| From: | Heikki Linnakangas <hlinnaka(at)iki(dot)fi> |
|---|---|
| To: | Peter Eisentraut <peter(at)eisentraut(dot)org>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: Let OpenSSL auto-select DH parameters by default |
| Date: | 2026-11-19 19:32:42 |
| Message-ID: | b45c5632-fcc3-4b5d-9d36-a03100d122a4@iki.fi |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On 11/08/2026 16:24, Peter Eisentraut wrote:
> I was curious about the hardcoded FILE_DH2048 in src/include/libpq/
> libpq-be.h -- how did it get there, do we ever need to update it, how do
> we know, etc. -- and it seems that there is a more modern approach that
> allows us to get rid of it.
>
> Previously, if ssl_dh_params_file was not set, the compiled-in default
> parameters in FILE_DH2048 were used. Instead, we can call
> SSL_CTX_set_dh_auto() and have OpenSSL choose appropriate, well-known,
> and appropriately sized parameters. In particular, the size of the
> parameters is chosen consistent with the size of the key and
> certificate, and so it can automatically grow with new standards and
> best practices. This facility was not available at the time FILE_DH2048
> was added, but it is now available in all relevant OpenSSL and LibreSSL
> versions.
>
> I think this new behavior is preferable (also consistent with
> recommendations in the OpenSSL documentation), so I have tweaked the
> documentation a bit in that direction.
>
> There was also some not-very-modern text about this in README.SSL,
> which I have deleted.
Interesting, I didn't know about SSL_CTX_set_dh_auto(). Makes much more
sense than having hardcoded parameters in the application (PostgreSQL in
this case).
> Note that if you use a very small certificate/key size, then this
> could mean that the size of the chosen DH parameters could also end up
> smaller than the previous hard-coded default. But that could only
> happen if you disregarded all current security best practices anyway.
Yeah, seems fine.
> Note a small behavior change: Previously, if loading the DH file
> failed during a server reload, the system would then fall back to
> using the built-in parameters, instead of failing the reload and
> keeping the old SSL context, which is what it would do if some other
> aspect of reinitializing SSL would fail. I think this is actually a
> bug. I have fixed this here as part of reflowing the relevant function.
>
> Anyway, this is all very complicated and obscure. Reviews are welcome.
LGTM. I wonder if we should just remove the ssl_dh_params_file option.
If I understand correctly, there's no reason for anyone to use it.
- Heikki
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Daniel Gustafsson | 2026-11-19 19:47:51 | Re: Assert "target_wal_endrec == divergerec" in pg_rewind |
| Previous Message | Manu | 2026-11-19 19:17:07 | Re: Use an index scan in RemoveSubscriptionRel() |