User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-GB; rv:18.104.22.168) Gecko/2008070208 Firefox/3.0.1 Build Identifier: This bug crashes the browser without any possibility for user intervention. This is exploitable in the browser. Do *not* click the given link if you want to keep anything in your current Firefox session. Reproducible: Always Steps to Reproduce: (new Date()).getMonth.call(new Function()) Actual Results: jsshell (or Firefox) crash. Expected Results: Exception. The built-in Date class doesn't check the class of its "this" object sufficiently. Affected methods are getYear(), getMonth(), getDate() and so on. This affects all of Date's methods that internally call GetLocalTime() (in jsdate.c) which doesn't check the class of object passed in (compare with JS_InstanceOf() call in GetUTCTime()). The object passed in needs to have at least 2 private slots, like a) the global object b) Function objects c) Iterator objects The value from the private slot is passed to JSVAL_TO_DOUBLE() without further checks in GetLocalTime(), leading to bogus pointer dereference for objects not of class Date. In addition to adding this to the test suite, this sort of thing could be added to jsparsefuzz.js (i.e. the fuzzer should be modified to use <method>.call() with random arguments if it's not already doing this). I don't have time right now to check out a current tree and post a patch and will not get around to it until some time next week.
Created attachment 336054 [details] [diff] [review] fix v1 The essence of the fix is the addition of the missing instanceof check for GetAndCacheLocalTime. I also took an opportunity in the patch to replace heavy JS_(Get|Set)ReservvedSlot calls with direct slot access for simpler code.
Igor's patch should fix the issue AFAICS. If needed I can report back next week when I get a chance to build trunk and verify.
Created attachment 336077 [details] testcase crashes trunk Preserving a copy of the testcase in Bugzilla.
Much nicer -- thanks, Igor. /be
landed on trunk - http://hg.mozilla.org/mozilla-central/rev/42b53b1f04e6
Comment on attachment 336054 [details] [diff] [review] fix v1 The patch applies to 1.9.0 branch as-is.
The bug only exists on 1.9.0 and later and is an regression from the bug 340992.
Is it worth doing a version of the patch that only has the fix and not the cleanup, for risk reduction on 1.9.0.x?
(In reply to comment #8) > Is it worth doing a version of the patch that only has the fix and not the > cleanup, for risk reduction on 1.9.0.x? There's no risk reduction in messing with a reviewed patch that cuts out overhead and doubt about reserved slot complexities that do not affect fslots' dimension or lifetime. Making more variation between trees adds tiny risk too, probably more than any risk in sticking this patch in as-is. But it's too tiny to estimate or measure, or to be worth arguing much about. /be
22.214.171.124 is in release testing, this will have to sail with the next ship.
Comment on attachment 336054 [details] [diff] [review] fix v1 Approved for 126.96.36.199, a=dveditz for release-drivers
Created attachment 344937 [details] [diff] [review] fix for 1.9.0 It turned out the patch required a trivial change for 1.9.0. It applies as-is, but the trunk version uses const to define constant names. But that does not work with in C sources as in C such constants are not real constant expressions. So I replaced constants with the preprocessor macros. Here is the plain diff between the trunk and the branch patches: diff /home/igor/s/fix_date.patch /home/igor/s/fix_date.1.9.0.patch 7c7 < +++ jsdate.c 27 Oct 2008 17:22:47 -0000 --- > +++ jsdate.c 27 Oct 2008 17:21:59 -0000 19,20c19,20 < +const uint32 JSSLOT_UTC_TIME = JSSLOT_PRIVATE; < +const uint32 JSSLOT_LOCAL_TIME = JSSLOT_PRIVATE + 1; --- > +#define JSSLOT_UTC_TIME JSSLOT_PRIVATE > +#define JSSLOT_LOCAL_TIME (JSSLOT_PRIVATE + 1) 22c22 < +const uint32 DATE_RESERVED_SLOTS = 2; --- > +#define DATE_RESERVED_SLOTS 2
I landed the patch from the comment 12 on 1.9.0 branch: Checking in jsdate.c; /cvsroot/mozilla/js/src/jsdate.c,v <-- jsdate.c new revision: 3.112; previous revision: 3.111 done
We now get the exception "TypeError on line 1: Date.prototype.getMonth called on incompatible Function" Verified for 188.8.131.52 with Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:184.108.40.206pre) Gecko/2008102804 GranParadiso/3.0.4pre. Same behavior in Trunk as well with Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9.1b2pre) Gecko/20081028 Minefield/3.1b2pre.
Checking in ecma_3/Date/regress-452786.js; /cvsroot/mozilla/js/tests/ecma_3/Date/regress-452786.js,v <-- regress-452786.js initial revision: 1.1 done