Re: How to prevent malicious linking/posting to webapps?

[Lincoln Yeoh <lyeoh () POP JARING MY>]

At 11:24 PM 11-09-2000 +0200, Bluefish (P.Magnusson) wrote:
> > And the web app has a command which is accessed by a url similar to
> > http://www.mydomain.com/webapp?command=deletefolder&folderid=1
> > (assuming using cookies for session authentication and the session is
active).
> So.... We would perhaps draw the conclusion that having the session id in
> a cookie is a bit... risky ;-) Given that you can fool someone to run
> your javascript or html file while browsing the site in another window....
> *argh*

But having the session id in the url (the alternative to cookies) means it
can be taken from the http-referer field, if a user is somehow directed to
a malicious website (less likely but not impossible).

I am also assuming that cookies could be stolen by the attacker through
stuff like javascript (ick!). However the attacker cannot directly use
these cookies if we tie the session-ids to the legitimate user's source IP
(risking problems for some legit users). Thus it's more likely for the
stolen cookie info to be used to launch a less direct attack - like
malicious links/redirects.

Well since I haven't had many responses I'll put some stuff up for discussion:

***url/cgi parameter checksumming
One option I'm considering is to use cryptographically secure checksums for
urls/parameters.

So the attacker must be able to guess the correct checksum which is only
valid for that session-id and that particular url. The url could also have
a randomized parameter (salt) to make things more interesting for the
attackers.

Example:
checksum=sha1(cookie.sessionid, cgi parameters and values, salt, secret).

The secret (nominal 160 bits of entropy) could be a constant, or if you're
paranoid it could be randomly generated for that session. Though the
entropy won't be 160 bits, the attacker is unlikely to brute force it
successfully.

The issue I'm having with this is- it's more resource intensive and a lot
less convenient! Every url or form on a webpage has to be checksummed. And
checking when it's a HTTP-POST is not as easy :(.

A potential issue remains:
Say a webapp has a view message page, and on that page there's a high
impact option like "delete folder now!".

If a webapp user happens to read a malicious message on that page, some
fancy javascript might be able to cause the user to visit "delete folder
now!" from the message view page. And naturally that link will checksum
correctly.

A solution would be to avoid having such options available when viewing
foreign content, in addition to the usual confirmation screens. Also
turning off stuff like javascript/ActiveX (triple yuck!) would be a good
idea.

If anyone has other solutions I'd be glad to hear it, especially if they
are easier to do :).

Cheerio,
Link.