2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2006-2007 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_tcpdebug.h"
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/limits.h>
52 #include <sys/malloc.h>
53 #include <sys/refcount.h>
54 #include <sys/kernel.h>
55 #include <sys/sysctl.h>
58 #include <sys/domain.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/protosw.h>
65 #include <sys/syslog.h>
72 #include <net/if_var.h>
73 #include <net/route.h>
76 #include <netinet/in.h>
77 #include <netinet/in_kdtrace.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip_var.h>
83 #include <netinet/ip6.h>
84 #include <netinet6/in6_pcb.h>
85 #include <netinet6/ip6_var.h>
86 #include <netinet6/scope6_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/tcp_log_buf.h>
94 #include <netinet/tcpip.h>
95 #include <netinet/cc/cc.h>
96 #include <netinet/tcp_fastopen.h>
97 #include <netinet/tcp_hpts.h>
99 #include <netinet/tcp_pcap.h>
102 #include <netinet/tcp_debug.h>
105 #include <netinet/tcp_offload.h>
107 #include <netipsec/ipsec_support.h>
110 * TCP protocol interface to socket abstraction.
112 static int tcp_attach(struct socket *);
114 static int tcp_connect(struct tcpcb *, struct sockaddr *,
118 static int tcp6_connect(struct tcpcb *, struct sockaddr *,
121 static void tcp_disconnect(struct tcpcb *);
122 static void tcp_usrclosed(struct tcpcb *);
123 static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
126 #define TCPDEBUG0 int ostate = 0
127 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
128 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
129 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
133 #define TCPDEBUG2(req)
137 * TCP attaches to socket via pru_attach(), reserving space,
138 * and an internet control block.
141 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
144 struct tcpcb *tp = NULL;
149 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
152 error = tcp_attach(so);
156 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
157 so->so_linger = TCP_LINGERTIME;
162 TCPDEBUG2(PRU_ATTACH);
163 TCP_PROBE2(debug__user, tp, PRU_ATTACH);
168 * tcp_detach is called when the socket layer loses its final reference
169 * to the socket, be it a file descriptor reference, a reference from TCP,
170 * etc. At this point, there is only one case in which we will keep around
171 * inpcb state: time wait.
173 * This function can probably be re-absorbed back into tcp_usr_detach() now
174 * that there is a single detach path.
177 tcp_detach(struct socket *so, struct inpcb *inp)
181 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
182 INP_WLOCK_ASSERT(inp);
184 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
185 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
189 if (inp->inp_flags & INP_TIMEWAIT) {
191 * There are two cases to handle: one in which the time wait
192 * state is being discarded (INP_DROPPED), and one in which
193 * this connection will remain in timewait. In the former,
194 * it is time to discard all state (except tcptw, which has
195 * already been discarded by the timewait close code, which
196 * should be further up the call stack somewhere). In the
197 * latter case, we detach from the socket, but leave the pcb
198 * present until timewait ends.
200 * XXXRW: Would it be cleaner to free the tcptw here?
202 * Astute question indeed, from twtcp perspective there are
203 * four cases to consider:
205 * #1 tcp_detach is called at tcptw creation time by
206 * tcp_twstart, then do not discard the newly created tcptw
207 * and leave inpcb present until timewait ends
208 * #2 tcp_detach is called at tcptw creation time by
209 * tcp_twstart, but connection is local and tw will be
210 * discarded immediately
211 * #3 tcp_detach is called at timewait end (or reuse) by
212 * tcp_twclose, then the tcptw has already been discarded
213 * (or reused) and inpcb is freed here
214 * #4 tcp_detach is called() after timewait ends (or reuse)
215 * (e.g. by soclose), then tcptw has already been discarded
216 * (or reused) and inpcb is freed here
218 * In all three cases the tcptw should not be freed here.
220 if (inp->inp_flags & INP_DROPPED) {
222 if (__predict_true(tp == NULL)) {
226 * This case should not happen as in TIMEWAIT
227 * state the inp should not be destroyed before
228 * its tcptw. If INVARIANTS is defined, panic.
231 panic("%s: Panic before an inp double-free: "
232 "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
235 log(LOG_ERR, "%s: Avoid an inp double-free: "
236 "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
247 * If the connection is not in timewait, we consider two
248 * two conditions: one in which no further processing is
249 * necessary (dropped || embryonic), and one in which TCP is
250 * not yet done, but no longer requires the socket, so the
251 * pcb will persist for the time being.
253 * XXXRW: Does the second case still occur?
255 if (inp->inp_flags & INP_DROPPED ||
256 tp->t_state < TCPS_SYN_SENT) {
268 * pru_detach() detaches the TCP protocol from the socket.
269 * If the protocol state is non-embryonic, then can't
270 * do this directly: have to initiate a pru_disconnect(),
271 * which may finish later; embryonic TCB's can just
275 tcp_usr_detach(struct socket *so)
279 struct epoch_tracker et;
282 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
283 if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
284 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
288 KASSERT(inp->inp_socket != NULL,
289 ("tcp_usr_detach: inp_socket == NULL"));
292 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
297 * Give the socket an address.
300 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
304 struct tcpcb *tp = NULL;
305 struct sockaddr_in *sinp;
307 sinp = (struct sockaddr_in *)nam;
308 if (nam->sa_len != sizeof (*sinp))
311 * Must check for multicast addresses and disallow binding
314 if (sinp->sin_family == AF_INET &&
315 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
316 return (EAFNOSUPPORT);
320 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
322 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
328 INP_HASH_WLOCK(&V_tcbinfo);
329 error = in_pcbbind(inp, nam, td->td_ucred);
330 INP_HASH_WUNLOCK(&V_tcbinfo);
333 TCP_PROBE2(debug__user, tp, PRU_BIND);
342 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
346 struct tcpcb *tp = NULL;
347 struct sockaddr_in6 *sin6p;
349 sin6p = (struct sockaddr_in6 *)nam;
350 if (nam->sa_len != sizeof (*sin6p))
353 * Must check for multicast addresses and disallow binding
356 if (sin6p->sin6_family == AF_INET6 &&
357 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
358 return (EAFNOSUPPORT);
362 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
364 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
370 INP_HASH_WLOCK(&V_tcbinfo);
371 inp->inp_vflag &= ~INP_IPV4;
372 inp->inp_vflag |= INP_IPV6;
374 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
375 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
376 inp->inp_vflag |= INP_IPV4;
377 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
378 struct sockaddr_in sin;
380 in6_sin6_2_sin(&sin, sin6p);
381 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
382 error = EAFNOSUPPORT;
383 INP_HASH_WUNLOCK(&V_tcbinfo);
386 inp->inp_vflag |= INP_IPV4;
387 inp->inp_vflag &= ~INP_IPV6;
388 error = in_pcbbind(inp, (struct sockaddr *)&sin,
390 INP_HASH_WUNLOCK(&V_tcbinfo);
395 error = in6_pcbbind(inp, nam, td->td_ucred);
396 INP_HASH_WUNLOCK(&V_tcbinfo);
399 TCP_PROBE2(debug__user, tp, PRU_BIND);
407 * Prepare to accept connections.
410 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
414 struct tcpcb *tp = NULL;
418 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
420 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
427 error = solisten_proto_check(so);
428 INP_HASH_WLOCK(&V_tcbinfo);
429 if (error == 0 && inp->inp_lport == 0)
430 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
431 INP_HASH_WUNLOCK(&V_tcbinfo);
433 tcp_state_change(tp, TCPS_LISTEN);
434 solisten_proto(so, backlog);
436 if ((so->so_options & SO_NO_OFFLOAD) == 0)
437 tcp_offload_listen_start(tp);
442 if (IS_FASTOPEN(tp->t_flags))
443 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
446 TCPDEBUG2(PRU_LISTEN);
447 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
455 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
459 struct tcpcb *tp = NULL;
463 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
465 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
472 error = solisten_proto_check(so);
473 INP_HASH_WLOCK(&V_tcbinfo);
474 if (error == 0 && inp->inp_lport == 0) {
475 inp->inp_vflag &= ~INP_IPV4;
476 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
477 inp->inp_vflag |= INP_IPV4;
478 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
480 INP_HASH_WUNLOCK(&V_tcbinfo);
482 tcp_state_change(tp, TCPS_LISTEN);
483 solisten_proto(so, backlog);
485 if ((so->so_options & SO_NO_OFFLOAD) == 0)
486 tcp_offload_listen_start(tp);
491 if (IS_FASTOPEN(tp->t_flags))
492 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
495 TCPDEBUG2(PRU_LISTEN);
496 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
504 * Initiate connection to peer.
505 * Create a template for use in transmissions on this connection.
506 * Enter SYN_SENT state, and mark socket as connecting.
507 * Start keep-alive timer, and seed output sequence space.
508 * Send initial segment on connection.
511 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
515 struct tcpcb *tp = NULL;
516 struct sockaddr_in *sinp;
518 sinp = (struct sockaddr_in *)nam;
519 if (nam->sa_len != sizeof (*sinp))
522 * Must disallow TCP ``connections'' to multicast addresses.
524 if (sinp->sin_family == AF_INET
525 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
526 return (EAFNOSUPPORT);
527 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
532 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
534 if (inp->inp_flags & INP_TIMEWAIT) {
538 if (inp->inp_flags & INP_DROPPED) {
539 error = ECONNREFUSED;
544 if ((error = tcp_connect(tp, nam, td)) != 0)
547 if (registered_toedevs > 0 &&
548 (so->so_options & SO_NO_OFFLOAD) == 0 &&
549 (error = tcp_offload_connect(so, nam)) == 0)
552 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
553 error = tp->t_fb->tfb_tcp_output(tp);
555 TCPDEBUG2(PRU_CONNECT);
556 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
564 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
568 struct tcpcb *tp = NULL;
569 struct sockaddr_in6 *sin6p;
573 sin6p = (struct sockaddr_in6 *)nam;
574 if (nam->sa_len != sizeof (*sin6p))
577 * Must disallow TCP ``connections'' to multicast addresses.
579 if (sin6p->sin6_family == AF_INET6
580 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
581 return (EAFNOSUPPORT);
584 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
586 if (inp->inp_flags & INP_TIMEWAIT) {
590 if (inp->inp_flags & INP_DROPPED) {
591 error = ECONNREFUSED;
598 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
599 * therefore probably require the hash lock, which isn't held here.
600 * Is this a significant problem?
602 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
603 struct sockaddr_in sin;
605 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
609 if ((inp->inp_vflag & INP_IPV4) == 0) {
610 error = EAFNOSUPPORT;
614 in6_sin6_2_sin(&sin, sin6p);
615 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
616 error = EAFNOSUPPORT;
619 inp->inp_vflag |= INP_IPV4;
620 inp->inp_vflag &= ~INP_IPV6;
621 if ((error = prison_remote_ip4(td->td_ucred,
622 &sin.sin_addr)) != 0)
624 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
627 if (registered_toedevs > 0 &&
628 (so->so_options & SO_NO_OFFLOAD) == 0 &&
629 (error = tcp_offload_connect(so, nam)) == 0)
632 error = tp->t_fb->tfb_tcp_output(tp);
635 if ((inp->inp_vflag & INP_IPV6) == 0) {
636 error = EAFNOSUPPORT;
641 inp->inp_vflag &= ~INP_IPV4;
642 inp->inp_vflag |= INP_IPV6;
643 inp->inp_inc.inc_flags |= INC_ISIPV6;
644 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
646 if ((error = tcp6_connect(tp, nam, td)) != 0)
649 if (registered_toedevs > 0 &&
650 (so->so_options & SO_NO_OFFLOAD) == 0 &&
651 (error = tcp_offload_connect(so, nam)) == 0)
654 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
655 error = tp->t_fb->tfb_tcp_output(tp);
658 TCPDEBUG2(PRU_CONNECT);
659 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
666 * Initiate disconnect from peer.
667 * If connection never passed embryonic stage, just drop;
668 * else if don't need to let data drain, then can just drop anyways,
669 * else have to begin TCP shutdown process: mark socket disconnecting,
670 * drain unread data, state switch to reflect user close, and
671 * send segment (e.g. FIN) to peer. Socket will be really disconnected
672 * when peer sends FIN and acks ours.
674 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
677 tcp_usr_disconnect(struct socket *so)
680 struct tcpcb *tp = NULL;
681 struct epoch_tracker et;
685 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
687 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
689 if (inp->inp_flags & INP_TIMEWAIT)
691 if (inp->inp_flags & INP_DROPPED) {
699 TCPDEBUG2(PRU_DISCONNECT);
700 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
702 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
708 * Accept a connection. Essentially all the work is done at higher levels;
709 * just return the address of the peer, storing through addr.
712 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
715 struct inpcb *inp = NULL;
716 struct tcpcb *tp = NULL;
721 if (so->so_state & SS_ISDISCONNECTED)
722 return (ECONNABORTED);
725 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
727 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
728 error = ECONNABORTED;
735 * We inline in_getpeeraddr and COMMON_END here, so that we can
736 * copy the data of interest and defer the malloc until after we
739 port = inp->inp_fport;
740 addr = inp->inp_faddr;
743 TCPDEBUG2(PRU_ACCEPT);
744 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
747 *nam = in_sockaddr(port, &addr);
754 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
756 struct inpcb *inp = NULL;
758 struct tcpcb *tp = NULL;
760 struct in6_addr addr6;
761 struct epoch_tracker et;
766 if (so->so_state & SS_ISDISCONNECTED)
767 return (ECONNABORTED);
770 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
771 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
773 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
774 error = ECONNABORTED;
781 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
782 * copy the data of interest and defer the malloc until after we
785 if (inp->inp_vflag & INP_IPV4) {
787 port = inp->inp_fport;
788 addr = inp->inp_faddr;
790 port = inp->inp_fport;
791 addr6 = inp->in6p_faddr;
795 TCPDEBUG2(PRU_ACCEPT);
796 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
798 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
801 *nam = in6_v4mapsin6_sockaddr(port, &addr);
803 *nam = in6_sockaddr(port, &addr6);
810 * Mark the connection as being incapable of further output.
813 tcp_usr_shutdown(struct socket *so)
817 struct tcpcb *tp = NULL;
818 struct epoch_tracker et;
821 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
823 KASSERT(inp != NULL, ("inp == NULL"));
825 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
833 if (!(inp->inp_flags & INP_DROPPED))
834 error = tp->t_fb->tfb_tcp_output(tp);
837 TCPDEBUG2(PRU_SHUTDOWN);
838 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
840 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
846 * After a receive, possibly send window update to peer.
849 tcp_usr_rcvd(struct socket *so, int flags)
852 struct tcpcb *tp = NULL;
857 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
859 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
866 * For passively-created TFO connections, don't attempt a window
867 * update while still in SYN_RECEIVED as this may trigger an early
868 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with
869 * application response data, or failing that, when the DELACK timer
872 if (IS_FASTOPEN(tp->t_flags) &&
873 (tp->t_state == TCPS_SYN_RECEIVED))
876 if (tp->t_flags & TF_TOE)
877 tcp_offload_rcvd(tp);
880 tp->t_fb->tfb_tcp_output(tp);
884 TCP_PROBE2(debug__user, tp, PRU_RCVD);
890 * Do a send by putting data in output queue and updating urgent
891 * marker if URG set. Possibly send more data. Unlike the other
892 * pru_*() routines, the mbuf chains are our responsibility. We
893 * must either enqueue them or free them. The other pru_* routines
894 * generally are caller-frees.
897 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
898 struct sockaddr *nam, struct mbuf *control, struct thread *td)
902 struct tcpcb *tp = NULL;
903 struct epoch_tracker net_et;
905 struct sockaddr_in sin, *sinp;
913 * We require the pcbinfo lock if we will close the socket as part of
916 if (flags & PRUS_EOF)
917 INP_INFO_RLOCK_ET(&V_tcbinfo, net_et);
919 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
921 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
925 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
926 * for freeing memory.
928 if (m && (flags & PRUS_NOTREADY) == 0)
935 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
936 switch (nam->sa_family) {
939 sinp = (struct sockaddr_in *)nam;
940 if (sinp->sin_len != sizeof(struct sockaddr_in)) {
946 if ((inp->inp_vflag & INP_IPV6) != 0) {
949 error = EAFNOSUPPORT;
952 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
955 error = EAFNOSUPPORT;
958 if ((error = prison_remote_ip4(td->td_ucred,
972 struct sockaddr_in6 *sin6p;
974 sin6p = (struct sockaddr_in6 *)nam;
975 if (sin6p->sin6_len != sizeof(struct sockaddr_in6)) {
981 if (IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
984 error = EAFNOSUPPORT;
987 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
989 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
995 if ((inp->inp_vflag & INP_IPV4) == 0) {
996 error = EAFNOSUPPORT;
1001 inp->inp_vflag &= ~INP_IPV6;
1003 in6_sin6_2_sin(sinp, sin6p);
1005 ntohl(sinp->sin_addr.s_addr))) {
1006 error = EAFNOSUPPORT;
1011 if ((error = prison_remote_ip4(td->td_ucred,
1012 &sinp->sin_addr))) {
1019 error = EAFNOSUPPORT;
1025 if ((inp->inp_vflag & INP_IPV6) == 0) {
1028 error = EAFNOSUPPORT;
1031 inp->inp_vflag &= ~INP_IPV4;
1032 inp->inp_inc.inc_flags |= INC_ISIPV6;
1033 if ((error = prison_remote_ip6(td->td_ucred,
1034 &sin6p->sin6_addr))) {
1047 error = EAFNOSUPPORT;
1052 /* TCP doesn't do control messages (rights, creds, etc) */
1053 if (control->m_len) {
1060 m_freem(control); /* empty control, just free it */
1062 if (!(flags & PRUS_OOB)) {
1063 sbappendstream(&so->so_snd, m, flags);
1064 if (nam && tp->t_state < TCPS_SYN_SENT) {
1066 * Do implied connect if not yet connected,
1067 * initialize window to default value, and
1068 * initialize maxseg using peer's cached MSS.
1072 error = tcp6_connect(tp, nam, td);
1074 #if defined(INET6) && defined(INET)
1078 error = tcp_connect(tp,
1079 (struct sockaddr *)sinp, td);
1083 if (IS_FASTOPEN(tp->t_flags))
1084 tcp_fastopen_connect(tp);
1086 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1090 if (flags & PRUS_EOF) {
1092 * Close the send side of the connection after
1095 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1099 if (!(inp->inp_flags & INP_DROPPED) &&
1100 !(flags & PRUS_NOTREADY)) {
1101 if (flags & PRUS_MORETOCOME)
1102 tp->t_flags |= TF_MORETOCOME;
1103 error = tp->t_fb->tfb_tcp_output(tp);
1104 if (flags & PRUS_MORETOCOME)
1105 tp->t_flags &= ~TF_MORETOCOME;
1109 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1111 SOCKBUF_LOCK(&so->so_snd);
1112 if (sbspace(&so->so_snd) < -512) {
1113 SOCKBUF_UNLOCK(&so->so_snd);
1119 * According to RFC961 (Assigned Protocols),
1120 * the urgent pointer points to the last octet
1121 * of urgent data. We continue, however,
1122 * to consider it to indicate the first octet
1123 * of data past the urgent section.
1124 * Otherwise, snd_up should be one lower.
1126 sbappendstream_locked(&so->so_snd, m, flags);
1127 SOCKBUF_UNLOCK(&so->so_snd);
1128 if (nam && tp->t_state < TCPS_SYN_SENT) {
1130 * Do implied connect if not yet connected,
1131 * initialize window to default value, and
1132 * initialize maxseg using peer's cached MSS.
1136 * Not going to contemplate SYN|URG
1138 if (IS_FASTOPEN(tp->t_flags))
1139 tp->t_flags &= ~TF_FASTOPEN;
1142 error = tcp6_connect(tp, nam, td);
1144 #if defined(INET6) && defined(INET)
1148 error = tcp_connect(tp,
1149 (struct sockaddr *)sinp, td);
1153 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1156 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1157 if (!(flags & PRUS_NOTREADY)) {
1158 tp->t_flags |= TF_FORCEDATA;
1159 error = tp->t_fb->tfb_tcp_output(tp);
1160 tp->t_flags &= ~TF_FORCEDATA;
1163 TCP_LOG_EVENT(tp, NULL,
1164 &inp->inp_socket->so_rcv,
1165 &inp->inp_socket->so_snd,
1166 TCP_LOG_USERSEND, error,
1169 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1170 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1171 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1172 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1174 if (flags & PRUS_EOF)
1175 INP_INFO_RUNLOCK_ET(&V_tcbinfo, net_et);
1180 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1186 inp = sotoinpcb(so);
1188 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1190 for (int i = 0; i < count; i++)
1192 return (ECONNRESET);
1194 tp = intotcpcb(inp);
1196 SOCKBUF_LOCK(&so->so_snd);
1197 error = sbready(&so->so_snd, m, count);
1198 SOCKBUF_UNLOCK(&so->so_snd);
1200 error = tp->t_fb->tfb_tcp_output(tp);
1207 * Abort the TCP. Drop the connection abruptly.
1210 tcp_usr_abort(struct socket *so)
1213 struct tcpcb *tp = NULL;
1214 struct epoch_tracker et;
1217 inp = sotoinpcb(so);
1218 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1220 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1222 KASSERT(inp->inp_socket != NULL,
1223 ("tcp_usr_abort: inp_socket == NULL"));
1226 * If we still have full TCP state, and we're not dropped, drop.
1228 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1229 !(inp->inp_flags & INP_DROPPED)) {
1230 tp = intotcpcb(inp);
1232 tp = tcp_drop(tp, ECONNABORTED);
1235 TCPDEBUG2(PRU_ABORT);
1236 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1238 if (!(inp->inp_flags & INP_DROPPED)) {
1240 so->so_state |= SS_PROTOREF;
1242 inp->inp_flags |= INP_SOCKREF;
1246 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1250 * TCP socket is closed. Start friendly disconnect.
1253 tcp_usr_close(struct socket *so)
1256 struct tcpcb *tp = NULL;
1257 struct epoch_tracker et;
1260 inp = sotoinpcb(so);
1261 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1263 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1265 KASSERT(inp->inp_socket != NULL,
1266 ("tcp_usr_close: inp_socket == NULL"));
1269 * If we still have full TCP state, and we're not dropped, initiate
1272 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1273 !(inp->inp_flags & INP_DROPPED)) {
1274 tp = intotcpcb(inp);
1277 TCPDEBUG2(PRU_CLOSE);
1278 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1280 if (!(inp->inp_flags & INP_DROPPED)) {
1282 so->so_state |= SS_PROTOREF;
1284 inp->inp_flags |= INP_SOCKREF;
1287 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1291 * Receive out-of-band data.
1294 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1298 struct tcpcb *tp = NULL;
1301 inp = sotoinpcb(so);
1302 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1304 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1308 tp = intotcpcb(inp);
1310 if ((so->so_oobmark == 0 &&
1311 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1312 so->so_options & SO_OOBINLINE ||
1313 tp->t_oobflags & TCPOOB_HADDATA) {
1317 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1318 error = EWOULDBLOCK;
1322 *mtod(m, caddr_t) = tp->t_iobc;
1323 if ((flags & MSG_PEEK) == 0)
1324 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1327 TCPDEBUG2(PRU_RCVOOB);
1328 TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1334 struct pr_usrreqs tcp_usrreqs = {
1335 .pru_abort = tcp_usr_abort,
1336 .pru_accept = tcp_usr_accept,
1337 .pru_attach = tcp_usr_attach,
1338 .pru_bind = tcp_usr_bind,
1339 .pru_connect = tcp_usr_connect,
1340 .pru_control = in_control,
1341 .pru_detach = tcp_usr_detach,
1342 .pru_disconnect = tcp_usr_disconnect,
1343 .pru_listen = tcp_usr_listen,
1344 .pru_peeraddr = in_getpeeraddr,
1345 .pru_rcvd = tcp_usr_rcvd,
1346 .pru_rcvoob = tcp_usr_rcvoob,
1347 .pru_send = tcp_usr_send,
1348 .pru_ready = tcp_usr_ready,
1349 .pru_shutdown = tcp_usr_shutdown,
1350 .pru_sockaddr = in_getsockaddr,
1351 .pru_sosetlabel = in_pcbsosetlabel,
1352 .pru_close = tcp_usr_close,
1357 struct pr_usrreqs tcp6_usrreqs = {
1358 .pru_abort = tcp_usr_abort,
1359 .pru_accept = tcp6_usr_accept,
1360 .pru_attach = tcp_usr_attach,
1361 .pru_bind = tcp6_usr_bind,
1362 .pru_connect = tcp6_usr_connect,
1363 .pru_control = in6_control,
1364 .pru_detach = tcp_usr_detach,
1365 .pru_disconnect = tcp_usr_disconnect,
1366 .pru_listen = tcp6_usr_listen,
1367 .pru_peeraddr = in6_mapped_peeraddr,
1368 .pru_rcvd = tcp_usr_rcvd,
1369 .pru_rcvoob = tcp_usr_rcvoob,
1370 .pru_send = tcp_usr_send,
1371 .pru_ready = tcp_usr_ready,
1372 .pru_shutdown = tcp_usr_shutdown,
1373 .pru_sockaddr = in6_mapped_sockaddr,
1374 .pru_sosetlabel = in_pcbsosetlabel,
1375 .pru_close = tcp_usr_close,
1381 * Common subroutine to open a TCP connection to remote host specified
1382 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1383 * port number if needed. Call in_pcbconnect_setup to do the routing and
1384 * to choose a local host address (interface). If there is an existing
1385 * incarnation of the same connection in TIME-WAIT state and if the remote
1386 * host was sending CC options and if the connection duration was < MSL, then
1387 * truncate the previous TIME-WAIT state and proceed.
1388 * Initialize connection parameters and enter SYN-SENT state.
1391 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1393 struct inpcb *inp = tp->t_inpcb, *oinp;
1394 struct socket *so = inp->inp_socket;
1395 struct in_addr laddr;
1399 INP_WLOCK_ASSERT(inp);
1400 INP_HASH_WLOCK(&V_tcbinfo);
1402 if (inp->inp_lport == 0) {
1403 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1409 * Cannot simply call in_pcbconnect, because there might be an
1410 * earlier incarnation of this same connection still in
1411 * TIME_WAIT state, creating an ADDRINUSE error.
1413 laddr = inp->inp_laddr;
1414 lport = inp->inp_lport;
1415 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1416 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1417 if (error && oinp == NULL)
1423 inp->inp_laddr = laddr;
1425 INP_HASH_WUNLOCK(&V_tcbinfo);
1428 * Compute window scaling to request:
1429 * Scale to fit into sweet spot. See tcp_syncache.c.
1430 * XXX: This should move to tcp_output().
1432 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1433 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1434 tp->request_r_scale++;
1437 TCPSTAT_INC(tcps_connattempt);
1438 tcp_state_change(tp, TCPS_SYN_SENT);
1439 tp->iss = tcp_new_isn(tp);
1440 tcp_sendseqinit(tp);
1445 INP_HASH_WUNLOCK(&V_tcbinfo);
1452 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1454 struct inpcb *inp = tp->t_inpcb;
1457 INP_WLOCK_ASSERT(inp);
1458 INP_HASH_WLOCK(&V_tcbinfo);
1460 if (inp->inp_lport == 0) {
1461 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1465 error = in6_pcbconnect(inp, nam, td->td_ucred);
1468 INP_HASH_WUNLOCK(&V_tcbinfo);
1470 /* Compute window scaling to request. */
1471 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1472 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1473 tp->request_r_scale++;
1475 soisconnecting(inp->inp_socket);
1476 TCPSTAT_INC(tcps_connattempt);
1477 tcp_state_change(tp, TCPS_SYN_SENT);
1478 tp->iss = tcp_new_isn(tp);
1479 tcp_sendseqinit(tp);
1484 INP_HASH_WUNLOCK(&V_tcbinfo);
1490 * Export TCP internal state information via a struct tcp_info, based on the
1491 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1492 * (TCP state machine, etc). We export all information using FreeBSD-native
1493 * constants -- for example, the numeric values for tcpi_state will differ
1497 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1500 INP_WLOCK_ASSERT(tp->t_inpcb);
1501 bzero(ti, sizeof(*ti));
1503 ti->tcpi_state = tp->t_state;
1504 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1505 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1506 if (tp->t_flags & TF_SACK_PERMIT)
1507 ti->tcpi_options |= TCPI_OPT_SACK;
1508 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1509 ti->tcpi_options |= TCPI_OPT_WSCALE;
1510 ti->tcpi_snd_wscale = tp->snd_scale;
1511 ti->tcpi_rcv_wscale = tp->rcv_scale;
1513 if (tp->t_flags & TF_ECN_PERMIT)
1514 ti->tcpi_options |= TCPI_OPT_ECN;
1516 ti->tcpi_rto = tp->t_rxtcur * tick;
1517 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1518 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1519 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1521 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1522 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1525 * FreeBSD-specific extension fields for tcp_info.
1527 ti->tcpi_rcv_space = tp->rcv_wnd;
1528 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1529 ti->tcpi_snd_wnd = tp->snd_wnd;
1530 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1531 ti->tcpi_snd_nxt = tp->snd_nxt;
1532 ti->tcpi_snd_mss = tp->t_maxseg;
1533 ti->tcpi_rcv_mss = tp->t_maxseg;
1534 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1535 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1536 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1538 if (tp->t_flags & TF_TOE) {
1539 ti->tcpi_options |= TCPI_OPT_TOE;
1540 tcp_offload_tcp_info(tp, ti);
1546 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1547 * socket option arguments. When it re-acquires the lock after the copy, it
1548 * has to revalidate that the connection is still valid for the socket
1551 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
1553 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1556 return (ECONNRESET); \
1558 tp = intotcpcb(inp); \
1560 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1563 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1568 struct tcp_function_block *blk;
1569 struct tcp_function_set fsn;
1572 inp = sotoinpcb(so);
1573 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1575 if (sopt->sopt_level != IPPROTO_TCP) {
1577 if (inp->inp_vflag & INP_IPV6PROTO) {
1579 error = ip6_ctloutput(so, sopt);
1582 #if defined(INET6) && defined(INET)
1588 error = ip_ctloutput(so, sopt);
1593 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1595 return (ECONNRESET);
1597 tp = intotcpcb(inp);
1599 * Protect the TCP option TCP_FUNCTION_BLK so
1600 * that a sub-function can *never* overwrite this.
1602 if ((sopt->sopt_dir == SOPT_SET) &&
1603 (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1605 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1609 INP_WLOCK_RECHECK(inp);
1610 blk = find_and_ref_tcp_functions(&fsn);
1615 if (tp->t_fb == blk) {
1616 /* You already have this */
1617 refcount_release(&blk->tfb_refcnt);
1621 if (tp->t_state != TCPS_CLOSED) {
1624 * The user has advanced the state
1625 * past the initial point, we may not
1626 * be able to switch.
1628 if (blk->tfb_tcp_handoff_ok != NULL) {
1630 * Does the stack provide a
1631 * query mechanism, if so it may
1632 * still be possible?
1634 error = (*blk->tfb_tcp_handoff_ok)(tp);
1637 refcount_release(&blk->tfb_refcnt);
1642 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1643 refcount_release(&blk->tfb_refcnt);
1648 * Release the old refcnt, the
1649 * lookup acquired a ref on the
1652 if (tp->t_fb->tfb_tcp_fb_fini) {
1654 * Tell the stack to cleanup with 0 i.e.
1655 * the tcb is not going away.
1657 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1660 /* Assure that we are not on any hpts */
1661 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1663 if (blk->tfb_tcp_fb_init) {
1664 error = (*blk->tfb_tcp_fb_init)(tp);
1666 refcount_release(&blk->tfb_refcnt);
1667 if (tp->t_fb->tfb_tcp_fb_init) {
1668 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) {
1669 /* Fall back failed, drop the connection */
1678 refcount_release(&tp->t_fb->tfb_refcnt);
1681 if (tp->t_flags & TF_TOE) {
1682 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1689 } else if ((sopt->sopt_dir == SOPT_GET) &&
1690 (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1691 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1692 TCP_FUNCTION_NAME_LEN_MAX);
1693 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1694 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1696 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1699 /* Pass in the INP locked, called must unlock it */
1700 return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1704 * If this assert becomes untrue, we need to change the size of the buf
1705 * variable in tcp_default_ctloutput().
1708 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1709 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1713 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1715 int error, opt, optval;
1718 struct cc_algo *algo;
1719 char *pbuf, buf[TCP_LOG_ID_LEN];
1723 * For TCP_CCALGOOPT forward the control to CC module, for both
1724 * SOPT_SET and SOPT_GET.
1726 switch (sopt->sopt_name) {
1729 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1730 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1731 sopt->sopt_valsize);
1736 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1737 if (CC_ALGO(tp)->ctl_output != NULL)
1738 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1742 if (error == 0 && sopt->sopt_dir == SOPT_GET)
1743 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1748 switch (sopt->sopt_dir) {
1750 switch (sopt->sopt_name) {
1751 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1753 if (!TCPMD5_ENABLED()) {
1755 return (ENOPROTOOPT);
1757 error = TCPMD5_PCBCTL(inp, sopt);
1760 goto unlock_and_done;
1766 error = sooptcopyin(sopt, &optval, sizeof optval,
1771 INP_WLOCK_RECHECK(inp);
1772 switch (sopt->sopt_name) {
1780 opt = 0; /* dead code to fool gcc */
1787 tp->t_flags &= ~opt;
1790 if (tp->t_flags & TF_TOE) {
1791 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1800 error = sooptcopyin(sopt, &optval, sizeof optval,
1805 INP_WLOCK_RECHECK(inp);
1807 tp->t_flags |= TF_NOPUSH;
1808 else if (tp->t_flags & TF_NOPUSH) {
1809 tp->t_flags &= ~TF_NOPUSH;
1810 if (TCPS_HAVEESTABLISHED(tp->t_state))
1811 error = tp->t_fb->tfb_tcp_output(tp);
1813 goto unlock_and_done;
1817 error = sooptcopyin(sopt, &optval, sizeof optval,
1822 INP_WLOCK_RECHECK(inp);
1823 if (optval > 0 && optval <= tp->t_maxseg &&
1824 optval + 40 >= V_tcp_minmss)
1825 tp->t_maxseg = optval;
1828 goto unlock_and_done;
1835 case TCP_CONGESTION:
1837 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1840 buf[sopt->sopt_valsize] = '\0';
1841 INP_WLOCK_RECHECK(inp);
1843 STAILQ_FOREACH(algo, &cc_list, entries)
1844 if (strncmp(buf, algo->name,
1845 TCP_CA_NAME_MAX) == 0)
1854 * We hold a write lock over the tcb so it's safe to
1855 * do these things without ordering concerns.
1857 if (CC_ALGO(tp)->cb_destroy != NULL)
1858 CC_ALGO(tp)->cb_destroy(tp->ccv);
1862 * If something goes pear shaped initialising the new
1863 * algo, fall back to newreno (which does not
1864 * require initialisation).
1866 if (algo->cb_init != NULL &&
1867 algo->cb_init(tp->ccv) != 0) {
1868 CC_ALGO(tp) = &newreno_cc_algo;
1870 * The only reason init should fail is
1871 * because of malloc.
1882 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1886 if (ui > (UINT_MAX / hz)) {
1892 INP_WLOCK_RECHECK(inp);
1893 switch (sopt->sopt_name) {
1895 tp->t_keepidle = ui;
1897 * XXX: better check current remaining
1898 * timeout and "merge" it with new value.
1900 if ((tp->t_state > TCPS_LISTEN) &&
1901 (tp->t_state <= TCPS_CLOSING))
1902 tcp_timer_activate(tp, TT_KEEP,
1906 tp->t_keepintvl = ui;
1907 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1908 (TP_MAXIDLE(tp) > 0))
1909 tcp_timer_activate(tp, TT_2MSL,
1913 tp->t_keepinit = ui;
1914 if (tp->t_state == TCPS_SYN_RECEIVED ||
1915 tp->t_state == TCPS_SYN_SENT)
1916 tcp_timer_activate(tp, TT_KEEP,
1920 goto unlock_and_done;
1924 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1928 INP_WLOCK_RECHECK(inp);
1930 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1931 (TP_MAXIDLE(tp) > 0))
1932 tcp_timer_activate(tp, TT_2MSL,
1934 goto unlock_and_done;
1940 error = sooptcopyin(sopt, &optval, sizeof optval,
1945 INP_WLOCK_RECHECK(inp);
1947 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1948 &(tp->t_outpkts) : &(tp->t_inpkts),
1952 goto unlock_and_done;
1955 case TCP_FASTOPEN: {
1956 struct tcp_fastopen tfo_optval;
1959 if (!V_tcp_fastopen_client_enable &&
1960 !V_tcp_fastopen_server_enable)
1963 error = sooptcopyin(sopt, &tfo_optval,
1964 sizeof(tfo_optval), sizeof(int));
1968 INP_WLOCK_RECHECK(inp);
1969 if (tfo_optval.enable) {
1970 if (tp->t_state == TCPS_LISTEN) {
1971 if (!V_tcp_fastopen_server_enable) {
1973 goto unlock_and_done;
1976 tp->t_flags |= TF_FASTOPEN;
1977 if (tp->t_tfo_pending == NULL)
1979 tcp_fastopen_alloc_counter();
1982 * If a pre-shared key was provided,
1983 * stash it in the client cookie
1984 * field of the tcpcb for use during
1987 if (sopt->sopt_valsize ==
1988 sizeof(tfo_optval)) {
1989 memcpy(tp->t_tfo_cookie.client,
1991 TCP_FASTOPEN_PSK_LEN);
1992 tp->t_tfo_client_cookie_len =
1993 TCP_FASTOPEN_PSK_LEN;
1995 tp->t_flags |= TF_FASTOPEN;
1998 tp->t_flags &= ~TF_FASTOPEN;
1999 goto unlock_and_done;
2005 error = sooptcopyin(sopt, &optval, sizeof optval,
2010 INP_WLOCK_RECHECK(inp);
2011 error = tcp_log_state_change(tp, optval);
2012 goto unlock_and_done;
2021 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2024 buf[sopt->sopt_valsize] = '\0';
2025 INP_WLOCK_RECHECK(inp);
2026 error = tcp_log_set_id(tp, buf);
2027 /* tcp_log_set_id() unlocks the INP. */
2034 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2037 buf[sopt->sopt_valsize] = '\0';
2038 INP_WLOCK_RECHECK(inp);
2039 if (sopt->sopt_name == TCP_LOGDUMP) {
2040 error = tcp_log_dump_tp_logbuf(tp, buf,
2044 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2046 * tcp_log_dump_tp_bucket_logbufs() drops the
2055 error = ENOPROTOOPT;
2061 tp = intotcpcb(inp);
2062 switch (sopt->sopt_name) {
2063 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2065 if (!TCPMD5_ENABLED()) {
2067 return (ENOPROTOOPT);
2069 error = TCPMD5_PCBCTL(inp, sopt);
2074 optval = tp->t_flags & TF_NODELAY;
2076 error = sooptcopyout(sopt, &optval, sizeof optval);
2079 optval = tp->t_maxseg;
2081 error = sooptcopyout(sopt, &optval, sizeof optval);
2084 optval = tp->t_flags & TF_NOOPT;
2086 error = sooptcopyout(sopt, &optval, sizeof optval);
2089 optval = tp->t_flags & TF_NOPUSH;
2091 error = sooptcopyout(sopt, &optval, sizeof optval);
2094 tcp_fill_info(tp, &ti);
2096 error = sooptcopyout(sopt, &ti, sizeof ti);
2098 case TCP_CONGESTION:
2099 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2101 error = sooptcopyout(sopt, buf, len + 1);
2107 switch (sopt->sopt_name) {
2109 ui = TP_KEEPIDLE(tp) / hz;
2112 ui = TP_KEEPINTVL(tp) / hz;
2115 ui = TP_KEEPINIT(tp) / hz;
2118 ui = TP_KEEPCNT(tp);
2122 error = sooptcopyout(sopt, &ui, sizeof(ui));
2127 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2128 &(tp->t_outpkts) : &(tp->t_inpkts));
2130 error = sooptcopyout(sopt, &optval, sizeof optval);
2134 optval = tp->t_flags & TF_FASTOPEN;
2136 error = sooptcopyout(sopt, &optval, sizeof optval);
2140 optval = tp->t_logstate;
2142 error = sooptcopyout(sopt, &optval, sizeof(optval));
2145 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2146 error = tcp_log_getlogbuf(sopt, tp);
2149 len = tcp_log_get_id(tp, buf);
2151 error = sooptcopyout(sopt, buf, len + 1);
2161 error = ENOPROTOOPT;
2168 #undef INP_WLOCK_RECHECK
2169 #undef INP_WLOCK_RECHECK_CLEANUP
2172 * Attach TCP protocol to socket, allocating
2173 * internet protocol control block, tcp control block,
2174 * bufer space, and entering LISTEN state if to accept connections.
2177 tcp_attach(struct socket *so)
2181 struct epoch_tracker et;
2184 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2185 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
2189 so->so_rcv.sb_flags |= SB_AUTOSIZE;
2190 so->so_snd.sb_flags |= SB_AUTOSIZE;
2191 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2192 error = in_pcballoc(so, &V_tcbinfo);
2194 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2197 inp = sotoinpcb(so);
2199 if (inp->inp_vflag & INP_IPV6PROTO) {
2200 inp->inp_vflag |= INP_IPV6;
2201 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
2202 inp->inp_vflag |= INP_IPV4;
2203 inp->in6p_hops = -1; /* use kernel default */
2207 inp->inp_vflag |= INP_IPV4;
2208 tp = tcp_newtcpcb(inp);
2212 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2215 tp->t_state = TCPS_CLOSED;
2217 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2218 TCPSTATES_INC(TCPS_CLOSED);
2223 * Initiate (or continue) disconnect.
2224 * If embryonic state, just send reset (once).
2225 * If in ``let data drain'' option and linger null, just drop.
2226 * Otherwise (hard), mark socket disconnecting and drop
2227 * current input data; switch states based on user close, and
2228 * send segment to peer (with FIN).
2231 tcp_disconnect(struct tcpcb *tp)
2233 struct inpcb *inp = tp->t_inpcb;
2234 struct socket *so = inp->inp_socket;
2236 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2237 INP_WLOCK_ASSERT(inp);
2240 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2241 * socket is still open.
2243 if (tp->t_state < TCPS_ESTABLISHED &&
2244 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2247 ("tcp_disconnect: tcp_close() returned NULL"));
2248 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2249 tp = tcp_drop(tp, 0);
2251 ("tcp_disconnect: tcp_drop() returned NULL"));
2253 soisdisconnecting(so);
2254 sbflush(&so->so_rcv);
2256 if (!(inp->inp_flags & INP_DROPPED))
2257 tp->t_fb->tfb_tcp_output(tp);
2262 * User issued close, and wish to trail through shutdown states:
2263 * if never received SYN, just forget it. If got a SYN from peer,
2264 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2265 * If already got a FIN from peer, then almost done; go to LAST_ACK
2266 * state. In all other cases, have already sent FIN to peer (e.g.
2267 * after PRU_SHUTDOWN), and just have to play tedious game waiting
2268 * for peer to send FIN or not respond to keep-alives, etc.
2269 * We can let the user exit from the close as soon as the FIN is acked.
2272 tcp_usrclosed(struct tcpcb *tp)
2275 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2276 INP_WLOCK_ASSERT(tp->t_inpcb);
2278 switch (tp->t_state) {
2281 tcp_offload_listen_stop(tp);
2283 tcp_state_change(tp, TCPS_CLOSED);
2288 * tcp_close() should never return NULL here as the socket is
2292 ("tcp_usrclosed: tcp_close() returned NULL"));
2296 case TCPS_SYN_RECEIVED:
2297 tp->t_flags |= TF_NEEDFIN;
2300 case TCPS_ESTABLISHED:
2301 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2304 case TCPS_CLOSE_WAIT:
2305 tcp_state_change(tp, TCPS_LAST_ACK);
2308 if (tp->t_state >= TCPS_FIN_WAIT_2) {
2309 soisdisconnected(tp->t_inpcb->inp_socket);
2310 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2311 if (tp->t_state == TCPS_FIN_WAIT_2) {
2314 timeout = (tcp_fast_finwait2_recycle) ?
2315 tcp_finwait2_timeout : TP_MAXIDLE(tp);
2316 tcp_timer_activate(tp, TT_2MSL, timeout);
2323 db_print_indent(int indent)
2327 for (i = 0; i < indent; i++)
2332 db_print_tstate(int t_state)
2337 db_printf("TCPS_CLOSED");
2341 db_printf("TCPS_LISTEN");
2345 db_printf("TCPS_SYN_SENT");
2348 case TCPS_SYN_RECEIVED:
2349 db_printf("TCPS_SYN_RECEIVED");
2352 case TCPS_ESTABLISHED:
2353 db_printf("TCPS_ESTABLISHED");
2356 case TCPS_CLOSE_WAIT:
2357 db_printf("TCPS_CLOSE_WAIT");
2360 case TCPS_FIN_WAIT_1:
2361 db_printf("TCPS_FIN_WAIT_1");
2365 db_printf("TCPS_CLOSING");
2369 db_printf("TCPS_LAST_ACK");
2372 case TCPS_FIN_WAIT_2:
2373 db_printf("TCPS_FIN_WAIT_2");
2376 case TCPS_TIME_WAIT:
2377 db_printf("TCPS_TIME_WAIT");
2381 db_printf("unknown");
2387 db_print_tflags(u_int t_flags)
2392 if (t_flags & TF_ACKNOW) {
2393 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2396 if (t_flags & TF_DELACK) {
2397 db_printf("%sTF_DELACK", comma ? ", " : "");
2400 if (t_flags & TF_NODELAY) {
2401 db_printf("%sTF_NODELAY", comma ? ", " : "");
2404 if (t_flags & TF_NOOPT) {
2405 db_printf("%sTF_NOOPT", comma ? ", " : "");
2408 if (t_flags & TF_SENTFIN) {
2409 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2412 if (t_flags & TF_REQ_SCALE) {
2413 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2416 if (t_flags & TF_RCVD_SCALE) {
2417 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2420 if (t_flags & TF_REQ_TSTMP) {
2421 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2424 if (t_flags & TF_RCVD_TSTMP) {
2425 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2428 if (t_flags & TF_SACK_PERMIT) {
2429 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2432 if (t_flags & TF_NEEDSYN) {
2433 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2436 if (t_flags & TF_NEEDFIN) {
2437 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2440 if (t_flags & TF_NOPUSH) {
2441 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2444 if (t_flags & TF_MORETOCOME) {
2445 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2448 if (t_flags & TF_LQ_OVERFLOW) {
2449 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2452 if (t_flags & TF_LASTIDLE) {
2453 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2456 if (t_flags & TF_RXWIN0SENT) {
2457 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2460 if (t_flags & TF_FASTRECOVERY) {
2461 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2464 if (t_flags & TF_CONGRECOVERY) {
2465 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2468 if (t_flags & TF_WASFRECOVERY) {
2469 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2472 if (t_flags & TF_SIGNATURE) {
2473 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2476 if (t_flags & TF_FORCEDATA) {
2477 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2480 if (t_flags & TF_TSO) {
2481 db_printf("%sTF_TSO", comma ? ", " : "");
2484 if (t_flags & TF_ECN_PERMIT) {
2485 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2488 if (t_flags & TF_FASTOPEN) {
2489 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2495 db_print_toobflags(char t_oobflags)
2500 if (t_oobflags & TCPOOB_HAVEDATA) {
2501 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2504 if (t_oobflags & TCPOOB_HADDATA) {
2505 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2511 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2514 db_print_indent(indent);
2515 db_printf("%s at %p\n", name, tp);
2519 db_print_indent(indent);
2520 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
2521 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2523 db_print_indent(indent);
2524 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
2525 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2527 db_print_indent(indent);
2528 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2529 &tp->t_timers->tt_delack, tp->t_inpcb);
2531 db_print_indent(indent);
2532 db_printf("t_state: %d (", tp->t_state);
2533 db_print_tstate(tp->t_state);
2536 db_print_indent(indent);
2537 db_printf("t_flags: 0x%x (", tp->t_flags);
2538 db_print_tflags(tp->t_flags);
2541 db_print_indent(indent);
2542 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
2543 tp->snd_una, tp->snd_max, tp->snd_nxt);
2545 db_print_indent(indent);
2546 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
2547 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2549 db_print_indent(indent);
2550 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
2551 tp->iss, tp->irs, tp->rcv_nxt);
2553 db_print_indent(indent);
2554 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
2555 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2557 db_print_indent(indent);
2558 db_printf("snd_wnd: %u snd_cwnd: %u\n",
2559 tp->snd_wnd, tp->snd_cwnd);
2561 db_print_indent(indent);
2562 db_printf("snd_ssthresh: %u snd_recover: "
2563 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2565 db_print_indent(indent);
2566 db_printf("t_rcvtime: %u t_startime: %u\n",
2567 tp->t_rcvtime, tp->t_starttime);
2569 db_print_indent(indent);
2570 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
2571 tp->t_rtttime, tp->t_rtseq);
2573 db_print_indent(indent);
2574 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
2575 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2577 db_print_indent(indent);
2578 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
2579 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2582 db_print_indent(indent);
2583 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n",
2584 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2586 db_print_indent(indent);
2587 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2588 db_print_toobflags(tp->t_oobflags);
2589 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
2591 db_print_indent(indent);
2592 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
2593 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2595 db_print_indent(indent);
2596 db_printf("ts_recent: %u ts_recent_age: %u\n",
2597 tp->ts_recent, tp->ts_recent_age);
2599 db_print_indent(indent);
2600 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
2601 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2603 db_print_indent(indent);
2604 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
2605 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2606 tp->snd_recover_prev, tp->t_badrxtwin);
2608 db_print_indent(indent);
2609 db_printf("snd_numholes: %d snd_holes first: %p\n",
2610 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2612 db_print_indent(indent);
2613 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: "
2614 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2616 /* Skip sackblks, sackhint. */
2618 db_print_indent(indent);
2619 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
2620 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2623 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2628 db_printf("usage: show tcpcb <addr>\n");
2631 tp = (struct tcpcb *)addr;
2633 db_print_tcpcb(tp, "tcpcb", 0);