The original TSQ algorithm limits the number of packets in qdisc/devices to
two packets / or ~1 ms. With this commit, two sysctl knobs are added to
allow tuning the number of packets or the ms value.
Signed-off-by: Natale Patriciello <natale.patricie...@gmail.com>
Cc: Carlo Augusto Grazia <carloaugusto.gra...@unimore.it>
Tested-by: Carlo Augusto Grazia <carloaugusto.gra...@unimore.it>
---
Documentation/networking/ip-sysctl.txt | 23 ++++++++++++++++++++++-
include/net/netns/ipv4.h | 2 ++
net/ipv4/sysctl_net_ipv4.c | 14 ++++++++++++++
net/ipv4/tcp_output.c | 5 ++++-
4 files changed, 42 insertions(+), 2 deletions(-)
diff --git a/Documentation/networking/ip-sysctl.txt
b/Documentation/networking/ip-sysctl.txt
index 3b530fe8a494..2510ef885746 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -721,9 +721,30 @@ tcp_limit_output_bytes - INTEGER
typical pfifo_fast qdiscs.
tcp_limit_output_bytes limits the number of bytes on qdisc
or device to reduce artificial RTT/cwnd and reduce bufferbloat.
- Set to -1 to disable.
+ The overall limit is given by the following (rate is in B/ms):
+ limit = min(output_bytes, max(output_pkt * mss, output_ms * rate)
+ Set to -1 to unconditionally disable TSQ, regardless of the
+ values of tcp_limit_output_ms and tcp_limit_output_pkt.
Default: 262144
+tcp_limit_output_ms - UNSIGNED INTEGER
+ Controls TCP Small Queue limit per TCP socket, under a time point
+ of view. Given a transmission rate, limit the bytes on qdisc or
+ device to a value that can be transmitted approximately in the
+ time provided in this parameter at the given rate. This limit
+ is doubled for retransmissions. The overall limit is given by
+ the following (rate is in B/ms):
+ limit = min(output_bytes, max(output_pkt * mss, output_ms * rate)
+ Default: 1
+
+tcp_limit_output_pkt - UNSIGNED INTEGER
+ Controls TCP Small Queue limit per tcp socket.
+ tcp_limit_output_pkt limits the number of packets queued in
+ qdisc/device. This limit is doubled for retransmissions.
+ The overall limit is given by the following (rate is in B/ms):
+ limit = min(output_bytes, max(output_pkt * mss, output_ms * rate)
+ Default: 2
+
tcp_challenge_ack_limit - INTEGER
Limits number of Challenge ACK sent per second, as recommended
in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index 44668c29701a..e2c06827d0bb 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -148,6 +148,8 @@ struct netns_ipv4 {
int sysctl_tcp_tso_win_divisor;
int sysctl_tcp_workaround_signed_windows;
int sysctl_tcp_limit_output_bytes;
+ unsigned int sysctl_tcp_limit_output_ms;
+ unsigned int sysctl_tcp_limit_output_pkt;
int sysctl_tcp_challenge_ack_limit;
int sysctl_tcp_min_tso_segs;
int sysctl_tcp_min_rtt_wlen;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 93e172118a94..775a4d079a9b 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -1094,6 +1094,20 @@ static struct ctl_table ipv4_net_table[] = {
.mode = 0644,
.proc_handler = proc_dointvec
},
+ {
+ .procname = "tcp_limit_output_ms",
+ .data = &init_net.ipv4.sysctl_tcp_limit_output_ms,
+ .maxlen = sizeof(unsigned int),
+ .mode = 0644,
+ .proc_handler = proc_douintvec
+ },
+ {
+ .procname = "tcp_limit_output_pkt",
+ .data = &init_net.ipv4.sysctl_tcp_limit_output_pkt,
+ .maxlen = sizeof(unsigned int),
+ .mode = 0644,
+ .proc_handler = proc_douintvec
+ },
{
.procname = "tcp_challenge_ack_limit",
.data = &init_net.ipv4.sysctl_tcp_challenge_ack_limit,
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 997a6fbdbe1a..eae715c4a005 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2201,7 +2201,10 @@ static bool tcp_small_queue_check(struct sock *sk, const
struct sk_buff *skb,
if (sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes < 0)
return false;
- limit = max(2 * skb->truesize, sk->sk_pacing_rate >> 10);
+ limit = sock_net(sk)->ipv4.sysctl_tcp_limit_output_ms *
+ (sk->sk_pacing_rate >> 10);
+ limit = max(sock_net(sk)->ipv4.sysctl_tcp_limit_output_pkt *
skb->truesize,
+ limit);
limit = min_t(u32, limit,
sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
limit <<= factor;
--
2.15.1
