Received-SPF: pass (google.com: domain of linux-kernel+bounces-40318-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Luis Henriques <lhenriques@suse.de>
To: Eric Biggers <ebiggers@kernel.org>
Cc: David Howells <dhowells@redhat.com>,  Jarkko Sakkinen
 <jarkko@kernel.org>,  keyrings@vger.kernel.org,
  linux-kernel@vger.kernel.org
Subject: Re: [RFC PATCH v2] keys: update key quotas in key_put()
In-Reply-To: <20240124221225.GD1088@sol.localdomain> (Eric Biggers's message
	of "Wed, 24 Jan 2024 14:12:25 -0800")
References: <2744563.1702303367@warthog.procyon.org.uk>
	<20240115120300.27606-1-lhenriques@suse.de>
	<20240124221225.GD1088@sol.localdomain>
Date: Fri, 26 Jan 2024 16:12:59 +0000
Message-ID: <87bk988450.fsf@suse.de>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable

Eric Biggers <ebiggers@kernel.org> writes:

> On Mon, Jan 15, 2024 at 12:03:00PM +0000, Luis Henriques wrote:
>> Delaying key quotas update when key's refcount reaches 0 in key_put() has
>> been causing some issues in fscrypt testing.  This patches fixes this te=
st
>> flakiness by dealing with the quotas immediately, but leaving all the ot=
her
>> clean-ups to the key garbage collector.  Unfortunately, this means that =
we
>> also need to switch to the irq-version of the spinlock that protects quo=
ta.
>>=20
>> Signed-off-by: Luis Henriques <lhenriques@suse.de>
>> ---
>> Hi David!
>>=20
>> I have these changes in my local disk for a while; I wanted to send them
>> before EOY break but... yeah, it didn't happen.  Anyway, I'm still sendi=
ng
>> it as an RFC as I'm probably missing something.
>>=20
>>  security/keys/gc.c     |  8 --------
>>  security/keys/key.c    | 32 ++++++++++++++++++++++----------
>>  security/keys/keyctl.c | 11 ++++++-----
>>  3 files changed, 28 insertions(+), 23 deletions(-)
>
> This patch seems reasonable to me, though I'm still thinking about changi=
ng
> fs/crypto/ to manage its key quotas itself which would avoid the issue en=
tirely.
>
> Note that as I said before, fs/crypto/ does key_put() on a whole keyring =
at
> once, in order to release the quota of the keys in the keyring.  Do you p=
lan to
> also change fs/crypto/ to keyring_clear() the keyring before putting it?
> Without that, I don't think the problem is solved, as the quota release w=
ill
> still happen asynchronously due to the keyring being cleared asynchronous=
ly.

Ah, good point.  In the meantime I had forgotten everything about this
code and missed that.  So, I can send another patch to fs/crypto to add
that extra call once (or if) this patch is accepted.

If I'm reading the code correctly, the only place where this extra call is
required is on fscrypt_put_master_key():

diff --git a/fs/crypto/keyring.c b/fs/crypto/keyring.c
index 0edf0b58daa7..4afd32f1aed9 100644
--- a/fs/crypto/keyring.c
+++ b/fs/crypto/keyring.c
@@ -74,6 +74,7 @@ void fscrypt_put_master_key(struct fscrypt_master_key *mk)
 	 * that concurrent keyring lookups can no longer find it.
 	 */
 	WARN_ON_ONCE(refcount_read(&mk->mk_active_refs) !=3D 0);
+	keyring_clear(mk->mk_users);
 	key_put(mk->mk_users);
 	mk->mk_users =3D NULL;
 	call_rcu(&mk->mk_rcu_head, fscrypt_free_master_key);

On the other hand, if you're really working towards dropping this code
entirely, maybe there's not point doing that.

Cheers,
--=20
Lu=C3=ADs