Re: [PATCH 2/6] lvm: Ensure LVM is initialised in non-hostonly mode.

[Colin Guthrie <colin-iHtYJnzJly4@public.gmane.org>]

'Twas brillig, and Harald Hoyer at 24/01/13 11:32 did gyre and gimble:
> Am 23.01.2013 21:15, schrieb Colin Guthrie:
>> This is a bit of a hack, but it's needed if /usr is on LVM.
>> ---
>>  modules.d/90lvm/module-setup.sh          | 3 +++
>>  modules.d/90lvm/non-hostonly-init-lvm.sh | 8 ++++++++
>>  2 files changed, 11 insertions(+)
>>  create mode 100644 modules.d/90lvm/non-hostonly-init-lvm.sh
>>
>> diff --git a/modules.d/90lvm/module-setup.sh b/modules.d/90lvm/module-setup.sh
>> index a537b1a..611379a 100755
>> --- a/modules.d/90lvm/module-setup.sh
>> +++ b/modules.d/90lvm/module-setup.sh
>> @@ -49,6 +49,9 @@ install() {
>>              sed -i -e 's/\(^[[:space:]]*\)use_lvmetad[[:space:]]*=[[:space:]]*[[:digit:]]/\1use_lvmetad = 0/' ${initdir}/etc/lvm/lvm.conf
>>          fi
>>      fi
>> +    if ! [[ $hostonly ]]; then
>> +        inst_hook pre-udev 30 "$moddir/non-hostonly-init-lvm.sh"
>> +    fi
>>  
>>      inst_rules 11-dm-lvm.rules
>>      # Gentoo ebuild for LVM2 prior to 2.02.63-r1 doesn't install above rules
>> diff --git a/modules.d/90lvm/non-hostonly-init-lvm.sh b/modules.d/90lvm/non-hostonly-init-lvm.sh
>> new file mode 100644
>> index 0000000..2e9212c
>> --- /dev/null
>> +++ b/modules.d/90lvm/non-hostonly-init-lvm.sh
>> @@ -0,0 +1,8 @@
>> +#!/bin/sh
>> +# -*- mode: shell-script; indent-tabs-mode: nil; sh-basic-offset: 4; -*-
>> +# ex: ts=8 sw=4 sts=4 et filetype=sh
>> +
>> +printf 'cancel_wait_for_dev /dev/non-hostonly-lvm; rm "$job";' \
>> +    "$hookdir" >> $hookdir/initqueue/timeout/non-hostonly-lvm.sh
>> +
>> +wait_for_dev "/dev/non-hostonly-lvm"
>>
> 
> huh? what does this try to do? This only ensures, that the timeout is reached.

Well..... it's been a while, so my memory is fuzzy, but this was meant
to "solve" the problem of a non-hostonly initrd (i.e. one without
etc/cmdline.d/90lvm.conf), but where the system has a /usr on LVM but /
on a normal partition. i.e. it should initialise all volume groups it
finds in order to mount /usr.

In that setup I don't think lvm was initialised properly because the
initqueue loop was never entered at all and thus nothing actually
triggered the lvm vgchange.

So all this does is ensure the loop is run at least once, which triggers
the LVM init.

I'm going from memory here so a bit hazy - the explanation might not be
totally bang on. I'll see if I can test a non-hostonly initrd with that
kind of disk layout in my VMs to confirm it's still needed as I guess
there is a chance the "never run the initqueue" issue has been factored
out since 017 which is when I think I wrote this :)

Cheers

Col

-- 

Colin Guthrie
  http://colin.guthr.ie/