js/xpconnect/crashtests/786142-iframe.html

Wed, 31 Dec 2014 06:09:35 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Wed, 31 Dec 2014 06:09:35 +0100
changeset 0
6474c204b198
permissions
-rw-r--r--

Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.

michael@0 1 <!DOCTYPE html>
michael@0 2 <html>
michael@0 3 <head>
michael@0 4 <script>
michael@0 5
michael@0 6 // Morph a slim wrapper into an XPCWN by generating a cross-compartment wrapper
michael@0 7 // for it (we Morph in WrapperFactory::PrepareForWrapping).
michael@0 8 function makeNonSlim(elem) {
michael@0 9 window.parent.someCCW = elem;
michael@0 10 delete window.parent.someCCW;
michael@0 11 }
michael@0 12
michael@0 13 function doTest() {
michael@0 14
michael@0 15 // Get a slim wrapper for |form|. This lives in scope 1.
michael@0 16 var form = document.getElementById('myform');
michael@0 17
michael@0 18 // The gist here is that we want to create an input element that isn't part
michael@0 19 // of a form, so that its PreCreate hook will cause it to be parented to the
michael@0 20 // document. However, there are several considerations that make this more
michael@0 21 // complicted.
michael@0 22 //
michael@0 23 // First, the crashtest becomes non-deterministics if we morph |form| before
michael@0 24 // getting to scope 3 (as explained below). This means that we can't trigger
michael@0 25 // the PreCreate hook for |input|, because that will call WrapNativeParent
michael@0 26 // on the form, which will end up making a cross-compartment wrapper, which
michael@0 27 // will morph form. But this puts us in a pickle, because appendChild returns
michael@0 28 // the apppended child, which will trigger the PreCreate hook in
michael@0 29 // NativeInterface2JSObject. So we do this little hack where we append a buch
michael@0 30 // of dummy <div> children to the form, and use replaceChild (which returns
michael@0 31 // the replacer, not the replacee) to stick the input elements as children of
michael@0 32 // the form.
michael@0 33 //
michael@0 34 // Second, the crashtest can also be non-deterministics if the final call to
michael@0 35 // MoveWrappers iterates over the hashtable in such a way that it fixes up
michael@0 36 // the Document before it fixes up the Input. If so, the Input will become
michael@0 37 // orphaned, and we'll detect it and fix things up at the top of MoveWrapper.
michael@0 38 // Since we can't control the hashtable ordering here, we just make 100 input
michael@0 39 // elements, to make it a near-certainty (statistically) that we'll encounter
michael@0 40 // one of them during iteration before encountering the Document.
michael@0 41 //
michael@0 42 // With all this, this testcase deterministically crashes on my machine. Whew!
michael@0 43
michael@0 44 // Create an input element. This isn't part of a form right now, so it gets
michael@0 45 // parented to the document.
michael@0 46 var inputs = [];
michael@0 47 var placeHolders = [];
michael@0 48 for (var i = 0; i < 100; ++i) {
michael@0 49 var dummyDiv = form.appendChild(document.createElement('div'));
michael@0 50 var input = document.createElement('input');
michael@0 51 makeNonSlim(input);
michael@0 52 inputs.push(input);
michael@0 53 placeHolders.push(dummyDiv);
michael@0 54 }
michael@0 55
michael@0 56 // Blow away the document, forcing a transplan of all the XPCWNs in scope. This
michael@0 57 // will transplant |input|, but |form| stays in the old scope (since it's slim).
michael@0 58 document.open();
michael@0 59 document.close();
michael@0 60
michael@0 61 // Now we're in scope 2. Associate |input| with |form| so that the next call to
michael@0 62 // PreCreate will parent |input| to |form| rather than the document. But make
michael@0 63 // sure to do it with replaceChild, rather than appendChild, so that we don't
michael@0 64 // end up triggering the PreCreate hook for |form| in scope 2, which would make
michael@0 65 // make it non-slim. If we didn't, the ensuing call to MoveWrappers might find
michael@0 66 // |form| before |input| while iterating over the hashtable. If so, |form|
michael@0 67 // would be rescued as an orphan and everything would be fixed before getting to // |input|.
michael@0 68 for (var i = 0; i < inputs.length; ++i)
michael@0 69 form.replaceChild(inputs[i], placeHolders[i]);
michael@0 70
michael@0 71 // Blow away the document a second time. This should cause the crash in
michael@0 72 // unpatched builds.
michael@0 73 document.open();
michael@0 74 document.close();
michael@0 75 }
michael@0 76 </script>
michael@0 77 </head>
michael@0 78 <body>
michael@0 79 <form id="myform"></form>
michael@0 80 </body>
michael@0 81 </html>

mercurial