705651 - (CVE-2012-0446) Frame scripts that access untrusted objects are exploitable

Status: VERIFIED FIXED

(Core :: Security, defect)

Description

[moz_bug_r_a4]

Frame scripts run on the special JS context for which we call SetSecurityManagerForJSContext with flags=0, thus if a frame script calls into an untrusted function, XPConnect does not do proper security checks.

Comment 4

[moz_bug_r_a4]

Attachment: screenshot of testcase 3

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

Nice :(

mrbkap, what kinds of flags can be passed to SetSecurityManagerForJSContext?
There is no documentation in nsIXPConnect nor in nsXPConnect.cpp

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

I'm pretty sure 0 is passed because it is (or was) passed also here
http://mxr.mozilla.org/mozilla2.0/source/dom/src/threads/nsDOMThreadService.cpp#1075

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

Ahaa, some documentation is here
http://mxr.mozilla.org/mozilla-central/source/js/xpconnect/idl/nsIXPCSecurityManager.idl#61
Should I pass HOOK_ALL ?

Comment 8

[Blake Kaplan (:mrbkap) (inactive)]

Hey Olli. I don't think you actually need to call SetSecurityManagerForJSContext at all. That function is useful if a given context needs to use a different security manager from the rest of the runtime. However, in most cases, we can just use the one that's installed on XPConnect itself. On the other hand, using HOOK_ALL would also fix this bug.

Comment 9

[Olli Pettay [:smaug][[e-mail]]]

Attachment: patch

Create a helper method.

Comment 10

[Blake Kaplan (:mrbkap) (inactive)]

Comment on attachment 577535 [details] [diff] [review]
patch

Review of attachment 577535 [details] [diff] [review]:
-----------------------------------------------------------------

I mentioned on IRC that the reason that web workers got away with this (before the recent rewrite) is that they truly used object capabilities to protect themselves. Furthermore, they couldn't use the script security manager off the main thread, and therefore wanted to *prevent* calls to the SSM. Because there was nothing around to protect, this was safe.

::: content/base/src/nsFrameMessageManager.cpp
@@ +840,5 @@
> +
> +  JSAutoRequest ar(cx);
> +  nsIXPConnect* xpc = nsContentUtils::XPConnect();
> +  const PRUint32 flags = nsIXPConnect::INIT_JS_STANDARD_CLASSES |
> +                         /*nsIXPConnect::OMIT_COMPONENTS_OBJECT ?  |*/

I doubt you want this flag. It's only useful if you want to cut the scripts you're going to be running off from XPCOM entirely.

Comment 11

[Olli Pettay [:smaug][[e-mail]]]

Yeah, that flag has been there since the beginning when it wasn't sure whether we want to
expose components to frame scripts or let them do only simpler things.
I'll remove that.

Comment 12

[Olli Pettay [:smaug][[e-mail]]]

Attachment: up-to-date

Comment 13

[Olli Pettay [:smaug][[e-mail]]]

https://hg.mozilla.org/mozilla-central/rev/5ce8d4e02e81

Comment 14

[Olli Pettay [:smaug][[e-mail]]]

I'll prepare patches for aurora and beta.

Comment 15

[Olli Pettay [:smaug][[e-mail]]]

Attachment: for aurora

The only difference to trunk is that trunk doesn't have
| JSOPTION_JIT

Comment 16

[Olli Pettay [:smaug][[e-mail]]]

In beta there is also JS_SetScriptStackQuota and  JSOPTION_ANONFUNFIX |

Comment 17

[Alex Keybl [:akeybl]]

Now tracking for FF9/10, and added the sec-review-needed keyword. Please address possible risk for taking on aurora/beta.

Comment 18

[Daniel Veditz [:dveditz]]

There is little risk here, we should take this for aurora (fx10)

Comment 19

[Daniel Veditz [:dveditz]]

little risk from the patch, I mean! It's a bad security hole that's probably exploitable through several addons and definitely exploitable through an XSS on AMO.

Comment 20

[Olli Pettay [:smaug][[e-mail]]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/677dfd331c5b

Comment 21

[u279076]

Verified fixed in Firefox 11.0b6.

Comment 22

[Jason Smith [:jsmith]]

Verified fixed in Firefox Aurora.

Comment 23

[Jason Smith [:jsmith]]

Verified fixed in Firefox Nightly.