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 * TCP attaches to socket via pru_attach(), reserving space,
176 * and an internet control block. This is likely occuring on
177 * cpu0 and may have to move later when we bind/connect.
180 tcp_usr_attach(netmsg_t msg)
182 struct socket *so = msg->base.nm_so;
183 struct pru_attach_info *ai = msg->attach.nm_ai;
186 struct tcpcb *tp = NULL;
197 error = tcp_attach(so, ai);
201 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
202 so->so_linger = TCP_LINGERTIME;
205 sofree(so); /* from ref above */
206 TCPDEBUG2(PRU_ATTACH);
207 lwkt_replymsg(&msg->lmsg, error);
211 * pru_detach() detaches the TCP protocol from the socket.
212 * If the protocol state is non-embryonic, then can't
213 * do this directly: have to initiate a pru_disconnect(),
214 * which may finish later; embryonic TCB's can just
218 tcp_usr_detach(netmsg_t msg)
220 struct socket *so = msg->base.nm_so;
229 * If the inp is already detached it may have been due to an async
230 * close. Just return as if no error occured.
232 * It's possible for the tcpcb (tp) to disconnect from the inp due
233 * to tcp_drop()->tcp_close() being called. This may occur *after*
234 * the detach message has been queued so we may find a NULL tp here.
237 if ((tp = intotcpcb(inp)) != NULL) {
239 tp = tcp_disconnect(tp);
240 TCPDEBUG2(PRU_DETACH);
243 lwkt_replymsg(&msg->lmsg, error);
247 * NOTE: ignore_error is non-zero for certain disconnection races
248 * which we want to silently allow, otherwise close() may return
249 * an unexpected error.
251 * NOTE: The variables (msg) and (tp) are assumed.
253 #define COMMON_START(so, inp, ignore_error) \
259 error = ignore_error ? 0 : EINVAL; \
263 tp = intotcpcb(inp); \
267 #define COMMON_END1(req, noreply) \
271 lwkt_replymsg(&msg->lmsg, error); \
275 #define COMMON_END(req) COMMON_END1((req), 0)
278 * Give the socket an address.
281 tcp_usr_bind(netmsg_t msg)
283 struct socket *so = msg->bind.base.nm_so;
284 struct sockaddr *nam = msg->bind.nm_nam;
285 struct thread *td = msg->bind.nm_td;
289 struct sockaddr_in *sinp;
291 COMMON_START(so, inp, 0);
294 * Must check for multicast addresses and disallow binding
297 sinp = (struct sockaddr_in *)nam;
298 if (sinp->sin_family == AF_INET &&
299 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
300 error = EAFNOSUPPORT;
303 error = in_pcbbind(inp, nam, td);
306 COMMON_END(PRU_BIND);
313 tcp6_usr_bind(netmsg_t msg)
315 struct socket *so = msg->bind.base.nm_so;
316 struct sockaddr *nam = msg->bind.nm_nam;
317 struct thread *td = msg->bind.nm_td;
321 struct sockaddr_in6 *sin6p;
323 COMMON_START(so, inp, 0);
326 * Must check for multicast addresses and disallow binding
329 sin6p = (struct sockaddr_in6 *)nam;
330 if (sin6p->sin6_family == AF_INET6 &&
331 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
332 error = EAFNOSUPPORT;
335 inp->inp_vflag &= ~INP_IPV4;
336 inp->inp_vflag |= INP_IPV6;
337 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
338 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
339 inp->inp_vflag |= INP_IPV4;
340 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
341 struct sockaddr_in sin;
343 in6_sin6_2_sin(&sin, sin6p);
344 inp->inp_vflag |= INP_IPV4;
345 inp->inp_vflag &= ~INP_IPV6;
346 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
350 error = in6_pcbbind(inp, nam, td);
353 COMMON_END(PRU_BIND);
357 struct netmsg_inswildcard {
358 struct netmsg_base base;
359 struct inpcb *nm_inp;
363 in_pcbinswildcardhash_handler(netmsg_t msg)
365 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
366 int cpu = mycpuid, nextcpu;
368 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
371 if (nextcpu < ncpus2)
372 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
374 lwkt_replymsg(&nm->base.lmsg, 0);
378 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
380 sosetport(((struct netmsg_base *)msg)->nm_so, port);
384 * Prepare to accept connections.
387 tcp_usr_listen(netmsg_t msg)
389 struct socket *so = msg->listen.base.nm_so;
390 struct thread *td = msg->listen.nm_td;
394 struct netmsg_inswildcard nm;
395 lwkt_port_t port0 = netisr_cpuport(0);
397 COMMON_START(so, inp, 0);
399 if (&curthread->td_msgport != port0) {
400 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
402 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
403 ("already asked to relink"));
405 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
406 msg->listen.nm_flags |= PRUL_RELINK;
408 /* See the related comment in tcp_connect() */
409 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
410 lwkt_forwardmsg(port0, lmsg);
411 /* msg invalid now */
414 KASSERT(so->so_port == port0, ("so_port is not netisr0"));
416 if (msg->listen.nm_flags & PRUL_RELINK) {
417 msg->listen.nm_flags &= ~PRUL_RELINK;
418 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
420 KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
422 if (tp->t_flags & TF_LISTEN)
425 if (inp->inp_lport == 0) {
426 error = in_pcbbind(inp, NULL, td);
431 tp->t_state = TCPS_LISTEN;
432 tp->t_flags |= TF_LISTEN;
433 tp->tt_msg = NULL; /* Catch any invalid timer usage */
437 * We have to set the flag because we can't have other cpus
438 * messing with our inp's flags.
440 KASSERT(!(inp->inp_flags & INP_CONNECTED),
441 ("already on connhash"));
442 KASSERT(!(inp->inp_flags & INP_WILDCARD),
443 ("already on wildcardhash"));
444 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
445 ("already on MP wildcardhash"));
446 inp->inp_flags |= INP_WILDCARD_MP;
448 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
449 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
451 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
453 in_pcbinswildcardhash(inp);
454 COMMON_END(PRU_LISTEN);
460 tcp6_usr_listen(netmsg_t msg)
462 struct socket *so = msg->listen.base.nm_so;
463 struct thread *td = msg->listen.nm_td;
467 struct netmsg_inswildcard nm;
469 COMMON_START(so, inp, 0);
471 if (tp->t_flags & TF_LISTEN)
474 if (inp->inp_lport == 0) {
475 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
476 inp->inp_vflag |= INP_IPV4;
478 inp->inp_vflag &= ~INP_IPV4;
479 error = in6_pcbbind(inp, NULL, td);
484 tp->t_state = TCPS_LISTEN;
485 tp->t_flags |= TF_LISTEN;
486 tp->tt_msg = NULL; /* Catch any invalid timer usage */
490 * We have to set the flag because we can't have other cpus
491 * messing with our inp's flags.
493 KASSERT(!(inp->inp_flags & INP_CONNECTED),
494 ("already on connhash"));
495 KASSERT(!(inp->inp_flags & INP_WILDCARD),
496 ("already on wildcardhash"));
497 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
498 ("already on MP wildcardhash"));
499 inp->inp_flags |= INP_WILDCARD_MP;
501 KKASSERT(so->so_port == netisr_cpuport(0));
502 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
503 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
505 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
506 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
508 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
510 in_pcbinswildcardhash(inp);
511 COMMON_END(PRU_LISTEN);
516 * Initiate connection to peer.
517 * Create a template for use in transmissions on this connection.
518 * Enter SYN_SENT state, and mark socket as connecting.
519 * Start keep-alive timer, and seed output sequence space.
520 * Send initial segment on connection.
523 tcp_usr_connect(netmsg_t msg)
525 struct socket *so = msg->connect.base.nm_so;
526 struct sockaddr *nam = msg->connect.nm_nam;
527 struct thread *td = msg->connect.nm_td;
531 struct sockaddr_in *sinp;
533 COMMON_START(so, inp, 0);
536 * Must disallow TCP ``connections'' to multicast addresses.
538 sinp = (struct sockaddr_in *)nam;
539 if (sinp->sin_family == AF_INET
540 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
541 error = EAFNOSUPPORT;
545 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
546 error = EAFNOSUPPORT; /* IPv6 only jail */
551 /* msg is invalid now */
554 if (msg->connect.nm_m) {
555 m_freem(msg->connect.nm_m);
556 msg->connect.nm_m = NULL;
558 if (msg->connect.nm_flags & PRUC_HELDTD)
560 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
561 so->so_error = error;
562 soisdisconnected(so);
564 lwkt_replymsg(&msg->lmsg, error);
570 tcp6_usr_connect(netmsg_t msg)
572 struct socket *so = msg->connect.base.nm_so;
573 struct sockaddr *nam = msg->connect.nm_nam;
574 struct thread *td = msg->connect.nm_td;
578 struct sockaddr_in6 *sin6p;
580 COMMON_START(so, inp, 0);
583 * Must disallow TCP ``connections'' to multicast addresses.
585 sin6p = (struct sockaddr_in6 *)nam;
586 if (sin6p->sin6_family == AF_INET6
587 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
588 error = EAFNOSUPPORT;
592 if (!prison_remote_ip(td, nam)) {
593 error = EAFNOSUPPORT; /* IPv4 only jail */
597 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
598 struct sockaddr_in *sinp;
600 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
604 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
605 in6_sin6_2_sin(sinp, sin6p);
606 inp->inp_vflag |= INP_IPV4;
607 inp->inp_vflag &= ~INP_IPV6;
608 msg->connect.nm_nam = (struct sockaddr *)sinp;
609 msg->connect.nm_flags |= PRUC_NAMALLOC;
611 /* msg is invalid now */
614 inp->inp_vflag &= ~INP_IPV4;
615 inp->inp_vflag |= INP_IPV6;
616 inp->inp_inc.inc_isipv6 = 1;
618 msg->connect.nm_flags |= PRUC_FALLBACK;
620 /* msg is invalid now */
623 if (msg->connect.nm_m) {
624 m_freem(msg->connect.nm_m);
625 msg->connect.nm_m = NULL;
627 lwkt_replymsg(&msg->lmsg, error);
633 * Initiate disconnect from peer.
634 * If connection never passed embryonic stage, just drop;
635 * else if don't need to let data drain, then can just drop anyways,
636 * else have to begin TCP shutdown process: mark socket disconnecting,
637 * drain unread data, state switch to reflect user close, and
638 * send segment (e.g. FIN) to peer. Socket will be really disconnected
639 * when peer sends FIN and acks ours.
641 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
644 tcp_usr_disconnect(netmsg_t msg)
646 struct socket *so = msg->disconnect.base.nm_so;
651 COMMON_START(so, inp, 1);
652 tp = tcp_disconnect(tp);
653 COMMON_END(PRU_DISCONNECT);
657 * Accept a connection. Essentially all the work is
658 * done at higher levels; just return the address
659 * of the peer, storing through addr.
662 tcp_usr_accept(netmsg_t msg)
664 struct socket *so = msg->accept.base.nm_so;
665 struct sockaddr **nam = msg->accept.nm_nam;
668 struct tcpcb *tp = NULL;
672 if (so->so_state & SS_ISDISCONNECTED) {
673 error = ECONNABORTED;
683 in_setpeeraddr(so, nam);
684 COMMON_END(PRU_ACCEPT);
689 tcp6_usr_accept(netmsg_t msg)
691 struct socket *so = msg->accept.base.nm_so;
692 struct sockaddr **nam = msg->accept.nm_nam;
695 struct tcpcb *tp = NULL;
700 if (so->so_state & SS_ISDISCONNECTED) {
701 error = ECONNABORTED;
710 in6_mapped_peeraddr(so, nam);
711 COMMON_END(PRU_ACCEPT);
715 * Mark the connection as being incapable of further output.
718 tcp_usr_shutdown(netmsg_t msg)
720 struct socket *so = msg->shutdown.base.nm_so;
725 COMMON_START(so, inp, 0);
727 tp = tcp_usrclosed(tp);
729 error = tcp_output(tp);
730 COMMON_END(PRU_SHUTDOWN);
734 * After a receive, possibly send window update to peer.
737 tcp_usr_rcvd(netmsg_t msg)
739 struct socket *so = msg->rcvd.base.nm_so;
740 int error = 0, noreply = 0;
744 COMMON_START(so, inp, 0);
746 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
748 so_async_rcvd_reply(so);
752 COMMON_END1(PRU_RCVD, noreply);
756 * Do a send by putting data in output queue and updating urgent
757 * marker if URG set. Possibly send more data. Unlike the other
758 * pru_*() routines, the mbuf chains are our responsibility. We
759 * must either enqueue them or free them. The other pru_* routines
760 * generally are caller-frees.
763 tcp_usr_send(netmsg_t msg)
765 struct socket *so = msg->send.base.nm_so;
766 int flags = msg->send.nm_flags;
767 struct mbuf *m = msg->send.nm_m;
773 KKASSERT(msg->send.nm_control == NULL);
774 KKASSERT(msg->send.nm_addr == NULL);
775 KKASSERT((flags & PRUS_FREEADDR) == 0);
781 * OOPS! we lost a race, the TCP session got reset after
782 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
783 * network interrupt in the non-critical section of sosend().
786 error = ECONNRESET; /* XXX EPIPE? */
796 * This is no longer necessary, since:
797 * - sosendtcp() has already checked it for us
798 * - It does not work with asynchronized send
802 * Don't let too much OOB data build up
804 if (flags & PRUS_OOB) {
805 if (ssb_space(&so->so_snd) < -512) {
814 * Pump the data into the socket.
817 ssb_appendstream(&so->so_snd, m);
820 if (flags & PRUS_OOB) {
822 * According to RFC961 (Assigned Protocols),
823 * the urgent pointer points to the last octet
824 * of urgent data. We continue, however,
825 * to consider it to indicate the first octet
826 * of data past the urgent section.
827 * Otherwise, snd_up should be one lower.
829 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
830 tp->t_flags |= TF_FORCE;
831 error = tcp_output(tp);
832 tp->t_flags &= ~TF_FORCE;
834 if (flags & PRUS_EOF) {
836 * Close the send side of the connection after
840 tp = tcp_usrclosed(tp);
842 if (tp != NULL && !tcp_output_pending(tp)) {
843 if (flags & PRUS_MORETOCOME)
844 tp->t_flags |= TF_MORETOCOME;
845 error = tcp_output_fair(tp);
846 if (flags & PRUS_MORETOCOME)
847 tp->t_flags &= ~TF_MORETOCOME;
850 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
851 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
852 (flags & PRUS_NOREPLY));
856 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
857 * will sofree() it when we return.
860 tcp_usr_abort(netmsg_t msg)
862 struct socket *so = msg->abort.base.nm_so;
867 COMMON_START(so, inp, 1);
868 tp = tcp_drop(tp, ECONNABORTED);
869 COMMON_END(PRU_ABORT);
873 * Receive out-of-band data.
876 tcp_usr_rcvoob(netmsg_t msg)
878 struct socket *so = msg->rcvoob.base.nm_so;
879 struct mbuf *m = msg->rcvoob.nm_m;
880 int flags = msg->rcvoob.nm_flags;
885 COMMON_START(so, inp, 0);
886 if ((so->so_oobmark == 0 &&
887 (so->so_state & SS_RCVATMARK) == 0) ||
888 so->so_options & SO_OOBINLINE ||
889 tp->t_oobflags & TCPOOB_HADDATA) {
893 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
898 *mtod(m, caddr_t) = tp->t_iobc;
899 if ((flags & MSG_PEEK) == 0)
900 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
901 COMMON_END(PRU_RCVOOB);
905 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
907 in_savefaddr(so, faddr);
912 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
914 in6_mapped_savefaddr(so, faddr);
919 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
920 struct thread *td __unused)
922 const struct sockaddr_in *sinp;
924 sinp = (const struct sockaddr_in *)nam;
925 if (sinp->sin_family == AF_INET &&
926 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
933 /* xxx - should be const */
934 struct pr_usrreqs tcp_usrreqs = {
935 .pru_abort = tcp_usr_abort,
936 .pru_accept = tcp_usr_accept,
937 .pru_attach = tcp_usr_attach,
938 .pru_bind = tcp_usr_bind,
939 .pru_connect = tcp_usr_connect,
940 .pru_connect2 = pr_generic_notsupp,
941 .pru_control = in_control_dispatch,
942 .pru_detach = tcp_usr_detach,
943 .pru_disconnect = tcp_usr_disconnect,
944 .pru_listen = tcp_usr_listen,
945 .pru_peeraddr = in_setpeeraddr_dispatch,
946 .pru_rcvd = tcp_usr_rcvd,
947 .pru_rcvoob = tcp_usr_rcvoob,
948 .pru_send = tcp_usr_send,
949 .pru_sense = pru_sense_null,
950 .pru_shutdown = tcp_usr_shutdown,
951 .pru_sockaddr = in_setsockaddr_dispatch,
952 .pru_sosend = sosendtcp,
953 .pru_soreceive = sorecvtcp,
954 .pru_savefaddr = tcp_usr_savefaddr,
955 .pru_preconnect = tcp_usr_preconnect
959 struct pr_usrreqs tcp6_usrreqs = {
960 .pru_abort = tcp_usr_abort,
961 .pru_accept = tcp6_usr_accept,
962 .pru_attach = tcp_usr_attach,
963 .pru_bind = tcp6_usr_bind,
964 .pru_connect = tcp6_usr_connect,
965 .pru_connect2 = pr_generic_notsupp,
966 .pru_control = in6_control_dispatch,
967 .pru_detach = tcp_usr_detach,
968 .pru_disconnect = tcp_usr_disconnect,
969 .pru_listen = tcp6_usr_listen,
970 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
971 .pru_rcvd = tcp_usr_rcvd,
972 .pru_rcvoob = tcp_usr_rcvoob,
973 .pru_send = tcp_usr_send,
974 .pru_sense = pru_sense_null,
975 .pru_shutdown = tcp_usr_shutdown,
976 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
977 .pru_sosend = sosendtcp,
978 .pru_soreceive = sorecvtcp,
979 .pru_savefaddr = tcp6_usr_savefaddr
984 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
985 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
987 struct inpcb *inp = tp->t_inpcb, *oinp;
988 struct socket *so = inp->inp_socket;
989 struct route *ro = &inp->inp_route;
991 KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
992 ("pcbinfo mismatch"));
994 oinp = in_pcblookup_hash(inp->inp_pcbinfo,
995 sin->sin_addr, sin->sin_port,
996 (inp->inp_laddr.s_addr != INADDR_ANY ?
997 inp->inp_laddr : if_sin->sin_addr),
998 inp->inp_lport, 0, NULL);
1001 return (EADDRINUSE);
1003 if (inp->inp_laddr.s_addr == INADDR_ANY)
1004 inp->inp_laddr = if_sin->sin_addr;
1005 inp->inp_faddr = sin->sin_addr;
1006 inp->inp_fport = sin->sin_port;
1007 in_pcbinsconnhash(inp);
1010 * We are now on the inpcb's owner CPU, if the cached route was
1011 * freed because the rtentry's owner CPU is not the current CPU
1012 * (e.g. in tcp_connect()), then we try to reallocate it here with
1013 * the hope that a rtentry may be cloned from a RTF_PRCLONING
1016 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1017 ro->ro_rt == NULL) {
1018 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1019 ro->ro_dst.sa_family = AF_INET;
1020 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1021 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1027 * Now that no more errors can occur, change the protocol processing
1028 * port to the current thread (which is the correct thread).
1030 * Create TCP timer message now; we are on the tcpcb's owner
1033 tcp_create_timermsg(tp, &curthread->td_msgport);
1036 * Compute window scaling to request. Use a larger scaling then
1037 * needed for the initial receive buffer in case the receive buffer
1040 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1041 tp->request_r_scale = TCP_MIN_WINSHIFT;
1042 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1043 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1045 tp->request_r_scale++;
1049 tcpstat.tcps_connattempt++;
1050 tp->t_state = TCPS_SYN_SENT;
1051 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1052 tp->iss = tcp_new_isn(tp);
1053 tcp_sendseqinit(tp);
1055 ssb_appendstream(&so->so_snd, m);
1057 if (flags & PRUS_OOB)
1058 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1062 * Close the send side of the connection after
1063 * the data is sent if flagged.
1065 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1067 tp = tcp_usrclosed(tp);
1069 return (tcp_output(tp));
1073 * Common subroutine to open a TCP connection to remote host specified
1074 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1075 * port number if needed. Call in_pcbladdr to do the routing and to choose
1076 * a local host address (interface).
1077 * Initialize connection parameters and enter SYN-SENT state.
1080 tcp_connect(netmsg_t msg)
1082 struct socket *so = msg->connect.base.nm_so;
1083 struct sockaddr *nam = msg->connect.nm_nam;
1084 struct thread *td = msg->connect.nm_td;
1085 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1086 struct sockaddr_in *if_sin = NULL;
1092 COMMON_START(so, inp, 0);
1095 * Reconnect our pcb if we have to
1097 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1098 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1099 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1103 * Bind if we have to
1105 if (inp->inp_lport == 0) {
1106 if (tcp_lport_extension) {
1107 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1109 error = in_pcbladdr(inp, nam, &if_sin, td);
1112 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1114 error = in_pcbbind_remote(inp, nam, td);
1118 msg->connect.nm_flags |= PRUC_HASLADDR;
1120 error = in_pcbbind(inp, NULL, td);
1126 if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1128 * Calculate the correct protocol processing thread. The
1129 * connect operation must run there. Set the forwarding
1130 * port before we forward the message or it will get bounced
1133 error = in_pcbladdr(inp, nam, &if_sin, td);
1137 KKASSERT(inp->inp_socket == so);
1139 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1140 (inp->inp_laddr.s_addr != INADDR_ANY ?
1141 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1144 if (port != &curthread->td_msgport) {
1145 struct route *ro = &inp->inp_route;
1146 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1149 * in_pcbladdr() may have allocated a route entry for us
1150 * on the current CPU, but we need a route entry on the
1151 * inpcb's owner CPU, so free it here.
1153 if (ro->ro_rt != NULL)
1155 bzero(ro, sizeof(*ro));
1158 * We are moving the protocol processing port the socket
1159 * is on, we have to unlink here and re-link on the
1162 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1163 msg->connect.nm_flags |= PRUC_RECONNECT;
1164 msg->connect.base.nm_dispatch = tcp_connect;
1167 * Use message put done receipt to change this socket's
1168 * so_port, i.e. _after_ this message was put onto the
1169 * target netisr's msgport but _before_ the message could
1170 * be pulled from the target netisr's msgport, so that:
1171 * - The upper half (socket code) will not see the new
1172 * msgport before this message reaches the new msgport
1173 * and messages for this socket will be ordered.
1174 * - This message will see the new msgport, when its
1175 * handler is called in the target netisr.
1178 * We MUST use messege put done receipt to change this
1180 * If we changed the so_port in this netisr after the
1181 * lwkt_forwardmsg (so messages for this socket will be
1182 * ordered) and changed the so_port in the target netisr
1183 * at the very beginning of this message's handler, we
1184 * would suffer so_port overwritten race, given this
1185 * message might be forwarded again.
1188 * This mechanism depends on that the netisr's msgport
1189 * is spin msgport (currently it is :).
1191 * If the upper half saw the new msgport before this
1192 * message reached the target netisr's msgport, the
1193 * messages sent from the upper half could reach the new
1194 * msgport before this message, thus there would be
1195 * message reordering. The worst case could be soclose()
1196 * saw the new msgport and the detach message could reach
1197 * the new msgport before this message, i.e. the inpcb
1198 * could have been destroyed when this message was still
1199 * pending on or on its way to the new msgport. Other
1200 * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1201 * since we have unlinked this inpcb from the current
1204 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1205 lwkt_forwardmsg(port, lmsg);
1206 /* msg invalid now */
1208 } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1210 * The original thread is no longer needed; release it.
1213 msg->connect.nm_flags &= ~PRUC_HELDTD;
1215 error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1216 msg->connect.nm_m, sin, if_sin);
1217 msg->connect.nm_m = NULL;
1219 if (msg->connect.nm_m) {
1220 m_freem(msg->connect.nm_m);
1221 msg->connect.nm_m = NULL;
1223 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1224 kfree(msg->connect.nm_nam, M_LWKTMSG);
1225 msg->connect.nm_nam = NULL;
1227 if (msg->connect.nm_flags & PRUC_HELDTD)
1229 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1230 so->so_error = error;
1231 soisdisconnected(so);
1233 lwkt_replymsg(&msg->connect.base.lmsg, error);
1234 /* msg invalid now */
1240 tcp6_connect(netmsg_t msg)
1243 struct socket *so = msg->connect.base.nm_so;
1244 struct sockaddr *nam = msg->connect.nm_nam;
1245 struct thread *td = msg->connect.nm_td;
1247 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1248 struct in6_addr *addr6;
1252 COMMON_START(so, inp, 0);
1255 * Reconnect our pcb if we have to
1257 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1258 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1259 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1263 * Bind if we have to
1265 if (inp->inp_lport == 0) {
1266 error = in6_pcbbind(inp, NULL, td);
1272 * Cannot simply call in_pcbconnect, because there might be an
1273 * earlier incarnation of this same connection still in
1274 * TIME_WAIT state, creating an ADDRINUSE error.
1276 error = in6_pcbladdr(inp, nam, &addr6, td);
1280 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1282 if (port != &curthread->td_msgport) {
1283 struct route *ro = &inp->inp_route;
1284 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1287 * in_pcbladdr() may have allocated a route entry for us
1288 * on the current CPU, but we need a route entry on the
1289 * inpcb's owner CPU, so free it here.
1291 if (ro->ro_rt != NULL)
1293 bzero(ro, sizeof(*ro));
1295 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1296 msg->connect.nm_flags |= PRUC_RECONNECT;
1297 msg->connect.base.nm_dispatch = tcp6_connect;
1299 /* See the related comment in tcp_connect() */
1300 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1301 lwkt_forwardmsg(port, lmsg);
1302 /* msg invalid now */
1305 error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1306 &msg->connect.nm_m, sin6, addr6);
1307 /* nm_m may still be intact */
1309 if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1311 /* msg invalid now */
1313 if (msg->connect.nm_m) {
1314 m_freem(msg->connect.nm_m);
1315 msg->connect.nm_m = NULL;
1317 if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1318 kfree(msg->connect.nm_nam, M_LWKTMSG);
1319 msg->connect.nm_nam = NULL;
1321 lwkt_replymsg(&msg->connect.base.lmsg, error);
1322 /* msg invalid now */
1327 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1328 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1330 struct mbuf *m = *mp;
1331 struct inpcb *inp = tp->t_inpcb;
1332 struct socket *so = inp->inp_socket;
1336 * Cannot simply call in_pcbconnect, because there might be an
1337 * earlier incarnation of this same connection still in
1338 * TIME_WAIT state, creating an ADDRINUSE error.
1340 oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1341 &sin6->sin6_addr, sin6->sin6_port,
1342 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1343 addr6 : &inp->in6p_laddr),
1344 inp->inp_lport, 0, NULL);
1346 return (EADDRINUSE);
1348 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1349 inp->in6p_laddr = *addr6;
1350 inp->in6p_faddr = sin6->sin6_addr;
1351 inp->inp_fport = sin6->sin6_port;
1352 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1353 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1354 in_pcbinsconnhash(inp);
1357 * Now that no more errors can occur, change the protocol processing
1358 * port to the current thread (which is the correct thread).
1360 * Create TCP timer message now; we are on the tcpcb's owner
1363 tcp_create_timermsg(tp, &curthread->td_msgport);
1365 /* Compute window scaling to request. */
1366 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1367 tp->request_r_scale = TCP_MIN_WINSHIFT;
1368 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1369 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1370 tp->request_r_scale++;
1374 tcpstat.tcps_connattempt++;
1375 tp->t_state = TCPS_SYN_SENT;
1376 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1377 tp->iss = tcp_new_isn(tp);
1378 tcp_sendseqinit(tp);
1380 ssb_appendstream(&so->so_snd, m);
1382 if (flags & PRUS_OOB)
1383 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1387 * Close the send side of the connection after
1388 * the data is sent if flagged.
1390 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1392 tp = tcp_usrclosed(tp);
1394 return (tcp_output(tp));
1400 * The new sockopt interface makes it possible for us to block in the
1401 * copyin/out step (if we take a page fault). Taking a page fault while
1402 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1403 * both now use TSM, there probably isn't any need for this function to
1404 * run in a critical section any more. This needs more examination.)
1407 tcp_ctloutput(netmsg_t msg)
1409 struct socket *so = msg->base.nm_so;
1410 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1411 int error, opt, optval, opthz;
1421 tp = intotcpcb(inp);
1423 /* Get socket's owner cpuid hint */
1424 if (sopt->sopt_level == SOL_SOCKET &&
1425 sopt->sopt_dir == SOPT_GET &&
1426 sopt->sopt_name == SO_CPUHINT) {
1427 if (tp->t_flags & TF_LISTEN) {
1429 * Listen sockets owner cpuid is always 0,
1430 * which does not make sense if SO_REUSEPORT
1433 if (so->so_options & SO_REUSEPORT)
1434 optval = (inp->inp_lgrpindex & ncpus2_mask);
1436 optval = -1; /* no hint */
1440 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1444 if (sopt->sopt_level != IPPROTO_TCP) {
1445 switch (sopt->sopt_name) {
1446 case IP_MULTICAST_IF:
1447 case IP_MULTICAST_VIF:
1448 case IP_MULTICAST_TTL:
1449 case IP_MULTICAST_LOOP:
1450 case IP_ADD_MEMBERSHIP:
1451 case IP_DROP_MEMBERSHIP:
1453 * Multicast does not make sense on TCP sockets.
1459 if (INP_CHECK_SOCKAF(so, AF_INET6))
1460 ip6_ctloutput_dispatch(msg);
1464 /* msg invalid now */
1468 switch (sopt->sopt_dir) {
1470 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1474 switch (sopt->sopt_name) {
1477 tp->t_keepidle = tp->t_keepintvl;
1479 tp->t_keepidle = tcp_keepidle;
1480 tcp_timer_keep_activity(tp, 0);
1482 #ifdef TCP_SIGNATURE
1483 case TCP_SIGNATURE_ENABLE:
1484 if (tp->t_state == TCPS_CLOSED) {
1486 * This is the only safe state that this
1487 * option could be changed. Some segments
1488 * could already have been sent in other
1492 tp->t_flags |= TF_SIGNATURE;
1494 tp->t_flags &= ~TF_SIGNATURE;
1499 #endif /* TCP_SIGNATURE */
1502 switch (sopt->sopt_name) {
1510 opt = 0; /* dead code to fool gcc */
1517 tp->t_flags &= ~opt;
1521 if (tcp_disable_nopush)
1524 tp->t_flags |= TF_NOPUSH;
1526 tp->t_flags &= ~TF_NOPUSH;
1527 error = tcp_output(tp);
1533 * Must be between 0 and maxseg. If the requested
1534 * maxseg is too small to satisfy the desired minmss,
1535 * pump it up (silently so sysctl modifications of
1536 * minmss do not create unexpected program failures).
1537 * Handle degenerate cases.
1539 if (optval > 0 && optval <= tp->t_maxseg) {
1540 if (optval + 40 < tcp_minmss) {
1541 optval = tcp_minmss - 40;
1545 tp->t_maxseg = optval;
1552 opthz = ((int64_t)optval * hz) / 1000;
1554 tp->t_keepinit = opthz;
1560 opthz = ((int64_t)optval * hz) / 1000;
1562 tp->t_keepidle = opthz;
1563 tcp_timer_keep_activity(tp, 0);
1570 opthz = ((int64_t)optval * hz) / 1000;
1572 tp->t_keepintvl = opthz;
1573 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1581 tp->t_keepcnt = optval;
1582 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1589 error = ENOPROTOOPT;
1595 switch (sopt->sopt_name) {
1596 #ifdef TCP_SIGNATURE
1597 case TCP_SIGNATURE_ENABLE:
1598 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1600 #endif /* TCP_SIGNATURE */
1602 optval = tp->t_flags & TF_NODELAY;
1605 optval = tp->t_maxseg;
1608 optval = tp->t_flags & TF_NOOPT;
1611 optval = tp->t_flags & TF_NOPUSH;
1614 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1617 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1620 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1623 optval = tp->t_keepcnt;
1626 error = ENOPROTOOPT;
1630 soopt_from_kbuf(sopt, &optval, sizeof optval);
1634 lwkt_replymsg(&msg->lmsg, error);
1638 * tcp_sendspace and tcp_recvspace are the default send and receive window
1639 * sizes, respectively. These are obsolescent (this information should
1640 * be set by the route).
1642 * Use a default that does not require tcp window scaling to be turned
1643 * on. Individual programs or the administrator can increase the default.
1645 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1646 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1647 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1648 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1649 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1650 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1653 * Attach TCP protocol to socket, allocating internet protocol control
1654 * block, tcp control block, buffer space, and entering CLOSED state.
1657 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1664 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1667 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1668 lwkt_gettoken(&so->so_rcv.ssb_token);
1669 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1671 lwkt_reltoken(&so->so_rcv.ssb_token);
1675 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1676 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
1677 cpu = mycpu->gd_cpuid;
1680 * Set the default port for protocol processing. This will likely
1681 * change when we connect.
1683 error = in_pcballoc(so, &tcbinfo[cpu]);
1689 inp->inp_vflag |= INP_IPV6;
1690 inp->in6p_hops = -1; /* use kernel default */
1694 inp->inp_vflag |= INP_IPV4;
1695 tp = tcp_newtcpcb(inp);
1698 * Make sure the socket is destroyed by the pcbdetach.
1707 sofree(so); /* from ref above */
1710 tp->t_state = TCPS_CLOSED;
1711 /* Keep a reference for asynchronized pru_rcvd */
1717 * Initiate (or continue) disconnect.
1718 * If embryonic state, just send reset (once).
1719 * If in ``let data drain'' option and linger null, just drop.
1720 * Otherwise (hard), mark socket disconnecting and drop
1721 * current input data; switch states based on user close, and
1722 * send segment to peer (with FIN).
1724 static struct tcpcb *
1725 tcp_disconnect(struct tcpcb *tp)
1727 struct socket *so = tp->t_inpcb->inp_socket;
1729 if (tp->t_state < TCPS_ESTABLISHED) {
1731 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1732 tp = tcp_drop(tp, 0);
1734 lwkt_gettoken(&so->so_rcv.ssb_token);
1735 soisdisconnecting(so);
1736 sbflush(&so->so_rcv.sb);
1737 tp = tcp_usrclosed(tp);
1740 lwkt_reltoken(&so->so_rcv.ssb_token);
1746 * User issued close, and wish to trail through shutdown states:
1747 * if never received SYN, just forget it. If got a SYN from peer,
1748 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1749 * If already got a FIN from peer, then almost done; go to LAST_ACK
1750 * state. In all other cases, have already sent FIN to peer (e.g.
1751 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1752 * for peer to send FIN or not respond to keep-alives, etc.
1753 * We can let the user exit from the close as soon as the FIN is acked.
1755 static struct tcpcb *
1756 tcp_usrclosed(struct tcpcb *tp)
1759 switch (tp->t_state) {
1763 tp->t_state = TCPS_CLOSED;
1768 case TCPS_SYN_RECEIVED:
1769 tp->t_flags |= TF_NEEDFIN;
1772 case TCPS_ESTABLISHED:
1773 tp->t_state = TCPS_FIN_WAIT_1;
1776 case TCPS_CLOSE_WAIT:
1777 tp->t_state = TCPS_LAST_ACK;
1780 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1781 soisdisconnected(tp->t_inpcb->inp_socket);
1782 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1783 if (tp->t_state == TCPS_FIN_WAIT_2) {
1784 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,