]> git.baikalelectronics.ru Git - kernel.git/commitdiff
mptcp: process MP_CAPABLE data option
authorChristoph Paasch <cpaasch@apple.com>
Wed, 22 Jan 2020 00:56:32 +0000 (16:56 -0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 24 Jan 2020 12:44:08 +0000 (13:44 +0100)
This patch implements the handling of MP_CAPABLE + data option, as per
RFC 6824 bis / RFC 8684: MPTCP v1.

On the server side we can receive the remote key after that the connection
is established. We need to explicitly track the 'missing remote key'
status and avoid emitting a mptcp ack until we get such info.

When a late/retransmitted/OoO pkt carrying MP_CAPABLE[+data] option
is received, we have to propagate the mptcp seq number info to
the msk socket. To avoid ABBA locking issue, explicitly check for
that in recvmsg(), where we own msk and subflow sock locks.

The above also means that an established mp_capable subflow - still
waiting for the remote key - can be 'downgraded' to plain TCP.

Such change could potentially block a reader waiting for new data
forever - as they hook to msk, while later wake-up after the downgrade
will be on subflow only.

The above issue is not handled here, we likely have to get rid of
msk->fallback to handle that cleanly.

Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/mptcp/options.c
net/mptcp/protocol.c
net/mptcp/protocol.h
net/mptcp/subflow.c

index 8f82ff9a5a8e4fbbbb325146d4678a9ed91398d3..45acd877bef345259ca2a84c58cd0d375f88386f 100644 (file)
@@ -243,6 +243,7 @@ void mptcp_rcv_synsent(struct sock *sk)
        pr_debug("subflow=%p", subflow);
        if (subflow->request_mptcp && tp->rx_opt.mptcp.mp_capable) {
                subflow->mp_capable = 1;
+               subflow->can_ack = 1;
                subflow->remote_key = tp->rx_opt.mptcp.sndr_key;
        } else {
                tcp_sk(sk)->is_mptcp = 0;
@@ -332,6 +333,7 @@ static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
        struct mptcp_ext *mpext;
        struct mptcp_sock *msk;
        unsigned int ack_size;
+       bool ret = false;
        u8 tcp_fin;
 
        if (skb) {
@@ -355,6 +357,14 @@ static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
                if (skb && tcp_fin &&
                    subflow->conn->sk_state != TCP_ESTABLISHED)
                        mptcp_write_data_fin(subflow, &opts->ext_copy);
+               ret = true;
+       }
+
+       opts->ext_copy.use_ack = 0;
+       msk = mptcp_sk(subflow->conn);
+       if (!msk || !READ_ONCE(msk->can_ack)) {
+               *size = ALIGN(dss_size, 4);
+               return ret;
        }
 
        ack_size = TCPOLEN_MPTCP_DSS_ACK64;
@@ -365,15 +375,7 @@ static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 
        dss_size += ack_size;
 
-       msk = mptcp_sk(mptcp_subflow_ctx(sk)->conn);
-       if (msk) {
-               opts->ext_copy.data_ack = msk->ack_seq;
-       } else {
-               mptcp_crypto_key_sha(mptcp_subflow_ctx(sk)->remote_key,
-                                    NULL, &opts->ext_copy.data_ack);
-               opts->ext_copy.data_ack++;
-       }
-
+       opts->ext_copy.data_ack = msk->ack_seq;
        opts->ext_copy.ack64 = 1;
        opts->ext_copy.use_ack = 1;
 
@@ -422,13 +424,46 @@ bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
        return false;
 }
 
+static bool check_fourth_ack(struct mptcp_subflow_context *subflow,
+                            struct sk_buff *skb,
+                            struct mptcp_options_received *mp_opt)
+{
+       /* here we can process OoO, in-window pkts, only in-sequence 4th ack
+        * are relevant
+        */
+       if (likely(subflow->fourth_ack ||
+                  TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1))
+               return true;
+
+       if (mp_opt->use_ack)
+               subflow->fourth_ack = 1;
+
+       if (subflow->can_ack)
+               return true;
+
+       /* If the first established packet does not contain MP_CAPABLE + data
+        * then fallback to TCP
+        */
+       if (!mp_opt->mp_capable) {
+               subflow->mp_capable = 0;
+               tcp_sk(mptcp_subflow_tcp_sock(subflow))->is_mptcp = 0;
+               return false;
+       }
+       subflow->remote_key = mp_opt->sndr_key;
+       subflow->can_ack = 1;
+       return true;
+}
+
 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
                            struct tcp_options_received *opt_rx)
 {
+       struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
        struct mptcp_options_received *mp_opt;
        struct mptcp_ext *mpext;
 
        mp_opt = &opt_rx->mptcp;
+       if (!check_fourth_ack(subflow, skb, mp_opt))
+               return;
 
        if (!mp_opt->dss)
                return;
@@ -441,9 +476,6 @@ void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
 
        if (mp_opt->use_map) {
                if (mp_opt->mpc_map) {
-                       struct mptcp_subflow_context *subflow =
-                               mptcp_subflow_ctx(sk);
-
                        /* this is an MP_CAPABLE carrying MPTCP data
                         * we know this map the first chunk of data
                         */
index 45e482864a195cbbf5409c85ab98f11dca44526b..1b64dfaa5f63aaf7f9fd31980f1ec7b75e2c4200 100644 (file)
@@ -30,7 +30,7 @@
  */
 static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
 {
-       if (!msk->subflow || mptcp_subflow_ctx(msk->subflow->sk)->fourth_ack)
+       if (!msk->subflow || READ_ONCE(msk->can_ack))
                return NULL;
 
        return msk->subflow;
@@ -651,17 +651,20 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
                __mptcp_init_sock(new_mptcp_sock);
 
                msk = mptcp_sk(new_mptcp_sock);
-               msk->remote_key = subflow->remote_key;
                msk->local_key = subflow->local_key;
                msk->token = subflow->token;
                msk->subflow = NULL;
 
                mptcp_token_update_accept(newsk, new_mptcp_sock);
 
-               mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
                msk->write_seq = subflow->idsn + 1;
-               ack_seq++;
-               msk->ack_seq = ack_seq;
+               if (subflow->can_ack) {
+                       msk->can_ack = true;
+                       msk->remote_key = subflow->remote_key;
+                       mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
+                       ack_seq++;
+                       msk->ack_seq = ack_seq;
+               }
                newsk = new_mptcp_sock;
                mptcp_copy_inaddrs(newsk, ssk);
                list_add(&subflow->node, &msk->conn_list);
@@ -678,8 +681,6 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
                 * the receive path and process the pending ones
                 */
                lock_sock(ssk);
-               subflow->map_seq = ack_seq;
-               subflow->map_subflow_seq = 1;
                subflow->rel_write_seq = 1;
                subflow->tcp_sock = ssk;
                subflow->conn = new_mptcp_sock;
@@ -795,6 +796,7 @@ void mptcp_finish_connect(struct sock *ssk)
        WRITE_ONCE(msk->token, subflow->token);
        WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
        WRITE_ONCE(msk->ack_seq, ack_seq);
+       WRITE_ONCE(msk->can_ack, 1);
 }
 
 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
index 36b90024d34d82459857d37f482f5e8d243580a6..10eaa7c7381b0443823d856049207aa0dc85795d 100644 (file)
@@ -69,6 +69,7 @@ struct mptcp_sock {
        u64             ack_seq;
        u32             token;
        unsigned long   flags;
+       bool            can_ack;
        struct list_head conn_list;
        struct skb_ext  *cached_ext;    /* for the next sendmsg */
        struct socket   *subflow; /* outgoing connect/listener/!mp_capable */
@@ -84,9 +85,10 @@ static inline struct mptcp_sock *mptcp_sk(const struct sock *sk)
 
 struct mptcp_subflow_request_sock {
        struct  tcp_request_sock sk;
-       u     mp_capable : 1,
+       u16     mp_capable : 1,
                mp_join : 1,
-               backup : 1;
+               backup : 1,
+               remote_key_valid : 1;
        u64     local_key;
        u64     remote_key;
        u64     idsn;
@@ -118,8 +120,10 @@ struct mptcp_subflow_context {
                fourth_ack : 1,     /* send initial DSS */
                conn_finished : 1,
                map_valid : 1,
+               mpc_map : 1,
                data_avail : 1,
-               rx_eof : 1;
+               rx_eof : 1,
+               can_ack : 1;        /* only after processing the remote a key */
 
        struct  sock *tcp_sock;     /* tcp sk backpointer */
        struct  sock *conn;         /* parent mptcp_sock */
index 8892855f4f52edb302c7b98c06400ca274427faa..8cfa1d29d59c82996b72ab6e1218abb2b3f25c88 100644 (file)
@@ -61,6 +61,7 @@ static void subflow_init_req(struct request_sock *req,
        mptcp_get_options(skb, &rx_opt);
 
        subflow_req->mp_capable = 0;
+       subflow_req->remote_key_valid = 0;
 
 #ifdef CONFIG_TCP_MD5SIG
        /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
@@ -185,17 +186,28 @@ static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 
        pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 
-       /* if the sk is MP_CAPABLE, we need to fetch the client key */
+       /* if the sk is MP_CAPABLE, we try to fetch the client key */
        subflow_req = mptcp_subflow_rsk(req);
        if (subflow_req->mp_capable) {
+               if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
+                       /* here we can receive and accept an in-window,
+                        * out-of-order pkt, which will not carry the MP_CAPABLE
+                        * opt even on mptcp enabled paths
+                        */
+                       goto create_child;
+               }
+
                opt_rx.mptcp.mp_capable = 0;
                mptcp_get_options(skb, &opt_rx);
-               if (!opt_rx.mptcp.mp_capable)
-                       subflow_req->mp_capable = 0;
-               else
+               if (opt_rx.mptcp.mp_capable) {
                        subflow_req->remote_key = opt_rx.mptcp.sndr_key;
+                       subflow_req->remote_key_valid = 1;
+               } else {
+                       subflow_req->mp_capable = 0;
+               }
        }
 
+create_child:
        child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
                                                     req_unhash, own_req);
 
@@ -377,6 +389,7 @@ static enum mapping_status get_mapping_status(struct sock *ssk)
        subflow->map_subflow_seq = mpext->subflow_seq;
        subflow->map_data_len = data_len;
        subflow->map_valid = 1;
+       subflow->mpc_map = mpext->mpc_map;
        pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
                 subflow->map_seq, subflow->map_subflow_seq,
                 subflow->map_data_len);
@@ -428,6 +441,19 @@ static bool subflow_check_data_avail(struct sock *ssk)
                if (WARN_ON_ONCE(!skb))
                        return false;
 
+               /* if msk lacks the remote key, this subflow must provide an
+                * MP_CAPABLE-based mapping
+                */
+               if (unlikely(!READ_ONCE(msk->can_ack))) {
+                       if (!subflow->mpc_map) {
+                               ssk->sk_err = EBADMSG;
+                               goto fatal;
+                       }
+                       WRITE_ONCE(msk->remote_key, subflow->remote_key);
+                       WRITE_ONCE(msk->ack_seq, subflow->map_seq);
+                       WRITE_ONCE(msk->can_ack, true);
+               }
+
                old_ack = READ_ONCE(msk->ack_seq);
                ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
                pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
@@ -752,13 +778,17 @@ static void subflow_ulp_clone(const struct request_sock *req,
                return;
        }
 
+       /* see comments in subflow_syn_recv_sock(), MPTCP connection is fully
+        * established only after we receive the remote key
+        */
        new_ctx->conn_finished = 1;
        new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
        new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
        new_ctx->tcp_state_change = old_ctx->tcp_state_change;
        new_ctx->tcp_write_space = old_ctx->tcp_write_space;
        new_ctx->mp_capable = 1;
-       new_ctx->fourth_ack = 1;
+       new_ctx->fourth_ack = subflow_req->remote_key_valid;
+       new_ctx->can_ack = subflow_req->remote_key_valid;
        new_ctx->remote_key = subflow_req->remote_key;
        new_ctx->local_key = subflow_req->local_key;
        new_ctx->token = subflow_req->token;