diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index 188121c..703bbbc 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -520,10 +520,12 @@
 	struct bio_list deferred_bios;
 	struct bio_list deferred_flush_bios;
 	struct list_head prepared_mappings;
+	struct list_head prepared_discards;
 
 	struct bio_list retry_on_resume_list;
 
 	struct deferred_set shared_read_ds;
+	struct deferred_set all_io_ds;
 
 	struct new_mapping *next_mapping;
 	mempool_t *mapping_pool;
@@ -621,6 +623,7 @@
 struct endio_hook {
 	struct thin_c *tc;
 	struct deferred_entry *shared_read_entry;
+	struct deferred_entry *all_io_entry;
 	struct new_mapping *overwrite_mapping;
 };
 
@@ -728,11 +731,12 @@
 
 	unsigned quiesced:1;
 	unsigned prepared:1;
+	unsigned pass_discard:1;
 
 	struct thin_c *tc;
 	dm_block_t virt_block;
 	dm_block_t data_block;
-	struct cell *cell;
+	struct cell *cell, *cell2;
 	int err;
 
 	/*
@@ -872,7 +876,30 @@
 	mempool_free(m, tc->pool->mapping_pool);
 }
 
-static void process_prepared_mappings(struct pool *pool)
+static void process_prepared_discard(struct new_mapping *m)
+{
+	int r;
+	struct thin_c *tc = m->tc;
+
+	r = dm_thin_remove_block(tc->td, m->virt_block);
+	if (r)
+		DMERR("dm_thin_remove_block() failed");
+
+	/*
+	 * Pass the discard down to the underlying device?
+	 */
+	if (m->pass_discard)
+		remap_and_issue(tc, m->bio, m->data_block);
+	else
+		bio_endio(m->bio, 0);
+
+	cell_defer_except(tc, m->cell);
+	cell_defer_except(tc, m->cell2);
+	mempool_free(m, tc->pool->mapping_pool);
+}
+
+static void process_prepared(struct pool *pool, struct list_head *head,
+			     void (*fn)(struct new_mapping *))
 {
 	unsigned long flags;
 	struct list_head maps;
@@ -880,21 +907,27 @@
 
 	INIT_LIST_HEAD(&maps);
 	spin_lock_irqsave(&pool->lock, flags);
-	list_splice_init(&pool->prepared_mappings, &maps);
+	list_splice_init(head, &maps);
 	spin_unlock_irqrestore(&pool->lock, flags);
 
 	list_for_each_entry_safe(m, tmp, &maps, list)
-		process_prepared_mapping(m);
+		fn(m);
 }
 
 /*
  * Deferred bio jobs.
  */
+static int io_overlaps_block(struct pool *pool, struct bio *bio)
+{
+	return !(bio->bi_sector & pool->offset_mask) &&
+		(bio->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) &&
-		!(bio->bi_sector & pool->offset_mask)) &&
-		(bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT));
+	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,
@@ -1134,6 +1167,86 @@
 		retry_on_resume(bio);
 }
 
+static void process_discard(struct thin_c *tc, struct bio *bio)
+{
+	int r;
+	struct pool *pool = tc->pool;
+	struct cell *cell, *cell2;
+	struct cell_key key, key2;
+	dm_block_t block = get_bio_block(tc, bio);
+	struct dm_thin_lookup_result lookup_result;
+	struct new_mapping *m;
+
+	build_virtual_key(tc->td, block, &key);
+	if (bio_detain(tc->pool->prison, &key, bio, &cell))
+		return;
+
+	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
+	switch (r) {
+	case 0:
+		/*
+		 * Check nobody is fiddling with this pool block.  This can
+		 * happen if someone's in the process of breaking sharing
+		 * on this block.
+		 */
+		build_data_key(tc->td, lookup_result.block, &key2);
+		if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) {
+			cell_release_singleton(cell, bio);
+			break;
+		}
+
+		if (io_overlaps_block(pool, bio)) {
+			/*
+			 * IO may still be going to the destination block.  We must
+			 * quiesce before we can do the removal.
+			 */
+			m = get_next_mapping(pool);
+			m->tc = tc;
+			m->pass_discard = !lookup_result.shared;
+			m->virt_block = block;
+			m->data_block = lookup_result.block;
+			m->cell = cell;
+			m->cell2 = cell2;
+			m->err = 0;
+			m->bio = bio;
+
+			if (!ds_add_work(&pool->all_io_ds, &m->list)) {
+				list_add(&m->list, &pool->prepared_discards);
+				wake_worker(pool);
+			}
+		} else {
+			/*
+			 * This path is hit if people are ignoring
+			 * limits->discard_granularity.  It ignores any
+			 * part of the discard that is in a subsequent
+			 * block.
+			 */
+			sector_t offset = bio->bi_sector - (block << pool->block_shift);
+			unsigned remaining = (pool->sectors_per_block - offset) << 9;
+			bio->bi_size = min(bio->bi_size, remaining);
+
+			cell_release_singleton(cell, bio);
+			cell_release_singleton(cell2, bio);
+			remap_and_issue(tc, bio, lookup_result.block);
+		}
+		break;
+
+	case -ENODATA:
+		/*
+		 * It isn't provisioned, just forget it.
+		 */
+		cell_release_singleton(cell, bio);
+		bio_endio(bio, 0);
+		break;
+
+	default:
+		DMERR("discard: find block unexpectedly returned %d", r);
+		cell_release_singleton(cell, bio);
+		bio_io_error(bio);
+		break;
+	}
+}
+
 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
 			  struct cell_key *key,
 			  struct dm_thin_lookup_result *lookup_result,
@@ -1279,6 +1392,7 @@
 
 	default:
 		DMERR("dm_thin_find_block() failed, error = %d", r);
+		cell_release_singleton(cell, bio);
 		bio_io_error(bio);
 		break;
 	}
@@ -1320,7 +1434,11 @@
 
 			break;
 		}
-		process_bio(tc, bio);
+
+		if (bio->bi_rw & REQ_DISCARD)
+			process_discard(tc, bio);
+		else
+			process_bio(tc, bio);
 	}
 
 	/*
@@ -1354,7 +1472,8 @@
 {
 	struct pool *pool = container_of(ws, struct pool, worker);
 
-	process_prepared_mappings(pool);
+	process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping);
+	process_prepared(pool, &pool->prepared_discards, process_prepared_discard);
 	process_deferred_bios(pool);
 }
 
@@ -1397,6 +1516,7 @@
 
 	h->tc = tc;
 	h->shared_read_entry = NULL;
+	h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds);
 	h->overwrite_mapping = NULL;
 
 	return h;
@@ -1415,7 +1535,7 @@
 	struct dm_thin_lookup_result result;
 
 	map_context->ptr = thin_hook_bio(tc, bio);
-	if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
+	if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
 		thin_defer_bio(tc, bio);
 		return DM_MAPIO_SUBMITTED;
 	}
@@ -1591,10 +1711,12 @@
 	bio_list_init(&pool->deferred_bios);
 	bio_list_init(&pool->deferred_flush_bios);
 	INIT_LIST_HEAD(&pool->prepared_mappings);
+	INIT_LIST_HEAD(&pool->prepared_discards);
 	pool->low_water_triggered = 0;
 	pool->no_free_space = 0;
 	bio_list_init(&pool->retry_on_resume_list);
 	ds_init(&pool->shared_read_ds);
+	ds_init(&pool->all_io_ds);
 
 	pool->next_mapping = NULL;
 	pool->mapping_pool =
@@ -1834,7 +1956,8 @@
 	pt->low_water_blocks = low_water_blocks;
 	pt->zero_new_blocks = pf.zero_new_blocks;
 	ti->num_flush_requests = 1;
-	ti->num_discard_requests = 0;
+	ti->num_discard_requests = 1;
+	ti->discards_supported = 1;
 	ti->private = pt;
 
 	pt->callbacks.congested_fn = pool_is_congested;
@@ -2227,6 +2350,17 @@
 	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
 }
 
+static void set_discard_limits(struct pool *pool, struct queue_limits *limits)
+{
+	limits->max_discard_sectors = pool->sectors_per_block;
+
+	/*
+	 * This is just a hint, and not enforced.  We have to cope with
+	 * bios that overlap 2 blocks.
+	 */
+	limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
+}
+
 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
 {
 	struct pool_c *pt = ti->private;
@@ -2234,6 +2368,7 @@
 
 	blk_limits_io_min(limits, 0);
 	blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
+	set_discard_limits(pool, limits);
 }
 
 static struct target_type pool_target = {
@@ -2350,8 +2485,8 @@
 
 	ti->split_io = tc->pool->sectors_per_block;
 	ti->num_flush_requests = 1;
-	ti->num_discard_requests = 0;
-	ti->discards_supported = 0;
+	ti->num_discard_requests = 1;
+	ti->discards_supported = 1;
 
 	dm_put(pool_md);
 
@@ -2407,6 +2542,13 @@
 		spin_unlock_irqrestore(&pool->lock, flags);
 	}
 
+	if (h->all_io_entry) {
+		INIT_LIST_HEAD(&work);
+		ds_dec(h->all_io_entry, &work);
+		list_for_each_entry_safe(m, tmp, &work, list)
+			list_add(&m->list, &pool->prepared_discards);
+	}
+
 	mempool_free(h, pool->endio_hook_pool);
 
 	return 0;
@@ -2487,9 +2629,11 @@
 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
 {
 	struct thin_c *tc = ti->private;
+	struct pool *pool = tc->pool;
 
 	blk_limits_io_min(limits, 0);
-	blk_limits_io_opt(limits, tc->pool->sectors_per_block << SECTOR_SHIFT);
+	blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
+	set_discard_limits(pool, limits);
 }
 
 static struct target_type thin_target = {