tcp: TCP Fast Open Server - main code path

This patch adds the main processing path to complete the TFO server
patches.

A TFO request (i.e., SYN+data packet with a TFO cookie option) first
gets processed in tcp_v4_conn_request(). If it passes the various TFO
checks by tcp_fastopen_check(), a child socket will be created right
away to be accepted by applications, rather than waiting for the 3WHS
to finish.

In additon to the use of TFO cookie, a simple max_qlen based scheme
is put in place to fend off spoofed TFO attack.

When a valid ACK comes back to tcp_rcv_state_process(), it will cause
the state of the child socket to switch from either TCP_SYN_RECV to
TCP_ESTABLISHED, or TCP_FIN_WAIT1 to TCP_FIN_WAIT2. At this time
retransmission will resume for any unack'ed (data, FIN,...) segments.

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>

2 files changed, 309 insertions, 27 deletions

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index d47d5fe..8c304a4 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -3127,6 +3127,12 @@ void tcp_rearm_rto(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
+	/* If the retrans timer is currently being used by Fast Open
+	 * for SYN-ACK retrans purpose, stay put.
+	 */
+	if (tp->fastopen_rsk)
+		return;
+
 	if (!tp->packets_out) {
 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
 	} else {
@@ -5895,7 +5901,9 @@ discard:
 		tcp_send_synack(sk);
 #if 0
 		/* Note, we could accept data and URG from this segment.
-		 * There are no obstacles to make this.
+		 * There are no obstacles to make this (except that we must
+		 * either change tcp_recvmsg() to prevent it from returning data
+		 * before 3WHS completes per RFC793, or employ TCP Fast Open).
 		 *
 		 * However, if we ignore data in ACKless segments sometimes,
 		 * we have no reasons to accept it sometimes.
@@ -5935,6 +5943,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct inet_connection_sock *icsk = inet_csk(sk);
+	struct request_sock *req;
 	int queued = 0;
 
 	tp->rx_opt.saw_tstamp = 0;
@@ -5990,7 +5999,14 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 		return 0;
 	}
 
-	if (!tcp_validate_incoming(sk, skb, th, 0))
+	req = tp->fastopen_rsk;
+	if (req != NULL) {
+		BUG_ON(sk->sk_state != TCP_SYN_RECV &&
+		    sk->sk_state != TCP_FIN_WAIT1);
+
+		if (tcp_check_req(sk, skb, req, NULL, true) == NULL)
+			goto discard;
+	} else if (!tcp_validate_incoming(sk, skb, th, 0))
 		return 0;
 
 	/* step 5: check the ACK field */
@@ -6000,7 +6016,22 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 		switch (sk->sk_state) {
 		case TCP_SYN_RECV:
 			if (acceptable) {
-				tp->copied_seq = tp->rcv_nxt;
+				/* Once we leave TCP_SYN_RECV, we no longer
+				 * need req so release it.
+				 */
+				if (req) {
+					reqsk_fastopen_remove(sk, req, false);
+				} else {
+					/* Make sure socket is routed, for
+					 * correct metrics.
+					 */
+					icsk->icsk_af_ops->rebuild_header(sk);
+					tcp_init_congestion_control(sk);
+
+					tcp_mtup_init(sk);
+					tcp_init_buffer_space(sk);
+					tp->copied_seq = tp->rcv_nxt;
+				}
 				smp_mb();
 				tcp_set_state(sk, TCP_ESTABLISHED);
 				sk->sk_state_change(sk);
@@ -6022,23 +6053,27 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 				if (tp->rx_opt.tstamp_ok)
 					tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
 
-				/* Make sure socket is routed, for
-				 * correct metrics.
-				 */
-				icsk->icsk_af_ops->rebuild_header(sk);
-
-				tcp_init_metrics(sk);
-
-				tcp_init_congestion_control(sk);
+				if (req) {
+					/* Re-arm the timer because data may
+					 * have been sent out. This is similar
+					 * to the regular data transmission case
+					 * when new data has just been ack'ed.
+					 *
+					 * (TFO) - we could try to be more
+					 * aggressive and retranmitting any data
+					 * sooner based on when they were sent
+					 * out.
+					 */
+					tcp_rearm_rto(sk);
+				} else
+					tcp_init_metrics(sk);
 
 				/* Prevent spurious tcp_cwnd_restart() on
 				 * first data packet.
 				 */
 				tp->lsndtime = tcp_time_stamp;
 
-				tcp_mtup_init(sk);
 				tcp_initialize_rcv_mss(sk);
-				tcp_init_buffer_space(sk);
 				tcp_fast_path_on(tp);
 			} else {
 				return 1;
@@ -6046,6 +6081,16 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 			break;
 
 		case TCP_FIN_WAIT1:
+			/* If we enter the TCP_FIN_WAIT1 state and we are a
+			 * Fast Open socket and this is the first acceptable
+			 * ACK we have received, this would have acknowledged
+			 * our SYNACK so stop the SYNACK timer.
+			 */
+			if (acceptable && req != NULL) {
+				/* We no longer need the request sock. */
+				reqsk_fastopen_remove(sk, req, false);
+				tcp_rearm_rto(sk);
+			}
 			if (tp->snd_una == tp->write_seq) {
 				struct dst_entry *dst;
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index bb148de..e64abed 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -352,6 +352,7 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 	const int code = icmp_hdr(icmp_skb)->code;
 	struct sock *sk;
 	struct sk_buff *skb;
+	struct request_sock *req;
 	__u32 seq;
 	__u32 remaining;
 	int err;
@@ -394,9 +395,12 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 
 	icsk = inet_csk(sk);
 	tp = tcp_sk(sk);
+	req = tp->fastopen_rsk;
 	seq = ntohl(th->seq);
 	if (sk->sk_state != TCP_LISTEN &&
-	    !between(seq, tp->snd_una, tp->snd_nxt)) {
+	    !between(seq, tp->snd_una, tp->snd_nxt) &&
+	    (req == NULL || seq != tcp_rsk(req)->snt_isn)) {
+		/* For a Fast Open socket, allow seq to be snt_isn. */
 		NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 		goto out;
 	}
@@ -435,6 +439,8 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 		    !icsk->icsk_backoff)
 			break;
 
+		/* XXX (TFO) - revisit the following logic for TFO */
+
 		if (sock_owned_by_user(sk))
 			break;
 
@@ -466,6 +472,14 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 		goto out;
 	}
 
+	/* XXX (TFO) - if it's a TFO socket and has been accepted, rather
+	 * than following the TCP_SYN_RECV case and closing the socket,
+	 * we ignore the ICMP error and keep trying like a fully established
+	 * socket. Is this the right thing to do?
+	 */
+	if (req && req->sk == NULL)
+		goto out;
+
 	switch (sk->sk_state) {
 		struct request_sock *req, **prev;
 	case TCP_LISTEN:
@@ -498,7 +512,8 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 
 	case TCP_SYN_SENT:
 	case TCP_SYN_RECV:  /* Cannot happen.
-			       It can f.e. if SYNs crossed.
+			       It can f.e. if SYNs crossed,
+			       or Fast Open.
 			     */
 		if (!sock_owned_by_user(sk)) {
 			sk->sk_err = err;
@@ -809,8 +824,12 @@ static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
 				  struct request_sock *req)
 {
-	tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
-			tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
+	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
+	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
+	 */
+	tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
+			tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
+			tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
 			req->ts_recent,
 			0,
 			tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
@@ -1272,6 +1291,178 @@ static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
 };
 #endif
 
+static bool tcp_fastopen_check(struct sock *sk, struct sk_buff *skb,
+			       struct request_sock *req,
+			       struct tcp_fastopen_cookie *foc,
+			       struct tcp_fastopen_cookie *valid_foc)
+{
+	bool skip_cookie = false;
+	struct fastopen_queue *fastopenq;
+
+	if (likely(!fastopen_cookie_present(foc))) {
+		/* See include/net/tcp.h for the meaning of these knobs */
+		if ((sysctl_tcp_fastopen & TFO_SERVER_ALWAYS) ||
+		    ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD) &&
+		    (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1)))
+			skip_cookie = true; /* no cookie to validate */
+		else
+			return false;
+	}
+	fastopenq = inet_csk(sk)->icsk_accept_queue.fastopenq;
+	/* A FO option is present; bump the counter. */
+	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVE);
+
+	/* Make sure the listener has enabled fastopen, and we don't
+	 * exceed the max # of pending TFO requests allowed before trying
+	 * to validating the cookie in order to avoid burning CPU cycles
+	 * unnecessarily.
+	 *
+	 * XXX (TFO) - The implication of checking the max_qlen before
+	 * processing a cookie request is that clients can't differentiate
+	 * between qlen overflow causing Fast Open to be disabled
+	 * temporarily vs a server not supporting Fast Open at all.
+	 */
+	if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) == 0 ||
+	    fastopenq == NULL || fastopenq->max_qlen == 0)
+		return false;
+
+	if (fastopenq->qlen >= fastopenq->max_qlen) {
+		struct request_sock *req1;
+		spin_lock(&fastopenq->lock);
+		req1 = fastopenq->rskq_rst_head;
+		if ((req1 == NULL) || time_after(req1->expires, jiffies)) {
+			spin_unlock(&fastopenq->lock);
+			NET_INC_STATS_BH(sock_net(sk),
+			    LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
+			/* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
+			foc->len = -1;
+			return false;
+		}
+		fastopenq->rskq_rst_head = req1->dl_next;
+		fastopenq->qlen--;
+		spin_unlock(&fastopenq->lock);
+		reqsk_free(req1);
+	}
+	if (skip_cookie) {
+		tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
+		return true;
+	}
+	if (foc->len == TCP_FASTOPEN_COOKIE_SIZE) {
+		if ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_CHKED) == 0) {
+			tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
+			if ((valid_foc->len != TCP_FASTOPEN_COOKIE_SIZE) ||
+			    memcmp(&foc->val[0], &valid_foc->val[0],
+			    TCP_FASTOPEN_COOKIE_SIZE) != 0)
+				return false;
+			valid_foc->len = -1;
+		}
+		/* Acknowledge the data received from the peer. */
+		tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
+		return true;
+	} else if (foc->len == 0) { /* Client requesting a cookie */
+		tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
+		NET_INC_STATS_BH(sock_net(sk),
+		    LINUX_MIB_TCPFASTOPENCOOKIEREQD);
+	} else {
+		/* Client sent a cookie with wrong size. Treat it
+		 * the same as invalid and return a valid one.
+		 */
+		tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
+	}
+	return false;
+}
+
+static int tcp_v4_conn_req_fastopen(struct sock *sk,
+				    struct sk_buff *skb,
+				    struct sk_buff *skb_synack,
+				    struct request_sock *req,
+				    struct request_values *rvp)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
+	const struct inet_request_sock *ireq = inet_rsk(req);
+	struct sock *child;
+
+	req->retrans = 0;
+	req->sk = NULL;
+
+	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
+	if (child == NULL) {
+		NET_INC_STATS_BH(sock_net(sk),
+				 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
+		kfree_skb(skb_synack);
+		return -1;
+	}
+	ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr,
+			ireq->rmt_addr, ireq->opt);
+	/* XXX (TFO) - is it ok to ignore error and continue? */
+
+	spin_lock(&queue->fastopenq->lock);
+	queue->fastopenq->qlen++;
+	spin_unlock(&queue->fastopenq->lock);
+
+	/* Initialize the child socket. Have to fix some values to take
+	 * into account the child is a Fast Open socket and is created
+	 * only out of the bits carried in the SYN packet.
+	 */
+	tp = tcp_sk(child);
+
+	tp->fastopen_rsk = req;
+	/* Do a hold on the listner sk so that if the listener is being
+	 * closed, the child that has been accepted can live on and still
+	 * access listen_lock.
+	 */
+	sock_hold(sk);
+	tcp_rsk(req)->listener = sk;
+
+	/* RFC1323: The window in SYN & SYN/ACK segments is never
+	 * scaled. So correct it appropriately.
+	 */
+	tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
+
+	/* Activate the retrans timer so that SYNACK can be retransmitted.
+	 * The request socket is not added to the SYN table of the parent
+	 * because it's been added to the accept queue directly.
+	 */
+	inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
+	    TCP_TIMEOUT_INIT, TCP_RTO_MAX);
+
+	/* Add the child socket directly into the accept queue */
+	inet_csk_reqsk_queue_add(sk, req, child);
+
+	/* Now finish processing the fastopen child socket. */
+	inet_csk(child)->icsk_af_ops->rebuild_header(child);
+	tcp_init_congestion_control(child);
+	tcp_mtup_init(child);
+	tcp_init_buffer_space(child);
+	tcp_init_metrics(child);
+
+	/* Queue the data carried in the SYN packet. We need to first
+	 * bump skb's refcnt because the caller will attempt to free it.
+	 *
+	 * XXX (TFO) - we honor a zero-payload TFO request for now.
+	 * (Any reason not to?)
+	 */
+	if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq + 1) {
+		/* Don't queue the skb if there is no payload in SYN.
+		 * XXX (TFO) - How about SYN+FIN?
+		 */
+		tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
+	} else {
+		skb = skb_get(skb);
+		skb_dst_drop(skb);
+		__skb_pull(skb, tcp_hdr(skb)->doff * 4);
+		skb_set_owner_r(skb, child);
+		__skb_queue_tail(&child->sk_receive_queue, skb);
+		tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
+	}
+	sk->sk_data_ready(sk, 0);
+	bh_unlock_sock(child);
+	sock_put(child);
+	WARN_ON(req->sk == NULL);
+	return 0;
+}
+
 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 {
 	struct tcp_extend_values tmp_ext;
@@ -1285,6 +1476,11 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 	__be32 daddr = ip_hdr(skb)->daddr;
 	__u32 isn = TCP_SKB_CB(skb)->when;
 	bool want_cookie = false;
+	struct flowi4 fl4;
+	struct tcp_fastopen_cookie foc = { .len = -1 };
+	struct tcp_fastopen_cookie valid_foc = { .len = -1 };
+	struct sk_buff *skb_synack;
+	int do_fastopen;
 
 	/* Never answer to SYNs send to broadcast or multicast */
 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
@@ -1319,7 +1515,8 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 	tcp_clear_options(&tmp_opt);
 	tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
 	tmp_opt.user_mss  = tp->rx_opt.user_mss;
-	tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
+	tcp_parse_options(skb, &tmp_opt, &hash_location, 0,
+	    want_cookie ? NULL : &foc);
 
 	if (tmp_opt.cookie_plus > 0 &&
 	    tmp_opt.saw_tstamp &&
@@ -1377,8 +1574,6 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 		isn = cookie_v4_init_sequence(sk, skb, &req->mss);
 		req->cookie_ts = tmp_opt.tstamp_ok;
 	} else if (!isn) {
-		struct flowi4 fl4;
-
 		/* VJ's idea. We save last timestamp seen
 		 * from the destination in peer table, when entering
 		 * state TIME-WAIT, and check against it before
@@ -1419,14 +1614,52 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 	tcp_rsk(req)->snt_isn = isn;
 	tcp_rsk(req)->snt_synack = tcp_time_stamp;
 
-	if (tcp_v4_send_synack(sk, dst, req,
-			       (struct request_values *)&tmp_ext,
-			       skb_get_queue_mapping(skb),
-			       want_cookie) ||
-	    want_cookie)
+	if (dst == NULL) {
+		dst = inet_csk_route_req(sk, &fl4, req);
+		if (dst == NULL)
+			goto drop_and_free;
+	}
+	do_fastopen = tcp_fastopen_check(sk, skb, req, &foc, &valid_foc);
+
+	/* We don't call tcp_v4_send_synack() directly because we need
+	 * to make sure a child socket can be created successfully before
+	 * sending back synack!
+	 *
+	 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
+	 * (or better yet, call tcp_send_synack() in the child context
+	 * directly, but will have to fix bunch of other code first)
+	 * after syn_recv_sock() except one will need to first fix the
+	 * latter to remove its dependency on the current implementation
+	 * of tcp_v4_send_synack()->tcp_select_initial_window().
+	 */
+	skb_synack = tcp_make_synack(sk, dst, req,
+	    (struct request_values *)&tmp_ext,
+	    fastopen_cookie_present(&valid_foc) ? &valid_foc : NULL);
+
+	if (skb_synack) {
+		__tcp_v4_send_check(skb_synack, ireq->loc_addr, ireq->rmt_addr);
+		skb_set_queue_mapping(skb_synack, skb_get_queue_mapping(skb));
+	} else
+		goto drop_and_free;
+
+	if (likely(!do_fastopen)) {
+		int err;
+		err = ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr,
+		     ireq->rmt_addr, ireq->opt);
+		err = net_xmit_eval(err);
+		if (err || want_cookie)
+			goto drop_and_free;
+
+		tcp_rsk(req)->listener = NULL;
+		/* Add the request_sock to the SYN table */
+		inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
+		if (fastopen_cookie_present(&foc) && foc.len != 0)
+			NET_INC_STATS_BH(sock_net(sk),
+			    LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
+	} else if (tcp_v4_conn_req_fastopen(sk, skb, skb_synack, req,
+	    (struct request_values *)&tmp_ext))
 		goto drop_and_free;
 
-	inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
 	return 0;
 
 drop_and_release:
@@ -1977,6 +2210,7 @@ void tcp_v4_destroy_sock(struct sock *sk)
 			 tcp_cookie_values_release);
 		tp->cookie_values = NULL;
 	}
+	BUG_ON(tp->fastopen_rsk != NULL);
 
 	/* If socket is aborted during connect operation */
 	tcp_free_fastopen_req(tp);
@@ -2425,6 +2659,7 @@ static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
 	const struct tcp_sock *tp = tcp_sk(sk);
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	const struct inet_sock *inet = inet_sk(sk);
+	struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
 	__be32 dest = inet->inet_daddr;
 	__be32 src = inet->inet_rcv_saddr;
 	__u16 destp = ntohs(inet->inet_dport);
@@ -2469,7 +2704,9 @@ static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
 		jiffies_to_clock_t(icsk->icsk_ack.ato),
 		(icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
 		tp->snd_cwnd,
-		tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
+		sk->sk_state == TCP_LISTEN ?
+		    (fastopenq ? fastopenq->max_qlen : 0) :
+		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh),
 		len);
 }