RtlGenRandom doesn't use enough entropy

From what I can understand, the latter half of /drivers/crypto/ksecdd/random.c calculates an environment hash using md4. The first half uses something that could at least be debatably cryptographic if it outputted one bit at a time. 

This can't be a problem since no open source project trusts Win32 random numbers and usually hashes them with various entropy sources. I'm pretty sure GnuPGP wouldn't generate weak public keys in ReactOS.

1. KsecRandomSeed should at least depend on the environment hash.
2. There ought be an entropy accumulator, using events such as the mouse position during mouse clicks, the time it took to complete a DHCP transaction, etc. When RtlGenRandom is called, it could be reseeded (the hash context updated) with GlobalMemoryStatus, GetCursorPos, and GetCurrentThreadId.
3. Anything is better than "KsecRandomSeed ^= _rotl(TickCount.LowPart, (KsecRandomSeed % 23));"
4. If the entropy accumulator is insufficiently seeded (say with less than 128 events), a calling thread should sleep for a hundred milliseconds. This wouldn't be severely blocking and presumably the accumulator has been seeded better by then. 

I don't have much knowledge in this field, but at the very least, there should be enough entropy for TCP sequence number security.

Edit: Extended MD4 allows for parallelism and doubles state size. I would recommend using Extended MD4 for the entropy accumulator. I have looked at some CRNGs, and what I'm suggesting would be both cross-platform portable and sufficient.