[OpenAFS] Re: Invalid tags in release stream?

[Andrew Deason]

On Fri, 1 Aug 2014 01:03:50 -0400
Nathaniel W Filardo <nwf@cs.jhu.edu> wrote:

> I'm wondering if we're the only cell on the planet seeing messages like
> 
> > Thu Jul 31 04:24:32 2014 ReadVnodes: unknown tag x96 found, skipping
> > Thu Jul 31 04:24:32 2014 ReadVnodes: unknown invalid tag x96, aborting
> 
> transiently while releasing volumes?  All hosts involved are Linux
> running Debian packaging 1.6.9-1 on amd64 machines.  Last the salvager
> ran on this particular volume, it came away happy, for whatever that's
> worth.

While I cannot guarantee that you're the only site seeing such a thing,
seeing messages like that for normal volume releases/moves/etc is not
normal, and is cause for concern. In almost all cases, yes, that means
the dump stream is corrupted in some way (especially if the reported
values change). While I have seen that before, in all cases it was
caused by either a bug or something mangling a dump incorrectly; in any
case, there's nothing I know of in 1.6.9 that would cause that without
something external messing with the dump stream.

What would be interesting to see for that would be a copy of a dump of
the volume. If there's nothing sensitive in the volume, you may want to
just share a copy of the dump itself. Or, you can examine the dump
yourself locally by running it through afsdump_scan, a tool to interpret
the contents of volume dumps. A copy of that with some extra fixes
applied to it can be found here:
<https://github.com/openafs-contrib/cmu-dumpscan/tree/fixes>

I would dump the volume using 'vos dump' (specifically the RO clone on
the RW site, since this is for a release), and run:

$ afsdump_scan -PBHVvi /path/to/volume.dump

And you should see a bunch of info about the files in the dump. If there
is a problem with the dump, afsdump_scan should print out a somewhat
similiar message as you saw, complaining about an unrecognized tag.

-- 
Andrew Deason
adeason@sinenomine.net