> On 2 Aug 2020, at 09:05, Michael Paquier <mich...@paquier.xyz> wrote: > > On Sat, Aug 01, 2020 at 11:48:23PM -0700, Noah Misch wrote: >> On Thu, Jul 30, 2020 at 11:42:16PM +0200, Daniel Gustafsson wrote: >>> Somewhat on topic though, 1.1.1 adds a RAND_priv_bytes function for random >>> numbers that are supposed to be private and extra protected via it's own >>> DRBG. >>> Maybe we should use that for SCRAM salts etc in case we detect 1.1.1? >> >> Maybe. Would you have a separate pg_private_random() function, or just use >> RAND_priv_bytes() for pg_strong_random()? No pg_strong_random() caller is >> clearly disinterested in privacy; gen_random_uuid() may come closest. > > FWIW, I am not sure that we need extra level of complexity when it > comes to random number generation, so having only one API to rule them > all sounds sensible to me, particularly if we know that the API used > has more private protections.
I would agree with that, especially since we might not be able to provide an equivalent implementation of a pg_private_random() function in non-OpenSSL builds. Will do a bit more reading and poking and post a patch. cheers ./daniel
