linux/drivers/md/dm-snap.c
<<
>>
Prefs
   1/*
   2 * dm-snapshot.c
   3 *
   4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
   5 *
   6 * This file is released under the GPL.
   7 */
   8
   9#include <linux/blkdev.h>
  10#include <linux/device-mapper.h>
  11#include <linux/delay.h>
  12#include <linux/fs.h>
  13#include <linux/init.h>
  14#include <linux/kdev_t.h>
  15#include <linux/list.h>
  16#include <linux/mempool.h>
  17#include <linux/module.h>
  18#include <linux/slab.h>
  19#include <linux/vmalloc.h>
  20#include <linux/log2.h>
  21#include <linux/dm-kcopyd.h>
  22
  23#include "dm-exception-store.h"
  24
  25#define DM_MSG_PREFIX "snapshots"
  26
  27static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
  28
  29#define dm_target_is_snapshot_merge(ti) \
  30        ((ti)->type->name == dm_snapshot_merge_target_name)
  31
  32/*
  33 * The size of the mempool used to track chunks in use.
  34 */
  35#define MIN_IOS 256
  36
  37#define DM_TRACKED_CHUNK_HASH_SIZE      16
  38#define DM_TRACKED_CHUNK_HASH(x)        ((unsigned long)(x) & \
  39                                         (DM_TRACKED_CHUNK_HASH_SIZE - 1))
  40
  41struct dm_exception_table {
  42        uint32_t hash_mask;
  43        unsigned hash_shift;
  44        struct list_head *table;
  45};
  46
  47struct dm_snapshot {
  48        struct rw_semaphore lock;
  49
  50        struct dm_dev *origin;
  51        struct dm_dev *cow;
  52
  53        struct dm_target *ti;
  54
  55        /* List of snapshots per Origin */
  56        struct list_head list;
  57
  58        /*
  59         * You can't use a snapshot if this is 0 (e.g. if full).
  60         * A snapshot-merge target never clears this.
  61         */
  62        int valid;
  63
  64        /* Origin writes don't trigger exceptions until this is set */
  65        int active;
  66
  67        atomic_t pending_exceptions_count;
  68
  69        mempool_t *pending_pool;
  70
  71        struct dm_exception_table pending;
  72        struct dm_exception_table complete;
  73
  74        /*
  75         * pe_lock protects all pending_exception operations and access
  76         * as well as the snapshot_bios list.
  77         */
  78        spinlock_t pe_lock;
  79
  80        /* Chunks with outstanding reads */
  81        spinlock_t tracked_chunk_lock;
  82        struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
  83
  84        /* The on disk metadata handler */
  85        struct dm_exception_store *store;
  86
  87        struct dm_kcopyd_client *kcopyd_client;
  88
  89        /* Wait for events based on state_bits */
  90        unsigned long state_bits;
  91
  92        /* Range of chunks currently being merged. */
  93        chunk_t first_merging_chunk;
  94        int num_merging_chunks;
  95
  96        /*
  97         * The merge operation failed if this flag is set.
  98         * Failure modes are handled as follows:
  99         * - I/O error reading the header
 100         *      => don't load the target; abort.
 101         * - Header does not have "valid" flag set
 102         *      => use the origin; forget about the snapshot.
 103         * - I/O error when reading exceptions
 104         *      => don't load the target; abort.
 105         *         (We can't use the intermediate origin state.)
 106         * - I/O error while merging
 107         *      => stop merging; set merge_failed; process I/O normally.
 108         */
 109        int merge_failed;
 110
 111        /*
 112         * Incoming bios that overlap with chunks being merged must wait
 113         * for them to be committed.
 114         */
 115        struct bio_list bios_queued_during_merge;
 116};
 117
 118/*
 119 * state_bits:
 120 *   RUNNING_MERGE  - Merge operation is in progress.
 121 *   SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
 122 *                    cleared afterwards.
 123 */
 124#define RUNNING_MERGE          0
 125#define SHUTDOWN_MERGE         1
 126
 127DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
 128                "A percentage of time allocated for copy on write");
 129
 130struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
 131{
 132        return s->origin;
 133}
 134EXPORT_SYMBOL(dm_snap_origin);
 135
 136struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
 137{
 138        return s->cow;
 139}
 140EXPORT_SYMBOL(dm_snap_cow);
 141
 142static sector_t chunk_to_sector(struct dm_exception_store *store,
 143                                chunk_t chunk)
 144{
 145        return chunk << store->chunk_shift;
 146}
 147
 148static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
 149{
 150        /*
 151         * There is only ever one instance of a particular block
 152         * device so we can compare pointers safely.
 153         */
 154        return lhs == rhs;
 155}
 156
 157struct dm_snap_pending_exception {
 158        struct dm_exception e;
 159
 160        /*
 161         * Origin buffers waiting for this to complete are held
 162         * in a bio list
 163         */
 164        struct bio_list origin_bios;
 165        struct bio_list snapshot_bios;
 166
 167        /* Pointer back to snapshot context */
 168        struct dm_snapshot *snap;
 169
 170        /*
 171         * 1 indicates the exception has already been sent to
 172         * kcopyd.
 173         */
 174        int started;
 175
 176        /*
 177         * For writing a complete chunk, bypassing the copy.
 178         */
 179        struct bio *full_bio;
 180        bio_end_io_t *full_bio_end_io;
 181        void *full_bio_private;
 182};
 183
 184/*
 185 * Hash table mapping origin volumes to lists of snapshots and
 186 * a lock to protect it
 187 */
 188static struct kmem_cache *exception_cache;
 189static struct kmem_cache *pending_cache;
 190
 191struct dm_snap_tracked_chunk {
 192        struct hlist_node node;
 193        chunk_t chunk;
 194};
 195
 196static void init_tracked_chunk(struct bio *bio)
 197{
 198        struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 199        INIT_HLIST_NODE(&c->node);
 200}
 201
 202static bool is_bio_tracked(struct bio *bio)
 203{
 204        struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 205        return !hlist_unhashed(&c->node);
 206}
 207
 208static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
 209{
 210        struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 211
 212        c->chunk = chunk;
 213
 214        spin_lock_irq(&s->tracked_chunk_lock);
 215        hlist_add_head(&c->node,
 216                       &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
 217        spin_unlock_irq(&s->tracked_chunk_lock);
 218}
 219
 220static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
 221{
 222        struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 223        unsigned long flags;
 224
 225        spin_lock_irqsave(&s->tracked_chunk_lock, flags);
 226        hlist_del(&c->node);
 227        spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
 228}
 229
 230static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
 231{
 232        struct dm_snap_tracked_chunk *c;
 233        int found = 0;
 234
 235        spin_lock_irq(&s->tracked_chunk_lock);
 236
 237        hlist_for_each_entry(c,
 238            &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
 239                if (c->chunk == chunk) {
 240                        found = 1;
 241                        break;
 242                }
 243        }
 244
 245        spin_unlock_irq(&s->tracked_chunk_lock);
 246
 247        return found;
 248}
 249
 250/*
 251 * This conflicting I/O is extremely improbable in the caller,
 252 * so msleep(1) is sufficient and there is no need for a wait queue.
 253 */
 254static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
 255{
 256        while (__chunk_is_tracked(s, chunk))
 257                msleep(1);
 258}
 259
 260/*
 261 * One of these per registered origin, held in the snapshot_origins hash
 262 */
 263struct origin {
 264        /* The origin device */
 265        struct block_device *bdev;
 266
 267        struct list_head hash_list;
 268
 269        /* List of snapshots for this origin */
 270        struct list_head snapshots;
 271};
 272
 273/*
 274 * Size of the hash table for origin volumes. If we make this
 275 * the size of the minors list then it should be nearly perfect
 276 */
 277#define ORIGIN_HASH_SIZE 256
 278#define ORIGIN_MASK      0xFF
 279static struct list_head *_origins;
 280static struct rw_semaphore _origins_lock;
 281
 282static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
 283static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
 284static uint64_t _pending_exceptions_done_count;
 285
 286static int init_origin_hash(void)
 287{
 288        int i;
 289
 290        _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
 291                           GFP_KERNEL);
 292        if (!_origins) {
 293                DMERR("unable to allocate memory");
 294                return -ENOMEM;
 295        }
 296
 297        for (i = 0; i < ORIGIN_HASH_SIZE; i++)
 298                INIT_LIST_HEAD(_origins + i);
 299        init_rwsem(&_origins_lock);
 300
 301        return 0;
 302}
 303
 304static void exit_origin_hash(void)
 305{
 306        kfree(_origins);
 307}
 308
 309static unsigned origin_hash(struct block_device *bdev)
 310{
 311        return bdev->bd_dev & ORIGIN_MASK;
 312}
 313
 314static struct origin *__lookup_origin(struct block_device *origin)
 315{
 316        struct list_head *ol;
 317        struct origin *o;
 318
 319        ol = &_origins[origin_hash(origin)];
 320        list_for_each_entry (o, ol, hash_list)
 321                if (bdev_equal(o->bdev, origin))
 322                        return o;
 323
 324        return NULL;
 325}
 326
 327static void __insert_origin(struct origin *o)
 328{
 329        struct list_head *sl = &_origins[origin_hash(o->bdev)];
 330        list_add_tail(&o->hash_list, sl);
 331}
 332
 333/*
 334 * _origins_lock must be held when calling this function.
 335 * Returns number of snapshots registered using the supplied cow device, plus:
 336 * snap_src - a snapshot suitable for use as a source of exception handover
 337 * snap_dest - a snapshot capable of receiving exception handover.
 338 * snap_merge - an existing snapshot-merge target linked to the same origin.
 339 *   There can be at most one snapshot-merge target. The parameter is optional.
 340 *
 341 * Possible return values and states of snap_src and snap_dest.
 342 *   0: NULL, NULL  - first new snapshot
 343 *   1: snap_src, NULL - normal snapshot
 344 *   2: snap_src, snap_dest  - waiting for handover
 345 *   2: snap_src, NULL - handed over, waiting for old to be deleted
 346 *   1: NULL, snap_dest - source got destroyed without handover
 347 */
 348static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
 349                                        struct dm_snapshot **snap_src,
 350                                        struct dm_snapshot **snap_dest,
 351                                        struct dm_snapshot **snap_merge)
 352{
 353        struct dm_snapshot *s;
 354        struct origin *o;
 355        int count = 0;
 356        int active;
 357
 358        o = __lookup_origin(snap->origin->bdev);
 359        if (!o)
 360                goto out;
 361
 362        list_for_each_entry(s, &o->snapshots, list) {
 363                if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
 364                        *snap_merge = s;
 365                if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
 366                        continue;
 367
 368                down_read(&s->lock);
 369                active = s->active;
 370                up_read(&s->lock);
 371
 372                if (active) {
 373                        if (snap_src)
 374                                *snap_src = s;
 375                } else if (snap_dest)
 376                        *snap_dest = s;
 377
 378                count++;
 379        }
 380
 381out:
 382        return count;
 383}
 384
 385/*
 386 * On success, returns 1 if this snapshot is a handover destination,
 387 * otherwise returns 0.
 388 */
 389static int __validate_exception_handover(struct dm_snapshot *snap)
 390{
 391        struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
 392        struct dm_snapshot *snap_merge = NULL;
 393
 394        /* Does snapshot need exceptions handed over to it? */
 395        if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
 396                                          &snap_merge) == 2) ||
 397            snap_dest) {
 398                snap->ti->error = "Snapshot cow pairing for exception "
 399                                  "table handover failed";
 400                return -EINVAL;
 401        }
 402
 403        /*
 404         * If no snap_src was found, snap cannot become a handover
 405         * destination.
 406         */
 407        if (!snap_src)
 408                return 0;
 409
 410        /*
 411         * Non-snapshot-merge handover?
 412         */
 413        if (!dm_target_is_snapshot_merge(snap->ti))
 414                return 1;
 415
 416        /*
 417         * Do not allow more than one merging snapshot.
 418         */
 419        if (snap_merge) {
 420                snap->ti->error = "A snapshot is already merging.";
 421                return -EINVAL;
 422        }
 423
 424        if (!snap_src->store->type->prepare_merge ||
 425            !snap_src->store->type->commit_merge) {
 426                snap->ti->error = "Snapshot exception store does not "
 427                                  "support snapshot-merge.";
 428                return -EINVAL;
 429        }
 430
 431        return 1;
 432}
 433
 434static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
 435{
 436        struct dm_snapshot *l;
 437
 438        /* Sort the list according to chunk size, largest-first smallest-last */
 439        list_for_each_entry(l, &o->snapshots, list)
 440                if (l->store->chunk_size < s->store->chunk_size)
 441                        break;
 442        list_add_tail(&s->list, &l->list);
 443}
 444
 445/*
 446 * Make a note of the snapshot and its origin so we can look it
 447 * up when the origin has a write on it.
 448 *
 449 * Also validate snapshot exception store handovers.
 450 * On success, returns 1 if this registration is a handover destination,
 451 * otherwise returns 0.
 452 */
 453static int register_snapshot(struct dm_snapshot *snap)
 454{
 455        struct origin *o, *new_o = NULL;
 456        struct block_device *bdev = snap->origin->bdev;
 457        int r = 0;
 458
 459        new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
 460        if (!new_o)
 461                return -ENOMEM;
 462
 463        down_write(&_origins_lock);
 464
 465        r = __validate_exception_handover(snap);
 466        if (r < 0) {
 467                kfree(new_o);
 468                goto out;
 469        }
 470
 471        o = __lookup_origin(bdev);
 472        if (o)
 473                kfree(new_o);
 474        else {
 475                /* New origin */
 476                o = new_o;
 477
 478                /* Initialise the struct */
 479                INIT_LIST_HEAD(&o->snapshots);
 480                o->bdev = bdev;
 481
 482                __insert_origin(o);
 483        }
 484
 485        __insert_snapshot(o, snap);
 486
 487out:
 488        up_write(&_origins_lock);
 489
 490        return r;
 491}
 492
 493/*
 494 * Move snapshot to correct place in list according to chunk size.
 495 */
 496static void reregister_snapshot(struct dm_snapshot *s)
 497{
 498        struct block_device *bdev = s->origin->bdev;
 499
 500        down_write(&_origins_lock);
 501
 502        list_del(&s->list);
 503        __insert_snapshot(__lookup_origin(bdev), s);
 504
 505        up_write(&_origins_lock);
 506}
 507
 508static void unregister_snapshot(struct dm_snapshot *s)
 509{
 510        struct origin *o;
 511
 512        down_write(&_origins_lock);
 513        o = __lookup_origin(s->origin->bdev);
 514
 515        list_del(&s->list);
 516        if (o && list_empty(&o->snapshots)) {
 517                list_del(&o->hash_list);
 518                kfree(o);
 519        }
 520
 521        up_write(&_origins_lock);
 522}
 523
 524/*
 525 * Implementation of the exception hash tables.
 526 * The lowest hash_shift bits of the chunk number are ignored, allowing
 527 * some consecutive chunks to be grouped together.
 528 */
 529static int dm_exception_table_init(struct dm_exception_table *et,
 530                                   uint32_t size, unsigned hash_shift)
 531{
 532        unsigned int i;
 533
 534        et->hash_shift = hash_shift;
 535        et->hash_mask = size - 1;
 536        et->table = dm_vcalloc(size, sizeof(struct list_head));
 537        if (!et->table)
 538                return -ENOMEM;
 539
 540        for (i = 0; i < size; i++)
 541                INIT_LIST_HEAD(et->table + i);
 542
 543        return 0;
 544}
 545
 546static void dm_exception_table_exit(struct dm_exception_table *et,
 547                                    struct kmem_cache *mem)
 548{
 549        struct list_head *slot;
 550        struct dm_exception *ex, *next;
 551        int i, size;
 552
 553        size = et->hash_mask + 1;
 554        for (i = 0; i < size; i++) {
 555                slot = et->table + i;
 556
 557                list_for_each_entry_safe (ex, next, slot, hash_list)
 558                        kmem_cache_free(mem, ex);
 559        }
 560
 561        vfree(et->table);
 562}
 563
 564static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
 565{
 566        return (chunk >> et->hash_shift) & et->hash_mask;
 567}
 568
 569static void dm_remove_exception(struct dm_exception *e)
 570{
 571        list_del(&e->hash_list);
 572}
 573
 574/*
 575 * Return the exception data for a sector, or NULL if not
 576 * remapped.
 577 */
 578static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
 579                                                chunk_t chunk)
 580{
 581        struct list_head *slot;
 582        struct dm_exception *e;
 583
 584        slot = &et->table[exception_hash(et, chunk)];
 585        list_for_each_entry (e, slot, hash_list)
 586                if (chunk >= e->old_chunk &&
 587                    chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
 588                        return e;
 589
 590        return NULL;
 591}
 592
 593static struct dm_exception *alloc_completed_exception(void)
 594{
 595        struct dm_exception *e;
 596
 597        e = kmem_cache_alloc(exception_cache, GFP_NOIO);
 598        if (!e)
 599                e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
 600
 601        return e;
 602}
 603
 604static void free_completed_exception(struct dm_exception *e)
 605{
 606        kmem_cache_free(exception_cache, e);
 607}
 608
 609static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
 610{
 611        struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
 612                                                             GFP_NOIO);
 613
 614        atomic_inc(&s->pending_exceptions_count);
 615        pe->snap = s;
 616
 617        return pe;
 618}
 619
 620static void free_pending_exception(struct dm_snap_pending_exception *pe)
 621{
 622        struct dm_snapshot *s = pe->snap;
 623
 624        mempool_free(pe, s->pending_pool);
 625        smp_mb__before_atomic_dec();
 626        atomic_dec(&s->pending_exceptions_count);
 627}
 628
 629static void dm_insert_exception(struct dm_exception_table *eh,
 630                                struct dm_exception *new_e)
 631{
 632        struct list_head *l;
 633        struct dm_exception *e = NULL;
 634
 635        l = &eh->table[exception_hash(eh, new_e->old_chunk)];
 636
 637        /* Add immediately if this table doesn't support consecutive chunks */
 638        if (!eh->hash_shift)
 639                goto out;
 640
 641        /* List is ordered by old_chunk */
 642        list_for_each_entry_reverse(e, l, hash_list) {
 643                /* Insert after an existing chunk? */
 644                if (new_e->old_chunk == (e->old_chunk +
 645                                         dm_consecutive_chunk_count(e) + 1) &&
 646                    new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
 647                                         dm_consecutive_chunk_count(e) + 1)) {
 648                        dm_consecutive_chunk_count_inc(e);
 649                        free_completed_exception(new_e);
 650                        return;
 651                }
 652
 653                /* Insert before an existing chunk? */
 654                if (new_e->old_chunk == (e->old_chunk - 1) &&
 655                    new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
 656                        dm_consecutive_chunk_count_inc(e);
 657                        e->old_chunk--;
 658                        e->new_chunk--;
 659                        free_completed_exception(new_e);
 660                        return;
 661                }
 662
 663                if (new_e->old_chunk > e->old_chunk)
 664                        break;
 665        }
 666
 667out:
 668        list_add(&new_e->hash_list, e ? &e->hash_list : l);
 669}
 670
 671/*
 672 * Callback used by the exception stores to load exceptions when
 673 * initialising.
 674 */
 675static int dm_add_exception(void *context, chunk_t old, chunk_t new)
 676{
 677        struct dm_snapshot *s = context;
 678        struct dm_exception *e;
 679
 680        e = alloc_completed_exception();
 681        if (!e)
 682                return -ENOMEM;
 683
 684        e->old_chunk = old;
 685
 686        /* Consecutive_count is implicitly initialised to zero */
 687        e->new_chunk = new;
 688
 689        dm_insert_exception(&s->complete, e);
 690
 691        return 0;
 692}
 693
 694/*
 695 * Return a minimum chunk size of all snapshots that have the specified origin.
 696 * Return zero if the origin has no snapshots.
 697 */
 698static uint32_t __minimum_chunk_size(struct origin *o)
 699{
 700        struct dm_snapshot *snap;
 701        unsigned chunk_size = 0;
 702
 703        if (o)
 704                list_for_each_entry(snap, &o->snapshots, list)
 705                        chunk_size = min_not_zero(chunk_size,
 706                                                  snap->store->chunk_size);
 707
 708        return (uint32_t) chunk_size;
 709}
 710
 711/*
 712 * Hard coded magic.
 713 */
 714static int calc_max_buckets(void)
 715{
 716        /* use a fixed size of 2MB */
 717        unsigned long mem = 2 * 1024 * 1024;
 718        mem /= sizeof(struct list_head);
 719
 720        return mem;
 721}
 722
 723/*
 724 * Allocate room for a suitable hash table.
 725 */
 726static int init_hash_tables(struct dm_snapshot *s)
 727{
 728        sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
 729
 730        /*
 731         * Calculate based on the size of the original volume or
 732         * the COW volume...
 733         */
 734        cow_dev_size = get_dev_size(s->cow->bdev);
 735        origin_dev_size = get_dev_size(s->origin->bdev);
 736        max_buckets = calc_max_buckets();
 737
 738        hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
 739        hash_size = min(hash_size, max_buckets);
 740
 741        if (hash_size < 64)
 742                hash_size = 64;
 743        hash_size = rounddown_pow_of_two(hash_size);
 744        if (dm_exception_table_init(&s->complete, hash_size,
 745                                    DM_CHUNK_CONSECUTIVE_BITS))
 746                return -ENOMEM;
 747
 748        /*
 749         * Allocate hash table for in-flight exceptions
 750         * Make this smaller than the real hash table
 751         */
 752        hash_size >>= 3;
 753        if (hash_size < 64)
 754                hash_size = 64;
 755
 756        if (dm_exception_table_init(&s->pending, hash_size, 0)) {
 757                dm_exception_table_exit(&s->complete, exception_cache);
 758                return -ENOMEM;
 759        }
 760
 761        return 0;
 762}
 763
 764static void merge_shutdown(struct dm_snapshot *s)
 765{
 766        clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
 767        smp_mb__after_clear_bit();
 768        wake_up_bit(&s->state_bits, RUNNING_MERGE);
 769}
 770
 771static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
 772{
 773        s->first_merging_chunk = 0;
 774        s->num_merging_chunks = 0;
 775
 776        return bio_list_get(&s->bios_queued_during_merge);
 777}
 778
 779/*
 780 * Remove one chunk from the index of completed exceptions.
 781 */
 782static int __remove_single_exception_chunk(struct dm_snapshot *s,
 783                                           chunk_t old_chunk)
 784{
 785        struct dm_exception *e;
 786
 787        e = dm_lookup_exception(&s->complete, old_chunk);
 788        if (!e) {
 789                DMERR("Corruption detected: exception for block %llu is "
 790                      "on disk but not in memory",
 791                      (unsigned long long)old_chunk);
 792                return -EINVAL;
 793        }
 794
 795        /*
 796         * If this is the only chunk using this exception, remove exception.
 797         */
 798        if (!dm_consecutive_chunk_count(e)) {
 799                dm_remove_exception(e);
 800                free_completed_exception(e);
 801                return 0;
 802        }
 803
 804        /*
 805         * The chunk may be either at the beginning or the end of a
 806         * group of consecutive chunks - never in the middle.  We are
 807         * removing chunks in the opposite order to that in which they
 808         * were added, so this should always be true.
 809         * Decrement the consecutive chunk counter and adjust the
 810         * starting point if necessary.
 811         */
 812        if (old_chunk == e->old_chunk) {
 813                e->old_chunk++;
 814                e->new_chunk++;
 815        } else if (old_chunk != e->old_chunk +
 816                   dm_consecutive_chunk_count(e)) {
 817                DMERR("Attempt to merge block %llu from the "
 818                      "middle of a chunk range [%llu - %llu]",
 819                      (unsigned long long)old_chunk,
 820                      (unsigned long long)e->old_chunk,
 821                      (unsigned long long)
 822                      e->old_chunk + dm_consecutive_chunk_count(e));
 823                return -EINVAL;
 824        }
 825
 826        dm_consecutive_chunk_count_dec(e);
 827
 828        return 0;
 829}
 830
 831static void flush_bios(struct bio *bio);
 832
 833static int remove_single_exception_chunk(struct dm_snapshot *s)
 834{
 835        struct bio *b = NULL;
 836        int r;
 837        chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
 838
 839        down_write(&s->lock);
 840
 841        /*
 842         * Process chunks (and associated exceptions) in reverse order
 843         * so that dm_consecutive_chunk_count_dec() accounting works.
 844         */
 845        do {
 846                r = __remove_single_exception_chunk(s, old_chunk);
 847                if (r)
 848                        goto out;
 849        } while (old_chunk-- > s->first_merging_chunk);
 850
 851        b = __release_queued_bios_after_merge(s);
 852
 853out:
 854        up_write(&s->lock);
 855        if (b)
 856                flush_bios(b);
 857
 858        return r;
 859}
 860
 861static int origin_write_extent(struct dm_snapshot *merging_snap,
 862                               sector_t sector, unsigned chunk_size);
 863
 864static void merge_callback(int read_err, unsigned long write_err,
 865                           void *context);
 866
 867static uint64_t read_pending_exceptions_done_count(void)
 868{
 869        uint64_t pending_exceptions_done;
 870
 871        spin_lock(&_pending_exceptions_done_spinlock);
 872        pending_exceptions_done = _pending_exceptions_done_count;
 873        spin_unlock(&_pending_exceptions_done_spinlock);
 874
 875        return pending_exceptions_done;
 876}
 877
 878static void increment_pending_exceptions_done_count(void)
 879{
 880        spin_lock(&_pending_exceptions_done_spinlock);
 881        _pending_exceptions_done_count++;
 882        spin_unlock(&_pending_exceptions_done_spinlock);
 883
 884        wake_up_all(&_pending_exceptions_done);
 885}
 886
 887static void snapshot_merge_next_chunks(struct dm_snapshot *s)
 888{
 889        int i, linear_chunks;
 890        chunk_t old_chunk, new_chunk;
 891        struct dm_io_region src, dest;
 892        sector_t io_size;
 893        uint64_t previous_count;
 894
 895        BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
 896        if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
 897                goto shut;
 898
 899        /*
 900         * valid flag never changes during merge, so no lock required.
 901         */
 902        if (!s->valid) {
 903                DMERR("Snapshot is invalid: can't merge");
 904                goto shut;
 905        }
 906
 907        linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
 908                                                      &new_chunk);
 909        if (linear_chunks <= 0) {
 910                if (linear_chunks < 0) {
 911                        DMERR("Read error in exception store: "
 912                              "shutting down merge");
 913                        down_write(&s->lock);
 914                        s->merge_failed = 1;
 915                        up_write(&s->lock);
 916                }
 917                goto shut;
 918        }
 919
 920        /* Adjust old_chunk and new_chunk to reflect start of linear region */
 921        old_chunk = old_chunk + 1 - linear_chunks;
 922        new_chunk = new_chunk + 1 - linear_chunks;
 923
 924        /*
 925         * Use one (potentially large) I/O to copy all 'linear_chunks'
 926         * from the exception store to the origin
 927         */
 928        io_size = linear_chunks * s->store->chunk_size;
 929
 930        dest.bdev = s->origin->bdev;
 931        dest.sector = chunk_to_sector(s->store, old_chunk);
 932        dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
 933
 934        src.bdev = s->cow->bdev;
 935        src.sector = chunk_to_sector(s->store, new_chunk);
 936        src.count = dest.count;
 937
 938        /*
 939         * Reallocate any exceptions needed in other snapshots then
 940         * wait for the pending exceptions to complete.
 941         * Each time any pending exception (globally on the system)
 942         * completes we are woken and repeat the process to find out
 943         * if we can proceed.  While this may not seem a particularly
 944         * efficient algorithm, it is not expected to have any
 945         * significant impact on performance.
 946         */
 947        previous_count = read_pending_exceptions_done_count();
 948        while (origin_write_extent(s, dest.sector, io_size)) {
 949                wait_event(_pending_exceptions_done,
 950                           (read_pending_exceptions_done_count() !=
 951                            previous_count));
 952                /* Retry after the wait, until all exceptions are done. */
 953                previous_count = read_pending_exceptions_done_count();
 954        }
 955
 956        down_write(&s->lock);
 957        s->first_merging_chunk = old_chunk;
 958        s->num_merging_chunks = linear_chunks;
 959        up_write(&s->lock);
 960
 961        /* Wait until writes to all 'linear_chunks' drain */
 962        for (i = 0; i < linear_chunks; i++)
 963                __check_for_conflicting_io(s, old_chunk + i);
 964
 965        dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
 966        return;
 967
 968shut:
 969        merge_shutdown(s);
 970}
 971
 972static void error_bios(struct bio *bio);
 973
 974static void merge_callback(int read_err, unsigned long write_err, void *context)
 975{
 976        struct dm_snapshot *s = context;
 977        struct bio *b = NULL;
 978
 979        if (read_err || write_err) {
 980                if (read_err)
 981                        DMERR("Read error: shutting down merge.");
 982                else
 983                        DMERR("Write error: shutting down merge.");
 984                goto shut;
 985        }
 986
 987        if (s->store->type->commit_merge(s->store,
 988                                         s->num_merging_chunks) < 0) {
 989                DMERR("Write error in exception store: shutting down merge");
 990                goto shut;
 991        }
 992
 993        if (remove_single_exception_chunk(s) < 0)
 994                goto shut;
 995
 996        snapshot_merge_next_chunks(s);
 997
 998        return;
 999
1000shut:
1001        down_write(&s->lock);
1002        s->merge_failed = 1;
1003        b = __release_queued_bios_after_merge(s);
1004        up_write(&s->lock);
1005        error_bios(b);
1006
1007        merge_shutdown(s);
1008}
1009
1010static void start_merge(struct dm_snapshot *s)
1011{
1012        if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1013                snapshot_merge_next_chunks(s);
1014}
1015
1016static int wait_schedule(void *ptr)
1017{
1018        schedule();
1019
1020        return 0;
1021}
1022
1023/*
1024 * Stop the merging process and wait until it finishes.
1025 */
1026static void stop_merge(struct dm_snapshot *s)
1027{
1028        set_bit(SHUTDOWN_MERGE, &s->state_bits);
1029        wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1030                    TASK_UNINTERRUPTIBLE);
1031        clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1032}
1033
1034/*
1035 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1036 */
1037static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1038{
1039        struct dm_snapshot *s;
1040        int i;
1041        int r = -EINVAL;
1042        char *origin_path, *cow_path;
1043        unsigned args_used, num_flush_bios = 1;
1044        fmode_t origin_mode = FMODE_READ;
1045
1046        if (argc != 4) {
1047                ti->error = "requires exactly 4 arguments";
1048                r = -EINVAL;
1049                goto bad;
1050        }
1051
1052        if (dm_target_is_snapshot_merge(ti)) {
1053                num_flush_bios = 2;
1054                origin_mode = FMODE_WRITE;
1055        }
1056
1057        s = kmalloc(sizeof(*s), GFP_KERNEL);
1058        if (!s) {
1059                ti->error = "Cannot allocate private snapshot structure";
1060                r = -ENOMEM;
1061                goto bad;
1062        }
1063
1064        origin_path = argv[0];
1065        argv++;
1066        argc--;
1067
1068        r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1069        if (r) {
1070                ti->error = "Cannot get origin device";
1071                goto bad_origin;
1072        }
1073
1074        cow_path = argv[0];
1075        argv++;
1076        argc--;
1077
1078        r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1079        if (r) {
1080                ti->error = "Cannot get COW device";
1081                goto bad_cow;
1082        }
1083
1084        r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1085        if (r) {
1086                ti->error = "Couldn't create exception store";
1087                r = -EINVAL;
1088                goto bad_store;
1089        }
1090
1091        argv += args_used;
1092        argc -= args_used;
1093
1094        s->ti = ti;
1095        s->valid = 1;
1096        s->active = 0;
1097        atomic_set(&s->pending_exceptions_count, 0);
1098        init_rwsem(&s->lock);
1099        INIT_LIST_HEAD(&s->list);
1100        spin_lock_init(&s->pe_lock);
1101        s->state_bits = 0;
1102        s->merge_failed = 0;
1103        s->first_merging_chunk = 0;
1104        s->num_merging_chunks = 0;
1105        bio_list_init(&s->bios_queued_during_merge);
1106
1107        /* Allocate hash table for COW data */
1108        if (init_hash_tables(s)) {
1109                ti->error = "Unable to allocate hash table space";
1110                r = -ENOMEM;
1111                goto bad_hash_tables;
1112        }
1113
1114        s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1115        if (IS_ERR(s->kcopyd_client)) {
1116                r = PTR_ERR(s->kcopyd_client);
1117                ti->error = "Could not create kcopyd client";
1118                goto bad_kcopyd;
1119        }
1120
1121        s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1122        if (!s->pending_pool) {
1123                ti->error = "Could not allocate mempool for pending exceptions";
1124                r = -ENOMEM;
1125                goto bad_pending_pool;
1126        }
1127
1128        for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1129                INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1130
1131        spin_lock_init(&s->tracked_chunk_lock);
1132
1133        ti->private = s;
1134        ti->num_flush_bios = num_flush_bios;
1135        ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1136
1137        /* Add snapshot to the list of snapshots for this origin */
1138        /* Exceptions aren't triggered till snapshot_resume() is called */
1139        r = register_snapshot(s);
1140        if (r == -ENOMEM) {
1141                ti->error = "Snapshot origin struct allocation failed";
1142                goto bad_load_and_register;
1143        } else if (r < 0) {
1144                /* invalid handover, register_snapshot has set ti->error */
1145                goto bad_load_and_register;
1146        }
1147
1148        /*
1149         * Metadata must only be loaded into one table at once, so skip this
1150         * if metadata will be handed over during resume.
1151         * Chunk size will be set during the handover - set it to zero to
1152         * ensure it's ignored.
1153         */
1154        if (r > 0) {
1155                s->store->chunk_size = 0;
1156                return 0;
1157        }
1158
1159        r = s->store->type->read_metadata(s->store, dm_add_exception,
1160                                          (void *)s);
1161        if (r < 0) {
1162                ti->error = "Failed to read snapshot metadata";
1163                goto bad_read_metadata;
1164        } else if (r > 0) {
1165                s->valid = 0;
1166                DMWARN("Snapshot is marked invalid.");
1167        }
1168
1169        if (!s->store->chunk_size) {
1170                ti->error = "Chunk size not set";
1171                goto bad_read_metadata;
1172        }
1173
1174        r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1175        if (r)
1176                goto bad_read_metadata;
1177
1178        return 0;
1179
1180bad_read_metadata:
1181        unregister_snapshot(s);
1182
1183bad_load_and_register:
1184        mempool_destroy(s->pending_pool);
1185
1186bad_pending_pool:
1187        dm_kcopyd_client_destroy(s->kcopyd_client);
1188
1189bad_kcopyd:
1190        dm_exception_table_exit(&s->pending, pending_cache);
1191        dm_exception_table_exit(&s->complete, exception_cache);
1192
1193bad_hash_tables:
1194        dm_exception_store_destroy(s->store);
1195
1196bad_store:
1197        dm_put_device(ti, s->cow);
1198
1199bad_cow:
1200        dm_put_device(ti, s->origin);
1201
1202bad_origin:
1203        kfree(s);
1204
1205bad:
1206        return r;
1207}
1208
1209static void __free_exceptions(struct dm_snapshot *s)
1210{
1211        dm_kcopyd_client_destroy(s->kcopyd_client);
1212        s->kcopyd_client = NULL;
1213
1214        dm_exception_table_exit(&s->pending, pending_cache);
1215        dm_exception_table_exit(&s->complete, exception_cache);
1216}
1217
1218static void __handover_exceptions(struct dm_snapshot *snap_src,
1219                                  struct dm_snapshot *snap_dest)
1220{
1221        union {
1222                struct dm_exception_table table_swap;
1223                struct dm_exception_store *store_swap;
1224        } u;
1225
1226        /*
1227         * Swap all snapshot context information between the two instances.
1228         */
1229        u.table_swap = snap_dest->complete;
1230        snap_dest->complete = snap_src->complete;
1231        snap_src->complete = u.table_swap;
1232
1233        u.store_swap = snap_dest->store;
1234        snap_dest->store = snap_src->store;
1235        snap_src->store = u.store_swap;
1236
1237        snap_dest->store->snap = snap_dest;
1238        snap_src->store->snap = snap_src;
1239
1240        snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1241        snap_dest->valid = snap_src->valid;
1242
1243        /*
1244         * Set source invalid to ensure it receives no further I/O.
1245         */
1246        snap_src->valid = 0;
1247}
1248
1249static void snapshot_dtr(struct dm_target *ti)
1250{
1251#ifdef CONFIG_DM_DEBUG
1252        int i;
1253#endif
1254        struct dm_snapshot *s = ti->private;
1255        struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1256
1257        down_read(&_origins_lock);
1258        /* Check whether exception handover must be cancelled */
1259        (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1260        if (snap_src && snap_dest && (s == snap_src)) {
1261                down_write(&snap_dest->lock);
1262                snap_dest->valid = 0;
1263                up_write(&snap_dest->lock);
1264                DMERR("Cancelling snapshot handover.");
1265        }
1266        up_read(&_origins_lock);
1267
1268        if (dm_target_is_snapshot_merge(ti))
1269                stop_merge(s);
1270
1271        /* Prevent further origin writes from using this snapshot. */
1272        /* After this returns there can be no new kcopyd jobs. */
1273        unregister_snapshot(s);
1274
1275        while (atomic_read(&s->pending_exceptions_count))
1276                msleep(1);
1277        /*
1278         * Ensure instructions in mempool_destroy aren't reordered
1279         * before atomic_read.
1280         */
1281        smp_mb();
1282
1283#ifdef CONFIG_DM_DEBUG
1284        for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1285                BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1286#endif
1287
1288        __free_exceptions(s);
1289
1290        mempool_destroy(s->pending_pool);
1291
1292        dm_exception_store_destroy(s->store);
1293
1294        dm_put_device(ti, s->cow);
1295
1296        dm_put_device(ti, s->origin);
1297
1298        kfree(s);
1299}
1300
1301/*
1302 * Flush a list of buffers.
1303 */
1304static void flush_bios(struct bio *bio)
1305{
1306        struct bio *n;
1307
1308        while (bio) {
1309                n = bio->bi_next;
1310                bio->bi_next = NULL;
1311                generic_make_request(bio);
1312                bio = n;
1313        }
1314}
1315
1316static int do_origin(struct dm_dev *origin, struct bio *bio);
1317
1318/*
1319 * Flush a list of buffers.
1320 */
1321static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1322{
1323        struct bio *n;
1324        int r;
1325
1326        while (bio) {
1327                n = bio->bi_next;
1328                bio->bi_next = NULL;
1329                r = do_origin(s->origin, bio);
1330                if (r == DM_MAPIO_REMAPPED)
1331                        generic_make_request(bio);
1332                bio = n;
1333        }
1334}
1335
1336/*
1337 * Error a list of buffers.
1338 */
1339static void error_bios(struct bio *bio)
1340{
1341        struct bio *n;
1342
1343        while (bio) {
1344                n = bio->bi_next;
1345                bio->bi_next = NULL;
1346                bio_io_error(bio);
1347                bio = n;
1348        }
1349}
1350
1351static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1352{
1353        if (!s->valid)
1354                return;
1355
1356        if (err == -EIO)
1357                DMERR("Invalidating snapshot: Error reading/writing.");
1358        else if (err == -ENOMEM)
1359                DMERR("Invalidating snapshot: Unable to allocate exception.");
1360
1361        if (s->store->type->drop_snapshot)
1362                s->store->type->drop_snapshot(s->store);
1363
1364        s->valid = 0;
1365
1366        dm_table_event(s->ti->table);
1367}
1368
1369static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1370{
1371        struct dm_exception *e;
1372        struct dm_snapshot *s = pe->snap;
1373        struct bio *origin_bios = NULL;
1374        struct bio *snapshot_bios = NULL;
1375        struct bio *full_bio = NULL;
1376        int error = 0;
1377
1378        if (!success) {
1379                /* Read/write error - snapshot is unusable */
1380                down_write(&s->lock);
1381                __invalidate_snapshot(s, -EIO);
1382                error = 1;
1383                goto out;
1384        }
1385
1386        e = alloc_completed_exception();
1387        if (!e) {
1388                down_write(&s->lock);
1389                __invalidate_snapshot(s, -ENOMEM);
1390                error = 1;
1391                goto out;
1392        }
1393        *e = pe->e;
1394
1395        down_write(&s->lock);
1396        if (!s->valid) {
1397                free_completed_exception(e);
1398                error = 1;
1399                goto out;
1400        }
1401
1402        /* Check for conflicting reads */
1403        __check_for_conflicting_io(s, pe->e.old_chunk);
1404
1405        /*
1406         * Add a proper exception, and remove the
1407         * in-flight exception from the list.
1408         */
1409        dm_insert_exception(&s->complete, e);
1410
1411out:
1412        dm_remove_exception(&pe->e);
1413        snapshot_bios = bio_list_get(&pe->snapshot_bios);
1414        origin_bios = bio_list_get(&pe->origin_bios);
1415        full_bio = pe->full_bio;
1416        if (full_bio) {
1417                full_bio->bi_end_io = pe->full_bio_end_io;
1418                full_bio->bi_private = pe->full_bio_private;
1419        }
1420        free_pending_exception(pe);
1421
1422        increment_pending_exceptions_done_count();
1423
1424        up_write(&s->lock);
1425
1426        /* Submit any pending write bios */
1427        if (error) {
1428                if (full_bio)
1429                        bio_io_error(full_bio);
1430                error_bios(snapshot_bios);
1431        } else {
1432                if (full_bio)
1433                        bio_endio(full_bio, 0);
1434                flush_bios(snapshot_bios);
1435        }
1436
1437        retry_origin_bios(s, origin_bios);
1438}
1439
1440static void commit_callback(void *context, int success)
1441{
1442        struct dm_snap_pending_exception *pe = context;
1443
1444        pending_complete(pe, success);
1445}
1446
1447/*
1448 * Called when the copy I/O has finished.  kcopyd actually runs
1449 * this code so don't block.
1450 */
1451static void copy_callback(int read_err, unsigned long write_err, void *context)
1452{
1453        struct dm_snap_pending_exception *pe = context;
1454        struct dm_snapshot *s = pe->snap;
1455
1456        if (read_err || write_err)
1457                pending_complete(pe, 0);
1458
1459        else
1460                /* Update the metadata if we are persistent */
1461                s->store->type->commit_exception(s->store, &pe->e,
1462                                                 commit_callback, pe);
1463}
1464
1465/*
1466 * Dispatches the copy operation to kcopyd.
1467 */
1468static void start_copy(struct dm_snap_pending_exception *pe)
1469{
1470        struct dm_snapshot *s = pe->snap;
1471        struct dm_io_region src, dest;
1472        struct block_device *bdev = s->origin->bdev;
1473        sector_t dev_size;
1474
1475        dev_size = get_dev_size(bdev);
1476
1477        src.bdev = bdev;
1478        src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1479        src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1480
1481        dest.bdev = s->cow->bdev;
1482        dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1483        dest.count = src.count;
1484
1485        /* Hand over to kcopyd */
1486        dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1487}
1488
1489static void full_bio_end_io(struct bio *bio, int error)
1490{
1491        void *callback_data = bio->bi_private;
1492
1493        dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1494}
1495
1496static void start_full_bio(struct dm_snap_pending_exception *pe,
1497                           struct bio *bio)
1498{
1499        struct dm_snapshot *s = pe->snap;
1500        void *callback_data;
1501
1502        pe->full_bio = bio;
1503        pe->full_bio_end_io = bio->bi_end_io;
1504        pe->full_bio_private = bio->bi_private;
1505
1506        callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1507                                                   copy_callback, pe);
1508
1509        bio->bi_end_io = full_bio_end_io;
1510        bio->bi_private = callback_data;
1511
1512        generic_make_request(bio);
1513}
1514
1515static struct dm_snap_pending_exception *
1516__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1517{
1518        struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1519
1520        if (!e)
1521                return NULL;
1522
1523        return container_of(e, struct dm_snap_pending_exception, e);
1524}
1525
1526/*
1527 * Looks to see if this snapshot already has a pending exception
1528 * for this chunk, otherwise it allocates a new one and inserts
1529 * it into the pending table.
1530 *
1531 * NOTE: a write lock must be held on snap->lock before calling
1532 * this.
1533 */
1534static struct dm_snap_pending_exception *
1535__find_pending_exception(struct dm_snapshot *s,
1536                         struct dm_snap_pending_exception *pe, chunk_t chunk)
1537{
1538        struct dm_snap_pending_exception *pe2;
1539
1540        pe2 = __lookup_pending_exception(s, chunk);
1541        if (pe2) {
1542                free_pending_exception(pe);
1543                return pe2;
1544        }
1545
1546        pe->e.old_chunk = chunk;
1547        bio_list_init(&pe->origin_bios);
1548        bio_list_init(&pe->snapshot_bios);
1549        pe->started = 0;
1550        pe->full_bio = NULL;
1551
1552        if (s->store->type->prepare_exception(s->store, &pe->e)) {
1553                free_pending_exception(pe);
1554                return NULL;
1555        }
1556
1557        dm_insert_exception(&s->pending, &pe->e);
1558
1559        return pe;
1560}
1561
1562static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1563                            struct bio *bio, chunk_t chunk)
1564{
1565        bio->bi_bdev = s->cow->bdev;
1566        bio->bi_sector = chunk_to_sector(s->store,
1567                                         dm_chunk_number(e->new_chunk) +
1568                                         (chunk - e->old_chunk)) +
1569                                         (bio->bi_sector &
1570                                          s->store->chunk_mask);
1571}
1572
1573static int snapshot_map(struct dm_target *ti, struct bio *bio)
1574{
1575        struct dm_exception *e;
1576        struct dm_snapshot *s = ti->private;
1577        int r = DM_MAPIO_REMAPPED;
1578        chunk_t chunk;
1579        struct dm_snap_pending_exception *pe = NULL;
1580
1581        init_tracked_chunk(bio);
1582
1583        if (bio->bi_rw & REQ_FLUSH) {
1584                bio->bi_bdev = s->cow->bdev;
1585                return DM_MAPIO_REMAPPED;
1586        }
1587
1588        chunk = sector_to_chunk(s->store, bio->bi_sector);
1589
1590        /* Full snapshots are not usable */
1591        /* To get here the table must be live so s->active is always set. */
1592        if (!s->valid)
1593                return -EIO;
1594
1595        /* FIXME: should only take write lock if we need
1596         * to copy an exception */
1597        down_write(&s->lock);
1598
1599        if (!s->valid) {
1600                r = -EIO;
1601                goto out_unlock;
1602        }
1603
1604        /* If the block is already remapped - use that, else remap it */
1605        e = dm_lookup_exception(&s->complete, chunk);
1606        if (e) {
1607                remap_exception(s, e, bio, chunk);
1608                goto out_unlock;
1609        }
1610
1611        /*
1612         * Write to snapshot - higher level takes care of RW/RO
1613         * flags so we should only get this if we are
1614         * writeable.
1615         */
1616        if (bio_rw(bio) == WRITE) {
1617                pe = __lookup_pending_exception(s, chunk);
1618                if (!pe) {
1619                        up_write(&s->lock);
1620                        pe = alloc_pending_exception(s);
1621                        down_write(&s->lock);
1622
1623                        if (!s->valid) {
1624                                free_pending_exception(pe);
1625                                r = -EIO;
1626                                goto out_unlock;
1627                        }
1628
1629                        e = dm_lookup_exception(&s->complete, chunk);
1630                        if (e) {
1631                                free_pending_exception(pe);
1632                                remap_exception(s, e, bio, chunk);
1633                                goto out_unlock;
1634                        }
1635
1636                        pe = __find_pending_exception(s, pe, chunk);
1637                        if (!pe) {
1638                                __invalidate_snapshot(s, -ENOMEM);
1639                                r = -EIO;
1640                                goto out_unlock;
1641                        }
1642                }
1643
1644                remap_exception(s, &pe->e, bio, chunk);
1645
1646                r = DM_MAPIO_SUBMITTED;
1647
1648                if (!pe->started &&
1649                    bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1650                        pe->started = 1;
1651                        up_write(&s->lock);
1652                        start_full_bio(pe, bio);
1653                        goto out;
1654                }
1655
1656                bio_list_add(&pe->snapshot_bios, bio);
1657
1658                if (!pe->started) {
1659                        /* this is protected by snap->lock */
1660                        pe->started = 1;
1661                        up_write(&s->lock);
1662                        start_copy(pe);
1663                        goto out;
1664                }
1665        } else {
1666                bio->bi_bdev = s->origin->bdev;
1667                track_chunk(s, bio, chunk);
1668        }
1669
1670out_unlock:
1671        up_write(&s->lock);
1672out:
1673        return r;
1674}
1675
1676/*
1677 * A snapshot-merge target behaves like a combination of a snapshot
1678 * target and a snapshot-origin target.  It only generates new
1679 * exceptions in other snapshots and not in the one that is being
1680 * merged.
1681 *
1682 * For each chunk, if there is an existing exception, it is used to
1683 * redirect I/O to the cow device.  Otherwise I/O is sent to the origin,
1684 * which in turn might generate exceptions in other snapshots.
1685 * If merging is currently taking place on the chunk in question, the
1686 * I/O is deferred by adding it to s->bios_queued_during_merge.
1687 */
1688static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1689{
1690        struct dm_exception *e;
1691        struct dm_snapshot *s = ti->private;
1692        int r = DM_MAPIO_REMAPPED;
1693        chunk_t chunk;
1694
1695        init_tracked_chunk(bio);
1696
1697        if (bio->bi_rw & REQ_FLUSH) {
1698                if (!dm_bio_get_target_bio_nr(bio))
1699                        bio->bi_bdev = s->origin->bdev;
1700                else
1701                        bio->bi_bdev = s->cow->bdev;
1702                return DM_MAPIO_REMAPPED;
1703        }
1704
1705        chunk = sector_to_chunk(s->store, bio->bi_sector);
1706
1707        down_write(&s->lock);
1708
1709        /* Full merging snapshots are redirected to the origin */
1710        if (!s->valid)
1711                goto redirect_to_origin;
1712
1713        /* If the block is already remapped - use that */
1714        e = dm_lookup_exception(&s->complete, chunk);
1715        if (e) {
1716                /* Queue writes overlapping with chunks being merged */
1717                if (bio_rw(bio) == WRITE &&
1718                    chunk >= s->first_merging_chunk &&
1719                    chunk < (s->first_merging_chunk +
1720                             s->num_merging_chunks)) {
1721                        bio->bi_bdev = s->origin->bdev;
1722                        bio_list_add(&s->bios_queued_during_merge, bio);
1723                        r = DM_MAPIO_SUBMITTED;
1724                        goto out_unlock;
1725                }
1726
1727                remap_exception(s, e, bio, chunk);
1728
1729                if (bio_rw(bio) == WRITE)
1730                        track_chunk(s, bio, chunk);
1731                goto out_unlock;
1732        }
1733
1734redirect_to_origin:
1735        bio->bi_bdev = s->origin->bdev;
1736
1737        if (bio_rw(bio) == WRITE) {
1738                up_write(&s->lock);
1739                return do_origin(s->origin, bio);
1740        }
1741
1742out_unlock:
1743        up_write(&s->lock);
1744
1745        return r;
1746}
1747
1748static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
1749{
1750        struct dm_snapshot *s = ti->private;
1751
1752        if (is_bio_tracked(bio))
1753                stop_tracking_chunk(s, bio);
1754
1755        return 0;
1756}
1757
1758static void snapshot_merge_presuspend(struct dm_target *ti)
1759{
1760        struct dm_snapshot *s = ti->private;
1761
1762        stop_merge(s);
1763}
1764
1765static int snapshot_preresume(struct dm_target *ti)
1766{
1767        int r = 0;
1768        struct dm_snapshot *s = ti->private;
1769        struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1770
1771        down_read(&_origins_lock);
1772        (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1773        if (snap_src && snap_dest) {
1774                down_read(&snap_src->lock);
1775                if (s == snap_src) {
1776                        DMERR("Unable to resume snapshot source until "
1777                              "handover completes.");
1778                        r = -EINVAL;
1779                } else if (!dm_suspended(snap_src->ti)) {
1780                        DMERR("Unable to perform snapshot handover until "
1781                              "source is suspended.");
1782                        r = -EINVAL;
1783                }
1784                up_read(&snap_src->lock);
1785        }
1786        up_read(&_origins_lock);
1787
1788        return r;
1789}
1790
1791static void snapshot_resume(struct dm_target *ti)
1792{
1793        struct dm_snapshot *s = ti->private;
1794        struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1795
1796        down_read(&_origins_lock);
1797        (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1798        if (snap_src && snap_dest) {
1799                down_write(&snap_src->lock);
1800                down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1801                __handover_exceptions(snap_src, snap_dest);
1802                up_write(&snap_dest->lock);
1803                up_write(&snap_src->lock);
1804        }
1805        up_read(&_origins_lock);
1806
1807        /* Now we have correct chunk size, reregister */
1808        reregister_snapshot(s);
1809
1810        down_write(&s->lock);
1811        s->active = 1;
1812        up_write(&s->lock);
1813}
1814
1815static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1816{
1817        uint32_t min_chunksize;
1818
1819        down_read(&_origins_lock);
1820        min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1821        up_read(&_origins_lock);
1822
1823        return min_chunksize;
1824}
1825
1826static void snapshot_merge_resume(struct dm_target *ti)
1827{
1828        struct dm_snapshot *s = ti->private;
1829
1830        /*
1831         * Handover exceptions from existing snapshot.
1832         */
1833        snapshot_resume(ti);
1834
1835        /*
1836         * snapshot-merge acts as an origin, so set ti->max_io_len
1837         */
1838        ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1839
1840        start_merge(s);
1841}
1842
1843static void snapshot_status(struct dm_target *ti, status_type_t type,
1844                            unsigned status_flags, char *result, unsigned maxlen)
1845{
1846        unsigned sz = 0;
1847        struct dm_snapshot *snap = ti->private;
1848
1849        switch (type) {
1850        case STATUSTYPE_INFO:
1851
1852                down_write(&snap->lock);
1853
1854                if (!snap->valid)
1855                        DMEMIT("Invalid");
1856                else if (snap->merge_failed)
1857                        DMEMIT("Merge failed");
1858                else {
1859                        if (snap->store->type->usage) {
1860                                sector_t total_sectors, sectors_allocated,
1861                                         metadata_sectors;
1862                                snap->store->type->usage(snap->store,
1863                                                         &total_sectors,
1864                                                         &sectors_allocated,
1865                                                         &metadata_sectors);
1866                                DMEMIT("%llu/%llu %llu",
1867                                       (unsigned long long)sectors_allocated,
1868                                       (unsigned long long)total_sectors,
1869                                       (unsigned long long)metadata_sectors);
1870                        }
1871                        else
1872                                DMEMIT("Unknown");
1873                }
1874
1875                up_write(&snap->lock);
1876
1877                break;
1878
1879        case STATUSTYPE_TABLE:
1880                /*
1881                 * kdevname returns a static pointer so we need
1882                 * to make private copies if the output is to
1883                 * make sense.
1884                 */
1885                DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1886                snap->store->type->status(snap->store, type, result + sz,
1887                                          maxlen - sz);
1888                break;
1889        }
1890}
1891
1892static int snapshot_iterate_devices(struct dm_target *ti,
1893                                    iterate_devices_callout_fn fn, void *data)
1894{
1895        struct dm_snapshot *snap = ti->private;
1896        int r;
1897
1898        r = fn(ti, snap->origin, 0, ti->len, data);
1899
1900        if (!r)
1901                r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1902
1903        return r;
1904}
1905
1906
1907/*-----------------------------------------------------------------
1908 * Origin methods
1909 *---------------------------------------------------------------*/
1910
1911/*
1912 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1913 * supplied bio was ignored.  The caller may submit it immediately.
1914 * (No remapping actually occurs as the origin is always a direct linear
1915 * map.)
1916 *
1917 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1918 * and any supplied bio is added to a list to be submitted once all
1919 * the necessary exceptions exist.
1920 */
1921static int __origin_write(struct list_head *snapshots, sector_t sector,
1922                          struct bio *bio)
1923{
1924        int r = DM_MAPIO_REMAPPED;
1925        struct dm_snapshot *snap;
1926        struct dm_exception *e;
1927        struct dm_snap_pending_exception *pe;
1928        struct dm_snap_pending_exception *pe_to_start_now = NULL;
1929        struct dm_snap_pending_exception *pe_to_start_last = NULL;
1930        chunk_t chunk;
1931
1932        /* Do all the snapshots on this origin */
1933        list_for_each_entry (snap, snapshots, list) {
1934                /*
1935                 * Don't make new exceptions in a merging snapshot
1936                 * because it has effectively been deleted
1937                 */
1938                if (dm_target_is_snapshot_merge(snap->ti))
1939                        continue;
1940
1941                down_write(&snap->lock);
1942
1943                /* Only deal with valid and active snapshots */
1944                if (!snap->valid || !snap->active)
1945                        goto next_snapshot;
1946
1947                /* Nothing to do if writing beyond end of snapshot */
1948                if (sector >= dm_table_get_size(snap->ti->table))
1949                        goto next_snapshot;
1950
1951                /*
1952                 * Remember, different snapshots can have
1953                 * different chunk sizes.
1954                 */
1955                chunk = sector_to_chunk(snap->store, sector);
1956
1957                /*
1958                 * Check exception table to see if block
1959                 * is already remapped in this snapshot
1960                 * and trigger an exception if not.
1961                 */
1962                e = dm_lookup_exception(&snap->complete, chunk);
1963                if (e)
1964                        goto next_snapshot;
1965
1966                pe = __lookup_pending_exception(snap, chunk);
1967                if (!pe) {
1968                        up_write(&snap->lock);
1969                        pe = alloc_pending_exception(snap);
1970                        down_write(&snap->lock);
1971
1972                        if (!snap->valid) {
1973                                free_pending_exception(pe);
1974                                goto next_snapshot;
1975                        }
1976
1977                        e = dm_lookup_exception(&snap->complete, chunk);
1978                        if (e) {
1979                                free_pending_exception(pe);
1980                                goto next_snapshot;
1981                        }
1982
1983                        pe = __find_pending_exception(snap, pe, chunk);
1984                        if (!pe) {
1985                                __invalidate_snapshot(snap, -ENOMEM);
1986                                goto next_snapshot;
1987                        }
1988                }
1989
1990                r = DM_MAPIO_SUBMITTED;
1991
1992                /*
1993                 * If an origin bio was supplied, queue it to wait for the
1994                 * completion of this exception, and start this one last,
1995                 * at the end of the function.
1996                 */
1997                if (bio) {
1998                        bio_list_add(&pe->origin_bios, bio);
1999                        bio = NULL;
2000
2001                        if (!pe->started) {
2002                                pe->started = 1;
2003                                pe_to_start_last = pe;
2004                        }
2005                }
2006
2007                if (!pe->started) {
2008                        pe->started = 1;
2009                        pe_to_start_now = pe;
2010                }
2011
2012next_snapshot:
2013                up_write(&snap->lock);
2014
2015                if (pe_to_start_now) {
2016                        start_copy(pe_to_start_now);
2017                        pe_to_start_now = NULL;
2018                }
2019        }
2020
2021        /*
2022         * Submit the exception against which the bio is queued last,
2023         * to give the other exceptions a head start.
2024         */
2025        if (pe_to_start_last)
2026                start_copy(pe_to_start_last);
2027
2028        return r;
2029}
2030
2031/*
2032 * Called on a write from the origin driver.
2033 */
2034static int do_origin(struct dm_dev *origin, struct bio *bio)
2035{
2036        struct origin *o;
2037        int r = DM_MAPIO_REMAPPED;
2038
2039        down_read(&_origins_lock);
2040        o = __lookup_origin(origin->bdev);
2041        if (o)
2042                r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2043        up_read(&_origins_lock);
2044
2045        return r;
2046}
2047
2048/*
2049 * Trigger exceptions in all non-merging snapshots.
2050 *
2051 * The chunk size of the merging snapshot may be larger than the chunk
2052 * size of some other snapshot so we may need to reallocate multiple
2053 * chunks in other snapshots.
2054 *
2055 * We scan all the overlapping exceptions in the other snapshots.
2056 * Returns 1 if anything was reallocated and must be waited for,
2057 * otherwise returns 0.
2058 *
2059 * size must be a multiple of merging_snap's chunk_size.
2060 */
2061static int origin_write_extent(struct dm_snapshot *merging_snap,
2062                               sector_t sector, unsigned size)
2063{
2064        int must_wait = 0;
2065        sector_t n;
2066        struct origin *o;
2067
2068        /*
2069         * The origin's __minimum_chunk_size() got stored in max_io_len
2070         * by snapshot_merge_resume().
2071         */
2072        down_read(&_origins_lock);
2073        o = __lookup_origin(merging_snap->origin->bdev);
2074        for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2075                if (__origin_write(&o->snapshots, sector + n, NULL) ==
2076                    DM_MAPIO_SUBMITTED)
2077                        must_wait = 1;
2078        up_read(&_origins_lock);
2079
2080        return must_wait;
2081}
2082
2083/*
2084 * Origin: maps a linear range of a device, with hooks for snapshotting.
2085 */
2086
2087/*
2088 * Construct an origin mapping: <dev_path>
2089 * The context for an origin is merely a 'struct dm_dev *'
2090 * pointing to the real device.
2091 */
2092static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2093{
2094        int r;
2095        struct dm_dev *dev;
2096
2097        if (argc != 1) {
2098                ti->error = "origin: incorrect number of arguments";
2099                return -EINVAL;
2100        }
2101
2102        r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2103        if (r) {
2104                ti->error = "Cannot get target device";
2105                return r;
2106        }
2107
2108        ti->private = dev;
2109        ti->num_flush_bios = 1;
2110
2111        return 0;
2112}
2113
2114static void origin_dtr(struct dm_target *ti)
2115{
2116        struct dm_dev *dev = ti->private;
2117        dm_put_device(ti, dev);
2118}
2119
2120static int origin_map(struct dm_target *ti, struct bio *bio)
2121{
2122        struct dm_dev *dev = ti->private;
2123        bio->bi_bdev = dev->bdev;
2124
2125        if (bio->bi_rw & REQ_FLUSH)
2126                return DM_MAPIO_REMAPPED;
2127
2128        /* Only tell snapshots if this is a write */
2129        return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2130}
2131
2132/*
2133 * Set the target "max_io_len" field to the minimum of all the snapshots'
2134 * chunk sizes.
2135 */
2136static void origin_resume(struct dm_target *ti)
2137{
2138        struct dm_dev *dev = ti->private;
2139
2140        ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
2141}
2142
2143static void origin_status(struct dm_target *ti, status_type_t type,
2144                          unsigned status_flags, char *result, unsigned maxlen)
2145{
2146        struct dm_dev *dev = ti->private;
2147
2148        switch (type) {
2149        case STATUSTYPE_INFO:
2150                result[0] = '\0';
2151                break;
2152
2153        case STATUSTYPE_TABLE:
2154                snprintf(result, maxlen, "%s", dev->name);
2155                break;
2156        }
2157}
2158
2159static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2160                        struct bio_vec *biovec, int max_size)
2161{
2162        struct dm_dev *dev = ti->private;
2163        struct request_queue *q = bdev_get_queue(dev->bdev);
2164
2165        if (!q->merge_bvec_fn)
2166                return max_size;
2167
2168        bvm->bi_bdev = dev->bdev;
2169
2170        return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2171}
2172
2173static int origin_iterate_devices(struct dm_target *ti,
2174                                  iterate_devices_callout_fn fn, void *data)
2175{
2176        struct dm_dev *dev = ti->private;
2177
2178        return fn(ti, dev, 0, ti->len, data);
2179}
2180
2181static struct target_type origin_target = {
2182        .name    = "snapshot-origin",
2183        .version = {1, 8, 1},
2184        .module  = THIS_MODULE,
2185        .ctr     = origin_ctr,
2186        .dtr     = origin_dtr,
2187        .map     = origin_map,
2188        .resume  = origin_resume,
2189        .status  = origin_status,
2190        .merge   = origin_merge,
2191        .iterate_devices = origin_iterate_devices,
2192};
2193
2194static struct target_type snapshot_target = {
2195        .name    = "snapshot",
2196        .version = {1, 11, 1},
2197        .module  = THIS_MODULE,
2198        .ctr     = snapshot_ctr,
2199        .dtr     = snapshot_dtr,
2200        .map     = snapshot_map,
2201        .end_io  = snapshot_end_io,
2202        .preresume  = snapshot_preresume,
2203        .resume  = snapshot_resume,
2204        .status  = snapshot_status,
2205        .iterate_devices = snapshot_iterate_devices,
2206};
2207
2208static struct target_type merge_target = {
2209        .name    = dm_snapshot_merge_target_name,
2210        .version = {1, 2, 0},
2211        .module  = THIS_MODULE,
2212        .ctr     = snapshot_ctr,
2213        .dtr     = snapshot_dtr,
2214        .map     = snapshot_merge_map,
2215        .end_io  = snapshot_end_io,
2216        .presuspend = snapshot_merge_presuspend,
2217        .preresume  = snapshot_preresume,
2218        .resume  = snapshot_merge_resume,
2219        .status  = snapshot_status,
2220        .iterate_devices = snapshot_iterate_devices,
2221};
2222
2223static int __init dm_snapshot_init(void)
2224{
2225        int r;
2226
2227        r = dm_exception_store_init();
2228        if (r) {
2229                DMERR("Failed to initialize exception stores");
2230                return r;
2231        }
2232
2233        r = dm_register_target(&snapshot_target);
2234        if (r < 0) {
2235                DMERR("snapshot target register failed %d", r);
2236                goto bad_register_snapshot_target;
2237        }
2238
2239        r = dm_register_target(&origin_target);
2240        if (r < 0) {
2241                DMERR("Origin target register failed %d", r);
2242                goto bad_register_origin_target;
2243        }
2244
2245        r = dm_register_target(&merge_target);
2246        if (r < 0) {
2247                DMERR("Merge target register failed %d", r);
2248                goto bad_register_merge_target;
2249        }
2250
2251        r = init_origin_hash();
2252        if (r) {
2253                DMERR("init_origin_hash failed.");
2254                goto bad_origin_hash;
2255        }
2256
2257        exception_cache = KMEM_CACHE(dm_exception, 0);
2258        if (!exception_cache) {
2259                DMERR("Couldn't create exception cache.");
2260                r = -ENOMEM;
2261                goto bad_exception_cache;
2262        }
2263
2264        pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2265        if (!pending_cache) {
2266                DMERR("Couldn't create pending cache.");
2267                r = -ENOMEM;
2268                goto bad_pending_cache;
2269        }
2270
2271        return 0;
2272
2273bad_pending_cache:
2274        kmem_cache_destroy(exception_cache);
2275bad_exception_cache:
2276        exit_origin_hash();
2277bad_origin_hash:
2278        dm_unregister_target(&merge_target);
2279bad_register_merge_target:
2280        dm_unregister_target(&origin_target);
2281bad_register_origin_target:
2282        dm_unregister_target(&snapshot_target);
2283bad_register_snapshot_target:
2284        dm_exception_store_exit();
2285
2286        return r;
2287}
2288
2289static void __exit dm_snapshot_exit(void)
2290{
2291        dm_unregister_target(&snapshot_target);
2292        dm_unregister_target(&origin_target);
2293        dm_unregister_target(&merge_target);
2294
2295        exit_origin_hash();
2296        kmem_cache_destroy(pending_cache);
2297        kmem_cache_destroy(exception_cache);
2298
2299        dm_exception_store_exit();
2300}
2301
2302/* Module hooks */
2303module_init(dm_snapshot_init);
2304module_exit(dm_snapshot_exit);
2305
2306MODULE_DESCRIPTION(DM_NAME " snapshot target");
2307MODULE_AUTHOR("Joe Thornber");
2308MODULE_LICENSE("GPL");
2309MODULE_ALIAS("dm-snapshot-origin");
2310MODULE_ALIAS("dm-snapshot-merge");
2311
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.