Patchwork [net-next,14/15] tcp: don't clear lost_skb_hint when not necessary

login
register
mail settings
Submitter Ilpo Järvinen
Date Sept. 20, 2008, 5:52 p.m.
Message ID <1221933153-11874-15-git-send-email-ilpo.jarvinen@helsinki.fi>
Download mbox | patch
Permalink /patch/732/
State Rejected
Headers show

Comments

Ilpo Järvinen - Sept. 20, 2008, 5:52 p.m.
Most importantly avoid doing it with cumulative ACK. However,
since we have lost_cnt_hint in the picture as well needing
adjustments, it's not as trivial as dealing with
retransmit_skb_hint (and cannot be done in the all place we
could trivially leave retransmit_skb_hint untouched).

With the previous patch, this should mostly remove O(n^2)
behavior while cumulative ACKs start flowing once rexmit
after a lossy round-trip made it through.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
---
 net/ipv4/tcp_input.c |   14 +++++++++++++-
 1 files changed, 13 insertions(+), 1 deletions(-)

Patch

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index f217b4f..9daeb07 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -2844,6 +2844,7 @@  static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets)
 	int flag = 0;
 	u32 pkts_acked = 0;
 	u32 reord = tp->packets_out;
+	u32 prior_sacked = tp->sacked_out;
 	s32 seq_rtt = -1;
 	s32 ca_seq_rtt = -1;
 	ktime_t last_ackt = net_invalid_timestamp();
@@ -2925,9 +2926,11 @@  static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets)
 
 		tcp_unlink_write_queue(skb, sk);
 		sk_wmem_free_skb(sk, skb);
-		tcp_clear_retrans_hints_partial(tp);
+		tp->scoreboard_skb_hint = NULL;
 		if (skb == tp->retransmit_skb_hint)
 			tp->retransmit_skb_hint = NULL;
+		if (skb == tp->lost_skb_hint)
+			tp->lost_skb_hint = NULL;
 	}
 
 	if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
@@ -2946,6 +2949,15 @@  static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets)
 			/* Non-retransmitted hole got filled? That's reordering */
 			if (reord < prior_fackets)
 				tcp_update_reordering(sk, tp->fackets_out - reord, 0);
+
+			/* No need to care for underflows here because
+			 * the lost_skb_hint gets NULLed if we're past it
+			 * (or something non-trivial happened)
+			 */
+			if (tcp_is_fack(tp))
+				tp->lost_cnt_hint -= pkts_acked;
+			else
+				tp->lost_cnt_hint -= prior_sacked - tp->sacked_out;
 		}
 
 		tp->fackets_out -= min(pkts_acked, tp->fackets_out);