On 06/27/18 16:50, Ming Lei wrote:
On Wed, Jun 27, 2018 at 01:12:31PM -0700, Bart Van Assche wrote:
Although __bio_clone_fast() copies bi_io_vec, it does not copy bi_vcnt,
the number of elements in bi_io_vec[] that contains data. Copy bi_vcnt
such that code that needs this member behaves identically for original
and for cloned requests.
Signed-off-by: Bart Van Assche <bart.vanassche@xxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Mike Snitzer <snitzer@xxxxxxxxxx>
Cc: Ming Lei <ming.lei@xxxxxxxxxx>
Cc: Hannes Reinecke <hare@xxxxxxxx>
Cc: Johannes Thumshirn <jthumshirn@xxxxxxx>
---
block/bio.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/block/bio.c b/block/bio.c
index f7e3d88bd0b6..55f8e0dedd69 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -605,6 +605,7 @@ void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
bio->bi_opf = bio_src->bi_opf;
bio->bi_write_hint = bio_src->bi_write_hint;
bio->bi_iter = bio_src->bi_iter;
+ bio->bi_vcnt = bio_src->bi_vcnt;
bio->bi_io_vec = bio_src->bi_io_vec;
No, don't do that.
Why not? I think it's a huge booby trap that cloned bio's have a
bi_io_vec but zero bi_vcnt.
For a cloned bio, both .bi_vcnt and .bi_io_vec can't be used directly,
and we have done huge such cleanup for supporting multipage bvec, that is
the great idea of immutable bvecs invented by Kent.
My understanding is that the above change doesn't conflict at all with
the immutable biovec work -- to the contrary.
Bart.
