Description of problem:
In dm-exception-store.c, when a metatdata chunk is filled in, the code firstly write to disk the current metadata chunk and then zero out the next metadata chunk. This could end up with a inconsistent exception data structure on disk if there is a power outage / kernel crash coming in between the two writes.
The bug has actually been fixed in the current mainstream kernel by Mikulas Patocka. See the following link:
http://lkml.indiana.edu/hypermail/linux/kernel/0810.3/0020.html
But I am not sure if it's going to be in the next RHEL4/RHEL5 updates. So here I am requesting to have this included in the next updates.
Version-Release number of selected component (if applicable):
It exists on all the stable RHEL4/RHEL5 releases
How reproducible:
It's hard to reproduce in real life, but we can certainly simulate this by manually put a "die" in between the two writes.
Additional info: