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, SYNCIO);
 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
 191 static void blk_trace_cleanup(struct blk_trace *bt)
 192 {
 193         debugfs_remove(bt->msg_file);
 194         debugfs_remove(bt->dropped_file);
 195         relay_close(bt->rchan);
 196         free_percpu(bt->sequence);
 197         free_percpu(bt->msg_data);
 198         kfree(bt);
 199         mutex_lock(&blk_probe_mutex);
 200         if (atomic_dec_and_test(&blk_probes_ref))
 201                 blk_unregister_tracepoints();
 202         mutex_unlock(&blk_probe_mutex);
 203 }
 204
 205 int blk_trace_remove(struct request_queue *q)
 206 {
 207         struct blk_trace *bt;
 208
 209         bt = xchg(&q->blk_trace, NULL);
 210         if (!bt)
 211                 return -EINVAL;
 212
 213         if (bt->trace_state == Blktrace_setup ||
 214             bt->trace_state == Blktrace_stopped)
 215                 blk_trace_cleanup(bt);
 216
 217         return 0;
 218 }
 219 EXPORT_SYMBOL_GPL(blk_trace_remove);
 220
 221 static int blk_dropped_open(struct inode *inode, struct file *filp)
 222 {
 223         filp->private_data = inode->i_private;
 224
 225         return 0;
 226 }
 227
 228 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 229                                 size_t count, loff_t *ppos)
 230 {
 231         struct blk_trace *bt = filp->private_data;
 232         char buf[16];
 233
 234         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 235
 236         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 237 }
 238
 239 static const struct file_operations blk_dropped_fops = {
 240         .owner =        THIS_MODULE,
 241         .open =         blk_dropped_open,
 242         .read =         blk_dropped_read,
 243 };
 244
 245 static int blk_msg_open(struct inode *inode, struct file *filp)
 246 {
 247         filp->private_data = inode->i_private;
 248
 249         return 0;
 250 }
 251
 252 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 253                                 size_t count, loff_t *ppos)
 254 {
 255         char *msg;
 256         struct blk_trace *bt;
 257
 258         if (count > BLK_TN_MAX_MSG)
 259                 return -EINVAL;
 260
 261         msg = kmalloc(count, GFP_KERNEL);
 262         if (msg == NULL)
 263                 return -ENOMEM;
 264
 265         if (copy_from_user(msg, buffer, count)) {
 266                 kfree(msg);
 267                 return -EFAULT;
 268         }
 269
 270         bt = filp->private_data;
 271         __trace_note_message(bt, "%s", msg);
 272         kfree(msg);
 273
 274         return count;
 275 }
 276
 277 static const struct file_operations blk_msg_fops = {
 278         .owner =        THIS_MODULE,
 279         .open =         blk_msg_open,
 280         .write =        blk_msg_write,
 281 };
 282
 283 /*
 284  * Keep track of how many times we encountered a full subbuffer, to aid
 285  * the user space app in telling how many lost events there were.
 286  */
 287 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 288                                      void *prev_subbuf, size_t prev_padding)
 289 {
 290         struct blk_trace *bt;
 291
 292         if (!relay_buf_full(buf))
 293                 return 1;
 294
 295         bt = buf->chan->private_data;
 296         atomic_inc(&bt->dropped);
 297         return 0;
 298 }
 299
 300 static int blk_remove_buf_file_callback(struct dentry *dentry)
 301 {
 302         struct dentry *parent = dentry->d_parent;
 303         debugfs_remove(dentry);
 304
 305         /*
 306         * this will fail for all but the last file, but that is ok. what we
 307         * care about is the top level buts->name directory going away, when
 308         * the last trace file is gone. Then we don't have to rmdir() that
 309         * manually on trace stop, so it nicely solves the issue with
 310         * force killing of running traces.
 311         */
 312
 313         debugfs_remove(parent);
 314         return 0;
 315 }
 316
 317 static struct dentry *blk_create_buf_file_callback(const char *filename,
 318                                                    struct dentry *parent,
 319                                                    int mode,
 320                                                    struct rchan_buf *buf,
 321                                                    int *is_global)
 322 {
 323         return debugfs_create_file(filename, mode, parent, buf,
 324                                         &relay_file_operations);
 325 }
 326
 327 static struct rchan_callbacks blk_relay_callbacks = {
 328         .subbuf_start           = blk_subbuf_start_callback,
 329         .create_buf_file        = blk_create_buf_file_callback,
 330         .remove_buf_file        = blk_remove_buf_file_callback,
 331 };
 332
 333 /*
 334  * Setup everything required to start tracing
 335  */
 336 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 337                         struct blk_user_trace_setup *buts)
 338 {
 339         struct blk_trace *old_bt, *bt = NULL;
 340         struct dentry *dir = NULL;
 341         int ret, i;
 342
 343         if (!buts->buf_size || !buts->buf_nr)
 344                 return -EINVAL;
 345
 346         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 347         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 348
 349         /*
 350          * some device names have larger paths - convert the slashes
 351          * to underscores for this to work as expected
 352          */
 353         for (i = 0; i < strlen(buts->name); i++)
 354                 if (buts->name[i] == '/')
 355                         buts->name[i] = '_';
 356
 357         ret = -ENOMEM;
 358         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 359         if (!bt)
 360                 goto err;
 361
 362         bt->sequence = alloc_percpu(unsigned long);
 363         if (!bt->sequence)
 364                 goto err;
 365
 366         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 367         if (!bt->msg_data)
 368                 goto err;
 369
 370         ret = -ENOENT;
 371
 372         if (!blk_tree_root) {
 373                 blk_tree_root = debugfs_create_dir("block", NULL);
 374                 if (!blk_tree_root)
 375                         return -ENOMEM;
 376         }
 377
 378         dir = debugfs_create_dir(buts->name, blk_tree_root);
 379
 380         if (!dir)
 381                 goto err;
 382
 383         bt->dir = dir;
 384         bt->dev = dev;
 385         atomic_set(&bt->dropped, 0);
 386
 387         ret = -EIO;
 388         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 389         if (!bt->dropped_file)
 390                 goto err;
 391
 392         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 393         if (!bt->msg_file)
 394                 goto err;
 395
 396         bt->rchan = relay_open("trace", dir, buts->buf_size,
 397                                 buts->buf_nr, &blk_relay_callbacks, bt);
 398         if (!bt->rchan)
 399                 goto err;
 400
 401         bt->act_mask = buts->act_mask;
 402         if (!bt->act_mask)
 403                 bt->act_mask = (u16) -1;
 404
 405         bt->start_lba = buts->start_lba;
 406         bt->end_lba = buts->end_lba;
 407         if (!bt->end_lba)
 408                 bt->end_lba = -1ULL;
 409
 410         bt->pid = buts->pid;
 411         bt->trace_state = Blktrace_setup;
 412
 413         mutex_lock(&blk_probe_mutex);
 414         if (atomic_add_return(1, &blk_probes_ref) == 1) {
 415                 ret = blk_register_tracepoints();
 416                 if (ret)
 417                         goto probe_err;
 418         }
 419         mutex_unlock(&blk_probe_mutex);
 420
 421         ret = -EBUSY;
 422         old_bt = xchg(&q->blk_trace, bt);
 423         if (old_bt) {
 424                 (void) xchg(&q->blk_trace, old_bt);
 425                 goto err;
 426         }
 427
 428         return 0;
 429 probe_err:
 430         atomic_dec(&blk_probes_ref);
 431         mutex_unlock(&blk_probe_mutex);
 432 err:
 433         if (bt) {
 434                 if (bt->msg_file)
 435                         debugfs_remove(bt->msg_file);
 436                 if (bt->dropped_file)
 437                         debugfs_remove(bt->dropped_file);
 438                 free_percpu(bt->sequence);
 439                 free_percpu(bt->msg_data);
 440                 if (bt->rchan)
 441                         relay_close(bt->rchan);
 442                 kfree(bt);
 443         }
 444         return ret;
 445 }
 446
 447 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 448                     char __user *arg)
 449 {
 450         struct blk_user_trace_setup buts;
 451         int ret;
 452
 453         ret = copy_from_user(&buts, arg, sizeof(buts));
 454         if (ret)
 455                 return -EFAULT;
 456
 457         ret = do_blk_trace_setup(q, name, dev, &buts);
 458         if (ret)
 459                 return ret;
 460
 461         if (copy_to_user(arg, &buts, sizeof(buts)))
 462                 return -EFAULT;
 463
 464         return 0;
 465 }
 466 EXPORT_SYMBOL_GPL(blk_trace_setup);
 467
 468 int blk_trace_startstop(struct request_queue *q, int start)
 469 {
 470         struct blk_trace *bt;
 471         int ret;
 472
 473         if ((bt = q->blk_trace) == NULL)
 474                 return -EINVAL;
 475
 476         /*
 477          * For starting a trace, we can transition from a setup or stopped
 478          * trace. For stopping a trace, the state must be running
 479          */
 480         ret = -EINVAL;
 481         if (start) {
 482                 if (bt->trace_state == Blktrace_setup ||
 483                     bt->trace_state == Blktrace_stopped) {
 484                         blktrace_seq++;
 485                         smp_mb();
 486                         bt->trace_state = Blktrace_running;
 487
 488                         trace_note_time(bt);
 489                         ret = 0;
 490                 }
 491         } else {
 492                 if (bt->trace_state == Blktrace_running) {
 493                         bt->trace_state = Blktrace_stopped;
 494                         relay_flush(bt->rchan);
 495                         ret = 0;
 496                 }
 497         }
 498
 499         return ret;
 500 }
 501 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 502
 503 /**
 504  * blk_trace_ioctl: - handle the ioctls associated with tracing
 505  * @bdev:       the block device
 506  * @cmd:        the ioctl cmd
 507  * @arg:        the argument data, if any
 508  *
 509  **/
 510 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 511 {
 512         struct request_queue *q;
 513         int ret, start = 0;
 514         char b[BDEVNAME_SIZE];
 515
 516         q = bdev_get_queue(bdev);
 517         if (!q)
 518                 return -ENXIO;
 519
 520         mutex_lock(&bdev->bd_mutex);
 521
 522         switch (cmd) {
 523         case BLKTRACESETUP:
 524                 bdevname(bdev, b);
 525                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 526                 break;
 527         case BLKTRACESTART:
 528                 start = 1;
 529         case BLKTRACESTOP:
 530                 ret = blk_trace_startstop(q, start);
 531                 break;
 532         case BLKTRACETEARDOWN:
 533                 ret = blk_trace_remove(q);
 534                 break;
 535         default:
 536                 ret = -ENOTTY;
 537                 break;
 538         }
 539
 540         mutex_unlock(&bdev->bd_mutex);
 541         return ret;
 542 }
 543
 544 /**
 545  * blk_trace_shutdown: - stop and cleanup trace structures
 546  * @q:    the request queue associated with the device
 547  *
 548  **/
 549 void blk_trace_shutdown(struct request_queue *q)
 550 {
 551         if (q->blk_trace) {
 552                 blk_trace_startstop(q, 0);
 553                 blk_trace_remove(q);
 554         }
 555 }
 556
 557 /*
 558  * blktrace probes
 559  */
 560
 561 /**
 562  * blk_add_trace_rq - Add a trace for a request oriented action
 563  * @q:          queue the io is for
 564  * @rq:         the source request
 565  * @what:       the action
 566  *
 567  * Description:
 568  *     Records an action against a request. Will log the bio offset + size.
 569  *
 570  **/
 571 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 572                                     u32 what)
 573 {
 574         struct blk_trace *bt = q->blk_trace;
 575         int rw = rq->cmd_flags & 0x03;
 576
 577         if (likely(!bt))
 578                 return;
 579
 580         if (blk_discard_rq(rq))
 581                 rw |= (1 << BIO_RW_DISCARD);
 582
 583         if (blk_pc_request(rq)) {
 584                 what |= BLK_TC_ACT(BLK_TC_PC);
 585                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 586                                 sizeof(rq->cmd), rq->cmd);
 587         } else  {
 588                 what |= BLK_TC_ACT(BLK_TC_FS);
 589                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 590                                 rw, what, rq->errors, 0, NULL);
 591         }
 592 }
 593
 594 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 595 {
 596         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 597 }
 598
 599 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 600 {
 601         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 602 }
 603
 604 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 605 {
 606         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 607 }
 608
 609 static void blk_add_trace_rq_requeue(struct request_queue *q, struct request *rq)
 610 {
 611         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 612 }
 613
 614 static void blk_add_trace_rq_complete(struct request_queue *q, struct request *rq)
 615 {
 616         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 617 }
 618
 619 /**
 620  * blk_add_trace_bio - Add a trace for a bio oriented action
 621  * @q:          queue the io is for
 622  * @bio:        the source bio
 623  * @what:       the action
 624  *
 625  * Description:
 626  *     Records an action against a bio. Will log the bio offset + size.
 627  *
 628  **/
 629 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 630                                      u32 what)
 631 {
 632         struct blk_trace *bt = q->blk_trace;
 633
 634         if (likely(!bt))
 635                 return;
 636
 637         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 638                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 639 }
 640
 641 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 642 {
 643         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 644 }
 645
 646 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 647 {
 648         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 649 }
 650
 651 static void blk_add_trace_bio_backmerge(struct request_queue *q, struct bio *bio)
 652 {
 653         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 654 }
 655
 656 static void blk_add_trace_bio_frontmerge(struct request_queue *q, struct bio *bio)
 657 {
 658         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 659 }
 660
 661 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 662 {
 663         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 664 }
 665
 666 static void blk_add_trace_getrq(struct request_queue *q, struct bio *bio, int rw)
 667 {
 668         if (bio)
 669                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 670         else {
 671                 struct blk_trace *bt = q->blk_trace;
 672
 673                 if (bt)
 674                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 675         }
 676 }
 677
 678
 679 static void blk_add_trace_sleeprq(struct request_queue *q, struct bio *bio, int rw)
 680 {
 681         if (bio)
 682                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 683         else {
 684                 struct blk_trace *bt = q->blk_trace;
 685
 686                 if (bt)
 687                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ, 0, 0, NULL);
 688         }
 689 }
 690
 691 static void blk_add_trace_plug(struct request_queue *q)
 692 {
 693         struct blk_trace *bt = q->blk_trace;
 694
 695         if (bt)
 696                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 697 }
 698
 699 static void blk_add_trace_unplug_io(struct request_queue *q)
 700 {
 701         struct blk_trace *bt = q->blk_trace;
 702
 703         if (bt) {
 704                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 705                 __be64 rpdu = cpu_to_be64(pdu);
 706
 707                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 708                                 sizeof(rpdu), &rpdu);
 709         }
 710 }
 711
 712 static void blk_add_trace_unplug_timer(struct request_queue *q)
 713 {
 714         struct blk_trace *bt = q->blk_trace;
 715
 716         if (bt) {
 717                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 718                 __be64 rpdu = cpu_to_be64(pdu);
 719
 720                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 721                                 sizeof(rpdu), &rpdu);
 722         }
 723 }
 724
 725 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 726                                 unsigned int pdu)
 727 {
 728         struct blk_trace *bt = q->blk_trace;
 729
 730         if (bt) {
 731                 __be64 rpdu = cpu_to_be64(pdu);
 732
 733                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 734                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 735                                 sizeof(rpdu), &rpdu);
 736         }
 737 }
 738
 739 /**
 740  * blk_add_trace_remap - Add a trace for a remap operation
 741  * @q:          queue the io is for
 742  * @bio:        the source bio
 743  * @dev:        target device
 744  * @from:       source sector
 745  * @to:         target sector
 746  *
 747  * Description:
 748  *     Device mapper or raid target sometimes need to split a bio because
 749  *     it spans a stripe (or similar). Add a trace for that action.
 750  *
 751  **/
 752 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 753                                        dev_t dev, sector_t from, sector_t to)
 754 {
 755         struct blk_trace *bt = q->blk_trace;
 756         struct blk_io_trace_remap r;
 757
 758         if (likely(!bt))
 759                 return;
 760
 761         r.device = cpu_to_be32(dev);
 762         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 763         r.sector = cpu_to_be64(to);
 764
 765         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 766                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 767 }
 768
 769 /**
 770  * blk_add_driver_data - Add binary message with driver-specific data
 771  * @q:          queue the io is for
 772  * @rq:         io request
 773  * @data:       driver-specific data
 774  * @len:        length of driver-specific data
 775  *
 776  * Description:
 777  *     Some drivers might want to write driver-specific data per request.
 778  *
 779  **/
 780 void blk_add_driver_data(struct request_queue *q,
 781                          struct request *rq,
 782                          void *data, size_t len)
 783 {
 784         struct blk_trace *bt = q->blk_trace;
 785
 786         if (likely(!bt))
 787                 return;
 788
 789         if (blk_pc_request(rq))
 790                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 791                                 rq->errors, len, data);
 792         else
 793                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 794                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 795 }
 796 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 797
 798 static int blk_register_tracepoints(void)
 799 {
 800         int ret;
 801
 802         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 803         WARN_ON(ret);
 804         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 805         WARN_ON(ret);
 806         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 807         WARN_ON(ret);
 808         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 809         WARN_ON(ret);
 810         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 811         WARN_ON(ret);
 812         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 813         WARN_ON(ret);
 814         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 815         WARN_ON(ret);
 816         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 817         WARN_ON(ret);
 818         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 819         WARN_ON(ret);
 820         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 821         WARN_ON(ret);
 822         ret = register_trace_block_getrq(blk_add_trace_getrq);
 823         WARN_ON(ret);
 824         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 825         WARN_ON(ret);
 826         ret = register_trace_block_plug(blk_add_trace_plug);
 827         WARN_ON(ret);
 828         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 829         WARN_ON(ret);
 830         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 831         WARN_ON(ret);
 832         ret = register_trace_block_split(blk_add_trace_split);
 833         WARN_ON(ret);
 834         ret = register_trace_block_remap(blk_add_trace_remap);
 835         WARN_ON(ret);
 836         return 0;
 837 }
 838
 839 static void blk_unregister_tracepoints(void)
 840 {
 841         unregister_trace_block_remap(blk_add_trace_remap);
 842         unregister_trace_block_split(blk_add_trace_split);
 843         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 844         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 845         unregister_trace_block_plug(blk_add_trace_plug);
 846         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 847         unregister_trace_block_getrq(blk_add_trace_getrq);
 848         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 849         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 850         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 851         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 852         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 853         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 854         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 855         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 856         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 857         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 858
 859         tracepoint_synchronize_unregister();
 860 }