HTTPS Cookie Hijacking Not Just For Gmail

mikepery writes with a followup to last month's mention of a security vulnerability affecting Gmail accounts, which it seems understated the problem. "I figure the Slashdot readership is the best place to reach a large number of slacking admins and developers, so I want to announce that it's been 30 days since my DEFCON presentation on HTTPS cookie hijacking, and as such, it's now time to release the tool to a much wider group. Despite what was initially reported, neither the attack nor the tool are gmail-specific, and many other websites are vulnerable. So, if you maintain any sort of reasonable looking website secured by any SSL certificate (Sorry Rupert, you lose on both counts), even if it is just self-signed, you can contact me and I will provide you with a copy of the tool. Be sure to put 'CookieMonster' in the subject, without a space." (More below.)

"I'd also like to encourage security professionals and consultants to request a copy of the tool for use in encouraging their clients to adopt SSL properly for their websites. There's no possible way for me to reach every site, but if convincing demonstrations can be given of the vulnerability on an individual basis, perhaps that will drive the issue home much more than the press alone has done. Heck, the tool might even land you a few new clients."

Easy to find out...

If you want to manually examine a site you visit:

Clear all cookies.

Log in.

Clear all cookies marked as "SECURE" (in firefox, preferences->privacy->show cookies. Delete JUST the cookies marked as "Encrypted connections only").

Go back to the site. Can you act as if you are logged in? If so, the site is COMPLETELY insecure.

Re:Easy to find out...

I would also like to point out you can test it further by adding your own cookies using the same cookie values on another computer via the webdeveloper plugin [mozilla.org].

Re:Easy to find out...

If you're suggesting that the site should be tying the cookie to an ip address, then I have to disagree. I've done just that, and I discovered that it's a very bad idea. Not just because some people have a different ip address from one session to the next, but because some of them have a different ip address from one page to the next. For example, there are some ISPs that send GET requests through one address, and POSTs through another.

Re:

I can get you there.
My ISP assigns me a new IP every few minutes even if I'm in the middle of a download. Makes using rapidshare impossible.
And yes, there's always the jackass developer who thinks he's smart locking sessions to an IP which for me just means either being logged out again and again or it locking up till the link to the old IP has expired.

Yes I know it's the ISP but they're the only game in town and I'm not going back to 56K

Re:

No, I was not suggesting that, but you know someone is doing it (or binding to useragent, but same concept). Moving the cookie to a different computer's browser is the real test in case they have something extra thrown into the mix like that. PS: I although I already posted once, but it's not showing after 3 mins, so sorry if you get two replies from me.

Re:Easy to find out...

if this is true (and I am able to follow directions correctly) then Bank of America has some explaining to do.

Re:Easy to find out...

I think the explanation is quite simple. "We don't know what we're doing."

Re:Easy to find out...

I did this with our own web-based product and found out that yes, indeed, we are insecure. It took a few minutes of poking around to find out how to secure our site.

So, for everybody else: if you are using PHP, you need to pay attention to Set_Cookie_Params() [php.net]. Here's the 1-liner call that we make in order to solve this problem for us, before any calls to session_register():

Session_Set_Cookie_Params(720, '/', $_SERVER['SERVER_NAME'], true);

Parameters:

1) 720: Our sessions timeout after 2 hours.

2) '/': the cookie applies to all paths within our site.

3) $_SERVER['SERVER_NAME']: applies only to the specific domain name originally called. (we use subdomains, so this is important)

4) true: (the most important one), this means that the cookies can only be used over SSL.

WTF?!?

If you are going to release a tool, just fucking do it. Give is a link and be done with it.

please tag itsatrap

Why don't YOU just e-mail the guy and be done with it?!

Perhaps it's... A TRAP!

Re:

Please install this tool on all servers and workstations as admin for maximum benifit.

Djagno and Secure Cookies

I run a few Django SSL-secured websites, and I noticed the default is to send insecure cookies for the session id (i.e. hijack-able cookies). I'm going to try to get on someone's case to make this information more widely available, because you have to turn on secure cookies with a "SESSION_COOKIE_SECURE = True" statement, which I never knew until I checked today. Doing this should secure any Django-powered site from this particular attack.

How do you secure a site?

There were a lot of 'email me's and talk about bad htps settings but not much content on really what needs to be done for fixing an existing site or properly setting up a new site to be secure.

Re:How do you secure a site?

There were a lot of 'email me's and talk about bad htps settings but not much content on really what needs to be done for fixing an existing site or properly setting up a new site to be secure.

Executive summary:

It is possible to set a single bit in a cookie sent to the browser which means "Only send this cookie over secured connections".

Many websites (including some important ones like online banking) don't set this bit for the cookie(s) used for session tracking. Hence, it is possible for an attacker to get the cookie with an invisible proxy that injects HTML which forces the browser to fetch something from the server which set up the cookie.

eg. I set up an invisible proxy on my wireless network which injects into every page and logs the cookie your browser sends when it attempts to connect to mail.google.com for the image.

I can now plug this cookie into Firefox and read your email.

Solution: if your website sets any cookies over an HTTPS connection, such cookies must set the bit meaning "Only send over secure connections". How one goes about doing this will depend on whether you're generating cookies yourself or using an existing framework.

The issue described

1. Look at DNS requests or do a IP-domain reverse lookup to know what websites the target is visiting over HTTPS. Automated tool can do this over time.

2. On next regular HTTP request by your target, be a man-in-the-middle and inject an image pointing to desired HTTPS site, except don't use HTTPS - just HTTP.

3. Browser will dutifully send the cookie along with the image request over plain HTTP (after all, the domain names match), even though the cookie was created and managed only via HTTPS by the original website.

4. Now your automated sniffer just picked up the supposedly "secure" cookie for the HTTPS site, even though you never even attempted to hack the HTTPS conversation. If the site stores your username/password, a session id, etc this could expose sensitive information.

5. Protect your applications by setting the encrypted session only flag on the cookie so the browser won't send it with plain HTTP requests. If you have HTTP and HTTPS areas of the site, keep separate cookies for both areas and make sure sensitive info is only stored in the HTTPS-only cookie.

Re:

Yep. The Very Best(tm) security is an air gap.
But it kinda limits your possibilities.

Re:yeah...

In what way? We run a lot of stuff over our SIPR [wikipedia.org] and I haven't really noticed any "limits" to what we've done. Hell, I even watch CNN over SIPR sometimes.

Re:

CNN, FoxNews, and a few other broadcast channels are re-broadcast on SIPR because that's where Intel gets their Intel.

Walk into any US Intel / Base Ops / Command Post in the world, and you'll find CNN on a big flat-screen up on the wall.

Re:

Walk into any US Intel / Base Ops / Command Post in the world, and you'll find CNN on a big flat-screen up on the wall.

I tried this, now i'm in gitmo.

Re:

He is. Rebroadcasting a satellite feed is simple.

Re:

Many (most?) classified government systems require TEMPEST [wikipedia.org] hardening, to specifically protect against your second point. Many government buildings which contain classified materials also require a similar hardening.

Re:

What kind of an email address is "mikeperry-cmviathefsckedorganization" ??

Am I missing something? Should I replace the - with a @ and make organization into .org ?
It's fscked.org from the link, so maybe just mikeperry-cm @ that?

Screw it... sent an email to everything... mikeperry-cm [at] fscked org didn't get returned... I'll assume that's it.

Re:new security vulnerability

cmvia is command modulated voice interface application. In other words you roll down your window and yell 'MIKE PERRY I NEED THAT FILE.' Eventually a carrier pigeon delivers it to you.

Re:

He was just trying to use basic Darwinism to filter out idiots. But some defender of white moths told them to fly away and take cover cause da smoke was a comin! Dam u!

Re:Secure by permissioning or secure by encryption

File Permissions: When the web server asks to write cookie, browser uses file permissions to create a group for that site, and allows only members of that group to read or write to that cookie. Either use the disk filesystem's permissioning, or reimplement a permissioning system within the browser profile, to be used only for cookies.

Requires giving the browser root access to the system (though you can mitigate this by running a jail or sandbox).

File Encryption: Using a public key encryption method, the content of cookie file is encrypted using the web site's private encryption key, and can only be decrypted by the web server.

I thought that's what SSL cookies do in the first place?