From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
cc: Aditya D <dsaditya91@gmail.com>,
        pgsql-admin <pgsql-admin@lists.postgresql.org>
Subject: Re: pg_restore -L reordering of the statements does not work
In-reply-to: <20231114222743.x2c2r74ttgom2cfx@awork3.anarazel.de>
References: 
 <CAEATte6vgZr6d2abZo9_wxTarwQP3TC_NLpCKTDzWW4E4OB9cw@mail.gmail.com>
 <1589825.1698431686@sss.pgh.pa.us>
 <20231114194933.ci3nxol3avtisyrs@awork3.anarazel.de>
 <2522830.1699994542@sss.pgh.pa.us>
 <20231114222743.x2c2r74ttgom2cfx@awork3.anarazel.de>
Comments: In-reply-to Andres Freund <andres@anarazel.de>
	message dated "Tue, 14 Nov 2023 14:27:43 -0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2537260.1700001602.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 14 Nov 2023 17:40:02 -0500
Message-ID: <2537261.1700001602@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/2537261.1700001602%40sss.pgh.pa.us>
Precedence: bulk

Andres Freund <andres@anarazel.de> writes:
> On 2023-11-14 15:42:22 -0500, Tom Lane wrote:
>> Moreover, reordering the GRANTs is no solution, because who promised th=
at
>> the schema owner granted you any permissions?

> I'm not quite following - the schema is created in the dump, so the gran=
t is
> part of it?

Yeah, but the GRANT will restore whatever permissions existed in the
source database.  If the restoring user isn't super, those permissions
don't necessarily grant him access.

>> I experimented with making the restoring user be a member with inherit
>> of the nosuper_N roles, and indeed I still see the failure above,
>> which makes me wonder if the ACL check is being done correctly for
>> that specific case.  The INHERIT bit ought to let it work.

> The check is for nosuper_2 to have permission on the schema

... no, it should be for the user executing the ALTER to have permission.

> ... and the check
> happens before the grant on the schema. For inherit to help, nosuper_2 w=
ould
> have to be granted membership to the presumably more privileged user doi=
ng the
> restore.

No, surely the other way?  Restoring user must be member of nosuper_2,
else the ALTER OWNER won't work either.

			regards, tom lane