It's safest to copy the data off and recreate the filesystem. The domain error in mirror-read reflects the CRC failure that checkmap found. The mirroring code is exiting on that error so you probably don't have a complete mirror copy. I would mount the filesystem and use cpdup to make the copy and see what errors (if any) cpdup reports.
-Matt On Sat, Jan 31, 2015 at 3:57 AM, Vasily Postnicov <[email protected]> wrote: > Hello. > > I have a possibly damaged HAMMER fs. Look at output of hammer checkmap > command: > > ressurected# hammer -f > /dev/serno/5VPD5Y9X.s1:/dev/serno/6PT5CW16.s1:/dev/serno/WD-WMASY1750086.s1d > checkmap > Volume header records=0 next_tid=0000000279fe6800 > bufoffset=0000000044040000 > Collecting allocation info from B-Tree: B NODE 80000005221c3000 cnt=32 > p=c000000522191000 type=L depth=4 mirror 0000000278289c40 { > done > > I think this was caused by damaged hardware (overheated CPU). > > I do not see any lost/inconsistent data yet, only hammer mirror-read > command fails with one of PFSes: > > ressurected# hammer mirror-read /home/shared > /dev/zero > Prescan to break up bulk transfer > Mirror-read /home/shared failed: Numerical argument out of domain > > Can I lose data if I continue to use this fs? May it be cured (e.g. by > copying data from /home/shared to newly created PFS and then destroying the > old one)? > > With best regards, Vasily. >
