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. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
63 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
66 #include "opt_ipsec.h"
68 #include "opt_inet6.h"
69 #include "opt_tcpdebug.h"
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
81 #include <sys/domain.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/socketops.h>
86 #include <sys/protosw.h>
88 #include <sys/thread2.h>
89 #include <sys/msgport2.h>
90 #include <sys/socketvar2.h>
93 #include <net/netisr.h>
94 #include <net/route.h>
96 #include <net/netmsg2.h>
97 #include <net/netisr2.h>
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
102 #include <netinet/ip6.h>
104 #include <netinet/in_pcb.h>
106 #include <netinet6/in6_pcb.h>
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/tcp6_var.h>
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_timer2.h>
119 #include <netinet/tcp_var.h>
120 #include <netinet/tcpip.h>
122 #include <netinet/tcp_debug.h>
126 #include <netinet6/ipsec.h>
130 * TCP protocol interface to socket abstraction.
132 extern char *tcpstates[]; /* XXX ??? */
134 static int tcp_attach (struct socket *, struct pru_attach_info *);
135 static void tcp_connect (netmsg_t msg);
137 static void tcp6_connect (netmsg_t msg);
138 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
140 struct sockaddr_in6 *sin6,
141 struct in6_addr *addr6);
143 static struct tcpcb *
144 tcp_disconnect (struct tcpcb *);
145 static struct tcpcb *
146 tcp_usrclosed (struct tcpcb *);
149 #define TCPDEBUG0 int ostate = 0
150 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
151 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
152 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
156 #define TCPDEBUG2(req)
159 static int tcp_lport_extension = 1;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
161 &tcp_lport_extension, 0, "");
164 * For some ill optimized programs, which try to use TCP_NOPUSH
165 * to improve performance, will have small amount of data sits
166 * in the sending buffer. These small amount of data will _not_
167 * be pushed into the network until more data are written into
168 * the socket or the socket write side is shutdown.
170 static int tcp_disable_nopush = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
172 &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
175 * TCP attaches to socket via pru_attach(), reserving space,
176 * and an internet control block. This is likely occuring on
177 * cpu0 and may have to move later when we bind/connect.
180 tcp_usr_attach(netmsg_t msg)
182 struct socket *so = msg->base.nm_so;
183 struct pru_attach_info *ai = msg->attach.nm_ai;
186 struct tcpcb *tp = NULL;
197 error = tcp_attach(so, ai);
201 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
202 so->so_linger = TCP_LINGERTIME;
205 sofree(so); /* from ref above */
206 TCPDEBUG2(PRU_ATTACH);
207 lwkt_replymsg(&msg->lmsg, error);
211 * pru_detach() detaches the TCP protocol from the socket.
212 * If the protocol state is non-embryonic, then can't
213 * do this directly: have to initiate a pru_disconnect(),
214 * which may finish later; embryonic TCB's can just
218 tcp_usr_detach(netmsg_t msg)
220 struct socket *so = msg->base.nm_so;
229 * If the inp is already detached it may have been due to an async
230 * close. Just return as if no error occured.
232 * It's possible for the tcpcb (tp) to disconnect from the inp due
233 * to tcp_drop()->tcp_close() being called. This may occur *after*
234 * the detach message has been queued so we may find a NULL tp here.
237 if ((tp = intotcpcb(inp)) != NULL) {
239 tp = tcp_disconnect(tp);
240 TCPDEBUG2(PRU_DETACH);
243 lwkt_replymsg(&msg->lmsg, error);
247 * NOTE: ignore_error is non-zero for certain disconnection races
248 * which we want to silently allow, otherwise close() may return
249 * an unexpected error.
251 * NOTE: The variables (msg) and (tp) are assumed.
253 #define COMMON_START(so, inp, ignore_error) \
259 error = ignore_error ? 0 : EINVAL; \
263 tp = intotcpcb(inp); \
267 #define COMMON_END1(req, noreply) \
271 lwkt_replymsg(&msg->lmsg, error); \
275 #define COMMON_END(req) COMMON_END1((req), 0)
278 * Give the socket an address.
281 tcp_usr_bind(netmsg_t msg)
283 struct socket *so = msg->bind.base.nm_so;
284 struct sockaddr *nam = msg->bind.nm_nam;
285 struct thread *td = msg->bind.nm_td;
289 struct sockaddr_in *sinp;
291 COMMON_START(so, inp, 0);
294 * Must check for multicast addresses and disallow binding
297 sinp = (struct sockaddr_in *)nam;
298 if (sinp->sin_family == AF_INET &&
299 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
300 error = EAFNOSUPPORT;
303 error = in_pcbbind(inp, nam, td);
306 COMMON_END(PRU_BIND);
313 tcp6_usr_bind(netmsg_t msg)
315 struct socket *so = msg->bind.base.nm_so;
316 struct sockaddr *nam = msg->bind.nm_nam;
317 struct thread *td = msg->bind.nm_td;
321 struct sockaddr_in6 *sin6p;
323 COMMON_START(so, inp, 0);
326 * Must check for multicast addresses and disallow binding
329 sin6p = (struct sockaddr_in6 *)nam;
330 if (sin6p->sin6_family == AF_INET6 &&
331 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
332 error = EAFNOSUPPORT;
335 inp->inp_vflag &= ~INP_IPV4;
336 inp->inp_vflag |= INP_IPV6;
337 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
338 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
339 inp->inp_vflag |= INP_IPV4;
340 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
341 struct sockaddr_in sin;
343 in6_sin6_2_sin(&sin, sin6p);
344 inp->inp_vflag |= INP_IPV4;
345 inp->inp_vflag &= ~INP_IPV6;
346 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
350 error = in6_pcbbind(inp, nam, td);
353 COMMON_END(PRU_BIND);
357 struct netmsg_inswildcard {
358 struct netmsg_base base;
359 struct inpcb *nm_inp;
363 in_pcbinswildcardhash_handler(netmsg_t msg)
365 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
366 int cpu = mycpuid, nextcpu;
368 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
371 if (nextcpu < ncpus2)
372 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
374 lwkt_replymsg(&nm->base.lmsg, 0);
378 * Prepare to accept connections.
381 tcp_usr_listen(netmsg_t msg)
383 struct socket *so = msg->listen.base.nm_so;
384 struct thread *td = msg->listen.nm_td;
388 struct netmsg_inswildcard nm;
390 COMMON_START(so, inp, 0);
392 if (tp->t_flags & TF_LISTEN)
395 if (inp->inp_lport == 0) {
396 error = in_pcbbind(inp, NULL, td);
401 tp->t_state = TCPS_LISTEN;
402 tp->t_flags |= TF_LISTEN;
403 tp->tt_msg = NULL; /* Catch any invalid timer usage */
407 * We have to set the flag because we can't have other cpus
408 * messing with our inp's flags.
410 KASSERT(!(inp->inp_flags & INP_CONNECTED),
411 ("already on connhash"));
412 KASSERT(!(inp->inp_flags & INP_WILDCARD),
413 ("already on wildcardhash"));
414 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
415 ("already on MP wildcardhash"));
416 inp->inp_flags |= INP_WILDCARD_MP;
418 KKASSERT(so->so_port == netisr_cpuport(0));
419 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
420 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
422 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
423 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
425 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
427 in_pcbinswildcardhash(inp);
428 COMMON_END(PRU_LISTEN);
434 tcp6_usr_listen(netmsg_t msg)
436 struct socket *so = msg->listen.base.nm_so;
437 struct thread *td = msg->listen.nm_td;
441 struct netmsg_inswildcard nm;
443 COMMON_START(so, inp, 0);
445 if (tp->t_flags & TF_LISTEN)
448 if (inp->inp_lport == 0) {
449 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
450 inp->inp_vflag |= INP_IPV4;
452 inp->inp_vflag &= ~INP_IPV4;
453 error = in6_pcbbind(inp, NULL, td);
458 tp->t_state = TCPS_LISTEN;
459 tp->t_flags |= TF_LISTEN;
460 tp->tt_msg = NULL; /* Catch any invalid timer usage */
464 * We have to set the flag because we can't have other cpus
465 * messing with our inp's flags.
467 KASSERT(!(inp->inp_flags & INP_CONNECTED),
468 ("already on connhash"));
469 KASSERT(!(inp->inp_flags & INP_WILDCARD),
470 ("already on wildcardhash"));
471 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
472 ("already on MP wildcardhash"));
473 inp->inp_flags |= INP_WILDCARD_MP;
475 KKASSERT(so->so_port == netisr_cpuport(0));
476 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
477 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
479 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
480 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
482 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
484 in_pcbinswildcardhash(inp);
485 COMMON_END(PRU_LISTEN);
490 * Initiate connection to peer.
491 * Create a template for use in transmissions on this connection.
492 * Enter SYN_SENT state, and mark socket as connecting.
493 * Start keep-alive timer, and seed output sequence space.
494 * Send initial segment on connection.
497 tcp_usr_connect(netmsg_t msg)
499 struct socket *so = msg->connect.base.nm_so;
500 struct sockaddr *nam = msg->connect.nm_nam;
501 struct thread *td = msg->connect.nm_td;
505 struct sockaddr_in *sinp;
507 COMMON_START(so, inp, 0);
510 * Must disallow TCP ``connections'' to multicast addresses.
512 sinp = (struct sockaddr_in *)nam;
513 if (sinp->sin_family == AF_INET
514 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
515 error = EAFNOSUPPORT;
519 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
520 error = EAFNOSUPPORT; /* IPv6 only jail */
525 /* msg is invalid now */
528 if (msg->connect.nm_m) {
529 m_freem(msg->connect.nm_m);
530 msg->connect.nm_m = NULL;
532 lwkt_replymsg(&msg->lmsg, error);
538 tcp6_usr_connect(netmsg_t msg)
540 struct socket *so = msg->connect.base.nm_so;
541 struct sockaddr *nam = msg->connect.nm_nam;
542 struct thread *td = msg->connect.nm_td;
546 struct sockaddr_in6 *sin6p;
548 COMMON_START(so, inp, 0);
551 * Must disallow TCP ``connections'' to multicast addresses.
553 sin6p = (struct sockaddr_in6 *)nam;
554 if (sin6p->sin6_family == AF_INET6
555 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
556 error = EAFNOSUPPORT;
560 if (!prison_remote_ip(td, nam)) {
561 error = EAFNOSUPPORT; /* IPv4 only jail */
565 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
566 struct sockaddr_in *sinp;
568 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
572 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
573 in6_sin6_2_sin(sinp, sin6p);
574 inp->inp_vflag |= INP_IPV4;
575 inp->inp_vflag &= ~INP_IPV6;
576 msg->connect.nm_nam = (struct sockaddr *)sinp;
577 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
579 /* msg is invalid now */
582 inp->inp_vflag &= ~INP_IPV4;
583 inp->inp_vflag |= INP_IPV6;
584 inp->inp_inc.inc_isipv6 = 1;
586 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
588 /* msg is invalid now */
591 if (msg->connect.nm_m) {
592 m_freem(msg->connect.nm_m);
593 msg->connect.nm_m = NULL;
595 lwkt_replymsg(&msg->lmsg, error);
601 * Initiate disconnect from peer.
602 * If connection never passed embryonic stage, just drop;
603 * else if don't need to let data drain, then can just drop anyways,
604 * else have to begin TCP shutdown process: mark socket disconnecting,
605 * drain unread data, state switch to reflect user close, and
606 * send segment (e.g. FIN) to peer. Socket will be really disconnected
607 * when peer sends FIN and acks ours.
609 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
612 tcp_usr_disconnect(netmsg_t msg)
614 struct socket *so = msg->disconnect.base.nm_so;
619 COMMON_START(so, inp, 1);
620 tp = tcp_disconnect(tp);
621 COMMON_END(PRU_DISCONNECT);
625 * Accept a connection. Essentially all the work is
626 * done at higher levels; just return the address
627 * of the peer, storing through addr.
630 tcp_usr_accept(netmsg_t msg)
632 struct socket *so = msg->accept.base.nm_so;
633 struct sockaddr **nam = msg->accept.nm_nam;
636 struct tcpcb *tp = NULL;
640 if (so->so_state & SS_ISDISCONNECTED) {
641 error = ECONNABORTED;
651 in_setpeeraddr(so, nam);
652 COMMON_END(PRU_ACCEPT);
657 tcp6_usr_accept(netmsg_t msg)
659 struct socket *so = msg->accept.base.nm_so;
660 struct sockaddr **nam = msg->accept.nm_nam;
663 struct tcpcb *tp = NULL;
668 if (so->so_state & SS_ISDISCONNECTED) {
669 error = ECONNABORTED;
678 in6_mapped_peeraddr(so, nam);
679 COMMON_END(PRU_ACCEPT);
683 * Mark the connection as being incapable of further output.
686 tcp_usr_shutdown(netmsg_t msg)
688 struct socket *so = msg->shutdown.base.nm_so;
693 COMMON_START(so, inp, 0);
695 tp = tcp_usrclosed(tp);
697 error = tcp_output(tp);
698 COMMON_END(PRU_SHUTDOWN);
702 * After a receive, possibly send window update to peer.
705 tcp_usr_rcvd(netmsg_t msg)
707 struct socket *so = msg->rcvd.base.nm_so;
708 int error = 0, noreply = 0;
712 COMMON_START(so, inp, 0);
714 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
716 so_async_rcvd_reply(so);
720 COMMON_END1(PRU_RCVD, noreply);
724 * Do a send by putting data in output queue and updating urgent
725 * marker if URG set. Possibly send more data. Unlike the other
726 * pru_*() routines, the mbuf chains are our responsibility. We
727 * must either enqueue them or free them. The other pru_* routines
728 * generally are caller-frees.
731 tcp_usr_send(netmsg_t msg)
733 struct socket *so = msg->send.base.nm_so;
734 int flags = msg->send.nm_flags;
735 struct mbuf *m = msg->send.nm_m;
741 KKASSERT(msg->send.nm_control == NULL);
742 KKASSERT(msg->send.nm_addr == NULL);
743 KKASSERT((flags & PRUS_FREEADDR) == 0);
749 * OOPS! we lost a race, the TCP session got reset after
750 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
751 * network interrupt in the non-critical section of sosend().
754 error = ECONNRESET; /* XXX EPIPE? */
764 * This is no longer necessary, since:
765 * - sosendtcp() has already checked it for us
766 * - It does not work with asynchronized send
770 * Don't let too much OOB data build up
772 if (flags & PRUS_OOB) {
773 if (ssb_space(&so->so_snd) < -512) {
782 * Pump the data into the socket.
785 ssb_appendstream(&so->so_snd, m);
786 if (flags & PRUS_OOB) {
788 * According to RFC961 (Assigned Protocols),
789 * the urgent pointer points to the last octet
790 * of urgent data. We continue, however,
791 * to consider it to indicate the first octet
792 * of data past the urgent section.
793 * Otherwise, snd_up should be one lower.
795 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
796 tp->t_flags |= TF_FORCE;
797 error = tcp_output(tp);
798 tp->t_flags &= ~TF_FORCE;
800 if (flags & PRUS_EOF) {
802 * Close the send side of the connection after
806 tp = tcp_usrclosed(tp);
808 if (tp != NULL && !tcp_output_pending(tp)) {
809 if (flags & PRUS_MORETOCOME)
810 tp->t_flags |= TF_MORETOCOME;
811 error = tcp_output_fair(tp);
812 if (flags & PRUS_MORETOCOME)
813 tp->t_flags &= ~TF_MORETOCOME;
816 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
817 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
818 (flags & PRUS_NOREPLY));
822 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
823 * will sofree() it when we return.
826 tcp_usr_abort(netmsg_t msg)
828 struct socket *so = msg->abort.base.nm_so;
833 COMMON_START(so, inp, 1);
834 tp = tcp_drop(tp, ECONNABORTED);
835 COMMON_END(PRU_ABORT);
839 * Receive out-of-band data.
842 tcp_usr_rcvoob(netmsg_t msg)
844 struct socket *so = msg->rcvoob.base.nm_so;
845 struct mbuf *m = msg->rcvoob.nm_m;
846 int flags = msg->rcvoob.nm_flags;
851 COMMON_START(so, inp, 0);
852 if ((so->so_oobmark == 0 &&
853 (so->so_state & SS_RCVATMARK) == 0) ||
854 so->so_options & SO_OOBINLINE ||
855 tp->t_oobflags & TCPOOB_HADDATA) {
859 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
864 *mtod(m, caddr_t) = tp->t_iobc;
865 if ((flags & MSG_PEEK) == 0)
866 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
867 COMMON_END(PRU_RCVOOB);
871 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
873 in_savefaddr(so, faddr);
878 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
880 in6_mapped_savefaddr(so, faddr);
884 /* xxx - should be const */
885 struct pr_usrreqs tcp_usrreqs = {
886 .pru_abort = tcp_usr_abort,
887 .pru_accept = tcp_usr_accept,
888 .pru_attach = tcp_usr_attach,
889 .pru_bind = tcp_usr_bind,
890 .pru_connect = tcp_usr_connect,
891 .pru_connect2 = pr_generic_notsupp,
892 .pru_control = in_control_dispatch,
893 .pru_detach = tcp_usr_detach,
894 .pru_disconnect = tcp_usr_disconnect,
895 .pru_listen = tcp_usr_listen,
896 .pru_peeraddr = in_setpeeraddr_dispatch,
897 .pru_rcvd = tcp_usr_rcvd,
898 .pru_rcvoob = tcp_usr_rcvoob,
899 .pru_send = tcp_usr_send,
900 .pru_sense = pru_sense_null,
901 .pru_shutdown = tcp_usr_shutdown,
902 .pru_sockaddr = in_setsockaddr_dispatch,
903 .pru_sosend = sosendtcp,
904 .pru_soreceive = sorecvtcp,
905 .pru_savefaddr = tcp_usr_savefaddr
909 struct pr_usrreqs tcp6_usrreqs = {
910 .pru_abort = tcp_usr_abort,
911 .pru_accept = tcp6_usr_accept,
912 .pru_attach = tcp_usr_attach,
913 .pru_bind = tcp6_usr_bind,
914 .pru_connect = tcp6_usr_connect,
915 .pru_connect2 = pr_generic_notsupp,
916 .pru_control = in6_control_dispatch,
917 .pru_detach = tcp_usr_detach,
918 .pru_disconnect = tcp_usr_disconnect,
919 .pru_listen = tcp6_usr_listen,
920 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
921 .pru_rcvd = tcp_usr_rcvd,
922 .pru_rcvoob = tcp_usr_rcvoob,
923 .pru_send = tcp_usr_send,
924 .pru_sense = pru_sense_null,
925 .pru_shutdown = tcp_usr_shutdown,
926 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
927 .pru_sosend = sosendtcp,
928 .pru_soreceive = sorecvtcp,
929 .pru_savefaddr = tcp6_usr_savefaddr
934 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
935 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
937 struct inpcb *inp = tp->t_inpcb, *oinp;
938 struct socket *so = inp->inp_socket;
939 struct route *ro = &inp->inp_route;
941 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
942 sin->sin_addr, sin->sin_port,
943 (inp->inp_laddr.s_addr != INADDR_ANY ?
944 inp->inp_laddr : if_sin->sin_addr),
945 inp->inp_lport, 0, NULL);
950 if (inp->inp_laddr.s_addr == INADDR_ANY)
951 inp->inp_laddr = if_sin->sin_addr;
952 inp->inp_faddr = sin->sin_addr;
953 inp->inp_fport = sin->sin_port;
954 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
955 in_pcbinsconnhash(inp);
958 * We are now on the inpcb's owner CPU, if the cached route was
959 * freed because the rtentry's owner CPU is not the current CPU
960 * (e.g. in tcp_connect()), then we try to reallocate it here with
961 * the hope that a rtentry may be cloned from a RTF_PRCLONING
964 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
966 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
967 ro->ro_dst.sa_family = AF_INET;
968 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
969 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
975 * Now that no more errors can occur, change the protocol processing
976 * port to the current thread (which is the correct thread).
978 * Create TCP timer message now; we are on the tcpcb's owner
981 tcp_create_timermsg(tp, &curthread->td_msgport);
984 * Compute window scaling to request. Use a larger scaling then
985 * needed for the initial receive buffer in case the receive buffer
988 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
989 tp->request_r_scale = TCP_MIN_WINSHIFT;
990 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
991 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
993 tp->request_r_scale++;
997 tcpstat.tcps_connattempt++;
998 tp->t_state = TCPS_SYN_SENT;
999 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1000 tp->iss = tcp_new_isn(tp);
1001 tcp_sendseqinit(tp);
1003 ssb_appendstream(&so->so_snd, m);
1005 if (flags & PRUS_OOB)
1006 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1010 * Close the send side of the connection after
1011 * the data is sent if flagged.
1013 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1015 tp = tcp_usrclosed(tp);
1017 return (tcp_output(tp));
1021 * Common subroutine to open a TCP connection to remote host specified
1022 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1023 * port number if needed. Call in_pcbladdr to do the routing and to choose
1024 * a local host address (interface).
1025 * Initialize connection parameters and enter SYN-SENT state.
1028 tcp_connect(netmsg_t msg)
1030 struct socket *so = msg->connect.base.nm_so;
1031 struct sockaddr *nam = msg->connect.nm_nam;
1032 struct thread *td = msg->connect.nm_td;
1033 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1034 struct sockaddr_in *if_sin;
1037 int error, calc_laddr = 1;
1040 COMMON_START(so, inp, 0);
1043 * Reconnect our pcb if we have to
1045 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1046 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1047 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1051 * Bind if we have to
1053 if (inp->inp_lport == 0) {
1054 if (tcp_lport_extension) {
1055 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1057 error = in_pcbladdr(inp, nam, &if_sin, td);
1060 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1062 error = in_pcbconn_bind(inp, nam, td);
1068 error = in_pcbbind(inp, NULL, td);
1076 * Calculate the correct protocol processing thread. The
1077 * connect operation must run there. Set the forwarding
1078 * port before we forward the message or it will get bounced
1081 error = in_pcbladdr(inp, nam, &if_sin, td);
1085 KKASSERT(inp->inp_socket == so);
1087 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1088 (inp->inp_laddr.s_addr ?
1089 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1092 if (port != &curthread->td_msgport) {
1093 struct route *ro = &inp->inp_route;
1096 * in_pcbladdr() may have allocated a route entry for us
1097 * on the current CPU, but we need a route entry on the
1098 * inpcb's owner CPU, so free it here.
1100 if (ro->ro_rt != NULL)
1102 bzero(ro, sizeof(*ro));
1105 * We are moving the protocol processing port the socket
1106 * is on, we have to unlink here and re-link on the
1109 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1110 sosetport(so, port);
1111 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1112 msg->connect.base.nm_dispatch = tcp_connect;
1114 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1115 /* msg invalid now */
1118 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1119 msg->connect.nm_m, sin, if_sin);
1120 msg->connect.nm_m = NULL;
1122 if (msg->connect.nm_m) {
1123 m_freem(msg->connect.nm_m);
1124 msg->connect.nm_m = NULL;
1126 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1127 kfree(msg->connect.nm_nam, M_LWKTMSG);
1128 msg->connect.nm_nam = NULL;
1130 lwkt_replymsg(&msg->connect.base.lmsg, error);
1131 /* msg invalid now */
1137 tcp6_connect(netmsg_t msg)
1140 struct socket *so = msg->connect.base.nm_so;
1141 struct sockaddr *nam = msg->connect.nm_nam;
1142 struct thread *td = msg->connect.nm_td;
1144 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1145 struct in6_addr *addr6;
1149 COMMON_START(so, inp, 0);
1152 * Reconnect our pcb if we have to
1154 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1155 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1156 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1160 * Bind if we have to
1162 if (inp->inp_lport == 0) {
1163 error = in6_pcbbind(inp, NULL, td);
1169 * Cannot simply call in_pcbconnect, because there might be an
1170 * earlier incarnation of this same connection still in
1171 * TIME_WAIT state, creating an ADDRINUSE error.
1173 error = in6_pcbladdr(inp, nam, &addr6, td);
1177 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1179 if (port != &curthread->td_msgport) {
1180 struct route *ro = &inp->inp_route;
1183 * in_pcbladdr() may have allocated a route entry for us
1184 * on the current CPU, but we need a route entry on the
1185 * inpcb's owner CPU, so free it here.
1187 if (ro->ro_rt != NULL)
1189 bzero(ro, sizeof(*ro));
1191 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1192 sosetport(so, port);
1193 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1194 msg->connect.base.nm_dispatch = tcp6_connect;
1196 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1197 /* msg invalid now */
1200 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1201 &msg->connect.nm_m, sin6, addr6);
1202 /* nm_m may still be intact */
1204 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1206 /* msg invalid now */
1208 if (msg->connect.nm_m) {
1209 m_freem(msg->connect.nm_m);
1210 msg->connect.nm_m = NULL;
1212 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1213 kfree(msg->connect.nm_nam, M_LWKTMSG);
1214 msg->connect.nm_nam = NULL;
1216 lwkt_replymsg(&msg->connect.base.lmsg, error);
1217 /* msg invalid now */
1222 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1223 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1225 struct mbuf *m = *mp;
1226 struct inpcb *inp = tp->t_inpcb;
1227 struct socket *so = inp->inp_socket;
1231 * Cannot simply call in_pcbconnect, because there might be an
1232 * earlier incarnation of this same connection still in
1233 * TIME_WAIT state, creating an ADDRINUSE error.
1235 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1236 &sin6->sin6_addr, sin6->sin6_port,
1237 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1238 addr6 : &inp->in6p_laddr),
1239 inp->inp_lport, 0, NULL);
1241 return (EADDRINUSE);
1243 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1244 inp->in6p_laddr = *addr6;
1245 inp->in6p_faddr = sin6->sin6_addr;
1246 inp->inp_fport = sin6->sin6_port;
1247 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1248 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1249 in_pcbinsconnhash(inp);
1252 * Now that no more errors can occur, change the protocol processing
1253 * port to the current thread (which is the correct thread).
1255 * Create TCP timer message now; we are on the tcpcb's owner
1258 tcp_create_timermsg(tp, &curthread->td_msgport);
1260 /* Compute window scaling to request. */
1261 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1262 tp->request_r_scale = TCP_MIN_WINSHIFT;
1263 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1264 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1265 tp->request_r_scale++;
1269 tcpstat.tcps_connattempt++;
1270 tp->t_state = TCPS_SYN_SENT;
1271 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1272 tp->iss = tcp_new_isn(tp);
1273 tcp_sendseqinit(tp);
1275 ssb_appendstream(&so->so_snd, m);
1277 if (flags & PRUS_OOB)
1278 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1282 * Close the send side of the connection after
1283 * the data is sent if flagged.
1285 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1287 tp = tcp_usrclosed(tp);
1289 return (tcp_output(tp));
1295 * The new sockopt interface makes it possible for us to block in the
1296 * copyin/out step (if we take a page fault). Taking a page fault while
1297 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1298 * both now use TSM, there probably isn't any need for this function to
1299 * run in a critical section any more. This needs more examination.)
1302 tcp_ctloutput(netmsg_t msg)
1304 struct socket *so = msg->base.nm_so;
1305 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1306 int error, opt, optval, opthz;
1317 if (sopt->sopt_level != IPPROTO_TCP) {
1319 if (INP_CHECK_SOCKAF(so, AF_INET6))
1320 ip6_ctloutput_dispatch(msg);
1324 /* msg invalid now */
1327 tp = intotcpcb(inp);
1329 switch (sopt->sopt_dir) {
1331 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1335 switch (sopt->sopt_name) {
1338 tp->t_keepidle = tp->t_keepintvl;
1340 tp->t_keepidle = tcp_keepidle;
1341 tcp_timer_keep_activity(tp, 0);
1343 #ifdef TCP_SIGNATURE
1344 case TCP_SIGNATURE_ENABLE:
1345 if (tp->t_state == TCPS_CLOSED) {
1347 * This is the only safe state that this
1348 * option could be changed. Some segments
1349 * could already have been sent in other
1353 tp->t_flags |= TF_SIGNATURE;
1355 tp->t_flags &= ~TF_SIGNATURE;
1360 #endif /* TCP_SIGNATURE */
1363 switch (sopt->sopt_name) {
1371 opt = 0; /* dead code to fool gcc */
1378 tp->t_flags &= ~opt;
1382 if (tcp_disable_nopush)
1385 tp->t_flags |= TF_NOPUSH;
1387 tp->t_flags &= ~TF_NOPUSH;
1388 error = tcp_output(tp);
1394 * Must be between 0 and maxseg. If the requested
1395 * maxseg is too small to satisfy the desired minmss,
1396 * pump it up (silently so sysctl modifications of
1397 * minmss do not create unexpected program failures).
1398 * Handle degenerate cases.
1400 if (optval > 0 && optval <= tp->t_maxseg) {
1401 if (optval + 40 < tcp_minmss) {
1402 optval = tcp_minmss - 40;
1406 tp->t_maxseg = optval;
1413 opthz = ((int64_t)optval * hz) / 1000;
1415 tp->t_keepinit = opthz;
1421 opthz = ((int64_t)optval * hz) / 1000;
1423 tp->t_keepidle = opthz;
1424 tcp_timer_keep_activity(tp, 0);
1431 opthz = ((int64_t)optval * hz) / 1000;
1433 tp->t_keepintvl = opthz;
1434 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1442 tp->t_keepcnt = optval;
1443 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1450 error = ENOPROTOOPT;
1456 switch (sopt->sopt_name) {
1457 #ifdef TCP_SIGNATURE
1458 case TCP_SIGNATURE_ENABLE:
1459 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1461 #endif /* TCP_SIGNATURE */
1463 optval = tp->t_flags & TF_NODELAY;
1466 optval = tp->t_maxseg;
1469 optval = tp->t_flags & TF_NOOPT;
1472 optval = tp->t_flags & TF_NOPUSH;
1475 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1478 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1481 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1484 optval = tp->t_keepcnt;
1487 error = ENOPROTOOPT;
1491 soopt_from_kbuf(sopt, &optval, sizeof optval);
1495 lwkt_replymsg(&msg->lmsg, error);
1499 * tcp_sendspace and tcp_recvspace are the default send and receive window
1500 * sizes, respectively. These are obsolescent (this information should
1501 * be set by the route).
1503 * Use a default that does not require tcp window scaling to be turned
1504 * on. Individual programs or the administrator can increase the default.
1506 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1507 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1508 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1509 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1510 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1511 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1514 * Attach TCP protocol to socket, allocating internet protocol control
1515 * block, tcp control block, bufer space, and entering LISTEN state
1516 * if to accept connections.
1519 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1526 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1529 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1530 lwkt_gettoken(&so->so_rcv.ssb_token);
1531 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1533 lwkt_reltoken(&so->so_rcv.ssb_token);
1537 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1538 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1539 cpu = mycpu->gd_cpuid;
1542 * Set the default port for protocol processing. This will likely
1543 * change when we connect.
1545 error = in_pcballoc(so, &tcbinfo[cpu]);
1551 inp->inp_vflag |= INP_IPV6;
1552 inp->in6p_hops = -1; /* use kernel default */
1556 inp->inp_vflag |= INP_IPV4;
1557 tp = tcp_newtcpcb(inp);
1560 * Make sure the socket is destroyed by the pcbdetach.
1569 sofree(so); /* from ref above */
1572 tp->t_state = TCPS_CLOSED;
1573 /* Keep a reference for asynchronized pru_rcvd */
1579 * Initiate (or continue) disconnect.
1580 * If embryonic state, just send reset (once).
1581 * If in ``let data drain'' option and linger null, just drop.
1582 * Otherwise (hard), mark socket disconnecting and drop
1583 * current input data; switch states based on user close, and
1584 * send segment to peer (with FIN).
1586 static struct tcpcb *
1587 tcp_disconnect(struct tcpcb *tp)
1589 struct socket *so = tp->t_inpcb->inp_socket;
1591 if (tp->t_state < TCPS_ESTABLISHED) {
1593 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1594 tp = tcp_drop(tp, 0);
1596 lwkt_gettoken(&so->so_rcv.ssb_token);
1597 soisdisconnecting(so);
1598 sbflush(&so->so_rcv.sb);
1599 tp = tcp_usrclosed(tp);
1602 lwkt_reltoken(&so->so_rcv.ssb_token);
1608 * User issued close, and wish to trail through shutdown states:
1609 * if never received SYN, just forget it. If got a SYN from peer,
1610 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1611 * If already got a FIN from peer, then almost done; go to LAST_ACK
1612 * state. In all other cases, have already sent FIN to peer (e.g.
1613 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1614 * for peer to send FIN or not respond to keep-alives, etc.
1615 * We can let the user exit from the close as soon as the FIN is acked.
1617 static struct tcpcb *
1618 tcp_usrclosed(struct tcpcb *tp)
1621 switch (tp->t_state) {
1625 tp->t_state = TCPS_CLOSED;
1630 case TCPS_SYN_RECEIVED:
1631 tp->t_flags |= TF_NEEDFIN;
1634 case TCPS_ESTABLISHED:
1635 tp->t_state = TCPS_FIN_WAIT_1;
1638 case TCPS_CLOSE_WAIT:
1639 tp->t_state = TCPS_LAST_ACK;
1642 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1643 soisdisconnected(tp->t_inpcb->inp_socket);
1644 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1645 if (tp->t_state == TCPS_FIN_WAIT_2) {
1646 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,