kernel/blk-trace-2.6.16-rc1-git-S0

   1 diff --git a/block/Kconfig b/block/Kconfig
   2 index 377f6dd..27eaed9 100644
   3 --- a/block/Kconfig
   4 +++ b/block/Kconfig
   5 @@ -11,4 +11,15 @@ config LBD
   6           your machine, or if you want to have a raid or loopback device
   7           bigger than 2TB.  Otherwise say N.
   8
   9 +config BLK_DEV_IO_TRACE
  10 +       bool "Support for tracing block io actions"
  11 +       select RELAYFS_FS
  12 +       help
  13 +         Say Y here, if you want to be able to trace the block layer actions
  14 +         on a given queue. Tracing allows you to see any traffic happening
  15 +         on a block device queue. For more information (and the user space
  16 +         support tools needed), fetch the blktrace app from:
  17 +
  18 +         git://brick.kernel.dk/data/git/blktrace.git
  19 +
  20  source block/Kconfig.iosched
  21 diff --git a/block/Makefile b/block/Makefile
  22 index 7e4f93e..c05de0e 100644
  23 --- a/block/Makefile
  24 +++ b/block/Makefile
  25 @@ -8,3 +8,5 @@ obj-$(CONFIG_IOSCHED_NOOP)      += noop-iosch
  26  obj-$(CONFIG_IOSCHED_AS)       += as-iosched.o
  27  obj-$(CONFIG_IOSCHED_DEADLINE) += deadline-iosched.o
  28  obj-$(CONFIG_IOSCHED_CFQ)      += cfq-iosched.o
  29 +
  30 +obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
  31 diff --git a/block/blktrace.c b/block/blktrace.c
  32 new file mode 100644
  33 index 0000000..5ffd46f
  34 --- /dev/null
  35 +++ b/block/blktrace.c
  36 @@ -0,0 +1,338 @@
  37 +#include <linux/config.h>
  38 +#include <linux/kernel.h>
  39 +#include <linux/blkdev.h>
  40 +#include <linux/blktrace_api.h>
  41 +#include <linux/percpu.h>
  42 +#include <linux/init.h>
  43 +#include <linux/mutex.h>
  44 +#include <asm/uaccess.h>
  45 +
  46 +static DEFINE_PER_CPU(unsigned long long, blk_trace_cpu_offset) = { 0, };
  47 +
  48 +void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
  49 +                    int rw, u32 what, int error, int pdu_len, void *pdu_data)
  50 +{
  51 +       struct blk_io_trace t;
  52 +       unsigned long flags;
  53 +       pid_t pid;
  54 +       int cpu;
  55 +
  56 +       if (rw & (1 << BIO_RW_BARRIER))
  57 +               what |= BLK_TC_ACT(BLK_TC_BARRIER);
  58 +       if (rw & (1 << BIO_RW_SYNC))
  59 +               what |= BLK_TC_ACT(BLK_TC_SYNC);
  60 +
  61 +       if (rw & WRITE)
  62 +               what |= BLK_TC_ACT(BLK_TC_WRITE);
  63 +       else
  64 +               what |= BLK_TC_ACT(BLK_TC_READ);
  65 +
  66 +       if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
  67 +               return;
  68 +       if (sector < bt->start_lba || sector > bt->end_lba)
  69 +               return;
  70 +
  71 +       pid = current->pid;
  72 +       if (bt->pid && pid != bt->pid)
  73 +               return;
  74 +
  75 +       t.magic         = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  76 +
  77 +       t.device        = bt->dev;
  78 +       t.sector        = sector;
  79 +       t.bytes         = bytes;
  80 +       t.action        = what;
  81 +       t.error         = error;
  82 +       t.pdu_len       = pdu_len;
  83 +
  84 +       t.pid           = pid;
  85 +       memcpy(t.comm, current->comm, sizeof(t.comm));
  86 +
  87 +       /*
  88 +        * need to serialize this part completely to prevent multiple CPUs
  89 +        * from misordering events
  90 +        */
  91 +       spin_lock_irqsave(&bt->lock, flags);
  92 +
  93 +       t.sequence      = ++bt->sequence;
  94 +
  95 +       cpu = smp_processor_id();
  96 +       t.cpu           = cpu;
  97 +       t.time          = sched_clock() - per_cpu(blk_trace_cpu_offset, cpu);
  98 +
  99 +       __relay_write(bt->rchan, &t, sizeof(t));
 100 +       if (pdu_len)
 101 +               __relay_write(bt->rchan, pdu_data, pdu_len);
 102 +
 103 +       spin_unlock_irqrestore(&bt->lock, flags);
 104 +}
 105 +
 106 +EXPORT_SYMBOL_GPL(__blk_add_trace);
 107 +
 108 +static struct dentry *blk_tree_root;
 109 +static struct mutex blk_tree_mutex;
 110 +
 111 +static inline void blk_remove_root(void)
 112 +{
 113 +       if (relayfs_remove_dir(blk_tree_root) != -ENOTEMPTY)
 114 +               blk_tree_root = NULL;
 115 +}
 116 +
 117 +static void blk_remove_tree(struct dentry *dir)
 118 +{
 119 +       mutex_lock(&blk_tree_mutex);
 120 +       relayfs_remove_dir(dir);
 121 +       blk_remove_root();
 122 +       mutex_unlock(&blk_tree_mutex);
 123 +}
 124 +
 125 +static struct dentry *blk_create_tree(const char *blk_name)
 126 +{
 127 +       struct dentry *dir = NULL;
 128 +
 129 +       mutex_lock(&blk_tree_mutex);
 130 +
 131 +       if (!blk_tree_root) {
 132 +               blk_tree_root = relayfs_create_dir("block", NULL);
 133 +               if (!blk_tree_root)
 134 +                       goto err;
 135 +       }
 136 +
 137 +       dir = relayfs_create_dir(blk_name, blk_tree_root);
 138 +       if (!dir)
 139 +               blk_remove_root();
 140 +
 141 +err:
 142 +       mutex_unlock(&blk_tree_mutex);
 143 +       return dir;
 144 +}
 145 +
 146 +void blk_cleanup_trace(struct blk_trace *bt)
 147 +{
 148 +       relay_close(bt->rchan);
 149 +       relayfs_remove_file(bt->dropped_file);
 150 +       blk_remove_tree(bt->dir);
 151 +       kfree(bt);
 152 +}
 153 +
 154 +int blk_stop_trace(struct block_device *bdev)
 155 +{
 156 +       request_queue_t *q = bdev_get_queue(bdev);
 157 +       struct blk_trace *bt = NULL;
 158 +       int ret = -EINVAL;
 159 +
 160 +       if (!q)
 161 +               return -ENXIO;
 162 +
 163 +       down(&bdev->bd_sem);
 164 +
 165 +       if (q->blk_trace) {
 166 +               bt = q->blk_trace;
 167 +               q->blk_trace = NULL;
 168 +               ret = 0;
 169 +       }
 170 +
 171 +       up(&bdev->bd_sem);
 172 +
 173 +       if (bt)
 174 +               blk_cleanup_trace(bt);
 175 +
 176 +       return ret;
 177 +}
 178 +
 179 +static int blk_dropped_open(struct inode *inode, struct file *filp)
 180 +{
 181 +       filp->private_data = inode->u.generic_ip;
 182 +
 183 +       return 0;
 184 +}
 185 +
 186 +static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 187 +                               size_t count, loff_t *ppos)
 188 +{
 189 +       struct blk_trace *bt = filp->private_data;
 190 +       char buf[16];
 191 +
 192 +       snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 193 +
 194 +       return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 195 +}
 196 +
 197 +static struct file_operations blk_dropped_fops = {
 198 +       .owner =        THIS_MODULE,
 199 +       .open =         blk_dropped_open,
 200 +       .read =         blk_dropped_read,
 201 +};
 202 +
 203 +static int blk_subbuf_start_callback (struct rchan_buf *buf,
 204 +                                     void *subbuf,
 205 +                                     void *prev_subbuf,
 206 +                                     size_t prev_padding)
 207 +{
 208 +       if (relay_buf_full(buf)) {
 209 +               struct blk_trace *bt = buf->chan->private_data;
 210 +
 211 +               atomic_inc(&bt->dropped);
 212 +               return 0;
 213 +       }
 214 +
 215 +       return 1;
 216 +}
 217 +
 218 +static struct rchan_callbacks blk_relay_callbacks = {
 219 +       .subbuf_start = blk_subbuf_start_callback,
 220 +};
 221 +
 222 +int blk_start_trace(struct block_device *bdev, char __user *arg)
 223 +{
 224 +       request_queue_t *q = bdev_get_queue(bdev);
 225 +       struct blk_user_trace_setup buts;
 226 +       struct blk_trace *bt = NULL;
 227 +       struct dentry *dir = NULL;
 228 +       char b[BDEVNAME_SIZE];
 229 +       int ret, i;
 230 +
 231 +       if (!q)
 232 +               return -ENXIO;
 233 +
 234 +       if (copy_from_user(&buts, arg, sizeof(buts)))
 235 +               return -EFAULT;
 236 +
 237 +       if (!buts.buf_size || !buts.buf_nr)
 238 +               return -EINVAL;
 239 +
 240 +       strcpy(buts.name, bdevname(bdev, b));
 241 +
 242 +       /*
 243 +        * some device names have larger paths - convert the slashes
 244 +        * to underscores for this to work as expected
 245 +        */
 246 +       for (i = 0; i < strlen(buts.name); i++)
 247 +               if (buts.name[i] == '/')
 248 +                       buts.name[i] = '_';
 249 +
 250 +       if (copy_to_user(arg, &buts, sizeof(buts)))
 251 +               return -EFAULT;
 252 +
 253 +       down(&bdev->bd_sem);
 254 +       ret = -EBUSY;
 255 +       if (q->blk_trace)
 256 +               goto err;
 257 +
 258 +       ret = -ENOMEM;
 259 +       bt = kmalloc(sizeof(*bt), GFP_KERNEL);
 260 +       if (!bt)
 261 +               goto err;
 262 +
 263 +       ret = -ENOENT;
 264 +       dir = blk_create_tree(buts.name);
 265 +       if (!dir)
 266 +               goto err;
 267 +
 268 +       bt->dir = dir;
 269 +       bt->dev = bdev->bd_dev;
 270 +       bt->sequence = 0;
 271 +       spin_lock_init(&bt->lock);
 272 +       atomic_set(&bt->dropped, 0);
 273 +
 274 +       ret = -EIO;
 275 +       bt->dropped_file = relayfs_create_file("dropped", dir, 0, &blk_dropped_fops, bt);
 276 +       if (!bt->dropped_file)
 277 +               goto err;
 278 +
 279 +       bt->rchan = relay_open("trace", dir, buts.buf_size, buts.buf_nr, &blk_relay_callbacks);
 280 +       if (!bt->rchan)
 281 +               goto err;
 282 +       bt->rchan->private_data = bt;
 283 +
 284 +       bt->act_mask = buts.act_mask;
 285 +       if (!bt->act_mask)
 286 +               bt->act_mask = (u16) -1;
 287 +
 288 +       bt->start_lba = buts.start_lba;
 289 +       bt->end_lba = buts.end_lba;
 290 +       if (!bt->end_lba)
 291 +               bt->end_lba = -1ULL;
 292 +
 293 +       bt->pid = buts.pid;
 294 +
 295 +       q->blk_trace = bt;
 296 +       up(&bdev->bd_sem);
 297 +       return 0;
 298 +err:
 299 +       up(&bdev->bd_sem);
 300 +       if (bt && bt->dropped_file)
 301 +               relayfs_remove_file(bt->dropped_file);
 302 +       if (dir)
 303 +               blk_remove_tree(dir);
 304 +       if (bt)
 305 +               kfree(bt);
 306 +       return ret;
 307 +}
 308 +
 309 +static void blk_check_time(unsigned long long *t)
 310 +{
 311 +       unsigned long long a, b;
 312 +       struct timeval tv;
 313 +
 314 +       a = sched_clock();
 315 +       do_gettimeofday(&tv);
 316 +       b = sched_clock();
 317 +
 318 +       *t = tv.tv_sec * 1000000000 + tv.tv_usec * 1000;
 319 +       *t -= (a + b) / 2;
 320 +}
 321 +
 322 +static void blk_trace_check_cpu_time(void *data)
 323 +{
 324 +       unsigned long long *t;
 325 +       int cpu = get_cpu();
 326 +
 327 +       t = &per_cpu(blk_trace_cpu_offset, cpu);
 328 +
 329 +       blk_check_time(t);
 330 +       blk_check_time(t);
 331 +
 332 +       put_cpu();
 333 +}
 334 +
 335 +static void blk_trace_calibrate_offsets(void)
 336 +{
 337 +       unsigned long flags;
 338 +
 339 +       smp_call_function(blk_trace_check_cpu_time, NULL, 1, 1);
 340 +       local_irq_save(flags);
 341 +       blk_trace_check_cpu_time(NULL);
 342 +       local_irq_restore(flags);
 343 +}
 344 +
 345 +static __init int blk_trace_init(void)
 346 +{
 347 +       int cpu, i;
 348 +
 349 +       mutex_init(&blk_tree_mutex);
 350 +       blk_trace_calibrate_offsets();
 351 +
 352 +       /*
 353 +        * now make sure HT siblings have the same time offset
 354 +        */
 355 +       preempt_disable();
 356 +       for_each_online_cpu(cpu) {
 357 +               unsigned long long *cpu_off, *sibling_off;
 358 +
 359 +               for_each_cpu_mask(i, cpu_sibling_map[cpu]) {
 360 +                       if (i == cpu)
 361 +                               continue;
 362 +
 363 +                       cpu_off = &per_cpu(blk_trace_cpu_offset, cpu);
 364 +                       sibling_off = &per_cpu(blk_trace_cpu_offset, i);
 365 +                       *sibling_off = *cpu_off;
 366 +               }
 367 +       }
 368 +       preempt_enable();
 369 +
 370 +       return 0;
 371 +}
 372 +
 373 +module_init(blk_trace_init);
 374 +
 375 diff --git a/block/elevator.c b/block/elevator.c
 376 index c9f424d..793c686 100644
 377 --- a/block/elevator.c
 378 +++ b/block/elevator.c
 379 @@ -33,6 +33,7 @@
 380  #include <linux/init.h>
 381  #include <linux/compiler.h>
 382  #include <linux/delay.h>
 383 +#include <linux/blktrace_api.h>
 384
 385  #include <asm/uaccess.h>
 386
 387 @@ -333,6 +334,8 @@ void __elv_add_request(request_queue_t *
 388         struct list_head *pos;
 389         unsigned ordseq;
 390
 391 +       blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 392 +
 393         if (q->ordcolor)
 394                 rq->flags |= REQ_ORDERED_COLOR;
 395
 396 @@ -491,6 +494,7 @@ struct request *elv_next_request(request
 397                          * not be passed by new incoming requests
 398                          */
 399                         rq->flags |= REQ_STARTED;
 400 +                       blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 401                 }
 402
 403                 if (!q->boundary_rq || q->boundary_rq == rq) {
 404 diff --git a/block/ioctl.c b/block/ioctl.c
 405 index e110949..63e67a2 100644
 406 --- a/block/ioctl.c
 407 +++ b/block/ioctl.c
 408 @@ -5,6 +5,7 @@
 409  #include <linux/backing-dev.h>
 410  #include <linux/buffer_head.h>
 411  #include <linux/smp_lock.h>
 412 +#include <linux/blktrace_api.h>
 413  #include <asm/uaccess.h>
 414
 415  static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
 416 @@ -189,6 +190,10 @@ static int blkdev_locked_ioctl(struct fi
 417                 return put_ulong(arg, bdev->bd_inode->i_size >> 9);
 418         case BLKGETSIZE64:
 419                 return put_u64(arg, bdev->bd_inode->i_size);
 420 +       case BLKSTARTTRACE:
 421 +               return blk_start_trace(bdev, (char __user *) arg);
 422 +       case BLKSTOPTRACE:
 423 +               return blk_stop_trace(bdev);
 424         }
 425         return -ENOIOCTLCMD;
 426  }
 427 diff --git a/block/ll_rw_blk.c b/block/ll_rw_blk.c
 428 index 8e27d0a..bfcde0f 100644
 429 --- a/block/ll_rw_blk.c
 430 +++ b/block/ll_rw_blk.c
 431 @@ -28,6 +28,7 @@
 432  #include <linux/writeback.h>
 433  #include <linux/interrupt.h>
 434  #include <linux/cpu.h>
 435 +#include <linux/blktrace_api.h>
 436
 437  /*
 438   * for max sense size
 439 @@ -1555,8 +1556,10 @@ void blk_plug_device(request_queue_t *q)
 440         if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
 441                 return;
 442
 443 -       if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
 444 +       if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
 445                 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
 446 +               blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
 447 +       }
 448  }
 449
 450  EXPORT_SYMBOL(blk_plug_device);
 451 @@ -1620,14 +1623,21 @@ static void blk_backing_dev_unplug(struc
 452         /*
 453          * devices don't necessarily have an ->unplug_fn defined
 454          */
 455 -       if (q->unplug_fn)
 456 +       if (q->unplug_fn) {
 457 +               blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
 458 +                                       q->rq.count[READ] + q->rq.count[WRITE]);
 459 +
 460                 q->unplug_fn(q);
 461 +       }
 462  }
 463
 464  static void blk_unplug_work(void *data)
 465  {
 466         request_queue_t *q = data;
 467
 468 +       blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
 469 +                               q->rq.count[READ] + q->rq.count[WRITE]);
 470 +
 471         q->unplug_fn(q);
 472  }
 473
 474 @@ -1635,6 +1645,9 @@ static void blk_unplug_timeout(unsigned
 475  {
 476         request_queue_t *q = (request_queue_t *)data;
 477
 478 +       blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
 479 +                               q->rq.count[READ] + q->rq.count[WRITE]);
 480 +
 481         kblockd_schedule_work(&q->unplug_work);
 482  }
 483
 484 @@ -1757,6 +1770,11 @@ void blk_cleanup_queue(request_queue_t *
 485         if (q->queue_tags)
 486                 __blk_queue_free_tags(q);
 487
 488 +       if (q->blk_trace) {
 489 +               blk_cleanup_trace(q->blk_trace);
 490 +               q->blk_trace = NULL;
 491 +       }
 492 +
 493         kmem_cache_free(requestq_cachep, q);
 494  }
 495
 496 @@ -2108,6 +2126,8 @@ rq_starved:
 497
 498         rq_init(q, rq);
 499         rq->rl = rl;
 500 +
 501 +       blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
 502  out:
 503         return rq;
 504  }
 505 @@ -2136,6 +2156,8 @@ static struct request *get_request_wait(
 506                 if (!rq) {
 507                         struct io_context *ioc;
 508
 509 +                       blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
 510 +
 511                         __generic_unplug_device(q);
 512                         spin_unlock_irq(q->queue_lock);
 513                         io_schedule();
 514 @@ -2189,6 +2211,8 @@ EXPORT_SYMBOL(blk_get_request);
 515   */
 516  void blk_requeue_request(request_queue_t *q, struct request *rq)
 517  {
 518 +       blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 519 +
 520         if (blk_rq_tagged(rq))
 521                 blk_queue_end_tag(q, rq);
 522
 523 @@ -2820,6 +2844,8 @@ static int __make_request(request_queue_
 524                         if (!q->back_merge_fn(q, req, bio))
 525                                 break;
 526
 527 +                       blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 528 +
 529                         req->biotail->bi_next = bio;
 530                         req->biotail = bio;
 531                         req->nr_sectors = req->hard_nr_sectors += nr_sectors;
 532 @@ -2835,6 +2861,8 @@ static int __make_request(request_queue_
 533                         if (!q->front_merge_fn(q, req, bio))
 534                                 break;
 535
 536 +                       blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 537 +
 538                         bio->bi_next = req->bio;
 539                         req->bio = bio;
 540
 541 @@ -2952,6 +2980,7 @@ void generic_make_request(struct bio *bi
 542         request_queue_t *q;
 543         sector_t maxsector;
 544         int ret, nr_sectors = bio_sectors(bio);
 545 +       dev_t old_dev;
 546
 547         might_sleep();
 548         /* Test device or partition size, when known. */
 549 @@ -2978,6 +3007,8 @@ void generic_make_request(struct bio *bi
 550          * NOTE: we don't repeat the blk_size check for each new device.
 551          * Stacking drivers are expected to know what they are doing.
 552          */
 553 +       maxsector = -1;
 554 +       old_dev = 0;
 555         do {
 556                 char b[BDEVNAME_SIZE];
 557
 558 @@ -3010,6 +3041,15 @@ end_io:
 559                  */
 560                 blk_partition_remap(bio);
 561
 562 +               if (maxsector != -1)
 563 +                       blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
 564 +                                           maxsector);
 565 +
 566 +               blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 567 +
 568 +               maxsector = bio->bi_sector;
 569 +               old_dev = bio->bi_bdev->bd_dev;
 570 +
 571                 ret = q->make_request_fn(q, bio);
 572         } while (ret);
 573  }
 574 @@ -3129,6 +3169,8 @@ static int __end_that_request_first(stru
 575         int total_bytes, bio_nbytes, error, next_idx = 0;
 576         struct bio *bio;
 577
 578 +       blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
 579 +
 580         /*
 581          * extend uptodate bool to allow < 0 value to be direct io error
 582          */
 583 diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
 584 index 12d7b9b..880892a 100644
 585 --- a/drivers/block/cciss.c
 586 +++ b/drivers/block/cciss.c
 587 @@ -38,6 +38,7 @@
 588  #include <linux/hdreg.h>
 589  #include <linux/spinlock.h>
 590  #include <linux/compat.h>
 591 +#include <linux/blktrace_api.h>
 592  #include <asm/uaccess.h>
 593  #include <asm/io.h>
 594
 595 @@ -2330,6 +2331,7 @@ static inline void complete_command( ctl
 596
 597         cmd->rq->completion_data = cmd;
 598         cmd->rq->errors = status;
 599 +       blk_add_trace_rq(cmd->rq->q, cmd->rq, BLK_TA_COMPLETE);
 600         blk_complete_request(cmd->rq);
 601  }
 602
 603 diff --git a/drivers/md/dm.c b/drivers/md/dm.c
 604 index 8c16359..8c979c2 100644
 605 --- a/drivers/md/dm.c
 606 +++ b/drivers/md/dm.c
 607 @@ -17,6 +17,7 @@
 608  #include <linux/mempool.h>
 609  #include <linux/slab.h>
 610  #include <linux/idr.h>
 611 +#include <linux/blktrace_api.h>
 612
 613  static const char *_name = DM_NAME;
 614
 615 @@ -303,6 +304,8 @@ static void dec_pending(struct dm_io *io
 616                         /* nudge anyone waiting on suspend queue */
 617                         wake_up(&io->md->wait);
 618
 619 +               blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
 620 +
 621                 bio_endio(io->bio, io->bio->bi_size, io->error);
 622                 free_io(io->md, io);
 623         }
 624 @@ -361,6 +364,7 @@ static void __map_bio(struct dm_target *
 625                       struct target_io *tio)
 626  {
 627         int r;
 628 +       sector_t sector;
 629
 630         /*
 631          * Sanity checks.
 632 @@ -376,10 +380,17 @@ static void __map_bio(struct dm_target *
 633          * this io.
 634          */
 635         atomic_inc(&tio->io->io_count);
 636 +       sector = clone->bi_sector;
 637         r = ti->type->map(ti, clone, &tio->info);
 638 -       if (r > 0)
 639 +       if (r > 0) {
 640                 /* the bio has been remapped so dispatch it */
 641 +
 642 +               blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
 643 +                                   tio->io->bio->bi_bdev->bd_dev, sector,
 644 +                                   clone->bi_sector);
 645 +
 646                 generic_make_request(clone);
 647 +       }
 648
 649         else if (r < 0) {
 650                 /* error the io and bail out */
 651 diff --git a/fs/bio.c b/fs/bio.c
 652 index bbc442b..8a1b0b6 100644
 653 --- a/fs/bio.c
 654 +++ b/fs/bio.c
 655 @@ -25,6 +25,7 @@
 656  #include <linux/module.h>
 657  #include <linux/mempool.h>
 658  #include <linux/workqueue.h>
 659 +#include <linux/blktrace_api.h>
 660  #include <scsi/sg.h>           /* for struct sg_iovec */
 661
 662  #define BIO_POOL_SIZE 256
 663 @@ -1094,6 +1095,9 @@ struct bio_pair *bio_split(struct bio *b
 664         if (!bp)
 665                 return bp;
 666
 667 +       blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
 668 +                               bi->bi_sector + first_sectors);
 669 +
 670         BUG_ON(bi->bi_vcnt != 1);
 671         BUG_ON(bi->bi_idx != 0);
 672         atomic_set(&bp->cnt, 3);
 673 diff --git a/fs/compat_ioctl.c b/fs/compat_ioctl.c
 674 index 5dd0207..010e02b 100644
 675 --- a/fs/compat_ioctl.c
 676 +++ b/fs/compat_ioctl.c
 677 @@ -72,6 +72,7 @@
 678  #include <linux/i2c-dev.h>
 679  #include <linux/wireless.h>
 680  #include <linux/atalk.h>
 681 +#include <linux/blktrace_api.h>
 682
 683  #include <net/sock.h>          /* siocdevprivate_ioctl */
 684  #include <net/bluetooth/bluetooth.h>
 685 diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
 686 index 02a585f..195c3b9 100644
 687 --- a/include/linux/blkdev.h
 688 +++ b/include/linux/blkdev.h
 689 @@ -22,6 +22,7 @@ typedef struct request_queue request_que
 690  struct elevator_queue;
 691  typedef struct elevator_queue elevator_t;
 692  struct request_pm_state;
 693 +struct blk_trace;
 694
 695  #define BLKDEV_MIN_RQ  4
 696  #define BLKDEV_MAX_RQ  128     /* Default maximum */
 697 @@ -416,6 +417,8 @@ struct request_queue
 698         unsigned int            sg_reserved_size;
 699         int                     node;
 700
 701 +       struct blk_trace        *blk_trace;
 702 +
 703         /*
 704          * reserved for flush operations
 705          */
 706 diff --git a/include/linux/blktrace_api.h b/include/linux/blktrace_api.h
 707 new file mode 100644
 708 index 0000000..d6b4317
 709 --- /dev/null
 710 +++ b/include/linux/blktrace_api.h
 711 @@ -0,0 +1,216 @@
 712 +#ifndef BLKTRACE_H
 713 +#define BLKTRACE_H
 714 +
 715 +#include <linux/config.h>
 716 +#include <linux/blkdev.h>
 717 +#include <linux/relayfs_fs.h>
 718 +
 719 +/*
 720 + * Trace categories
 721 + */
 722 +enum blktrace_cat {
 723 +       BLK_TC_READ     = 1 << 0,       /* reads */
 724 +       BLK_TC_WRITE    = 1 << 1,       /* writes */
 725 +       BLK_TC_BARRIER  = 1 << 2,       /* barrier */
 726 +       BLK_TC_SYNC     = 1 << 3,       /* barrier */
 727 +       BLK_TC_QUEUE    = 1 << 4,       /* queueing/merging */
 728 +       BLK_TC_REQUEUE  = 1 << 5,       /* requeueing */
 729 +       BLK_TC_ISSUE    = 1 << 6,       /* issue */
 730 +       BLK_TC_COMPLETE = 1 << 7,       /* completions */
 731 +       BLK_TC_FS       = 1 << 8,       /* fs requests */
 732 +       BLK_TC_PC       = 1 << 9,       /* pc requests */
 733 +
 734 +       BLK_TC_END      = 1 << 15,      /* only 16-bits, reminder */
 735 +};
 736 +
 737 +#define BLK_TC_SHIFT           (16)
 738 +#define BLK_TC_ACT(act)                ((act) << BLK_TC_SHIFT)
 739 +
 740 +/*
 741 + * Basic trace actions
 742 + */
 743 +enum blktrace_act {
 744 +       __BLK_TA_QUEUE = 1,             /* queued */
 745 +       __BLK_TA_BACKMERGE,             /* back merged to existing rq */
 746 +       __BLK_TA_FRONTMERGE,            /* front merge to existing rq */
 747 +       __BLK_TA_GETRQ,                 /* allocated new request */
 748 +       __BLK_TA_SLEEPRQ,               /* sleeping on rq allocation */
 749 +       __BLK_TA_REQUEUE,               /* request requeued */
 750 +       __BLK_TA_ISSUE,                 /* sent to driver */
 751 +       __BLK_TA_COMPLETE,              /* completed by driver */
 752 +       __BLK_TA_PLUG,                  /* queue was plugged */
 753 +       __BLK_TA_UNPLUG_IO,             /* queue was unplugged by io */
 754 +       __BLK_TA_UNPLUG_TIMER,          /* queue was unplugged by timer */
 755 +       __BLK_TA_INSERT,                /* insert request */
 756 +       __BLK_TA_SPLIT,                 /* bio was split */
 757 +       __BLK_TA_BOUNCE,                /* bio was bounced */
 758 +       __BLK_TA_REMAP,                 /* bio was remapped */
 759 +};
 760 +
 761 +/*
 762 + * Trace actions in full. Additionally, read or write is masked
 763 + */
 764 +#define BLK_TA_QUEUE           (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
 765 +#define BLK_TA_BACKMERGE       (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
 766 +#define BLK_TA_FRONTMERGE      (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
 767 +#define        BLK_TA_GETRQ            (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
 768 +#define        BLK_TA_SLEEPRQ          (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
 769 +#define        BLK_TA_REQUEUE          (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
 770 +#define BLK_TA_ISSUE           (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
 771 +#define BLK_TA_COMPLETE                (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
 772 +#define BLK_TA_PLUG            (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
 773 +#define BLK_TA_UNPLUG_IO       (__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
 774 +#define BLK_TA_UNPLUG_TIMER    (__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
 775 +#define BLK_TA_INSERT          (__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
 776 +#define BLK_TA_SPLIT           (__BLK_TA_SPLIT)
 777 +#define BLK_TA_BOUNCE          (__BLK_TA_BOUNCE)
 778 +#define BLK_TA_REMAP           (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
 779 +
 780 +#define BLK_IO_TRACE_MAGIC     0x65617400
 781 +#define BLK_IO_TRACE_VERSION   0x05
 782 +
 783 +/*
 784 + * The trace itself
 785 + */
 786 +struct blk_io_trace {
 787 +       u32 magic;              /* MAGIC << 8 | version */
 788 +       u32 sequence;           /* event number */
 789 +       u64 time;               /* in microseconds */
 790 +       u64 sector;             /* disk offset */
 791 +       u32 bytes;              /* transfer length */
 792 +       u32 action;             /* what happened */
 793 +       u32 pid;                /* who did it */
 794 +       u32 cpu;                /* on what cpu did it happen */
 795 +       u16 error;              /* completion error */
 796 +       u16 pdu_len;            /* length of data after this trace */
 797 +       u32 device;             /* device number */
 798 +       char comm[16];          /* task command name (TASK_COMM_LEN) */
 799 +};
 800 +
 801 +/*
 802 + * The remap event
 803 + */
 804 +struct blk_io_trace_remap {
 805 +       u32 device;
 806 +       u32 __pad;
 807 +       u64 sector;
 808 +};
 809 +
 810 +struct blk_trace {
 811 +       struct dentry *dir;
 812 +       struct rchan *rchan;
 813 +       struct dentry *dropped_file;
 814 +       atomic_t dropped;
 815 +       spinlock_t lock;
 816 +       unsigned long sequence;
 817 +       u32 dev;
 818 +       u16 act_mask;
 819 +       u64 start_lba;
 820 +       u64 end_lba;
 821 +       u32 pid;
 822 +};
 823 +
 824 +/*
 825 + * User setup structure passed with BLKSTARTTRACE
 826 + */
 827 +struct blk_user_trace_setup {
 828 +       char name[BDEVNAME_SIZE];       /* output */
 829 +       u16 act_mask;                   /* input */
 830 +       u32 buf_size;                   /* input */
 831 +       u32 buf_nr;                     /* input */
 832 +       u64 start_lba;
 833 +       u64 end_lba;
 834 +       u32 pid;
 835 +};
 836 +
 837 +#if defined(CONFIG_BLK_DEV_IO_TRACE)
 838 +extern int blk_start_trace(struct block_device *, char __user *);
 839 +extern int blk_stop_trace(struct block_device *);
 840 +extern void blk_cleanup_trace(struct blk_trace *);
 841 +extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
 842 +
 843 +static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 844 +                                   u32 what)
 845 +{
 846 +       struct blk_trace *bt = q->blk_trace;
 847 +       int rw = rq->flags & 0x07;
 848 +
 849 +       if (likely(!bt))
 850 +               return;
 851 +
 852 +       if (blk_pc_request(rq)) {
 853 +               what |= BLK_TC_ACT(BLK_TC_PC);
 854 +               __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
 855 +       } else  {
 856 +               what |= BLK_TC_ACT(BLK_TC_FS);
 857 +               __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
 858 +       }
 859 +}
 860 +
 861 +static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 862 +                                    u32 what)
 863 +{
 864 +       struct blk_trace *bt = q->blk_trace;
 865 +
 866 +       if (likely(!bt))
 867 +               return;
 868 +
 869 +       __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 870 +}
 871 +
 872 +static inline void blk_add_trace_generic(struct request_queue *q,
 873 +                                        struct bio *bio, int rw, u32 what)
 874 +{
 875 +       struct blk_trace *bt = q->blk_trace;
 876 +
 877 +       if (likely(!bt))
 878 +               return;
 879 +
 880 +       if (bio)
 881 +               blk_add_trace_bio(q, bio, what);
 882 +       else
 883 +               __blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
 884 +}
 885 +
 886 +static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
 887 +                                        struct bio *bio, unsigned int pdu)
 888 +{
 889 +       struct blk_trace *bt = q->blk_trace;
 890 +       u64 rpdu = cpu_to_be64(pdu);
 891 +
 892 +       if (likely(!bt))
 893 +               return;
 894 +
 895 +       if (bio)
 896 +               __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
 897 +       else
 898 +               __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 899 +}
 900 +
 901 +static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 902 +                                      dev_t dev, sector_t from, sector_t to)
 903 +{
 904 +       struct blk_trace *bt = q->blk_trace;
 905 +       struct blk_io_trace_remap r;
 906 +
 907 +       if (likely(!bt))
 908 +               return;
 909 +
 910 +       r.device = cpu_to_be32(dev);
 911 +       r.sector = cpu_to_be64(to);
 912 +
 913 +       __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 914 +}
 915 +
 916 +#else /* !CONFIG_BLK_DEV_IO_TRACE */
 917 +#define blk_start_trace(bdev, arg)             (-EINVAL)
 918 +#define blk_stop_trace(bdev)                   (-EINVAL)
 919 +#define blk_cleanup_trace(bt)                  do { } while (0)
 920 +#define blk_add_trace_rq(q, rq, what)          do { } while (0)
 921 +#define blk_add_trace_bio(q, rq, what)         do { } while (0)
 922 +#define blk_add_trace_generic(q, rq, rw, what) do { } while (0)
 923 +#define blk_add_trace_pdu_int(q, what, bio, pdu)       do { } while (0)
 924 +#define blk_add_trace_remap(q, bio, dev, f, t) do {} while (0)
 925 +#endif /* CONFIG_BLK_DEV_IO_TRACE */
 926 +
 927 +#endif
 928 diff --git a/include/linux/compat_ioctl.h b/include/linux/compat_ioctl.h
 929 index 8fad50f..5bed09a 100644
 930 --- a/include/linux/compat_ioctl.h
 931 +++ b/include/linux/compat_ioctl.h
 932 @@ -97,6 +97,8 @@ COMPATIBLE_IOCTL(BLKRRPART)
 933  COMPATIBLE_IOCTL(BLKFLSBUF)
 934  COMPATIBLE_IOCTL(BLKSECTSET)
 935  COMPATIBLE_IOCTL(BLKSSZGET)
 936 +COMPATIBLE_IOCTL(BLKSTARTTRACE)
 937 +COMPATIBLE_IOCTL(BLKSTOPTRACE)
 938  ULONG_IOCTL(BLKRASET)
 939  ULONG_IOCTL(BLKFRASET)
 940  /* RAID */
 941 diff --git a/include/linux/fs.h b/include/linux/fs.h
 942 index b77f260..5575284 100644
 943 --- a/include/linux/fs.h
 944 +++ b/include/linux/fs.h
 945 @@ -196,6 +196,8 @@ extern int dir_notify_enable;
 946  #define BLKBSZGET  _IOR(0x12,112,size_t)
 947  #define BLKBSZSET  _IOW(0x12,113,size_t)
 948  #define BLKGETSIZE64 _IOR(0x12,114,size_t)     /* return device size in bytes (u64 *arg) */
 949 +#define BLKSTARTTRACE _IOWR(0x12,115,struct blk_user_trace_setup)
 950 +#define BLKSTOPTRACE _IO(0x12,116)
 951
 952  #define BMAP_IOCTL 1           /* obsolete - kept for compatibility */
 953  #define FIBMAP    _IO(0x00,1)  /* bmap access */
 954 diff --git a/mm/highmem.c b/mm/highmem.c
 955 index ce2e7e8..d0ea1ee 100644
 956 --- a/mm/highmem.c
 957 +++ b/mm/highmem.c
 958 @@ -26,6 +26,7 @@
 959  #include <linux/init.h>
 960  #include <linux/hash.h>
 961  #include <linux/highmem.h>
 962 +#include <linux/blktrace_api.h>
 963  #include <asm/tlbflush.h>
 964
 965  static mempool_t *page_pool, *isa_page_pool;
 966 @@ -483,6 +484,8 @@ void blk_queue_bounce(request_queue_t *q
 967                 pool = isa_page_pool;
 968         }
 969
 970 +       blk_add_trace_bio(q, *bio_orig, BLK_TA_BOUNCE);
 971 +
 972         /*
 973          * slow path
 974          */