[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Installer doesn't seem to allow partitioning of /dev/mdX



I am raising this here for discussion as to whether this is intentional behaviour or not.

I am installing Debian Wheezy on a system with two 1TB hard drives, configured in the BIOS as a RAID1 pair and a single volume, "Volume0". This motherboard uses IMSM metadata.

Thanks to http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=684708 this is now detected correctly using mdadm instead of dmraid:

* /dev/md127 (also /dev/md/imsm0) is the container
* /dev/md126 (also /dev/md/Volume0) is the data

and md126 starts resyncing immediately.

However the installer does not let me partition this. It shows:

RAID1 device #126 - 950.2 GB Software RAID device
    #1     950.2GB
             512.0 B     unusable

I can select #1, I can nominate what kind of filesystem it is (or "do not use"), I can wipe it with random data, but I cannot delete #1 nor create a partition within it.

What I wanted was to create:

/dev/md126p1  - for /boot filesystem
/dev/md126p2 - for alternate /boot filesystem (e.g. installing future version of Linux)
/dev/md126p3  - for LVM

I know from previous experience that md volumes can be partitioned in this way, but the installer doesn't appear to let me.

Workaround: go back into the BIOS and create distinct volumes within the IMSM container.

This should be fine for my usage case, but if instead of using IMSM I had wanted to create a simple /dev/md127 using native metadata, and then partition it, I think I would be stuck.

Workaround 2: you can partition drives and then mirror the partitions as different /dev/mdX devices. But this means when you come to swap a drive, you cannot simply replace the failed drive: you must first prepare the new one by manually partitioning it in the same way as its partner, and then you must re-add the mirror pairs individually.

Has this been discussed before? Is partitioning of md devices intentionally blocked in the installer?

Thanks,

Brian Candler.


Reply to: