Bug 12779

Hi,

I'm running a RH6.2 Linux box called ``bunker'' with 

</etc/inetd.conf excerpt>
...
time    stream  tcp     nowait  root    internal
...

bunker is used by other hosts on the LAN to get a correct
time with rdate (Solaris and Linux rdate clients running 
``rdate bunker'' in a cron).

Looking at open files (with /usr/sbin/lsof) you can see a
new entry at each connection  of a client.

inetd      3163 root   11u  IPv4     403354               TCP
bunker.izinet.au:time->bonnie.izinet.au:1697 (CLOSE_WAIT)

After some time the table of open files is full and other
applications stop working properly (syslog, cvs server (connection
reset by server: too many open files) resulting in typical denial
of service attack possibility.

Reproduce it by running on the server:

watch "/usr/sbin/lsof |grep time |wc -l"

and on a client

while true; do rdate bunker; done

[As a fix I thought that the socket should close itself when
the client calls close() or when the client process exits.
This does not happen and it seems there's no timeout on server
side socket open in TCP_WAIT state.]