bpf: Allow a TCP CC to write sk_pacing_rate and sk_pacing_status

A CC that implements tcp_congestion_ops.cong_control() should be able to
control sk_pacing_rate and set sk_pacing_status, since
tcp_update_pacing_rate() is never called in this case. A built-in CC or
one from a kernel module is already able to write to both struct sock
members. For a BPF program, write access has not been allowed, yet.

Signed-off-by: Jörn-Thorben Hinz <jthinz@mailbox.tu-berlin.de>
Reviewed-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/20220622191227.898118-2-jthinz@mailbox.tu-berlin.de
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

net/ipv4/bpf_tcp_ca.c

@@ -111,6 +111,12 @@ static int bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log *log,
 	}
 
 	switch (off) {
+	case offsetof(struct sock, sk_pacing_rate):
+		end = offsetofend(struct sock, sk_pacing_rate);
+		break;
+	case offsetof(struct sock, sk_pacing_status):
+		end = offsetofend(struct sock, sk_pacing_status);
+		break;
 	case bpf_ctx_range(struct inet_connection_sock, icsk_ca_priv):
 		end = offsetofend(struct inet_connection_sock, icsk_ca_priv);
 		break;