Date: Fri, 6 Mar 2026 16:01:50 -0600
From: Nathan Bossart <nathandbossart@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Robert Haas <robertmhaas@gmail.com>,
	Peter Eisentraut <peter@eisentraut.org>,
	"David G. Johnston" <david.g.johnston@gmail.com>,
	"Ing. Marijo Kristo" <marijo.kristo@icloud.com>,
	PostgreSQL Bug List <pgsql-bugs@lists.postgresql.org>
Subject: Re: Revoke Connect Privilege from Database not working
Message-ID: <aatOzgie9RlzbGoo@nathan>
References: 
 <CAKFQuwa7m2smqqpgPetw=i8Aj-xqg9Zjc5Z2aX3AUwNh96WnXw@mail.gmail.com>
 <d9bf666c-4d11-4196-99a8-b71d01d9ad40@me.com>
 <CAKFQuwbB-ZKtN_p_y5sWa2MrTuy5=pRNPWSj1Ud4HHvTuhb54w@mail.gmail.com>
 <3467676.1744041977@sss.pgh.pa.us>
 <CAKFQuwbpC5w6sUq8gZQATrviZUT4bYpxW+=2uH6sWWMg7fWjzg@mail.gmail.com>
 <aRYLkTpazxKhnS_w@nathan>
 <1933586.1768950341@sss.pgh.pa.us>
 <aXDwtbXCu42Fdmrn@nathan>
 <2222571.1769014621@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <2222571.1769014621@sss.pgh.pa.us>
Archived-At: <https://www.postgresql.org/message-id/aatOzgie9RlzbGoo%40nathan>
Precedence: bulk

On Wed, Jan 21, 2026 at 11:57:01AM -0500, Tom Lane wrote:
> Nathan Bossart <nathandbossart@gmail.com> writes:
>> Yeah, I think doing most of the work in select_best_grantor() is obviously
>> better.  I recall wondering whether we should check for INHERIT or SET
>> privilege (or both) on the grantor role, and IIRC I settled on INHERIT
>> because select_best_grantor() searches through roles we have INHERIT on.
> 
> Yeah, I mentally had that point as something to check on.  Clearly,
> if you have SET ROLE privilege then you can become the target role
> and then issue the GRANT, so if we define GRANTED BY like that
> then we're not allowing anything that can't be done today.  However,
> it feels like INHERIT is a more natural test to make, since AIUI
> that is what permits "automatic" use of a role's privileges, and that
> seems to be what we'd be doing here.

Agreed.

> I'd be interested to hear Robert's opinion on this, or somebody
> else who worked on the SET/INHERIT splitup.
> 
> Also cc'ing Peter, who put in the current effectively-a-noise-clause
> behavior of GRANTED BY, to see if he has standards-compliance or
> other concerns about changing this.

Robert/Peter, do you have any thoughts about expanding GRANT/REVOKE GRANTED
BY like this?  I think it would've helped with a couple of reports received
during this development cycle, and IMHO it'd be a nice little feature for
v19.

-- 
nathan