Hmmm, this is the second time in 1 year that I had a corrupt filesystem on my raid partition. I saw no other option than mke2fs -j /dev/md7.

My first problem with XFS started on my fileserver, which only had 256 MB of memory. Turns out XFS was OOM-ing inside the kernel, this in turned messed up the filesystem. xfs_repair was also running out of memory. When I finally got enough virtual memory in this box xfs_repair was kind enough to segfault. Ok, shit happens, mkfs.xfs and try again.

I’ve moved the raid services off this under powered box and put the disks in my normal server with 2 GB of memory. But today I wanted to remove some old backups which heavily uses hardlinks. my guess is that there are a few million on that (backup/raid) partition alone. This failed, it started to spew out errors after having ran for an hour.

OK.

I killed the rm, umounted the filesystem, mounted in again so that XFS could recheck it. This worked; i.e. the filesystem was still OK. Next I thought it would be a good idea to run xfs_repair again on it to double check.

Then I went to work.

After coming home again, it still wasn’t finished. OK. I got my USB hard disk out, cp-ed the stuff over and:

mke2fs -j /dev/md7