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)
163 * TCP attaches to socket via pru_attach(), reserving space,
164 * and an internet control block. This is likely occuring on
165 * cpu0 and may have to move later when we bind/connect.
168 tcp_usr_attach(netmsg_t msg)
170 struct socket *so = msg->base.nm_so;
171 struct pru_attach_info *ai = msg->attach.nm_ai;
174 struct tcpcb *tp = 0;
185 error = tcp_attach(so, ai);
189 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
190 so->so_linger = TCP_LINGERTIME;
193 sofree(so); /* from ref above */
194 TCPDEBUG2(PRU_ATTACH);
195 lwkt_replymsg(&msg->lmsg, error);
199 * pru_detach() detaches the TCP protocol from the socket.
200 * If the protocol state is non-embryonic, then can't
201 * do this directly: have to initiate a pru_disconnect(),
202 * which may finish later; embryonic TCB's can just
206 tcp_usr_detach(netmsg_t msg)
208 struct socket *so = msg->base.nm_so;
217 * If the inp is already detached it may have been due to an async
218 * close. Just return as if no error occured.
220 * It's possible for the tcpcb (tp) to disconnect from the inp due
221 * to tcp_drop()->tcp_close() being called. This may occur *after*
222 * the detach message has been queued so we may find a NULL tp here.
225 if ((tp = intotcpcb(inp)) != NULL) {
227 tp = tcp_disconnect(tp);
228 TCPDEBUG2(PRU_DETACH);
231 lwkt_replymsg(&msg->lmsg, error);
235 * NOTE: ignore_error is non-zero for certain disconnection races
236 * which we want to silently allow, otherwise close() may return
237 * an unexpected error.
239 * NOTE: The variables (msg) and (tp) are assumed.
241 #define COMMON_START(so, inp, ignore_error) \
247 error = ignore_error ? 0 : EINVAL; \
251 tp = intotcpcb(inp); \
255 #define COMMON_END(req) \
258 lwkt_replymsg(&msg->lmsg, error); \
263 * Give the socket an address.
266 tcp_usr_bind(netmsg_t msg)
268 struct socket *so = msg->bind.base.nm_so;
269 struct sockaddr *nam = msg->bind.nm_nam;
270 struct thread *td = msg->bind.nm_td;
274 struct sockaddr_in *sinp;
276 COMMON_START(so, inp, 0);
279 * Must check for multicast addresses and disallow binding
282 sinp = (struct sockaddr_in *)nam;
283 if (sinp->sin_family == AF_INET &&
284 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
285 error = EAFNOSUPPORT;
288 error = in_pcbbind(inp, nam, td);
291 COMMON_END(PRU_BIND);
298 tcp6_usr_bind(netmsg_t msg)
300 struct socket *so = msg->bind.base.nm_so;
301 struct sockaddr *nam = msg->bind.nm_nam;
302 struct thread *td = msg->bind.nm_td;
306 struct sockaddr_in6 *sin6p;
308 COMMON_START(so, inp, 0);
311 * Must check for multicast addresses and disallow binding
314 sin6p = (struct sockaddr_in6 *)nam;
315 if (sin6p->sin6_family == AF_INET6 &&
316 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
317 error = EAFNOSUPPORT;
320 inp->inp_vflag &= ~INP_IPV4;
321 inp->inp_vflag |= INP_IPV6;
322 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
323 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
324 inp->inp_vflag |= INP_IPV4;
325 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
326 struct sockaddr_in sin;
328 in6_sin6_2_sin(&sin, sin6p);
329 inp->inp_vflag |= INP_IPV4;
330 inp->inp_vflag &= ~INP_IPV6;
331 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
335 error = in6_pcbbind(inp, nam, td);
338 COMMON_END(PRU_BIND);
344 struct netmsg_inswildcard {
345 struct netmsg_base base;
346 struct inpcb *nm_inp;
350 in_pcbinswildcardhash_handler(netmsg_t msg)
352 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
353 int cpu = mycpuid, nextcpu;
355 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
358 if (nextcpu < ncpus2)
359 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
361 lwkt_replymsg(&nm->base.lmsg, 0);
367 * Prepare to accept connections.
370 tcp_usr_listen(netmsg_t msg)
372 struct socket *so = msg->listen.base.nm_so;
373 struct thread *td = msg->listen.nm_td;
378 struct netmsg_inswildcard nm;
381 COMMON_START(so, inp, 0);
382 if (inp->inp_lport == 0) {
383 error = in_pcbbind(inp, NULL, td);
388 tp->t_state = TCPS_LISTEN;
389 tp->t_flags |= TF_SYNCACHE;
390 tp->tt_msg = NULL; /* Catch any invalid timer usage */
395 * We have to set the flag because we can't have other cpus
396 * messing with our inp's flags.
398 inp->inp_flags |= INP_WILDCARD_MP;
400 KKASSERT(so->so_port == cpu_portfn(0));
401 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
402 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
404 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
405 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
407 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
410 in_pcbinswildcardhash(inp);
411 COMMON_END(PRU_LISTEN);
417 tcp6_usr_listen(netmsg_t msg)
419 struct socket *so = msg->listen.base.nm_so;
420 struct thread *td = msg->listen.nm_td;
425 struct netmsg_inswildcard nm;
428 COMMON_START(so, inp, 0);
429 if (inp->inp_lport == 0) {
430 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
431 inp->inp_vflag |= INP_IPV4;
433 inp->inp_vflag &= ~INP_IPV4;
434 error = in6_pcbbind(inp, NULL, td);
439 tp->t_state = TCPS_LISTEN;
440 tp->t_flags |= TF_SYNCACHE;
441 tp->tt_msg = NULL; /* Catch any invalid timer usage */
446 * We have to set the flag because we can't have other cpus
447 * messing with our inp's flags.
449 inp->inp_flags |= INP_WILDCARD_MP;
451 KKASSERT(so->so_port == cpu_portfn(0));
452 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
453 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
455 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
456 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
458 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
461 in_pcbinswildcardhash(inp);
462 COMMON_END(PRU_LISTEN);
467 * Initiate connection to peer.
468 * Create a template for use in transmissions on this connection.
469 * Enter SYN_SENT state, and mark socket as connecting.
470 * Start keep-alive timer, and seed output sequence space.
471 * Send initial segment on connection.
474 tcp_usr_connect(netmsg_t msg)
476 struct socket *so = msg->connect.base.nm_so;
477 struct sockaddr *nam = msg->connect.nm_nam;
478 struct thread *td = msg->connect.nm_td;
482 struct sockaddr_in *sinp;
484 COMMON_START(so, inp, 0);
487 * Must disallow TCP ``connections'' to multicast addresses.
489 sinp = (struct sockaddr_in *)nam;
490 if (sinp->sin_family == AF_INET
491 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
492 error = EAFNOSUPPORT;
496 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
497 error = EAFNOSUPPORT; /* IPv6 only jail */
502 /* msg is invalid now */
505 if (msg->connect.nm_m) {
506 m_freem(msg->connect.nm_m);
507 msg->connect.nm_m = NULL;
509 lwkt_replymsg(&msg->lmsg, error);
515 tcp6_usr_connect(netmsg_t msg)
517 struct socket *so = msg->connect.base.nm_so;
518 struct sockaddr *nam = msg->connect.nm_nam;
519 struct thread *td = msg->connect.nm_td;
523 struct sockaddr_in6 *sin6p;
525 COMMON_START(so, inp, 0);
528 * Must disallow TCP ``connections'' to multicast addresses.
530 sin6p = (struct sockaddr_in6 *)nam;
531 if (sin6p->sin6_family == AF_INET6
532 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
533 error = EAFNOSUPPORT;
537 if (!prison_remote_ip(td, nam)) {
538 error = EAFNOSUPPORT; /* IPv4 only jail */
542 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
543 struct sockaddr_in *sinp;
545 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
549 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
550 in6_sin6_2_sin(sinp, sin6p);
551 inp->inp_vflag |= INP_IPV4;
552 inp->inp_vflag &= ~INP_IPV6;
553 msg->connect.nm_nam = (struct sockaddr *)sinp;
554 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
556 /* msg is invalid now */
559 inp->inp_vflag &= ~INP_IPV4;
560 inp->inp_vflag |= INP_IPV6;
561 inp->inp_inc.inc_isipv6 = 1;
563 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
565 /* msg is invalid now */
568 if (msg->connect.nm_m) {
569 m_freem(msg->connect.nm_m);
570 msg->connect.nm_m = NULL;
572 lwkt_replymsg(&msg->lmsg, error);
578 * Initiate disconnect from peer.
579 * If connection never passed embryonic stage, just drop;
580 * else if don't need to let data drain, then can just drop anyways,
581 * else have to begin TCP shutdown process: mark socket disconnecting,
582 * drain unread data, state switch to reflect user close, and
583 * send segment (e.g. FIN) to peer. Socket will be really disconnected
584 * when peer sends FIN and acks ours.
586 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
589 tcp_usr_disconnect(netmsg_t msg)
591 struct socket *so = msg->disconnect.base.nm_so;
596 COMMON_START(so, inp, 1);
597 tp = tcp_disconnect(tp);
598 COMMON_END(PRU_DISCONNECT);
602 * Accept a connection. Essentially all the work is
603 * done at higher levels; just return the address
604 * of the peer, storing through addr.
607 tcp_usr_accept(netmsg_t msg)
609 struct socket *so = msg->accept.base.nm_so;
610 struct sockaddr **nam = msg->accept.nm_nam;
613 struct tcpcb *tp = NULL;
617 if (so->so_state & SS_ISDISCONNECTED) {
618 error = ECONNABORTED;
628 in_setpeeraddr(so, nam);
629 COMMON_END(PRU_ACCEPT);
634 tcp6_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;
645 if (so->so_state & SS_ISDISCONNECTED) {
646 error = ECONNABORTED;
655 in6_mapped_peeraddr(so, nam);
656 COMMON_END(PRU_ACCEPT);
660 * Mark the connection as being incapable of further output.
663 tcp_usr_shutdown(netmsg_t msg)
665 struct socket *so = msg->shutdown.base.nm_so;
670 COMMON_START(so, inp, 0);
672 tp = tcp_usrclosed(tp);
674 error = tcp_output(tp);
675 COMMON_END(PRU_SHUTDOWN);
679 * After a receive, possibly send window update to peer.
682 tcp_usr_rcvd(netmsg_t msg)
684 struct socket *so = msg->rcvd.base.nm_so;
689 COMMON_START(so, inp, 0);
691 COMMON_END(PRU_RCVD);
695 * Do a send by putting data in output queue and updating urgent
696 * marker if URG set. Possibly send more data. Unlike the other
697 * pru_*() routines, the mbuf chains are our responsibility. We
698 * must either enqueue them or free them. The other pru_* routines
699 * generally are caller-frees.
702 tcp_usr_send(netmsg_t msg)
704 struct socket *so = msg->send.base.nm_so;
705 int flags = msg->send.nm_flags;
706 struct mbuf *m = msg->send.nm_m;
707 struct sockaddr *nam = msg->send.nm_addr;
708 struct mbuf *control = msg->send.nm_control;
709 struct thread *td = msg->send.nm_td;
722 * OOPS! we lost a race, the TCP session got reset after
723 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
724 * network interrupt in the non-critical section of sosend().
729 error = ECONNRESET; /* XXX EPIPE? */
735 isipv6 = nam && nam->sa_family == AF_INET6;
740 /* TCP doesn't do control messages (rights, creds, etc) */
741 if (control->m_len) {
747 m_freem(control); /* empty control, just free it */
751 * Don't let too much OOB data build up
753 if (flags & PRUS_OOB) {
754 if (ssb_space(&so->so_snd) < -512) {
762 * Do implied connect if not yet connected. Any data sent
763 * with the connect is handled by tcp_connect() and friends.
765 * NOTE! PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
767 if (nam && tp->t_state < TCPS_SYN_SENT) {
768 kprintf("implied fallback\n");
769 msg->connect.nm_nam = nam;
770 msg->connect.nm_td = td;
771 msg->connect.nm_m = m;
772 msg->connect.nm_flags = flags;
773 msg->connect.nm_reconnect = NMSG_RECONNECT_FALLBACK;
780 /* msg invalid now */
785 * Pump the data into the socket.
788 ssb_appendstream(&so->so_snd, m);
789 if (flags & PRUS_OOB) {
791 * According to RFC961 (Assigned Protocols),
792 * the urgent pointer points to the last octet
793 * of urgent data. We continue, however,
794 * to consider it to indicate the first octet
795 * of data past the urgent section.
796 * Otherwise, snd_up should be one lower.
798 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
799 tp->t_flags |= TF_FORCE;
800 error = tcp_output(tp);
801 tp->t_flags &= ~TF_FORCE;
803 if (flags & PRUS_EOF) {
805 * Close the send side of the connection after
809 tp = tcp_usrclosed(tp);
812 if (flags & PRUS_MORETOCOME)
813 tp->t_flags |= TF_MORETOCOME;
814 error = tcp_output(tp);
815 if (flags & PRUS_MORETOCOME)
816 tp->t_flags &= ~TF_MORETOCOME;
819 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
820 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
824 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
825 * will sofree() it when we return.
828 tcp_usr_abort(netmsg_t msg)
830 struct socket *so = msg->abort.base.nm_so;
835 COMMON_START(so, inp, 1);
836 tp = tcp_drop(tp, ECONNABORTED);
837 COMMON_END(PRU_ABORT);
841 * Receive out-of-band data.
844 tcp_usr_rcvoob(netmsg_t msg)
846 struct socket *so = msg->rcvoob.base.nm_so;
847 struct mbuf *m = msg->rcvoob.nm_m;
848 int flags = msg->rcvoob.nm_flags;
853 COMMON_START(so, inp, 0);
854 if ((so->so_oobmark == 0 &&
855 (so->so_state & SS_RCVATMARK) == 0) ||
856 so->so_options & SO_OOBINLINE ||
857 tp->t_oobflags & TCPOOB_HADDATA) {
861 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
866 *mtod(m, caddr_t) = tp->t_iobc;
867 if ((flags & MSG_PEEK) == 0)
868 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
869 COMMON_END(PRU_RCVOOB);
872 /* xxx - should be const */
873 struct pr_usrreqs tcp_usrreqs = {
874 .pru_abort = tcp_usr_abort,
875 .pru_accept = tcp_usr_accept,
876 .pru_attach = tcp_usr_attach,
877 .pru_bind = tcp_usr_bind,
878 .pru_connect = tcp_usr_connect,
879 .pru_connect2 = pr_generic_notsupp,
880 .pru_control = in_control_dispatch,
881 .pru_detach = tcp_usr_detach,
882 .pru_disconnect = tcp_usr_disconnect,
883 .pru_listen = tcp_usr_listen,
884 .pru_peeraddr = in_setpeeraddr_dispatch,
885 .pru_rcvd = tcp_usr_rcvd,
886 .pru_rcvoob = tcp_usr_rcvoob,
887 .pru_send = tcp_usr_send,
888 .pru_sense = pru_sense_null,
889 .pru_shutdown = tcp_usr_shutdown,
890 .pru_sockaddr = in_setsockaddr_dispatch,
891 .pru_sosend = sosend,
892 .pru_soreceive = soreceive
896 struct pr_usrreqs tcp6_usrreqs = {
897 .pru_abort = tcp_usr_abort,
898 .pru_accept = tcp6_usr_accept,
899 .pru_attach = tcp_usr_attach,
900 .pru_bind = tcp6_usr_bind,
901 .pru_connect = tcp6_usr_connect,
902 .pru_connect2 = pr_generic_notsupp,
903 .pru_control = in6_control_dispatch,
904 .pru_detach = tcp_usr_detach,
905 .pru_disconnect = tcp_usr_disconnect,
906 .pru_listen = tcp6_usr_listen,
907 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
908 .pru_rcvd = tcp_usr_rcvd,
909 .pru_rcvoob = tcp_usr_rcvoob,
910 .pru_send = tcp_usr_send,
911 .pru_sense = pru_sense_null,
912 .pru_shutdown = tcp_usr_shutdown,
913 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
914 .pru_sosend = sosend,
915 .pru_soreceive = soreceive
920 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
921 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
923 struct inpcb *inp = tp->t_inpcb, *oinp;
924 struct socket *so = inp->inp_socket;
925 struct route *ro = &inp->inp_route;
927 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
928 sin->sin_addr, sin->sin_port,
929 (inp->inp_laddr.s_addr != INADDR_ANY ?
930 inp->inp_laddr : if_sin->sin_addr),
931 inp->inp_lport, 0, NULL);
936 if (inp->inp_laddr.s_addr == INADDR_ANY)
937 inp->inp_laddr = if_sin->sin_addr;
938 inp->inp_faddr = sin->sin_addr;
939 inp->inp_fport = sin->sin_port;
940 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
941 in_pcbinsconnhash(inp);
944 * We are now on the inpcb's owner CPU, if the cached route was
945 * freed because the rtentry's owner CPU is not the current CPU
946 * (e.g. in tcp_connect()), then we try to reallocate it here with
947 * the hope that a rtentry may be cloned from a RTF_PRCLONING
950 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
952 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
953 ro->ro_dst.sa_family = AF_INET;
954 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
955 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
961 * Now that no more errors can occur, change the protocol processing
962 * port to the current thread (which is the correct thread).
964 * Create TCP timer message now; we are on the tcpcb's owner
967 tcp_create_timermsg(tp, &curthread->td_msgport);
970 * Compute window scaling to request. Use a larger scaling then
971 * needed for the initial receive buffer in case the receive buffer
974 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
975 tp->request_r_scale = TCP_MIN_WINSHIFT;
976 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
977 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
979 tp->request_r_scale++;
983 tcpstat.tcps_connattempt++;
984 tp->t_state = TCPS_SYN_SENT;
985 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
986 tp->iss = tcp_new_isn(tp);
989 ssb_appendstream(&so->so_snd, m);
991 if (flags & PRUS_OOB)
992 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
996 * Close the send side of the connection after
997 * the data is sent if flagged.
999 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1001 tp = tcp_usrclosed(tp);
1003 return (tcp_output(tp));
1007 * Common subroutine to open a TCP connection to remote host specified
1008 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1009 * port number if needed. Call in_pcbladdr to do the routing and to choose
1010 * a local host address (interface).
1011 * Initialize connection parameters and enter SYN-SENT state.
1014 tcp_connect(netmsg_t msg)
1016 struct socket *so = msg->connect.base.nm_so;
1017 struct sockaddr *nam = msg->connect.nm_nam;
1018 struct thread *td = msg->connect.nm_td;
1019 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1020 struct sockaddr_in *if_sin;
1028 COMMON_START(so, inp, 0);
1031 * Reconnect our pcb if we have to
1033 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1034 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1035 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1039 * Bind if we have to
1041 if (inp->inp_lport == 0) {
1042 error = in_pcbbind(inp, NULL, td);
1046 so = inp->inp_socket;
1050 * Calculate the correct protocol processing thread. The connect
1051 * operation must run there. Set the forwarding port before we
1052 * forward the message or it will get bounced right back to us.
1054 error = in_pcbladdr(inp, nam, &if_sin, td);
1059 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1060 (inp->inp_laddr.s_addr ?
1061 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1064 if (port != &curthread->td_msgport) {
1065 struct route *ro = &inp->inp_route;
1068 * in_pcbladdr() may have allocated a route entry for us
1069 * on the current CPU, but we need a route entry on the
1070 * inpcb's owner CPU, so free it here.
1072 if (ro->ro_rt != NULL)
1074 bzero(ro, sizeof(*ro));
1077 * We are moving the protocol processing port the socket
1078 * is on, we have to unlink here and re-link on the
1081 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1082 sosetport(so, port);
1083 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1084 msg->connect.base.nm_dispatch = tcp_connect;
1086 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1087 /* msg invalid now */
1091 KKASSERT(so->so_port == &curthread->td_msgport);
1093 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1094 msg->connect.nm_m, sin, if_sin);
1095 msg->connect.nm_m = NULL;
1097 if (msg->connect.nm_m) {
1098 m_freem(msg->connect.nm_m);
1099 msg->connect.nm_m = NULL;
1101 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1102 kfree(msg->connect.nm_nam, M_LWKTMSG);
1103 msg->connect.nm_nam = NULL;
1105 lwkt_replymsg(&msg->connect.base.lmsg, error);
1106 /* msg invalid now */
1112 tcp6_connect(netmsg_t msg)
1115 struct socket *so = msg->connect.base.nm_so;
1116 struct sockaddr *nam = msg->connect.nm_nam;
1117 struct thread *td = msg->connect.nm_td;
1119 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1120 struct in6_addr *addr6;
1126 COMMON_START(so, inp, 0);
1129 * Reconnect our pcb if we have to
1131 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1132 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1133 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1137 * Bind if we have to
1139 if (inp->inp_lport == 0) {
1140 error = in6_pcbbind(inp, NULL, td);
1146 * Cannot simply call in_pcbconnect, because there might be an
1147 * earlier incarnation of this same connection still in
1148 * TIME_WAIT state, creating an ADDRINUSE error.
1150 error = in6_pcbladdr(inp, nam, &addr6, td);
1155 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1157 if (port != &curthread->td_msgport) {
1158 struct route *ro = &inp->inp_route;
1161 * in_pcbladdr() may have allocated a route entry for us
1162 * on the current CPU, but we need a route entry on the
1163 * inpcb's owner CPU, so free it here.
1165 if (ro->ro_rt != NULL)
1167 bzero(ro, sizeof(*ro));
1169 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1170 sosetport(so, port);
1171 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1172 msg->connect.base.nm_dispatch = tcp6_connect;
1174 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1175 /* msg invalid now */
1179 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1180 &msg->connect.nm_m, sin6, addr6);
1181 /* nm_m may still be intact */
1183 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1185 /* msg invalid now */
1187 if (msg->connect.nm_m) {
1188 m_freem(msg->connect.nm_m);
1189 msg->connect.nm_m = NULL;
1191 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1192 kfree(msg->connect.nm_nam, M_LWKTMSG);
1193 msg->connect.nm_nam = NULL;
1195 lwkt_replymsg(&msg->connect.base.lmsg, error);
1196 /* msg invalid now */
1201 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1202 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1204 struct mbuf *m = *mp;
1205 struct inpcb *inp = tp->t_inpcb;
1206 struct socket *so = inp->inp_socket;
1210 * Cannot simply call in_pcbconnect, because there might be an
1211 * earlier incarnation of this same connection still in
1212 * TIME_WAIT state, creating an ADDRINUSE error.
1214 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1215 &sin6->sin6_addr, sin6->sin6_port,
1216 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1217 addr6 : &inp->in6p_laddr),
1218 inp->inp_lport, 0, NULL);
1220 return (EADDRINUSE);
1222 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1223 inp->in6p_laddr = *addr6;
1224 inp->in6p_faddr = sin6->sin6_addr;
1225 inp->inp_fport = sin6->sin6_port;
1226 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1227 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1228 in_pcbinsconnhash(inp);
1231 * Now that no more errors can occur, change the protocol processing
1232 * port to the current thread (which is the correct thread).
1234 * Create TCP timer message now; we are on the tcpcb's owner
1237 tcp_create_timermsg(tp, &curthread->td_msgport);
1239 /* Compute window scaling to request. */
1240 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1241 tp->request_r_scale = TCP_MIN_WINSHIFT;
1242 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1243 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1244 tp->request_r_scale++;
1248 tcpstat.tcps_connattempt++;
1249 tp->t_state = TCPS_SYN_SENT;
1250 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1251 tp->iss = tcp_new_isn(tp);
1252 tcp_sendseqinit(tp);
1254 ssb_appendstream(&so->so_snd, m);
1256 if (flags & PRUS_OOB)
1257 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1261 * Close the send side of the connection after
1262 * the data is sent if flagged.
1264 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1266 tp = tcp_usrclosed(tp);
1268 return (tcp_output(tp));
1274 * The new sockopt interface makes it possible for us to block in the
1275 * copyin/out step (if we take a page fault). Taking a page fault while
1276 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1277 * both now use TSM, there probably isn't any need for this function to
1278 * run in a critical section any more. This needs more examination.)
1281 tcp_ctloutput(netmsg_t msg)
1283 struct socket *so = msg->base.nm_so;
1284 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1285 int error, opt, optval;
1296 if (sopt->sopt_level != IPPROTO_TCP) {
1298 if (INP_CHECK_SOCKAF(so, AF_INET6))
1299 ip6_ctloutput_dispatch(msg);
1303 /* msg invalid now */
1306 tp = intotcpcb(inp);
1308 switch (sopt->sopt_dir) {
1310 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1314 switch (sopt->sopt_name) {
1317 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1318 tp->t_flags |= TF_FASTKEEP;
1319 tcp_timer_keep_activity(tp, 0);
1322 tp->t_flags &= ~TF_FASTKEEP;
1325 #ifdef TCP_SIGNATURE
1326 case TCP_SIGNATURE_ENABLE:
1328 tp->t_flags |= TF_SIGNATURE;
1330 tp->t_flags &= ~TF_SIGNATURE;
1332 #endif /* TCP_SIGNATURE */
1335 switch (sopt->sopt_name) {
1343 opt = 0; /* dead code to fool gcc */
1350 tp->t_flags &= ~opt;
1355 tp->t_flags |= TF_NOPUSH;
1357 tp->t_flags &= ~TF_NOPUSH;
1358 error = tcp_output(tp);
1364 * Must be between 0 and maxseg. If the requested
1365 * maxseg is too small to satisfy the desired minmss,
1366 * pump it up (silently so sysctl modifications of
1367 * minmss do not create unexpected program failures).
1368 * Handle degenerate cases.
1370 if (optval > 0 && optval <= tp->t_maxseg) {
1371 if (optval + 40 < tcp_minmss) {
1372 optval = tcp_minmss - 40;
1376 tp->t_maxseg = optval;
1383 error = ENOPROTOOPT;
1389 switch (sopt->sopt_name) {
1390 #ifdef TCP_SIGNATURE
1391 case TCP_SIGNATURE_ENABLE:
1392 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1394 #endif /* TCP_SIGNATURE */
1396 optval = tp->t_flags & TF_NODELAY;
1399 optval = tp->t_maxseg;
1402 optval = tp->t_flags & TF_NOOPT;
1405 optval = tp->t_flags & TF_NOPUSH;
1408 error = ENOPROTOOPT;
1412 soopt_from_kbuf(sopt, &optval, sizeof optval);
1416 lwkt_replymsg(&msg->lmsg, error);
1420 * tcp_sendspace and tcp_recvspace are the default send and receive window
1421 * sizes, respectively. These are obsolescent (this information should
1422 * be set by the route).
1424 * Use a default that does not require tcp window scaling to be turned
1425 * on. Individual programs or the administrator can increase the default.
1427 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1428 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1429 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1430 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1431 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1432 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1435 * Attach TCP protocol to socket, allocating internet protocol control
1436 * block, tcp control block, bufer space, and entering LISTEN state
1437 * if to accept connections.
1440 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1447 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1450 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1451 lwkt_gettoken(&so->so_rcv.ssb_token);
1452 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1454 lwkt_reltoken(&so->so_rcv.ssb_token);
1458 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1459 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1460 cpu = mycpu->gd_cpuid;
1463 * Set the default port for protocol processing. This will likely
1464 * change when we connect.
1466 error = in_pcballoc(so, &tcbinfo[cpu]);
1472 inp->inp_vflag |= INP_IPV6;
1473 inp->in6p_hops = -1; /* use kernel default */
1477 inp->inp_vflag |= INP_IPV4;
1478 tp = tcp_newtcpcb(inp);
1481 * Make sure the socket is destroyed by the pcbdetach.
1490 sofree(so); /* from ref above */
1493 tp->t_state = TCPS_CLOSED;
1498 * Initiate (or continue) disconnect.
1499 * If embryonic state, just send reset (once).
1500 * If in ``let data drain'' option and linger null, just drop.
1501 * Otherwise (hard), mark socket disconnecting and drop
1502 * current input data; switch states based on user close, and
1503 * send segment to peer (with FIN).
1505 static struct tcpcb *
1506 tcp_disconnect(struct tcpcb *tp)
1508 struct socket *so = tp->t_inpcb->inp_socket;
1510 if (tp->t_state < TCPS_ESTABLISHED) {
1512 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1513 tp = tcp_drop(tp, 0);
1515 lwkt_gettoken(&so->so_rcv.ssb_token);
1516 soisdisconnecting(so);
1517 sbflush(&so->so_rcv.sb);
1518 tp = tcp_usrclosed(tp);
1521 lwkt_reltoken(&so->so_rcv.ssb_token);
1527 * User issued close, and wish to trail through shutdown states:
1528 * if never received SYN, just forget it. If got a SYN from peer,
1529 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1530 * If already got a FIN from peer, then almost done; go to LAST_ACK
1531 * state. In all other cases, have already sent FIN to peer (e.g.
1532 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1533 * for peer to send FIN or not respond to keep-alives, etc.
1534 * We can let the user exit from the close as soon as the FIN is acked.
1536 static struct tcpcb *
1537 tcp_usrclosed(struct tcpcb *tp)
1540 switch (tp->t_state) {
1544 tp->t_state = TCPS_CLOSED;
1549 case TCPS_SYN_RECEIVED:
1550 tp->t_flags |= TF_NEEDFIN;
1553 case TCPS_ESTABLISHED:
1554 tp->t_state = TCPS_FIN_WAIT_1;
1557 case TCPS_CLOSE_WAIT:
1558 tp->t_state = TCPS_LAST_ACK;
1561 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1562 soisdisconnected(tp->t_inpcb->inp_socket);
1563 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1564 if (tp->t_state == TCPS_FIN_WAIT_2) {
1565 tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,