Re: [Gnu-arch-users] in-tree pristines fatally wounded (merge-fest etc)

[Robert Collins]

On Tue, 2003-12-02 at 12:37, Tom Lord wrote:
>     > From: "Stephen J. Turnbull" <address@hidden>
>     > I gather from the later discussion in your post that library corruption
>     > is detected via the "inode hack".  Is that understanding correct?  Are
>     > there other tests applied?
> 
> Correct, I think.
> 
> Whenever tla needs a library revision, it does a quick inventory of
> that revision and compares the inode data on the disk to the inode
> data recorded when the revision was created.  If they differ, tla
> aborts.   Mostly this is to catch cases where you made a project tree
> (aka "working directory") with hardlinks to the library, and then
> modified a file without breaking the link.
> 
> It doesn't do anything like read every file and verify a checksum for
> obvious performance reasons.   That's the kind of functionality that
> in a mission-critical context you'd want to write a 30-line shell
> script cron job for.

Yep, it detects mismatches between the inode signature and the library.

The only way to break it I can think of is to hardlink the inode
signature into a tree, and then alter it unsafely - but the inode
signature code is hardlink safe AFAICT, so this won't happen 'by
mistake'. It's possible for someone to write a script that will
deliberately do the wrong thing, but hey - anything is possible.

Rob
-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.

signature.asc
Description: This is a digitally signed message part