1 files changed, 39 insertions, 27 deletions

diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index 2c0cf51..ee29037 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -640,7 +640,9 @@ static void process_prepared_mapping(struct dm_thin_new_mapping *m)
 	 */
 	r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
 	if (r) {
-		DMERR_LIMIT("dm_thin_insert_block() failed");
+		DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
+			    dm_device_name(pool->pool_md), r);
+		set_pool_mode(pool, PM_READ_ONLY);
 		cell_error(pool, m->cell);
 		goto out;
 	}
@@ -881,32 +883,23 @@ static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
 	}
 }
 
-static int commit(struct pool *pool)
-{
-	int r;
-
-	r = dm_pool_commit_metadata(pool->pmd);
-	if (r)
-		DMERR_LIMIT("%s: commit failed: error = %d",
-			    dm_device_name(pool->pool_md), r);
-
-	return r;
-}
-
 /*
  * A non-zero return indicates read_only or fail_io mode.
  * Many callers don't care about the return value.
  */
-static int commit_or_fallback(struct pool *pool)
+static int commit(struct pool *pool)
 {
 	int r;
 
 	if (get_pool_mode(pool) != PM_WRITE)
 		return -EINVAL;
 
-	r = commit(pool);
-	if (r)
+	r = dm_pool_commit_metadata(pool->pmd);
+	if (r) {
+		DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
+			    dm_device_name(pool->pool_md), r);
 		set_pool_mode(pool, PM_READ_ONLY);
+	}
 
 	return r;
 }
@@ -943,7 +936,9 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
 		 * Try to commit to see if that will free up some
 		 * more space.
 		 */
-		(void) commit_or_fallback(pool);
+		r = commit(pool);
+		if (r)
+			return r;
 
 		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
 		if (r)
@@ -957,7 +952,7 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
 		 * table reload).
 		 */
 		if (!free_blocks) {
-			DMWARN("%s: no free space available.",
+			DMWARN("%s: no free data space available.",
 			       dm_device_name(pool->pool_md));
 			spin_lock_irqsave(&pool->lock, flags);
 			pool->no_free_space = 1;
@@ -967,8 +962,16 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
 	}
 
 	r = dm_pool_alloc_data_block(pool->pmd, result);
-	if (r)
+	if (r) {
+		if (r == -ENOSPC &&
+		    !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
+		    !free_blocks) {
+			DMWARN("%s: no free metadata space available.",
+			       dm_device_name(pool->pool_md));
+			set_pool_mode(pool, PM_READ_ONLY);
+		}
 		return r;
+	}
 
 	return 0;
 }
@@ -1349,7 +1352,7 @@ static void process_deferred_bios(struct pool *pool)
 	if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
 		return;
 
-	if (commit_or_fallback(pool)) {
+	if (commit(pool)) {
 		while ((bio = bio_list_pop(&bios)))
 			bio_io_error(bio);
 		return;
@@ -1397,6 +1400,7 @@ static void set_pool_mode(struct pool *pool, enum pool_mode mode)
 	case PM_FAIL:
 		DMERR("%s: switching pool to failure mode",
 		      dm_device_name(pool->pool_md));
+		dm_pool_metadata_read_only(pool->pmd);
 		pool->process_bio = process_bio_fail;
 		pool->process_discard = process_bio_fail;
 		pool->process_prepared_mapping = process_prepared_mapping_fail;
@@ -1421,6 +1425,7 @@ static void set_pool_mode(struct pool *pool, enum pool_mode mode)
 		break;
 
 	case PM_WRITE:
+		dm_pool_metadata_read_write(pool->pmd);
 		pool->process_bio = process_bio;
 		pool->process_discard = process_discard;
 		pool->process_prepared_mapping = process_prepared_mapping;
@@ -1637,12 +1642,19 @@ static int bind_control_target(struct pool *pool, struct dm_target *ti)
 	struct pool_c *pt = ti->private;
 
 	/*
-	 * We want to make sure that degraded pools are never upgraded.
+	 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
 	 */
 	enum pool_mode old_mode = pool->pf.mode;
 	enum pool_mode new_mode = pt->adjusted_pf.mode;
 
-	if (old_mode > new_mode)
+	/*
+	 * If we were in PM_FAIL mode, rollback of metadata failed.  We're
+	 * not going to recover without a thin_repair.  So we never let the
+	 * pool move out of the old mode.  On the other hand a PM_READ_ONLY
+	 * may have been due to a lack of metadata or data space, and may
+	 * now work (ie. if the underlying devices have been resized).
+	 */
+	if (old_mode == PM_FAIL)
 		new_mode = old_mode;
 
 	pool->ti = ti;
@@ -2266,7 +2278,7 @@ static int pool_preresume(struct dm_target *ti)
 		return r;
 
 	if (need_commit1 || need_commit2)
-		(void) commit_or_fallback(pool);
+		(void) commit(pool);
 
 	return 0;
 }
@@ -2293,7 +2305,7 @@ static void pool_postsuspend(struct dm_target *ti)
 
 	cancel_delayed_work(&pool->waker);
 	flush_workqueue(pool->wq);
-	(void) commit_or_fallback(pool);
+	(void) commit(pool);
 }
 
 static int check_arg_count(unsigned argc, unsigned args_required)
@@ -2427,7 +2439,7 @@ static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct
 	if (r)
 		return r;
 
-	(void) commit_or_fallback(pool);
+	(void) commit(pool);
 
 	r = dm_pool_reserve_metadata_snap(pool->pmd);
 	if (r)
@@ -2489,7 +2501,7 @@ static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
 		DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
 
 	if (!r)
-		(void) commit_or_fallback(pool);
+		(void) commit(pool);
 
 	return r;
 }
@@ -2544,7 +2556,7 @@ static void pool_status(struct dm_target *ti, status_type_t type,
 
 		/* Commit to ensure statistics aren't out-of-date */
 		if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
-			(void) commit_or_fallback(pool);
+			(void) commit(pool);
 
 		r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
 		if (r) {