1 files changed, 137 insertions, 15 deletions

diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index a1cc797fe88f..d52ea584e0bc 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -200,7 +200,13 @@ struct dm_thin_new_mapping;
 enum pool_mode {
 	PM_WRITE,		/* metadata may be changed */
 	PM_OUT_OF_DATA_SPACE,	/* metadata may be changed, though data may not be allocated */
+
+	/*
+	 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
+	 */
+	PM_OUT_OF_METADATA_SPACE,
 	PM_READ_ONLY,		/* metadata may not be changed */
+
 	PM_FAIL,		/* all I/O fails */
 };
 
@@ -250,6 +256,7 @@ struct pool {
 
 	spinlock_t lock;
 	struct bio_list deferred_flush_bios;
+	struct bio_list deferred_flush_completions;
 	struct list_head prepared_mappings;
 	struct list_head prepared_discards;
 	struct list_head active_thins;
@@ -914,6 +921,39 @@ static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
 	mempool_free(m, m->tc->pool->mapping_pool);
 }
 
+static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
+{
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	/*
+	 * If the bio has the REQ_FUA flag set we must commit the metadata
+	 * before signaling its completion.
+	 */
+	if (!bio_triggers_commit(tc, bio)) {
+		bio_endio(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_completions, bio);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
 {
 	struct thin_c *tc = m->tc;
@@ -946,7 +986,7 @@ static void process_prepared_mapping(struct dm_thin_new_mapping *m)
 	 */
 	if (bio) {
 		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
-		bio_endio(bio);
+		complete_overwrite_bio(tc, bio);
 	} else {
 		inc_all_io_entry(tc->pool, m->cell->holder);
 		remap_and_issue(tc, m->cell->holder, m->data_block);
@@ -1299,7 +1339,37 @@ static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
 
 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
 
-static void check_for_space(struct pool *pool)
+static void requeue_bios(struct pool *pool);
+
+static bool is_read_only_pool_mode(enum pool_mode mode)
+{
+	return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
+}
+
+static bool is_read_only(struct pool *pool)
+{
+	return is_read_only_pool_mode(get_pool_mode(pool));
+}
+
+static void check_for_metadata_space(struct pool *pool)
+{
+	int r;
+	const char *ooms_reason = NULL;
+	dm_block_t nr_free;
+
+	r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
+	if (r)
+		ooms_reason = "Could not get free metadata blocks";
+	else if (!nr_free)
+		ooms_reason = "No free metadata blocks";
+
+	if (ooms_reason && !is_read_only(pool)) {
+		DMERR("%s", ooms_reason);
+		set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
+	}
+}
+
+static void check_for_data_space(struct pool *pool)
 {
 	int r;
 	dm_block_t nr_free;
@@ -1311,8 +1381,10 @@ static void check_for_space(struct pool *pool)
 	if (r)
 		return;
 
-	if (nr_free)
+	if (nr_free) {
 		set_pool_mode(pool, PM_WRITE);
+		requeue_bios(pool);
+	}
 }
 
 /*
@@ -1323,14 +1395,16 @@ static int commit(struct pool *pool)
 {
 	int r;
 
-	if (get_pool_mode(pool) >= PM_READ_ONLY)
+	if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
 		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);
+	else {
+		check_for_metadata_space(pool);
+		check_for_data_space(pool);
+	}
 
 	return r;
 }
@@ -1389,10 +1463,26 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
 
 	r = dm_pool_alloc_data_block(pool->pmd, result);
 	if (r) {
-		metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
+		if (r == -ENOSPC)
+			set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
+		else
+			metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
+		return r;
+	}
+
+	r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
+	if (r) {
+		metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
 		return r;
 	}
 
+	if (!free_blocks) {
+		/* Let's commit before we use up the metadata reserve. */
+		r = commit(pool);
+		if (r)
+			return r;
+	}
+
 	return 0;
 }
 
@@ -1424,6 +1514,7 @@ static int should_error_unserviceable_bio(struct pool *pool)
 	case PM_OUT_OF_DATA_SPACE:
 		return pool->pf.error_if_no_space ? -ENOSPC : 0;
 
+	case PM_OUT_OF_METADATA_SPACE:
 	case PM_READ_ONLY:
 	case PM_FAIL:
 		return -EIO;
@@ -2171,7 +2262,7 @@ static void process_deferred_bios(struct pool *pool)
 {
 	unsigned long flags;
 	struct bio *bio;
-	struct bio_list bios;
+	struct bio_list bios, bio_completions;
 	struct thin_c *tc;
 
 	tc = get_first_thin(pool);
@@ -2182,26 +2273,36 @@ static void process_deferred_bios(struct pool *pool)
 	}
 
 	/*
-	 * If there are any deferred flush bios, we must commit
-	 * the metadata before issuing them.
+	 * If there are any deferred flush bios, we must commit the metadata
+	 * before issuing them or signaling their completion.
 	 */
 	bio_list_init(&bios);
+	bio_list_init(&bio_completions);
+
 	spin_lock_irqsave(&pool->lock, flags);
 	bio_list_merge(&bios, &pool->deferred_flush_bios);
 	bio_list_init(&pool->deferred_flush_bios);
+
+	bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
+	bio_list_init(&pool->deferred_flush_completions);
 	spin_unlock_irqrestore(&pool->lock, flags);
 
-	if (bio_list_empty(&bios) &&
+	if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
 	    !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
 		return;
 
 	if (commit(pool)) {
+		bio_list_merge(&bios, &bio_completions);
+
 		while ((bio = bio_list_pop(&bios)))
 			bio_io_error(bio);
 		return;
 	}
 	pool->last_commit_jiffies = jiffies;
 
+	while ((bio = bio_list_pop(&bio_completions)))
+		bio_endio(bio);
+
 	while ((bio = bio_list_pop(&bios)))
 		generic_make_request(bio);
 }
@@ -2394,8 +2495,9 @@ static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
 		error_retry_list(pool);
 		break;
 
+	case PM_OUT_OF_METADATA_SPACE:
 	case PM_READ_ONLY:
-		if (old_mode != new_mode)
+		if (!is_read_only_pool_mode(old_mode))
 			notify_of_pool_mode_change(pool, "read-only");
 		dm_pool_metadata_read_only(pool->pmd);
 		pool->process_bio = process_bio_read_only;
@@ -2827,6 +2929,7 @@ static struct pool *pool_create(struct mapped_device *pool_md,
 	INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
 	spin_lock_init(&pool->lock);
 	bio_list_init(&pool->deferred_flush_bios);
+	bio_list_init(&pool->deferred_flush_completions);
 	INIT_LIST_HEAD(&pool->prepared_mappings);
 	INIT_LIST_HEAD(&pool->prepared_discards);
 	INIT_LIST_HEAD(&pool->active_thins);
@@ -3107,6 +3210,13 @@ static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
 	as.argc = argc;
 	as.argv = argv;
 
+	/* make sure metadata and data are different devices */
+	if (!strcmp(argv[0], argv[1])) {
+		ti->error = "Error setting metadata or data device";
+		r = -EINVAL;
+		goto out_unlock;
+	}
+
 	/*
 	 * Set default pool features.
 	 */
@@ -3326,6 +3436,10 @@ static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
 		DMINFO("%s: growing the metadata device from %llu to %llu blocks",
 		       dm_device_name(pool->pool_md),
 		       sb_metadata_dev_size, metadata_dev_size);
+
+		if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
+			set_pool_mode(pool, PM_WRITE);
+
 		r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
 		if (r) {
 			metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
@@ -3629,7 +3743,7 @@ static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
 	struct pool_c *pt = ti->private;
 	struct pool *pool = pt->pool;
 
-	if (get_pool_mode(pool) >= PM_READ_ONLY) {
+	if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
 		DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
 		      dm_device_name(pool->pool_md));
 		return -EOPNOTSUPP;
@@ -3703,6 +3817,7 @@ static void pool_status(struct dm_target *ti, status_type_t type,
 	dm_block_t nr_blocks_data;
 	dm_block_t nr_blocks_metadata;
 	dm_block_t held_root;
+	enum pool_mode mode;
 	char buf[BDEVNAME_SIZE];
 	char buf2[BDEVNAME_SIZE];
 	struct pool_c *pt = ti->private;
@@ -3773,9 +3888,10 @@ static void pool_status(struct dm_target *ti, status_type_t type,
 		else
 			DMEMIT("- ");
 
-		if (pool->pf.mode == PM_OUT_OF_DATA_SPACE)
+		mode = get_pool_mode(pool);
+		if (mode == PM_OUT_OF_DATA_SPACE)
 			DMEMIT("out_of_data_space ");
-		else if (pool->pf.mode == PM_READ_ONLY)
+		else if (is_read_only_pool_mode(mode))
 			DMEMIT("ro ");
 		else
 			DMEMIT("rw ");
@@ -3983,6 +4099,12 @@ static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
 	tc->sort_bio_list = RB_ROOT;
 
 	if (argc == 3) {
+		if (!strcmp(argv[0], argv[2])) {
+			ti->error = "Error setting origin device";
+			r = -EINVAL;
+			goto bad_origin_dev;
+		}
+
 		r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
 		if (r) {
 			ti->error = "Error opening origin device";