drivers/md/dm-region-hash.c

   1 /*
   2  * Copyright (C) 2003 Sistina Software Limited.
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include <linux/dm-dirty-log.h>
   9 #include <linux/dm-region-hash.h>
  10
  11 #include <linux/ctype.h>
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/vmalloc.h>
  16
  17 #include "dm.h"
  18
  19 #define DM_MSG_PREFIX   "region hash"
  20
  21 /*-----------------------------------------------------------------
  22  * Region hash
  23  *
  24  * The mirror splits itself up into discrete regions.  Each
  25  * region can be in one of three states: clean, dirty,
  26  * nosync.  There is no need to put clean regions in the hash.
  27  *
  28  * In addition to being present in the hash table a region _may_
  29  * be present on one of three lists.
  30  *
  31  *   clean_regions: Regions on this list have no io pending to
  32  *   them, they are in sync, we are no longer interested in them,
  33  *   they are dull.  dm_rh_update_states() will remove them from the
  34  *   hash table.
  35  *
  36  *   quiesced_regions: These regions have been spun down, ready
  37  *   for recovery.  rh_recovery_start() will remove regions from
  38  *   this list and hand them to kmirrord, which will schedule the
  39  *   recovery io with kcopyd.
  40  *
  41  *   recovered_regions: Regions that kcopyd has successfully
  42  *   recovered.  dm_rh_update_states() will now schedule any delayed
  43  *   io, up the recovery_count, and remove the region from the
  44  *   hash.
  45  *
  46  * There are 2 locks:
  47  *   A rw spin lock 'hash_lock' protects just the hash table,
  48  *   this is never held in write mode from interrupt context,
  49  *   which I believe means that we only have to disable irqs when
  50  *   doing a write lock.
  51  *
  52  *   An ordinary spin lock 'region_lock' that protects the three
  53  *   lists in the region_hash, with the 'state', 'list' and
  54  *   'delayed_bios' fields of the regions.  This is used from irq
  55  *   context, so all other uses will have to suspend local irqs.
  56  *---------------------------------------------------------------*/
  57 struct dm_region_hash {
  58         uint32_t region_size;
  59         unsigned region_shift;
  60
  61         /* holds persistent region state */
  62         struct dm_dirty_log *log;
  63
  64         /* hash table */
  65         rwlock_t hash_lock;
  66         mempool_t *region_pool;
  67         unsigned mask;
  68         unsigned nr_buckets;
  69         unsigned prime;
  70         unsigned shift;
  71         struct list_head *buckets;
  72
  73         unsigned max_recovery; /* Max # of regions to recover in parallel */
  74
  75         spinlock_t region_lock;
  76         atomic_t recovery_in_flight;
  77         struct semaphore recovery_count;
  78         struct list_head clean_regions;
  79         struct list_head quiesced_regions;
  80         struct list_head recovered_regions;
  81         struct list_head failed_recovered_regions;
  82
  83         /*
  84          * If there was a flush failure no regions can be marked clean.
  85          */
  86         int flush_failure;
  87
  88         void *context;
  89         sector_t target_begin;
  90
  91         /* Callback function to schedule bios writes */
  92         void (*dispatch_bios)(void *context, struct bio_list *bios);
  93
  94         /* Callback function to wakeup callers worker thread. */
  95         void (*wakeup_workers)(void *context);
  96
  97         /* Callback function to wakeup callers recovery waiters. */
  98         void (*wakeup_all_recovery_waiters)(void *context);
  99 };
 100
 101 struct dm_region {
 102         struct dm_region_hash *rh;      /* FIXME: can we get rid of this ? */
 103         region_t key;
 104         int state;
 105
 106         struct list_head hash_list;
 107         struct list_head list;
 108
 109         atomic_t pending;
 110         struct bio_list delayed_bios;
 111 };
 112
 113 /*
 114  * Conversion fns
 115  */
 116 static region_t dm_rh_sector_to_region(struct dm_region_hash *rh, sector_t sector)
 117 {
 118         return sector >> rh->region_shift;
 119 }
 120
 121 sector_t dm_rh_region_to_sector(struct dm_region_hash *rh, region_t region)
 122 {
 123         return region << rh->region_shift;
 124 }
 125 EXPORT_SYMBOL_GPL(dm_rh_region_to_sector);
 126
 127 region_t dm_rh_bio_to_region(struct dm_region_hash *rh, struct bio *bio)
 128 {
 129         return dm_rh_sector_to_region(rh, bio->bi_sector - rh->target_begin);
 130 }
 131 EXPORT_SYMBOL_GPL(dm_rh_bio_to_region);
 132
 133 void *dm_rh_region_context(struct dm_region *reg)
 134 {
 135         return reg->rh->context;
 136 }
 137 EXPORT_SYMBOL_GPL(dm_rh_region_context);
 138
 139 region_t dm_rh_get_region_key(struct dm_region *reg)
 140 {
 141         return reg->key;
 142 }
 143 EXPORT_SYMBOL_GPL(dm_rh_get_region_key);
 144
 145 sector_t dm_rh_get_region_size(struct dm_region_hash *rh)
 146 {
 147         return rh->region_size;
 148 }
 149 EXPORT_SYMBOL_GPL(dm_rh_get_region_size);
 150
 151 /*
 152  * FIXME: shall we pass in a structure instead of all these args to
 153  * dm_region_hash_create()????
 154  */
 155 #define RH_HASH_MULT 2654435387U
 156 #define RH_HASH_SHIFT 12
 157
 158 #define MIN_REGIONS 64
 159 struct dm_region_hash *dm_region_hash_create(
 160                 void *context, void (*dispatch_bios)(void *context,
 161                                                      struct bio_list *bios),
 162                 void (*wakeup_workers)(void *context),
 163                 void (*wakeup_all_recovery_waiters)(void *context),
 164                 sector_t target_begin, unsigned max_recovery,
 165                 struct dm_dirty_log *log, uint32_t region_size,
 166                 region_t nr_regions)
 167 {
 168         struct dm_region_hash *rh;
 169         unsigned nr_buckets, max_buckets;
 170         size_t i;
 171
 172         /*
 173          * Calculate a suitable number of buckets for our hash
 174          * table.
 175          */
 176         max_buckets = nr_regions >> 6;
 177         for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
 178                 ;
 179         nr_buckets >>= 1;
 180
 181         rh = kmalloc(sizeof(*rh), GFP_KERNEL);
 182         if (!rh) {
 183                 DMERR("unable to allocate region hash memory");
 184                 return ERR_PTR(-ENOMEM);
 185         }
 186
 187         rh->context = context;
 188         rh->dispatch_bios = dispatch_bios;
 189         rh->wakeup_workers = wakeup_workers;
 190         rh->wakeup_all_recovery_waiters = wakeup_all_recovery_waiters;
 191         rh->target_begin = target_begin;
 192         rh->max_recovery = max_recovery;
 193         rh->log = log;
 194         rh->region_size = region_size;
 195         rh->region_shift = ffs(region_size) - 1;
 196         rwlock_init(&rh->hash_lock);
 197         rh->mask = nr_buckets - 1;
 198         rh->nr_buckets = nr_buckets;
 199
 200         rh->shift = RH_HASH_SHIFT;
 201         rh->prime = RH_HASH_MULT;
 202
 203         rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
 204         if (!rh->buckets) {
 205                 DMERR("unable to allocate region hash bucket memory");
 206                 kfree(rh);
 207                 return ERR_PTR(-ENOMEM);
 208         }
 209
 210         for (i = 0; i < nr_buckets; i++)
 211                 INIT_LIST_HEAD(rh->buckets + i);
 212
 213         spin_lock_init(&rh->region_lock);
 214         sema_init(&rh->recovery_count, 0);
 215         atomic_set(&rh->recovery_in_flight, 0);
 216         INIT_LIST_HEAD(&rh->clean_regions);
 217         INIT_LIST_HEAD(&rh->quiesced_regions);
 218         INIT_LIST_HEAD(&rh->recovered_regions);
 219         INIT_LIST_HEAD(&rh->failed_recovered_regions);
 220         rh->flush_failure = 0;
 221
 222         rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
 223                                                       sizeof(struct dm_region));
 224         if (!rh->region_pool) {
 225                 vfree(rh->buckets);
 226                 kfree(rh);
 227                 rh = ERR_PTR(-ENOMEM);
 228         }
 229
 230         return rh;
 231 }
 232 EXPORT_SYMBOL_GPL(dm_region_hash_create);
 233
 234 void dm_region_hash_destroy(struct dm_region_hash *rh)
 235 {
 236         unsigned h;
 237         struct dm_region *reg, *nreg;
 238
 239         BUG_ON(!list_empty(&rh->quiesced_regions));
 240         for (h = 0; h < rh->nr_buckets; h++) {
 241                 list_for_each_entry_safe(reg, nreg, rh->buckets + h,
 242                                          hash_list) {
 243                         BUG_ON(atomic_read(&reg->pending));
 244                         mempool_free(reg, rh->region_pool);
 245                 }
 246         }
 247
 248         if (rh->log)
 249                 dm_dirty_log_destroy(rh->log);
 250
 251         if (rh->region_pool)
 252                 mempool_destroy(rh->region_pool);
 253
 254         vfree(rh->buckets);
 255         kfree(rh);
 256 }
 257 EXPORT_SYMBOL_GPL(dm_region_hash_destroy);
 258
 259 struct dm_dirty_log *dm_rh_dirty_log(struct dm_region_hash *rh)
 260 {
 261         return rh->log;
 262 }
 263 EXPORT_SYMBOL_GPL(dm_rh_dirty_log);
 264
 265 static unsigned rh_hash(struct dm_region_hash *rh, region_t region)
 266 {
 267         return (unsigned) ((region * rh->prime) >> rh->shift) & rh->mask;
 268 }
 269
 270 static struct dm_region *__rh_lookup(struct dm_region_hash *rh, region_t region)
 271 {
 272         struct dm_region *reg;
 273         struct list_head *bucket = rh->buckets + rh_hash(rh, region);
 274
 275         list_for_each_entry(reg, bucket, hash_list)
 276                 if (reg->key == region)
 277                         return reg;
 278
 279         return NULL;
 280 }
 281
 282 static void __rh_insert(struct dm_region_hash *rh, struct dm_region *reg)
 283 {
 284         list_add(&reg->hash_list, rh->buckets + rh_hash(rh, reg->key));
 285 }
 286
 287 static struct dm_region *__rh_alloc(struct dm_region_hash *rh, region_t region)
 288 {
 289         struct dm_region *reg, *nreg;
 290
 291         nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
 292         if (unlikely(!nreg))
 293                 nreg = kmalloc(sizeof(*nreg), GFP_NOIO | __GFP_NOFAIL);
 294
 295         nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
 296                       DM_RH_CLEAN : DM_RH_NOSYNC;
 297         nreg->rh = rh;
 298         nreg->key = region;
 299         INIT_LIST_HEAD(&nreg->list);
 300         atomic_set(&nreg->pending, 0);
 301         bio_list_init(&nreg->delayed_bios);
 302
 303         write_lock_irq(&rh->hash_lock);
 304         reg = __rh_lookup(rh, region);
 305         if (reg)
 306                 /* We lost the race. */
 307                 mempool_free(nreg, rh->region_pool);
 308         else {
 309                 __rh_insert(rh, nreg);
 310                 if (nreg->state == DM_RH_CLEAN) {
 311                         spin_lock(&rh->region_lock);
 312                         list_add(&nreg->list, &rh->clean_regions);
 313                         spin_unlock(&rh->region_lock);
 314                 }
 315
 316                 reg = nreg;
 317         }
 318         write_unlock_irq(&rh->hash_lock);
 319
 320         return reg;
 321 }
 322
 323 static struct dm_region *__rh_find(struct dm_region_hash *rh, region_t region)
 324 {
 325         struct dm_region *reg;
 326
 327         reg = __rh_lookup(rh, region);
 328         if (!reg) {
 329                 read_unlock(&rh->hash_lock);
 330                 reg = __rh_alloc(rh, region);
 331                 read_lock(&rh->hash_lock);
 332         }
 333
 334         return reg;
 335 }
 336
 337 int dm_rh_get_state(struct dm_region_hash *rh, region_t region, int may_block)
 338 {
 339         int r;
 340         struct dm_region *reg;
 341
 342         read_lock(&rh->hash_lock);
 343         reg = __rh_lookup(rh, region);
 344         read_unlock(&rh->hash_lock);
 345
 346         if (reg)
 347                 return reg->state;
 348
 349         /*
 350          * The region wasn't in the hash, so we fall back to the
 351          * dirty log.
 352          */
 353         r = rh->log->type->in_sync(rh->log, region, may_block);
 354
 355         /*
 356          * Any error from the dirty log (eg. -EWOULDBLOCK) gets
 357          * taken as a DM_RH_NOSYNC
 358          */
 359         return r == 1 ? DM_RH_CLEAN : DM_RH_NOSYNC;
 360 }
 361 EXPORT_SYMBOL_GPL(dm_rh_get_state);
 362
 363 static void complete_resync_work(struct dm_region *reg, int success)
 364 {
 365         struct dm_region_hash *rh = reg->rh;
 366
 367         rh->log->type->set_region_sync(rh->log, reg->key, success);
 368
 369         /*
 370          * Dispatch the bios before we call 'wake_up_all'.
 371          * This is important because if we are suspending,
 372          * we want to know that recovery is complete and
 373          * the work queue is flushed.  If we wake_up_all
 374          * before we dispatch_bios (queue bios and call wake()),
 375          * then we risk suspending before the work queue
 376          * has been properly flushed.
 377          */
 378         rh->dispatch_bios(rh->context, &reg->delayed_bios);
 379         if (atomic_dec_and_test(&rh->recovery_in_flight))
 380                 rh->wakeup_all_recovery_waiters(rh->context);
 381         up(&rh->recovery_count);
 382 }
 383
 384 /* dm_rh_mark_nosync
 385  * @ms
 386  * @bio
 387  *
 388  * The bio was written on some mirror(s) but failed on other mirror(s).
 389  * We can successfully endio the bio but should avoid the region being
 390  * marked clean by setting the state DM_RH_NOSYNC.
 391  *
 392  * This function is _not_ safe in interrupt context!
 393  */
 394 void dm_rh_mark_nosync(struct dm_region_hash *rh, struct bio *bio)
 395 {
 396         unsigned long flags;
 397         struct dm_dirty_log *log = rh->log;
 398         struct dm_region *reg;
 399         region_t region = dm_rh_bio_to_region(rh, bio);
 400         int recovering = 0;
 401
 402         if (bio->bi_rw & REQ_FLUSH) {
 403                 rh->flush_failure = 1;
 404                 return;
 405         }
 406
 407         /* We must inform the log that the sync count has changed. */
 408         log->type->set_region_sync(log, region, 0);
 409
 410         read_lock(&rh->hash_lock);
 411         reg = __rh_find(rh, region);
 412         read_unlock(&rh->hash_lock);
 413
 414         /* region hash entry should exist because write was in-flight */
 415         BUG_ON(!reg);
 416         BUG_ON(!list_empty(&reg->list));
 417
 418         spin_lock_irqsave(&rh->region_lock, flags);
 419         /*
 420          * Possible cases:
 421          *   1) DM_RH_DIRTY
 422          *   2) DM_RH_NOSYNC: was dirty, other preceding writes failed
 423          *   3) DM_RH_RECOVERING: flushing pending writes
 424          * Either case, the region should have not been connected to list.
 425          */
 426         recovering = (reg->state == DM_RH_RECOVERING);
 427         reg->state = DM_RH_NOSYNC;
 428         BUG_ON(!list_empty(&reg->list));
 429         spin_unlock_irqrestore(&rh->region_lock, flags);
 430
 431         if (recovering)
 432                 complete_resync_work(reg, 0);
 433 }
 434 EXPORT_SYMBOL_GPL(dm_rh_mark_nosync);
 435
 436 void dm_rh_update_states(struct dm_region_hash *rh, int errors_handled)
 437 {
 438         struct dm_region *reg, *next;
 439
 440         LIST_HEAD(clean);
 441         LIST_HEAD(recovered);
 442         LIST_HEAD(failed_recovered);
 443
 444         /*
 445          * Quickly grab the lists.
 446          */
 447         write_lock_irq(&rh->hash_lock);
 448         spin_lock(&rh->region_lock);
 449         if (!list_empty(&rh->clean_regions)) {
 450                 list_splice_init(&rh->clean_regions, &clean);
 451
 452                 list_for_each_entry(reg, &clean, list)
 453                         list_del(&reg->hash_list);
 454         }
 455
 456         if (!list_empty(&rh->recovered_regions)) {
 457                 list_splice_init(&rh->recovered_regions, &recovered);
 458
 459                 list_for_each_entry(reg, &recovered, list)
 460                         list_del(&reg->hash_list);
 461         }
 462
 463         if (!list_empty(&rh->failed_recovered_regions)) {
 464                 list_splice_init(&rh->failed_recovered_regions,
 465                                  &failed_recovered);
 466
 467                 list_for_each_entry(reg, &failed_recovered, list)
 468                         list_del(&reg->hash_list);
 469         }
 470
 471         spin_unlock(&rh->region_lock);
 472         write_unlock_irq(&rh->hash_lock);
 473
 474         /*
 475          * All the regions on the recovered and clean lists have
 476          * now been pulled out of the system, so no need to do
 477          * any more locking.
 478          */
 479         list_for_each_entry_safe(reg, next, &recovered, list) {
 480                 rh->log->type->clear_region(rh->log, reg->key);
 481                 complete_resync_work(reg, 1);
 482                 mempool_free(reg, rh->region_pool);
 483         }
 484
 485         list_for_each_entry_safe(reg, next, &failed_recovered, list) {
 486                 complete_resync_work(reg, errors_handled ? 0 : 1);
 487                 mempool_free(reg, rh->region_pool);
 488         }
 489
 490         list_for_each_entry_safe(reg, next, &clean, list) {
 491                 rh->log->type->clear_region(rh->log, reg->key);
 492                 mempool_free(reg, rh->region_pool);
 493         }
 494
 495         rh->log->type->flush(rh->log);
 496 }
 497 EXPORT_SYMBOL_GPL(dm_rh_update_states);
 498
 499 static void rh_inc(struct dm_region_hash *rh, region_t region)
 500 {
 501         struct dm_region *reg;
 502
 503         read_lock(&rh->hash_lock);
 504         reg = __rh_find(rh, region);
 505
 506         spin_lock_irq(&rh->region_lock);
 507         atomic_inc(&reg->pending);
 508
 509         if (reg->state == DM_RH_CLEAN) {
 510                 reg->state = DM_RH_DIRTY;
 511                 list_del_init(&reg->list);      /* take off the clean list */
 512                 spin_unlock_irq(&rh->region_lock);
 513
 514                 rh->log->type->mark_region(rh->log, reg->key);
 515         } else
 516                 spin_unlock_irq(&rh->region_lock);
 517
 518
 519         read_unlock(&rh->hash_lock);
 520 }
 521
 522 void dm_rh_inc_pending(struct dm_region_hash *rh, struct bio_list *bios)
 523 {
 524         struct bio *bio;
 525
 526         for (bio = bios->head; bio; bio = bio->bi_next) {
 527                 if (bio->bi_rw & REQ_FLUSH)
 528                         continue;
 529                 rh_inc(rh, dm_rh_bio_to_region(rh, bio));
 530         }
 531 }
 532 EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
 533
 534 void dm_rh_dec(struct dm_region_hash *rh, region_t region)
 535 {
 536         unsigned long flags;
 537         struct dm_region *reg;
 538         int should_wake = 0;
 539
 540         read_lock(&rh->hash_lock);
 541         reg = __rh_lookup(rh, region);
 542         read_unlock(&rh->hash_lock);
 543
 544         spin_lock_irqsave(&rh->region_lock, flags);
 545         if (atomic_dec_and_test(&reg->pending)) {
 546                 /*
 547                  * There is no pending I/O for this region.
 548                  * We can move the region to corresponding list for next action.
 549                  * At this point, the region is not yet connected to any list.
 550                  *
 551                  * If the state is DM_RH_NOSYNC, the region should be kept off
 552                  * from clean list.
 553                  * The hash entry for DM_RH_NOSYNC will remain in memory
 554                  * until the region is recovered or the map is reloaded.
 555                  */
 556
 557                 /* do nothing for DM_RH_NOSYNC */
 558                 if (unlikely(rh->flush_failure)) {
 559                         /*
 560                          * If a write flush failed some time ago, we
 561                          * don't know whether or not this write made it
 562                          * to the disk, so we must resync the device.
 563                          */
 564                         reg->state = DM_RH_NOSYNC;
 565                 } else if (reg->state == DM_RH_RECOVERING) {
 566                         list_add_tail(&reg->list, &rh->quiesced_regions);
 567                 } else if (reg->state == DM_RH_DIRTY) {
 568                         reg->state = DM_RH_CLEAN;
 569                         list_add(&reg->list, &rh->clean_regions);
 570                 }
 571                 should_wake = 1;
 572         }
 573         spin_unlock_irqrestore(&rh->region_lock, flags);
 574
 575         if (should_wake)
 576                 rh->wakeup_workers(rh->context);
 577 }
 578 EXPORT_SYMBOL_GPL(dm_rh_dec);
 579
 580 /*
 581  * Starts quiescing a region in preparation for recovery.
 582  */
 583 static int __rh_recovery_prepare(struct dm_region_hash *rh)
 584 {
 585         int r;
 586         region_t region;
 587         struct dm_region *reg;
 588
 589         /*
 590          * Ask the dirty log what's next.
 591          */
 592         r = rh->log->type->get_resync_work(rh->log, &region);
 593         if (r <= 0)
 594                 return r;
 595
 596         /*
 597          * Get this region, and start it quiescing by setting the
 598          * recovering flag.
 599          */
 600         read_lock(&rh->hash_lock);
 601         reg = __rh_find(rh, region);
 602         read_unlock(&rh->hash_lock);
 603
 604         spin_lock_irq(&rh->region_lock);
 605         reg->state = DM_RH_RECOVERING;
 606
 607         /* Already quiesced ? */
 608         if (atomic_read(&reg->pending))
 609                 list_del_init(&reg->list);
 610         else
 611                 list_move(&reg->list, &rh->quiesced_regions);
 612
 613         spin_unlock_irq(&rh->region_lock);
 614
 615         return 1;
 616 }
 617
 618 void dm_rh_recovery_prepare(struct dm_region_hash *rh)
 619 {
 620         /* Extra reference to avoid race with dm_rh_stop_recovery */
 621         atomic_inc(&rh->recovery_in_flight);
 622
 623         while (!down_trylock(&rh->recovery_count)) {
 624                 atomic_inc(&rh->recovery_in_flight);
 625                 if (__rh_recovery_prepare(rh) <= 0) {
 626                         atomic_dec(&rh->recovery_in_flight);
 627                         up(&rh->recovery_count);
 628                         break;
 629                 }
 630         }
 631
 632         /* Drop the extra reference */
 633         if (atomic_dec_and_test(&rh->recovery_in_flight))
 634                 rh->wakeup_all_recovery_waiters(rh->context);
 635 }
 636 EXPORT_SYMBOL_GPL(dm_rh_recovery_prepare);
 637
 638 /*
 639  * Returns any quiesced regions.
 640  */
 641 struct dm_region *dm_rh_recovery_start(struct dm_region_hash *rh)
 642 {
 643         struct dm_region *reg = NULL;
 644
 645         spin_lock_irq(&rh->region_lock);
 646         if (!list_empty(&rh->quiesced_regions)) {
 647                 reg = list_entry(rh->quiesced_regions.next,
 648                                  struct dm_region, list);
 649                 list_del_init(&reg->list);  /* remove from the quiesced list */
 650         }
 651         spin_unlock_irq(&rh->region_lock);
 652
 653         return reg;
 654 }
 655 EXPORT_SYMBOL_GPL(dm_rh_recovery_start);
 656
 657 void dm_rh_recovery_end(struct dm_region *reg, int success)
 658 {
 659         struct dm_region_hash *rh = reg->rh;
 660
 661         spin_lock_irq(&rh->region_lock);
 662         if (success)
 663                 list_add(&reg->list, &reg->rh->recovered_regions);
 664         else
 665                 list_add(&reg->list, &reg->rh->failed_recovered_regions);
 666
 667         spin_unlock_irq(&rh->region_lock);
 668
 669         rh->wakeup_workers(rh->context);
 670 }
 671 EXPORT_SYMBOL_GPL(dm_rh_recovery_end);
 672
 673 /* Return recovery in flight count. */
 674 int dm_rh_recovery_in_flight(struct dm_region_hash *rh)
 675 {
 676         return atomic_read(&rh->recovery_in_flight);
 677 }
 678 EXPORT_SYMBOL_GPL(dm_rh_recovery_in_flight);
 679
 680 int dm_rh_flush(struct dm_region_hash *rh)
 681 {
 682         return rh->log->type->flush(rh->log);
 683 }
 684 EXPORT_SYMBOL_GPL(dm_rh_flush);
 685
 686 void dm_rh_delay(struct dm_region_hash *rh, struct bio *bio)
 687 {
 688         struct dm_region *reg;
 689
 690         read_lock(&rh->hash_lock);
 691         reg = __rh_find(rh, dm_rh_bio_to_region(rh, bio));
 692         bio_list_add(&reg->delayed_bios, bio);
 693         read_unlock(&rh->hash_lock);
 694 }
 695 EXPORT_SYMBOL_GPL(dm_rh_delay);
 696
 697 void dm_rh_stop_recovery(struct dm_region_hash *rh)
 698 {
 699         int i;
 700
 701         /* wait for any recovering regions */
 702         for (i = 0; i < rh->max_recovery; i++)
 703                 down(&rh->recovery_count);
 704 }
 705 EXPORT_SYMBOL_GPL(dm_rh_stop_recovery);
 706
 707 void dm_rh_start_recovery(struct dm_region_hash *rh)
 708 {
 709         int i;
 710
 711         for (i = 0; i < rh->max_recovery; i++)
 712                 up(&rh->recovery_count);
 713
 714         rh->wakeup_workers(rh->context);
 715 }
 716 EXPORT_SYMBOL_GPL(dm_rh_start_recovery);
 717
 718 MODULE_DESCRIPTION(DM_NAME " region hash");
 719 MODULE_AUTHOR("Joe Thornber/Heinz Mauelshagen <dm-devel@redhat.com>");
 720 MODULE_LICENSE("GPL");