Update: chip suggests dmraid -r -D in rescue mode. I haven't tried it -- wiping both disks with dd didn't help, but booting one disk, installing, and then adding the second disk gave me a usable install.

My latest Linux problem is odd, but at least somewhat interesting. I'm installing CentOS 5.3 onto an old PowerEdge 1950 with a couple 750gb SATA disks. My kickstart configuration failed, complaining that the system couldn't repartition sda. I tried hda, with no more luck. Confusingly, I do see hda when booting, but it looks like that's the CD-ROM drive, and anaconda skips it for partitioning.

anaconda: error -- cannot find physical disk

When I ran anaconda (the Red Hat / CentOS installer) manually, it showed me only a single device: /dev/mapper/ddf1_Virtual Disk 0. This is clearly some sort of logical device -- LVM, software RAID, or something like that. I booted into the Dell SAS BIOS and confirmed it wasn't presenting logical devices -- it's not a PERC controller, and doesn't appear to support hardware RAID at all.

anaconda: cannot manage LVM volume

I booted into linux rescue mode, and the system found and mounted /dev/mapper/ddf1_Virtual Disk 0p5 and various associated partitions, but I was able to destroy and recreate the partition tables with fdisk.

linux rescue mode: logical and physical disks

I filed a bug against CentOS (since I haven't tried to reproduce this in RHEL, although I'm pretty sure it's their issue).

Unfortunately, when I booted back into anaconda, the virtual disk remained the only available device. Someone on #centos suggested that this was an LVM volume, and the system was finding LVM superblocks scattered across the disk (so not dependent on the partition table). Next attempt is ongoing right now (writing 12,002,501,984,256 zeroes takes a while):

  • time dd if=/dev/zero of=/dev/sda bs=1G
  • time dd if=/dev/zero of=/dev/sdb bs=1G