[OpenAFS] windows client/rx connections

Jeffrey Hutzelman jhutz@cmu.edu
Fri, 25 Feb 2005 12:54:16 -0500


On Friday, February 25, 2005 12:40:04 PM -0500 "John S. Bucy" 
<bucy-openafs-info@gloop.org> wrote:

>
> I seem to remember hearing at a celladmin meeting a few months ago
> that something changed in the Windows client that caused it to start
> using "more" rx connections and that this was causing fileserver
> crashes.  What were the particulars?

Prior to 1.3.74 or so, there were similar bugs in both the Windows and UNIX 
clients related to locking of outgoing connections to fileservers.  The 
effect of this was that it was possible for a connection to be freed just 
as the client was about to make a call on it, resulting in corruption of 
various data structures.  There was also a similar bug in the fileserver, 
related to tracking of outgoing callback connections.  In both cases, a 
high turnover rate in the number of connections made the bug much more 
likely to be triggered.  IIRC, the fileserver and unix client variants of 
this bug were one of the main reasons for the 1.2.13 release.

Also prior to 1.3.74 or so (and all the way back to pre-OpenAFS days), 
there was a bug in the Windows client which had the effect that if the 
"always encrypt" flag was set, the client would create a new connection for 
more or less every call.  This generated a high connection turnover rate, 
making both the client and the fileserver it was talking to more likely to 
crash.

The thing that changed "recently" is that somewhere around 1.3.70, the 
Windows client started shipping with the always-encrypt flag turned on.


-- Jeff