
2539         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2540         /* Make sure there's enough room to replace the string with the value */
2541         if (len < elen)
2542                 return NULL;
2543
2544         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2545
2546         /* Get the rest of the string of ptr */
2547         rlen = strlen(ptr + len);
2548         memmove(ptr + elen, ptr + len, rlen);
2549         /* Make sure we end the new string */
2550         ptr[elen + rlen] = 0;
2551
2552         return ptr + elen;
2553 }
2554
2555 static void update_event_printk(struct trace_event_call *call,
2556                                 struct trace_eval_map *map)
2557 {
2558         char *ptr;
2559         int quote = 0;
2560         int len = strlen(map->eval_string);
2561
2562         for (ptr = call->print_fmt; *ptr; ptr++) {
2563                 if (*ptr == '\\') {
2564                         ptr++;
2565                         /* paranoid */
2566                         if (!*ptr)
2567                                 break;
2568                         continue;
2569                 }
2570                 if (*ptr == '"') {
2571                         quote ^= 1;
2572                         continue;
2573                 }
2574                 if (quote)
2575                         continue;
2576                 if (isdigit(*ptr)) {
2577                         /* skip numbers */
2578                         do {
2579                                 ptr++;
2580                                 /* Check for alpha chars like ULL */
2581                         } while (isalnum(*ptr));
2582                         if (!*ptr)
2583                                 break;
2584                         /*
2585                          * A number must have some kind of delimiter after
2586                          * it, and we can ignore that too.
2587                          */
2588                         continue;
2589                 }
2590                 if (isalpha(*ptr) || *ptr == '_') {
2591                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2592                             !isalnum(ptr[len]) && ptr[len] != '_') {
2593                                 ptr = eval_replace(ptr, map, len);
2594                                 /* enum/sizeof string smaller than value */
2595                                 if (WARN_ON_ONCE(!ptr))
2596                                         return;
2597                                 /*
2598                                  * No need to decrement here, as eval_replace()
2599                                  * returns the pointer to the character passed
2600                                  * the eval, and two evals can not be placed
2601                                  * back to back without something in between.
2602                                  * We can skip that something in between.
2603                                  */
2604                                 continue;
2605                         }
2606                 skip_more:
2607                         do {
2608                                 ptr++;
2609                         } while (isalnum(*ptr) || *ptr == '_');
2610                         if (!*ptr)
2611                                 break;
2612                         /*
2613                          * If what comes after this variable is a '.' or
2614                          * '->' then we can continue to ignore that string.
2615                          */
2616                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2617                                 ptr += *ptr == '.' ? 1 : 2;
2618                                 if (!*ptr)
2619                                         break;
2620                                 goto skip_more;
2621                         }
2622                         /*
2623                          * Once again, we can skip the delimiter that came
2624                          * after the string.
2625                          */
2626                         continue;
2627                 }
2628         }
2629 }
2630
2631 void trace_event_eval_update(struct trace_eval_map **map, int len)
2632 {
2633         struct trace_event_call *call, *p;
2634         const char *last_system = NULL;
2635         bool first = false;
2636         int last_i;
2637         int i;
2638
2639         down_write(&trace_event_sem);
2640         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2641                 /* events are usually grouped together with systems */
2642                 if (!last_system || call->class->system != last_system) {
2643                         first = true;
2644                         last_i = 0;
2645                         last_system = call->class->system;
2646                 }
2647
2648                 /*
2649                  * Since calls are grouped by systems, the likelihood that the
2650                  * next call in the iteration belongs to the same system as the
2651                  * previous call is high. As an optimization, we skip searching
2652                  * for a map[] that matches the call's system if the last call
2653                  * was from the same system. That's what last_i is for. If the
2654                  * call has the same system as the previous call, then last_i
2655                  * will be the index of the first map[] that has a matching
2656                  * system.
2657                  */
2658                 for (i = last_i; i < len; i++) {
2659                         if (call->class->system == map[i]->system) {
2660                                 /* Save the first system if need be */
2661                                 if (first) {
2662                                         last_i = i;
2663                                         first = false;
2664                                 }
2665                                 update_event_printk(call, map[i]);
2666                         }
2667                 }
2668         }
2669         up_write(&trace_event_sem);
2670 }
2671
2672 static struct trace_event_file *
2673 trace_create_new_event(struct trace_event_call *call,
2674                        struct trace_array *tr)
2675 {
2676         struct trace_event_file *file;
2677
2678         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2679         if (!file)
2680                 return NULL;
2681
2682         file->event_call = call;
2683         file->tr = tr;
2684         atomic_set(&file->sm_ref, 0);
2685         atomic_set(&file->tm_ref, 0);
2686         INIT_LIST_HEAD(&file->triggers);
2687         list_add(&file->list, &tr->events);
2688
2689         return file;
2690 }
2691
2692 /* Add an event to a trace directory */
2693 static int
2694 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2695 {
2696         struct trace_event_file *file;
2697
2698         file = trace_create_new_event(call, tr);
2699         if (!file)
2700                 return -ENOMEM;
2701
2702         if (eventdir_initialized)
2703                 return event_create_dir(tr->event_dir, file);
2704         else
2705                 return event_define_fields(call);
2706 }
2707
2708 /*
2709  * Just create a descriptor for early init. A descriptor is required
2710  * for enabling events at boot. We want to enable events before
2711  * the filesystem is initialized.
2712  */
2713 static int
2714 __trace_early_add_new_event(struct trace_event_call *call,
2715                             struct trace_array *tr)
2716 {
2717         struct trace_event_file *file;
2718
2719         file = trace_create_new_event(call, tr);
2720         if (!file)
2721                 return -ENOMEM;
2722
2723         return event_define_fields(call);
2724 }
2725
2726 struct ftrace_module_file_ops;
2727 static void __add_event_to_tracers(struct trace_event_call *call);
2728
2729 /* Add an additional event_call dynamically */
2730 int trace_add_event_call(struct trace_event_call *call)
2731 {
2732         int ret;
2733         lockdep_assert_held(&event_mutex);
2734
2735         mutex_lock(&trace_types_lock);
2736
2737         ret = __register_event(call, NULL);
2738         if (ret >= 0)
2739                 __add_event_to_tracers(call);
2740
2741         mutex_unlock(&trace_types_lock);
2742         return ret;
2743 }
2744
2745 /*
2746  * Must be called under locking of trace_types_lock, event_mutex and
2747  * trace_event_sem.
2748  */
2749 static void __trace_remove_event_call(struct trace_event_call *call)
2750 {
2751         event_remove(call);
2752         trace_destroy_fields(call);
2753         free_event_filter(call->filter);
2754         call->filter = NULL;
2755 }
2756
2757 static int probe_remove_event_call(struct trace_event_call *call)
2758 {
2759         struct trace_array *tr;
2760         struct trace_event_file *file;
2761
2762 #ifdef CONFIG_PERF_EVENTS
2763         if (call->perf_refcount)
2764                 return -EBUSY;
2765 #endif
2766         do_for_each_event_file(tr, file) {
2767                 if (file->event_call != call)
2768                         continue;
2769                 /*
2770                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2771                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2772                  * TRACE_REG_UNREGISTER.
2773                  */
2774                 if (file->flags & EVENT_FILE_FL_ENABLED)
2775                         return -EBUSY;
2776                 /*
2777                  * The do_for_each_event_file_safe() is
2778                  * a double loop. After finding the call for this
2779                  * trace_array, we use break to jump to the next
2780                  * trace_array.
2781                  */
2782                 break;
2783         } while_for_each_event_file();
2784
2785         __trace_remove_event_call(call);
2786
2787         return 0;
2788 }
2789
2790 /* Remove an event_call */
2791 int trace_remove_event_call(struct trace_event_call *call)
2792 {
2793         int ret;
2794
2795         lockdep_assert_held(&event_mutex);
2796
2797         mutex_lock(&trace_types_lock);
2798         down_write(&trace_event_sem);
2799         ret = probe_remove_event_call(call);
2800         up_write(&trace_event_sem);
2801         mutex_unlock(&trace_types_lock);
2802
2803         return ret;
2804 }
2805
2806 #define for_each_event(event, start, end)                       \
2807         for (event = start;                                     \
2808              (unsigned long)event < (unsigned long)end;         \
2809              event++)
2810
2811 #ifdef CONFIG_MODULES
2812
2813 static void trace_module_add_events(struct module *mod)
2814 {
2815         struct trace_event_call **call, **start, **end;
2816
2817         if (!mod->num_trace_events)
2818                 return;
2819
2820         /* Don't add infrastructure for mods without tracepoints */
2821         if (trace_module_has_bad_taint(mod)) {
2822                 pr_err("%s: module has bad taint, not creating trace events\n",
2823                        mod->name);
2824                 return;
2825         }
2826
2827         start = mod->trace_events;
2828         end = mod->trace_events + mod->num_trace_events;
2829
2830         for_each_event(call, start, end) {
2831                 __register_event(*call, mod);
2832                 __add_event_to_tracers(*call);
2833         }
2834 }
2835
2836 static void trace_module_remove_events(struct module *mod)
2837 {
2838         struct trace_event_call *call, *p;
2839
2840         down_write(&trace_event_sem);
2841         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2842                 if (call->mod == mod)
2843                         __trace_remove_event_call(call);
2844         }
2845         up_write(&trace_event_sem);
2846
2847         /*
2848          * It is safest to reset the ring buffer if the module being unloaded
2849          * registered any events that were used. The only worry is if
2850          * a new module gets loaded, and takes on the same id as the events
2851          * of this module. When printing out the buffer, traced events left
2852          * over from this module may be passed to the new module events and
2853          * unexpected results may occur.
2854          */
2855         tracing_reset_all_online_cpus();
2856 }
2857
2858 static int trace_module_notify(struct notifier_block *self,
2859                                unsigned long val, void *data)
2860 {
2861         struct module *mod = data;
2862
2863         mutex_lock(&event_mutex);
2864         mutex_lock(&trace_types_lock);
2865         switch (val) {
2866         case MODULE_STATE_COMING:
2867                 trace_module_add_events(mod);
2868                 break;
2869         case MODULE_STATE_GOING:
2870                 trace_module_remove_events(mod);
2871                 break;
2872         }
2873         mutex_unlock(&trace_types_lock);
2874         mutex_unlock(&event_mutex);
2875
2876         return NOTIFY_OK;
2877 }
2878
2879 static struct notifier_block trace_module_nb = {
2880         .notifier_call = trace_module_notify,
2881         .priority = 1, /* higher than trace.c module notify */
2882 };
2883 #endif /* CONFIG_MODULES */
2884
2885 /* Create a new event directory structure for a trace directory. */
2886 static void
2887 __trace_add_event_dirs(struct trace_array *tr)
2888 {
2889         struct trace_event_call *call;
2890         int ret;
2891
2892         list_for_each_entry(call, &ftrace_events, list) {
2893                 ret = __trace_add_new_event(call, tr);
2894                 if (ret < 0)
2895                         pr_warn("Could not create directory for event %s\n",
2896                                 trace_event_name(call));
2897         }
2898 }
2899
2900 /* Returns any file that matches the system and event */
2901 struct trace_event_file *
2902 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2903 {
2904         struct trace_event_file *file;
2905         struct trace_event_call *call;
2906         const char *name;
2907
2908         list_for_each_entry(file, &tr->events, list) {
2909
2910                 call = file->event_call;
2911                 name = trace_event_name(call);
2912
2913                 if (!name || !call->class)
2914                         continue;
2915
2916                 if (strcmp(event, name) == 0 &&
2917                     strcmp(system, call->class->system) == 0)
2918                         return file;
2919         }
2920         return NULL;
2921 }
2922
2923 /* Returns valid trace event files that match system and event */
2924 struct trace_event_file *
2925 find_event_file(struct trace_array *tr, const char *system, const char *event)
2926 {
2927         struct trace_event_file *file;
2928
2929         file = __find_event_file(tr, system, event);
2930         if (!file || !file->event_call->class->reg ||
2931             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2932                 return NULL;
2933
2934         return file;
2935 }
2936
2937 /**
2938  * trace_get_event_file - Find and return a trace event file
2939  * @instance: The name of the trace instance containing the event
2940  * @system: The name of the system containing the event
2941  * @event: The name of the event
2942  *
2943  * Return a trace event file given the trace instance name, trace
2944  * system, and trace event name.  If the instance name is NULL, it
2945  * refers to the top-level trace array.
2946  *
2947  * This function will look it up and return it if found, after calling
2948  * trace_array_get() to prevent the instance from going away, and
2949  * increment the event's module refcount to prevent it from being
2950  * removed.
2951  *
2952  * To release the file, call trace_put_event_file(), which will call
2953  * trace_array_put() and decrement the event's module refcount.
2954  *
2955  * Return: The trace event on success, ERR_PTR otherwise.
2956  */
2957 struct trace_event_file *trace_get_event_file(const char *instance,
2958                                               const char *system,
2959                                               const char *event)
2960 {
2961         struct trace_array *tr = top_trace_array();
2962         struct trace_event_file *file = NULL;
2963         int ret = -EINVAL;
2964
2965         if (instance) {
2966                 tr = trace_array_find_get(instance);
2967                 if (!tr)
2968                         return ERR_PTR(-ENOENT);
2969         } else {
2970                 ret = trace_array_get(tr);
2971                 if (ret)
2972                         return ERR_PTR(ret);
2973         }
2974
2975         mutex_lock(&event_mutex);
2976
2977         file = find_event_file(tr, system, event);
2978         if (!file) {
2979                 trace_array_put(tr);
2980                 ret = -EINVAL;
2981                 goto out;
2982         }
2983
2984         /* Don't let event modules unload while in use */
2985         ret = try_module_get(file->event_call->mod);
2986         if (!ret) {
2987                 trace_array_put(tr);
2988                 ret = -EBUSY;
2989                 goto out;
2990         }
2991
2992         ret = 0;
2993  out:
2994         mutex_unlock(&event_mutex);
2995
2996         if (ret)
2997                 file = ERR_PTR(ret);
2998
2999         return file;
3000 }
3001 EXPORT_SYMBOL_GPL(trace_get_event_file);
3002
3003 /**
3004  * trace_put_event_file - Release a file from trace_get_event_file()
3005  * @file: The trace event file
3006  *
3007  * If a file was retrieved using trace_get_event_file(), this should
3008  * be called when it's no longer needed.  It will cancel the previous
3009  * trace_array_get() called by that function, and decrement the
3010  * event's module refcount.
3011  */
3012 void trace_put_event_file(struct trace_event_file *file)
3013 {
3014         mutex_lock(&event_mutex);
3015         module_put(file->event_call->mod);
3016         mutex_unlock(&event_mutex);
3017
3018         trace_array_put(file->tr);
3019 }
3020 EXPORT_SYMBOL_GPL(trace_put_event_file);
3021
3022 #ifdef CONFIG_DYNAMIC_FTRACE
3023
3024 /* Avoid typos */
3025 #define ENABLE_EVENT_STR        "enable_event"
3026 #define DISABLE_EVENT_STR       "disable_event"
3027
3028 struct event_probe_data {
3029         struct trace_event_file *file;
3030         unsigned long                   count;
3031         int                             ref;
3032         bool                            enable;
3033 };
3034
3035 static void update_event_probe(struct event_probe_data *data)
3036 {
3037         if (data->enable)
3038                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3039         else
3040                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3041 }
3042
3043 static void
3044 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3045                    struct trace_array *tr, struct ftrace_probe_ops *ops,
3046                    void *data)
3047 {
3048         struct ftrace_func_mapper *mapper = data;
3049         struct event_probe_data *edata;
3050         void **pdata;
3051
3052         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3053         if (!pdata || !*pdata)
3054                 return;
3055
3056         edata = *pdata;
3057         update_event_probe(edata);
3058 }
3059
3060 static void
3061 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3062                          struct trace_array *tr, struct ftrace_probe_ops *ops,
3063                          void *data)
3064 {
3065         struct ftrace_func_mapper *mapper = data;
3066         struct event_probe_data *edata;
3067         void **pdata;
3068
3069         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3070         if (!pdata || !*pdata)
3071                 return;
3072
3073         edata = *pdata;
3074
3075         if (!edata->count)
3076                 return;
3077
3078         /* Skip if the event is in a state we want to switch to */
3079         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3080                 return;
3081
3082         if (edata->count != -1)
3083                 (edata->count)--;
3084
3085         update_event_probe(edata);
3086 }
3087
3088 static int
3089 event_enable_print(struct seq_file *m, unsigned long ip,
3090                    struct ftrace_probe_ops *ops, void *data)
3091 {
3092         struct ftrace_func_mapper *mapper = data;
3093         struct event_probe_data *edata;
3094         void **pdata;
3095
3096         pdata = ftrace_func_mapper_find_ip(mapper, ip);
3097
3098         if (WARN_ON_ONCE(!pdata || !*pdata))
3099                 return 0;
3100
3101         edata = *pdata;
3102
3103         seq_printf(m, "%ps:", (void *)ip);
3104
3105         seq_printf(m, "%s:%s:%s",
3106                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3107                    edata->file->event_call->class->system,
3108                    trace_event_name(edata->file->event_call));
3109
3110         if (edata->count == -1)
3111                 seq_puts(m, ":unlimited\n");
3112         else
3113                 seq_printf(m, ":count=%ld\n", edata->count);
3114
3115         return 0;
3116 }
3117
3118 static int
3119 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3120                   unsigned long ip, void *init_data, void **data)
3121 {
3122         struct ftrace_func_mapper *mapper = *data;
3123         struct event_probe_data *edata = init_data;
3124         int ret;
3125
3126         if (!mapper) {
3127                 mapper = allocate_ftrace_func_mapper();
3128                 if (!mapper)
3129                         return -ENODEV;
3130                 *data = mapper;
3131         }
3132
3133         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3134         if (ret < 0)
3135                 return ret;
3136
3137         edata->ref++;
3138
3139         return 0;
3140 }
3141
3142 static int free_probe_data(void *data)
3143 {
3144         struct event_probe_data *edata = data;
3145
3146         edata->ref--;
3147         if (!edata->ref) {
3148                 /* Remove the SOFT_MODE flag */
3149                 __ftrace_event_enable_disable(edata->file, 0, 1);
3150                 module_put(edata->file->event_call->mod);
3151                 kfree(edata);
3152         }
3153         return 0;
3154 }
3155
3156 static void
3157 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3158                   unsigned long ip, void *data)
3159 {
3160         struct ftrace_func_mapper *mapper = data;
3161         struct event_probe_data *edata;
3162
3163         if (!ip) {
3164                 if (!mapper)
3165                         return;
3166                 free_ftrace_func_mapper(mapper, free_probe_data);
3167                 return;
3168         }
3169
3170         edata = ftrace_func_mapper_remove_ip(mapper, ip);
3171
3172         if (WARN_ON_ONCE(!edata))
3173                 return;
3174
3175         if (WARN_ON_ONCE(edata->ref <= 0))
3176                 return;
3177
3178         free_probe_data(edata);
3179 }
3180
3181 static struct ftrace_probe_ops event_enable_probe_ops = {
3182         .func                   = event_enable_probe,
3183         .print                  = event_enable_print,
3184         .init                   = event_enable_init,
3185         .free                   = event_enable_free,
3186 };
3187
3188 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3189         .func                   = event_enable_count_probe,
3190         .print                  = event_enable_print,
3191         .init                   = event_enable_init,
3192         .free                   = event_enable_free,
3193 };
3194
3195 static struct ftrace_probe_ops event_disable_probe_ops = {
3196         .func                   = event_enable_probe,
3197         .print                  = event_enable_print,
3198         .init                   = event_enable_init,
3199         .free                   = event_enable_free,
3200 };
3201
3202 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3203         .func                   = event_enable_count_probe,
3204         .print                  = event_enable_print,
3205         .init                   = event_enable_init,
3206         .free                   = event_enable_free,
3207 };
3208
3209 static int
3210 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3211                   char *glob, char *cmd, char *param, int enabled)
3212 {
3213         struct trace_event_file *file;
3214         struct ftrace_probe_ops *ops;
3215         struct event_probe_data *data;
3216         const char *system;
3217         const char *event;
3218         char *number;
3219         bool enable;
3220         int ret;
3221
3222         if (!tr)
3223                 return -ENODEV;
3224
3225         /* hash funcs only work with set_ftrace_filter */
3226         if (!enabled || !param)
3227                 return -EINVAL;
3228
3229         system = strsep(&param, ":");
3230         if (!param)
3231                 return -EINVAL;
3232
3233         event = strsep(&param, ":");
3234
3235         mutex_lock(&event_mutex);
3236
3237         ret = -EINVAL;
3238         file = find_event_file(tr, system, event);
3239         if (!file)
3240                 goto out;
3241
3242         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3243
3244         if (enable)
3245                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3246         else
3247                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3248
3249         if (glob[0] == '!') {
3250                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3251                 goto out;
3252         }
3253
3254         ret = -ENOMEM;
3255
3256         data = kzalloc(sizeof(*data), GFP_KERNEL);
3257         if (!data)
3258                 goto out;
3259
3260         data->enable = enable;
3261         data->count = -1;
3262         data->file = file;
3263
3264         if (!param)
3265                 goto out_reg;
3266
3267         number = strsep(&param, ":");
3268
3269         ret = -EINVAL;
3270         if (!strlen(number))
3271                 goto out_free;
3272
3273         /*
3274          * We use the callback data field (which is a pointer)
3275          * as our counter.
3276          */
3277         ret = kstrtoul(number, 0, &data->count);
3278         if (ret)
3279                 goto out_free;
3280
3281  out_reg:
3282         /* Don't let event modules unload while probe registered */
3283         ret = try_module_get(file->event_call->mod);
3284         if (!ret) {
3285                 ret = -EBUSY;
3286                 goto out_free;
3287         }
3288
3289         ret = __ftrace_event_enable_disable(file, 1, 1);
3290         if (ret < 0)
3291                 goto out_put;
3292
3293         ret = register_ftrace_function_probe(glob, tr, ops, data);
3294         /*
3295          * The above returns on success the # of functions enabled,
3296          * but if it didn't find any functions it returns zero.
3297          * Consider no functions a failure too.
3298          */
3299         if (!ret) {
3300                 ret = -ENOENT;
3301                 goto out_disable;
3302         } else if (ret < 0)
3303                 goto out_disable;
3304         /* Just return zero, not the number of enabled functions */
3305         ret = 0;
3306  out:
3307         mutex_unlock(&event_mutex);
3308         return ret;
3309
3310  out_disable:
3311         __ftrace_event_enable_disable(file, 0, 1);
3312  out_put:
3313         module_put(file->event_call->mod);
3314  out_free:
3315         kfree(data);
3316         goto out;
3317 }
3318
3319 static struct ftrace_func_command event_enable_cmd = {
3320         .name                   = ENABLE_EVENT_STR,
3321         .func                   = event_enable_func,
3322 };
3323
3324 static struct ftrace_func_command event_disable_cmd = {
3325         .name                   = DISABLE_EVENT_STR,
3326         .func                   = event_enable_func,
3327 };
3328
3329 static __init int register_event_cmds(void)
3330 {
3331         int ret;
3332
3333         ret = register_ftrace_command(&event_enable_cmd);
3334         if (WARN_ON(ret < 0))
3335                 return ret;
3336         ret = register_ftrace_command(&event_disable_cmd);
3337         if (WARN_ON(ret < 0))
3338                 unregister_ftrace_command(&event_enable_cmd);
3339         return ret;
3340 }
3341 #else
3342 static inline int register_event_cmds(void) { return 0; }
3343 #endif /* CONFIG_DYNAMIC_FTRACE */
3344
3345 /*
3346  * The top level array and trace arrays created by boot-time tracing
3347  * have already had its trace_event_file descriptors created in order
3348  * to allow for early events to be recorded.
3349  * This function is called after the tracefs has been initialized,
3350  * and we now have to create the files associated to the events.
3351  */
3352 static void __trace_early_add_event_dirs(struct trace_array *tr)
3353 {
3354         struct trace_event_file *file;
3355         int ret;
3356
3357
3358         list_for_each_entry(file, &tr->events, list) {
3359                 ret = event_create_dir(tr->event_dir, file);
3360                 if (ret < 0)
3361                         pr_warn("Could not create directory for event %s\n",
3362                                 trace_event_name(file->event_call));
3363         }
3364 }
3365
3366 /*
3367  * For early boot up, the top trace array and the trace arrays created
3368  * by boot-time tracing require to have a list of events that can be
3369  * enabled. This must be done before the filesystem is set up in order
3370  * to allow events to be traced early.
3371  */
3372 void __trace_early_add_events(struct trace_array *tr)
3373 {
3374         struct trace_event_call *call;
3375         int ret;
3376
3377         list_for_each_entry(call, &ftrace_events, list) {
3378                 /* Early boot up should not have any modules loaded */
3379                 if (WARN_ON_ONCE(call->mod))
3380                         continue;
3381
3382                 ret = __trace_early_add_new_event(call, tr);
3383                 if (ret < 0)
3384                         pr_warn("Could not create early event %s\n",
3385                                 trace_event_name(call));
3386         }
3387 }
3388
3389 /* Remove the event directory structure for a trace directory. */
3390 static void
3391 __trace_remove_event_dirs(struct trace_array *tr)
3392 {
3393         struct trace_event_file *file, *next;
3394
3395         list_for_each_entry_safe(file, next, &tr->events, list)
3396                 remove_event_file_dir(file);
3397 }
3398
3399 static void __add_event_to_tracers(struct trace_event_call *call)
3400 {
3401         struct trace_array *tr;
3402
3403         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3404                 __trace_add_new_event(call, tr);
3405 }
3406
3407 extern struct trace_event_call *__start_ftrace_events[];
3408 extern struct trace_event_call *__stop_ftrace_events[];
3409
3410 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3411
3412 static __init int setup_trace_event(char *str)
3413 {
3414         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3415         ring_buffer_expanded = true;
3416         disable_tracing_selftest("running event tracing");
3417
3418         return 1;
3419 }
3420 __setup("trace_event=", setup_trace_event);
3421
3422 /* Expects to have event_mutex held when called */
3423 static int
3424 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3425 {
3426         struct dentry *d_events;
3427         struct dentry *entry;
3428
3429         entry = tracefs_create_file("set_event", 0644, parent,
3430                                     tr, &ftrace_set_event_fops);
3431         if (!entry) {
3432                 pr_warn("Could not create tracefs 'set_event' entry\n");
3433                 return -ENOMEM;
3434         }
3435
3436         d_events = tracefs_create_dir("events", parent);
3437         if (!d_events) {
3438                 pr_warn("Could not create tracefs 'events' directory\n");
3439                 return -ENOMEM;
3440         }
3441
3442         entry = trace_create_file("enable", 0644, d_events,
3443                                   tr, &ftrace_tr_enable_fops);
3444         if (!entry) {
3445                 pr_warn("Could not create tracefs 'enable' entry\n");
3446                 return -ENOMEM;
3447         }
3448
3449         /* There are not as crucial, just warn if they are not created */
3450
3451         entry = tracefs_create_file("set_event_pid", 0644, parent,
3452                                     tr, &ftrace_set_event_pid_fops);
3453         if (!entry)
3454                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3455
3456         entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3457                                     tr, &ftrace_set_event_notrace_pid_fops);
3458         if (!entry)
3459                 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3460
3461         /* ring buffer internal formats */
3462         entry = trace_create_file("header_page", 0444, d_events,
3463                                   ring_buffer_print_page_header,
3464                                   &ftrace_show_header_fops);
3465         if (!entry)
3466                 pr_warn("Could not create tracefs 'header_page' entry\n");
3467
3468         entry = trace_create_file("header_event", 0444, d_events,
3469                                   ring_buffer_print_entry_header,
3470                                   &ftrace_show_header_fops);
3471         if (!entry)
3472                 pr_warn("Could not create tracefs 'header_event' entry\n");
3473
3474         tr->event_dir = d_events;
3475
3476         return 0;
3477 }
3478
3479 /**
3480  * event_trace_add_tracer - add a instance of a trace_array to events
3481  * @parent: The parent dentry to place the files/directories for events in
3482  * @tr: The trace array associated with these events
3483  *
3484  * When a new instance is created, it needs to set up its events
3485  * directory, as well as other files associated with events. It also
3486  * creates the event hierarchy in the @parent/events directory.
3487  *
3488  * Returns 0 on success.
3489  *
3490  * Must be called with event_mutex held.
3491  */
3492 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3493 {
3494         int ret;
3495
3496         lockdep_assert_held(&event_mutex);
3497
3498         ret = create_event_toplevel_files(parent, tr);
3499         if (ret)
3500                 goto out;
3501
3502         down_write(&trace_event_sem);
3503         /* If tr already has the event list, it is initialized in early boot. */
3504         if (unlikely(!list_empty(&tr->events)))
3505                 __trace_early_add_event_dirs(tr);
3506         else
3507                 __trace_add_event_dirs(tr);
3508         up_write(&trace_event_sem);
3509
3510  out:
3511         return ret;
3512 }
3513
3514 /*
3515  * The top trace array already had its file descriptors created.
3516  * Now the files themselves need to be created.
3517  */
3518 static __init int
3519 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3520 {
3521         int ret;
3522
3523         mutex_lock(&event_mutex);
3524
3525         ret = create_event_toplevel_files(parent, tr);
3526         if (ret)
3527                 goto out_unlock;
3528
3529         down_write(&trace_event_sem);
3530         __trace_early_add_event_dirs(tr);
3531         up_write(&trace_event_sem);
3532
3533  out_unlock:
3534         mutex_unlock(&event_mutex);
3535
3536         return ret;
3537 }
3538
3539 /* Must be called with event_mutex held */
3540 int event_trace_del_tracer(struct trace_array *tr)
3541 {
3542         lockdep_assert_held(&event_mutex);
3543
3544         /* Disable any event triggers and associated soft-disabled events */
3545         clear_event_triggers(tr);
3546
3547         /* Clear the pid list */
3548         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3549
3550         /* Disable any running events */
3551         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3552
3553         /* Make sure no more events are being executed */
3554         tracepoint_synchronize_unregister();
3555
3556         down_write(&trace_event_sem);
3557         __trace_remove_event_dirs(tr);
3558         tracefs_remove(tr->event_dir);
3559         up_write(&trace_event_sem);
3560
3561         tr->event_dir = NULL;
3562
3563         return 0;
3564 }
3565
3566 static __init int event_trace_memsetup(void)
3567 {
3568         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3569         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3570         return 0;
3571 }
3572
3573 static __init void
3574 early_enable_events(struct trace_array *tr, bool disable_first)
3575 {
3576         char *buf = bootup_event_buf;
3577         char *token;
3578         int ret;
3579
3580         while (true) {
3581                 token = strsep(&buf, ",");
3582
3583                 if (!token)
3584                         break;
3585
3586                 if (*token) {
3587                         /* Restarting syscalls requires that we stop them first */
3588                         if (disable_first)
3589                                 ftrace_set_clr_event(tr, token, 0);
3590
3591                         ret = ftrace_set_clr_event(tr, token, 1);
3592                         if (ret)
3593                                 pr_warn("Failed to enable trace event: %s\n", token);
3594                 }
3595
3596                 /* Put back the comma to allow this to be called again */
3597                 if (buf)
3598                         *(buf - 1) = ',';
3599         }
3600 }
3601
3602 static __init int event_trace_enable(void)
3603 {
3604         struct trace_array *tr = top_trace_array();
3605         struct trace_event_call **iter, *call;
3606         int ret;
3607
3608         if (!tr)
3609                 return -ENODEV;
3610
3611         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3612
3613                 call = *iter;
3614                 ret = event_init(call);
3615                 if (!ret)
3616                         list_add(&call->list, &ftrace_events);
3617         }
3618
3619         /*
3620          * We need the top trace array to have a working set of trace
3621          * points at early init, before the debug files and directories
3622          * are created. Create the file entries now, and attach them
3623          * to the actual file dentries later.
3624          */
3625         __trace_early_add_events(tr);
3626
3627         early_enable_events(tr, false);
3628
3629         trace_printk_start_comm();
3630
3631         register_event_cmds();
3632
3633         register_trigger_cmds();
3634
3635         return 0;
3636 }
3637
3638 /*
3639  * event_trace_enable() is called from trace_event_init() first to
3640  * initialize events and perhaps start any events that are on the
3641  * command line. Unfortunately, there are some events that will not
3642  * start this early, like the system call tracepoints that need
3643  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3644  * event_trace_enable() is called before pid 1 starts, and this flag
3645  * is never set, making the syscall tracepoint never get reached, but
3646  * the event is enabled regardless (and not doing anything).
3647  */
3648 static __init int event_trace_enable_again(void)
3649 {
3650         struct trace_array *tr;
3651
3652         tr = top_trace_array();
3653         if (!tr)
3654                 return -ENODEV;
3655
3656         early_enable_events(tr, true);
3657
3658         return 0;
3659 }
3660
3661 early_initcall(event_trace_enable_again);
3662
3663 /* Init fields which doesn't related to the tracefs */
3664 static __init int event_trace_init_fields(void)
3665 {
3666         if (trace_define_generic_fields())
3667                 pr_warn("tracing: Failed to allocated generic fields");
3668
3669         if (trace_define_common_fields())
3670                 pr_warn("tracing: Failed to allocate common fields");
3671
3672         return 0;
3673 }
3674
3675 __init int event_trace_init(void)
3676 {
3677         struct trace_array *tr;
3678         struct dentry *entry;
3679         int ret;
3680
3681         tr = top_trace_array();
3682         if (!tr)
3683                 return -ENODEV;
3684
3685         entry = tracefs_create_file("available_events", 0444, NULL,
3686                                     tr, &ftrace_avail_fops);
3687         if (!entry)
3688                 pr_warn("Could not create tracefs 'available_events' entry\n");
3689
3690         ret = early_event_add_tracer(NULL, tr);
3691         if (ret)
3692                 return ret;
3693
3694 #ifdef CONFIG_MODULES
3695         ret = register_module_notifier(&trace_module_nb);
3696         if (ret)
3697                 pr_warn("Failed to register trace events module notifier\n");
3698 #endif
3699
3700         eventdir_initialized = true;
3701
3702         return 0;
3703 }
3704
3705 void __init trace_event_init(void)
3706 {
3707         event_trace_memsetup();
3708         init_ftrace_syscalls();
3709         event_trace_enable();
3710         event_trace_init_fields();
3711 }
3712
3713 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3714
3715 static DEFINE_SPINLOCK(test_spinlock);
3716 static DEFINE_SPINLOCK(test_spinlock_irq);
3717 static DEFINE_MUTEX(test_mutex);
3718
3719 static __init void test_work(struct work_struct *dummy)
3720 {
3721         spin_lock(&test_spinlock);
3722         spin_lock_irq(&test_spinlock_irq);
3723         udelay(1);
3724         spin_unlock_irq(&test_spinlock_irq);
3725         spin_unlock(&test_spinlock);
3726
3727         mutex_lock(&test_mutex);
3728         msleep(1);
3729         mutex_unlock(&test_mutex);
3730 }
3731
3732 static __init int event_test_thread(void *unused)
3733 {
3734         void *test_malloc;
3735
3736         test_malloc = kmalloc(1234, GFP_KERNEL);
3737         if (!test_malloc)
3738                 pr_info("failed to kmalloc\n");
3739
3740         schedule_on_each_cpu(test_work);
3741
3742         kfree(test_malloc);
3743
3744         set_current_state(TASK_INTERRUPTIBLE);
3745         while (!kthread_should_stop()) {
3746                 schedule();
3747                 set_current_state(TASK_INTERRUPTIBLE);
3748         }
3749         __set_current_state(TASK_RUNNING);
3750
3751         return 0;
3752 }
3753
3754 /*
3755  * Do various things that may trigger events.
3756  */
3757 static __init void event_test_stuff(void)
3758 {
3759         struct task_struct *test_thread;
3760
3761         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3762         msleep(1);
3763         kthread_stop(test_thread);
3764 }
3765
3766 /*
3767  * For every trace event defined, we will test each trace point separately,
3768  * and then by groups, and finally all trace points.
3769  */
3770 static __init void event_trace_self_tests(void)
3771 {
3772         struct trace_subsystem_dir *dir;
3773         struct trace_event_file *file;
3774         struct trace_event_call *call;
3775         struct event_subsystem *system;
3776         struct trace_array *tr;
3777         int ret;
3778
3779         tr = top_trace_array();
3780         if (!tr)
3781                 return;
3782
3783         pr_info("Running tests on trace events:\n");
3784
3785         list_for_each_entry(file, &tr->events, list) {
3786
3787                 call = file->event_call;
3788
3789                 /* Only test those that have a probe */
3790                 if (!call->class || !call->class->probe)
3791                         continue;
3792
3793 /*
3794  * Testing syscall events here is pretty useless, but
3795  * we still do it if configured. But this is time consuming.
3796  * What we really need is a user thread to perform the
3797  * syscalls as we test.
3798  */
3799 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3800                 if (call->class->system &&
3801                     strcmp(call->class->system, "syscalls") == 0)
3802                         continue;
3803 #endif
3804
3805                 pr_info("Testing event %s: ", trace_event_name(call));
3806
3807                 /*
3808                  * If an event is already enabled, someone is using
3809                  * it and the self test should not be on.
3810                  */
3811                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3812                         pr_warn("Enabled event during self test!\n");
3813                         WARN_ON_ONCE(1);
3814                         continue;
3815                 }
3816
3817                 ftrace_event_enable_disable(file, 1);
3818                 event_test_stuff();
3819                 ftrace_event_enable_disable(file, 0);
3820
3821                 pr_cont("OK\n");
3822         }
3823
3824         /* Now test at the sub system level */
3825
3826         pr_info("Running tests on trace event systems:\n");
3827
3828         list_for_each_entry(dir, &tr->systems, list) {
3829
3830                 system = dir->subsystem;
3831
3832                 /* the ftrace system is special, skip it */
3833                 if (strcmp(system->name, "ftrace") == 0)
3834                         continue;
3835
3836                 pr_info("Testing event system %s: ", system->name);
3837
3838                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3839                 if (WARN_ON_ONCE(ret)) {
3840                         pr_warn("error enabling system %s\n",
3841                                 system->name);
3842                         continue;
3843                 }
3844
3845                 event_test_stuff();
3846
3847                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3848                 if (WARN_ON_ONCE(ret)) {
3849                         pr_warn("error disabling system %s\n",
3850                                 system->name);
3851                         continue;
3852                 }
3853
3854                 pr_cont("OK\n");
3855         }
3856
3857         /* Test with all events enabled */
3858
3859         pr_info("Running tests on all trace events:\n");
3860         pr_info("Testing all events: ");
3861
3862         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3863         if (WARN_ON_ONCE(ret)) {
3864                 pr_warn("error enabling all events\n");
3865                 return;
3866         }
3867
3868         event_test_stuff();
3869
3870         /* reset sysname */
3871         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3872         if (WARN_ON_ONCE(ret)) {
3873                 pr_warn("error disabling all events\n");
3874                 return;
3875         }
3876
3877         pr_cont("OK\n");
3878 }
3879
3880 #ifdef CONFIG_FUNCTION_TRACER
3881
3882 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3883
3884 static struct trace_event_file event_trace_file __initdata;
3885
3886 static void __init
3887 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3888                           struct ftrace_ops *op, struct ftrace_regs *regs)
3889 {
3890         struct trace_buffer *buffer;
3891         struct ring_buffer_event *event;
3892         struct ftrace_entry *entry;
3893         unsigned int trace_ctx;
3894         long disabled;
3895         int cpu;
3896
3897         trace_ctx = tracing_gen_ctx();
3898         preempt_disable_notrace();
3899         cpu = raw_smp_processor_id();
3900         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3901
3902         if (disabled != 1)
3903                 goto out;
3904
3905         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3906                                                 TRACE_FN, sizeof(*entry),
3907                                                 trace_ctx);
3908         if (!event)
3909                 goto out;
3910         entry   = ring_buffer_event_data(event);
3911         entry->ip                       = ip;
3912         entry->parent_ip                = parent_ip;
3913
3914         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3915                                     entry, trace_ctx);
3916  out:
3917         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3918         preempt_enable_notrace();
3919 }
3920
3921 static struct ftrace_ops trace_ops __initdata  =
3922 {
3923         .func = function_test_events_call,
3924 };
3925
3926 static __init void event_trace_self_test_with_function(void)
3927 {
3928         int ret;
3929
3930         event_trace_file.tr = top_trace_array();
3931         if (WARN_ON(!event_trace_file.tr))
3932                 return;
3933
3934         ret = register_ftrace_function(&trace_ops);
3935         if (WARN_ON(ret < 0)) {
3936                 pr_info("Failed to enable function tracer for event tests\n");
3937                 return;
3938         }
3939         pr_info("Running tests again, along with the function tracer\n");
3940         event_trace_self_tests();
3941         unregister_ftrace_function(&trace_ops);
3942 }
3943 #else
3944 static __init void event_trace_self_test_with_function(void)
3945 {
3946 }
3947 #endif
3948
3949 static __init int event_trace_self_tests_init(void)
3950 {
3951         if (!tracing_selftest_disabled) {
3952                 event_trace_self_tests();
3953                 event_trace_self_test_with_function();
3954         }
3955
3956         return 0;
3957 }
3958
3959 late_initcall(event_trace_self_tests_init);
3960
3961 #endif