Chakra: CFG bypass by overwriting JavaScript bytecode Assume an attacker has the ability to overwrite Chakra's bytecode, either through a read/write primitive or through an overflow type vulnerability. Let's take a look at the ArgOut_A bytecode instruction which is implmented in InterpreterStackFrame::OP_ArgOut_A: template void InterpreterStackFrame::OP_ArgOut_A(const unaligned T * playout) { SetOut(playout->Arg, GetReg(playout->Reg)); } And let's take a look at SetOut and GetReg functions: inline void InterpreterStackFrame::SetOut(ArgSlot outRegisterID, Var aValue) { Assert(m_outParams + outRegisterID < m_outSp); m_outParams[outRegisterID] = aValue; } template Var InterpreterStackFrame::GetReg(RegSlotType localRegisterID) const { Var value = m_localSlots[localRegisterID]; ValidateRegValue(value); return value; } Note there is an Assert() in SetOut() but that won't affect the behavior of the release version. Similarly, ValidateRegValue will only actually perform validation in the debug build. If we can corrupt the bytecode of the ArgOut_A instruction (pointed to by playout) we can *read* an out-of-bounds value in GetReg() and then also *write* it out-of-bounds in SetOut. Note that out-of-bounds read/write here is limited in the sense that both playout->Arg and playout->Reg are 8-bit values but it is sufficient for our purpose. If we already have a read/write primitive, this normally wouldn't give us new capabilities. But note that both m_outParams and m_localSlots actually point on the stack. Thus modifying the ArgOut_A layout will also give us a limited read/write primitive on the stack, which allows us to bypass CFG. For example, we might read some value we control in GetReg() and then use SetOut() to overwrite a return address, thus bypassing CFG. We might also read out a memory from the stack using GetReg() and then write it somewhere where we can extract it using JavaScript. Note #1: There are other instructions that behave similarly such as ProfiledArgOut_A and ArgOut_Env Note #2: These are most likely not the only bytecode instructions that allow an attacker to bypass CFG. In fact by dumb-fuzzing the bytecode and then interpreting it it is fairly easy to get RIP to point outside of the executable memory. This bug is subject to a 90 day disclosure deadline. After 90 days elapse or a patch has been made broadly available, the bug report will become visible to the public. Found by: ifratric