2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1993
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
63 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
66 #include "opt_ipsec.h"
68 #include "opt_inet6.h"
69 #include "opt_tcpdebug.h"
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
81 #include <sys/domain.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/socketops.h>
86 #include <sys/protosw.h>
88 #include <sys/thread2.h>
89 #include <sys/msgport2.h>
90 #include <sys/socketvar2.h>
93 #include <net/netisr.h>
94 #include <net/route.h>
96 #include <net/netmsg2.h>
97 #include <net/netisr2.h>
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
102 #include <netinet/ip6.h>
104 #include <netinet/in_pcb.h>
106 #include <netinet6/in6_pcb.h>
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/tcp6_var.h>
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_timer2.h>
119 #include <netinet/tcp_var.h>
120 #include <netinet/tcpip.h>
122 #include <netinet/tcp_debug.h>
126 #include <netinet6/ipsec.h>
130 * TCP protocol interface to socket abstraction.
132 extern char *tcpstates[]; /* XXX ??? */
134 static int tcp_attach (struct socket *, struct pru_attach_info *);
135 static void tcp_connect (netmsg_t msg);
137 static void tcp6_connect (netmsg_t msg);
138 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
140 struct sockaddr_in6 *sin6,
141 struct in6_addr *addr6);
143 static struct tcpcb *
144 tcp_disconnect (struct tcpcb *);
145 static struct tcpcb *
146 tcp_usrclosed (struct tcpcb *);
149 #define TCPDEBUG0 int ostate = 0
150 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
151 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
152 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
156 #define TCPDEBUG2(req)
159 static int tcp_lport_extension = 1;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
161 &tcp_lport_extension, 0, "");
164 * For some ill optimized programs, which try to use TCP_NOPUSH
165 * to improve performance, will have small amount of data sits
166 * in the sending buffer. These small amount of data will _not_
167 * be pushed into the network until more data are written into
168 * the socket or the socket write side is shutdown.
170 static int tcp_disable_nopush = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
172 &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
175 * Allocate socket buffer space.
178 tcp_usr_preattach(struct socket *so, int proto __unused,
179 struct pru_attach_info *ai)
183 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
184 error = soreserve(so, tcp_sendspace, tcp_recvspace,
189 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
190 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
196 * TCP attaches to socket via pru_attach(), reserving space,
197 * and an internet control block. This socket may move to
198 * other CPU later when we bind/connect.
201 tcp_usr_attach(netmsg_t msg)
203 struct socket *so = msg->base.nm_so;
204 struct pru_attach_info *ai = msg->attach.nm_ai;
207 struct tcpcb *tp = NULL;
212 KASSERT(inp == NULL, ("tcp socket attached"));
215 error = tcp_attach(so, ai);
219 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
220 so->so_linger = TCP_LINGERTIME;
223 sofree(so); /* from ref above */
224 TCPDEBUG2(PRU_ATTACH);
225 lwkt_replymsg(&msg->lmsg, error);
229 * pru_detach() detaches the TCP protocol from the socket.
230 * If the protocol state is non-embryonic, then can't
231 * do this directly: have to initiate a pru_disconnect(),
232 * which may finish later; embryonic TCB's can just
236 tcp_usr_detach(netmsg_t msg)
238 struct socket *so = msg->base.nm_so;
247 * If the inp is already detached or never attached, it may have
248 * been due to an async close or async attach failure. Just return
249 * as if no error occured.
251 * It's possible for the tcpcb (tp) to disconnect from the inp due
252 * to tcp_drop()->tcp_close() being called. This may occur *after*
253 * the detach message has been queued so we may find a NULL tp here.
256 if ((tp = intotcpcb(inp)) != NULL) {
258 tp = tcp_disconnect(tp);
259 TCPDEBUG2(PRU_DETACH);
262 lwkt_replymsg(&msg->lmsg, error);
266 * NOTE: ignore_error is non-zero for certain disconnection races
267 * which we want to silently allow, otherwise close() may return
268 * an unexpected error.
270 * NOTE: The variables (msg) and (tp) are assumed.
272 #define COMMON_START(so, inp, ignore_error) \
278 error = ignore_error ? 0 : EINVAL; \
282 tp = intotcpcb(inp); \
286 #define COMMON_END1(req, noreply) \
290 lwkt_replymsg(&msg->lmsg, error); \
294 #define COMMON_END(req) COMMON_END1((req), 0)
297 * Give the socket an address.
300 tcp_usr_bind(netmsg_t msg)
302 struct socket *so = msg->bind.base.nm_so;
303 struct sockaddr *nam = msg->bind.nm_nam;
304 struct thread *td = msg->bind.nm_td;
308 struct sockaddr_in *sinp;
310 COMMON_START(so, inp, 0);
313 * Must check for multicast addresses and disallow binding
316 sinp = (struct sockaddr_in *)nam;
317 if (sinp->sin_family == AF_INET &&
318 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
319 error = EAFNOSUPPORT;
322 error = in_pcbbind(inp, nam, td);
326 COMMON_END(PRU_BIND);
332 tcp6_usr_bind(netmsg_t msg)
334 struct socket *so = msg->bind.base.nm_so;
335 struct sockaddr *nam = msg->bind.nm_nam;
336 struct thread *td = msg->bind.nm_td;
340 struct sockaddr_in6 *sin6p;
342 COMMON_START(so, inp, 0);
345 * Must check for multicast addresses and disallow binding
348 sin6p = (struct sockaddr_in6 *)nam;
349 if (sin6p->sin6_family == AF_INET6 &&
350 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
351 error = EAFNOSUPPORT;
354 error = in6_pcbbind(inp, nam, td);
357 COMMON_END(PRU_BIND);
361 struct netmsg_inswildcard {
362 struct netmsg_base base;
363 struct inpcb *nm_inp;
367 in_pcbinswildcardhash_handler(netmsg_t msg)
369 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
370 int cpu = mycpuid, nextcpu;
372 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
375 if (nextcpu < ncpus2)
376 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
378 lwkt_replymsg(&nm->base.lmsg, 0);
382 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
384 sosetport(((struct netmsg_base *)msg)->nm_so, port);
388 * Prepare to accept connections.
391 tcp_usr_listen(netmsg_t msg)
393 struct socket *so = msg->listen.base.nm_so;
394 struct thread *td = msg->listen.nm_td;
398 struct netmsg_inswildcard nm;
399 lwkt_port_t port0 = netisr_cpuport(0);
401 COMMON_START(so, inp, 0);
403 if (&curthread->td_msgport != port0) {
404 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
406 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
407 ("already asked to relink"));
409 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
410 msg->listen.nm_flags |= PRUL_RELINK;
412 /* See the related comment in tcp_connect() */
413 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
414 lwkt_forwardmsg(port0, lmsg);
415 /* msg invalid now */
418 KASSERT(so->so_port == port0, ("so_port is not netisr0"));
420 if (msg->listen.nm_flags & PRUL_RELINK) {
421 msg->listen.nm_flags &= ~PRUL_RELINK;
422 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
424 KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
426 if (tp->t_flags & TF_LISTEN)
429 if (inp->inp_lport == 0) {
430 error = in_pcbbind(inp, NULL, td);
435 tp->t_state = TCPS_LISTEN;
436 tp->t_flags |= TF_LISTEN;
437 tp->tt_msg = NULL; /* Catch any invalid timer usage */
441 * Put this inpcb into wildcard hash on other cpus.
443 ASSERT_INP_NOTINHASH(inp);
444 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
445 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
447 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
449 in_pcbinswildcardhash(inp);
450 COMMON_END(PRU_LISTEN);
456 tcp6_usr_listen(netmsg_t msg)
458 struct socket *so = msg->listen.base.nm_so;
459 struct thread *td = msg->listen.nm_td;
463 struct netmsg_inswildcard nm;
465 COMMON_START(so, inp, 0);
467 if (tp->t_flags & TF_LISTEN)
470 if (inp->inp_lport == 0) {
471 error = in6_pcbbind(inp, NULL, td);
476 tp->t_state = TCPS_LISTEN;
477 tp->t_flags |= TF_LISTEN;
478 tp->tt_msg = NULL; /* Catch any invalid timer usage */
482 * Put this inpcb into wildcard hash on other cpus.
484 KKASSERT(so->so_port == netisr_cpuport(0));
486 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
487 ASSERT_INP_NOTINHASH(inp);
489 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
490 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
492 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
494 in_pcbinswildcardhash(inp);
495 COMMON_END(PRU_LISTEN);
500 * Initiate connection to peer.
501 * Create a template for use in transmissions on this connection.
502 * Enter SYN_SENT state, and mark socket as connecting.
503 * Start keep-alive timer, and seed output sequence space.
504 * Send initial segment on connection.
507 tcp_usr_connect(netmsg_t msg)
509 struct socket *so = msg->connect.base.nm_so;
510 struct sockaddr *nam = msg->connect.nm_nam;
511 struct thread *td = msg->connect.nm_td;
515 struct sockaddr_in *sinp;
517 COMMON_START(so, inp, 0);
520 * Must disallow TCP ``connections'' to multicast addresses.
522 sinp = (struct sockaddr_in *)nam;
523 if (sinp->sin_family == AF_INET
524 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
525 error = EAFNOSUPPORT;
529 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
530 error = EAFNOSUPPORT; /* IPv6 only jail */
535 /* msg is invalid now */
538 if (msg->connect.nm_m) {
539 m_freem(msg->connect.nm_m);
540 msg->connect.nm_m = NULL;
542 if (msg->connect.nm_flags & PRUC_HELDTD)
544 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
545 so->so_error = error;
546 soisdisconnected(so);
548 lwkt_replymsg(&msg->lmsg, error);
554 tcp6_usr_connect(netmsg_t msg)
556 struct socket *so = msg->connect.base.nm_so;
557 struct sockaddr *nam = msg->connect.nm_nam;
558 struct thread *td = msg->connect.nm_td;
562 struct sockaddr_in6 *sin6p;
564 COMMON_START(so, inp, 0);
567 * Must disallow TCP ``connections'' to multicast addresses.
569 sin6p = (struct sockaddr_in6 *)nam;
570 if (sin6p->sin6_family == AF_INET6
571 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
572 error = EAFNOSUPPORT;
576 if (!prison_remote_ip(td, nam)) {
577 error = EAFNOSUPPORT; /* IPv4 only jail */
581 /* Reject v4-mapped address */
582 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
583 error = EADDRNOTAVAIL;
587 inp->inp_inc.inc_isipv6 = 1;
589 /* msg is invalid now */
592 if (msg->connect.nm_m) {
593 m_freem(msg->connect.nm_m);
594 msg->connect.nm_m = NULL;
596 lwkt_replymsg(&msg->lmsg, error);
602 * Initiate disconnect from peer.
603 * If connection never passed embryonic stage, just drop;
604 * else if don't need to let data drain, then can just drop anyways,
605 * else have to begin TCP shutdown process: mark socket disconnecting,
606 * drain unread data, state switch to reflect user close, and
607 * send segment (e.g. FIN) to peer. Socket will be really disconnected
608 * when peer sends FIN and acks ours.
610 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
613 tcp_usr_disconnect(netmsg_t msg)
615 struct socket *so = msg->disconnect.base.nm_so;
620 COMMON_START(so, inp, 1);
621 tp = tcp_disconnect(tp);
622 COMMON_END(PRU_DISCONNECT);
626 * Accept a connection. Essentially all the work is
627 * done at higher levels; just return the address
628 * of the peer, storing through addr.
631 tcp_usr_accept(netmsg_t msg)
633 struct socket *so = msg->accept.base.nm_so;
634 struct sockaddr **nam = msg->accept.nm_nam;
637 struct tcpcb *tp = NULL;
641 if (so->so_state & SS_ISDISCONNECTED) {
642 error = ECONNABORTED;
652 in_setpeeraddr(so, nam);
653 COMMON_END(PRU_ACCEPT);
658 tcp6_usr_accept(netmsg_t msg)
660 struct socket *so = msg->accept.base.nm_so;
661 struct sockaddr **nam = msg->accept.nm_nam;
664 struct tcpcb *tp = NULL;
669 if (so->so_state & SS_ISDISCONNECTED) {
670 error = ECONNABORTED;
679 in6_setpeeraddr(so, nam);
680 COMMON_END(PRU_ACCEPT);
685 * Mark the connection as being incapable of further output.
688 tcp_usr_shutdown(netmsg_t msg)
690 struct socket *so = msg->shutdown.base.nm_so;
695 COMMON_START(so, inp, 0);
697 tp = tcp_usrclosed(tp);
699 error = tcp_output(tp);
700 COMMON_END(PRU_SHUTDOWN);
704 * After a receive, possibly send window update to peer.
707 tcp_usr_rcvd(netmsg_t msg)
709 struct socket *so = msg->rcvd.base.nm_so;
710 int error = 0, noreply = 0;
714 COMMON_START(so, inp, 0);
716 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
718 so_async_rcvd_reply(so);
722 COMMON_END1(PRU_RCVD, noreply);
726 * Do a send by putting data in output queue and updating urgent
727 * marker if URG set. Possibly send more data. Unlike the other
728 * pru_*() routines, the mbuf chains are our responsibility. We
729 * must either enqueue them or free them. The other pru_* routines
730 * generally are caller-frees.
733 tcp_usr_send(netmsg_t msg)
735 struct socket *so = msg->send.base.nm_so;
736 int flags = msg->send.nm_flags;
737 struct mbuf *m = msg->send.nm_m;
743 KKASSERT(msg->send.nm_control == NULL);
744 KKASSERT(msg->send.nm_addr == NULL);
745 KKASSERT((flags & PRUS_FREEADDR) == 0);
751 * OOPS! we lost a race, the TCP session got reset after
752 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
753 * network interrupt in the non-critical section of sosend().
756 error = ECONNRESET; /* XXX EPIPE? */
766 * This is no longer necessary, since:
767 * - sosendtcp() has already checked it for us
768 * - It does not work with asynchronized send
772 * Don't let too much OOB data build up
774 if (flags & PRUS_OOB) {
775 if (ssb_space(&so->so_snd) < -512) {
784 * Pump the data into the socket.
787 ssb_appendstream(&so->so_snd, m);
790 if (flags & PRUS_OOB) {
792 * According to RFC961 (Assigned Protocols),
793 * the urgent pointer points to the last octet
794 * of urgent data. We continue, however,
795 * to consider it to indicate the first octet
796 * of data past the urgent section.
797 * Otherwise, snd_up should be one lower.
799 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
800 tp->t_flags |= TF_FORCE;
801 error = tcp_output(tp);
802 tp->t_flags &= ~TF_FORCE;
804 if (flags & PRUS_EOF) {
806 * Close the send side of the connection after
810 tp = tcp_usrclosed(tp);
812 if (tp != NULL && !tcp_output_pending(tp)) {
813 if (flags & PRUS_MORETOCOME)
814 tp->t_flags |= TF_MORETOCOME;
815 error = tcp_output_fair(tp);
816 if (flags & PRUS_MORETOCOME)
817 tp->t_flags &= ~TF_MORETOCOME;
820 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
821 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
822 (flags & PRUS_NOREPLY));
826 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
827 * will sofree() it when we return.
830 tcp_usr_abort(netmsg_t msg)
832 struct socket *so = msg->abort.base.nm_so;
837 COMMON_START(so, inp, 1);
838 tp = tcp_drop(tp, ECONNABORTED);
839 COMMON_END(PRU_ABORT);
843 * Receive out-of-band data.
846 tcp_usr_rcvoob(netmsg_t msg)
848 struct socket *so = msg->rcvoob.base.nm_so;
849 struct mbuf *m = msg->rcvoob.nm_m;
850 int flags = msg->rcvoob.nm_flags;
855 COMMON_START(so, inp, 0);
856 if ((so->so_oobmark == 0 &&
857 (so->so_state & SS_RCVATMARK) == 0) ||
858 so->so_options & SO_OOBINLINE ||
859 tp->t_oobflags & TCPOOB_HADDATA) {
863 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
868 *mtod(m, caddr_t) = tp->t_iobc;
869 if ((flags & MSG_PEEK) == 0)
870 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
871 COMMON_END(PRU_RCVOOB);
875 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
877 in_savefaddr(so, faddr);
882 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
884 in6_savefaddr(so, faddr);
889 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
890 struct thread *td __unused)
892 const struct sockaddr_in *sinp;
894 sinp = (const struct sockaddr_in *)nam;
895 if (sinp->sin_family == AF_INET &&
896 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
903 /* xxx - should be const */
904 struct pr_usrreqs tcp_usrreqs = {
905 .pru_abort = tcp_usr_abort,
906 .pru_accept = tcp_usr_accept,
907 .pru_attach = tcp_usr_attach,
908 .pru_bind = tcp_usr_bind,
909 .pru_connect = tcp_usr_connect,
910 .pru_connect2 = pr_generic_notsupp,
911 .pru_control = in_control_dispatch,
912 .pru_detach = tcp_usr_detach,
913 .pru_disconnect = tcp_usr_disconnect,
914 .pru_listen = tcp_usr_listen,
915 .pru_peeraddr = in_setpeeraddr_dispatch,
916 .pru_rcvd = tcp_usr_rcvd,
917 .pru_rcvoob = tcp_usr_rcvoob,
918 .pru_send = tcp_usr_send,
919 .pru_sense = pru_sense_null,
920 .pru_shutdown = tcp_usr_shutdown,
921 .pru_sockaddr = in_setsockaddr_dispatch,
922 .pru_sosend = sosendtcp,
923 .pru_soreceive = sorecvtcp,
924 .pru_savefaddr = tcp_usr_savefaddr,
925 .pru_preconnect = tcp_usr_preconnect,
926 .pru_preattach = tcp_usr_preattach
930 struct pr_usrreqs tcp6_usrreqs = {
931 .pru_abort = tcp_usr_abort,
932 .pru_accept = tcp6_usr_accept,
933 .pru_attach = tcp_usr_attach,
934 .pru_bind = tcp6_usr_bind,
935 .pru_connect = tcp6_usr_connect,
936 .pru_connect2 = pr_generic_notsupp,
937 .pru_control = in6_control_dispatch,
938 .pru_detach = tcp_usr_detach,
939 .pru_disconnect = tcp_usr_disconnect,
940 .pru_listen = tcp6_usr_listen,
941 .pru_peeraddr = in6_setpeeraddr_dispatch,
942 .pru_rcvd = tcp_usr_rcvd,
943 .pru_rcvoob = tcp_usr_rcvoob,
944 .pru_send = tcp_usr_send,
945 .pru_sense = pru_sense_null,
946 .pru_shutdown = tcp_usr_shutdown,
947 .pru_sockaddr = in6_setsockaddr_dispatch,
948 .pru_sosend = sosendtcp,
949 .pru_soreceive = sorecvtcp,
950 .pru_savefaddr = tcp6_usr_savefaddr
955 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
956 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
958 struct inpcb *inp = tp->t_inpcb, *oinp;
959 struct socket *so = inp->inp_socket;
960 struct route *ro = &inp->inp_route;
962 KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
963 ("pcbinfo mismatch"));
965 oinp = in_pcblookup_hash(inp->inp_pcbinfo,
966 sin->sin_addr, sin->sin_port,
967 (inp->inp_laddr.s_addr != INADDR_ANY ?
968 inp->inp_laddr : if_sin->sin_addr),
969 inp->inp_lport, 0, NULL);
974 if (inp->inp_laddr.s_addr == INADDR_ANY)
975 inp->inp_laddr = if_sin->sin_addr;
976 inp->inp_faddr = sin->sin_addr;
977 inp->inp_fport = sin->sin_port;
978 in_pcbinsconnhash(inp);
981 * We are now on the inpcb's owner CPU, if the cached route was
982 * freed because the rtentry's owner CPU is not the current CPU
983 * (e.g. in tcp_connect()), then we try to reallocate it here with
984 * the hope that a rtentry may be cloned from a RTF_PRCLONING
987 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
989 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
990 ro->ro_dst.sa_family = AF_INET;
991 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
992 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
998 * Now that no more errors can occur, change the protocol processing
999 * port to the current thread (which is the correct thread).
1001 * Create TCP timer message now; we are on the tcpcb's owner
1004 tcp_create_timermsg(tp, &curthread->td_msgport);
1007 * Compute window scaling to request. Use a larger scaling then
1008 * needed for the initial receive buffer in case the receive buffer
1011 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1012 tp->request_r_scale = TCP_MIN_WINSHIFT;
1013 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1014 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1016 tp->request_r_scale++;
1020 tcpstat.tcps_connattempt++;
1021 tp->t_state = TCPS_SYN_SENT;
1022 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1023 tp->iss = tcp_new_isn(tp);
1024 tcp_sendseqinit(tp);
1026 ssb_appendstream(&so->so_snd, m);
1028 if (flags & PRUS_OOB)
1029 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1033 * Close the send side of the connection after
1034 * the data is sent if flagged.
1036 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1038 tp = tcp_usrclosed(tp);
1040 return (tcp_output(tp));
1044 * Common subroutine to open a TCP connection to remote host specified
1045 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1046 * port number if needed. Call in_pcbladdr to do the routing and to choose
1047 * a local host address (interface).
1048 * Initialize connection parameters and enter SYN-SENT state.
1051 tcp_connect(netmsg_t msg)
1053 struct socket *so = msg->connect.base.nm_so;
1054 struct sockaddr *nam = msg->connect.nm_nam;
1055 struct thread *td = msg->connect.nm_td;
1056 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1057 struct sockaddr_in *if_sin = NULL;
1063 COMMON_START(so, inp, 0);
1066 * Reconnect our pcb if we have to
1068 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1069 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1070 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1074 * Bind if we have to
1076 if (inp->inp_lport == 0) {
1077 if (tcp_lport_extension) {
1078 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1080 error = in_pcbladdr(inp, nam, &if_sin, td);
1083 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1085 error = in_pcbbind_remote(inp, nam, td);
1089 msg->connect.nm_flags |= PRUC_HASLADDR;
1091 error = in_pcbbind(inp, NULL, td);
1097 if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1099 * Calculate the correct protocol processing thread. The
1100 * connect operation must run there. Set the forwarding
1101 * port before we forward the message or it will get bounced
1104 error = in_pcbladdr(inp, nam, &if_sin, td);
1108 KKASSERT(inp->inp_socket == so);
1110 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1111 (inp->inp_laddr.s_addr != INADDR_ANY ?
1112 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1115 if (port != &curthread->td_msgport) {
1116 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1119 * in_pcbladdr() may have allocated a route entry for us
1120 * on the current CPU, but we need a route entry on the
1121 * inpcb's owner CPU, so free it here.
1123 in_pcbresetroute(inp);
1126 * We are moving the protocol processing port the socket
1127 * is on, we have to unlink here and re-link on the
1130 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1131 msg->connect.nm_flags |= PRUC_RECONNECT;
1132 msg->connect.base.nm_dispatch = tcp_connect;
1135 * Use message put done receipt to change this socket's
1136 * so_port, i.e. _after_ this message was put onto the
1137 * target netisr's msgport but _before_ the message could
1138 * be pulled from the target netisr's msgport, so that:
1139 * - The upper half (socket code) will not see the new
1140 * msgport before this message reaches the new msgport
1141 * and messages for this socket will be ordered.
1142 * - This message will see the new msgport, when its
1143 * handler is called in the target netisr.
1146 * We MUST use messege put done receipt to change this
1148 * If we changed the so_port in this netisr after the
1149 * lwkt_forwardmsg (so messages for this socket will be
1150 * ordered) and changed the so_port in the target netisr
1151 * at the very beginning of this message's handler, we
1152 * would suffer so_port overwritten race, given this
1153 * message might be forwarded again.
1156 * This mechanism depends on that the netisr's msgport
1157 * is spin msgport (currently it is :).
1159 * If the upper half saw the new msgport before this
1160 * message reached the target netisr's msgport, the
1161 * messages sent from the upper half could reach the new
1162 * msgport before this message, thus there would be
1163 * message reordering. The worst case could be soclose()
1164 * saw the new msgport and the detach message could reach
1165 * the new msgport before this message, i.e. the inpcb
1166 * could have been destroyed when this message was still
1167 * pending on or on its way to the new msgport. Other
1168 * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1169 * since we have unlinked this inpcb from the current
1172 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1173 lwkt_forwardmsg(port, lmsg);
1174 /* msg invalid now */
1176 } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1178 * The original thread is no longer needed; release it.
1181 msg->connect.nm_flags &= ~PRUC_HELDTD;
1183 error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1184 msg->connect.nm_m, sin, if_sin);
1185 msg->connect.nm_m = NULL;
1187 if (msg->connect.nm_m) {
1188 m_freem(msg->connect.nm_m);
1189 msg->connect.nm_m = NULL;
1191 if (msg->connect.nm_flags & PRUC_HELDTD)
1193 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1194 so->so_error = error;
1195 soisdisconnected(so);
1197 lwkt_replymsg(&msg->connect.base.lmsg, error);
1198 /* msg invalid now */
1204 tcp6_connect(netmsg_t msg)
1207 struct socket *so = msg->connect.base.nm_so;
1208 struct sockaddr *nam = msg->connect.nm_nam;
1209 struct thread *td = msg->connect.nm_td;
1211 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1212 struct in6_addr *addr6;
1216 COMMON_START(so, inp, 0);
1219 * Reconnect our pcb if we have to
1221 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1222 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1223 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1227 * Bind if we have to
1229 if (inp->inp_lport == 0) {
1230 error = in6_pcbbind(inp, NULL, td);
1236 * Cannot simply call in_pcbconnect, because there might be an
1237 * earlier incarnation of this same connection still in
1238 * TIME_WAIT state, creating an ADDRINUSE error.
1240 error = in6_pcbladdr(inp, nam, &addr6, td);
1244 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1246 if (port != &curthread->td_msgport) {
1247 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1250 * in_pcbladdr() may have allocated a route entry for us
1251 * on the current CPU, but we need a route entry on the
1252 * inpcb's owner CPU, so free it here.
1254 in_pcbresetroute(inp);
1256 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1257 msg->connect.nm_flags |= PRUC_RECONNECT;
1258 msg->connect.base.nm_dispatch = tcp6_connect;
1260 /* See the related comment in tcp_connect() */
1261 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1262 lwkt_forwardmsg(port, lmsg);
1263 /* msg invalid now */
1266 error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1267 &msg->connect.nm_m, sin6, addr6);
1268 /* nm_m may still be intact */
1270 if (msg->connect.nm_m) {
1271 m_freem(msg->connect.nm_m);
1272 msg->connect.nm_m = NULL;
1274 lwkt_replymsg(&msg->connect.base.lmsg, error);
1275 /* msg invalid now */
1279 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1280 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1282 struct mbuf *m = *mp;
1283 struct inpcb *inp = tp->t_inpcb;
1284 struct socket *so = inp->inp_socket;
1288 * Cannot simply call in_pcbconnect, because there might be an
1289 * earlier incarnation of this same connection still in
1290 * TIME_WAIT state, creating an ADDRINUSE error.
1292 oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1293 &sin6->sin6_addr, sin6->sin6_port,
1294 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1295 addr6 : &inp->in6p_laddr),
1296 inp->inp_lport, 0, NULL);
1298 return (EADDRINUSE);
1300 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1301 inp->in6p_laddr = *addr6;
1302 inp->in6p_faddr = sin6->sin6_addr;
1303 inp->inp_fport = sin6->sin6_port;
1304 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1305 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1306 in_pcbinsconnhash(inp);
1309 * Now that no more errors can occur, change the protocol processing
1310 * port to the current thread (which is the correct thread).
1312 * Create TCP timer message now; we are on the tcpcb's owner
1315 tcp_create_timermsg(tp, &curthread->td_msgport);
1317 /* Compute window scaling to request. */
1318 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1319 tp->request_r_scale = TCP_MIN_WINSHIFT;
1320 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1321 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1322 tp->request_r_scale++;
1326 tcpstat.tcps_connattempt++;
1327 tp->t_state = TCPS_SYN_SENT;
1328 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1329 tp->iss = tcp_new_isn(tp);
1330 tcp_sendseqinit(tp);
1332 ssb_appendstream(&so->so_snd, m);
1334 if (flags & PRUS_OOB)
1335 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1339 * Close the send side of the connection after
1340 * the data is sent if flagged.
1342 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1344 tp = tcp_usrclosed(tp);
1346 return (tcp_output(tp));
1352 * The new sockopt interface makes it possible for us to block in the
1353 * copyin/out step (if we take a page fault). Taking a page fault while
1354 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1355 * both now use TSM, there probably isn't any need for this function to
1356 * run in a critical section any more. This needs more examination.)
1359 tcp_ctloutput(netmsg_t msg)
1361 struct socket *so = msg->base.nm_so;
1362 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1363 int error, opt, optval, opthz;
1373 tp = intotcpcb(inp);
1375 /* Get socket's owner cpuid hint */
1376 if (sopt->sopt_level == SOL_SOCKET &&
1377 sopt->sopt_dir == SOPT_GET &&
1378 sopt->sopt_name == SO_CPUHINT) {
1379 if (tp->t_flags & TF_LISTEN) {
1381 * Listen sockets owner cpuid is always 0,
1382 * which does not make sense if SO_REUSEPORT
1385 if (so->so_options & SO_REUSEPORT)
1386 optval = (inp->inp_lgrpindex & ncpus2_mask);
1388 optval = -1; /* no hint */
1392 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1396 if (sopt->sopt_level != IPPROTO_TCP) {
1397 if (sopt->sopt_level == IPPROTO_IP) {
1398 switch (sopt->sopt_name) {
1399 case IP_MULTICAST_IF:
1400 case IP_MULTICAST_VIF:
1401 case IP_MULTICAST_TTL:
1402 case IP_MULTICAST_LOOP:
1403 case IP_ADD_MEMBERSHIP:
1404 case IP_DROP_MEMBERSHIP:
1406 * Multicast does not make sense on
1414 if (INP_CHECK_SOCKAF(so, AF_INET6))
1415 ip6_ctloutput_dispatch(msg);
1419 /* msg invalid now */
1423 switch (sopt->sopt_dir) {
1425 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1429 switch (sopt->sopt_name) {
1432 tp->t_keepidle = tp->t_keepintvl;
1434 tp->t_keepidle = tcp_keepidle;
1435 tcp_timer_keep_activity(tp, 0);
1437 #ifdef TCP_SIGNATURE
1438 case TCP_SIGNATURE_ENABLE:
1439 if (tp->t_state == TCPS_CLOSED) {
1441 * This is the only safe state that this
1442 * option could be changed. Some segments
1443 * could already have been sent in other
1447 tp->t_flags |= TF_SIGNATURE;
1449 tp->t_flags &= ~TF_SIGNATURE;
1454 #endif /* TCP_SIGNATURE */
1457 switch (sopt->sopt_name) {
1465 opt = 0; /* dead code to fool gcc */
1472 tp->t_flags &= ~opt;
1476 if (tcp_disable_nopush)
1479 tp->t_flags |= TF_NOPUSH;
1481 tp->t_flags &= ~TF_NOPUSH;
1482 error = tcp_output(tp);
1488 * Must be between 0 and maxseg. If the requested
1489 * maxseg is too small to satisfy the desired minmss,
1490 * pump it up (silently so sysctl modifications of
1491 * minmss do not create unexpected program failures).
1492 * Handle degenerate cases.
1494 if (optval > 0 && optval <= tp->t_maxseg) {
1495 if (optval + 40 < tcp_minmss) {
1496 optval = tcp_minmss - 40;
1500 tp->t_maxseg = optval;
1507 opthz = ((int64_t)optval * hz) / 1000;
1509 tp->t_keepinit = opthz;
1515 opthz = ((int64_t)optval * hz) / 1000;
1517 tp->t_keepidle = opthz;
1518 tcp_timer_keep_activity(tp, 0);
1525 opthz = ((int64_t)optval * hz) / 1000;
1527 tp->t_keepintvl = opthz;
1528 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1536 tp->t_keepcnt = optval;
1537 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1544 error = ENOPROTOOPT;
1550 switch (sopt->sopt_name) {
1551 #ifdef TCP_SIGNATURE
1552 case TCP_SIGNATURE_ENABLE:
1553 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1555 #endif /* TCP_SIGNATURE */
1557 optval = tp->t_flags & TF_NODELAY;
1560 optval = tp->t_maxseg;
1563 optval = tp->t_flags & TF_NOOPT;
1566 optval = tp->t_flags & TF_NOPUSH;
1569 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1572 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1575 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1578 optval = tp->t_keepcnt;
1581 error = ENOPROTOOPT;
1585 soopt_from_kbuf(sopt, &optval, sizeof optval);
1589 lwkt_replymsg(&msg->lmsg, error);
1593 * tcp_sendspace and tcp_recvspace are the default send and receive window
1594 * sizes, respectively. These are obsolescent (this information should
1595 * be set by the route).
1597 * Use a default that does not require tcp window scaling to be turned
1598 * on. Individual programs or the administrator can increase the default.
1600 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1601 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1602 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1603 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1604 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1605 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1608 * Attach TCP protocol to socket, allocating internet protocol control
1609 * block, tcp control block, buffer space, and entering CLOSED state.
1612 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1619 boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1623 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1627 /* Post attach; do nothing */
1630 cpu = mycpu->gd_cpuid;
1633 * Set the default pcbinfo. This will likely change when we
1636 error = in_pcballoc(so, &tcbinfo[cpu]);
1642 inp->in6p_hops = -1; /* use kernel default */
1644 tp = tcp_newtcpcb(inp);
1647 * Make sure the socket is destroyed by the pcbdetach.
1656 sofree(so); /* from ref above */
1659 tp->t_state = TCPS_CLOSED;
1660 /* Keep a reference for asynchronized pru_rcvd */
1666 * Initiate (or continue) disconnect.
1667 * If embryonic state, just send reset (once).
1668 * If in ``let data drain'' option and linger null, just drop.
1669 * Otherwise (hard), mark socket disconnecting and drop
1670 * current input data; switch states based on user close, and
1671 * send segment to peer (with FIN).
1673 static struct tcpcb *
1674 tcp_disconnect(struct tcpcb *tp)
1676 struct socket *so = tp->t_inpcb->inp_socket;
1678 if (tp->t_state < TCPS_ESTABLISHED) {
1680 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1681 tp = tcp_drop(tp, 0);
1683 lwkt_gettoken(&so->so_rcv.ssb_token);
1684 soisdisconnecting(so);
1685 sbflush(&so->so_rcv.sb);
1686 tp = tcp_usrclosed(tp);
1689 lwkt_reltoken(&so->so_rcv.ssb_token);
1695 * User issued close, and wish to trail through shutdown states:
1696 * if never received SYN, just forget it. If got a SYN from peer,
1697 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1698 * If already got a FIN from peer, then almost done; go to LAST_ACK
1699 * state. In all other cases, have already sent FIN to peer (e.g.
1700 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1701 * for peer to send FIN or not respond to keep-alives, etc.
1702 * We can let the user exit from the close as soon as the FIN is acked.
1704 static struct tcpcb *
1705 tcp_usrclosed(struct tcpcb *tp)
1708 switch (tp->t_state) {
1712 tp->t_state = TCPS_CLOSED;
1717 case TCPS_SYN_RECEIVED:
1718 tp->t_flags |= TF_NEEDFIN;
1721 case TCPS_ESTABLISHED:
1722 tp->t_state = TCPS_FIN_WAIT_1;
1725 case TCPS_CLOSE_WAIT:
1726 tp->t_state = TCPS_LAST_ACK;
1729 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1730 soisdisconnected(tp->t_inpcb->inp_socket);
1731 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1732 if (tp->t_state == TCPS_FIN_WAIT_2) {
1733 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,