Re: [PATCH] io_uring: retry in case of short read on block device
From: Pavel Begunkov <asml.silence@gmail.com>
Date: 2021-08-31 10:07:55
On 8/21/21 4:07 PM, Ming Lei wrote:
In case of buffered reading from block device, when short read happens,
we should retry to read more, otherwise the IO will be completed
partially, for example, the following fio expects to read 2MB, but it
can only read 1M or less bytes:
fio --name=onessd --filename=/dev/nvme0n1 --filesize=2M \
--rw=randread --bs=2M --direct=0 --overwrite=0 --numjobs=1 \
--iodepth=1 --time_based=0 --runtime=2 --ioengine=io_uring \
--registerfiles --fixedbufs --gtod_reduce=1 --group_reporting
Fix the issue by allowing short read retry for block device, which sets
FMODE_BUF_RASYNC really.Should note that overhead on touching inode shouldn't be of concern at this point, so all looks good Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
quoted hunk ↗ jump to hunk
Fixes: 9a173346bd9e ("io_uring: fix short read retries for non-reg files") Cc: Pavel Begunkov <asml.silence@gmail.com> Signed-off-by: Ming Lei <redacted> --- fs/io_uring.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)diff --git a/fs/io_uring.c b/fs/io_uring.c index bf548af0426c..bbcd1a9e75e5 100644 --- a/fs/io_uring.c +++ b/fs/io_uring.c@@ -3268,6 +3268,12 @@ static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) return -EINVAL; } +static bool need_read_all(struct io_kiocb *req) +{ + return req->flags & REQ_F_ISREG || + S_ISBLK(file_inode(req->file)->i_mode); +} + static int io_read(struct io_kiocb *req, unsigned int issue_flags) { struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;@@ -3322,7 +3328,7 @@ static int io_read(struct io_kiocb *req, unsigned int issue_flags) } else if (ret == -EIOCBQUEUED) { goto out_free; } else if (ret <= 0 || ret == io_size || !force_nonblock || - (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { + (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { /* read all, failed, already did sync or don't want to retry */ goto done; }
-- Pavel Begunkov