I'm struggling with a rather odd phenomenon. It seems to be that
in networks making use of AMD & NIS (AMD maps exported via NIS), a client
can hang the entire server simply by making a POP3 request to it. This
does not happen every time; perhaps once to a dozen times daily on a busy
network. The kernel versions I've tried were from 2.0.18 to 2.0.30,
starting with RedHat 2.1 and going up to 4.2. This happens with
libc-5.3.12 through 5.4.33, ypclients of RedHat 2.1 through 970318, etc.
I've tried several versions of AMD, including building it myself. Nothing
cures this problem.
The symptom is that users who leave something like netscape
running, that periodically POP's the server (who is also NIS/AMD server),
eventually one of the pop requests leaves a stuck ipop3d running on
server. Then, as the other netscape's try through the night, the process
table fills up on the server and you have to reboot it. In the meantime,
server refuses to access any existing mount points, or mount new ones
through AMD. It's very easy to bring a system down with this attack. Is
this normal? Need I write a little C process reaper that has to watch the
process table and kill things that have too many copies? Sounds like a
bug in AMD to me, but I don't know...
Anyone have any suggestions?
-bp
-- B. James Phillippe Seattle Software Labs, Inc Network Administrator Phone: (206) 521-8346 NIC Handle: BJP4 Fax: (206) 521-8340 http://w3.terran.org/~bryan http://www.sealabs.com