block: don't ignore REQ_NOWAIT for direct IO
authorPavel Begunkov <asml.silence@gmail.com>
Fri, 20 Nov 2020 17:10:28 +0000 (17:10 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 16 Apr 2021 09:46:38 +0000 (11:46 +0200)
[ Upstream commit f8b78caf21d5bc3fcfc40c18898f9d52ed1451a5 ]

If IOCB_NOWAIT is set on submission, then that needs to get propagated to
REQ_NOWAIT on the block side. Otherwise we completely lose this
information, and any issuer of IOCB_NOWAIT IO will potentially end up
blocking on eg request allocation on the storage side.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/block_dev.c

index 79272cdbe82773fe1d99ac80a74c8f56fb62cf9d..bd93563477a43609db620e059be2a90606199886 100644 (file)
@@ -246,6 +246,8 @@ __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
                bio.bi_opf = dio_bio_write_op(iocb);
                task_io_account_write(ret);
        }
+       if (iocb->ki_flags & IOCB_NOWAIT)
+               bio.bi_opf |= REQ_NOWAIT;
        if (iocb->ki_flags & IOCB_HIPRI)
                bio_set_polled(&bio, iocb);
 
@@ -399,6 +401,8 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
                        bio->bi_opf = dio_bio_write_op(iocb);
                        task_io_account_write(bio->bi_iter.bi_size);
                }
+               if (iocb->ki_flags & IOCB_NOWAIT)
+                       bio->bi_opf |= REQ_NOWAIT;
 
                dio->size += bio->bi_iter.bi_size;
                pos += bio->bi_iter.bi_size;