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_kern_tls.h"
48 #include "opt_tcpdebug.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/limits.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
56 #include <sys/kernel.h>
58 #include <sys/qmath.h>
59 #include <sys/sysctl.h>
62 #include <sys/domain.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
69 #include <sys/stats.h>
76 #include <net/if_var.h>
77 #include <net/route.h>
80 #include <netinet/in.h>
81 #include <netinet/in_kdtrace.h>
82 #include <netinet/in_pcb.h>
83 #include <netinet/in_systm.h>
84 #include <netinet/in_var.h>
85 #include <netinet/ip.h>
86 #include <netinet/ip_var.h>
88 #include <netinet/ip6.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_log_buf.h>
99 #include <netinet/tcpip.h>
100 #include <netinet/cc/cc.h>
101 #include <netinet/tcp_fastopen.h>
102 #include <netinet/tcp_hpts.h>
104 #include <netinet/tcp_pcap.h>
107 #include <netinet/tcp_debug.h>
110 #include <netinet/tcp_offload.h>
112 #include <netipsec/ipsec_support.h>
115 #include <vm/vm_param.h>
117 #include <vm/vm_extern.h>
118 #include <vm/vm_map.h>
119 #include <vm/vm_page.h>
122 * TCP protocol interface to socket abstraction.
125 static int tcp_connect(struct tcpcb *, struct sockaddr *,
129 static int tcp6_connect(struct tcpcb *, struct sockaddr *,
132 static void tcp_disconnect(struct tcpcb *);
133 static void tcp_usrclosed(struct tcpcb *);
134 static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
136 static int tcp_pru_options_support(struct tcpcb *tp, int flags);
139 #define TCPDEBUG0 int ostate = 0
140 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
141 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
142 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
146 #define TCPDEBUG2(req)
150 * tcp_require_unique port requires a globally-unique source port for each
151 * outgoing connection. The default is to require the 4-tuple to be unique.
153 VNET_DEFINE(int, tcp_require_unique_port) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port,
155 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0,
156 "Require globally-unique ephemeral port for outgoing connections");
157 #define V_tcp_require_unique_port VNET(tcp_require_unique_port)
160 * TCP attaches to socket via pru_attach(), reserving space,
161 * and an internet control block.
164 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
167 struct tcpcb *tp = NULL;
172 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
175 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
176 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
181 so->so_rcv.sb_flags |= SB_AUTOSIZE;
182 so->so_snd.sb_flags |= SB_AUTOSIZE;
183 error = in_pcballoc(so, &V_tcbinfo);
188 if (inp->inp_vflag & INP_IPV6PROTO) {
189 inp->inp_vflag |= INP_IPV6;
190 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
191 inp->inp_vflag |= INP_IPV4;
192 inp->in6p_hops = -1; /* use kernel default */
196 inp->inp_vflag |= INP_IPV4;
197 tp = tcp_newtcpcb(inp);
204 tp->t_state = TCPS_CLOSED;
206 TCPSTATES_INC(TCPS_CLOSED);
208 TCPDEBUG2(PRU_ATTACH);
209 TCP_PROBE2(debug__user, tp, PRU_ATTACH);
214 * tcp_usr_detach is called when the socket layer loses its final reference
215 * to the socket, be it a file descriptor reference, a reference from TCP,
216 * etc. At this point, there is only one case in which we will keep around
217 * inpcb state: time wait.
220 tcp_usr_detach(struct socket *so)
226 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
228 KASSERT(so->so_pcb == inp && inp->inp_socket == so,
229 ("%s: socket %p inp %p mismatch", __func__, so, inp));
233 if (inp->inp_flags & INP_TIMEWAIT) {
235 * There are two cases to handle: one in which the time wait
236 * state is being discarded (INP_DROPPED), and one in which
237 * this connection will remain in timewait. In the former,
238 * it is time to discard all state (except tcptw, which has
239 * already been discarded by the timewait close code, which
240 * should be further up the call stack somewhere). In the
241 * latter case, we detach from the socket, but leave the pcb
242 * present until timewait ends.
244 * XXXRW: Would it be cleaner to free the tcptw here?
246 * Astute question indeed, from twtcp perspective there are
247 * four cases to consider:
249 * #1 tcp_usr_detach is called at tcptw creation time by
250 * tcp_twstart, then do not discard the newly created tcptw
251 * and leave inpcb present until timewait ends
252 * #2 tcp_usr_detach is called at tcptw creation time by
253 * tcp_twstart, but connection is local and tw will be
254 * discarded immediately
255 * #3 tcp_usr_detach is called at timewait end (or reuse) by
256 * tcp_twclose, then the tcptw has already been discarded
257 * (or reused) and inpcb is freed here
258 * #4 tcp_usr_detach is called() after timewait ends (or reuse)
259 * (e.g. by soclose), then tcptw has already been discarded
260 * (or reused) and inpcb is freed here
262 * In all three cases the tcptw should not be freed here.
264 if (inp->inp_flags & INP_DROPPED) {
265 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
266 "INP_DROPPED && tp != NULL"));
275 * If the connection is not in timewait, we consider two
276 * two conditions: one in which no further processing is
277 * necessary (dropped || embryonic), and one in which TCP is
278 * not yet done, but no longer requires the socket, so the
279 * pcb will persist for the time being.
281 * XXXRW: Does the second case still occur?
283 if (inp->inp_flags & INP_DROPPED ||
284 tp->t_state < TCPS_SYN_SENT) {
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_family != AF_INET) {
310 * Preserve compatibility with old programs.
312 if (nam->sa_family != AF_UNSPEC ||
313 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
314 sinp->sin_addr.s_addr != INADDR_ANY)
315 return (EAFNOSUPPORT);
316 nam->sa_family = AF_INET;
318 if (nam->sa_len != sizeof(*sinp))
322 * Must check for multicast addresses and disallow binding
325 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
326 return (EAFNOSUPPORT);
330 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
332 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
338 INP_HASH_WLOCK(&V_tcbinfo);
339 error = in_pcbbind(inp, nam, td->td_ucred);
340 INP_HASH_WUNLOCK(&V_tcbinfo);
343 TCP_PROBE2(debug__user, tp, PRU_BIND);
352 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
356 struct tcpcb *tp = NULL;
357 struct sockaddr_in6 *sin6;
360 sin6 = (struct sockaddr_in6 *)nam;
361 if (nam->sa_family != AF_INET6)
362 return (EAFNOSUPPORT);
363 if (nam->sa_len != sizeof(*sin6))
367 * Must check for multicast addresses and disallow binding
370 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
371 return (EAFNOSUPPORT);
375 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
377 vflagsav = inp->inp_vflag;
378 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
384 INP_HASH_WLOCK(&V_tcbinfo);
385 inp->inp_vflag &= ~INP_IPV4;
386 inp->inp_vflag |= INP_IPV6;
388 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
389 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
390 inp->inp_vflag |= INP_IPV4;
391 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
392 struct sockaddr_in sin;
394 in6_sin6_2_sin(&sin, sin6);
395 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
396 error = EAFNOSUPPORT;
397 INP_HASH_WUNLOCK(&V_tcbinfo);
400 inp->inp_vflag |= INP_IPV4;
401 inp->inp_vflag &= ~INP_IPV6;
402 error = in_pcbbind(inp, (struct sockaddr *)&sin,
404 INP_HASH_WUNLOCK(&V_tcbinfo);
409 error = in6_pcbbind(inp, nam, td->td_ucred);
410 INP_HASH_WUNLOCK(&V_tcbinfo);
413 inp->inp_vflag = vflagsav;
415 TCP_PROBE2(debug__user, tp, PRU_BIND);
423 * Prepare to accept connections.
426 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
430 struct tcpcb *tp = NULL;
434 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
436 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
443 error = solisten_proto_check(so);
448 if (inp->inp_lport == 0) {
449 INP_HASH_WLOCK(&V_tcbinfo);
450 error = in_pcbbind(inp, NULL, td->td_ucred);
451 INP_HASH_WUNLOCK(&V_tcbinfo);
454 tcp_state_change(tp, TCPS_LISTEN);
455 solisten_proto(so, backlog);
457 if ((so->so_options & SO_NO_OFFLOAD) == 0)
458 tcp_offload_listen_start(tp);
461 solisten_proto_abort(so);
465 if (IS_FASTOPEN(tp->t_flags))
466 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
469 TCPDEBUG2(PRU_LISTEN);
470 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
478 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
482 struct tcpcb *tp = NULL;
487 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
489 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
493 vflagsav = inp->inp_vflag;
497 error = solisten_proto_check(so);
502 INP_HASH_WLOCK(&V_tcbinfo);
503 if (inp->inp_lport == 0) {
504 inp->inp_vflag &= ~INP_IPV4;
505 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
506 inp->inp_vflag |= INP_IPV4;
507 error = in6_pcbbind(inp, NULL, td->td_ucred);
509 INP_HASH_WUNLOCK(&V_tcbinfo);
511 tcp_state_change(tp, TCPS_LISTEN);
512 solisten_proto(so, backlog);
514 if ((so->so_options & SO_NO_OFFLOAD) == 0)
515 tcp_offload_listen_start(tp);
518 solisten_proto_abort(so);
522 if (IS_FASTOPEN(tp->t_flags))
523 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
526 inp->inp_vflag = vflagsav;
529 TCPDEBUG2(PRU_LISTEN);
530 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
538 * Initiate connection to peer.
539 * Create a template for use in transmissions on this connection.
540 * Enter SYN_SENT state, and mark socket as connecting.
541 * Start keep-alive timer, and seed output sequence space.
542 * Send initial segment on connection.
545 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
547 struct epoch_tracker et;
550 struct tcpcb *tp = NULL;
551 struct sockaddr_in *sinp;
553 sinp = (struct sockaddr_in *)nam;
554 if (nam->sa_family != AF_INET)
555 return (EAFNOSUPPORT);
556 if (nam->sa_len != sizeof (*sinp))
560 * Must disallow TCP ``connections'' to multicast addresses.
562 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
563 return (EAFNOSUPPORT);
564 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST)
566 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
571 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
573 if (inp->inp_flags & INP_TIMEWAIT) {
577 if (inp->inp_flags & INP_DROPPED) {
578 error = ECONNREFUSED;
581 if (SOLISTENING(so)) {
588 if ((error = tcp_connect(tp, nam, td)) != 0)
591 if (registered_toedevs > 0 &&
592 (so->so_options & SO_NO_OFFLOAD) == 0 &&
593 (error = tcp_offload_connect(so, nam)) == 0)
596 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
597 error = tcp_output(tp);
598 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
599 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
603 TCPDEBUG2(PRU_CONNECT);
604 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
612 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
614 struct epoch_tracker et;
617 struct tcpcb *tp = NULL;
618 struct sockaddr_in6 *sin6;
624 sin6 = (struct sockaddr_in6 *)nam;
625 if (nam->sa_family != AF_INET6)
626 return (EAFNOSUPPORT);
627 if (nam->sa_len != sizeof (*sin6))
631 * Must disallow TCP ``connections'' to multicast addresses.
633 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
634 return (EAFNOSUPPORT);
637 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
639 vflagsav = inp->inp_vflag;
640 incflagsav = inp->inp_inc.inc_flags;
641 if (inp->inp_flags & INP_TIMEWAIT) {
645 if (inp->inp_flags & INP_DROPPED) {
646 error = ECONNREFUSED;
649 if (SOLISTENING(so)) {
657 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
658 * therefore probably require the hash lock, which isn't held here.
659 * Is this a significant problem?
661 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
662 struct sockaddr_in sin;
664 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
668 if ((inp->inp_vflag & INP_IPV4) == 0) {
669 error = EAFNOSUPPORT;
673 in6_sin6_2_sin(&sin, sin6);
674 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
675 error = EAFNOSUPPORT;
678 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
682 if ((error = prison_remote_ip4(td->td_ucred,
683 &sin.sin_addr)) != 0)
685 inp->inp_vflag |= INP_IPV4;
686 inp->inp_vflag &= ~INP_IPV6;
688 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
691 if (registered_toedevs > 0 &&
692 (so->so_options & SO_NO_OFFLOAD) == 0 &&
693 (error = tcp_offload_connect(so, nam)) == 0)
696 error = tcp_output(tp);
699 if ((inp->inp_vflag & INP_IPV6) == 0) {
700 error = EAFNOSUPPORT;
705 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
707 inp->inp_vflag &= ~INP_IPV4;
708 inp->inp_vflag |= INP_IPV6;
709 inp->inp_inc.inc_flags |= INC_ISIPV6;
710 if ((error = tcp6_connect(tp, nam, td)) != 0)
713 if (registered_toedevs > 0 &&
714 (so->so_options & SO_NO_OFFLOAD) == 0 &&
715 (error = tcp_offload_connect(so, nam)) == 0)
718 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
720 error = tcp_output(tp);
726 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
727 ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
729 * If the implicit bind in the connect call fails, restore
730 * the flags we modified.
732 if (error != 0 && inp->inp_lport == 0) {
733 inp->inp_vflag = vflagsav;
734 inp->inp_inc.inc_flags = incflagsav;
737 TCPDEBUG2(PRU_CONNECT);
738 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
745 * Initiate disconnect from peer.
746 * If connection never passed embryonic stage, just drop;
747 * else if don't need to let data drain, then can just drop anyways,
748 * else have to begin TCP shutdown process: mark socket disconnecting,
749 * drain unread data, state switch to reflect user close, and
750 * send segment (e.g. FIN) to peer. Socket will be really disconnected
751 * when peer sends FIN and acks ours.
753 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
756 tcp_usr_disconnect(struct socket *so)
759 struct tcpcb *tp = NULL;
760 struct epoch_tracker et;
766 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
768 if (inp->inp_flags & INP_TIMEWAIT)
770 if (inp->inp_flags & INP_DROPPED) {
778 TCPDEBUG2(PRU_DISCONNECT);
779 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
787 * Accept a connection. Essentially all the work is done at higher levels;
788 * just return the address of the peer, storing through addr.
791 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
794 struct inpcb *inp = NULL;
795 struct tcpcb *tp = NULL;
800 if (so->so_state & SS_ISDISCONNECTED)
801 return (ECONNABORTED);
804 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
806 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
807 error = ECONNABORTED;
814 * We inline in_getpeeraddr and COMMON_END here, so that we can
815 * copy the data of interest and defer the malloc until after we
818 port = inp->inp_fport;
819 addr = inp->inp_faddr;
822 TCPDEBUG2(PRU_ACCEPT);
823 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
826 *nam = in_sockaddr(port, &addr);
833 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
835 struct inpcb *inp = NULL;
837 struct tcpcb *tp = NULL;
839 struct in6_addr addr6;
840 struct epoch_tracker et;
845 if (so->so_state & SS_ISDISCONNECTED)
846 return (ECONNABORTED);
849 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
852 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
853 error = ECONNABORTED;
860 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
861 * copy the data of interest and defer the malloc until after we
864 if (inp->inp_vflag & INP_IPV4) {
866 port = inp->inp_fport;
867 addr = inp->inp_faddr;
869 port = inp->inp_fport;
870 addr6 = inp->in6p_faddr;
874 TCPDEBUG2(PRU_ACCEPT);
875 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
880 *nam = in6_v4mapsin6_sockaddr(port, &addr);
882 *nam = in6_sockaddr(port, &addr6);
889 * Mark the connection as being incapable of further output.
892 tcp_usr_shutdown(struct socket *so)
896 struct tcpcb *tp = NULL;
897 struct epoch_tracker et;
901 KASSERT(inp != NULL, ("inp == NULL"));
903 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
912 if (!(inp->inp_flags & INP_DROPPED))
913 error = tcp_output_nodrop(tp);
914 TCPDEBUG2(PRU_SHUTDOWN);
915 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
916 error = tcp_unlock_or_drop(tp, error);
923 * After a receive, possibly send window update to peer.
926 tcp_usr_rcvd(struct socket *so, int flags)
928 struct epoch_tracker et;
930 struct tcpcb *tp = NULL;
931 int outrv = 0, error = 0;
935 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
937 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
945 * For passively-created TFO connections, don't attempt a window
946 * update while still in SYN_RECEIVED as this may trigger an early
947 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with
948 * application response data, or failing that, when the DELACK timer
951 if (IS_FASTOPEN(tp->t_flags) &&
952 (tp->t_state == TCPS_SYN_RECEIVED))
955 if (tp->t_flags & TF_TOE)
956 tcp_offload_rcvd(tp);
959 outrv = tcp_output_nodrop(tp);
962 TCP_PROBE2(debug__user, tp, PRU_RCVD);
963 (void) tcp_unlock_or_drop(tp, outrv);
969 * Do a send by putting data in output queue and updating urgent
970 * marker if URG set. Possibly send more data. Unlike the other
971 * pru_*() routines, the mbuf chains are our responsibility. We
972 * must either enqueue them or free them. The other pru_* routines
973 * generally are caller-frees.
976 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
977 struct sockaddr *nam, struct mbuf *control, struct thread *td)
979 struct epoch_tracker et;
982 struct tcpcb *tp = NULL;
985 struct sockaddr_in sin;
987 struct sockaddr_in *sinp;
997 if (control != NULL) {
998 /* TCP doesn't do control messages (rights, creds, etc) */
999 if (control->m_len) {
1003 m_freem(control); /* empty control, just free it */
1006 inp = sotoinpcb(so);
1007 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
1009 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1010 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1013 return (ECONNRESET);
1016 vflagsav = inp->inp_vflag;
1017 incflagsav = inp->inp_inc.inc_flags;
1018 restoreflags = false;
1019 tp = intotcpcb(inp);
1021 NET_EPOCH_ENTER(et);
1022 if ((flags & PRUS_OOB) != 0 &&
1023 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1027 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1028 if (tp->t_state == TCPS_LISTEN) {
1032 switch (nam->sa_family) {
1035 sinp = (struct sockaddr_in *)nam;
1036 if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1040 if ((inp->inp_vflag & INP_IPV6) != 0) {
1041 error = EAFNOSUPPORT;
1044 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1045 error = EAFNOSUPPORT;
1048 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1052 if ((error = prison_remote_ip4(td->td_ucred,
1063 struct sockaddr_in6 *sin6;
1065 sin6 = (struct sockaddr_in6 *)nam;
1066 if (sin6->sin6_len != sizeof(*sin6)) {
1070 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1071 error = EAFNOSUPPORT;
1074 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1075 error = EAFNOSUPPORT;
1078 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1080 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1084 if ((inp->inp_vflag & INP_IPV4) == 0) {
1085 error = EAFNOSUPPORT;
1088 restoreflags = true;
1089 inp->inp_vflag &= ~INP_IPV6;
1091 in6_sin6_2_sin(sinp, sin6);
1093 ntohl(sinp->sin_addr.s_addr))) {
1094 error = EAFNOSUPPORT;
1097 if ((error = prison_remote_ip4(td->td_ucred,
1102 error = EAFNOSUPPORT;
1106 if ((inp->inp_vflag & INP_IPV6) == 0) {
1107 error = EAFNOSUPPORT;
1110 restoreflags = true;
1111 inp->inp_vflag &= ~INP_IPV4;
1112 inp->inp_inc.inc_flags |= INC_ISIPV6;
1113 if ((error = prison_remote_ip6(td->td_ucred,
1122 error = EAFNOSUPPORT;
1126 if (!(flags & PRUS_OOB)) {
1127 sbappendstream(&so->so_snd, m, flags);
1129 if (nam && tp->t_state < TCPS_SYN_SENT) {
1130 KASSERT(tp->t_state == TCPS_CLOSED,
1131 ("%s: tp %p is listening", __func__, tp));
1134 * Do implied connect if not yet connected,
1135 * initialize window to default value, and
1136 * initialize maxseg using peer's cached MSS.
1140 error = tcp6_connect(tp, nam, td);
1142 #if defined(INET6) && defined(INET)
1146 error = tcp_connect(tp,
1147 (struct sockaddr *)sinp, td);
1150 * The bind operation in tcp_connect succeeded. We
1151 * no longer want to restore the flags if later
1154 if (error == 0 || inp->inp_lport != 0)
1155 restoreflags = false;
1158 /* m is freed if PRUS_NOTREADY is unset. */
1159 sbflush(&so->so_snd);
1162 if (IS_FASTOPEN(tp->t_flags))
1163 tcp_fastopen_connect(tp);
1165 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1169 if (flags & PRUS_EOF) {
1171 * Close the send side of the connection after
1177 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1178 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1179 (tp->t_fbyte_out == 0) &&
1180 (so->so_snd.sb_ccc > 0)) {
1181 tp->t_fbyte_out = ticks;
1182 if (tp->t_fbyte_out == 0)
1183 tp->t_fbyte_out = 1;
1184 if (tp->t_fbyte_out && tp->t_fbyte_in)
1185 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1187 if (!(inp->inp_flags & INP_DROPPED) &&
1188 !(flags & PRUS_NOTREADY)) {
1189 if (flags & PRUS_MORETOCOME)
1190 tp->t_flags |= TF_MORETOCOME;
1191 error = tcp_output_nodrop(tp);
1192 if (flags & PRUS_MORETOCOME)
1193 tp->t_flags &= ~TF_MORETOCOME;
1197 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1199 SOCKBUF_LOCK(&so->so_snd);
1200 if (sbspace(&so->so_snd) < -512) {
1201 SOCKBUF_UNLOCK(&so->so_snd);
1206 * According to RFC961 (Assigned Protocols),
1207 * the urgent pointer points to the last octet
1208 * of urgent data. We continue, however,
1209 * to consider it to indicate the first octet
1210 * of data past the urgent section.
1211 * Otherwise, snd_up should be one lower.
1213 sbappendstream_locked(&so->so_snd, m, flags);
1214 SOCKBUF_UNLOCK(&so->so_snd);
1216 if (nam && tp->t_state < TCPS_SYN_SENT) {
1218 * Do implied connect if not yet connected,
1219 * initialize window to default value, and
1220 * initialize maxseg using peer's cached MSS.
1224 * Not going to contemplate SYN|URG
1226 if (IS_FASTOPEN(tp->t_flags))
1227 tp->t_flags &= ~TF_FASTOPEN;
1230 error = tcp6_connect(tp, nam, td);
1232 #if defined(INET6) && defined(INET)
1236 error = tcp_connect(tp,
1237 (struct sockaddr *)sinp, td);
1240 * The bind operation in tcp_connect succeeded. We
1241 * no longer want to restore the flags if later
1244 if (error == 0 || inp->inp_lport != 0)
1245 restoreflags = false;
1248 /* m is freed if PRUS_NOTREADY is unset. */
1249 sbflush(&so->so_snd);
1252 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1255 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1256 if ((flags & PRUS_NOTREADY) == 0) {
1257 tp->t_flags |= TF_FORCEDATA;
1258 error = tcp_output_nodrop(tp);
1259 tp->t_flags &= ~TF_FORCEDATA;
1262 TCP_LOG_EVENT(tp, NULL,
1263 &inp->inp_socket->so_rcv,
1264 &inp->inp_socket->so_snd,
1265 TCP_LOG_USERSEND, error,
1270 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1271 * responsible for freeing memory.
1273 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1277 * If the request was unsuccessful and we changed flags,
1278 * restore the original flags.
1280 if (error != 0 && restoreflags) {
1281 inp->inp_vflag = vflagsav;
1282 inp->inp_inc.inc_flags = incflagsav;
1284 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1285 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1286 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1287 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1288 error = tcp_unlock_or_drop(tp, error);
1294 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1296 struct epoch_tracker et;
1301 inp = sotoinpcb(so);
1303 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1305 mb_free_notready(m, count);
1306 return (ECONNRESET);
1308 tp = intotcpcb(inp);
1310 SOCKBUF_LOCK(&so->so_snd);
1311 error = sbready(&so->so_snd, m, count);
1312 SOCKBUF_UNLOCK(&so->so_snd);
1317 NET_EPOCH_ENTER(et);
1318 error = tcp_output_unlock(tp);
1325 * Abort the TCP. Drop the connection abruptly.
1328 tcp_usr_abort(struct socket *so)
1331 struct tcpcb *tp = NULL;
1332 struct epoch_tracker et;
1335 inp = sotoinpcb(so);
1336 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1338 NET_EPOCH_ENTER(et);
1340 KASSERT(inp->inp_socket != NULL,
1341 ("tcp_usr_abort: inp_socket == NULL"));
1344 * If we still have full TCP state, and we're not dropped, drop.
1346 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1347 !(inp->inp_flags & INP_DROPPED)) {
1348 tp = intotcpcb(inp);
1350 tp = tcp_drop(tp, ECONNABORTED);
1353 TCPDEBUG2(PRU_ABORT);
1354 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1356 if (!(inp->inp_flags & INP_DROPPED)) {
1358 so->so_state |= SS_PROTOREF;
1360 inp->inp_flags |= INP_SOCKREF;
1368 * TCP socket is closed. Start friendly disconnect.
1371 tcp_usr_close(struct socket *so)
1374 struct tcpcb *tp = NULL;
1375 struct epoch_tracker et;
1378 inp = sotoinpcb(so);
1379 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1381 NET_EPOCH_ENTER(et);
1383 KASSERT(inp->inp_socket != NULL,
1384 ("tcp_usr_close: inp_socket == NULL"));
1387 * If we still have full TCP state, and we're not dropped, initiate
1390 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1391 !(inp->inp_flags & INP_DROPPED)) {
1392 tp = intotcpcb(inp);
1395 TCPDEBUG2(PRU_CLOSE);
1396 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1398 if (!(inp->inp_flags & INP_DROPPED)) {
1400 so->so_state |= SS_PROTOREF;
1402 inp->inp_flags |= INP_SOCKREF;
1409 tcp_pru_options_support(struct tcpcb *tp, int flags)
1412 * If the specific TCP stack has a pru_options
1413 * specified then it does not always support
1414 * all the PRU_XX options and we must ask it.
1415 * If the function is not specified then all
1416 * of the PRU_XX options are supported.
1420 if (tp->t_fb->tfb_pru_options) {
1421 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1427 * Receive out-of-band data.
1430 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1434 struct tcpcb *tp = NULL;
1437 inp = sotoinpcb(so);
1438 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1440 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1444 tp = intotcpcb(inp);
1445 error = tcp_pru_options_support(tp, PRUS_OOB);
1450 if ((so->so_oobmark == 0 &&
1451 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1452 so->so_options & SO_OOBINLINE ||
1453 tp->t_oobflags & TCPOOB_HADDATA) {
1457 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1458 error = EWOULDBLOCK;
1462 *mtod(m, caddr_t) = tp->t_iobc;
1463 if ((flags & MSG_PEEK) == 0)
1464 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1467 TCPDEBUG2(PRU_RCVOOB);
1468 TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1474 struct pr_usrreqs tcp_usrreqs = {
1475 .pru_abort = tcp_usr_abort,
1476 .pru_accept = tcp_usr_accept,
1477 .pru_attach = tcp_usr_attach,
1478 .pru_bind = tcp_usr_bind,
1479 .pru_connect = tcp_usr_connect,
1480 .pru_control = in_control,
1481 .pru_detach = tcp_usr_detach,
1482 .pru_disconnect = tcp_usr_disconnect,
1483 .pru_listen = tcp_usr_listen,
1484 .pru_peeraddr = in_getpeeraddr,
1485 .pru_rcvd = tcp_usr_rcvd,
1486 .pru_rcvoob = tcp_usr_rcvoob,
1487 .pru_send = tcp_usr_send,
1488 .pru_ready = tcp_usr_ready,
1489 .pru_shutdown = tcp_usr_shutdown,
1490 .pru_sockaddr = in_getsockaddr,
1491 .pru_sosetlabel = in_pcbsosetlabel,
1492 .pru_close = tcp_usr_close,
1497 struct pr_usrreqs tcp6_usrreqs = {
1498 .pru_abort = tcp_usr_abort,
1499 .pru_accept = tcp6_usr_accept,
1500 .pru_attach = tcp_usr_attach,
1501 .pru_bind = tcp6_usr_bind,
1502 .pru_connect = tcp6_usr_connect,
1503 .pru_control = in6_control,
1504 .pru_detach = tcp_usr_detach,
1505 .pru_disconnect = tcp_usr_disconnect,
1506 .pru_listen = tcp6_usr_listen,
1507 .pru_peeraddr = in6_mapped_peeraddr,
1508 .pru_rcvd = tcp_usr_rcvd,
1509 .pru_rcvoob = tcp_usr_rcvoob,
1510 .pru_send = tcp_usr_send,
1511 .pru_ready = tcp_usr_ready,
1512 .pru_shutdown = tcp_usr_shutdown,
1513 .pru_sockaddr = in6_mapped_sockaddr,
1514 .pru_sosetlabel = in_pcbsosetlabel,
1515 .pru_close = tcp_usr_close,
1521 * Common subroutine to open a TCP connection to remote host specified
1522 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1523 * port number if needed. Call in_pcbconnect_setup to do the routing and
1524 * to choose a local host address (interface). If there is an existing
1525 * incarnation of the same connection in TIME-WAIT state and if the remote
1526 * host was sending CC options and if the connection duration was < MSL, then
1527 * truncate the previous TIME-WAIT state and proceed.
1528 * Initialize connection parameters and enter SYN-SENT state.
1531 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1533 struct inpcb *inp = tp->t_inpcb, *oinp;
1534 struct socket *so = inp->inp_socket;
1535 struct in_addr laddr;
1540 INP_WLOCK_ASSERT(inp);
1541 INP_HASH_WLOCK(&V_tcbinfo);
1543 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1544 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1550 * Cannot simply call in_pcbconnect, because there might be an
1551 * earlier incarnation of this same connection still in
1552 * TIME_WAIT state, creating an ADDRINUSE error.
1554 laddr = inp->inp_laddr;
1555 lport = inp->inp_lport;
1556 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1557 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1558 if (error && oinp == NULL)
1564 /* Handle initial bind if it hadn't been done in advance. */
1565 if (inp->inp_lport == 0) {
1566 inp->inp_lport = lport;
1567 if (in_pcbinshash(inp) != 0) {
1573 inp->inp_laddr = laddr;
1575 INP_HASH_WUNLOCK(&V_tcbinfo);
1578 * Compute window scaling to request:
1579 * Scale to fit into sweet spot. See tcp_syncache.c.
1580 * XXX: This should move to tcp_output().
1582 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1583 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1584 tp->request_r_scale++;
1587 TCPSTAT_INC(tcps_connattempt);
1588 tcp_state_change(tp, TCPS_SYN_SENT);
1589 tp->iss = tcp_new_isn(&inp->inp_inc);
1590 if (tp->t_flags & TF_REQ_TSTMP)
1591 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1592 tcp_sendseqinit(tp);
1597 INP_HASH_WUNLOCK(&V_tcbinfo);
1604 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1606 struct inpcb *inp = tp->t_inpcb;
1609 INP_WLOCK_ASSERT(inp);
1610 INP_HASH_WLOCK(&V_tcbinfo);
1612 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1613 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1617 error = in6_pcbconnect(inp, nam, td->td_ucred);
1620 INP_HASH_WUNLOCK(&V_tcbinfo);
1622 /* Compute window scaling to request. */
1623 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1624 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1625 tp->request_r_scale++;
1627 soisconnecting(inp->inp_socket);
1628 TCPSTAT_INC(tcps_connattempt);
1629 tcp_state_change(tp, TCPS_SYN_SENT);
1630 tp->iss = tcp_new_isn(&inp->inp_inc);
1631 if (tp->t_flags & TF_REQ_TSTMP)
1632 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1633 tcp_sendseqinit(tp);
1638 INP_HASH_WUNLOCK(&V_tcbinfo);
1644 * Export TCP internal state information via a struct tcp_info, based on the
1645 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1646 * (TCP state machine, etc). We export all information using FreeBSD-native
1647 * constants -- for example, the numeric values for tcpi_state will differ
1651 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1654 INP_WLOCK_ASSERT(tp->t_inpcb);
1655 bzero(ti, sizeof(*ti));
1657 ti->tcpi_state = tp->t_state;
1658 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1659 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1660 if (tp->t_flags & TF_SACK_PERMIT)
1661 ti->tcpi_options |= TCPI_OPT_SACK;
1662 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1663 ti->tcpi_options |= TCPI_OPT_WSCALE;
1664 ti->tcpi_snd_wscale = tp->snd_scale;
1665 ti->tcpi_rcv_wscale = tp->rcv_scale;
1667 if (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT))
1668 ti->tcpi_options |= TCPI_OPT_ECN;
1670 ti->tcpi_rto = tp->t_rxtcur * tick;
1671 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1672 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1673 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1675 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1676 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1679 * FreeBSD-specific extension fields for tcp_info.
1681 ti->tcpi_rcv_space = tp->rcv_wnd;
1682 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1683 ti->tcpi_snd_wnd = tp->snd_wnd;
1684 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1685 ti->tcpi_snd_nxt = tp->snd_nxt;
1686 ti->tcpi_snd_mss = tp->t_maxseg;
1687 ti->tcpi_rcv_mss = tp->t_maxseg;
1688 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1689 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1690 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1692 if (tp->t_flags & TF_TOE) {
1693 ti->tcpi_options |= TCPI_OPT_TOE;
1694 tcp_offload_tcp_info(tp, ti);
1700 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1701 * socket option arguments. When it re-acquires the lock after the copy, it
1702 * has to revalidate that the connection is still valid for the socket
1705 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
1707 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1710 return (ECONNRESET); \
1712 tp = intotcpcb(inp); \
1714 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1717 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
1719 struct socket *so = inp->inp_socket;
1720 struct tcpcb *tp = intotcpcb(inp);
1723 MPASS(sopt->sopt_dir == SOPT_SET);
1724 INP_WLOCK_ASSERT(inp);
1725 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1726 ("inp_flags == %x", inp->inp_flags));
1727 KASSERT(so != NULL, ("inp_socket == NULL"));
1729 if (sopt->sopt_level != IPPROTO_TCP) {
1732 if (inp->inp_vflag & INP_IPV6PROTO)
1733 error = ip6_ctloutput(so, sopt);
1735 #if defined(INET6) && defined(INET)
1739 error = ip_ctloutput(so, sopt);
1742 * When an IP-level socket option affects TCP, pass control
1743 * down to stack tfb_tcp_ctloutput, otherwise return what
1744 * IP level returned.
1746 switch (sopt->sopt_level) {
1749 if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
1751 switch (sopt->sopt_name) {
1753 /* Notify tcp stacks that care (e.g. RACK). */
1755 case IPV6_USE_MIN_MTU:
1756 /* Update t_maxseg accordingly. */
1765 switch (sopt->sopt_name) {
1767 inp->inp_ip_tos &= ~IPTOS_ECN_MASK;
1770 /* Notify tcp stacks that care (e.g. RACK). */
1781 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1783 return (ECONNRESET);
1785 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
1787 * Protect the TCP option TCP_FUNCTION_BLK so
1788 * that a sub-function can *never* overwrite this.
1790 struct tcp_function_set fsn;
1791 struct tcp_function_block *blk;
1794 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
1799 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1801 return (ECONNRESET);
1803 tp = intotcpcb(inp);
1805 blk = find_and_ref_tcp_functions(&fsn);
1810 if (tp->t_fb == blk) {
1811 /* You already have this */
1812 refcount_release(&blk->tfb_refcnt);
1816 if (tp->t_state != TCPS_CLOSED) {
1818 * The user has advanced the state
1819 * past the initial point, we may not
1820 * be able to switch.
1822 if (blk->tfb_tcp_handoff_ok != NULL) {
1824 * Does the stack provide a
1825 * query mechanism, if so it may
1826 * still be possible?
1828 error = (*blk->tfb_tcp_handoff_ok)(tp);
1832 refcount_release(&blk->tfb_refcnt);
1837 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1838 refcount_release(&blk->tfb_refcnt);
1843 * Release the old refcnt, the
1844 * lookup acquired a ref on the
1847 if (tp->t_fb->tfb_tcp_fb_fini) {
1848 struct epoch_tracker et;
1850 * Tell the stack to cleanup with 0 i.e.
1851 * the tcb is not going away.
1853 NET_EPOCH_ENTER(et);
1854 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1858 /* Assure that we are not on any hpts */
1859 tcp_hpts_remove(tp->t_inpcb);
1861 if (blk->tfb_tcp_fb_init) {
1862 error = (*blk->tfb_tcp_fb_init)(tp);
1864 refcount_release(&blk->tfb_refcnt);
1865 if (tp->t_fb->tfb_tcp_fb_init) {
1866 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) {
1867 /* Fall back failed, drop the connection */
1876 refcount_release(&tp->t_fb->tfb_refcnt);
1879 if (tp->t_flags & TF_TOE) {
1880 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1889 /* Pass in the INP locked, callee must unlock it. */
1890 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1894 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
1896 struct socket *so = inp->inp_socket;
1897 struct tcpcb *tp = intotcpcb(inp);
1900 MPASS(sopt->sopt_dir == SOPT_GET);
1901 INP_WLOCK_ASSERT(inp);
1902 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1903 ("inp_flags == %x", inp->inp_flags));
1904 KASSERT(so != NULL, ("inp_socket == NULL"));
1906 if (sopt->sopt_level != IPPROTO_TCP) {
1909 if (inp->inp_vflag & INP_IPV6PROTO)
1910 error = ip6_ctloutput(so, sopt);
1912 #if defined(INET6) && defined(INET)
1916 error = ip_ctloutput(so, sopt);
1920 if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
1921 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
1922 struct tcp_function_set fsn;
1924 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
1925 memset(&fsn, 0, sizeof(fsn));
1926 find_tcp_function_alias(tp->t_fb, &fsn);
1928 strncpy(fsn.function_set_name,
1929 tp->t_fb->tfb_tcp_block_name,
1930 TCP_FUNCTION_NAME_LEN_MAX);
1931 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1933 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1935 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1939 /* Pass in the INP locked, callee must unlock it. */
1940 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1944 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1948 inp = sotoinpcb(so);
1949 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1952 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1954 return (ECONNRESET);
1956 if (sopt->sopt_dir == SOPT_SET)
1957 return (tcp_ctloutput_set(inp, sopt));
1958 else if (sopt->sopt_dir == SOPT_GET)
1959 return (tcp_ctloutput_get(inp, sopt));
1961 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
1965 * If this assert becomes untrue, we need to change the size of the buf
1966 * variable in tcp_default_ctloutput().
1969 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1970 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1975 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1977 struct tls_enable_v0 tls_v0;
1980 if (sopt->sopt_valsize == sizeof(tls_v0)) {
1981 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1985 memset(tls, 0, sizeof(*tls));
1986 tls->cipher_key = tls_v0.cipher_key;
1987 tls->iv = tls_v0.iv;
1988 tls->auth_key = tls_v0.auth_key;
1989 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1990 tls->cipher_key_len = tls_v0.cipher_key_len;
1991 tls->iv_len = tls_v0.iv_len;
1992 tls->auth_algorithm = tls_v0.auth_algorithm;
1993 tls->auth_key_len = tls_v0.auth_key_len;
1994 tls->flags = tls_v0.flags;
1995 tls->tls_vmajor = tls_v0.tls_vmajor;
1996 tls->tls_vminor = tls_v0.tls_vminor;
2000 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
2004 extern struct cc_algo newreno_cc_algo;
2007 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt)
2009 struct cc_algo *algo;
2012 struct cc_var cc_mem;
2013 char buf[TCP_CA_NAME_MAX];
2018 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2021 buf[sopt->sopt_valsize] = '\0';
2023 STAILQ_FOREACH(algo, &cc_list, entries) {
2024 if (strncmp(buf, algo->name,
2025 TCP_CA_NAME_MAX) == 0) {
2026 if (algo->flags & CC_MODULE_BEING_REMOVED) {
2027 /* We can't "see" modules being unloaded */
2038 * With a reference the algorithm cannot be removed
2039 * so we hold a reference through the change process.
2043 if (algo->cb_init != NULL) {
2044 /* We can now pre-get the memory for the CC */
2045 mem_sz = (*algo->cc_data_sz)();
2049 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
2056 * Make sure its all clean and zero and also get
2059 memset(&cc_mem, 0, sizeof(cc_mem));
2061 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2064 free(ptr, M_CC_MEM);
2065 /* Release our temp reference */
2069 return (ECONNRESET);
2071 tp = intotcpcb(inp);
2073 memset(ptr, 0, mem_sz);
2074 cc_mem.ccvc.tcp = tp;
2076 * We once again hold a write lock over the tcb so it's
2077 * safe to do these things without ordering concerns.
2078 * Note here we init into stack memory.
2080 if (algo->cb_init != NULL)
2081 error = algo->cb_init(&cc_mem, ptr);
2085 * The CC algorithms, when given their memory
2086 * should not fail we could in theory have a
2091 * Touchdown, lets go ahead and move the
2092 * connection to the new CC module by
2093 * copying in the cc_mem after we call
2094 * the old ones cleanup (if any).
2096 if (CC_ALGO(tp)->cb_destroy != NULL)
2097 CC_ALGO(tp)->cb_destroy(tp->ccv);
2098 /* Detach the old CC from the tcpcb */
2100 /* Copy in our temp memory that was inited */
2101 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2102 /* Now attach the new, which takes a reference */
2103 cc_attach(tp, algo);
2104 /* Ok now are we where we have gotten past any conn_init? */
2105 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
2106 /* Yep run the connection init for the new CC */
2107 CC_ALGO(tp)->conn_init(tp->ccv);
2110 free(ptr, M_CC_MEM);
2112 /* Now lets release our temp reference */
2120 tcp_default_ctloutput(struct inpcb *inp, struct sockopt *sopt)
2122 struct tcpcb *tp = intotcpcb(inp);
2123 int error, opt, optval;
2127 struct tls_enable tls;
2128 struct socket *so = inp->inp_socket;
2130 char *pbuf, buf[TCP_LOG_ID_LEN];
2132 struct statsblob *sbp;
2136 INP_WLOCK_ASSERT(inp);
2137 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
2138 ("inp_flags == %x", inp->inp_flags));
2139 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL"));
2141 switch (sopt->sopt_level) {
2144 MPASS(inp->inp_vflag & INP_IPV6PROTO);
2145 switch (sopt->sopt_name) {
2146 case IPV6_USE_MIN_MTU:
2147 tcp6_use_min_mtu(tp);
2161 * For TCP_CCALGOOPT forward the control to CC module, for both
2162 * SOPT_SET and SOPT_GET.
2164 switch (sopt->sopt_name) {
2167 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2169 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2170 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2171 sopt->sopt_valsize);
2176 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2177 if (CC_ALGO(tp)->ctl_output != NULL)
2178 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
2182 if (error == 0 && sopt->sopt_dir == SOPT_GET)
2183 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2188 switch (sopt->sopt_dir) {
2190 switch (sopt->sopt_name) {
2191 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2193 if (!TCPMD5_ENABLED()) {
2195 return (ENOPROTOOPT);
2197 error = TCPMD5_PCBCTL(inp, sopt);
2200 goto unlock_and_done;
2207 error = sooptcopyin(sopt, &optval, sizeof optval,
2212 INP_WLOCK_RECHECK(inp);
2213 switch (sopt->sopt_name) {
2224 opt = 0; /* dead code to fool gcc */
2231 tp->t_flags &= ~opt;
2234 if (tp->t_flags & TF_TOE) {
2235 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2244 error = sooptcopyin(sopt, &optval, sizeof optval,
2249 INP_WLOCK_RECHECK(inp);
2251 tp->t_flags |= TF_NOPUSH;
2252 else if (tp->t_flags & TF_NOPUSH) {
2253 tp->t_flags &= ~TF_NOPUSH;
2254 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2255 struct epoch_tracker et;
2257 NET_EPOCH_ENTER(et);
2258 error = tcp_output_nodrop(tp);
2262 goto unlock_and_done;
2264 case TCP_REMOTE_UDP_ENCAPS_PORT:
2266 error = sooptcopyin(sopt, &optval, sizeof optval,
2270 if ((optval < TCP_TUNNELING_PORT_MIN) ||
2271 (optval > TCP_TUNNELING_PORT_MAX)) {
2272 /* Its got to be in range */
2275 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2276 /* You have to have enabled a UDP tunneling port first */
2279 INP_WLOCK_RECHECK(inp);
2280 if (tp->t_state != TCPS_CLOSED) {
2281 /* You can't change after you are connected */
2284 /* Ok we are all good set the port */
2285 tp->t_port = htons(optval);
2287 goto unlock_and_done;
2291 error = sooptcopyin(sopt, &optval, sizeof optval,
2296 INP_WLOCK_RECHECK(inp);
2297 if (optval > 0 && optval <= tp->t_maxseg &&
2298 optval + 40 >= V_tcp_minmss)
2299 tp->t_maxseg = optval;
2302 goto unlock_and_done;
2312 error = sooptcopyin(sopt, &optval, sizeof optval,
2318 sbp = stats_blob_alloc(
2319 V_tcp_perconn_stats_dflt_tpl, 0);
2323 INP_WLOCK_RECHECK(inp);
2324 if ((tp->t_stats != NULL && sbp == NULL) ||
2325 (tp->t_stats == NULL && sbp != NULL)) {
2326 struct statsblob *t = tp->t_stats;
2332 stats_blob_destroy(sbp);
2334 return (EOPNOTSUPP);
2338 case TCP_CONGESTION:
2339 error = tcp_set_cc_mod(inp, sopt);
2342 case TCP_REUSPORT_LB_NUMA:
2344 error = sooptcopyin(sopt, &optval, sizeof(optval),
2346 INP_WLOCK_RECHECK(inp);
2348 error = in_pcblbgroup_numa(inp, optval);
2353 case TCP_TXTLS_ENABLE:
2355 error = copyin_tls_enable(sopt, &tls);
2358 error = ktls_enable_tx(so, &tls);
2360 case TCP_TXTLS_MODE:
2362 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2366 INP_WLOCK_RECHECK(inp);
2367 error = ktls_set_tx_mode(so, ui);
2370 case TCP_RXTLS_ENABLE:
2372 error = sooptcopyin(sopt, &tls, sizeof(tls),
2376 error = ktls_enable_rx(so, &tls);
2384 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2388 if (ui > (UINT_MAX / hz)) {
2394 INP_WLOCK_RECHECK(inp);
2395 switch (sopt->sopt_name) {
2397 tp->t_keepidle = ui;
2399 * XXX: better check current remaining
2400 * timeout and "merge" it with new value.
2402 if ((tp->t_state > TCPS_LISTEN) &&
2403 (tp->t_state <= TCPS_CLOSING))
2404 tcp_timer_activate(tp, TT_KEEP,
2408 tp->t_keepintvl = ui;
2409 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2410 (TP_MAXIDLE(tp) > 0))
2411 tcp_timer_activate(tp, TT_2MSL,
2415 tp->t_keepinit = ui;
2416 if (tp->t_state == TCPS_SYN_RECEIVED ||
2417 tp->t_state == TCPS_SYN_SENT)
2418 tcp_timer_activate(tp, TT_KEEP,
2422 goto unlock_and_done;
2426 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2430 INP_WLOCK_RECHECK(inp);
2432 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2433 (TP_MAXIDLE(tp) > 0))
2434 tcp_timer_activate(tp, TT_2MSL,
2436 goto unlock_and_done;
2442 error = sooptcopyin(sopt, &optval, sizeof optval,
2447 INP_WLOCK_RECHECK(inp);
2449 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2450 &(tp->t_outpkts) : &(tp->t_inpkts),
2454 goto unlock_and_done;
2457 case TCP_FASTOPEN: {
2458 struct tcp_fastopen tfo_optval;
2461 if (!V_tcp_fastopen_client_enable &&
2462 !V_tcp_fastopen_server_enable)
2465 error = sooptcopyin(sopt, &tfo_optval,
2466 sizeof(tfo_optval), sizeof(int));
2470 INP_WLOCK_RECHECK(inp);
2471 if ((tp->t_state != TCPS_CLOSED) &&
2472 (tp->t_state != TCPS_LISTEN)) {
2474 goto unlock_and_done;
2476 if (tfo_optval.enable) {
2477 if (tp->t_state == TCPS_LISTEN) {
2478 if (!V_tcp_fastopen_server_enable) {
2480 goto unlock_and_done;
2483 if (tp->t_tfo_pending == NULL)
2485 tcp_fastopen_alloc_counter();
2488 * If a pre-shared key was provided,
2489 * stash it in the client cookie
2490 * field of the tcpcb for use during
2493 if (sopt->sopt_valsize ==
2494 sizeof(tfo_optval)) {
2495 memcpy(tp->t_tfo_cookie.client,
2497 TCP_FASTOPEN_PSK_LEN);
2498 tp->t_tfo_client_cookie_len =
2499 TCP_FASTOPEN_PSK_LEN;
2502 tp->t_flags |= TF_FASTOPEN;
2504 tp->t_flags &= ~TF_FASTOPEN;
2505 goto unlock_and_done;
2511 error = sooptcopyin(sopt, &optval, sizeof optval,
2516 INP_WLOCK_RECHECK(inp);
2517 error = tcp_log_state_change(tp, optval);
2518 goto unlock_and_done;
2527 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2530 buf[sopt->sopt_valsize] = '\0';
2531 INP_WLOCK_RECHECK(inp);
2532 error = tcp_log_set_id(tp, buf);
2533 /* tcp_log_set_id() unlocks the INP. */
2540 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2543 buf[sopt->sopt_valsize] = '\0';
2544 INP_WLOCK_RECHECK(inp);
2545 if (sopt->sopt_name == TCP_LOGDUMP) {
2546 error = tcp_log_dump_tp_logbuf(tp, buf,
2550 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2552 * tcp_log_dump_tp_bucket_logbufs() drops the
2561 error = ENOPROTOOPT;
2567 tp = intotcpcb(inp);
2568 switch (sopt->sopt_name) {
2569 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2571 if (!TCPMD5_ENABLED()) {
2573 return (ENOPROTOOPT);
2575 error = TCPMD5_PCBCTL(inp, sopt);
2580 optval = tp->t_flags & TF_NODELAY;
2582 error = sooptcopyout(sopt, &optval, sizeof optval);
2585 optval = tp->t_maxseg;
2587 error = sooptcopyout(sopt, &optval, sizeof optval);
2589 case TCP_REMOTE_UDP_ENCAPS_PORT:
2590 optval = ntohs(tp->t_port);
2592 error = sooptcopyout(sopt, &optval, sizeof optval);
2595 optval = tp->t_flags & TF_NOOPT;
2597 error = sooptcopyout(sopt, &optval, sizeof optval);
2600 optval = tp->t_flags & TF_NOPUSH;
2602 error = sooptcopyout(sopt, &optval, sizeof optval);
2605 tcp_fill_info(tp, &ti);
2607 error = sooptcopyout(sopt, &ti, sizeof ti);
2613 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2616 socklen_t outsbsz = sopt->sopt_valsize;
2617 if (tp->t_stats == NULL)
2619 else if (outsbsz >= tp->t_stats->cursz)
2620 outsbsz = tp->t_stats->cursz;
2621 else if (outsbsz >= sizeof(struct statsblob))
2622 outsbsz = sizeof(struct statsblob);
2629 sbp = sopt->sopt_val;
2630 nheld = atop(round_page(((vm_offset_t)sbp) +
2631 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2632 vm_page_t ma[nheld];
2633 if (vm_fault_quick_hold_pages(
2634 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2635 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2641 if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2642 SIZEOF_MEMBER(struct statsblob, flags))))
2645 INP_WLOCK_RECHECK(inp);
2646 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2647 sbflags | SB_CLONE_USRDSTNOFAULT);
2649 sopt->sopt_valsize = outsbsz;
2651 vm_page_unhold_pages(ma, nheld);
2658 case TCP_CONGESTION:
2659 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2661 error = sooptcopyout(sopt, buf, len + 1);
2667 switch (sopt->sopt_name) {
2669 ui = TP_KEEPIDLE(tp) / hz;
2672 ui = TP_KEEPINTVL(tp) / hz;
2675 ui = TP_KEEPINIT(tp) / hz;
2678 ui = TP_KEEPCNT(tp);
2682 error = sooptcopyout(sopt, &ui, sizeof(ui));
2687 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2688 &(tp->t_outpkts) : &(tp->t_inpkts));
2690 error = sooptcopyout(sopt, &optval, sizeof optval);
2694 optval = tp->t_flags & TF_FASTOPEN;
2696 error = sooptcopyout(sopt, &optval, sizeof optval);
2700 optval = tp->t_logstate;
2702 error = sooptcopyout(sopt, &optval, sizeof(optval));
2705 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2706 error = tcp_log_getlogbuf(sopt, tp);
2709 len = tcp_log_get_id(tp, buf);
2711 error = sooptcopyout(sopt, buf, len + 1);
2720 case TCP_TXTLS_MODE:
2721 error = ktls_get_tx_mode(so, &optval);
2724 error = sooptcopyout(sopt, &optval,
2727 case TCP_RXTLS_MODE:
2728 error = ktls_get_rx_mode(so, &optval);
2731 error = sooptcopyout(sopt, &optval,
2736 optval = tp->t_flags & TF_LRD;
2738 error = sooptcopyout(sopt, &optval, sizeof optval);
2742 error = ENOPROTOOPT;
2749 #undef INP_WLOCK_RECHECK
2750 #undef INP_WLOCK_RECHECK_CLEANUP
2753 * Initiate (or continue) disconnect.
2754 * If embryonic state, just send reset (once).
2755 * If in ``let data drain'' option and linger null, just drop.
2756 * Otherwise (hard), mark socket disconnecting and drop
2757 * current input data; switch states based on user close, and
2758 * send segment to peer (with FIN).
2761 tcp_disconnect(struct tcpcb *tp)
2763 struct inpcb *inp = tp->t_inpcb;
2764 struct socket *so = inp->inp_socket;
2767 INP_WLOCK_ASSERT(inp);
2770 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2771 * socket is still open.
2773 if (tp->t_state < TCPS_ESTABLISHED &&
2774 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2777 ("tcp_disconnect: tcp_close() returned NULL"));
2778 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2779 tp = tcp_drop(tp, 0);
2781 ("tcp_disconnect: tcp_drop() returned NULL"));
2783 soisdisconnecting(so);
2784 sbflush(&so->so_rcv);
2786 if (!(inp->inp_flags & INP_DROPPED))
2787 /* Ignore stack's drop request, we already at it. */
2788 (void)tcp_output_nodrop(tp);
2793 * User issued close, and wish to trail through shutdown states:
2794 * if never received SYN, just forget it. If got a SYN from peer,
2795 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2796 * If already got a FIN from peer, then almost done; go to LAST_ACK
2797 * state. In all other cases, have already sent FIN to peer (e.g.
2798 * after PRU_SHUTDOWN), and just have to play tedious game waiting
2799 * for peer to send FIN or not respond to keep-alives, etc.
2800 * We can let the user exit from the close as soon as the FIN is acked.
2803 tcp_usrclosed(struct tcpcb *tp)
2807 INP_WLOCK_ASSERT(tp->t_inpcb);
2809 switch (tp->t_state) {
2812 tcp_offload_listen_stop(tp);
2814 tcp_state_change(tp, TCPS_CLOSED);
2819 * tcp_close() should never return NULL here as the socket is
2823 ("tcp_usrclosed: tcp_close() returned NULL"));
2827 case TCPS_SYN_RECEIVED:
2828 tp->t_flags |= TF_NEEDFIN;
2831 case TCPS_ESTABLISHED:
2832 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2835 case TCPS_CLOSE_WAIT:
2836 tcp_state_change(tp, TCPS_LAST_ACK);
2839 if (tp->t_state >= TCPS_FIN_WAIT_2) {
2840 soisdisconnected(tp->t_inpcb->inp_socket);
2841 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2842 if (tp->t_state == TCPS_FIN_WAIT_2) {
2845 timeout = (tcp_fast_finwait2_recycle) ?
2846 tcp_finwait2_timeout : TP_MAXIDLE(tp);
2847 tcp_timer_activate(tp, TT_2MSL, timeout);
2854 db_print_indent(int indent)
2858 for (i = 0; i < indent; i++)
2863 db_print_tstate(int t_state)
2868 db_printf("TCPS_CLOSED");
2872 db_printf("TCPS_LISTEN");
2876 db_printf("TCPS_SYN_SENT");
2879 case TCPS_SYN_RECEIVED:
2880 db_printf("TCPS_SYN_RECEIVED");
2883 case TCPS_ESTABLISHED:
2884 db_printf("TCPS_ESTABLISHED");
2887 case TCPS_CLOSE_WAIT:
2888 db_printf("TCPS_CLOSE_WAIT");
2891 case TCPS_FIN_WAIT_1:
2892 db_printf("TCPS_FIN_WAIT_1");
2896 db_printf("TCPS_CLOSING");
2900 db_printf("TCPS_LAST_ACK");
2903 case TCPS_FIN_WAIT_2:
2904 db_printf("TCPS_FIN_WAIT_2");
2907 case TCPS_TIME_WAIT:
2908 db_printf("TCPS_TIME_WAIT");
2912 db_printf("unknown");
2918 db_print_tflags(u_int t_flags)
2923 if (t_flags & TF_ACKNOW) {
2924 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2927 if (t_flags & TF_DELACK) {
2928 db_printf("%sTF_DELACK", comma ? ", " : "");
2931 if (t_flags & TF_NODELAY) {
2932 db_printf("%sTF_NODELAY", comma ? ", " : "");
2935 if (t_flags & TF_NOOPT) {
2936 db_printf("%sTF_NOOPT", comma ? ", " : "");
2939 if (t_flags & TF_SENTFIN) {
2940 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2943 if (t_flags & TF_REQ_SCALE) {
2944 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2947 if (t_flags & TF_RCVD_SCALE) {
2948 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2951 if (t_flags & TF_REQ_TSTMP) {
2952 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2955 if (t_flags & TF_RCVD_TSTMP) {
2956 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2959 if (t_flags & TF_SACK_PERMIT) {
2960 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2963 if (t_flags & TF_NEEDSYN) {
2964 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2967 if (t_flags & TF_NEEDFIN) {
2968 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2971 if (t_flags & TF_NOPUSH) {
2972 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2975 if (t_flags & TF_PREVVALID) {
2976 db_printf("%sTF_PREVVALID", comma ? ", " : "");
2979 if (t_flags & TF_MORETOCOME) {
2980 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2983 if (t_flags & TF_LQ_OVERFLOW) {
2984 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2987 if (t_flags & TF_LASTIDLE) {
2988 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2991 if (t_flags & TF_RXWIN0SENT) {
2992 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2995 if (t_flags & TF_FASTRECOVERY) {
2996 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2999 if (t_flags & TF_CONGRECOVERY) {
3000 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
3003 if (t_flags & TF_WASFRECOVERY) {
3004 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
3007 if (t_flags & TF_WASCRECOVERY) {
3008 db_printf("%sTF_WASCRECOVERY", comma ? ", " : "");
3011 if (t_flags & TF_SIGNATURE) {
3012 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
3015 if (t_flags & TF_FORCEDATA) {
3016 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
3019 if (t_flags & TF_TSO) {
3020 db_printf("%sTF_TSO", comma ? ", " : "");
3023 if (t_flags & TF_FASTOPEN) {
3024 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
3030 db_print_tflags2(u_int t_flags2)
3035 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) {
3036 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : "");
3039 if (t_flags2 & TF2_PLPMTU_PMTUD) {
3040 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : "");
3043 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) {
3044 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : "");
3047 if (t_flags2 & TF2_LOG_AUTO) {
3048 db_printf("%sTF2_LOG_AUTO", comma ? ", " : "");
3051 if (t_flags2 & TF2_DROP_AF_DATA) {
3052 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : "");
3055 if (t_flags2 & TF2_ECN_PERMIT) {
3056 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
3059 if (t_flags2 & TF2_ECN_SND_CWR) {
3060 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : "");
3063 if (t_flags2 & TF2_ECN_SND_ECE) {
3064 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : "");
3067 if (t_flags2 & TF2_ACE_PERMIT) {
3068 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : "");
3071 if (t_flags2 & TF2_FBYTES_COMPLETE) {
3072 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : "");
3078 db_print_toobflags(char t_oobflags)
3083 if (t_oobflags & TCPOOB_HAVEDATA) {
3084 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
3087 if (t_oobflags & TCPOOB_HADDATA) {
3088 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
3094 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
3097 db_print_indent(indent);
3098 db_printf("%s at %p\n", name, tp);
3102 db_print_indent(indent);
3103 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
3104 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
3106 db_print_indent(indent);
3107 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
3108 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
3110 db_print_indent(indent);
3111 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
3112 &tp->t_timers->tt_delack, tp->t_inpcb);
3114 db_print_indent(indent);
3115 db_printf("t_state: %d (", tp->t_state);
3116 db_print_tstate(tp->t_state);
3119 db_print_indent(indent);
3120 db_printf("t_flags: 0x%x (", tp->t_flags);
3121 db_print_tflags(tp->t_flags);
3124 db_print_indent(indent);
3125 db_printf("t_flags2: 0x%x (", tp->t_flags2);
3126 db_print_tflags2(tp->t_flags2);
3129 db_print_indent(indent);
3130 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
3131 tp->snd_una, tp->snd_max, tp->snd_nxt);
3133 db_print_indent(indent);
3134 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
3135 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
3137 db_print_indent(indent);
3138 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
3139 tp->iss, tp->irs, tp->rcv_nxt);
3141 db_print_indent(indent);
3142 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
3143 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
3145 db_print_indent(indent);
3146 db_printf("snd_wnd: %u snd_cwnd: %u\n",
3147 tp->snd_wnd, tp->snd_cwnd);
3149 db_print_indent(indent);
3150 db_printf("snd_ssthresh: %u snd_recover: "
3151 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
3153 db_print_indent(indent);
3154 db_printf("t_rcvtime: %u t_startime: %u\n",
3155 tp->t_rcvtime, tp->t_starttime);
3157 db_print_indent(indent);
3158 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
3159 tp->t_rtttime, tp->t_rtseq);
3161 db_print_indent(indent);
3162 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
3163 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
3165 db_print_indent(indent);
3166 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
3167 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
3170 db_print_indent(indent);
3171 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n",
3172 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
3174 db_print_indent(indent);
3175 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
3176 db_print_toobflags(tp->t_oobflags);
3177 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
3179 db_print_indent(indent);
3180 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
3181 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
3183 db_print_indent(indent);
3184 db_printf("ts_recent: %u ts_recent_age: %u\n",
3185 tp->ts_recent, tp->ts_recent_age);
3187 db_print_indent(indent);
3188 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
3189 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
3191 db_print_indent(indent);
3192 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
3193 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
3194 tp->snd_recover_prev, tp->t_badrxtwin);
3196 db_print_indent(indent);
3197 db_printf("snd_numholes: %d snd_holes first: %p\n",
3198 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
3200 db_print_indent(indent);
3201 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n",
3202 tp->snd_fack, tp->rcv_numsacks);
3204 /* Skip sackblks, sackhint. */
3206 db_print_indent(indent);
3207 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
3208 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
3211 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
3216 db_printf("usage: show tcpcb <addr>\n");
3219 tp = (struct tcpcb *)addr;
3221 db_print_tcpcb(tp, "tcpcb", 0);