Blocking nimda probes with a content-layer switch

Has anybody tried to block nimda HTTP GET probes using URL
pattern matches in a "layer-4-7"[1] switch?

The ideal result is to prevent nimda GET probes from ever
reaching the destination address, but causing the session
to be reset towards the server after the open handshake but
before the GET can be sent to the server would be acceptably
useful.

Particularly whether it's possible on a cisco/Arrowpoint switch,
but it would be interesting to know about other vendors too.

Please reply directly, will summarise if there are answers to
share.

Thanks!

[1] substitute phrase-du-jour as appropriate

# Detect W32.nimda worm and move to /var/tmp/nimda.DATE.username
# w32.nimda.amm

Has anybody tried to block nimda HTTP GET probes using URL
pattern matches in a "layer-4-7"[1] switch?

note that this worm uses multiple methods for spreading - email and HTTP. these methods are only addressing the HTTP side of things:

NBAR (Network Based Application Recognition) in IOS:
         existing CodeRed NBAR settings will block the HTTP probes.
         see http://www.cisco.com/warp/public/63/nbar_acl_codered.shtml

Cisco Content Engine:
         rule enable
         rule block url-regex http://.*/cmd\.exe
         rule block url-regex http://.*/root\.exe

Cisco CSS11K (Content Services switch):
         service dummy
           ip address 10.1.1.1
           keepalive type none
           active
         !
         header-field-group cmd.exe
           header-field cmd.exe request-line contain "cmd.exe"
         header-field-group root.exe
          header-field root.exe request-line contain "root.exe"
         !
         content block_cmd.exe
          protocol tcp
          port 80
          url "/*"
          header-field-rule cmd.exe
          add service dummy
          active
         content block_root.exe
          protocol tcp
          port 80
          url "/*"
          header-field-rule root.exe
          add service dummy
          active

The ideal result is to prevent nimda GET probes from ever
reaching the destination address, but causing the session
to be reset towards the server after the open handshake but
before the GET can be sent to the server would be acceptably
useful.

note that only the CSS and the CE in the above will actually prevent the 'tcp syn' from propogating, and they'll establish the tcp 3-way handshake, receive the HTTP headers and then match the request to be an "attack".
NBAR will only match the headers on connections that successfully establish a 3-way handshake. (NBAR won't cause the router to spoof the TCP setup - so-called delayed-binding).

cheers,

lincoln.

If you're using snort for network intrusion monitoring, you can do something
like this:

alert tcp any any -> any 80 (content: "cmd.exe";msg: "cmd.exe access in HTTP!!";react: block;)
alert tcp any any -> any 80 (content: "root.exe";msg: "root.exe access in HTTP!!";react: block;)

If the snort box is on the same subnet as your webservers, the webserver will
still get the request, but will be sent a RST and won't respond. Not as good
as dropping the packets at the front door, but better than your webserver
responding.

If you have the snort box in the data path, then you can actually drop the
packets instead of just sent the RSTs.