kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 DEFINE_MUTEX(event_storage_mutex);
  31 EXPORT_SYMBOL_GPL(event_storage_mutex);
  32
  33 char event_storage[EVENT_STORAGE_SIZE];
  34 EXPORT_SYMBOL_GPL(event_storage);
  35
  36 LIST_HEAD(ftrace_events);
  37 LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 /* Double loops, do not use break, only goto's work */
  45 #define do_for_each_event_file(tr, file)                        \
  46         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  47                 list_for_each_entry(file, &tr->events, list)
  48
  49 #define do_for_each_event_file_safe(tr, file)                   \
  50         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  51                 struct ftrace_event_file *___n;                         \
  52                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  53
  54 #define while_for_each_event_file()             \
  55         }
  56
  57 struct list_head *
  58 trace_get_fields(struct ftrace_event_call *event_call)
  59 {
  60         if (!event_call->class->get_fields)
  61                 return &event_call->class->fields;
  62         return event_call->class->get_fields(event_call);
  63 }
  64
  65 static int __trace_define_field(struct list_head *head, const char *type,
  66                                 const char *name, int offset, int size,
  67                                 int is_signed, int filter_type)
  68 {
  69         struct ftrace_event_field *field;
  70
  71         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
  72         if (!field)
  73                 goto err;
  74
  75         field->name = name;
  76         field->type = type;
  77
  78         if (filter_type == FILTER_OTHER)
  79                 field->filter_type = filter_assign_type(type);
  80         else
  81                 field->filter_type = filter_type;
  82
  83         field->offset = offset;
  84         field->size = size;
  85         field->is_signed = is_signed;
  86
  87         list_add(&field->link, head);
  88
  89         return 0;
  90
  91 err:
  92         kmem_cache_free(field_cachep, field);
  93
  94         return -ENOMEM;
  95 }
  96
  97 int trace_define_field(struct ftrace_event_call *call, const char *type,
  98                        const char *name, int offset, int size, int is_signed,
  99                        int filter_type)
 100 {
 101         struct list_head *head;
 102
 103         if (WARN_ON(!call->class))
 104                 return 0;
 105
 106         head = trace_get_fields(call);
 107         return __trace_define_field(head, type, name, offset, size,
 108                                     is_signed, filter_type);
 109 }
 110 EXPORT_SYMBOL_GPL(trace_define_field);
 111
 112 #define __common_field(type, item)                                      \
 113         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 114                                    "common_" #item,                     \
 115                                    offsetof(typeof(ent), item),         \
 116                                    sizeof(ent.item),                    \
 117                                    is_signed_type(type), FILTER_OTHER); \
 118         if (ret)                                                        \
 119                 return ret;
 120
 121 static int trace_define_common_fields(void)
 122 {
 123         int ret;
 124         struct trace_entry ent;
 125
 126         __common_field(unsigned short, type);
 127         __common_field(unsigned char, flags);
 128         __common_field(unsigned char, preempt_count);
 129         __common_field(int, pid);
 130
 131         return ret;
 132 }
 133
 134 void trace_destroy_fields(struct ftrace_event_call *call)
 135 {
 136         struct ftrace_event_field *field, *next;
 137         struct list_head *head;
 138
 139         head = trace_get_fields(call);
 140         list_for_each_entry_safe(field, next, head, link) {
 141                 list_del(&field->link);
 142                 kmem_cache_free(field_cachep, field);
 143         }
 144 }
 145
 146 int trace_event_raw_init(struct ftrace_event_call *call)
 147 {
 148         int id;
 149
 150         id = register_ftrace_event(&call->event);
 151         if (!id)
 152                 return -ENODEV;
 153
 154         return 0;
 155 }
 156 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 157
 158 int ftrace_event_reg(struct ftrace_event_call *call,
 159                      enum trace_reg type, void *data)
 160 {
 161         struct ftrace_event_file *file = data;
 162
 163         switch (type) {
 164         case TRACE_REG_REGISTER:
 165                 return tracepoint_probe_register(call->name,
 166                                                  call->class->probe,
 167                                                  file);
 168         case TRACE_REG_UNREGISTER:
 169                 tracepoint_probe_unregister(call->name,
 170                                             call->class->probe,
 171                                             file);
 172                 return 0;
 173
 174 #ifdef CONFIG_PERF_EVENTS
 175         case TRACE_REG_PERF_REGISTER:
 176                 return tracepoint_probe_register(call->name,
 177                                                  call->class->perf_probe,
 178                                                  call);
 179         case TRACE_REG_PERF_UNREGISTER:
 180                 tracepoint_probe_unregister(call->name,
 181                                             call->class->perf_probe,
 182                                             call);
 183                 return 0;
 184         case TRACE_REG_PERF_OPEN:
 185         case TRACE_REG_PERF_CLOSE:
 186         case TRACE_REG_PERF_ADD:
 187         case TRACE_REG_PERF_DEL:
 188                 return 0;
 189 #endif
 190         }
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 194
 195 void trace_event_enable_cmd_record(bool enable)
 196 {
 197         struct ftrace_event_file *file;
 198         struct trace_array *tr;
 199
 200         mutex_lock(&event_mutex);
 201         do_for_each_event_file(tr, file) {
 202
 203                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 204                         continue;
 205
 206                 if (enable) {
 207                         tracing_start_cmdline_record();
 208                         file->flags |= FTRACE_EVENT_FL_RECORDED_CMD;
 209                 } else {
 210                         tracing_stop_cmdline_record();
 211                         file->flags &= ~FTRACE_EVENT_FL_RECORDED_CMD;
 212                 }
 213         } while_for_each_event_file();
 214         mutex_unlock(&event_mutex);
 215 }
 216
 217 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 218                                        int enable)
 219 {
 220         struct ftrace_event_call *call = file->event_call;
 221         int ret = 0;
 222
 223         switch (enable) {
 224         case 0:
 225                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
 226                         file->flags &= ~FTRACE_EVENT_FL_ENABLED;
 227                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 228                                 tracing_stop_cmdline_record();
 229                                 file->flags &= ~FTRACE_EVENT_FL_RECORDED_CMD;
 230                         }
 231                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 232                 }
 233                 break;
 234         case 1:
 235                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 236                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 237                                 tracing_start_cmdline_record();
 238                                 file->flags |= FTRACE_EVENT_FL_RECORDED_CMD;
 239                         }
 240                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 241                         if (ret) {
 242                                 tracing_stop_cmdline_record();
 243                                 pr_info("event trace: Could not enable event "
 244                                         "%s\n", call->name);
 245                                 break;
 246                         }
 247                         file->flags |= FTRACE_EVENT_FL_ENABLED;
 248
 249                         /* WAS_ENABLED gets set but never cleared. */
 250                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 251                 }
 252                 break;
 253         }
 254
 255         return ret;
 256 }
 257
 258 static void ftrace_clear_events(struct trace_array *tr)
 259 {
 260         struct ftrace_event_file *file;
 261
 262         mutex_lock(&event_mutex);
 263         list_for_each_entry(file, &tr->events, list) {
 264                 ftrace_event_enable_disable(file, 0);
 265         }
 266         mutex_unlock(&event_mutex);
 267 }
 268
 269 static void __put_system(struct event_subsystem *system)
 270 {
 271         struct event_filter *filter = system->filter;
 272
 273         WARN_ON_ONCE(system->ref_count == 0);
 274         if (--system->ref_count)
 275                 return;
 276
 277         list_del(&system->list);
 278
 279         if (filter) {
 280                 kfree(filter->filter_string);
 281                 kfree(filter);
 282         }
 283         kfree(system);
 284 }
 285
 286 static void __get_system(struct event_subsystem *system)
 287 {
 288         WARN_ON_ONCE(system->ref_count == 0);
 289         system->ref_count++;
 290 }
 291
 292 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 293 {
 294         WARN_ON_ONCE(dir->ref_count == 0);
 295         dir->ref_count++;
 296         __get_system(dir->subsystem);
 297 }
 298
 299 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 300 {
 301         WARN_ON_ONCE(dir->ref_count == 0);
 302         /* If the subsystem is about to be freed, the dir must be too */
 303         WARN_ON_ONCE(dir->subsystem->ref_count == 1 && dir->ref_count != 1);
 304
 305         __put_system(dir->subsystem);
 306         if (!--dir->ref_count)
 307                 kfree(dir);
 308 }
 309
 310 static void put_system(struct ftrace_subsystem_dir *dir)
 311 {
 312         mutex_lock(&event_mutex);
 313         __put_system_dir(dir);
 314         mutex_unlock(&event_mutex);
 315 }
 316
 317 /*
 318  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 319  */
 320 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 321                                   const char *sub, const char *event, int set)
 322 {
 323         struct ftrace_event_file *file;
 324         struct ftrace_event_call *call;
 325         int ret = -EINVAL;
 326
 327         mutex_lock(&event_mutex);
 328         list_for_each_entry(file, &tr->events, list) {
 329
 330                 call = file->event_call;
 331
 332                 if (!call->name || !call->class || !call->class->reg)
 333                         continue;
 334
 335                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 336                         continue;
 337
 338                 if (match &&
 339                     strcmp(match, call->name) != 0 &&
 340                     strcmp(match, call->class->system) != 0)
 341                         continue;
 342
 343                 if (sub && strcmp(sub, call->class->system) != 0)
 344                         continue;
 345
 346                 if (event && strcmp(event, call->name) != 0)
 347                         continue;
 348
 349                 ftrace_event_enable_disable(file, set);
 350
 351                 ret = 0;
 352         }
 353         mutex_unlock(&event_mutex);
 354
 355         return ret;
 356 }
 357
 358 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 359 {
 360         char *event = NULL, *sub = NULL, *match;
 361
 362         /*
 363          * The buf format can be <subsystem>:<event-name>
 364          *  *:<event-name> means any event by that name.
 365          *  :<event-name> is the same.
 366          *
 367          *  <subsystem>:* means all events in that subsystem
 368          *  <subsystem>: means the same.
 369          *
 370          *  <name> (no ':') means all events in a subsystem with
 371          *  the name <name> or any event that matches <name>
 372          */
 373
 374         match = strsep(&buf, ":");
 375         if (buf) {
 376                 sub = match;
 377                 event = buf;
 378                 match = NULL;
 379
 380                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 381                         sub = NULL;
 382                 if (!strlen(event) || strcmp(event, "*") == 0)
 383                         event = NULL;
 384         }
 385
 386         return __ftrace_set_clr_event(tr, match, sub, event, set);
 387 }
 388
 389 /**
 390  * trace_set_clr_event - enable or disable an event
 391  * @system: system name to match (NULL for any system)
 392  * @event: event name to match (NULL for all events, within system)
 393  * @set: 1 to enable, 0 to disable
 394  *
 395  * This is a way for other parts of the kernel to enable or disable
 396  * event recording.
 397  *
 398  * Returns 0 on success, -EINVAL if the parameters do not match any
 399  * registered events.
 400  */
 401 int trace_set_clr_event(const char *system, const char *event, int set)
 402 {
 403         struct trace_array *tr = top_trace_array();
 404
 405         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 406 }
 407 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 408
 409 /* 128 should be much more than enough */
 410 #define EVENT_BUF_SIZE          127
 411
 412 static ssize_t
 413 ftrace_event_write(struct file *file, const char __user *ubuf,
 414                    size_t cnt, loff_t *ppos)
 415 {
 416         struct trace_parser parser;
 417         struct seq_file *m = file->private_data;
 418         struct trace_array *tr = m->private;
 419         ssize_t read, ret;
 420
 421         if (!cnt)
 422                 return 0;
 423
 424         ret = tracing_update_buffers();
 425         if (ret < 0)
 426                 return ret;
 427
 428         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 429                 return -ENOMEM;
 430
 431         read = trace_get_user(&parser, ubuf, cnt, ppos);
 432
 433         if (read >= 0 && trace_parser_loaded((&parser))) {
 434                 int set = 1;
 435
 436                 if (*parser.buffer == '!')
 437                         set = 0;
 438
 439                 parser.buffer[parser.idx] = 0;
 440
 441                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 442                 if (ret)
 443                         goto out_put;
 444         }
 445
 446         ret = read;
 447
 448  out_put:
 449         trace_parser_put(&parser);
 450
 451         return ret;
 452 }
 453
 454 static void *
 455 t_next(struct seq_file *m, void *v, loff_t *pos)
 456 {
 457         struct ftrace_event_file *file = v;
 458         struct ftrace_event_call *call;
 459         struct trace_array *tr = m->private;
 460
 461         (*pos)++;
 462
 463         list_for_each_entry_continue(file, &tr->events, list) {
 464                 call = file->event_call;
 465                 /*
 466                  * The ftrace subsystem is for showing formats only.
 467                  * They can not be enabled or disabled via the event files.
 468                  */
 469                 if (call->class && call->class->reg)
 470                         return file;
 471         }
 472
 473         return NULL;
 474 }
 475
 476 static void *t_start(struct seq_file *m, loff_t *pos)
 477 {
 478         struct ftrace_event_file *file;
 479         struct trace_array *tr = m->private;
 480         loff_t l;
 481
 482         mutex_lock(&event_mutex);
 483
 484         file = list_entry(&tr->events, struct ftrace_event_file, list);
 485         for (l = 0; l <= *pos; ) {
 486                 file = t_next(m, file, &l);
 487                 if (!file)
 488                         break;
 489         }
 490         return file;
 491 }
 492
 493 static void *
 494 s_next(struct seq_file *m, void *v, loff_t *pos)
 495 {
 496         struct ftrace_event_file *file = v;
 497         struct trace_array *tr = m->private;
 498
 499         (*pos)++;
 500
 501         list_for_each_entry_continue(file, &tr->events, list) {
 502                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 503                         return file;
 504         }
 505
 506         return NULL;
 507 }
 508
 509 static void *s_start(struct seq_file *m, loff_t *pos)
 510 {
 511         struct ftrace_event_file *file;
 512         struct trace_array *tr = m->private;
 513         loff_t l;
 514
 515         mutex_lock(&event_mutex);
 516
 517         file = list_entry(&tr->events, struct ftrace_event_file, list);
 518         for (l = 0; l <= *pos; ) {
 519                 file = s_next(m, file, &l);
 520                 if (!file)
 521                         break;
 522         }
 523         return file;
 524 }
 525
 526 static int t_show(struct seq_file *m, void *v)
 527 {
 528         struct ftrace_event_file *file = v;
 529         struct ftrace_event_call *call = file->event_call;
 530
 531         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 532                 seq_printf(m, "%s:", call->class->system);
 533         seq_printf(m, "%s\n", call->name);
 534
 535         return 0;
 536 }
 537
 538 static void t_stop(struct seq_file *m, void *p)
 539 {
 540         mutex_unlock(&event_mutex);
 541 }
 542
 543 static ssize_t
 544 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 545                   loff_t *ppos)
 546 {
 547         struct ftrace_event_file *file = filp->private_data;
 548         char *buf;
 549
 550         if (file->flags & FTRACE_EVENT_FL_ENABLED)
 551                 buf = "1\n";
 552         else
 553                 buf = "0\n";
 554
 555         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 556 }
 557
 558 static ssize_t
 559 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 560                    loff_t *ppos)
 561 {
 562         struct ftrace_event_file *file = filp->private_data;
 563         unsigned long val;
 564         int ret;
 565
 566         if (!file)
 567                 return -EINVAL;
 568
 569         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 570         if (ret)
 571                 return ret;
 572
 573         ret = tracing_update_buffers();
 574         if (ret < 0)
 575                 return ret;
 576
 577         switch (val) {
 578         case 0:
 579         case 1:
 580                 mutex_lock(&event_mutex);
 581                 ret = ftrace_event_enable_disable(file, val);
 582                 mutex_unlock(&event_mutex);
 583                 break;
 584
 585         default:
 586                 return -EINVAL;
 587         }
 588
 589         *ppos += cnt;
 590
 591         return ret ? ret : cnt;
 592 }
 593
 594 static ssize_t
 595 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 596                    loff_t *ppos)
 597 {
 598         const char set_to_char[4] = { '?', '0', '1', 'X' };
 599         struct ftrace_subsystem_dir *dir = filp->private_data;
 600         struct event_subsystem *system = dir->subsystem;
 601         struct ftrace_event_call *call;
 602         struct ftrace_event_file *file;
 603         struct trace_array *tr = dir->tr;
 604         char buf[2];
 605         int set = 0;
 606         int ret;
 607
 608         mutex_lock(&event_mutex);
 609         list_for_each_entry(file, &tr->events, list) {
 610                 call = file->event_call;
 611                 if (!call->name || !call->class || !call->class->reg)
 612                         continue;
 613
 614                 if (system && strcmp(call->class->system, system->name) != 0)
 615                         continue;
 616
 617                 /*
 618                  * We need to find out if all the events are set
 619                  * or if all events or cleared, or if we have
 620                  * a mixture.
 621                  */
 622                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 623
 624                 /*
 625                  * If we have a mixture, no need to look further.
 626                  */
 627                 if (set == 3)
 628                         break;
 629         }
 630         mutex_unlock(&event_mutex);
 631
 632         buf[0] = set_to_char[set];
 633         buf[1] = '\n';
 634
 635         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 636
 637         return ret;
 638 }
 639
 640 static ssize_t
 641 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 642                     loff_t *ppos)
 643 {
 644         struct ftrace_subsystem_dir *dir = filp->private_data;
 645         struct event_subsystem *system = dir->subsystem;
 646         const char *name = NULL;
 647         unsigned long val;
 648         ssize_t ret;
 649
 650         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 651         if (ret)
 652                 return ret;
 653
 654         ret = tracing_update_buffers();
 655         if (ret < 0)
 656                 return ret;
 657
 658         if (val != 0 && val != 1)
 659                 return -EINVAL;
 660
 661         /*
 662          * Opening of "enable" adds a ref count to system,
 663          * so the name is safe to use.
 664          */
 665         if (system)
 666                 name = system->name;
 667
 668         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 669         if (ret)
 670                 goto out;
 671
 672         ret = cnt;
 673
 674 out:
 675         *ppos += cnt;
 676
 677         return ret;
 678 }
 679
 680 enum {
 681         FORMAT_HEADER           = 1,
 682         FORMAT_FIELD_SEPERATOR  = 2,
 683         FORMAT_PRINTFMT         = 3,
 684 };
 685
 686 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 687 {
 688         struct ftrace_event_call *call = m->private;
 689         struct ftrace_event_field *field;
 690         struct list_head *common_head = &ftrace_common_fields;
 691         struct list_head *head = trace_get_fields(call);
 692
 693         (*pos)++;
 694
 695         switch ((unsigned long)v) {
 696         case FORMAT_HEADER:
 697                 if (unlikely(list_empty(common_head)))
 698                         return NULL;
 699
 700                 field = list_entry(common_head->prev,
 701                                    struct ftrace_event_field, link);
 702                 return field;
 703
 704         case FORMAT_FIELD_SEPERATOR:
 705                 if (unlikely(list_empty(head)))
 706                         return NULL;
 707
 708                 field = list_entry(head->prev, struct ftrace_event_field, link);
 709                 return field;
 710
 711         case FORMAT_PRINTFMT:
 712                 /* all done */
 713                 return NULL;
 714         }
 715
 716         field = v;
 717         if (field->link.prev == common_head)
 718                 return (void *)FORMAT_FIELD_SEPERATOR;
 719         else if (field->link.prev == head)
 720                 return (void *)FORMAT_PRINTFMT;
 721
 722         field = list_entry(field->link.prev, struct ftrace_event_field, link);
 723
 724         return field;
 725 }
 726
 727 static void *f_start(struct seq_file *m, loff_t *pos)
 728 {
 729         loff_t l = 0;
 730         void *p;
 731
 732         /* Start by showing the header */
 733         if (!*pos)
 734                 return (void *)FORMAT_HEADER;
 735
 736         p = (void *)FORMAT_HEADER;
 737         do {
 738                 p = f_next(m, p, &l);
 739         } while (p && l < *pos);
 740
 741         return p;
 742 }
 743
 744 static int f_show(struct seq_file *m, void *v)
 745 {
 746         struct ftrace_event_call *call = m->private;
 747         struct ftrace_event_field *field;
 748         const char *array_descriptor;
 749
 750         switch ((unsigned long)v) {
 751         case FORMAT_HEADER:
 752                 seq_printf(m, "name: %s\n", call->name);
 753                 seq_printf(m, "ID: %d\n", call->event.type);
 754                 seq_printf(m, "format:\n");
 755                 return 0;
 756
 757         case FORMAT_FIELD_SEPERATOR:
 758                 seq_putc(m, '\n');
 759                 return 0;
 760
 761         case FORMAT_PRINTFMT:
 762                 seq_printf(m, "\nprint fmt: %s\n",
 763                            call->print_fmt);
 764                 return 0;
 765         }
 766
 767         field = v;
 768
 769         /*
 770          * Smartly shows the array type(except dynamic array).
 771          * Normal:
 772          *      field:TYPE VAR
 773          * If TYPE := TYPE[LEN], it is shown:
 774          *      field:TYPE VAR[LEN]
 775          */
 776         array_descriptor = strchr(field->type, '[');
 777
 778         if (!strncmp(field->type, "__data_loc", 10))
 779                 array_descriptor = NULL;
 780
 781         if (!array_descriptor)
 782                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 783                            field->type, field->name, field->offset,
 784                            field->size, !!field->is_signed);
 785         else
 786                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 787                            (int)(array_descriptor - field->type),
 788                            field->type, field->name,
 789                            array_descriptor, field->offset,
 790                            field->size, !!field->is_signed);
 791
 792         return 0;
 793 }
 794
 795 static void f_stop(struct seq_file *m, void *p)
 796 {
 797 }
 798
 799 static const struct seq_operations trace_format_seq_ops = {
 800         .start          = f_start,
 801         .next           = f_next,
 802         .stop           = f_stop,
 803         .show           = f_show,
 804 };
 805
 806 static int trace_format_open(struct inode *inode, struct file *file)
 807 {
 808         struct ftrace_event_call *call = inode->i_private;
 809         struct seq_file *m;
 810         int ret;
 811
 812         ret = seq_open(file, &trace_format_seq_ops);
 813         if (ret < 0)
 814                 return ret;
 815
 816         m = file->private_data;
 817         m->private = call;
 818
 819         return 0;
 820 }
 821
 822 static ssize_t
 823 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 824 {
 825         struct ftrace_event_call *call = filp->private_data;
 826         struct trace_seq *s;
 827         int r;
 828
 829         if (*ppos)
 830                 return 0;
 831
 832         s = kmalloc(sizeof(*s), GFP_KERNEL);
 833         if (!s)
 834                 return -ENOMEM;
 835
 836         trace_seq_init(s);
 837         trace_seq_printf(s, "%d\n", call->event.type);
 838
 839         r = simple_read_from_buffer(ubuf, cnt, ppos,
 840                                     s->buffer, s->len);
 841         kfree(s);
 842         return r;
 843 }
 844
 845 static ssize_t
 846 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 847                   loff_t *ppos)
 848 {
 849         struct ftrace_event_call *call = filp->private_data;
 850         struct trace_seq *s;
 851         int r;
 852
 853         if (*ppos)
 854                 return 0;
 855
 856         s = kmalloc(sizeof(*s), GFP_KERNEL);
 857         if (!s)
 858                 return -ENOMEM;
 859
 860         trace_seq_init(s);
 861
 862         print_event_filter(call, s);
 863         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 864
 865         kfree(s);
 866
 867         return r;
 868 }
 869
 870 static ssize_t
 871 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 872                    loff_t *ppos)
 873 {
 874         struct ftrace_event_call *call = filp->private_data;
 875         char *buf;
 876         int err;
 877
 878         if (cnt >= PAGE_SIZE)
 879                 return -EINVAL;
 880
 881         buf = (char *)__get_free_page(GFP_TEMPORARY);
 882         if (!buf)
 883                 return -ENOMEM;
 884
 885         if (copy_from_user(buf, ubuf, cnt)) {
 886                 free_page((unsigned long) buf);
 887                 return -EFAULT;
 888         }
 889         buf[cnt] = '\0';
 890
 891         err = apply_event_filter(call, buf);
 892         free_page((unsigned long) buf);
 893         if (err < 0)
 894                 return err;
 895
 896         *ppos += cnt;
 897
 898         return cnt;
 899 }
 900
 901 static LIST_HEAD(event_subsystems);
 902
 903 static int subsystem_open(struct inode *inode, struct file *filp)
 904 {
 905         struct event_subsystem *system = NULL;
 906         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
 907         struct trace_array *tr;
 908         int ret;
 909
 910         /* Make sure the system still exists */
 911         mutex_lock(&event_mutex);
 912         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
 913                 list_for_each_entry(dir, &tr->systems, list) {
 914                         if (dir == inode->i_private) {
 915                                 /* Don't open systems with no events */
 916                                 if (dir->nr_events) {
 917                                         __get_system_dir(dir);
 918                                         system = dir->subsystem;
 919                                 }
 920                                 goto exit_loop;
 921                         }
 922                 }
 923         }
 924  exit_loop:
 925         mutex_unlock(&event_mutex);
 926
 927         if (!system)
 928                 return -ENODEV;
 929
 930         /* Some versions of gcc think dir can be uninitialized here */
 931         WARN_ON(!dir);
 932
 933         ret = tracing_open_generic(inode, filp);
 934         if (ret < 0)
 935                 put_system(dir);
 936
 937         return ret;
 938 }
 939
 940 static int system_tr_open(struct inode *inode, struct file *filp)
 941 {
 942         struct ftrace_subsystem_dir *dir;
 943         struct trace_array *tr = inode->i_private;
 944         int ret;
 945
 946         /* Make a temporary dir that has no system but points to tr */
 947         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
 948         if (!dir)
 949                 return -ENOMEM;
 950
 951         dir->tr = tr;
 952
 953         ret = tracing_open_generic(inode, filp);
 954         if (ret < 0)
 955                 kfree(dir);
 956
 957         filp->private_data = dir;
 958
 959         return ret;
 960 }
 961
 962 static int subsystem_release(struct inode *inode, struct file *file)
 963 {
 964         struct ftrace_subsystem_dir *dir = file->private_data;
 965
 966         /*
 967          * If dir->subsystem is NULL, then this is a temporary
 968          * descriptor that was made for a trace_array to enable
 969          * all subsystems.
 970          */
 971         if (dir->subsystem)
 972                 put_system(dir);
 973         else
 974                 kfree(dir);
 975
 976         return 0;
 977 }
 978
 979 static ssize_t
 980 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 981                       loff_t *ppos)
 982 {
 983         struct ftrace_subsystem_dir *dir = filp->private_data;
 984         struct event_subsystem *system = dir->subsystem;
 985         struct trace_seq *s;
 986         int r;
 987
 988         if (*ppos)
 989                 return 0;
 990
 991         s = kmalloc(sizeof(*s), GFP_KERNEL);
 992         if (!s)
 993                 return -ENOMEM;
 994
 995         trace_seq_init(s);
 996
 997         print_subsystem_event_filter(system, s);
 998         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 999
1000         kfree(s);
1001
1002         return r;
1003 }
1004
1005 static ssize_t
1006 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1007                        loff_t *ppos)
1008 {
1009         struct ftrace_subsystem_dir *dir = filp->private_data;
1010         char *buf;
1011         int err;
1012
1013         if (cnt >= PAGE_SIZE)
1014                 return -EINVAL;
1015
1016         buf = (char *)__get_free_page(GFP_TEMPORARY);
1017         if (!buf)
1018                 return -ENOMEM;
1019
1020         if (copy_from_user(buf, ubuf, cnt)) {
1021                 free_page((unsigned long) buf);
1022                 return -EFAULT;
1023         }
1024         buf[cnt] = '\0';
1025
1026         err = apply_subsystem_event_filter(dir, buf);
1027         free_page((unsigned long) buf);
1028         if (err < 0)
1029                 return err;
1030
1031         *ppos += cnt;
1032
1033         return cnt;
1034 }
1035
1036 static ssize_t
1037 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1038 {
1039         int (*func)(struct trace_seq *s) = filp->private_data;
1040         struct trace_seq *s;
1041         int r;
1042
1043         if (*ppos)
1044                 return 0;
1045
1046         s = kmalloc(sizeof(*s), GFP_KERNEL);
1047         if (!s)
1048                 return -ENOMEM;
1049
1050         trace_seq_init(s);
1051
1052         func(s);
1053         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1054
1055         kfree(s);
1056
1057         return r;
1058 }
1059
1060 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1061 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1062
1063 static const struct seq_operations show_event_seq_ops = {
1064         .start = t_start,
1065         .next = t_next,
1066         .show = t_show,
1067         .stop = t_stop,
1068 };
1069
1070 static const struct seq_operations show_set_event_seq_ops = {
1071         .start = s_start,
1072         .next = s_next,
1073         .show = t_show,
1074         .stop = t_stop,
1075 };
1076
1077 static const struct file_operations ftrace_avail_fops = {
1078         .open = ftrace_event_avail_open,
1079         .read = seq_read,
1080         .llseek = seq_lseek,
1081         .release = seq_release,
1082 };
1083
1084 static const struct file_operations ftrace_set_event_fops = {
1085         .open = ftrace_event_set_open,
1086         .read = seq_read,
1087         .write = ftrace_event_write,
1088         .llseek = seq_lseek,
1089         .release = seq_release,
1090 };
1091
1092 static const struct file_operations ftrace_enable_fops = {
1093         .open = tracing_open_generic,
1094         .read = event_enable_read,
1095         .write = event_enable_write,
1096         .llseek = default_llseek,
1097 };
1098
1099 static const struct file_operations ftrace_event_format_fops = {
1100         .open = trace_format_open,
1101         .read = seq_read,
1102         .llseek = seq_lseek,
1103         .release = seq_release,
1104 };
1105
1106 static const struct file_operations ftrace_event_id_fops = {
1107         .open = tracing_open_generic,
1108         .read = event_id_read,
1109         .llseek = default_llseek,
1110 };
1111
1112 static const struct file_operations ftrace_event_filter_fops = {
1113         .open = tracing_open_generic,
1114         .read = event_filter_read,
1115         .write = event_filter_write,
1116         .llseek = default_llseek,
1117 };
1118
1119 static const struct file_operations ftrace_subsystem_filter_fops = {
1120         .open = subsystem_open,
1121         .read = subsystem_filter_read,
1122         .write = subsystem_filter_write,
1123         .llseek = default_llseek,
1124         .release = subsystem_release,
1125 };
1126
1127 static const struct file_operations ftrace_system_enable_fops = {
1128         .open = subsystem_open,
1129         .read = system_enable_read,
1130         .write = system_enable_write,
1131         .llseek = default_llseek,
1132         .release = subsystem_release,
1133 };
1134
1135 static const struct file_operations ftrace_tr_enable_fops = {
1136         .open = system_tr_open,
1137         .read = system_enable_read,
1138         .write = system_enable_write,
1139         .llseek = default_llseek,
1140         .release = subsystem_release,
1141 };
1142
1143 static const struct file_operations ftrace_show_header_fops = {
1144         .open = tracing_open_generic,
1145         .read = show_header,
1146         .llseek = default_llseek,
1147 };
1148
1149 static int
1150 ftrace_event_open(struct inode *inode, struct file *file,
1151                   const struct seq_operations *seq_ops)
1152 {
1153         struct seq_file *m;
1154         int ret;
1155
1156         ret = seq_open(file, seq_ops);
1157         if (ret < 0)
1158                 return ret;
1159         m = file->private_data;
1160         /* copy tr over to seq ops */
1161         m->private = inode->i_private;
1162
1163         return ret;
1164 }
1165
1166 static int
1167 ftrace_event_avail_open(struct inode *inode, struct file *file)
1168 {
1169         const struct seq_operations *seq_ops = &show_event_seq_ops;
1170
1171         return ftrace_event_open(inode, file, seq_ops);
1172 }
1173
1174 static int
1175 ftrace_event_set_open(struct inode *inode, struct file *file)
1176 {
1177         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1178         struct trace_array *tr = inode->i_private;
1179
1180         if ((file->f_mode & FMODE_WRITE) &&
1181             (file->f_flags & O_TRUNC))
1182                 ftrace_clear_events(tr);
1183
1184         return ftrace_event_open(inode, file, seq_ops);
1185 }
1186
1187 static struct event_subsystem *
1188 create_new_subsystem(const char *name)
1189 {
1190         struct event_subsystem *system;
1191
1192         /* need to create new entry */
1193         system = kmalloc(sizeof(*system), GFP_KERNEL);
1194         if (!system)
1195                 return NULL;
1196
1197         system->ref_count = 1;
1198         system->name = name;
1199
1200         system->filter = NULL;
1201
1202         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1203         if (!system->filter)
1204                 goto out_free;
1205
1206         list_add(&system->list, &event_subsystems);
1207
1208         return system;
1209
1210  out_free:
1211         kfree(system);
1212         return NULL;
1213 }
1214
1215 static struct dentry *
1216 event_subsystem_dir(struct trace_array *tr, const char *name,
1217                     struct ftrace_event_file *file, struct dentry *parent)
1218 {
1219         struct ftrace_subsystem_dir *dir;
1220         struct event_subsystem *system;
1221         struct dentry *entry;
1222
1223         /* First see if we did not already create this dir */
1224         list_for_each_entry(dir, &tr->systems, list) {
1225                 system = dir->subsystem;
1226                 if (strcmp(system->name, name) == 0) {
1227                         dir->nr_events++;
1228                         file->system = dir;
1229                         return dir->entry;
1230                 }
1231         }
1232
1233         /* Now see if the system itself exists. */
1234         list_for_each_entry(system, &event_subsystems, list) {
1235                 if (strcmp(system->name, name) == 0)
1236                         break;
1237         }
1238         /* Reset system variable when not found */
1239         if (&system->list == &event_subsystems)
1240                 system = NULL;
1241
1242         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1243         if (!dir)
1244                 goto out_fail;
1245
1246         if (!system) {
1247                 system = create_new_subsystem(name);
1248                 if (!system)
1249                         goto out_free;
1250         } else
1251                 __get_system(system);
1252
1253         dir->entry = debugfs_create_dir(name, parent);
1254         if (!dir->entry) {
1255                 pr_warning("Failed to create system directory %s\n", name);
1256                 __put_system(system);
1257                 goto out_free;
1258         }
1259
1260         dir->tr = tr;
1261         dir->ref_count = 1;
1262         dir->nr_events = 1;
1263         dir->subsystem = system;
1264         file->system = dir;
1265
1266         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1267                                     &ftrace_subsystem_filter_fops);
1268         if (!entry) {
1269                 kfree(system->filter);
1270                 system->filter = NULL;
1271                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1272         }
1273
1274         trace_create_file("enable", 0644, dir->entry, dir,
1275                           &ftrace_system_enable_fops);
1276
1277         list_add(&dir->list, &tr->systems);
1278
1279         return dir->entry;
1280
1281  out_free:
1282         kfree(dir);
1283  out_fail:
1284         /* Only print this message if failed on memory allocation */
1285         if (!dir || !system)
1286                 pr_warning("No memory to create event subsystem %s\n",
1287                            name);
1288         return NULL;
1289 }
1290
1291 static int
1292 event_create_dir(struct dentry *parent,
1293                  struct ftrace_event_file *file,
1294                  const struct file_operations *id,
1295                  const struct file_operations *enable,
1296                  const struct file_operations *filter,
1297                  const struct file_operations *format)
1298 {
1299         struct ftrace_event_call *call = file->event_call;
1300         struct trace_array *tr = file->tr;
1301         struct list_head *head;
1302         struct dentry *d_events;
1303         int ret;
1304
1305         /*
1306          * If the trace point header did not define TRACE_SYSTEM
1307          * then the system would be called "TRACE_SYSTEM".
1308          */
1309         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1310                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1311                 if (!d_events)
1312                         return -ENOMEM;
1313         } else
1314                 d_events = parent;
1315
1316         file->dir = debugfs_create_dir(call->name, d_events);
1317         if (!file->dir) {
1318                 pr_warning("Could not create debugfs '%s' directory\n",
1319                            call->name);
1320                 return -1;
1321         }
1322
1323         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1324                 trace_create_file("enable", 0644, file->dir, file,
1325                                   enable);
1326
1327 #ifdef CONFIG_PERF_EVENTS
1328         if (call->event.type && call->class->reg)
1329                 trace_create_file("id", 0444, file->dir, call,
1330                                   id);
1331 #endif
1332
1333         /*
1334          * Other events may have the same class. Only update
1335          * the fields if they are not already defined.
1336          */
1337         head = trace_get_fields(call);
1338         if (list_empty(head)) {
1339                 ret = call->class->define_fields(call);
1340                 if (ret < 0) {
1341                         pr_warning("Could not initialize trace point"
1342                                    " events/%s\n", call->name);
1343                         return -1;
1344                 }
1345         }
1346         trace_create_file("filter", 0644, file->dir, call,
1347                           filter);
1348
1349         trace_create_file("format", 0444, file->dir, call,
1350                           format);
1351
1352         return 0;
1353 }
1354
1355 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
1356 {
1357         if (!dir)
1358                 return;
1359
1360         if (!--dir->nr_events) {
1361                 debugfs_remove_recursive(dir->entry);
1362                 list_del(&dir->list);
1363                 __put_system_dir(dir);
1364         }
1365 }
1366
1367 static void remove_event_from_tracers(struct ftrace_event_call *call)
1368 {
1369         struct ftrace_event_file *file;
1370         struct trace_array *tr;
1371
1372         do_for_each_event_file_safe(tr, file) {
1373
1374                 if (file->event_call != call)
1375                         continue;
1376
1377                 list_del(&file->list);
1378                 debugfs_remove_recursive(file->dir);
1379                 remove_subsystem(file->system);
1380                 kmem_cache_free(file_cachep, file);
1381
1382                 /*
1383                  * The do_for_each_event_file_safe() is
1384                  * a double loop. After finding the call for this
1385                  * trace_array, we use break to jump to the next
1386                  * trace_array.
1387                  */
1388                 break;
1389         } while_for_each_event_file();
1390 }
1391
1392 static void event_remove(struct ftrace_event_call *call)
1393 {
1394         struct trace_array *tr;
1395         struct ftrace_event_file *file;
1396
1397         do_for_each_event_file(tr, file) {
1398                 if (file->event_call != call)
1399                         continue;
1400                 ftrace_event_enable_disable(file, 0);
1401                 /*
1402                  * The do_for_each_event_file() is
1403                  * a double loop. After finding the call for this
1404                  * trace_array, we use break to jump to the next
1405                  * trace_array.
1406                  */
1407                 break;
1408         } while_for_each_event_file();
1409
1410         if (call->event.funcs)
1411                 __unregister_ftrace_event(&call->event);
1412         remove_event_from_tracers(call);
1413         list_del(&call->list);
1414 }
1415
1416 static int event_init(struct ftrace_event_call *call)
1417 {
1418         int ret = 0;
1419
1420         if (WARN_ON(!call->name))
1421                 return -EINVAL;
1422
1423         if (call->class->raw_init) {
1424                 ret = call->class->raw_init(call);
1425                 if (ret < 0 && ret != -ENOSYS)
1426                         pr_warn("Could not initialize trace events/%s\n",
1427                                 call->name);
1428         }
1429
1430         return ret;
1431 }
1432
1433 static int
1434 __register_event(struct ftrace_event_call *call, struct module *mod)
1435 {
1436         int ret;
1437
1438         ret = event_init(call);
1439         if (ret < 0)
1440                 return ret;
1441
1442         list_add(&call->list, &ftrace_events);
1443         call->mod = mod;
1444
1445         return 0;
1446 }
1447
1448 /* Add an event to a trace directory */
1449 static int
1450 __trace_add_new_event(struct ftrace_event_call *call,
1451                       struct trace_array *tr,
1452                       const struct file_operations *id,
1453                       const struct file_operations *enable,
1454                       const struct file_operations *filter,
1455                       const struct file_operations *format)
1456 {
1457         struct ftrace_event_file *file;
1458
1459         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1460         if (!file)
1461                 return -ENOMEM;
1462
1463         file->event_call = call;
1464         file->tr = tr;
1465         list_add(&file->list, &tr->events);
1466
1467         return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1468 }
1469
1470 /*
1471  * Just create a decriptor for early init. A descriptor is required
1472  * for enabling events at boot. We want to enable events before
1473  * the filesystem is initialized.
1474  */
1475 static __init int
1476 __trace_early_add_new_event(struct ftrace_event_call *call,
1477                             struct trace_array *tr)
1478 {
1479         struct ftrace_event_file *file;
1480
1481         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1482         if (!file)
1483                 return -ENOMEM;
1484
1485         file->event_call = call;
1486         file->tr = tr;
1487         list_add(&file->list, &tr->events);
1488
1489         return 0;
1490 }
1491
1492 struct ftrace_module_file_ops;
1493 static void __add_event_to_tracers(struct ftrace_event_call *call,
1494                                    struct ftrace_module_file_ops *file_ops);
1495
1496 /* Add an additional event_call dynamically */
1497 int trace_add_event_call(struct ftrace_event_call *call)
1498 {
1499         int ret;
1500         mutex_lock(&event_mutex);
1501
1502         ret = __register_event(call, NULL);
1503         if (ret >= 0)
1504                 __add_event_to_tracers(call, NULL);
1505
1506         mutex_unlock(&event_mutex);
1507         return ret;
1508 }
1509
1510 /*
1511  * Must be called under locking both of event_mutex and trace_event_mutex.
1512  */
1513 static void __trace_remove_event_call(struct ftrace_event_call *call)
1514 {
1515         event_remove(call);
1516         trace_destroy_fields(call);
1517         destroy_preds(call);
1518 }
1519
1520 /* Remove an event_call */
1521 void trace_remove_event_call(struct ftrace_event_call *call)
1522 {
1523         mutex_lock(&event_mutex);
1524         down_write(&trace_event_mutex);
1525         __trace_remove_event_call(call);
1526         up_write(&trace_event_mutex);
1527         mutex_unlock(&event_mutex);
1528 }
1529
1530 #define for_each_event(event, start, end)                       \
1531         for (event = start;                                     \
1532              (unsigned long)event < (unsigned long)end;         \
1533              event++)
1534
1535 #ifdef CONFIG_MODULES
1536
1537 static LIST_HEAD(ftrace_module_file_list);
1538
1539 /*
1540  * Modules must own their file_operations to keep up with
1541  * reference counting.
1542  */
1543 struct ftrace_module_file_ops {
1544         struct list_head                list;
1545         struct module                   *mod;
1546         struct file_operations          id;
1547         struct file_operations          enable;
1548         struct file_operations          format;
1549         struct file_operations          filter;
1550 };
1551
1552 static struct ftrace_module_file_ops *
1553 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1554 {
1555         /*
1556          * As event_calls are added in groups by module,
1557          * when we find one file_ops, we don't need to search for
1558          * each call in that module, as the rest should be the
1559          * same. Only search for a new one if the last one did
1560          * not match.
1561          */
1562         if (file_ops && mod == file_ops->mod)
1563                 return file_ops;
1564
1565         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1566                 if (file_ops->mod == mod)
1567                         return file_ops;
1568         }
1569         return NULL;
1570 }
1571
1572 static struct ftrace_module_file_ops *
1573 trace_create_file_ops(struct module *mod)
1574 {
1575         struct ftrace_module_file_ops *file_ops;
1576
1577         /*
1578          * This is a bit of a PITA. To allow for correct reference
1579          * counting, modules must "own" their file_operations.
1580          * To do this, we allocate the file operations that will be
1581          * used in the event directory.
1582          */
1583
1584         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1585         if (!file_ops)
1586                 return NULL;
1587
1588         file_ops->mod = mod;
1589
1590         file_ops->id = ftrace_event_id_fops;
1591         file_ops->id.owner = mod;
1592
1593         file_ops->enable = ftrace_enable_fops;
1594         file_ops->enable.owner = mod;
1595
1596         file_ops->filter = ftrace_event_filter_fops;
1597         file_ops->filter.owner = mod;
1598
1599         file_ops->format = ftrace_event_format_fops;
1600         file_ops->format.owner = mod;
1601
1602         list_add(&file_ops->list, &ftrace_module_file_list);
1603
1604         return file_ops;
1605 }
1606
1607 static void trace_module_add_events(struct module *mod)
1608 {
1609         struct ftrace_module_file_ops *file_ops = NULL;
1610         struct ftrace_event_call **call, **start, **end;
1611
1612         start = mod->trace_events;
1613         end = mod->trace_events + mod->num_trace_events;
1614
1615         if (start == end)
1616                 return;
1617
1618         file_ops = trace_create_file_ops(mod);
1619         if (!file_ops)
1620                 return;
1621
1622         for_each_event(call, start, end) {
1623                 __register_event(*call, mod);
1624                 __add_event_to_tracers(*call, file_ops);
1625         }
1626 }
1627
1628 static void trace_module_remove_events(struct module *mod)
1629 {
1630         struct ftrace_module_file_ops *file_ops;
1631         struct ftrace_event_call *call, *p;
1632         bool clear_trace = false;
1633
1634         down_write(&trace_event_mutex);
1635         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1636                 if (call->mod == mod) {
1637                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1638                                 clear_trace = true;
1639                         __trace_remove_event_call(call);
1640                 }
1641         }
1642
1643         /* Now free the file_operations */
1644         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1645                 if (file_ops->mod == mod)
1646                         break;
1647         }
1648         if (&file_ops->list != &ftrace_module_file_list) {
1649                 list_del(&file_ops->list);
1650                 kfree(file_ops);
1651         }
1652         up_write(&trace_event_mutex);
1653
1654         /*
1655          * It is safest to reset the ring buffer if the module being unloaded
1656          * registered any events that were used. The only worry is if
1657          * a new module gets loaded, and takes on the same id as the events
1658          * of this module. When printing out the buffer, traced events left
1659          * over from this module may be passed to the new module events and
1660          * unexpected results may occur.
1661          */
1662         if (clear_trace)
1663                 tracing_reset_all_online_cpus();
1664 }
1665
1666 static int trace_module_notify(struct notifier_block *self,
1667                                unsigned long val, void *data)
1668 {
1669         struct module *mod = data;
1670
1671         mutex_lock(&event_mutex);
1672         switch (val) {
1673         case MODULE_STATE_COMING:
1674                 trace_module_add_events(mod);
1675                 break;
1676         case MODULE_STATE_GOING:
1677                 trace_module_remove_events(mod);
1678                 break;
1679         }
1680         mutex_unlock(&event_mutex);
1681
1682         return 0;
1683 }
1684
1685 static int
1686 __trace_add_new_mod_event(struct ftrace_event_call *call,
1687                           struct trace_array *tr,
1688                           struct ftrace_module_file_ops *file_ops)
1689 {
1690         return __trace_add_new_event(call, tr,
1691                                      &file_ops->id, &file_ops->enable,
1692                                      &file_ops->filter, &file_ops->format);
1693 }
1694
1695 #else
1696 static inline struct ftrace_module_file_ops *
1697 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1698 {
1699         return NULL;
1700 }
1701 static inline int trace_module_notify(struct notifier_block *self,
1702                                       unsigned long val, void *data)
1703 {
1704         return 0;
1705 }
1706 static inline int
1707 __trace_add_new_mod_event(struct ftrace_event_call *call,
1708                           struct trace_array *tr,
1709                           struct ftrace_module_file_ops *file_ops)
1710 {
1711         return -ENODEV;
1712 }
1713 #endif /* CONFIG_MODULES */
1714
1715 /* Create a new event directory structure for a trace directory. */
1716 static void
1717 __trace_add_event_dirs(struct trace_array *tr)
1718 {
1719         struct ftrace_module_file_ops *file_ops = NULL;
1720         struct ftrace_event_call *call;
1721         int ret;
1722
1723         list_for_each_entry(call, &ftrace_events, list) {
1724                 if (call->mod) {
1725                         /*
1726                          * Directories for events by modules need to
1727                          * keep module ref counts when opened (as we don't
1728                          * want the module to disappear when reading one
1729                          * of these files). The file_ops keep account of
1730                          * the module ref count.
1731                          */
1732                         file_ops = find_ftrace_file_ops(file_ops, call->mod);
1733                         if (!file_ops)
1734                                 continue; /* Warn? */
1735                         ret = __trace_add_new_mod_event(call, tr, file_ops);
1736                         if (ret < 0)
1737                                 pr_warning("Could not create directory for event %s\n",
1738                                            call->name);
1739                         continue;
1740                 }
1741                 ret = __trace_add_new_event(call, tr,
1742                                             &ftrace_event_id_fops,
1743                                             &ftrace_enable_fops,
1744                                             &ftrace_event_filter_fops,
1745                                             &ftrace_event_format_fops);
1746                 if (ret < 0)
1747                         pr_warning("Could not create directory for event %s\n",
1748                                    call->name);
1749         }
1750 }
1751
1752 /*
1753  * The top level array has already had its ftrace_event_file
1754  * descriptors created in order to allow for early events to
1755  * be recorded. This function is called after the debugfs has been
1756  * initialized, and we now have to create the files associated
1757  * to the events.
1758  */
1759 static __init void
1760 __trace_early_add_event_dirs(struct trace_array *tr)
1761 {
1762         struct ftrace_event_file *file;
1763         int ret;
1764
1765
1766         list_for_each_entry(file, &tr->events, list) {
1767                 ret = event_create_dir(tr->event_dir, file,
1768                                        &ftrace_event_id_fops,
1769                                        &ftrace_enable_fops,
1770                                        &ftrace_event_filter_fops,
1771                                        &ftrace_event_format_fops);
1772                 if (ret < 0)
1773                         pr_warning("Could not create directory for event %s\n",
1774                                    file->event_call->name);
1775         }
1776 }
1777
1778 /*
1779  * For early boot up, the top trace array requires to have
1780  * a list of events that can be enabled. This must be done before
1781  * the filesystem is set up in order to allow events to be traced
1782  * early.
1783  */
1784 static __init void
1785 __trace_early_add_events(struct trace_array *tr)
1786 {
1787         struct ftrace_event_call *call;
1788         int ret;
1789
1790         list_for_each_entry(call, &ftrace_events, list) {
1791                 /* Early boot up should not have any modules loaded */
1792                 if (WARN_ON_ONCE(call->mod))
1793                         continue;
1794
1795                 ret = __trace_early_add_new_event(call, tr);
1796                 if (ret < 0)
1797                         pr_warning("Could not create early event %s\n",
1798                                    call->name);
1799         }
1800 }
1801
1802 /* Remove the event directory structure for a trace directory. */
1803 static void
1804 __trace_remove_event_dirs(struct trace_array *tr)
1805 {
1806         struct ftrace_event_file *file, *next;
1807
1808         list_for_each_entry_safe(file, next, &tr->events, list) {
1809                 list_del(&file->list);
1810                 debugfs_remove_recursive(file->dir);
1811                 remove_subsystem(file->system);
1812                 kmem_cache_free(file_cachep, file);
1813         }
1814 }
1815
1816 static void
1817 __add_event_to_tracers(struct ftrace_event_call *call,
1818                        struct ftrace_module_file_ops *file_ops)
1819 {
1820         struct trace_array *tr;
1821
1822         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1823                 if (file_ops)
1824                         __trace_add_new_mod_event(call, tr, file_ops);
1825                 else
1826                         __trace_add_new_event(call, tr,
1827                                               &ftrace_event_id_fops,
1828                                               &ftrace_enable_fops,
1829                                               &ftrace_event_filter_fops,
1830                                               &ftrace_event_format_fops);
1831         }
1832 }
1833
1834 static struct notifier_block trace_module_nb = {
1835         .notifier_call = trace_module_notify,
1836         .priority = 0,
1837 };
1838
1839 extern struct ftrace_event_call *__start_ftrace_events[];
1840 extern struct ftrace_event_call *__stop_ftrace_events[];
1841
1842 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1843
1844 static __init int setup_trace_event(char *str)
1845 {
1846         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1847         ring_buffer_expanded = 1;
1848         tracing_selftest_disabled = 1;
1849
1850         return 1;
1851 }
1852 __setup("trace_event=", setup_trace_event);
1853
1854 /* Expects to have event_mutex held when called */
1855 static int
1856 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
1857 {
1858         struct dentry *d_events;
1859         struct dentry *entry;
1860
1861         entry = debugfs_create_file("set_event", 0644, parent,
1862                                     tr, &ftrace_set_event_fops);
1863         if (!entry) {
1864                 pr_warning("Could not create debugfs 'set_event' entry\n");
1865                 return -ENOMEM;
1866         }
1867
1868         d_events = debugfs_create_dir("events", parent);
1869         if (!d_events) {
1870                 pr_warning("Could not create debugfs 'events' directory\n");
1871                 return -ENOMEM;
1872         }
1873
1874         /* ring buffer internal formats */
1875         trace_create_file("header_page", 0444, d_events,
1876                           ring_buffer_print_page_header,
1877                           &ftrace_show_header_fops);
1878
1879         trace_create_file("header_event", 0444, d_events,
1880                           ring_buffer_print_entry_header,
1881                           &ftrace_show_header_fops);
1882
1883         trace_create_file("enable", 0644, d_events,
1884                           tr, &ftrace_tr_enable_fops);
1885
1886         tr->event_dir = d_events;
1887
1888         return 0;
1889 }
1890
1891 /**
1892  * event_trace_add_tracer - add a instance of a trace_array to events
1893  * @parent: The parent dentry to place the files/directories for events in
1894  * @tr: The trace array associated with these events
1895  *
1896  * When a new instance is created, it needs to set up its events
1897  * directory, as well as other files associated with events. It also
1898  * creates the event hierachry in the @parent/events directory.
1899  *
1900  * Returns 0 on success.
1901  */
1902 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
1903 {
1904         int ret;
1905
1906         mutex_lock(&event_mutex);
1907
1908         ret = create_event_toplevel_files(parent, tr);
1909         if (ret)
1910                 goto out_unlock;
1911
1912         down_write(&trace_event_mutex);
1913         __trace_add_event_dirs(tr);
1914         up_write(&trace_event_mutex);
1915
1916  out_unlock:
1917         mutex_unlock(&event_mutex);
1918
1919         return ret;
1920 }
1921
1922 /*
1923  * The top trace array already had its file descriptors created.
1924  * Now the files themselves need to be created.
1925  */
1926 static __init int
1927 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
1928 {
1929         int ret;
1930
1931         mutex_lock(&event_mutex);
1932
1933         ret = create_event_toplevel_files(parent, tr);
1934         if (ret)
1935                 goto out_unlock;
1936
1937         down_write(&trace_event_mutex);
1938         __trace_early_add_event_dirs(tr);
1939         up_write(&trace_event_mutex);
1940
1941  out_unlock:
1942         mutex_unlock(&event_mutex);
1943
1944         return ret;
1945 }
1946
1947 int event_trace_del_tracer(struct trace_array *tr)
1948 {
1949         /* Disable any running events */
1950         __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
1951
1952         mutex_lock(&event_mutex);
1953
1954         down_write(&trace_event_mutex);
1955         __trace_remove_event_dirs(tr);
1956         debugfs_remove_recursive(tr->event_dir);
1957         up_write(&trace_event_mutex);
1958
1959         tr->event_dir = NULL;
1960
1961         mutex_unlock(&event_mutex);
1962
1963         return 0;
1964 }
1965
1966 static __init int event_trace_memsetup(void)
1967 {
1968         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
1969         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
1970         return 0;
1971 }
1972
1973 static __init int event_trace_enable(void)
1974 {
1975         struct trace_array *tr = top_trace_array();
1976         struct ftrace_event_call **iter, *call;
1977         char *buf = bootup_event_buf;
1978         char *token;
1979         int ret;
1980
1981         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
1982
1983                 call = *iter;
1984                 ret = event_init(call);
1985                 if (!ret)
1986                         list_add(&call->list, &ftrace_events);
1987         }
1988
1989         /*
1990          * We need the top trace array to have a working set of trace
1991          * points at early init, before the debug files and directories
1992          * are created. Create the file entries now, and attach them
1993          * to the actual file dentries later.
1994          */
1995         __trace_early_add_events(tr);
1996
1997         while (true) {
1998                 token = strsep(&buf, ",");
1999
2000                 if (!token)
2001                         break;
2002                 if (!*token)
2003                         continue;
2004
2005                 ret = ftrace_set_clr_event(tr, token, 1);
2006                 if (ret)
2007                         pr_warn("Failed to enable trace event: %s\n", token);
2008         }
2009
2010         trace_printk_start_comm();
2011
2012         return 0;
2013 }
2014
2015 static __init int event_trace_init(void)
2016 {
2017         struct trace_array *tr;
2018         struct dentry *d_tracer;
2019         struct dentry *entry;
2020         int ret;
2021
2022         tr = top_trace_array();
2023
2024         d_tracer = tracing_init_dentry();
2025         if (!d_tracer)
2026                 return 0;
2027
2028         entry = debugfs_create_file("available_events", 0444, d_tracer,
2029                                     tr, &ftrace_avail_fops);
2030         if (!entry)
2031                 pr_warning("Could not create debugfs "
2032                            "'available_events' entry\n");
2033
2034         if (trace_define_common_fields())
2035                 pr_warning("tracing: Failed to allocate common fields");
2036
2037         ret = early_event_add_tracer(d_tracer, tr);
2038         if (ret)
2039                 return ret;
2040
2041         ret = register_module_notifier(&trace_module_nb);
2042         if (ret)
2043                 pr_warning("Failed to register trace events module notifier\n");
2044
2045         return 0;
2046 }
2047 early_initcall(event_trace_memsetup);
2048 core_initcall(event_trace_enable);
2049 fs_initcall(event_trace_init);
2050
2051 #ifdef CONFIG_FTRACE_STARTUP_TEST
2052
2053 static DEFINE_SPINLOCK(test_spinlock);
2054 static DEFINE_SPINLOCK(test_spinlock_irq);
2055 static DEFINE_MUTEX(test_mutex);
2056
2057 static __init void test_work(struct work_struct *dummy)
2058 {
2059         spin_lock(&test_spinlock);
2060         spin_lock_irq(&test_spinlock_irq);
2061         udelay(1);
2062         spin_unlock_irq(&test_spinlock_irq);
2063         spin_unlock(&test_spinlock);
2064
2065         mutex_lock(&test_mutex);
2066         msleep(1);
2067         mutex_unlock(&test_mutex);
2068 }
2069
2070 static __init int event_test_thread(void *unused)
2071 {
2072         void *test_malloc;
2073
2074         test_malloc = kmalloc(1234, GFP_KERNEL);
2075         if (!test_malloc)
2076                 pr_info("failed to kmalloc\n");
2077
2078         schedule_on_each_cpu(test_work);
2079
2080         kfree(test_malloc);
2081
2082         set_current_state(TASK_INTERRUPTIBLE);
2083         while (!kthread_should_stop())
2084                 schedule();
2085
2086         return 0;
2087 }
2088
2089 /*
2090  * Do various things that may trigger events.
2091  */
2092 static __init void event_test_stuff(void)
2093 {
2094         struct task_struct *test_thread;
2095
2096         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2097         msleep(1);
2098         kthread_stop(test_thread);
2099 }
2100
2101 /*
2102  * For every trace event defined, we will test each trace point separately,
2103  * and then by groups, and finally all trace points.
2104  */
2105 static __init void event_trace_self_tests(void)
2106 {
2107         struct ftrace_subsystem_dir *dir;
2108         struct ftrace_event_file *file;
2109         struct ftrace_event_call *call;
2110         struct event_subsystem *system;
2111         struct trace_array *tr;
2112         int ret;
2113
2114         tr = top_trace_array();
2115
2116         pr_info("Running tests on trace events:\n");
2117
2118         list_for_each_entry(file, &tr->events, list) {
2119
2120                 call = file->event_call;
2121
2122                 /* Only test those that have a probe */
2123                 if (!call->class || !call->class->probe)
2124                         continue;
2125
2126 /*
2127  * Testing syscall events here is pretty useless, but
2128  * we still do it if configured. But this is time consuming.
2129  * What we really need is a user thread to perform the
2130  * syscalls as we test.
2131  */
2132 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2133                 if (call->class->system &&
2134                     strcmp(call->class->system, "syscalls") == 0)
2135                         continue;
2136 #endif
2137
2138                 pr_info("Testing event %s: ", call->name);
2139
2140                 /*
2141                  * If an event is already enabled, someone is using
2142                  * it and the self test should not be on.
2143                  */
2144                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2145                         pr_warning("Enabled event during self test!\n");
2146                         WARN_ON_ONCE(1);
2147                         continue;
2148                 }
2149
2150                 ftrace_event_enable_disable(file, 1);
2151                 event_test_stuff();
2152                 ftrace_event_enable_disable(file, 0);
2153
2154                 pr_cont("OK\n");
2155         }
2156
2157         /* Now test at the sub system level */
2158
2159         pr_info("Running tests on trace event systems:\n");
2160
2161         list_for_each_entry(dir, &tr->systems, list) {
2162
2163                 system = dir->subsystem;
2164
2165                 /* the ftrace system is special, skip it */
2166                 if (strcmp(system->name, "ftrace") == 0)
2167                         continue;
2168
2169                 pr_info("Testing event system %s: ", system->name);
2170
2171                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2172                 if (WARN_ON_ONCE(ret)) {
2173                         pr_warning("error enabling system %s\n",
2174                                    system->name);
2175                         continue;
2176                 }
2177
2178                 event_test_stuff();
2179
2180                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2181                 if (WARN_ON_ONCE(ret)) {
2182                         pr_warning("error disabling system %s\n",
2183                                    system->name);
2184                         continue;
2185                 }
2186
2187                 pr_cont("OK\n");
2188         }
2189
2190         /* Test with all events enabled */
2191
2192         pr_info("Running tests on all trace events:\n");
2193         pr_info("Testing all events: ");
2194
2195         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2196         if (WARN_ON_ONCE(ret)) {
2197                 pr_warning("error enabling all events\n");
2198                 return;
2199         }
2200
2201         event_test_stuff();
2202
2203         /* reset sysname */
2204         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2205         if (WARN_ON_ONCE(ret)) {
2206                 pr_warning("error disabling all events\n");
2207                 return;
2208         }
2209
2210         pr_cont("OK\n");
2211 }
2212
2213 #ifdef CONFIG_FUNCTION_TRACER
2214
2215 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2216
2217 static void
2218 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2219                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2220 {
2221         struct ring_buffer_event *event;
2222         struct ring_buffer *buffer;
2223         struct ftrace_entry *entry;
2224         unsigned long flags;
2225         long disabled;
2226         int cpu;
2227         int pc;
2228
2229         pc = preempt_count();
2230         preempt_disable_notrace();
2231         cpu = raw_smp_processor_id();
2232         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2233
2234         if (disabled != 1)
2235                 goto out;
2236
2237         local_save_flags(flags);
2238
2239         event = trace_current_buffer_lock_reserve(&buffer,
2240                                                   TRACE_FN, sizeof(*entry),
2241                                                   flags, pc);
2242         if (!event)
2243                 goto out;
2244         entry   = ring_buffer_event_data(event);
2245         entry->ip                       = ip;
2246         entry->parent_ip                = parent_ip;
2247
2248         trace_buffer_unlock_commit(buffer, event, flags, pc);
2249
2250  out:
2251         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2252         preempt_enable_notrace();
2253 }
2254
2255 static struct ftrace_ops trace_ops __initdata  =
2256 {
2257         .func = function_test_events_call,
2258         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2259 };
2260
2261 static __init void event_trace_self_test_with_function(void)
2262 {
2263         int ret;
2264         ret = register_ftrace_function(&trace_ops);
2265         if (WARN_ON(ret < 0)) {
2266                 pr_info("Failed to enable function tracer for event tests\n");
2267                 return;
2268         }
2269         pr_info("Running tests again, along with the function tracer\n");
2270         event_trace_self_tests();
2271         unregister_ftrace_function(&trace_ops);
2272 }
2273 #else
2274 static __init void event_trace_self_test_with_function(void)
2275 {
2276 }
2277 #endif
2278
2279 static __init int event_trace_self_tests_init(void)
2280 {
2281         if (!tracing_selftest_disabled) {
2282                 event_trace_self_tests();
2283                 event_trace_self_test_with_function();
2284         }
2285
2286         return 0;
2287 }
2288
2289 late_initcall(event_trace_self_tests_init);
2290
2291 #endif