patch to implement RFC3517 in linux 2.4.22

From: Douglas Leith (doug_at_eee.strath.ac.uk)
Date: 10/09/03

  • Next message: Russell King: "Re: [RFC] disable_irq()/enable_irq() semantics and ide-probe.c" 
    Date:	Thu, 9 Oct 2003 07:27:15 -0000
To: <linux-net@vger.kernel.org>

    Attached is a short patch against version 2.4.22 to implement the recent
    rfc3517 for SACK in TCP (see ftp://ftp.rfc-editor.org/in-notes/rfc3517.txt).
    This modifies how packets which fail to be sacked are marked as lost. At
    present, any packets falling into sack holes are marked lost after the first
    successful retransmit. The new code marks packets in sack holes as lost as
    soon as tp->reordering or more packets with higher sequence number are
    sacked. This is consistent with how reordering is handled elsewhere and
    should lead to faster retransmission and recovery when multiple drops occur.

    Comments appreciated.

    Regards,

    Doug Leith

    www.hamilton.ie

    diff -u -wbrP --exclude=config.save linux-2.4.22/net/ipv4/tcp_input.c linux-
    2.4.22-retrans/net/ipv4/tcp_input.c
    --- linux-2.4.22/net/ipv4/tcp_input.c 2003-06-13 15:51:39.000000000 +0100
    +++ linux-2.4.22-retrans/net/ipv4/tcp_input.c 2003-10-09 07:59:48.000000000
    +0100
    @@ -61,6 +61,7 @@
      * Panu Kuhlberg: Experimental audit of TCP (re)
    transmission
      * engine. Lots of bugs are found.
      * Pasi Sarolahti: F-RTO for dealing with spurious RTOs
    + * Doug Leith: Early retransmission of packets
    falling into sack holes.
      */

     #include <linux/config.h>
    @@ -757,6 +758,10 @@
      * for retransmitted and already SACKed segment -> reordering..
      * Both of these heuristics are not used in Loss state, when we cannot
      * account for retransmits accurately.
    + *
    + * Retransmission of packets
    + * -------------------------
    + * Extension of Hoe's retransmit to allow early retransmission of packets
    which fall into sack holes
      */
     static int
     tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32
    prior_snd_una)
    @@ -770,6 +775,7 @@
            u32 lost_retrans = 0;
            int flag = 0;
            int i;
    + u32 prior_sacked_out=tp->sacked_out;

            if (!tp->sacked_out)
                    tp->fackets_out = 0;
    @@ -781,6 +787,7 @@
                    __u32 end_seq = ntohl(sp->end_seq);
                    int fack_count = 0;
                    int dup_sack = 0;
    + u32 sack_count=0;

                    /* Check for D-SACK. */
                    if (i == 0) {
    @@ -828,6 +835,8 @@
                                    break;

                            fack_count++;
    + if (sacked&TCPCB_RETRANS)
    + sack_count++;

                            in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
                                    !before(end_seq, TCP_SKB_CB(skb)->end_seq);
    @@ -860,8 +869,14 @@
                                (!lost_retrans || after(end_seq, lost_retrans)))
                                    lost_retrans = end_seq;

    - if (!in_sack)
    + if (!in_sack) {
    + /* force early retransmit ? */
    + if (!(sacked&TCPCB_TAGBITS) && (prior_sacked_out
    > sack_count + tp->reordering)) {
    + TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
    + tp->lost_out++;
    + }
                                    continue;
    + }

                            if (!(sacked&TCPCB_SACKED_ACKED)) {
                                    if (sacked & TCPCB_SACKED_RETRANS) {
    @@ -1599,7 +1614,7 @@
            if ((flag&FLAG_DATA_LOST) &&
                before(tp->snd_una, tp->high_seq) &&
                tp->ca_state != TCP_CA_Open &&
    - tp->fackets_out > tp->reordering) {
    + tp->fackets_out > tp->reordering && IsReno(tp) ) {
                    tcp_mark_head_lost(sk, tp, tp->fackets_out-tp->reordering, tp-
    >high_seq);
                    NET_INC_STATS_BH(TCPLoss);
            }
    @@ -1714,7 +1729,7 @@
                    tp->ca_state = TCP_CA_Recovery;
            }

    - if (is_dupack || tcp_head_timedout(sk, tp))
    + if ( (is_dupack || tcp_head_timedout(sk, tp)) && IsReno(tp) )
                    tcp_update_scoreboard(sk, tp);
            tcp_cwnd_down(tp);
            tcp_xmit_retransmit_queue(sk);

    diff -u -wbrP --exclude=config.save linux-2.4.22/net/ipv4/tcp_input.c linux-2.4.22-retrans/net/ipv4/tcp_input.c
    --- linux-2.4.22/net/ipv4/tcp_input.c 2003-06-13 15:51:39.000000000 +0100
    +++ linux-2.4.22-retrans/net/ipv4/tcp_input.c 2003-10-09 07:59:48.000000000 +0100
    @@ -61,6 +61,7 @@
      * Panu Kuhlberg: Experimental audit of TCP (re)transmission
      * engine. Lots of bugs are found.
      * Pasi Sarolahti: F-RTO for dealing with spurious RTOs
    + * Doug Leith: Early retransmission of packets falling into sack holes.
      */
     
     #include <linux/config.h>
    @@ -757,6 +758,10 @@
      * for retransmitted and already SACKed segment -> reordering..
      * Both of these heuristics are not used in Loss state, when we cannot
      * account for retransmits accurately.
    + *
    + * Retransmission of packets
    + * -------------------------
    + * Extension of Hoe's retransmit to allow early retransmission of packets which fall into sack holes
      */
     static int
     tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_una)
    @@ -770,6 +775,7 @@
             u32 lost_retrans = 0;
             int flag = 0;
             int i;
    + u32 prior_sacked_out=tp->sacked_out;
     
             if (!tp->sacked_out)
                     tp->fackets_out = 0;
    @@ -781,6 +787,7 @@
                     __u32 end_seq = ntohl(sp->end_seq);
                     int fack_count = 0;
                     int dup_sack = 0;
    + u32 sack_count=0;
     
                     /* Check for D-SACK. */
                     if (i == 0) {
    @@ -828,6 +835,8 @@
                                     break;
     
                             fack_count++;
    + if (sacked&TCPCB_RETRANS)
    + sack_count++;
     
                             in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
                                     !before(end_seq, TCP_SKB_CB(skb)->end_seq);
    @@ -860,8 +869,14 @@
                                 (!lost_retrans || after(end_seq, lost_retrans)))
                                     lost_retrans = end_seq;
     
    - if (!in_sack)
    + if (!in_sack) {
    + /* force early retransmit ? */
    + if (!(sacked&TCPCB_TAGBITS) && (prior_sacked_out > sack_count + tp->reordering)) {
    + TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
    + tp->lost_out++;
    + }
                                     continue;
    + }
     
                             if (!(sacked&TCPCB_SACKED_ACKED)) {
                                     if (sacked & TCPCB_SACKED_RETRANS) {
    @@ -1599,7 +1614,7 @@
             if ((flag&FLAG_DATA_LOST) &&
                 before(tp->snd_una, tp->high_seq) &&
                 tp->ca_state != TCP_CA_Open &&
    - tp->fackets_out > tp->reordering) {
    + tp->fackets_out > tp->reordering && IsReno(tp) ) {
                     tcp_mark_head_lost(sk, tp, tp->fackets_out-tp->reordering, tp->high_seq);
                     NET_INC_STATS_BH(TCPLoss);
             }
    @@ -1714,7 +1729,7 @@
                     tp->ca_state = TCP_CA_Recovery;
             }
     
    - if (is_dupack || tcp_head_timedout(sk, tp))
    + if ( (is_dupack || tcp_head_timedout(sk, tp)) && IsReno(tp) )
                     tcp_update_scoreboard(sk, tp);
             tcp_cwnd_down(tp);
             tcp_xmit_retransmit_queue(sk);
    NrybXǧv^)޺{.n+{zXܨ}Ơz&j:+vzZ+2ޙ&)ߡaGhj:+vw٥>WiaxPjm -+d

  • Next message: Russell King: "Re: [RFC] disable_irq()/enable_irq() semantics and ide-probe.c"

    Relevant Pages

    • Re: TCP keepalive timer problem
      ... the normal retransmission takes place and no further keepalive probe will be sent. ... We tried TCP implementation on Windows XP SP3, the keepalive and retransmission don't intervene. ... I found one problem in Linux TCP keepalive timer processing, ... Keep-alive packets MUST only be sent when no data or acknowledgement packets have ...
      (Linux-Kernel)
    • Re: Long Range Wireless Network
      ... packet acknowledgement or retransmission. ... The automatic-retransmission timeouts in the cards are set to a fairly ... If the transmitting card's timeout value is set to less ... received packets will be forwarded to their destination, ...
      (rec.radio.amateur.homebrew)
    • Re: Atheros, hardware access layer, collisions
      ... > I think there is still collision detection happening on the hardware ... I think I have to disable the retransmission of frames ... > hosts are sending packets to the middle guy at the same time. ... you can stop the hardware doing it. ...
      (freebsd-hackers)
    • Re: Multicast log question
      ... c-quality value drops to 60 percent, and it doesn't go up again for the ... My guess is that when the client lost 13 packets that weren't recoverable ... >> Based on past testing we've found that WMP can only recover one lost packet ...
      (microsoft.public.windowsmedia.server)
    • Re: [patch] Real-Time Preemption, -RT-2.6.10-rc1-mm2-V0.7.1
      ... There was a burst of lost packets between seq 1777 and 1959, ... The display lockup on the top test was a little odd. ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)