This is a regression from bug 754202. (fx16,17,18 are affected.)
When a chrome code calls an untrusted function via CrossCompartmentWrapper/WaiveXrayWrapper, GetSubjectPrincipal returns a content principal, but GetPrincipalAndFrame returns the system principal. nsLocation::CheckURL gets an owner principal by calling GetCxSubjectPrincipalAndFrame, which calls GetPrincipalAndFrame.
By using this bug, content can perform an XSS attack, and if there is an extension that exposes a chrome window to content, content can run arbitrary code with chrome privileges.
Created attachment 663341 [details]
testcase 1 - XSS
This tries to get cookies for www.mozilla.org.
Created attachment 663342 [details]
testcase 2 - arbitrary code execution with Error Console
I decided to do this work over in public over in bug 797204, since the security issue at hand is non-obvious.
We've already gone to build with beta6 so the window on 16 landings has closed. Wontfixing for 16.
fixed on trunk in bug 797204
Re-noming for 16 in case we consider this fix for a 16.0.2 "location cleanup" chemspill. In light of bug 801305 though it might be incomplete and/or regression prone; Firefox 17 might be as early as we want to push it.
bug 797204 has the uplift approvals, just need them to land on branches.
(In reply to Lukas Blakk [:lsblakk] from comment #7)
> bug 797204 has the uplift approvals, just need them to land on branches.
Unfortunately we determined that bug 797204 doesn't fix this bug. I've got more targeted patches over in bug 801305.
(In reply to Bobby Holley (:bholley) from comment #8)
> (In reply to Lukas Blakk [:lsblakk] from comment #7)
> > bug 797204 has the uplift approvals, just need them to land on branches.
> Unfortunately we determined that bug 797204 doesn't fix this bug. I've got
> more targeted patches over in bug 801305.
Can we mark FF17-19 as fixed now that bug 801305 has landed?
Created attachment 675226 [details]
Testcase 1 -- XSS (updated)
The original XSS testcase doesn't work because www.mozilla.org now has an X-Frame-Options header to prevent framing. This is functionally equivalent, at least until blog.mozilla.org does the same thing.
I'm downgrading to sec-high because it's not at the moment a known drive-by exploit. addons are already on dangerous ground using unwrapped DOM objects and should not be injecting content that way.
bug 801305 hadn't been checked in on branches when comment 9 was asked, but it has now and I've verified the fix in Nightly and Aurora.