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 $
67 #include "opt_inet6.h"
68 #include "opt_tcpdebug.h"
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
73 #include <sys/malloc.h>
74 #include <sys/sysctl.h>
75 #include <sys/globaldata.h>
76 #include <sys/thread.h>
80 #include <sys/domain.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/socketops.h>
85 #include <sys/protosw.h>
88 #include <sys/msgport2.h>
89 #include <sys/socketvar2.h>
92 #include <net/netisr.h>
93 #include <net/route.h>
95 #include <net/netmsg2.h>
96 #include <net/netisr2.h>
98 #include <netinet/in.h>
99 #include <netinet/in_systm.h>
101 #include <netinet/ip6.h>
103 #include <netinet/in_pcb.h>
105 #include <netinet6/in6_pcb.h>
107 #include <netinet/in_var.h>
108 #include <netinet/ip_var.h>
110 #include <netinet6/ip6_var.h>
111 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcp.h>
114 #include <netinet/tcp_fsm.h>
115 #include <netinet/tcp_seq.h>
116 #include <netinet/tcp_timer.h>
117 #include <netinet/tcp_timer2.h>
118 #include <netinet/tcp_var.h>
119 #include <netinet/tcpip.h>
121 #include <netinet/tcp_debug.h>
123 #include <machine/limits.h>
126 * Limits for TCP_KEEP* options (we will adopt the same limits that linux
129 #define MAXKEEPALIVE 32767
130 #define MAXKEEPCNT 127
133 * TCP protocol interface to socket abstraction.
135 extern char *tcpstates[]; /* XXX ??? */
137 static int tcp_attach (struct socket *, struct pru_attach_info *);
138 static void tcp_connect (netmsg_t msg);
140 static void tcp6_connect (netmsg_t msg);
141 static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
143 struct sockaddr_in6 *sin6,
144 struct in6_addr *addr6);
146 static struct tcpcb *
147 tcp_disconnect (struct tcpcb *);
148 static struct tcpcb *
149 tcp_usrclosed (struct tcpcb *);
152 #define TCPDEBUG0 int ostate = 0
153 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
154 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
155 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
159 #define TCPDEBUG2(req)
163 * For some ill optimized programs, which try to use TCP_NOPUSH
164 * to improve performance, will have small amount of data sits
165 * in the sending buffer. These small amount of data will _not_
166 * be pushed into the network until more data are written into
167 * the socket or the socket write side is shutdown.
169 static int tcp_disable_nopush = 1;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
171 &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
174 * Allocate socket buffer space.
177 tcp_usr_preattach(struct socket *so, int proto __unused,
178 struct pru_attach_info *ai)
182 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
183 error = soreserve(so, tcp_sendspace, tcp_recvspace,
188 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
189 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
195 * TCP attaches to socket via pru_attach(), reserving space,
196 * and an internet control block. This socket may move to
197 * other CPU later when we bind/connect.
200 tcp_usr_attach(netmsg_t msg)
202 struct socket *so = msg->base.nm_so;
203 struct pru_attach_info *ai = msg->attach.nm_ai;
206 struct tcpcb *tp = NULL;
210 KASSERT(inp == NULL, ("tcp socket attached"));
213 error = tcp_attach(so, ai);
217 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
218 so->so_linger = TCP_LINGERTIME;
221 TCPDEBUG2(PRU_ATTACH);
222 lwkt_replymsg(&msg->lmsg, error);
226 * pru_detach() detaches the TCP protocol from the socket.
227 * If the protocol state is non-embryonic, then can't
228 * do this directly: have to initiate a pru_disconnect(),
229 * which may finish later; embryonic TCB's can just
233 tcp_usr_detach(netmsg_t msg)
235 struct socket *so = msg->base.nm_so;
244 * If the inp is already detached or never attached, it may have
245 * been due to an async close or async attach failure. Just return
246 * as if no error occured.
250 KASSERT(tp != NULL, ("tcp_usr_detach: tp is NULL"));
252 tp = tcp_disconnect(tp);
253 TCPDEBUG2(PRU_DETACH);
255 lwkt_replymsg(&msg->lmsg, error);
259 * NOTE: ignore_error is non-zero for certain disconnection races
260 * which we want to silently allow, otherwise close() may return
261 * an unexpected error.
263 * NOTE: The variables (msg) and (tp) are assumed.
265 #define COMMON_START(so, inp, ignore_error) \
271 error = ignore_error ? 0 : EINVAL; \
275 tp = intotcpcb(inp); \
279 #define COMMON_END1(req, noreply) \
283 lwkt_replymsg(&msg->lmsg, error); \
287 #define COMMON_END(req) COMMON_END1((req), 0)
290 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
292 sosetport(((struct netmsg_base *)msg)->nm_so, port);
296 * Give the socket an address.
299 tcp_usr_bind(netmsg_t msg)
301 struct socket *so = msg->bind.base.nm_so;
302 struct sockaddr *nam = msg->bind.nm_nam;
303 struct thread *td = msg->bind.nm_td;
307 struct sockaddr_in *sinp;
308 lwkt_port_t port0 = netisr_cpuport(0);
310 COMMON_START(so, inp, 0);
313 * Must check for multicast addresses and disallow binding
316 sinp = (struct sockaddr_in *)nam;
317 if (sinp->sin_family == AF_INET &&
318 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
319 error = EAFNOSUPPORT;
324 * Check "already bound" here (in_pcbbind() does the same check
325 * though), so we don't forward a connected socket to netisr0,
326 * which would panic in the following in_pcbunlink().
328 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) {
329 error = EINVAL; /* already bound */
334 * Use netisr0 to serialize in_pcbbind(), so that pru_detach and
335 * pru_bind for different sockets on the same local port could be
336 * properly ordered. The original race is illustrated here for
341 * close(s1); <----- asynchronous
345 * All will expect bind(s2, *.PORT) to succeed. However, it will
346 * fail, if following sequence happens due to random socket initial
347 * msgport and asynchronous close(2):
351 * : pru_bind(s2) [*.PORT is used by s1]
354 if (&curthread->td_msgport != port0) {
355 lwkt_msg_t lmsg = &msg->bind.base.lmsg;
357 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0,
358 ("already asked to relink"));
360 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
361 msg->bind.nm_flags |= PRUB_RELINK;
363 TCP_STATE_MIGRATE_START(tp);
365 /* See the related comment in tcp_connect() */
366 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
367 lwkt_forwardmsg(port0, lmsg);
368 /* msg invalid now */
371 KASSERT(so->so_port == port0, ("so_port is not netisr0"));
373 if (msg->bind.nm_flags & PRUB_RELINK) {
374 msg->bind.nm_flags &= ~PRUB_RELINK;
375 TCP_STATE_MIGRATE_END(tp);
376 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
378 KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
380 error = in_pcbbind(inp, nam, td);
384 COMMON_END(PRU_BIND);
390 tcp6_usr_bind(netmsg_t msg)
392 struct socket *so = msg->bind.base.nm_so;
393 struct sockaddr *nam = msg->bind.nm_nam;
394 struct thread *td = msg->bind.nm_td;
398 struct sockaddr_in6 *sin6p;
400 COMMON_START(so, inp, 0);
403 * Must check for multicast addresses and disallow binding
406 sin6p = (struct sockaddr_in6 *)nam;
407 if (sin6p->sin6_family == AF_INET6 &&
408 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
409 error = EAFNOSUPPORT;
412 error = in6_pcbbind(inp, nam, td);
415 COMMON_END(PRU_BIND);
419 struct netmsg_inswildcard {
420 struct netmsg_base base;
421 struct inpcb *nm_inp;
425 in_pcbinswildcardhash_handler(netmsg_t msg)
427 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
428 int cpu = mycpuid, nextcpu;
430 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
433 if (nextcpu < netisr_ncpus)
434 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
436 lwkt_replymsg(&nm->base.lmsg, 0);
440 * Prepare to accept connections.
443 tcp_usr_listen(netmsg_t msg)
445 struct socket *so = msg->listen.base.nm_so;
446 struct thread *td = msg->listen.nm_td;
450 struct netmsg_inswildcard nm;
451 lwkt_port_t port0 = netisr_cpuport(0);
453 COMMON_START(so, inp, 0);
455 if (&curthread->td_msgport != port0) {
456 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
458 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
459 ("already asked to relink"));
461 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
462 msg->listen.nm_flags |= PRUL_RELINK;
464 TCP_STATE_MIGRATE_START(tp);
466 /* See the related comment in tcp_connect() */
467 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
468 lwkt_forwardmsg(port0, lmsg);
469 /* msg invalid now */
472 KASSERT(so->so_port == port0, ("so_port is not netisr0"));
474 if (msg->listen.nm_flags & PRUL_RELINK) {
475 msg->listen.nm_flags &= ~PRUL_RELINK;
476 TCP_STATE_MIGRATE_END(tp);
477 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
479 KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
481 if (tp->t_flags & TF_LISTEN)
484 if (inp->inp_lport == 0) {
485 error = in_pcbbind(inp, NULL, td);
490 TCP_STATE_CHANGE(tp, TCPS_LISTEN);
491 tp->t_flags |= TF_LISTEN;
492 tp->tt_msg = NULL; /* Catch any invalid timer usage */
495 * Create tcpcb per-cpu port cache
498 * This _must_ be done before installing this inpcb into
501 tcp_pcbport_create(tp);
503 if (netisr_ncpus > 1) {
505 * Put this inpcb into wildcard hash on other cpus.
507 ASSERT_INP_NOTINHASH(inp);
508 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
509 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
511 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
513 in_pcbinswildcardhash(inp);
514 COMMON_END(PRU_LISTEN);
520 tcp6_usr_listen(netmsg_t msg)
522 struct socket *so = msg->listen.base.nm_so;
523 struct thread *td = msg->listen.nm_td;
527 struct netmsg_inswildcard nm;
529 COMMON_START(so, inp, 0);
531 if (tp->t_flags & TF_LISTEN)
534 if (inp->inp_lport == 0) {
535 error = in6_pcbbind(inp, NULL, td);
540 TCP_STATE_CHANGE(tp, TCPS_LISTEN);
541 tp->t_flags |= TF_LISTEN;
542 tp->tt_msg = NULL; /* Catch any invalid timer usage */
545 * Create tcpcb per-cpu port cache
548 * This _must_ be done before installing this inpcb into
551 tcp_pcbport_create(tp);
553 if (netisr_ncpus > 1) {
555 * Put this inpcb into wildcard hash on other cpus.
557 KKASSERT(so->so_port == netisr_cpuport(0));
559 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
560 ASSERT_INP_NOTINHASH(inp);
562 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
563 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
565 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
567 in_pcbinswildcardhash(inp);
568 COMMON_END(PRU_LISTEN);
573 * Initiate connection to peer.
574 * Create a template for use in transmissions on this connection.
575 * Enter SYN_SENT state, and mark socket as connecting.
576 * Start keep-alive timer, and seed output sequence space.
577 * Send initial segment on connection.
580 tcp_usr_connect(netmsg_t msg)
582 struct socket *so = msg->connect.base.nm_so;
583 struct sockaddr *nam = msg->connect.nm_nam;
584 struct thread *td = msg->connect.nm_td;
588 struct sockaddr_in *sinp;
590 ASSERT_NETISR_NCPUS(mycpuid);
592 COMMON_START(so, inp, 0);
595 * Must disallow TCP ``connections'' to multicast addresses.
597 sinp = (struct sockaddr_in *)nam;
598 if (sinp->sin_family == AF_INET
599 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
600 error = EAFNOSUPPORT;
604 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
605 error = EAFNOSUPPORT; /* IPv6 only jail */
610 /* msg is invalid now */
613 if (msg->connect.nm_m) {
614 m_freem(msg->connect.nm_m);
615 msg->connect.nm_m = NULL;
617 if (msg->connect.nm_flags & PRUC_HELDTD)
619 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
620 so->so_error = error;
621 soisdisconnected(so);
623 lwkt_replymsg(&msg->lmsg, error);
629 tcp6_usr_connect(netmsg_t msg)
631 struct socket *so = msg->connect.base.nm_so;
632 struct sockaddr *nam = msg->connect.nm_nam;
633 struct thread *td = msg->connect.nm_td;
637 struct sockaddr_in6 *sin6p;
639 ASSERT_NETISR_NCPUS(mycpuid);
641 COMMON_START(so, inp, 0);
644 * Must disallow TCP ``connections'' to multicast addresses.
646 sin6p = (struct sockaddr_in6 *)nam;
647 if (sin6p->sin6_family == AF_INET6
648 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
649 error = EAFNOSUPPORT;
653 if (!prison_remote_ip(td, nam)) {
654 error = EAFNOSUPPORT; /* IPv4 only jail */
658 /* Reject v4-mapped address */
659 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
660 error = EADDRNOTAVAIL;
664 inp->inp_inc.inc_isipv6 = 1;
666 /* msg is invalid now */
669 if (msg->connect.nm_m) {
670 m_freem(msg->connect.nm_m);
671 msg->connect.nm_m = NULL;
673 lwkt_replymsg(&msg->lmsg, error);
679 * Initiate disconnect from peer.
680 * If connection never passed embryonic stage, just drop;
681 * else if don't need to let data drain, then can just drop anyways,
682 * else have to begin TCP shutdown process: mark socket disconnecting,
683 * drain unread data, state switch to reflect user close, and
684 * send segment (e.g. FIN) to peer. Socket will be really disconnected
685 * when peer sends FIN and acks ours.
687 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
690 tcp_usr_disconnect(netmsg_t msg)
692 struct socket *so = msg->disconnect.base.nm_so;
697 COMMON_START(so, inp, 1);
698 tp = tcp_disconnect(tp);
699 COMMON_END(PRU_DISCONNECT);
703 * Accept a connection. Essentially all the work is
704 * done at higher levels; just return the address
705 * of the peer, storing through addr.
708 tcp_usr_accept(netmsg_t msg)
710 struct socket *so = msg->accept.base.nm_so;
711 struct sockaddr **nam = msg->accept.nm_nam;
714 struct tcpcb *tp = NULL;
718 if (so->so_state & SS_ISDISCONNECTED) {
719 error = ECONNABORTED;
729 in_setpeeraddr(so, nam);
730 COMMON_END(PRU_ACCEPT);
735 tcp6_usr_accept(netmsg_t msg)
737 struct socket *so = msg->accept.base.nm_so;
738 struct sockaddr **nam = msg->accept.nm_nam;
741 struct tcpcb *tp = NULL;
746 if (so->so_state & SS_ISDISCONNECTED) {
747 error = ECONNABORTED;
756 in6_setpeeraddr(so, nam);
757 COMMON_END(PRU_ACCEPT);
762 * Mark the connection as being incapable of further output.
765 tcp_usr_shutdown(netmsg_t msg)
767 struct socket *so = msg->shutdown.base.nm_so;
772 COMMON_START(so, inp, 0);
774 tp = tcp_usrclosed(tp);
776 error = tcp_output(tp);
777 COMMON_END(PRU_SHUTDOWN);
781 * After a receive, possibly send window update to peer.
784 tcp_usr_rcvd(netmsg_t msg)
786 struct socket *so = msg->rcvd.base.nm_so;
787 int error = 0, noreply = 0;
791 COMMON_START(so, inp, 0);
793 if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
795 so_async_rcvd_reply(so);
799 COMMON_END1(PRU_RCVD, noreply);
803 * Do a send by putting data in output queue and updating urgent
804 * marker if URG set. Possibly send more data. Unlike the other
805 * pru_*() routines, the mbuf chains are our responsibility. We
806 * must either enqueue them or free them. The other pru_* routines
807 * generally are caller-frees.
810 tcp_usr_send(netmsg_t msg)
812 struct socket *so = msg->send.base.nm_so;
813 int flags = msg->send.nm_flags;
814 struct mbuf *m = msg->send.nm_m;
820 KKASSERT(msg->send.nm_control == NULL);
821 KKASSERT(msg->send.nm_addr == NULL);
822 KKASSERT((flags & PRUS_FREEADDR) == 0);
828 * OOPS! we lost a race, the TCP session got reset after
829 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
830 * network interrupt in the non-critical section of sosend().
833 error = ECONNRESET; /* XXX EPIPE? */
843 * This is no longer necessary, since:
844 * - sosendtcp() has already checked it for us
845 * - It does not work with asynchronized send
849 * Don't let too much OOB data build up
851 if (flags & PRUS_OOB) {
852 if (ssb_space(&so->so_snd) < -512) {
861 * Pump the data into the socket.
864 ssb_appendstream(&so->so_snd, m);
867 if (flags & PRUS_OOB) {
869 * According to RFC961 (Assigned Protocols),
870 * the urgent pointer points to the last octet
871 * of urgent data. We continue, however,
872 * to consider it to indicate the first octet
873 * of data past the urgent section.
874 * Otherwise, snd_up should be one lower.
876 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
877 tp->t_flags |= TF_FORCE;
878 error = tcp_output(tp);
879 tp->t_flags &= ~TF_FORCE;
881 if (flags & PRUS_EOF) {
883 * Close the send side of the connection after
887 tp = tcp_usrclosed(tp);
889 if (tp != NULL && !tcp_output_pending(tp)) {
890 if (flags & PRUS_MORETOCOME)
891 tp->t_flags |= TF_MORETOCOME;
892 error = tcp_output_fair(tp);
893 if (flags & PRUS_MORETOCOME)
894 tp->t_flags &= ~TF_MORETOCOME;
897 COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
898 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
899 (flags & PRUS_NOREPLY));
903 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
904 * will sofree() it when we return.
907 tcp_usr_abort(netmsg_t msg)
909 struct socket *so = msg->abort.base.nm_so;
914 COMMON_START(so, inp, 1);
915 tp = tcp_drop(tp, ECONNABORTED);
916 COMMON_END(PRU_ABORT);
920 * Receive out-of-band data.
923 tcp_usr_rcvoob(netmsg_t msg)
925 struct socket *so = msg->rcvoob.base.nm_so;
926 struct mbuf *m = msg->rcvoob.nm_m;
927 int flags = msg->rcvoob.nm_flags;
932 COMMON_START(so, inp, 0);
933 if ((so->so_oobmark == 0 &&
934 (so->so_state & SS_RCVATMARK) == 0) ||
935 so->so_options & SO_OOBINLINE ||
936 tp->t_oobflags & TCPOOB_HADDATA) {
940 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
945 *mtod(m, caddr_t) = tp->t_iobc;
946 if ((flags & MSG_PEEK) == 0)
947 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
948 COMMON_END(PRU_RCVOOB);
952 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
954 in_savefaddr(so, faddr);
959 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
961 in6_savefaddr(so, faddr);
966 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
967 struct thread *td __unused)
969 const struct sockaddr_in *sinp;
971 sinp = (const struct sockaddr_in *)nam;
972 if (sinp->sin_family == AF_INET &&
973 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
980 /* xxx - should be const */
981 struct pr_usrreqs tcp_usrreqs = {
982 .pru_abort = tcp_usr_abort,
983 .pru_accept = tcp_usr_accept,
984 .pru_attach = tcp_usr_attach,
985 .pru_bind = tcp_usr_bind,
986 .pru_connect = tcp_usr_connect,
987 .pru_connect2 = pr_generic_notsupp,
988 .pru_control = in_control_dispatch,
989 .pru_detach = tcp_usr_detach,
990 .pru_disconnect = tcp_usr_disconnect,
991 .pru_listen = tcp_usr_listen,
992 .pru_peeraddr = in_setpeeraddr_dispatch,
993 .pru_rcvd = tcp_usr_rcvd,
994 .pru_rcvoob = tcp_usr_rcvoob,
995 .pru_send = tcp_usr_send,
996 .pru_sense = pru_sense_null,
997 .pru_shutdown = tcp_usr_shutdown,
998 .pru_sockaddr = in_setsockaddr_dispatch,
999 .pru_sosend = sosendtcp,
1000 .pru_soreceive = sorecvtcp,
1001 .pru_savefaddr = tcp_usr_savefaddr,
1002 .pru_preconnect = tcp_usr_preconnect,
1003 .pru_preattach = tcp_usr_preattach
1007 struct pr_usrreqs tcp6_usrreqs = {
1008 .pru_abort = tcp_usr_abort,
1009 .pru_accept = tcp6_usr_accept,
1010 .pru_attach = tcp_usr_attach,
1011 .pru_bind = tcp6_usr_bind,
1012 .pru_connect = tcp6_usr_connect,
1013 .pru_connect2 = pr_generic_notsupp,
1014 .pru_control = in6_control_dispatch,
1015 .pru_detach = tcp_usr_detach,
1016 .pru_disconnect = tcp_usr_disconnect,
1017 .pru_listen = tcp6_usr_listen,
1018 .pru_peeraddr = in6_setpeeraddr_dispatch,
1019 .pru_rcvd = tcp_usr_rcvd,
1020 .pru_rcvoob = tcp_usr_rcvoob,
1021 .pru_send = tcp_usr_send,
1022 .pru_sense = pru_sense_null,
1023 .pru_shutdown = tcp_usr_shutdown,
1024 .pru_sockaddr = in6_setsockaddr_dispatch,
1025 .pru_sosend = sosendtcp,
1026 .pru_soreceive = sorecvtcp,
1027 .pru_savefaddr = tcp6_usr_savefaddr
1032 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1033 const struct sockaddr_in *sin, struct sockaddr_in *if_sin,
1036 struct inpcb *inp = tp->t_inpcb, *oinp;
1037 struct socket *so = inp->inp_socket;
1038 struct route *ro = &inp->inp_route;
1040 KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
1041 ("pcbinfo mismatch"));
1043 oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1044 sin->sin_addr, sin->sin_port,
1045 (inp->inp_laddr.s_addr != INADDR_ANY ?
1046 inp->inp_laddr : if_sin->sin_addr),
1047 inp->inp_lport, 0, NULL);
1050 return (EADDRINUSE);
1052 if (inp->inp_laddr.s_addr == INADDR_ANY)
1053 inp->inp_laddr = if_sin->sin_addr;
1054 KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1055 ("faddr mismatch for reconnect"));
1056 KASSERT(inp->inp_fport == sin->sin_port,
1057 ("fport mismatch for reconnect"));
1058 in_pcbinsconnhash(inp);
1060 inp->inp_flags |= INP_HASH;
1061 inp->inp_hashval = hash;
1064 * We are now on the inpcb's owner CPU, if the cached route was
1065 * freed because the rtentry's owner CPU is not the current CPU
1066 * (e.g. in tcp_connect()), then we try to reallocate it here with
1067 * the hope that a rtentry may be cloned from a RTF_PRCLONING
1070 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1071 ro->ro_rt == NULL) {
1072 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1073 ro->ro_dst.sa_family = AF_INET;
1074 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1075 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1081 * Now that no more errors can occur, change the protocol processing
1082 * port to the current thread (which is the correct thread).
1084 * Create TCP timer message now; we are on the tcpcb's owner
1087 tcp_create_timermsg(tp, &curthread->td_msgport);
1090 * Compute window scaling to request. Use a larger scaling then
1091 * needed for the initial receive buffer in case the receive buffer
1094 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1095 tp->request_r_scale = TCP_MIN_WINSHIFT;
1096 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1097 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1099 tp->request_r_scale++;
1103 tcpstat.tcps_connattempt++;
1104 TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1105 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1106 tp->iss = tcp_new_isn(tp);
1107 tcp_sendseqinit(tp);
1109 ssb_appendstream(&so->so_snd, m);
1111 if (flags & PRUS_OOB)
1112 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1116 * Close the send side of the connection after
1117 * the data is sent if flagged.
1119 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1121 tp = tcp_usrclosed(tp);
1123 return (tcp_output(tp));
1127 * Common subroutine to open a TCP connection to remote host specified
1128 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1129 * port number if needed. Call in_pcbladdr to do the routing and to choose
1130 * a local host address (interface).
1131 * Initialize connection parameters and enter SYN-SENT state.
1134 tcp_connect(netmsg_t msg)
1136 struct socket *so = msg->connect.base.nm_so;
1137 struct sockaddr *nam = msg->connect.nm_nam;
1138 struct thread *td = msg->connect.nm_td;
1139 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1140 struct sockaddr_in *if_sin = NULL;
1147 COMMON_START(so, inp, 0);
1150 * Reconnect our pcb if we have to
1152 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1153 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1154 TCP_STATE_MIGRATE_END(tp);
1155 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1157 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1158 kprintf("inpcb %p, double-connect race\n", inp);
1160 if (so->so_state & SS_ISCONNECTING)
1164 KASSERT(inp->inp_fport == 0, ("invalid fport"));
1168 * Select local port, if it is not yet selected.
1170 if (inp->inp_lport == 0) {
1171 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1173 error = in_pcbladdr(inp, nam, &if_sin, td);
1176 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1177 msg->connect.nm_flags |= PRUC_HASLADDR;
1180 * Install faddr/fport earlier, so that when this
1181 * inpcb is installed on to the lport hash, the
1182 * 4-tuple contains correct value.
1184 * NOTE: The faddr/fport will have to be installed
1185 * after the in_pcbladdr(), which may change them.
1187 inp->inp_faddr = sin->sin_addr;
1188 inp->inp_fport = sin->sin_port;
1190 error = in_pcbbind_remote(inp, nam, td);
1195 if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1198 * This inpcb was bound before this connect.
1200 error = in_pcbladdr(inp, nam, &if_sin, td);
1205 * Save or refresh the faddr/fport, since they may
1206 * be changed by in_pcbladdr().
1208 inp->inp_faddr = sin->sin_addr;
1209 inp->inp_fport = sin->sin_port;
1213 KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1214 ("faddr mismatch for reconnect"));
1215 KASSERT(inp->inp_fport == sin->sin_port,
1216 ("fport mismatch for reconnect"));
1219 KKASSERT(inp->inp_socket == so);
1221 hash = tcp_addrhash(sin->sin_addr.s_addr, sin->sin_port,
1222 (inp->inp_laddr.s_addr != INADDR_ANY ?
1223 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1225 port = netisr_hashport(hash);
1227 if (port != &curthread->td_msgport) {
1228 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1231 * in_pcbladdr() may have allocated a route entry for us
1232 * on the current CPU, but we need a route entry on the
1233 * inpcb's owner CPU, so free it here.
1235 in_pcbresetroute(inp);
1238 * We are moving the protocol processing port the socket
1239 * is on, we have to unlink here and re-link on the
1242 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1243 msg->connect.nm_flags |= PRUC_RECONNECT;
1244 msg->connect.base.nm_dispatch = tcp_connect;
1246 TCP_STATE_MIGRATE_START(tp);
1249 * Use message put done receipt to change this socket's
1250 * so_port, i.e. _after_ this message was put onto the
1251 * target netisr's msgport but _before_ the message could
1252 * be pulled from the target netisr's msgport, so that:
1253 * - The upper half (socket code) will not see the new
1254 * msgport before this message reaches the new msgport
1255 * and messages for this socket will be ordered.
1256 * - This message will see the new msgport, when its
1257 * handler is called in the target netisr.
1260 * We MUST use messege put done receipt to change this
1262 * If we changed the so_port in this netisr after the
1263 * lwkt_forwardmsg (so messages for this socket will be
1264 * ordered) and changed the so_port in the target netisr
1265 * at the very beginning of this message's handler, we
1266 * would suffer so_port overwritten race, given this
1267 * message might be forwarded again.
1270 * This mechanism depends on that the netisr's msgport
1271 * is spin msgport (currently it is :).
1273 * If the upper half saw the new msgport before this
1274 * message reached the target netisr's msgport, the
1275 * messages sent from the upper half could reach the new
1276 * msgport before this message, thus there would be
1277 * message reordering. The worst case could be soclose()
1278 * saw the new msgport and the detach message could reach
1279 * the new msgport before this message, i.e. the inpcb
1280 * could have been destroyed when this message was still
1281 * pending on or on its way to the new msgport. Other
1282 * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1283 * since we have unlinked this inpcb from the current
1286 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1287 lwkt_forwardmsg(port, lmsg);
1288 /* msg invalid now */
1290 } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1292 * The original thread is no longer needed; release it.
1295 msg->connect.nm_flags &= ~PRUC_HELDTD;
1297 error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1298 msg->connect.nm_m, sin, if_sin, hash);
1299 msg->connect.nm_m = NULL;
1301 if (msg->connect.nm_m) {
1302 m_freem(msg->connect.nm_m);
1303 msg->connect.nm_m = NULL;
1305 if (msg->connect.nm_flags & PRUC_HELDTD)
1307 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1308 so->so_error = error;
1309 soisdisconnected(so);
1311 lwkt_replymsg(&msg->connect.base.lmsg, error);
1312 /* msg invalid now */
1318 tcp6_connect(netmsg_t msg)
1321 struct socket *so = msg->connect.base.nm_so;
1322 struct sockaddr *nam = msg->connect.nm_nam;
1323 struct thread *td = msg->connect.nm_td;
1325 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1326 struct in6_addr *addr6;
1330 COMMON_START(so, inp, 0);
1333 * Reconnect our pcb if we have to
1335 if (msg->connect.nm_flags & PRUC_RECONNECT) {
1336 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1337 TCP_STATE_MIGRATE_END(tp);
1338 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1342 * Bind if we have to
1344 if (inp->inp_lport == 0) {
1345 error = in6_pcbbind(inp, NULL, td);
1351 * Cannot simply call in_pcbconnect, because there might be an
1352 * earlier incarnation of this same connection still in
1353 * TIME_WAIT state, creating an ADDRINUSE error.
1355 error = in6_pcbladdr(inp, nam, &addr6, td);
1359 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1361 if (port != &curthread->td_msgport) {
1362 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1365 * in_pcbladdr() may have allocated a route entry for us
1366 * on the current CPU, but we need a route entry on the
1367 * inpcb's owner CPU, so free it here.
1369 in_pcbresetroute(inp);
1371 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1372 msg->connect.nm_flags |= PRUC_RECONNECT;
1373 msg->connect.base.nm_dispatch = tcp6_connect;
1375 TCP_STATE_MIGRATE_START(tp);
1377 /* See the related comment in tcp_connect() */
1378 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1379 lwkt_forwardmsg(port, lmsg);
1380 /* msg invalid now */
1383 error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1384 &msg->connect.nm_m, sin6, addr6);
1385 /* nm_m may still be intact */
1387 if (msg->connect.nm_m) {
1388 m_freem(msg->connect.nm_m);
1389 msg->connect.nm_m = NULL;
1391 lwkt_replymsg(&msg->connect.base.lmsg, error);
1392 /* msg invalid now */
1396 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1397 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1399 struct mbuf *m = *mp;
1400 struct inpcb *inp = tp->t_inpcb;
1401 struct socket *so = inp->inp_socket;
1405 * Cannot simply call in_pcbconnect, because there might be an
1406 * earlier incarnation of this same connection still in
1407 * TIME_WAIT state, creating an ADDRINUSE error.
1409 oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1410 &sin6->sin6_addr, sin6->sin6_port,
1411 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1412 addr6 : &inp->in6p_laddr),
1413 inp->inp_lport, 0, NULL);
1415 return (EADDRINUSE);
1417 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1418 inp->in6p_laddr = *addr6;
1419 inp->in6p_faddr = sin6->sin6_addr;
1420 inp->inp_fport = sin6->sin6_port;
1421 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1422 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1423 in_pcbinsconnhash(inp);
1426 * Now that no more errors can occur, change the protocol processing
1427 * port to the current thread (which is the correct thread).
1429 * Create TCP timer message now; we are on the tcpcb's owner
1432 tcp_create_timermsg(tp, &curthread->td_msgport);
1434 /* Compute window scaling to request. */
1435 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1436 tp->request_r_scale = TCP_MIN_WINSHIFT;
1437 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1438 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1439 tp->request_r_scale++;
1443 tcpstat.tcps_connattempt++;
1444 TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1445 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1446 tp->iss = tcp_new_isn(tp);
1447 tcp_sendseqinit(tp);
1449 ssb_appendstream(&so->so_snd, m);
1451 if (flags & PRUS_OOB)
1452 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1456 * Close the send side of the connection after
1457 * the data is sent if flagged.
1459 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1461 tp = tcp_usrclosed(tp);
1463 return (tcp_output(tp));
1469 * The new sockopt interface makes it possible for us to block in the
1470 * copyin/out step (if we take a page fault). Taking a page fault while
1471 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1472 * both now use TSM, there probably isn't any need for this function to
1473 * run in a critical section any more. This needs more examination.)
1476 tcp_ctloutput(netmsg_t msg)
1478 struct socket *so = msg->base.nm_so;
1479 struct sockopt *sopt = msg->ctloutput.nm_sopt;
1480 struct thread *td = NULL;
1481 int error, opt, optval, opthz;
1485 if (msg->ctloutput.nm_flags & PRCO_HELDTD)
1494 tp = intotcpcb(inp);
1496 /* Get socket's owner cpuid hint */
1497 if (sopt->sopt_level == SOL_SOCKET &&
1498 sopt->sopt_dir == SOPT_GET &&
1499 sopt->sopt_name == SO_CPUHINT) {
1500 if (tp->t_flags & TF_LISTEN) {
1502 * Listen sockets owner cpuid is always 0,
1503 * which does not make sense if SO_REUSEPORT
1506 * NOTE: inp_lgrpindex is _not_ assigned in jail.
1508 if ((so->so_options & SO_REUSEPORT) &&
1509 inp->inp_lgrpindex >= 0)
1510 optval = inp->inp_lgrpindex % netisr_ncpus;
1512 optval = -1; /* no hint */
1516 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1520 if (sopt->sopt_level != IPPROTO_TCP) {
1521 if (sopt->sopt_level == IPPROTO_IP) {
1522 switch (sopt->sopt_name) {
1523 case IP_MULTICAST_IF:
1524 case IP_MULTICAST_VIF:
1525 case IP_MULTICAST_TTL:
1526 case IP_MULTICAST_LOOP:
1527 case IP_ADD_MEMBERSHIP:
1528 case IP_DROP_MEMBERSHIP:
1530 * Multicast does not make sense on
1538 if (INP_CHECK_SOCKAF(so, AF_INET6))
1539 ip6_ctloutput_dispatch(msg);
1543 /* msg invalid now */
1549 switch (sopt->sopt_dir) {
1551 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1555 switch (sopt->sopt_name) {
1558 tp->t_keepidle = tp->t_keepintvl;
1560 tp->t_keepidle = tcp_keepidle;
1561 tcp_timer_keep_activity(tp, 0);
1563 #ifdef TCP_SIGNATURE
1564 case TCP_SIGNATURE_ENABLE:
1565 if (tp->t_state == TCPS_CLOSED) {
1567 * This is the only safe state that this
1568 * option could be changed. Some segments
1569 * could already have been sent in other
1573 tp->t_flags |= TF_SIGNATURE;
1575 tp->t_flags &= ~TF_SIGNATURE;
1580 #endif /* TCP_SIGNATURE */
1583 switch (sopt->sopt_name) {
1591 opt = 0; /* dead code to fool gcc */
1598 tp->t_flags &= ~opt;
1602 if (tcp_disable_nopush)
1605 tp->t_flags |= TF_NOPUSH;
1607 tp->t_flags &= ~TF_NOPUSH;
1608 error = tcp_output(tp);
1614 * Must be between 0 and maxseg. If the requested
1615 * maxseg is too small to satisfy the desired minmss,
1616 * pump it up (silently so sysctl modifications of
1617 * minmss do not create unexpected program failures).
1618 * Handle degenerate cases.
1620 if (optval > 0 && optval <= tp->t_maxseg) {
1621 if (optval + 40 < tcp_minmss) {
1622 optval = tcp_minmss - 40;
1626 tp->t_maxseg = optval;
1635 if (optval < 1 || optval > MAXKEEPALIVE) {
1639 opthz = optval * hz;
1641 switch (sopt->sopt_name) {
1643 tp->t_keepinit = opthz;
1646 tp->t_keepidle = opthz;
1647 tcp_timer_keep_activity(tp, 0);
1650 tp->t_keepintvl = opthz;
1651 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1657 if (optval < 1 || optval > MAXKEEPCNT) {
1661 tp->t_keepcnt = optval;
1662 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1666 error = ENOPROTOOPT;
1672 switch (sopt->sopt_name) {
1673 #ifdef TCP_SIGNATURE
1674 case TCP_SIGNATURE_ENABLE:
1675 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1677 #endif /* TCP_SIGNATURE */
1679 optval = tp->t_flags & TF_NODELAY;
1682 optval = tp->t_maxseg;
1685 optval = tp->t_flags & TF_NOOPT;
1688 optval = tp->t_flags & TF_NOPUSH;
1691 optval = tp->t_keepinit / hz;
1694 optval = tp->t_keepidle / hz;
1697 optval = tp->t_keepintvl / hz;
1700 optval = tp->t_keepcnt;
1703 error = ENOPROTOOPT;
1707 soopt_from_kbuf(sopt, &optval, sizeof optval);
1713 lwkt_replymsg(&msg->lmsg, error);
1716 struct netmsg_tcp_ctloutput {
1717 struct netmsg_pr_ctloutput ctloutput;
1718 struct sockopt sopt;
1723 * Allocate netmsg_pr_ctloutput for asynchronous tcp_ctloutput.
1725 struct netmsg_pr_ctloutput *
1726 tcp_ctloutmsg(struct sockopt *sopt)
1728 struct netmsg_tcp_ctloutput *msg;
1729 int flags = 0, error;
1731 KASSERT(sopt->sopt_dir == SOPT_SET, ("not from ctloutput"));
1733 /* Only small set of options allows asynchronous setting. */
1734 if (sopt->sopt_level != IPPROTO_TCP)
1736 switch (sopt->sopt_name) {
1746 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_NULLOK);
1748 /* Fallback to synchronous tcp_ctloutput */
1752 /* Save the sockopt */
1755 /* Fixup the sopt.sopt_val ptr */
1756 error = sooptcopyin(sopt, &msg->sopt_val,
1757 sizeof(msg->sopt_val), sizeof(msg->sopt_val));
1759 kfree(msg, M_LWKTMSG);
1762 msg->sopt.sopt_val = &msg->sopt_val;
1764 /* Hold the current thread */
1765 if (msg->sopt.sopt_td != NULL) {
1766 flags |= PRCO_HELDTD;
1767 lwkt_hold(msg->sopt.sopt_td);
1770 msg->ctloutput.nm_flags = flags;
1771 msg->ctloutput.nm_sopt = &msg->sopt;
1773 return &msg->ctloutput;
1777 * tcp_sendspace and tcp_recvspace are the default send and receive window
1778 * sizes, respectively. These are obsolescent (this information should
1779 * be set by the route).
1781 * Use a default that does not require tcp window scaling to be turned
1782 * on. Individual programs or the administrator can increase the default.
1784 u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1785 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1786 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1787 u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1788 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1789 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1792 * Attach TCP protocol to socket, allocating internet protocol control
1793 * block, tcp control block, buffer space, and entering CLOSED state.
1796 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1802 boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1806 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1810 /* Post attach; do nothing */
1813 cpu = mycpu->gd_cpuid;
1816 * Set the default pcbinfo. This will likely change when we
1819 error = in_pcballoc(so, &tcbinfo[cpu]);
1825 inp->in6p_hops = -1; /* use kernel default */
1828 /* Keep a reference for asynchronized pru_rcvd */
1834 * Initiate (or continue) disconnect.
1835 * If embryonic state, just send reset (once).
1836 * If in ``let data drain'' option and linger null, just drop.
1837 * Otherwise (hard), mark socket disconnecting and drop
1838 * current input data; switch states based on user close, and
1839 * send segment to peer (with FIN).
1841 static struct tcpcb *
1842 tcp_disconnect(struct tcpcb *tp)
1844 struct socket *so = tp->t_inpcb->inp_socket;
1846 if (tp->t_state < TCPS_ESTABLISHED) {
1848 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1849 tp = tcp_drop(tp, 0);
1851 lwkt_gettoken(&so->so_rcv.ssb_token);
1852 soisdisconnecting(so);
1853 sbflush(&so->so_rcv.sb);
1854 tp = tcp_usrclosed(tp);
1857 lwkt_reltoken(&so->so_rcv.ssb_token);
1863 * User issued close, and wish to trail through shutdown states:
1864 * if never received SYN, just forget it. If got a SYN from peer,
1865 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1866 * If already got a FIN from peer, then almost done; go to LAST_ACK
1867 * state. In all other cases, have already sent FIN to peer (e.g.
1868 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1869 * for peer to send FIN or not respond to keep-alives, etc.
1870 * We can let the user exit from the close as soon as the FIN is acked.
1872 static struct tcpcb *
1873 tcp_usrclosed(struct tcpcb *tp)
1876 switch (tp->t_state) {
1880 TCP_STATE_CHANGE(tp, TCPS_CLOSED);
1885 case TCPS_SYN_RECEIVED:
1886 tp->t_flags |= TF_NEEDFIN;
1889 case TCPS_ESTABLISHED:
1890 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1893 case TCPS_CLOSE_WAIT:
1894 TCP_STATE_CHANGE(tp, TCPS_LAST_ACK);
1897 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1898 soisdisconnected(tp->t_inpcb->inp_socket);
1899 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1900 if (tp->t_state == TCPS_FIN_WAIT_2) {
1901 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,