tcp: early retransmit

This patch implements RFC 5827 early retransmit (ER) for TCP.
It reduces DUPACK threshold (dupthresh) if outstanding packets are
less than 4 to recover losses by fast recovery instead of timeout.

While the algorithm is simple, small but frequent network reordering
makes this feature dangerous: the connection repeatedly enter
false recovery and degrade performance. Therefore we implement
a mitigation suggested in the appendix of the RFC that delays
entering fast recovery by a small interval, i.e., RTT/4. Currently
ER is conservative and is disabled for the rest of the connection
after the first reordering event. A large scale web server
experiment on the performance impact of ER is summarized in
section 6 of the paper "Proportional Rate Reduction for TCP”,
IMC 2011. http://conferences.sigcomm.org/imc/2011/docs/p155.pdf

Note that Linux has a similar feature called THIN_DUPACK. The
differences are THIN_DUPACK do not mitigate reorderings and is only
used after slow start. Currently ER is disabled if THIN_DUPACK is
enabled. I would be happy to merge THIN_DUPACK feature with ER if
people think it's a good idea.

ER is enabled by sysctl_tcp_early_retrans:
  0: Disables ER

  1: Reduce dupthresh to packets_out - 1 when outstanding packets < 4.

  2: (Default) reduce dupthresh like mode 1. In addition, delay
     entering fast recovery by RTT/4.

Note: mode 2 is implemented in the third part of this patch series.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

7 files changed, 59 insertions, 0 deletions

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index 9b569a2d9c60..34916e792d9d 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -190,6 +190,20 @@ tcp_cookie_size - INTEGER
 tcp_dsack - BOOLEAN
         Allows TCP to send "duplicate" SACKs.
 
+tcp_early_retrans - INTEGER
+        Enable Early Retransmit (ER), per RFC 5827. ER lowers the threshold
+        for triggering fast retransmit when the amount of outstanding data is
+        small and when no previously unsent data can be transmitted (such
+        that limited transmit could be used).
+        Possible values:
+                0 disables ER
+                1 enables ER
+                2 enables ER but delays fast recovery and fast retransmit
+                  by a fourth of RTT. This mitigates connection falsely
+                  recovers when network has a small degree of reordering
+                  (less than 3 packets).
+        Default: 2
+
 tcp_ecn - INTEGER
         Enable Explicit Congestion Notification (ECN) in TCP. ECN is only
         used when both ends of the TCP flow support it. It is useful to
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 278af9ea42d4..7859b416d46e 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -372,6 +372,7 @@ struct tcp_sock {
                 repair      : 1,
                 unused      : 1;
         u8      repair_queue;
+        u8      do_early_retrans:1;/* Enable RFC5827 early-retransmit  */
 
 /* RTT measurement */
         u32     srtt;           /* smoothed round trip time << 3        */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0fb84de6da36..685437a16c97 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -252,6 +252,7 @@ extern int sysctl_tcp_max_ssthresh;
 extern int sysctl_tcp_cookie_size;
 extern int sysctl_tcp_thin_linear_timeouts;
 extern int sysctl_tcp_thin_dupack;
+extern int sysctl_tcp_early_retrans;
 
 extern atomic_long_t tcp_memory_allocated;
 extern struct percpu_counter tcp_sockets_allocated;
@@ -797,6 +798,20 @@ static inline void tcp_enable_fack(struct tcp_sock *tp)
         tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
 }
 
+/* TCP early-retransmit (ER) is similar to but more conservative than
+ * the thin-dupack feature.  Enable ER only if thin-dupack is disabled.
+ */
+static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
+{
+        tp->do_early_retrans = sysctl_tcp_early_retrans &&
+                !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
+}
+
+static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
+{
+        tp->do_early_retrans = 0;
+}
+
 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
 {
         return tp->sacked_out + tp->lost_out;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 33417f84e07f..ef32956ed655 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -27,6 +27,7 @@
 #include <net/tcp_memcontrol.h>
 
 static int zero;
+static int two = 2;
 static int tcp_retr1_max = 255;
 static int ip_local_port_range_min[] = { 1, 1 };
 static int ip_local_port_range_max[] = { 65535, 65535 };
@@ -677,6 +678,15 @@ static struct ctl_table ipv4_table[] = {
                 .proc_handler   = proc_dointvec
         },
         {
+                .procname       = "tcp_early_retrans",
+                .data           = &sysctl_tcp_early_retrans,
+                .maxlen         = sizeof(int),
+                .mode           = 0644,
+                .proc_handler   = proc_dointvec_minmax,
+                .extra1         = &zero,
+                .extra2         = &two,
+        },
+        {
                 .procname       = "udp_mem",
                 .data           = &sysctl_udp_mem,
                 .maxlen         = sizeof(sysctl_udp_mem),
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 9670af341931..6802c89bc44d 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -395,6 +395,7 @@ void tcp_init_sock(struct sock *sk)
         tp->mss_cache = TCP_MSS_DEFAULT;
 
         tp->reordering = sysctl_tcp_reordering;
+        tcp_enable_early_retrans(tp);
         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
 
         sk->sk_state = TCP_CLOSE;
@@ -2495,6 +2496,8 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
                         err = -EINVAL;
                 else
                         tp->thin_dupack = val;
+                        if (tp->thin_dupack)
+                                tcp_disable_early_retrans(tp);
                 break;
 
         case TCP_REPAIR:
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index be8e09d2c6b1..e042cabb695e 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -99,6 +99,7 @@ int sysctl_tcp_thin_dupack __read_mostly;
 
 int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
 int sysctl_tcp_abc __read_mostly;
+int sysctl_tcp_early_retrans __read_mostly = 2;
 
 #define FLAG_DATA               0x01 /* Incoming frame contained data.          */
 #define FLAG_WIN_UPDATE         0x02 /* Incoming ACK was a window update.       */
@@ -906,6 +907,7 @@ static void tcp_init_metrics(struct sock *sk)
         if (dst_metric(dst, RTAX_REORDERING) &&
             tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
                 tcp_disable_fack(tp);
+                tcp_disable_early_retrans(tp);
                 tp->reordering = dst_metric(dst, RTAX_REORDERING);
         }
 
@@ -988,6 +990,9 @@ static void tcp_update_reordering(struct sock *sk, const int metric,
 #endif
                 tcp_disable_fack(tp);
         }
+
+        if (metric > 0)
+                tcp_disable_early_retrans(tp);
 }
 
 /* This must be called before lost_out is incremented */
@@ -2492,6 +2497,16 @@ static int tcp_time_to_recover(struct sock *sk)
             tcp_is_sack(tp) && !tcp_send_head(sk))
                 return 1;
 
+        /* Trick#6: TCP early retransmit, per RFC5827.  To avoid spurious
+         * retransmissions due to small network reorderings, we implement
+         * Mitigation A.3 in the RFC and delay the retransmission for a short
+         * interval if appropriate.
+         */
+        if (tp->do_early_retrans && !tp->retrans_out && tp->sacked_out &&
+            (tp->packets_out == (tp->sacked_out + 1) && tp->packets_out < 4) &&
+            !tcp_may_send_now(sk))
+                return 1;
+
         return 0;
 }
 
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 3cabafb5cdd1..6f6a91832826 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -482,6 +482,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req,
                 newtp->sacked_out = 0;
                 newtp->fackets_out = 0;
                 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
+                tcp_enable_early_retrans(newtp);
 
                 /* So many TCP implementations out there (incorrectly) count the
                  * initial SYN frame in their delayed-ACK and congestion control