block/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/time.h>
  26 #include <trace/block.h>
  27 #include <asm/uaccess.h>
  28
  29 static unsigned int blktrace_seq __read_mostly = 1;
  30
  31 /* Global reference count of probes */
  32 static DEFINE_MUTEX(blk_probe_mutex);
  33 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  34
  35 static int blk_register_tracepoints(void);
  36 static void blk_unregister_tracepoints(void);
  37
  38 /*
  39  * Send out a notify message.
  40  */
  41 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  42                        const void *data, size_t len)
  43 {
  44         struct blk_io_trace *t;
  45
  46         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  47         if (t) {
  48                 const int cpu = smp_processor_id();
  49
  50                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  51                 t->time = ktime_to_ns(ktime_get());
  52                 t->device = bt->dev;
  53                 t->action = action;
  54                 t->pid = pid;
  55                 t->cpu = cpu;
  56                 t->pdu_len = len;
  57                 memcpy((void *) t + sizeof(*t), data, len);
  58         }
  59 }
  60
  61 /*
  62  * Send out a notify for this process, if we haven't done so since a trace
  63  * started
  64  */
  65 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
  66 {
  67         tsk->btrace_seq = blktrace_seq;
  68         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
  69 }
  70
  71 static void trace_note_time(struct blk_trace *bt)
  72 {
  73         struct timespec now;
  74         unsigned long flags;
  75         u32 words[2];
  76
  77         getnstimeofday(&now);
  78         words[0] = now.tv_sec;
  79         words[1] = now.tv_nsec;
  80
  81         local_irq_save(flags);
  82         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
  83         local_irq_restore(flags);
  84 }
  85
  86 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
  87 {
  88         int n;
  89         va_list args;
  90         unsigned long flags;
  91         char *buf;
  92
  93         local_irq_save(flags);
  94         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
  95         va_start(args, fmt);
  96         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
  97         va_end(args);
  98
  99         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 100         local_irq_restore(flags);
 101 }
 102 EXPORT_SYMBOL_GPL(__trace_note_message);
 103
 104 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 105                          pid_t pid)
 106 {
 107         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 108                 return 1;
 109         if (sector < bt->start_lba || sector > bt->end_lba)
 110                 return 1;
 111         if (bt->pid && pid != bt->pid)
 112                 return 1;
 113
 114         return 0;
 115 }
 116
 117 /*
 118  * Data direction bit lookup
 119  */
 120 static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), BLK_TC_ACT(BLK_TC_WRITE) };
 121
 122 /* The ilog2() calls fall out because they're constant */
 123 #define MASK_TC_BIT(rw, __name) ( (rw & (1 << BIO_RW_ ## __name)) << \
 124           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name) )
 125
 126 /*
 127  * The worker for the various blk_add_trace*() types. Fills out a
 128  * blk_io_trace structure and places it in a per-cpu subbuffer.
 129  */
 130 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 131                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 132 {
 133         struct task_struct *tsk = current;
 134         struct blk_io_trace *t;
 135         unsigned long flags;
 136         unsigned long *sequence;
 137         pid_t pid;
 138         int cpu;
 139
 140         if (unlikely(bt->trace_state != Blktrace_running))
 141                 return;
 142
 143         what |= ddir_act[rw & WRITE];
 144         what |= MASK_TC_BIT(rw, BARRIER);
 145         what |= MASK_TC_BIT(rw, SYNC);
 146         what |= MASK_TC_BIT(rw, AHEAD);
 147         what |= MASK_TC_BIT(rw, META);
 148         what |= MASK_TC_BIT(rw, DISCARD);
 149
 150         pid = tsk->pid;
 151         if (unlikely(act_log_check(bt, what, sector, pid)))
 152                 return;
 153
 154         /*
 155          * A word about the locking here - we disable interrupts to reserve
 156          * some space in the relay per-cpu buffer, to prevent an irq
 157          * from coming in and stepping on our toes.
 158          */
 159         local_irq_save(flags);
 160
 161         if (unlikely(tsk->btrace_seq != blktrace_seq))
 162                 trace_note_tsk(bt, tsk);
 163
 164         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 165         if (t) {
 166                 cpu = smp_processor_id();
 167                 sequence = per_cpu_ptr(bt->sequence, cpu);
 168
 169                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 170                 t->sequence = ++(*sequence);
 171                 t->time = ktime_to_ns(ktime_get());
 172                 t->sector = sector;
 173                 t->bytes = bytes;
 174                 t->action = what;
 175                 t->pid = pid;
 176                 t->device = bt->dev;
 177                 t->cpu = cpu;
 178                 t->error = error;
 179                 t->pdu_len = pdu_len;
 180
 181                 if (pdu_len)
 182                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 183         }
 184
 185         local_irq_restore(flags);
 186 }
 187
 188 static struct dentry *blk_tree_root;
 189 static DEFINE_MUTEX(blk_tree_mutex);
 190 static unsigned int root_users;
 191
 192 static inline void blk_remove_root(void)
 193 {
 194         if (blk_tree_root) {
 195                 debugfs_remove(blk_tree_root);
 196                 blk_tree_root = NULL;
 197         }
 198 }
 199
 200 static void blk_remove_tree(struct dentry *dir)
 201 {
 202         mutex_lock(&blk_tree_mutex);
 203         debugfs_remove(dir);
 204         if (--root_users == 0)
 205                 blk_remove_root();
 206         mutex_unlock(&blk_tree_mutex);
 207 }
 208
 209 static struct dentry *blk_create_tree(const char *blk_name)
 210 {
 211         struct dentry *dir = NULL;
 212         int created = 0;
 213
 214         mutex_lock(&blk_tree_mutex);
 215
 216         if (!blk_tree_root) {
 217                 blk_tree_root = debugfs_create_dir("block", NULL);
 218                 if (!blk_tree_root)
 219                         goto err;
 220                 created = 1;
 221         }
 222
 223         dir = debugfs_create_dir(blk_name, blk_tree_root);
 224         if (dir)
 225                 root_users++;
 226         else {
 227                 /* Delete root only if we created it */
 228                 if (created)
 229                         blk_remove_root();
 230         }
 231
 232 err:
 233         mutex_unlock(&blk_tree_mutex);
 234         return dir;
 235 }
 236
 237 static void blk_trace_cleanup(struct blk_trace *bt)
 238 {
 239         relay_close(bt->rchan);
 240         debugfs_remove(bt->msg_file);
 241         debugfs_remove(bt->dropped_file);
 242         blk_remove_tree(bt->dir);
 243         free_percpu(bt->sequence);
 244         free_percpu(bt->msg_data);
 245         kfree(bt);
 246         mutex_lock(&blk_probe_mutex);
 247         if (atomic_dec_and_test(&blk_probes_ref))
 248                 blk_unregister_tracepoints();
 249         mutex_unlock(&blk_probe_mutex);
 250 }
 251
 252 int blk_trace_remove(struct request_queue *q)
 253 {
 254         struct blk_trace *bt;
 255
 256         bt = xchg(&q->blk_trace, NULL);
 257         if (!bt)
 258                 return -EINVAL;
 259
 260         if (bt->trace_state == Blktrace_setup ||
 261             bt->trace_state == Blktrace_stopped)
 262                 blk_trace_cleanup(bt);
 263
 264         return 0;
 265 }
 266 EXPORT_SYMBOL_GPL(blk_trace_remove);
 267
 268 static int blk_dropped_open(struct inode *inode, struct file *filp)
 269 {
 270         filp->private_data = inode->i_private;
 271
 272         return 0;
 273 }
 274
 275 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 276                                 size_t count, loff_t *ppos)
 277 {
 278         struct blk_trace *bt = filp->private_data;
 279         char buf[16];
 280
 281         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 282
 283         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 284 }
 285
 286 static const struct file_operations blk_dropped_fops = {
 287         .owner =        THIS_MODULE,
 288         .open =         blk_dropped_open,
 289         .read =         blk_dropped_read,
 290 };
 291
 292 static int blk_msg_open(struct inode *inode, struct file *filp)
 293 {
 294         filp->private_data = inode->i_private;
 295
 296         return 0;
 297 }
 298
 299 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 300                                 size_t count, loff_t *ppos)
 301 {
 302         char *msg;
 303         struct blk_trace *bt;
 304
 305         if (count > BLK_TN_MAX_MSG)
 306                 return -EINVAL;
 307
 308         msg = kmalloc(count, GFP_KERNEL);
 309         if (msg == NULL)
 310                 return -ENOMEM;
 311
 312         if (copy_from_user(msg, buffer, count)) {
 313                 kfree(msg);
 314                 return -EFAULT;
 315         }
 316
 317         bt = filp->private_data;
 318         __trace_note_message(bt, "%s", msg);
 319         kfree(msg);
 320
 321         return count;
 322 }
 323
 324 static const struct file_operations blk_msg_fops = {
 325         .owner =        THIS_MODULE,
 326         .open =         blk_msg_open,
 327         .write =        blk_msg_write,
 328 };
 329
 330 /*
 331  * Keep track of how many times we encountered a full subbuffer, to aid
 332  * the user space app in telling how many lost events there were.
 333  */
 334 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 335                                      void *prev_subbuf, size_t prev_padding)
 336 {
 337         struct blk_trace *bt;
 338
 339         if (!relay_buf_full(buf))
 340                 return 1;
 341
 342         bt = buf->chan->private_data;
 343         atomic_inc(&bt->dropped);
 344         return 0;
 345 }
 346
 347 static int blk_remove_buf_file_callback(struct dentry *dentry)
 348 {
 349         debugfs_remove(dentry);
 350         return 0;
 351 }
 352
 353 static struct dentry *blk_create_buf_file_callback(const char *filename,
 354                                                    struct dentry *parent,
 355                                                    int mode,
 356                                                    struct rchan_buf *buf,
 357                                                    int *is_global)
 358 {
 359         return debugfs_create_file(filename, mode, parent, buf,
 360                                         &relay_file_operations);
 361 }
 362
 363 static struct rchan_callbacks blk_relay_callbacks = {
 364         .subbuf_start           = blk_subbuf_start_callback,
 365         .create_buf_file        = blk_create_buf_file_callback,
 366         .remove_buf_file        = blk_remove_buf_file_callback,
 367 };
 368
 369 /*
 370  * Setup everything required to start tracing
 371  */
 372 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 373                         struct blk_user_trace_setup *buts)
 374 {
 375         struct blk_trace *old_bt, *bt = NULL;
 376         struct dentry *dir = NULL;
 377         int ret, i;
 378
 379         if (!buts->buf_size || !buts->buf_nr)
 380                 return -EINVAL;
 381
 382         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 383         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 384
 385         /*
 386          * some device names have larger paths - convert the slashes
 387          * to underscores for this to work as expected
 388          */
 389         for (i = 0; i < strlen(buts->name); i++)
 390                 if (buts->name[i] == '/')
 391                         buts->name[i] = '_';
 392
 393         ret = -ENOMEM;
 394         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 395         if (!bt)
 396                 goto err;
 397
 398         bt->sequence = alloc_percpu(unsigned long);
 399         if (!bt->sequence)
 400                 goto err;
 401
 402         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 403         if (!bt->msg_data)
 404                 goto err;
 405
 406         ret = -ENOENT;
 407         dir = blk_create_tree(buts->name);
 408         if (!dir)
 409                 goto err;
 410
 411         bt->dir = dir;
 412         bt->dev = dev;
 413         atomic_set(&bt->dropped, 0);
 414
 415         ret = -EIO;
 416         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 417         if (!bt->dropped_file)
 418                 goto err;
 419
 420         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 421         if (!bt->msg_file)
 422                 goto err;
 423
 424         bt->rchan = relay_open("trace", dir, buts->buf_size,
 425                                 buts->buf_nr, &blk_relay_callbacks, bt);
 426         if (!bt->rchan)
 427                 goto err;
 428
 429         bt->act_mask = buts->act_mask;
 430         if (!bt->act_mask)
 431                 bt->act_mask = (u16) -1;
 432
 433         bt->start_lba = buts->start_lba;
 434         bt->end_lba = buts->end_lba;
 435         if (!bt->end_lba)
 436                 bt->end_lba = -1ULL;
 437
 438         bt->pid = buts->pid;
 439         bt->trace_state = Blktrace_setup;
 440
 441         mutex_lock(&blk_probe_mutex);
 442         if (atomic_add_return(1, &blk_probes_ref) == 1) {
 443                 ret = blk_register_tracepoints();
 444                 if (ret)
 445                         goto probe_err;
 446         }
 447         mutex_unlock(&blk_probe_mutex);
 448
 449         ret = -EBUSY;
 450         old_bt = xchg(&q->blk_trace, bt);
 451         if (old_bt) {
 452                 (void) xchg(&q->blk_trace, old_bt);
 453                 goto err;
 454         }
 455
 456         return 0;
 457 probe_err:
 458         atomic_dec(&blk_probes_ref);
 459         mutex_unlock(&blk_probe_mutex);
 460 err:
 461         if (dir)
 462                 blk_remove_tree(dir);
 463         if (bt) {
 464                 if (bt->msg_file)
 465                         debugfs_remove(bt->msg_file);
 466                 if (bt->dropped_file)
 467                         debugfs_remove(bt->dropped_file);
 468                 free_percpu(bt->sequence);
 469                 free_percpu(bt->msg_data);
 470                 if (bt->rchan)
 471                         relay_close(bt->rchan);
 472                 kfree(bt);
 473         }
 474         return ret;
 475 }
 476
 477 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 478                     char __user *arg)
 479 {
 480         struct blk_user_trace_setup buts;
 481         int ret;
 482
 483         ret = copy_from_user(&buts, arg, sizeof(buts));
 484         if (ret)
 485                 return -EFAULT;
 486
 487         ret = do_blk_trace_setup(q, name, dev, &buts);
 488         if (ret)
 489                 return ret;
 490
 491         if (copy_to_user(arg, &buts, sizeof(buts)))
 492                 return -EFAULT;
 493
 494         return 0;
 495 }
 496 EXPORT_SYMBOL_GPL(blk_trace_setup);
 497
 498 int blk_trace_startstop(struct request_queue *q, int start)
 499 {
 500         struct blk_trace *bt;
 501         int ret;
 502
 503         if ((bt = q->blk_trace) == NULL)
 504                 return -EINVAL;
 505
 506         /*
 507          * For starting a trace, we can transition from a setup or stopped
 508          * trace. For stopping a trace, the state must be running
 509          */
 510         ret = -EINVAL;
 511         if (start) {
 512                 if (bt->trace_state == Blktrace_setup ||
 513                     bt->trace_state == Blktrace_stopped) {
 514                         blktrace_seq++;
 515                         smp_mb();
 516                         bt->trace_state = Blktrace_running;
 517
 518                         trace_note_time(bt);
 519                         ret = 0;
 520                 }
 521         } else {
 522                 if (bt->trace_state == Blktrace_running) {
 523                         bt->trace_state = Blktrace_stopped;
 524                         relay_flush(bt->rchan);
 525                         ret = 0;
 526                 }
 527         }
 528
 529         return ret;
 530 }
 531 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 532
 533 /**
 534  * blk_trace_ioctl: - handle the ioctls associated with tracing
 535  * @bdev:       the block device
 536  * @cmd:        the ioctl cmd
 537  * @arg:        the argument data, if any
 538  *
 539  **/
 540 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 541 {
 542         struct request_queue *q;
 543         int ret, start = 0;
 544         char b[BDEVNAME_SIZE];
 545
 546         q = bdev_get_queue(bdev);
 547         if (!q)
 548                 return -ENXIO;
 549
 550         mutex_lock(&bdev->bd_mutex);
 551
 552         switch (cmd) {
 553         case BLKTRACESETUP:
 554                 bdevname(bdev, b);
 555                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 556                 break;
 557         case BLKTRACESTART:
 558                 start = 1;
 559         case BLKTRACESTOP:
 560                 ret = blk_trace_startstop(q, start);
 561                 break;
 562         case BLKTRACETEARDOWN:
 563                 ret = blk_trace_remove(q);
 564                 break;
 565         default:
 566                 ret = -ENOTTY;
 567                 break;
 568         }
 569
 570         mutex_unlock(&bdev->bd_mutex);
 571         return ret;
 572 }
 573
 574 /**
 575  * blk_trace_shutdown: - stop and cleanup trace structures
 576  * @q:    the request queue associated with the device
 577  *
 578  **/
 579 void blk_trace_shutdown(struct request_queue *q)
 580 {
 581         if (q->blk_trace) {
 582                 blk_trace_startstop(q, 0);
 583                 blk_trace_remove(q);
 584         }
 585 }
 586
 587 /*
 588  * blktrace probes
 589  */
 590
 591 /**
 592  * blk_add_trace_rq - Add a trace for a request oriented action
 593  * @q:          queue the io is for
 594  * @rq:         the source request
 595  * @what:       the action
 596  *
 597  * Description:
 598  *     Records an action against a request. Will log the bio offset + size.
 599  *
 600  **/
 601 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 602                                     u32 what)
 603 {
 604         struct blk_trace *bt = q->blk_trace;
 605         int rw = rq->cmd_flags & 0x03;
 606
 607         if (likely(!bt))
 608                 return;
 609
 610         if (blk_discard_rq(rq))
 611                 rw |= (1 << BIO_RW_DISCARD);
 612
 613         if (blk_pc_request(rq)) {
 614                 what |= BLK_TC_ACT(BLK_TC_PC);
 615                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 616                                 sizeof(rq->cmd), rq->cmd);
 617         } else  {
 618                 what |= BLK_TC_ACT(BLK_TC_FS);
 619                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 620                                 rw, what, rq->errors, 0, NULL);
 621         }
 622 }
 623
 624 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 625 {
 626         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 627 }
 628
 629 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 630 {
 631         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 632 }
 633
 634 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 635 {
 636         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 637 }
 638
 639 static void blk_add_trace_rq_requeue(struct request_queue *q, struct request *rq)
 640 {
 641         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 642 }
 643
 644 static void blk_add_trace_rq_complete(struct request_queue *q, struct request *rq)
 645 {
 646         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 647 }
 648
 649 /**
 650  * blk_add_trace_bio - Add a trace for a bio oriented action
 651  * @q:          queue the io is for
 652  * @bio:        the source bio
 653  * @what:       the action
 654  *
 655  * Description:
 656  *     Records an action against a bio. Will log the bio offset + size.
 657  *
 658  **/
 659 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 660                                      u32 what)
 661 {
 662         struct blk_trace *bt = q->blk_trace;
 663
 664         if (likely(!bt))
 665                 return;
 666
 667         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 668                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 669 }
 670
 671 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 672 {
 673         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 674 }
 675
 676 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 677 {
 678         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 679 }
 680
 681 static void blk_add_trace_bio_backmerge(struct request_queue *q, struct bio *bio)
 682 {
 683         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 684 }
 685
 686 static void blk_add_trace_bio_frontmerge(struct request_queue *q, struct bio *bio)
 687 {
 688         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 689 }
 690
 691 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 692 {
 693         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 694 }
 695
 696 static void blk_add_trace_getrq(struct request_queue *q, struct bio *bio, int rw)
 697 {
 698         if (bio)
 699                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 700         else {
 701                 struct blk_trace *bt = q->blk_trace;
 702
 703                 if (bt)
 704                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 705         }
 706 }
 707
 708
 709 static void blk_add_trace_sleeprq(struct request_queue *q, struct bio *bio, int rw)
 710 {
 711         if (bio)
 712                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 713         else {
 714                 struct blk_trace *bt = q->blk_trace;
 715
 716                 if (bt)
 717                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ, 0, 0, NULL);
 718         }
 719 }
 720
 721 static void blk_add_trace_plug(struct request_queue *q)
 722 {
 723         struct blk_trace *bt = q->blk_trace;
 724
 725         if (bt)
 726                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 727 }
 728
 729 static void blk_add_trace_unplug_io(struct request_queue *q)
 730 {
 731         struct blk_trace *bt = q->blk_trace;
 732
 733         if (bt) {
 734                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 735                 __be64 rpdu = cpu_to_be64(pdu);
 736
 737                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 738                                 sizeof(rpdu), &rpdu);
 739         }
 740 }
 741
 742 static void blk_add_trace_unplug_timer(struct request_queue *q)
 743 {
 744         struct blk_trace *bt = q->blk_trace;
 745
 746         if (bt) {
 747                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 748                 __be64 rpdu = cpu_to_be64(pdu);
 749
 750                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 751                                 sizeof(rpdu), &rpdu);
 752         }
 753 }
 754
 755 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 756                                 unsigned int pdu)
 757 {
 758         struct blk_trace *bt = q->blk_trace;
 759
 760         if (bt) {
 761                 __be64 rpdu = cpu_to_be64(pdu);
 762
 763                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 764                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 765                                 sizeof(rpdu), &rpdu);
 766         }
 767 }
 768
 769 /**
 770  * blk_add_trace_remap - Add a trace for a remap operation
 771  * @q:          queue the io is for
 772  * @bio:        the source bio
 773  * @dev:        target device
 774  * @from:       source sector
 775  * @to:         target sector
 776  *
 777  * Description:
 778  *     Device mapper or raid target sometimes need to split a bio because
 779  *     it spans a stripe (or similar). Add a trace for that action.
 780  *
 781  **/
 782 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 783                                        dev_t dev, sector_t from, sector_t to)
 784 {
 785         struct blk_trace *bt = q->blk_trace;
 786         struct blk_io_trace_remap r;
 787
 788         if (likely(!bt))
 789                 return;
 790
 791         r.device = cpu_to_be32(dev);
 792         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 793         r.sector = cpu_to_be64(to);
 794
 795         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 796                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 797 }
 798
 799 /**
 800  * blk_add_driver_data - Add binary message with driver-specific data
 801  * @q:          queue the io is for
 802  * @rq:         io request
 803  * @data:       driver-specific data
 804  * @len:        length of driver-specific data
 805  *
 806  * Description:
 807  *     Some drivers might want to write driver-specific data per request.
 808  *
 809  **/
 810 void blk_add_driver_data(struct request_queue *q,
 811                          struct request *rq,
 812                          void *data, size_t len)
 813 {
 814         struct blk_trace *bt = q->blk_trace;
 815
 816         if (likely(!bt))
 817                 return;
 818
 819         if (blk_pc_request(rq))
 820                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 821                                 rq->errors, len, data);
 822         else
 823                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 824                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 825 }
 826 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 827
 828 static int blk_register_tracepoints(void)
 829 {
 830         int ret;
 831
 832         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 833         WARN_ON(ret);
 834         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 835         WARN_ON(ret);
 836         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 837         WARN_ON(ret);
 838         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 839         WARN_ON(ret);
 840         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 841         WARN_ON(ret);
 842         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 843         WARN_ON(ret);
 844         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 845         WARN_ON(ret);
 846         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 847         WARN_ON(ret);
 848         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 849         WARN_ON(ret);
 850         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 851         WARN_ON(ret);
 852         ret = register_trace_block_getrq(blk_add_trace_getrq);
 853         WARN_ON(ret);
 854         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 855         WARN_ON(ret);
 856         ret = register_trace_block_plug(blk_add_trace_plug);
 857         WARN_ON(ret);
 858         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 859         WARN_ON(ret);
 860         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 861         WARN_ON(ret);
 862         ret = register_trace_block_split(blk_add_trace_split);
 863         WARN_ON(ret);
 864         ret = register_trace_block_remap(blk_add_trace_remap);
 865         WARN_ON(ret);
 866         return 0;
 867 }
 868
 869 static void blk_unregister_tracepoints(void)
 870 {
 871         unregister_trace_block_remap(blk_add_trace_remap);
 872         unregister_trace_block_split(blk_add_trace_split);
 873         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 874         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 875         unregister_trace_block_plug(blk_add_trace_plug);
 876         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 877         unregister_trace_block_getrq(blk_add_trace_getrq);
 878         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 879         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 880         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 881         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 882         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 883         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 884         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 885         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 886         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 887         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 888
 889         tracepoint_synchronize_unregister();
 890 }