[lug] more RAID0 on /

[D. Stimits]

David Morris wrote:
> On Fri, Nov 19, 2004 at 08:24:30PM -0700, D. Stimits wrote:
> 
>>Unless there is something in init to raidstart autodetect that is itself 
>>initially stored and loaded from "/", I can't see a reason why this 
>>would fail. Hints or clues? Do I have to custom build initrd differently 
>>when root partition is RAID0 in order to raidstart despite autodetect?
> 
> 
> I've setup several systems to use RAID5 for the root partition
> and RIAD1 for the /boot partition.  I've found raid
> autodetect is a great idea in theory, but on some systems it
> just doean't work.  Rather than bother with fighting
> autodetect, I usually create a custom initrd which calls
> raidstart before mounting root and doing the pivot.
> 
> YMMV, its been about two years since I did the last system
> like this but I don't think much has changed.
> 
> If you're interested, I can try and dig up my initrd
> creation instructions....

I've been playing with a custom initrd but have been unsuccessful. The 
reason why is that I need a statically linked raidstart, which is next 
to impossible. The chicken-and-egg dilemma is: Since raidstart is in 
/sbin/, which in turn is on the md0 device, how the heck do I run 
raidstart to activate the root partition with /sbin/? If raidstart were 
statically linked then this wouldn't be a huge problem, I could for 
example put it in /boot/ or in the initrd, but a dynamically linked 
raidstart is not so easy to work with. Making a statically linked 
version available then making minor edits to linuxrc would do wonders.

Did your custom initrd's have raidstart in them? Or was it just kernel 
module support with linuxrc updates?

That leads me to yet another thought...perhaps the loading of raid0 
modules triggers the kernel to attempt autodetect even without 
raidstart...I'm starting to wonder if making it a module in initrd 
rather than statically built into the kernel would trigger "nice things" 
to happen.

D. Stimits, stimits AT comcast DOT net