COMMAND
inetd
SYSTEMS AFFECTED
Solaris, Linux
PROBLEM
Alla Bezroutchko stumbled upon something that looks like a bug in
inetd on Solaris. If a Solaris box is portscaned by nmap with -T
Insane option (very quick scan) daemons that are started by inetd
stop responding. That is you can connect to them, connection get
accepted, by they don't display any banner or answer in any way.
It stays that way until inetd is restarted. Other daemons (not
started by inetd) seem to be unaffected by this.
The effect depends on number of daemons enabled in inetd
configuration. If only one daemon (ftp in my case) is enabled,
nothing happens at all. Inetd with two daemons does hang but not
always. Five daemons enabled make it hang every time.
This was tested over a 10Mbps LAN against Solaris 7 and 8 on Sparc
and Solaris 7 on Intel. All three were affected.
All linux versions are vulnerable as well, you can kill inetd over
a 28.8 modem in less than 40 seconds ... you just need to connect
and disconnect really fast .....
SOLUTION
Properly patched Solaris doesn't seem to react to intensive
portscan in any way. Unpatched Solaris inetd does hang and
doesn't seem to do it on purpose. It doesn't log anything and it
doesn't answer to any host connecting to it, not only the one
that did the scan. Inetd sleeps on accept syscall (normally it
sleeps on poll) and stays that way until restarted.