535926 - "ASSERTION: This is unsafe" with <iframe>, XBL

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Jesse Ruderman]

Attachment: testcase

###!!! ASSERTION: This is unsafe!: 'Error', file /Users/jruderman/mozilla-central/content/events/src/nsEventDispatcher.cpp, line 485

This is the assertion just added in bug 531176.

I'm not sure why I need the setTimeout.

Comment 1

[Jesse Ruderman]

Attachment: stack trace

Comment 2

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

So nsBrowserStatusFilter::OnStateChange calls some JS code which end up to
progressmeter.xml which dispatches ValueChange.

In this particular case I wonder why we even call OnStateChange.
I believe the document is already loaded but
#21 0x00007f82d1f3e2ec in nsDocument::BlockOnload (this=0x261b070) at /home/smaug/mozilla/mozilla_cvs/hg/src/content/base/src/nsDocument.cpp:7037
#22 0x00007f82d20e85c8 in nsBindingManager::AddToAttachedQueue (this=0x2158530, aBinding=0x226a5e0)
    at /home/smaug/mozilla/mozilla_cvs/hg/src/content/xbl/src/nsBindingManager.cpp:955
adds new request to load group which triggers the method call.

Investigating...

Comment 3

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

But still, I think we can't easily prevent progress listeners to dispatch
events.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Hmm.  We should avoid calling progress listeners for BlockOnload calls if we can do it...

Maybe we should stop faking the loadgroup thing and just directly have BlockOnload/UnblockOnload poke the docloader or something?

Comment 5

[Bob Clary [:bc] (inactive)]

An example of this assertion can be seen on Windows 1.9.3 at least at http://www.slacker.com/

Comment 6

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

Not sure if this is sg:critical (at least when using the testcase).
That is happening in chrome, and during time we might get dom modifications anyway.

Comment 7

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

I wonder if, at least for now, the assertion should fire only if the
target of the event is in content.

Comment 8

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

Repeating this:
> Not sure if this is sg:critical (at least when using the testcase).
> That is happening in chrome, and during time we might get dom modifications
> anyway.

What in this bug makes this sg:critical ?

Comment 9

[Jesse Ruderman]

I only marked it as sg:critical because the assertion was added in an sg:critical bug.  If this bug isn't a security bug, let's unmark the whiteboard and make it public.

Comment 10

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

Well, I wouldn't remove the security flag from this bug, but
this just may not be sg:critical.

Comment 11

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Olli, if this only results in events being fired at chrome windows, then why not remove the security flag?

Comment 12

[Johnny Stenback (:jst)]

Olli, any more thoughts on whether this needs to be sg:critical, or if it's even a security bug any more?

Comment 13

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

comment 4 says what should be done here, I think.

I think this doesn't need to be sg:critical, since there is no clear evidence
that content can trigger anything bad - unless ff ui or some extension starts
to handle status updates in a different way.
But this is still a security bug.

Comment 14

[Damon Sicore (:damons)]

Looks like we should probably remove the sg:crit rating here, but there's ongoing discussion of what the ration should be.  Thoughts?

Comment 15

[Brandon Sterne (:bsterne)]

If this is a critical vulnerability only exposed to extension code presently, we should lower the severity to sg:high.  If we think it's likely that FF UI code would add changes triggering this vuln, then sg:critical severity is appropriate.

Comment 16

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

Or we can just fix this and not care too much about sg:*. :)

Taking. I'll try to write the patch on Thursday.

Comment 17

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

Um, I can't reproduce this atm.
I've tried on Linux and OSX, with and without html5 parser.

The assertion in this case was changed to a warning, but I can't see that either.

I wonder if bug 557398 changed something here.
Testing...

Comment 18

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

Backing out bug 557398 doesn't seem to affect to this bug.

Comment 19

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

Ah, perhaps LazyFrameCtor changed timing here. Which means this would still
need to be fixed in 1.9.2.

Comment 20

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

Not LazyFrameCtor. Still trying to find a build where I get the error

Comment 21

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

I can reproduce this at least on 1.9.2, when I added the same assertion/warning
what trunk has.
But patch needs to wait at least til tomorrow.

Comment 22

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

Attachment: patch

I posted this to tryserver

Comment 23

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

Comment on attachment 441281 [details] [diff] [review]
patch

Passed tests on tryserver.

Boris, what do you think about this approach?
A bit ugly, but since I think we want this to 1.9.2 too, I didn't
want to make huge changes. And I'm not even sure what would be
the best way to change nsDocLoader or something.
It would be quite hackish to add onloadblocker-type-of-requests
and handle them specially in docloader.
 
This relies on script blockers to be on stack and 
PostUnblockOnloadEvent to handle things truly async.

Comment 24

[Boris Zbarsky [:bzbarsky]]

Would it make sense to do:

  if (!nsContentUtils::IsSafeToRunScript()) {
    ++mAsyncOnloadBlockCount;
    if (mAsyncOnloadBlockCount == 1) {
      nsContentUtils::AddScriptRunner(
         NS_NEW_RUNNABLE_METHOD(nsDocument, this, AsyncBlockOnload));
    }
    return;
  }

and have AsyncBlockOnload call BlockOnload mAsyncOnLoadBlockCount times, so as not to thrash things with lots of scriptrunners?

Having an API on docloader (whose job it is to keep track of when onload should fire) to block onload would be a lot less hacky than having this fake request, btw.  But making it work right might be a pain...

Why is the new block in DoUnblockOnload needed?  How can we get there with mAsyncOnloadBlockCount != 0?

Comment 25

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

(In reply to comment #24)

> Would it make sense to do:
> ...
> and have AsyncBlockOnload call BlockOnload mAsyncOnLoadBlockCount times, so as
> not to thrash things with lots of scriptrunners?
That should work.


> Why is the new block in DoUnblockOnload needed?  How can we get there with
> mAsyncOnloadBlockCount != 0?
If some script runner (which is added before the AsyncBlockOnload script runner, so it runs before it) processes event loop.

Comment 26

[Boris Zbarsky [:bzbarsky]]

Ah, ok.  Do you want to post a patch with my suggestion from the beginning of comment 24?

Comment 27

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

Yeah, I will, in a minute.

Comment 28

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

Attachment: v2

Like this?

Comment 29

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 441519 [details] [diff] [review]
v2

Space after "while", please.

Comment 30

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

Attachment: +space

uh, how did I miss the space :/

Anyway, this is a security bug, so sr is needed.

Comment 31

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 441526 [details] [diff] [review]
+space

Please add documentation on the member describing how it works. Took reading this a few times to figure it out.

And add a comment to ::BlockOnload describing *why* we're entering the if-statement.

Comment 32

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

Attachment: +some comments

Comment 33

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

http://hg.mozilla.org/mozilla-central/rev/6f7405a55a48
http://hg.mozilla.org/mozilla-central/rev/e5a075d4e5ba

Comment 34

[Reed Loden [:reed]]

Just wanting to get this on the wanted/blocking list for future 1.9.2 release, as per earlier comments stating that this also affects 1.9.2.

Comment 35

[LegNeato]

Does this affect 1.9.1 as well?

Comment 36

[Timothy Nikkel (:tnikkel)]

Bug 569674 is another regression from this, we need to figure that out before landing this on branches. Sorry I didn't mark it sooner.

Comment 37

[Timothy Nikkel (:tnikkel)]

Disregard comment 36, I was thinking this was a different bug.

Comment 38

[LegNeato]

Comment on attachment 441743 [details] [diff] [review]
+some comments

a=LegNeato for 1.9.2.6 and 1.9.1.11. Please land this on mozilla-1.9.2 default and mozilla-1.9.1 default.

Code freeze is this Friday @ 11:59 pm PST.

Comment 39

[Reed Loden [:reed]]

http://hg.mozilla.org/releases/mozilla-1.9.2/rev/336a77f067c8
http://hg.mozilla.org/releases/mozilla-1.9.1/rev/866db4f8b8f3

Comment 40

[Reed Loden [:reed]]

http://hg.mozilla.org/releases/mozilla-1.9.2/rev/2c5245d56edb
http://hg.mozilla.org/releases/mozilla-1.9.1/rev/e2498d527265

Comment 41

[Al Billings [:abillings - ex-MoCo]]

Verified for 1.9.2 with Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2.7pre) Gecko/20100630 Namoroka/3.6.7pre ( .NET CLR 3.5.30729) using the attached testcase.

Comment 42

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Comment on attachment 441743 [details] [diff] [review]
+some comments

Requesting approval1.9.0.next on this patch so that we can take it in upcoming Camino 2.0.x security and stability updates.  If approved, I'll handle the checkins, unless the patch author requests otherwise.

Comment 43

[Daniel Veditz [:dveditz]]

Comment on attachment 441743 [details] [diff] [review]
+some comments

Approved for 1.9.0.20, a=dveditz

Comment 44

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Checking in content/base/src/nsDocument.cpp;
/cvsroot/mozilla/content/base/src/nsDocument.cpp,v  <--  nsDocument.cpp
new revision: 3.837; previous revision: 3.836
done
Checking in content/base/src/nsDocument.h;
/cvsroot/mozilla/content/base/src/nsDocument.h,v  <--  nsDocument.h
new revision: 3.381; previous revision: 3.380
done

Comment 45

[Jesse Ruderman]

Crashtest: http://hg.mozilla.org/mozilla-central/rev/524a5670133b