Currently xfs_buf_iodone_callbacks() keep retrying forever to submit
a buffer IO, and rely on another part of the filesystem to shutdown
it in case of an IO error.
But, if before some other work happens, the filesystem is unmounted,
it get stuck forever, retrying to submit buffer IOs, and the umount
command will stuck in D state waiting for it.
Also, no extra action from the user can be taken in this situation, since
here, we already removed the mount entry from the namespace, so, not
even xfs_io "shutdown" command can be used.
To fix this, I created a flag to notify any other filesystem task, that it
is being unmounted, so it can be used to stop retrying to submit buffer IOs
to a failed device, in case of the filesystem is being umounted.
I'm not sure if this is a good solution, or if it's more a hack, but I
do believe we need to have some solution like this, otherwise the user can
hit a lockup situation. And sending the patch might be a good start to discuss
about a solution for this.
Signed-off-by: Carlos Maiolino <cmaiolino@xxxxxxxxxx>
---
fs/xfs/xfs_buf_item.c | 9 +++++++++
fs/xfs/xfs_mount.c | 3 +++
fs/xfs/xfs_mount.h | 2 ++
3 files changed, 14 insertions(+)
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index 99e91a0..5cd9c6c 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -1076,6 +1076,15 @@ xfs_buf_iodone_callbacks(
time_after(jiffies, (lasttime + 5*HZ))) {
lasttime = jiffies;
xfs_buf_ioerror_alert(bp, __func__);
+
+ /*
+ * If we have a metadata error, and user tried to unmount
+ * the filesystem, stop retrying to submit buffers forever
+ * and shutdown the filesystem here, instead of rely in
+ * another layer to do so.
+ */
+ if (mp->m_flags & XFS_MOUNT_FS_UNMOUNT)
+ xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
}
lasttarg = bp->b_target;
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index 536a0ee..5b740df 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -987,6 +987,9 @@ xfs_unmountfs(
__uint64_t resblks;
int error;
+ /* filesystem is being unmounted */
+ mp->m_flags |= XFS_MOUNT_FS_UNMOUNT;
+
cancel_delayed_work_sync(&mp->m_eofblocks_work);
xfs_qm_unmount_quotas(mp);
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index bac6b34..562e5fc 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -191,6 +191,8 @@ typedef struct xfs_mount {
#define XFS_MOUNT_FILESTREAMS (1ULL << 24) /* enable the filestreams
allocator */
#define XFS_MOUNT_NOATTR2 (1ULL << 25) /* disable use of attr2 format */
+#define XFS_MOUNT_FS_UNMOUNT (1ULL << 26) /* Notify filesystem is being
+ unmounted */
#define XFS_MOUNT_DAX (1ULL << 62) /* TEST ONLY! */
--
2.4.3
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
