[linux-lvm] LVM non-bug: md10 and higher broken

[William Blunn]

Doesn't seem to be a bug in LVM, but it seems that the problem is
tickled when migrating from a multiple-MD-devices system to an LVM
system.

LVM 1.0.1 / kernel 2.2.20

I am using /dev/md0 through /dev/md10 for my existing (non-LVM) system.

I put in additional disks, set them up on /dev/md11, and created a PV on
/dev/md11.

When trying to create the VG, it wouldn't find the PV.

Suspiciously, /proc/partitions shows MD devices listed as

   ...
   /dev/md8
   /dev/md9
   /dev/md:
   /dev/md;

instead of what you would expect

   ...
   /dev/md8
   /dev/md9
   /dev/md10
   /dev/md11

i.e. (sloppily) implemented as translating the MD number into a single
character instead of generating a (potentially multiple character)
string of the decimal representation.

Swapping /dev/md0 with /dev/md11 makes it all work.

So I guess that LVM is accessing /proc/partitions and not seeing any MD
devices 10 or higher.

That would make it a bug in the kernel, but since people are likely to
come across it through LVM I thought it would be a good idea to mention
it here.

Does anyone know if anyone has made/submitted a (2.2) kernel fix for
this?

Another little snag is that the code for handling this is right next to
where the LVM patch gets applied, so making a patch would likely (I
guess) make the LVM patch not apply properly.

Bill