netdev - [PATCH net-next 2/2] tcp: defer xmit timer reset in tcp_xmit_retransmit

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20200504182750.176486-3-edumazet@google.com>
Date:   Mon,  4 May 2020 11:27:50 -0700
From:   Eric Dumazet <edumazet@...gle.com>
To:     "David S . Miller" <davem@...emloft.net>
Cc:     netdev <netdev@...r.kernel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        Eric Dumazet <eric.dumazet@...il.com>,
        Neal Cardwell <ncardwell@...gle.com>,
        Yuchung Cheng <ycheng@...gle.com>
Subject: [PATCH net-next 2/2] tcp: defer xmit timer reset in tcp_xmit_retransmit_queue()

As hinted in prior change ("tcp: refine tcp_pacing_delay()
for very low pacing rates"), it is probably best arming
the xmit timer only when all the packets have been scheduled,
rather than when the head of rtx queue has been re-sent.

This does matter for flows having extremely low pacing rates,
since their tp->tcp_wstamp_ns could be far in the future.

Note that the regular xmit path has a stronger limit
in tcp_small_queue_check(), meaning it is less likely to
go beyond the pacing horizon.

Signed-off-by: Eric Dumazet <edumazet@...gle.com>
---
 net/ipv4/tcp_output.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 32c9db902f180aad3776b85bcdeb482443a9a5be..a50e1990a845a258d4cc6a2a989d09068ea3a973 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3112,6 +3112,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	struct sk_buff *skb, *rtx_head, *hole = NULL;
 	struct tcp_sock *tp = tcp_sk(sk);
+	bool rearm_timer = false;
 	u32 max_segs;
 	int mib_idx;
 
@@ -3134,7 +3135,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 
 		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
 		if (segs <= 0)
-			return;
+			break;
 		sacked = TCP_SKB_CB(skb)->sacked;
 		/* In case tcp_shift_skb_data() have aggregated large skbs,
 		 * we need to make sure not sending too bigs TSO packets
@@ -3159,10 +3160,10 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 			continue;
 
 		if (tcp_small_queue_check(sk, skb, 1))
-			return;
+			break;
 
 		if (tcp_retransmit_skb(sk, skb, segs))
-			return;
+			break;
 
 		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
 
@@ -3171,10 +3172,13 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 
 		if (skb == rtx_head &&
 		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
-			tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-					     inet_csk(sk)->icsk_rto,
-					     TCP_RTO_MAX);
+			rearm_timer = true;
+
 	}
+	if (rearm_timer)
+		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+				     inet_csk(sk)->icsk_rto,
+				     TCP_RTO_MAX);
 }
 
 /* We allow to exceed memory limits for FIN packets to expedite
-- 
2.26.2.526.g744177e7f7-goog