[PATCH] src/min_dio_alignment: don't return invalid dio_offset_align

[PATCH] src/min_dio_alignment: don't return invalid dio_offset_align

[Chao Yu]

If returned parameters of statx() are: a)STATX_DIOALIGN is set in
stx_mask, b)stx.stx_dio_offset_align is zero, it indicates filesystem
supports DIO, but the file doesn't.

It needs to avoid returning zeroed stx.stx_dio_offset_align value,
instead, we can fallthrough to get alignment size of block device or
page size, otherwise, it may cause potential deadloop, e.g.
generic/465:

align=stx_dio_offset_align(it equals to zero)
page_size=4096
while [ $align -le $page_size ]; do
    echo "$AIO_TEST -a $align -d $testfile.$align" >> $seqres.full
    $AIO_TEST -a $align -d $testfile.$align 2>&1 | tee -a $seqres.full
    align=$((align * 2))
done

Cc: Christoph Hellwig <hch@lst.de>
Cc: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Chao Yu <chao@kernel.org>
---
 src/min_dio_alignment.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/min_dio_alignment.c b/src/min_dio_alignment.c
index 131f6023..c1d7112d 100644
--- a/src/min_dio_alignment.c
+++ b/src/min_dio_alignment.c
@@ -28,7 +28,7 @@ static int min_dio_alignment(const char *mntpnt, const char *devname)
 	fd = open(mntpnt, O_TMPFILE | O_RDWR | O_EXCL, 0600);
 	if (fd >= 0 &&
 	    xfstests_statx(fd, "", AT_EMPTY_PATH, STATX_DIOALIGN, &stx) == 0 &&
-	    (stx.stx_mask & STATX_DIOALIGN))
+	    (stx.stx_mask & STATX_DIOALIGN) && stx.stx_dio_offset_align != 0)
 		return stx.stx_dio_offset_align;
 
 	/*
-- 
2.40.1

Re: [PATCH] src/min_dio_alignment: don't return invalid dio_offset_align

[Christoph Hellwig]

Looks good:

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

although one day we might want to do something smarted with this
information.  That being said a file system not supporting direct
I/O is a bit silly these days.  What is running into this?

Re: [PATCH] src/min_dio_alignment: don't return invalid dio_offset_align

[Chao Yu]

On 2024/9/11 15:46, Christoph Hellwig wrote:
> although one day we might want to do something smarted with this
> information.  That being said a file system not supporting direct
> I/O is a bit silly these days.  What is running into this?

After applying commit 5c8764f8679e ("f2fs: fix to force buffered IO
on inline_data inode"), inline_data enabled tmpfile shows it doesn't
support DIO, then it results in deadloop of generic/465.