On Fri, 2003-07-11 at 19:07, et@cyberspace.org wrote:
> The reason why i did a prototype is because i had to modify the internal
Yes, that is a limitation, the new acl_helper approach in squid 2.5 is
somewhat better than the old redirectors because you can get arbitrary
headers (once you know which ones you are looking for that is). True it
is far from perfect, and it would certainly be very useful to see the
POST data for forms. That too has room for evasion techniques and would
probably require normalization of of some sort (at least be decoded).
However, for a simple signature matching IDS/filter, you can glean a lot
just from the URL/method/couple of headers and I can think of few
attacks that wouldn't have some kind of identifiable pattern in there.
That said, imagine trying to reliably catch exploits in the transfer
encoding and request footers, and all the other crazy parts of the HTTP
protocol. I'm not even sure what squid does with chunked encodings and
the like, it probably passes them straight through. A totally
comprehensive solution would have to handle all of the protocol decode I
guess.
--
// Gianni Tedesco (gianni at scaramanga dot co dot uk)
lynx --source www.scaramanga.co.uk/gianni-at-ecsc.asc | gpg --import
8646BE7D: 6D9F 2287 870E A2C9 8F60 3A3C 91B5 7669 8646 BE7D
Received on Mon Jul 14 19:43:13 2003
This archive was generated by hypermail 2.1.8
: Wed Aug 23 2006 - 14:01:16 EDT
|