Bug 14998

we have a major problem with our FreeBSD 3.3 boxes (they are
planned to be proxy servers)
 
Try the following commands and tell me your experiences (and
possibly a fix/workaround):
 
- telnet from a remote machine to your FreeBSD box
- login as ordinary user (not as root!)
- su (all following tasks are run under root account)
- run top
 
The process chain is now: telnetd - shell - su - top
the process owners are:    root     user    root root
 
Now kill the corresponding telnetd task from another root shell
and voila: telnetd has disappeared, but after some seconds
top sucks the CPU time 99% (on a single CPU machine)
 
Please don't ask "which fool would kill telnetd?"
exactly this happens, if you telnet from another FreeBSD box
in an xterm, then start "su" and "top" and finally kill the xterm 
without having killed the "top" task.
(there is this neat M$ish "X" button on the top right of the window :-(
top reports lots of IO errors in an endless loop (can be viewed with
truss)

I feel there's a major problem with FreeBSDs signal handling or
something

How-To-Repeat: telnet from a remote machine to your FreeBSD box
login as ordinary user (not as root!)
su (all following tasks are run under root account)
run top (or "more /etc/passwd")
Now kill the corresponding telnetd task from another root shell
and voila: telnetd has disappeared, but after some seconds
top (or more) sucks the CPU time 99% (on a single CPU machine)