OpenSSL RAND_poll 'good enough'

Question

We are using OpenSSL for generating IV's and other random data. We initialize it using RAND_poll and, on Windows, RAND_screen.

The question is though, if the seed generated by those methods is good enough (RAND_screen at least sounds to me like it is). Also, I was wondering how exactly RAND_poll works, as I could not find much documentation about it ant if it does behave differently depending on platform, like RAND_screen which is only available on Windows.

Answer 1

RAND_bytes() automatically calls RAND_poll() if it has not already been done at least once. So you do not have to call it yourself. RAND_poll() feeds on what the operating system provides: on Linux, Solaris, FreeBSD and similar Unix-like systems, it will use /dev/urandom (or /dev/random if there is no /dev/urandom) to obtain a cryptographically secure initial seed; on Windows, it will call CryptGenRandom() for the same effect.

RAND_screen() is provided by OpenSSL only for backward compatibility with (much) older code which may call it (that was before OpenSSL used proper OS-based seed initialization).

So the "normal" way of dealing with RAND_poll() and RAND_screen() is to call neither. Just use RAND_bytes() and be happy.