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>
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 int tcp_connect (struct tcpcb *, int flags, struct mbuf *m,
138 struct sockaddr *, struct thread *);
140 static int tcp6_connect (struct tcpcb *, int flags, struct mbuf *m,
141 struct sockaddr *, struct thread *);
142 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
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.
167 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
171 struct tcpcb *tp = 0;
182 error = tcp_attach(so, ai);
186 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
187 so->so_linger = TCP_LINGERTIME;
190 TCPDEBUG2(PRU_ATTACH);
196 * pru_detach() detaches the TCP protocol from the socket.
197 * If the protocol state is non-embryonic, then can't
198 * do this directly: have to initiate a pru_disconnect(),
199 * which may finish later; embryonic TCB's can just
203 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.
223 * It's possible for the tcpcb (tp) to disconnect from the inp due
224 * to tcp_drop()->tcp_close() being called. This may occur *after*
225 * the detach message has been queued so we may find a NULL tp here.
227 if ((tp = intotcpcb(inp)) != NULL) {
229 tp = tcp_disconnect(tp);
230 TCPDEBUG2(PRU_DETACH);
237 * Note: ignore_error is non-zero for certain disconnection races
238 * which we want to silently allow, otherwise close() may return
239 * an unexpected error.
241 #define COMMON_START(so, inp, ignore_error) \
249 return (ignore_error ? 0 : EINVAL); \
251 tp = intotcpcb(inp); \
255 #define COMMON_END(req) out: TCPDEBUG2(req); crit_exit(); return error; goto out
259 * Give the socket an address.
262 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
267 struct sockaddr_in *sinp;
269 COMMON_START(so, inp, 0);
272 * Must check for multicast addresses and disallow binding
275 sinp = (struct sockaddr_in *)nam;
276 if (sinp->sin_family == AF_INET &&
277 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
278 error = EAFNOSUPPORT;
281 error = in_pcbbind(inp, nam, td);
284 COMMON_END(PRU_BIND);
290 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
295 struct sockaddr_in6 *sin6p;
297 COMMON_START(so, inp, 0);
300 * Must check for multicast addresses and disallow binding
303 sin6p = (struct sockaddr_in6 *)nam;
304 if (sin6p->sin6_family == AF_INET6 &&
305 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
306 error = EAFNOSUPPORT;
309 inp->inp_vflag &= ~INP_IPV4;
310 inp->inp_vflag |= INP_IPV6;
311 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
312 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
313 inp->inp_vflag |= INP_IPV4;
314 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
315 struct sockaddr_in sin;
317 in6_sin6_2_sin(&sin, sin6p);
318 inp->inp_vflag |= INP_IPV4;
319 inp->inp_vflag &= ~INP_IPV6;
320 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
324 error = in6_pcbbind(inp, nam, td);
327 COMMON_END(PRU_BIND);
332 struct netmsg_inswildcard {
333 struct netmsg nm_netmsg;
334 struct inpcb *nm_inp;
335 struct inpcbinfo *nm_pcbinfo;
339 in_pcbinswildcardhash_handler(struct netmsg *msg0)
341 struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
343 in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
344 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
349 * Prepare to accept connections.
352 tcp_usr_listen(struct socket *so, struct thread *td)
361 COMMON_START(so, inp, 0);
362 if (inp->inp_lport == 0) {
363 error = in_pcbbind(inp, NULL, td);
368 tp->t_state = TCPS_LISTEN;
369 tp->tt_msg = NULL; /* Catch any invalid timer usage */
372 * We have to set the flag because we can't have other cpus
373 * messing with our inp's flags.
375 inp->inp_flags |= INP_WILDCARD_MP;
376 for (cpu = 0; cpu < ncpus2; cpu++) {
377 struct netmsg_inswildcard *msg;
379 if (cpu == mycpu->gd_cpuid) {
380 in_pcbinswildcardhash(inp);
384 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
386 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
387 0, in_pcbinswildcardhash_handler);
389 msg->nm_pcbinfo = &tcbinfo[cpu];
390 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
393 in_pcbinswildcardhash(inp);
395 COMMON_END(PRU_LISTEN);
400 tcp6_usr_listen(struct socket *so, struct thread *td)
409 COMMON_START(so, inp, 0);
410 if (inp->inp_lport == 0) {
411 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
412 inp->inp_vflag |= INP_IPV4;
414 inp->inp_vflag &= ~INP_IPV4;
415 error = in6_pcbbind(inp, NULL, td);
418 tp->t_state = TCPS_LISTEN;
421 * We have to set the flag because we can't have other cpus
422 * messing with our inp's flags.
424 inp->inp_flags |= INP_WILDCARD_MP;
425 for (cpu = 0; cpu < ncpus2; cpu++) {
426 struct netmsg_inswildcard *msg;
428 if (cpu == mycpu->gd_cpuid) {
429 in_pcbinswildcardhash(inp);
433 msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
435 netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
436 0, in_pcbinswildcardhash_handler);
438 msg->nm_pcbinfo = &tcbinfo[cpu];
439 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
442 in_pcbinswildcardhash(inp);
444 COMMON_END(PRU_LISTEN);
449 * Initiate connection to peer.
450 * Create a template for use in transmissions on this connection.
451 * Enter SYN_SENT state, and mark socket as connecting.
452 * Start keep-alive timer, and seed output sequence space.
453 * Send initial segment on connection.
456 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
461 struct sockaddr_in *sinp;
463 COMMON_START(so, inp, 0);
466 * Must disallow TCP ``connections'' to multicast addresses.
468 sinp = (struct sockaddr_in *)nam;
469 if (sinp->sin_family == AF_INET
470 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
471 error = EAFNOSUPPORT;
475 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
476 error = EAFNOSUPPORT; /* IPv6 only jail */
480 if ((error = tcp_connect(tp, 0, NULL, nam, td)) != 0)
482 COMMON_END(PRU_CONNECT);
487 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
492 struct sockaddr_in6 *sin6p;
494 COMMON_START(so, inp, 0);
497 * Must disallow TCP ``connections'' to multicast addresses.
499 sin6p = (struct sockaddr_in6 *)nam;
500 if (sin6p->sin6_family == AF_INET6
501 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
502 error = EAFNOSUPPORT;
506 if (!prison_remote_ip(td, nam)) {
507 error = EAFNOSUPPORT; /* IPv4 only jail */
511 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
512 struct sockaddr_in sin;
514 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
519 in6_sin6_2_sin(&sin, sin6p);
520 inp->inp_vflag |= INP_IPV4;
521 inp->inp_vflag &= ~INP_IPV6;
522 error = tcp_connect(tp, 0, NULL, (struct sockaddr *)&sin, td);
527 inp->inp_vflag &= ~INP_IPV4;
528 inp->inp_vflag |= INP_IPV6;
529 inp->inp_inc.inc_isipv6 = 1;
530 if ((error = tcp6_connect(tp, 0, NULL, nam, td)) != 0)
532 error = tcp_output(tp);
533 COMMON_END(PRU_CONNECT);
538 * Initiate disconnect from peer.
539 * If connection never passed embryonic stage, just drop;
540 * else if don't need to let data drain, then can just drop anyways,
541 * else have to begin TCP shutdown process: mark socket disconnecting,
542 * drain unread data, state switch to reflect user close, and
543 * send segment (e.g. FIN) to peer. Socket will be really disconnected
544 * when peer sends FIN and acks ours.
546 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
549 tcp_usr_disconnect(struct socket *so)
555 COMMON_START(so, inp, 1);
556 tp = tcp_disconnect(tp);
557 COMMON_END(PRU_DISCONNECT);
561 * Accept a connection. Essentially all the work is
562 * done at higher levels; just return the address
563 * of the peer, storing through addr.
566 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
570 struct tcpcb *tp = NULL;
575 if (so->so_state & SS_ISDISCONNECTED) {
576 error = ECONNABORTED;
585 in_setpeeraddr(so, nam);
586 COMMON_END(PRU_ACCEPT);
591 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
595 struct tcpcb *tp = NULL;
601 if (so->so_state & SS_ISDISCONNECTED) {
602 error = ECONNABORTED;
611 in6_mapped_peeraddr(so, nam);
612 COMMON_END(PRU_ACCEPT);
616 * Mark the connection as being incapable of further output.
619 tcp_usr_shutdown(struct socket *so)
625 COMMON_START(so, inp, 0);
627 tp = tcp_usrclosed(tp);
629 error = tcp_output(tp);
630 COMMON_END(PRU_SHUTDOWN);
634 * After a receive, possibly send window update to peer.
637 tcp_usr_rcvd(struct socket *so, int flags)
643 COMMON_START(so, inp, 0);
645 COMMON_END(PRU_RCVD);
649 * Do a send by putting data in output queue and updating urgent
650 * marker if URG set. Possibly send more data. Unlike the other
651 * pru_*() routines, the mbuf chains are our responsibility. We
652 * must either enqueue them or free them. The other pru_* routines
653 * generally are caller-frees.
656 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
657 struct sockaddr *nam, struct mbuf *control, struct thread *td)
672 * OOPS! we lost a race, the TCP session got reset after
673 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
674 * network interrupt in the non-critical section of sosend().
679 error = ECONNRESET; /* XXX EPIPE? */
685 isipv6 = nam && nam->sa_family == AF_INET6;
690 /* TCP doesn't do control messages (rights, creds, etc) */
691 if (control->m_len) {
697 m_freem(control); /* empty control, just free it */
701 * Don't let too much OOB data build up
703 if (flags & PRUS_OOB) {
704 if (ssb_space(&so->so_snd) < -512) {
712 * Do implied connect if not yet connected. Any data sent
713 * with the connect is handled by tcp_connect() and friends.
715 * NOTE! PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
717 if (nam && tp->t_state < TCPS_SYN_SENT) {
720 error = tcp6_connect(tp, flags, m, nam, td);
723 error = tcp_connect(tp, flags, m, nam, td);
725 /* WTF is this doing here? */
726 tp->snd_wnd = TTCP_CLIENT_SND_WND;
733 * Pump the data into the socket.
736 ssb_appendstream(&so->so_snd, m);
737 if (flags & PRUS_OOB) {
739 * According to RFC961 (Assigned Protocols),
740 * the urgent pointer points to the last octet
741 * of urgent data. We continue, however,
742 * to consider it to indicate the first octet
743 * of data past the urgent section.
744 * Otherwise, snd_up should be one lower.
746 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
747 tp->t_flags |= TF_FORCE;
748 error = tcp_output(tp);
749 tp->t_flags &= ~TF_FORCE;
751 if (flags & PRUS_EOF) {
753 * Close the send side of the connection after
757 tp = tcp_usrclosed(tp);
760 if (flags & PRUS_MORETOCOME)
761 tp->t_flags |= TF_MORETOCOME;
762 error = tcp_output(tp);
763 if (flags & PRUS_MORETOCOME)
764 tp->t_flags &= ~TF_MORETOCOME;
767 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
768 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
775 tcp_usr_abort(struct socket *so)
781 COMMON_START(so, inp, 1);
782 tp = tcp_drop(tp, ECONNABORTED);
783 COMMON_END(PRU_ABORT);
787 * Receive out-of-band data.
790 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
796 COMMON_START(so, inp, 0);
797 if ((so->so_oobmark == 0 &&
798 (so->so_state & SS_RCVATMARK) == 0) ||
799 so->so_options & SO_OOBINLINE ||
800 tp->t_oobflags & TCPOOB_HADDATA) {
804 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
809 *mtod(m, caddr_t) = tp->t_iobc;
810 if ((flags & MSG_PEEK) == 0)
811 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
812 COMMON_END(PRU_RCVOOB);
815 /* xxx - should be const */
816 struct pr_usrreqs tcp_usrreqs = {
817 .pru_abort = tcp_usr_abort,
818 .pru_accept = tcp_usr_accept,
819 .pru_attach = tcp_usr_attach,
820 .pru_bind = tcp_usr_bind,
821 .pru_connect = tcp_usr_connect,
822 .pru_connect2 = pru_connect2_notsupp,
823 .pru_control = in_control,
824 .pru_detach = tcp_usr_detach,
825 .pru_disconnect = tcp_usr_disconnect,
826 .pru_listen = tcp_usr_listen,
827 .pru_peeraddr = in_setpeeraddr,
828 .pru_rcvd = tcp_usr_rcvd,
829 .pru_rcvoob = tcp_usr_rcvoob,
830 .pru_send = tcp_usr_send,
831 .pru_sense = pru_sense_null,
832 .pru_shutdown = tcp_usr_shutdown,
833 .pru_sockaddr = in_setsockaddr,
834 .pru_sosend = sosend,
835 .pru_soreceive = soreceive
839 struct pr_usrreqs tcp6_usrreqs = {
840 .pru_abort = tcp_usr_abort,
841 .pru_accept = tcp6_usr_accept,
842 .pru_attach = tcp_usr_attach,
843 .pru_bind = tcp6_usr_bind,
844 .pru_connect = tcp6_usr_connect,
845 .pru_connect2 = pru_connect2_notsupp,
846 .pru_control = in6_control,
847 .pru_detach = tcp_usr_detach,
848 .pru_disconnect = tcp_usr_disconnect,
849 .pru_listen = tcp6_usr_listen,
850 .pru_peeraddr = in6_mapped_peeraddr,
851 .pru_rcvd = tcp_usr_rcvd,
852 .pru_rcvoob = tcp_usr_rcvoob,
853 .pru_send = tcp_usr_send,
854 .pru_sense = pru_sense_null,
855 .pru_shutdown = tcp_usr_shutdown,
856 .pru_sockaddr = in6_mapped_sockaddr,
857 .pru_sosend = sosend,
858 .pru_soreceive = soreceive
863 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
864 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
866 struct inpcb *inp = tp->t_inpcb, *oinp;
867 struct socket *so = inp->inp_socket;
868 struct route *ro = &inp->inp_route;
870 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
871 sin->sin_addr, sin->sin_port,
872 inp->inp_laddr.s_addr != INADDR_ANY ?
873 inp->inp_laddr : if_sin->sin_addr,
874 inp->inp_lport, 0, NULL);
879 if (inp->inp_laddr.s_addr == INADDR_ANY)
880 inp->inp_laddr = if_sin->sin_addr;
881 inp->inp_faddr = sin->sin_addr;
882 inp->inp_fport = sin->sin_port;
883 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
884 in_pcbinsconnhash(inp);
887 * We are now on the inpcb's owner CPU, if the cached route was
888 * freed because the rtentry's owner CPU is not the current CPU
889 * (e.g. in tcp_connect()), then we try to reallocate it here with
890 * the hope that a rtentry may be cloned from a RTF_PRCLONING
893 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
895 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
896 ro->ro_dst.sa_family = AF_INET;
897 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
898 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
904 * Now that no more errors can occur, change the protocol processing
905 * port to the current thread (which is the correct thread).
907 * Create TCP timer message now; we are on the tcpcb's owner
910 sosetport(so, &curthread->td_msgport);
911 tcp_create_timermsg(tp, &curthread->td_msgport);
914 * Compute window scaling to request. Use a larger scaling then
915 * needed for the initial receive buffer in case the receive buffer
918 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
919 tp->request_r_scale = TCP_MIN_WINSHIFT;
920 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
921 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
923 tp->request_r_scale++;
927 tcpstat.tcps_connattempt++;
928 tp->t_state = TCPS_SYN_SENT;
929 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
930 tp->iss = tcp_new_isn(tp);
933 ssb_appendstream(&so->so_snd, m);
935 if (flags & PRUS_OOB)
936 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
940 * Close the send side of the connection after
941 * the data is sent if flagged.
943 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
945 tp = tcp_usrclosed(tp);
947 return (tcp_output(tp));
952 struct netmsg_tcp_connect {
953 struct netmsg nm_netmsg;
955 struct sockaddr_in *nm_sin;
956 struct sockaddr_in *nm_ifsin;
962 tcp_connect_handler(netmsg_t netmsg)
964 struct netmsg_tcp_connect *msg = (void *)netmsg;
967 error = tcp_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
968 msg->nm_sin, msg->nm_ifsin);
969 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
972 struct netmsg_tcp6_connect {
973 struct netmsg nm_netmsg;
975 struct sockaddr_in6 *nm_sin6;
976 struct in6_addr *nm_addr6;
983 tcp6_connect_handler(netmsg_t netmsg)
985 struct netmsg_tcp6_connect *msg = (void *)netmsg;
988 error = tcp6_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
989 msg->nm_sin6, msg->nm_addr6);
990 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
997 * Common subroutine to open a TCP connection to remote host specified
998 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
999 * port number if needed. Call in_pcbladdr to do the routing and to choose
1000 * a local host address (interface).
1001 * Initialize connection parameters and enter SYN-SENT state.
1004 tcp_connect(struct tcpcb *tp, int flags, struct mbuf *m,
1005 struct sockaddr *nam, struct thread *td)
1007 struct inpcb *inp = tp->t_inpcb;
1008 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1009 struct sockaddr_in *if_sin;
1016 * Bind if we have to
1018 if (inp->inp_lport == 0) {
1019 error = in_pcbbind(inp, NULL, td);
1027 * Calculate the correct protocol processing thread. The connect
1028 * operation must run there.
1030 error = in_pcbladdr(inp, nam, &if_sin, td);
1037 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1038 inp->inp_laddr.s_addr ?
1039 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
1042 if (port != &curthread->td_msgport) {
1043 struct netmsg_tcp_connect msg;
1044 struct route *ro = &inp->inp_route;
1047 * in_pcbladdr() may have allocated a route entry for us
1048 * on the current CPU, but we need a route entry on the
1049 * inpcb's owner CPU, so free it here.
1051 if (ro->ro_rt != NULL)
1053 bzero(ro, sizeof(*ro));
1056 * NOTE: We haven't set so->so_port yet do not pass so
1057 * to netmsg_init() or it will be improperly forwarded.
1059 netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1060 0, tcp_connect_handler);
1063 msg.nm_ifsin = if_sin;
1064 msg.nm_flags = flags;
1066 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1068 error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1071 error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1079 tcp6_connect(struct tcpcb *tp, int flags, struct mbuf *m,
1080 struct sockaddr *nam, struct thread *td)
1082 struct inpcb *inp = tp->t_inpcb;
1083 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1084 struct in6_addr *addr6;
1090 if (inp->inp_lport == 0) {
1091 error = in6_pcbbind(inp, NULL, td);
1099 * Cannot simply call in_pcbconnect, because there might be an
1100 * earlier incarnation of this same connection still in
1101 * TIME_WAIT state, creating an ADDRINUSE error.
1103 error = in6_pcbladdr(inp, nam, &addr6, td);
1110 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1112 if (port != &curthread->td_msgport) {
1113 struct netmsg_tcp6_connect msg;
1114 struct route *ro = &inp->inp_route;
1117 * in_pcbladdr() may have allocated a route entry for us
1118 * on the current CPU, but we need a route entry on the
1119 * inpcb's owner CPU, so free it here.
1121 if (ro->ro_rt != NULL)
1123 bzero(ro, sizeof(*ro));
1125 netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1126 0, tcp6_connect_handler);
1129 msg.nm_addr6 = addr6;
1130 msg.nm_flags = flags;
1132 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1134 error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1137 error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1143 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1144 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1146 struct inpcb *inp = tp->t_inpcb;
1147 struct socket *so = inp->inp_socket;
1151 * Cannot simply call in_pcbconnect, because there might be an
1152 * earlier incarnation of this same connection still in
1153 * TIME_WAIT state, creating an ADDRINUSE error.
1155 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1156 &sin6->sin6_addr, sin6->sin6_port,
1157 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1158 addr6 : &inp->in6p_laddr,
1159 inp->inp_lport, 0, NULL);
1162 return (EADDRINUSE);
1164 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1165 inp->in6p_laddr = *addr6;
1166 inp->in6p_faddr = sin6->sin6_addr;
1167 inp->inp_fport = sin6->sin6_port;
1168 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1169 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1170 in_pcbinsconnhash(inp);
1173 * Now that no more errors can occur, change the protocol processing
1174 * port to the current thread (which is the correct thread).
1176 * Create TCP timer message now; we are on the tcpcb's owner
1179 sosetport(so, &curthread->td_msgport);
1180 tcp_create_timermsg(tp, &curthread->td_msgport);
1182 /* Compute window scaling to request. */
1183 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1184 tp->request_r_scale = TCP_MIN_WINSHIFT;
1185 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1186 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1187 tp->request_r_scale++;
1191 tcpstat.tcps_connattempt++;
1192 tp->t_state = TCPS_SYN_SENT;
1193 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1194 tp->iss = tcp_new_isn(tp);
1195 tcp_sendseqinit(tp);
1197 ssb_appendstream(&so->so_snd, m);
1199 if (flags & PRUS_OOB)
1200 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1204 * Close the send side of the connection after
1205 * the data is sent if flagged.
1207 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1209 tp = tcp_usrclosed(tp);
1211 return (tcp_output(tp));
1217 * The new sockopt interface makes it possible for us to block in the
1218 * copyin/out step (if we take a page fault). Taking a page fault while
1219 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1220 * both now use TSM, there probably isn't any need for this function to
1221 * run in a critical section any more. This needs more examination.)
1224 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1226 int error, opt, optval;
1231 crit_enter(); /* XXX */
1235 return (ECONNRESET);
1237 if (sopt->sopt_level != IPPROTO_TCP) {
1239 if (INP_CHECK_SOCKAF(so, AF_INET6))
1240 error = ip6_ctloutput(so, sopt);
1243 error = ip_ctloutput(so, sopt);
1247 tp = intotcpcb(inp);
1249 switch (sopt->sopt_dir) {
1251 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1255 switch (sopt->sopt_name) {
1256 #ifdef TCP_SIGNATURE
1257 case TCP_SIGNATURE_ENABLE:
1259 tp->t_flags |= TF_SIGNATURE;
1261 tp->t_flags &= ~TF_SIGNATURE;
1263 #endif /* TCP_SIGNATURE */
1266 switch (sopt->sopt_name) {
1274 opt = 0; /* dead code to fool gcc */
1281 tp->t_flags &= ~opt;
1286 tp->t_flags |= TF_NOPUSH;
1288 tp->t_flags &= ~TF_NOPUSH;
1289 error = tcp_output(tp);
1295 * Must be between 0 and maxseg. If the requested
1296 * maxseg is too small to satisfy the desired minmss,
1297 * pump it up (silently so sysctl modifications of
1298 * minmss do not create unexpected program failures).
1299 * Handle degenerate cases.
1301 if (optval > 0 && optval <= tp->t_maxseg) {
1302 if (optval + 40 < tcp_minmss) {
1303 optval = tcp_minmss - 40;
1307 tp->t_maxseg = optval;
1314 error = ENOPROTOOPT;
1320 switch (sopt->sopt_name) {
1321 #ifdef TCP_SIGNATURE
1322 case TCP_SIGNATURE_ENABLE:
1323 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1325 #endif /* TCP_SIGNATURE */
1327 optval = tp->t_flags & TF_NODELAY;
1330 optval = tp->t_maxseg;
1333 optval = tp->t_flags & TF_NOOPT;
1336 optval = tp->t_flags & TF_NOPUSH;
1339 error = ENOPROTOOPT;
1343 soopt_from_kbuf(sopt, &optval, sizeof optval);
1351 * tcp_sendspace and tcp_recvspace are the default send and receive window
1352 * sizes, respectively. These are obsolescent (this information should
1353 * be set by the route).
1355 * Use a default that does not require tcp window scaling to be turned
1356 * on. Individual programs or the administrator can increase the default.
1358 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1359 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1360 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1361 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1362 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1363 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1366 * Attach TCP protocol to socket, allocating
1367 * internet protocol control block, tcp control block,
1368 * bufer space, and entering LISTEN state if to accept connections.
1371 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1378 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1381 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1382 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1387 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1388 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1389 cpu = mycpu->gd_cpuid;
1390 error = in_pcballoc(so, &tcbinfo[cpu]);
1396 inp->inp_vflag |= INP_IPV6;
1397 inp->in6p_hops = -1; /* use kernel default */
1401 inp->inp_vflag |= INP_IPV4;
1402 tp = tcp_newtcpcb(inp);
1404 int nofd = so->so_state & SS_NOFDREF; /* XXX */
1406 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */
1413 so->so_state |= nofd;
1416 tp->t_state = TCPS_CLOSED;
1417 so->so_port = tcp_soport_attach(so);
1422 * Initiate (or continue) disconnect.
1423 * If embryonic state, just send reset (once).
1424 * If in ``let data drain'' option and linger null, just drop.
1425 * Otherwise (hard), mark socket disconnecting and drop
1426 * current input data; switch states based on user close, and
1427 * send segment to peer (with FIN).
1429 static struct tcpcb *
1430 tcp_disconnect(struct tcpcb *tp)
1432 struct socket *so = tp->t_inpcb->inp_socket;
1434 if (tp->t_state < TCPS_ESTABLISHED)
1436 else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1437 tp = tcp_drop(tp, 0);
1439 soisdisconnecting(so);
1440 sbflush(&so->so_rcv.sb);
1441 tp = tcp_usrclosed(tp);
1449 * User issued close, and wish to trail through shutdown states:
1450 * if never received SYN, just forget it. If got a SYN from peer,
1451 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1452 * If already got a FIN from peer, then almost done; go to LAST_ACK
1453 * state. In all other cases, have already sent FIN to peer (e.g.
1454 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1455 * for peer to send FIN or not respond to keep-alives, etc.
1456 * We can let the user exit from the close as soon as the FIN is acked.
1458 static struct tcpcb *
1459 tcp_usrclosed(struct tcpcb *tp)
1462 switch (tp->t_state) {
1466 tp->t_state = TCPS_CLOSED;
1471 case TCPS_SYN_RECEIVED:
1472 tp->t_flags |= TF_NEEDFIN;
1475 case TCPS_ESTABLISHED:
1476 tp->t_state = TCPS_FIN_WAIT_1;
1479 case TCPS_CLOSE_WAIT:
1480 tp->t_state = TCPS_LAST_ACK;
1483 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1484 soisdisconnected(tp->t_inpcb->inp_socket);
1485 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1486 if (tp->t_state == TCPS_FIN_WAIT_2) {
1487 tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,