[OpenAFS-devel] Sock memory leak

[Troy Benjegerdes]

On Mon, Jan 02, 2006 at 01:15:37PM +0100, Jerry Lundstr?m wrote:
> Hi,
> 
> We are running 1.2.13 on 6 machines (2.4.29) serving IMAP over AFS and
> we have noticed large leaks in sock buffers. This means we need to
> reboot the machines after ~2 month and I remember that someone sent a
> mail saying an afs client should be rebooted after 200days and this can
> be one of the reasons.
> 
> Looking at a machine that has been up around 40days we get this:
> 
> # cat /proc/slabinfo |grep sock
> sock              320484 320484   1024 80121 80121    1 :  124   62
> 
> At the time of this check the machine had 700 open sockets and that
> doesnt explain the 320meg memory allocated for socks.
> 
> Shutting down the IMAP server and AFS does not release anything.
> 
> Is this a known issue in 1.2.x and/or fixed in 1.4.x?

Out of curiousity, what IMAP server are you running, and how big is your
AFS cache? Are these machines SMP machines? (I used to have a lot of
trouble with SMP machines and 1.2.x series AFS clients having race
conditions with courier-imap).