kernel/drivers/md/dm-snap.c

   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.h"
  24
  25 #include "dm-exception-store.h"
  26
  27 #define DM_MSG_PREFIX "snapshots"
  28
  29 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
  30
  31 #define dm_target_is_snapshot_merge(ti) \
  32         ((ti)->type->name == dm_snapshot_merge_target_name)
  33
  34 /*
  35  * The size of the mempool used to track chunks in use.
  36  */
  37 #define MIN_IOS 256
  38
  39 #define DM_TRACKED_CHUNK_HASH_SIZE      16
  40 #define DM_TRACKED_CHUNK_HASH(x)        ((unsigned long)(x) & \
  41                                          (DM_TRACKED_CHUNK_HASH_SIZE - 1))
  42
  43 struct dm_exception_table {
  44         uint32_t hash_mask;
  45         unsigned hash_shift;
  46         struct list_head *table;
  47 };
  48
  49 struct dm_snapshot {
  50         struct rw_semaphore lock;
  51
  52         struct dm_dev *origin;
  53         struct dm_dev *cow;
  54
  55         struct dm_target *ti;
  56
  57         /* List of snapshots per Origin */
  58         struct list_head list;
  59
  60         /*
  61          * You can't use a snapshot if this is 0 (e.g. if full).
  62          * A snapshot-merge target never clears this.
  63          */
  64         int valid;
  65
  66         /*
  67          * The snapshot overflowed because of a write to the snapshot device.
  68          * We don't have to invalidate the snapshot in this case, but we need
  69          * to prevent further writes.
  70          */
  71         int snapshot_overflowed;
  72
  73         /* Origin writes don't trigger exceptions until this is set */
  74         int active;
  75
  76         atomic_t pending_exceptions_count;
  77
  78         /* Protected by "lock" */
  79         sector_t exception_start_sequence;
  80
  81         /* Protected by kcopyd single-threaded callback */
  82         sector_t exception_complete_sequence;
  83
  84         /*
  85          * A list of pending exceptions that completed out of order.
  86          * Protected by kcopyd single-threaded callback.
  87          */
  88         struct list_head out_of_order_list;
  89
  90         mempool_t *pending_pool;
  91
  92         struct dm_exception_table pending;
  93         struct dm_exception_table complete;
  94
  95         /*
  96          * pe_lock protects all pending_exception operations and access
  97          * as well as the snapshot_bios list.
  98          */
  99         spinlock_t pe_lock;
 100
 101         /* Chunks with outstanding reads */
 102         spinlock_t tracked_chunk_lock;
 103         struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
 104
 105         /* The on disk metadata handler */
 106         struct dm_exception_store *store;
 107
 108         struct dm_kcopyd_client *kcopyd_client;
 109
 110         /* Wait for events based on state_bits */
 111         unsigned long state_bits;
 112
 113         /* Range of chunks currently being merged. */
 114         chunk_t first_merging_chunk;
 115         int num_merging_chunks;
 116
 117         /*
 118          * The merge operation failed if this flag is set.
 119          * Failure modes are handled as follows:
 120          * - I/O error reading the header
 121          *      => don't load the target; abort.
 122          * - Header does not have "valid" flag set
 123          *      => use the origin; forget about the snapshot.
 124          * - I/O error when reading exceptions
 125          *      => don't load the target; abort.
 126          *         (We can't use the intermediate origin state.)
 127          * - I/O error while merging
 128          *      => stop merging; set merge_failed; process I/O normally.
 129          */
 130         int merge_failed;
 131
 132         /*
 133          * Incoming bios that overlap with chunks being merged must wait
 134          * for them to be committed.
 135          */
 136         struct bio_list bios_queued_during_merge;
 137 };
 138
 139 /*
 140  * state_bits:
 141  *   RUNNING_MERGE  - Merge operation is in progress.
 142  *   SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
 143  *                    cleared afterwards.
 144  */
 145 #define RUNNING_MERGE          0
 146 #define SHUTDOWN_MERGE         1
 147
 148 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
 149                 "A percentage of time allocated for copy on write");
 150
 151 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
 152 {
 153         return s->origin;
 154 }
 155 EXPORT_SYMBOL(dm_snap_origin);
 156
 157 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
 158 {
 159         return s->cow;
 160 }
 161 EXPORT_SYMBOL(dm_snap_cow);
 162
 163 static sector_t chunk_to_sector(struct dm_exception_store *store,
 164                                 chunk_t chunk)
 165 {
 166         return chunk << store->chunk_shift;
 167 }
 168
 169 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
 170 {
 171         /*
 172          * There is only ever one instance of a particular block
 173          * device so we can compare pointers safely.
 174          */
 175         return lhs == rhs;
 176 }
 177
 178 struct dm_snap_pending_exception {
 179         struct dm_exception e;
 180
 181         /*
 182          * Origin buffers waiting for this to complete are held
 183          * in a bio list
 184          */
 185         struct bio_list origin_bios;
 186         struct bio_list snapshot_bios;
 187
 188         /* Pointer back to snapshot context */
 189         struct dm_snapshot *snap;
 190
 191         /*
 192          * 1 indicates the exception has already been sent to
 193          * kcopyd.
 194          */
 195         int started;
 196
 197         /* There was copying error. */
 198         int copy_error;
 199
 200         /* A sequence number, it is used for in-order completion. */
 201         sector_t exception_sequence;
 202
 203         struct list_head out_of_order_entry;
 204
 205         /*
 206          * For writing a complete chunk, bypassing the copy.
 207          */
 208         struct bio *full_bio;
 209         bio_end_io_t *full_bio_end_io;
 210         void *full_bio_private;
 211 };
 212
 213 /*
 214  * Hash table mapping origin volumes to lists of snapshots and
 215  * a lock to protect it
 216  */
 217 static struct kmem_cache *exception_cache;
 218 static struct kmem_cache *pending_cache;
 219
 220 struct dm_snap_tracked_chunk {
 221         struct hlist_node node;
 222         chunk_t chunk;
 223 };
 224
 225 static void init_tracked_chunk(struct bio *bio)
 226 {
 227         struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 228         INIT_HLIST_NODE(&c->node);
 229 }
 230
 231 static bool is_bio_tracked(struct bio *bio)
 232 {
 233         struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 234         return !hlist_unhashed(&c->node);
 235 }
 236
 237 static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
 238 {
 239         struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 240
 241         c->chunk = chunk;
 242
 243         spin_lock_irq(&s->tracked_chunk_lock);
 244         hlist_add_head(&c->node,
 245                        &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
 246         spin_unlock_irq(&s->tracked_chunk_lock);
 247 }
 248
 249 static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
 250 {
 251         struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
 252         unsigned long flags;
 253
 254         spin_lock_irqsave(&s->tracked_chunk_lock, flags);
 255         hlist_del(&c->node);
 256         spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
 257 }
 258
 259 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
 260 {
 261         struct dm_snap_tracked_chunk *c;
 262         int found = 0;
 263
 264         spin_lock_irq(&s->tracked_chunk_lock);
 265
 266         hlist_for_each_entry(c,
 267             &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
 268                 if (c->chunk == chunk) {
 269                         found = 1;
 270                         break;
 271                 }
 272         }
 273
 274         spin_unlock_irq(&s->tracked_chunk_lock);
 275
 276         return found;
 277 }
 278
 279 /*
 280  * This conflicting I/O is extremely improbable in the caller,
 281  * so msleep(1) is sufficient and there is no need for a wait queue.
 282  */
 283 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
 284 {
 285         while (__chunk_is_tracked(s, chunk))
 286                 msleep(1);
 287 }
 288
 289 /*
 290  * One of these per registered origin, held in the snapshot_origins hash
 291  */
 292 struct origin {
 293         /* The origin device */
 294         struct block_device *bdev;
 295
 296         struct list_head hash_list;
 297
 298         /* List of snapshots for this origin */
 299         struct list_head snapshots;
 300 };
 301
 302 /*
 303  * This structure is allocated for each origin target
 304  */
 305 struct dm_origin {
 306         struct dm_dev *dev;
 307         struct dm_target *ti;
 308         unsigned split_boundary;
 309         struct list_head hash_list;
 310 };
 311
 312 /*
 313  * Size of the hash table for origin volumes. If we make this
 314  * the size of the minors list then it should be nearly perfect
 315  */
 316 #define ORIGIN_HASH_SIZE 256
 317 #define ORIGIN_MASK      0xFF
 318 static struct list_head *_origins;
 319 static struct list_head *_dm_origins;
 320 static struct rw_semaphore _origins_lock;
 321
 322 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
 323 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
 324 static uint64_t _pending_exceptions_done_count;
 325
 326 static int init_origin_hash(void)
 327 {
 328         int i;
 329
 330         _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
 331                            GFP_KERNEL);
 332         if (!_origins) {
 333                 DMERR("unable to allocate memory for _origins");
 334                 return -ENOMEM;
 335         }
 336         for (i = 0; i < ORIGIN_HASH_SIZE; i++)
 337                 INIT_LIST_HEAD(_origins + i);
 338
 339         _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
 340                               GFP_KERNEL);
 341         if (!_dm_origins) {
 342                 DMERR("unable to allocate memory for _dm_origins");
 343                 kfree(_origins);
 344                 return -ENOMEM;
 345         }
 346         for (i = 0; i < ORIGIN_HASH_SIZE; i++)
 347                 INIT_LIST_HEAD(_dm_origins + i);
 348
 349         init_rwsem(&_origins_lock);
 350
 351         return 0;
 352 }
 353
 354 static void exit_origin_hash(void)
 355 {
 356         kfree(_origins);
 357         kfree(_dm_origins);
 358 }
 359
 360 static unsigned origin_hash(struct block_device *bdev)
 361 {
 362         return bdev->bd_dev & ORIGIN_MASK;
 363 }
 364
 365 static struct origin *__lookup_origin(struct block_device *origin)
 366 {
 367         struct list_head *ol;
 368         struct origin *o;
 369
 370         ol = &_origins[origin_hash(origin)];
 371         list_for_each_entry (o, ol, hash_list)
 372                 if (bdev_equal(o->bdev, origin))
 373                         return o;
 374
 375         return NULL;
 376 }
 377
 378 static void __insert_origin(struct origin *o)
 379 {
 380         struct list_head *sl = &_origins[origin_hash(o->bdev)];
 381         list_add_tail(&o->hash_list, sl);
 382 }
 383
 384 static struct dm_origin *__lookup_dm_origin(struct block_device *origin)
 385 {
 386         struct list_head *ol;
 387         struct dm_origin *o;
 388
 389         ol = &_dm_origins[origin_hash(origin)];
 390         list_for_each_entry (o, ol, hash_list)
 391                 if (bdev_equal(o->dev->bdev, origin))
 392                         return o;
 393
 394         return NULL;
 395 }
 396
 397 static void __insert_dm_origin(struct dm_origin *o)
 398 {
 399         struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)];
 400         list_add_tail(&o->hash_list, sl);
 401 }
 402
 403 static void __remove_dm_origin(struct dm_origin *o)
 404 {
 405         list_del(&o->hash_list);
 406 }
 407
 408 /*
 409  * _origins_lock must be held when calling this function.
 410  * Returns number of snapshots registered using the supplied cow device, plus:
 411  * snap_src - a snapshot suitable for use as a source of exception handover
 412  * snap_dest - a snapshot capable of receiving exception handover.
 413  * snap_merge - an existing snapshot-merge target linked to the same origin.
 414  *   There can be at most one snapshot-merge target. The parameter is optional.
 415  *
 416  * Possible return values and states of snap_src and snap_dest.
 417  *   0: NULL, NULL  - first new snapshot
 418  *   1: snap_src, NULL - normal snapshot
 419  *   2: snap_src, snap_dest  - waiting for handover
 420  *   2: snap_src, NULL - handed over, waiting for old to be deleted
 421  *   1: NULL, snap_dest - source got destroyed without handover
 422  */
 423 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
 424                                         struct dm_snapshot **snap_src,
 425                                         struct dm_snapshot **snap_dest,
 426                                         struct dm_snapshot **snap_merge)
 427 {
 428         struct dm_snapshot *s;
 429         struct origin *o;
 430         int count = 0;
 431         int active;
 432
 433         o = __lookup_origin(snap->origin->bdev);
 434         if (!o)
 435                 goto out;
 436
 437         list_for_each_entry(s, &o->snapshots, list) {
 438                 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
 439                         *snap_merge = s;
 440                 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
 441                         continue;
 442
 443                 down_read(&s->lock);
 444                 active = s->active;
 445                 up_read(&s->lock);
 446
 447                 if (active) {
 448                         if (snap_src)
 449                                 *snap_src = s;
 450                 } else if (snap_dest)
 451                         *snap_dest = s;
 452
 453                 count++;
 454         }
 455
 456 out:
 457         return count;
 458 }
 459
 460 /*
 461  * On success, returns 1 if this snapshot is a handover destination,
 462  * otherwise returns 0.
 463  */
 464 static int __validate_exception_handover(struct dm_snapshot *snap)
 465 {
 466         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
 467         struct dm_snapshot *snap_merge = NULL;
 468
 469         /* Does snapshot need exceptions handed over to it? */
 470         if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
 471                                           &snap_merge) == 2) ||
 472             snap_dest) {
 473                 snap->ti->error = "Snapshot cow pairing for exception "
 474                                   "table handover failed";
 475                 return -EINVAL;
 476         }
 477
 478         /*
 479          * If no snap_src was found, snap cannot become a handover
 480          * destination.
 481          */
 482         if (!snap_src)
 483                 return 0;
 484
 485         /*
 486          * Non-snapshot-merge handover?
 487          */
 488         if (!dm_target_is_snapshot_merge(snap->ti))
 489                 return 1;
 490
 491         /*
 492          * Do not allow more than one merging snapshot.
 493          */
 494         if (snap_merge) {
 495                 snap->ti->error = "A snapshot is already merging.";
 496                 return -EINVAL;
 497         }
 498
 499         if (!snap_src->store->type->prepare_merge ||
 500             !snap_src->store->type->commit_merge) {
 501                 snap->ti->error = "Snapshot exception store does not "
 502                                   "support snapshot-merge.";
 503                 return -EINVAL;
 504         }
 505
 506         return 1;
 507 }
 508
 509 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
 510 {
 511         struct dm_snapshot *l;
 512
 513         /* Sort the list according to chunk size, largest-first smallest-last */
 514         list_for_each_entry(l, &o->snapshots, list)
 515                 if (l->store->chunk_size < s->store->chunk_size)
 516                         break;
 517         list_add_tail(&s->list, &l->list);
 518 }
 519
 520 /*
 521  * Make a note of the snapshot and its origin so we can look it
 522  * up when the origin has a write on it.
 523  *
 524  * Also validate snapshot exception store handovers.
 525  * On success, returns 1 if this registration is a handover destination,
 526  * otherwise returns 0.
 527  */
 528 static int register_snapshot(struct dm_snapshot *snap)
 529 {
 530         struct origin *o, *new_o = NULL;
 531         struct block_device *bdev = snap->origin->bdev;
 532         int r = 0;
 533
 534         new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
 535         if (!new_o)
 536                 return -ENOMEM;
 537
 538         down_write(&_origins_lock);
 539
 540         r = __validate_exception_handover(snap);
 541         if (r < 0) {
 542                 kfree(new_o);
 543                 goto out;
 544         }
 545
 546         o = __lookup_origin(bdev);
 547         if (o)
 548                 kfree(new_o);
 549         else {
 550                 /* New origin */
 551                 o = new_o;
 552
 553                 /* Initialise the struct */
 554                 INIT_LIST_HEAD(&o->snapshots);
 555                 o->bdev = bdev;
 556
 557                 __insert_origin(o);
 558         }
 559
 560         __insert_snapshot(o, snap);
 561
 562 out:
 563         up_write(&_origins_lock);
 564
 565         return r;
 566 }
 567
 568 /*
 569  * Move snapshot to correct place in list according to chunk size.
 570  */
 571 static void reregister_snapshot(struct dm_snapshot *s)
 572 {
 573         struct block_device *bdev = s->origin->bdev;
 574
 575         down_write(&_origins_lock);
 576
 577         list_del(&s->list);
 578         __insert_snapshot(__lookup_origin(bdev), s);
 579
 580         up_write(&_origins_lock);
 581 }
 582
 583 static void unregister_snapshot(struct dm_snapshot *s)
 584 {
 585         struct origin *o;
 586
 587         down_write(&_origins_lock);
 588         o = __lookup_origin(s->origin->bdev);
 589
 590         list_del(&s->list);
 591         if (o && list_empty(&o->snapshots)) {
 592                 list_del(&o->hash_list);
 593                 kfree(o);
 594         }
 595
 596         up_write(&_origins_lock);
 597 }
 598
 599 /*
 600  * Implementation of the exception hash tables.
 601  * The lowest hash_shift bits of the chunk number are ignored, allowing
 602  * some consecutive chunks to be grouped together.
 603  */
 604 static int dm_exception_table_init(struct dm_exception_table *et,
 605                                    uint32_t size, unsigned hash_shift)
 606 {
 607         unsigned int i;
 608
 609         et->hash_shift = hash_shift;
 610         et->hash_mask = size - 1;
 611         et->table = dm_vcalloc(size, sizeof(struct list_head));
 612         if (!et->table)
 613                 return -ENOMEM;
 614
 615         for (i = 0; i < size; i++)
 616                 INIT_LIST_HEAD(et->table + i);
 617
 618         return 0;
 619 }
 620
 621 static void dm_exception_table_exit(struct dm_exception_table *et,
 622                                     struct kmem_cache *mem)
 623 {
 624         struct list_head *slot;
 625         struct dm_exception *ex, *next;
 626         int i, size;
 627
 628         size = et->hash_mask + 1;
 629         for (i = 0; i < size; i++) {
 630                 slot = et->table + i;
 631
 632                 list_for_each_entry_safe (ex, next, slot, hash_list)
 633                         kmem_cache_free(mem, ex);
 634         }
 635
 636         vfree(et->table);
 637 }
 638
 639 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
 640 {
 641         return (chunk >> et->hash_shift) & et->hash_mask;
 642 }
 643
 644 static void dm_remove_exception(struct dm_exception *e)
 645 {
 646         list_del(&e->hash_list);
 647 }
 648
 649 /*
 650  * Return the exception data for a sector, or NULL if not
 651  * remapped.
 652  */
 653 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
 654                                                 chunk_t chunk)
 655 {
 656         struct list_head *slot;
 657         struct dm_exception *e;
 658
 659         slot = &et->table[exception_hash(et, chunk)];
 660         list_for_each_entry (e, slot, hash_list)
 661                 if (chunk >= e->old_chunk &&
 662                     chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
 663                         return e;
 664
 665         return NULL;
 666 }
 667
 668 static struct dm_exception *alloc_completed_exception(gfp_t gfp)
 669 {
 670         struct dm_exception *e;
 671
 672         e = kmem_cache_alloc(exception_cache, gfp);
 673         if (!e && gfp == GFP_NOIO)
 674                 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
 675
 676         return e;
 677 }
 678
 679 static void free_completed_exception(struct dm_exception *e)
 680 {
 681         kmem_cache_free(exception_cache, e);
 682 }
 683
 684 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
 685 {
 686         struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
 687                                                              GFP_NOIO);
 688
 689         atomic_inc(&s->pending_exceptions_count);
 690         pe->snap = s;
 691
 692         return pe;
 693 }
 694
 695 static void free_pending_exception(struct dm_snap_pending_exception *pe)
 696 {
 697         struct dm_snapshot *s = pe->snap;
 698
 699         mempool_free(pe, s->pending_pool);
 700         smp_mb__before_atomic();
 701         atomic_dec(&s->pending_exceptions_count);
 702 }
 703
 704 static void dm_insert_exception(struct dm_exception_table *eh,
 705                                 struct dm_exception *new_e)
 706 {
 707         struct list_head *l;
 708         struct dm_exception *e = NULL;
 709
 710         l = &eh->table[exception_hash(eh, new_e->old_chunk)];
 711
 712         /* Add immediately if this table doesn't support consecutive chunks */
 713         if (!eh->hash_shift)
 714                 goto out;
 715
 716         /* List is ordered by old_chunk */
 717         list_for_each_entry_reverse(e, l, hash_list) {
 718                 /* Insert after an existing chunk? */
 719                 if (new_e->old_chunk == (e->old_chunk +
 720                                          dm_consecutive_chunk_count(e) + 1) &&
 721                     new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
 722                                          dm_consecutive_chunk_count(e) + 1)) {
 723                         dm_consecutive_chunk_count_inc(e);
 724                         free_completed_exception(new_e);
 725                         return;
 726                 }
 727
 728                 /* Insert before an existing chunk? */
 729                 if (new_e->old_chunk == (e->old_chunk - 1) &&
 730                     new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
 731                         dm_consecutive_chunk_count_inc(e);
 732                         e->old_chunk--;
 733                         e->new_chunk--;
 734                         free_completed_exception(new_e);
 735                         return;
 736                 }
 737
 738                 if (new_e->old_chunk > e->old_chunk)
 739                         break;
 740         }
 741
 742 out:
 743         list_add(&new_e->hash_list, e ? &e->hash_list : l);
 744 }
 745
 746 /*
 747  * Callback used by the exception stores to load exceptions when
 748  * initialising.
 749  */
 750 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
 751 {
 752         struct dm_snapshot *s = context;
 753         struct dm_exception *e;
 754
 755         e = alloc_completed_exception(GFP_KERNEL);
 756         if (!e)
 757                 return -ENOMEM;
 758
 759         e->old_chunk = old;
 760
 761         /* Consecutive_count is implicitly initialised to zero */
 762         e->new_chunk = new;
 763
 764         dm_insert_exception(&s->complete, e);
 765
 766         return 0;
 767 }
 768
 769 /*
 770  * Return a minimum chunk size of all snapshots that have the specified origin.
 771  * Return zero if the origin has no snapshots.
 772  */
 773 static uint32_t __minimum_chunk_size(struct origin *o)
 774 {
 775         struct dm_snapshot *snap;
 776         unsigned chunk_size = 0;
 777
 778         if (o)
 779                 list_for_each_entry(snap, &o->snapshots, list)
 780                         chunk_size = min_not_zero(chunk_size,
 781                                                   snap->store->chunk_size);
 782
 783         return (uint32_t) chunk_size;
 784 }
 785
 786 /*
 787  * Hard coded magic.
 788  */
 789 static int calc_max_buckets(void)
 790 {
 791         /* use a fixed size of 2MB */
 792         unsigned long mem = 2 * 1024 * 1024;
 793         mem /= sizeof(struct list_head);
 794
 795         return mem;
 796 }
 797
 798 /*
 799  * Allocate room for a suitable hash table.
 800  */
 801 static int init_hash_tables(struct dm_snapshot *s)
 802 {
 803         sector_t hash_size, cow_dev_size, max_buckets;
 804
 805         /*
 806          * Calculate based on the size of the original volume or
 807          * the COW volume...
 808          */
 809         cow_dev_size = get_dev_size(s->cow->bdev);
 810         max_buckets = calc_max_buckets();
 811
 812         hash_size = cow_dev_size >> s->store->chunk_shift;
 813         hash_size = min(hash_size, max_buckets);
 814
 815         if (hash_size < 64)
 816                 hash_size = 64;
 817         hash_size = rounddown_pow_of_two(hash_size);
 818         if (dm_exception_table_init(&s->complete, hash_size,
 819                                     DM_CHUNK_CONSECUTIVE_BITS))
 820                 return -ENOMEM;
 821
 822         /*
 823          * Allocate hash table for in-flight exceptions
 824          * Make this smaller than the real hash table
 825          */
 826         hash_size >>= 3;
 827         if (hash_size < 64)
 828                 hash_size = 64;
 829
 830         if (dm_exception_table_init(&s->pending, hash_size, 0)) {
 831                 dm_exception_table_exit(&s->complete, exception_cache);
 832                 return -ENOMEM;
 833         }
 834
 835         return 0;
 836 }
 837
 838 static void merge_shutdown(struct dm_snapshot *s)
 839 {
 840         clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
 841         smp_mb__after_atomic();
 842         wake_up_bit(&s->state_bits, RUNNING_MERGE);
 843 }
 844
 845 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
 846 {
 847         s->first_merging_chunk = 0;
 848         s->num_merging_chunks = 0;
 849
 850         return bio_list_get(&s->bios_queued_during_merge);
 851 }
 852
 853 /*
 854  * Remove one chunk from the index of completed exceptions.
 855  */
 856 static int __remove_single_exception_chunk(struct dm_snapshot *s,
 857                                            chunk_t old_chunk)
 858 {
 859         struct dm_exception *e;
 860
 861         e = dm_lookup_exception(&s->complete, old_chunk);
 862         if (!e) {
 863                 DMERR("Corruption detected: exception for block %llu is "
 864                       "on disk but not in memory",
 865                       (unsigned long long)old_chunk);
 866                 return -EINVAL;
 867         }
 868
 869         /*
 870          * If this is the only chunk using this exception, remove exception.
 871          */
 872         if (!dm_consecutive_chunk_count(e)) {
 873                 dm_remove_exception(e);
 874                 free_completed_exception(e);
 875                 return 0;
 876         }
 877
 878         /*
 879          * The chunk may be either at the beginning or the end of a
 880          * group of consecutive chunks - never in the middle.  We are
 881          * removing chunks in the opposite order to that in which they
 882          * were added, so this should always be true.
 883          * Decrement the consecutive chunk counter and adjust the
 884          * starting point if necessary.
 885          */
 886         if (old_chunk == e->old_chunk) {
 887                 e->old_chunk++;
 888                 e->new_chunk++;
 889         } else if (old_chunk != e->old_chunk +
 890                    dm_consecutive_chunk_count(e)) {
 891                 DMERR("Attempt to merge block %llu from the "
 892                       "middle of a chunk range [%llu - %llu]",
 893                       (unsigned long long)old_chunk,
 894                       (unsigned long long)e->old_chunk,
 895                       (unsigned long long)
 896                       e->old_chunk + dm_consecutive_chunk_count(e));
 897                 return -EINVAL;
 898         }
 899
 900         dm_consecutive_chunk_count_dec(e);
 901
 902         return 0;
 903 }
 904
 905 static void flush_bios(struct bio *bio);
 906
 907 static int remove_single_exception_chunk(struct dm_snapshot *s)
 908 {
 909         struct bio *b = NULL;
 910         int r;
 911         chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
 912
 913         down_write(&s->lock);
 914
 915         /*
 916          * Process chunks (and associated exceptions) in reverse order
 917          * so that dm_consecutive_chunk_count_dec() accounting works.
 918          */
 919         do {
 920                 r = __remove_single_exception_chunk(s, old_chunk);
 921                 if (r)
 922                         goto out;
 923         } while (old_chunk-- > s->first_merging_chunk);
 924
 925         b = __release_queued_bios_after_merge(s);
 926
 927 out:
 928         up_write(&s->lock);
 929         if (b)
 930                 flush_bios(b);
 931
 932         return r;
 933 }
 934
 935 static int origin_write_extent(struct dm_snapshot *merging_snap,
 936                                sector_t sector, unsigned chunk_size);
 937
 938 static void merge_callback(int read_err, unsigned long write_err,
 939                            void *context);
 940
 941 static uint64_t read_pending_exceptions_done_count(void)
 942 {
 943         uint64_t pending_exceptions_done;
 944
 945         spin_lock(&_pending_exceptions_done_spinlock);
 946         pending_exceptions_done = _pending_exceptions_done_count;
 947         spin_unlock(&_pending_exceptions_done_spinlock);
 948
 949         return pending_exceptions_done;
 950 }
 951
 952 static void increment_pending_exceptions_done_count(void)
 953 {
 954         spin_lock(&_pending_exceptions_done_spinlock);
 955         _pending_exceptions_done_count++;
 956         spin_unlock(&_pending_exceptions_done_spinlock);
 957
 958         wake_up_all(&_pending_exceptions_done);
 959 }
 960
 961 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
 962 {
 963         int i, linear_chunks;
 964         chunk_t old_chunk, new_chunk;
 965         struct dm_io_region src, dest;
 966         sector_t io_size;
 967         uint64_t previous_count;
 968
 969         BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
 970         if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
 971                 goto shut;
 972
 973         /*
 974          * valid flag never changes during merge, so no lock required.
 975          */
 976         if (!s->valid) {
 977                 DMERR("Snapshot is invalid: can't merge");
 978                 goto shut;
 979         }
 980
 981         linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
 982                                                       &new_chunk);
 983         if (linear_chunks <= 0) {
 984                 if (linear_chunks < 0) {
 985                         DMERR("Read error in exception store: "
 986                               "shutting down merge");
 987                         down_write(&s->lock);
 988                         s->merge_failed = 1;
 989                         up_write(&s->lock);
 990                 }
 991                 goto shut;
 992         }
 993
 994         /* Adjust old_chunk and new_chunk to reflect start of linear region */
 995         old_chunk = old_chunk + 1 - linear_chunks;
 996         new_chunk = new_chunk + 1 - linear_chunks;
 997
 998         /*
 999          * Use one (potentially large) I/O to copy all 'linear_chunks'
1000          * from the exception store to the origin
1001          */
1002         io_size = linear_chunks * s->store->chunk_size;
1003
1004         dest.bdev = s->origin->bdev;
1005         dest.sector = chunk_to_sector(s->store, old_chunk);
1006         dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
1007
1008         src.bdev = s->cow->bdev;
1009         src.sector = chunk_to_sector(s->store, new_chunk);
1010         src.count = dest.count;
1011
1012         /*
1013          * Reallocate any exceptions needed in other snapshots then
1014          * wait for the pending exceptions to complete.
1015          * Each time any pending exception (globally on the system)
1016          * completes we are woken and repeat the process to find out
1017          * if we can proceed.  While this may not seem a particularly
1018          * efficient algorithm, it is not expected to have any
1019          * significant impact on performance.
1020          */
1021         previous_count = read_pending_exceptions_done_count();
1022         while (origin_write_extent(s, dest.sector, io_size)) {
1023                 wait_event(_pending_exceptions_done,
1024                            (read_pending_exceptions_done_count() !=
1025                             previous_count));
1026                 /* Retry after the wait, until all exceptions are done. */
1027                 previous_count = read_pending_exceptions_done_count();
1028         }
1029
1030         down_write(&s->lock);
1031         s->first_merging_chunk = old_chunk;
1032         s->num_merging_chunks = linear_chunks;
1033         up_write(&s->lock);
1034
1035         /* Wait until writes to all 'linear_chunks' drain */
1036         for (i = 0; i < linear_chunks; i++)
1037                 __check_for_conflicting_io(s, old_chunk + i);
1038
1039         dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
1040         return;
1041
1042 shut:
1043         merge_shutdown(s);
1044 }
1045
1046 static void error_bios(struct bio *bio);
1047
1048 static void merge_callback(int read_err, unsigned long write_err, void *context)
1049 {
1050         struct dm_snapshot *s = context;
1051         struct bio *b = NULL;
1052
1053         if (read_err || write_err) {
1054                 if (read_err)
1055                         DMERR("Read error: shutting down merge.");
1056                 else
1057                         DMERR("Write error: shutting down merge.");
1058                 goto shut;
1059         }
1060
1061         if (s->store->type->commit_merge(s->store,
1062                                          s->num_merging_chunks) < 0) {
1063                 DMERR("Write error in exception store: shutting down merge");
1064                 goto shut;
1065         }
1066
1067         if (remove_single_exception_chunk(s) < 0)
1068                 goto shut;
1069
1070         snapshot_merge_next_chunks(s);
1071
1072         return;
1073
1074 shut:
1075         down_write(&s->lock);
1076         s->merge_failed = 1;
1077         b = __release_queued_bios_after_merge(s);
1078         up_write(&s->lock);
1079         error_bios(b);
1080
1081         merge_shutdown(s);
1082 }
1083
1084 static void start_merge(struct dm_snapshot *s)
1085 {
1086         if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1087                 snapshot_merge_next_chunks(s);
1088 }
1089
1090 /*
1091  * Stop the merging process and wait until it finishes.
1092  */
1093 static void stop_merge(struct dm_snapshot *s)
1094 {
1095         set_bit(SHUTDOWN_MERGE, &s->state_bits);
1096         wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
1097         clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1098 }
1099
1100 /*
1101  * Construct a snapshot mapping: <origin_dev> <COW-dev> <p|po|n> <chunk-size>
1102  */
1103 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1104 {
1105         struct dm_snapshot *s;
1106         int i;
1107         int r = -EINVAL;
1108         char *origin_path, *cow_path;
1109         dev_t origin_dev, cow_dev;
1110         unsigned args_used, num_flush_bios = 1;
1111         fmode_t origin_mode = FMODE_READ;
1112
1113         if (argc != 4) {
1114                 ti->error = "requires exactly 4 arguments";
1115                 r = -EINVAL;
1116                 goto bad;
1117         }
1118
1119         if (dm_target_is_snapshot_merge(ti)) {
1120                 num_flush_bios = 2;
1121                 origin_mode = FMODE_WRITE;
1122         }
1123
1124         s = kmalloc(sizeof(*s), GFP_KERNEL);
1125         if (!s) {
1126                 ti->error = "Cannot allocate private snapshot structure";
1127                 r = -ENOMEM;
1128                 goto bad;
1129         }
1130
1131         origin_path = argv[0];
1132         argv++;
1133         argc--;
1134
1135         r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1136         if (r) {
1137                 ti->error = "Cannot get origin device";
1138                 goto bad_origin;
1139         }
1140         origin_dev = s->origin->bdev->bd_dev;
1141
1142         cow_path = argv[0];
1143         argv++;
1144         argc--;
1145
1146         cow_dev = dm_get_dev_t(cow_path);
1147         if (cow_dev && cow_dev == origin_dev) {
1148                 ti->error = "COW device cannot be the same as origin device";
1149                 r = -EINVAL;
1150                 goto bad_cow;
1151         }
1152
1153         r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1154         if (r) {
1155                 ti->error = "Cannot get COW device";
1156                 goto bad_cow;
1157         }
1158
1159         r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1160         if (r) {
1161                 ti->error = "Couldn't create exception store";
1162                 r = -EINVAL;
1163                 goto bad_store;
1164         }
1165
1166         argv += args_used;
1167         argc -= args_used;
1168
1169         s->ti = ti;
1170         s->valid = 1;
1171         s->snapshot_overflowed = 0;
1172         s->active = 0;
1173         atomic_set(&s->pending_exceptions_count, 0);
1174         s->exception_start_sequence = 0;
1175         s->exception_complete_sequence = 0;
1176         INIT_LIST_HEAD(&s->out_of_order_list);
1177         init_rwsem(&s->lock);
1178         INIT_LIST_HEAD(&s->list);
1179         spin_lock_init(&s->pe_lock);
1180         s->state_bits = 0;
1181         s->merge_failed = 0;
1182         s->first_merging_chunk = 0;
1183         s->num_merging_chunks = 0;
1184         bio_list_init(&s->bios_queued_during_merge);
1185
1186         /* Allocate hash table for COW data */
1187         if (init_hash_tables(s)) {
1188                 ti->error = "Unable to allocate hash table space";
1189                 r = -ENOMEM;
1190                 goto bad_hash_tables;
1191         }
1192
1193         s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1194         if (IS_ERR(s->kcopyd_client)) {
1195                 r = PTR_ERR(s->kcopyd_client);
1196                 ti->error = "Could not create kcopyd client";
1197                 goto bad_kcopyd;
1198         }
1199
1200         s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1201         if (!s->pending_pool) {
1202                 ti->error = "Could not allocate mempool for pending exceptions";
1203                 r = -ENOMEM;
1204                 goto bad_pending_pool;
1205         }
1206
1207         for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1208                 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1209
1210         spin_lock_init(&s->tracked_chunk_lock);
1211
1212         ti->private = s;
1213         ti->num_flush_bios = num_flush_bios;
1214         ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1215
1216         /* Add snapshot to the list of snapshots for this origin */
1217         /* Exceptions aren't triggered till snapshot_resume() is called */
1218         r = register_snapshot(s);
1219         if (r == -ENOMEM) {
1220                 ti->error = "Snapshot origin struct allocation failed";
1221                 goto bad_load_and_register;
1222         } else if (r < 0) {
1223                 /* invalid handover, register_snapshot has set ti->error */
1224                 goto bad_load_and_register;
1225         }
1226
1227         /*
1228          * Metadata must only be loaded into one table at once, so skip this
1229          * if metadata will be handed over during resume.
1230          * Chunk size will be set during the handover - set it to zero to
1231          * ensure it's ignored.
1232          */
1233         if (r > 0) {
1234                 s->store->chunk_size = 0;
1235                 return 0;
1236         }
1237
1238         r = s->store->type->read_metadata(s->store, dm_add_exception,
1239                                           (void *)s);
1240         if (r < 0) {
1241                 ti->error = "Failed to read snapshot metadata";
1242                 goto bad_read_metadata;
1243         } else if (r > 0) {
1244                 s->valid = 0;
1245                 DMWARN("Snapshot is marked invalid.");
1246         }
1247
1248         if (!s->store->chunk_size) {
1249                 ti->error = "Chunk size not set";
1250                 goto bad_read_metadata;
1251         }
1252
1253         r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1254         if (r)
1255                 goto bad_read_metadata;
1256
1257         return 0;
1258
1259 bad_read_metadata:
1260         unregister_snapshot(s);
1261
1262 bad_load_and_register:
1263         mempool_destroy(s->pending_pool);
1264
1265 bad_pending_pool:
1266         dm_kcopyd_client_destroy(s->kcopyd_client);
1267
1268 bad_kcopyd:
1269         dm_exception_table_exit(&s->pending, pending_cache);
1270         dm_exception_table_exit(&s->complete, exception_cache);
1271
1272 bad_hash_tables:
1273         dm_exception_store_destroy(s->store);
1274
1275 bad_store:
1276         dm_put_device(ti, s->cow);
1277
1278 bad_cow:
1279         dm_put_device(ti, s->origin);
1280
1281 bad_origin:
1282         kfree(s);
1283
1284 bad:
1285         return r;
1286 }
1287
1288 static void __free_exceptions(struct dm_snapshot *s)
1289 {
1290         dm_kcopyd_client_destroy(s->kcopyd_client);
1291         s->kcopyd_client = NULL;
1292
1293         dm_exception_table_exit(&s->pending, pending_cache);
1294         dm_exception_table_exit(&s->complete, exception_cache);
1295 }
1296
1297 static void __handover_exceptions(struct dm_snapshot *snap_src,
1298                                   struct dm_snapshot *snap_dest)
1299 {
1300         union {
1301                 struct dm_exception_table table_swap;
1302                 struct dm_exception_store *store_swap;
1303         } u;
1304
1305         /*
1306          * Swap all snapshot context information between the two instances.
1307          */
1308         u.table_swap = snap_dest->complete;
1309         snap_dest->complete = snap_src->complete;
1310         snap_src->complete = u.table_swap;
1311
1312         u.store_swap = snap_dest->store;
1313         snap_dest->store = snap_src->store;
1314         snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow;
1315         snap_src->store = u.store_swap;
1316
1317         snap_dest->store->snap = snap_dest;
1318         snap_src->store->snap = snap_src;
1319
1320         snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1321         snap_dest->valid = snap_src->valid;
1322         snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed;
1323
1324         /*
1325          * Set source invalid to ensure it receives no further I/O.
1326          */
1327         snap_src->valid = 0;
1328 }
1329
1330 static void snapshot_dtr(struct dm_target *ti)
1331 {
1332 #ifdef CONFIG_DM_DEBUG
1333         int i;
1334 #endif
1335         struct dm_snapshot *s = ti->private;
1336         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1337
1338         down_read(&_origins_lock);
1339         /* Check whether exception handover must be cancelled */
1340         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1341         if (snap_src && snap_dest && (s == snap_src)) {
1342                 down_write(&snap_dest->lock);
1343                 snap_dest->valid = 0;
1344                 up_write(&snap_dest->lock);
1345                 DMERR("Cancelling snapshot handover.");
1346         }
1347         up_read(&_origins_lock);
1348
1349         if (dm_target_is_snapshot_merge(ti))
1350                 stop_merge(s);
1351
1352         /* Prevent further origin writes from using this snapshot. */
1353         /* After this returns there can be no new kcopyd jobs. */
1354         unregister_snapshot(s);
1355
1356         while (atomic_read(&s->pending_exceptions_count))
1357                 msleep(1);
1358         /*
1359          * Ensure instructions in mempool_destroy aren't reordered
1360          * before atomic_read.
1361          */
1362         smp_mb();
1363
1364 #ifdef CONFIG_DM_DEBUG
1365         for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1366                 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1367 #endif
1368
1369         __free_exceptions(s);
1370
1371         mempool_destroy(s->pending_pool);
1372
1373         dm_exception_store_destroy(s->store);
1374
1375         dm_put_device(ti, s->cow);
1376
1377         dm_put_device(ti, s->origin);
1378
1379         kfree(s);
1380 }
1381
1382 /*
1383  * Flush a list of buffers.
1384  */
1385 static void flush_bios(struct bio *bio)
1386 {
1387         struct bio *n;
1388
1389         while (bio) {
1390                 n = bio->bi_next;
1391                 bio->bi_next = NULL;
1392                 generic_make_request(bio);
1393                 bio = n;
1394         }
1395 }
1396
1397 static int do_origin(struct dm_dev *origin, struct bio *bio);
1398
1399 /*
1400  * Flush a list of buffers.
1401  */
1402 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1403 {
1404         struct bio *n;
1405         int r;
1406
1407         while (bio) {
1408                 n = bio->bi_next;
1409                 bio->bi_next = NULL;
1410                 r = do_origin(s->origin, bio);
1411                 if (r == DM_MAPIO_REMAPPED)
1412                         generic_make_request(bio);
1413                 bio = n;
1414         }
1415 }
1416
1417 /*
1418  * Error a list of buffers.
1419  */
1420 static void error_bios(struct bio *bio)
1421 {
1422         struct bio *n;
1423
1424         while (bio) {
1425                 n = bio->bi_next;
1426                 bio->bi_next = NULL;
1427                 bio_io_error(bio);
1428                 bio = n;
1429         }
1430 }
1431
1432 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1433 {
1434         if (!s->valid)
1435                 return;
1436
1437         if (err == -EIO)
1438                 DMERR("Invalidating snapshot: Error reading/writing.");
1439         else if (err == -ENOMEM)
1440                 DMERR("Invalidating snapshot: Unable to allocate exception.");
1441
1442         if (s->store->type->drop_snapshot)
1443                 s->store->type->drop_snapshot(s->store);
1444
1445         s->valid = 0;
1446
1447         dm_table_event(s->ti->table);
1448 }
1449
1450 static void pending_complete(void *context, int success)
1451 {
1452         struct dm_snap_pending_exception *pe = context;
1453         struct dm_exception *e;
1454         struct dm_snapshot *s = pe->snap;
1455         struct bio *origin_bios = NULL;
1456         struct bio *snapshot_bios = NULL;
1457         struct bio *full_bio = NULL;
1458         int error = 0;
1459
1460         if (!success) {
1461                 /* Read/write error - snapshot is unusable */
1462                 down_write(&s->lock);
1463                 __invalidate_snapshot(s, -EIO);
1464                 error = 1;
1465                 goto out;
1466         }
1467
1468         e = alloc_completed_exception(GFP_NOIO);
1469         if (!e) {
1470                 down_write(&s->lock);
1471                 __invalidate_snapshot(s, -ENOMEM);
1472                 error = 1;
1473                 goto out;
1474         }
1475         *e = pe->e;
1476
1477         down_write(&s->lock);
1478         if (!s->valid) {
1479                 free_completed_exception(e);
1480                 error = 1;
1481                 goto out;
1482         }
1483
1484         /* Check for conflicting reads */
1485         __check_for_conflicting_io(s, pe->e.old_chunk);
1486
1487         /*
1488          * Add a proper exception, and remove the
1489          * in-flight exception from the list.
1490          */
1491         dm_insert_exception(&s->complete, e);
1492
1493 out:
1494         dm_remove_exception(&pe->e);
1495         snapshot_bios = bio_list_get(&pe->snapshot_bios);
1496         origin_bios = bio_list_get(&pe->origin_bios);
1497         full_bio = pe->full_bio;
1498         if (full_bio) {
1499                 full_bio->bi_end_io = pe->full_bio_end_io;
1500                 full_bio->bi_private = pe->full_bio_private;
1501         }
1502         increment_pending_exceptions_done_count();
1503
1504         up_write(&s->lock);
1505
1506         /* Submit any pending write bios */
1507         if (error) {
1508                 if (full_bio)
1509                         bio_io_error(full_bio);
1510                 error_bios(snapshot_bios);
1511         } else {
1512                 if (full_bio)
1513                         bio_endio(full_bio);
1514                 flush_bios(snapshot_bios);
1515         }
1516
1517         retry_origin_bios(s, origin_bios);
1518
1519         free_pending_exception(pe);
1520 }
1521
1522 static void complete_exception(struct dm_snap_pending_exception *pe)
1523 {
1524         struct dm_snapshot *s = pe->snap;
1525
1526         /* Update the metadata if we are persistent */
1527         s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error,
1528                                          pending_complete, pe);
1529 }
1530
1531 /*
1532  * Called when the copy I/O has finished.  kcopyd actually runs
1533  * this code so don't block.
1534  */
1535 static void copy_callback(int read_err, unsigned long write_err, void *context)
1536 {
1537         struct dm_snap_pending_exception *pe = context;
1538         struct dm_snapshot *s = pe->snap;
1539
1540         pe->copy_error = read_err || write_err;
1541
1542         if (pe->exception_sequence == s->exception_complete_sequence) {
1543                 s->exception_complete_sequence++;
1544                 complete_exception(pe);
1545
1546                 while (!list_empty(&s->out_of_order_list)) {
1547                         pe = list_entry(s->out_of_order_list.next,
1548                                         struct dm_snap_pending_exception, out_of_order_entry);
1549                         if (pe->exception_sequence != s->exception_complete_sequence)
1550                                 break;
1551                         s->exception_complete_sequence++;
1552                         list_del(&pe->out_of_order_entry);
1553                         complete_exception(pe);
1554                 }
1555         } else {
1556                 struct list_head *lh;
1557                 struct dm_snap_pending_exception *pe2;
1558
1559                 list_for_each_prev(lh, &s->out_of_order_list) {
1560                         pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
1561                         if (pe2->exception_sequence < pe->exception_sequence)
1562                                 break;
1563                 }
1564                 list_add(&pe->out_of_order_entry, lh);
1565         }
1566 }
1567
1568 /*
1569  * Dispatches the copy operation to kcopyd.
1570  */
1571 static void start_copy(struct dm_snap_pending_exception *pe)
1572 {
1573         struct dm_snapshot *s = pe->snap;
1574         struct dm_io_region src, dest;
1575         struct block_device *bdev = s->origin->bdev;
1576         sector_t dev_size;
1577
1578         dev_size = get_dev_size(bdev);
1579
1580         src.bdev = bdev;
1581         src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1582         src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1583
1584         dest.bdev = s->cow->bdev;
1585         dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1586         dest.count = src.count;
1587
1588         /* Hand over to kcopyd */
1589         dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1590 }
1591
1592 static void full_bio_end_io(struct bio *bio)
1593 {
1594         void *callback_data = bio->bi_private;
1595
1596         dm_kcopyd_do_callback(callback_data, 0, bio->bi_error ? 1 : 0);
1597 }
1598
1599 static void start_full_bio(struct dm_snap_pending_exception *pe,
1600                            struct bio *bio)
1601 {
1602         struct dm_snapshot *s = pe->snap;
1603         void *callback_data;
1604
1605         pe->full_bio = bio;
1606         pe->full_bio_end_io = bio->bi_end_io;
1607         pe->full_bio_private = bio->bi_private;
1608
1609         callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1610                                                    copy_callback, pe);
1611
1612         bio->bi_end_io = full_bio_end_io;
1613         bio->bi_private = callback_data;
1614
1615         generic_make_request(bio);
1616 }
1617
1618 static struct dm_snap_pending_exception *
1619 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1620 {
1621         struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1622
1623         if (!e)
1624                 return NULL;
1625
1626         return container_of(e, struct dm_snap_pending_exception, e);
1627 }
1628
1629 /*
1630  * Looks to see if this snapshot already has a pending exception
1631  * for this chunk, otherwise it allocates a new one and inserts
1632  * it into the pending table.
1633  *
1634  * NOTE: a write lock must be held on snap->lock before calling
1635  * this.
1636  */
1637 static struct dm_snap_pending_exception *
1638 __find_pending_exception(struct dm_snapshot *s,
1639                          struct dm_snap_pending_exception *pe, chunk_t chunk)
1640 {
1641         struct dm_snap_pending_exception *pe2;
1642
1643         pe2 = __lookup_pending_exception(s, chunk);
1644         if (pe2) {
1645                 free_pending_exception(pe);
1646                 return pe2;
1647         }
1648
1649         pe->e.old_chunk = chunk;
1650         bio_list_init(&pe->origin_bios);
1651         bio_list_init(&pe->snapshot_bios);
1652         pe->started = 0;
1653         pe->full_bio = NULL;
1654
1655         if (s->store->type->prepare_exception(s->store, &pe->e)) {
1656                 free_pending_exception(pe);
1657                 return NULL;
1658         }
1659
1660         pe->exception_sequence = s->exception_start_sequence++;
1661
1662         dm_insert_exception(&s->pending, &pe->e);
1663
1664         return pe;
1665 }
1666
1667 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1668                             struct bio *bio, chunk_t chunk)
1669 {
1670         bio->bi_bdev = s->cow->bdev;
1671         bio->bi_iter.bi_sector =
1672                 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1673                                 (chunk - e->old_chunk)) +
1674                 (bio->bi_iter.bi_sector & s->store->chunk_mask);
1675 }
1676
1677 static int snapshot_map(struct dm_target *ti, struct bio *bio)
1678 {
1679         struct dm_exception *e;
1680         struct dm_snapshot *s = ti->private;
1681         int r = DM_MAPIO_REMAPPED;
1682         chunk_t chunk;
1683         struct dm_snap_pending_exception *pe = NULL;
1684
1685         init_tracked_chunk(bio);
1686
1687         if (bio->bi_rw & REQ_FLUSH) {
1688                 bio->bi_bdev = s->cow->bdev;
1689                 return DM_MAPIO_REMAPPED;
1690         }
1691
1692         chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1693
1694         /* Full snapshots are not usable */
1695         /* To get here the table must be live so s->active is always set. */
1696         if (!s->valid)
1697                 return -EIO;
1698
1699         /* FIXME: should only take write lock if we need
1700          * to copy an exception */
1701         down_write(&s->lock);
1702
1703         if (!s->valid || (unlikely(s->snapshot_overflowed) && bio_rw(bio) == WRITE)) {
1704                 r = -EIO;
1705                 goto out_unlock;
1706         }
1707
1708         /* If the block is already remapped - use that, else remap it */
1709         e = dm_lookup_exception(&s->complete, chunk);
1710         if (e) {
1711                 remap_exception(s, e, bio, chunk);
1712                 goto out_unlock;
1713         }
1714
1715         /*
1716          * Write to snapshot - higher level takes care of RW/RO
1717          * flags so we should only get this if we are
1718          * writeable.
1719          */
1720         if (bio_rw(bio) == WRITE) {
1721                 pe = __lookup_pending_exception(s, chunk);
1722                 if (!pe) {
1723                         up_write(&s->lock);
1724                         pe = alloc_pending_exception(s);
1725                         down_write(&s->lock);
1726
1727                         if (!s->valid || s->snapshot_overflowed) {
1728                                 free_pending_exception(pe);
1729                                 r = -EIO;
1730                                 goto out_unlock;
1731                         }
1732
1733                         e = dm_lookup_exception(&s->complete, chunk);
1734                         if (e) {
1735                                 free_pending_exception(pe);
1736                                 remap_exception(s, e, bio, chunk);
1737                                 goto out_unlock;
1738                         }
1739
1740                         pe = __find_pending_exception(s, pe, chunk);
1741                         if (!pe) {
1742                                 if (s->store->userspace_supports_overflow) {
1743                                         s->snapshot_overflowed = 1;
1744                                         DMERR("Snapshot overflowed: Unable to allocate exception.");
1745                                 } else
1746                                         __invalidate_snapshot(s, -ENOMEM);
1747                                 r = -EIO;
1748                                 goto out_unlock;
1749                         }
1750                 }
1751
1752                 remap_exception(s, &pe->e, bio, chunk);
1753
1754                 r = DM_MAPIO_SUBMITTED;
1755
1756                 if (!pe->started &&
1757                     bio->bi_iter.bi_size ==
1758                     (s->store->chunk_size << SECTOR_SHIFT)) {
1759                         pe->started = 1;
1760                         up_write(&s->lock);
1761                         start_full_bio(pe, bio);
1762                         goto out;
1763                 }
1764
1765                 bio_list_add(&pe->snapshot_bios, bio);
1766
1767                 if (!pe->started) {
1768                         /* this is protected by snap->lock */
1769                         pe->started = 1;
1770                         up_write(&s->lock);
1771                         start_copy(pe);
1772                         goto out;
1773                 }
1774         } else {
1775                 bio->bi_bdev = s->origin->bdev;
1776                 track_chunk(s, bio, chunk);
1777         }
1778
1779 out_unlock:
1780         up_write(&s->lock);
1781 out:
1782         return r;
1783 }
1784
1785 /*
1786  * A snapshot-merge target behaves like a combination of a snapshot
1787  * target and a snapshot-origin target.  It only generates new
1788  * exceptions in other snapshots and not in the one that is being
1789  * merged.
1790  *
1791  * For each chunk, if there is an existing exception, it is used to
1792  * redirect I/O to the cow device.  Otherwise I/O is sent to the origin,
1793  * which in turn might generate exceptions in other snapshots.
1794  * If merging is currently taking place on the chunk in question, the
1795  * I/O is deferred by adding it to s->bios_queued_during_merge.
1796  */
1797 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1798 {
1799         struct dm_exception *e;
1800         struct dm_snapshot *s = ti->private;
1801         int r = DM_MAPIO_REMAPPED;
1802         chunk_t chunk;
1803
1804         init_tracked_chunk(bio);
1805
1806         if (bio->bi_rw & REQ_FLUSH) {
1807                 if (!dm_bio_get_target_bio_nr(bio))
1808                         bio->bi_bdev = s->origin->bdev;
1809                 else
1810                         bio->bi_bdev = s->cow->bdev;
1811                 return DM_MAPIO_REMAPPED;
1812         }
1813
1814         chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1815
1816         down_write(&s->lock);
1817
1818         /* Full merging snapshots are redirected to the origin */
1819         if (!s->valid)
1820                 goto redirect_to_origin;
1821
1822         /* If the block is already remapped - use that */
1823         e = dm_lookup_exception(&s->complete, chunk);
1824         if (e) {
1825                 /* Queue writes overlapping with chunks being merged */
1826                 if (bio_rw(bio) == WRITE &&
1827                     chunk >= s->first_merging_chunk &&
1828                     chunk < (s->first_merging_chunk +
1829                              s->num_merging_chunks)) {
1830                         bio->bi_bdev = s->origin->bdev;
1831                         bio_list_add(&s->bios_queued_during_merge, bio);
1832                         r = DM_MAPIO_SUBMITTED;
1833                         goto out_unlock;
1834                 }
1835
1836                 remap_exception(s, e, bio, chunk);
1837
1838                 if (bio_rw(bio) == WRITE)
1839                         track_chunk(s, bio, chunk);
1840                 goto out_unlock;
1841         }
1842
1843 redirect_to_origin:
1844         bio->bi_bdev = s->origin->bdev;
1845
1846         if (bio_rw(bio) == WRITE) {
1847                 up_write(&s->lock);
1848                 return do_origin(s->origin, bio);
1849         }
1850
1851 out_unlock:
1852         up_write(&s->lock);
1853
1854         return r;
1855 }
1856
1857 static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
1858 {
1859         struct dm_snapshot *s = ti->private;
1860
1861         if (is_bio_tracked(bio))
1862                 stop_tracking_chunk(s, bio);
1863
1864         return 0;
1865 }
1866
1867 static void snapshot_merge_presuspend(struct dm_target *ti)
1868 {
1869         struct dm_snapshot *s = ti->private;
1870
1871         stop_merge(s);
1872 }
1873
1874 static int snapshot_preresume(struct dm_target *ti)
1875 {
1876         int r = 0;
1877         struct dm_snapshot *s = ti->private;
1878         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1879
1880         down_read(&_origins_lock);
1881         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1882         if (snap_src && snap_dest) {
1883                 down_read(&snap_src->lock);
1884                 if (s == snap_src) {
1885                         DMERR("Unable to resume snapshot source until "
1886                               "handover completes.");
1887                         r = -EINVAL;
1888                 } else if (!dm_suspended(snap_src->ti)) {
1889                         DMERR("Unable to perform snapshot handover until "
1890                               "source is suspended.");
1891                         r = -EINVAL;
1892                 }
1893                 up_read(&snap_src->lock);
1894         }
1895         up_read(&_origins_lock);
1896
1897         return r;
1898 }
1899
1900 static void snapshot_resume(struct dm_target *ti)
1901 {
1902         struct dm_snapshot *s = ti->private;
1903         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL;
1904         struct dm_origin *o;
1905         struct mapped_device *origin_md = NULL;
1906         bool must_restart_merging = false;
1907
1908         down_read(&_origins_lock);
1909
1910         o = __lookup_dm_origin(s->origin->bdev);
1911         if (o)
1912                 origin_md = dm_table_get_md(o->ti->table);
1913         if (!origin_md) {
1914                 (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging);
1915                 if (snap_merging)
1916                         origin_md = dm_table_get_md(snap_merging->ti->table);
1917         }
1918         if (origin_md == dm_table_get_md(ti->table))
1919                 origin_md = NULL;
1920         if (origin_md) {
1921                 if (dm_hold(origin_md))
1922                         origin_md = NULL;
1923         }
1924
1925         up_read(&_origins_lock);
1926
1927         if (origin_md) {
1928                 dm_internal_suspend_fast(origin_md);
1929                 if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) {
1930                         must_restart_merging = true;
1931                         stop_merge(snap_merging);
1932                 }
1933         }
1934
1935         down_read(&_origins_lock);
1936
1937         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1938         if (snap_src && snap_dest) {
1939                 down_write(&snap_src->lock);
1940                 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1941                 __handover_exceptions(snap_src, snap_dest);
1942                 up_write(&snap_dest->lock);
1943                 up_write(&snap_src->lock);
1944         }
1945
1946         up_read(&_origins_lock);
1947
1948         if (origin_md) {
1949                 if (must_restart_merging)
1950                         start_merge(snap_merging);
1951                 dm_internal_resume_fast(origin_md);
1952                 dm_put(origin_md);
1953         }
1954
1955         /* Now we have correct chunk size, reregister */
1956         reregister_snapshot(s);
1957
1958         down_write(&s->lock);
1959         s->active = 1;
1960         up_write(&s->lock);
1961 }
1962
1963 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1964 {
1965         uint32_t min_chunksize;
1966
1967         down_read(&_origins_lock);
1968         min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1969         up_read(&_origins_lock);
1970
1971         return min_chunksize;
1972 }
1973
1974 static void snapshot_merge_resume(struct dm_target *ti)
1975 {
1976         struct dm_snapshot *s = ti->private;
1977
1978         /*
1979          * Handover exceptions from existing snapshot.
1980          */
1981         snapshot_resume(ti);
1982
1983         /*
1984          * snapshot-merge acts as an origin, so set ti->max_io_len
1985          */
1986         ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1987
1988         start_merge(s);
1989 }
1990
1991 static void snapshot_status(struct dm_target *ti, status_type_t type,
1992                             unsigned status_flags, char *result, unsigned maxlen)
1993 {
1994         unsigned sz = 0;
1995         struct dm_snapshot *snap = ti->private;
1996
1997         switch (type) {
1998         case STATUSTYPE_INFO:
1999
2000                 down_write(&snap->lock);
2001
2002                 if (!snap->valid)
2003                         DMEMIT("Invalid");
2004                 else if (snap->merge_failed)
2005                         DMEMIT("Merge failed");
2006                 else if (snap->snapshot_overflowed)
2007                         DMEMIT("Overflow");
2008                 else {
2009                         if (snap->store->type->usage) {
2010                                 sector_t total_sectors, sectors_allocated,
2011                                          metadata_sectors;
2012                                 snap->store->type->usage(snap->store,
2013                                                          &total_sectors,
2014                                                          &sectors_allocated,
2015                                                          &metadata_sectors);
2016                                 DMEMIT("%llu/%llu %llu",
2017                                        (unsigned long long)sectors_allocated,
2018                                        (unsigned long long)total_sectors,
2019                                        (unsigned long long)metadata_sectors);
2020                         }
2021                         else
2022                                 DMEMIT("Unknown");
2023                 }
2024
2025                 up_write(&snap->lock);
2026
2027                 break;
2028
2029         case STATUSTYPE_TABLE:
2030                 /*
2031                  * kdevname returns a static pointer so we need
2032                  * to make private copies if the output is to
2033                  * make sense.
2034                  */
2035                 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
2036                 snap->store->type->status(snap->store, type, result + sz,
2037                                           maxlen - sz);
2038                 break;
2039         }
2040 }
2041
2042 static int snapshot_iterate_devices(struct dm_target *ti,
2043                                     iterate_devices_callout_fn fn, void *data)
2044 {
2045         struct dm_snapshot *snap = ti->private;
2046         int r;
2047
2048         r = fn(ti, snap->origin, 0, ti->len, data);
2049
2050         if (!r)
2051                 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
2052
2053         return r;
2054 }
2055
2056
2057 /*-----------------------------------------------------------------
2058  * Origin methods
2059  *---------------------------------------------------------------*/
2060
2061 /*
2062  * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
2063  * supplied bio was ignored.  The caller may submit it immediately.
2064  * (No remapping actually occurs as the origin is always a direct linear
2065  * map.)
2066  *
2067  * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
2068  * and any supplied bio is added to a list to be submitted once all
2069  * the necessary exceptions exist.
2070  */
2071 static int __origin_write(struct list_head *snapshots, sector_t sector,
2072                           struct bio *bio)
2073 {
2074         int r = DM_MAPIO_REMAPPED;
2075         struct dm_snapshot *snap;
2076         struct dm_exception *e;
2077         struct dm_snap_pending_exception *pe;
2078         struct dm_snap_pending_exception *pe_to_start_now = NULL;
2079         struct dm_snap_pending_exception *pe_to_start_last = NULL;
2080         chunk_t chunk;
2081
2082         /* Do all the snapshots on this origin */
2083         list_for_each_entry (snap, snapshots, list) {
2084                 /*
2085                  * Don't make new exceptions in a merging snapshot
2086                  * because it has effectively been deleted
2087                  */
2088                 if (dm_target_is_snapshot_merge(snap->ti))
2089                         continue;
2090
2091                 down_write(&snap->lock);
2092
2093                 /* Only deal with valid and active snapshots */
2094                 if (!snap->valid || !snap->active)
2095                         goto next_snapshot;
2096
2097                 /* Nothing to do if writing beyond end of snapshot */
2098                 if (sector >= dm_table_get_size(snap->ti->table))
2099                         goto next_snapshot;
2100
2101                 /*
2102                  * Remember, different snapshots can have
2103                  * different chunk sizes.
2104                  */
2105                 chunk = sector_to_chunk(snap->store, sector);
2106
2107                 /*
2108                  * Check exception table to see if block
2109                  * is already remapped in this snapshot
2110                  * and trigger an exception if not.
2111                  */
2112                 e = dm_lookup_exception(&snap->complete, chunk);
2113                 if (e)
2114                         goto next_snapshot;
2115
2116                 pe = __lookup_pending_exception(snap, chunk);
2117                 if (!pe) {
2118                         up_write(&snap->lock);
2119                         pe = alloc_pending_exception(snap);
2120                         down_write(&snap->lock);
2121
2122                         if (!snap->valid) {
2123                                 free_pending_exception(pe);
2124                                 goto next_snapshot;
2125                         }
2126
2127                         e = dm_lookup_exception(&snap->complete, chunk);
2128                         if (e) {
2129                                 free_pending_exception(pe);
2130                                 goto next_snapshot;
2131                         }
2132
2133                         pe = __find_pending_exception(snap, pe, chunk);
2134                         if (!pe) {
2135                                 __invalidate_snapshot(snap, -ENOMEM);
2136                                 goto next_snapshot;
2137                         }
2138                 }
2139
2140                 r = DM_MAPIO_SUBMITTED;
2141
2142                 /*
2143                  * If an origin bio was supplied, queue it to wait for the
2144                  * completion of this exception, and start this one last,
2145                  * at the end of the function.
2146                  */
2147                 if (bio) {
2148                         bio_list_add(&pe->origin_bios, bio);
2149                         bio = NULL;
2150
2151                         if (!pe->started) {
2152                                 pe->started = 1;
2153                                 pe_to_start_last = pe;
2154                         }
2155                 }
2156
2157                 if (!pe->started) {
2158                         pe->started = 1;
2159                         pe_to_start_now = pe;
2160                 }
2161
2162 next_snapshot:
2163                 up_write(&snap->lock);
2164
2165                 if (pe_to_start_now) {
2166                         start_copy(pe_to_start_now);
2167                         pe_to_start_now = NULL;
2168                 }
2169         }
2170
2171         /*
2172          * Submit the exception against which the bio is queued last,
2173          * to give the other exceptions a head start.
2174          */
2175         if (pe_to_start_last)
2176                 start_copy(pe_to_start_last);
2177
2178         return r;
2179 }
2180
2181 /*
2182  * Called on a write from the origin driver.
2183  */
2184 static int do_origin(struct dm_dev *origin, struct bio *bio)
2185 {
2186         struct origin *o;
2187         int r = DM_MAPIO_REMAPPED;
2188
2189         down_read(&_origins_lock);
2190         o = __lookup_origin(origin->bdev);
2191         if (o)
2192                 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
2193         up_read(&_origins_lock);
2194
2195         return r;
2196 }
2197
2198 /*
2199  * Trigger exceptions in all non-merging snapshots.
2200  *
2201  * The chunk size of the merging snapshot may be larger than the chunk
2202  * size of some other snapshot so we may need to reallocate multiple
2203  * chunks in other snapshots.
2204  *
2205  * We scan all the overlapping exceptions in the other snapshots.
2206  * Returns 1 if anything was reallocated and must be waited for,
2207  * otherwise returns 0.
2208  *
2209  * size must be a multiple of merging_snap's chunk_size.
2210  */
2211 static int origin_write_extent(struct dm_snapshot *merging_snap,
2212                                sector_t sector, unsigned size)
2213 {
2214         int must_wait = 0;
2215         sector_t n;
2216         struct origin *o;
2217
2218         /*
2219          * The origin's __minimum_chunk_size() got stored in max_io_len
2220          * by snapshot_merge_resume().
2221          */
2222         down_read(&_origins_lock);
2223         o = __lookup_origin(merging_snap->origin->bdev);
2224         for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2225                 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2226                     DM_MAPIO_SUBMITTED)
2227                         must_wait = 1;
2228         up_read(&_origins_lock);
2229
2230         return must_wait;
2231 }
2232
2233 /*
2234  * Origin: maps a linear range of a device, with hooks for snapshotting.
2235  */
2236
2237 /*
2238  * Construct an origin mapping: <dev_path>
2239  * The context for an origin is merely a 'struct dm_dev *'
2240  * pointing to the real device.
2241  */
2242 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2243 {
2244         int r;
2245         struct dm_origin *o;
2246
2247         if (argc != 1) {
2248                 ti->error = "origin: incorrect number of arguments";
2249                 return -EINVAL;
2250         }
2251
2252         o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2253         if (!o) {
2254                 ti->error = "Cannot allocate private origin structure";
2255                 r = -ENOMEM;
2256                 goto bad_alloc;
2257         }
2258
2259         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2260         if (r) {
2261                 ti->error = "Cannot get target device";
2262                 goto bad_open;
2263         }
2264
2265         o->ti = ti;
2266         ti->private = o;
2267         ti->num_flush_bios = 1;
2268
2269         return 0;
2270
2271 bad_open:
2272         kfree(o);
2273 bad_alloc:
2274         return r;
2275 }
2276
2277 static void origin_dtr(struct dm_target *ti)
2278 {
2279         struct dm_origin *o = ti->private;
2280
2281         dm_put_device(ti, o->dev);
2282         kfree(o);
2283 }
2284
2285 static int origin_map(struct dm_target *ti, struct bio *bio)
2286 {
2287         struct dm_origin *o = ti->private;
2288         unsigned available_sectors;
2289
2290         bio->bi_bdev = o->dev->bdev;
2291
2292         if (unlikely(bio->bi_rw & REQ_FLUSH))
2293                 return DM_MAPIO_REMAPPED;
2294
2295         if (bio_rw(bio) != WRITE)
2296                 return DM_MAPIO_REMAPPED;
2297
2298         available_sectors = o->split_boundary -
2299                 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2300
2301         if (bio_sectors(bio) > available_sectors)
2302                 dm_accept_partial_bio(bio, available_sectors);
2303
2304         /* Only tell snapshots if this is a write */
2305         return do_origin(o->dev, bio);
2306 }
2307
2308 /*
2309  * Set the target "max_io_len" field to the minimum of all the snapshots'
2310  * chunk sizes.
2311  */
2312 static void origin_resume(struct dm_target *ti)
2313 {
2314         struct dm_origin *o = ti->private;
2315
2316         o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
2317
2318         down_write(&_origins_lock);
2319         __insert_dm_origin(o);
2320         up_write(&_origins_lock);
2321 }
2322
2323 static void origin_postsuspend(struct dm_target *ti)
2324 {
2325         struct dm_origin *o = ti->private;
2326
2327         down_write(&_origins_lock);
2328         __remove_dm_origin(o);
2329         up_write(&_origins_lock);
2330 }
2331
2332 static void origin_status(struct dm_target *ti, status_type_t type,
2333                           unsigned status_flags, char *result, unsigned maxlen)
2334 {
2335         struct dm_origin *o = ti->private;
2336
2337         switch (type) {
2338         case STATUSTYPE_INFO:
2339                 result[0] = '\0';
2340                 break;
2341
2342         case STATUSTYPE_TABLE:
2343                 snprintf(result, maxlen, "%s", o->dev->name);
2344                 break;
2345         }
2346 }
2347
2348 static int origin_iterate_devices(struct dm_target *ti,
2349                                   iterate_devices_callout_fn fn, void *data)
2350 {
2351         struct dm_origin *o = ti->private;
2352
2353         return fn(ti, o->dev, 0, ti->len, data);
2354 }
2355
2356 static struct target_type origin_target = {
2357         .name    = "snapshot-origin",
2358         .version = {1, 9, 0},
2359         .module  = THIS_MODULE,
2360         .ctr     = origin_ctr,
2361         .dtr     = origin_dtr,
2362         .map     = origin_map,
2363         .resume  = origin_resume,
2364         .postsuspend = origin_postsuspend,
2365         .status  = origin_status,
2366         .iterate_devices = origin_iterate_devices,
2367 };
2368
2369 static struct target_type snapshot_target = {
2370         .name    = "snapshot",
2371         .version = {1, 15, 0},
2372         .module  = THIS_MODULE,
2373         .ctr     = snapshot_ctr,
2374         .dtr     = snapshot_dtr,
2375         .map     = snapshot_map,
2376         .end_io  = snapshot_end_io,
2377         .preresume  = snapshot_preresume,
2378         .resume  = snapshot_resume,
2379         .status  = snapshot_status,
2380         .iterate_devices = snapshot_iterate_devices,
2381 };
2382
2383 static struct target_type merge_target = {
2384         .name    = dm_snapshot_merge_target_name,
2385         .version = {1, 4, 0},
2386         .module  = THIS_MODULE,
2387         .ctr     = snapshot_ctr,
2388         .dtr     = snapshot_dtr,
2389         .map     = snapshot_merge_map,
2390         .end_io  = snapshot_end_io,
2391         .presuspend = snapshot_merge_presuspend,
2392         .preresume  = snapshot_preresume,
2393         .resume  = snapshot_merge_resume,
2394         .status  = snapshot_status,
2395         .iterate_devices = snapshot_iterate_devices,
2396 };
2397
2398 static int __init dm_snapshot_init(void)
2399 {
2400         int r;
2401
2402         r = dm_exception_store_init();
2403         if (r) {
2404                 DMERR("Failed to initialize exception stores");
2405                 return r;
2406         }
2407
2408         r = dm_register_target(&snapshot_target);
2409         if (r < 0) {
2410                 DMERR("snapshot target register failed %d", r);
2411                 goto bad_register_snapshot_target;
2412         }
2413
2414         r = dm_register_target(&origin_target);
2415         if (r < 0) {
2416                 DMERR("Origin target register failed %d", r);
2417                 goto bad_register_origin_target;
2418         }
2419
2420         r = dm_register_target(&merge_target);
2421         if (r < 0) {
2422                 DMERR("Merge target register failed %d", r);
2423                 goto bad_register_merge_target;
2424         }
2425
2426         r = init_origin_hash();
2427         if (r) {
2428                 DMERR("init_origin_hash failed.");
2429                 goto bad_origin_hash;
2430         }
2431
2432         exception_cache = KMEM_CACHE(dm_exception, 0);
2433         if (!exception_cache) {
2434                 DMERR("Couldn't create exception cache.");
2435                 r = -ENOMEM;
2436                 goto bad_exception_cache;
2437         }
2438
2439         pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2440         if (!pending_cache) {
2441                 DMERR("Couldn't create pending cache.");
2442                 r = -ENOMEM;
2443                 goto bad_pending_cache;
2444         }
2445
2446         return 0;
2447
2448 bad_pending_cache:
2449         kmem_cache_destroy(exception_cache);
2450 bad_exception_cache:
2451         exit_origin_hash();
2452 bad_origin_hash:
2453         dm_unregister_target(&merge_target);
2454 bad_register_merge_target:
2455         dm_unregister_target(&origin_target);
2456 bad_register_origin_target:
2457         dm_unregister_target(&snapshot_target);
2458 bad_register_snapshot_target:
2459         dm_exception_store_exit();
2460
2461         return r;
2462 }
2463
2464 static void __exit dm_snapshot_exit(void)
2465 {
2466         dm_unregister_target(&snapshot_target);
2467         dm_unregister_target(&origin_target);
2468         dm_unregister_target(&merge_target);
2469
2470         exit_origin_hash();
2471         kmem_cache_destroy(pending_cache);
2472         kmem_cache_destroy(exception_cache);
2473
2474         dm_exception_store_exit();
2475 }
2476
2477 /* Module hooks */
2478 module_init(dm_snapshot_init);
2479 module_exit(dm_snapshot_exit);
2480
2481 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2482 MODULE_AUTHOR("Joe Thornber");
2483 MODULE_LICENSE("GPL");
2484 MODULE_ALIAS("dm-snapshot-origin");
2485 MODULE_ALIAS("dm-snapshot-merge");