[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: e2fsprogs 1.23 problem handling 2.2 version 1 format journals

On Tue, 2001-10-02 at 14:15, Theodore Tso wrote:
> Yes, I'll fix e2fsck to ignore these fields.  (Actually, what I should
> do given the paragraph below is to change e2fsck to *zero* those
> fields for v1 superblocks.)

Good stuff!
> > Except it looks to me as though the upgrade code is *not* zeroing out
> > those fields (from an attempt yesterday when I saw the problem to do a
> > V1->V2 upgrade using a 2.2 kernel).  This means ignoring the misset
> > fields just leaves a bomb in place to explode when the journal is
> > upgraded (at this point e2fsck explodes when it sees the journal).
> > 
> > Time to have a dig into that code...
> Yeah, the kernel code should definitely be fixed.  On the e2fsck side,
> I'll look into figuring out some good huerstic to distinguish between
> a the case of a buggy upgraded V2 superblock and a legitimate shared
> journal case.  (i.e., looking at the fields and deciding that they're
> insane, and thus offering the user the opportunity to clear them out.)

I've found out what actually caused the problem here....
I've been using ext3 on production boxes for quite some time now, and
the build process for those boxes has modified over time.

Early boxes had a slightly hacked RH kickstart which created the journal
file stubs during install, and then created the journals themselves on
first boot.

Then the journal build was modified to use tune2fs -j to build the
journals all in the kickstart %post section.

It appears that the affected system was one that had just the latter
build method, and I guess used pretty much the first tune2fs journal
build support - I guess one of Stephen's 1.20 WIP sets of e2fsprogs.  
I haven't pinned down the exact versions yet - and the copy I looked at
appears to build a V2 journal after zeroing the block its building the
structure into...

So I think I have a few machines with corrupt V2 format journals (they
are hidden as well to make life slightly more complex).  However
e2fsprogs prior to 1.22 do not detect their is a problem, and the
kernels I have used do not see a problem.

I guess I could build some scripting to strip and recreate the
journals... but this is all looking nasty :-(


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]