I got the following assert when reading an articles on http://www.economist.com/printedition. I could not reproduce it at will, but i managed to get it 3 times.
Assertion failure: hookPrincipals == compPrincipals || (hookPrincipals->subsume(hookPrincipals, compPrincipals) && compPrincipals->subsume(compPrincipals, hookPrincipals)), at /scratch/igor/m/tm/j/src/jsobjinlines.h:877
Created attachment 527152 [details]
This happens when I browsed using a VNC session on a server. I guess I can keep the debugger running for some time there.
To Luke: can it be a fallout from bug 602994?
Well, that assert was definitely added by bug 602994. But its a property that should generally be true: the compartment containing an object should have equivalent principals to those found through findObjectPrincipals. What are the two principals involved?
Well good to see this isn't some terrible chrome privilege leaking. Igor: Blake was suspecting document.domain at play; do you see any evidence of this?
(In reply to comment #6)
> Blake was suspecting document.domain at play; do you see any evidence of this?
document.domain is set to economist.com on http://www.economist.com/printedition/ but not on articles like http://www.economist.com/node/18560709?story_id=18560709
I got the same assert when browsing at gazeta.ru (a Russian news site). It contains document.domain = 'gazeta.ru'.
Oh right, I talked to mrbkap about this IRL but then we got interrupted and the consensus was never recorded.
Blake explained that this is a problem where, even if two documents have the same domain, if only one of them sets document.domain, they get different (non-equivalent) principals. Due to the lack of compartments==globals (bug 650353), this can lead to objects in the same compartment but with non-equivalent principals. Blake explains this isn't a security hole because they must have started with the same principals, at which point, any evil actions could have happened. Thus, the resolution was to delete the assert since there isn't anything weaker we could assert. Furthermore, after bug 650353, the assert also wouldn't have much meaning since, IIUC, it would essentially be true by definition.
Created attachment 528974 [details] [diff] [review]
But I thought you convinced me that the compartment's principals were good enough? Or have you convinced yourself otherwise now?
Comment on attachment 528974 [details] [diff] [review]
IRL answer from mrbkap: 'yes'
So I see this assertion reliably while restoring a session. In particular, I'm asserting on a script in:
in a tab.
When I apply the fix (attachment 528974 [details] [diff] [review]), however, I still hit a fatal assertion... just a different one:
Assertion failure: callerPrincipals->subsume(callerPrincipals, calleePrincipals), at /home/dbaron/builds/ssd/mozilla-central/mozilla/js/src/jsobj.cpp:1344
In this case, in js::PrincipalsForCompiledCode, calleePrincipals->codebase
but unfortunately callerPrincipals is <value optimized out>.
Is that a continuation of this bug, or another one?
I've seen Assertion failure: hookPrincipals == compPrincipals on a number of urls on Windows, Mac and 32bit as well as 64 bit Linux. It is quite common and will negatively impact crash testing the trunk. It would be very very nice if this landed soon.
I'm getting the same problem as dbaron, also with this patch applied. It's easy to reproduce with this URL (from bug 654590):
The assertion is the same, and it also seems to be a Facebook thing.
callerPrincipals is "http://www.facebook.com/plugins/comments.php?api_key=5919951452&channel_url=http%3A%2F%2Fstatic.ak.fbcdn.net%2Fconnect%2Fxd_proxy.php%3Fversion%3D0%23cb%3Df1eb9e020c0c55e%26origin%3Dhttp%253A%252F%252"...
calleePrincipals is "http://www.facebook.com/extern/login_status.php?api_key=5919951452&app_id=5919951452&channel_url=http%3A%2F%2Fstatic.ak.fbcdn.net%2Fconnect%2Fxd_proxy.php%3Fversion%3D0%23cb%3Df2954ed90346292%26origin"...
We should open a second bug for the other compartment mismatch.
cdleary-bot mozilla-central merge info:
It looks like the follow-up was just filed with bug 656171.