[PATCH v2 0/7] TCP-AO fixes

[PATCH v2 0/7] TCP-AO fixes

[Dmitry Safonov]

Hi,

Changes from v1:
- Use tcp_can_repair_sock() helper to limit TCP_AO_REPAIR (Eric)
- Instead of hook to listen() syscall, allow removing current/rnext keys
  on TCP_LISTEN (addressing Eric's objection)
- Add sne_lock to protect snd_sne/rcv_sne
- Don't move used_tcp_ao in struct tcp_request_sock (Eric)

I've been working on TCP-AO key-rotation selftests and as a result
exercised some corner-cases that are not usually met in production.

Here are a bunch of semi-related fixes:
- Documentation typo (reported by Markus Elfring)
- Proper alignment for TCP-AO option in TCP header that has MAC length
  of non 4 bytes (now a selftest with randomized maclen/algorithm/etc
  passes)
- 3 uAPI restricting patches that disallow more things to userspace in
  order to prevent it shooting itself in any parts of the body
- SNEs READ_ONCE()/WRITE_ONCE() that went missing by my human factor
- Avoid storing MAC length from SYN header as SYN-ACK will use
  rnext_key.maclen (drops an extra check that fails on new selftests)

Please, consider applying/pulling.

The following changes since commit d3fa86b1a7b4cdc4367acacea16b72e0a200b3d7:

  Merge tag 'net-6.7-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net (2023-11-23 10:40:13 -0800)

are available in the Git repository at:

  git@github.com:0x7f454c46/linux.git tcp-ao-post-merge-v2

for you to fetch changes up to c5e4cecfcdc7f996acae740812d9ab2ebcd90517:

  net/tcp: Don't store TCP-AO maclen on reqsk (2023-11-23 20:54:54 +0000)

----------------------------------------------------------------

Thanks,
             Dmitry

Cc: David Ahern <dsahern@kernel.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Dmitry Safonov <0x7f454c46@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Francesco Ruggeri <fruggeri05@gmail.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Salam Noureddine <noureddine@arista.com>
Cc: Simon Horman <horms@kernel.org>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org

Dmitry Safonov (7):
  Documentation/tcp: Fix an obvious typo
  net/tcp: Consistently align TCP-AO option in the header
  net/tcp: Limit TCP_AO_REPAIR to non-listen sockets
  net/tcp: Allow removing current/rnext TCP-AO keys on TCP_LISTEN sockets
  net/tcp: Don't add key with non-matching VRF on connected sockets
  net/tcp: Add sne_lock to access SNEs
  net/tcp: Don't store TCP-AO maclen on reqsk

 Documentation/networking/tcp_ao.rst |  2 +-
 include/linux/tcp.h                 |  8 +---
 include/net/tcp_ao.h                |  8 +++-
 net/ipv4/tcp.c                      |  6 +++
 net/ipv4/tcp_ao.c                   | 57 +++++++++++++++++++----------
 net/ipv4/tcp_input.c                | 21 +++++++++--
 net/ipv4/tcp_ipv4.c                 |  4 +-
 net/ipv4/tcp_minisocks.c            |  2 +-
 net/ipv4/tcp_output.c               | 15 +++-----
 net/ipv6/tcp_ipv6.c                 |  2 +-
 10 files changed, 81 insertions(+), 44 deletions(-)


base-commit: d3fa86b1a7b4cdc4367acacea16b72e0a200b3d7
-- 
2.43.0

[PATCH v2 1/7] Documentation/tcp: Fix an obvious typo

[Dmitry Safonov]

Yep, my VIM spellchecker is not good enough for typos like this one.

Fixes: 7fe0e38bb669 ("Documentation/tcp: Add TCP-AO documentation")
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Reported-by: Markus Elfring <Markus.Elfring@web.de>
Closes: https://lore.kernel.org/all/2745ab4e-acac-40d4-83bf-37f2600d0c3d@web.de/
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 Documentation/networking/tcp_ao.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Documentation/networking/tcp_ao.rst b/Documentation/networking/tcp_ao.rst
index cfa5bf1cc542..8a58321acce7 100644
--- a/Documentation/networking/tcp_ao.rst
+++ b/Documentation/networking/tcp_ao.rst
@@ -99,7 +99,7 @@ also [6.1]::
    when it is no longer considered permitted.
 
 Linux TCP-AO will try its best to prevent you from removing a key that's
-being used, considering it a key management failure. But sine keeping
+being used, considering it a key management failure. But since keeping
 an outdated key may become a security issue and as a peer may
 unintentionally prevent the removal of an old key by always setting
 it as RNextKeyID - a forced key removal mechanism is provided, where
-- 
2.43.0

[PATCH v2 2/7] net/tcp: Consistently align TCP-AO option in the header

[Dmitry Safonov]

Currently functions that pre-calculate TCP header options length use
unaligned TCP-AO header + MAC-length for skb reservation.
And the functions that actually write TCP-AO options into skb do align
the header. Nothing good can come out of this for ((maclen % 4) != 0).

Provide tcp_ao_len_aligned() helper and use it everywhere for TCP
header options space calculations.

Fixes: 1e03d32bea8e ("net/tcp: Add TCP-AO sign to outgoing packets")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 include/net/tcp_ao.h     | 6 ++++++
 net/ipv4/tcp_ao.c        | 4 ++--
 net/ipv4/tcp_ipv4.c      | 4 ++--
 net/ipv4/tcp_minisocks.c | 2 +-
 net/ipv4/tcp_output.c    | 6 +++---
 net/ipv6/tcp_ipv6.c      | 2 +-
 6 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/include/net/tcp_ao.h b/include/net/tcp_ao.h
index b56be10838f0..647781080613 100644
--- a/include/net/tcp_ao.h
+++ b/include/net/tcp_ao.h
@@ -62,11 +62,17 @@ static inline int tcp_ao_maclen(const struct tcp_ao_key *key)
 	return key->maclen;
 }
 
+/* Use tcp_ao_len_aligned() for TCP header calculations */
 static inline int tcp_ao_len(const struct tcp_ao_key *key)
 {
 	return tcp_ao_maclen(key) + sizeof(struct tcp_ao_hdr);
 }
 
+static inline int tcp_ao_len_aligned(const struct tcp_ao_key *key)
+{
+	return round_up(tcp_ao_len(key), 4);
+}
+
 static inline unsigned int tcp_ao_digest_size(struct tcp_ao_key *key)
 {
 	return key->digest_size;
diff --git a/net/ipv4/tcp_ao.c b/net/ipv4/tcp_ao.c
index 7696417d0640..c8be1d526eac 100644
--- a/net/ipv4/tcp_ao.c
+++ b/net/ipv4/tcp_ao.c
@@ -1100,7 +1100,7 @@ void tcp_ao_connect_init(struct sock *sk)
 			ao_info->current_key = key;
 		if (!ao_info->rnext_key)
 			ao_info->rnext_key = key;
-		tp->tcp_header_len += tcp_ao_len(key);
+		tp->tcp_header_len += tcp_ao_len_aligned(key);
 
 		ao_info->lisn = htonl(tp->write_seq);
 		ao_info->snd_sne = 0;
@@ -1346,7 +1346,7 @@ static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key)
 	syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED;
 	syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED;
 	syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED;
-	if (tcp_ao_len(key) > syn_tcp_option_space) {
+	if (tcp_ao_len_aligned(key) > syn_tcp_option_space) {
 		err = -EMSGSIZE;
 		goto err_kfree;
 	}
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 5f693bbd578d..0c50c5a32b84 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -690,7 +690,7 @@ static bool tcp_v4_ao_sign_reset(const struct sock *sk, struct sk_buff *skb,
 
 	reply_options[0] = htonl((TCPOPT_AO << 24) | (tcp_ao_len(key) << 16) |
 				 (aoh->rnext_keyid << 8) | keyid);
-	arg->iov[0].iov_len += round_up(tcp_ao_len(key), 4);
+	arg->iov[0].iov_len += tcp_ao_len_aligned(key);
 	reply->doff = arg->iov[0].iov_len / 4;
 
 	if (tcp_ao_hash_hdr(AF_INET, (char *)&reply_options[1],
@@ -978,7 +978,7 @@ static void tcp_v4_send_ack(const struct sock *sk,
 					  (tcp_ao_len(key->ao_key) << 16) |
 					  (key->ao_key->sndid << 8) |
 					  key->rcv_next);
-		arg.iov[0].iov_len += round_up(tcp_ao_len(key->ao_key), 4);
+		arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
 		rep.th.doff = arg.iov[0].iov_len / 4;
 
 		tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index a9807eeb311c..9e85f2a0bddd 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -615,7 +615,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
 	ao_key = treq->af_specific->ao_lookup(sk, req,
 				tcp_rsk(req)->ao_keyid, -1);
 	if (ao_key)
-		newtp->tcp_header_len += tcp_ao_len(ao_key);
+		newtp->tcp_header_len += tcp_ao_len_aligned(ao_key);
  #endif
 	if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
 		newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index eb13a55d660c..93eef1dbbc55 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -825,7 +825,7 @@ static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
 		timestamps = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps);
 		if (tcp_key_is_ao(key)) {
 			opts->options |= OPTION_AO;
-			remaining -= tcp_ao_len(key->ao_key);
+			remaining -= tcp_ao_len_aligned(key->ao_key);
 		}
 	}
 
@@ -915,7 +915,7 @@ static unsigned int tcp_synack_options(const struct sock *sk,
 			ireq->tstamp_ok &= !ireq->sack_ok;
 	} else if (tcp_key_is_ao(key)) {
 		opts->options |= OPTION_AO;
-		remaining -= tcp_ao_len(key->ao_key);
+		remaining -= tcp_ao_len_aligned(key->ao_key);
 		ireq->tstamp_ok &= !ireq->sack_ok;
 	}
 
@@ -982,7 +982,7 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb
 		size += TCPOLEN_MD5SIG_ALIGNED;
 	} else if (tcp_key_is_ao(key)) {
 		opts->options |= OPTION_AO;
-		size += tcp_ao_len(key->ao_key);
+		size += tcp_ao_len_aligned(key->ao_key);
 	}
 
 	if (likely(tp->rx_opt.tstamp_ok)) {
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 937a02c2e534..8c6623496dd7 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -881,7 +881,7 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32
 	if (tcp_key_is_md5(key))
 		tot_len += TCPOLEN_MD5SIG_ALIGNED;
 	if (tcp_key_is_ao(key))
-		tot_len += tcp_ao_len(key->ao_key);
+		tot_len += tcp_ao_len_aligned(key->ao_key);
 
 #ifdef CONFIG_MPTCP
 	if (rst && !tcp_key_is_md5(key)) {
-- 
2.43.0

[PATCH v2 3/7] net/tcp: Limit TCP_AO_REPAIR to non-listen sockets

[Dmitry Safonov]

Listen socket is not an established TCP connection, so
setsockopt(TCP_AO_REPAIR) doesn't have any impact.

Restrict this uAPI for listen sockets.

Fixes: faadfaba5e01 ("net/tcp: Add TCP_AO_REPAIR")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 net/ipv4/tcp.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 53bcc17c91e4..b1fe4eb01829 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3594,6 +3594,10 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,
 		break;
 
 	case TCP_AO_REPAIR:
+		if (!tcp_can_repair_sock(sk)) {
+			err = -EPERM;
+			break;
+		}
 		err = tcp_ao_set_repair(sk, optval, optlen);
 		break;
 #ifdef CONFIG_TCP_AO
@@ -4293,6 +4297,8 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 	}
 #endif
 	case TCP_AO_REPAIR:
+		if (!tcp_can_repair_sock(sk))
+			return -EPERM;
 		return tcp_ao_get_repair(sk, optval, optlen);
 	case TCP_AO_GET_KEYS:
 	case TCP_AO_INFO: {
-- 
2.43.0

[PATCH v2 4/7] net/tcp: Allow removing current/rnext TCP-AO keys on TCP_LISTEN sockets

[Dmitry Safonov]

TCP_LISTEN sockets are not connected to any peer, so having
current_key/rnext_key doesn't make sense.

The userspace may falter over this issue by setting current or rnext
TCP-AO key before listen() syscall. setsockopt(TCP_AO_DEL_KEY) doesn't
allow removing a key that is in use (in accordance to RFC 5925), so
it might be inconvenient to have keys that can be destroyed only with
listener socket.

Fixes: 4954f17ddefc ("net/tcp: Introduce TCP_AO setsockopt()s")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 net/ipv4/tcp_ao.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_ao.c b/net/ipv4/tcp_ao.c
index c8be1d526eac..bf41be6d4721 100644
--- a/net/ipv4/tcp_ao.c
+++ b/net/ipv4/tcp_ao.c
@@ -1818,8 +1818,16 @@ static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family,
 		if (!new_rnext)
 			return -ENOENT;
 	}
-	if (cmd.del_async && sk->sk_state != TCP_LISTEN)
-		return -EINVAL;
+	if (sk->sk_state == TCP_LISTEN) {
+		/* Cleaning up possible "stale" current/rnext keys state,
+		 * that may have preserved from TCP_CLOSE, before sys_listen()
+		 */
+		ao_info->current_key = NULL;
+		ao_info->rnext_key = NULL;
+	} else {
+		if (cmd.del_async)
+			return -EINVAL;
+	}
 
 	if (family == AF_INET) {
 		struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr;
-- 
2.43.0

[PATCH v2 5/7] net/tcp: Don't add key with non-matching VRF on connected sockets

[Dmitry Safonov]

If the connection was established, don't allow adding TCP-AO keys that
don't match the peer. Currently, there are checks for ip-address
matching, but L3 index check is missing. Add it to restrict userspace
shooting itself somewhere.

Fixes: 248411b8cb89 ("net/tcp: Wire up l3index to TCP-AO")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 net/ipv4/tcp_ao.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/net/ipv4/tcp_ao.c b/net/ipv4/tcp_ao.c
index bf41be6d4721..2d000e275ce7 100644
--- a/net/ipv4/tcp_ao.c
+++ b/net/ipv4/tcp_ao.c
@@ -1608,6 +1608,9 @@ static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family,
 		if (!dev || !l3index)
 			return -EINVAL;
 
+		if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
+			return -EINVAL;
+
 		/* It's still possible to bind after adding keys or even
 		 * re-bind to a different dev (with CAP_NET_RAW).
 		 * So, no reason to return error here, rather try to be
-- 
2.43.0

[PATCH v2 6/7] net/tcp: Add sne_lock to access SNEs

[Dmitry Safonov]

RFC 5925 (6.2):
> TCP-AO emulates a 64-bit sequence number space by inferring when to
> increment the high-order 32-bit portion (the SNE) based on
> transitions in the low-order portion (the TCP sequence number).

snd_sne and rcv_sne are the upper 4 bytes of extended SEQ number.
Unfortunately, reading two 4-bytes pointers can't be performed
atomically (without synchronization).

Let's keep it KISS and add an rwlock - that shouldn't create much
contention as SNE are updated every 4Gb of traffic and the atomic region
is quite small.

Fixes: 64382c71a557 ("net/tcp: Add TCP-AO SNE support")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 include/net/tcp_ao.h |  2 +-
 net/ipv4/tcp_ao.c    | 34 +++++++++++++++++++++-------------
 net/ipv4/tcp_input.c | 16 ++++++++++++++--
 3 files changed, 36 insertions(+), 16 deletions(-)

diff --git a/include/net/tcp_ao.h b/include/net/tcp_ao.h
index 647781080613..beea3e6b39e2 100644
--- a/include/net/tcp_ao.h
+++ b/include/net/tcp_ao.h
@@ -123,6 +123,7 @@ struct tcp_ao_info {
 	 */
 	u32			snd_sne;
 	u32			rcv_sne;
+	rwlock_t		sne_lock;
 	refcount_t		refcnt;		/* Protects twsk destruction */
 	struct rcu_head		rcu;
 };
@@ -212,7 +213,6 @@ enum skb_drop_reason tcp_inbound_ao_hash(struct sock *sk,
 			const struct sk_buff *skb, unsigned short int family,
 			const struct request_sock *req, int l3index,
 			const struct tcp_ao_hdr *aoh);
-u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq);
 struct tcp_ao_key *tcp_ao_do_lookup(const struct sock *sk, int l3index,
 				    const union tcp_ao_addr *addr,
 				    int family, int sndid, int rcvid);
diff --git a/net/ipv4/tcp_ao.c b/net/ipv4/tcp_ao.c
index 2d000e275ce7..74db80aeeef3 100644
--- a/net/ipv4/tcp_ao.c
+++ b/net/ipv4/tcp_ao.c
@@ -230,6 +230,7 @@ static struct tcp_ao_info *tcp_ao_alloc_info(gfp_t flags)
 		return NULL;
 	INIT_HLIST_HEAD(&ao->head);
 	refcount_set(&ao->refcnt, 1);
+	rwlock_init(&ao->sne_lock);
 
 	return ao;
 }
@@ -472,10 +473,8 @@ static int tcp_ao_hash_pseudoheader(unsigned short int family,
 	return -EAFNOSUPPORT;
 }
 
-u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq)
+static u32 tcp_ao_compute_sne(u32 sne, u32 next_seq, u32 seq)
 {
-	u32 sne = next_sne;
-
 	if (before(seq, next_seq)) {
 		if (seq > next_seq)
 			sne--;
@@ -483,7 +482,6 @@ u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq)
 		if (seq < next_seq)
 			sne++;
 	}
-
 	return sne;
 }
 
@@ -763,14 +761,15 @@ int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
 		*keyid = (*key)->rcvid;
 	} else {
 		struct tcp_ao_key *rnext_key;
-		u32 snd_basis;
+		const u32 *snd_basis;
+		unsigned long flags;
 
 		if (sk->sk_state == TCP_TIME_WAIT) {
 			ao_info = rcu_dereference(tcp_twsk(sk)->ao_info);
-			snd_basis = tcp_twsk(sk)->tw_snd_nxt;
+			snd_basis = &tcp_twsk(sk)->tw_snd_nxt;
 		} else {
 			ao_info = rcu_dereference(tcp_sk(sk)->ao_info);
-			snd_basis = tcp_sk(sk)->snd_una;
+			snd_basis = &tcp_sk(sk)->snd_una;
 		}
 		if (!ao_info)
 			return -ENOENT;
@@ -781,8 +780,10 @@ int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
 		*traffic_key = snd_other_key(*key);
 		rnext_key = READ_ONCE(ao_info->rnext_key);
 		*keyid = rnext_key->rcvid;
-		*sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
-					  snd_basis, seq);
+		read_lock_irqsave(&ao_info->sne_lock, flags);
+		*sne = tcp_ao_compute_sne(ao_info->snd_sne,
+					  READ_ONCE(*snd_basis), seq);
+		read_unlock_irqrestore(&ao_info->sne_lock, flags);
 	}
 	return 0;
 }
@@ -795,6 +796,7 @@ int tcp_ao_transmit_skb(struct sock *sk, struct sk_buff *skb,
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct tcp_ao_info *ao;
 	void *tkey_buf = NULL;
+	unsigned long flags;
 	u8 *traffic_key;
 	u32 sne;
 
@@ -816,8 +818,10 @@ int tcp_ao_transmit_skb(struct sock *sk, struct sk_buff *skb,
 		tp->af_specific->ao_calc_key_sk(key, traffic_key,
 						sk, ao->lisn, disn, true);
 	}
-	sne = tcp_ao_compute_sne(READ_ONCE(ao->snd_sne), READ_ONCE(tp->snd_una),
-				 ntohl(th->seq));
+	read_lock_irqsave(&ao->sne_lock, flags);
+	sne = tcp_ao_compute_sne(ao->snd_sne,
+				 READ_ONCE(tp->snd_una), ntohl(th->seq));
+	read_unlock_irqrestore(&ao->sne_lock, flags);
 	tp->af_specific->calc_ao_hash(hash_location, key, sk, skb, traffic_key,
 				      hash_location - (u8 *)th, sne);
 	kfree(tkey_buf);
@@ -938,8 +942,9 @@ tcp_inbound_ao_hash(struct sock *sk, const struct sk_buff *skb,
 
 	/* Fast-path */
 	if (likely((1 << sk->sk_state) & TCP_AO_ESTABLISHED)) {
-		enum skb_drop_reason err;
 		struct tcp_ao_key *current_key;
+		enum skb_drop_reason err;
+		unsigned long flags;
 
 		/* Check if this socket's rnext_key matches the keyid in the
 		 * packet. If not we lookup the key based on the keyid
@@ -956,8 +961,11 @@ tcp_inbound_ao_hash(struct sock *sk, const struct sk_buff *skb,
 		if (unlikely(th->syn && !th->ack))
 			goto verify_hash;
 
-		sne = tcp_ao_compute_sne(info->rcv_sne, tcp_sk(sk)->rcv_nxt,
+		read_lock_irqsave(&info->sne_lock, flags);
+		sne = tcp_ao_compute_sne(info->rcv_sne,
+					 READ_ONCE(tcp_sk(sk)->rcv_nxt),
 					 ntohl(th->seq));
+		read_unlock_irqrestore(&info->sne_lock, flags);
 		/* Established socket, traffic key are cached */
 		traffic_key = rcv_other_key(key);
 		err = tcp_ao_verify_hash(sk, skb, family, info, aoh, key,
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index bcb55d98004c..fc3c27ce2b73 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -3582,8 +3582,14 @@ static void tcp_snd_sne_update(struct tcp_sock *tp, u32 ack)
 
 	ao = rcu_dereference_protected(tp->ao_info,
 				       lockdep_sock_is_held((struct sock *)tp));
-	if (ao && ack < tp->snd_una)
+	if (ao && ack < tp->snd_una) {
+		unsigned long flags;
+
+		write_lock_irqsave(&ao->sne_lock, flags);
 		ao->snd_sne++;
+		tp->snd_una = ack;
+		write_unlock_irqrestore(&ao->sne_lock, flags);
+	}
 #endif
 }
 
@@ -3608,8 +3614,14 @@ static void tcp_rcv_sne_update(struct tcp_sock *tp, u32 seq)
 
 	ao = rcu_dereference_protected(tp->ao_info,
 				       lockdep_sock_is_held((struct sock *)tp));
-	if (ao && seq < tp->rcv_nxt)
+	if (ao && seq < tp->rcv_nxt) {
+		unsigned long flags;
+
+		write_lock_irqsave(&ao->sne_lock, flags);
 		ao->rcv_sne++;
+		WRITE_ONCE(tp->rcv_nxt, seq);
+		write_unlock_irqrestore(&ao->sne_lock, flags);
+	}
 #endif
 }
 
-- 
2.43.0

[PATCH v2 7/7] net/tcp: Don't store TCP-AO maclen on reqsk

[Dmitry Safonov]

This extra check doesn't work for a handshake when SYN segment has
(current_key.maclen != rnext_key.maclen). It could be amended to
preserve rnext_key.maclen instead of current_key.maclen, but that
requires a lookup on listen socket.

Originally, this extra maclen check was introduced just because it was
cheap. Drop it and convert tcp_request_sock::maclen into boolean
tcp_request_sock::used_tcp_ao.

Fixes: 06b22ef29591 ("net/tcp: Wire TCP-AO to request sockets")
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 include/linux/tcp.h   | 8 ++------
 net/ipv4/tcp_ao.c     | 4 ++--
 net/ipv4/tcp_input.c  | 5 +++--
 net/ipv4/tcp_output.c | 9 +++------
 4 files changed, 10 insertions(+), 16 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 68f3d315d2e1..b646b574b060 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -169,7 +169,7 @@ struct tcp_request_sock {
 #ifdef CONFIG_TCP_AO
 	u8				ao_keyid;
 	u8				ao_rcv_next;
-	u8				maclen;
+	bool				used_tcp_ao;
 #endif
 };
 
@@ -180,14 +180,10 @@ static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
 
 static inline bool tcp_rsk_used_ao(const struct request_sock *req)
 {
-	/* The real length of MAC is saved in the request socket,
-	 * signing anything with zero-length makes no sense, so here is
-	 * a little hack..
-	 */
 #ifndef CONFIG_TCP_AO
 	return false;
 #else
-	return tcp_rsk(req)->maclen != 0;
+	return tcp_rsk(req)->used_tcp_ao;
 #endif
 }
 
diff --git a/net/ipv4/tcp_ao.c b/net/ipv4/tcp_ao.c
index 74db80aeeef3..cfa264c320a7 100644
--- a/net/ipv4/tcp_ao.c
+++ b/net/ipv4/tcp_ao.c
@@ -855,7 +855,7 @@ void tcp_ao_syncookie(struct sock *sk, const struct sk_buff *skb,
 	const struct tcp_ao_hdr *aoh;
 	struct tcp_ao_key *key;
 
-	treq->maclen = 0;
+	treq->used_tcp_ao = false;
 
 	if (tcp_parse_auth_options(th, NULL, &aoh) || !aoh)
 		return;
@@ -867,7 +867,7 @@ void tcp_ao_syncookie(struct sock *sk, const struct sk_buff *skb,
 
 	treq->ao_rcv_next = aoh->keyid;
 	treq->ao_keyid = aoh->rnext_keyid;
-	treq->maclen = tcp_ao_maclen(key);
+	treq->used_tcp_ao = true;
 }
 
 static enum skb_drop_reason
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index fc3c27ce2b73..0135a6c6f600 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -7194,11 +7194,12 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
 	if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
 		goto drop_and_release; /* Invalid TCP options */
 	if (aoh) {
-		tcp_rsk(req)->maclen = aoh->length - sizeof(struct tcp_ao_hdr);
+		tcp_rsk(req)->used_tcp_ao = true;
 		tcp_rsk(req)->ao_rcv_next = aoh->keyid;
 		tcp_rsk(req)->ao_keyid = aoh->rnext_keyid;
+
 	} else {
-		tcp_rsk(req)->maclen = 0;
+		tcp_rsk(req)->used_tcp_ao = false;
 	}
 #endif
 	tcp_rsk(req)->snt_isn = isn;
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 93eef1dbbc55..f5ef15e1d9ac 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3720,7 +3720,6 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 	if (tcp_rsk_used_ao(req)) {
 #ifdef CONFIG_TCP_AO
 		struct tcp_ao_key *ao_key = NULL;
-		u8 maclen = tcp_rsk(req)->maclen;
 		u8 keyid = tcp_rsk(req)->ao_keyid;
 
 		ao_key = tcp_sk(sk)->af_specific->ao_lookup(sk, req_to_sk(req),
@@ -3730,13 +3729,11 @@ struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 		 * for another peer-matching key, but the peer has requested
 		 * ao_keyid (RFC5925 RNextKeyID), so let's keep it simple here.
 		 */
-		if (unlikely(!ao_key || tcp_ao_maclen(ao_key) != maclen)) {
-			u8 key_maclen = ao_key ? tcp_ao_maclen(ao_key) : 0;
-
+		if (unlikely(!ao_key)) {
 			rcu_read_unlock();
 			kfree_skb(skb);
-			net_warn_ratelimited("TCP-AO: the keyid %u with maclen %u|%u from SYN packet is not present - not sending SYNACK\n",
-					     keyid, maclen, key_maclen);
+			net_warn_ratelimited("TCP-AO: the keyid %u from SYN packet is not present - not sending SYNACK\n",
+					     keyid);
 			return NULL;
 		}
 		key.ao_key = ao_key;
-- 
2.43.0

Re: [PATCH v2 6/7] net/tcp: Add sne_lock to access SNEs

[Paolo Abeni]

On Fri, 2023-11-24 at 00:27 +0000, Dmitry Safonov wrote:
> RFC 5925 (6.2):
> > TCP-AO emulates a 64-bit sequence number space by inferring when to
> > increment the high-order 32-bit portion (the SNE) based on
> > transitions in the low-order portion (the TCP sequence number).
> 
> snd_sne and rcv_sne are the upper 4 bytes of extended SEQ number.
> Unfortunately, reading two 4-bytes pointers can't be performed
> atomically (without synchronization).
> 
> Let's keep it KISS and add an rwlock - that shouldn't create much
> contention as SNE are updated every 4Gb of traffic and the atomic region
> is quite small.
> 
> Fixes: 64382c71a557 ("net/tcp: Add TCP-AO SNE support")
> Signed-off-by: Dmitry Safonov <dima@arista.com>
> ---
>  include/net/tcp_ao.h |  2 +-
>  net/ipv4/tcp_ao.c    | 34 +++++++++++++++++++++-------------
>  net/ipv4/tcp_input.c | 16 ++++++++++++++--
>  3 files changed, 36 insertions(+), 16 deletions(-)
> 
> diff --git a/include/net/tcp_ao.h b/include/net/tcp_ao.h
> index 647781080613..beea3e6b39e2 100644
> --- a/include/net/tcp_ao.h
> +++ b/include/net/tcp_ao.h
> @@ -123,6 +123,7 @@ struct tcp_ao_info {
>  	 */
>  	u32			snd_sne;
>  	u32			rcv_sne;
> +	rwlock_t		sne_lock;

RW lock are problematic in the networking code, see commit
dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65.

I think you can use a plain spinlock here, as both read and write
appears to be in the fastpath (?!?)

> @@ -781,8 +780,10 @@ int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
>  		*traffic_key = snd_other_key(*key);
>  		rnext_key = READ_ONCE(ao_info->rnext_key);
>  		*keyid = rnext_key->rcvid;
> -		*sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
> -					  snd_basis, seq);
> +		read_lock_irqsave(&ao_info->sne_lock, flags);
> +		*sne = tcp_ao_compute_sne(ao_info->snd_sne,
> +					  READ_ONCE(*snd_basis), seq);
> +		read_unlock_irqrestore(&ao_info->sne_lock, flags);

Why are you using the irqsave variant? bh should suffice.

Cheers,

Paolo

Re: [PATCH v2 6/7] net/tcp: Add sne_lock to access SNEs

[Dmitry Safonov]

On 11/27/23 11:41, Paolo Abeni wrote:
> On Fri, 2023-11-24 at 00:27 +0000, Dmitry Safonov wrote:
>> RFC 5925 (6.2):
>>> TCP-AO emulates a 64-bit sequence number space by inferring when to
>>> increment the high-order 32-bit portion (the SNE) based on
>>> transitions in the low-order portion (the TCP sequence number).
>>
>> snd_sne and rcv_sne are the upper 4 bytes of extended SEQ number.
>> Unfortunately, reading two 4-bytes pointers can't be performed
>> atomically (without synchronization).
>>
>> Let's keep it KISS and add an rwlock - that shouldn't create much
>> contention as SNE are updated every 4Gb of traffic and the atomic region
>> is quite small.
>>
>> Fixes: 64382c71a557 ("net/tcp: Add TCP-AO SNE support")
>> Signed-off-by: Dmitry Safonov <dima@arista.com>
>> ---
>>  include/net/tcp_ao.h |  2 +-
>>  net/ipv4/tcp_ao.c    | 34 +++++++++++++++++++++-------------
>>  net/ipv4/tcp_input.c | 16 ++++++++++++++--
>>  3 files changed, 36 insertions(+), 16 deletions(-)
>>
>> diff --git a/include/net/tcp_ao.h b/include/net/tcp_ao.h
>> index 647781080613..beea3e6b39e2 100644
>> --- a/include/net/tcp_ao.h
>> +++ b/include/net/tcp_ao.h
>> @@ -123,6 +123,7 @@ struct tcp_ao_info {
>>  	 */
>>  	u32			snd_sne;
>>  	u32			rcv_sne;
>> +	rwlock_t		sne_lock;
> 
> RW lock are problematic in the networking code, see commit
> dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65.

Thanks, was not aware of this pitfall.

> I think you can use a plain spinlock here, as both read and write
> appears to be in the fastpath (?!?)

Yeah, I wanted to avoid to RX concurrency here as writing happens only
once in 4Gb. I'll take another attempt to prevent that in v3.

>> @@ -781,8 +780,10 @@ int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
>>  		*traffic_key = snd_other_key(*key);
>>  		rnext_key = READ_ONCE(ao_info->rnext_key);
>>  		*keyid = rnext_key->rcvid;
>> -		*sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
>> -					  snd_basis, seq);
>> +		read_lock_irqsave(&ao_info->sne_lock, flags);
>> +		*sne = tcp_ao_compute_sne(ao_info->snd_sne,
>> +					  READ_ONCE(*snd_basis), seq);
>> +		read_unlock_irqrestore(&ao_info->sne_lock, flags);
> 
> Why are you using the irqsave variant? bh should suffice.

It should, yes :)

> 
> Cheers,
> 
> Paolo
> 

Thanks,
             Dmitry