tcp: Namespace-ify sysctl_tcp_pacing_ss_ratio
Also remove an obsolete comment about TCP pacing.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index db4d458..29539d3 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -767,13 +767,6 @@ static void tcp_rtt_estimator(struct sock *sk, long mrtt_us)
tp->srtt_us = max(1U, srtt);
}
-/* Set the sk_pacing_rate to allow proper sizing of TSO packets.
- * Note: TCP stack does not yet implement pacing.
- * FQ packet scheduler can be used to implement cheap but effective
- * TCP pacing, to smooth the burst on large writes when packets
- * in flight is significantly lower than cwnd (or rwin)
- */
-int sysctl_tcp_pacing_ss_ratio __read_mostly = 200;
int sysctl_tcp_pacing_ca_ratio __read_mostly = 120;
static void tcp_update_pacing_rate(struct sock *sk)
@@ -793,7 +786,7 @@ static void tcp_update_pacing_rate(struct sock *sk)
* end of slow start and should slow down.
*/
if (tp->snd_cwnd < tp->snd_ssthresh / 2)
- rate *= sysctl_tcp_pacing_ss_ratio;
+ rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio;
else
rate *= sysctl_tcp_pacing_ca_ratio;
