block: handle the null_mapped flag correctly in blk_rq_map_user_iov
authorChristoph Hellwig <hch@lst.de>
Wed, 11 Feb 2015 13:07:49 +0000 (14:07 +0100)
committerJens Axboe <axboe@fb.com>
Wed, 11 Feb 2015 18:24:12 +0000 (11:24 -0700)
The tape drivers (and the sg driver in a special case that doesn't matter
here) use the null_mapped flag to tell blk_rq_map_user to not copy around
any data into or out of the bounce buffers.  blk_rq_map_user_iov never
got that treatment, which didn't matter until I refactored blk_rq_map_user
to be implemented in terms of blk_rq_map_user_iov.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Fixes: ddad8dd0a162 ("block: use blk_rq_map_user_iov to implement blk_rq_map_user")
Signed-off-by: Jens Axboe <axboe@fb.com>
block/blk-map.c

index 0f22911f17dc42f8e969c9f69a2d493c7821b16a..b8d2725324a6b88391db4c8a5daf7a9c1309dd86 100644 (file)
@@ -93,6 +93,9 @@ int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
        if (IS_ERR(bio))
                return PTR_ERR(bio);
 
+       if (map_data && map_data->null_mapped)
+               bio->bi_flags |= (1 << BIO_NULL_MAPPED);
+
        if (bio->bi_iter.bi_size != iter->count) {
                /*
                 * Grab an extra reference to this bio, as bio_unmap_user()