[PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[Konstantin Khlebnikov]

This is required to avoid waiting in lower layers.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
---
 fs/iomap/direct-io.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index 20dde5aadcdd..9b53fa7651e3 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -63,6 +63,8 @@ static void iomap_dio_submit_bio(struct iomap_dio *dio, struct iomap *iomap,
 {
 	atomic_inc(&dio->ref);
 
+	if (dio->iocb->ki_flags & IOCB_NOWAIT)
+		bio->bi_opf |= REQ_NOWAIT;
 	if (dio->iocb->ki_flags & IOCB_HIPRI)
 		bio_set_polled(bio, dio->iocb);
 

[PATCH RFC 2/2] fs/direct-io: pass NOWAIT to also for read requests

[Konstantin Khlebnikov]

For some reason NOWAIT currently is passed only for writes.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Fixes: 03a07c92a9ed ("block: return on congested block device")
---
 fs/direct-io.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/direct-io.c b/fs/direct-io.c
index 00b4d15bb811..dbb6afef6be9 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -1234,11 +1234,11 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	if (iov_iter_rw(iter) == WRITE) {
 		dio->op = REQ_OP_WRITE;
 		dio->op_flags = REQ_SYNC | REQ_IDLE;
-		if (iocb->ki_flags & IOCB_NOWAIT)
-			dio->op_flags |= REQ_NOWAIT;
 	} else {
 		dio->op = REQ_OP_READ;
 	}
+	if (iocb->ki_flags & IOCB_NOWAIT)
+		dio->op_flags |= REQ_NOWAIT;
 	if (iocb->ki_flags & IOCB_HIPRI)
 		dio->op_flags |= REQ_HIPRI;
 

Re: [PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[Christoph Hellwig]

On Mon, May 04, 2020 at 06:54:53PM +0300, Konstantin Khlebnikov wrote:
> This is required to avoid waiting in lower layers.
> 
> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>

This looks sensible.  Did you run this through xfstests?

Re: [PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[Konstantin Khlebnikov]

On 04/05/2020 19.00, Christoph Hellwig wrote:
> On Mon, May 04, 2020 at 06:54:53PM +0300, Konstantin Khlebnikov wrote:
>> This is required to avoid waiting in lower layers.
>>
>> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
> 
> This looks sensible.  Did you run this through xfstests?
> 

Nope. It seems xfstests has one trivial test for NOWAIT - generic/471
It tests only write with/without extent, nothing about contention.

I've added nowait into fio and played with it a little.
https://github.com/axboe/fio/pull/972

With these patches I see EAGAINs when queue is flooded.

Re: [PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[Christoph Hellwig]

On Mon, May 04, 2020 at 07:23:50PM +0300, Konstantin Khlebnikov wrote:
> On 04/05/2020 19.00, Christoph Hellwig wrote:
> > On Mon, May 04, 2020 at 06:54:53PM +0300, Konstantin Khlebnikov wrote:
> > > This is required to avoid waiting in lower layers.
> > > 
> > > Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
> > 
> > This looks sensible.  Did you run this through xfstests?
> > 
> 
> Nope. It seems xfstests has one trivial test for NOWAIT - generic/471
> It tests only write with/without extent, nothing about contention.
> 
> I've added nowait into fio and played with it a little.
> https://github.com/axboe/fio/pull/972
> 
> With these patches I see EAGAINs when queue is flooded.

Once the fio changes land, can you add a simple fio based test
to xfstests?

Re: [PATCH RFC 1/2] fs/iomap/direct-io: pass NOWAIT to bio flags

[Christoph Hellwig]

On Mon, May 04, 2020 at 06:54:53PM +0300, Konstantin Khlebnikov wrote:
> This is required to avoid waiting in lower layers.
> 
> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH RFC 2/2] fs/direct-io: pass NOWAIT to also for read requests

[Christoph Hellwig]

On Mon, May 04, 2020 at 06:54:57PM +0300, Konstantin Khlebnikov wrote:
> For some reason NOWAIT currently is passed only for writes.
> 
> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
> Fixes: 03a07c92a9ed ("block: return on congested block device")

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>