[PATCH v2] xfs: fix log recovery erroring out on refcount recovery failure

[PATCH v2] xfs: fix log recovery erroring out on refcount recovery failure

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Per the comment in the error case of xfs_reflink_recover_cow, zero out
any error (after shutting down the log) so that we actually kill any new
intent items that might have gotten logged by later recovery steps.
Discovered by xfs/434, which few people actually seem to run.

Fixes: 2c1e31ed5c88 ("xfs: place intent recovery under NOFS allocation context")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
---
v2: actually send it to the list this time
---
 fs/xfs/xfs_log_recover.c |    1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c
index cd134830a695..13f1d2e91540 100644
--- a/fs/xfs/xfs_log_recover.c
+++ b/fs/xfs/xfs_log_recover.c
@@ -3532,6 +3532,7 @@ xlog_recover_finish(
 		 * and AIL.
 		 */
 		xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
+		error = 0;
 		goto out_error;
 	}
 

Re: [PATCH v2] xfs: fix log recovery erroring out on refcount recovery failure

[Christoph Hellwig]

On Thu, Feb 22, 2024 at 09:48:17PM -0800, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> Per the comment in the error case of xfs_reflink_recover_cow, zero out
> any error (after shutting down the log) so that we actually kill any new
> intent items that might have gotten logged by later recovery steps.
> Discovered by xfs/434, which few people actually seem to run.

Talking about xfs/434, it just caused me some nasty debugging headache
last week.  It is the first of only three tests requiring a loadable
xfs module, and when there is an xfs.ko for the uname around but xfs
is built into the kernel as it is for my usual test configs it fails
in completely weird and hard to debug ways.

The fact that it requires a modular xfs might also be the reason why
it's rarely run.

The fix itself looks good:

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