On Fri, Apr 12, 2019 at 9:53 AM Jason Yan <yanaijie@xxxxxxxxxx> wrote:
>
> If the last bio returned is not dio->bio, the status of the bio will
> not assigned to dio->bio if it is error. This will cause the whole IO
> status wrong.
>
> ksoftirqd/21-117 [021] ..s. 4017.966090: 8,0 C N 4883648 [0]
> <idle>-0 [018] ..s. 4017.970888: 8,0 C WS 4924800 + 1024 [0]
> <idle>-0 [018] ..s. 4017.970909: 8,0 D WS 4935424 + 1024 [<idle>]
> <idle>-0 [018] ..s. 4017.970924: 8,0 D WS 4936448 + 321 [<idle>]
> ksoftirqd/21-117 [021] ..s. 4017.995033: 8,0 C R 4883648 + 336 [65475]
> ksoftirqd/21-117 [021] d.s. 4018.001988: myprobe1: (blkdev_bio_end_io+0x0/0x168) bi_status=7
> ksoftirqd/21-117 [021] d.s. 4018.001992: myprobe: (aio_complete_rw+0x0/0x148) x0=0xffff802f2595ad80 res=0x12a000 res2=0x0
>
> We always have to assign bio->bi_status to dio->bio.bi_status because we
> will only check dio->bio.bi_status when we return the whole IO to
> the upper layer.
>
> Fixes: 542ff7bf18c6 ("block: new direct I/O implementation")
> Cc: Christoph Hellwig <hch@xxxxxx>
> Cc: Jens Axboe <axboe@xxxxxxxxx>
> Cc: Ming Lei <ming.lei@xxxxxxxxxx>
> Signed-off-by: Jason Yan <yanaijie@xxxxxxxxxx>
> ---
>
> v2: remove white space changes
>
> fs/block_dev.c | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 78d3257435c0..24615c76c1d0 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -307,10 +307,10 @@ static void blkdev_bio_end_io(struct bio *bio)
> struct blkdev_dio *dio = bio->bi_private;
> bool should_dirty = dio->should_dirty;
>
> - if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
> - if (bio->bi_status && !dio->bio.bi_status)
> - dio->bio.bi_status = bio->bi_status;
> - } else {
> + if (bio->bi_status && !dio->bio.bi_status)
> + dio->bio.bi_status = bio->bi_status;
> +
> + if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
> if (!dio->is_sync) {
> struct kiocb *iocb = dio->iocb;
> ssize_t ret;
> --
> 2.17.2
>
Looks fine:
Reviewed-by: Ming Lei <ming.lei@xxxxxxxxxx>
--
Ming Lei
