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. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
67 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
68 * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
71 #include "opt_ipsec.h"
73 #include "opt_inet6.h"
74 #include "opt_tcpdebug.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/malloc.h>
80 #include <sys/sysctl.h>
81 #include <sys/globaldata.h>
82 #include <sys/thread.h>
86 #include <sys/domain.h>
88 #include <sys/socket.h>
89 #include <sys/socketvar.h>
90 #include <sys/protosw.h>
92 #include <sys/thread2.h>
93 #include <sys/msgport2.h>
94 #include <sys/socketvar2.h>
97 #include <net/netisr.h>
98 #include <net/route.h>
100 #include <net/netmsg2.h>
102 #include <netinet/in.h>
103 #include <netinet/in_systm.h>
105 #include <netinet/ip6.h>
107 #include <netinet/in_pcb.h>
109 #include <netinet6/in6_pcb.h>
111 #include <netinet/in_var.h>
112 #include <netinet/ip_var.h>
114 #include <netinet6/ip6_var.h>
115 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcp.h>
118 #include <netinet/tcp_fsm.h>
119 #include <netinet/tcp_seq.h>
120 #include <netinet/tcp_timer.h>
121 #include <netinet/tcp_timer2.h>
122 #include <netinet/tcp_var.h>
123 #include <netinet/tcpip.h>
125 #include <netinet/tcp_debug.h>
129 #include <netinet6/ipsec.h>
133 * TCP protocol interface to socket abstraction.
135 extern char *tcpstates[]; /* XXX ??? */
137 static int tcp_attach (struct socket *, struct pru_attach_info *);
138 static void tcp_connect (netmsg_t msg);
140 static void tcp6_connect (netmsg_t msg);
141 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
143 struct sockaddr_in6 *sin6,
144 struct in6_addr *addr6);
146 static struct tcpcb *
147 tcp_disconnect (struct tcpcb *);
148 static struct tcpcb *
149 tcp_usrclosed (struct tcpcb *);
152 #define TCPDEBUG0 int ostate = 0
153 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
154 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
155 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
159 #define TCPDEBUG2(req)
162 static int tcp_lport_extension = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
165 &tcp_lport_extension, 0, "");
168 * TCP attaches to socket via pru_attach(), reserving space,
169 * and an internet control block. This is likely occuring on
170 * cpu0 and may have to move later when we bind/connect.
173 tcp_usr_attach(netmsg_t msg)
175 struct socket *so = msg->base.nm_so;
176 struct pru_attach_info *ai = msg->attach.nm_ai;
179 struct tcpcb *tp = 0;
190 error = tcp_attach(so, ai);
194 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
195 so->so_linger = TCP_LINGERTIME;
198 sofree(so); /* from ref above */
199 TCPDEBUG2(PRU_ATTACH);
200 lwkt_replymsg(&msg->lmsg, error);
204 * pru_detach() detaches the TCP protocol from the socket.
205 * If the protocol state is non-embryonic, then can't
206 * do this directly: have to initiate a pru_disconnect(),
207 * which may finish later; embryonic TCB's can just
211 tcp_usr_detach(netmsg_t msg)
213 struct socket *so = msg->base.nm_so;
222 * If the inp is already detached it may have been due to an async
223 * close. Just return as if no error occured.
225 * It's possible for the tcpcb (tp) to disconnect from the inp due
226 * to tcp_drop()->tcp_close() being called. This may occur *after*
227 * the detach message has been queued so we may find a NULL tp here.
230 if ((tp = intotcpcb(inp)) != NULL) {
232 tp = tcp_disconnect(tp);
233 TCPDEBUG2(PRU_DETACH);
236 lwkt_replymsg(&msg->lmsg, error);
240 * NOTE: ignore_error is non-zero for certain disconnection races
241 * which we want to silently allow, otherwise close() may return
242 * an unexpected error.
244 * NOTE: The variables (msg) and (tp) are assumed.
246 #define COMMON_START(so, inp, ignore_error) \
252 error = ignore_error ? 0 : EINVAL; \
256 tp = intotcpcb(inp); \
260 #define COMMON_END(req) \
263 lwkt_replymsg(&msg->lmsg, error); \
268 * Give the socket an address.
271 tcp_usr_bind(netmsg_t msg)
273 struct socket *so = msg->bind.base.nm_so;
274 struct sockaddr *nam = msg->bind.nm_nam;
275 struct thread *td = msg->bind.nm_td;
279 struct sockaddr_in *sinp;
281 COMMON_START(so, inp, 0);
284 * Must check for multicast addresses and disallow binding
287 sinp = (struct sockaddr_in *)nam;
288 if (sinp->sin_family == AF_INET &&
289 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
290 error = EAFNOSUPPORT;
293 error = in_pcbbind(inp, nam, td);
296 COMMON_END(PRU_BIND);
303 tcp6_usr_bind(netmsg_t msg)
305 struct socket *so = msg->bind.base.nm_so;
306 struct sockaddr *nam = msg->bind.nm_nam;
307 struct thread *td = msg->bind.nm_td;
311 struct sockaddr_in6 *sin6p;
313 COMMON_START(so, inp, 0);
316 * Must check for multicast addresses and disallow binding
319 sin6p = (struct sockaddr_in6 *)nam;
320 if (sin6p->sin6_family == AF_INET6 &&
321 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
322 error = EAFNOSUPPORT;
325 inp->inp_vflag &= ~INP_IPV4;
326 inp->inp_vflag |= INP_IPV6;
327 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
328 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
329 inp->inp_vflag |= INP_IPV4;
330 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
331 struct sockaddr_in sin;
333 in6_sin6_2_sin(&sin, sin6p);
334 inp->inp_vflag |= INP_IPV4;
335 inp->inp_vflag &= ~INP_IPV6;
336 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
340 error = in6_pcbbind(inp, nam, td);
343 COMMON_END(PRU_BIND);
349 struct netmsg_inswildcard {
350 struct netmsg_base base;
351 struct inpcb *nm_inp;
355 in_pcbinswildcardhash_handler(netmsg_t msg)
357 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
358 int cpu = mycpuid, nextcpu;
360 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
363 if (nextcpu < ncpus2)
364 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
366 lwkt_replymsg(&nm->base.lmsg, 0);
372 * Prepare to accept connections.
375 tcp_usr_listen(netmsg_t msg)
377 struct socket *so = msg->listen.base.nm_so;
378 struct thread *td = msg->listen.nm_td;
383 struct netmsg_inswildcard nm;
386 COMMON_START(so, inp, 0);
388 if (tp->t_flags & TF_LISTEN)
391 if (inp->inp_lport == 0) {
392 error = in_pcbbind(inp, NULL, td);
397 tp->t_state = TCPS_LISTEN;
398 tp->t_flags |= TF_LISTEN;
399 tp->tt_msg = NULL; /* Catch any invalid timer usage */
404 * We have to set the flag because we can't have other cpus
405 * messing with our inp's flags.
407 KASSERT(!(inp->inp_flags & INP_CONNECTED),
408 ("already on connhash\n"));
409 KASSERT(!(inp->inp_flags & INP_WILDCARD),
410 ("already on wildcardhash\n"));
411 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
412 ("already on MP wildcardhash\n"));
413 inp->inp_flags |= INP_WILDCARD_MP;
415 KKASSERT(so->so_port == cpu_portfn(0));
416 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
417 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
419 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
420 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
422 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
425 in_pcbinswildcardhash(inp);
426 COMMON_END(PRU_LISTEN);
432 tcp6_usr_listen(netmsg_t msg)
434 struct socket *so = msg->listen.base.nm_so;
435 struct thread *td = msg->listen.nm_td;
440 struct netmsg_inswildcard nm;
443 COMMON_START(so, inp, 0);
445 if (tp->t_flags & TF_LISTEN)
448 if (inp->inp_lport == 0) {
449 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
450 inp->inp_vflag |= INP_IPV4;
452 inp->inp_vflag &= ~INP_IPV4;
453 error = in6_pcbbind(inp, NULL, td);
458 tp->t_state = TCPS_LISTEN;
459 tp->t_flags |= TF_LISTEN;
460 tp->tt_msg = NULL; /* Catch any invalid timer usage */
465 * We have to set the flag because we can't have other cpus
466 * messing with our inp's flags.
468 KASSERT(!(inp->inp_flags & INP_CONNECTED),
469 ("already on connhash\n"));
470 KASSERT(!(inp->inp_flags & INP_WILDCARD),
471 ("already on wildcardhash\n"));
472 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
473 ("already on MP wildcardhash\n"));
474 inp->inp_flags |= INP_WILDCARD_MP;
476 KKASSERT(so->so_port == cpu_portfn(0));
477 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
478 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
480 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
481 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
483 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
486 in_pcbinswildcardhash(inp);
487 COMMON_END(PRU_LISTEN);
492 * Initiate connection to peer.
493 * Create a template for use in transmissions on this connection.
494 * Enter SYN_SENT state, and mark socket as connecting.
495 * Start keep-alive timer, and seed output sequence space.
496 * Send initial segment on connection.
499 tcp_usr_connect(netmsg_t msg)
501 struct socket *so = msg->connect.base.nm_so;
502 struct sockaddr *nam = msg->connect.nm_nam;
503 struct thread *td = msg->connect.nm_td;
507 struct sockaddr_in *sinp;
509 COMMON_START(so, inp, 0);
512 * Must disallow TCP ``connections'' to multicast addresses.
514 sinp = (struct sockaddr_in *)nam;
515 if (sinp->sin_family == AF_INET
516 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
517 error = EAFNOSUPPORT;
521 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
522 error = EAFNOSUPPORT; /* IPv6 only jail */
527 /* msg is invalid now */
530 if (msg->connect.nm_m) {
531 m_freem(msg->connect.nm_m);
532 msg->connect.nm_m = NULL;
534 lwkt_replymsg(&msg->lmsg, error);
540 tcp6_usr_connect(netmsg_t msg)
542 struct socket *so = msg->connect.base.nm_so;
543 struct sockaddr *nam = msg->connect.nm_nam;
544 struct thread *td = msg->connect.nm_td;
548 struct sockaddr_in6 *sin6p;
550 COMMON_START(so, inp, 0);
553 * Must disallow TCP ``connections'' to multicast addresses.
555 sin6p = (struct sockaddr_in6 *)nam;
556 if (sin6p->sin6_family == AF_INET6
557 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
558 error = EAFNOSUPPORT;
562 if (!prison_remote_ip(td, nam)) {
563 error = EAFNOSUPPORT; /* IPv4 only jail */
567 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
568 struct sockaddr_in *sinp;
570 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
574 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
575 in6_sin6_2_sin(sinp, sin6p);
576 inp->inp_vflag |= INP_IPV4;
577 inp->inp_vflag &= ~INP_IPV6;
578 msg->connect.nm_nam = (struct sockaddr *)sinp;
579 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
581 /* msg is invalid now */
584 inp->inp_vflag &= ~INP_IPV4;
585 inp->inp_vflag |= INP_IPV6;
586 inp->inp_inc.inc_isipv6 = 1;
588 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
590 /* msg is invalid now */
593 if (msg->connect.nm_m) {
594 m_freem(msg->connect.nm_m);
595 msg->connect.nm_m = NULL;
597 lwkt_replymsg(&msg->lmsg, error);
603 * Initiate disconnect from peer.
604 * If connection never passed embryonic stage, just drop;
605 * else if don't need to let data drain, then can just drop anyways,
606 * else have to begin TCP shutdown process: mark socket disconnecting,
607 * drain unread data, state switch to reflect user close, and
608 * send segment (e.g. FIN) to peer. Socket will be really disconnected
609 * when peer sends FIN and acks ours.
611 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
614 tcp_usr_disconnect(netmsg_t msg)
616 struct socket *so = msg->disconnect.base.nm_so;
621 COMMON_START(so, inp, 1);
622 tp = tcp_disconnect(tp);
623 COMMON_END(PRU_DISCONNECT);
627 * Accept a connection. Essentially all the work is
628 * done at higher levels; just return the address
629 * of the peer, storing through addr.
632 tcp_usr_accept(netmsg_t msg)
634 struct socket *so = msg->accept.base.nm_so;
635 struct sockaddr **nam = msg->accept.nm_nam;
638 struct tcpcb *tp = NULL;
642 if (so->so_state & SS_ISDISCONNECTED) {
643 error = ECONNABORTED;
653 in_setpeeraddr(so, nam);
654 COMMON_END(PRU_ACCEPT);
659 tcp6_usr_accept(netmsg_t msg)
661 struct socket *so = msg->accept.base.nm_so;
662 struct sockaddr **nam = msg->accept.nm_nam;
665 struct tcpcb *tp = NULL;
670 if (so->so_state & SS_ISDISCONNECTED) {
671 error = ECONNABORTED;
680 in6_mapped_peeraddr(so, nam);
681 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;
714 COMMON_START(so, inp, 0);
716 COMMON_END(PRU_RCVD);
720 * Do a send by putting data in output queue and updating urgent
721 * marker if URG set. Possibly send more data. Unlike the other
722 * pru_*() routines, the mbuf chains are our responsibility. We
723 * must either enqueue them or free them. The other pru_* routines
724 * generally are caller-frees.
727 tcp_usr_send(netmsg_t msg)
729 struct socket *so = msg->send.base.nm_so;
730 int flags = msg->send.nm_flags;
731 struct mbuf *m = msg->send.nm_m;
732 struct mbuf *control = msg->send.nm_control;
742 * OOPS! we lost a race, the TCP session got reset after
743 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
744 * network interrupt in the non-critical section of sosend().
749 error = ECONNRESET; /* XXX EPIPE? */
757 /* TCP doesn't do control messages (rights, creds, etc) */
758 if (control->m_len) {
764 m_freem(control); /* empty control, just free it */
768 * Don't let too much OOB data build up
770 if (flags & PRUS_OOB) {
771 if (ssb_space(&so->so_snd) < -512) {
779 * Pump the data into the socket.
782 ssb_appendstream(&so->so_snd, m);
783 if (flags & PRUS_OOB) {
785 * According to RFC961 (Assigned Protocols),
786 * the urgent pointer points to the last octet
787 * of urgent data. We continue, however,
788 * to consider it to indicate the first octet
789 * of data past the urgent section.
790 * Otherwise, snd_up should be one lower.
792 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
793 tp->t_flags |= TF_FORCE;
794 error = tcp_output(tp);
795 tp->t_flags &= ~TF_FORCE;
797 if (flags & PRUS_EOF) {
799 * Close the send side of the connection after
803 tp = tcp_usrclosed(tp);
806 if (flags & PRUS_MORETOCOME)
807 tp->t_flags |= TF_MORETOCOME;
808 error = tcp_output(tp);
809 if (flags & PRUS_MORETOCOME)
810 tp->t_flags &= ~TF_MORETOCOME;
813 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
814 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
818 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
819 * will sofree() it when we return.
822 tcp_usr_abort(netmsg_t msg)
824 struct socket *so = msg->abort.base.nm_so;
829 COMMON_START(so, inp, 1);
830 tp = tcp_drop(tp, ECONNABORTED);
831 COMMON_END(PRU_ABORT);
835 * Receive out-of-band data.
838 tcp_usr_rcvoob(netmsg_t msg)
840 struct socket *so = msg->rcvoob.base.nm_so;
841 struct mbuf *m = msg->rcvoob.nm_m;
842 int flags = msg->rcvoob.nm_flags;
847 COMMON_START(so, inp, 0);
848 if ((so->so_oobmark == 0 &&
849 (so->so_state & SS_RCVATMARK) == 0) ||
850 so->so_options & SO_OOBINLINE ||
851 tp->t_oobflags & TCPOOB_HADDATA) {
855 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
860 *mtod(m, caddr_t) = tp->t_iobc;
861 if ((flags & MSG_PEEK) == 0)
862 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
863 COMMON_END(PRU_RCVOOB);
866 /* xxx - should be const */
867 struct pr_usrreqs tcp_usrreqs = {
868 .pru_abort = tcp_usr_abort,
869 .pru_accept = tcp_usr_accept,
870 .pru_attach = tcp_usr_attach,
871 .pru_bind = tcp_usr_bind,
872 .pru_connect = tcp_usr_connect,
873 .pru_connect2 = pr_generic_notsupp,
874 .pru_control = in_control_dispatch,
875 .pru_detach = tcp_usr_detach,
876 .pru_disconnect = tcp_usr_disconnect,
877 .pru_listen = tcp_usr_listen,
878 .pru_peeraddr = in_setpeeraddr_dispatch,
879 .pru_rcvd = tcp_usr_rcvd,
880 .pru_rcvoob = tcp_usr_rcvoob,
881 .pru_send = tcp_usr_send,
882 .pru_sense = pru_sense_null,
883 .pru_shutdown = tcp_usr_shutdown,
884 .pru_sockaddr = in_setsockaddr_dispatch,
885 .pru_sosend = sosend,
886 .pru_soreceive = soreceive
890 struct pr_usrreqs tcp6_usrreqs = {
891 .pru_abort = tcp_usr_abort,
892 .pru_accept = tcp6_usr_accept,
893 .pru_attach = tcp_usr_attach,
894 .pru_bind = tcp6_usr_bind,
895 .pru_connect = tcp6_usr_connect,
896 .pru_connect2 = pr_generic_notsupp,
897 .pru_control = in6_control_dispatch,
898 .pru_detach = tcp_usr_detach,
899 .pru_disconnect = tcp_usr_disconnect,
900 .pru_listen = tcp6_usr_listen,
901 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
902 .pru_rcvd = tcp_usr_rcvd,
903 .pru_rcvoob = tcp_usr_rcvoob,
904 .pru_send = tcp_usr_send,
905 .pru_sense = pru_sense_null,
906 .pru_shutdown = tcp_usr_shutdown,
907 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
908 .pru_sosend = sosend,
909 .pru_soreceive = soreceive
914 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
915 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
917 struct inpcb *inp = tp->t_inpcb, *oinp;
918 struct socket *so = inp->inp_socket;
919 struct route *ro = &inp->inp_route;
921 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
922 sin->sin_addr, sin->sin_port,
923 (inp->inp_laddr.s_addr != INADDR_ANY ?
924 inp->inp_laddr : if_sin->sin_addr),
925 inp->inp_lport, 0, NULL);
930 if (inp->inp_laddr.s_addr == INADDR_ANY)
931 inp->inp_laddr = if_sin->sin_addr;
932 inp->inp_faddr = sin->sin_addr;
933 inp->inp_fport = sin->sin_port;
934 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
935 in_pcbinsconnhash(inp);
938 * We are now on the inpcb's owner CPU, if the cached route was
939 * freed because the rtentry's owner CPU is not the current CPU
940 * (e.g. in tcp_connect()), then we try to reallocate it here with
941 * the hope that a rtentry may be cloned from a RTF_PRCLONING
944 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
946 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
947 ro->ro_dst.sa_family = AF_INET;
948 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
949 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
955 * Now that no more errors can occur, change the protocol processing
956 * port to the current thread (which is the correct thread).
958 * Create TCP timer message now; we are on the tcpcb's owner
961 tcp_create_timermsg(tp, &curthread->td_msgport);
964 * Compute window scaling to request. Use a larger scaling then
965 * needed for the initial receive buffer in case the receive buffer
968 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
969 tp->request_r_scale = TCP_MIN_WINSHIFT;
970 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
971 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
973 tp->request_r_scale++;
977 tcpstat.tcps_connattempt++;
978 tp->t_state = TCPS_SYN_SENT;
979 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
980 tp->iss = tcp_new_isn(tp);
983 ssb_appendstream(&so->so_snd, m);
985 if (flags & PRUS_OOB)
986 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
990 * Close the send side of the connection after
991 * the data is sent if flagged.
993 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
995 tp = tcp_usrclosed(tp);
997 return (tcp_output(tp));
1001 * Common subroutine to open a TCP connection to remote host specified
1002 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1003 * port number if needed. Call in_pcbladdr to do the routing and to choose
1004 * a local host address (interface).
1005 * Initialize connection parameters and enter SYN-SENT state.
1008 tcp_connect(netmsg_t msg)
1010 struct socket *so = msg->connect.base.nm_so;
1011 struct sockaddr *nam = msg->connect.nm_nam;
1012 struct thread *td = msg->connect.nm_td;
1013 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1014 struct sockaddr_in *if_sin;
1017 int error, calc_laddr = 1;
1022 COMMON_START(so, inp, 0);
1025 * Reconnect our pcb if we have to
1027 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1028 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1029 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1033 * Bind if we have to
1035 if (inp->inp_lport == 0) {
1036 if (tcp_lport_extension) {
1037 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1039 error = in_pcbladdr(inp, nam, &if_sin, td);
1042 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1044 error = in_pcbconn_bind(inp, nam, td);
1050 error = in_pcbbind(inp, NULL, td);
1058 * Calculate the correct protocol processing thread. The
1059 * connect operation must run there. Set the forwarding
1060 * port before we forward the message or it will get bounced
1063 error = in_pcbladdr(inp, nam, &if_sin, td);
1067 KKASSERT(inp->inp_socket == so);
1070 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1071 (inp->inp_laddr.s_addr ?
1072 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1075 if (port != &curthread->td_msgport) {
1076 struct route *ro = &inp->inp_route;
1079 * in_pcbladdr() may have allocated a route entry for us
1080 * on the current CPU, but we need a route entry on the
1081 * inpcb's owner CPU, so free it here.
1083 if (ro->ro_rt != NULL)
1085 bzero(ro, sizeof(*ro));
1088 * We are moving the protocol processing port the socket
1089 * is on, we have to unlink here and re-link on the
1092 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1093 sosetport(so, port);
1094 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1095 msg->connect.base.nm_dispatch = tcp_connect;
1097 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1098 /* msg invalid now */
1102 KKASSERT(so->so_port == &curthread->td_msgport);
1104 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1105 msg->connect.nm_m, sin, if_sin);
1106 msg->connect.nm_m = NULL;
1108 if (msg->connect.nm_m) {
1109 m_freem(msg->connect.nm_m);
1110 msg->connect.nm_m = NULL;
1112 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1113 kfree(msg->connect.nm_nam, M_LWKTMSG);
1114 msg->connect.nm_nam = NULL;
1116 lwkt_replymsg(&msg->connect.base.lmsg, error);
1117 /* msg invalid now */
1123 tcp6_connect(netmsg_t msg)
1126 struct socket *so = msg->connect.base.nm_so;
1127 struct sockaddr *nam = msg->connect.nm_nam;
1128 struct thread *td = msg->connect.nm_td;
1130 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1131 struct in6_addr *addr6;
1137 COMMON_START(so, inp, 0);
1140 * Reconnect our pcb if we have to
1142 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1143 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1144 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1148 * Bind if we have to
1150 if (inp->inp_lport == 0) {
1151 error = in6_pcbbind(inp, NULL, td);
1157 * Cannot simply call in_pcbconnect, because there might be an
1158 * earlier incarnation of this same connection still in
1159 * TIME_WAIT state, creating an ADDRINUSE error.
1161 error = in6_pcbladdr(inp, nam, &addr6, td);
1166 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1168 if (port != &curthread->td_msgport) {
1169 struct route *ro = &inp->inp_route;
1172 * in_pcbladdr() may have allocated a route entry for us
1173 * on the current CPU, but we need a route entry on the
1174 * inpcb's owner CPU, so free it here.
1176 if (ro->ro_rt != NULL)
1178 bzero(ro, sizeof(*ro));
1180 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1181 sosetport(so, port);
1182 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1183 msg->connect.base.nm_dispatch = tcp6_connect;
1185 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1186 /* msg invalid now */
1190 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1191 &msg->connect.nm_m, sin6, addr6);
1192 /* nm_m may still be intact */
1194 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1196 /* msg invalid now */
1198 if (msg->connect.nm_m) {
1199 m_freem(msg->connect.nm_m);
1200 msg->connect.nm_m = NULL;
1202 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1203 kfree(msg->connect.nm_nam, M_LWKTMSG);
1204 msg->connect.nm_nam = NULL;
1206 lwkt_replymsg(&msg->connect.base.lmsg, error);
1207 /* msg invalid now */
1212 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1213 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1215 struct mbuf *m = *mp;
1216 struct inpcb *inp = tp->t_inpcb;
1217 struct socket *so = inp->inp_socket;
1221 * Cannot simply call in_pcbconnect, because there might be an
1222 * earlier incarnation of this same connection still in
1223 * TIME_WAIT state, creating an ADDRINUSE error.
1225 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1226 &sin6->sin6_addr, sin6->sin6_port,
1227 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1228 addr6 : &inp->in6p_laddr),
1229 inp->inp_lport, 0, NULL);
1231 return (EADDRINUSE);
1233 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1234 inp->in6p_laddr = *addr6;
1235 inp->in6p_faddr = sin6->sin6_addr;
1236 inp->inp_fport = sin6->sin6_port;
1237 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1238 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1239 in_pcbinsconnhash(inp);
1242 * Now that no more errors can occur, change the protocol processing
1243 * port to the current thread (which is the correct thread).
1245 * Create TCP timer message now; we are on the tcpcb's owner
1248 tcp_create_timermsg(tp, &curthread->td_msgport);
1250 /* Compute window scaling to request. */
1251 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1252 tp->request_r_scale = TCP_MIN_WINSHIFT;
1253 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1254 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1255 tp->request_r_scale++;
1259 tcpstat.tcps_connattempt++;
1260 tp->t_state = TCPS_SYN_SENT;
1261 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1262 tp->iss = tcp_new_isn(tp);
1263 tcp_sendseqinit(tp);
1265 ssb_appendstream(&so->so_snd, m);
1267 if (flags & PRUS_OOB)
1268 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1272 * Close the send side of the connection after
1273 * the data is sent if flagged.
1275 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1277 tp = tcp_usrclosed(tp);
1279 return (tcp_output(tp));
1285 * The new sockopt interface makes it possible for us to block in the
1286 * copyin/out step (if we take a page fault). Taking a page fault while
1287 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1288 * both now use TSM, there probably isn't any need for this function to
1289 * run in a critical section any more. This needs more examination.)
1292 tcp_ctloutput(netmsg_t msg)
1294 struct socket *so = msg->base.nm_so;
1295 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1296 int error, opt, optval, opthz;
1307 if (sopt->sopt_level != IPPROTO_TCP) {
1309 if (INP_CHECK_SOCKAF(so, AF_INET6))
1310 ip6_ctloutput_dispatch(msg);
1314 /* msg invalid now */
1317 tp = intotcpcb(inp);
1319 switch (sopt->sopt_dir) {
1321 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1325 switch (sopt->sopt_name) {
1328 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1329 tp->t_flags |= TF_FASTKEEP;
1330 tcp_timer_keep_activity(tp, 0);
1333 tp->t_flags &= ~TF_FASTKEEP;
1336 #ifdef TCP_SIGNATURE
1337 case TCP_SIGNATURE_ENABLE:
1339 tp->t_flags |= TF_SIGNATURE;
1341 tp->t_flags &= ~TF_SIGNATURE;
1343 #endif /* TCP_SIGNATURE */
1346 switch (sopt->sopt_name) {
1354 opt = 0; /* dead code to fool gcc */
1361 tp->t_flags &= ~opt;
1366 tp->t_flags |= TF_NOPUSH;
1368 tp->t_flags &= ~TF_NOPUSH;
1369 error = tcp_output(tp);
1375 * Must be between 0 and maxseg. If the requested
1376 * maxseg is too small to satisfy the desired minmss,
1377 * pump it up (silently so sysctl modifications of
1378 * minmss do not create unexpected program failures).
1379 * Handle degenerate cases.
1381 if (optval > 0 && optval <= tp->t_maxseg) {
1382 if (optval + 40 < tcp_minmss) {
1383 optval = tcp_minmss - 40;
1387 tp->t_maxseg = optval;
1394 opthz = ((int64_t)optval * hz) / 1000;
1396 tp->t_keepinit = opthz;
1402 opthz = ((int64_t)optval * hz) / 1000;
1404 tp->t_keepidle = opthz;
1410 opthz = ((int64_t)optval * hz) / 1000;
1412 tp->t_keepintvl = opthz;
1413 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1421 tp->t_keepcnt = optval;
1422 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1429 error = ENOPROTOOPT;
1435 switch (sopt->sopt_name) {
1436 #ifdef TCP_SIGNATURE
1437 case TCP_SIGNATURE_ENABLE:
1438 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1440 #endif /* TCP_SIGNATURE */
1442 optval = tp->t_flags & TF_NODELAY;
1445 optval = tp->t_maxseg;
1448 optval = tp->t_flags & TF_NOOPT;
1451 optval = tp->t_flags & TF_NOPUSH;
1454 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1457 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1460 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1463 optval = tp->t_keepcnt;
1466 error = ENOPROTOOPT;
1470 soopt_from_kbuf(sopt, &optval, sizeof optval);
1474 lwkt_replymsg(&msg->lmsg, error);
1478 * tcp_sendspace and tcp_recvspace are the default send and receive window
1479 * sizes, respectively. These are obsolescent (this information should
1480 * be set by the route).
1482 * Use a default that does not require tcp window scaling to be turned
1483 * on. Individual programs or the administrator can increase the default.
1485 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1486 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1487 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1488 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1489 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1490 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1493 * Attach TCP protocol to socket, allocating internet protocol control
1494 * block, tcp control block, bufer space, and entering LISTEN state
1495 * if to accept connections.
1498 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1505 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1508 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1509 lwkt_gettoken(&so->so_rcv.ssb_token);
1510 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1512 lwkt_reltoken(&so->so_rcv.ssb_token);
1516 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1517 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1518 cpu = mycpu->gd_cpuid;
1521 * Set the default port for protocol processing. This will likely
1522 * change when we connect.
1524 error = in_pcballoc(so, &tcbinfo[cpu]);
1530 inp->inp_vflag |= INP_IPV6;
1531 inp->in6p_hops = -1; /* use kernel default */
1535 inp->inp_vflag |= INP_IPV4;
1536 tp = tcp_newtcpcb(inp);
1539 * Make sure the socket is destroyed by the pcbdetach.
1548 sofree(so); /* from ref above */
1551 tp->t_state = TCPS_CLOSED;
1556 * Initiate (or continue) disconnect.
1557 * If embryonic state, just send reset (once).
1558 * If in ``let data drain'' option and linger null, just drop.
1559 * Otherwise (hard), mark socket disconnecting and drop
1560 * current input data; switch states based on user close, and
1561 * send segment to peer (with FIN).
1563 static struct tcpcb *
1564 tcp_disconnect(struct tcpcb *tp)
1566 struct socket *so = tp->t_inpcb->inp_socket;
1568 if (tp->t_state < TCPS_ESTABLISHED) {
1570 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1571 tp = tcp_drop(tp, 0);
1573 lwkt_gettoken(&so->so_rcv.ssb_token);
1574 soisdisconnecting(so);
1575 sbflush(&so->so_rcv.sb);
1576 tp = tcp_usrclosed(tp);
1579 lwkt_reltoken(&so->so_rcv.ssb_token);
1585 * User issued close, and wish to trail through shutdown states:
1586 * if never received SYN, just forget it. If got a SYN from peer,
1587 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1588 * If already got a FIN from peer, then almost done; go to LAST_ACK
1589 * state. In all other cases, have already sent FIN to peer (e.g.
1590 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1591 * for peer to send FIN or not respond to keep-alives, etc.
1592 * We can let the user exit from the close as soon as the FIN is acked.
1594 static struct tcpcb *
1595 tcp_usrclosed(struct tcpcb *tp)
1598 switch (tp->t_state) {
1602 tp->t_state = TCPS_CLOSED;
1607 case TCPS_SYN_RECEIVED:
1608 tp->t_flags |= TF_NEEDFIN;
1611 case TCPS_ESTABLISHED:
1612 tp->t_state = TCPS_FIN_WAIT_1;
1615 case TCPS_CLOSE_WAIT:
1616 tp->t_state = TCPS_LAST_ACK;
1619 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1620 soisdisconnected(tp->t_inpcb->inp_socket);
1621 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1622 if (tp->t_state == TCPS_FIN_WAIT_2) {
1623 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,