Date: Thu, 19 Sep 2024 16:52:02 -0500
From: Nathan Bossart <nathandbossart@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Alexander Lakhin <exclusion@gmail.com>,
	pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: Should rolpassword be toastable?
Message-ID: <ZuydAj4nrt4cpGU5@nathan>
References: <89e8649c-eb74-db25-7945-6d6b23992394@gmail.com>
 <1284094.1695479962@sss.pgh.pa.us>
 <67793ae1-3adc-a6e0-0bb3-c23a28c0627a@gmail.com>
 <a9524f94-316c-349a-6e4d-79b878706f82@gmail.com>
 <Zuwzp09o9uiG0f0F@nathan>
 <1986859.1726756275@sss.pgh.pa.us>
 <ZuxjAIOJ4GSa34KC@nathan>
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="OPwpp7vWUTj0xjp/"
Content-Disposition: inline
In-Reply-To: <ZuxjAIOJ4GSa34KC@nathan>
Archived-At: <https://www.postgresql.org/message-id/ZuydAj4nrt4cpGU5%40nathan>
Precedence: bulk


--OPwpp7vWUTj0xjp/
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Thu, Sep 19, 2024 at 12:44:32PM -0500, Nathan Bossart wrote:
> On Thu, Sep 19, 2024 at 10:31:15AM -0400, Tom Lane wrote:
>> We could put an arbitrary limit (say, half of BLCKSZ) on the length of
>> passwords.
> 
> Something like that could be good enough.  I was thinking about actually
> validating that the hash had the correct form, but that might be a little
> more complex than is warranted here.

Oh, actually, I see that we are already validating the hash, but you can
create valid SCRAM-SHA-256 hashes that are really long.  So putting an
arbitrary limit (patch attached) is probably the correct path forward.  I'd
also remove pg_authid's TOAST table while at it.

-- 
nathan

--OPwpp7vWUTj0xjp/
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename="fail_for_long_scram_hash.patch"

diff --git a/src/backend/libpq/crypt.c b/src/backend/libpq/crypt.c
index 629e51e00b..a723e8219a 100644
--- a/src/backend/libpq/crypt.c
+++ b/src/backend/libpq/crypt.c
@@ -121,6 +121,19 @@ encrypt_password(PasswordType target_type, const char *role,
 
 	if (guessed_type != PASSWORD_TYPE_PLAINTEXT)
 	{
+		/*
+		 * Valid SCRAM-SHA-256 hashes can have very long "iterations" and
+		 * "salt" fields, but we don't want to store anything that might get
+		 * TOASTed, since de-TOASTing won't work during authentication because
+		 * we haven't selected a database yet and cannot read pg_class.  256
+		 * bytes should be more than enough for all practical use, so fail for
+		 * anything longer.
+		 */
+		if (guessed_type == PASSWORD_TYPE_SCRAM_SHA_256 &&
+			strlen(password) > 256)
+			ereport(ERROR,
+					(errmsg("SCRAM-SHA-256 hash is too long")));
+
 		/*
 		 * Cannot convert an already-encrypted password from one format to
 		 * another, so return it as it is.

--OPwpp7vWUTj0xjp/--