[OpenAFS] Evaluating OpenAFS

Russ Allbery rra@stanford.edu
Tue, 17 May 2011 12:34:30 -0700

Stephan Wiesand <stephan.wiesand@desy.de> writes:
> On May 17, 2011, at 19:31 , Russ Allbery wrote:

>> If you're willing to test on Debian unstable, the version of the
>> OpenAFS packages currently in Debian unstable should be close to what
>> will show up as 1.6.0pre6 (and has the data corruption bug already
>> fixed).

> If the issue is what 4d3e83e86abd0bb6f70e50f208c63bcc248c6673 fixes (?),
> I wouldn't call that a data corruption bug. It seems like an ordinary
> bug in the client, preventing it from retrieving the true content of
> certain files stored safely and correctly on the fileserver.

Data corruption is somewhat strong, but note that, in the presence of this
bug, the client will have the correct file length and will think it has
the complete file, but everything after the first chunk will be nuls.  So
if you manipulate the file in a way that causes the client to write it out
again (like copying it somewhere, or even editing it), the data will be
corrupted on the server.

When it comes to any potential for data loss in a file system, I tend to
err on the side of scaring people away.  :)

Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>