2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1993
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
67 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
70 #include "opt_ipsec.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/sysctl.h>
80 #include <sys/globaldata.h>
81 #include <sys/thread.h>
85 #include <sys/domain.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/socketops.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>
101 #include <net/netisr2.h>
103 #include <netinet/in.h>
104 #include <netinet/in_systm.h>
106 #include <netinet/ip6.h>
108 #include <netinet/in_pcb.h>
110 #include <netinet6/in6_pcb.h>
112 #include <netinet/in_var.h>
113 #include <netinet/ip_var.h>
115 #include <netinet6/ip6_var.h>
116 #include <netinet6/tcp6_var.h>
118 #include <netinet/tcp.h>
119 #include <netinet/tcp_fsm.h>
120 #include <netinet/tcp_seq.h>
121 #include <netinet/tcp_timer.h>
122 #include <netinet/tcp_timer2.h>
123 #include <netinet/tcp_var.h>
124 #include <netinet/tcpip.h>
126 #include <netinet/tcp_debug.h>
130 #include <netinet6/ipsec.h>
134 * TCP protocol interface to socket abstraction.
136 extern char *tcpstates[]; /* XXX ??? */
138 static int tcp_attach (struct socket *, struct pru_attach_info *);
139 static void tcp_connect (netmsg_t msg);
141 static void tcp6_connect (netmsg_t msg);
142 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
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)
163 static int tcp_lport_extension = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
165 &tcp_lport_extension, 0, "");
168 * For some ill optimized programs, which try to use TCP_NOPUSH
169 * to improve performance, will have small amount of data sits
170 * in the sending buffer. These small amount of data will _not_
171 * be pushed into the network until more data are written into
172 * the socket or the socket write side is shutdown.
174 static int tcp_disable_nopush = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
176 &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
179 * TCP attaches to socket via pru_attach(), reserving space,
180 * and an internet control block. This is likely occuring on
181 * cpu0 and may have to move later when we bind/connect.
184 tcp_usr_attach(netmsg_t msg)
186 struct socket *so = msg->base.nm_so;
187 struct pru_attach_info *ai = msg->attach.nm_ai;
190 struct tcpcb *tp = NULL;
201 error = tcp_attach(so, ai);
205 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
206 so->so_linger = TCP_LINGERTIME;
209 sofree(so); /* from ref above */
210 TCPDEBUG2(PRU_ATTACH);
211 lwkt_replymsg(&msg->lmsg, error);
215 * pru_detach() detaches the TCP protocol from the socket.
216 * If the protocol state is non-embryonic, then can't
217 * do this directly: have to initiate a pru_disconnect(),
218 * which may finish later; embryonic TCB's can just
222 tcp_usr_detach(netmsg_t msg)
224 struct socket *so = msg->base.nm_so;
233 * If the inp is already detached it may have been due to an async
234 * close. Just return as if no error occured.
236 * It's possible for the tcpcb (tp) to disconnect from the inp due
237 * to tcp_drop()->tcp_close() being called. This may occur *after*
238 * the detach message has been queued so we may find a NULL tp here.
241 if ((tp = intotcpcb(inp)) != NULL) {
243 tp = tcp_disconnect(tp);
244 TCPDEBUG2(PRU_DETACH);
247 lwkt_replymsg(&msg->lmsg, error);
251 * NOTE: ignore_error is non-zero for certain disconnection races
252 * which we want to silently allow, otherwise close() may return
253 * an unexpected error.
255 * NOTE: The variables (msg) and (tp) are assumed.
257 #define COMMON_START(so, inp, ignore_error) \
263 error = ignore_error ? 0 : EINVAL; \
267 tp = intotcpcb(inp); \
271 #define COMMON_END1(req, noreply) \
275 lwkt_replymsg(&msg->lmsg, error); \
279 #define COMMON_END(req) COMMON_END1((req), 0)
282 * Give the socket an address.
285 tcp_usr_bind(netmsg_t msg)
287 struct socket *so = msg->bind.base.nm_so;
288 struct sockaddr *nam = msg->bind.nm_nam;
289 struct thread *td = msg->bind.nm_td;
293 struct sockaddr_in *sinp;
295 COMMON_START(so, inp, 0);
298 * Must check for multicast addresses and disallow binding
301 sinp = (struct sockaddr_in *)nam;
302 if (sinp->sin_family == AF_INET &&
303 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
304 error = EAFNOSUPPORT;
307 error = in_pcbbind(inp, nam, td);
310 COMMON_END(PRU_BIND);
317 tcp6_usr_bind(netmsg_t msg)
319 struct socket *so = msg->bind.base.nm_so;
320 struct sockaddr *nam = msg->bind.nm_nam;
321 struct thread *td = msg->bind.nm_td;
325 struct sockaddr_in6 *sin6p;
327 COMMON_START(so, inp, 0);
330 * Must check for multicast addresses and disallow binding
333 sin6p = (struct sockaddr_in6 *)nam;
334 if (sin6p->sin6_family == AF_INET6 &&
335 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
336 error = EAFNOSUPPORT;
339 inp->inp_vflag &= ~INP_IPV4;
340 inp->inp_vflag |= INP_IPV6;
341 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
342 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
343 inp->inp_vflag |= INP_IPV4;
344 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
345 struct sockaddr_in sin;
347 in6_sin6_2_sin(&sin, sin6p);
348 inp->inp_vflag |= INP_IPV4;
349 inp->inp_vflag &= ~INP_IPV6;
350 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
354 error = in6_pcbbind(inp, nam, td);
357 COMMON_END(PRU_BIND);
361 struct netmsg_inswildcard {
362 struct netmsg_base base;
363 struct inpcb *nm_inp;
367 in_pcbinswildcardhash_handler(netmsg_t msg)
369 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
370 int cpu = mycpuid, nextcpu;
372 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
375 if (nextcpu < ncpus2)
376 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
378 lwkt_replymsg(&nm->base.lmsg, 0);
382 * Prepare to accept connections.
385 tcp_usr_listen(netmsg_t msg)
387 struct socket *so = msg->listen.base.nm_so;
388 struct thread *td = msg->listen.nm_td;
392 struct netmsg_inswildcard nm;
394 COMMON_START(so, inp, 0);
396 if (tp->t_flags & TF_LISTEN)
399 if (inp->inp_lport == 0) {
400 error = in_pcbbind(inp, NULL, td);
405 tp->t_state = TCPS_LISTEN;
406 tp->t_flags |= TF_LISTEN;
407 tp->tt_msg = NULL; /* Catch any invalid timer usage */
411 * We have to set the flag because we can't have other cpus
412 * messing with our inp's flags.
414 KASSERT(!(inp->inp_flags & INP_CONNECTED),
415 ("already on connhash"));
416 KASSERT(!(inp->inp_flags & INP_WILDCARD),
417 ("already on wildcardhash"));
418 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
419 ("already on MP wildcardhash"));
420 inp->inp_flags |= INP_WILDCARD_MP;
422 KKASSERT(so->so_port == netisr_cpuport(0));
423 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
424 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
426 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
427 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
429 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
431 in_pcbinswildcardhash(inp);
432 COMMON_END(PRU_LISTEN);
438 tcp6_usr_listen(netmsg_t msg)
440 struct socket *so = msg->listen.base.nm_so;
441 struct thread *td = msg->listen.nm_td;
445 struct netmsg_inswildcard nm;
447 COMMON_START(so, inp, 0);
449 if (tp->t_flags & TF_LISTEN)
452 if (inp->inp_lport == 0) {
453 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
454 inp->inp_vflag |= INP_IPV4;
456 inp->inp_vflag &= ~INP_IPV4;
457 error = in6_pcbbind(inp, NULL, td);
462 tp->t_state = TCPS_LISTEN;
463 tp->t_flags |= TF_LISTEN;
464 tp->tt_msg = NULL; /* Catch any invalid timer usage */
468 * We have to set the flag because we can't have other cpus
469 * messing with our inp's flags.
471 KASSERT(!(inp->inp_flags & INP_CONNECTED),
472 ("already on connhash"));
473 KASSERT(!(inp->inp_flags & INP_WILDCARD),
474 ("already on wildcardhash"));
475 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
476 ("already on MP wildcardhash"));
477 inp->inp_flags |= INP_WILDCARD_MP;
479 KKASSERT(so->so_port == netisr_cpuport(0));
480 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
481 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
483 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
484 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
486 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
488 in_pcbinswildcardhash(inp);
489 COMMON_END(PRU_LISTEN);
494 * Initiate connection to peer.
495 * Create a template for use in transmissions on this connection.
496 * Enter SYN_SENT state, and mark socket as connecting.
497 * Start keep-alive timer, and seed output sequence space.
498 * Send initial segment on connection.
501 tcp_usr_connect(netmsg_t msg)
503 struct socket *so = msg->connect.base.nm_so;
504 struct sockaddr *nam = msg->connect.nm_nam;
505 struct thread *td = msg->connect.nm_td;
509 struct sockaddr_in *sinp;
511 COMMON_START(so, inp, 0);
514 * Must disallow TCP ``connections'' to multicast addresses.
516 sinp = (struct sockaddr_in *)nam;
517 if (sinp->sin_family == AF_INET
518 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
519 error = EAFNOSUPPORT;
523 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
524 error = EAFNOSUPPORT; /* IPv6 only jail */
529 /* msg is invalid now */
532 if (msg->connect.nm_m) {
533 m_freem(msg->connect.nm_m);
534 msg->connect.nm_m = NULL;
536 lwkt_replymsg(&msg->lmsg, error);
542 tcp6_usr_connect(netmsg_t msg)
544 struct socket *so = msg->connect.base.nm_so;
545 struct sockaddr *nam = msg->connect.nm_nam;
546 struct thread *td = msg->connect.nm_td;
550 struct sockaddr_in6 *sin6p;
552 COMMON_START(so, inp, 0);
555 * Must disallow TCP ``connections'' to multicast addresses.
557 sin6p = (struct sockaddr_in6 *)nam;
558 if (sin6p->sin6_family == AF_INET6
559 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
560 error = EAFNOSUPPORT;
564 if (!prison_remote_ip(td, nam)) {
565 error = EAFNOSUPPORT; /* IPv4 only jail */
569 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
570 struct sockaddr_in *sinp;
572 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
576 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
577 in6_sin6_2_sin(sinp, sin6p);
578 inp->inp_vflag |= INP_IPV4;
579 inp->inp_vflag &= ~INP_IPV6;
580 msg->connect.nm_nam = (struct sockaddr *)sinp;
581 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
583 /* msg is invalid now */
586 inp->inp_vflag &= ~INP_IPV4;
587 inp->inp_vflag |= INP_IPV6;
588 inp->inp_inc.inc_isipv6 = 1;
590 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
592 /* msg is invalid now */
595 if (msg->connect.nm_m) {
596 m_freem(msg->connect.nm_m);
597 msg->connect.nm_m = NULL;
599 lwkt_replymsg(&msg->lmsg, error);
605 * Initiate disconnect from peer.
606 * If connection never passed embryonic stage, just drop;
607 * else if don't need to let data drain, then can just drop anyways,
608 * else have to begin TCP shutdown process: mark socket disconnecting,
609 * drain unread data, state switch to reflect user close, and
610 * send segment (e.g. FIN) to peer. Socket will be really disconnected
611 * when peer sends FIN and acks ours.
613 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
616 tcp_usr_disconnect(netmsg_t msg)
618 struct socket *so = msg->disconnect.base.nm_so;
623 COMMON_START(so, inp, 1);
624 tp = tcp_disconnect(tp);
625 COMMON_END(PRU_DISCONNECT);
629 * Accept a connection. Essentially all the work is
630 * done at higher levels; just return the address
631 * of the peer, storing through addr.
634 tcp_usr_accept(netmsg_t msg)
636 struct socket *so = msg->accept.base.nm_so;
637 struct sockaddr **nam = msg->accept.nm_nam;
640 struct tcpcb *tp = NULL;
644 if (so->so_state & SS_ISDISCONNECTED) {
645 error = ECONNABORTED;
655 in_setpeeraddr(so, nam);
656 COMMON_END(PRU_ACCEPT);
661 tcp6_usr_accept(netmsg_t msg)
663 struct socket *so = msg->accept.base.nm_so;
664 struct sockaddr **nam = msg->accept.nm_nam;
667 struct tcpcb *tp = NULL;
672 if (so->so_state & SS_ISDISCONNECTED) {
673 error = ECONNABORTED;
682 in6_mapped_peeraddr(so, nam);
683 COMMON_END(PRU_ACCEPT);
687 * Mark the connection as being incapable of further output.
690 tcp_usr_shutdown(netmsg_t msg)
692 struct socket *so = msg->shutdown.base.nm_so;
697 COMMON_START(so, inp, 0);
699 tp = tcp_usrclosed(tp);
701 error = tcp_output(tp);
702 COMMON_END(PRU_SHUTDOWN);
706 * After a receive, possibly send window update to peer.
709 tcp_usr_rcvd(netmsg_t msg)
711 struct socket *so = msg->rcvd.base.nm_so;
712 int error = 0, noreply = 0;
716 COMMON_START(so, inp, 0);
718 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
720 so_async_rcvd_reply(so);
724 COMMON_END1(PRU_RCVD, noreply);
728 * Do a send by putting data in output queue and updating urgent
729 * marker if URG set. Possibly send more data. Unlike the other
730 * pru_*() routines, the mbuf chains are our responsibility. We
731 * must either enqueue them or free them. The other pru_* routines
732 * generally are caller-frees.
735 tcp_usr_send(netmsg_t msg)
737 struct socket *so = msg->send.base.nm_so;
738 int flags = msg->send.nm_flags;
739 struct mbuf *m = msg->send.nm_m;
745 KKASSERT(msg->send.nm_control == NULL);
746 KKASSERT(msg->send.nm_addr == NULL);
747 KKASSERT((flags & PRUS_FREEADDR) == 0);
753 * OOPS! we lost a race, the TCP session got reset after
754 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
755 * network interrupt in the non-critical section of sosend().
758 error = ECONNRESET; /* XXX EPIPE? */
768 * This is no longer necessary, since:
769 * - sosendtcp() has already checked it for us
770 * - It does not work with asynchronized send
774 * Don't let too much OOB data build up
776 if (flags & PRUS_OOB) {
777 if (ssb_space(&so->so_snd) < -512) {
786 * Pump the data into the socket.
789 ssb_appendstream(&so->so_snd, m);
790 if (flags & PRUS_OOB) {
792 * According to RFC961 (Assigned Protocols),
793 * the urgent pointer points to the last octet
794 * of urgent data. We continue, however,
795 * to consider it to indicate the first octet
796 * of data past the urgent section.
797 * Otherwise, snd_up should be one lower.
799 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
800 tp->t_flags |= TF_FORCE;
801 error = tcp_output(tp);
802 tp->t_flags &= ~TF_FORCE;
804 if (flags & PRUS_EOF) {
806 * Close the send side of the connection after
810 tp = tcp_usrclosed(tp);
812 if (tp != NULL && !tcp_output_pending(tp)) {
813 if (flags & PRUS_MORETOCOME)
814 tp->t_flags |= TF_MORETOCOME;
815 error = tcp_output_fair(tp);
816 if (flags & PRUS_MORETOCOME)
817 tp->t_flags &= ~TF_MORETOCOME;
820 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
821 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
822 (flags & PRUS_NOREPLY));
826 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
827 * will sofree() it when we return.
830 tcp_usr_abort(netmsg_t msg)
832 struct socket *so = msg->abort.base.nm_so;
837 COMMON_START(so, inp, 1);
838 tp = tcp_drop(tp, ECONNABORTED);
839 COMMON_END(PRU_ABORT);
843 * Receive out-of-band data.
846 tcp_usr_rcvoob(netmsg_t msg)
848 struct socket *so = msg->rcvoob.base.nm_so;
849 struct mbuf *m = msg->rcvoob.nm_m;
850 int flags = msg->rcvoob.nm_flags;
855 COMMON_START(so, inp, 0);
856 if ((so->so_oobmark == 0 &&
857 (so->so_state & SS_RCVATMARK) == 0) ||
858 so->so_options & SO_OOBINLINE ||
859 tp->t_oobflags & TCPOOB_HADDATA) {
863 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
868 *mtod(m, caddr_t) = tp->t_iobc;
869 if ((flags & MSG_PEEK) == 0)
870 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
871 COMMON_END(PRU_RCVOOB);
875 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
877 in_savefaddr(so, faddr);
882 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
884 in6_mapped_savefaddr(so, faddr);
888 /* xxx - should be const */
889 struct pr_usrreqs tcp_usrreqs = {
890 .pru_abort = tcp_usr_abort,
891 .pru_accept = tcp_usr_accept,
892 .pru_attach = tcp_usr_attach,
893 .pru_bind = tcp_usr_bind,
894 .pru_connect = tcp_usr_connect,
895 .pru_connect2 = pr_generic_notsupp,
896 .pru_control = in_control_dispatch,
897 .pru_detach = tcp_usr_detach,
898 .pru_disconnect = tcp_usr_disconnect,
899 .pru_listen = tcp_usr_listen,
900 .pru_peeraddr = in_setpeeraddr_dispatch,
901 .pru_rcvd = tcp_usr_rcvd,
902 .pru_rcvoob = tcp_usr_rcvoob,
903 .pru_send = tcp_usr_send,
904 .pru_sense = pru_sense_null,
905 .pru_shutdown = tcp_usr_shutdown,
906 .pru_sockaddr = in_setsockaddr_dispatch,
907 .pru_sosend = sosendtcp,
908 .pru_soreceive = sorecvtcp,
909 .pru_savefaddr = tcp_usr_savefaddr
913 struct pr_usrreqs tcp6_usrreqs = {
914 .pru_abort = tcp_usr_abort,
915 .pru_accept = tcp6_usr_accept,
916 .pru_attach = tcp_usr_attach,
917 .pru_bind = tcp6_usr_bind,
918 .pru_connect = tcp6_usr_connect,
919 .pru_connect2 = pr_generic_notsupp,
920 .pru_control = in6_control_dispatch,
921 .pru_detach = tcp_usr_detach,
922 .pru_disconnect = tcp_usr_disconnect,
923 .pru_listen = tcp6_usr_listen,
924 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
925 .pru_rcvd = tcp_usr_rcvd,
926 .pru_rcvoob = tcp_usr_rcvoob,
927 .pru_send = tcp_usr_send,
928 .pru_sense = pru_sense_null,
929 .pru_shutdown = tcp_usr_shutdown,
930 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
931 .pru_sosend = sosendtcp,
932 .pru_soreceive = sorecvtcp,
933 .pru_savefaddr = tcp6_usr_savefaddr
938 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
939 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
941 struct inpcb *inp = tp->t_inpcb, *oinp;
942 struct socket *so = inp->inp_socket;
943 struct route *ro = &inp->inp_route;
945 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
946 sin->sin_addr, sin->sin_port,
947 (inp->inp_laddr.s_addr != INADDR_ANY ?
948 inp->inp_laddr : if_sin->sin_addr),
949 inp->inp_lport, 0, NULL);
954 if (inp->inp_laddr.s_addr == INADDR_ANY)
955 inp->inp_laddr = if_sin->sin_addr;
956 inp->inp_faddr = sin->sin_addr;
957 inp->inp_fport = sin->sin_port;
958 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
959 in_pcbinsconnhash(inp);
962 * We are now on the inpcb's owner CPU, if the cached route was
963 * freed because the rtentry's owner CPU is not the current CPU
964 * (e.g. in tcp_connect()), then we try to reallocate it here with
965 * the hope that a rtentry may be cloned from a RTF_PRCLONING
968 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
970 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
971 ro->ro_dst.sa_family = AF_INET;
972 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
973 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
979 * Now that no more errors can occur, change the protocol processing
980 * port to the current thread (which is the correct thread).
982 * Create TCP timer message now; we are on the tcpcb's owner
985 tcp_create_timermsg(tp, &curthread->td_msgport);
988 * Compute window scaling to request. Use a larger scaling then
989 * needed for the initial receive buffer in case the receive buffer
992 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
993 tp->request_r_scale = TCP_MIN_WINSHIFT;
994 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
995 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
997 tp->request_r_scale++;
1001 tcpstat.tcps_connattempt++;
1002 tp->t_state = TCPS_SYN_SENT;
1003 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1004 tp->iss = tcp_new_isn(tp);
1005 tcp_sendseqinit(tp);
1007 ssb_appendstream(&so->so_snd, m);
1009 if (flags & PRUS_OOB)
1010 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1014 * Close the send side of the connection after
1015 * the data is sent if flagged.
1017 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1019 tp = tcp_usrclosed(tp);
1021 return (tcp_output(tp));
1025 * Common subroutine to open a TCP connection to remote host specified
1026 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1027 * port number if needed. Call in_pcbladdr to do the routing and to choose
1028 * a local host address (interface).
1029 * Initialize connection parameters and enter SYN-SENT state.
1032 tcp_connect(netmsg_t msg)
1034 struct socket *so = msg->connect.base.nm_so;
1035 struct sockaddr *nam = msg->connect.nm_nam;
1036 struct thread *td = msg->connect.nm_td;
1037 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1038 struct sockaddr_in *if_sin;
1041 int error, calc_laddr = 1;
1044 COMMON_START(so, inp, 0);
1047 * Reconnect our pcb if we have to
1049 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1050 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1051 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1055 * Bind if we have to
1057 if (inp->inp_lport == 0) {
1058 if (tcp_lport_extension) {
1059 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1061 error = in_pcbladdr(inp, nam, &if_sin, td);
1064 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1066 error = in_pcbconn_bind(inp, nam, td);
1072 error = in_pcbbind(inp, NULL, td);
1080 * Calculate the correct protocol processing thread. The
1081 * connect operation must run there. Set the forwarding
1082 * port before we forward the message or it will get bounced
1085 error = in_pcbladdr(inp, nam, &if_sin, td);
1089 KKASSERT(inp->inp_socket == so);
1091 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1092 (inp->inp_laddr.s_addr ?
1093 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1096 if (port != &curthread->td_msgport) {
1097 struct route *ro = &inp->inp_route;
1100 * in_pcbladdr() may have allocated a route entry for us
1101 * on the current CPU, but we need a route entry on the
1102 * inpcb's owner CPU, so free it here.
1104 if (ro->ro_rt != NULL)
1106 bzero(ro, sizeof(*ro));
1109 * We are moving the protocol processing port the socket
1110 * is on, we have to unlink here and re-link on the
1113 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1114 sosetport(so, port);
1115 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1116 msg->connect.base.nm_dispatch = tcp_connect;
1118 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1119 /* msg invalid now */
1122 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1123 msg->connect.nm_m, sin, if_sin);
1124 msg->connect.nm_m = NULL;
1126 if (msg->connect.nm_m) {
1127 m_freem(msg->connect.nm_m);
1128 msg->connect.nm_m = NULL;
1130 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1131 kfree(msg->connect.nm_nam, M_LWKTMSG);
1132 msg->connect.nm_nam = NULL;
1134 lwkt_replymsg(&msg->connect.base.lmsg, error);
1135 /* msg invalid now */
1141 tcp6_connect(netmsg_t msg)
1144 struct socket *so = msg->connect.base.nm_so;
1145 struct sockaddr *nam = msg->connect.nm_nam;
1146 struct thread *td = msg->connect.nm_td;
1148 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1149 struct in6_addr *addr6;
1153 COMMON_START(so, inp, 0);
1156 * Reconnect our pcb if we have to
1158 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1159 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1160 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1164 * Bind if we have to
1166 if (inp->inp_lport == 0) {
1167 error = in6_pcbbind(inp, NULL, td);
1173 * Cannot simply call in_pcbconnect, because there might be an
1174 * earlier incarnation of this same connection still in
1175 * TIME_WAIT state, creating an ADDRINUSE error.
1177 error = in6_pcbladdr(inp, nam, &addr6, td);
1181 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1183 if (port != &curthread->td_msgport) {
1184 struct route *ro = &inp->inp_route;
1187 * in_pcbladdr() may have allocated a route entry for us
1188 * on the current CPU, but we need a route entry on the
1189 * inpcb's owner CPU, so free it here.
1191 if (ro->ro_rt != NULL)
1193 bzero(ro, sizeof(*ro));
1195 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1196 sosetport(so, port);
1197 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1198 msg->connect.base.nm_dispatch = tcp6_connect;
1200 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1201 /* msg invalid now */
1204 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1205 &msg->connect.nm_m, sin6, addr6);
1206 /* nm_m may still be intact */
1208 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1210 /* msg invalid now */
1212 if (msg->connect.nm_m) {
1213 m_freem(msg->connect.nm_m);
1214 msg->connect.nm_m = NULL;
1216 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1217 kfree(msg->connect.nm_nam, M_LWKTMSG);
1218 msg->connect.nm_nam = NULL;
1220 lwkt_replymsg(&msg->connect.base.lmsg, error);
1221 /* msg invalid now */
1226 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1227 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1229 struct mbuf *m = *mp;
1230 struct inpcb *inp = tp->t_inpcb;
1231 struct socket *so = inp->inp_socket;
1235 * Cannot simply call in_pcbconnect, because there might be an
1236 * earlier incarnation of this same connection still in
1237 * TIME_WAIT state, creating an ADDRINUSE error.
1239 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1240 &sin6->sin6_addr, sin6->sin6_port,
1241 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1242 addr6 : &inp->in6p_laddr),
1243 inp->inp_lport, 0, NULL);
1245 return (EADDRINUSE);
1247 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1248 inp->in6p_laddr = *addr6;
1249 inp->in6p_faddr = sin6->sin6_addr;
1250 inp->inp_fport = sin6->sin6_port;
1251 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1252 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1253 in_pcbinsconnhash(inp);
1256 * Now that no more errors can occur, change the protocol processing
1257 * port to the current thread (which is the correct thread).
1259 * Create TCP timer message now; we are on the tcpcb's owner
1262 tcp_create_timermsg(tp, &curthread->td_msgport);
1264 /* Compute window scaling to request. */
1265 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1266 tp->request_r_scale = TCP_MIN_WINSHIFT;
1267 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1268 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1269 tp->request_r_scale++;
1273 tcpstat.tcps_connattempt++;
1274 tp->t_state = TCPS_SYN_SENT;
1275 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1276 tp->iss = tcp_new_isn(tp);
1277 tcp_sendseqinit(tp);
1279 ssb_appendstream(&so->so_snd, m);
1281 if (flags & PRUS_OOB)
1282 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1286 * Close the send side of the connection after
1287 * the data is sent if flagged.
1289 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1291 tp = tcp_usrclosed(tp);
1293 return (tcp_output(tp));
1299 * The new sockopt interface makes it possible for us to block in the
1300 * copyin/out step (if we take a page fault). Taking a page fault while
1301 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1302 * both now use TSM, there probably isn't any need for this function to
1303 * run in a critical section any more. This needs more examination.)
1306 tcp_ctloutput(netmsg_t msg)
1308 struct socket *so = msg->base.nm_so;
1309 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1310 int error, opt, optval, opthz;
1321 if (sopt->sopt_level != IPPROTO_TCP) {
1323 if (INP_CHECK_SOCKAF(so, AF_INET6))
1324 ip6_ctloutput_dispatch(msg);
1328 /* msg invalid now */
1331 tp = intotcpcb(inp);
1333 switch (sopt->sopt_dir) {
1335 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1339 switch (sopt->sopt_name) {
1342 tp->t_keepidle = tp->t_keepintvl;
1344 tp->t_keepidle = tcp_keepidle;
1345 tcp_timer_keep_activity(tp, 0);
1347 #ifdef TCP_SIGNATURE
1348 case TCP_SIGNATURE_ENABLE:
1349 if (tp->t_state == TCPS_CLOSED) {
1351 * This is the only safe state that this
1352 * option could be changed. Some segments
1353 * could already have been sent in other
1357 tp->t_flags |= TF_SIGNATURE;
1359 tp->t_flags &= ~TF_SIGNATURE;
1364 #endif /* TCP_SIGNATURE */
1367 switch (sopt->sopt_name) {
1375 opt = 0; /* dead code to fool gcc */
1382 tp->t_flags &= ~opt;
1386 if (tcp_disable_nopush)
1389 tp->t_flags |= TF_NOPUSH;
1391 tp->t_flags &= ~TF_NOPUSH;
1392 error = tcp_output(tp);
1398 * Must be between 0 and maxseg. If the requested
1399 * maxseg is too small to satisfy the desired minmss,
1400 * pump it up (silently so sysctl modifications of
1401 * minmss do not create unexpected program failures).
1402 * Handle degenerate cases.
1404 if (optval > 0 && optval <= tp->t_maxseg) {
1405 if (optval + 40 < tcp_minmss) {
1406 optval = tcp_minmss - 40;
1410 tp->t_maxseg = optval;
1417 opthz = ((int64_t)optval * hz) / 1000;
1419 tp->t_keepinit = opthz;
1425 opthz = ((int64_t)optval * hz) / 1000;
1427 tp->t_keepidle = opthz;
1428 tcp_timer_keep_activity(tp, 0);
1435 opthz = ((int64_t)optval * hz) / 1000;
1437 tp->t_keepintvl = opthz;
1438 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1446 tp->t_keepcnt = optval;
1447 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1454 error = ENOPROTOOPT;
1460 switch (sopt->sopt_name) {
1461 #ifdef TCP_SIGNATURE
1462 case TCP_SIGNATURE_ENABLE:
1463 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1465 #endif /* TCP_SIGNATURE */
1467 optval = tp->t_flags & TF_NODELAY;
1470 optval = tp->t_maxseg;
1473 optval = tp->t_flags & TF_NOOPT;
1476 optval = tp->t_flags & TF_NOPUSH;
1479 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1482 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1485 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1488 optval = tp->t_keepcnt;
1491 error = ENOPROTOOPT;
1495 soopt_from_kbuf(sopt, &optval, sizeof optval);
1499 lwkt_replymsg(&msg->lmsg, error);
1503 * tcp_sendspace and tcp_recvspace are the default send and receive window
1504 * sizes, respectively. These are obsolescent (this information should
1505 * be set by the route).
1507 * Use a default that does not require tcp window scaling to be turned
1508 * on. Individual programs or the administrator can increase the default.
1510 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1511 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1512 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1513 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1514 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1515 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1518 * Attach TCP protocol to socket, allocating internet protocol control
1519 * block, tcp control block, bufer space, and entering LISTEN state
1520 * if to accept connections.
1523 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1530 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1533 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1534 lwkt_gettoken(&so->so_rcv.ssb_token);
1535 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1537 lwkt_reltoken(&so->so_rcv.ssb_token);
1541 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1542 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1543 cpu = mycpu->gd_cpuid;
1546 * Set the default port for protocol processing. This will likely
1547 * change when we connect.
1549 error = in_pcballoc(so, &tcbinfo[cpu]);
1555 inp->inp_vflag |= INP_IPV6;
1556 inp->in6p_hops = -1; /* use kernel default */
1560 inp->inp_vflag |= INP_IPV4;
1561 tp = tcp_newtcpcb(inp);
1564 * Make sure the socket is destroyed by the pcbdetach.
1573 sofree(so); /* from ref above */
1576 tp->t_state = TCPS_CLOSED;
1577 /* Keep a reference for asynchronized pru_rcvd */
1583 * Initiate (or continue) disconnect.
1584 * If embryonic state, just send reset (once).
1585 * If in ``let data drain'' option and linger null, just drop.
1586 * Otherwise (hard), mark socket disconnecting and drop
1587 * current input data; switch states based on user close, and
1588 * send segment to peer (with FIN).
1590 static struct tcpcb *
1591 tcp_disconnect(struct tcpcb *tp)
1593 struct socket *so = tp->t_inpcb->inp_socket;
1595 if (tp->t_state < TCPS_ESTABLISHED) {
1597 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1598 tp = tcp_drop(tp, 0);
1600 lwkt_gettoken(&so->so_rcv.ssb_token);
1601 soisdisconnecting(so);
1602 sbflush(&so->so_rcv.sb);
1603 tp = tcp_usrclosed(tp);
1606 lwkt_reltoken(&so->so_rcv.ssb_token);
1612 * User issued close, and wish to trail through shutdown states:
1613 * if never received SYN, just forget it. If got a SYN from peer,
1614 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1615 * If already got a FIN from peer, then almost done; go to LAST_ACK
1616 * state. In all other cases, have already sent FIN to peer (e.g.
1617 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1618 * for peer to send FIN or not respond to keep-alives, etc.
1619 * We can let the user exit from the close as soon as the FIN is acked.
1621 static struct tcpcb *
1622 tcp_usrclosed(struct tcpcb *tp)
1625 switch (tp->t_state) {
1629 tp->t_state = TCPS_CLOSED;
1634 case TCPS_SYN_RECEIVED:
1635 tp->t_flags |= TF_NEEDFIN;
1638 case TCPS_ESTABLISHED:
1639 tp->t_state = TCPS_FIN_WAIT_1;
1642 case TCPS_CLOSE_WAIT:
1643 tp->t_state = TCPS_LAST_ACK;
1646 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1647 soisdisconnected(tp->t_inpcb->inp_socket);
1648 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1649 if (tp->t_state == TCPS_FIN_WAIT_2) {
1650 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,