Patchwork RFC2988bis + taking RTT sample from 3WHS for the passive open side

login
register
mail settings
Submitter Jerry Chu
Date June 8, 2011, 6:04 p.m.
Message ID <BANLkTimXgK03tM1e3FYe+wausHT15W6H-2eMNCWaj8LLxRvbZw@mail.gmail.com>
Download mbox | patch
Permalink /patch/99523/
State Superseded
Delegated to: David Miller
Headers show

Comments

Jerry Chu - June 8, 2011, 6:04 p.m.
[resent to cc netdev]

This patch lowers the default initRTO from 3secs to 1sec per
RFC2988bis. It falls back to 3secs if the SYN or SYN-ACK packet
has been retransmitted, AND the TCP timestamp option is not on.

It also adds support to take RTT sample during 3WHS on the passive
open side, just like its active open counterpart, and uses it, if
valid, to seed the initRTO for the data transmission phase.

The patch also resets ssthresh to its initial default at the
beginning of the data transmission phase, and reduces cwnd to 1 if
there has been MORE THAN ONE retransmission during 3WHS per RFC5681.

Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
---
 include/linux/tcp.h      |    1 +
 include/net/tcp.h        |   11 +++++++++--
 net/ipv4/syncookies.c    |    1 +
 net/ipv4/tcp_input.c     |   46 +++++++++++++++++++++++++---------------------
 net/ipv4/tcp_ipv4.c      |   11 ++++++++---
 net/ipv4/tcp_minisocks.c |    6 +++++-
 net/ipv6/syncookies.c    |    1 +
 net/ipv6/tcp_ipv6.c      |    5 +++++
 8 files changed, 55 insertions(+), 27 deletions(-)


@@ -1509,6 +1510,10 @@ static struct sock *
tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
       tcp_sync_mss(newsk, dst_mtu(dst));
       newtp->advmss = dst_metric_advmss(dst);
       tcp_initialize_rcv_mss(newsk);
+       if (tcp_rsk(req)->snt_synack)
+               tcp_valid_rtt_meas(newsk,
+                   tcp_time_stamp - tcp_rsk(req)->snt_synack);
+       newtp->total_retrans = req->retrans;

       newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
       newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
--
1.7.3.1
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Yuchung Cheng - June 8, 2011, 8:54 p.m.
Acked-by: Yuchung Cheng <ycheng@google.com>

On Wed, Jun 8, 2011 at 11:04 AM, Jerry Chu <hkchu@google.com> wrote:
>
> [resent to cc netdev]
>
> This patch lowers the default initRTO from 3secs to 1sec per
> RFC2988bis. It falls back to 3secs if the SYN or SYN-ACK packet
> has been retransmitted, AND the TCP timestamp option is not on.
>
> It also adds support to take RTT sample during 3WHS on the passive
> open side, just like its active open counterpart, and uses it, if
> valid, to seed the initRTO for the data transmission phase.
>
> The patch also resets ssthresh to its initial default at the
> beginning of the data transmission phase, and reduces cwnd to 1 if
> there has been MORE THAN ONE retransmission during 3WHS per RFC5681.
>
> Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
> ---
>  include/linux/tcp.h      |    1 +
>  include/net/tcp.h        |   11 +++++++++--
>  net/ipv4/syncookies.c    |    1 +
>  net/ipv4/tcp_input.c     |   46 +++++++++++++++++++++++++---------------------
>  net/ipv4/tcp_ipv4.c      |   11 ++++++++---
>  net/ipv4/tcp_minisocks.c |    6 +++++-
>  net/ipv6/syncookies.c    |    1 +
>  net/ipv6/tcp_ipv6.c      |    5 +++++
>  8 files changed, 55 insertions(+), 27 deletions(-)
>
> diff --git a/include/linux/tcp.h b/include/linux/tcp.h
> index e64f4c6..531ede8 100644
> --- a/include/linux/tcp.h
> +++ b/include/linux/tcp.h
> @@ -282,6 +282,7 @@ struct tcp_request_sock {
>  #endif
>        u32                             rcv_isn;
>        u32                             snt_isn;
> +       u32                             snt_synack; /* synack sent time */
>  };
>
>  static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index cda30ea..149a415 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -122,7 +122,13 @@ extern void tcp_time_wait(struct sock *sk, int
> state, int timeo);
>  #endif
>  #define TCP_RTO_MAX    ((unsigned)(120*HZ))
>  #define TCP_RTO_MIN    ((unsigned)(HZ/5))
> -#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ))    /* RFC 1122 initial
> RTO value   */
> +#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))    /* RFC2988bis initial
> RTO value */
> +#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))        /* RFC 1122
> initial RTO value, now
> +                                                * used as a fallback
> RTO for the
> +                                                * initial data
> transmission if no
> +                                                * valid RTT sample
> has been acquired,
> +                                                * most likely due to
> retrans in 3WHS.
> +                                                */
>
>  #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal
> interval between probes
>                                                         * for local resources.
> @@ -295,7 +301,7 @@ static inline void tcp_synq_overflow(struct sock *sk)
>  static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
>  {
>        unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
> -       return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
> +       return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
>  }
>
>  extern struct proto tcp_prot;
> @@ -508,6 +514,7 @@ extern void tcp_initialize_rcv_mss(struct sock *sk);
>  extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
>  extern int tcp_mss_to_mtu(struct sock *sk, int mss);
>  extern void tcp_mtup_init(struct sock *sk);
> +extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
>
>  static inline void tcp_bound_rto(const struct sock *sk)
>  {
> diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
> index 2646149..92bb943 100644
> --- a/net/ipv4/syncookies.c
> +++ b/net/ipv4/syncookies.c
> @@ -316,6 +316,7 @@ struct sock *cookie_v4_check(struct sock *sk,
> struct sk_buff *skb,
>        ireq->wscale_ok         = tcp_opt.wscale_ok;
>        ireq->tstamp_ok         = tcp_opt.saw_tstamp;
>        req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
> +       treq->snt_synack        = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
>
>        /* We throwed the options of the initial SYN away, so we hope
>         * the ACK carries the same options again (see RFC1122 4.2.3.8)
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index bef9f04..ea0d218 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -880,6 +880,11 @@ static void tcp_init_metrics(struct sock *sk)
>                tp->snd_ssthresh = dst_metric(dst, RTAX_SSTHRESH);
>                if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
>                        tp->snd_ssthresh = tp->snd_cwnd_clamp;
> +       } else {
> +               /* ssthresh may have been reduced unnecessarily during.
> +                * 3WHS. Restore it back to its initial default.
> +                */
> +               tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
>        }
>        if (dst_metric(dst, RTAX_REORDERING) &&
>            tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
> @@ -887,10 +892,7 @@ static void tcp_init_metrics(struct sock *sk)
>                tp->reordering = dst_metric(dst, RTAX_REORDERING);
>        }
>
> -       if (dst_metric(dst, RTAX_RTT) == 0)
> -               goto reset;
> -
> -       if (!tp->srtt && dst_metric_rtt(dst, RTAX_RTT) <
> (TCP_TIMEOUT_INIT << 3))
> +       if (dst_metric(dst, RTAX_RTT) == 0 || tp->srtt == 0)
>                goto reset;
>
>        /* Initial rtt is determined from SYN,SYN-ACK.
> @@ -916,19 +918,26 @@ static void tcp_init_metrics(struct sock *sk)
>                tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
>        }
>        tcp_set_rto(sk);
> -       if (inet_csk(sk)->icsk_rto < TCP_TIMEOUT_INIT &&
> !tp->rx_opt.saw_tstamp) {
>  reset:
> -               /* Play conservative. If timestamps are not
> -                * supported, TCP will fail to recalculate correct
> -                * rtt, if initial rto is too small. FORGET ALL AND RESET!
> +       if (tp->srtt == 0) {
> +               /* RFC2988bis: We've failed to get a valid RTT sample from
> +                * 3WHS. This is most likely due to retransmission,
> +                * including spurious one. Reset the RTO back to 3secs
> +                * from the more aggressive 1sec to avoid more spurious
> +                * retransmission.
>                 */
> -               if (!tp->rx_opt.saw_tstamp && tp->srtt) {
> -                       tp->srtt = 0;
> -                       tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_INIT;
> -                       inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
> -               }
> +               tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_FALLBACK;
> +               inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
>        }
> -       tp->snd_cwnd = tcp_init_cwnd(tp, dst);
> +       /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
> +        * retransmitted. In light of RFC2988bis' more aggressive 1sec
> +        * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
> +        * retransmission has occurred.
> +        */
> +       if (tp->total_retrans > 1)
> +               tp->snd_cwnd = 1;
> +       else
> +               tp->snd_cwnd = tcp_init_cwnd(tp, dst);
>        tp->snd_cwnd_stamp = tcp_time_stamp;
>  }
>
> @@ -3112,12 +3121,13 @@ static void tcp_fastretrans_alert(struct sock
> *sk, int pkts_acked, int flag)
>        tcp_xmit_retransmit_queue(sk);
>  }
>
> -static void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
> +void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
>  {
>        tcp_rtt_estimator(sk, seq_rtt);
>        tcp_set_rto(sk);
>        inet_csk(sk)->icsk_backoff = 0;
>  }
> +EXPORT_SYMBOL(tcp_valid_rtt_meas);
>
>  /* Read draft-ietf-tcplw-high-performance before mucking
>  * with this code. (Supersedes RFC1323)
> @@ -5806,12 +5816,6 @@ int tcp_rcv_state_process(struct sock *sk,
> struct sk_buff *skb,
>                                              tp->rx_opt.snd_wscale;
>                                tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);
>
> -                               /* tcp_ack considers this ACK as duplicate
> -                                * and does not calculate rtt.
> -                                * Force it here.
> -                                */
> -                               tcp_ack_update_rtt(sk, 0, 0);
> -
>                                if (tp->rx_opt.tstamp_ok)
>                                        tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
>
> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> index 3c8d9b6..5fb504b 100644
> --- a/net/ipv4/tcp_ipv4.c
> +++ b/net/ipv4/tcp_ipv4.c
> @@ -429,8 +429,8 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
>                        break;
>
>                icsk->icsk_backoff--;
> -               inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
> -                                        icsk->icsk_backoff;
> +               inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
> +                       TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
>                tcp_bound_rto(sk);
>
>                skb = tcp_write_queue_head(sk);
> @@ -1384,6 +1384,7 @@ int tcp_v4_conn_request(struct sock *sk, struct
> sk_buff *skb)
>                isn = tcp_v4_init_sequence(skb);
>        }
>        tcp_rsk(req)->snt_isn = isn;
> +       tcp_rsk(req)->snt_synack = tcp_time_stamp;
>
>        if (tcp_v4_send_synack(sk, dst, req,
>                               (struct request_values *)&tmp_ext) ||
> @@ -1458,6 +1459,10 @@ struct sock *tcp_v4_syn_recv_sock(struct sock
> *sk, struct sk_buff *skb,
>                newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
>
>        tcp_initialize_rcv_mss(newsk);
> +       if (tcp_rsk(req)->snt_synack)
> +               tcp_valid_rtt_meas(newsk,
> +                   tcp_time_stamp - tcp_rsk(req)->snt_synack);
> +       newtp->total_retrans = req->retrans;
>
>  #ifdef CONFIG_TCP_MD5SIG
>        /* Copy over the MD5 key from the original socket */
> @@ -1854,7 +1859,7 @@ static int tcp_v4_init_sock(struct sock *sk)
>         * algorithms that we must have the following bandaid to talk
>         * efficiently to them.  -DaveM
>         */
> -       tp->snd_cwnd = 2;
> +       tp->snd_cwnd = TCP_INIT_CWND;
>
>        /* See draft-stevens-tcpca-spec-01 for discussion of the
>         * initialization of these values.
> diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
> index 80b1f80..d2fe4e0 100644
> --- a/net/ipv4/tcp_minisocks.c
> +++ b/net/ipv4/tcp_minisocks.c
> @@ -486,7 +486,7 @@ struct sock *tcp_create_openreq_child(struct sock
> *sk, struct request_sock *req,
>                 * algorithms that we must have the following bandaid to talk
>                 * efficiently to them.  -DaveM
>                 */
> -               newtp->snd_cwnd = 2;
> +               newtp->snd_cwnd = TCP_INIT_CWND;
>                newtp->snd_cwnd_cnt = 0;
>                newtp->bytes_acked = 0;
>
> @@ -720,6 +720,10 @@ struct sock *tcp_check_req(struct sock *sk,
> struct sk_buff *skb,
>                NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
>                return NULL;
>        }
> +       if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
> +               tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
> +       else if (req->retrans) /* don't take RTT sample if retrans && ~TS */
> +               tcp_rsk(req)->snt_synack = 0;
>
>        /* OK, ACK is valid, create big socket and
>         * feed this segment to it. It will repeat all
> diff --git a/net/ipv6/syncookies.c b/net/ipv6/syncookies.c
> index 8b9644a..89d5bf8 100644
> --- a/net/ipv6/syncookies.c
> +++ b/net/ipv6/syncookies.c
> @@ -223,6 +223,7 @@ struct sock *cookie_v6_check(struct sock *sk,
> struct sk_buff *skb)
>        ireq->wscale_ok         = tcp_opt.wscale_ok;
>        ireq->tstamp_ok         = tcp_opt.saw_tstamp;
>        req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
> +       treq->snt_synack        = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
>        treq->rcv_isn = ntohl(th->seq) - 1;
>        treq->snt_isn = cookie;
>
> diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> index 8683664..e7d47e4 100644
> --- a/net/ipv6/tcp_ipv6.c
> +++ b/net/ipv6/tcp_ipv6.c
> @@ -1341,6 +1341,7 @@ static int tcp_v6_conn_request(struct sock *sk,
> struct sk_buff *skb)
>        }
>  have_isn:
>        tcp_rsk(req)->snt_isn = isn;
> +       tcp_rsk(req)->snt_synack = tcp_time_stamp;
>
>        security_inet_conn_request(sk, skb, req);
>
> @@ -1509,6 +1510,10 @@ static struct sock *
> tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
>        tcp_sync_mss(newsk, dst_mtu(dst));
>        newtp->advmss = dst_metric_advmss(dst);
>        tcp_initialize_rcv_mss(newsk);
> +       if (tcp_rsk(req)->snt_synack)
> +               tcp_valid_rtt_meas(newsk,
> +                   tcp_time_stamp - tcp_rsk(req)->snt_synack);
> +       newtp->total_retrans = req->retrans;
>
>        newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
>        newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
> --
> 1.7.3.1
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Eric Dumazet - June 8, 2011, 9:03 p.m.
Le mercredi 08 juin 2011 à 11:04 -0700, Jerry Chu a écrit :
> [resent to cc netdev]
> 

Well, the first copy I received this morning was OK ;)

But this second one is mangled by your mailer or something ?

> This patch lowers the default initRTO from 3secs to 1sec per
> RFC2988bis. It falls back to 3secs if the SYN or SYN-ACK packet
> has been retransmitted, AND the TCP timestamp option is not on.
> 
> It also adds support to take RTT sample during 3WHS on the passive
> open side, just like its active open counterpart, and uses it, if
> valid, to seed the initRTO for the data transmission phase.
> 
> The patch also resets ssthresh to its initial default at the
> beginning of the data transmission phase, and reduces cwnd to 1 if
> there has been MORE THAN ONE retransmission during 3WHS per RFC5681.
> 
> Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
> ---



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Hagen Paul Pfeifer - June 8, 2011, 9:16 p.m.
Hello Jerry

* Jerry Chu | 2011-06-08 11:04:22 [-0700]:

>+       /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
>+        * retransmitted. In light of RFC2988bis' more aggressive 1sec
>+        * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
>+        * retransmission has occurred.
>+        */
>+       if (tp->total_retrans > 1)
>+               tp->snd_cwnd = 1;
>+       else
>+               tp->snd_cwnd = tcp_init_cwnd(tp, dst);

Is this behavior somewhere documented? In RFC2988bis I cannot find something
similar.


Cheers, Hagen
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Jerry Chu - June 8, 2011, 10:06 p.m.
Hi Hagen,

On Wed, Jun 8, 2011 at 2:16 PM, Hagen Paul Pfeifer <hagen@jauu.net> wrote:
> Hello Jerry
>
> * Jerry Chu | 2011-06-08 11:04:22 [-0700]:
>
>>+       /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
>>+        * retransmitted. In light of RFC2988bis' more aggressive 1sec
>>+        * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
>>+        * retransmission has occurred.
>>+        */
>>+       if (tp->total_retrans > 1)
>>+               tp->snd_cwnd = 1;
>>+       else
>>+               tp->snd_cwnd = tcp_init_cwnd(tp, dst);
>
> Is this behavior somewhere documented? In RFC2988bis I cannot find something
> similar.

See my 5/20/11 mail on this subject. The short answer is, no, it's not
part of RFC2988bis. The long answer is, when we (Vern, Mark and I)
were working
on RFC2988bis we shared a common design goal to minimize the negative
impact to those paths with RTT > 1sec as much as possible.

RFC3390 and later RFC5681 have the following clause:
"... Further, if the SYN or SYN/ACK is lost, the initial window used by a
sender after a correctly transmitted SYN MUST be one segment
consisting of at most SMSS bytes."

Unfortunately, with initRTO reduced to 1sec, unless the stack can reliably
detect spurious retransmissions of SYN or SYN-ACK, the paths with RTT >
1sec may be penalized unnecessarily with initcwnd = 1 if the stack
implementation follows the above clause faithfully. We can't really amend
the clause as part of RFC2988bis because it doesn't belong there, as the
latter only covers RTO computation, not the congestion control algorithm.
In the end Vern/Mark suggested me to add the amendment to the IW10
proposal I was working on (now becomes draft-ietf-tcpm-initcwnd-01.txt)
So I did, with following paragraph:
"...
   Furthermore, RFC 3390 and RFC 5681 [RFC5681] state that

         "If the SYN or SYN/ACK is lost, the initial window used by a
         sender after a correctly transmitted SYN MUST be one segment
         consisting of MSS bytes."

   The proposed change to reduce the default RTO to 1 second [PACS11]
   increases the chance for spurious SYN or SYN/ACK retransmission, thus
   unnecessarily penalizing connections with RTT > 1 second if their
   initial window is reduced to 1 segment. For this reason, it is
   RECOMMENDED that implementations refrain from resetting the initial
   window to 1 segment, unless either there have been multiple SYN or
   SYN/ACK retransmissions, or true loss detection has been made."

The patch follows the intent of the co-authors. It also moves the
implementation closer to the spec. Without the patch the existing
implementation will reduce ssthresh (active open side only), but not the
cwnd upon SYN or SYN-ACK retransmission.

Best,

Jerry

>
>
> Cheers, Hagen
>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Miller - June 8, 2011, 10:13 p.m.
From: Yuchung Cheng <ycheng@google.com>
Date: Wed, 8 Jun 2011 13:54:41 -0700

> Acked-by: Yuchung Cheng <ycheng@google.com>

Please do not top-post.  Quote the commit log message portion, delete
the patch itself since you're not giving any specific feedback, and
then insert your Acked-by/Tested-by/Reviewed-by/whatever right
afterwards.

I know that people are trained not to properly put context before the
things they say in email replied, and edit out the unnecessary
portions of the quoted material, simply because many popular email
clients tend to make this difficult and make top-posting way too easy,
but that is not our problem.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index e64f4c6..531ede8 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -282,6 +282,7 @@  struct tcp_request_sock {
 #endif
       u32                             rcv_isn;
       u32                             snt_isn;
+       u32                             snt_synack; /* synack sent time */
 };

 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
diff --git a/include/net/tcp.h b/include/net/tcp.h
index cda30ea..149a415 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -122,7 +122,13 @@  extern void tcp_time_wait(struct sock *sk, int
state, int timeo);
 #endif
 #define TCP_RTO_MAX    ((unsigned)(120*HZ))
 #define TCP_RTO_MIN    ((unsigned)(HZ/5))
-#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ))    /* RFC 1122 initial
RTO value   */
+#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))    /* RFC2988bis initial
RTO value */
+#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))        /* RFC 1122
initial RTO value, now
+                                                * used as a fallback
RTO for the
+                                                * initial data
transmission if no
+                                                * valid RTT sample
has been acquired,
+                                                * most likely due to
retrans in 3WHS.
+                                                */

 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal
interval between probes
                                                        * for local resources.
@@ -295,7 +301,7 @@  static inline void tcp_synq_overflow(struct sock *sk)
 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
 {
       unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
-       return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
+       return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
 }

 extern struct proto tcp_prot;
@@ -508,6 +514,7 @@  extern void tcp_initialize_rcv_mss(struct sock *sk);
 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
 extern void tcp_mtup_init(struct sock *sk);
+extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);

 static inline void tcp_bound_rto(const struct sock *sk)
 {
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 2646149..92bb943 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -316,6 +316,7 @@  struct sock *cookie_v4_check(struct sock *sk,
struct sk_buff *skb,
       ireq->wscale_ok         = tcp_opt.wscale_ok;
       ireq->tstamp_ok         = tcp_opt.saw_tstamp;
       req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
+       treq->snt_synack        = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;

       /* We throwed the options of the initial SYN away, so we hope
        * the ACK carries the same options again (see RFC1122 4.2.3.8)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index bef9f04..ea0d218 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -880,6 +880,11 @@  static void tcp_init_metrics(struct sock *sk)
               tp->snd_ssthresh = dst_metric(dst, RTAX_SSTHRESH);
               if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
                       tp->snd_ssthresh = tp->snd_cwnd_clamp;
+       } else {
+               /* ssthresh may have been reduced unnecessarily during.
+                * 3WHS. Restore it back to its initial default.
+                */
+               tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
       }
       if (dst_metric(dst, RTAX_REORDERING) &&
           tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
@@ -887,10 +892,7 @@  static void tcp_init_metrics(struct sock *sk)
               tp->reordering = dst_metric(dst, RTAX_REORDERING);
       }

-       if (dst_metric(dst, RTAX_RTT) == 0)
-               goto reset;
-
-       if (!tp->srtt && dst_metric_rtt(dst, RTAX_RTT) <
(TCP_TIMEOUT_INIT << 3))
+       if (dst_metric(dst, RTAX_RTT) == 0 || tp->srtt == 0)
               goto reset;

       /* Initial rtt is determined from SYN,SYN-ACK.
@@ -916,19 +918,26 @@  static void tcp_init_metrics(struct sock *sk)
               tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
       }
       tcp_set_rto(sk);
-       if (inet_csk(sk)->icsk_rto < TCP_TIMEOUT_INIT &&
!tp->rx_opt.saw_tstamp) {
 reset:
-               /* Play conservative. If timestamps are not
-                * supported, TCP will fail to recalculate correct
-                * rtt, if initial rto is too small. FORGET ALL AND RESET!
+       if (tp->srtt == 0) {
+               /* RFC2988bis: We've failed to get a valid RTT sample from
+                * 3WHS. This is most likely due to retransmission,
+                * including spurious one. Reset the RTO back to 3secs
+                * from the more aggressive 1sec to avoid more spurious
+                * retransmission.
                */
-               if (!tp->rx_opt.saw_tstamp && tp->srtt) {
-                       tp->srtt = 0;
-                       tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_INIT;
-                       inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
-               }
+               tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_FALLBACK;
+               inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
       }
-       tp->snd_cwnd = tcp_init_cwnd(tp, dst);
+       /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
+        * retransmitted. In light of RFC2988bis' more aggressive 1sec
+        * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
+        * retransmission has occurred.
+        */
+       if (tp->total_retrans > 1)
+               tp->snd_cwnd = 1;
+       else
+               tp->snd_cwnd = tcp_init_cwnd(tp, dst);
       tp->snd_cwnd_stamp = tcp_time_stamp;
 }

@@ -3112,12 +3121,13 @@  static void tcp_fastretrans_alert(struct sock
*sk, int pkts_acked, int flag)
       tcp_xmit_retransmit_queue(sk);
 }

-static void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
+void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
 {
       tcp_rtt_estimator(sk, seq_rtt);
       tcp_set_rto(sk);
       inet_csk(sk)->icsk_backoff = 0;
 }
+EXPORT_SYMBOL(tcp_valid_rtt_meas);

 /* Read draft-ietf-tcplw-high-performance before mucking
 * with this code. (Supersedes RFC1323)
@@ -5806,12 +5816,6 @@  int tcp_rcv_state_process(struct sock *sk,
struct sk_buff *skb,
                                             tp->rx_opt.snd_wscale;
                               tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);

-                               /* tcp_ack considers this ACK as duplicate
-                                * and does not calculate rtt.
-                                * Force it here.
-                                */
-                               tcp_ack_update_rtt(sk, 0, 0);
-
                               if (tp->rx_opt.tstamp_ok)
                                       tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;

diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 3c8d9b6..5fb504b 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -429,8 +429,8 @@  void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
                       break;

               icsk->icsk_backoff--;
-               inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
-                                        icsk->icsk_backoff;
+               inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
+                       TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
               tcp_bound_rto(sk);

               skb = tcp_write_queue_head(sk);
@@ -1384,6 +1384,7 @@  int tcp_v4_conn_request(struct sock *sk, struct
sk_buff *skb)
               isn = tcp_v4_init_sequence(skb);
       }
       tcp_rsk(req)->snt_isn = isn;
+       tcp_rsk(req)->snt_synack = tcp_time_stamp;

       if (tcp_v4_send_synack(sk, dst, req,
                              (struct request_values *)&tmp_ext) ||
@@ -1458,6 +1459,10 @@  struct sock *tcp_v4_syn_recv_sock(struct sock
*sk, struct sk_buff *skb,
               newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;

       tcp_initialize_rcv_mss(newsk);
+       if (tcp_rsk(req)->snt_synack)
+               tcp_valid_rtt_meas(newsk,
+                   tcp_time_stamp - tcp_rsk(req)->snt_synack);
+       newtp->total_retrans = req->retrans;

 #ifdef CONFIG_TCP_MD5SIG
       /* Copy over the MD5 key from the original socket */
@@ -1854,7 +1859,7 @@  static int tcp_v4_init_sock(struct sock *sk)
        * algorithms that we must have the following bandaid to talk
        * efficiently to them.  -DaveM
        */
-       tp->snd_cwnd = 2;
+       tp->snd_cwnd = TCP_INIT_CWND;

       /* See draft-stevens-tcpca-spec-01 for discussion of the
        * initialization of these values.
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 80b1f80..d2fe4e0 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -486,7 +486,7 @@  struct sock *tcp_create_openreq_child(struct sock
*sk, struct request_sock *req,
                * algorithms that we must have the following bandaid to talk
                * efficiently to them.  -DaveM
                */
-               newtp->snd_cwnd = 2;
+               newtp->snd_cwnd = TCP_INIT_CWND;
               newtp->snd_cwnd_cnt = 0;
               newtp->bytes_acked = 0;

@@ -720,6 +720,10 @@  struct sock *tcp_check_req(struct sock *sk,
struct sk_buff *skb,
               NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
               return NULL;
       }
+       if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
+               tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
+       else if (req->retrans) /* don't take RTT sample if retrans && ~TS */
+               tcp_rsk(req)->snt_synack = 0;

       /* OK, ACK is valid, create big socket and
        * feed this segment to it. It will repeat all
diff --git a/net/ipv6/syncookies.c b/net/ipv6/syncookies.c
index 8b9644a..89d5bf8 100644
--- a/net/ipv6/syncookies.c
+++ b/net/ipv6/syncookies.c
@@ -223,6 +223,7 @@  struct sock *cookie_v6_check(struct sock *sk,
struct sk_buff *skb)
       ireq->wscale_ok         = tcp_opt.wscale_ok;
       ireq->tstamp_ok         = tcp_opt.saw_tstamp;
       req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
+       treq->snt_synack        = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
       treq->rcv_isn = ntohl(th->seq) - 1;
       treq->snt_isn = cookie;

diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 8683664..e7d47e4 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1341,6 +1341,7 @@  static int tcp_v6_conn_request(struct sock *sk,
struct sk_buff *skb)
       }
 have_isn:
       tcp_rsk(req)->snt_isn = isn;
+       tcp_rsk(req)->snt_synack = tcp_time_stamp;

       security_inet_conn_request(sk, skb, req);