Skip to contentSkip to conversation

tcp: fix tcp_ofo_queue() to avoid including too much DUP SACK range

PATCH In review xin.guo 2 participants 3 messages Jun 17, 2025 – Jun 24, 2025

Subsystems (3)
lore
Share & tools
lore

xin.guo

author
#0 lore raw

PATCH 1/1 tcp: fix tcp_ofo_queue() to avoid including too much DUP SACK range

If the new coming segment covers more than one skbs in the ofo queue, and which seq is equal to rcv_nxt , then the sequence range that is not duplicated will be sent as DUP SACK, the detail as below, in step6, the {501,2001} range is clearly including too much DUP SACK range:
1. client.43629 > server.8080: Flags [.], seq 501:1001, ack 1325288529, win 20000, length 500: HTTP
2. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 1 {501:1001}], length 0
3. Iclient.43629 > server.8080: Flags [.], seq 1501:2001, ack 1325288529, win 20000, length 500: HTTP
4. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 2 {1501:2001} {501:1001}], length 0
5. client.43629 > server.8080: Flags [.], seq 1:2001,
ack 1325288529, win 20000, length 2000: HTTP
6. server.8080 > client.43629: Flags [.], ack 2001, win 65535, options [nop,nop,TS val 269383722 ecr 200,nop,nop,sack 1 {501:2001}], length 0

---
 net/ipv4/tcp_input.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/tcp_input.c +1 -1
index 8ec92dec321a..6194ddf46024 100644
@@ -4973,7 +4973,7 @@ static void tcp_ofo_queue(struct sock *sk)
 		if (before(TCP_SKB_CB(skb)->seq, dsack_high)) {
 			__u32 dsack = dsack_high;
 			if (before(TCP_SKB_CB(skb)->end_seq, dsack_high))
-				dsack_high = TCP_SKB_CB(skb)->end_seq;
+				dsack = TCP_SKB_CB(skb)->end_seq;
 			tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack);
 		}
 		p = rb_next(p);
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 11 lines · On Tue, 17 Jun 2025 23:37:06 +0800 xin.guo wrote:

If the new coming segment covers more than one skbs in the ofo queue, and which seq is equal to rcv_nxt , then the sequence range that is not duplicated will be sent as DUP SACK, the detail as below, in step6, the {501,2001} range is clearly including too much DUP SACK range:
1. client.43629 > server.8080: Flags [.], seq 501:1001, ack 1325288529, win 20000, length 500: HTTP
2. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 1 {501:1001}], length 0
3. Iclient.43629 > server.8080: Flags [.], seq 1501:2001, ack 1325288529, win 20000, length 500: HTTP

Quoted 8 lines

4. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 2 {1501:2001} {501:1001}], length 0
5. client.43629 > server.8080: Flags [.], seq 1:2001,
ack 1325288529, win 20000, length 2000: HTTP
6. server.8080 > client.43629: Flags [.], ack 2001, win 65535, options [nop,nop,TS val 269383722 ecr 200,nop,nop,sack 1 {501:2001}], length 0

Looks reasonable, AFAICT, tho perhaps there's some implicit benefit from reporting end of the DSACK == rcv_next..

Could you please add the info about how the packet from step 6 looks "after" this patch? So we have before / after comparison?

With that please resend and make sure you include _all_ maintainers that the get_maintainer script points out. You missed CCing Eric D.

Signature · 1 line
pw-bot: cr
author
#2 lore raw

Thanks Jakub, I will add more info and resend.

Regards
Guo Xin.

Quoted 32 lines · Jakub Kicinski
Quoted 11 lines · On Tue, 17 Jun 2025 23:37:06 +0800 xin.guo wrote:

If the new coming segment covers more than one skbs in the ofo queue, and which seq is equal to rcv_nxt , then the sequence range that is not duplicated will be sent as DUP SACK, the detail as below, in step6, the {501,2001} range is clearly including too much DUP SACK range:
1. client.43629 > server.8080: Flags [.], seq 501:1001, ack 1325288529, win 20000, length 500: HTTP
2. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 1 {501:1001}], length 0
3. Iclient.43629 > server.8080: Flags [.], seq 1501:2001, ack 1325288529, win 20000, length 500: HTTP

Quoted 8 lines

4. server.8080 > client.43629: Flags [.], ack 1, win 65535, options [nop,nop,TS val 269383721 ecr 200,nop,nop,sack 2 {1501:2001} {501:1001}], length 0
5. client.43629 > server.8080: Flags [.], seq 1:2001,
ack 1325288529, win 20000, length 2000: HTTP
6. server.8080 > client.43629: Flags [.], ack 2001, win 65535, options [nop,nop,TS val 269383722 ecr 200,nop,nop,sack 1 {501:2001}], length 0

Looks reasonable, AFAICT, tho perhaps there's some implicit benefit from reporting end of the DSACK == rcv_next..

Could you please add the info about how the packet from step 6 looks "after" this patch? So we have before / after comparison?

With that please resend and make sure you include _all_ maintainers that the get_maintainer script points out. You missed CCing Eric D. --
pw-bot: cr