dm-snap.c

Linux Kernel 2.6.9 for OMAP1710

			s->valid = 0;
			flush = __flush_bios(pe);
			up_write(&s->lock);

			error_bios(bio_list_get(&pe->snapshot_bios));
			goto out;
		}
		*e = pe->e;

		/*
		 * Add a proper exception, and remove the
		 * in-flight exception from the list.
		 */
		down_write(&s->lock);
		insert_exception(&s->complete, e);
		remove_exception(&pe->e);
		flush = __flush_bios(pe);

		/* Submit any pending write bios */
		up_write(&s->lock);

		flush_bios(bio_list_get(&pe->snapshot_bios));
	} else {
		/* Read/write error - snapshot is unusable */
		down_write(&s->lock);
		if (s->valid)
			DMERR("Error reading/writing snapshot");
		s->store.drop_snapshot(&s->store);
		s->valid = 0;
		remove_exception(&pe->e);
		flush = __flush_bios(pe);
		up_write(&s->lock);

		error_bios(bio_list_get(&pe->snapshot_bios));

		dm_table_event(s->table);
	}

 out:
	free_pending_exception(pe);

	if (flush)
		flush_bios(flush);
}

static void commit_callback(void *context, int success)
{
	struct pending_exception *pe = (struct pending_exception *) context;
	pending_complete(pe, success);
}

/*
 * Called when the copy I/O has finished.  kcopyd actually runs
 * this code so don't block.
 */
static void copy_callback(int read_err, unsigned int write_err, void *context)
{
	struct pending_exception *pe = (struct pending_exception *) context;
	struct dm_snapshot *s = pe->snap;

	if (read_err || write_err)
		pending_complete(pe, 0);

	else
		/* Update the metadata if we are persistent */
		s->store.commit_exception(&s->store, &pe->e, commit_callback,
					  pe);
}

/*
 * Dispatches the copy operation to kcopyd.
 */
static inline void start_copy(struct pending_exception *pe)
{
	struct dm_snapshot *s = pe->snap;
	struct io_region src, dest;
	struct block_device *bdev = s->origin->bdev;
	sector_t dev_size;

	dev_size = get_dev_size(bdev);

	src.bdev = bdev;
	src.sector = chunk_to_sector(s, pe->e.old_chunk);
	src.count = min(s->chunk_size, dev_size - src.sector);

	dest.bdev = s->cow->bdev;
	dest.sector = chunk_to_sector(s, pe->e.new_chunk);
	dest.count = src.count;

	/* Hand over to kcopyd */
	kcopyd_copy(s->kcopyd_client,
		    &src, 1, &dest, 0, copy_callback, pe);
}

/*
 * Looks to see if this snapshot already has a pending exception
 * for this chunk, otherwise it allocates a new one and inserts
 * it into the pending table.
 *
 * NOTE: a write lock must be held on snap->lock before calling
 * this.
 */
static struct pending_exception *
__find_pending_exception(struct dm_snapshot *s, struct bio *bio)
{
	struct exception *e;
	struct pending_exception *pe;
	chunk_t chunk = sector_to_chunk(s, bio->bi_sector);

	/*
	 * Is there a pending exception for this already ?
	 */
	e = lookup_exception(&s->pending, chunk);
	if (e) {
		/* cast the exception to a pending exception */
		pe = container_of(e, struct pending_exception, e);

	} else {
		/*
		 * Create a new pending exception, we don't want
		 * to hold the lock while we do this.
		 */
		up_write(&s->lock);
		pe = alloc_pending_exception();
		down_write(&s->lock);

		e = lookup_exception(&s->pending, chunk);
		if (e) {
			free_pending_exception(pe);
			pe = container_of(e, struct pending_exception, e);
		} else {
			pe->e.old_chunk = chunk;
			bio_list_init(&pe->origin_bios);
			bio_list_init(&pe->snapshot_bios);
			INIT_LIST_HEAD(&pe->siblings);
			pe->snap = s;
			pe->started = 0;

			if (s->store.prepare_exception(&s->store, &pe->e)) {
				free_pending_exception(pe);
				s->valid = 0;
				return NULL;
			}

			insert_exception(&s->pending, &pe->e);
		}
	}

	return pe;
}

static inline void remap_exception(struct dm_snapshot *s, struct exception *e,
				   struct bio *bio)
{
	bio->bi_bdev = s->cow->bdev;
	bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
		(bio->bi_sector & s->chunk_mask);
}

static int snapshot_map(struct dm_target *ti, struct bio *bio,
			union map_info *map_context)
{
	struct exception *e;
	struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
	int r = 1;
	chunk_t chunk;
	struct pending_exception *pe;

	chunk = sector_to_chunk(s, bio->bi_sector);

	/* Full snapshots are not usable */
	if (!s->valid)
		return -1;

	/*
	 * Write to snapshot - higher level takes care of RW/RO
	 * flags so we should only get this if we are
	 * writeable.
	 */
	if (bio_rw(bio) == WRITE) {

		/* FIXME: should only take write lock if we need
		 * to copy an exception */
		down_write(&s->lock);

		/* If the block is already remapped - use that, else remap it */
		e = lookup_exception(&s->complete, chunk);
		if (e) {
			remap_exception(s, e, bio);
			up_write(&s->lock);

		} else {
			pe = __find_pending_exception(s, bio);

			if (!pe) {
				if (s->store.drop_snapshot)
					s->store.drop_snapshot(&s->store);
				s->valid = 0;
				r = -EIO;
				up_write(&s->lock);
			} else {
				remap_exception(s, &pe->e, bio);
				bio_list_add(&pe->snapshot_bios, bio);

				if (!pe->started) {
					/* this is protected by snap->lock */
					pe->started = 1;
					up_write(&s->lock);
					start_copy(pe);
				} else
					up_write(&s->lock);
				r = 0;
			}
		}

	} else {
		/*
		 * FIXME: this read path scares me because we
		 * always use the origin when we have a pending
		 * exception.  However I can't think of a
		 * situation where this is wrong - ejt.
		 */

		/* Do reads */
		down_read(&s->lock);

		/* See if it it has been remapped */
		e = lookup_exception(&s->complete, chunk);
		if (e)
			remap_exception(s, e, bio);
		else
			bio->bi_bdev = s->origin->bdev;

		up_read(&s->lock);
	}

	return r;
}

static void snapshot_resume(struct dm_target *ti)
{
	struct dm_snapshot *s = (struct dm_snapshot *) ti->private;

	if (s->have_metadata)
		return;

	if (s->store.read_metadata(&s->store)) {
		down_write(&s->lock);
		s->valid = 0;
		up_write(&s->lock);
	}

	s->have_metadata = 1;
}

static int snapshot_status(struct dm_target *ti, status_type_t type,
			   char *result, unsigned int maxlen)
{
	struct dm_snapshot *snap = (struct dm_snapshot *) ti->private;
	char cow[32];
	char org[32];

	switch (type) {
	case STATUSTYPE_INFO:
		if (!snap->valid)
			snprintf(result, maxlen, "Invalid");
		else {
			if (snap->store.fraction_full) {
				sector_t numerator, denominator;
				snap->store.fraction_full(&snap->store,
							  &numerator,
							  &denominator);
				snprintf(result, maxlen,
					 SECTOR_FORMAT "/" SECTOR_FORMAT,
					 numerator, denominator);
			}
			else
				snprintf(result, maxlen, "Unknown");
		}
		break;

	case STATUSTYPE_TABLE:
		/*
		 * kdevname returns a static pointer so we need
		 * to make private copies if the output is to
		 * make sense.
		 */
		format_dev_t(cow, snap->cow->bdev->bd_dev);
		format_dev_t(org, snap->origin->bdev->bd_dev);
		snprintf(result, maxlen, "%s %s %c " SECTOR_FORMAT, org, cow,
			 snap->type, snap->chunk_size);
		break;
	}

	return 0;
}

/*-----------------------------------------------------------------
 * Origin methods
 *---------------------------------------------------------------*/
static void list_merge(struct list_head *l1, struct list_head *l2)
{
	struct list_head *l1_n, *l2_p;

	l1_n = l1->next;
	l2_p = l2->prev;

	l1->next = l2;
	l2->prev = l1;

	l2_p->next = l1_n;
	l1_n->prev = l2_p;
}

static int __origin_write(struct list_head *snapshots, struct bio *bio)
{
	int r = 1, first = 1;
	struct dm_snapshot *snap;
	struct exception *e;
	struct pending_exception *pe, *last = NULL;
	chunk_t chunk;

	/* Do all the snapshots on this origin */
	list_for_each_entry (snap, snapshots, list) {

		/* Only deal with valid snapshots */
		if (!snap->valid)
			continue;

		down_write(&snap->lock);

		/*
		 * Remember, different snapshots can have
		 * different chunk sizes.
		 */
		chunk = sector_to_chunk(snap, bio->bi_sector);

		/*
		 * Check exception table to see if block
		 * is already remapped in this snapshot
		 * and trigger an exception if not.
		 */
		e = lookup_exception(&snap->complete, chunk);
		if (!e) {
			pe = __find_pending_exception(snap, bio);
			if (!pe) {
				snap->store.drop_snapshot(&snap->store);
				snap->valid = 0;

			} else {
				if (last)
					list_merge(&pe->siblings,
						   &last->siblings);

				last = pe;
				r = 0;
			}
		}

		up_write(&snap->lock);
	}

	/*
	 * Now that we have a complete pe list we can start the copying.
	 */
	if (last) {
		pe = last;
		do {
			down_write(&pe->snap->lock);
			if (first)
				bio_list_add(&pe->origin_bios, bio);
			if (!pe->started) {
				pe->started = 1;
				up_write(&pe->snap->lock);
				start_copy(pe);
			} else
				up_write(&pe->snap->lock);
			first = 0;
			pe = list_entry(pe->siblings.next,
					struct pending_exception, siblings);

		} while (pe != last);
	}

	return r;
}

/*
 * Called on a write from the origin driver.
 */
static int do_origin(struct dm_dev *origin, struct bio *bio)
{
	struct origin *o;
	int r = 1;

	down_read(&_origins_lock);
	o = __lookup_origin(origin->bdev);
	if (o)
		r = __origin_write(&o->snapshots, bio);
	up_read(&_origins_lock);

	return r;
}

/*
 * Origin: maps a linear range of a device, with hooks for snapshotting.
 */

/*
 * Construct an origin mapping: <dev_path>
 * The context for an origin is merely a 'struct dm_dev *'
 * pointing to the real device.
 */
static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	int r;
	struct dm_dev *dev;

	if (argc != 1) {
		ti->error = "dm-origin: incorrect number of arguments";
		return -EINVAL;
	}

	r = dm_get_device(ti, argv[0], 0, ti->len,
			  dm_table_get_mode(ti->table), &dev);
	if (r) {
		ti->error = "Cannot get target device";
		return r;
	}

	ti->private = dev;
	return 0;
}

static void origin_dtr(struct dm_target *ti)
{
	struct dm_dev *dev = (struct dm_dev *) ti->private;
	dm_put_device(ti, dev);
}

static int origin_map(struct dm_target *ti, struct bio *bio,
		      union map_info *map_context)
{
	struct dm_dev *dev = (struct dm_dev *) ti->private;
	bio->bi_bdev = dev->bdev;

	/* Only tell snapshots if this is a write */
	return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1;
}

#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))

/*
 * Set the target "split_io" field to the minimum of all the snapshots'
 * chunk sizes.
 */
static void origin_resume(struct dm_target *ti)
{
	struct dm_dev *dev = (struct dm_dev *) ti->private;
	struct dm_snapshot *snap;
	struct origin *o;
	chunk_t chunk_size = 0;

	down_read(&_origins_lock);
	o = __lookup_origin(dev->bdev);
	if (o)
		list_for_each_entry (snap, &o->snapshots, list)
			chunk_size = min_not_zero(chunk_size, snap->chunk_size);
	up_read(&_origins_lock);

	ti->split_io = chunk_size;
}

static int origin_status(struct dm_target *ti, status_type_t type, char *result,
			 unsigned int maxlen)
{
	struct dm_dev *dev = (struct dm_dev *) ti->private;
	char buffer[32];

	switch (type) {
	case STATUSTYPE_INFO:
		result[0] = '\0';
		break;

	case STATUSTYPE_TABLE:
		format_dev_t(buffer, dev->bdev->bd_dev);
		snprintf(result, maxlen, "%s", buffer);
		break;
	}

	return 0;
}

static struct target_type origin_target = {
	.name    = "snapshot-origin",
	.version = {1, 0, 1},
	.module  = THIS_MODULE,
	.ctr     = origin_ctr,
	.dtr     = origin_dtr,
	.map     = origin_map,
	.resume  = origin_resume,
	.status  = origin_status,
};

static struct target_type snapshot_target = {
	.name    = "snapshot",
	.version = {1, 0, 1},
	.module  = THIS_MODULE,
	.ctr     = snapshot_ctr,
	.dtr     = snapshot_dtr,
	.map     = snapshot_map,
	.resume  = snapshot_resume,
	.status  = snapshot_status,
};

static int __init dm_snapshot_init(void)
{
	int r;

	r = dm_register_target(&snapshot_target);
	if (r) {
		DMERR("snapshot target register failed %d", r);
		return r;
	}

	r = dm_register_target(&origin_target);
	if (r < 0) {
		DMERR("Device mapper: Origin: register failed %d\n", r);
		goto bad1;
	}

	r = init_origin_hash();
	if (r) {
		DMERR("init_origin_hash failed.");
		goto bad2;
	}

	exception_cache = kmem_cache_create("dm-snapshot-ex",
					    sizeof(struct exception),
					    __alignof__(struct exception),
					    0, NULL, NULL);
	if (!exception_cache) {
		DMERR("Couldn't create exception cache.");
		r = -ENOMEM;
		goto bad3;
	}

	pending_cache =
	    kmem_cache_create("dm-snapshot-in",
			      sizeof(struct pending_exception),
			      __alignof__(struct pending_exception),
			      0, NULL, NULL);
	if (!pending_cache) {
		DMERR("Couldn't create pending cache.");
		r = -ENOMEM;
		goto bad4;
	}

	pending_pool = mempool_create(128, mempool_alloc_slab,
				      mempool_free_slab, pending_cache);
	if (!pending_pool) {
		DMERR("Couldn't create pending pool.");
		r = -ENOMEM;
		goto bad5;
	}

	return 0;

      bad5:
	kmem_cache_destroy(pending_cache);
      bad4:
	kmem_cache_destroy(exception_cache);
      bad3:
	exit_origin_hash();
      bad2:
	dm_unregister_target(&origin_target);
      bad1:
	dm_unregister_target(&snapshot_target);
	return r;
}

static void __exit dm_snapshot_exit(void)
{
	int r;

	r = dm_unregister_target(&snapshot_target);
	if (r)
		DMERR("snapshot unregister failed %d", r);

	r = dm_unregister_target(&origin_target);
	if (r)
		DMERR("origin unregister failed %d", r);

	exit_origin_hash();
	mempool_destroy(pending_pool);
	kmem_cache_destroy(pending_cache);
	kmem_cache_destroy(exception_cache);
}

/* Module hooks */
module_init(dm_snapshot_init);
module_exit(dm_snapshot_exit);

MODULE_DESCRIPTION(DM_NAME " snapshot target");
MODULE_AUTHOR("Joe Thornber");
MODULE_LICENSE("GPL");