On Tue, Aug 23, 2011 at 11:55 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:

Le mardi 23 août 2011 à 23:31 +0300, Daniel Baluta a écrit :

...

Hello,

Please help me understanding the behavior of the following TCP conversation.

You can find bellow a snippet of the (FTP) conversation captured both on client (C) and server (S).

[client]$ tcpdump  -n  -r client-6-conv.cap [P1] 49.045690 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [P.], seq 197:220, ack 81, win 757, length 23 [P2] 49.046600 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 220, win 738, length 0 [P3] 49.047462 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [P.], seq 81:87, ack 220, win 738, length 6 [P5] 49.048757 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [F.], seq 242, ack 87, win 757, length 0 [P6] 49.048794 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 220, win 738, options [nop,nop,sack 1 {242:243}], length 0 [P4] 49.048801 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [P.], seq 220:242, ack 87, win 757, length 22 [P7] 49.048833 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 243, win 715, length 0 [P8] 49.049566 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [F.], seq 87, ack 243, win 715, length 0 [P9] 49.050889 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [R], seq 1672731590, win 0, length 0

[server]$ tcpdump  -n  -r server-6-conv.cap [P1] 49.059740 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [P.], seq 197:220, ack 81, win 757, length 23 [P2] 49.061394 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 220, win 738, length 0 [P3] 49.061760 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [P.], seq 81:87, ack 220, win 738, length 6 [P4] 49.062794 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [P.], seq 220:242, ack 87, win 757, length 22 [P5] 49.062843 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [F.], seq 242, ack 87, win 757, length 0 [P6] 49.063808 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 220, win 738, options [nop,nop,sack 1 {242:243}], length 0 [P7] 49.063823 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [.], ack 243, win 715, length 0 [P8] 49.064271 IP 10.10.0.20.58277 > 10.10.0.1.21: Flags [F.], seq 87, ack 243, win 715, length 0 [P9] 49.064481 IP 10.10.0.1.21 > 10.10.0.20.58277: Flags [R], seq 1672731590, win 0, length 0

What happens is that servers sends packets P4 and P5, but client receives P5 before P4. Since SACK is enabled, client will send a SACK (P6) ack-ing P5.

Then client sees P4, and will send an ack (P7) for P4, then sends P8 with FIN flag set. What I don't understand, is why server responds with RST (P9) instead of ACK?

This is was obtained on 2.6.32.43. I have also attached full capture files.

I am reading TCP's RFC and kernel code, but so far I haven't reached a conclusion.

TCP in RFC 1122 section 4.2.2.13:

 "A host MAY implement a "half-duplex" TCP close sequence, so that an  application that has called CLOSE cannot continue to read data from the  connection. If such a host issues a CLOSE call while received data is  still pending in TCP, or if new data is received after CLOSE is called,  its TCP SHOULD send a RST to show that data was lost."

So, this means that server's CLOSE operation is issued while received data is still pending? I will analyze ftp's server code, but this is strange since P4 [221 Have a nice day!\r\n] it is generated as a response for P3 [QUIT\r\n]. So P4 must have been fully received. Also, looking at the capture no data is received from the client after server calls CLOSE (P5) (there are only ACKs and FIN - P6, P7, P8).

This is why some apps first call shutdown(), then drain receive queue, then close()

This makes sense :). thanks Eric. Daniel.