[PATCH] xfs: restore shutdown check in mapped write fault path

[PATCH] xfs: restore shutdown check in mapped write fault path

[Brian Foster]

XFS triggers an iomap warning in the write fault path due to a
!PageUptodate() page if a write fault happens to occur on a page
that recently failed writeback. The iomap writeback error handling
code can clear the Uptodate flag if no portion of the page is
submitted for I/O. This is reproduced by fstest generic/019, which
combines various forms of I/O with simulated disk failures that
inevitably lead to filesystem shutdown (which then unconditionally
fails page writeback).

This is a regression introduced by commit f150b4234397 ("xfs: split
the iomap ops for buffered vs direct writes") due to the removal of
a shutdown check and explicit error return in the ->iomap_begin()
path used by the write fault path. The explicit error return
historically translated to a SIGBUS, but now carries on with iomap
processing where it complains about the unexpected state. Restore
the shutdown check to xfs_buffered_write_iomap_begin() to restore
historical behavior.

Fixes: f150b4234397 ("xfs: split the iomap ops for buffered vs direct writes")
Signed-off-by: Brian Foster <bfoster@redhat.com>
---
 fs/xfs/xfs_iomap.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 70c341658c01..6594f572096e 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -860,6 +860,9 @@ xfs_buffered_write_iomap_begin(
 	int			allocfork = XFS_DATA_FORK;
 	int			error = 0;
 
+	if (XFS_FORCED_SHUTDOWN(mp))
+		return -EIO;
+
 	/* we can't use delayed allocations when using extent size hints */
 	if (xfs_get_extsz_hint(ip))
 		return xfs_direct_write_iomap_begin(inode, offset, count,
-- 
2.26.2

Re: [PATCH] xfs: restore shutdown check in mapped write fault path

[Eric Sandeen]

On 2/10/21 11:01 AM, Brian Foster wrote:
> XFS triggers an iomap warning in the write fault path due to a
> !PageUptodate() page if a write fault happens to occur on a page
> that recently failed writeback. The iomap writeback error handling
> code can clear the Uptodate flag if no portion of the page is
> submitted for I/O. This is reproduced by fstest generic/019, which
> combines various forms of I/O with simulated disk failures that
> inevitably lead to filesystem shutdown (which then unconditionally
> fails page writeback).
> 
> This is a regression introduced by commit f150b4234397 ("xfs: split
> the iomap ops for buffered vs direct writes") due to the removal of
> a shutdown check and explicit error return in the ->iomap_begin()
> path used by the write fault path. The explicit error return
> historically translated to a SIGBUS, but now carries on with iomap
> processing where it complains about the unexpected state. Restore
> the shutdown check to xfs_buffered_write_iomap_begin() to restore
> historical behavior.
> 
> Fixes: f150b4234397 ("xfs: split the iomap ops for buffered vs direct writes")
> Signed-off-by: Brian Foster <bfoster@redhat.com>

Looks good to me

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

> ---
>  fs/xfs/xfs_iomap.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index 70c341658c01..6594f572096e 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -860,6 +860,9 @@ xfs_buffered_write_iomap_begin(
>  	int			allocfork = XFS_DATA_FORK;
>  	int			error = 0;
>  
> +	if (XFS_FORCED_SHUTDOWN(mp))
> +		return -EIO;
> +
>  	/* we can't use delayed allocations when using extent size hints */
>  	if (xfs_get_extsz_hint(ip))
>  		return xfs_direct_write_iomap_begin(inode, offset, count,
> 

Re: [PATCH] xfs: restore shutdown check in mapped write fault path

[Darrick J. Wong]

On Wed, Feb 10, 2021 at 12:01:12PM -0500, Brian Foster wrote:
> XFS triggers an iomap warning in the write fault path due to a
> !PageUptodate() page if a write fault happens to occur on a page
> that recently failed writeback. The iomap writeback error handling
> code can clear the Uptodate flag if no portion of the page is
> submitted for I/O. This is reproduced by fstest generic/019, which
> combines various forms of I/O with simulated disk failures that
> inevitably lead to filesystem shutdown (which then unconditionally
> fails page writeback).
> 
> This is a regression introduced by commit f150b4234397 ("xfs: split
> the iomap ops for buffered vs direct writes") due to the removal of
> a shutdown check and explicit error return in the ->iomap_begin()
> path used by the write fault path. The explicit error return
> historically translated to a SIGBUS, but now carries on with iomap
> processing where it complains about the unexpected state. Restore
> the shutdown check to xfs_buffered_write_iomap_begin() to restore
> historical behavior.
> 
> Fixes: f150b4234397 ("xfs: split the iomap ops for buffered vs direct writes")
> Signed-off-by: Brian Foster <bfoster@redhat.com>

Looks fine,
Reviewed-by: Darrick J. Wong <djwong@kernel.org>

--D

> ---
>  fs/xfs/xfs_iomap.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index 70c341658c01..6594f572096e 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -860,6 +860,9 @@ xfs_buffered_write_iomap_begin(
>  	int			allocfork = XFS_DATA_FORK;
>  	int			error = 0;
>  
> +	if (XFS_FORCED_SHUTDOWN(mp))
> +		return -EIO;
> +
>  	/* we can't use delayed allocations when using extent size hints */
>  	if (xfs_get_extsz_hint(ip))
>  		return xfs_direct_write_iomap_begin(inode, offset, count,
> -- 
> 2.26.2
> 

Re: [PATCH] xfs: restore shutdown check in mapped write fault path

[Gao Xiang]

On Wed, Feb 10, 2021 at 12:01:12PM -0500, Brian Foster wrote:
> XFS triggers an iomap warning in the write fault path due to a
> !PageUptodate() page if a write fault happens to occur on a page
> that recently failed writeback. The iomap writeback error handling
> code can clear the Uptodate flag if no portion of the page is
> submitted for I/O. This is reproduced by fstest generic/019, which
> combines various forms of I/O with simulated disk failures that
> inevitably lead to filesystem shutdown (which then unconditionally
> fails page writeback).
> 
> This is a regression introduced by commit f150b4234397 ("xfs: split
> the iomap ops for buffered vs direct writes") due to the removal of
> a shutdown check and explicit error return in the ->iomap_begin()
> path used by the write fault path. The explicit error return
> historically translated to a SIGBUS, but now carries on with iomap
> processing where it complains about the unexpected state. Restore
> the shutdown check to xfs_buffered_write_iomap_begin() to restore
> historical behavior.
> 
> Fixes: f150b4234397 ("xfs: split the iomap ops for buffered vs direct writes")
> Signed-off-by: Brian Foster <bfoster@redhat.com>

Taking some time on the internal discussion thread, I also think
SIGBUS is also appropriate after fs shutdown rather than dirty
this page anyway.

Reviewed-by: Gao Xiang <hsiangkao@redhat.com>

Thanks,
Gao Xiang