When XFS underlying disk fails, it could take several milliseconds for the
FS to be marked
shutdown. xfs_buf_iodone_callbacks() retries buf upon first failure by
submitting it once
again. But if the buf fails 2nd time before FS is marked for shutdown, it
just releases the
buf with xfs_buf_relse(). This is flawed that nobody is releasing the
XFS_IFLOCK on the inode.
Because of this AIL tasks repeated effort to xfs_inode_item_push() will
see that xfs_iflock()
cannot be acquired. This blocks XFS from being unmounted as
xfs_ail_push_all_sync() will keep
looping without progress.
Fix this by marking the FS for shutdown if we have a permanent failure &
resubmit the buf.
xfs_buf_submit() will see FS marked for shutdown & invoke the callback
which releases
XFS_IFLOCK.
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index 1a6c9b9..6f73ee0 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -1100,7 +1100,12 @@ xfs_buf_iodone_callbacks(
XBF_DONE | XBF_WRITE_FAIL;
xfs_buf_submit(bp);
} else {
- xfs_buf_relse(bp);
+ /*
+ * if we have the buf fail 2nd time, force a FS
shutdown & resubmit
+ * the buf for it to be failed back immediately
+ */
+ xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
+ xfs_buf_submit(bp);
}
return;
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
