[OpenAFS-devel] Re: [OpenAFS-announce] OpenAFS 1.2.11 release available

[Derrick J Brashear]

On Mon, 12 Jan 2004, Michael Niksch wrote:

> RELNOTES-1.2.11 mentions another bug having been fixed:
>
> - A bug where replicating volumes on systems with namei fileservers
>    could result in an empty volume was fixed.
>
> Can anyone tell a bit more about this bug? I have never seen it until
> today, after installing the OpenAFS 1.2.11 database server binaries, but
> still running the OpenAFS 1.2.10 filserver and volserver binaries. In
> that stage, I created a new volume to hold all the OpenAFS 1.2.11 stuff.
> The release seemed to work, but hours later I found that machines trying
> to get the data from a replica found no data at all until I did a 'vos
> release -f'.
>
> Am I in danger to have existing replicas corrupted before I get to
> restarting my fs processes to activate the OpenAFS 1.2.11 binaries? Or
> is this a bug that affects only new volumes? Should I do a 'vos release
> -f' on all volumes just to be safe? This would take ages, of course.

It's a namei fileserver problem. In theory it was always present, but was
masked by another bug before 1.2.(8 or 9). You shouldn't see it on
platforms not using the namei fileserver.

vos release -f actually isn't guaranteed to help. the problem is a
reference to the root vnode in the volume is leaked, and then reused,
writing into the unlinked file, despite a new file being created. if you
release an additional time, the right thing happens. i don't remember if
another release after that then loses.

a volserver restart of course loses the bad state, and it should only
happen after a volCreateRoot, e.g. when a volume is first replicated to a
server.