360529 - Arbitrary code execution using XSS hole and feed preview page

Status: RESOLVED FIXED

(Firefox Graveyard :: RSS Discovery and Preview, defect, P2)

Description

[moz_bug_r_a4]

Please see bug 353266.

Array.prototype methods trick (bug 344495) and document.(open|write) trick (bug
346659) and XSS holes (bug 351370, bug 359137) are available on fx2.0. (bug
359137 was already fixed.)

In a feed preview page, an event handler that was registered by content script
can be called by chrome script via elem.doCommand() or elem.dispatchEvent(). 
In this case, the event handler's scripted caller is chrome script.  Thus, an
attacker can run arbitrary code with chrome privileges by using Array.prototype
methods trick or document.write trick.

--
  document.getElementById("handlersMenuList")
   .addEventListener("command", x, true);

  SubscribeHandler._feedWriter.write(window);
   or
  SubscribeHandler._feedWriter
   .QueryInterface(Components.interfaces.nsIObserver)
   .observe({}, "nsPref:changed", "browser.feeds.handler.default");

x() (if x is a function) or x.handleEvent() is called in
FeedWriter._setSelectedHandler() via handlers[0].doCommand(),
selectedAppMenuItem.doCommand() or liveBookmarksMenuItem.doCommand().

  document.getElementById("alwaysUse")
   .addEventListener("CheckboxStateChange", y, true);

  SubscribeHandler._feedWriter.write(window);
   or
  SubscribeHandler._feedWriter
   .QueryInterface(Components.interfaces.nsIObserver)
   .observe({}, "nsPref:changed", "browser.feeds.handler");

y() or y.handleEvent() is called in FeedWriter._setCheckboxCheckedState() via
aCheckbox.dispatchEvent(event).
--

* Fx1.5.0.x is not affected since it does not have feed preview feature.
* The trunk should be exploitable if an XSS hole existed.

Comment 3

[moz_bug_r_a4]

The old testcases no longer work, since those try to use the XSS hole (bug
351370) that has been already fixed.  I'm attaching new testcases that use
another XSS hole (Bug 363597).

Comment 6

[moz_bug_r_a4]

Is there any chance this can be fixed without relying on fixing bug 344495?

Comment 7

[Martijn Wargers (dead)]

Is this still a problem in current trunk/branch build?

Comment 8

[moz_bug_r_a4]

Yes, at least on 1.8 branch.  It's still possible to use the feed preview page
to turn an XSS bug into an arbitrary code execution bug.

Comment 11

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

Repeating the question:
(In reply to comment #7)
> Is this still a problem in current trunk/branch build?

If this is branch only, could the bug be marked as gecko-1.8branch/ff-2.0branch?

Comment 12

[Daniel Veditz [:dveditz]]

Whether or not it's a trunk problem depends on whether there are any known universal-xss vulnerabilities on the trunk, or will be any found (or made) in the future. We've gone to a lot of trouble to make sure content can't open any chrome-privileged URIs precisely because we didn't trust that there were no such holes, but being able to open frames on arbitrary feeds means that web content _can_ still open chrome-privileged frames.

The only true solution is to make the Feed preview non-privileged. The privileged part could be a separate chrome-privileged sheet-like thing much like an oversized infobar.

Comment 13

[Mike Beltzner [:beltzner, not reading bugmail]]

Not gonna block Firefox 3 on this for now, but if you find that potential XSS vuln or figure out a fix for branch, please renom.

Comment 14

[Samuel Sidler (old account; do not CC)]

sayrer, is this something you can look at?

Comment 15

[Robert Sayre]

(In reply to comment #14)
> sayrer, is this something you can look at?
> 

Mano dealt with this stuff last time, and came up with the Fx2 solution. I am really busy with other stuff--a summary of the attributes and elements used for the attack would be much appreciated.

Comment 16

[Samuel Sidler (old account; do not CC)]

Mano, do you have time to look at this in the next week? We're about a week away from our branch freeze.

Comment 17

[Daniel Veditz [:dveditz]]

(In reply to comment #13)
> Not gonna block Firefox 3 on this for now, but if you find that potential XSS
> vuln or figure out a fix for branch, please renom.

We find XSS fairly regularly, it needs to get fixed.

Comment 18

[Asaf Romano (gone)]

Attachment: patch

Comment 19

[Mike Beltzner [:beltzner, not reading bugmail]]

Comment on attachment 306777 [details] [diff] [review]
patch

a1.9b4=beltzner

Comment 20

[Asaf Romano (gone)]

mozilla/browser/components/feeds/src/FeedWriter.js 1.53

Comment 21

[Samuel Sidler (old account; do not CC)]

Mano, does this same patch apply to the branch? If so, can you request approval? If not, a branch-specific one would be great.

Comment 22

[Asaf Romano (gone)]

Attachment: untested branch patch

I don't have a tree to this one on.

Comment 23

[moz_bug_r_a4]

It seems that appendChild(), removeAttribute(), etc. are also exploitable,
since those methods fire mutation events.

Comment 25

[Brendan Eich [:brendan]]

Followup bug would be better unless the landed patch is backed out.

/be

Comment 26

[Asaf Romano (gone)]

Let's keep using this one.

Comment 27

[Samuel Sidler (old account; do not CC)]

This needs a complete trunk patch before we can take it on the branch. Punting to the next release.

Comment 28

[Asaf Romano (gone)]

Mike and I agreed to push this to early-post-beta given the scale of my upcoming rewrite.

Comment 29

[Asaf Romano (gone)]

So, after thinking about this a little more, I'm going to just rely on the fact that DOMAttrModified and other mutation events are not being dispatched for nodes which are not yet inserted into the document (and call appendChild within the sandbox).

Comment 30

[Asaf Romano (gone)]

In the js console:
var a=function() { alert('foo'); }; document.addEventListener("DOMAttrModified", a, true); var b = document.createElement("b"); b.setAttribute("f", "b"); document.documentElement.appendChild(b); b.setAttribute("m", "f");

alerts only for the second setAttribute call. Boris, Jonas,  Johnny: Can the feedwriter code rely on this? W3C's documentation is pretty vague about mutation events behavior for nodes which are not yet inserted into the document.

Comment 31

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

Mutation events do get dispatched, but in your testcase the event can't 
propagate to document, because element isn't in document.
But if you add mutation event listener to the element, the listener will be
called.

Comment 32

[Asaf Romano (gone)]

I'm fine with that, we create the elements and set the attributes on the same time. Would it be enough than to call appendChild and all  _post_ node-insertion, attribute-modifications methods  in the sandbox?

Comment 33

[Asaf Romano (gone)]

Attachment: wip

Comment 34

[Asaf Romano (gone)]

Attachment: patch

Comment 35

[Mike Connor [:mconnor]]

Comment on attachment 311175 [details] [diff] [review]
patch

>-      feedTitleLink.setAttribute("title", titleText);
>+      this._contentSandbox.feedTitleLink = feedTitleLink;
>+      this._contentSandbox.titleText = titleText;
>+      var codeStr = "feedTitleLink.setAttribute('title', titleText);";
>+      Cu.evalInSandbox(codeStr, this._contentSandbox);
>       this._safeSetURIAttribute(feedTitleLink, "href", 
>                                 parts.getPropertyAsAString("link"));

can null these after to clean up

>   _setSubscribeUsingLabel: function FW__setSubscribeUsingLabel() {
>-    var stringLabel = null;
>-
>-    switch (this._getFeedType()) {
>+    var stringLabel;
>+    var feedType = this._getFeedType();
>+    switch (feedType) {

you don't use feedType elsewhere, don't understand the change here.

>       case Ci.nsIFeed.TYPE_VIDEO:
>         stringLabel = "subscribeVideoPodcastUsing";
>         break;
> 
>       case Ci.nsIFeed.TYPE_AUDIO:
>         stringLabel = "subscribeAudioPodcastUsing";
>         break;
> 
>       default:
>         stringLabel = "subscribeFeedUsing";
>     }

you could just init stringLabel to this and it'd be a little less code overall, but that's being nitpicky.

r=me, this should be pretty safe, just needs to get tested thoroughly by QA tomorrow...

Comment 36

[Mike Connor [:mconnor]]

Comment on attachment 311175 [details] [diff] [review]
patch

a1.9b5 on this patch, as its fixing an sg:critical bug and has potential regression impact in the wild.

Please make sure that the existing litmus tests cover the functionality we're touching here, and send mail to [e-mail] asking that they be run ASAP on all platforms.  If there are missing tests, please tell QA what the tests should cover.

Comment 37

[Shawn Wilsher :sdwilsh]

It seems to me that this wouldn't be too hard to automate testing on? (not that I'm volunteering - my plate is plenty full)

The testcases seem to basically be already written (some modification so they don't use an alert box maybe, but that shouldn't be too hard either...)

Comment 38

[Asaf Romano (gone)]

No, you cannot test this, we generally fix XSS holes :p

Comment 39

[Shawn Wilsher :sdwilsh]

right but, if we can exploit said vulnerability in a test, the test should fail, right?

Comment 40

[Asaf Romano (gone)]

But we cannot. As far as i can tell, none of the tests attached here can be used to test this bug on current trunk.

Comment 41

[Boris Zbarsky [:bzbarsky]]

XSS-like conditions can be created by code with UniversalXPConnect privileges, no?

Comment 42

[Asaf Romano (gone)]

I couldn't find a way (even with privileges enabled) to call about:feed's SubscribeHandler.init on trunk without using the console / urlbar as described in testcase 6. 

Comment 43

[Asaf Romano (gone)]

Attachment: for checkin

I'll write some tests later as we figure out how to do so...

Comment 44

[Asaf Romano (gone)]

mozilla/browser/components/feeds/src/FeedWriter.js 1.54

Mail sent to qa@m.c.

Comment 45

[Boris Zbarsky [:bzbarsky]]

> without using the console / urlbar as described in testcase 6. 

So if you inject code into the Firefox window that calls eval(content.code), it doesn't work?

Comment 46

[Asaf Romano (gone)]

Firefox window as in browser.xul? I guess that might work, though it requires the chrome-testing framework....

Comment 47

[Boris Zbarsky [:bzbarsky]]

Why?  You can do that from inside a mochitest if you really want to...

Doing it from chrome mochitests might work as long as you can get the feed preview page loaded first.

Comment 48

[Samuel Sidler (old account; do not CC)]

Is there any work being done here for an updated branch patch, presumably one that also includes a fix for bug 430658?

Comment 49

[Samuel Sidler (old account; do not CC)]

Still don't have a branch patch and given the size of the trunk patch we don't want to take it at the last minute. Moving to 1.8.1.16, but we need to get a patch soon so we can land it early in the cycle.

Comment 52

[Samuel Sidler (old account; do not CC)]

Mano, we could really use a branch patch here...

Comment 53

[Asaf Romano (gone)]

Attachment: combined branch patch

Comment 54

[Mike Connor [:mconnor]]

Comment on attachment 334770 [details] [diff] [review]
combined branch patch

This feels like pain.

Comment 55

[Daniel Veditz [:dveditz]]

qanote: testcase 7 can't be used to verify because the stepping-stone bug was fixed in FF2.0.0.15. Either we need a 2.0.0.16-working testcase that uses another yet-unfixed XSS bug, or this patch needs to be verified by applying the patch to a 2.0.0.14 source tree.

Comment 56

[Daniel Veditz [:dveditz]]

Comment on attachment 334770 [details] [diff] [review]
combined branch patch

It'd be good to get a second review given the size of the patch and the limited testing time we've got left for this release. Johnny: you reviewed an earlier iteration, you up for it?

Comment 58

[Samuel Sidler (old account; do not CC)]

Comment on attachment 334770 [details] [diff] [review]
combined branch patch

Approved for 1.8.1.17. Please land ASAP on the branch. a=ss

Comment 59

[Asaf Romano (gone)]

Checking in src/FeedWriter.js;
/cvsroot/mozilla/browser/components/feeds/src/FeedWriter.js,v  <--  FeedWriter.js
new revision: 1.2.2.33; previous revision: 1.2.2.32
done

Comment 60

[moz_bug_r_a4]

The fx2 patch caused a syntax error on Windows.

Here is an extra "}".
http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/browser/components/feeds/src/FeedWriter.js&rev=1.2.2.33&mark=788#782


Error: missing catch or finally after try
Source File: file:///C:/test/fx-2.0.0.17pre-2008-08-26-03/firefox/components/FeedWriter.js
Line: 736, Column: 6
Source Code:
      else {

Error: BrowserFeedWriter is not defined
Source File: chrome://browser/content/feeds/subscribe.js
Line: 45

Comment 61

[Mike Shaver (:shaver emeritus)]

Preeeetty sure we need the syntax error fixed before we can ship 2.0.0.17, so unmarking it as branch-fixed until that's addressed.

How did this code ever work in testing the #ifdef XP_WIN block?  We really, *really* can't be checking untested code into the stable branches, so I'm very much hoping that something else happened. :(

Comment 62

[Asaf Romano (gone)]

Bad merge :-/

Comment 63

[:Gavin Sharp [email: [e-mail]]]

With that fixed I get:
Error: uncaught exception: [Exception... "'[JavaScript Error: "result has no properties" {file: "file:///I:/mozb/mozilla/ff-opt/dist/bin/components/FeedWriter.js" line: 381}]' when calling method: [nsIFeedWriter::write]"  nsresult: "0x80570021 (NS_ERROR_XPC_JAVASCRIPT_ERROR_WITH_DETAILS)"  location: "JS frame :: chrome://browser/content/feeds/subscribe.js :: SH_init :: line 46"  data: yes]

with all of the testcases, along with:
Security Error: Content at about:feeds may not load or link to chrome://global/content/mozilla.xhtml.

I assume this is the expected failure (I get the same on trunk).

Comment 64

[:Gavin Sharp [email: [e-mail]]]

mozilla/browser/components/feeds/src/FeedWriter.js 	1.2.2.34 

Comment 65

[juan becerra [:juanb]]

Verified on Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.17) Gecko/2008082909 Firefox/2.0.0.17 using test cases here. This is the output in the error console while running test case 8, in comment #57:

Error: result has no properties
Source File: file:///C:/Program%20Files/Mozilla%20Firefox/components/FeedWriter.js
Line: 354

Error: uncaught exception: [Exception... "'[JavaScript Error: "result has no properties" {file: "file:///C:/Program%20Files/Mozilla%20Firefox/components/FeedWriter.js" line: 354}]' when calling method: [nsIFeedWriter::write]"  nsresult: "0x80570021 (NS_ERROR_XPC_JAVASCRIPT_ERROR_WITH_DETAILS)"  location: "JS frame :: chrome://browser/content/feeds/subscribe.js :: SH_init :: line 46"  data: yes]

Error: [Exception... "Component returned failure code: 0x80004005 (NS_ERROR_FAILURE) [nsIDOMLocation.href]"  nsresult: "0x80004005 (NS_ERROR_FAILURE)"  location: "native frame :: <unknown filename> :: <TOP_LEVEL> :: line 0"  data: no]

Comment 66

[Alexander Sack]

doesnt apply on 1.8.0 from what i can see. If i am mistaking flip wanted1.8.0.x or blocking1.8.0.15 to ? again. Thanks!

Comment 67

[juan becerra [:juanb]]

In 20015 test case #8 opens up an alert window, and in 20017 it doesn't.

Comment 70

[:Gavin Sharp [email: [e-mail]]]

moz_bug, could you file a new bug on the omissions? Tracking multiple patches in one bug just leads to confusion.

Comment 71

[Martijn Wargers (dead)]

And it seems to me that should block the fx2.0.0.17 release, right?

Comment 72

[Mike Beltzner [:beltzner, not reading bugmail]]

(In reply to comment #70)
> moz_bug, could you file a new bug on the omissions? Tracking multiple patches
> in one bug just leads to confusion.

I have been convinced that's the right thing to do, instead of continually expanding the scope of this single bug. We'll leave this part of the issue as fixed & verified, and stamp out the others in future releases.

(In reply to comment #71)
> And it seems to me that should block the fx2.0.0.17 release, right?

No, we've waited before, and can continue to do so again.

Comment 75

[Asaf Romano (gone)]

CC me on it please if you wanst me to fix it.