[Mach-O] Catch exceptions thrown in universal transform header parsing

[emesare]

Fixes potential OOB reads (which in the binary reader cause exceptions) caused by malformed data. We might want to instead catch exceptions higher up the call stack, not sure. If we instead just want to guard every read with some bounds check or something im amenable to that, but honestly that might just bloat code up with potential bad guards 🤣 (and we typically only do that when we have a recoverable situation)

Fixes https://github.com/Vector35/binaryninja/issues/1911

[bpotchik]

The exception fix prevents the crash, but it looks like there is a root cause here. There is an integer overflow that allows the out of bounds read. The bug is in two places, copied from the original FAT macho view.

• universaltransform.cpp
  • size_t requiredFatHeaderSize = fatHeader.nfat_arch * (isFat64 ? 32 : 20) + 8;
• fatmachoview.cpp
  • size_t expectSize = header.nfat_arch * ....

Just add a (size_t) to extend before the multipl on both: fatHeader.nfat_arch and header.nfat_arch

Also, yes I think catching higher up is the right call. Look in api/transform.cpp. Following the same patter in BinaryView::InitCallback(void* ctxt)

You could apply the pattern below to each of DecodeCallback, EncodeCallback, and CanDecodeCallback.

try {
....
catch (const std::exception& e)
{
    LogError("Transform::DecodeWithContext failed: %s", e.what());
    return false;
}
catch (...)
{
    LogError("Transform::DecodeWithContext failed with unknown exception");
    return false;
}