Re: git 1.5.3.5 error over NFS (pack data corruption)

From: Alex Riesen <raa.lkml@gmail.com>
Date: Nov 13, 2007, 21:48 UTC
Message-ID: <20071113214802.GH3268@steel.home>
In-Reply-To: <18233.47472.890658.729250@lisa.zopyra.com>

Bill Lear, Tue, Nov 13, 2007 15:49:20 +0100:

Show 11 quoted lines

> >> >> We have an NFS-mounted filesystem, and git pull is choking on it.
> >> >> 
> >> >> % uname -a
> >> >> Linux uhlr.zopyra.com 2.6.9-42.0.2.ELsmp #1 SMP Wed Aug 23 13:38:27 BST 2006 x86_64 x86_64 x86_64 GNU/Linux
> >
> >It is a really old kernel... Maybe you could try with some of the
> >recent ones?
> 
> I'll see if we can: the machine in question is a high security one,
> and not easy to change.  We haven't seen this sort of problem
> elsewhere on our newer systems as far as I know.

Suspected that :)

Show 10 quoted lines

> >I extend the part you quoted. The file is opened here:
> >...
> >This is strange. The current git should not produce anything like
> >this (and does not, here). ...
> 
> You are absolutely correct.  My comrade ran this with 1.5.0.1 by
> mistake.  He reran the strace with 1.5.3.5, and I have replaced
> the tarball on my server:
> 
>     http://www.zopyra.com/~rael/git/git-trace.tar.bz2

Still looks like v1.5.0.1:

    write(2, "fatal: ", 7)                  = 7
    write(2, "cannot pread pack file: No such "..., 49) = 49

> With this minor correction (!), here is the last part from
> the last file (strace.out.26001):

I think it got worse:

> write(3, "\360]|\204\205\317|\352\336C\342\316\206\334\326\201\211"..., 4096) = 4096
> write(3, "\20\t\33if (cached_vias.insert(die,\227\17"..., 4096) = 4096

The data in packs is zlib-compressed. I don't think you should be able to see any readable strings in them (except for "PACK" in the header). I believe you have a data corruption now.

Does git-fsck --all say anything about that, BTW?

Besides, could you try with a git compiled with NO_PREAD?