91043 - document.write while another page is loading can bypass same-origin check

Status: RESOLVED FIXED

(Core :: Security, defect, P1)

Description

[Jesse Ruderman]

document.write is currently allAccess.  If an attacker can time a document.write
to fire while an HTML document from another site is loading, he can insert
whatever code he wants into the other document, and Mozilla will assume that
code came from the other site.

Possible fixes:

1. Make document.write be sameOrigin.  Web scripts will then have to explicitly
call document.open if they want to write a new document in a window they don't
currently own.

2. Make the document.write code check who's calling it.  If the caller's
principal doesn't match the target document's principal, call document.open to
create a new document owned by the attacker, or deny access.

I vote for 1, because it's simpler and seems less likely to lead to confusing
problems.  Note that IE doesn't allow access to document.write *or*
document.open in another domain.

Comment 1

[Jesse Ruderman]

Attachment: demo

Comment 2

[Bradley Baetz (:bbaetz)]

document.write has code to stop this - see bug 32088, revision 3.207 in
nsHTMLDocument.cpp. Why is this failing?

Can this be turned into a universalXPConnect vulnerabilty in combination with
bug 88087? I couldn't trigger it, but I'll try on a 133Mhz machine tomorrow.

Comment 3

[Jesse Ruderman]

Netscape 4 allows access to document.write across domains, but IE doesn't.  So 
if we blocked access to document.write across domains, we would be breaking 
compatibility only with Netscape 4.

By the way, Netscape 4 crashes if I try to test for this security hole.  I guess 
it isn't vulnerable...

Comment 4

[Mitchell Stoltz (not reading bugmail)]

Attachment: Patch - make document.write(ln) same-origin access only

Comment 5

[Mitchell Stoltz (not reading bugmail)]

Attachment: Patch 2 - ignore the last one.

Comment 6

[Jesse Ruderman]

r=jesse

Comment 7

[John Bandhauer]

sr=jband

Comment 8

[Mitchell Stoltz (not reading bugmail)]

Checked in on trunk. Adding vtrunk. I would like to get this onto the branch.

Comment 9

[selmer (gone)]

ckritzer: pls verify on trunk builds when available.

Comment 10

[selmer (gone)]

PDT+ per pdt mtg w/mstoltz

Comment 11

[Mitchell Stoltz (not reading bugmail)]

Fixed on branch.

Comment 12

[ckritzer (gone)]

Okay, this bug is VERIFIED FIXED on:

Win98SE 2001-07-19-09-trunkMacOS91 2001-07-19-08-trunk
LinRH62 2001-07-19-08-trunk

And since Mitch checked into the branch tonight, I will verify tomorrow's branch
bits when they are available.

Comment 13

[ckritzer (gone)]

Marking VERIFIED FIXED on:
MacOS91 2001-07-20-06-0.9.2
Win98SE 2001-07-20-03-0.9.2
LinRH62 2001-07-20-04-0.9.2

Comment 14

[Gayatri Rimola]

Nominating since this is a regression from Mach V.

Comment 15

[Daniel Veditz [:dveditz]]

Seems to crash in layout now, after the exploit has already run.

Comment 16

[Michael Buckland]

Mitch can you say if you think this is required for Blackbird?

Comment 17

[Mitchell Stoltz (not reading bugmail)]

Yes, I think this should be required for Blackbird.

It looks like I solved the problem originally by making
document.open/write/writeln same-origin access only. Then, in bug 122317, I made
them allAccess again, because the restriction was breaking some convoluted JS
usage in a Siebel app - I had forgotten about the security implications at the
time (doh).

The quick, easy fix is to check in the above patch again, which simply adds a
few lines to the security prefs. However, that will break the Siebel app again.
Security is more important in this case, but I would at least like to explore
some alternatives before we break them again. If I can't get any information
before the end of the day, I'll change the prefs back.

Comment 18

[Mitchell Stoltz (not reading bugmail)]

OK, the Siebel issue has a quick workaround - we will work with Evangelism to
see that they get the message. In the menatime, let's check in the above patch
again. It removes the security prefs for document.write and writeln, which will
make them default to sameOrigin.

Can I get r/sr for removing those prefs?

Comment 19

[Mitchell Stoltz (not reading bugmail)]

We should probably get this into 1.2 as well...

Comment 20

[Asa Dotzler [:asa]]

Comment on attachment 42806 [details] [diff] [review]
Patch 2 - ignore the last one.

a=asa for checkin to the 1.2 branch (on behalf of drivers)

Comment 21

[Daniel Veditz [:dveditz]]

adt approval for the 1.0.2 branch, please get drivers approval and check in tonight.

Comment 22

[Daniel Veditz [:dveditz]]

Comment on attachment 42806 [details] [diff] [review]
Patch 2 - ignore the last one.

sr=dveditz

Comment 23

[Mitchell Stoltz (not reading bugmail)]

verbal a=chofmann for 1.0 branch.

Comment 24

[Daniel Veditz [:dveditz]]

Mitch checked this into the 1.0 branch earlier tonight.

Comment 25

[bsharma]

Verified on 2002-11-14-branch build.
The crash is still happening. The crash happens after a while now but it still
consistently happens.

    Steps to reproduce:
    1. Open the demo page from the bug report.
    2. Click on the first button. This will open the new window with large file,
which will keep on loading.
    3. Come back to the demo page and click on the next two buttons one after
the other, nothing happens.
    4. Go back to that large page and it is still loading. Come back to the demo
page and click on the last two buttons again for couple of time. The crash will
happen after 10 seconds or so.

    Please let me know if you need more details.

Comment 26

[Jimmy Lee]

Removing fixed1.0.2 keyword to get on radar.  I can reproduce what Bindu is seeing.

Comment 27

[Gayatri Rimola]

Mitch, can you please give us a test case which does not use innerHTML since
that is crashing and Bindu can't verify the fix with confidence. Since the bug
has been re-opened in the branch, add back the fixed1.0.2 keyword when you
provide the test case. Thanks.

Comment 28

[Sivakiran Tummala]

Does not look it is a problem with innerHTML, tested with build:
2002-11-14-09-1.0 Attaching a testcase to demonstate innerHTML is working
properly. And i could not reproduce the crash.

Comment 29

[Sivakiran Tummala]

Attachment: testcase showing innerHTML is working fine

Comment 30

[Mitchell Stoltz (not reading bugmail)]

I think the crash is a separate bug, independent of the security problem, and
not a stop-ship. harishd is working on the crash and has filed a separate bug.
For now, we can ignore the crash. What to look for is the alert() produced by
the testcase. The testcase should not produce an alert(), but rather an
exception in the JS console. If that's the case, the security bug is fixed. I'm
putting this back to fixed1.0.2; please call me if you're still having trouble
verifying.

Comment 31

[bsharma]

Verified on 2002-11-14-branch build on Win2000.

The demo page loads with the exception for each document.write in the JS console.

Comment 32

[Ben M.]

This is odd... The browser security check at
http://bcheck.scanit.be/bcheck/index.php is saying my copy of Mozilla 1.3 still
has this vurnerability.
Is it wrong?

Comment 33

[Marek Stępień [:marcoos, inactive]]

Ben:
I have 0/0/0 errors on bcheck.scanit.be while using yesterday's nightly:
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4a) Gecko/20030327

worksforme.

Comment 34

[Flo C]

Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.3) Gecko/20030312
has the vulnerability according to the test at
http://bcheck.scanit.be/bcheck/index.php 

Still there with the latest nightly build of Phoenix:
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.4a) Gecko/20030409 Phoenix/0.5+

Comment 35

[Lucky Pozzo]

I went through the browser security test at
http://bcheck.scanit.be/bcheck/index.php and it showed this bug. Using Mozilla
1.3 on Redhat 8.0

Comment 36

[Sam Hulick]

confirmed.. this security issue is present on builds as late as 2003041704

Comment 37

[Johnny Stenback (:jst)]

Reopening to make sure this gets re-checked.

Comment 38

[Mitchell Stoltz (not reading bugmail)]

Or we could just wait for Godot, Lucky :)

Comment 39

[Chris Brooking]

Just tested on Mozilla 1.4b
Mozilla/5.0 (Windows; U; WinNT4.0; en-US; rv:1.4b) Gecko/20030507
(using http://bcheck.scanit.be/bcheck)
bug still present.

Comment 40

[Mitchell Stoltz (not reading bugmail)]

I don't believe this bug is still present. The document.write function is still
same-origin access only, and both the original testcase attached to this bug and
the scanit.be testcase come up negative for me. I can't see the source of the
scanit.be test, since it doesn't stay loaded long enough to grab the source, but
my guess is that the test on that site is flawed.

I'm marking the bug Fixed, but I'd like to get some redundant verification.
Brent or Charles, could you please try the testcase mentioned above, and the
testcase attached to this bug?

Johnny, do you see any suspicious behavior still occurring?

Comment 41

[Alla Bezroutchko]

I think there was a timing condition in http://bcheck.scanit.be/bcheck/ test. It
might have caused false positives. Now I think I fixed it. According to my
testing 1.4b is not vulnerable, neither is 1.3a, or 1.2.1. 

1.1 is vulnerable. Did not test other versions.

Sorry for the confusion caused by the test.

Here is how I test now:

win = window.open("http://otherdomain/test-moz91043bis.php");
// This window contains a document from another domain that takes
// long time to load
var j = setInterval("catchme()", 100);

function catchme() {
        try {
                var a = win.document.title; // Test if the window is
                                         // still in our domain
        } catch (e) { // Window is not in our domain anymore
                clearInterval(j);
                try {
                        win.document.write("stuffstuffstuff");
                        vulnerable();
                } catch (e) {
                        notvulnerable();
                }
        }

}

Comment 42

[tangalanga]

Sorry, i have Mozilla/5.0 (Windows; U; Windows NT 5.1; es-AR; rv:1.9.2a1pre) Gecko/20090708 Minefield/3.6a1pre  and the same bug appeared in http://bcheck.scanit.be/old-bcheck/

Comment 43

[Daniel Veditz [:dveditz]]

Not sure what you're talking about. The (very) old version of the scanit.be test in the "old-bcheck" directory does have a test for this bug, but all recent versions of Mozilla browsers pass those tests.