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 $
70 #include "opt_ipsec.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/sysctl.h>
80 #include <sys/globaldata.h>
81 #include <sys/thread.h>
85 #include <sys/domain.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/protosw.h>
91 #include <sys/thread2.h>
92 #include <sys/msgport2.h>
93 #include <sys/socketvar2.h>
96 #include <net/netisr.h>
97 #include <net/route.h>
99 #include <net/netmsg2.h>
101 #include <netinet/in.h>
102 #include <netinet/in_systm.h>
104 #include <netinet/ip6.h>
106 #include <netinet/in_pcb.h>
108 #include <netinet6/in6_pcb.h>
110 #include <netinet/in_var.h>
111 #include <netinet/ip_var.h>
113 #include <netinet6/ip6_var.h>
114 #include <netinet6/tcp6_var.h>
116 #include <netinet/tcp.h>
117 #include <netinet/tcp_fsm.h>
118 #include <netinet/tcp_seq.h>
119 #include <netinet/tcp_timer.h>
120 #include <netinet/tcp_timer2.h>
121 #include <netinet/tcp_var.h>
122 #include <netinet/tcpip.h>
124 #include <netinet/tcp_debug.h>
128 #include <netinet6/ipsec.h>
132 * TCP protocol interface to socket abstraction.
134 extern char *tcpstates[]; /* XXX ??? */
136 static int tcp_attach (struct socket *, struct pru_attach_info *);
137 static void tcp_connect (netmsg_t msg);
139 static void tcp6_connect (netmsg_t msg);
140 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
142 struct sockaddr_in6 *sin6,
143 struct in6_addr *addr6);
145 static struct tcpcb *
146 tcp_disconnect (struct tcpcb *);
147 static struct tcpcb *
148 tcp_usrclosed (struct tcpcb *);
151 #define TCPDEBUG0 int ostate = 0
152 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
153 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
154 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
158 #define TCPDEBUG2(req)
161 static int tcp_lport_extension = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
164 &tcp_lport_extension, 0, "");
167 * TCP attaches to socket via pru_attach(), reserving space,
168 * and an internet control block. This is likely occuring on
169 * cpu0 and may have to move later when we bind/connect.
172 tcp_usr_attach(netmsg_t msg)
174 struct socket *so = msg->base.nm_so;
175 struct pru_attach_info *ai = msg->attach.nm_ai;
178 struct tcpcb *tp = 0;
189 error = tcp_attach(so, ai);
193 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
194 so->so_linger = TCP_LINGERTIME;
197 sofree(so); /* from ref above */
198 TCPDEBUG2(PRU_ATTACH);
199 lwkt_replymsg(&msg->lmsg, error);
203 * pru_detach() detaches the TCP protocol from the socket.
204 * If the protocol state is non-embryonic, then can't
205 * do this directly: have to initiate a pru_disconnect(),
206 * which may finish later; embryonic TCB's can just
210 tcp_usr_detach(netmsg_t msg)
212 struct socket *so = msg->base.nm_so;
221 * If the inp is already detached it may have been due to an async
222 * close. Just return as if no error occured.
224 * It's possible for the tcpcb (tp) to disconnect from the inp due
225 * to tcp_drop()->tcp_close() being called. This may occur *after*
226 * the detach message has been queued so we may find a NULL tp here.
229 if ((tp = intotcpcb(inp)) != NULL) {
231 tp = tcp_disconnect(tp);
232 TCPDEBUG2(PRU_DETACH);
235 lwkt_replymsg(&msg->lmsg, error);
239 * NOTE: ignore_error is non-zero for certain disconnection races
240 * which we want to silently allow, otherwise close() may return
241 * an unexpected error.
243 * NOTE: The variables (msg) and (tp) are assumed.
245 #define COMMON_START(so, inp, ignore_error) \
251 error = ignore_error ? 0 : EINVAL; \
255 tp = intotcpcb(inp); \
259 #define COMMON_END1(req, noreply) \
263 lwkt_replymsg(&msg->lmsg, error); \
267 #define COMMON_END(req) COMMON_END1((req), 0)
270 * Give the socket an address.
273 tcp_usr_bind(netmsg_t msg)
275 struct socket *so = msg->bind.base.nm_so;
276 struct sockaddr *nam = msg->bind.nm_nam;
277 struct thread *td = msg->bind.nm_td;
281 struct sockaddr_in *sinp;
283 COMMON_START(so, inp, 0);
286 * Must check for multicast addresses and disallow binding
289 sinp = (struct sockaddr_in *)nam;
290 if (sinp->sin_family == AF_INET &&
291 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
292 error = EAFNOSUPPORT;
295 error = in_pcbbind(inp, nam, td);
298 COMMON_END(PRU_BIND);
305 tcp6_usr_bind(netmsg_t msg)
307 struct socket *so = msg->bind.base.nm_so;
308 struct sockaddr *nam = msg->bind.nm_nam;
309 struct thread *td = msg->bind.nm_td;
313 struct sockaddr_in6 *sin6p;
315 COMMON_START(so, inp, 0);
318 * Must check for multicast addresses and disallow binding
321 sin6p = (struct sockaddr_in6 *)nam;
322 if (sin6p->sin6_family == AF_INET6 &&
323 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
324 error = EAFNOSUPPORT;
327 inp->inp_vflag &= ~INP_IPV4;
328 inp->inp_vflag |= INP_IPV6;
329 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
330 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
331 inp->inp_vflag |= INP_IPV4;
332 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
333 struct sockaddr_in sin;
335 in6_sin6_2_sin(&sin, sin6p);
336 inp->inp_vflag |= INP_IPV4;
337 inp->inp_vflag &= ~INP_IPV6;
338 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
342 error = in6_pcbbind(inp, nam, td);
345 COMMON_END(PRU_BIND);
351 struct netmsg_inswildcard {
352 struct netmsg_base base;
353 struct inpcb *nm_inp;
357 in_pcbinswildcardhash_handler(netmsg_t msg)
359 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
360 int cpu = mycpuid, nextcpu;
362 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
365 if (nextcpu < ncpus2)
366 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
368 lwkt_replymsg(&nm->base.lmsg, 0);
374 * Prepare to accept connections.
377 tcp_usr_listen(netmsg_t msg)
379 struct socket *so = msg->listen.base.nm_so;
380 struct thread *td = msg->listen.nm_td;
385 struct netmsg_inswildcard nm;
388 COMMON_START(so, inp, 0);
390 if (tp->t_flags & TF_LISTEN)
393 if (inp->inp_lport == 0) {
394 error = in_pcbbind(inp, NULL, td);
399 tp->t_state = TCPS_LISTEN;
400 tp->t_flags |= TF_LISTEN;
401 tp->tt_msg = NULL; /* Catch any invalid timer usage */
406 * We have to set the flag because we can't have other cpus
407 * messing with our inp's flags.
409 KASSERT(!(inp->inp_flags & INP_CONNECTED),
410 ("already on connhash\n"));
411 KASSERT(!(inp->inp_flags & INP_WILDCARD),
412 ("already on wildcardhash\n"));
413 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
414 ("already on MP wildcardhash\n"));
415 inp->inp_flags |= INP_WILDCARD_MP;
417 KKASSERT(so->so_port == cpu_portfn(0));
418 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
419 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
421 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
422 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
424 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
427 in_pcbinswildcardhash(inp);
428 COMMON_END(PRU_LISTEN);
434 tcp6_usr_listen(netmsg_t msg)
436 struct socket *so = msg->listen.base.nm_so;
437 struct thread *td = msg->listen.nm_td;
442 struct netmsg_inswildcard nm;
445 COMMON_START(so, inp, 0);
447 if (tp->t_flags & TF_LISTEN)
450 if (inp->inp_lport == 0) {
451 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
452 inp->inp_vflag |= INP_IPV4;
454 inp->inp_vflag &= ~INP_IPV4;
455 error = in6_pcbbind(inp, NULL, td);
460 tp->t_state = TCPS_LISTEN;
461 tp->t_flags |= TF_LISTEN;
462 tp->tt_msg = NULL; /* Catch any invalid timer usage */
467 * We have to set the flag because we can't have other cpus
468 * messing with our inp's flags.
470 KASSERT(!(inp->inp_flags & INP_CONNECTED),
471 ("already on connhash\n"));
472 KASSERT(!(inp->inp_flags & INP_WILDCARD),
473 ("already on wildcardhash\n"));
474 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
475 ("already on MP wildcardhash\n"));
476 inp->inp_flags |= INP_WILDCARD_MP;
478 KKASSERT(so->so_port == cpu_portfn(0));
479 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
480 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
482 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
483 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
485 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
488 in_pcbinswildcardhash(inp);
489 COMMON_END(PRU_LISTEN);
494 * Initiate connection to peer.
495 * Create a template for use in transmissions on this connection.
496 * Enter SYN_SENT state, and mark socket as connecting.
497 * Start keep-alive timer, and seed output sequence space.
498 * Send initial segment on connection.
501 tcp_usr_connect(netmsg_t msg)
503 struct socket *so = msg->connect.base.nm_so;
504 struct sockaddr *nam = msg->connect.nm_nam;
505 struct thread *td = msg->connect.nm_td;
509 struct sockaddr_in *sinp;
511 COMMON_START(so, inp, 0);
514 * Must disallow TCP ``connections'' to multicast addresses.
516 sinp = (struct sockaddr_in *)nam;
517 if (sinp->sin_family == AF_INET
518 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
519 error = EAFNOSUPPORT;
523 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
524 error = EAFNOSUPPORT; /* IPv6 only jail */
529 /* msg is invalid now */
532 if (msg->connect.nm_m) {
533 m_freem(msg->connect.nm_m);
534 msg->connect.nm_m = NULL;
536 lwkt_replymsg(&msg->lmsg, error);
542 tcp6_usr_connect(netmsg_t msg)
544 struct socket *so = msg->connect.base.nm_so;
545 struct sockaddr *nam = msg->connect.nm_nam;
546 struct thread *td = msg->connect.nm_td;
550 struct sockaddr_in6 *sin6p;
552 COMMON_START(so, inp, 0);
555 * Must disallow TCP ``connections'' to multicast addresses.
557 sin6p = (struct sockaddr_in6 *)nam;
558 if (sin6p->sin6_family == AF_INET6
559 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
560 error = EAFNOSUPPORT;
564 if (!prison_remote_ip(td, nam)) {
565 error = EAFNOSUPPORT; /* IPv4 only jail */
569 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
570 struct sockaddr_in *sinp;
572 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
576 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
577 in6_sin6_2_sin(sinp, sin6p);
578 inp->inp_vflag |= INP_IPV4;
579 inp->inp_vflag &= ~INP_IPV6;
580 msg->connect.nm_nam = (struct sockaddr *)sinp;
581 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
583 /* msg is invalid now */
586 inp->inp_vflag &= ~INP_IPV4;
587 inp->inp_vflag |= INP_IPV6;
588 inp->inp_inc.inc_isipv6 = 1;
590 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
592 /* msg is invalid now */
595 if (msg->connect.nm_m) {
596 m_freem(msg->connect.nm_m);
597 msg->connect.nm_m = NULL;
599 lwkt_replymsg(&msg->lmsg, error);
605 * Initiate disconnect from peer.
606 * If connection never passed embryonic stage, just drop;
607 * else if don't need to let data drain, then can just drop anyways,
608 * else have to begin TCP shutdown process: mark socket disconnecting,
609 * drain unread data, state switch to reflect user close, and
610 * send segment (e.g. FIN) to peer. Socket will be really disconnected
611 * when peer sends FIN and acks ours.
613 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
616 tcp_usr_disconnect(netmsg_t msg)
618 struct socket *so = msg->disconnect.base.nm_so;
623 COMMON_START(so, inp, 1);
624 tp = tcp_disconnect(tp);
625 COMMON_END(PRU_DISCONNECT);
629 * Accept a connection. Essentially all the work is
630 * done at higher levels; just return the address
631 * of the peer, storing through addr.
634 tcp_usr_accept(netmsg_t msg)
636 struct socket *so = msg->accept.base.nm_so;
637 struct sockaddr **nam = msg->accept.nm_nam;
640 struct tcpcb *tp = NULL;
644 if (so->so_state & SS_ISDISCONNECTED) {
645 error = ECONNABORTED;
655 in_setpeeraddr(so, nam);
656 COMMON_END(PRU_ACCEPT);
661 tcp6_usr_accept(netmsg_t msg)
663 struct socket *so = msg->accept.base.nm_so;
664 struct sockaddr **nam = msg->accept.nm_nam;
667 struct tcpcb *tp = NULL;
672 if (so->so_state & SS_ISDISCONNECTED) {
673 error = ECONNABORTED;
682 in6_mapped_peeraddr(so, nam);
683 COMMON_END(PRU_ACCEPT);
687 * Mark the connection as being incapable of further output.
690 tcp_usr_shutdown(netmsg_t msg)
692 struct socket *so = msg->shutdown.base.nm_so;
697 COMMON_START(so, inp, 0);
699 tp = tcp_usrclosed(tp);
701 error = tcp_output(tp);
702 COMMON_END(PRU_SHUTDOWN);
706 * After a receive, possibly send window update to peer.
709 tcp_usr_rcvd(netmsg_t msg)
711 struct socket *so = msg->rcvd.base.nm_so;
716 COMMON_START(so, inp, 0);
718 COMMON_END(PRU_RCVD);
722 * Do a send by putting data in output queue and updating urgent
723 * marker if URG set. Possibly send more data. Unlike the other
724 * pru_*() routines, the mbuf chains are our responsibility. We
725 * must either enqueue them or free them. The other pru_* routines
726 * generally are caller-frees.
729 tcp_usr_send(netmsg_t msg)
731 struct socket *so = msg->send.base.nm_so;
732 int flags = msg->send.nm_flags;
733 struct mbuf *m = msg->send.nm_m;
734 struct mbuf *control __debugvar = msg->send.nm_control;
740 KKASSERT(control == NULL);
746 * OOPS! we lost a race, the TCP session got reset after
747 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
748 * network interrupt in the non-critical section of sosend().
751 error = ECONNRESET; /* XXX EPIPE? */
760 * Don't let too much OOB data build up
762 if (flags & PRUS_OOB) {
763 if (ssb_space(&so->so_snd) < -512) {
771 * Pump the data into the socket.
774 ssb_appendstream(&so->so_snd, m);
775 if (flags & PRUS_OOB) {
777 * According to RFC961 (Assigned Protocols),
778 * the urgent pointer points to the last octet
779 * of urgent data. We continue, however,
780 * to consider it to indicate the first octet
781 * of data past the urgent section.
782 * Otherwise, snd_up should be one lower.
784 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
785 tp->t_flags |= TF_FORCE;
786 error = tcp_output(tp);
787 tp->t_flags &= ~TF_FORCE;
789 if (flags & PRUS_EOF) {
791 * Close the send side of the connection after
795 tp = tcp_usrclosed(tp);
798 if (flags & PRUS_MORETOCOME)
799 tp->t_flags |= TF_MORETOCOME;
800 error = tcp_output(tp);
801 if (flags & PRUS_MORETOCOME)
802 tp->t_flags &= ~TF_MORETOCOME;
805 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
806 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
807 (flags & PRUS_NOREPLY));
811 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
812 * will sofree() it when we return.
815 tcp_usr_abort(netmsg_t msg)
817 struct socket *so = msg->abort.base.nm_so;
822 COMMON_START(so, inp, 1);
823 tp = tcp_drop(tp, ECONNABORTED);
824 COMMON_END(PRU_ABORT);
828 * Receive out-of-band data.
831 tcp_usr_rcvoob(netmsg_t msg)
833 struct socket *so = msg->rcvoob.base.nm_so;
834 struct mbuf *m = msg->rcvoob.nm_m;
835 int flags = msg->rcvoob.nm_flags;
840 COMMON_START(so, inp, 0);
841 if ((so->so_oobmark == 0 &&
842 (so->so_state & SS_RCVATMARK) == 0) ||
843 so->so_options & SO_OOBINLINE ||
844 tp->t_oobflags & TCPOOB_HADDATA) {
848 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
853 *mtod(m, caddr_t) = tp->t_iobc;
854 if ((flags & MSG_PEEK) == 0)
855 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
856 COMMON_END(PRU_RCVOOB);
859 /* xxx - should be const */
860 struct pr_usrreqs tcp_usrreqs = {
861 .pru_abort = tcp_usr_abort,
862 .pru_accept = tcp_usr_accept,
863 .pru_attach = tcp_usr_attach,
864 .pru_bind = tcp_usr_bind,
865 .pru_connect = tcp_usr_connect,
866 .pru_connect2 = pr_generic_notsupp,
867 .pru_control = in_control_dispatch,
868 .pru_detach = tcp_usr_detach,
869 .pru_disconnect = tcp_usr_disconnect,
870 .pru_listen = tcp_usr_listen,
871 .pru_peeraddr = in_setpeeraddr_dispatch,
872 .pru_rcvd = tcp_usr_rcvd,
873 .pru_rcvoob = tcp_usr_rcvoob,
874 .pru_send = tcp_usr_send,
875 .pru_sense = pru_sense_null,
876 .pru_shutdown = tcp_usr_shutdown,
877 .pru_sockaddr = in_setsockaddr_dispatch,
878 .pru_sosend = sosendtcp,
879 .pru_soreceive = soreceive
883 struct pr_usrreqs tcp6_usrreqs = {
884 .pru_abort = tcp_usr_abort,
885 .pru_accept = tcp6_usr_accept,
886 .pru_attach = tcp_usr_attach,
887 .pru_bind = tcp6_usr_bind,
888 .pru_connect = tcp6_usr_connect,
889 .pru_connect2 = pr_generic_notsupp,
890 .pru_control = in6_control_dispatch,
891 .pru_detach = tcp_usr_detach,
892 .pru_disconnect = tcp_usr_disconnect,
893 .pru_listen = tcp6_usr_listen,
894 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
895 .pru_rcvd = tcp_usr_rcvd,
896 .pru_rcvoob = tcp_usr_rcvoob,
897 .pru_send = tcp_usr_send,
898 .pru_sense = pru_sense_null,
899 .pru_shutdown = tcp_usr_shutdown,
900 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
901 .pru_sosend = sosendtcp,
902 .pru_soreceive = soreceive
907 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
908 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
910 struct inpcb *inp = tp->t_inpcb, *oinp;
911 struct socket *so = inp->inp_socket;
912 struct route *ro = &inp->inp_route;
914 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
915 sin->sin_addr, sin->sin_port,
916 (inp->inp_laddr.s_addr != INADDR_ANY ?
917 inp->inp_laddr : if_sin->sin_addr),
918 inp->inp_lport, 0, NULL);
923 if (inp->inp_laddr.s_addr == INADDR_ANY)
924 inp->inp_laddr = if_sin->sin_addr;
925 inp->inp_faddr = sin->sin_addr;
926 inp->inp_fport = sin->sin_port;
927 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
928 in_pcbinsconnhash(inp);
931 * We are now on the inpcb's owner CPU, if the cached route was
932 * freed because the rtentry's owner CPU is not the current CPU
933 * (e.g. in tcp_connect()), then we try to reallocate it here with
934 * the hope that a rtentry may be cloned from a RTF_PRCLONING
937 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
939 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
940 ro->ro_dst.sa_family = AF_INET;
941 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
942 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
948 * Now that no more errors can occur, change the protocol processing
949 * port to the current thread (which is the correct thread).
951 * Create TCP timer message now; we are on the tcpcb's owner
954 tcp_create_timermsg(tp, &curthread->td_msgport);
957 * Compute window scaling to request. Use a larger scaling then
958 * needed for the initial receive buffer in case the receive buffer
961 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
962 tp->request_r_scale = TCP_MIN_WINSHIFT;
963 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
964 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
966 tp->request_r_scale++;
970 tcpstat.tcps_connattempt++;
971 tp->t_state = TCPS_SYN_SENT;
972 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
973 tp->iss = tcp_new_isn(tp);
976 ssb_appendstream(&so->so_snd, m);
978 if (flags & PRUS_OOB)
979 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
983 * Close the send side of the connection after
984 * the data is sent if flagged.
986 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
988 tp = tcp_usrclosed(tp);
990 return (tcp_output(tp));
994 * Common subroutine to open a TCP connection to remote host specified
995 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
996 * port number if needed. Call in_pcbladdr to do the routing and to choose
997 * a local host address (interface).
998 * Initialize connection parameters and enter SYN-SENT state.
1001 tcp_connect(netmsg_t msg)
1003 struct socket *so = msg->connect.base.nm_so;
1004 struct sockaddr *nam = msg->connect.nm_nam;
1005 struct thread *td = msg->connect.nm_td;
1006 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1007 struct sockaddr_in *if_sin;
1010 int error, calc_laddr = 1;
1015 COMMON_START(so, inp, 0);
1018 * Reconnect our pcb if we have to
1020 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1021 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1022 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1026 * Bind if we have to
1028 if (inp->inp_lport == 0) {
1029 if (tcp_lport_extension) {
1030 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1032 error = in_pcbladdr(inp, nam, &if_sin, td);
1035 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1037 error = in_pcbconn_bind(inp, nam, td);
1043 error = in_pcbbind(inp, NULL, td);
1051 * Calculate the correct protocol processing thread. The
1052 * connect operation must run there. Set the forwarding
1053 * port before we forward the message or it will get bounced
1056 error = in_pcbladdr(inp, nam, &if_sin, td);
1060 KKASSERT(inp->inp_socket == so);
1063 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1064 (inp->inp_laddr.s_addr ?
1065 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1068 if (port != &curthread->td_msgport) {
1069 struct route *ro = &inp->inp_route;
1072 * in_pcbladdr() may have allocated a route entry for us
1073 * on the current CPU, but we need a route entry on the
1074 * inpcb's owner CPU, so free it here.
1076 if (ro->ro_rt != NULL)
1078 bzero(ro, sizeof(*ro));
1081 * We are moving the protocol processing port the socket
1082 * is on, we have to unlink here and re-link on the
1085 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1086 sosetport(so, port);
1087 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1088 msg->connect.base.nm_dispatch = tcp_connect;
1090 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1091 /* msg invalid now */
1095 KKASSERT(so->so_port == &curthread->td_msgport);
1097 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1098 msg->connect.nm_m, sin, if_sin);
1099 msg->connect.nm_m = NULL;
1101 if (msg->connect.nm_m) {
1102 m_freem(msg->connect.nm_m);
1103 msg->connect.nm_m = NULL;
1105 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1106 kfree(msg->connect.nm_nam, M_LWKTMSG);
1107 msg->connect.nm_nam = NULL;
1109 lwkt_replymsg(&msg->connect.base.lmsg, error);
1110 /* msg invalid now */
1116 tcp6_connect(netmsg_t msg)
1119 struct socket *so = msg->connect.base.nm_so;
1120 struct sockaddr *nam = msg->connect.nm_nam;
1121 struct thread *td = msg->connect.nm_td;
1123 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1124 struct in6_addr *addr6;
1130 COMMON_START(so, inp, 0);
1133 * Reconnect our pcb if we have to
1135 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1136 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1137 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1141 * Bind if we have to
1143 if (inp->inp_lport == 0) {
1144 error = in6_pcbbind(inp, NULL, td);
1150 * Cannot simply call in_pcbconnect, because there might be an
1151 * earlier incarnation of this same connection still in
1152 * TIME_WAIT state, creating an ADDRINUSE error.
1154 error = in6_pcbladdr(inp, nam, &addr6, td);
1159 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1161 if (port != &curthread->td_msgport) {
1162 struct route *ro = &inp->inp_route;
1165 * in_pcbladdr() may have allocated a route entry for us
1166 * on the current CPU, but we need a route entry on the
1167 * inpcb's owner CPU, so free it here.
1169 if (ro->ro_rt != NULL)
1171 bzero(ro, sizeof(*ro));
1173 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1174 sosetport(so, port);
1175 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1176 msg->connect.base.nm_dispatch = tcp6_connect;
1178 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1179 /* msg invalid now */
1183 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1184 &msg->connect.nm_m, sin6, addr6);
1185 /* nm_m may still be intact */
1187 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1189 /* msg invalid now */
1191 if (msg->connect.nm_m) {
1192 m_freem(msg->connect.nm_m);
1193 msg->connect.nm_m = NULL;
1195 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1196 kfree(msg->connect.nm_nam, M_LWKTMSG);
1197 msg->connect.nm_nam = NULL;
1199 lwkt_replymsg(&msg->connect.base.lmsg, error);
1200 /* msg invalid now */
1205 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1206 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1208 struct mbuf *m = *mp;
1209 struct inpcb *inp = tp->t_inpcb;
1210 struct socket *so = inp->inp_socket;
1214 * Cannot simply call in_pcbconnect, because there might be an
1215 * earlier incarnation of this same connection still in
1216 * TIME_WAIT state, creating an ADDRINUSE error.
1218 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1219 &sin6->sin6_addr, sin6->sin6_port,
1220 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1221 addr6 : &inp->in6p_laddr),
1222 inp->inp_lport, 0, NULL);
1224 return (EADDRINUSE);
1226 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1227 inp->in6p_laddr = *addr6;
1228 inp->in6p_faddr = sin6->sin6_addr;
1229 inp->inp_fport = sin6->sin6_port;
1230 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1231 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1232 in_pcbinsconnhash(inp);
1235 * Now that no more errors can occur, change the protocol processing
1236 * port to the current thread (which is the correct thread).
1238 * Create TCP timer message now; we are on the tcpcb's owner
1241 tcp_create_timermsg(tp, &curthread->td_msgport);
1243 /* Compute window scaling to request. */
1244 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1245 tp->request_r_scale = TCP_MIN_WINSHIFT;
1246 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1247 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1248 tp->request_r_scale++;
1252 tcpstat.tcps_connattempt++;
1253 tp->t_state = TCPS_SYN_SENT;
1254 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1255 tp->iss = tcp_new_isn(tp);
1256 tcp_sendseqinit(tp);
1258 ssb_appendstream(&so->so_snd, m);
1260 if (flags & PRUS_OOB)
1261 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1265 * Close the send side of the connection after
1266 * the data is sent if flagged.
1268 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1270 tp = tcp_usrclosed(tp);
1272 return (tcp_output(tp));
1278 * The new sockopt interface makes it possible for us to block in the
1279 * copyin/out step (if we take a page fault). Taking a page fault while
1280 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1281 * both now use TSM, there probably isn't any need for this function to
1282 * run in a critical section any more. This needs more examination.)
1285 tcp_ctloutput(netmsg_t msg)
1287 struct socket *so = msg->base.nm_so;
1288 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1289 int error, opt, optval, opthz;
1300 if (sopt->sopt_level != IPPROTO_TCP) {
1302 if (INP_CHECK_SOCKAF(so, AF_INET6))
1303 ip6_ctloutput_dispatch(msg);
1307 /* msg invalid now */
1310 tp = intotcpcb(inp);
1312 switch (sopt->sopt_dir) {
1314 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1318 switch (sopt->sopt_name) {
1321 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1322 tp->t_flags |= TF_FASTKEEP;
1323 tcp_timer_keep_activity(tp, 0);
1326 tp->t_flags &= ~TF_FASTKEEP;
1329 #ifdef TCP_SIGNATURE
1330 case TCP_SIGNATURE_ENABLE:
1332 tp->t_flags |= TF_SIGNATURE;
1334 tp->t_flags &= ~TF_SIGNATURE;
1336 #endif /* TCP_SIGNATURE */
1339 switch (sopt->sopt_name) {
1347 opt = 0; /* dead code to fool gcc */
1354 tp->t_flags &= ~opt;
1359 tp->t_flags |= TF_NOPUSH;
1361 tp->t_flags &= ~TF_NOPUSH;
1362 error = tcp_output(tp);
1368 * Must be between 0 and maxseg. If the requested
1369 * maxseg is too small to satisfy the desired minmss,
1370 * pump it up (silently so sysctl modifications of
1371 * minmss do not create unexpected program failures).
1372 * Handle degenerate cases.
1374 if (optval > 0 && optval <= tp->t_maxseg) {
1375 if (optval + 40 < tcp_minmss) {
1376 optval = tcp_minmss - 40;
1380 tp->t_maxseg = optval;
1387 opthz = ((int64_t)optval * hz) / 1000;
1389 tp->t_keepinit = opthz;
1395 opthz = ((int64_t)optval * hz) / 1000;
1397 tp->t_keepidle = opthz;
1403 opthz = ((int64_t)optval * hz) / 1000;
1405 tp->t_keepintvl = opthz;
1406 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1414 tp->t_keepcnt = optval;
1415 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1422 error = ENOPROTOOPT;
1428 switch (sopt->sopt_name) {
1429 #ifdef TCP_SIGNATURE
1430 case TCP_SIGNATURE_ENABLE:
1431 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1433 #endif /* TCP_SIGNATURE */
1435 optval = tp->t_flags & TF_NODELAY;
1438 optval = tp->t_maxseg;
1441 optval = tp->t_flags & TF_NOOPT;
1444 optval = tp->t_flags & TF_NOPUSH;
1447 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1450 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1453 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1456 optval = tp->t_keepcnt;
1459 error = ENOPROTOOPT;
1463 soopt_from_kbuf(sopt, &optval, sizeof optval);
1467 lwkt_replymsg(&msg->lmsg, error);
1471 * tcp_sendspace and tcp_recvspace are the default send and receive window
1472 * sizes, respectively. These are obsolescent (this information should
1473 * be set by the route).
1475 * Use a default that does not require tcp window scaling to be turned
1476 * on. Individual programs or the administrator can increase the default.
1478 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1479 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1480 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1481 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1482 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1483 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1486 * Attach TCP protocol to socket, allocating internet protocol control
1487 * block, tcp control block, bufer space, and entering LISTEN state
1488 * if to accept connections.
1491 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1498 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1501 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1502 lwkt_gettoken(&so->so_rcv.ssb_token);
1503 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1505 lwkt_reltoken(&so->so_rcv.ssb_token);
1509 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1510 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1511 cpu = mycpu->gd_cpuid;
1514 * Set the default port for protocol processing. This will likely
1515 * change when we connect.
1517 error = in_pcballoc(so, &tcbinfo[cpu]);
1523 inp->inp_vflag |= INP_IPV6;
1524 inp->in6p_hops = -1; /* use kernel default */
1528 inp->inp_vflag |= INP_IPV4;
1529 tp = tcp_newtcpcb(inp);
1532 * Make sure the socket is destroyed by the pcbdetach.
1541 sofree(so); /* from ref above */
1544 tp->t_state = TCPS_CLOSED;
1549 * Initiate (or continue) disconnect.
1550 * If embryonic state, just send reset (once).
1551 * If in ``let data drain'' option and linger null, just drop.
1552 * Otherwise (hard), mark socket disconnecting and drop
1553 * current input data; switch states based on user close, and
1554 * send segment to peer (with FIN).
1556 static struct tcpcb *
1557 tcp_disconnect(struct tcpcb *tp)
1559 struct socket *so = tp->t_inpcb->inp_socket;
1561 if (tp->t_state < TCPS_ESTABLISHED) {
1563 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1564 tp = tcp_drop(tp, 0);
1566 lwkt_gettoken(&so->so_rcv.ssb_token);
1567 soisdisconnecting(so);
1568 sbflush(&so->so_rcv.sb);
1569 tp = tcp_usrclosed(tp);
1572 lwkt_reltoken(&so->so_rcv.ssb_token);
1578 * User issued close, and wish to trail through shutdown states:
1579 * if never received SYN, just forget it. If got a SYN from peer,
1580 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1581 * If already got a FIN from peer, then almost done; go to LAST_ACK
1582 * state. In all other cases, have already sent FIN to peer (e.g.
1583 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1584 * for peer to send FIN or not respond to keep-alives, etc.
1585 * We can let the user exit from the close as soon as the FIN is acked.
1587 static struct tcpcb *
1588 tcp_usrclosed(struct tcpcb *tp)
1591 switch (tp->t_state) {
1595 tp->t_state = TCPS_CLOSED;
1600 case TCPS_SYN_RECEIVED:
1601 tp->t_flags |= TF_NEEDFIN;
1604 case TCPS_ESTABLISHED:
1605 tp->t_state = TCPS_FIN_WAIT_1;
1608 case TCPS_CLOSE_WAIT:
1609 tp->t_state = TCPS_LAST_ACK;
1612 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1613 soisdisconnected(tp->t_inpcb->inp_socket);
1614 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1615 if (tp->t_state == TCPS_FIN_WAIT_2) {
1616 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,