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/debugfs.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/module.h>
  14 #include <linux/ctype.h>
  15
  16 #include "trace.h"
  17
  18 #define TRACE_SYSTEM "TRACE_SYSTEM"
  19
  20 static DEFINE_MUTEX(event_mutex);
  21
  22 #define events_for_each(event)                                          \
  23         for (event = __start_ftrace_events;                             \
  24              (unsigned long)event < (unsigned long)__stop_ftrace_events; \
  25              event++)
  26
  27 void event_trace_printk(unsigned long ip, const char *fmt, ...)
  28 {
  29         va_list ap;
  30
  31         va_start(ap, fmt);
  32         tracing_record_cmdline(current);
  33         trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
  34         va_end(ap);
  35 }
  36
  37 static void ftrace_clear_events(void)
  38 {
  39         struct ftrace_event_call *call = (void *)__start_ftrace_events;
  40
  41
  42         while ((unsigned long)call < (unsigned long)__stop_ftrace_events) {
  43
  44                 if (call->enabled) {
  45                         call->enabled = 0;
  46                         call->unregfunc();
  47                 }
  48                 call++;
  49         }
  50 }
  51
  52 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
  53                                         int enable)
  54 {
  55
  56         switch (enable) {
  57         case 0:
  58                 if (call->enabled) {
  59                         call->enabled = 0;
  60                         call->unregfunc();
  61                 }
  62                 if (call->raw_enabled) {
  63                         call->raw_enabled = 0;
  64                         call->raw_unreg();
  65                 }
  66                 break;
  67         case 1:
  68                 if (!call->enabled &&
  69                     (call->type & TRACE_EVENT_TYPE_PRINTF)) {
  70                         call->enabled = 1;
  71                         call->regfunc();
  72                 }
  73                 if (!call->raw_enabled &&
  74                     (call->type & TRACE_EVENT_TYPE_RAW)) {
  75                         call->raw_enabled = 1;
  76                         call->raw_reg();
  77                 }
  78                 break;
  79         }
  80 }
  81
  82 static int ftrace_set_clr_event(char *buf, int set)
  83 {
  84         struct ftrace_event_call *call = __start_ftrace_events;
  85         char *event = NULL, *sub = NULL, *match;
  86         int ret = -EINVAL;
  87
  88         /*
  89          * The buf format can be <subsystem>:<event-name>
  90          *  *:<event-name> means any event by that name.
  91          *  :<event-name> is the same.
  92          *
  93          *  <subsystem>:* means all events in that subsystem
  94          *  <subsystem>: means the same.
  95          *
  96          *  <name> (no ':') means all events in a subsystem with
  97          *  the name <name> or any event that matches <name>
  98          */
  99
 100         match = strsep(&buf, ":");
 101         if (buf) {
 102                 sub = match;
 103                 event = buf;
 104                 match = NULL;
 105
 106                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 107                         sub = NULL;
 108                 if (!strlen(event) || strcmp(event, "*") == 0)
 109                         event = NULL;
 110         }
 111
 112         mutex_lock(&event_mutex);
 113         events_for_each(call) {
 114
 115                 if (!call->name)
 116                         continue;
 117
 118                 if (match &&
 119                     strcmp(match, call->name) != 0 &&
 120                     strcmp(match, call->system) != 0)
 121                         continue;
 122
 123                 if (sub && strcmp(sub, call->system) != 0)
 124                         continue;
 125
 126                 if (event && strcmp(event, call->name) != 0)
 127                         continue;
 128
 129                 ftrace_event_enable_disable(call, set);
 130
 131                 ret = 0;
 132         }
 133         mutex_unlock(&event_mutex);
 134
 135         return ret;
 136 }
 137
 138 /* 128 should be much more than enough */
 139 #define EVENT_BUF_SIZE          127
 140
 141 static ssize_t
 142 ftrace_event_write(struct file *file, const char __user *ubuf,
 143                    size_t cnt, loff_t *ppos)
 144 {
 145         size_t read = 0;
 146         int i, set = 1;
 147         ssize_t ret;
 148         char *buf;
 149         char ch;
 150
 151         if (!cnt || cnt < 0)
 152                 return 0;
 153
 154         ret = get_user(ch, ubuf++);
 155         if (ret)
 156                 return ret;
 157         read++;
 158         cnt--;
 159
 160         /* skip white space */
 161         while (cnt && isspace(ch)) {
 162                 ret = get_user(ch, ubuf++);
 163                 if (ret)
 164                         return ret;
 165                 read++;
 166                 cnt--;
 167         }
 168
 169         /* Only white space found? */
 170         if (isspace(ch)) {
 171                 file->f_pos += read;
 172                 ret = read;
 173                 return ret;
 174         }
 175
 176         buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
 177         if (!buf)
 178                 return -ENOMEM;
 179
 180         if (cnt > EVENT_BUF_SIZE)
 181                 cnt = EVENT_BUF_SIZE;
 182
 183         i = 0;
 184         while (cnt && !isspace(ch)) {
 185                 if (!i && ch == '!')
 186                         set = 0;
 187                 else
 188                         buf[i++] = ch;
 189
 190                 ret = get_user(ch, ubuf++);
 191                 if (ret)
 192                         goto out_free;
 193                 read++;
 194                 cnt--;
 195         }
 196         buf[i] = 0;
 197
 198         file->f_pos += read;
 199
 200         ret = ftrace_set_clr_event(buf, set);
 201         if (ret)
 202                 goto out_free;
 203
 204         ret = read;
 205
 206  out_free:
 207         kfree(buf);
 208
 209         return ret;
 210 }
 211
 212 static void *
 213 t_next(struct seq_file *m, void *v, loff_t *pos)
 214 {
 215         struct ftrace_event_call *call = m->private;
 216         struct ftrace_event_call *next = call;
 217
 218         (*pos)++;
 219
 220         if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
 221                 return NULL;
 222
 223         m->private = ++next;
 224
 225         return call;
 226 }
 227
 228 static void *t_start(struct seq_file *m, loff_t *pos)
 229 {
 230         return t_next(m, NULL, pos);
 231 }
 232
 233 static void *
 234 s_next(struct seq_file *m, void *v, loff_t *pos)
 235 {
 236         struct ftrace_event_call *call = m->private;
 237         struct ftrace_event_call *next;
 238
 239         (*pos)++;
 240
 241  retry:
 242         if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
 243                 return NULL;
 244
 245         if (!call->enabled) {
 246                 call++;
 247                 goto retry;
 248         }
 249
 250         next = call;
 251         m->private = ++next;
 252
 253         return call;
 254 }
 255
 256 static void *s_start(struct seq_file *m, loff_t *pos)
 257 {
 258         return s_next(m, NULL, pos);
 259 }
 260
 261 static int t_show(struct seq_file *m, void *v)
 262 {
 263         struct ftrace_event_call *call = v;
 264
 265         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 266                 seq_printf(m, "%s:", call->system);
 267         seq_printf(m, "%s\n", call->name);
 268
 269         return 0;
 270 }
 271
 272 static void t_stop(struct seq_file *m, void *p)
 273 {
 274 }
 275
 276 static int
 277 ftrace_event_seq_open(struct inode *inode, struct file *file)
 278 {
 279         int ret;
 280         const struct seq_operations *seq_ops;
 281
 282         if ((file->f_mode & FMODE_WRITE) &&
 283             !(file->f_flags & O_APPEND))
 284                 ftrace_clear_events();
 285
 286         seq_ops = inode->i_private;
 287         ret = seq_open(file, seq_ops);
 288         if (!ret) {
 289                 struct seq_file *m = file->private_data;
 290
 291                 m->private = __start_ftrace_events;
 292         }
 293         return ret;
 294 }
 295
 296 static ssize_t
 297 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 298                   loff_t *ppos)
 299 {
 300         struct ftrace_event_call *call = filp->private_data;
 301         char *buf;
 302
 303         if (call->enabled || call->raw_enabled)
 304                 buf = "1\n";
 305         else
 306                 buf = "0\n";
 307
 308         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 309 }
 310
 311 static ssize_t
 312 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 313                    loff_t *ppos)
 314 {
 315         struct ftrace_event_call *call = filp->private_data;
 316         char buf[64];
 317         unsigned long val;
 318         int ret;
 319
 320         if (cnt >= sizeof(buf))
 321                 return -EINVAL;
 322
 323         if (copy_from_user(&buf, ubuf, cnt))
 324                 return -EFAULT;
 325
 326         buf[cnt] = 0;
 327
 328         ret = strict_strtoul(buf, 10, &val);
 329         if (ret < 0)
 330                 return ret;
 331
 332         switch (val) {
 333         case 0:
 334         case 1:
 335                 mutex_lock(&event_mutex);
 336                 ftrace_event_enable_disable(call, val);
 337                 mutex_unlock(&event_mutex);
 338                 break;
 339
 340         default:
 341                 return -EINVAL;
 342         }
 343
 344         *ppos += cnt;
 345
 346         return cnt;
 347 }
 348
 349 static ssize_t
 350 event_type_read(struct file *filp, char __user *ubuf, size_t cnt,
 351                 loff_t *ppos)
 352 {
 353         struct ftrace_event_call *call = filp->private_data;
 354         char buf[16];
 355         int r = 0;
 356
 357         if (call->type & TRACE_EVENT_TYPE_PRINTF)
 358                 r += sprintf(buf, "printf\n");
 359
 360         if (call->type & TRACE_EVENT_TYPE_RAW)
 361                 r += sprintf(buf+r, "raw\n");
 362
 363         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 364 }
 365
 366 static ssize_t
 367 event_type_write(struct file *filp, const char __user *ubuf, size_t cnt,
 368                  loff_t *ppos)
 369 {
 370         struct ftrace_event_call *call = filp->private_data;
 371         char buf[64];
 372
 373         /*
 374          * If there's only one type, we can't change it.
 375          * And currently we always have printf type, and we
 376          * may or may not have raw type.
 377          *
 378          * This is a redundant check, the file should be read
 379          * only if this is the case anyway.
 380          */
 381
 382         if (!call->raw_init)
 383                 return -EPERM;
 384
 385         if (cnt >= sizeof(buf))
 386                 return -EINVAL;
 387
 388         if (copy_from_user(&buf, ubuf, cnt))
 389                 return -EFAULT;
 390
 391         buf[cnt] = 0;
 392
 393         if (!strncmp(buf, "printf", 6) &&
 394             (!buf[6] || isspace(buf[6]))) {
 395
 396                 call->type = TRACE_EVENT_TYPE_PRINTF;
 397
 398                 /*
 399                  * If raw enabled, the disable it and enable
 400                  * printf type.
 401                  */
 402                 if (call->raw_enabled) {
 403                         call->raw_enabled = 0;
 404                         call->raw_unreg();
 405
 406                         call->enabled = 1;
 407                         call->regfunc();
 408                 }
 409
 410         } else if (!strncmp(buf, "raw", 3) &&
 411             (!buf[3] || isspace(buf[3]))) {
 412
 413                 call->type = TRACE_EVENT_TYPE_RAW;
 414
 415                 /*
 416                  * If printf enabled, the disable it and enable
 417                  * raw type.
 418                  */
 419                 if (call->enabled) {
 420                         call->enabled = 0;
 421                         call->unregfunc();
 422
 423                         call->raw_enabled = 1;
 424                         call->raw_reg();
 425                 }
 426         } else
 427                 return -EINVAL;
 428
 429         *ppos += cnt;
 430
 431         return cnt;
 432 }
 433
 434 static ssize_t
 435 event_available_types_read(struct file *filp, char __user *ubuf, size_t cnt,
 436                            loff_t *ppos)
 437 {
 438         struct ftrace_event_call *call = filp->private_data;
 439         char buf[16];
 440         int r = 0;
 441
 442         r += sprintf(buf, "printf\n");
 443
 444         if (call->raw_init)
 445                 r += sprintf(buf+r, "raw\n");
 446
 447         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 448 }
 449
 450 static ssize_t
 451 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 452                   loff_t *ppos)
 453 {
 454         struct ftrace_event_call *call = filp->private_data;
 455         struct trace_seq *s;
 456         char *buf;
 457         int r;
 458
 459         s = kmalloc(sizeof(*s), GFP_KERNEL);
 460         if (!s)
 461                 return -ENOMEM;
 462
 463         trace_seq_init(s);
 464
 465         if (*ppos)
 466                 return 0;
 467
 468         r = call->show_format(s);
 469         if (!r) {
 470                 /*
 471                  * ug!  The format output is bigger than a PAGE!!
 472                  */
 473                 buf = "FORMAT TOO BIG\n";
 474                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 475                                               buf, strlen(buf));
 476                 goto out;
 477         }
 478
 479         r = simple_read_from_buffer(ubuf, cnt, ppos,
 480                                     s->buffer, s->len);
 481  out:
 482         kfree(s);
 483         return r;
 484 }
 485
 486 static const struct seq_operations show_event_seq_ops = {
 487         .start = t_start,
 488         .next = t_next,
 489         .show = t_show,
 490         .stop = t_stop,
 491 };
 492
 493 static const struct seq_operations show_set_event_seq_ops = {
 494         .start = s_start,
 495         .next = s_next,
 496         .show = t_show,
 497         .stop = t_stop,
 498 };
 499
 500 static const struct file_operations ftrace_avail_fops = {
 501         .open = ftrace_event_seq_open,
 502         .read = seq_read,
 503         .llseek = seq_lseek,
 504         .release = seq_release,
 505 };
 506
 507 static const struct file_operations ftrace_set_event_fops = {
 508         .open = ftrace_event_seq_open,
 509         .read = seq_read,
 510         .write = ftrace_event_write,
 511         .llseek = seq_lseek,
 512         .release = seq_release,
 513 };
 514
 515 static const struct file_operations ftrace_enable_fops = {
 516         .open = tracing_open_generic,
 517         .read = event_enable_read,
 518         .write = event_enable_write,
 519 };
 520
 521 static const struct file_operations ftrace_type_fops = {
 522         .open = tracing_open_generic,
 523         .read = event_type_read,
 524         .write = event_type_write,
 525 };
 526
 527 static const struct file_operations ftrace_available_types_fops = {
 528         .open = tracing_open_generic,
 529         .read = event_available_types_read,
 530 };
 531
 532 static const struct file_operations ftrace_event_format_fops = {
 533         .open = tracing_open_generic,
 534         .read = event_format_read,
 535 };
 536
 537 static struct dentry *event_trace_events_dir(void)
 538 {
 539         static struct dentry *d_tracer;
 540         static struct dentry *d_events;
 541
 542         if (d_events)
 543                 return d_events;
 544
 545         d_tracer = tracing_init_dentry();
 546         if (!d_tracer)
 547                 return NULL;
 548
 549         d_events = debugfs_create_dir("events", d_tracer);
 550         if (!d_events)
 551                 pr_warning("Could not create debugfs "
 552                            "'events' directory\n");
 553
 554         return d_events;
 555 }
 556
 557 struct event_subsystem {
 558         struct list_head        list;
 559         const char              *name;
 560         struct dentry           *entry;
 561 };
 562
 563 static LIST_HEAD(event_subsystems);
 564
 565 static struct dentry *
 566 event_subsystem_dir(const char *name, struct dentry *d_events)
 567 {
 568         struct event_subsystem *system;
 569
 570         /* First see if we did not already create this dir */
 571         list_for_each_entry(system, &event_subsystems, list) {
 572                 if (strcmp(system->name, name) == 0)
 573                         return system->entry;
 574         }
 575
 576         /* need to create new entry */
 577         system = kmalloc(sizeof(*system), GFP_KERNEL);
 578         if (!system) {
 579                 pr_warning("No memory to create event subsystem %s\n",
 580                            name);
 581                 return d_events;
 582         }
 583
 584         system->entry = debugfs_create_dir(name, d_events);
 585         if (!system->entry) {
 586                 pr_warning("Could not create event subsystem %s\n",
 587                            name);
 588                 kfree(system);
 589                 return d_events;
 590         }
 591
 592         system->name = name;
 593         list_add(&system->list, &event_subsystems);
 594
 595         return system->entry;
 596 }
 597
 598 static int
 599 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
 600 {
 601         struct dentry *entry;
 602         int ret;
 603
 604         /*
 605          * If the trace point header did not define TRACE_SYSTEM
 606          * then the system would be called "TRACE_SYSTEM".
 607          */
 608         if (strcmp(call->system, "TRACE_SYSTEM") != 0)
 609                 d_events = event_subsystem_dir(call->system, d_events);
 610
 611         if (call->raw_init) {
 612                 ret = call->raw_init();
 613                 if (ret < 0) {
 614                         pr_warning("Could not initialize trace point"
 615                                    " events/%s\n", call->name);
 616                         return ret;
 617                 }
 618         }
 619
 620         /* default the output to printf */
 621         call->type = TRACE_EVENT_TYPE_PRINTF;
 622
 623         call->dir = debugfs_create_dir(call->name, d_events);
 624         if (!call->dir) {
 625                 pr_warning("Could not create debugfs "
 626                            "'%s' directory\n", call->name);
 627                 return -1;
 628         }
 629
 630         entry = debugfs_create_file("enable", 0644, call->dir, call,
 631                                     &ftrace_enable_fops);
 632         if (!entry)
 633                 pr_warning("Could not create debugfs "
 634                            "'%s/enable' entry\n", call->name);
 635
 636         /* Only let type be writable, if we can change it */
 637         entry = debugfs_create_file("type",
 638                                     call->raw_init ? 0644 : 0444,
 639                                     call->dir, call,
 640                                     &ftrace_type_fops);
 641         if (!entry)
 642                 pr_warning("Could not create debugfs "
 643                            "'%s/type' entry\n", call->name);
 644
 645         entry = debugfs_create_file("available_types", 0444, call->dir, call,
 646                                     &ftrace_available_types_fops);
 647         if (!entry)
 648                 pr_warning("Could not create debugfs "
 649                            "'%s/available_types' entry\n", call->name);
 650
 651         /* A trace may not want to export its format */
 652         if (!call->show_format)
 653                 return 0;
 654
 655         entry = debugfs_create_file("format", 0444, call->dir, call,
 656                                     &ftrace_event_format_fops);
 657         if (!entry)
 658                 pr_warning("Could not create debugfs "
 659                            "'%s/format' entry\n", call->name);
 660
 661         return 0;
 662 }
 663
 664 static __init int event_trace_init(void)
 665 {
 666         struct ftrace_event_call *call = __start_ftrace_events;
 667         struct dentry *d_tracer;
 668         struct dentry *entry;
 669         struct dentry *d_events;
 670
 671         d_tracer = tracing_init_dentry();
 672         if (!d_tracer)
 673                 return 0;
 674
 675         entry = debugfs_create_file("available_events", 0444, d_tracer,
 676                                     (void *)&show_event_seq_ops,
 677                                     &ftrace_avail_fops);
 678         if (!entry)
 679                 pr_warning("Could not create debugfs "
 680                            "'available_events' entry\n");
 681
 682         entry = debugfs_create_file("set_event", 0644, d_tracer,
 683                                     (void *)&show_set_event_seq_ops,
 684                                     &ftrace_set_event_fops);
 685         if (!entry)
 686                 pr_warning("Could not create debugfs "
 687                            "'set_event' entry\n");
 688
 689         d_events = event_trace_events_dir();
 690         if (!d_events)
 691                 return 0;
 692
 693         events_for_each(call) {
 694                 /* The linker may leave blanks */
 695                 if (!call->name)
 696                         continue;
 697                 event_create_dir(call, d_events);
 698         }
 699
 700         return 0;
 701 }
 702 fs_initcall(event_trace_init);