linux/drivers/md/dm-thin.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2011-2012 Red Hat UK.
   3 *
   4 * This file is released under the GPL.
   5 */
   6
   7#include "dm-thin-metadata.h"
   8#include "dm-bio-prison.h"
   9#include "dm.h"
  10
  11#include <linux/device-mapper.h>
  12#include <linux/dm-io.h>
  13#include <linux/dm-kcopyd.h>
  14#include <linux/list.h>
  15#include <linux/init.h>
  16#include <linux/module.h>
  17#include <linux/slab.h>
  18
  19#define DM_MSG_PREFIX   "thin"
  20
  21/*
  22 * Tunable constants
  23 */
  24#define ENDIO_HOOK_POOL_SIZE 1024
  25#define MAPPING_POOL_SIZE 1024
  26#define PRISON_CELLS 1024
  27#define COMMIT_PERIOD HZ
  28
  29DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
  30                "A percentage of time allocated for copy on write");
  31
  32/*
  33 * The block size of the device holding pool data must be
  34 * between 64KB and 1GB.
  35 */
  36#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
  37#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
  38
  39/*
  40 * Device id is restricted to 24 bits.
  41 */
  42#define MAX_DEV_ID ((1 << 24) - 1)
  43
  44/*
  45 * How do we handle breaking sharing of data blocks?
  46 * =================================================
  47 *
  48 * We use a standard copy-on-write btree to store the mappings for the
  49 * devices (note I'm talking about copy-on-write of the metadata here, not
  50 * the data).  When you take an internal snapshot you clone the root node
  51 * of the origin btree.  After this there is no concept of an origin or a
  52 * snapshot.  They are just two device trees that happen to point to the
  53 * same data blocks.
  54 *
  55 * When we get a write in we decide if it's to a shared data block using
  56 * some timestamp magic.  If it is, we have to break sharing.
  57 *
  58 * Let's say we write to a shared block in what was the origin.  The
  59 * steps are:
  60 *
  61 * i) plug io further to this physical block. (see bio_prison code).
  62 *
  63 * ii) quiesce any read io to that shared data block.  Obviously
  64 * including all devices that share this block.  (see dm_deferred_set code)
  65 *
  66 * iii) copy the data block to a newly allocate block.  This step can be
  67 * missed out if the io covers the block. (schedule_copy).
  68 *
  69 * iv) insert the new mapping into the origin's btree
  70 * (process_prepared_mapping).  This act of inserting breaks some
  71 * sharing of btree nodes between the two devices.  Breaking sharing only
  72 * effects the btree of that specific device.  Btrees for the other
  73 * devices that share the block never change.  The btree for the origin
  74 * device as it was after the last commit is untouched, ie. we're using
  75 * persistent data structures in the functional programming sense.
  76 *
  77 * v) unplug io to this physical block, including the io that triggered
  78 * the breaking of sharing.
  79 *
  80 * Steps (ii) and (iii) occur in parallel.
  81 *
  82 * The metadata _doesn't_ need to be committed before the io continues.  We
  83 * get away with this because the io is always written to a _new_ block.
  84 * If there's a crash, then:
  85 *
  86 * - The origin mapping will point to the old origin block (the shared
  87 * one).  This will contain the data as it was before the io that triggered
  88 * the breaking of sharing came in.
  89 *
  90 * - The snap mapping still points to the old block.  As it would after
  91 * the commit.
  92 *
  93 * The downside of this scheme is the timestamp magic isn't perfect, and
  94 * will continue to think that data block in the snapshot device is shared
  95 * even after the write to the origin has broken sharing.  I suspect data
  96 * blocks will typically be shared by many different devices, so we're
  97 * breaking sharing n + 1 times, rather than n, where n is the number of
  98 * devices that reference this data block.  At the moment I think the
  99 * benefits far, far outweigh the disadvantages.
 100 */
 101
 102/*----------------------------------------------------------------*/
 103
 104/*
 105 * Key building.
 106 */
 107static void build_data_key(struct dm_thin_device *td,
 108                           dm_block_t b, struct dm_cell_key *key)
 109{
 110        key->virtual = 0;
 111        key->dev = dm_thin_dev_id(td);
 112        key->block = b;
 113}
 114
 115static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
 116                              struct dm_cell_key *key)
 117{
 118        key->virtual = 1;
 119        key->dev = dm_thin_dev_id(td);
 120        key->block = b;
 121}
 122
 123/*----------------------------------------------------------------*/
 124
 125/*
 126 * A pool device ties together a metadata device and a data device.  It
 127 * also provides the interface for creating and destroying internal
 128 * devices.
 129 */
 130struct dm_thin_new_mapping;
 131
 132/*
 133 * The pool runs in 3 modes.  Ordered in degraded order for comparisons.
 134 */
 135enum pool_mode {
 136        PM_WRITE,               /* metadata may be changed */
 137        PM_READ_ONLY,           /* metadata may not be changed */
 138        PM_FAIL,                /* all I/O fails */
 139};
 140
 141struct pool_features {
 142        enum pool_mode mode;
 143
 144        bool zero_new_blocks:1;
 145        bool discard_enabled:1;
 146        bool discard_passdown:1;
 147};
 148
 149struct thin_c;
 150typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
 151typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
 152
 153struct pool {
 154        struct list_head list;
 155        struct dm_target *ti;   /* Only set if a pool target is bound */
 156
 157        struct mapped_device *pool_md;
 158        struct block_device *md_dev;
 159        struct dm_pool_metadata *pmd;
 160
 161        dm_block_t low_water_blocks;
 162        uint32_t sectors_per_block;
 163        int sectors_per_block_shift;
 164
 165        struct pool_features pf;
 166        unsigned low_water_triggered:1; /* A dm event has been sent */
 167        unsigned no_free_space:1;       /* A -ENOSPC warning has been issued */
 168
 169        struct dm_bio_prison *prison;
 170        struct dm_kcopyd_client *copier;
 171
 172        struct workqueue_struct *wq;
 173        struct work_struct worker;
 174        struct delayed_work waker;
 175
 176        unsigned long last_commit_jiffies;
 177        unsigned ref_count;
 178
 179        spinlock_t lock;
 180        struct bio_list deferred_bios;
 181        struct bio_list deferred_flush_bios;
 182        struct list_head prepared_mappings;
 183        struct list_head prepared_discards;
 184
 185        struct bio_list retry_on_resume_list;
 186
 187        struct dm_deferred_set *shared_read_ds;
 188        struct dm_deferred_set *all_io_ds;
 189
 190        struct dm_thin_new_mapping *next_mapping;
 191        mempool_t *mapping_pool;
 192
 193        process_bio_fn process_bio;
 194        process_bio_fn process_discard;
 195
 196        process_mapping_fn process_prepared_mapping;
 197        process_mapping_fn process_prepared_discard;
 198};
 199
 200static enum pool_mode get_pool_mode(struct pool *pool);
 201static void set_pool_mode(struct pool *pool, enum pool_mode mode);
 202
 203/*
 204 * Target context for a pool.
 205 */
 206struct pool_c {
 207        struct dm_target *ti;
 208        struct pool *pool;
 209        struct dm_dev *data_dev;
 210        struct dm_dev *metadata_dev;
 211        struct dm_target_callbacks callbacks;
 212
 213        dm_block_t low_water_blocks;
 214        struct pool_features requested_pf; /* Features requested during table load */
 215        struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
 216};
 217
 218/*
 219 * Target context for a thin.
 220 */
 221struct thin_c {
 222        struct dm_dev *pool_dev;
 223        struct dm_dev *origin_dev;
 224        dm_thin_id dev_id;
 225
 226        struct pool *pool;
 227        struct dm_thin_device *td;
 228};
 229
 230/*----------------------------------------------------------------*/
 231
 232/*
 233 * wake_worker() is used when new work is queued and when pool_resume is
 234 * ready to continue deferred IO processing.
 235 */
 236static void wake_worker(struct pool *pool)
 237{
 238        queue_work(pool->wq, &pool->worker);
 239}
 240
 241/*----------------------------------------------------------------*/
 242
 243static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
 244                      struct dm_bio_prison_cell **cell_result)
 245{
 246        int r;
 247        struct dm_bio_prison_cell *cell_prealloc;
 248
 249        /*
 250         * Allocate a cell from the prison's mempool.
 251         * This might block but it can't fail.
 252         */
 253        cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
 254
 255        r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
 256        if (r)
 257                /*
 258                 * We reused an old cell; we can get rid of
 259                 * the new one.
 260                 */
 261                dm_bio_prison_free_cell(pool->prison, cell_prealloc);
 262
 263        return r;
 264}
 265
 266static void cell_release(struct pool *pool,
 267                         struct dm_bio_prison_cell *cell,
 268                         struct bio_list *bios)
 269{
 270        dm_cell_release(pool->prison, cell, bios);
 271        dm_bio_prison_free_cell(pool->prison, cell);
 272}
 273
 274static void cell_release_no_holder(struct pool *pool,
 275                                   struct dm_bio_prison_cell *cell,
 276                                   struct bio_list *bios)
 277{
 278        dm_cell_release_no_holder(pool->prison, cell, bios);
 279        dm_bio_prison_free_cell(pool->prison, cell);
 280}
 281
 282static void cell_defer_no_holder_no_free(struct thin_c *tc,
 283                                         struct dm_bio_prison_cell *cell)
 284{
 285        struct pool *pool = tc->pool;
 286        unsigned long flags;
 287
 288        spin_lock_irqsave(&pool->lock, flags);
 289        dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
 290        spin_unlock_irqrestore(&pool->lock, flags);
 291
 292        wake_worker(pool);
 293}
 294
 295static void cell_error(struct pool *pool,
 296                       struct dm_bio_prison_cell *cell)
 297{
 298        dm_cell_error(pool->prison, cell);
 299        dm_bio_prison_free_cell(pool->prison, cell);
 300}
 301
 302/*----------------------------------------------------------------*/
 303
 304/*
 305 * A global list of pools that uses a struct mapped_device as a key.
 306 */
 307static struct dm_thin_pool_table {
 308        struct mutex mutex;
 309        struct list_head pools;
 310} dm_thin_pool_table;
 311
 312static void pool_table_init(void)
 313{
 314        mutex_init(&dm_thin_pool_table.mutex);
 315        INIT_LIST_HEAD(&dm_thin_pool_table.pools);
 316}
 317
 318static void __pool_table_insert(struct pool *pool)
 319{
 320        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
 321        list_add(&pool->list, &dm_thin_pool_table.pools);
 322}
 323
 324static void __pool_table_remove(struct pool *pool)
 325{
 326        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
 327        list_del(&pool->list);
 328}
 329
 330static struct pool *__pool_table_lookup(struct mapped_device *md)
 331{
 332        struct pool *pool = NULL, *tmp;
 333
 334        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
 335
 336        list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
 337                if (tmp->pool_md == md) {
 338                        pool = tmp;
 339                        break;
 340                }
 341        }
 342
 343        return pool;
 344}
 345
 346static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
 347{
 348        struct pool *pool = NULL, *tmp;
 349
 350        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
 351
 352        list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
 353                if (tmp->md_dev == md_dev) {
 354                        pool = tmp;
 355                        break;
 356                }
 357        }
 358
 359        return pool;
 360}
 361
 362/*----------------------------------------------------------------*/
 363
 364struct dm_thin_endio_hook {
 365        struct thin_c *tc;
 366        struct dm_deferred_entry *shared_read_entry;
 367        struct dm_deferred_entry *all_io_entry;
 368        struct dm_thin_new_mapping *overwrite_mapping;
 369};
 370
 371static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
 372{
 373        struct bio *bio;
 374        struct bio_list bios;
 375
 376        bio_list_init(&bios);
 377        bio_list_merge(&bios, master);
 378        bio_list_init(master);
 379
 380        while ((bio = bio_list_pop(&bios))) {
 381                struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 382
 383                if (h->tc == tc)
 384                        bio_endio(bio, DM_ENDIO_REQUEUE);
 385                else
 386                        bio_list_add(master, bio);
 387        }
 388}
 389
 390static void requeue_io(struct thin_c *tc)
 391{
 392        struct pool *pool = tc->pool;
 393        unsigned long flags;
 394
 395        spin_lock_irqsave(&pool->lock, flags);
 396        __requeue_bio_list(tc, &pool->deferred_bios);
 397        __requeue_bio_list(tc, &pool->retry_on_resume_list);
 398        spin_unlock_irqrestore(&pool->lock, flags);
 399}
 400
 401/*
 402 * This section of code contains the logic for processing a thin device's IO.
 403 * Much of the code depends on pool object resources (lists, workqueues, etc)
 404 * but most is exclusively called from the thin target rather than the thin-pool
 405 * target.
 406 */
 407
 408static bool block_size_is_power_of_two(struct pool *pool)
 409{
 410        return pool->sectors_per_block_shift >= 0;
 411}
 412
 413static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
 414{
 415        struct pool *pool = tc->pool;
 416        sector_t block_nr = bio->bi_sector;
 417
 418        if (block_size_is_power_of_two(pool))
 419                block_nr >>= pool->sectors_per_block_shift;
 420        else
 421                (void) sector_div(block_nr, pool->sectors_per_block);
 422
 423        return block_nr;
 424}
 425
 426static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
 427{
 428        struct pool *pool = tc->pool;
 429        sector_t bi_sector = bio->bi_sector;
 430
 431        bio->bi_bdev = tc->pool_dev->bdev;
 432        if (block_size_is_power_of_two(pool))
 433                bio->bi_sector = (block << pool->sectors_per_block_shift) |
 434                                (bi_sector & (pool->sectors_per_block - 1));
 435        else
 436                bio->bi_sector = (block * pool->sectors_per_block) +
 437                                 sector_div(bi_sector, pool->sectors_per_block);
 438}
 439
 440static void remap_to_origin(struct thin_c *tc, struct bio *bio)
 441{
 442        bio->bi_bdev = tc->origin_dev->bdev;
 443}
 444
 445static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
 446{
 447        return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
 448                dm_thin_changed_this_transaction(tc->td);
 449}
 450
 451static void inc_all_io_entry(struct pool *pool, struct bio *bio)
 452{
 453        struct dm_thin_endio_hook *h;
 454
 455        if (bio->bi_rw & REQ_DISCARD)
 456                return;
 457
 458        h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 459        h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
 460}
 461
 462static void issue(struct thin_c *tc, struct bio *bio)
 463{
 464        struct pool *pool = tc->pool;
 465        unsigned long flags;
 466
 467        if (!bio_triggers_commit(tc, bio)) {
 468                generic_make_request(bio);
 469                return;
 470        }
 471
 472        /*
 473         * Complete bio with an error if earlier I/O caused changes to
 474         * the metadata that can't be committed e.g, due to I/O errors
 475         * on the metadata device.
 476         */
 477        if (dm_thin_aborted_changes(tc->td)) {
 478                bio_io_error(bio);
 479                return;
 480        }
 481
 482        /*
 483         * Batch together any bios that trigger commits and then issue a
 484         * single commit for them in process_deferred_bios().
 485         */
 486        spin_lock_irqsave(&pool->lock, flags);
 487        bio_list_add(&pool->deferred_flush_bios, bio);
 488        spin_unlock_irqrestore(&pool->lock, flags);
 489}
 490
 491static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
 492{
 493        remap_to_origin(tc, bio);
 494        issue(tc, bio);
 495}
 496
 497static void remap_and_issue(struct thin_c *tc, struct bio *bio,
 498                            dm_block_t block)
 499{
 500        remap(tc, bio, block);
 501        issue(tc, bio);
 502}
 503
 504/*----------------------------------------------------------------*/
 505
 506/*
 507 * Bio endio functions.
 508 */
 509struct dm_thin_new_mapping {
 510        struct list_head list;
 511
 512        unsigned quiesced:1;
 513        unsigned prepared:1;
 514        unsigned pass_discard:1;
 515
 516        struct thin_c *tc;
 517        dm_block_t virt_block;
 518        dm_block_t data_block;
 519        struct dm_bio_prison_cell *cell, *cell2;
 520        int err;
 521
 522        /*
 523         * If the bio covers the whole area of a block then we can avoid
 524         * zeroing or copying.  Instead this bio is hooked.  The bio will
 525         * still be in the cell, so care has to be taken to avoid issuing
 526         * the bio twice.
 527         */
 528        struct bio *bio;
 529        bio_end_io_t *saved_bi_end_io;
 530};
 531
 532static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
 533{
 534        struct pool *pool = m->tc->pool;
 535
 536        if (m->quiesced && m->prepared) {
 537                list_add(&m->list, &pool->prepared_mappings);
 538                wake_worker(pool);
 539        }
 540}
 541
 542static void copy_complete(int read_err, unsigned long write_err, void *context)
 543{
 544        unsigned long flags;
 545        struct dm_thin_new_mapping *m = context;
 546        struct pool *pool = m->tc->pool;
 547
 548        m->err = read_err || write_err ? -EIO : 0;
 549
 550        spin_lock_irqsave(&pool->lock, flags);
 551        m->prepared = 1;
 552        __maybe_add_mapping(m);
 553        spin_unlock_irqrestore(&pool->lock, flags);
 554}
 555
 556static void overwrite_endio(struct bio *bio, int err)
 557{
 558        unsigned long flags;
 559        struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 560        struct dm_thin_new_mapping *m = h->overwrite_mapping;
 561        struct pool *pool = m->tc->pool;
 562
 563        m->err = err;
 564
 565        spin_lock_irqsave(&pool->lock, flags);
 566        m->prepared = 1;
 567        __maybe_add_mapping(m);
 568        spin_unlock_irqrestore(&pool->lock, flags);
 569}
 570
 571/*----------------------------------------------------------------*/
 572
 573/*
 574 * Workqueue.
 575 */
 576
 577/*
 578 * Prepared mapping jobs.
 579 */
 580
 581/*
 582 * This sends the bios in the cell back to the deferred_bios list.
 583 */
 584static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
 585{
 586        struct pool *pool = tc->pool;
 587        unsigned long flags;
 588
 589        spin_lock_irqsave(&pool->lock, flags);
 590        cell_release(pool, cell, &pool->deferred_bios);
 591        spin_unlock_irqrestore(&tc->pool->lock, flags);
 592
 593        wake_worker(pool);
 594}
 595
 596/*
 597 * Same as cell_defer above, except it omits the original holder of the cell.
 598 */
 599static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
 600{
 601        struct pool *pool = tc->pool;
 602        unsigned long flags;
 603
 604        spin_lock_irqsave(&pool->lock, flags);
 605        cell_release_no_holder(pool, cell, &pool->deferred_bios);
 606        spin_unlock_irqrestore(&pool->lock, flags);
 607
 608        wake_worker(pool);
 609}
 610
 611static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
 612{
 613        if (m->bio)
 614                m->bio->bi_end_io = m->saved_bi_end_io;
 615        cell_error(m->tc->pool, m->cell);
 616        list_del(&m->list);
 617        mempool_free(m, m->tc->pool->mapping_pool);
 618}
 619
 620static void process_prepared_mapping(struct dm_thin_new_mapping *m)
 621{
 622        struct thin_c *tc = m->tc;
 623        struct pool *pool = tc->pool;
 624        struct bio *bio;
 625        int r;
 626
 627        bio = m->bio;
 628        if (bio)
 629                bio->bi_end_io = m->saved_bi_end_io;
 630
 631        if (m->err) {
 632                cell_error(pool, m->cell);
 633                goto out;
 634        }
 635
 636        /*
 637         * Commit the prepared block into the mapping btree.
 638         * Any I/O for this block arriving after this point will get
 639         * remapped to it directly.
 640         */
 641        r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
 642        if (r) {
 643                DMERR_LIMIT("dm_thin_insert_block() failed");
 644                cell_error(pool, m->cell);
 645                goto out;
 646        }
 647
 648        /*
 649         * Release any bios held while the block was being provisioned.
 650         * If we are processing a write bio that completely covers the block,
 651         * we already processed it so can ignore it now when processing
 652         * the bios in the cell.
 653         */
 654        if (bio) {
 655                cell_defer_no_holder(tc, m->cell);
 656                bio_endio(bio, 0);
 657        } else
 658                cell_defer(tc, m->cell);
 659
 660out:
 661        list_del(&m->list);
 662        mempool_free(m, pool->mapping_pool);
 663}
 664
 665static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
 666{
 667        struct thin_c *tc = m->tc;
 668
 669        bio_io_error(m->bio);
 670        cell_defer_no_holder(tc, m->cell);
 671        cell_defer_no_holder(tc, m->cell2);
 672        mempool_free(m, tc->pool->mapping_pool);
 673}
 674
 675static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
 676{
 677        struct thin_c *tc = m->tc;
 678
 679        inc_all_io_entry(tc->pool, m->bio);
 680        cell_defer_no_holder(tc, m->cell);
 681        cell_defer_no_holder(tc, m->cell2);
 682
 683        if (m->pass_discard)
 684                remap_and_issue(tc, m->bio, m->data_block);
 685        else
 686                bio_endio(m->bio, 0);
 687
 688        mempool_free(m, tc->pool->mapping_pool);
 689}
 690
 691static void process_prepared_discard(struct dm_thin_new_mapping *m)
 692{
 693        int r;
 694        struct thin_c *tc = m->tc;
 695
 696        r = dm_thin_remove_block(tc->td, m->virt_block);
 697        if (r)
 698                DMERR_LIMIT("dm_thin_remove_block() failed");
 699
 700        process_prepared_discard_passdown(m);
 701}
 702
 703static void process_prepared(struct pool *pool, struct list_head *head,
 704                             process_mapping_fn *fn)
 705{
 706        unsigned long flags;
 707        struct list_head maps;
 708        struct dm_thin_new_mapping *m, *tmp;
 709
 710        INIT_LIST_HEAD(&maps);
 711        spin_lock_irqsave(&pool->lock, flags);
 712        list_splice_init(head, &maps);
 713        spin_unlock_irqrestore(&pool->lock, flags);
 714
 715        list_for_each_entry_safe(m, tmp, &maps, list)
 716                (*fn)(m);
 717}
 718
 719/*
 720 * Deferred bio jobs.
 721 */
 722static int io_overlaps_block(struct pool *pool, struct bio *bio)
 723{
 724        return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
 725}
 726
 727static int io_overwrites_block(struct pool *pool, struct bio *bio)
 728{
 729        return (bio_data_dir(bio) == WRITE) &&
 730                io_overlaps_block(pool, bio);
 731}
 732
 733static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
 734                               bio_end_io_t *fn)
 735{
 736        *save = bio->bi_end_io;
 737        bio->bi_end_io = fn;
 738}
 739
 740static int ensure_next_mapping(struct pool *pool)
 741{
 742        if (pool->next_mapping)
 743                return 0;
 744
 745        pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
 746
 747        return pool->next_mapping ? 0 : -ENOMEM;
 748}
 749
 750static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
 751{
 752        struct dm_thin_new_mapping *r = pool->next_mapping;
 753
 754        BUG_ON(!pool->next_mapping);
 755
 756        pool->next_mapping = NULL;
 757
 758        return r;
 759}
 760
 761static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
 762                          struct dm_dev *origin, dm_block_t data_origin,
 763                          dm_block_t data_dest,
 764                          struct dm_bio_prison_cell *cell, struct bio *bio)
 765{
 766        int r;
 767        struct pool *pool = tc->pool;
 768        struct dm_thin_new_mapping *m = get_next_mapping(pool);
 769
 770        INIT_LIST_HEAD(&m->list);
 771        m->quiesced = 0;
 772        m->prepared = 0;
 773        m->tc = tc;
 774        m->virt_block = virt_block;
 775        m->data_block = data_dest;
 776        m->cell = cell;
 777        m->err = 0;
 778        m->bio = NULL;
 779
 780        if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
 781                m->quiesced = 1;
 782
 783        /*
 784         * IO to pool_dev remaps to the pool target's data_dev.
 785         *
 786         * If the whole block of data is being overwritten, we can issue the
 787         * bio immediately. Otherwise we use kcopyd to clone the data first.
 788         */
 789        if (io_overwrites_block(pool, bio)) {
 790                struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 791
 792                h->overwrite_mapping = m;
 793                m->bio = bio;
 794                save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
 795                inc_all_io_entry(pool, bio);
 796                remap_and_issue(tc, bio, data_dest);
 797        } else {
 798                struct dm_io_region from, to;
 799
 800                from.bdev = origin->bdev;
 801                from.sector = data_origin * pool->sectors_per_block;
 802                from.count = pool->sectors_per_block;
 803
 804                to.bdev = tc->pool_dev->bdev;
 805                to.sector = data_dest * pool->sectors_per_block;
 806                to.count = pool->sectors_per_block;
 807
 808                r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
 809                                   0, copy_complete, m);
 810                if (r < 0) {
 811                        mempool_free(m, pool->mapping_pool);
 812                        DMERR_LIMIT("dm_kcopyd_copy() failed");
 813                        cell_error(pool, cell);
 814                }
 815        }
 816}
 817
 818static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
 819                                   dm_block_t data_origin, dm_block_t data_dest,
 820                                   struct dm_bio_prison_cell *cell, struct bio *bio)
 821{
 822        schedule_copy(tc, virt_block, tc->pool_dev,
 823                      data_origin, data_dest, cell, bio);
 824}
 825
 826static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
 827                                   dm_block_t data_dest,
 828                                   struct dm_bio_prison_cell *cell, struct bio *bio)
 829{
 830        schedule_copy(tc, virt_block, tc->origin_dev,
 831                      virt_block, data_dest, cell, bio);
 832}
 833
 834static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
 835                          dm_block_t data_block, struct dm_bio_prison_cell *cell,
 836                          struct bio *bio)
 837{
 838        struct pool *pool = tc->pool;
 839        struct dm_thin_new_mapping *m = get_next_mapping(pool);
 840
 841        INIT_LIST_HEAD(&m->list);
 842        m->quiesced = 1;
 843        m->prepared = 0;
 844        m->tc = tc;
 845        m->virt_block = virt_block;
 846        m->data_block = data_block;
 847        m->cell = cell;
 848        m->err = 0;
 849        m->bio = NULL;
 850
 851        /*
 852         * If the whole block of data is being overwritten or we are not
 853         * zeroing pre-existing data, we can issue the bio immediately.
 854         * Otherwise we use kcopyd to zero the data first.
 855         */
 856        if (!pool->pf.zero_new_blocks)
 857                process_prepared_mapping(m);
 858
 859        else if (io_overwrites_block(pool, bio)) {
 860                struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 861
 862                h->overwrite_mapping = m;
 863                m->bio = bio;
 864                save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
 865                inc_all_io_entry(pool, bio);
 866                remap_and_issue(tc, bio, data_block);
 867        } else {
 868                int r;
 869                struct dm_io_region to;
 870
 871                to.bdev = tc->pool_dev->bdev;
 872                to.sector = data_block * pool->sectors_per_block;
 873                to.count = pool->sectors_per_block;
 874
 875                r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
 876                if (r < 0) {
 877                        mempool_free(m, pool->mapping_pool);
 878                        DMERR_LIMIT("dm_kcopyd_zero() failed");
 879                        cell_error(pool, cell);
 880                }
 881        }
 882}
 883
 884static int commit(struct pool *pool)
 885{
 886        int r;
 887
 888        r = dm_pool_commit_metadata(pool->pmd);
 889        if (r)
 890                DMERR_LIMIT("%s: commit failed: error = %d",
 891                            dm_device_name(pool->pool_md), r);
 892
 893        return r;
 894}
 895
 896/*
 897 * A non-zero return indicates read_only or fail_io mode.
 898 * Many callers don't care about the return value.
 899 */
 900static int commit_or_fallback(struct pool *pool)
 901{
 902        int r;
 903
 904        if (get_pool_mode(pool) != PM_WRITE)
 905                return -EINVAL;
 906
 907        r = commit(pool);
 908        if (r)
 909                set_pool_mode(pool, PM_READ_ONLY);
 910
 911        return r;
 912}
 913
 914static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
 915{
 916        int r;
 917        dm_block_t free_blocks;
 918        unsigned long flags;
 919        struct pool *pool = tc->pool;
 920
 921        /*
 922         * Once no_free_space is set we must not allow allocation to succeed.
 923         * Otherwise it is difficult to explain, debug, test and support.
 924         */
 925        if (pool->no_free_space)
 926                return -ENOSPC;
 927
 928        r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
 929        if (r)
 930                return r;
 931
 932        if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
 933                DMWARN("%s: reached low water mark for data device: sending event.",
 934                       dm_device_name(pool->pool_md));
 935                spin_lock_irqsave(&pool->lock, flags);
 936                pool->low_water_triggered = 1;
 937                spin_unlock_irqrestore(&pool->lock, flags);
 938                dm_table_event(pool->ti->table);
 939        }
 940
 941        if (!free_blocks) {
 942                /*
 943                 * Try to commit to see if that will free up some
 944                 * more space.
 945                 */
 946                (void) commit_or_fallback(pool);
 947
 948                r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
 949                if (r)
 950                        return r;
 951
 952                /*
 953                 * If we still have no space we set a flag to avoid
 954                 * doing all this checking and return -ENOSPC.  This
 955                 * flag serves as a latch that disallows allocations from
 956                 * this pool until the admin takes action (e.g. resize or
 957                 * table reload).
 958                 */
 959                if (!free_blocks) {
 960                        DMWARN("%s: no free space available.",
 961                               dm_device_name(pool->pool_md));
 962                        spin_lock_irqsave(&pool->lock, flags);
 963                        pool->no_free_space = 1;
 964                        spin_unlock_irqrestore(&pool->lock, flags);
 965                        return -ENOSPC;
 966                }
 967        }
 968
 969        r = dm_pool_alloc_data_block(pool->pmd, result);
 970        if (r)
 971                return r;
 972
 973        return 0;
 974}
 975
 976/*
 977 * If we have run out of space, queue bios until the device is
 978 * resumed, presumably after having been reloaded with more space.
 979 */
 980static void retry_on_resume(struct bio *bio)
 981{
 982        struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
 983        struct thin_c *tc = h->tc;
 984        struct pool *pool = tc->pool;
 985        unsigned long flags;
 986
 987        spin_lock_irqsave(&pool->lock, flags);
 988        bio_list_add(&pool->retry_on_resume_list, bio);
 989        spin_unlock_irqrestore(&pool->lock, flags);
 990}
 991
 992static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
 993{
 994        struct bio *bio;
 995        struct bio_list bios;
 996
 997        bio_list_init(&bios);
 998        cell_release(pool, cell, &bios);
 999
1000        while ((bio = bio_list_pop(&bios)))
1001                retry_on_resume(bio);
1002}
1003
1004static void process_discard(struct thin_c *tc, struct bio *bio)
1005{
1006        int r;
1007        unsigned long flags;
1008        struct pool *pool = tc->pool;
1009        struct dm_bio_prison_cell *cell, *cell2;
1010        struct dm_cell_key key, key2;
1011        dm_block_t block = get_bio_block(tc, bio);
1012        struct dm_thin_lookup_result lookup_result;
1013        struct dm_thin_new_mapping *m;
1014
1015        build_virtual_key(tc->td, block, &key);
1016        if (bio_detain(tc->pool, &key, bio, &cell))
1017                return;
1018
1019        r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1020        switch (r) {
1021        case 0:
1022                /*
1023                 * Check nobody is fiddling with this pool block.  This can
1024                 * happen if someone's in the process of breaking sharing
1025                 * on this block.
1026                 */
1027                build_data_key(tc->td, lookup_result.block, &key2);
1028                if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1029                        cell_defer_no_holder(tc, cell);
1030                        break;
1031                }
1032
1033                if (io_overlaps_block(pool, bio)) {
1034                        /*
1035                         * IO may still be going to the destination block.  We must
1036                         * quiesce before we can do the removal.
1037                         */
1038                        m = get_next_mapping(pool);
1039                        m->tc = tc;
1040                        m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
1041                        m->virt_block = block;
1042                        m->data_block = lookup_result.block;
1043                        m->cell = cell;
1044                        m->cell2 = cell2;
1045                        m->err = 0;
1046                        m->bio = bio;
1047
1048                        if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1049                                spin_lock_irqsave(&pool->lock, flags);
1050                                list_add(&m->list, &pool->prepared_discards);
1051                                spin_unlock_irqrestore(&pool->lock, flags);
1052                                wake_worker(pool);
1053                        }
1054                } else {
1055                        inc_all_io_entry(pool, bio);
1056                        cell_defer_no_holder(tc, cell);
1057                        cell_defer_no_holder(tc, cell2);
1058
1059                        /*
1060                         * The DM core makes sure that the discard doesn't span
1061                         * a block boundary.  So we submit the discard of a
1062                         * partial block appropriately.
1063                         */
1064                        if ((!lookup_result.shared) && pool->pf.discard_passdown)
1065                                remap_and_issue(tc, bio, lookup_result.block);
1066                        else
1067                                bio_endio(bio, 0);
1068                }
1069                break;
1070
1071        case -ENODATA:
1072                /*
1073                 * It isn't provisioned, just forget it.
1074                 */
1075                cell_defer_no_holder(tc, cell);
1076                bio_endio(bio, 0);
1077                break;
1078
1079        default:
1080                DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1081                            __func__, r);
1082                cell_defer_no_holder(tc, cell);
1083                bio_io_error(bio);
1084                break;
1085        }
1086}
1087
1088static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1089                          struct dm_cell_key *key,
1090                          struct dm_thin_lookup_result *lookup_result,
1091                          struct dm_bio_prison_cell *cell)
1092{
1093        int r;
1094        dm_block_t data_block;
1095        struct pool *pool = tc->pool;
1096
1097        r = alloc_data_block(tc, &data_block);
1098        switch (r) {
1099        case 0:
1100                schedule_internal_copy(tc, block, lookup_result->block,
1101                                       data_block, cell, bio);
1102                break;
1103
1104        case -ENOSPC:
1105                no_space(pool, cell);
1106                break;
1107
1108        default:
1109                DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1110                            __func__, r);
1111                set_pool_mode(pool, PM_READ_ONLY);
1112                cell_error(pool, cell);
1113                break;
1114        }
1115}
1116
1117static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1118                               dm_block_t block,
1119                               struct dm_thin_lookup_result *lookup_result)
1120{
1121        struct dm_bio_prison_cell *cell;
1122        struct pool *pool = tc->pool;
1123        struct dm_cell_key key;
1124
1125        /*
1126         * If cell is already occupied, then sharing is already in the process
1127         * of being broken so we have nothing further to do here.
1128         */
1129        build_data_key(tc->td, lookup_result->block, &key);
1130        if (bio_detain(pool, &key, bio, &cell))
1131                return;
1132
1133        if (bio_data_dir(bio) == WRITE && bio->bi_size)
1134                break_sharing(tc, bio, block, &key, lookup_result, cell);
1135        else {
1136                struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1137
1138                h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1139                inc_all_io_entry(pool, bio);
1140                cell_defer_no_holder(tc, cell);
1141
1142                remap_and_issue(tc, bio, lookup_result->block);
1143        }
1144}
1145
1146static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1147                            struct dm_bio_prison_cell *cell)
1148{
1149        int r;
1150        dm_block_t data_block;
1151        struct pool *pool = tc->pool;
1152
1153        /*
1154         * Remap empty bios (flushes) immediately, without provisioning.
1155         */
1156        if (!bio->bi_size) {
1157                inc_all_io_entry(pool, bio);
1158                cell_defer_no_holder(tc, cell);
1159
1160                remap_and_issue(tc, bio, 0);
1161                return;
1162        }
1163
1164        /*
1165         * Fill read bios with zeroes and complete them immediately.
1166         */
1167        if (bio_data_dir(bio) == READ) {
1168                zero_fill_bio(bio);
1169                cell_defer_no_holder(tc, cell);
1170                bio_endio(bio, 0);
1171                return;
1172        }
1173
1174        r = alloc_data_block(tc, &data_block);
1175        switch (r) {
1176        case 0:
1177                if (tc->origin_dev)
1178                        schedule_external_copy(tc, block, data_block, cell, bio);
1179                else
1180                        schedule_zero(tc, block, data_block, cell, bio);
1181                break;
1182
1183        case -ENOSPC:
1184                no_space(pool, cell);
1185                break;
1186
1187        default:
1188                DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1189                            __func__, r);
1190                set_pool_mode(pool, PM_READ_ONLY);
1191                cell_error(pool, cell);
1192                break;
1193        }
1194}
1195
1196static void process_bio(struct thin_c *tc, struct bio *bio)
1197{
1198        int r;
1199        struct pool *pool = tc->pool;
1200        dm_block_t block = get_bio_block(tc, bio);
1201        struct dm_bio_prison_cell *cell;
1202        struct dm_cell_key key;
1203        struct dm_thin_lookup_result lookup_result;
1204
1205        /*
1206         * If cell is already occupied, then the block is already
1207         * being provisioned so we have nothing further to do here.
1208         */
1209        build_virtual_key(tc->td, block, &key);
1210        if (bio_detain(pool, &key, bio, &cell))
1211                return;
1212
1213        r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1214        switch (r) {
1215        case 0:
1216                if (lookup_result.shared) {
1217                        process_shared_bio(tc, bio, block, &lookup_result);
1218                        cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1219                } else {
1220                        inc_all_io_entry(pool, bio);
1221                        cell_defer_no_holder(tc, cell);
1222
1223                        remap_and_issue(tc, bio, lookup_result.block);
1224                }
1225                break;
1226
1227        case -ENODATA:
1228                if (bio_data_dir(bio) == READ && tc->origin_dev) {
1229                        inc_all_io_entry(pool, bio);
1230                        cell_defer_no_holder(tc, cell);
1231
1232                        remap_to_origin_and_issue(tc, bio);
1233                } else
1234                        provision_block(tc, bio, block, cell);
1235                break;
1236
1237        default:
1238                DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1239                            __func__, r);
1240                cell_defer_no_holder(tc, cell);
1241                bio_io_error(bio);
1242                break;
1243        }
1244}
1245
1246static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1247{
1248        int r;
1249        int rw = bio_data_dir(bio);
1250        dm_block_t block = get_bio_block(tc, bio);
1251        struct dm_thin_lookup_result lookup_result;
1252
1253        r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1254        switch (r) {
1255        case 0:
1256                if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1257                        bio_io_error(bio);
1258                else {
1259                        inc_all_io_entry(tc->pool, bio);
1260                        remap_and_issue(tc, bio, lookup_result.block);
1261                }
1262                break;
1263
1264        case -ENODATA:
1265                if (rw != READ) {
1266                        bio_io_error(bio);
1267                        break;
1268                }
1269
1270                if (tc->origin_dev) {
1271                        inc_all_io_entry(tc->pool, bio);
1272                        remap_to_origin_and_issue(tc, bio);
1273                        break;
1274                }
1275
1276                zero_fill_bio(bio);
1277                bio_endio(bio, 0);
1278                break;
1279
1280        default:
1281                DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1282                            __func__, r);
1283                bio_io_error(bio);
1284                break;
1285        }
1286}
1287
1288static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1289{
1290        bio_io_error(bio);
1291}
1292
1293/*
1294 * FIXME: should we also commit due to size of transaction, measured in
1295 * metadata blocks?
1296 */
1297static int need_commit_due_to_time(struct pool *pool)
1298{
1299        return jiffies < pool->last_commit_jiffies ||
1300               jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1301}
1302
1303static void process_deferred_bios(struct pool *pool)
1304{
1305        unsigned long flags;
1306        struct bio *bio;
1307        struct bio_list bios;
1308
1309        bio_list_init(&bios);
1310
1311        spin_lock_irqsave(&pool->lock, flags);
1312        bio_list_merge(&bios, &pool->deferred_bios);
1313        bio_list_init(&pool->deferred_bios);
1314        spin_unlock_irqrestore(&pool->lock, flags);
1315
1316        while ((bio = bio_list_pop(&bios))) {
1317                struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1318                struct thin_c *tc = h->tc;
1319
1320                /*
1321                 * If we've got no free new_mapping structs, and processing
1322                 * this bio might require one, we pause until there are some
1323                 * prepared mappings to process.
1324                 */
1325                if (ensure_next_mapping(pool)) {
1326                        spin_lock_irqsave(&pool->lock, flags);
1327                        bio_list_merge(&pool->deferred_bios, &bios);
1328                        spin_unlock_irqrestore(&pool->lock, flags);
1329
1330                        break;
1331                }
1332
1333                if (bio->bi_rw & REQ_DISCARD)
1334                        pool->process_discard(tc, bio);
1335                else
1336                        pool->process_bio(tc, bio);
1337        }
1338
1339        /*
1340         * If there are any deferred flush bios, we must commit
1341         * the metadata before issuing them.
1342         */
1343        bio_list_init(&bios);
1344        spin_lock_irqsave(&pool->lock, flags);
1345        bio_list_merge(&bios, &pool->deferred_flush_bios);
1346        bio_list_init(&pool->deferred_flush_bios);
1347        spin_unlock_irqrestore(&pool->lock, flags);
1348
1349        if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
1350                return;
1351
1352        if (commit_or_fallback(pool)) {
1353                while ((bio = bio_list_pop(&bios)))
1354                        bio_io_error(bio);
1355                return;
1356        }
1357        pool->last_commit_jiffies = jiffies;
1358
1359        while ((bio = bio_list_pop(&bios)))
1360                generic_make_request(bio);
1361}
1362
1363static void do_worker(struct work_struct *ws)
1364{
1365        struct pool *pool = container_of(ws, struct pool, worker);
1366
1367        process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1368        process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1369        process_deferred_bios(pool);
1370}
1371
1372/*
1373 * We want to commit periodically so that not too much
1374 * unwritten data builds up.
1375 */
1376static void do_waker(struct work_struct *ws)
1377{
1378        struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1379        wake_worker(pool);
1380        queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1381}
1382
1383/*----------------------------------------------------------------*/
1384
1385static enum pool_mode get_pool_mode(struct pool *pool)
1386{
1387        return pool->pf.mode;
1388}
1389
1390static void set_pool_mode(struct pool *pool, enum pool_mode mode)
1391{
1392        int r;
1393
1394        pool->pf.mode = mode;
1395
1396        switch (mode) {
1397        case PM_FAIL:
1398                DMERR("%s: switching pool to failure mode",
1399                      dm_device_name(pool->pool_md));
1400                pool->process_bio = process_bio_fail;
1401                pool->process_discard = process_bio_fail;
1402                pool->process_prepared_mapping = process_prepared_mapping_fail;
1403                pool->process_prepared_discard = process_prepared_discard_fail;
1404                break;
1405
1406        case PM_READ_ONLY:
1407                DMERR("%s: switching pool to read-only mode",
1408                      dm_device_name(pool->pool_md));
1409                r = dm_pool_abort_metadata(pool->pmd);
1410                if (r) {
1411                        DMERR("%s: aborting transaction failed",
1412                              dm_device_name(pool->pool_md));
1413                        set_pool_mode(pool, PM_FAIL);
1414                } else {
1415                        dm_pool_metadata_read_only(pool->pmd);
1416                        pool->process_bio = process_bio_read_only;
1417                        pool->process_discard = process_discard;
1418                        pool->process_prepared_mapping = process_prepared_mapping_fail;
1419                        pool->process_prepared_discard = process_prepared_discard_passdown;
1420                }
1421                break;
1422
1423        case PM_WRITE:
1424                pool->process_bio = process_bio;
1425                pool->process_discard = process_discard;
1426                pool->process_prepared_mapping = process_prepared_mapping;
1427                pool->process_prepared_discard = process_prepared_discard;
1428                break;
1429        }
1430}
1431
1432/*----------------------------------------------------------------*/
1433
1434/*
1435 * Mapping functions.
1436 */
1437
1438/*
1439 * Called only while mapping a thin bio to hand it over to the workqueue.
1440 */
1441static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1442{
1443        unsigned long flags;
1444        struct pool *pool = tc->pool;
1445
1446        spin_lock_irqsave(&pool->lock, flags);
1447        bio_list_add(&pool->deferred_bios, bio);
1448        spin_unlock_irqrestore(&pool->lock, flags);
1449
1450        wake_worker(pool);
1451}
1452
1453static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1454{
1455        struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1456
1457        h->tc = tc;
1458        h->shared_read_entry = NULL;
1459        h->all_io_entry = NULL;
1460        h->overwrite_mapping = NULL;
1461}
1462
1463/*
1464 * Non-blocking function called from the thin target's map function.
1465 */
1466static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1467{
1468        int r;
1469        struct thin_c *tc = ti->private;
1470        dm_block_t block = get_bio_block(tc, bio);
1471        struct dm_thin_device *td = tc->td;
1472        struct dm_thin_lookup_result result;
1473        struct dm_bio_prison_cell cell1, cell2;
1474        struct dm_bio_prison_cell *cell_result;
1475        struct dm_cell_key key;
1476
1477        thin_hook_bio(tc, bio);
1478
1479        if (get_pool_mode(tc->pool) == PM_FAIL) {
1480                bio_io_error(bio);
1481                return DM_MAPIO_SUBMITTED;
1482        }
1483
1484        if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1485                thin_defer_bio(tc, bio);
1486                return DM_MAPIO_SUBMITTED;
1487        }
1488
1489        r = dm_thin_find_block(td, block, 0, &result);
1490
1491        /*
1492         * Note that we defer readahead too.
1493         */
1494        switch (r) {
1495        case 0:
1496                if (unlikely(result.shared)) {
1497                        /*
1498                         * We have a race condition here between the
1499                         * result.shared value returned by the lookup and
1500                         * snapshot creation, which may cause new
1501                         * sharing.
1502                         *
1503                         * To avoid this always quiesce the origin before
1504                         * taking the snap.  You want to do this anyway to
1505                         * ensure a consistent application view
1506                         * (i.e. lockfs).
1507                         *
1508                         * More distant ancestors are irrelevant. The
1509                         * shared flag will be set in their case.
1510                         */
1511                        thin_defer_bio(tc, bio);
1512                        return DM_MAPIO_SUBMITTED;
1513                }
1514
1515                build_virtual_key(tc->td, block, &key);
1516                if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
1517                        return DM_MAPIO_SUBMITTED;
1518
1519                build_data_key(tc->td, result.block, &key);
1520                if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1521                        cell_defer_no_holder_no_free(tc, &cell1);
1522                        return DM_MAPIO_SUBMITTED;
1523                }
1524
1525                inc_all_io_entry(tc->pool, bio);
1526                cell_defer_no_holder_no_free(tc, &cell2);
1527                cell_defer_no_holder_no_free(tc, &cell1);
1528
1529                remap(tc, bio, result.block);
1530                return DM_MAPIO_REMAPPED;
1531
1532        case -ENODATA:
1533                if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1534                        /*
1535                         * This block isn't provisioned, and we have no way
1536                         * of doing so.  Just error it.
1537                         */
1538                        bio_io_error(bio);
1539                        return DM_MAPIO_SUBMITTED;
1540                }
1541                /* fall through */
1542
1543        case -EWOULDBLOCK:
1544                /*
1545                 * In future, the failed dm_thin_find_block above could
1546                 * provide the hint to load the metadata into cache.
1547                 */
1548                thin_defer_bio(tc, bio);
1549                return DM_MAPIO_SUBMITTED;
1550
1551        default:
1552                /*
1553                 * Must always call bio_io_error on failure.
1554                 * dm_thin_find_block can fail with -EINVAL if the
1555                 * pool is switched to fail-io mode.
1556                 */
1557                bio_io_error(bio);
1558                return DM_MAPIO_SUBMITTED;
1559        }
1560}
1561
1562static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1563{
1564        int r;
1565        unsigned long flags;
1566        struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1567
1568        spin_lock_irqsave(&pt->pool->lock, flags);
1569        r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1570        spin_unlock_irqrestore(&pt->pool->lock, flags);
1571
1572        if (!r) {
1573                struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1574                r = bdi_congested(&q->backing_dev_info, bdi_bits);
1575        }
1576
1577        return r;
1578}
1579
1580static void __requeue_bios(struct pool *pool)
1581{
1582        bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1583        bio_list_init(&pool->retry_on_resume_list);
1584}
1585
1586/*----------------------------------------------------------------
1587 * Binding of control targets to a pool object
1588 *--------------------------------------------------------------*/
1589static bool data_dev_supports_discard(struct pool_c *pt)
1590{
1591        struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1592
1593        return q && blk_queue_discard(q);
1594}
1595
1596static bool is_factor(sector_t block_size, uint32_t n)
1597{
1598        return !sector_div(block_size, n);
1599}
1600
1601/*
1602 * If discard_passdown was enabled verify that the data device
1603 * supports discards.  Disable discard_passdown if not.
1604 */
1605static void disable_passdown_if_not_supported(struct pool_c *pt)
1606{
1607        struct pool *pool = pt->pool;
1608        struct block_device *data_bdev = pt->data_dev->bdev;
1609        struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1610        sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1611        const char *reason = NULL;
1612        char buf[BDEVNAME_SIZE];
1613
1614        if (!pt->adjusted_pf.discard_passdown)
1615                return;
1616
1617        if (!data_dev_supports_discard(pt))
1618                reason = "discard unsupported";
1619
1620        else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1621                reason = "max discard sectors smaller than a block";
1622
1623        else if (data_limits->discard_granularity > block_size)
1624                reason = "discard granularity larger than a block";
1625
1626        else if (!is_factor(block_size, data_limits->discard_granularity))
1627                reason = "discard granularity not a factor of block size";
1628
1629        if (reason) {
1630                DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1631                pt->adjusted_pf.discard_passdown = false;
1632        }
1633}
1634
1635static int bind_control_target(struct pool *pool, struct dm_target *ti)
1636{
1637        struct pool_c *pt = ti->private;
1638
1639        /*
1640         * We want to make sure that degraded pools are never upgraded.
1641         */
1642        enum pool_mode old_mode = pool->pf.mode;
1643        enum pool_mode new_mode = pt->adjusted_pf.mode;
1644
1645        if (old_mode > new_mode)
1646                new_mode = old_mode;
1647
1648        pool->ti = ti;
1649        pool->low_water_blocks = pt->low_water_blocks;
1650        pool->pf = pt->adjusted_pf;
1651
1652        set_pool_mode(pool, new_mode);
1653
1654        return 0;
1655}
1656
1657static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1658{
1659        if (pool->ti == ti)
1660                pool->ti = NULL;
1661}
1662
1663/*----------------------------------------------------------------
1664 * Pool creation
1665 *--------------------------------------------------------------*/
1666/* Initialize pool features. */
1667static void pool_features_init(struct pool_features *pf)
1668{
1669        pf->mode = PM_WRITE;
1670        pf->zero_new_blocks = true;
1671        pf->discard_enabled = true;
1672        pf->discard_passdown = true;
1673}
1674
1675static void __pool_destroy(struct pool *pool)
1676{
1677        __pool_table_remove(pool);
1678
1679        if (dm_pool_metadata_close(pool->pmd) < 0)
1680                DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1681
1682        dm_bio_prison_destroy(pool->prison);
1683        dm_kcopyd_client_destroy(pool->copier);
1684
1685        if (pool->wq)
1686                destroy_workqueue(pool->wq);
1687
1688        if (pool->next_mapping)
1689                mempool_free(pool->next_mapping, pool->mapping_pool);
1690        mempool_destroy(pool->mapping_pool);
1691        dm_deferred_set_destroy(pool->shared_read_ds);
1692        dm_deferred_set_destroy(pool->all_io_ds);
1693        kfree(pool);
1694}
1695
1696static struct kmem_cache *_new_mapping_cache;
1697
1698static struct pool *pool_create(struct mapped_device *pool_md,
1699                                struct block_device *metadata_dev,
1700                                unsigned long block_size,
1701                                int read_only, char **error)
1702{
1703        int r;
1704        void *err_p;
1705        struct pool *pool;
1706        struct dm_pool_metadata *pmd;
1707        bool format_device = read_only ? false : true;
1708
1709        pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
1710        if (IS_ERR(pmd)) {
1711                *error = "Error creating metadata object";
1712                return (struct pool *)pmd;
1713        }
1714
1715        pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1716        if (!pool) {
1717                *error = "Error allocating memory for pool";
1718                err_p = ERR_PTR(-ENOMEM);
1719                goto bad_pool;
1720        }
1721
1722        pool->pmd = pmd;
1723        pool->sectors_per_block = block_size;
1724        if (block_size & (block_size - 1))
1725                pool->sectors_per_block_shift = -1;
1726        else
1727                pool->sectors_per_block_shift = __ffs(block_size);
1728        pool->low_water_blocks = 0;
1729        pool_features_init(&pool->pf);
1730        pool->prison = dm_bio_prison_create(PRISON_CELLS);
1731        if (!pool->prison) {
1732                *error = "Error creating pool's bio prison";
1733                err_p = ERR_PTR(-ENOMEM);
1734                goto bad_prison;
1735        }
1736
1737        pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1738        if (IS_ERR(pool->copier)) {
1739                r = PTR_ERR(pool->copier);
1740                *error = "Error creating pool's kcopyd client";
1741                err_p = ERR_PTR(r);
1742                goto bad_kcopyd_client;
1743        }
1744
1745        /*
1746         * Create singlethreaded workqueue that will service all devices
1747         * that use this metadata.
1748         */
1749        pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1750        if (!pool->wq) {
1751                *error = "Error creating pool's workqueue";
1752                err_p = ERR_PTR(-ENOMEM);
1753                goto bad_wq;
1754        }
1755
1756        INIT_WORK(&pool->worker, do_worker);
1757        INIT_DELAYED_WORK(&pool->waker, do_waker);
1758        spin_lock_init(&pool->lock);
1759        bio_list_init(&pool->deferred_bios);
1760        bio_list_init(&pool->deferred_flush_bios);
1761        INIT_LIST_HEAD(&pool->prepared_mappings);
1762        INIT_LIST_HEAD(&pool->prepared_discards);
1763        pool->low_water_triggered = 0;
1764        pool->no_free_space = 0;
1765        bio_list_init(&pool->retry_on_resume_list);
1766
1767        pool->shared_read_ds = dm_deferred_set_create();
1768        if (!pool->shared_read_ds) {
1769                *error = "Error creating pool's shared read deferred set";
1770                err_p = ERR_PTR(-ENOMEM);
1771                goto bad_shared_read_ds;
1772        }
1773
1774        pool->all_io_ds = dm_deferred_set_create();
1775        if (!pool->all_io_ds) {
1776                *error = "Error creating pool's all io deferred set";
1777                err_p = ERR_PTR(-ENOMEM);
1778                goto bad_all_io_ds;
1779        }
1780
1781        pool->next_mapping = NULL;
1782        pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1783                                                      _new_mapping_cache);
1784        if (!pool->mapping_pool) {
1785                *error = "Error creating pool's mapping mempool";
1786                err_p = ERR_PTR(-ENOMEM);
1787                goto bad_mapping_pool;
1788        }
1789
1790        pool->ref_count = 1;
1791        pool->last_commit_jiffies = jiffies;
1792        pool->pool_md = pool_md;
1793        pool->md_dev = metadata_dev;
1794        __pool_table_insert(pool);
1795
1796        return pool;
1797
1798bad_mapping_pool:
1799        dm_deferred_set_destroy(pool->all_io_ds);
1800bad_all_io_ds:
1801        dm_deferred_set_destroy(pool->shared_read_ds);
1802bad_shared_read_ds:
1803        destroy_workqueue(pool->wq);
1804bad_wq:
1805        dm_kcopyd_client_destroy(pool->copier);
1806bad_kcopyd_client:
1807        dm_bio_prison_destroy(pool->prison);
1808bad_prison:
1809        kfree(pool);
1810bad_pool:
1811        if (dm_pool_metadata_close(pmd))
1812                DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1813
1814        return err_p;
1815}
1816
1817static void __pool_inc(struct pool *pool)
1818{
1819        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1820        pool->ref_count++;
1821}
1822
1823static void __pool_dec(struct pool *pool)
1824{
1825        BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1826        BUG_ON(!pool->ref_count);
1827        if (!--pool->ref_count)
1828                __pool_destroy(pool);
1829}
1830
1831static struct pool *__pool_find(struct mapped_device *pool_md,
1832                                struct block_device *metadata_dev,
1833                                unsigned long block_size, int read_only,
1834                                char **error, int *created)
1835{
1836        struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1837
1838        if (pool) {
1839                if (pool->pool_md != pool_md) {
1840                        *error = "metadata device already in use by a pool";
1841                        return ERR_PTR(-EBUSY);
1842                }
1843                __pool_inc(pool);
1844
1845        } else {
1846                pool = __pool_table_lookup(pool_md);
1847                if (pool) {
1848                        if (pool->md_dev != metadata_dev) {
1849                                *error = "different pool cannot replace a pool";
1850                                return ERR_PTR(-EINVAL);
1851                        }
1852                        __pool_inc(pool);
1853
1854                } else {
1855                        pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
1856                        *created = 1;
1857                }
1858        }
1859
1860        return pool;
1861}
1862
1863/*----------------------------------------------------------------
1864 * Pool target methods
1865 *--------------------------------------------------------------*/
1866static void pool_dtr(struct dm_target *ti)
1867{
1868        struct pool_c *pt = ti->private;
1869
1870        mutex_lock(&dm_thin_pool_table.mutex);
1871
1872        unbind_control_target(pt->pool, ti);
1873        __pool_dec(pt->pool);
1874        dm_put_device(ti, pt->metadata_dev);
1875        dm_put_device(ti, pt->data_dev);
1876        kfree(pt);
1877
1878        mutex_unlock(&dm_thin_pool_table.mutex);
1879}
1880
1881static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1882                               struct dm_target *ti)
1883{
1884        int r;
1885        unsigned argc;
1886        const char *arg_name;
1887
1888        static struct dm_arg _args[] = {
1889                {0, 3, "Invalid number of pool feature arguments"},
1890        };
1891
1892        /*
1893         * No feature arguments supplied.
1894         */
1895        if (!as->argc)
1896                return 0;
1897
1898        r = dm_read_arg_group(_args, as, &argc, &ti->error);
1899        if (r)
1900                return -EINVAL;
1901
1902        while (argc && !r) {
1903                arg_name = dm_shift_arg(as);
1904                argc--;
1905
1906                if (!strcasecmp(arg_name, "skip_block_zeroing"))
1907                        pf->zero_new_blocks = false;
1908
1909                else if (!strcasecmp(arg_name, "ignore_discard"))
1910                        pf->discard_enabled = false;
1911
1912                else if (!strcasecmp(arg_name, "no_discard_passdown"))
1913                        pf->discard_passdown = false;
1914
1915                else if (!strcasecmp(arg_name, "read_only"))
1916                        pf->mode = PM_READ_ONLY;
1917
1918                else {
1919                        ti->error = "Unrecognised pool feature requested";
1920                        r = -EINVAL;
1921                        break;
1922                }
1923        }
1924
1925        return r;
1926}
1927
1928static void metadata_low_callback(void *context)
1929{
1930        struct pool *pool = context;
1931
1932        DMWARN("%s: reached low water mark for metadata device: sending event.",
1933               dm_device_name(pool->pool_md));
1934
1935        dm_table_event(pool->ti->table);
1936}
1937
1938static sector_t get_metadata_dev_size(struct block_device *bdev)
1939{
1940        sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1941        char buffer[BDEVNAME_SIZE];
1942
1943        if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
1944                DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1945                       bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
1946                metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
1947        }
1948
1949        return metadata_dev_size;
1950}
1951
1952static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
1953{
1954        sector_t metadata_dev_size = get_metadata_dev_size(bdev);
1955
1956        sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
1957
1958        return metadata_dev_size;
1959}
1960
1961/*
1962 * When a metadata threshold is crossed a dm event is triggered, and
1963 * userland should respond by growing the metadata device.  We could let
1964 * userland set the threshold, like we do with the data threshold, but I'm
1965 * not sure they know enough to do this well.
1966 */
1967static dm_block_t calc_metadata_threshold(struct pool_c *pt)
1968{
1969        /*
1970         * 4M is ample for all ops with the possible exception of thin
1971         * device deletion which is harmless if it fails (just retry the
1972         * delete after you've grown the device).
1973         */
1974        dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
1975        return min((dm_block_t)1024ULL /* 4M */, quarter);
1976}
1977
1978/*
1979 * thin-pool <metadata dev> <data dev>
1980 *           <data block size (sectors)>
1981 *           <low water mark (blocks)>
1982 *           [<#feature args> [<arg>]*]
1983 *
1984 * Optional feature arguments are:
1985 *           skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1986 *           ignore_discard: disable discard
1987 *           no_discard_passdown: don't pass discards down to the data device
1988 */
1989static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1990{
1991        int r, pool_created = 0;
1992        struct pool_c *pt;
1993        struct pool *pool;
1994        struct pool_features pf;
1995        struct dm_arg_set as;
1996        struct dm_dev *data_dev;
1997        unsigned long block_size;
1998        dm_block_t low_water_blocks;
1999        struct dm_dev *metadata_dev;
2000        fmode_t metadata_mode;
2001
2002        /*
2003         * FIXME Remove validation from scope of lock.
2004         */
2005        mutex_lock(&dm_thin_pool_table.mutex);
2006
2007        if (argc < 4) {
2008                ti->error = "Invalid argument count";
2009                r = -EINVAL;
2010                goto out_unlock;
2011        }
2012
2013        as.argc = argc;
2014        as.argv = argv;
2015
2016        /*
2017         * Set default pool features.
2018         */
2019        pool_features_init(&pf);
2020
2021        dm_consume_args(&as, 4);
2022        r = parse_pool_features(&as, &pf, ti);
2023        if (r)
2024                goto out_unlock;
2025
2026        metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2027        r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2028        if (r) {
2029                ti->error = "Error opening metadata block device";
2030                goto out_unlock;
2031        }
2032
2033        /*
2034         * Run for the side-effect of possibly issuing a warning if the
2035         * device is too big.
2036         */
2037        (void) get_metadata_dev_size(metadata_dev->bdev);
2038
2039        r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2040        if (r) {
2041                ti->error = "Error getting data device";
2042                goto out_metadata;
2043        }
2044
2045        if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2046            block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2047            block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2048            block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2049                ti->error = "Invalid block size";
2050                r = -EINVAL;
2051                goto out;
2052        }
2053
2054        if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2055                ti->error = "Invalid low water mark";
2056                r = -EINVAL;
2057                goto out;
2058        }
2059
2060        pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2061        if (!pt) {
2062                r = -ENOMEM;
2063                goto out;
2064        }
2065
2066        pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2067                           block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2068        if (IS_ERR(pool)) {
2069                r = PTR_ERR(pool);
2070                goto out_free_pt;
2071        }
2072
2073        /*
2074         * 'pool_created' reflects whether this is the first table load.
2075         * Top level discard support is not allowed to be changed after
2076         * initial load.  This would require a pool reload to trigger thin
2077         * device changes.
2078         */
2079        if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2080                ti->error = "Discard support cannot be disabled once enabled";
2081                r = -EINVAL;
2082                goto out_flags_changed;
2083        }
2084
2085        pt->pool = pool;
2086        pt->ti = ti;
2087        pt->metadata_dev = metadata_dev;
2088        pt->data_dev = data_dev;
2089        pt->low_water_blocks = low_water_blocks;
2090        pt->adjusted_pf = pt->requested_pf = pf;
2091        ti->num_flush_bios = 1;
2092
2093        /*
2094         * Only need to enable discards if the pool should pass
2095         * them down to the data device.  The thin device's discard
2096         * processing will cause mappings to be removed from the btree.
2097         */
2098        ti->discard_zeroes_data_unsupported = true;
2099        if (pf.discard_enabled && pf.discard_passdown) {
2100                ti->num_discard_bios = 1;
2101
2102                /*
2103                 * Setting 'discards_supported' circumvents the normal
2104                 * stacking of discard limits (this keeps the pool and
2105                 * thin devices' discard limits consistent).
2106                 */
2107                ti->discards_supported = true;
2108        }
2109        ti->private = pt;
2110
2111        r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2112                                                calc_metadata_threshold(pt),
2113                                                metadata_low_callback,
2114                                                pool);
2115        if (r)
2116                goto out_free_pt;
2117
2118        pt->callbacks.congested_fn = pool_is_congested;
2119        dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2120
2121        mutex_unlock(&dm_thin_pool_table.mutex);
2122
2123        return 0;
2124
2125out_flags_changed:
2126        __pool_dec(pool);
2127out_free_pt:
2128        kfree(pt);
2129out:
2130        dm_put_device(ti, data_dev);
2131out_metadata:
2132        dm_put_device(ti, metadata_dev);
2133out_unlock:
2134        mutex_unlock(&dm_thin_pool_table.mutex);
2135
2136        return r;
2137}
2138
2139static int pool_map(struct dm_target *ti, struct bio *bio)
2140{
2141        int r;
2142        struct pool_c *pt = ti->private;
2143        struct pool *pool = pt->pool;
2144        unsigned long flags;
2145
2146        /*
2147         * As this is a singleton target, ti->begin is always zero.
2148         */
2149        spin_lock_irqsave(&pool->lock, flags);
2150        bio->bi_bdev = pt->data_dev->bdev;
2151        r = DM_MAPIO_REMAPPED;
2152        spin_unlock_irqrestore(&pool->lock, flags);
2153
2154        return r;
2155}
2156
2157static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2158{
2159        int r;
2160        struct pool_c *pt = ti->private;
2161        struct pool *pool = pt->pool;
2162        sector_t data_size = ti->len;
2163        dm_block_t sb_data_size;
2164
2165        *need_commit = false;
2166
2167        (void) sector_div(data_size, pool->sectors_per_block);
2168
2169        r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2170        if (r) {
2171                DMERR("%s: failed to retrieve data device size",
2172                      dm_device_name(pool->pool_md));
2173                return r;
2174        }
2175
2176        if (data_size < sb_data_size) {
2177                DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2178                      dm_device_name(pool->pool_md),
2179                      (unsigned long long)data_size, sb_data_size);
2180                return -EINVAL;
2181
2182        } else if (data_size > sb_data_size) {
2183                r = dm_pool_resize_data_dev(pool->pmd, data_size);
2184                if (r) {
2185                        DMERR("%s: failed to resize data device",
2186                              dm_device_name(pool->pool_md));
2187                        set_pool_mode(pool, PM_READ_ONLY);
2188                        return r;
2189                }
2190
2191                *need_commit = true;
2192        }
2193
2194        return 0;
2195}
2196
2197static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2198{
2199        int r;
2200        struct pool_c *pt = ti->private;
2201        struct pool *pool = pt->pool;
2202        dm_block_t metadata_dev_size, sb_metadata_dev_size;
2203
2204        *need_commit = false;
2205
2206        metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2207
2208        r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2209        if (r) {
2210                DMERR("%s: failed to retrieve metadata device size",
2211                      dm_device_name(pool->pool_md));
2212                return r;
2213        }
2214
2215        if (metadata_dev_size < sb_metadata_dev_size) {
2216                DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2217                      dm_device_name(pool->pool_md),
2218                      metadata_dev_size, sb_metadata_dev_size);
2219                return -EINVAL;
2220
2221        } else if (metadata_dev_size > sb_metadata_dev_size) {
2222                r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2223                if (r) {
2224                        DMERR("%s: failed to resize metadata device",
2225                              dm_device_name(pool->pool_md));
2226                        return r;
2227                }
2228
2229                *need_commit = true;
2230        }
2231
2232        return 0;
2233}
2234
2235/*
2236 * Retrieves the number of blocks of the data device from
2237 * the superblock and compares it to the actual device size,
2238 * thus resizing the data device in case it has grown.
2239 *
2240 * This both copes with opening preallocated data devices in the ctr
2241 * being followed by a resume
2242 * -and-
2243 * calling the resume method individually after userspace has
2244 * grown the data device in reaction to a table event.
2245 */
2246static int pool_preresume(struct dm_target *ti)
2247{
2248        int r;
2249        bool need_commit1, need_commit2;
2250        struct pool_c *pt = ti->private;
2251        struct pool *pool = pt->pool;
2252
2253        /*
2254         * Take control of the pool object.
2255         */
2256        r = bind_control_target(pool, ti);
2257        if (r)
2258                return r;
2259
2260        r = maybe_resize_data_dev(ti, &need_commit1);
2261        if (r)
2262                return r;
2263
2264        r = maybe_resize_metadata_dev(ti, &need_commit2);
2265        if (r)
2266                return r;
2267
2268        if (need_commit1 || need_commit2)
2269                (void) commit_or_fallback(pool);
2270
2271        return 0;
2272}
2273
2274static void pool_resume(struct dm_target *ti)
2275{
2276        struct pool_c *pt = ti->private;
2277        struct pool *pool = pt->pool;
2278        unsigned long flags;
2279
2280        spin_lock_irqsave(&pool->lock, flags);
2281        pool->low_water_triggered = 0;
2282        pool->no_free_space = 0;
2283        __requeue_bios(pool);
2284        spin_unlock_irqrestore(&pool->lock, flags);
2285
2286        do_waker(&pool->waker.work);
2287}
2288
2289static void pool_postsuspend(struct dm_target *ti)
2290{
2291        struct pool_c *pt = ti->private;
2292        struct pool *pool = pt->pool;
2293
2294        cancel_delayed_work(&pool->waker);
2295        flush_workqueue(pool->wq);
2296        (void) commit_or_fallback(pool);
2297}
2298
2299static int check_arg_count(unsigned argc, unsigned args_required)
2300{
2301        if (argc != args_required) {
2302                DMWARN("Message received with %u arguments instead of %u.",
2303                       argc, args_required);
2304                return -EINVAL;
2305        }
2306
2307        return 0;
2308}
2309
2310static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2311{
2312        if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2313            *dev_id <= MAX_DEV_ID)
2314                return 0;
2315
2316        if (warning)
2317                DMWARN("Message received with invalid device id: %s", arg);
2318
2319        return -EINVAL;
2320}
2321
2322static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2323{
2324        dm_thin_id dev_id;
2325        int r;
2326
2327        r = check_arg_count(argc, 2);
2328        if (r)
2329                return r;
2330
2331        r = read_dev_id(argv[1], &dev_id, 1);
2332        if (r)
2333                return r;
2334
2335        r = dm_pool_create_thin(pool->pmd, dev_id);
2336        if (r) {
2337                DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2338                       argv[1]);
2339                return r;
2340        }
2341
2342        return 0;
2343}
2344
2345static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2346{
2347        dm_thin_id dev_id;
2348        dm_thin_id origin_dev_id;
2349        int r;
2350
2351        r = check_arg_count(argc, 3);
2352        if (r)
2353                return r;
2354
2355        r = read_dev_id(argv[1], &dev_id, 1);
2356        if (r)
2357                return r;
2358
2359        r = read_dev_id(argv[2], &origin_dev_id, 1);
2360        if (r)
2361                return r;
2362
2363        r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2364        if (r) {
2365                DMWARN("Creation of new snapshot %s of device %s failed.",
2366                       argv[1], argv[2]);
2367                return r;
2368        }
2369
2370        return 0;
2371}
2372
2373static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2374{
2375        dm_thin_id dev_id;
2376        int r;
2377
2378        r = check_arg_count(argc, 2);
2379        if (r)
2380                return r;
2381
2382        r = read_dev_id(argv[1], &dev_id, 1);
2383        if (r)
2384                return r;
2385
2386        r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2387        if (r)
2388                DMWARN("Deletion of thin device %s failed.", argv[1]);
2389
2390        return r;
2391}
2392
2393static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2394{
2395        dm_thin_id old_id, new_id;
2396        int r;
2397
2398        r = check_arg_count(argc, 3);
2399        if (r)
2400                return r;
2401
2402        if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2403                DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2404                return -EINVAL;
2405        }
2406
2407        if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2408                DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2409                return -EINVAL;
2410        }
2411
2412        r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2413        if (r) {
2414                DMWARN("Failed to change transaction id from %s to %s.",
2415                       argv[1], argv[2]);
2416                return r;
2417        }
2418
2419        return 0;
2420}
2421
2422static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2423{
2424        int r;
2425
2426        r = check_arg_count(argc, 1);
2427        if (r)
2428                return r;
2429
2430        (void) commit_or_fallback(pool);
2431
2432        r = dm_pool_reserve_metadata_snap(pool->pmd);
2433        if (r)
2434                DMWARN("reserve_metadata_snap message failed.");
2435
2436        return r;
2437}
2438
2439static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2440{
2441        int r;
2442
2443        r = check_arg_count(argc, 1);
2444        if (r)
2445                return r;
2446
2447        r = dm_pool_release_metadata_snap(pool->pmd);
2448        if (r)
2449                DMWARN("release_metadata_snap message failed.");
2450
2451        return r;
2452}
2453
2454/*
2455 * Messages supported:
2456 *   create_thin        <dev_id>
2457 *   create_snap        <dev_id> <origin_id>
2458 *   delete             <dev_id>
2459 *   trim               <dev_id> <new_size_in_sectors>
2460 *   set_transaction_id <current_trans_id> <new_trans_id>
2461 *   reserve_metadata_snap
2462 *   release_metadata_snap
2463 */
2464static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2465{
2466        int r = -EINVAL;
2467        struct pool_c *pt = ti->private;
2468        struct pool *pool = pt->pool;
2469
2470        if (!strcasecmp(argv[0], "create_thin"))
2471                r = process_create_thin_mesg(argc, argv, pool);
2472
2473        else if (!strcasecmp(argv[0], "create_snap"))
2474                r = process_create_snap_mesg(argc, argv, pool);
2475
2476        else if (!strcasecmp(argv[0], "delete"))
2477                r = process_delete_mesg(argc, argv, pool);
2478
2479        else if (!strcasecmp(argv[0], "set_transaction_id"))
2480                r = process_set_transaction_id_mesg(argc, argv, pool);
2481
2482        else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2483                r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2484
2485        else if (!strcasecmp(argv[0], "release_metadata_snap"))
2486                r = process_release_metadata_snap_mesg(argc, argv, pool);
2487
2488        else
2489                DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2490
2491        if (!r)
2492                (void) commit_or_fallback(pool);
2493
2494        return r;
2495}
2496
2497static void emit_flags(struct pool_features *pf, char *result,
2498                       unsigned sz, unsigned maxlen)
2499{
2500        unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2501                !pf->discard_passdown + (pf->mode == PM_READ_ONLY);
2502        DMEMIT("%u ", count);
2503
2504        if (!pf->zero_new_blocks)
2505                DMEMIT("skip_block_zeroing ");
2506
2507        if (!pf->discard_enabled)
2508                DMEMIT("ignore_discard ");
2509
2510        if (!pf->discard_passdown)
2511                DMEMIT("no_discard_passdown ");
2512
2513        if (pf->mode == PM_READ_ONLY)
2514                DMEMIT("read_only ");
2515}
2516
2517/*
2518 * Status line is:
2519 *    <transaction id> <used metadata sectors>/<total metadata sectors>
2520 *    <used data sectors>/<total data sectors> <held metadata root>
2521 */
2522static void pool_status(struct dm_target *ti, status_type_t type,
2523                        unsigned status_flags, char *result, unsigned maxlen)
2524{
2525        int r;
2526        unsigned sz = 0;
2527        uint64_t transaction_id;
2528        dm_block_t nr_free_blocks_data;
2529        dm_block_t nr_free_blocks_metadata;
2530        dm_block_t nr_blocks_data;
2531        dm_block_t nr_blocks_metadata;
2532        dm_block_t held_root;
2533        char buf[BDEVNAME_SIZE];
2534        char buf2[BDEVNAME_SIZE];
2535        struct pool_c *pt = ti->private;
2536        struct pool *pool = pt->pool;
2537
2538        switch (type) {
2539        case STATUSTYPE_INFO:
2540                if (get_pool_mode(pool) == PM_FAIL) {
2541                        DMEMIT("Fail");
2542                        break;
2543                }
2544
2545                /* Commit to ensure statistics aren't out-of-date */
2546                if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
2547                        (void) commit_or_fallback(pool);
2548
2549                r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2550                if (r) {
2551                        DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2552                              dm_device_name(pool->pool_md), r);
2553                        goto err;
2554                }
2555
2556                r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2557                if (r) {
2558                        DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2559                              dm_device_name(pool->pool_md), r);
2560                        goto err;
2561                }
2562
2563                r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2564                if (r) {
2565                        DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2566                              dm_device_name(pool->pool_md), r);
2567                        goto err;
2568                }
2569
2570                r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2571                if (r) {
2572                        DMERR("%s: dm_pool_get_free_block_count returned %d",
2573                              dm_device_name(pool->pool_md), r);
2574                        goto err;
2575                }
2576
2577                r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2578                if (r) {
2579                        DMERR("%s: dm_pool_get_data_dev_size returned %d",
2580                              dm_device_name(pool->pool_md), r);
2581                        goto err;
2582                }
2583
2584                r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
2585                if (r) {
2586                        DMERR("%s: dm_pool_get_metadata_snap returned %d",
2587                              dm_device_name(pool->pool_md), r);
2588                        goto err;
2589                }
2590
2591                DMEMIT("%llu %llu/%llu %llu/%llu ",
2592                       (unsigned long long)transaction_id,
2593                       (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2594                       (unsigned long long)nr_blocks_metadata,
2595                       (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2596                       (unsigned long long)nr_blocks_data);
2597
2598                if (held_root)
2599                        DMEMIT("%llu ", held_root);
2600                else
2601                        DMEMIT("- ");
2602
2603                if (pool->pf.mode == PM_READ_ONLY)
2604                        DMEMIT("ro ");
2605                else
2606                        DMEMIT("rw ");
2607
2608                if (!pool->pf.discard_enabled)
2609                        DMEMIT("ignore_discard");
2610                else if (pool->pf.discard_passdown)
2611                        DMEMIT("discard_passdown");
2612                else
2613                        DMEMIT("no_discard_passdown");
2614
2615                break;
2616
2617        case STATUSTYPE_TABLE:
2618                DMEMIT("%s %s %lu %llu ",
2619                       format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2620                       format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2621                       (unsigned long)pool->sectors_per_block,
2622                       (unsigned long long)pt->low_water_blocks);
2623                emit_flags(&pt->requested_pf, result, sz, maxlen);
2624                break;
2625        }
2626        return;
2627
2628err:
2629        DMEMIT("Error");
2630}
2631
2632static int pool_iterate_devices(struct dm_target *ti,
2633                                iterate_devices_callout_fn fn, void *data)
2634{
2635        struct pool_c *pt = ti->private;
2636
2637        return fn(ti, pt->data_dev, 0, ti->len, data);
2638}
2639
2640static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2641                      struct bio_vec *biovec, int max_size)
2642{
2643        struct pool_c *pt = ti->private;
2644        struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2645
2646        if (!q->merge_bvec_fn)
2647                return max_size;
2648
2649        bvm->bi_bdev = pt->data_dev->bdev;
2650
2651        return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2652}
2653
2654static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
2655{
2656        struct pool *pool = pt->pool;
2657        struct queue_limits *data_limits;
2658
2659        limits->max_discard_sectors = pool->sectors_per_block;
2660
2661        /*
2662         * discard_granularity is just a hint, and not enforced.
2663         */
2664        if (pt->adjusted_pf.discard_passdown) {
2665                data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2666                limits->discard_granularity = data_limits->discard_granularity;
2667        } else
2668                limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
2669}
2670
2671static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2672{
2673        struct pool_c *pt = ti->private;
2674        struct pool *pool = pt->pool;
2675        uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
2676
2677        /*
2678         * If the system-determined stacked limits are compatible with the
2679         * pool's blocksize (io_opt is a factor) do not override them.
2680         */
2681        if (io_opt_sectors < pool->sectors_per_block ||
2682            do_div(io_opt_sectors, pool->sectors_per_block)) {
2683                blk_limits_io_min(limits, 0);
2684                blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2685        }
2686
2687        /*
2688         * pt->adjusted_pf is a staging area for the actual features to use.
2689         * They get transferred to the live pool in bind_control_target()
2690         * called from pool_preresume().
2691         */
2692        if (!pt->adjusted_pf.discard_enabled) {
2693                /*
2694                 * Must explicitly disallow stacking discard limits otherwise the
2695                 * block layer will stack them if pool's data device has support.
2696                 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2697                 * user to see that, so make sure to set all discard limits to 0.
2698                 */
2699                limits->discard_granularity = 0;
2700                return;
2701        }
2702
2703        disable_passdown_if_not_supported(pt);
2704
2705        set_discard_limits(pt, limits);
2706}
2707
2708static struct target_type pool_target = {
2709        .name = "thin-pool",
2710        .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2711                    DM_TARGET_IMMUTABLE,
2712        .version = {1, 9, 0},
2713        .module = THIS_MODULE,
2714        .ctr = pool_ctr,
2715        .dtr = pool_dtr,
2716        .map = pool_map,
2717        .postsuspend = pool_postsuspend,
2718        .preresume = pool_preresume,
2719        .resume = pool_resume,
2720        .message = pool_message,
2721        .status = pool_status,
2722        .merge = pool_merge,
2723        .iterate_devices = pool_iterate_devices,
2724        .io_hints = pool_io_hints,
2725};
2726
2727/*----------------------------------------------------------------
2728 * Thin target methods
2729 *--------------------------------------------------------------*/
2730static void thin_dtr(struct dm_target *ti)
2731{
2732        struct thin_c *tc = ti->private;
2733
2734        mutex_lock(&dm_thin_pool_table.mutex);
2735
2736        __pool_dec(tc->pool);
2737        dm_pool_close_thin_device(tc->td);
2738        dm_put_device(ti, tc->pool_dev);
2739        if (tc->origin_dev)
2740                dm_put_device(ti, tc->origin_dev);
2741        kfree(tc);
2742
2743        mutex_unlock(&dm_thin_pool_table.mutex);
2744}
2745
2746/*
2747 * Thin target parameters:
2748 *
2749 * <pool_dev> <dev_id> [origin_dev]
2750 *
2751 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2752 * dev_id: the internal device identifier
2753 * origin_dev: a device external to the pool that should act as the origin
2754 *
2755 * If the pool device has discards disabled, they get disabled for the thin
2756 * device as well.
2757 */
2758static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2759{
2760        int r;
2761        struct thin_c *tc;
2762        struct dm_dev *pool_dev, *origin_dev;
2763        struct mapped_device *pool_md;
2764
2765        mutex_lock(&dm_thin_pool_table.mutex);
2766
2767        if (argc != 2 && argc != 3) {
2768                ti->error = "Invalid argument count";
2769                r = -EINVAL;
2770                goto out_unlock;
2771        }
2772
2773        tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2774        if (!tc) {
2775                ti->error = "Out of memory";
2776                r = -ENOMEM;
2777                goto out_unlock;
2778        }
2779
2780        if (argc == 3) {
2781                r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2782                if (r) {
2783                        ti->error = "Error opening origin device";
2784                        goto bad_origin_dev;
2785                }
2786                tc->origin_dev = origin_dev;
2787        }
2788
2789        r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2790        if (r) {
2791                ti->error = "Error opening pool device";
2792                goto bad_pool_dev;
2793        }
2794        tc->pool_dev = pool_dev;
2795
2796        if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2797                ti->error = "Invalid device id";
2798                r = -EINVAL;
2799                goto bad_common;
2800        }
2801
2802        pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2803        if (!pool_md) {
2804                ti->error = "Couldn't get pool mapped device";
2805                r = -EINVAL;
2806                goto bad_common;
2807        }
2808
2809        tc->pool = __pool_table_lookup(pool_md);
2810        if (!tc->pool) {
2811                ti->error = "Couldn't find pool object";
2812                r = -EINVAL;
2813                goto bad_pool_lookup;
2814        }
2815        __pool_inc(tc->pool);
2816
2817        if (get_pool_mode(tc->pool) == PM_FAIL) {
2818                ti->error = "Couldn't open thin device, Pool is in fail mode";
2819                goto bad_thin_open;
2820        }
2821
2822        r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2823        if (r) {
2824                ti->error = "Couldn't open thin internal device";
2825                goto bad_thin_open;
2826        }
2827
2828        r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2829        if (r)
2830                goto bad_thin_open;
2831
2832        ti->num_flush_bios = 1;
2833        ti->flush_supported = true;
2834        ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
2835
2836        /* In case the pool supports discards, pass them on. */
2837        ti->discard_zeroes_data_unsupported = true;
2838        if (tc->pool->pf.discard_enabled) {
2839                ti->discards_supported = true;
2840                ti->num_discard_bios = 1;
2841                /* Discard bios must be split on a block boundary */
2842                ti->split_discard_bios = true;
2843        }
2844
2845        dm_put(pool_md);
2846
2847        mutex_unlock(&dm_thin_pool_table.mutex);
2848
2849        return 0;
2850
2851bad_thin_open:
2852        __pool_dec(tc->pool);
2853bad_pool_lookup:
2854        dm_put(pool_md);
2855bad_common:
2856        dm_put_device(ti, tc->pool_dev);
2857bad_pool_dev:
2858        if (tc->origin_dev)
2859                dm_put_device(ti, tc->origin_dev);
2860bad_origin_dev:
2861        kfree(tc);
2862out_unlock:
2863        mutex_unlock(&dm_thin_pool_table.mutex);
2864
2865        return r;
2866}
2867
2868static int thin_map(struct dm_target *ti, struct bio *bio)
2869{
2870        bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
2871
2872        return thin_bio_map(ti, bio);
2873}
2874
2875static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
2876{
2877        unsigned long flags;
2878        struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2879        struct list_head work;
2880        struct dm_thin_new_mapping *m, *tmp;
2881        struct pool *pool = h->tc->pool;
2882
2883        if (h->shared_read_entry) {
2884                INIT_LIST_HEAD(&work);
2885                dm_deferred_entry_dec(h->shared_read_entry, &work);
2886
2887                spin_lock_irqsave(&pool->lock, flags);
2888                list_for_each_entry_safe(m, tmp, &work, list) {
2889                        list_del(&m->list);
2890                        m->quiesced = 1;
2891                        __maybe_add_mapping(m);
2892                }
2893                spin_unlock_irqrestore(&pool->lock, flags);
2894        }
2895
2896        if (h->all_io_entry) {
2897                INIT_LIST_HEAD(&work);
2898                dm_deferred_entry_dec(h->all_io_entry, &work);
2899                if (!list_empty(&work)) {
2900                        spin_lock_irqsave(&pool->lock, flags);
2901                        list_for_each_entry_safe(m, tmp, &work, list)
2902                                list_add(&m->list, &pool->prepared_discards);
2903                        spin_unlock_irqrestore(&pool->lock, flags);
2904                        wake_worker(pool);
2905                }
2906        }
2907
2908        return 0;
2909}
2910
2911static void thin_postsuspend(struct dm_target *ti)
2912{
2913        if (dm_noflush_suspending(ti))
2914                requeue_io((struct thin_c *)ti->private);
2915}
2916
2917/*
2918 * <nr mapped sectors> <highest mapped sector>
2919 */
2920static void thin_status(struct dm_target *ti, status_type_t type,
2921                        unsigned status_flags, char *result, unsigned maxlen)
2922{
2923        int r;
2924        ssize_t sz = 0;
2925        dm_block_t mapped, highest;
2926        char buf[BDEVNAME_SIZE];
2927        struct thin_c *tc = ti->private;
2928
2929        if (get_pool_mode(tc->pool) == PM_FAIL) {
2930                DMEMIT("Fail");
2931                return;
2932        }
2933
2934        if (!tc->td)
2935                DMEMIT("-");
2936        else {
2937                switch (type) {
2938                case STATUSTYPE_INFO:
2939                        r = dm_thin_get_mapped_count(tc->td, &mapped);
2940                        if (r) {
2941                                DMERR("dm_thin_get_mapped_count returned %d", r);
2942                                goto err;
2943                        }
2944
2945                        r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2946                        if (r < 0) {
2947                                DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2948                                goto err;
2949                        }
2950
2951                        DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2952                        if (r)
2953                                DMEMIT("%llu", ((highest + 1) *
2954                                                tc->pool->sectors_per_block) - 1);
2955                        else
2956                                DMEMIT("-");
2957                        break;
2958
2959                case STATUSTYPE_TABLE:
2960                        DMEMIT("%s %lu",
2961                               format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2962                               (unsigned long) tc->dev_id);
2963                        if (tc->origin_dev)
2964                                DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
2965                        break;
2966                }
2967        }
2968
2969        return;
2970
2971err:
2972        DMEMIT("Error");
2973}
2974
2975static int thin_iterate_devices(struct dm_target *ti,
2976                                iterate_devices_callout_fn fn, void *data)
2977{
2978        sector_t blocks;
2979        struct thin_c *tc = ti->private;
2980        struct pool *pool = tc->pool;
2981
2982        /*
2983         * We can't call dm_pool_get_data_dev_size() since that blocks.  So
2984         * we follow a more convoluted path through to the pool's target.
2985         */
2986        if (!pool->ti)
2987                return 0;       /* nothing is bound */
2988
2989        blocks = pool->ti->len;
2990        (void) sector_div(blocks, pool->sectors_per_block);
2991        if (blocks)
2992                return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
2993
2994        return 0;
2995}
2996
2997static struct target_type thin_target = {
2998        .name = "thin",
2999        .version = {1, 9, 0},
3000        .module = THIS_MODULE,
3001        .ctr = thin_ctr,
3002        .dtr = thin_dtr,
3003        .map = thin_map,
3004        .end_io = thin_endio,
3005        .postsuspend = thin_postsuspend,
3006        .status = thin_status,
3007        .iterate_devices = thin_iterate_devices,
3008};
3009
3010/*----------------------------------------------------------------*/
3011
3012static int __init dm_thin_init(void)
3013{
3014        int r;
3015
3016        pool_table_init();
3017
3018        r = dm_register_target(&thin_target);
3019        if (r)
3020                return r;
3021
3022        r = dm_register_target(&pool_target);
3023        if (r)
3024                goto bad_pool_target;
3025
3026        r = -ENOMEM;
3027
3028        _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3029        if (!_new_mapping_cache)
3030                goto bad_new_mapping_cache;
3031
3032        return 0;
3033
3034bad_new_mapping_cache:
3035        dm_unregister_target(&pool_target);
3036bad_pool_target:
3037        dm_unregister_target(&thin_target);
3038
3039        return r;
3040}
3041
3042static void dm_thin_exit(void)
3043{
3044        dm_unregister_target(&thin_target);
3045        dm_unregister_target(&pool_target);
3046
3047        kmem_cache_destroy(_new_mapping_cache);
3048}
3049
3050module_init(dm_thin_init);
3051module_exit(dm_thin_exit);
3052
3053MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3054MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3055MODULE_LICENSE("GPL");
3056
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.