This is an attempt to parallelize recovery. A new structure
ocfs2_recovery_node manages the recovery and replaces the recovery
map. More details in individual patches.
Changes:
- Use a pre-allocated list to avoid ENOMEM
- Use work queue to offload recovery procedure