2.4.18+O(1) tcp timewait zombies

Hi all,

After upgrading from stock 2.4.17 to 2.4.18 (with O(1) scheduler patch) I'm seeing some strange networking behavior. Basically, upon exit rsh connections seem to go into TIME_WAIT, count down to zero, and then stick around forever. For example:

compy$ rsh hyena echo hello < /dev/null
hello
compy$

This leaves a socket connnection in TIME_WAIT state:

tcp ... TIME_WAIT timewait (51.75/0/0)

However, the strange thing is that after the connection times out, it sticks around. I've got hundreds of these connections open that look like this:

tcp ... TIME_WAIT timewait (0.00/0/0)

And never seem to go away. Eventually, this causes me to get this message whenever I try to do an new rsh:

socket: All ports in use

Can anyone offer me any advice on how to debug this?

Take care,
Jason
--
Jason Baietto
[email protected]