On Tue, Jun 20, 2017 at 01:24:28AM +0100, Al Viro wrote:

On Mon, Jun 19, 2017 at 05:36:05PM -0600, Jens Axboe wrote:

quoted

On 06/19/2017 05:34 PM, Al Viro wrote:

quoted

On Mon, Jun 19, 2017 at 05:15:16PM -0600, Jens Axboe wrote:

quoted

On 06/19/2017 10:33 AM, Goldwyn Rodrigues wrote:

quoted

Jens,

As Christoph suggested, I am sending the patches against the block
tree for merge since the block layer changes had the most conflicts.
My tree is at https://github.com/goldwynr/linux/tree/nowait-block

I can merge it for 4.13, but I need Al to be happy with it first.

FWIW, it's really a matter of who gets more conflicts - I'm OK with the
general shape of that code and it could go through either tree (or both,
for that matter, if an immutable branch is promised).

I'm still hopeful that the write hint patchset will make it as well,
which will definitely conflict with this. So if you don't mind, I'd
like to queue it through the block tree. The block for-4.x/ branches
are not rebased.

Fine by me; the only problem I see there at the moment is with ext4 and
xfs ->read_iter() not buggering off on nowait (xfs one has
        xfs_ilock(ip, XFS_IOLOCK_SHARED);
done before it gets to generic_file_read_iter()).

Something like the patch below should take care of the XFS, free feel
to fold.

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index e159eb381d9f..c74847f71b3d 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c

@@ -216,7 +216,11 @@ xfs_file_dio_aio_read(
 
 	file_accessed(iocb->ki_filp);
 
-	xfs_ilock(ip, XFS_IOLOCK_SHARED);
+	if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) {
+		if (iocb->ki_flags & IOCB_NOWAIT)
+			return -EAGAIN;
+		xfs_ilock(ip, XFS_IOLOCK_SHARED);
+	}
 	ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 

@@ -237,7 +241,11 @@ xfs_file_dax_read(
 	if (!count)
 		return 0; /* skip atime */
 
-	xfs_ilock(ip, XFS_IOLOCK_SHARED);
+	if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) {
+		if (iocb->ki_flags & IOCB_NOWAIT)
+			return -EAGAIN;
+		xfs_ilock(ip, XFS_IOLOCK_SHARED);
+	}
 	ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);