Support for X-Frame-Options: Allow-From [uri]

• Component: Frames
• Status: RESOLVED FIXED

[Phil Ames]

Related to https://bugs.webkit.org/show_bug.cgi?id=23907 (hence the initial CC's)

Attached is a patch which enables support for “Allow-From [uri]” notation in the X-Frame-Options header.  I compared the behavior of patched Chrome with IE9 with the below values (syntax:IE permitted/patched Chrome permitted/header-value) which was sent by a cross-origin site (the top origin was http://www.intra.net/)

Probably the most substantial difference is that IE doesn’t seem to regard the port number.  A similar patch for Firefox which I will be submitting soon (to https://bugzilla.mozilla.org/show_bug.cgi?id=690168) produces the same behavior as this patch to Chrome, though.

N/N/X-Frame-Options: Allow-From
N/N/X-Frame-Options: Allow-From xyz
N/N/X-Frame-Options: Allow-From www.intra.net
Y/Y/X-Frame-Options: Allow-From http://www.intra.net
Y/Y/X-Frame-Options: Allow-From http://www.intra.net/
Y/Y/X-Frame-Options: Allow-From http://www.intra.net:80/
Y/N/X-Frame-Options: Allow-From http://www.intra.net:81/
Y/N/X-Frame-Options: Allow-From http://www.intra.net:443/
Y/Y/X-Frame-Options: Allow-From http://www.intra.net/foo/bar/

Please also note that although I modelled the LayoutTests after the ones that I found in the existing tree, I was not actually able to get them to run on my development environment, so they may require some tweaking (sorry, if anyone is at Google and wants to spend a few minutes helping me troubleshoot / get it working, I’d be happy to test and send an additional diff). The results in the table above are solely from ‘empirical’ testing/observation.

[Phil Ames]

Created attachment 160219 [details]
Patch

[Phil Ames]

[trying to cc myself]

[Adam Barth]

Note: We're still discussing this feature in the IETF websec working group.  Generally, our position has been that we shouldn't be adding new features to X-Frame-Options.  Instead, we should create a CSP directive that satisfies these use cases.

[Phil Ames]

Created attachment 161116 [details]
Updated patch.

FYI: I managed to get the layout tests passing, and also addressed that this would accept something like "Allow-Fromhttp://foo.com" instead of "Allow-From http://foo.com".

[Adam Barth]

@Phil: I appreciate your enthusiasm, but it's unlikely that we'll accept this patch until the standards issues are resolved.

[Phil Ames]

Understood, I just figured it made sense to have something up-to-date and attached here rather than something broken.

[Phil Ames]

Created attachment 197858 [details]
X-F-O Allow From implementation

I noticed this code had been refactored, so I've updated the patch.

[Brady Eidson]

In radar as <rdar://problem/13658368>

I'll try to take a look at this patch soon.

[Adam Barth]

(In reply to comment #8)
> In radar as <rdar://problem/13658368>
> 
> I'll try to take a look at this patch soon.

You should be aware that this feature is controversial in the W3C WebAppSec working group.  Before implementing it, you might want to confer with the working group.

[Brady Eidson]

(In reply to comment #9)
> (In reply to comment #8)
> > In radar as <rdar://problem/13658368>
> > 
> > I'll try to take a look at this patch soon.
> 
> You should be aware that this feature is controversial in the W3C WebAppSec working group.  Before implementing it, you might want to confer with the working group.

Interesting.

We were looking at it because it's been in the IETF draft for some time (http://tools.ietf.org/html/draft-ietf-websec-x-frame-options-01) and has utility we're interested in.

I'm failing at finding relevant threads in the webappsec archives that demonstrate this controversy.  Have any pointers?

[Adam Barth]

> We were looking at it because it's been in the IETF draft for some time (http://tools.ietf.org/html/draft-ietf-websec-x-frame-options-01) and has utility we're interested in.

The IETF websec working group has not addressed my WGLC feedback on this aspect of draft-ietf-websec-x-frame-options:

http://www.ietf.org/mail-archive/web/websec/current/msg01459.html

> I'm failing at finding relevant threads in the webappsec archives that demonstrate this controversy.  Have any pointers?

I couldn't find a good email thread in quick search.  The basic issue is that allow-from is basically the same thing as a source-expression from CSP but uses an incompatible syntax.

The current agreement in both the IETF websec working group and the W3C WebAppSec working group is to not add any new features to X-Frame-Options (including allow-from) and instead make frame-options into a CSP directive.

The final point of controversy is whether to let the web site specify multiple values for the frame-options directive (i.e., whether to allow a source-list like other CSP directives or whether to restrict frame-options to a single source-expression).

I'd recommend not implementing allow-from in X-Frame-Options until these issues are resolved.  I'd expect the likely outcome to be a frame-options CSP directive that either takes a source-list or a source-expression.

The W3C WebAppSec working group has a face-to-face meeting on April 25-26 where I'd expect these issues to be hammered out.  If you're interested in these topics, I'd encourage you to attend.

[Brady Eidson]

I'll try to be on IRC.

[Adam Hooper]

Has the standards process stalled here?

In October 2013, IETF published http://tools.ietf.org/html/rfc7034 (informative) related to X-Frame-Options.

Meanwhile, the CSP working draft shows nothing but a TODO: https://dvcs.w3.org/hg/content-security-policy/raw-file/5b353a8ac072/csp-specification.dev.html#frame-options--experimental

Allow-From solves a real problem. Because WebKit ignores it, websites can be forced to abandon X-Frame-Options altogether. Given there's an RFC on the topic, could we please have Allow-From?

[Alexey Proskuryakov]

CSP 1.1 now supports this use case, directive name is frame-ancestors. It's not implemented in WebKit yet.

[Mike Hemesath]

Is there an issue logged for implementing frame-ancestors in webkit?

[Brent Fulgham]

(In reply to comment #15)
> Is there an issue logged for implementing frame-ancestors in webkit?

Yes. This work was done under Bug 154345, and is now complete.

[Brent Fulgham]

Since the new CSP feature has been implemented, this issue is resolved.