Received-SPF: pass (google.com: domain of linux-kernel+bounces-184812-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Tue, 21 May 2024 19:34:38 +0900
From: Yewon Choi <woni9911@gmail.com>
To: Boris Pismenny <borisp@nvidia.com>,
	John Fastabend <john.fastabend@gmail.com>,
	Jakub Kicinski <kuba@kernel.org>,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>, Paolo Abeni <pabeni@redhat.com>,
	Jakub Sitnicki <jakub@cloudflare.com>, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org
Cc: "Dae R. Jeong" <threeearcat@gmail.com>
Subject: [PATCH net v2] tls: fix missing memory barrier in tls_init
Message-ID: <Zkx4vjSFp0mfpjQ2@libra05>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

From: Dae R. Jeong <threeearcat@gmail.com>

In tls_init(), a write memory barrier is missing, and store-store
reordering may cause NULL dereference in tls_{setsockopt,getsockopt}.

CPU0                               CPU1
-----                              -----
// In tls_init()
// In tls_ctx_create()
ctx = kzalloc()
ctx->sk_proto = READ_ONCE(sk->sk_prot) -(1)

// In update_sk_prot()
WRITE_ONCE(sk->sk_prot, tls_prots)     -(2)

                                   // In sock_common_setsockopt()
                                   READ_ONCE(sk->sk_prot)->setsockopt()

                                   // In tls_{setsockopt,getsockopt}()
                                   ctx->sk_proto->setsockopt()    -(3)

In the above scenario, when (1) and (2) are reordered, (3) can observe
the NULL value of ctx->sk_proto, causing NULL dereference.

To fix it, we rely on rcu_assign_pointer() which implies the release
barrier semantic. By moving rcu_assign_pointer() after ctx->sk_proto is
initialized, we can ensure that ctx->sk_proto are visible when
changing sk->sk_prot.

Fixes: d5bee7374b68 ("net/tls: Annotate access to sk_prot with READ_ONCE/WRITE_ONCE")
Signed-off-by: Yewon Choi <woni9911@gmail.com>
Signed-off-by: Dae R. Jeong <threeearcat@gmail.com>
Link: https://lore.kernel.org/netdev/ZU4OJG56g2V9z_H7@dragonet/T/
---
v2:
  - We don't get rid of tls_ctx_create() because it is called in multiple 
    places (tls_init(), tls_toe_bypass()). Instead, just move 
    rcu_assign_pointer() to the last of tls_ctx_create(). If needed, removing 
    tls_ctx_create() can be considered as later patch.
  - Added Fixes tag
v1: https://lore.kernel.org/all/ZU4Mk_RfzvRpwkmX@dragonet/

 net/tls/tls_main.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c
index b4674f03d71a..90b7f253d363 100644
--- a/net/tls/tls_main.c
+++ b/net/tls/tls_main.c
@@ -816,9 +816,17 @@ struct tls_context *tls_ctx_create(struct sock *sk)
 		return NULL;
 
 	mutex_init(&ctx->tx_lock);
-	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
 	ctx->sk_proto = READ_ONCE(sk->sk_prot);
 	ctx->sk = sk;
+	/* Release semantic of rcu_assign_pointer() ensures that
+	 * ctx->sk_proto is visible before changing sk->sk_prot in
+	 * update_sk_prot(), and prevents reading uninitialized value in
+	 * tls_{getsockopt, setsockopt}. Note that we do not need a
+	 * read barrier in tls_{getsockopt,setsockopt} as there is an
+	 * address dependency between sk->sk_proto->{getsockopt,setsockopt}
+	 * and ctx->sk_proto.
+	 */
+	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
 	return ctx;
 }
 
-- 
2.43.0