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 int tcp_connect (struct tcpcb *, int flags, struct mbuf *m,
139 struct sockaddr *, struct thread *);
141 static int tcp6_connect (struct tcpcb *, int flags, struct mbuf *m,
142 struct sockaddr *, struct thread *);
143 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
144 struct sockaddr_in6 *sin6,
145 struct in6_addr *addr6);
147 static struct tcpcb *
148 tcp_disconnect (struct tcpcb *);
149 static struct tcpcb *
150 tcp_usrclosed (struct tcpcb *);
153 #define TCPDEBUG0 int ostate = 0
154 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
155 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
156 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
160 #define TCPDEBUG2(req)
164 * TCP attaches to socket via pru_attach(), reserving space,
165 * and an internet control block.
168 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
172 struct tcpcb *tp = 0;
183 error = tcp_attach(so, ai);
187 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
188 so->so_linger = TCP_LINGERTIME;
191 sofree(so); /* from ref above */
192 TCPDEBUG2(PRU_ATTACH);
197 * pru_detach() detaches the TCP protocol from the socket.
198 * If the protocol state is non-embryonic, then can't
199 * do this directly: have to initiate a pru_disconnect(),
200 * which may finish later; embryonic TCB's can just
204 tcp_usr_detach(struct socket *so)
214 * If the inp is already detached it may have been due to an async
215 * close. Just return as if no error occured.
221 * It's possible for the tcpcb (tp) to disconnect from the inp due
222 * to tcp_drop()->tcp_close() being called. This may occur *after*
223 * 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);
234 * Note: ignore_error is non-zero for certain disconnection races
235 * which we want to silently allow, otherwise close() may return
236 * an unexpected error.
238 #define COMMON_START(so, inp, ignore_error) \
244 return (ignore_error ? 0 : EINVAL); \
246 tp = intotcpcb(inp); \
250 #define COMMON_END(req) out: TCPDEBUG2(req); return error; goto out
253 * Give the socket an address.
256 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
261 struct sockaddr_in *sinp;
263 COMMON_START(so, inp, 0);
266 * Must check for multicast addresses and disallow binding
269 sinp = (struct sockaddr_in *)nam;
270 if (sinp->sin_family == AF_INET &&
271 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
272 error = EAFNOSUPPORT;
275 error = in_pcbbind(inp, nam, td);
278 COMMON_END(PRU_BIND);
284 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
289 struct sockaddr_in6 *sin6p;
291 COMMON_START(so, inp, 0);
294 * Must check for multicast addresses and disallow binding
297 sin6p = (struct sockaddr_in6 *)nam;
298 if (sin6p->sin6_family == AF_INET6 &&
299 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
300 error = EAFNOSUPPORT;
303 inp->inp_vflag &= ~INP_IPV4;
304 inp->inp_vflag |= INP_IPV6;
305 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
306 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
307 inp->inp_vflag |= INP_IPV4;
308 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
309 struct sockaddr_in sin;
311 in6_sin6_2_sin(&sin, sin6p);
312 inp->inp_vflag |= INP_IPV4;
313 inp->inp_vflag &= ~INP_IPV6;
314 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
318 error = in6_pcbbind(inp, nam, td);
321 COMMON_END(PRU_BIND);
326 struct netmsg_inswildcard {
327 struct netmsg nm_netmsg;
328 struct inpcb *nm_inp;
329 struct inpcbinfo *nm_pcbinfo;
333 in_pcbinswildcardhash_handler(struct netmsg *msg0)
335 struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
337 in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
338 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
343 * Prepare to accept connections.
346 tcp_usr_listen(struct socket *so, struct thread *td)
355 COMMON_START(so, inp, 0);
356 if (inp->inp_lport == 0) {
357 error = in_pcbbind(inp, NULL, td);
362 tp->t_state = TCPS_LISTEN;
363 tp->tt_msg = NULL; /* Catch any invalid timer usage */
366 * We have to set the flag because we can't have other cpus
367 * messing with our inp's flags.
369 inp->inp_flags |= INP_WILDCARD_MP;
370 for (cpu = 0; cpu < ncpus2; cpu++) {
371 struct netmsg_inswildcard *msg;
373 if (cpu == mycpu->gd_cpuid) {
374 in_pcbinswildcardhash(inp);
378 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
380 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
381 0, in_pcbinswildcardhash_handler);
383 msg->nm_pcbinfo = &tcbinfo[cpu];
384 lwkt_sendmsg(cpu_portfn(cpu), &msg->nm_netmsg.nm_lmsg);
387 in_pcbinswildcardhash(inp);
389 COMMON_END(PRU_LISTEN);
394 tcp6_usr_listen(struct socket *so, struct thread *td)
403 COMMON_START(so, inp, 0);
404 if (inp->inp_lport == 0) {
405 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
406 inp->inp_vflag |= INP_IPV4;
408 inp->inp_vflag &= ~INP_IPV4;
409 error = in6_pcbbind(inp, NULL, td);
412 tp->t_state = TCPS_LISTEN;
415 * We have to set the flag because we can't have other cpus
416 * messing with our inp's flags.
418 inp->inp_flags |= INP_WILDCARD_MP;
419 for (cpu = 0; cpu < ncpus2; cpu++) {
420 struct netmsg_inswildcard *msg;
422 if (cpu == mycpu->gd_cpuid) {
423 in_pcbinswildcardhash(inp);
427 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
429 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
430 0, in_pcbinswildcardhash_handler);
432 msg->nm_pcbinfo = &tcbinfo[cpu];
433 lwkt_sendmsg(cpu_portfn(cpu), &msg->nm_netmsg.nm_lmsg);
436 in_pcbinswildcardhash(inp);
438 COMMON_END(PRU_LISTEN);
443 * Initiate connection to peer.
444 * Create a template for use in transmissions on this connection.
445 * Enter SYN_SENT state, and mark socket as connecting.
446 * Start keep-alive timer, and seed output sequence space.
447 * Send initial segment on connection.
450 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
455 struct sockaddr_in *sinp;
457 COMMON_START(so, inp, 0);
460 * Must disallow TCP ``connections'' to multicast addresses.
462 sinp = (struct sockaddr_in *)nam;
463 if (sinp->sin_family == AF_INET
464 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
465 error = EAFNOSUPPORT;
469 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
470 error = EAFNOSUPPORT; /* IPv6 only jail */
474 if ((error = tcp_connect(tp, 0, NULL, nam, td)) != 0)
476 COMMON_END(PRU_CONNECT);
481 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
486 struct sockaddr_in6 *sin6p;
488 COMMON_START(so, inp, 0);
491 * Must disallow TCP ``connections'' to multicast addresses.
493 sin6p = (struct sockaddr_in6 *)nam;
494 if (sin6p->sin6_family == AF_INET6
495 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
496 error = EAFNOSUPPORT;
500 if (!prison_remote_ip(td, nam)) {
501 error = EAFNOSUPPORT; /* IPv4 only jail */
505 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
506 struct sockaddr_in sin;
508 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
513 in6_sin6_2_sin(&sin, sin6p);
514 inp->inp_vflag |= INP_IPV4;
515 inp->inp_vflag &= ~INP_IPV6;
516 error = tcp_connect(tp, 0, NULL, (struct sockaddr *)&sin, td);
521 inp->inp_vflag &= ~INP_IPV4;
522 inp->inp_vflag |= INP_IPV6;
523 inp->inp_inc.inc_isipv6 = 1;
524 if ((error = tcp6_connect(tp, 0, NULL, nam, td)) != 0)
526 error = tcp_output(tp);
527 COMMON_END(PRU_CONNECT);
532 * Initiate disconnect from peer.
533 * If connection never passed embryonic stage, just drop;
534 * else if don't need to let data drain, then can just drop anyways,
535 * else have to begin TCP shutdown process: mark socket disconnecting,
536 * drain unread data, state switch to reflect user close, and
537 * send segment (e.g. FIN) to peer. Socket will be really disconnected
538 * when peer sends FIN and acks ours.
540 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
543 tcp_usr_disconnect(struct socket *so)
549 COMMON_START(so, inp, 1);
550 tp = tcp_disconnect(tp);
551 COMMON_END(PRU_DISCONNECT);
555 * Accept a connection. Essentially all the work is
556 * done at higher levels; just return the address
557 * of the peer, storing through addr.
560 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
564 struct tcpcb *tp = NULL;
568 if (so->so_state & SS_ISDISCONNECTED) {
569 error = ECONNABORTED;
577 in_setpeeraddr(so, nam);
578 COMMON_END(PRU_ACCEPT);
583 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
587 struct tcpcb *tp = NULL;
592 if (so->so_state & SS_ISDISCONNECTED) {
593 error = ECONNABORTED;
600 in6_mapped_peeraddr(so, nam);
601 COMMON_END(PRU_ACCEPT);
605 * Mark the connection as being incapable of further output.
608 tcp_usr_shutdown(struct socket *so)
614 COMMON_START(so, inp, 0);
616 tp = tcp_usrclosed(tp);
618 error = tcp_output(tp);
619 COMMON_END(PRU_SHUTDOWN);
623 * After a receive, possibly send window update to peer.
626 tcp_usr_rcvd(struct socket *so, int flags)
632 COMMON_START(so, inp, 0);
634 COMMON_END(PRU_RCVD);
638 * Do a send by putting data in output queue and updating urgent
639 * marker if URG set. Possibly send more data. Unlike the other
640 * pru_*() routines, the mbuf chains are our responsibility. We
641 * must either enqueue them or free them. The other pru_* routines
642 * generally are caller-frees.
645 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
646 struct sockaddr *nam, struct mbuf *control, struct thread *td)
660 * OOPS! we lost a race, the TCP session got reset after
661 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
662 * network interrupt in the non-critical section of sosend().
667 error = ECONNRESET; /* XXX EPIPE? */
673 isipv6 = nam && nam->sa_family == AF_INET6;
678 /* TCP doesn't do control messages (rights, creds, etc) */
679 if (control->m_len) {
685 m_freem(control); /* empty control, just free it */
689 * Don't let too much OOB data build up
691 if (flags & PRUS_OOB) {
692 if (ssb_space(&so->so_snd) < -512) {
700 * Do implied connect if not yet connected. Any data sent
701 * with the connect is handled by tcp_connect() and friends.
703 * NOTE! PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
705 if (nam && tp->t_state < TCPS_SYN_SENT) {
708 error = tcp6_connect(tp, flags, m, nam, td);
711 error = tcp_connect(tp, flags, m, nam, td);
713 /* WTF is this doing here? */
714 tp->snd_wnd = TTCP_CLIENT_SND_WND;
721 * Pump the data into the socket.
724 ssb_appendstream(&so->so_snd, m);
725 if (flags & PRUS_OOB) {
727 * According to RFC961 (Assigned Protocols),
728 * the urgent pointer points to the last octet
729 * of urgent data. We continue, however,
730 * to consider it to indicate the first octet
731 * of data past the urgent section.
732 * Otherwise, snd_up should be one lower.
734 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
735 tp->t_flags |= TF_FORCE;
736 error = tcp_output(tp);
737 tp->t_flags &= ~TF_FORCE;
739 if (flags & PRUS_EOF) {
741 * Close the send side of the connection after
745 tp = tcp_usrclosed(tp);
748 if (flags & PRUS_MORETOCOME)
749 tp->t_flags |= TF_MORETOCOME;
750 error = tcp_output(tp);
751 if (flags & PRUS_MORETOCOME)
752 tp->t_flags &= ~TF_MORETOCOME;
755 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
756 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
760 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
761 * will sofree() it when we return.
764 tcp_usr_abort(struct socket *so)
770 COMMON_START(so, inp, 1);
771 tp = tcp_drop(tp, ECONNABORTED);
772 COMMON_END(PRU_ABORT);
776 * Receive out-of-band data.
779 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
785 COMMON_START(so, inp, 0);
786 if ((so->so_oobmark == 0 &&
787 (so->so_state & SS_RCVATMARK) == 0) ||
788 so->so_options & SO_OOBINLINE ||
789 tp->t_oobflags & TCPOOB_HADDATA) {
793 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
798 *mtod(m, caddr_t) = tp->t_iobc;
799 if ((flags & MSG_PEEK) == 0)
800 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
801 COMMON_END(PRU_RCVOOB);
804 /* xxx - should be const */
805 struct pr_usrreqs tcp_usrreqs = {
806 .pru_abort = tcp_usr_abort,
807 .pru_accept = tcp_usr_accept,
808 .pru_attach = tcp_usr_attach,
809 .pru_bind = tcp_usr_bind,
810 .pru_connect = tcp_usr_connect,
811 .pru_connect2 = pru_connect2_notsupp,
812 .pru_control = in_control,
813 .pru_detach = tcp_usr_detach,
814 .pru_disconnect = tcp_usr_disconnect,
815 .pru_listen = tcp_usr_listen,
816 .pru_peeraddr = in_setpeeraddr,
817 .pru_rcvd = tcp_usr_rcvd,
818 .pru_rcvoob = tcp_usr_rcvoob,
819 .pru_send = tcp_usr_send,
820 .pru_sense = pru_sense_null,
821 .pru_shutdown = tcp_usr_shutdown,
822 .pru_sockaddr = in_setsockaddr,
823 .pru_sosend = sosend,
824 .pru_soreceive = soreceive
828 struct pr_usrreqs tcp6_usrreqs = {
829 .pru_abort = tcp_usr_abort,
830 .pru_accept = tcp6_usr_accept,
831 .pru_attach = tcp_usr_attach,
832 .pru_bind = tcp6_usr_bind,
833 .pru_connect = tcp6_usr_connect,
834 .pru_connect2 = pru_connect2_notsupp,
835 .pru_control = in6_control,
836 .pru_detach = tcp_usr_detach,
837 .pru_disconnect = tcp_usr_disconnect,
838 .pru_listen = tcp6_usr_listen,
839 .pru_peeraddr = in6_mapped_peeraddr,
840 .pru_rcvd = tcp_usr_rcvd,
841 .pru_rcvoob = tcp_usr_rcvoob,
842 .pru_send = tcp_usr_send,
843 .pru_sense = pru_sense_null,
844 .pru_shutdown = tcp_usr_shutdown,
845 .pru_sockaddr = in6_mapped_sockaddr,
846 .pru_sosend = sosend,
847 .pru_soreceive = soreceive
852 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
853 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
855 struct inpcb *inp = tp->t_inpcb, *oinp;
856 struct socket *so = inp->inp_socket;
857 struct route *ro = &inp->inp_route;
859 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
860 sin->sin_addr, sin->sin_port,
861 inp->inp_laddr.s_addr != INADDR_ANY ?
862 inp->inp_laddr : if_sin->sin_addr,
863 inp->inp_lport, 0, NULL);
868 if (inp->inp_laddr.s_addr == INADDR_ANY)
869 inp->inp_laddr = if_sin->sin_addr;
870 inp->inp_faddr = sin->sin_addr;
871 inp->inp_fport = sin->sin_port;
872 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
873 in_pcbinsconnhash(inp);
876 * We are now on the inpcb's owner CPU, if the cached route was
877 * freed because the rtentry's owner CPU is not the current CPU
878 * (e.g. in tcp_connect()), then we try to reallocate it here with
879 * the hope that a rtentry may be cloned from a RTF_PRCLONING
882 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
884 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
885 ro->ro_dst.sa_family = AF_INET;
886 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
887 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
893 * Now that no more errors can occur, change the protocol processing
894 * port to the current thread (which is the correct thread).
896 * Create TCP timer message now; we are on the tcpcb's owner
899 sosetport(so, &curthread->td_msgport);
900 tcp_create_timermsg(tp, &curthread->td_msgport);
903 * Compute window scaling to request. Use a larger scaling then
904 * needed for the initial receive buffer in case the receive buffer
907 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
908 tp->request_r_scale = TCP_MIN_WINSHIFT;
909 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
910 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
912 tp->request_r_scale++;
916 tcpstat.tcps_connattempt++;
917 tp->t_state = TCPS_SYN_SENT;
918 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
919 tp->iss = tcp_new_isn(tp);
922 ssb_appendstream(&so->so_snd, m);
924 if (flags & PRUS_OOB)
925 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
929 * Close the send side of the connection after
930 * the data is sent if flagged.
932 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
934 tp = tcp_usrclosed(tp);
936 return (tcp_output(tp));
941 struct netmsg_tcp_connect {
942 struct netmsg nm_netmsg;
944 struct sockaddr_in *nm_sin;
945 struct sockaddr_in *nm_ifsin;
951 tcp_connect_handler(netmsg_t netmsg)
953 struct netmsg_tcp_connect *msg = (void *)netmsg;
956 error = tcp_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
957 msg->nm_sin, msg->nm_ifsin);
958 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
961 struct netmsg_tcp6_connect {
962 struct netmsg nm_netmsg;
964 struct sockaddr_in6 *nm_sin6;
965 struct in6_addr *nm_addr6;
972 tcp6_connect_handler(netmsg_t netmsg)
974 struct netmsg_tcp6_connect *msg = (void *)netmsg;
977 error = tcp6_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
978 msg->nm_sin6, msg->nm_addr6);
979 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
986 * Common subroutine to open a TCP connection to remote host specified
987 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
988 * port number if needed. Call in_pcbladdr to do the routing and to choose
989 * a local host address (interface).
990 * Initialize connection parameters and enter SYN-SENT state.
993 tcp_connect(struct tcpcb *tp, int flags, struct mbuf *m,
994 struct sockaddr *nam, struct thread *td)
996 struct inpcb *inp = tp->t_inpcb;
997 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
998 struct sockaddr_in *if_sin;
1005 * Bind if we have to
1007 if (inp->inp_lport == 0) {
1008 error = in_pcbbind(inp, NULL, td);
1016 * Calculate the correct protocol processing thread. The connect
1017 * operation must run there.
1019 error = in_pcbladdr(inp, nam, &if_sin, td);
1026 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1027 inp->inp_laddr.s_addr ?
1028 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
1031 if (port != &curthread->td_msgport) {
1032 struct netmsg_tcp_connect msg;
1033 struct route *ro = &inp->inp_route;
1036 * in_pcbladdr() may have allocated a route entry for us
1037 * on the current CPU, but we need a route entry on the
1038 * inpcb's owner CPU, so free it here.
1040 if (ro->ro_rt != NULL)
1042 bzero(ro, sizeof(*ro));
1045 * NOTE: We haven't set so->so_port yet do not pass so
1046 * to netmsg_init() or it will be improperly forwarded.
1048 netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1049 0, tcp_connect_handler);
1052 msg.nm_ifsin = if_sin;
1053 msg.nm_flags = flags;
1055 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1057 error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1060 error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1068 tcp6_connect(struct tcpcb *tp, int flags, struct mbuf *m,
1069 struct sockaddr *nam, struct thread *td)
1071 struct inpcb *inp = tp->t_inpcb;
1072 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1073 struct in6_addr *addr6;
1079 if (inp->inp_lport == 0) {
1080 error = in6_pcbbind(inp, NULL, td);
1088 * Cannot simply call in_pcbconnect, because there might be an
1089 * earlier incarnation of this same connection still in
1090 * TIME_WAIT state, creating an ADDRINUSE error.
1092 error = in6_pcbladdr(inp, nam, &addr6, td);
1099 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1101 if (port != &curthread->td_msgport) {
1102 struct netmsg_tcp6_connect msg;
1103 struct route *ro = &inp->inp_route;
1106 * in_pcbladdr() may have allocated a route entry for us
1107 * on the current CPU, but we need a route entry on the
1108 * inpcb's owner CPU, so free it here.
1110 if (ro->ro_rt != NULL)
1112 bzero(ro, sizeof(*ro));
1114 netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1115 0, tcp6_connect_handler);
1118 msg.nm_addr6 = addr6;
1119 msg.nm_flags = flags;
1121 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1123 error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1126 error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1132 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1133 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1135 struct inpcb *inp = tp->t_inpcb;
1136 struct socket *so = inp->inp_socket;
1140 * Cannot simply call in_pcbconnect, because there might be an
1141 * earlier incarnation of this same connection still in
1142 * TIME_WAIT state, creating an ADDRINUSE error.
1144 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1145 &sin6->sin6_addr, sin6->sin6_port,
1146 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1147 addr6 : &inp->in6p_laddr,
1148 inp->inp_lport, 0, NULL);
1151 return (EADDRINUSE);
1153 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1154 inp->in6p_laddr = *addr6;
1155 inp->in6p_faddr = sin6->sin6_addr;
1156 inp->inp_fport = sin6->sin6_port;
1157 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1158 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1159 in_pcbinsconnhash(inp);
1162 * Now that no more errors can occur, change the protocol processing
1163 * port to the current thread (which is the correct thread).
1165 * Create TCP timer message now; we are on the tcpcb's owner
1168 sosetport(so, &curthread->td_msgport);
1169 tcp_create_timermsg(tp, &curthread->td_msgport);
1171 /* Compute window scaling to request. */
1172 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1173 tp->request_r_scale = TCP_MIN_WINSHIFT;
1174 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1175 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1176 tp->request_r_scale++;
1180 tcpstat.tcps_connattempt++;
1181 tp->t_state = TCPS_SYN_SENT;
1182 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1183 tp->iss = tcp_new_isn(tp);
1184 tcp_sendseqinit(tp);
1186 ssb_appendstream(&so->so_snd, m);
1188 if (flags & PRUS_OOB)
1189 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1193 * Close the send side of the connection after
1194 * the data is sent if flagged.
1196 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1198 tp = tcp_usrclosed(tp);
1200 return (tcp_output(tp));
1206 * The new sockopt interface makes it possible for us to block in the
1207 * copyin/out step (if we take a page fault). Taking a page fault while
1208 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1209 * both now use TSM, there probably isn't any need for this function to
1210 * run in a critical section any more. This needs more examination.)
1213 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1215 int error, opt, optval;
1222 return (ECONNRESET);
1224 if (sopt->sopt_level != IPPROTO_TCP) {
1226 if (INP_CHECK_SOCKAF(so, AF_INET6))
1227 error = ip6_ctloutput(so, sopt);
1230 error = ip_ctloutput(so, sopt);
1233 tp = intotcpcb(inp);
1235 switch (sopt->sopt_dir) {
1237 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1241 switch (sopt->sopt_name) {
1242 #ifdef TCP_SIGNATURE
1243 case TCP_SIGNATURE_ENABLE:
1245 tp->t_flags |= TF_SIGNATURE;
1247 tp->t_flags &= ~TF_SIGNATURE;
1249 #endif /* TCP_SIGNATURE */
1252 switch (sopt->sopt_name) {
1260 opt = 0; /* dead code to fool gcc */
1267 tp->t_flags &= ~opt;
1272 tp->t_flags |= TF_NOPUSH;
1274 tp->t_flags &= ~TF_NOPUSH;
1275 error = tcp_output(tp);
1281 * Must be between 0 and maxseg. If the requested
1282 * maxseg is too small to satisfy the desired minmss,
1283 * pump it up (silently so sysctl modifications of
1284 * minmss do not create unexpected program failures).
1285 * Handle degenerate cases.
1287 if (optval > 0 && optval <= tp->t_maxseg) {
1288 if (optval + 40 < tcp_minmss) {
1289 optval = tcp_minmss - 40;
1293 tp->t_maxseg = optval;
1300 error = ENOPROTOOPT;
1306 switch (sopt->sopt_name) {
1307 #ifdef TCP_SIGNATURE
1308 case TCP_SIGNATURE_ENABLE:
1309 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1311 #endif /* TCP_SIGNATURE */
1313 optval = tp->t_flags & TF_NODELAY;
1316 optval = tp->t_maxseg;
1319 optval = tp->t_flags & TF_NOOPT;
1322 optval = tp->t_flags & TF_NOPUSH;
1325 error = ENOPROTOOPT;
1329 soopt_from_kbuf(sopt, &optval, sizeof optval);
1336 * tcp_sendspace and tcp_recvspace are the default send and receive window
1337 * sizes, respectively. These are obsolescent (this information should
1338 * be set by the route).
1340 * Use a default that does not require tcp window scaling to be turned
1341 * on. Individual programs or the administrator can increase the default.
1343 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1344 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1345 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1346 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1347 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1348 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1351 * Attach TCP protocol to socket, allocating
1352 * internet protocol control block, tcp control block,
1353 * bufer space, and entering LISTEN state if to accept connections.
1356 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1363 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1366 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1367 lwkt_gettoken(&so->so_rcv.ssb_token);
1368 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1370 lwkt_reltoken(&so->so_rcv.ssb_token);
1374 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1375 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1376 cpu = mycpu->gd_cpuid;
1377 error = in_pcballoc(so, &tcbinfo[cpu]);
1383 inp->inp_vflag |= INP_IPV6;
1384 inp->in6p_hops = -1; /* use kernel default */
1388 inp->inp_vflag |= INP_IPV4;
1389 tp = tcp_newtcpcb(inp);
1392 * Make sure the socket is destroyed by the pcbdetach.
1401 sofree(so); /* from ref above */
1404 tp->t_state = TCPS_CLOSED;
1405 so->so_port = tcp_soport_attach(so);
1410 * Initiate (or continue) disconnect.
1411 * If embryonic state, just send reset (once).
1412 * If in ``let data drain'' option and linger null, just drop.
1413 * Otherwise (hard), mark socket disconnecting and drop
1414 * current input data; switch states based on user close, and
1415 * send segment to peer (with FIN).
1417 static struct tcpcb *
1418 tcp_disconnect(struct tcpcb *tp)
1420 struct socket *so = tp->t_inpcb->inp_socket;
1422 if (tp->t_state < TCPS_ESTABLISHED) {
1424 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1425 tp = tcp_drop(tp, 0);
1427 lwkt_gettoken(&so->so_rcv.ssb_token);
1428 soisdisconnecting(so);
1429 sbflush(&so->so_rcv.sb);
1430 tp = tcp_usrclosed(tp);
1433 lwkt_reltoken(&so->so_rcv.ssb_token);
1439 * User issued close, and wish to trail through shutdown states:
1440 * if never received SYN, just forget it. If got a SYN from peer,
1441 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1442 * If already got a FIN from peer, then almost done; go to LAST_ACK
1443 * state. In all other cases, have already sent FIN to peer (e.g.
1444 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1445 * for peer to send FIN or not respond to keep-alives, etc.
1446 * We can let the user exit from the close as soon as the FIN is acked.
1448 static struct tcpcb *
1449 tcp_usrclosed(struct tcpcb *tp)
1452 switch (tp->t_state) {
1456 tp->t_state = TCPS_CLOSED;
1461 case TCPS_SYN_RECEIVED:
1462 tp->t_flags |= TF_NEEDFIN;
1465 case TCPS_ESTABLISHED:
1466 tp->t_state = TCPS_FIN_WAIT_1;
1469 case TCPS_CLOSE_WAIT:
1470 tp->t_state = TCPS_LAST_ACK;
1473 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1474 soisdisconnected(tp->t_inpcb->inp_socket);
1475 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1476 if (tp->t_state == TCPS_FIN_WAIT_2) {
1477 tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,