tcp: Fix data-races around sk_pacing_rate.

While reading sysctl_tcp_pacing_(ss|ca)_ratio, they can be changed
concurrently.  Thus, we need to add READ_ONCE() to their readers.

Fixes: 43e122b014c9 ("tcp: refine pacing rate determination")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.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 c799f39cb774d9961dc90ec0c5d48ea790cd4cb0..dd05238f79f69c5ea9f0e96c4e1af8138fbeb9b8 100644 (file)
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -910,9 +910,9 @@ static void tcp_update_pacing_rate(struct sock *sk)
         *       end of slow start and should slow down.
         */
        if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2)
-               rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio;
+               rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio);
        else
-               rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio;
+               rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio);
 
        rate *= max(tcp_snd_cwnd(tp), tp->packets_out);