tcp: TCP Fast Open Server - support TFO listeners

This patch builds on top of the previous patch to add the support
for TFO listeners. This includes -

1. allocating, properly initializing, and managing the per listener
fastopen_queue structure when TFO is enabled

2. changes to the inet_csk_accept code to support TFO. E.g., the
request_sock can no longer be freed upon accept(), not until 3WHS
finishes

3. allowing a TCP_SYN_RECV socket to properly poll() and sendmsg()
if it's a TFO socket

4. properly closing a TFO listener, and a TFO socket before 3WHS
finishes

5. supporting TCP_FASTOPEN socket option

6. modifying tcp_check_req() to use to check a TFO socket as well
as request_sock

7. supporting TCP's TFO cookie option

8. adding a new SYN-ACK retransmit handler to use the timer directly
off the TFO socket rather than the listener socket. Note that TFO
server side will not retransmit anything other than SYN-ACK until
the 3WHS is completed.

The patch also contains an important function
"reqsk_fastopen_remove()" to manage the somewhat complex relation
between a listener, its request_sock, and the corresponding child
socket. See the comment above the function for the detail.

Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

13 files changed, 330 insertions, 50 deletions

diff --git a/include/net/request_sock.h b/include/net/request_sock.h
index c3cdd6c..b01d8dd 100644
--- a/include/net/request_sock.h
+++ b/include/net/request_sock.h
@@ -226,19 +226,6 @@ static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue
 	return req;
 }
 
-static inline struct sock *reqsk_queue_get_child(struct request_sock_queue *queue,
-						 struct sock *parent)
-{
-	struct request_sock *req = reqsk_queue_remove(queue);
-	struct sock *child = req->sk;
-
-	WARN_ON(child == NULL);
-
-	sk_acceptq_removed(parent);
-	__reqsk_free(req);
-	return child;
-}
-
 static inline int reqsk_queue_removed(struct request_sock_queue *queue,
 				      struct request_sock *req)
 {
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 9f8821e..1421b02 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -424,7 +424,8 @@ extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *
 						     const struct tcphdr *th);
 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
 				   struct request_sock *req,
-				   struct request_sock **prev);
+				   struct request_sock **prev,
+				   bool fastopen);
 extern int tcp_child_process(struct sock *parent, struct sock *child,
 			     struct sk_buff *skb);
 extern bool tcp_use_frto(struct sock *sk);
@@ -478,7 +479,8 @@ extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
 extern int tcp_connect(struct sock *sk);
 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 					struct request_sock *req,
-					struct request_values *rvp);
+					struct request_values *rvp,
+					struct tcp_fastopen_cookie *foc);
 extern int tcp_disconnect(struct sock *sk, int flags);
 
 void tcp_connect_init(struct sock *sk);
diff --git a/net/core/request_sock.c b/net/core/request_sock.c
index 9b570a6..c31d9e8 100644
--- a/net/core/request_sock.c
+++ b/net/core/request_sock.c
@@ -15,6 +15,7 @@
 #include <linux/random.h>
 #include <linux/slab.h>
 #include <linux/string.h>
+#include <linux/tcp.h>
 #include <linux/vmalloc.h>
 
 #include <net/request_sock.h>
@@ -130,3 +131,97 @@ void reqsk_queue_destroy(struct request_sock_queue *queue)
 		kfree(lopt);
 }
 
+/*
+ * This function is called to set a Fast Open socket's "fastopen_rsk" field
+ * to NULL when a TFO socket no longer needs to access the request_sock.
+ * This happens only after 3WHS has been either completed or aborted (e.g.,
+ * RST is received).
+ *
+ * Before TFO, a child socket is created only after 3WHS is completed,
+ * hence it never needs to access the request_sock. things get a lot more
+ * complex with TFO. A child socket, accepted or not, has to access its
+ * request_sock for 3WHS processing, e.g., to retransmit SYN-ACK pkts,
+ * until 3WHS is either completed or aborted. Afterwards the req will stay
+ * until either the child socket is accepted, or in the rare case when the
+ * listener is closed before the child is accepted.
+ *
+ * In short, a request socket is only freed after BOTH 3WHS has completed
+ * (or aborted) and the child socket has been accepted (or listener closed).
+ * When a child socket is accepted, its corresponding req->sk is set to
+ * NULL since it's no longer needed. More importantly, "req->sk == NULL"
+ * will be used by the code below to determine if a child socket has been
+ * accepted or not, and the check is protected by the fastopenq->lock
+ * described below.
+ *
+ * Note that fastopen_rsk is only accessed from the child socket's context
+ * with its socket lock held. But a request_sock (req) can be accessed by
+ * both its child socket through fastopen_rsk, and a listener socket through
+ * icsk_accept_queue.rskq_accept_head. To protect the access a simple spin
+ * lock per listener "icsk->icsk_accept_queue.fastopenq->lock" is created.
+ * only in the rare case when both the listener and the child locks are held,
+ * e.g., in inet_csk_listen_stop() do we not need to acquire the lock.
+ * The lock also protects other fields such as fastopenq->qlen, which is
+ * decremented by this function when fastopen_rsk is no longer needed.
+ *
+ * Note that another solution was to simply use the existing socket lock
+ * from the listener. But first socket lock is difficult to use. It is not
+ * a simple spin lock - one must consider sock_owned_by_user() and arrange
+ * to use sk_add_backlog() stuff. But what really makes it infeasible is the
+ * locking hierarchy violation. E.g., inet_csk_listen_stop() may try to
+ * acquire a child's lock while holding listener's socket lock. A corner
+ * case might also exist in tcp_v4_hnd_req() that will trigger this locking
+ * order.
+ *
+ * When a TFO req is created, it needs to sock_hold its listener to prevent
+ * the latter data structure from going away.
+ *
+ * This function also sets "treq->listener" to NULL and unreference listener
+ * socket. treq->listener is used by the listener so it is protected by the
+ * fastopenq->lock in this function.
+ */
+void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
+			   bool reset)
+{
+	struct sock *lsk = tcp_rsk(req)->listener;
+	struct fastopen_queue *fastopenq =
+	    inet_csk(lsk)->icsk_accept_queue.fastopenq;
+
+	BUG_ON(!spin_is_locked(&sk->sk_lock.slock) && !sock_owned_by_user(sk));
+
+	tcp_sk(sk)->fastopen_rsk = NULL;
+	spin_lock_bh(&fastopenq->lock);
+	fastopenq->qlen--;
+	tcp_rsk(req)->listener = NULL;
+	if (req->sk)	/* the child socket hasn't been accepted yet */
+		goto out;
+
+	if (!reset || lsk->sk_state != TCP_LISTEN) {
+		/* If the listener has been closed don't bother with the
+		 * special RST handling below.
+		 */
+		spin_unlock_bh(&fastopenq->lock);
+		sock_put(lsk);
+		reqsk_free(req);
+		return;
+	}
+	/* Wait for 60secs before removing a req that has triggered RST.
+	 * This is a simple defense against TFO spoofing attack - by
+	 * counting the req against fastopen.max_qlen, and disabling
+	 * TFO when the qlen exceeds max_qlen.
+	 *
+	 * For more details see CoNext'11 "TCP Fast Open" paper.
+	 */
+	req->expires = jiffies + 60*HZ;
+	if (fastopenq->rskq_rst_head == NULL)
+		fastopenq->rskq_rst_head = req;
+	else
+		fastopenq->rskq_rst_tail->dl_next = req;
+
+	req->dl_next = NULL;
+	fastopenq->rskq_rst_tail = req;
+	fastopenq->qlen++;
+out:
+	spin_unlock_bh(&fastopenq->lock);
+	sock_put(lsk);
+	return;
+}
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 6681ccf..4f70ef0 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -149,6 +149,11 @@ void inet_sock_destruct(struct sock *sk)
 		pr_err("Attempt to release alive inet socket %p\n", sk);
 		return;
 	}
+	if (sk->sk_type == SOCK_STREAM) {
+		struct fastopen_queue *fastopenq =
+			inet_csk(sk)->icsk_accept_queue.fastopenq;
+		kfree(fastopenq);
+	}
 
 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
@@ -212,6 +217,26 @@ int inet_listen(struct socket *sock, int backlog)
 	 * we can only allow the backlog to be adjusted.
 	 */
 	if (old_state != TCP_LISTEN) {
+		/* Check special setups for testing purpose to enable TFO w/o
+		 * requiring TCP_FASTOPEN sockopt.
+		 * Note that only TCP sockets (SOCK_STREAM) will reach here.
+		 * Also fastopenq may already been allocated because this
+		 * socket was in TCP_LISTEN state previously but was
+		 * shutdown() (rather than close()).
+		 */
+		if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
+		    inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
+			if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
+				err = fastopen_init_queue(sk, backlog);
+			else if ((sysctl_tcp_fastopen &
+				  TFO_SERVER_WO_SOCKOPT2) != 0)
+				err = fastopen_init_queue(sk,
+				    ((uint)sysctl_tcp_fastopen) >> 16);
+			else
+				err = 0;
+			if (err)
+				goto out;
+		}
 		err = inet_csk_listen_start(sk, backlog);
 		if (err)
 			goto out;
@@ -701,7 +726,8 @@ int inet_accept(struct socket *sock, struct socket *newsock, int flags)
 
 	sock_rps_record_flow(sk2);
 	WARN_ON(!((1 << sk2->sk_state) &
-		  (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
+		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
+		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 
 	sock_graft(sk2, newsock);
 
diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
index 7f75f21..8464b79 100644
--- a/net/ipv4/inet_connection_sock.c
+++ b/net/ipv4/inet_connection_sock.c
@@ -283,7 +283,9 @@ static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
+	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 	struct sock *newsk;
+	struct request_sock *req;
 	int error;
 
 	lock_sock(sk);
@@ -296,7 +298,7 @@ struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 		goto out_err;
 
 	/* Find already established connection */
-	if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
+	if (reqsk_queue_empty(queue)) {
 		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 
 		/* If this is a non blocking socket don't sleep */
@@ -308,14 +310,32 @@ struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 		if (error)
 			goto out_err;
 	}
-
-	newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
-	WARN_ON(newsk->sk_state == TCP_SYN_RECV);
+	req = reqsk_queue_remove(queue);
+	newsk = req->sk;
+
+	sk_acceptq_removed(sk);
+	if (sk->sk_type == SOCK_STREAM && queue->fastopenq != NULL) {
+		spin_lock_bh(&queue->fastopenq->lock);
+		if (tcp_rsk(req)->listener) {
+			/* We are still waiting for the final ACK from 3WHS
+			 * so can't free req now. Instead, we set req->sk to
+			 * NULL to signify that the child socket is taken
+			 * so reqsk_fastopen_remove() will free the req
+			 * when 3WHS finishes (or is aborted).
+			 */
+			req->sk = NULL;
+			req = NULL;
+		}
+		spin_unlock_bh(&queue->fastopenq->lock);
+	}
 out:
 	release_sock(sk);
+	if (req)
+		__reqsk_free(req);
 	return newsk;
 out_err:
 	newsk = NULL;
+	req = NULL;
 	*err = error;
 	goto out;
 }
@@ -720,13 +740,14 @@ EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 void inet_csk_listen_stop(struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
+	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 	struct request_sock *acc_req;
 	struct request_sock *req;
 
 	inet_csk_delete_keepalive_timer(sk);
 
 	/* make all the listen_opt local to us */
-	acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
+	acc_req = reqsk_queue_yank_acceptq(queue);
 
 	/* Following specs, it would be better either to send FIN
 	 * (and enter FIN-WAIT-1, it is normal close)
@@ -736,7 +757,7 @@ void inet_csk_listen_stop(struct sock *sk)
 	 * To be honest, we are not able to make either
 	 * of the variants now.			--ANK
 	 */
-	reqsk_queue_destroy(&icsk->icsk_accept_queue);
+	reqsk_queue_destroy(queue);
 
 	while ((req = acc_req) != NULL) {
 		struct sock *child = req->sk;
@@ -754,6 +775,19 @@ void inet_csk_listen_stop(struct sock *sk)
 
 		percpu_counter_inc(sk->sk_prot->orphan_count);
 
+		if (sk->sk_type == SOCK_STREAM && tcp_rsk(req)->listener) {
+			BUG_ON(tcp_sk(child)->fastopen_rsk != req);
+			BUG_ON(sk != tcp_rsk(req)->listener);
+
+			/* Paranoid, to prevent race condition if
+			 * an inbound pkt destined for child is
+			 * blocked by sock lock in tcp_v4_rcv().
+			 * Also to satisfy an assertion in
+			 * tcp_v4_destroy_sock().
+			 */
+			tcp_sk(child)->fastopen_rsk = NULL;
+			sock_put(sk);
+		}
 		inet_csk_destroy_sock(child);
 
 		bh_unlock_sock(child);
@@ -763,6 +797,17 @@ void inet_csk_listen_stop(struct sock *sk)
 		sk_acceptq_removed(sk);
 		__reqsk_free(req);
 	}
+	if (queue->fastopenq != NULL) {
+		/* Free all the reqs queued in rskq_rst_head. */
+		spin_lock_bh(&queue->fastopenq->lock);
+		acc_req = queue->fastopenq->rskq_rst_head;
+		queue->fastopenq->rskq_rst_head = NULL;
+		spin_unlock_bh(&queue->fastopenq->lock);
+		while ((req = acc_req) != NULL) {
+			acc_req = req->dl_next;
+			__reqsk_free(req);
+		}
+	}
 	WARN_ON(sk->sk_ack_backlog);
 }
 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 650e152..ba48e79 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -319,6 +319,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 	ireq->tstamp_ok		= tcp_opt.saw_tstamp;
 	req->ts_recent		= tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
 	treq->snt_synack	= tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
+	treq->listener		= NULL;
 
 	/* We throwed the options of the initial SYN away, so we hope
 	 * the ACK carries the same options again (see RFC1122 4.2.3.8)
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 2109ff4..df83d74 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -486,8 +486,9 @@ unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
 	if (sk->sk_shutdown & RCV_SHUTDOWN)
 		mask |= POLLIN | POLLRDNORM | POLLRDHUP;
 
-	/* Connected? */
-	if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
+	/* Connected or passive Fast Open socket? */
+	if (sk->sk_state != TCP_SYN_SENT &&
+	    (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) {
 		int target = sock_rcvlowat(sk, 0, INT_MAX);
 
 		if (tp->urg_seq == tp->copied_seq &&
@@ -840,10 +841,15 @@ static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffse
 	ssize_t copied;
 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 
-	/* Wait for a connection to finish. */
-	if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
+	/* Wait for a connection to finish. One exception is TCP Fast Open
+	 * (passive side) where data is allowed to be sent before a connection
+	 * is fully established.
+	 */
+	if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
+	    !tcp_passive_fastopen(sk)) {
 		if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
 			goto out_err;
+	}
 
 	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 
@@ -1042,10 +1048,15 @@ int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 
 	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 
-	/* Wait for a connection to finish. */
-	if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
+	/* Wait for a connection to finish. One exception is TCP Fast Open
+	 * (passive side) where data is allowed to be sent before a connection
+	 * is fully established.
+	 */
+	if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
+	    !tcp_passive_fastopen(sk)) {
 		if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
 			goto do_error;
+	}
 
 	if (unlikely(tp->repair)) {
 		if (tp->repair_queue == TCP_RECV_QUEUE) {
@@ -2144,6 +2155,10 @@ void tcp_close(struct sock *sk, long timeout)
 		 * they look as CLOSING or LAST_ACK for Linux)
 		 * Probably, I missed some more holelets.
 		 * 						--ANK
+		 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
+		 * in a single packet! (May consider it later but will
+		 * probably need API support or TCP_CORK SYN-ACK until
+		 * data is written and socket is closed.)
 		 */
 		tcp_send_fin(sk);
 	}
@@ -2215,8 +2230,16 @@ adjudge_to_death:
 		}
 	}
 
-	if (sk->sk_state == TCP_CLOSE)
+	if (sk->sk_state == TCP_CLOSE) {
+		struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
+		/* We could get here with a non-NULL req if the socket is
+		 * aborted (e.g., closed with unread data) before 3WHS
+		 * finishes.
+		 */
+		if (req != NULL)
+			reqsk_fastopen_remove(sk, req, false);
 		inet_csk_destroy_sock(sk);
+	}
 	/* Otherwise, socket is reprieved until protocol close. */
 
 out:
@@ -2688,6 +2711,14 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
 		else
 			icsk->icsk_user_timeout = msecs_to_jiffies(val);
 		break;
+
+	case TCP_FASTOPEN:
+		if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
+		    TCPF_LISTEN)))
+			err = fastopen_init_queue(sk, val);
+		else
+			err = -EINVAL;
+		break;
 	default:
 		err = -ENOPROTOOPT;
 		break;
@@ -3501,11 +3532,15 @@ EXPORT_SYMBOL(tcp_cookie_generator);
 
 void tcp_done(struct sock *sk)
 {
+	struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
+
 	if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
 
 	tcp_set_state(sk, TCP_CLOSE);
 	tcp_clear_xmit_timers(sk);
+	if (req != NULL)
+		reqsk_fastopen_remove(sk, req, false);
 
 	sk->sk_shutdown = SHUTDOWN_MASK;
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 36f02f9..bb148de 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -839,7 +839,7 @@ static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
 	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
 		return -1;
 
-	skb = tcp_make_synack(sk, dst, req, rvp);
+	skb = tcp_make_synack(sk, dst, req, rvp, NULL);
 
 	if (skb) {
 		__tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
@@ -1554,7 +1554,7 @@ static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 	struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
 						       iph->saddr, iph->daddr);
 	if (req)
-		return tcp_check_req(sk, skb, req, prev);
+		return tcp_check_req(sk, skb, req, prev, false);
 
 	nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
 			th->source, iph->daddr, th->dest, inet_iif(skb));
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 6ff7f10..e965319 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -507,6 +507,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
 			newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
 		newtp->rx_opt.mss_clamp = req->mss;
 		TCP_ECN_openreq_child(newtp, req);
+		newtp->fastopen_rsk = NULL;
 
 		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
 	}
@@ -515,13 +516,18 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
 EXPORT_SYMBOL(tcp_create_openreq_child);
 
 /*
- *	Process an incoming packet for SYN_RECV sockets represented
- *	as a request_sock.
+ * Process an incoming packet for SYN_RECV sockets represented as a
+ * request_sock. Normally sk is the listener socket but for TFO it
+ * points to the child socket.
+ *
+ * XXX (TFO) - The current impl contains a special check for ack
+ * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
  */
 
 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 			   struct request_sock *req,
-			   struct request_sock **prev)
+			   struct request_sock **prev,
+			   bool fastopen)
 {
 	struct tcp_options_received tmp_opt;
 	const u8 *hash_location;
@@ -530,6 +536,8 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 	__be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
 	bool paws_reject = false;
 
+	BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
+
 	tmp_opt.saw_tstamp = 0;
 	if (th->doff > (sizeof(struct tcphdr)>>2)) {
 		tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
@@ -565,6 +573,9 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 		 *
 		 * Enforce "SYN-ACK" according to figure 8, figure 6
 		 * of RFC793, fixed by RFC1122.
+		 *
+		 * Note that even if there is new data in the SYN packet
+		 * they will be thrown away too.
 		 */
 		req->rsk_ops->rtx_syn_ack(sk, req, NULL);
 		return NULL;
@@ -622,9 +633,12 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 	 *                  sent (the segment carries an unacceptable ACK) ...
 	 *                  a reset is sent."
 	 *
-	 * Invalid ACK: reset will be sent by listening socket
+	 * Invalid ACK: reset will be sent by listening socket.
+	 * Note that the ACK validity check for a Fast Open socket is done
+	 * elsewhere and is checked directly against the child socket rather
+	 * than req because user data may have been sent out.
 	 */
-	if ((flg & TCP_FLAG_ACK) &&
+	if ((flg & TCP_FLAG_ACK) && !fastopen &&
 	    (TCP_SKB_CB(skb)->ack_seq !=
 	     tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk))))
 		return sk;
@@ -637,7 +651,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 	/* RFC793: "first check sequence number". */
 
 	if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
-					  tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) {
+					  tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
 		/* Out of window: send ACK and drop. */
 		if (!(flg & TCP_FLAG_RST))
 			req->rsk_ops->send_ack(sk, skb, req);
@@ -648,7 +662,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 
 	/* In sequence, PAWS is OK. */
 
-	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1))
+	if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
 		req->ts_recent = tmp_opt.rcv_tsval;
 
 	if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
@@ -667,10 +681,19 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 
 	/* ACK sequence verified above, just make sure ACK is
 	 * set.  If ACK not set, just silently drop the packet.
+	 *
+	 * XXX (TFO) - if we ever allow "data after SYN", the
+	 * following check needs to be removed.
 	 */
 	if (!(flg & TCP_FLAG_ACK))
 		return NULL;
 
+	/* For Fast Open no more processing is needed (sk is the
+	 * child socket).
+	 */
+	if (fastopen)
+		return sk;
+
 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
 	if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
@@ -706,11 +729,21 @@ listen_overflow:
 	}
 
 embryonic_reset:
-	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
-	if (!(flg & TCP_FLAG_RST))
+	if (!(flg & TCP_FLAG_RST)) {
+		/* Received a bad SYN pkt - for TFO We try not to reset
+		 * the local connection unless it's really necessary to
+		 * avoid becoming vulnerable to outside attack aiming at
+		 * resetting legit local connections.
+		 */
 		req->rsk_ops->send_reset(sk, skb);
-
-	inet_csk_reqsk_queue_drop(sk, req, prev);
+	} else if (fastopen) { /* received a valid RST pkt */
+		reqsk_fastopen_remove(sk, req, true);
+		tcp_reset(sk);
+	}
+	if (!fastopen) {
+		inet_csk_reqsk_queue_drop(sk, req, prev);
+		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
+	}
 	return NULL;
 }
 EXPORT_SYMBOL(tcp_check_req);
@@ -719,6 +752,12 @@ EXPORT_SYMBOL(tcp_check_req);
  * Queue segment on the new socket if the new socket is active,
  * otherwise we just shortcircuit this and continue with
  * the new socket.
+ *
+ * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
+ * when entering. But other states are possible due to a race condition
+ * where after __inet_lookup_established() fails but before the listener
+ * locked is obtained, other packets cause the same connection to
+ * be created.
  */
 
 int tcp_child_process(struct sock *parent, struct sock *child,
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index d046326..9383b51 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -702,7 +702,8 @@ static unsigned int tcp_synack_options(struct sock *sk,
 				   unsigned int mss, struct sk_buff *skb,
 				   struct tcp_out_options *opts,
 				   struct tcp_md5sig_key **md5,
-				   struct tcp_extend_values *xvp)
+				   struct tcp_extend_values *xvp,
+				   struct tcp_fastopen_cookie *foc)
 {
 	struct inet_request_sock *ireq = inet_rsk(req);
 	unsigned int remaining = MAX_TCP_OPTION_SPACE;
@@ -747,7 +748,15 @@ static unsigned int tcp_synack_options(struct sock *sk,
 		if (unlikely(!ireq->tstamp_ok))
 			remaining -= TCPOLEN_SACKPERM_ALIGNED;
 	}
-
+	if (foc != NULL) {
+		u32 need = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
+		need = (need + 3) & ~3U;  /* Align to 32 bits */
+		if (remaining >= need) {
+			opts->options |= OPTION_FAST_OPEN_COOKIE;
+			opts->fastopen_cookie = foc;
+			remaining -= need;
+		}
+	}
 	/* Similar rationale to tcp_syn_options() applies here, too.
 	 * If the <SYN> options fit, the same options should fit now!
 	 */
@@ -2658,7 +2667,8 @@ int tcp_send_synack(struct sock *sk)
  */
 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 				struct request_sock *req,
-				struct request_values *rvp)
+				struct request_values *rvp,
+				struct tcp_fastopen_cookie *foc)
 {
 	struct tcp_out_options opts;
 	struct tcp_extend_values *xvp = tcp_xv(rvp);
@@ -2718,7 +2728,7 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 #endif
 	TCP_SKB_CB(skb)->when = tcp_time_stamp;
 	tcp_header_size = tcp_synack_options(sk, req, mss,
-					     skb, &opts, &md5, xvp)
+					     skb, &opts, &md5, xvp, foc)
 			+ sizeof(*th);
 
 	skb_push(skb, tcp_header_size);
@@ -2772,7 +2782,8 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 	}
 
 	th->seq = htonl(TCP_SKB_CB(skb)->seq);
-	th->ack_seq = htonl(tcp_rsk(req)->rcv_isn + 1);
+	/* XXX data is queued and acked as is. No buffer/window check */
+	th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt);
 
 	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
 	th->window = htons(min(req->rcv_wnd, 65535U));
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index b774a03..fc04711 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -305,6 +305,35 @@ static void tcp_probe_timer(struct sock *sk)
 }
 
 /*
+ *	Timer for Fast Open socket to retransmit SYNACK. Note that the
+ *	sk here is the child socket, not the parent (listener) socket.
+ */
+static void tcp_fastopen_synack_timer(struct sock *sk)
+{
+	struct inet_connection_sock *icsk = inet_csk(sk);
+	int max_retries = icsk->icsk_syn_retries ? :
+	    sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
+	struct request_sock *req;
+
+	req = tcp_sk(sk)->fastopen_rsk;
+	req->rsk_ops->syn_ack_timeout(sk, req);
+
+	if (req->retrans >= max_retries) {
+		tcp_write_err(sk);
+		return;
+	}
+	/* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
+	 * returned from rtx_syn_ack() to make it more persistent like
+	 * regular retransmit because if the child socket has been accepted
+	 * it's not good to give up too easily.
+	 */
+	req->rsk_ops->rtx_syn_ack(sk, req, NULL);
+	req->retrans++;
+	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+			  TCP_TIMEOUT_INIT << req->retrans, TCP_RTO_MAX);
+}
+
+/*
  *	The TCP retransmit timer.
  */
 
@@ -317,7 +346,15 @@ void tcp_retransmit_timer(struct sock *sk)
 		tcp_resume_early_retransmit(sk);
 		return;
 	}
-
+	if (tp->fastopen_rsk) {
+		BUG_ON(sk->sk_state != TCP_SYN_RECV &&
+		    sk->sk_state != TCP_FIN_WAIT1);
+		tcp_fastopen_synack_timer(sk);
+		/* Before we receive ACK to our SYN-ACK don't retransmit
+		 * anything else (e.g., data or FIN segments).
+		 */
+		return;
+	}
 	if (!tp->packets_out)
 		goto out;
 
diff --git a/net/ipv6/syncookies.c b/net/ipv6/syncookies.c
index bb46061..182ab9a 100644
--- a/net/ipv6/syncookies.c
+++ b/net/ipv6/syncookies.c
@@ -190,6 +190,7 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
 	ireq = inet_rsk(req);
 	ireq6 = inet6_rsk(req);
 	treq = tcp_rsk(req);
+	treq->listener = NULL;
 
 	if (security_inet_conn_request(sk, skb, req))
 		goto out_free;
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index f99b81d..09078b9 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -475,7 +475,7 @@ static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
 	if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
 		goto done;
 
-	skb = tcp_make_synack(sk, dst, req, rvp);
+	skb = tcp_make_synack(sk, dst, req, rvp, NULL);
 
 	if (skb) {
 		__tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
@@ -987,7 +987,7 @@ static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
 				   &ipv6_hdr(skb)->saddr,
 				   &ipv6_hdr(skb)->daddr, inet6_iif(skb));
 	if (req)
-		return tcp_check_req(sk, skb, req, prev);
+		return tcp_check_req(sk, skb, req, prev, false);
 
 	nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
 			&ipv6_hdr(skb)->saddr, th->source,
@@ -1179,6 +1179,7 @@ have_isn:
 	    want_cookie)
 		goto drop_and_free;
 
+	tcp_rsk(req)->listener = NULL;
 	inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
 	return 0;