Bug 97836 – JSC: CallFrameClosure::resetCallFrame() can corrupt the previous frame.

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Mark Lam]

Currently, CallFrameClosure::resetCallFrame() looks like this:

    void resetCallFrame()
    {
        newCallFrame->setScope(scope);
        for (int i = argumentCountIncludingThis; i < parameterCountIncludingThis; ++i)
            newCallFrame->setArgument(i, jsUndefined());
    }

However, CallFrame::setArgument() takes an arg index that starts from 0 and does not include the 'this' value.  Since both argumentCountIncludingThis and parameterCountIncludingThis, we need to subtract 1 from both values in order to make the iterator i be a valid arg index for setArgument().

If this is not corrected, then the last setArgument() would be writing to the stack slot before the beginning of the current frame (i.e. corruption the last stack of the previous frame).  In addition, it is also not properly initializing the 1st un-passed argument to 'undefined'.

[Mark Lam]

Created attachment 166106 [details]
Fix.

[Mark Lam]

Landed in http://trac.webkit.org/changeset/129827.

[Geoffrey Garen]

> In addition, it is also not properly initializing the 1st un-passed argument to 'undefined'.

Please add a regression test for this.

[Mark Lam]

Reopening to add a regression test.

[Mark Lam]

Created attachment 166156 [details]
Adding regression test.

[Geoffrey Garen]

Comment on attachment 166156 [details]
Adding regression test.

r=me

[Mark Lam]

Regression test landed in http://trac.webkit.org/changeset/129902.