dm-thin.c

/*
 * Copyright (C) 2011-2012 Red Hat UK.
 *
 * This file is released under the GPL.
 */

#include "dm-thin-metadata.h"
#include "dm-bio-prison.h"
#include "dm.h"

#include <linux/device-mapper.h>
#include <linux/dm-io.h>
#include <linux/dm-kcopyd.h>
#include <linux/jiffies.h>
#include <linux/log2.h>
#include <linux/list.h>
#include <linux/rculist.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/sort.h>
#include <linux/rbtree.h>

#define	DM_MSG_PREFIX	"thin"

/*
 * Tunable constants
 */
#define ENDIO_HOOK_POOL_SIZE 1024
#define MAPPING_POOL_SIZE 1024
#define COMMIT_PERIOD HZ
#define NO_SPACE_TIMEOUT_SECS 60

static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;

DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
		"A percentage of time allocated for copy on write");

/*
 * The block size of the device holding pool data must be
 * between 64KB and 1GB.
 */
#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)

/*
 * Device id is restricted to 24 bits.
 */
#define MAX_DEV_ID ((1 << 24) - 1)

/*
 * How do we handle breaking sharing of data blocks?
 * =================================================
 *
 * We use a standard copy-on-write btree to store the mappings for the
 * devices (note I'm talking about copy-on-write of the metadata here, not
 * the data).  When you take an internal snapshot you clone the root node
 * of the origin btree.  After this there is no concept of an origin or a
 * snapshot.  They are just two device trees that happen to point to the
 * same data blocks.
 *
 * When we get a write in we decide if it's to a shared data block using
 * some timestamp magic.  If it is, we have to break sharing.
 *
 * Let's say we write to a shared block in what was the origin.  The
 * steps are:
 *
 * i) plug io further to this physical block. (see bio_prison code).
 *
 * ii) quiesce any read io to that shared data block.  Obviously
 * including all devices that share this block.  (see dm_deferred_set code)
 *
 * iii) copy the data block to a newly allocate block.  This step can be
 * missed out if the io covers the block. (schedule_copy).
 *
 * iv) insert the new mapping into the origin's btree
 * (process_prepared_mapping).  This act of inserting breaks some
 * sharing of btree nodes between the two devices.  Breaking sharing only
 * effects the btree of that specific device.  Btrees for the other
 * devices that share the block never change.  The btree for the origin
 * device as it was after the last commit is untouched, ie. we're using
 * persistent data structures in the functional programming sense.
 *
 * v) unplug io to this physical block, including the io that triggered
 * the breaking of sharing.
 *
 * Steps (ii) and (iii) occur in parallel.
 *
 * The metadata _doesn't_ need to be committed before the io continues.  We
 * get away with this because the io is always written to a _new_ block.
 * If there's a crash, then:
 *
 * - The origin mapping will point to the old origin block (the shared
 * one).  This will contain the data as it was before the io that triggered
 * the breaking of sharing came in.
 *
 * - The snap mapping still points to the old block.  As it would after
 * the commit.
 *
 * The downside of this scheme is the timestamp magic isn't perfect, and
 * will continue to think that data block in the snapshot device is shared
 * even after the write to the origin has broken sharing.  I suspect data
 * blocks will typically be shared by many different devices, so we're
 * breaking sharing n + 1 times, rather than n, where n is the number of
 * devices that reference this data block.  At the moment I think the
 * benefits far, far outweigh the disadvantages.
 */

/*----------------------------------------------------------------*/

/*
 * Key building.
 */
static void build_data_key(struct dm_thin_device *td,
			   dm_block_t b, struct dm_cell_key *key)
{
	key->virtual = 0;
	key->dev = dm_thin_dev_id(td);
	key->block_begin = b;
	key->block_end = b + 1ULL;
}

static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
			      struct dm_cell_key *key)
{
	key->virtual = 1;
	key->dev = dm_thin_dev_id(td);
	key->block_begin = b;
	key->block_end = b + 1ULL;
}

/*----------------------------------------------------------------*/

#define THROTTLE_THRESHOLD (1 * HZ)

struct throttle {
	struct rw_semaphore lock;
	unsigned long threshold;
	bool throttle_applied;
};

static void throttle_init(struct throttle *t)
{
	init_rwsem(&t->lock);
	t->throttle_applied = false;
}

static void throttle_work_start(struct throttle *t)
{
	t->threshold = jiffies + THROTTLE_THRESHOLD;
}

static void throttle_work_update(struct throttle *t)
{
	if (!t->throttle_applied && jiffies > t->threshold) {
		down_write(&t->lock);
		t->throttle_applied = true;
	}
}

static void throttle_work_complete(struct throttle *t)
{
	if (t->throttle_applied) {
		t->throttle_applied = false;
		up_write(&t->lock);
	}
}

static void throttle_lock(struct throttle *t)
{
	down_read(&t->lock);
}

static void throttle_unlock(struct throttle *t)
{
	up_read(&t->lock);
}

/*----------------------------------------------------------------*/

/*
 * A pool device ties together a metadata device and a data device.  It
 * also provides the interface for creating and destroying internal
 * devices.
 */
struct dm_thin_new_mapping;

/*
 * The pool runs in 4 modes.  Ordered in degraded order for comparisons.
 */
enum pool_mode {
	PM_WRITE,		/* metadata may be changed */
	PM_OUT_OF_DATA_SPACE,	/* metadata may be changed, though data may not be allocated */
	PM_READ_ONLY,		/* metadata may not be changed */
	PM_FAIL,		/* all I/O fails */
};

struct pool_features {
	enum pool_mode mode;

	bool zero_new_blocks:1;
	bool discard_enabled:1;
	bool discard_passdown:1;
	bool error_if_no_space:1;
};

struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);

#define CELL_SORT_ARRAY_SIZE 8192

struct pool {
	struct list_head list;
	struct dm_target *ti;	/* Only set if a pool target is bound */

	struct mapped_device *pool_md;
	struct block_device *md_dev;
	struct dm_pool_metadata *pmd;

	dm_block_t low_water_blocks;
	uint32_t sectors_per_block;
	int sectors_per_block_shift;

	struct pool_features pf;
	bool low_water_triggered:1;	/* A dm event has been sent */
	bool suspended:1;

	struct dm_bio_prison *prison;
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
	struct throttle throttle;
	struct work_struct worker;
	struct delayed_work waker;
	struct delayed_work no_space_timeout;

	unsigned long last_commit_jiffies;
	unsigned ref_count;

	spinlock_t lock;
	struct bio_list deferred_flush_bios;
	struct list_head prepared_mappings;
	struct list_head prepared_discards;
	struct list_head active_thins;

	struct dm_deferred_set *shared_read_ds;
	struct dm_deferred_set *all_io_ds;

	struct dm_thin_new_mapping *next_mapping;
	mempool_t *mapping_pool;

	process_bio_fn process_bio;
	process_bio_fn process_discard;

	process_cell_fn process_cell;
	process_cell_fn process_discard_cell;

	process_mapping_fn process_prepared_mapping;
	process_mapping_fn process_prepared_discard;

	struct dm_bio_prison_cell *cell_sort_array[CELL_SORT_ARRAY_SIZE];
};

static enum pool_mode get_pool_mode(struct pool *pool);
static void metadata_operation_failed(struct pool *pool, const char *op, int r);

/*
 * Target context for a pool.
 */
struct pool_c {
	struct dm_target *ti;
	struct pool *pool;
	struct dm_dev *data_dev;
	struct dm_dev *metadata_dev;
	struct dm_target_callbacks callbacks;

	dm_block_t low_water_blocks;
	struct pool_features requested_pf; /* Features requested during table load */
	struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
};

/*
 * Target context for a thin.
 */
struct thin_c {
	struct list_head list;
	struct dm_dev *pool_dev;
	struct dm_dev *origin_dev;
	sector_t origin_size;
	dm_thin_id dev_id;

	struct pool *pool;
	struct dm_thin_device *td;
	struct mapped_device *thin_md;

	bool requeue_mode:1;
	spinlock_t lock;
	struct list_head deferred_cells;
	struct bio_list deferred_bio_list;
	struct bio_list retry_on_resume_list;
	struct rb_root sort_bio_list; /* sorted list of deferred bios */

	/*
	 * Ensures the thin is not destroyed until the worker has finished
	 * iterating the active_thins list.
	 */
	atomic_t refcount;
	struct completion can_destroy;
};

/*----------------------------------------------------------------*/

/*
 * wake_worker() is used when new work is queued and when pool_resume is
 * ready to continue deferred IO processing.
 */
static void wake_worker(struct pool *pool)
{
	queue_work(pool->wq, &pool->worker);
}

/*----------------------------------------------------------------*/

static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
		      struct dm_bio_prison_cell **cell_result)
{
	int r;
	struct dm_bio_prison_cell *cell_prealloc;

	/*
	 * Allocate a cell from the prison's mempool.
	 * This might block but it can't fail.
	 */
	cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);

	r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
	if (r)
		/*
		 * We reused an old cell; we can get rid of
		 * the new one.
		 */
		dm_bio_prison_free_cell(pool->prison, cell_prealloc);

	return r;
}

static void cell_release(struct pool *pool,
			 struct dm_bio_prison_cell *cell,
			 struct bio_list *bios)
{
	dm_cell_release(pool->prison, cell, bios);
	dm_bio_prison_free_cell(pool->prison, cell);
}

static void cell_visit_release(struct pool *pool,
			       void (*fn)(void *, struct dm_bio_prison_cell *),
			       void *context,
			       struct dm_bio_prison_cell *cell)
{
	dm_cell_visit_release(pool->prison, fn, context, cell);
	dm_bio_prison_free_cell(pool->prison, cell);
}

static void cell_release_no_holder(struct pool *pool,
				   struct dm_bio_prison_cell *cell,
				   struct bio_list *bios)
{
	dm_cell_release_no_holder(pool->prison, cell, bios);
	dm_bio_prison_free_cell(pool->prison, cell);
}

static void cell_error_with_code(struct pool *pool,
				 struct dm_bio_prison_cell *cell, int error_code)
{
	dm_cell_error(pool->prison, cell, error_code);
	dm_bio_prison_free_cell(pool->prison, cell);
}

static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, -EIO);
}

static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, 0);
}

static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE);
}

/*----------------------------------------------------------------*/

/*
 * A global list of pools that uses a struct mapped_device as a key.
 */
static struct dm_thin_pool_table {
	struct mutex mutex;
	struct list_head pools;
} dm_thin_pool_table;

static void pool_table_init(void)
{
	mutex_init(&dm_thin_pool_table.mutex);
	INIT_LIST_HEAD(&dm_thin_pool_table.pools);
}

static void __pool_table_insert(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	list_add(&pool->list, &dm_thin_pool_table.pools);
}

static void __pool_table_remove(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	list_del(&pool->list);
}

static struct pool *__pool_table_lookup(struct mapped_device *md)
{
	struct pool *pool = NULL, *tmp;

	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));

	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
		if (tmp->pool_md == md) {
			pool = tmp;
			break;
		}
	}

	return pool;
}

static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
{
	struct pool *pool = NULL, *tmp;

	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));

	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
		if (tmp->md_dev == md_dev) {
			pool = tmp;
			break;
		}
	}

	return pool;
}

/*----------------------------------------------------------------*/

struct dm_thin_endio_hook {
	struct thin_c *tc;
	struct dm_deferred_entry *shared_read_entry;
	struct dm_deferred_entry *all_io_entry;
	struct dm_thin_new_mapping *overwrite_mapping;
	struct rb_node rb_node;
};

static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
{
	bio_list_merge(bios, master);
	bio_list_init(master);
}

static void error_bio_list(struct bio_list *bios, int error)
{
	struct bio *bio;

	while ((bio = bio_list_pop(bios)))
		bio_endio(bio, error);
}

static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
{
	struct bio_list bios;
	unsigned long flags;

	bio_list_init(&bios);

	spin_lock_irqsave(&tc->lock, flags);
	__merge_bio_list(&bios, master);
	spin_unlock_irqrestore(&tc->lock, flags);

	error_bio_list(&bios, error);
}

static void requeue_deferred_cells(struct thin_c *tc)
{
	struct pool *pool = tc->pool;
	unsigned long flags;
	struct list_head cells;
	struct dm_bio_prison_cell *cell, *tmp;

	INIT_LIST_HEAD(&cells);

	spin_lock_irqsave(&tc->lock, flags);
	list_splice_init(&tc->deferred_cells, &cells);
	spin_unlock_irqrestore(&tc->lock, flags);

	list_for_each_entry_safe(cell, tmp, &cells, user_list)
		cell_requeue(pool, cell);
}

static void requeue_io(struct thin_c *tc)
{
	struct bio_list bios;
	unsigned long flags;

	bio_list_init(&bios);

	spin_lock_irqsave(&tc->lock, flags);
	__merge_bio_list(&bios, &tc->deferred_bio_list);
	__merge_bio_list(&bios, &tc->retry_on_resume_list);
	spin_unlock_irqrestore(&tc->lock, flags);

	error_bio_list(&bios, DM_ENDIO_REQUEUE);
	requeue_deferred_cells(tc);
}

static void error_retry_list(struct pool *pool)
{
	struct thin_c *tc;

	rcu_read_lock();
	list_for_each_entry_rcu(tc, &pool->active_thins, list)
		error_thin_bio_list(tc, &tc->retry_on_resume_list, -EIO);
	rcu_read_unlock();
}

/*
 * This section of code contains the logic for processing a thin device's IO.
 * Much of the code depends on pool object resources (lists, workqueues, etc)
 * but most is exclusively called from the thin target rather than the thin-pool
 * target.
 */

static bool block_size_is_power_of_two(struct pool *pool)
{
	return pool->sectors_per_block_shift >= 0;
}

static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
	struct pool *pool = tc->pool;
	sector_t block_nr = bio->bi_iter.bi_sector;

	if (block_size_is_power_of_two(pool))
		block_nr >>= pool->sectors_per_block_shift;
	else
		(void) sector_div(block_nr, pool->sectors_per_block);

	return block_nr;
}

static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
	sector_t bi_sector = bio->bi_iter.bi_sector;

	bio->bi_bdev = tc->pool_dev->bdev;
	if (block_size_is_power_of_two(pool))
		bio->bi_iter.bi_sector =
			(block << pool->sectors_per_block_shift) |
			(bi_sector & (pool->sectors_per_block - 1));
	else
		bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
				 sector_div(bi_sector, pool->sectors_per_block);
}

static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
{
	return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
		dm_thin_changed_this_transaction(tc->td);
}

static void inc_all_io_entry(struct pool *pool, struct bio *bio)
{
	struct dm_thin_endio_hook *h;

	if (bio->bi_rw & REQ_DISCARD)
		return;

	h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
	h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}

static void issue(struct thin_c *tc, struct bio *bio)
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	if (!bio_triggers_commit(tc, bio)) {
		generic_make_request(bio);
		return;
	}

	/*
	 * Complete bio with an error if earlier I/O caused changes to
	 * the metadata that can't be committed e.g, due to I/O errors
	 * on the metadata device.
	 */
	if (dm_thin_aborted_changes(tc->td)) {
		bio_io_error(bio);
		return;
	}

	/*
	 * Batch together any bios that trigger commits and then issue a
	 * single commit for them in process_deferred_bios().
	 */
	spin_lock_irqsave(&pool->lock, flags);
	bio_list_add(&pool->deferred_flush_bios, bio);
	spin_unlock_irqrestore(&pool->lock, flags);
}

static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
{
	remap_to_origin(tc, bio);
	issue(tc, bio);
}

static void remap_and_issue(struct thin_c *tc, struct bio *bio,
			    dm_block_t block)
{
	remap(tc, bio, block);
	issue(tc, bio);
}

/*----------------------------------------------------------------*/

/*
 * Bio endio functions.
 */
struct dm_thin_new_mapping {
	struct list_head list;

	bool pass_discard:1;
	bool definitely_not_shared:1;

	/*
	 * Track quiescing, copying and zeroing preparation actions.  When this
	 * counter hits zero the block is prepared and can be inserted into the
	 * btree.
	 */
	atomic_t prepare_actions;

	int err;
	struct thin_c *tc;
	dm_block_t virt_block;
	dm_block_t data_block;
	struct dm_bio_prison_cell *cell, *cell2;

	/*
	 * If the bio covers the whole area of a block then we can avoid
	 * zeroing or copying.  Instead this bio is hooked.  The bio will
	 * still be in the cell, so care has to be taken to avoid issuing
	 * the bio twice.
	 */
	struct bio *bio;
	bio_end_io_t *saved_bi_end_io;
};

static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
{
	struct pool *pool = m->tc->pool;

	if (atomic_dec_and_test(&m->prepare_actions)) {
		list_add_tail(&m->list, &pool->prepared_mappings);
		wake_worker(pool);
	}
}

static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
{
	unsigned long flags;
	struct pool *pool = m->tc->pool;

	spin_lock_irqsave(&pool->lock, flags);
	__complete_mapping_preparation(m);
	spin_unlock_irqrestore(&pool->lock, flags);
}

static void copy_complete(int read_err, unsigned long write_err, void *context)
{
	struct dm_thin_new_mapping *m = context;

	m->err = read_err || write_err ? -EIO : 0;
	complete_mapping_preparation(m);
}

static void overwrite_endio(struct bio *bio, int err)
{
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
	struct dm_thin_new_mapping *m = h->overwrite_mapping;

	m->err = err;
	complete_mapping_preparation(m);
}

/*----------------------------------------------------------------*/

/*
 * Workqueue.
 */

/*
 * Prepared mapping jobs.
 */

/*
 * This sends the bios in the cell, except the original holder, back
 * to the deferred_bios list.
 */
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&tc->lock, flags);
	cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
	spin_unlock_irqrestore(&tc->lock, flags);

	wake_worker(pool);
}

static void thin_defer_bio(struct thin_c *tc, struct bio *bio);

struct remap_info {
	struct thin_c *tc;
	struct bio_list defer_bios;
	struct bio_list issue_bios;
};

static void __inc_remap_and_issue_cell(void *context,
				       struct dm_bio_prison_cell *cell)
{
	struct remap_info *info = context;
	struct bio *bio;

	while ((bio = bio_list_pop(&cell->bios))) {
		if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))
			bio_list_add(&info->defer_bios, bio);
		else {
			inc_all_io_entry(info->tc->pool, bio);

			/*
			 * We can't issue the bios with the bio prison lock
			 * held, so we add them to a list to issue on
			 * return from this function.
			 */
			bio_list_add(&info->issue_bios, bio);
		}
	}
}

static void inc_remap_and_issue_cell(struct thin_c *tc,
				     struct dm_bio_prison_cell *cell,
				     dm_block_t block)
{
	struct bio *bio;
	struct remap_info info;

	info.tc = tc;
	bio_list_init(&info.defer_bios);
	bio_list_init(&info.issue_bios);

	/*
	 * We have to be careful to inc any bios we're about to issue
	 * before the cell is released, and avoid a race with new bios
	 * being added to the cell.
	 */
	cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
			   &info, cell);

	while ((bio = bio_list_pop(&info.defer_bios)))
		thin_defer_bio(tc, bio);

	while ((bio = bio_list_pop(&info.issue_bios)))
		remap_and_issue(info.tc, bio, block);
}

static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
	if (m->bio)
		m->bio->bi_end_io = m->saved_bi_end_io;

	cell_error(m->tc->pool, m->cell);
	list_del(&m->list);
	mempool_free(m, m->tc->pool->mapping_pool);
}

static void process_prepared_mapping(struct dm_thin_new_mapping *m)
{
	struct thin_c *tc = m->tc;
	struct pool *pool = tc->pool;
	struct bio *bio;
	int r;

	bio = m->bio;
	if (bio)
		bio->bi_end_io = m->saved_bi_end_io;

	if (m->err) {
		cell_error(pool, m->cell);
		goto out;
	}

	/*
	 * Commit the prepared block into the mapping btree.
	 * Any I/O for this block arriving after this point will get
	 * remapped to it directly.
	 */
	r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
	if (r) {
		metadata_operation_failed(pool, "dm_thin_insert_block", r);
		cell_error(pool, m->cell);
		goto out;
	}

	/*
	 * Release any bios held while the block was being provisioned.
	 * If we are processing a write bio that completely covers the block,
	 * we already processed it so can ignore it now when processing
	 * the bios in the cell.
	 */
	if (bio) {
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
		bio_endio(bio, 0);
	} else {
		inc_all_io_entry(tc->pool, m->cell->holder);
		remap_and_issue(tc, m->cell->holder, m->data_block);
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
	}

out:
	list_del(&m->list);
	mempool_free(m, pool->mapping_pool);
}

static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
{
	struct thin_c *tc = m->tc;

	bio_io_error(m->bio);
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
	mempool_free(m, tc->pool->mapping_pool);
}

static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
{
	struct thin_c *tc = m->tc;

	inc_all_io_entry(tc->pool, m->bio);
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);

	if (m->pass_discard)
		if (m->definitely_not_shared)
			remap_and_issue(tc, m->bio, m->data_block);
		else {
			bool used = false;
			if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
				bio_endio(m->bio, 0);
			else
				remap_and_issue(tc, m->bio, m->data_block);
		}
	else
		bio_endio(m->bio, 0);

	mempool_free(m, tc->pool->mapping_pool);
}

static void process_prepared_discard(struct dm_thin_new_mapping *m)
{
	int r;
	struct thin_c *tc = m->tc;

	r = dm_thin_remove_block(tc->td, m->virt_block);
	if (r)
		DMERR_LIMIT("dm_thin_remove_block() failed");

	process_prepared_discard_passdown(m);
}

static void process_prepared(struct pool *pool, struct list_head *head,
			     process_mapping_fn *fn)
{
	unsigned long flags;
	struct list_head maps;
	struct dm_thin_new_mapping *m, *tmp;

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
	list_splice_init(head, &maps);
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
		(*fn)(m);
}

/*
 * Deferred bio jobs.
 */
static int io_overlaps_block(struct pool *pool, struct bio *bio)
{
	return bio->bi_iter.bi_size ==
		(pool->sectors_per_block << SECTOR_SHIFT);
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
}

static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
			       bio_end_io_t *fn)
{
	*save = bio->bi_end_io;
	bio->bi_end_io = fn;
}

static int ensure_next_mapping(struct pool *pool)
{
	if (pool->next_mapping)
		return 0;

	pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);

	return pool->next_mapping ? 0 : -ENOMEM;
}

static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
{
	struct dm_thin_new_mapping *m = pool->next_mapping;

	BUG_ON(!pool->next_mapping);

	memset(m, 0, sizeof(struct dm_thin_new_mapping));
	INIT_LIST_HEAD(&m->list);
	m->bio = NULL;

	pool->next_mapping = NULL;

	return m;
}

static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
		    sector_t begin, sector_t end)
{
	int r;
	struct dm_io_region to;

	to.bdev = tc->pool_dev->bdev;
	to.sector = begin;
	to.count = end - begin;

	r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
	if (r < 0) {
		DMERR_LIMIT("dm_kcopyd_zero() failed");
		copy_complete(1, 1, m);
	}
}

static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
				      dm_block_t data_block,
				      struct dm_thin_new_mapping *m)
{
	struct pool *pool = tc->pool;
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));

	h->overwrite_mapping = m;
	m->bio = bio;
	save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
	inc_all_io_entry(pool, bio);
	remap_and_issue(tc, bio, data_block);
}

/*
 * A partial copy also needs to zero the uncopied region.
 */
static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
			  struct dm_bio_prison_cell *cell, struct bio *bio,
			  sector_t len)
{
	int r;
	struct pool *pool = tc->pool;
	struct dm_thin_new_mapping *m = get_next_mapping(pool);

	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_dest;
	m->cell = cell;

	/*
	 * quiesce action + copy action + an extra reference held for the
	 * duration of this function (we may need to inc later for a
	 * partial zero).
	 */
	atomic_set(&m->prepare_actions, 3);

	if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
		complete_mapping_preparation(m); /* already quiesced */

	/*
	 * IO to pool_dev remaps to the pool target's data_dev.
	 *
	 * If the whole block of data is being overwritten, we can issue the
	 * bio immediately. Otherwise we use kcopyd to clone the data first.
	 */
	if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_dest, m);
	else {
		struct dm_io_region from, to;

		from.bdev = origin->bdev;
		from.sector = data_origin * pool->sectors_per_block;
		from.count = len;

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_dest * pool->sectors_per_block;
		to.count = len;

		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
				   0, copy_complete, m);
		if (r < 0) {
			DMERR_LIMIT("dm_kcopyd_copy() failed");
			copy_complete(1, 1, m);

			/*
			 * We allow the zero to be issued, to simplify the
			 * error path.  Otherwise we'd need to start
			 * worrying about decrementing the prepare_actions
			 * counter.
			 */
		}

		/*
		 * Do we need to zero a tail region?
		 */
		if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
			atomic_inc(&m->prepare_actions);
			ll_zero(tc, m,
				data_dest * pool->sectors_per_block + len,
				(data_dest + 1) * pool->sectors_per_block);
		}
	}

	complete_mapping_preparation(m); /* drop our ref */
}

static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
				   dm_block_t data_origin, dm_block_t data_dest,
				   struct dm_bio_prison_cell *cell, struct bio *bio)
{
	schedule_copy(tc, virt_block, tc->pool_dev,
		      data_origin, data_dest, cell, bio,
		      tc->pool->sectors_per_block);
}

static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
	struct dm_thin_new_mapping *m = get_next_mapping(pool);

	atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_block;
	m->cell = cell;

	/*
	 * If the whole block of data is being overwritten or we are not
	 * zeroing pre-existing data, we can issue the bio immediately.
	 * Otherwise we use kcopyd to zero the data first.
	 */
	if (!pool->pf.zero_new_blocks)
		process_prepared_mapping(m);

	else if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_block, m);

	else
		ll_zero(tc, m,
			data_block * pool->sectors_per_block,
			(data_block + 1) * pool->sectors_per_block);
}

static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
				   dm_block_t data_dest,
				   struct dm_bio_prison_cell *cell, struct bio *bio)
{
	struct pool *pool = tc->pool;
	sector_t virt_block_begin = virt_block * pool->sectors_per_block;
	sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;

	if (virt_block_end <= tc->origin_size)
		schedule_copy(tc, virt_block, tc->origin_dev,
			      virt_block, data_dest, cell, bio,
			      pool->sectors_per_block);

	else if (virt_block_begin < tc->origin_size)
		schedule_copy(tc, virt_block, tc->origin_dev,
			      virt_block, data_dest, cell, bio,
			      tc->origin_size - virt_block_begin);

	else
		schedule_zero(tc, virt_block, data_dest, cell, bio);
}

static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);

static void check_for_space(struct pool *pool)
{
	int r;
	dm_block_t nr_free;

	if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
		return;

	r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
	if (r)
		return;

	if (nr_free)
		set_pool_mode(pool, PM_WRITE);
}

/*
 * A non-zero return indicates read_only or fail_io mode.
 * Many callers don't care about the return value.
 */
static int commit(struct pool *pool)
{
	int r;

	if (get_pool_mode(pool) >= PM_READ_ONLY)
		return -EINVAL;

	r = dm_pool_commit_metadata(pool->pmd);
	if (r)
		metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
	else
		check_for_space(pool);

	return r;
}

static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
{
	unsigned long flags;

	if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
		DMWARN("%s: reached low water mark for data device: sending event.",
		       dm_device_name(pool->pool_md));
		spin_lock_irqsave(&pool->lock, flags);
		pool->low_water_triggered = true;
		spin_unlock_irqrestore(&pool->lock, flags);
		dm_table_event(pool->ti->table);
	}
}

static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
	int r;
	dm_block_t free_blocks;
	struct pool *pool = tc->pool;

	if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
		return -EINVAL;

	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
	if (r) {
		metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
		return r;
	}

	check_low_water_mark(pool, free_blocks);

	if (!free_blocks) {
		/*
		 * Try to commit to see if that will free up some
		 * more space.
		 */
		r = commit(pool);
		if (r)
			return r;

		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
		if (r) {
			metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
			return r;
		}

		if (!free_blocks) {
			set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
			return -ENOSPC;
		}
	}

	r = dm_pool_alloc_data_block(pool->pmd, result);
	if (r) {
		metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
		return r;
	}

	return 0;
}

/*
 * If we have run out of space, queue bios until the device is
 * resumed, presumably after having been reloaded with more space.
 */
static void retry_on_resume(struct bio *bio)
{
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
	struct thin_c *tc = h->tc;
	unsigned long flags;

	spin_lock_irqsave(&tc->lock, flags);
	bio_list_add(&tc->retry_on_resume_list, bio);
	spin_unlock_irqrestore(&tc->lock, flags);
}

static int should_error_unserviceable_bio(struct pool *pool)
{
	enum pool_mode m = get_pool_mode(pool);

	switch (m) {
	case PM_WRITE:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
		return -EIO;

	case PM_OUT_OF_DATA_SPACE:
		return pool->pf.error_if_no_space ? -ENOSPC : 0;

	case PM_READ_ONLY:
	case PM_FAIL:
		return -EIO;
	default:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
		return -EIO;
	}
}

static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
{
	int error = should_error_unserviceable_bio(pool);

	if (error)
		bio_endio(bio, error);
	else
		retry_on_resume(bio);
}

static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	struct bio *bio;
	struct bio_list bios;
	int error;

	error = should_error_unserviceable_bio(pool);
	if (error) {
		cell_error_with_code(pool, cell, error);
		return;
	}

	bio_list_init(&bios);
	cell_release(pool, cell, &bios);

	while ((bio = bio_list_pop(&bios)))
		retry_on_resume(bio);
}

static void process_discard_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
{
	int r;
	struct bi