657292 - Firefox 6.0a1 Crash Report [@ xpc::XrayWrapper<JSCrossCompartmentWrapper>::createHolder(JSContext*, JSObject*, JSObject*) ][@ xpc::XrayWrapper<JSCrossCompartmentWrapper>::createHolder ]

Status: VERIFIED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan Honza Odvarko [:Honza] (always need-info? me)]

CRASH: http://hg.mozilla.org/mozilla-central/rev/8b6d6c128641
Report: https://crash-stats.mozilla.com/report/index/bp-bff87130-9724-4465-974e-d3b432110516

Last working build:
http://hg.mozilla.org/mozilla-central/rev/ad1fa68dcaf5

STR:

1) Install Firebug (attached XPI)
2) Restart and open Firebug UI (F12)
3) CRASH

Honza

Comment 1

[Jan Honza Odvarko [:Honza] (always need-info? me)]

Attachment: Firebug test XPI

Comment 2

[Jan Honza Odvarko [:Honza] (always need-info? me)]

Marking as Firebug-p1 since this is blocking Firebug team from testing on nightlies.

Honza

Comment 3

[Jan Honza Odvarko [:Honza] (always need-info? me)]

Yet another way how to see the exact regression range:
http://getfirebug.com/testresults
(started crashing on Sat: 14 May 2011 01:19:01 GMT, App Changeset: c2bea9fde3b0)

Honza

Comment 4

[Boris Zbarsky [:bzbarsky]]

Regression range from comment 0: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=ad1fa68dcaf5&tochange=8b6d6c128641

Nothing in there jumps out at me....

Comment 5

[jorge alves]

There's a tracemonkey merge in there...

Trying out hourly builds I got to this:

http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=7f1c22983222&tochange=5b6a2be34c55

Comment 6

[Robert Kaiser]

As of yesterday, this is the #1 crash on trunk with roughly 10% of all trunk crashes!
All the 6.0a1 crashes I checked in https://crash-stats.mozilla.com/report/list?signature=xpc%3A%3AXrayWrapper%3CJSCrossCompartmentWrapper%3E%3A%3AcreateHolder%28JSContext%2A%2C%20JSObject%2A%2C%20JSObject%2A%29 do have Firebug installed - but there are a few 4.0.1 crashes with the same signature (probably some other cause) and those don't have Firebug installed.

Still, the Nightly+Firebug problem is the immediate one, we should get that fixed, ideally before this code goes into Aurora next week.

Comment 7

[Boris Zbarsky [:bzbarsky]]

jorge: can you try Tracemonkey nightlies and/or hourlies?

Comment 8

[Boris Zbarsky [:bzbarsky]]

ccing some of the people in that range too.

Comment 9

[Marcia Knous [:marcia]]

Happens on Mac as well, 
[@ xpc::XrayWrapper<JSCrossCompartmentWrapper>::createHolder ]

Comment 10

[jorge alves]

(In reply to comment #7)
> jorge: can you try Tracemonkey nightlies and/or hourlies?

Works (5e69cb4c544b):
http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/tracemonkey-win32/1305224281/

Crashes (673f93bb84aa):
http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/tracemonkey-win32/1305227465/

pushlog:
http://hg.mozilla.org/tracemonkey/pushloghtml?fromchange=5e69cb4c544b&tochange=673f93bb84aa

Comment 11

[jorge alves]

BTW, is there an easier way to browse and test these builds?
Doing it manually is a bit error-prone.

Comment 12

[Boris Zbarsky [:bzbarsky]]

jorge, http://harthur.github.com/mozregression/ might work.

Comment 13

[Boris Zbarsky [:bzbarsky]]

Blake, this looks like a regression from bug 641342.

All the crashes are at 0x10, which looks like a null deref somewhere here.

This code:


    2.15 +        JSObject *inner = obj;
    2.16 +        OBJ_TO_INNER_OBJECT(cx, inner);
    2.17 +        XPCWrappedNative *wn;
    2.18 +        if (IS_WN_WRAPPER(inner) &&
    2.19 +            (wn = static_cast<XPCWrappedNative *>(inner->getPrivate()))->HasProto() &&
    2.20 +            wn->GetProto()->ClassIsDOMObject()) {
    2.21 +            typedef XrayWrapper<JSCrossCompartmentWrapper> Xray;
    2.22 +            wrapper = &FilteringWrapper<Xray,
    2.23 +                                        CrossOriginAccessiblePropertiesOnly>::singleton;
    2.24 +            xrayHolder = Xray::createHolder(cx, obj, parent);

looks to me like a gc hazard if |inner| can go away while |obj| stays around, since nothing is using |inner| in this case after we get |wn|.  Can inner go away that way?

Comment 14

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: Proposed fix

In practice inner and outer are related, so there's no GC hazard there. That being said, I cleaned up the code a little bit. The actual fix is checking targetdata before trying to create the Xray wrapper.

Do note that I haven't actually been able to reproduce the crash. But based on crash stacks, this has to be the fix.

Comment 15

[jorge alves]

(In reply to comment #14)
Applying the patch to current mozilla-central fixes the problem on my end.

Latest nightly still crashes.

Comment 16

[Blake Kaplan (:mrbkap) (inactive)]

http://hg.mozilla.org/tracemonkey/rev/60d467ab6b69

Comment 17

[Jan Honza Odvarko [:Honza] (always need-info? me)]

Still crashes in the latest Nightly channel. This bug blocks me to test Firebug on Nightlies and so, I have no idea whether there are any other problems raising.

Honza

Comment 18

[Blake Kaplan (:mrbkap) (inactive)]

http://hg.mozilla.org/mozilla-central/rev/5b6059eff741

Comment 19

[Robert Kaiser]

(In reply to comment #18)
> http://hg.mozilla.org/mozilla-central/rev/5b6059eff741

Nice! If this works out, we'll want it on aurora as well, I guess, as this crash should be on Aurora 6 now.

Comment 22

[Sebastian Zartner [:sebo]]

For me it is still not fixed yet. See https://crash-stats.mozilla.com/report/index/bp-d0481b69-d9a3-485f-8a92-86d332110526
Or didn't your fix make it into the latest nightly yet?

Comment 23

[Boris Zbarsky [:bzbarsky]]

Yep.  Blake's push was at 2011-05-26 06:15 Pacific, while that crash report is from a nightly that was pulled at 2011-05-26 03:05 Pacific.

Comment 24

[Sebastian Zartner [:sebo]]

Thanks, now the nightlies are finally working again. Though it seems the error now leaked into Aurora: https://crash-stats.mozilla.com/report/index/bp-6e26da3c-aa80-402a-985a-2cb102110527
So hopefully there'll be an update soon.

Comment 25

[Robert Kaiser]

(In reply to comment #24)
> Though it seems the
> error now leaked into Aurora

That's why approval‑mozilla‑aurora had been requested and now granted, the latter meaning that the fix will soon be pushed there (a comment here will be made about this) and then fixed Aurora builds will appear.

Comment 27

[Robert Kaiser]

Blake, can you please land this on Aurora? The last builds on trunk that are seeing this crash are the nightlies from before you checked in on m-c, but Aurora is still hitting this significantly. This crash now takes the two top spots in 6.0a2 topcrashes...

Comment 29

[Blake Kaplan (:mrbkap) (inactive)]

http://hg.mozilla.org/releases/mozilla-aurora/rev/155dea3ad17f

Comment 30

[Blake Kaplan (:mrbkap) (inactive)]

http://hg.mozilla.org/releases/mozilla-beta/rev/9eb6a0741123

Comment 31

[George Carstoiu]

Verified on Mozilla/5.0 (Windows NT 5.1; rv:5.0) Gecko/20100101 Firefox/5.0 - 5b5 and also on Mac OS X 10.6, Ubuntu 10.10 x86 and Win7 using the steps from Description.

Issue no longer present - setting status to Verified Fixed.