summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorEric Dumazet <edumazet@google.com>2016-04-21 10:55:23 -0700
committerDavid S. Miller <davem@davemloft.net>2016-04-24 14:43:59 -0400
commit10d3be569243def8d92ac3722395ef5a59c504e6 (patch)
treeac01b70cff99ad2e59c54e80ddcd524eaa9691a8
parent8cee83dd29dea4e7d27fda3b170381059f628868 (diff)
downloadlinux-10d3be569243def8d92ac3722395ef5a59c504e6.tar.gz
linux-10d3be569243def8d92ac3722395ef5a59c504e6.tar.bz2
linux-10d3be569243def8d92ac3722395ef5a59c504e6.zip
tcp-tso: do not split TSO packets at retransmit time
Linux TCP stack painfully segments all TSO/GSO packets before retransmits. This was fine back in the days when TSO/GSO were emerging, with their bugs, but we believe the dark age is over. Keeping big packets in write queues, but also in stack traversal has a lot of benefits. - Less memory overhead, because write queues have less skbs - Less cpu overhead at ACK processing. - Better SACK processing, as lot of studies mentioned how awful linux was at this ;) - Less cpu overhead to send the rtx packets (IP stack traversal, netfilter traversal, drivers...) - Better latencies in presence of losses. - Smaller spikes in fq like packet schedulers, as retransmits are not constrained by TCP Small Queues. 1 % packet losses are common today, and at 100Gbit speeds, this translates to ~80,000 losses per second. Losses are often correlated, and we see many retransmit events leading to 1-MSS train of packets, at the time hosts are already under stress. Signed-off-by: Eric Dumazet <edumazet@google.com> Acked-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
-rw-r--r--include/net/tcp.h4
-rw-r--r--net/ipv4/tcp_input.c2
-rw-r--r--net/ipv4/tcp_output.c64
-rw-r--r--net/ipv4/tcp_timer.c4
4 files changed, 34 insertions, 40 deletions
diff --git a/include/net/tcp.h b/include/net/tcp.h
index c0ef0544dfcf..7f2553da10d1 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -538,8 +538,8 @@ __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
int nonagle);
bool tcp_may_send_now(struct sock *sk);
-int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
-int tcp_retransmit_skb(struct sock *, struct sk_buff *);
+int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
+int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
void tcp_retransmit_timer(struct sock *sk);
void tcp_xmit_retransmit_queue(struct sock *);
void tcp_simple_retransmit(struct sock *);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 75e8336f6ecd..dcad8f9f96eb 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5545,7 +5545,7 @@ static bool tcp_rcv_fastopen_synack(struct sock *sk, struct sk_buff *synack,
if (data) { /* Retransmit unacked data in SYN */
tcp_for_write_queue_from(data, sk) {
if (data == tcp_send_head(sk) ||
- __tcp_retransmit_skb(sk, data))
+ __tcp_retransmit_skb(sk, data, 1))
break;
}
tcp_rearm_rto(sk);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index a6e4a8353b02..9d3b4b364652 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2268,7 +2268,7 @@ void tcp_send_loss_probe(struct sock *sk)
if (WARN_ON(!skb || !tcp_skb_pcount(skb)))
goto rearm_timer;
- if (__tcp_retransmit_skb(sk, skb))
+ if (__tcp_retransmit_skb(sk, skb, 1))
goto rearm_timer;
/* Record snd_nxt for loss detection. */
@@ -2571,17 +2571,17 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to,
* state updates are done by the caller. Returns non-zero if an
* error occurred which prevented the send.
*/
-int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
+int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
{
- struct tcp_sock *tp = tcp_sk(sk);
struct inet_connection_sock *icsk = inet_csk(sk);
+ struct tcp_sock *tp = tcp_sk(sk);
unsigned int cur_mss;
- int err;
+ int diff, len, err;
+
- /* Inconslusive MTU probe */
- if (icsk->icsk_mtup.probe_size) {
+ /* Inconclusive MTU probe */
+ if (icsk->icsk_mtup.probe_size)
icsk->icsk_mtup.probe_size = 0;
- }
/* Do not sent more than we queued. 1/4 is reserved for possible
* copying overhead: fragmentation, tunneling, mangling etc.
@@ -2614,30 +2614,27 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
TCP_SKB_CB(skb)->seq != tp->snd_una)
return -EAGAIN;
- if (skb->len > cur_mss) {
- if (tcp_fragment(sk, skb, cur_mss, cur_mss, GFP_ATOMIC))
+ len = cur_mss * segs;
+ if (skb->len > len) {
+ if (tcp_fragment(sk, skb, len, cur_mss, GFP_ATOMIC))
return -ENOMEM; /* We'll try again later. */
} else {
- int oldpcount = tcp_skb_pcount(skb);
+ if (skb_unclone(skb, GFP_ATOMIC))
+ return -ENOMEM;
- if (unlikely(oldpcount > 1)) {
- if (skb_unclone(skb, GFP_ATOMIC))
- return -ENOMEM;
- tcp_init_tso_segs(skb, cur_mss);
- tcp_adjust_pcount(sk, skb, oldpcount - tcp_skb_pcount(skb));
- }
+ diff = tcp_skb_pcount(skb);
+ tcp_set_skb_tso_segs(skb, cur_mss);
+ diff -= tcp_skb_pcount(skb);
+ if (diff)
+ tcp_adjust_pcount(sk, skb, diff);
+ if (skb->len < cur_mss)
+ tcp_retrans_try_collapse(sk, skb, cur_mss);
}
/* RFC3168, section 6.1.1.1. ECN fallback */
if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN)
tcp_ecn_clear_syn(sk, skb);
- tcp_retrans_try_collapse(sk, skb, cur_mss);
-
- /* Make a copy, if the first transmission SKB clone we made
- * is still in somebody's hands, else make a clone.
- */
-
/* make sure skb->data is aligned on arches that require it
* and check if ack-trimming & collapsing extended the headroom
* beyond what csum_start can cover.
@@ -2653,20 +2650,22 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
}
if (likely(!err)) {
+ segs = tcp_skb_pcount(skb);
+
TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
/* Update global TCP statistics. */
- TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
+ TCP_ADD_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS, segs);
if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
- tp->total_retrans++;
+ tp->total_retrans += segs;
}
return err;
}
-int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
+int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
{
struct tcp_sock *tp = tcp_sk(sk);
- int err = __tcp_retransmit_skb(sk, skb);
+ int err = __tcp_retransmit_skb(sk, skb, segs);
if (err == 0) {
#if FASTRETRANS_DEBUG > 0
@@ -2757,6 +2756,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
tcp_for_write_queue_from(skb, sk) {
__u8 sacked = TCP_SKB_CB(skb)->sacked;
+ int segs;
if (skb == tcp_send_head(sk))
break;
@@ -2764,14 +2764,8 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
if (!hole)
tp->retransmit_skb_hint = skb;
- /* Assume this retransmit will generate
- * only one packet for congestion window
- * calculation purposes. This works because
- * tcp_retransmit_skb() will chop up the
- * packet to be MSS sized and all the
- * packet counting works out.
- */
- if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
+ segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
+ if (segs <= 0)
return;
if (fwd_rexmitting) {
@@ -2808,7 +2802,7 @@ begin_fwd:
if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
continue;
- if (tcp_retransmit_skb(sk, skb))
+ if (tcp_retransmit_skb(sk, skb, segs))
return;
NET_INC_STATS_BH(sock_net(sk), mib_idx);
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 49bc474f8e35..373b03e78aaa 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -404,7 +404,7 @@ void tcp_retransmit_timer(struct sock *sk)
goto out;
}
tcp_enter_loss(sk);
- tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
+ tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1);
__sk_dst_reset(sk);
goto out_reset_timer;
}
@@ -436,7 +436,7 @@ void tcp_retransmit_timer(struct sock *sk)
tcp_enter_loss(sk);
- if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
+ if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1) > 0) {
/* Retransmission failed because of local congestion,
* do not backoff.
*/