Subject: Re: RAID0 (I know, I know) reconstruction on another drive pair.
To: None <"Undisclosed.Recipients">
From: Marc Tooley <netbsdMLpostNO@spam.quake.ca>
List: netbsd-help
Date: 08/25/2005 16:45:05
On Thursday 25 August 2005 14:54, Greg Oster wrote:

> My guess is it's a BIOS thing of some sort...  That, or /boot somehow
> got corrupted (or it now can't find /boot)

Sounds about right.

> You might want to try "conv=noerror,sync" with the 'dd'?   I might be
> kind enough to reply to the list, but I"m not sure I'm going to be of
> much help :-}

That itself was tremendously helpful actually. Doh! After reading the 
noerror paragraph in man dd again I realised that my original intention 
(to replace errors with nuls) probably wasn't being fulfilled. I 
could've sworn noerror did that all by itself. Let that be a lesson to 
me: Read the manpage, don't skim it. :-)

> I'd also attempt to do a 'dump' from the original RAID set... (you do
> have backups too, right? :) )

Mounting it is now impossible. I wanted to steal an image before the 
whole drive just up and died, and it looks like doing complete reads 
with dd have done some naughty things to it. Hrm..   dump -F looks 
promising though.

I do have backups, but they're from August 15. Recent enough to be 
useful, not recent enough as I'd like, and so I'm beating myself up 
wringing out these drives for every last ounce of information I'd 
otherwise be losing. :-)

> It's sounding like you're missing some metadata bits from the
> original filesystem, and that's going to cause all sorts of grief,
> no matter how you cut it...

If my dd conv=noerror didn't pad cleanly with nuls, that's hopefully my 
problem right there.

I can only see the top-level directory. cd'ing anywhere, or looking at 
any other files but those in the top-level directory, gives me garbage. 
It's tantalising..! :-)

> BIOSisms is my guess... that, or /boot (or a pointer to it) is
> munged.

This makes sense.

> When a RAID Level 0 set has a component failure there's not much that
> can be done to recover the missing bits... and if some of those
> missing bits are metadata, then backups become the prime recovery
> method...

Well, if worse comes to worse, I have two backups I'll just use.

Thanks again for the note, I appreciate your time. :-)

-Marc