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_END1(req, noreply) \
264 lwkt_replymsg(&msg->lmsg, error); \
268 #define COMMON_END(req) COMMON_END1((req), 0)
271 * Give the socket an address.
274 tcp_usr_bind(netmsg_t msg)
276 struct socket *so = msg->bind.base.nm_so;
277 struct sockaddr *nam = msg->bind.nm_nam;
278 struct thread *td = msg->bind.nm_td;
282 struct sockaddr_in *sinp;
284 COMMON_START(so, inp, 0);
287 * Must check for multicast addresses and disallow binding
290 sinp = (struct sockaddr_in *)nam;
291 if (sinp->sin_family == AF_INET &&
292 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
293 error = EAFNOSUPPORT;
296 error = in_pcbbind(inp, nam, td);
299 COMMON_END(PRU_BIND);
306 tcp6_usr_bind(netmsg_t msg)
308 struct socket *so = msg->bind.base.nm_so;
309 struct sockaddr *nam = msg->bind.nm_nam;
310 struct thread *td = msg->bind.nm_td;
314 struct sockaddr_in6 *sin6p;
316 COMMON_START(so, inp, 0);
319 * Must check for multicast addresses and disallow binding
322 sin6p = (struct sockaddr_in6 *)nam;
323 if (sin6p->sin6_family == AF_INET6 &&
324 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
325 error = EAFNOSUPPORT;
328 inp->inp_vflag &= ~INP_IPV4;
329 inp->inp_vflag |= INP_IPV6;
330 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
331 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
332 inp->inp_vflag |= INP_IPV4;
333 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
334 struct sockaddr_in sin;
336 in6_sin6_2_sin(&sin, sin6p);
337 inp->inp_vflag |= INP_IPV4;
338 inp->inp_vflag &= ~INP_IPV6;
339 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
343 error = in6_pcbbind(inp, nam, td);
346 COMMON_END(PRU_BIND);
352 struct netmsg_inswildcard {
353 struct netmsg_base base;
354 struct inpcb *nm_inp;
358 in_pcbinswildcardhash_handler(netmsg_t msg)
360 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
361 int cpu = mycpuid, nextcpu;
363 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
366 if (nextcpu < ncpus2)
367 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
369 lwkt_replymsg(&nm->base.lmsg, 0);
375 * Prepare to accept connections.
378 tcp_usr_listen(netmsg_t msg)
380 struct socket *so = msg->listen.base.nm_so;
381 struct thread *td = msg->listen.nm_td;
386 struct netmsg_inswildcard nm;
389 COMMON_START(so, inp, 0);
391 if (tp->t_flags & TF_LISTEN)
394 if (inp->inp_lport == 0) {
395 error = in_pcbbind(inp, NULL, td);
400 tp->t_state = TCPS_LISTEN;
401 tp->t_flags |= TF_LISTEN;
402 tp->tt_msg = NULL; /* Catch any invalid timer usage */
407 * We have to set the flag because we can't have other cpus
408 * messing with our inp's flags.
410 KASSERT(!(inp->inp_flags & INP_CONNECTED),
411 ("already on connhash\n"));
412 KASSERT(!(inp->inp_flags & INP_WILDCARD),
413 ("already on wildcardhash\n"));
414 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
415 ("already on MP wildcardhash\n"));
416 inp->inp_flags |= INP_WILDCARD_MP;
418 KKASSERT(so->so_port == cpu_portfn(0));
419 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
420 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
422 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
423 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
425 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
428 in_pcbinswildcardhash(inp);
429 COMMON_END(PRU_LISTEN);
435 tcp6_usr_listen(netmsg_t msg)
437 struct socket *so = msg->listen.base.nm_so;
438 struct thread *td = msg->listen.nm_td;
443 struct netmsg_inswildcard nm;
446 COMMON_START(so, inp, 0);
448 if (tp->t_flags & TF_LISTEN)
451 if (inp->inp_lport == 0) {
452 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
453 inp->inp_vflag |= INP_IPV4;
455 inp->inp_vflag &= ~INP_IPV4;
456 error = in6_pcbbind(inp, NULL, td);
461 tp->t_state = TCPS_LISTEN;
462 tp->t_flags |= TF_LISTEN;
463 tp->tt_msg = NULL; /* Catch any invalid timer usage */
468 * We have to set the flag because we can't have other cpus
469 * messing with our inp's flags.
471 KASSERT(!(inp->inp_flags & INP_CONNECTED),
472 ("already on connhash\n"));
473 KASSERT(!(inp->inp_flags & INP_WILDCARD),
474 ("already on wildcardhash\n"));
475 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
476 ("already on MP wildcardhash\n"));
477 inp->inp_flags |= INP_WILDCARD_MP;
479 KKASSERT(so->so_port == cpu_portfn(0));
480 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
481 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
483 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
484 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
486 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
489 in_pcbinswildcardhash(inp);
490 COMMON_END(PRU_LISTEN);
495 * Initiate connection to peer.
496 * Create a template for use in transmissions on this connection.
497 * Enter SYN_SENT state, and mark socket as connecting.
498 * Start keep-alive timer, and seed output sequence space.
499 * Send initial segment on connection.
502 tcp_usr_connect(netmsg_t msg)
504 struct socket *so = msg->connect.base.nm_so;
505 struct sockaddr *nam = msg->connect.nm_nam;
506 struct thread *td = msg->connect.nm_td;
510 struct sockaddr_in *sinp;
512 COMMON_START(so, inp, 0);
515 * Must disallow TCP ``connections'' to multicast addresses.
517 sinp = (struct sockaddr_in *)nam;
518 if (sinp->sin_family == AF_INET
519 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
520 error = EAFNOSUPPORT;
524 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
525 error = EAFNOSUPPORT; /* IPv6 only jail */
530 /* msg is invalid now */
533 if (msg->connect.nm_m) {
534 m_freem(msg->connect.nm_m);
535 msg->connect.nm_m = NULL;
537 lwkt_replymsg(&msg->lmsg, error);
543 tcp6_usr_connect(netmsg_t msg)
545 struct socket *so = msg->connect.base.nm_so;
546 struct sockaddr *nam = msg->connect.nm_nam;
547 struct thread *td = msg->connect.nm_td;
551 struct sockaddr_in6 *sin6p;
553 COMMON_START(so, inp, 0);
556 * Must disallow TCP ``connections'' to multicast addresses.
558 sin6p = (struct sockaddr_in6 *)nam;
559 if (sin6p->sin6_family == AF_INET6
560 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
561 error = EAFNOSUPPORT;
565 if (!prison_remote_ip(td, nam)) {
566 error = EAFNOSUPPORT; /* IPv4 only jail */
570 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
571 struct sockaddr_in *sinp;
573 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
577 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
578 in6_sin6_2_sin(sinp, sin6p);
579 inp->inp_vflag |= INP_IPV4;
580 inp->inp_vflag &= ~INP_IPV6;
581 msg->connect.nm_nam = (struct sockaddr *)sinp;
582 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
584 /* msg is invalid now */
587 inp->inp_vflag &= ~INP_IPV4;
588 inp->inp_vflag |= INP_IPV6;
589 inp->inp_inc.inc_isipv6 = 1;
591 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
593 /* msg is invalid now */
596 if (msg->connect.nm_m) {
597 m_freem(msg->connect.nm_m);
598 msg->connect.nm_m = NULL;
600 lwkt_replymsg(&msg->lmsg, error);
606 * Initiate disconnect from peer.
607 * If connection never passed embryonic stage, just drop;
608 * else if don't need to let data drain, then can just drop anyways,
609 * else have to begin TCP shutdown process: mark socket disconnecting,
610 * drain unread data, state switch to reflect user close, and
611 * send segment (e.g. FIN) to peer. Socket will be really disconnected
612 * when peer sends FIN and acks ours.
614 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
617 tcp_usr_disconnect(netmsg_t msg)
619 struct socket *so = msg->disconnect.base.nm_so;
624 COMMON_START(so, inp, 1);
625 tp = tcp_disconnect(tp);
626 COMMON_END(PRU_DISCONNECT);
630 * Accept a connection. Essentially all the work is
631 * done at higher levels; just return the address
632 * of the peer, storing through addr.
635 tcp_usr_accept(netmsg_t msg)
637 struct socket *so = msg->accept.base.nm_so;
638 struct sockaddr **nam = msg->accept.nm_nam;
641 struct tcpcb *tp = NULL;
645 if (so->so_state & SS_ISDISCONNECTED) {
646 error = ECONNABORTED;
656 in_setpeeraddr(so, nam);
657 COMMON_END(PRU_ACCEPT);
662 tcp6_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;
673 if (so->so_state & SS_ISDISCONNECTED) {
674 error = ECONNABORTED;
683 in6_mapped_peeraddr(so, nam);
684 COMMON_END(PRU_ACCEPT);
688 * Mark the connection as being incapable of further output.
691 tcp_usr_shutdown(netmsg_t msg)
693 struct socket *so = msg->shutdown.base.nm_so;
698 COMMON_START(so, inp, 0);
700 tp = tcp_usrclosed(tp);
702 error = tcp_output(tp);
703 COMMON_END(PRU_SHUTDOWN);
707 * After a receive, possibly send window update to peer.
710 tcp_usr_rcvd(netmsg_t msg)
712 struct socket *so = msg->rcvd.base.nm_so;
717 COMMON_START(so, inp, 0);
719 COMMON_END(PRU_RCVD);
723 * Do a send by putting data in output queue and updating urgent
724 * marker if URG set. Possibly send more data. Unlike the other
725 * pru_*() routines, the mbuf chains are our responsibility. We
726 * must either enqueue them or free them. The other pru_* routines
727 * generally are caller-frees.
730 tcp_usr_send(netmsg_t msg)
732 struct socket *so = msg->send.base.nm_so;
733 int flags = msg->send.nm_flags;
734 struct mbuf *m = msg->send.nm_m;
735 struct mbuf *control = msg->send.nm_control;
741 KKASSERT(control == NULL);
747 * OOPS! we lost a race, the TCP session got reset after
748 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
749 * network interrupt in the non-critical section of sosend().
752 error = ECONNRESET; /* XXX EPIPE? */
761 * Don't let too much OOB data build up
763 if (flags & PRUS_OOB) {
764 if (ssb_space(&so->so_snd) < -512) {
772 * Pump the data into the socket.
775 ssb_appendstream(&so->so_snd, m);
776 if (flags & PRUS_OOB) {
778 * According to RFC961 (Assigned Protocols),
779 * the urgent pointer points to the last octet
780 * of urgent data. We continue, however,
781 * to consider it to indicate the first octet
782 * of data past the urgent section.
783 * Otherwise, snd_up should be one lower.
785 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
786 tp->t_flags |= TF_FORCE;
787 error = tcp_output(tp);
788 tp->t_flags &= ~TF_FORCE;
790 if (flags & PRUS_EOF) {
792 * Close the send side of the connection after
796 tp = tcp_usrclosed(tp);
799 if (flags & PRUS_MORETOCOME)
800 tp->t_flags |= TF_MORETOCOME;
801 error = tcp_output(tp);
802 if (flags & PRUS_MORETOCOME)
803 tp->t_flags &= ~TF_MORETOCOME;
806 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
807 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
808 (flags & PRUS_NOREPLY));
812 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
813 * will sofree() it when we return.
816 tcp_usr_abort(netmsg_t msg)
818 struct socket *so = msg->abort.base.nm_so;
823 COMMON_START(so, inp, 1);
824 tp = tcp_drop(tp, ECONNABORTED);
825 COMMON_END(PRU_ABORT);
829 * Receive out-of-band data.
832 tcp_usr_rcvoob(netmsg_t msg)
834 struct socket *so = msg->rcvoob.base.nm_so;
835 struct mbuf *m = msg->rcvoob.nm_m;
836 int flags = msg->rcvoob.nm_flags;
841 COMMON_START(so, inp, 0);
842 if ((so->so_oobmark == 0 &&
843 (so->so_state & SS_RCVATMARK) == 0) ||
844 so->so_options & SO_OOBINLINE ||
845 tp->t_oobflags & TCPOOB_HADDATA) {
849 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
854 *mtod(m, caddr_t) = tp->t_iobc;
855 if ((flags & MSG_PEEK) == 0)
856 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
857 COMMON_END(PRU_RCVOOB);
860 /* xxx - should be const */
861 struct pr_usrreqs tcp_usrreqs = {
862 .pru_abort = tcp_usr_abort,
863 .pru_accept = tcp_usr_accept,
864 .pru_attach = tcp_usr_attach,
865 .pru_bind = tcp_usr_bind,
866 .pru_connect = tcp_usr_connect,
867 .pru_connect2 = pr_generic_notsupp,
868 .pru_control = in_control_dispatch,
869 .pru_detach = tcp_usr_detach,
870 .pru_disconnect = tcp_usr_disconnect,
871 .pru_listen = tcp_usr_listen,
872 .pru_peeraddr = in_setpeeraddr_dispatch,
873 .pru_rcvd = tcp_usr_rcvd,
874 .pru_rcvoob = tcp_usr_rcvoob,
875 .pru_send = tcp_usr_send,
876 .pru_sense = pru_sense_null,
877 .pru_shutdown = tcp_usr_shutdown,
878 .pru_sockaddr = in_setsockaddr_dispatch,
879 .pru_sosend = sosendtcp,
880 .pru_soreceive = soreceive
884 struct pr_usrreqs tcp6_usrreqs = {
885 .pru_abort = tcp_usr_abort,
886 .pru_accept = tcp6_usr_accept,
887 .pru_attach = tcp_usr_attach,
888 .pru_bind = tcp6_usr_bind,
889 .pru_connect = tcp6_usr_connect,
890 .pru_connect2 = pr_generic_notsupp,
891 .pru_control = in6_control_dispatch,
892 .pru_detach = tcp_usr_detach,
893 .pru_disconnect = tcp_usr_disconnect,
894 .pru_listen = tcp6_usr_listen,
895 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
896 .pru_rcvd = tcp_usr_rcvd,
897 .pru_rcvoob = tcp_usr_rcvoob,
898 .pru_send = tcp_usr_send,
899 .pru_sense = pru_sense_null,
900 .pru_shutdown = tcp_usr_shutdown,
901 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
902 .pru_sosend = sosendtcp,
903 .pru_soreceive = soreceive
908 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
909 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
911 struct inpcb *inp = tp->t_inpcb, *oinp;
912 struct socket *so = inp->inp_socket;
913 struct route *ro = &inp->inp_route;
915 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
916 sin->sin_addr, sin->sin_port,
917 (inp->inp_laddr.s_addr != INADDR_ANY ?
918 inp->inp_laddr : if_sin->sin_addr),
919 inp->inp_lport, 0, NULL);
924 if (inp->inp_laddr.s_addr == INADDR_ANY)
925 inp->inp_laddr = if_sin->sin_addr;
926 inp->inp_faddr = sin->sin_addr;
927 inp->inp_fport = sin->sin_port;
928 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
929 in_pcbinsconnhash(inp);
932 * We are now on the inpcb's owner CPU, if the cached route was
933 * freed because the rtentry's owner CPU is not the current CPU
934 * (e.g. in tcp_connect()), then we try to reallocate it here with
935 * the hope that a rtentry may be cloned from a RTF_PRCLONING
938 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
940 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
941 ro->ro_dst.sa_family = AF_INET;
942 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
943 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
949 * Now that no more errors can occur, change the protocol processing
950 * port to the current thread (which is the correct thread).
952 * Create TCP timer message now; we are on the tcpcb's owner
955 tcp_create_timermsg(tp, &curthread->td_msgport);
958 * Compute window scaling to request. Use a larger scaling then
959 * needed for the initial receive buffer in case the receive buffer
962 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
963 tp->request_r_scale = TCP_MIN_WINSHIFT;
964 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
965 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
967 tp->request_r_scale++;
971 tcpstat.tcps_connattempt++;
972 tp->t_state = TCPS_SYN_SENT;
973 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
974 tp->iss = tcp_new_isn(tp);
977 ssb_appendstream(&so->so_snd, m);
979 if (flags & PRUS_OOB)
980 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
984 * Close the send side of the connection after
985 * the data is sent if flagged.
987 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
989 tp = tcp_usrclosed(tp);
991 return (tcp_output(tp));
995 * Common subroutine to open a TCP connection to remote host specified
996 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
997 * port number if needed. Call in_pcbladdr to do the routing and to choose
998 * a local host address (interface).
999 * Initialize connection parameters and enter SYN-SENT state.
1002 tcp_connect(netmsg_t msg)
1004 struct socket *so = msg->connect.base.nm_so;
1005 struct sockaddr *nam = msg->connect.nm_nam;
1006 struct thread *td = msg->connect.nm_td;
1007 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1008 struct sockaddr_in *if_sin;
1011 int error, calc_laddr = 1;
1016 COMMON_START(so, inp, 0);
1019 * Reconnect our pcb if we have to
1021 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1022 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1023 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1027 * Bind if we have to
1029 if (inp->inp_lport == 0) {
1030 if (tcp_lport_extension) {
1031 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1033 error = in_pcbladdr(inp, nam, &if_sin, td);
1036 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1038 error = in_pcbconn_bind(inp, nam, td);
1044 error = in_pcbbind(inp, NULL, td);
1052 * Calculate the correct protocol processing thread. The
1053 * connect operation must run there. Set the forwarding
1054 * port before we forward the message or it will get bounced
1057 error = in_pcbladdr(inp, nam, &if_sin, td);
1061 KKASSERT(inp->inp_socket == so);
1064 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1065 (inp->inp_laddr.s_addr ?
1066 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1069 if (port != &curthread->td_msgport) {
1070 struct route *ro = &inp->inp_route;
1073 * in_pcbladdr() may have allocated a route entry for us
1074 * on the current CPU, but we need a route entry on the
1075 * inpcb's owner CPU, so free it here.
1077 if (ro->ro_rt != NULL)
1079 bzero(ro, sizeof(*ro));
1082 * We are moving the protocol processing port the socket
1083 * is on, we have to unlink here and re-link on the
1086 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1087 sosetport(so, port);
1088 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1089 msg->connect.base.nm_dispatch = tcp_connect;
1091 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1092 /* msg invalid now */
1096 KKASSERT(so->so_port == &curthread->td_msgport);
1098 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1099 msg->connect.nm_m, sin, if_sin);
1100 msg->connect.nm_m = NULL;
1102 if (msg->connect.nm_m) {
1103 m_freem(msg->connect.nm_m);
1104 msg->connect.nm_m = NULL;
1106 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1107 kfree(msg->connect.nm_nam, M_LWKTMSG);
1108 msg->connect.nm_nam = NULL;
1110 lwkt_replymsg(&msg->connect.base.lmsg, error);
1111 /* msg invalid now */
1117 tcp6_connect(netmsg_t msg)
1120 struct socket *so = msg->connect.base.nm_so;
1121 struct sockaddr *nam = msg->connect.nm_nam;
1122 struct thread *td = msg->connect.nm_td;
1124 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1125 struct in6_addr *addr6;
1131 COMMON_START(so, inp, 0);
1134 * Reconnect our pcb if we have to
1136 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1137 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1138 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1142 * Bind if we have to
1144 if (inp->inp_lport == 0) {
1145 error = in6_pcbbind(inp, NULL, td);
1151 * Cannot simply call in_pcbconnect, because there might be an
1152 * earlier incarnation of this same connection still in
1153 * TIME_WAIT state, creating an ADDRINUSE error.
1155 error = in6_pcbladdr(inp, nam, &addr6, td);
1160 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1162 if (port != &curthread->td_msgport) {
1163 struct route *ro = &inp->inp_route;
1166 * in_pcbladdr() may have allocated a route entry for us
1167 * on the current CPU, but we need a route entry on the
1168 * inpcb's owner CPU, so free it here.
1170 if (ro->ro_rt != NULL)
1172 bzero(ro, sizeof(*ro));
1174 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1175 sosetport(so, port);
1176 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1177 msg->connect.base.nm_dispatch = tcp6_connect;
1179 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1180 /* msg invalid now */
1184 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1185 &msg->connect.nm_m, sin6, addr6);
1186 /* nm_m may still be intact */
1188 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1190 /* msg invalid now */
1192 if (msg->connect.nm_m) {
1193 m_freem(msg->connect.nm_m);
1194 msg->connect.nm_m = NULL;
1196 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1197 kfree(msg->connect.nm_nam, M_LWKTMSG);
1198 msg->connect.nm_nam = NULL;
1200 lwkt_replymsg(&msg->connect.base.lmsg, error);
1201 /* msg invalid now */
1206 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1207 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1209 struct mbuf *m = *mp;
1210 struct inpcb *inp = tp->t_inpcb;
1211 struct socket *so = inp->inp_socket;
1215 * Cannot simply call in_pcbconnect, because there might be an
1216 * earlier incarnation of this same connection still in
1217 * TIME_WAIT state, creating an ADDRINUSE error.
1219 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1220 &sin6->sin6_addr, sin6->sin6_port,
1221 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1222 addr6 : &inp->in6p_laddr),
1223 inp->inp_lport, 0, NULL);
1225 return (EADDRINUSE);
1227 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1228 inp->in6p_laddr = *addr6;
1229 inp->in6p_faddr = sin6->sin6_addr;
1230 inp->inp_fport = sin6->sin6_port;
1231 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1232 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1233 in_pcbinsconnhash(inp);
1236 * Now that no more errors can occur, change the protocol processing
1237 * port to the current thread (which is the correct thread).
1239 * Create TCP timer message now; we are on the tcpcb's owner
1242 tcp_create_timermsg(tp, &curthread->td_msgport);
1244 /* Compute window scaling to request. */
1245 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1246 tp->request_r_scale = TCP_MIN_WINSHIFT;
1247 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1248 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1249 tp->request_r_scale++;
1253 tcpstat.tcps_connattempt++;
1254 tp->t_state = TCPS_SYN_SENT;
1255 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1256 tp->iss = tcp_new_isn(tp);
1257 tcp_sendseqinit(tp);
1259 ssb_appendstream(&so->so_snd, m);
1261 if (flags & PRUS_OOB)
1262 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1266 * Close the send side of the connection after
1267 * the data is sent if flagged.
1269 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1271 tp = tcp_usrclosed(tp);
1273 return (tcp_output(tp));
1279 * The new sockopt interface makes it possible for us to block in the
1280 * copyin/out step (if we take a page fault). Taking a page fault while
1281 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1282 * both now use TSM, there probably isn't any need for this function to
1283 * run in a critical section any more. This needs more examination.)
1286 tcp_ctloutput(netmsg_t msg)
1288 struct socket *so = msg->base.nm_so;
1289 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1290 int error, opt, optval, opthz;
1301 if (sopt->sopt_level != IPPROTO_TCP) {
1303 if (INP_CHECK_SOCKAF(so, AF_INET6))
1304 ip6_ctloutput_dispatch(msg);
1308 /* msg invalid now */
1311 tp = intotcpcb(inp);
1313 switch (sopt->sopt_dir) {
1315 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1319 switch (sopt->sopt_name) {
1322 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1323 tp->t_flags |= TF_FASTKEEP;
1324 tcp_timer_keep_activity(tp, 0);
1327 tp->t_flags &= ~TF_FASTKEEP;
1330 #ifdef TCP_SIGNATURE
1331 case TCP_SIGNATURE_ENABLE:
1333 tp->t_flags |= TF_SIGNATURE;
1335 tp->t_flags &= ~TF_SIGNATURE;
1337 #endif /* TCP_SIGNATURE */
1340 switch (sopt->sopt_name) {
1348 opt = 0; /* dead code to fool gcc */
1355 tp->t_flags &= ~opt;
1360 tp->t_flags |= TF_NOPUSH;
1362 tp->t_flags &= ~TF_NOPUSH;
1363 error = tcp_output(tp);
1369 * Must be between 0 and maxseg. If the requested
1370 * maxseg is too small to satisfy the desired minmss,
1371 * pump it up (silently so sysctl modifications of
1372 * minmss do not create unexpected program failures).
1373 * Handle degenerate cases.
1375 if (optval > 0 && optval <= tp->t_maxseg) {
1376 if (optval + 40 < tcp_minmss) {
1377 optval = tcp_minmss - 40;
1381 tp->t_maxseg = optval;
1388 opthz = ((int64_t)optval * hz) / 1000;
1390 tp->t_keepinit = opthz;
1396 opthz = ((int64_t)optval * hz) / 1000;
1398 tp->t_keepidle = opthz;
1404 opthz = ((int64_t)optval * hz) / 1000;
1406 tp->t_keepintvl = opthz;
1407 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1415 tp->t_keepcnt = optval;
1416 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1423 error = ENOPROTOOPT;
1429 switch (sopt->sopt_name) {
1430 #ifdef TCP_SIGNATURE
1431 case TCP_SIGNATURE_ENABLE:
1432 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1434 #endif /* TCP_SIGNATURE */
1436 optval = tp->t_flags & TF_NODELAY;
1439 optval = tp->t_maxseg;
1442 optval = tp->t_flags & TF_NOOPT;
1445 optval = tp->t_flags & TF_NOPUSH;
1448 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1451 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1454 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1457 optval = tp->t_keepcnt;
1460 error = ENOPROTOOPT;
1464 soopt_from_kbuf(sopt, &optval, sizeof optval);
1468 lwkt_replymsg(&msg->lmsg, error);
1472 * tcp_sendspace and tcp_recvspace are the default send and receive window
1473 * sizes, respectively. These are obsolescent (this information should
1474 * be set by the route).
1476 * Use a default that does not require tcp window scaling to be turned
1477 * on. Individual programs or the administrator can increase the default.
1479 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1480 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1481 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1482 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1483 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1484 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1487 * Attach TCP protocol to socket, allocating internet protocol control
1488 * block, tcp control block, bufer space, and entering LISTEN state
1489 * if to accept connections.
1492 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1499 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1502 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1503 lwkt_gettoken(&so->so_rcv.ssb_token);
1504 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1506 lwkt_reltoken(&so->so_rcv.ssb_token);
1510 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1511 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1512 cpu = mycpu->gd_cpuid;
1515 * Set the default port for protocol processing. This will likely
1516 * change when we connect.
1518 error = in_pcballoc(so, &tcbinfo[cpu]);
1524 inp->inp_vflag |= INP_IPV6;
1525 inp->in6p_hops = -1; /* use kernel default */
1529 inp->inp_vflag |= INP_IPV4;
1530 tp = tcp_newtcpcb(inp);
1533 * Make sure the socket is destroyed by the pcbdetach.
1542 sofree(so); /* from ref above */
1545 tp->t_state = TCPS_CLOSED;
1550 * Initiate (or continue) disconnect.
1551 * If embryonic state, just send reset (once).
1552 * If in ``let data drain'' option and linger null, just drop.
1553 * Otherwise (hard), mark socket disconnecting and drop
1554 * current input data; switch states based on user close, and
1555 * send segment to peer (with FIN).
1557 static struct tcpcb *
1558 tcp_disconnect(struct tcpcb *tp)
1560 struct socket *so = tp->t_inpcb->inp_socket;
1562 if (tp->t_state < TCPS_ESTABLISHED) {
1564 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1565 tp = tcp_drop(tp, 0);
1567 lwkt_gettoken(&so->so_rcv.ssb_token);
1568 soisdisconnecting(so);
1569 sbflush(&so->so_rcv.sb);
1570 tp = tcp_usrclosed(tp);
1573 lwkt_reltoken(&so->so_rcv.ssb_token);
1579 * User issued close, and wish to trail through shutdown states:
1580 * if never received SYN, just forget it. If got a SYN from peer,
1581 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1582 * If already got a FIN from peer, then almost done; go to LAST_ACK
1583 * state. In all other cases, have already sent FIN to peer (e.g.
1584 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1585 * for peer to send FIN or not respond to keep-alives, etc.
1586 * We can let the user exit from the close as soon as the FIN is acked.
1588 static struct tcpcb *
1589 tcp_usrclosed(struct tcpcb *tp)
1592 switch (tp->t_state) {
1596 tp->t_state = TCPS_CLOSED;
1601 case TCPS_SYN_RECEIVED:
1602 tp->t_flags |= TF_NEEDFIN;
1605 case TCPS_ESTABLISHED:
1606 tp->t_state = TCPS_FIN_WAIT_1;
1609 case TCPS_CLOSE_WAIT:
1610 tp->t_state = TCPS_LAST_ACK;
1613 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1614 soisdisconnected(tp->t_inpcb->inp_socket);
1615 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1616 if (tp->t_state == TCPS_FIN_WAIT_2) {
1617 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,