Red Hat Bugzilla – Bug 316031
dm-mirror: incorrect order of mirror presuspend ops causes cluster mirror hang
Last modified: 2007-11-16 20:14:56 EST
Cluster mirrors do not (and can not) allow writes to mirror regions which have
not recovered yet.
When a mirror suspends it:
1) stops recovery
2) flushes the work queue (finishing outstanding writes)
3) calls log presuspend
The log has no way of knowing that the mirror is suspending until #3, so it
blocks any outstanding writes to regions that are not-in-sync. This could hang
the mirror because recovery has been stopped and the not-in-sync regions will
never be cleared - delaying the writes indefinitely.
The solution is to switch the order of operations to:
1) stop recovery
2) call log presuspend
3) flush work queue
By doing this, the log knows that recovery has been stopped and that there will
be no collisions with writes. Therefore, it can allow writes to regions which
have not yet been recovered.
committed in stream U6 build 61. A test kernel with this patch is available from
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.