Re: [websec] X-Frame-Options and SSL

[Tobias Gondrom <[e-mail]>]

Hello,

so far I agree with Brad to leave the option open to be framed by a 
http-site even though you are https.  But what I also take from this, is 
that we definitely should expand on this in the Security Considerations 
Section and point out explicitly the risks of such.

So todo for the next version will be to add an Security Considerations 
section explaining why you SHOULD NOT (RFC2119) do it and what potential 
security risks will arise if you do it.

Thanks a lot for your reviews and your feedback pointing that out.

Tobias



On 23/07/11 05:20, Devdatta Akhawe wrote:
> I guess I am looking at it from a more 'what-is-the-invariant'
> perspective. The fact that HSTS makes everything HTTPS is a nice
> simple invariant that I feel is worth maintaining.
>
> Re. the economics argument: seems to me that you are
> assuming rational behavior; that Alice.com knowingly (and fully
> understanding the implications) chose to say "allow http site to frame
> me."
>
> I fear that the reason alice.com said "allow http://bob.com to frame
> me" might be that alice didn't realize the implicaions.
> Alice thought HSTS made her secure against active network
> attacker and doesn't understand the dangers of being framed by
> untrusted party.
>
> This is similar to why people created https pages with http scripts*
> in them: they didn't do this because 'the total money earned by
> attackers by hijacking the script to be on the wire is less than the
> cost of being on the wire'. They did it because they forgot/didn't
> think of the threat.
>
> Maintaining one simple invariant is a good thing. Having a separate
> mechanism for 'make me safe against active network attacker' reminds
> me of the 'secure' cookie: one had to use https as well as remember to
> use 'secure' cookies to really be secure against the active network
> attackers.
> ---
>
> But the scenario you presented also makes sense to me. I think my
> confusion was that I thought HSTS mandated browsers to block mixed
> content. Rereading the draft, it seems HSTS only suggests browsers do
> this (Section #10) and doesn't mandate any thing like this.
>
> How about a similar language for HSTS enabled sites being framed by HTTP sites?
>
>
> Thanks
> devdatta
>
>
>
>
>
>
>
>
> On 22 July 2011 15:34, Hill, Brad<[e-mail]>  wrote:
>> No, the clickjacking risk against an individual resource can still be managed independently of HSTS.
>>
>> It's not just about the attacker, it's about the asset.  HSTS can be used to protect high value assets like credentials and sessions without implying that every inbound click is high value or not subject to other controls.
>>
>>
>> Let's take a purely hypothetical scenario where alice.com wants to put a button in a frame on example.com. Let's also say that every hijacked click to alice.com against a U.S. user of example.com can generate a revenue of $0.02 for an attacker.  Alice.com deploys HSTS but example.com doesn't deploy https.   Alice.com believes it can generate revenues of $500/day from example.com.  Example.com believes it can generate revenue of $500/day from the partnership, but can't be convinced to move to https because that would cost $1000/day - more than the total value of the partnership.
>>
>> If we let "https://alice.com/exampleDotComButton" set XFO to "ALLOW http://example.com", there is the possibility that an attacker can set up in a U.S. coffee shop as an active MITM.  Such an attacker might make $0.50 a day.  This is probably not worth the cost of mounting the attack, and alice.com might notice the attack and add controls based on the originating IP or address block before any significant money flows to the attacker.  Because of this, alice.com's expected loss with XFO to the insecure site is maybe only $0.06/day and it is willing to accept this risk.
>>
>> If alice.com doesn't deploy any XFO headers, attackers can operate out of countries with a low cost of living, ensnare many more users, and generate fraud revenues of $100/day.
>>
>> If we change XFO so that sites that deploy HSTS require https-only framing, alice.com is left with only bad choices:  bear the risk of clickjacking, turn off HSTS, or don't partner and forgo the revenue from example.com.
>>
>> -Brad
>>
>>
>> -----Original Message-----
>> From: Devdatta Akhawe [mailto:[e-mail]]
>> Sent: Friday, July 22, 2011 3:31 PM
>> To: Hill, Brad
>> Cc: Adam Barth; [e-mail]
>> Subject: Re: [websec] X-Frame-Options and SSL
>>
>> Yes, but IE9 (e.g.) blocks mixed stylesheets, frames, objects too.
>>
>> Anyways, the point of the thread was to bring up discussion on this possible weakness. I (really) don't care much on what should be done in the default case. Whatever decision is taken by the WG, I think a note should be added in the "Security Considerations" section to that effect.
>>
>> What I do feel strongly about is the case where the secure site being framed is HSTS enabled. In such a scenario, the browser should block the insecure frame from framing the secure site. Clearly, in the HSTS case, the site has told the browser "I really care about the active network attacker scenario" and as such the browser should be proactive against the active network attacker. Do you agree?
>>
>>
>> -devdatta
>>
>> On 22 July 2011 14:16, Hill, Brad<[e-mail]>  wrote:
>>> In this case alice.com can only include http://bob.com.  I'm not saying we should force inclusion of http://bob.com, I'm just saying we shouldn't ban it, because it's a lot better than nothing.
>>>
>>> It is alike in concept to mixed content, but on a greatly reduced attack surface:  Routing a click to a location on a page vs. (in the case of an mixed script src) full control of the secure DOM.  For a business deploying XFO, that difference is pretty important.  Alice.com may want to be able to have a "like" or "pay" button from a secure source framed by the insecure bob.com because the risk/fraud rates from only active attackers may be acceptable or amenable to other compensating controls, where those from generalized remote clickjacking may not.
>>>
>>> Brad
>>> -----Original Message-----
>>> From: Devdatta Akhawe [mailto:[e-mail]]
>>> Sent: Friday, July 22, 2011 2:59 PM
>>> To: Hill, Brad
>>> Cc: Adam Barth; [e-mail]
>>> Subject: Re: [websec] X-Frame-Options and SSL
>>>
>>>> I guess I'm leaning towards "this is a weakness, but we shouldn't do anything about it."
>>> Say alice.com does care about this. My problem with the above position is alice.com can do nothing about this weakness. There is no flag it can send to say "hey don't allow insecure things to frame me: I am really worried about the active network attacker." It can do some weird haxoring similar to JS framebusting code but it is a pretty bad thing.
>>>
>>> To me, this is similar to the original motivations for XFO.
>>>
>>> From
>>> http://lists.whatwg.org/pipermail/whatwg-whatwg.org/2008-September/016
>>> 284.html
>>> ----8<--------
>>> "For a couple of months now, along with a number of my colleagues at Google, we were investigating a security problem that we feel is very difficult or impossible to avoid on application side, and might be best addressed on HTML or HTTP level in contemporary browsers."
>>>
>>> ---->8----------
>>>
>>> My view is that this weakness is similar to mixed content. Modern browsers (e.g., IE9) block mixed content by default while still allowing a click to enable it. I feel in the same style this should be blocked, while still allowing the user to click through.
>>>
>>> And in the same spirit, if the secure page being framed is HSTS enabled then there shouldn't be an option to the user to click through.
>>>
>>>
>>>
>>> =Devdatta
>>>
>>>> 1) Don't add the invariant, because moving the threat from remote to active is still a significant reduction in attack surface while maintaining compatibility with existing usages.  If we break that, the alternative is probably to have no protections.
>>>>
>>>> 2) Add the invariant because existing uses of this pattern are
>>>> already
>>>> broken: the user can't readily verify the origin of or that the
>>>> framed content is, in fact, secure.  This is only a Spoofing risk,
>>>> however, and so is less severe than the direct Elevation of Privilege
>>>> allowed by clickjacking.  (spoofing a "like" or "pay" button doesn't
>>>> get an attacker much)
>>>>
>>>> I guess I'm leaning towards "this is a weakness, but we shouldn't do anything about it."
>>>>
>>>> Brad
>>>>
>>>> From: [e-mail] [mailto:[e-mail]] On
>>>> Behalf Of Devdatta Akhawe
>>>> Sent: Wednesday, July 20, 2011 2:34 PM
>>>> To: Adam Barth
>>>> Cc: [e-mail]
>>>> Subject: Re: [websec] X-Frame-Options and SSL
>>>>
>>>> In case of http bob including https jquery, the HTTPS Jquery will run with the privileges of http bob.
>>>>
>>>> In the other case, https alice frame will run with the privileges of
>>>> https alice
>>>>
>>>>
>>>> =dev
>>>> On 20 July 2011 13:30, Adam Barth<[e-mail]>  wrote:
>>>> Why is that?  We're talking about HTTP Bob including HTTPS Alice,
>>>> just like we're talking about an HTTP page including HTTPS jQuery.
>>>>
>>>> Adam
>>>>
>>>>
>>>> On Wed, Jul 20, 2011 at 1:26 PM, Devdatta Akhawe<[e-mail]>  wrote:
>>>>> The invariant I am talking about is more comparable to an https page
>>>>> including jquery with an http URL, something afaik is considered not
>>>>> safe and blocked by browsers.
>>>>>
>>>>> -devdatta
>>>>>
>>>>> On 20 July 2011 13:24, Adam Barth<[e-mail]>  wrote:
>>>>>> I'm not sure that invariant makes sense.  As another example, it
>>>>>> seems entirely reasonable for an HTTP page to include a copy of
>>>>>> jQuery from an HTTPS URL.
>>>>>>
>>>>>> Adam
>>>>>>
>>>>>>
>>>>>> On Wed, Jul 20, 2011 at 1:16 PM, Devdatta Akhawe
>>>>>> <[e-mail]>
>>>>>> wrote:
>>>>>>> Hi folks
>>>>>>>
>>>>>>> Consider a site at www.alice.com that wants to only be framed by
>>>>>>> their friends at www.bob.com.
>>>>>>>
>>>>>>> Say, a request to https://www.alice.com might respond with a
>>>>>>> X-Frame-Options: allow-from http://www.bob.com
>>>>>>>
>>>>>>> Clearly, the https://www.alice.com has the privileges to act with
>>>>>>> the 'secure' cookie for alice.com. In this scenario,
>>>>>>> http://www.bob.com might actually be MITM'ed by Mallory and
>>>>>>> contain malicious code. In this scenario, does it make sense to
>>>>>>> allow http://www.bob.example to frame https://www.alice.example?
>>>>>>> I think this is wrong behavior: a more higher level invariant
>>>>>>> that should be maintained (at least in the newer specs
>>>>>>> :) is
>>>>>>> that only HTTPS content has access to secure cookie privileges.
>>>>>>>
>>>>>>> Thus, I think the right thing to do is :
>>>>>>> Enforce https for all the origins in the list returned in
>>>>>>> allow-from by https://www.alice.com. Even if
>>>>>>> https://www.alice.com responds with http://www.bob.com in its
>>>>>>> X-Frame-Options, the browser should only allow
>>>>>>> https://www.bob.com to frame https://www.alice.com
>>>>>>>
>>>>>>>
>>>>>>> I think this is even more compelling in case alice.com has
>>>>>>> enforced HSTS.
>>>>>>>
>>>>>>> What do others think ?
>>>>>>>
>>>>>>>
>>>>>>> thanks
>>>>>>> devdatta
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> websec mailing list
>>>>>>> [e-mail]
>>>>>>> https://www.ietf.org/mailman/listinfo/websec
>>>>>>>
>>>>>>>
>>>>>
>>>>
> _______________________________________________
> websec mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/websec