Re: [OSR-devel] Remounting New Root After Booting

[Gordan B. <go...@bo...>]

On Thu, 11 Oct 2007, Marc Grimme wrote:

>>>>> So when the initrd gets the iscsi disk as scsi-disk all the rest should
>>>>> be done automatically.
>>>>
>>>> Where is the template startup script for the initrd? A few things
>>>> require a bit of manual intervention, because for some reason iscsid
>>>> doesn't trigger loading of iscsi_tcp module which is required _before_
>>>> iscsid loads.
>>>
>>> linuxrc.generic.sh is the main init-script.
>>
>> Aha! I see now what all the talk about iscsi-lib.sh was. This is what gets
>> the parameters from the cluster.conf about the root source on iscsi stuff.
>>
>> But what can this do without the iscsi tools being included in the
>> standard initrd?
>>
>> On a separte note, is this "old" iSCSI speciffic (i.e. pre-Open-iSCSI, as
>> i see mentions of Cisco), or is it just a convenient "dump your iSCSI
>> initialisation stuff here" script?
>
> You can ignore or use anything in iscsi-lib.sh. It is very old. And long ago
> RHEL3 and I can't remember much of it. As it was only a "study" to prove it
> works.

Wow! This goes back as far as RHEL3? I'm impressed. :-)

>>>> And, obviously, I'd need to make all this happen before the normal boot
>>>> process tries to mount the root disk.
>>>
>>> Do it before or right after scsi-start is called.
>>
>> I can't find any reference to "scsi-start". Where does this happen?
>
> It's scsi_start in linux.generic.sh Line 239.

D'oh! What a difference an _ makes.

I think the iSCSI stuff should be added before the dm_start stuff, which 
is just before the scsi_start.

Right, that seems to have worked. Almost there. Now I get:

Mounting LABEL=/ on /mnt/newroot..device not found error [FAILED]
Mounting /cdsl.local on /cluster/cdsl/3..no cdsldir found 
"/mnt/mewroot//cluster/cdsl"
Could not mount cdsl /cdsl.local to /cluster/cdsl/3. Exiting

Where does it get the LABEL=/ from? I'm pretty sure I didn't set any.

>>>> Or will specifying the iscsi stuff in cluster.conf make all this work?
>>>> Given that the iscsi components weren't in the standard initrd, I'm not
>>>> too convinced this will work (although there's a distinct possibility
>>>> that this is totally unrelated and irrelevant in this case).
>>>
>>> Yes I agree. Make it work for you and we'll integrate it so it'll only be
>>> called if iscsi is selected.
>>
>> LOL! I must say I'm not used to getting integrated into the development
>> effort this quickly. Not that I don't mind. :-)
>
> We'll integrate it in preview and tag it as experimental. But as iscsi
> integration is on our list but only in next major version we are happy to use
> anything that is provided by others.

It surprises me that the whole concept didn't start with iSCSI. How else 
can you provide a shared GFS root for a cluster? NBDs? Shared SCSI bus (2 
controllers in 2 machines with the drives shared on the chain)?

Gordan