Bug#512475: mdadm fails to reassemble arrays stopped mid-reshape

[Will Glynn]

Package: mdadm
Version: 2.6.7.1-1
Severity: grave
Justification: causes non-serious data loss

When an array is grown and the grow is interrupted, mdadm is unable to  
reassemble the array, preventing the reshape from completing and  
denying access to the stored data.

I had a 5-disk RAID 6 array that I wanted to grow, so:

# mdadm --add /dev/md2 /dev/sdg2
# mdadm --grow /dev/md2 --raid-devices=6

The reshape began and passed the critical section without incident.  
Later, I needed to shut off the system. It was unbootable when I  
returned. (/dev/md2 is used with LVM and contains the root  
filesystem.) Booting into a Lenny RC1 rescue console, mdadm would  
refuse to assemble md2, saying "superblock on <first component>  
doesn't match others".

I believe this is fixed by a particular upstream commit, already  
included in 2.6.8:
   6e9eac4f089a44e9091967a3e6198aeae7e845a3

Regardless, I compiled a 2.6.8 binary and copied it to to the rescue  
environment. The same assemble command succeed with the new build, and  
the array resumed the reshape operation from where it left off.