On Mon, Jun 19, 2023 at 11:18:55PM -0700, Christoph Hellwig wrote:
> So it turns out this gets into the way of my planned cleanup to move
> all the tatic FMODE_ flags out of this basically full field into a new
> static one in file_operations. Do you think it is ok to go back to
> always claiming FMODE_NOWAIT for block devices and then just punt for
> the drivers that don't support it like the patch below?
Except that the version I posted if of course completly broken as my
testing rig told me. This is the version that actually works:
---
>From 7916df7434e1570978b9c81c65aa1ec1f3396b13 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@xxxxxx>
Date: Tue, 20 Jun 2023 07:53:13 +0200
Subject: block: always set FMODE_NOWAIT
Block devices are the only file_operations that do not set FMODE_NOWAIT
unconditionall in ->open and thus get in the way of a planned cleanup to
move this flags into a static field in file_operations. Switch to
always set FMODE_NOWAIT and just return -EAGAIN if it isn't actually
supported. This just affects minor ->submit_bio based drivers as all
blk-mq drivers and the important remappers (dm, md, nvme-multipath)
support nowait I/O.
Signed-off-by: Christoph Hellwig <hch@xxxxxx>
---
block/fops.c | 11 +++++++----
1 file changed, 7 insertions(+), 4 deletions(-)
diff --git a/block/fops.c b/block/fops.c
index 555b1b9ecd2cb9..58c2f65ae4a57e 100644
--- a/block/fops.c
+++ b/block/fops.c
@@ -505,7 +505,7 @@ static int blkdev_open(struct inode *inode, struct file *filp)
* during an unstable branch.
*/
filp->f_flags |= O_LARGEFILE;
- filp->f_mode |= FMODE_BUF_RASYNC;
+ filp->f_mode |= FMODE_BUF_RASYNC | FMODE_NOWAIT;
/*
* Use the file private data to store the holder for exclusive openes.
@@ -519,9 +519,6 @@ static int blkdev_open(struct inode *inode, struct file *filp)
if (IS_ERR(bdev))
return PTR_ERR(bdev);
- if (bdev_nowait(bdev))
- filp->f_mode |= FMODE_NOWAIT;
-
filp->f_mapping = bdev->bd_inode->i_mapping;
filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
return 0;
@@ -563,6 +560,9 @@ static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
return -EOPNOTSUPP;
+ if ((iocb->ki_flags & IOCB_NOWAIT) && !bdev_nowait(bdev))
+ return -EAGAIN;
+
size -= iocb->ki_pos;
if (iov_iter_count(from) > size) {
shorted = iov_iter_count(from) - size;
@@ -585,6 +585,9 @@ static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
ssize_t ret = 0;
size_t count;
+ if ((iocb->ki_flags & IOCB_NOWAIT) && !bdev_nowait(bdev))
+ return -EAGAIN;
+
if (unlikely(pos + iov_iter_count(to) > size)) {
if (pos >= size)
return 0;
--
2.39.2
