Hi all,
I've finally managed to compile my notes on WAS Protect (and
the HTTP server side protection in general). I also now have time
in my schedule to work on this element - I expect we can have
it completed in time. I can start and finish the work on my own
but, for things like this, I prefer to have input from other
people - it's always useful to consider various points of view.
The document is a result of my work on mod_security, its limitations,
analysis of other similar attempts, and various server side
programming APIs. Mod_security limitations were particularly
useful.
In my opinion, WAS Protect should consist of a specification of
an execution engine, with a data model and a (simple)
programming language. The capabilities of the engine are
specified in the document.
I have the following concerns:
1) The scope of the effort
On one level, WAS Protect can and should be used to protect
from known and unknown threats. On the next language, it can
also be designed to be a more generic server-side protection
language, which people can use to secure their web servers
and applications. I've done both for mod_security but generic
protection may not make sense for WAS Protect. It would be
cool but would it serve a purpose? Would anyone implement it?
Personally, I prefer iterative development and wouldn't
mind leaving the fancy stuff for later. Another thing to
consider is that a simpler standard will be more likely
to be accepted, or accepted faster.
Some of the features mentioned in the document are an order
of magnitude harder to implement. For example, stateful
operation. This functionality is a good candidate for a
later version.
Stateful operation is necessary for these:
* Detect brute force attacks
* Session fixation detection
* Session hijacking detection
* IP address / user blacklisting
2) Overlap with WAS Test
There's a large overlap in functionality between Test and
Protect. Test creates requests, and analyses responses. Protect
analyses requests, and responses, possibly creates responses.
Therefore a single language should be used for both. From
what I've seen, VulnXML cannot support Protect. So we need
to resolve this issue.
3) The question of format
XML is obviously the flavor of the day, for good reasons. But
is it a good choice to store engine statements/instructions?
XML is good for storage and parsing, but it's not very
user friendly. And if we use XML it is unlikely anyone will
want to write Protect by hand. Is having more than one
representation a good way of solving this problem?
PS. In case the attachment does not go through the email, I've
uploaded it here:
http://www.webkreator.com/download/was-protect-1.pdf
--
ModSecurity (http://www.modsecurity.org)
[ Open source IDS for Web applications ]
