Re: Move OpenSSL random under USE_OPENSSL_RANDOM

Michael Paquier <michael@paquier.xyz>

From: Michael Paquier <michael@paquier.xyz>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Magnus Hagander <magnus@hagander.net>, Daniel Gustafsson <daniel@yesql.se>, Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2020-11-16T00:20:16Z
Lists: pgsql-hackers

Attachments

On Sun, Nov 15, 2020 at 12:16:56PM -0500, Tom Lane wrote:
> The obvious problem with this is that if !USE_OPENSSL, we will not have
> pulled in openssl's headers.

FWIW, I argued upthread against including this part because it is
useless: if not building with OpenSSL, we'll never have the base to be
able to use RAND_poll().

> However ... all these machines are pointing at line 96, which is not
> that one but the one under "#if defined(USE_OPENSSL)".  So I'm not sure
> what to make of that, except that a bit more finesse seems required.

The build scripts of src/tools/msvc/ choose to not use OpenSSL as
strong random source even if building with OpenSSL.  The top of the
file only includes openssl/rand.h if using USE_OPENSSL_RANDOM.

Thinking about that afresh, I think that we got that wrong here on
three points:
- If attempting to use OpenSSL on Windows, let's just bite the bullet
and use OpenSSL as random source, using Windows as source only when
not building with OpenSSL.
- Instead of using a call to RAND_poll() that we know will never work,
let's just issue a compilation failure if attempting to use
USE_OPENSSL_RANDOM without USE_OPENSSL.
- rand.h needs to be included under USE_OPENSSL.
--
Michael

Commits

  1. Remove ability to independently select random number generator

  2. Add pg_strong_random_init function to initialize random number generator