[Ocfs2-tools-devel] [PATCH 2/3] fsck.ocfs2: Add extent list check for discontig bg.

[Joel Becker]

On Thu, Jul 22, 2010 at 11:57:28AM +0800, Tao Ma wrote:
> On 07/22/2010 11:17 AM, Joel Becker wrote:
> >	Are you thinking that if e_leaf_clusters>cpg, we know it's
> >corrupted, but if it's just big enough to add up to greater than cpg, we
> >have no idea what happened?
> yeah, so I think in both case, we need to drop the group because:
> in the 1st case, sure it is corrupted.
> in the 2nd case, what if the 1st extent rec's e_leaf_clusters is
> corrupted and trigger the final number to be greater than cpg or it
> is really a rare case for us to have a corrupted extent record
> before the last one?

	Good point.  Ok, in both cases we drop the group.

> That makes me now think that can we really fix a e_leaf_clusters
> error based on the sum of other *looked sane* extent records?

	Yes, I think we can.  If you have multiple places corrupt, you
likely have someone scribbling over things with dd(1) or something.
It will tend to be really corrupt.  But if you have exactly one thing
corrupt, you probably have a code error or some other corruption that
just twiddled a bit or a byte.
	So when we see e_leaf_clusters is completely impossible, we know
that particular record got corrupted.  If everything else looks perfect,
it's very likely a single corruption.  That's why I think we can fix it.
	When e_leaf_clusters is possible but just adds up to more than
cpg, we don't know which e_leaf_clusters was corrupted, including
possibly more than one.  So we can't fix it.

Joel

-- 

"I am working for the time when unqualified blacks, browns, and
 women join the unqualified men in running our overnment."
	- Sissy Farenthold

Joel Becker
Consulting Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127