xnu-3789.31.2.tar.gz

[apple/xnu.git] / bsd / netinet / tcp_usrreq.c

diff --git a/bsd/netinet/tcp_usrreq.c b/bsd/netinet/tcp_usrreq.c
index 96b17ba5afc36dcc5d686091895d01c405f08f98..f50103ea04447d01573738b11aff33858056a226 100644 (file)
--- a/bsd/netinet/tcp_usrreq.c
+++ b/bsd/netinet/tcp_usrreq.c
@@ -1,8 +1,8 @@
 /*
 /*

- * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2016 Apple Inc. All rights reserved.

  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@

- * 
+ *

  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in


@@ -11,10 +11,10 @@
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.

- * 
+ *

  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.

- * 
+ *

  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,


@@ -22,7 +22,7 @@
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.

- * 
+ *

  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*


@@ -136,13 +136,8 @@ static int tcp6_connect(struct tcpcb *, struct sockaddr *, struct proc *);
 static int     tcp6_usr_connect(struct socket *, struct sockaddr *,
                    struct proc *);
 #endif /* INET6 */
 static int     tcp6_usr_connect(struct socket *, struct sockaddr *,
                    struct proc *);
 #endif /* INET6 */

-static struct tcpcb *
-               tcp_disconnect(struct tcpcb *);
-static struct tcpcb *
-               tcp_usrclosed(struct tcpcb *);
-
-extern uint32_t tcp_autorcvbuf_max;
-
+static struct tcpcb *tcp_disconnect(struct tcpcb *);
+static struct tcpcb *tcp_usrclosed(struct tcpcb *);

 extern void tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sb);
 
 #if TCPDEBUG
 extern void tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sb);
 
 #if TCPDEBUG


@@ -183,7 +178,7 @@ tcp_usr_attach(struct socket *so, __unused int proto, struct proc *p)
                error = EISCONN;
                goto out;
        }
                error = EISCONN;
                goto out;
        }

-       
+

        error = tcp_attach(so, p);
        if (error)
                goto out;
        error = tcp_attach(so, p);
        if (error)
                goto out;


@@ -457,8 +452,6 @@ tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p)
                        error = ENETDOWN;
                }
 
                        error = ENETDOWN;
                }
 

-               /* Disable PRECONNECT_DATA, as we don't need to send a SYN anymore. */
-               so->so_flags1 &= ~SOF1_PRECONNECT_DATA;

                return error;
        }
 #endif /* FLOW_DIVERT */
                return error;
        }
 #endif /* FLOW_DIVERT */


@@ -671,6 +664,7 @@ tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p)
                } else {
                        error = ENETDOWN;
                }
                } else {
                        error = ENETDOWN;
                }

+

                return error;
        }
 #endif /* FLOW_DIVERT */
                return error;
        }
 #endif /* FLOW_DIVERT */


@@ -1180,6 +1174,7 @@ tcp_usr_abort(struct socket *so)
         if (tp == NULL)
                goto out;
        tp = tcp_drop(tp, ECONNABORTED);
         if (tp == NULL)
                goto out;
        tp = tcp_drop(tp, ECONNABORTED);

+       VERIFY(so->so_usecount > 0);

        so->so_usecount--;
        COMMON_END(PRU_ABORT);
 }
        so->so_usecount--;
        COMMON_END(PRU_ABORT);
 }


@@ -1213,6 +1208,7 @@ tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
        }
        m->m_len = 1;
        *mtod(m, caddr_t) = tp->t_iobc;
        }
        m->m_len = 1;
        *mtod(m, caddr_t) = tp->t_iobc;

+       so->so_state &= ~SS_RCVATMARK;

        if ((flags & MSG_PEEK) == 0)
                tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
        COMMON_END(PRU_RCVOOB);
        if ((flags & MSG_PEEK) == 0)
                tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
        COMMON_END(PRU_RCVOOB);


@@ -1221,7 +1217,20 @@ tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
 static int
 tcp_usr_preconnect(struct socket *so)
 {
 static int
 tcp_usr_preconnect(struct socket *so)
 {

-       int error = tcp_output(sototcpcb(so));
+       struct inpcb *inp = sotoinpcb(so);
+       int error = 0;
+
+#if NECP
+       if (necp_socket_should_use_flow_divert(inp)) {
+               /* May happen, if in tcp_usr_connect we did not had a chance
+                * to set the usrreqs (due to some error). So, let's get out
+                * of here.
+                */
+               goto out;
+       }
+#endif /* NECP */
+
+       error = tcp_output(sototcpcb(so));

 
        /* One read has been done. This was enough. Get back to "normal" behavior. */
        so->so_flags1 &= ~SOF1_PRECONNECT_DATA;
 
        /* One read has been done. This was enough. Get back to "normal" behavior. */
        so->so_flags1 &= ~SOF1_PRECONNECT_DATA;


@@ -1303,18 +1312,13 @@ struct pr_usrreqs tcp6_usrreqs = {
  *     in_pcbladdr:EADDRNOTAVAIL       Address not available
  */
 static int
  *     in_pcbladdr:EADDRNOTAVAIL       Address not available
  */
 static int

-tcp_connect(tp, nam, p)
-       register struct tcpcb *tp;
-       struct sockaddr *nam;
-       struct proc *p;
+tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct proc *p)

 {
        struct inpcb *inp = tp->t_inpcb, *oinp;
        struct socket *so = inp->inp_socket;
        struct tcpcb *otp;
        struct sockaddr_in *sin = (struct sockaddr_in *)(void *)nam;
        struct in_addr laddr;
 {
        struct inpcb *inp = tp->t_inpcb, *oinp;
        struct socket *so = inp->inp_socket;
        struct tcpcb *otp;
        struct sockaddr_in *sin = (struct sockaddr_in *)(void *)nam;
        struct in_addr laddr;

-       struct rmxp_tao *taop;
-       struct rmxp_tao tao_noncached;

        int error = 0;
        struct ifnet *outif = NULL;
 
        int error = 0;
        struct ifnet *outif = NULL;
 


@@ -1329,7 +1333,7 @@ tcp_connect(tp, nam, p)
         * earlier incarnation of this same connection still in
         * TIME_WAIT state, creating an ADDRINUSE error.
         */
         * earlier incarnation of this same connection still in
         * TIME_WAIT state, creating an ADDRINUSE error.
         */

-       error = in_pcbladdr(inp, nam, &laddr, IFSCOPE_NONE, &outif);
+       error = in_pcbladdr(inp, nam, &laddr, IFSCOPE_NONE, &outif, 0);

        if (error)
                goto done;
 
        if (error)
                goto done;
 


@@ -1392,7 +1396,7 @@ skip_oinp:
        if (inp->inp_flowhash == 0)
                inp->inp_flowhash = inp_calc_flowhash(inp);
 
        if (inp->inp_flowhash == 0)
                inp->inp_flowhash = inp_calc_flowhash(inp);
 

-       tcp_set_max_rwinscale(tp, so);
+       tcp_set_max_rwinscale(tp, so, TCP_AUTORCVBUF_MAX(outif));

 
        soisconnecting(so);
        tcpstat.tcps_connattempt++;
 
        soisconnecting(so);
        tcpstat.tcps_connattempt++;


@@ -1403,24 +1407,6 @@ skip_oinp:
        if (nstat_collect)
                nstat_route_connect_attempt(inp->inp_route.ro_rt);
 
        if (nstat_collect)
                nstat_route_connect_attempt(inp->inp_route.ro_rt);
 

-       /*
-        * Generate a CC value for this connection and
-        * check whether CC or CCnew should be used.
-        */
-       if ((taop = tcp_gettaocache(tp->t_inpcb)) == NULL) {
-               taop = &tao_noncached;
-               bzero(taop, sizeof(*taop));
-       }
-
-       tp->cc_send = CC_INC(tcp_ccgen);
-       if (taop->tao_ccsent != 0 &&
-           CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
-               taop->tao_ccsent = tp->cc_send;
-       } else {
-               taop->tao_ccsent = 0;
-               tp->t_flags |= TF_SENDCCNEW;
-       }
-

 done:
        if (outif != NULL)
                ifnet_release(outif);
 done:
        if (outif != NULL)
                ifnet_release(outif);


@@ -1430,18 +1416,13 @@ done:
 
 #if INET6
 static int
 
 #if INET6
 static int

-tcp6_connect(tp, nam, p)
-       register struct tcpcb *tp;
-       struct sockaddr *nam;
-       struct proc *p;
+tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct proc *p)

 {
        struct inpcb *inp = tp->t_inpcb, *oinp;
        struct socket *so = inp->inp_socket;
        struct tcpcb *otp;
        struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)(void *)nam;
        struct in6_addr addr6;
 {
        struct inpcb *inp = tp->t_inpcb, *oinp;
        struct socket *so = inp->inp_socket;
        struct tcpcb *otp;
        struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)(void *)nam;
        struct in6_addr addr6;

-       struct rmxp_tao *taop;
-       struct rmxp_tao tao_noncached;

        int error = 0;
        struct ifnet *outif = NULL;
 
        int error = 0;
        struct ifnet *outif = NULL;
 


@@ -1509,7 +1490,7 @@ tcp6_connect(tp, nam, p)
                    (htonl(inp->inp_flowhash) & IPV6_FLOWLABEL_MASK);
        }
 
                    (htonl(inp->inp_flowhash) & IPV6_FLOWLABEL_MASK);
        }
 

-       tcp_set_max_rwinscale(tp, so);
+       tcp_set_max_rwinscale(tp, so, TCP_AUTORCVBUF_MAX(outif));

 
        soisconnecting(so);
        tcpstat.tcps_connattempt++;
 
        soisconnecting(so);
        tcpstat.tcps_connattempt++;


@@ -1521,24 +1502,6 @@ tcp6_connect(tp, nam, p)
        if (nstat_collect)
                nstat_route_connect_attempt(inp->inp_route.ro_rt);
 
        if (nstat_collect)
                nstat_route_connect_attempt(inp->inp_route.ro_rt);
 

-       /*
-        * Generate a CC value for this connection and
-        * check whether CC or CCnew should be used.
-        */
-       if ((taop = tcp_gettaocache(tp->t_inpcb)) == NULL) {
-               taop = &tao_noncached;
-               bzero(taop, sizeof(*taop));
-       }
-
-       tp->cc_send = CC_INC(tcp_ccgen);
-       if (taop->tao_ccsent != 0 &&
-           CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
-               taop->tao_ccsent = tp->cc_send;
-       } else {
-               taop->tao_ccsent = 0;
-               tp->t_flags |= TF_SENDCCNEW;
-       }
-

 done:
        if (outif != NULL)
                ifnet_release(outif);
 done:
        if (outif != NULL)
                ifnet_release(outif);


@@ -1558,7 +1521,8 @@ tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
        bzero(ti, sizeof(*ti));
 
        ti->tcpi_state = tp->t_state;
        bzero(ti, sizeof(*ti));
 
        ti->tcpi_state = tp->t_state;

-       
+       ti->tcpi_flowhash = inp->inp_flowhash;
+

        if (tp->t_state > TCPS_LISTEN) {
                if (TSTMP_SUPPORTED(tp))
                        ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
        if (tp->t_state > TCPS_LISTEN) {
                if (TSTMP_SUPPORTED(tp))
                        ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;


@@ -1569,11 +1533,16 @@ tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
                        ti->tcpi_snd_wscale = tp->snd_scale;
                        ti->tcpi_rcv_wscale = tp->rcv_scale;
                }
                        ti->tcpi_snd_wscale = tp->snd_scale;
                        ti->tcpi_rcv_wscale = tp->rcv_scale;
                }

+               if (TCP_ECN_ENABLED(tp))
+                       ti->tcpi_options |= TCPI_OPT_ECN;

 
                /* Are we in retranmission episode */
                if (IN_FASTRECOVERY(tp) || tp->t_rxtshift > 0)
                        ti->tcpi_flags |= TCPI_FLAG_LOSSRECOVERY;
 
 
                /* Are we in retranmission episode */
                if (IN_FASTRECOVERY(tp) || tp->t_rxtshift > 0)
                        ti->tcpi_flags |= TCPI_FLAG_LOSSRECOVERY;
 

+               if (tp->t_flags & TF_STREAMING_ON)
+                       ti->tcpi_flags |= TCPI_FLAG_STREAMING_ON;
+

                ti->tcpi_rto = tp->t_timer[TCPT_REXMT] ? tp->t_rxtcur : 0;
                ti->tcpi_snd_mss = tp->t_maxseg;
                ti->tcpi_rcv_mss = tp->t_maxseg;
                ti->tcpi_rto = tp->t_timer[TCPT_REXMT] ? tp->t_rxtcur : 0;
                ti->tcpi_snd_mss = tp->t_maxseg;
                ti->tcpi_rcv_mss = tp->t_maxseg;


@@ -1643,6 +1612,36 @@ tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
                ti->tcpi_tfo_syn_data_sent = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_SENT);
                ti->tcpi_tfo_syn_data_acked = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_ACKED);
                ti->tcpi_tfo_syn_loss = !!(tp->t_tfo_stats & TFO_S_SYN_LOSS);
                ti->tcpi_tfo_syn_data_sent = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_SENT);
                ti->tcpi_tfo_syn_data_acked = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_ACKED);
                ti->tcpi_tfo_syn_loss = !!(tp->t_tfo_stats & TFO_S_SYN_LOSS);

+               ti->tcpi_tfo_cookie_wrong = !!(tp->t_tfo_stats & TFO_S_COOKIE_WRONG);
+               ti->tcpi_tfo_no_cookie_rcv = !!(tp->t_tfo_stats & TFO_S_NO_COOKIE_RCV);
+               ti->tcpi_tfo_heuristics_disable = !!(tp->t_tfo_stats & TFO_S_HEURISTICS_DISABLE);
+               ti->tcpi_tfo_send_blackhole = !!(tp->t_tfo_stats & TFO_S_SEND_BLACKHOLE);
+               ti->tcpi_tfo_recv_blackhole = !!(tp->t_tfo_stats & TFO_S_RECV_BLACKHOLE);
+
+               ti->tcpi_ecn_client_setup = !!(tp->ecn_flags & TE_SETUPSENT);
+               ti->tcpi_ecn_server_setup = !!(tp->ecn_flags & TE_SETUPRECEIVED);
+               ti->tcpi_ecn_success = (tp->ecn_flags & TE_ECN_ON) == TE_ECN_ON ? 1 : 0;
+               ti->tcpi_ecn_lost_syn = !!(tp->ecn_flags & TE_LOST_SYN);
+               ti->tcpi_ecn_lost_synack = !!(tp->ecn_flags & TE_LOST_SYNACK);
+
+               ti->tcpi_local_peer = !!(tp->t_flags & TF_LOCAL);
+
+               if (tp->t_inpcb->inp_last_outifp != NULL) {
+                       if (IFNET_IS_CELLULAR(tp->t_inpcb->inp_last_outifp))
+                               ti->tcpi_if_cell = 1;
+                       else if (IFNET_IS_WIFI(tp->t_inpcb->inp_last_outifp))
+                               ti->tcpi_if_wifi = 1;
+               }
+
+               ti->tcpi_ecn_recv_ce = tp->t_ecn_recv_ce;
+               ti->tcpi_ecn_recv_cwr = tp->t_ecn_recv_cwr;
+
+               ti->tcpi_rcvoopack = tp->t_rcvoopack;
+               ti->tcpi_pawsdrop = tp->t_pawsdrop;
+               ti->tcpi_sack_recovery_episode = tp->t_sack_recovery_episode;
+               ti->tcpi_reordered_pkts = tp->t_reordered_pkts;
+               ti->tcpi_dsack_sent = tp->t_dsack_sent;
+               ti->tcpi_dsack_recvd = tp->t_dsack_recvd;

        }
 }
 
        }
 }
 


@@ -1757,6 +1756,11 @@ tcp_connection_fill_info(struct tcpcb *tp, struct tcp_connection_info *tci)
                tci->tcpi_tfo_syn_data_sent = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_SENT);
                tci->tcpi_tfo_syn_data_acked = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_ACKED);
                tci->tcpi_tfo_syn_loss = !!(tp->t_tfo_stats & TFO_S_SYN_LOSS);
                tci->tcpi_tfo_syn_data_sent = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_SENT);
                tci->tcpi_tfo_syn_data_acked = !!(tp->t_tfo_stats & TFO_S_SYN_DATA_ACKED);
                tci->tcpi_tfo_syn_loss = !!(tp->t_tfo_stats & TFO_S_SYN_LOSS);

+               tci->tcpi_tfo_cookie_wrong = !!(tp->t_tfo_stats & TFO_S_COOKIE_WRONG);
+               tci->tcpi_tfo_no_cookie_rcv = !!(tp->t_tfo_stats & TFO_S_NO_COOKIE_RCV);
+               tci->tcpi_tfo_heuristics_disable = !!(tp->t_tfo_stats & TFO_S_HEURISTICS_DISABLE);
+               tci->tcpi_tfo_send_blackhole = !!(tp->t_tfo_stats & TFO_S_SEND_BLACKHOLE);
+               tci->tcpi_tfo_recv_blackhole = !!(tp->t_tfo_stats & TFO_S_RECV_BLACKHOLE);

        }
 }
 
        }
 }
 


@@ -1875,9 +1879,7 @@ tcp_getconninfo(struct socket *so, struct conninfo_tcp *tcp_ci)
  * splnet() any more.  This needs more examination.)
  */
 int
  * splnet() any more.  This needs more examination.)
  */
 int

-tcp_ctloutput(so, sopt)
-       struct socket *so;
-       struct sockopt *sopt;
+tcp_ctloutput(struct socket *so, struct sockopt *sopt)

 {
        int     error, opt, optval;
        struct  inpcb *inp;
 {
        int     error, opt, optval;
        struct  inpcb *inp;


@@ -1913,7 +1915,6 @@ tcp_ctloutput(so, sopt)
                case TCP_NODELAY:
                case TCP_NOOPT:
                case TCP_NOPUSH:
                case TCP_NODELAY:
                case TCP_NOOPT:
                case TCP_NOPUSH:

-               case TCP_ENABLE_ECN:

                        error = sooptcopyin(sopt, &optval, sizeof optval,
                                            sizeof optval);
                        if (error)
                        error = sooptcopyin(sopt, &optval, sizeof optval,
                                            sizeof optval);
                        if (error)


@@ -1929,9 +1930,6 @@ tcp_ctloutput(so, sopt)
                        case TCP_NOPUSH:
                                opt = TF_NOPUSH;
                                break;
                        case TCP_NOPUSH:
                                opt = TF_NOPUSH;
                                break;

-                       case TCP_ENABLE_ECN:
-                               opt = TF_ENABLE_ECN;
-                               break;

                        default:
                                opt = 0; /* dead code to fool gcc */
                                break;
                        default:
                                opt = 0; /* dead code to fool gcc */
                                break;


@@ -2099,6 +2097,21 @@ tcp_ctloutput(so, sopt)
                        }
                        break;
 
                        }
                        break;
 

+               case TCP_KEEPALIVE_OFFLOAD:
+                       error = sooptcopyin(sopt, &optval, sizeof(optval),
+                               sizeof(optval));
+                       if (error)
+                               break;
+                       if (optval < 0 || optval > INT32_MAX) {
+                               error = EINVAL;
+                               break;
+                       } 
+                       if (optval != 0)
+                               inp->inp_flags2 |= INP2_KEEPALIVE_OFFLOAD;
+                       else
+                               inp->inp_flags2 &= ~INP2_KEEPALIVE_OFFLOAD;
+                       break;
+

                case PERSIST_TIMEOUT:
                        error = sooptcopyin(sopt, &optval, sizeof optval,
                                                sizeof optval);
                case PERSIST_TIMEOUT:
                        error = sooptcopyin(sopt, &optval, sizeof optval,
                                                sizeof optval);


@@ -2184,13 +2197,13 @@ tcp_ctloutput(so, sopt)
                                }
 
                                /*
                                }
 
                                /*

-                                * allocate memory for storing message 
+                                * allocate memory for storing message

                                 * related state
                                 */
                                VERIFY(so->so_msg_state == NULL);
                                 * related state
                                 */
                                VERIFY(so->so_msg_state == NULL);

-                               MALLOC(so->so_msg_state, 
+                               MALLOC(so->so_msg_state,

                                        struct msg_state *,
                                        struct msg_state *,

-                                       sizeof(struct msg_state), 
+                                       sizeof(struct msg_state),

                                        M_TEMP, M_WAITOK | M_ZERO);
                                if (so->so_msg_state == NULL) {
                                        error = ENOMEM;
                                        M_TEMP, M_WAITOK | M_ZERO);
                                if (so->so_msg_state == NULL) {
                                        error = ENOMEM;


@@ -2200,9 +2213,9 @@ tcp_ctloutput(so, sopt)
                                /* Enable message delivery */
                                so->so_flags |= SOF_ENABLE_MSGS;
                        } else {
                                /* Enable message delivery */
                                so->so_flags |= SOF_ENABLE_MSGS;
                        } else {

-                               /* 
-                                * Can't disable message delivery on socket 
-                                * because of restrictions imposed by 
+                               /*
+                                * Can't disable message delivery on socket
+                                * because of restrictions imposed by

                                 * encoding/decoding
                                 */
                                error = EINVAL;
                                 * encoding/decoding
                                 */
                                error = EINVAL;


@@ -2260,6 +2273,60 @@ tcp_ctloutput(so, sopt)
                        else
                                tcp_disable_tfo(tp);
                        break;
                        else
                                tcp_disable_tfo(tp);
                        break;

+               case TCP_ENABLE_ECN:
+                       error = sooptcopyin(sopt, &optval, sizeof optval,
+                                           sizeof optval);
+                       if (error)
+                               break;
+                       if (optval) {
+                               tp->ecn_flags |= TE_ECN_MODE_ENABLE;
+                               tp->ecn_flags &= ~TE_ECN_MODE_DISABLE;
+                       } else {
+                               tp->ecn_flags &= ~TE_ECN_MODE_ENABLE;
+                       }
+                       break;
+               case TCP_ECN_MODE:
+                       error = sooptcopyin(sopt, &optval, sizeof optval,
+                                           sizeof optval);
+                       if (error)
+                               break;
+                       if (optval == ECN_MODE_DEFAULT) {
+                               tp->ecn_flags &= ~TE_ECN_MODE_ENABLE;
+                               tp->ecn_flags &= ~TE_ECN_MODE_DISABLE;
+                       } else if (optval == ECN_MODE_ENABLE) {
+                               tp->ecn_flags |= TE_ECN_MODE_ENABLE;
+                               tp->ecn_flags &= ~TE_ECN_MODE_DISABLE;
+                       } else if (optval == ECN_MODE_DISABLE) {
+                               tp->ecn_flags &= ~TE_ECN_MODE_ENABLE;
+                               tp->ecn_flags |= TE_ECN_MODE_DISABLE;
+                       } else {
+                               error = EINVAL;
+                       }
+                       break;
+               case TCP_NOTIFY_ACKNOWLEDGEMENT:
+                       error = sooptcopyin(sopt, &optval,
+                           sizeof(optval), sizeof(optval));
+                       if (error)
+                               break;
+                       if (optval <= 0) {
+                               error = EINVAL;
+                               break;
+                       }
+                       if (tp->t_notify_ack_count >= TCP_MAX_NOTIFY_ACK) {
+                               error = ETOOMANYREFS;
+                               break;
+                       }
+
+                       /*
+                        * validate that the given marker id is not
+                        * a duplicate to avoid ambiguity
+                        */
+                       if ((error = tcp_notify_ack_id_valid(tp, so,
+                           optval)) != 0) {
+                               break;
+                       }
+                       error = tcp_add_notify_ack_marker(tp, optval);
+                       break;

                case SO_FLUSH:
                        if ((error = sooptcopyin(sopt, &optval, sizeof (optval),
                            sizeof (optval))) != 0)
                case SO_FLUSH:
                        if ((error = sooptcopyin(sopt, &optval, sizeof (optval),
                            sizeof (optval))) != 0)


@@ -2297,13 +2364,25 @@ tcp_ctloutput(so, sopt)
                        optval = tp->t_maxseg;
                        break;
                case TCP_KEEPALIVE:
                        optval = tp->t_maxseg;
                        break;
                case TCP_KEEPALIVE:

-                       optval = tp->t_keepidle / TCP_RETRANSHZ;
+                       if (tp->t_keepidle > 0)
+                               optval = tp->t_keepidle / TCP_RETRANSHZ;
+                       else
+                               optval = tcp_keepidle  / TCP_RETRANSHZ;

                        break;
                case TCP_KEEPINTVL:
                        break;
                case TCP_KEEPINTVL:

-                       optval = tp->t_keepintvl / TCP_RETRANSHZ;
+                       if (tp->t_keepintvl > 0)
+                               optval = tp->t_keepintvl / TCP_RETRANSHZ;
+                       else
+                               optval = tcp_keepintvl / TCP_RETRANSHZ;

                        break;
                case TCP_KEEPCNT:
                        break;
                case TCP_KEEPCNT:

-                       optval = tp->t_keepcnt;
+                       if (tp->t_keepcnt > 0)
+                               optval = tp->t_keepcnt;
+                       else
+                               optval = tcp_keepcnt;
+                       break;
+               case TCP_KEEPALIVE_OFFLOAD:
+                       optval = !!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD);

                        break;
                case TCP_NOOPT:
                        optval = tp->t_flags & TF_NOOPT;
                        break;
                case TCP_NOOPT:
                        optval = tp->t_flags & TF_NOOPT;


@@ -2312,7 +2391,15 @@ tcp_ctloutput(so, sopt)
                        optval = tp->t_flags & TF_NOPUSH;
                        break;
                case TCP_ENABLE_ECN:
                        optval = tp->t_flags & TF_NOPUSH;
                        break;
                case TCP_ENABLE_ECN:

-                       optval = (tp->t_flags & TF_ENABLE_ECN) ? 1 : 0; 
+                       optval = (tp->ecn_flags & TE_ECN_MODE_ENABLE) ? 1 : 0;
+                       break;
+               case TCP_ECN_MODE:
+                       if (tp->ecn_flags & TE_ECN_MODE_ENABLE)
+                               optval = ECN_MODE_ENABLE;
+                       else if (tp->ecn_flags & TE_ECN_MODE_DISABLE)
+                               optval = ECN_MODE_DISABLE;
+                       else
+                               optval = ECN_MODE_DEFAULT;

                        break;
                case TCP_CONNECTIONTIMEOUT:
                        optval = tp->t_keepinit / TCP_RETRANSHZ;
                        break;
                case TCP_CONNECTIONTIMEOUT:
                        optval = tp->t_keepinit / TCP_RETRANSHZ;


@@ -2408,9 +2495,24 @@ tcp_ctloutput(so, sopt)
                        optval = tp->t_adaptive_wtimo;
                        break;
                case SO_TRAFFIC_MGT_BACKGROUND:
                        optval = tp->t_adaptive_wtimo;
                        break;
                case SO_TRAFFIC_MGT_BACKGROUND:

-                       optval = (so->so_traffic_mgt_flags &
-                           TRAFFIC_MGT_SO_BACKGROUND) ? 1 : 0;
+                       optval = (so->so_flags1 &
+                           SOF1_TRAFFIC_MGT_SO_BACKGROUND) ? 1 : 0;

                        break;
                        break;

+               case TCP_NOTIFY_ACKNOWLEDGEMENT: {
+                       struct tcp_notify_ack_complete retid;
+
+                       if (sopt->sopt_valsize != sizeof (retid)) {
+                               error = EINVAL;
+                               break;
+                       }
+                       bzero(&retid, sizeof (retid));
+                       tcp_get_notify_ack_count(tp, &retid);
+                       if (retid.notify_complete_count > 0)
+                               tcp_get_notify_ack_ids(tp, &retid);
+
+                       error = sooptcopyout(sopt, &retid, sizeof (retid));
+                       goto done;
+               }

                default:
                        error = ENOPROTOOPT;
                        break;
                default:
                        error = ENOPROTOOPT;
                        break;


@@ -2470,15 +2572,6 @@ SYSCTL_PROC(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLTYPE_INT | CTLFLAG_RW
 SYSCTL_PROC(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_recvspace , 0, &sysctl_tcp_sospace, "IU", "Maximum incoming TCP datagram size");
 
 SYSCTL_PROC(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_recvspace , 0, &sysctl_tcp_sospace, "IU", "Maximum incoming TCP datagram size");
 

-/* Sysctl for testing and tuning the connectx with data api */
-#define TCP_PRECONNECT_SBSZ_MAX 1460
-#define TCP_PRECONNECT_SBSZ_MIN (TCP_MSS)
-#define TCP_PRECONNECT_SBSZ_DEF        (TCP6_MSS)
-static int tcp_preconnect_sbspace = TCP_PRECONNECT_SBSZ_DEF;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, preconn_sbsz, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &tcp_preconnect_sbspace, 0, "Maximum preconnect space");
-
-

 /*
  * Attach TCP protocol to socket, allocating
  * internet protocol control block, tcp control block,
 /*
  * Attach TCP protocol to socket, allocating
  * internet protocol control block, tcp control block,


@@ -2491,11 +2584,9 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, preconn_sbsz, CTLFLAG_RW | CTLFLAG_LOCKED,
  *     soreserve:ENOBUFS
  */
 static int
  *     soreserve:ENOBUFS
  */
 static int

-tcp_attach(so, p)
-       struct socket *so;
-       struct proc *p;
+tcp_attach(struct socket *so, struct proc *p)

 {
 {

-       register struct tcpcb *tp;
+       struct tcpcb *tp;

        struct inpcb *inp;
        int error;
 #if INET6
        struct inpcb *inp;
        int error;
 #if INET6


@@ -2515,8 +2606,7 @@ tcp_attach(so, p)
        }
 
        if (so->so_snd.sb_preconn_hiwat == 0) {
        }
 
        if (so->so_snd.sb_preconn_hiwat == 0) {

-               soreserve_preconnect(so, imin(TCP_PRECONNECT_SBSZ_MAX,
-                   imax(tcp_preconnect_sbspace, TCP_PRECONNECT_SBSZ_MIN)));
+               soreserve_preconnect(so, 2048);

        }
 
        if ((so->so_rcv.sb_flags & SB_USRSIZE) == 0)
        }
 
        if ((so->so_rcv.sb_flags & SB_USRSIZE) == 0)


@@ -2561,8 +2651,7 @@ tcp_attach(so, p)
  * send segment to peer (with FIN).
  */
 static struct tcpcb *
  * send segment to peer (with FIN).
  */
 static struct tcpcb *

-tcp_disconnect(tp)
-       register struct tcpcb *tp;
+tcp_disconnect(struct tcpcb *tp)

 {
        struct socket *so = tp->t_inpcb->inp_socket;
 
 {
        struct socket *so = tp->t_inpcb->inp_socket;
 


@@ -2597,10 +2686,8 @@ tcp_disconnect(tp)
  * We can let the user exit from the close as soon as the FIN is acked.
  */
 static struct tcpcb *
  * We can let the user exit from the close as soon as the FIN is acked.
  */
 static struct tcpcb *

-tcp_usrclosed(tp)
-       register struct tcpcb *tp;
+tcp_usrclosed(struct tcpcb *tp)

 {
 {

-

        switch (tp->t_state) {
 
        case TCPS_CLOSED:
        switch (tp->t_state) {
 
        case TCPS_CLOSED: