[gentoo-user] Run command after root mounted ro?

[gentoo-user] Run command after root mounted ro?

[Daniel Frey]

Greetings,

I discovered a possible reason for mdadm always marking my raid array as
bad on reboot. The problem is I don't know how to check for the issue.

When using root on a native mdadm raid, the kernel handles the array
state transition at shutdown, so it's not a big deal. However, it's
different when using external metadata - mdadm must be called before the
reboot.

Apparently mdadm must be called after the root filesystem is mounted
read-only. Then it can mark the array as clean and not rebuild every
single reboot.

Where do I check this? The /etc/init.d/mdadm script happens well before
root is mounted read-only.

Apparently `mdadm --wait-clean --scan` needs to be called so the
external metadata is updated correctly. If you use this command with a
native mdadm raid it does nothing, but it's critical apparently to the
health of other metadata types.

I've grepped through /etc/init.d/ and found nothing.

Dan

Re: [gentoo-user] Run command after root mounted ro?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 1673 bytes --]

Am 09.08.2011 21:30, schrieb Daniel Frey:
> Greetings,
> 
> I discovered a possible reason for mdadm always marking my raid array as
> bad on reboot. The problem is I don't know how to check for the issue.
> 
> When using root on a native mdadm raid, the kernel handles the array
> state transition at shutdown, so it's not a big deal. However, it's
> different when using external metadata - mdadm must be called before the
> reboot.
> 
> Apparently mdadm must be called after the root filesystem is mounted
> read-only. Then it can mark the array as clean and not rebuild every
> single reboot.
> 
> Where do I check this? The /etc/init.d/mdadm script happens well before
> root is mounted read-only.
> 
> Apparently `mdadm --wait-clean --scan` needs to be called so the
> external metadata is updated correctly. If you use this command with a
> native mdadm raid it does nothing, but it's critical apparently to the
> health of other metadata types.
> 
> I've grepped through /etc/init.d/ and found nothing.
> 
> Dan
> 

Remounting root read-only is done by an init script called mount-ro
which is started in runlevel shutdown. Try to add a custom init script
to your /etc/init.d directory with the following content:

#!/sbin/runscript
depend()
{
 after mount-ro
}
start()
{
 ebegin 'Shutting down mdadm'
 mdadm --wait-clean --scan
 eend $?
}

Add it to the runlevel with `rc-update add <your-script> shutdown` and
don't forget to mark it executable.

Disclaimer: I've not tried this (obviously) and if the script eats your
dog and wreaks your system, it is entirely your fault ;)

Hope this helps,
Florian Philipp


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: Re: [gentoo-user] Run command after root mounted ro?

[Daniel Frey]

On 01/-10/37 11:59, Florian Philipp wrote:
> Remounting root read-only is done by an init script called mount-ro
> which is started in runlevel shutdown. Try to add a custom init script
> to your /etc/init.d directory with the following content:
> 
> #!/sbin/runscript
> depend()
> {
>  after mount-ro
> }
> start()
> {
>  ebegin 'Shutting down mdadm'
>  mdadm --wait-clean --scan
>  eend $?
> }
> 
> Add it to the runlevel with `rc-update add <your-script> shutdown` and
> don't forget to mark it executable.
> 
> Disclaimer: I've not tried this (obviously) and if the script eats your
> dog and wreaks your system, it is entirely your fault ;)


Thanks, I tried this out, and while it does run after mounting ro, it
just hangs. I've noticed that it's supposed to be monitoring
/proc/mdstat but it was (presumably) unmounted long ago.

I guess I have to do some more experimenting.

Dan

Re: [gentoo-user] Run command after root mounted ro?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 1228 bytes --]

Am 10.08.2011 03:04, schrieb Daniel Frey:
> On 01/-10/37 11:59, Florian Philipp wrote:
>> Remounting root read-only is done by an init script called mount-ro
>> which is started in runlevel shutdown. Try to add a custom init script
>> to your /etc/init.d directory with the following content:
>>
>> #!/sbin/runscript
>> depend()
>> {
>>  after mount-ro
>> }
>> start()
>> {
>>  ebegin 'Shutting down mdadm'
>>  mdadm --wait-clean --scan
>>  eend $?
>> }
>>
>> Add it to the runlevel with `rc-update add <your-script> shutdown` and
>> don't forget to mark it executable.
>>
>> Disclaimer: I've not tried this (obviously) and if the script eats your
>> dog and wreaks your system, it is entirely your fault ;)
> 
> 
> Thanks, I tried this out, and while it does run after mounting ro, it
> just hangs. I've noticed that it's supposed to be monitoring
> /proc/mdstat but it was (presumably) unmounted long ago.
> 
> I guess I have to do some more experimenting.
> 
> Dan
> 
> 
> 

Try remounting proc in your new init script. You have to advice mount to
not try adding it to /etc/mtab since that would obviously fail.
mount --no-mtab -t proc proc /proc

Hope this helps,
Florian Philipp


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]