Обсуждение: Support a`--with-copy-program` compile flag

On 12/11/2025 20:07, Steve Chavez wrote:
> Hello hackers,
> 
> Postgres provides the `COPY .. TO/FROM PROGRAM` statement. This is 
> dangerous from a security perspective because it allows users to escape 
> from the SQL sandbox and gain shell access on the instance.
> 
> Now there's the `pg_execute_server_program` predefined role to restrict 
> access to `COPY.. TO/FROM PROGRAM` but if somehow a pg user gains 
> superuser privileges then the predefined role is of no use.
> 
> So I wonder if we could remove the possibility of shell access by 
> providing a `--with-copy-program` compile flag.

If you are superuser, there are many other ways you can gain shell 
access. There is no security boundary there.

See e.g. 
https://www.postgresql.org/about/news/cve-2019-9193-not-a-security-vulnerability-1935/

- Heikki

Hi,

On 2025-11-12 13:07:27 -0500, Steve Chavez wrote:
> Postgres provides the `COPY .. TO/FROM PROGRAM` statement. This is
> dangerous from a security perspective because it allows users to escape
> from the SQL sandbox and gain shell access on the instance.
> 
> Now there's the `pg_execute_server_program` predefined role to restrict
> access to `COPY.. TO/FROM PROGRAM` but if somehow a pg user gains superuser
> privileges then the predefined role is of no use.
> 
> So I wonder if we could remove the possibility of shell access by providing
> a `--with-copy-program` compile flag.

If a user has superuser, the game is already lost. There are *dozens* of ways
to execute arbitrary code at that point.

Greetings,

Andres Freund

On Wed, Nov 12, 2025 at 01:07:27PM -0500, Steve Chavez wrote:
> So I wonder if we could remove the possibility of shell access by providing
> a `--with-copy-program` compile flag.

You might be interested in this past discussion for a similar idea:

    https://postgr.es/m/flat/20220520225619.GA876272%40nathanxps13

As others have already pointed out, there's no real boundary between
database superusers and the OS user running Postgres.  I think many would
like there to be one, but I'm unaware of any serious efforts in that area,
and I doubt there's much appetite for it in the community.

-- 
nathan