From 255cac91c3c9ce7dca7713b93ab03c75b7902e0e Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ilpo=20J=C3=A4rvinen?= Date: Mon, 4 May 2009 11:07:36 -0700 Subject: tcp: extend ECN sysctl to allow server-side only ECN MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This should be very safe compared with full enabled, so I see no reason why it shouldn't be done right away. As ECN can only be negotiated if the SYN sending party is also supporting it, somebody in the loop probably knows what he/she is doing. If SYN does not ask for ECN, the server side SYN-ACK is identical to what it is without ECN. Thus it's quite safe. The chosen value is safe w.r.t to existing configs which choose to currently set manually either 0 or 1 but silently upgrades those who have not explicitly requested ECN off. Whether to just enable both sides comes up time to time but unless that gets done now we can at least make the servers aware of ECN already. As there are some known problems to occur if ECN is enabled, it's currently questionable whether there's any real gain from enabling clients as servers mostly won't support it anyway (so we'd hit just the negative sides). After enabling the servers and getting that deployed, the client end enable really has some potential gain too. Signed-off-by: Ilpo Järvinen Signed-off-by: David S. Miller --- Documentation/networking/ip-sysctl.txt | 11 ++++++++++- net/ipv4/tcp_input.c | 2 +- net/ipv4/tcp_output.c | 2 +- 3 files changed, 12 insertions(+), 3 deletions(-) diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt index ec5de02f543..7f98aa3962c 100644 --- a/Documentation/networking/ip-sysctl.txt +++ b/Documentation/networking/ip-sysctl.txt @@ -168,7 +168,16 @@ tcp_dsack - BOOLEAN Allows TCP to send "duplicate" SACKs. tcp_ecn - BOOLEAN - Enable Explicit Congestion Notification in TCP. + Enable Explicit Congestion Notification (ECN) in TCP. ECN is only + used when both ends of the TCP flow support it. It is useful to + avoid losses due to congestion (when the bottleneck router supports + ECN). + Possible values are: + 0 disable ECN + 1 ECN enabled + 2 Only server-side ECN enabled. If the other end does + not support ECN, behavior is like with ECN disabled. + Default: 2 tcp_fack - BOOLEAN Enable FACK congestion avoidance and fast retransmission. diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index c96a6bb2543..56dcf97a97f 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -77,7 +77,7 @@ int sysctl_tcp_window_scaling __read_mostly = 1; int sysctl_tcp_sack __read_mostly = 1; int sysctl_tcp_fack __read_mostly = 1; int sysctl_tcp_reordering __read_mostly = TCP_FASTRETRANS_THRESH; -int sysctl_tcp_ecn __read_mostly; +int sysctl_tcp_ecn __read_mostly = 2; int sysctl_tcp_dsack __read_mostly = 1; int sysctl_tcp_app_win __read_mostly = 31; int sysctl_tcp_adv_win_scale __read_mostly = 2; diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 59aec609cec..79c39dc9b01 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -288,7 +288,7 @@ static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb) struct tcp_sock *tp = tcp_sk(sk); tp->ecn_flags = 0; - if (sysctl_tcp_ecn) { + if (sysctl_tcp_ecn == 1) { TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE | TCPCB_FLAG_CWR; tp->ecn_flags = TCP_ECN_OK; } -- cgit v1.2.3