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_var.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_pcb.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/scope6_var.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcp_log_buf.h>
98 #include <netinet/tcpip.h>
99 #include <netinet/cc/cc.h>
100 #include <netinet/tcp_fastopen.h>
101 #include <netinet/tcp_hpts.h>
103 #include <netinet/tcp_pcap.h>
106 #include <netinet/tcp_debug.h>
109 #include <netinet/tcp_offload.h>
111 #include <netipsec/ipsec_support.h>
114 #include <vm/vm_param.h>
116 #include <vm/vm_extern.h>
117 #include <vm/vm_map.h>
118 #include <vm/vm_page.h>
121 * TCP protocol interface to socket abstraction.
124 static int tcp_connect(struct tcpcb *, struct sockaddr *,
128 static int tcp6_connect(struct tcpcb *, struct sockaddr *,
131 static void tcp_disconnect(struct tcpcb *);
132 static void tcp_usrclosed(struct tcpcb *);
133 static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
135 static int tcp_pru_options_support(struct tcpcb *tp, int flags);
138 #define TCPDEBUG0 int ostate = 0
139 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
140 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
141 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
145 #define TCPDEBUG2(req)
149 * tcp_require_unique port requires a globally-unique source port for each
150 * outgoing connection. The default is to require the 4-tuple to be unique.
152 VNET_DEFINE(int, tcp_require_unique_port) = 0;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port,
154 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0,
155 "Require globally-unique ephemeral port for outgoing connections");
156 #define V_tcp_require_unique_port VNET(tcp_require_unique_port)
159 * TCP attaches to socket via pru_attach(), reserving space,
160 * and an internet control block.
163 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
166 struct tcpcb *tp = NULL;
171 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
174 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
175 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
180 so->so_rcv.sb_flags |= SB_AUTOSIZE;
181 so->so_snd.sb_flags |= SB_AUTOSIZE;
182 error = in_pcballoc(so, &V_tcbinfo);
187 if (inp->inp_vflag & INP_IPV6PROTO) {
188 inp->inp_vflag |= INP_IPV6;
189 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
190 inp->inp_vflag |= INP_IPV4;
191 inp->in6p_hops = -1; /* use kernel default */
195 inp->inp_vflag |= INP_IPV4;
196 tp = tcp_newtcpcb(inp);
203 tp->t_state = TCPS_CLOSED;
205 TCPSTATES_INC(TCPS_CLOSED);
207 TCPDEBUG2(PRU_ATTACH);
208 TCP_PROBE2(debug__user, tp, PRU_ATTACH);
213 * tcp_usr_detach is called when the socket layer loses its final reference
214 * to the socket, be it a file descriptor reference, a reference from TCP,
215 * etc. At this point, there is only one case in which we will keep around
216 * inpcb state: time wait.
219 tcp_usr_detach(struct socket *so)
225 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
227 KASSERT(so->so_pcb == inp && inp->inp_socket == so,
228 ("%s: socket %p inp %p mismatch", __func__, so, inp));
232 if (inp->inp_flags & INP_TIMEWAIT) {
234 * There are two cases to handle: one in which the time wait
235 * state is being discarded (INP_DROPPED), and one in which
236 * this connection will remain in timewait. In the former,
237 * it is time to discard all state (except tcptw, which has
238 * already been discarded by the timewait close code, which
239 * should be further up the call stack somewhere). In the
240 * latter case, we detach from the socket, but leave the pcb
241 * present until timewait ends.
243 * XXXRW: Would it be cleaner to free the tcptw here?
245 * Astute question indeed, from twtcp perspective there are
246 * four cases to consider:
248 * #1 tcp_usr_detach is called at tcptw creation time by
249 * tcp_twstart, then do not discard the newly created tcptw
250 * and leave inpcb present until timewait ends
251 * #2 tcp_usr_detach is called at tcptw creation time by
252 * tcp_twstart, but connection is local and tw will be
253 * discarded immediately
254 * #3 tcp_usr_detach is called at timewait end (or reuse) by
255 * tcp_twclose, then the tcptw has already been discarded
256 * (or reused) and inpcb is freed here
257 * #4 tcp_usr_detach is called() after timewait ends (or reuse)
258 * (e.g. by soclose), then tcptw has already been discarded
259 * (or reused) and inpcb is freed here
261 * In all three cases the tcptw should not be freed here.
263 if (inp->inp_flags & INP_DROPPED) {
264 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
265 "INP_DROPPED && tp != NULL"));
274 * If the connection is not in timewait, we consider two
275 * two conditions: one in which no further processing is
276 * necessary (dropped || embryonic), and one in which TCP is
277 * not yet done, but no longer requires the socket, so the
278 * pcb will persist for the time being.
280 * XXXRW: Does the second case still occur?
282 if (inp->inp_flags & INP_DROPPED ||
283 tp->t_state < TCPS_SYN_SENT) {
296 * Give the socket an address.
299 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
303 struct tcpcb *tp = NULL;
304 struct sockaddr_in *sinp;
306 sinp = (struct sockaddr_in *)nam;
307 if (nam->sa_family != AF_INET) {
309 * Preserve compatibility with old programs.
311 if (nam->sa_family != AF_UNSPEC ||
312 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
313 sinp->sin_addr.s_addr != INADDR_ANY)
314 return (EAFNOSUPPORT);
315 nam->sa_family = AF_INET;
317 if (nam->sa_len != sizeof(*sinp))
321 * Must check for multicast addresses and disallow binding
324 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
325 return (EAFNOSUPPORT);
329 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
331 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
337 INP_HASH_WLOCK(&V_tcbinfo);
338 error = in_pcbbind(inp, nam, td->td_ucred);
339 INP_HASH_WUNLOCK(&V_tcbinfo);
342 TCP_PROBE2(debug__user, tp, PRU_BIND);
351 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
355 struct tcpcb *tp = NULL;
356 struct sockaddr_in6 *sin6;
359 sin6 = (struct sockaddr_in6 *)nam;
360 if (nam->sa_family != AF_INET6)
361 return (EAFNOSUPPORT);
362 if (nam->sa_len != sizeof(*sin6))
366 * Must check for multicast addresses and disallow binding
369 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
370 return (EAFNOSUPPORT);
374 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
376 vflagsav = inp->inp_vflag;
377 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
383 INP_HASH_WLOCK(&V_tcbinfo);
384 inp->inp_vflag &= ~INP_IPV4;
385 inp->inp_vflag |= INP_IPV6;
387 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
388 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
389 inp->inp_vflag |= INP_IPV4;
390 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
391 struct sockaddr_in sin;
393 in6_sin6_2_sin(&sin, sin6);
394 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
395 error = EAFNOSUPPORT;
396 INP_HASH_WUNLOCK(&V_tcbinfo);
399 inp->inp_vflag |= INP_IPV4;
400 inp->inp_vflag &= ~INP_IPV6;
401 error = in_pcbbind(inp, (struct sockaddr *)&sin,
403 INP_HASH_WUNLOCK(&V_tcbinfo);
408 error = in6_pcbbind(inp, nam, td->td_ucred);
409 INP_HASH_WUNLOCK(&V_tcbinfo);
412 inp->inp_vflag = vflagsav;
414 TCP_PROBE2(debug__user, tp, PRU_BIND);
422 * Prepare to accept connections.
425 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
429 struct tcpcb *tp = NULL;
433 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
435 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
442 error = solisten_proto_check(so);
447 if (inp->inp_lport == 0) {
448 INP_HASH_WLOCK(&V_tcbinfo);
449 error = in_pcbbind(inp, NULL, td->td_ucred);
450 INP_HASH_WUNLOCK(&V_tcbinfo);
453 tcp_state_change(tp, TCPS_LISTEN);
454 solisten_proto(so, backlog);
456 if ((so->so_options & SO_NO_OFFLOAD) == 0)
457 tcp_offload_listen_start(tp);
460 solisten_proto_abort(so);
464 if (IS_FASTOPEN(tp->t_flags))
465 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
468 TCPDEBUG2(PRU_LISTEN);
469 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
477 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
481 struct tcpcb *tp = NULL;
486 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
488 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
492 vflagsav = inp->inp_vflag;
496 error = solisten_proto_check(so);
501 INP_HASH_WLOCK(&V_tcbinfo);
502 if (inp->inp_lport == 0) {
503 inp->inp_vflag &= ~INP_IPV4;
504 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
505 inp->inp_vflag |= INP_IPV4;
506 error = in6_pcbbind(inp, NULL, td->td_ucred);
508 INP_HASH_WUNLOCK(&V_tcbinfo);
510 tcp_state_change(tp, TCPS_LISTEN);
511 solisten_proto(so, backlog);
513 if ((so->so_options & SO_NO_OFFLOAD) == 0)
514 tcp_offload_listen_start(tp);
517 solisten_proto_abort(so);
521 if (IS_FASTOPEN(tp->t_flags))
522 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
525 inp->inp_vflag = vflagsav;
528 TCPDEBUG2(PRU_LISTEN);
529 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
537 * Initiate connection to peer.
538 * Create a template for use in transmissions on this connection.
539 * Enter SYN_SENT state, and mark socket as connecting.
540 * Start keep-alive timer, and seed output sequence space.
541 * Send initial segment on connection.
544 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
546 struct epoch_tracker et;
549 struct tcpcb *tp = NULL;
550 struct sockaddr_in *sinp;
552 sinp = (struct sockaddr_in *)nam;
553 if (nam->sa_family != AF_INET)
554 return (EAFNOSUPPORT);
555 if (nam->sa_len != sizeof (*sinp))
559 * Must disallow TCP ``connections'' to multicast addresses.
561 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
562 return (EAFNOSUPPORT);
563 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST)
565 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
570 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
572 if (inp->inp_flags & INP_TIMEWAIT) {
576 if (inp->inp_flags & INP_DROPPED) {
577 error = ECONNREFUSED;
580 if (SOLISTENING(so)) {
587 if ((error = tcp_connect(tp, nam, td)) != 0)
590 if (registered_toedevs > 0 &&
591 (so->so_options & SO_NO_OFFLOAD) == 0 &&
592 (error = tcp_offload_connect(so, nam)) == 0)
595 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
596 error = tcp_output(tp);
597 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()",
598 tp->t_fb->tfb_tcp_block_name, tp));
602 TCPDEBUG2(PRU_CONNECT);
603 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
611 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
613 struct epoch_tracker et;
616 struct tcpcb *tp = NULL;
617 struct sockaddr_in6 *sin6;
623 sin6 = (struct sockaddr_in6 *)nam;
624 if (nam->sa_family != AF_INET6)
625 return (EAFNOSUPPORT);
626 if (nam->sa_len != sizeof (*sin6))
630 * Must disallow TCP ``connections'' to multicast addresses.
632 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
633 return (EAFNOSUPPORT);
636 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
638 vflagsav = inp->inp_vflag;
639 incflagsav = inp->inp_inc.inc_flags;
640 if (inp->inp_flags & INP_TIMEWAIT) {
644 if (inp->inp_flags & INP_DROPPED) {
645 error = ECONNREFUSED;
648 if (SOLISTENING(so)) {
656 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
657 * therefore probably require the hash lock, which isn't held here.
658 * Is this a significant problem?
660 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
661 struct sockaddr_in sin;
663 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
667 if ((inp->inp_vflag & INP_IPV4) == 0) {
668 error = EAFNOSUPPORT;
672 in6_sin6_2_sin(&sin, sin6);
673 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
674 error = EAFNOSUPPORT;
677 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
681 if ((error = prison_remote_ip4(td->td_ucred,
682 &sin.sin_addr)) != 0)
684 inp->inp_vflag |= INP_IPV4;
685 inp->inp_vflag &= ~INP_IPV6;
687 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
690 if (registered_toedevs > 0 &&
691 (so->so_options & SO_NO_OFFLOAD) == 0 &&
692 (error = tcp_offload_connect(so, nam)) == 0)
695 error = tcp_output(tp);
698 if ((inp->inp_vflag & INP_IPV6) == 0) {
699 error = EAFNOSUPPORT;
704 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
706 inp->inp_vflag &= ~INP_IPV4;
707 inp->inp_vflag |= INP_IPV6;
708 inp->inp_inc.inc_flags |= INC_ISIPV6;
709 if ((error = tcp6_connect(tp, nam, td)) != 0)
712 if (registered_toedevs > 0 &&
713 (so->so_options & SO_NO_OFFLOAD) == 0 &&
714 (error = tcp_offload_connect(so, nam)) == 0)
717 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
719 error = tcp_output(tp);
725 KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()",
726 tp->t_fb->tfb_tcp_block_name, tp));
728 * If the implicit bind in the connect call fails, restore
729 * the flags we modified.
731 if (error != 0 && inp->inp_lport == 0) {
732 inp->inp_vflag = vflagsav;
733 inp->inp_inc.inc_flags = incflagsav;
736 TCPDEBUG2(PRU_CONNECT);
737 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
744 * Initiate disconnect from peer.
745 * If connection never passed embryonic stage, just drop;
746 * else if don't need to let data drain, then can just drop anyways,
747 * else have to begin TCP shutdown process: mark socket disconnecting,
748 * drain unread data, state switch to reflect user close, and
749 * send segment (e.g. FIN) to peer. Socket will be really disconnected
750 * when peer sends FIN and acks ours.
752 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
755 tcp_usr_disconnect(struct socket *so)
758 struct tcpcb *tp = NULL;
759 struct epoch_tracker et;
765 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
767 if (inp->inp_flags & INP_TIMEWAIT)
769 if (inp->inp_flags & INP_DROPPED) {
777 TCPDEBUG2(PRU_DISCONNECT);
778 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
786 * Accept a connection. Essentially all the work is done at higher levels;
787 * just return the address of the peer, storing through addr.
790 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
793 struct inpcb *inp = NULL;
794 struct tcpcb *tp = NULL;
799 if (so->so_state & SS_ISDISCONNECTED)
800 return (ECONNABORTED);
803 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
805 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
806 error = ECONNABORTED;
813 * We inline in_getpeeraddr and COMMON_END here, so that we can
814 * copy the data of interest and defer the malloc until after we
817 port = inp->inp_fport;
818 addr = inp->inp_faddr;
821 TCPDEBUG2(PRU_ACCEPT);
822 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
825 *nam = in_sockaddr(port, &addr);
832 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
834 struct inpcb *inp = NULL;
836 struct tcpcb *tp = NULL;
838 struct in6_addr addr6;
839 struct epoch_tracker et;
844 if (so->so_state & SS_ISDISCONNECTED)
845 return (ECONNABORTED);
848 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
851 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
852 error = ECONNABORTED;
859 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
860 * copy the data of interest and defer the malloc until after we
863 if (inp->inp_vflag & INP_IPV4) {
865 port = inp->inp_fport;
866 addr = inp->inp_faddr;
868 port = inp->inp_fport;
869 addr6 = inp->in6p_faddr;
873 TCPDEBUG2(PRU_ACCEPT);
874 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
879 *nam = in6_v4mapsin6_sockaddr(port, &addr);
881 *nam = in6_sockaddr(port, &addr6);
888 * Mark the connection as being incapable of further output.
891 tcp_usr_shutdown(struct socket *so)
895 struct tcpcb *tp = NULL;
896 struct epoch_tracker et;
900 KASSERT(inp != NULL, ("inp == NULL"));
902 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
911 if (!(inp->inp_flags & INP_DROPPED))
912 error = tcp_output_nodrop(tp);
913 TCPDEBUG2(PRU_SHUTDOWN);
914 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
915 error = tcp_unlock_or_drop(tp, error);
922 * After a receive, possibly send window update to peer.
925 tcp_usr_rcvd(struct socket *so, int flags)
927 struct epoch_tracker et;
929 struct tcpcb *tp = NULL;
930 int outrv = 0, error = 0;
934 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
936 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
944 * For passively-created TFO connections, don't attempt a window
945 * update while still in SYN_RECEIVED as this may trigger an early
946 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with
947 * application response data, or failing that, when the DELACK timer
950 if (IS_FASTOPEN(tp->t_flags) &&
951 (tp->t_state == TCPS_SYN_RECEIVED))
954 if (tp->t_flags & TF_TOE)
955 tcp_offload_rcvd(tp);
958 outrv = tcp_output_nodrop(tp);
961 TCP_PROBE2(debug__user, tp, PRU_RCVD);
962 (void) tcp_unlock_or_drop(tp, outrv);
968 * Do a send by putting data in output queue and updating urgent
969 * marker if URG set. Possibly send more data. Unlike the other
970 * pru_*() routines, the mbuf chains are our responsibility. We
971 * must either enqueue them or free them. The other pru_* routines
972 * generally are caller-frees.
975 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
976 struct sockaddr *nam, struct mbuf *control, struct thread *td)
978 struct epoch_tracker et;
981 struct tcpcb *tp = NULL;
984 struct sockaddr_in sin;
986 struct sockaddr_in *sinp;
996 if (control != NULL) {
997 /* TCP doesn't do control messages (rights, creds, etc) */
998 if (control->m_len) {
1002 m_freem(control); /* empty control, just free it */
1005 inp = sotoinpcb(so);
1006 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
1008 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1009 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1012 return (ECONNRESET);
1015 vflagsav = inp->inp_vflag;
1016 incflagsav = inp->inp_inc.inc_flags;
1017 restoreflags = false;
1018 tp = intotcpcb(inp);
1020 NET_EPOCH_ENTER(et);
1021 if ((flags & PRUS_OOB) != 0 &&
1022 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1026 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1027 if (tp->t_state == TCPS_LISTEN) {
1031 switch (nam->sa_family) {
1034 sinp = (struct sockaddr_in *)nam;
1035 if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1039 if ((inp->inp_vflag & INP_IPV6) != 0) {
1040 error = EAFNOSUPPORT;
1043 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1044 error = EAFNOSUPPORT;
1047 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1051 if ((error = prison_remote_ip4(td->td_ucred,
1062 struct sockaddr_in6 *sin6;
1064 sin6 = (struct sockaddr_in6 *)nam;
1065 if (sin6->sin6_len != sizeof(*sin6)) {
1069 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1070 error = EAFNOSUPPORT;
1073 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1074 error = EAFNOSUPPORT;
1077 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1079 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1083 if ((inp->inp_vflag & INP_IPV4) == 0) {
1084 error = EAFNOSUPPORT;
1087 restoreflags = true;
1088 inp->inp_vflag &= ~INP_IPV6;
1090 in6_sin6_2_sin(sinp, sin6);
1092 ntohl(sinp->sin_addr.s_addr))) {
1093 error = EAFNOSUPPORT;
1096 if ((error = prison_remote_ip4(td->td_ucred,
1101 error = EAFNOSUPPORT;
1105 if ((inp->inp_vflag & INP_IPV6) == 0) {
1106 error = EAFNOSUPPORT;
1109 restoreflags = true;
1110 inp->inp_vflag &= ~INP_IPV4;
1111 inp->inp_inc.inc_flags |= INC_ISIPV6;
1112 if ((error = prison_remote_ip6(td->td_ucred,
1121 error = EAFNOSUPPORT;
1125 if (!(flags & PRUS_OOB)) {
1126 sbappendstream(&so->so_snd, m, flags);
1128 if (nam && tp->t_state < TCPS_SYN_SENT) {
1129 KASSERT(tp->t_state == TCPS_CLOSED,
1130 ("%s: tp %p is listening", __func__, tp));
1133 * Do implied connect if not yet connected,
1134 * initialize window to default value, and
1135 * initialize maxseg using peer's cached MSS.
1139 error = tcp6_connect(tp, nam, td);
1141 #if defined(INET6) && defined(INET)
1145 error = tcp_connect(tp,
1146 (struct sockaddr *)sinp, td);
1149 * The bind operation in tcp_connect succeeded. We
1150 * no longer want to restore the flags if later
1153 if (error == 0 || inp->inp_lport != 0)
1154 restoreflags = false;
1157 /* m is freed if PRUS_NOTREADY is unset. */
1158 sbflush(&so->so_snd);
1161 if (IS_FASTOPEN(tp->t_flags))
1162 tcp_fastopen_connect(tp);
1164 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1168 if (flags & PRUS_EOF) {
1170 * Close the send side of the connection after
1176 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1177 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1178 (tp->t_fbyte_out == 0) &&
1179 (so->so_snd.sb_ccc > 0)) {
1180 tp->t_fbyte_out = ticks;
1181 if (tp->t_fbyte_out == 0)
1182 tp->t_fbyte_out = 1;
1183 if (tp->t_fbyte_out && tp->t_fbyte_in)
1184 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1186 if (!(inp->inp_flags & INP_DROPPED) &&
1187 !(flags & PRUS_NOTREADY)) {
1188 if (flags & PRUS_MORETOCOME)
1189 tp->t_flags |= TF_MORETOCOME;
1190 error = tcp_output_nodrop(tp);
1191 if (flags & PRUS_MORETOCOME)
1192 tp->t_flags &= ~TF_MORETOCOME;
1196 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1198 SOCKBUF_LOCK(&so->so_snd);
1199 if (sbspace(&so->so_snd) < -512) {
1200 SOCKBUF_UNLOCK(&so->so_snd);
1205 * According to RFC961 (Assigned Protocols),
1206 * the urgent pointer points to the last octet
1207 * of urgent data. We continue, however,
1208 * to consider it to indicate the first octet
1209 * of data past the urgent section.
1210 * Otherwise, snd_up should be one lower.
1212 sbappendstream_locked(&so->so_snd, m, flags);
1213 SOCKBUF_UNLOCK(&so->so_snd);
1215 if (nam && tp->t_state < TCPS_SYN_SENT) {
1217 * Do implied connect if not yet connected,
1218 * initialize window to default value, and
1219 * initialize maxseg using peer's cached MSS.
1223 * Not going to contemplate SYN|URG
1225 if (IS_FASTOPEN(tp->t_flags))
1226 tp->t_flags &= ~TF_FASTOPEN;
1229 error = tcp6_connect(tp, nam, td);
1231 #if defined(INET6) && defined(INET)
1235 error = tcp_connect(tp,
1236 (struct sockaddr *)sinp, td);
1239 * The bind operation in tcp_connect succeeded. We
1240 * no longer want to restore the flags if later
1243 if (error == 0 || inp->inp_lport != 0)
1244 restoreflags = false;
1247 /* m is freed if PRUS_NOTREADY is unset. */
1248 sbflush(&so->so_snd);
1251 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1254 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1255 if ((flags & PRUS_NOTREADY) == 0) {
1256 tp->t_flags |= TF_FORCEDATA;
1257 error = tcp_output_nodrop(tp);
1258 tp->t_flags &= ~TF_FORCEDATA;
1261 TCP_LOG_EVENT(tp, NULL,
1262 &inp->inp_socket->so_rcv,
1263 &inp->inp_socket->so_snd,
1264 TCP_LOG_USERSEND, error,
1269 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1270 * responsible for freeing memory.
1272 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1276 * If the request was unsuccessful and we changed flags,
1277 * restore the original flags.
1279 if (error != 0 && restoreflags) {
1280 inp->inp_vflag = vflagsav;
1281 inp->inp_inc.inc_flags = incflagsav;
1283 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1284 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1285 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1286 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1287 error = tcp_unlock_or_drop(tp, error);
1293 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1295 struct epoch_tracker et;
1300 inp = sotoinpcb(so);
1302 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1304 mb_free_notready(m, count);
1305 return (ECONNRESET);
1307 tp = intotcpcb(inp);
1309 SOCKBUF_LOCK(&so->so_snd);
1310 error = sbready(&so->so_snd, m, count);
1311 SOCKBUF_UNLOCK(&so->so_snd);
1316 NET_EPOCH_ENTER(et);
1317 error = tcp_output_unlock(tp);
1324 * Abort the TCP. Drop the connection abruptly.
1327 tcp_usr_abort(struct socket *so)
1330 struct tcpcb *tp = NULL;
1331 struct epoch_tracker et;
1334 inp = sotoinpcb(so);
1335 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1337 NET_EPOCH_ENTER(et);
1339 KASSERT(inp->inp_socket != NULL,
1340 ("tcp_usr_abort: inp_socket == NULL"));
1343 * If we still have full TCP state, and we're not dropped, drop.
1345 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1346 !(inp->inp_flags & INP_DROPPED)) {
1347 tp = intotcpcb(inp);
1349 tp = tcp_drop(tp, ECONNABORTED);
1352 TCPDEBUG2(PRU_ABORT);
1353 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1355 if (!(inp->inp_flags & INP_DROPPED)) {
1357 so->so_state |= SS_PROTOREF;
1359 inp->inp_flags |= INP_SOCKREF;
1367 * TCP socket is closed. Start friendly disconnect.
1370 tcp_usr_close(struct socket *so)
1373 struct tcpcb *tp = NULL;
1374 struct epoch_tracker et;
1377 inp = sotoinpcb(so);
1378 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1380 NET_EPOCH_ENTER(et);
1382 KASSERT(inp->inp_socket != NULL,
1383 ("tcp_usr_close: inp_socket == NULL"));
1386 * If we still have full TCP state, and we're not dropped, initiate
1389 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1390 !(inp->inp_flags & INP_DROPPED)) {
1391 tp = intotcpcb(inp);
1394 TCPDEBUG2(PRU_CLOSE);
1395 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1397 if (!(inp->inp_flags & INP_DROPPED)) {
1399 so->so_state |= SS_PROTOREF;
1401 inp->inp_flags |= INP_SOCKREF;
1408 tcp_pru_options_support(struct tcpcb *tp, int flags)
1411 * If the specific TCP stack has a pru_options
1412 * specified then it does not always support
1413 * all the PRU_XX options and we must ask it.
1414 * If the function is not specified then all
1415 * of the PRU_XX options are supported.
1419 if (tp->t_fb->tfb_pru_options) {
1420 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1426 * Receive out-of-band data.
1429 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1433 struct tcpcb *tp = NULL;
1436 inp = sotoinpcb(so);
1437 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1439 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1443 tp = intotcpcb(inp);
1444 error = tcp_pru_options_support(tp, PRUS_OOB);
1449 if ((so->so_oobmark == 0 &&
1450 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1451 so->so_options & SO_OOBINLINE ||
1452 tp->t_oobflags & TCPOOB_HADDATA) {
1456 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1457 error = EWOULDBLOCK;
1461 *mtod(m, caddr_t) = tp->t_iobc;
1462 if ((flags & MSG_PEEK) == 0)
1463 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1466 TCPDEBUG2(PRU_RCVOOB);
1467 TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1473 struct pr_usrreqs tcp_usrreqs = {
1474 .pru_abort = tcp_usr_abort,
1475 .pru_accept = tcp_usr_accept,
1476 .pru_attach = tcp_usr_attach,
1477 .pru_bind = tcp_usr_bind,
1478 .pru_connect = tcp_usr_connect,
1479 .pru_control = in_control,
1480 .pru_detach = tcp_usr_detach,
1481 .pru_disconnect = tcp_usr_disconnect,
1482 .pru_listen = tcp_usr_listen,
1483 .pru_peeraddr = in_getpeeraddr,
1484 .pru_rcvd = tcp_usr_rcvd,
1485 .pru_rcvoob = tcp_usr_rcvoob,
1486 .pru_send = tcp_usr_send,
1487 .pru_ready = tcp_usr_ready,
1488 .pru_shutdown = tcp_usr_shutdown,
1489 .pru_sockaddr = in_getsockaddr,
1490 .pru_sosetlabel = in_pcbsosetlabel,
1491 .pru_close = tcp_usr_close,
1496 struct pr_usrreqs tcp6_usrreqs = {
1497 .pru_abort = tcp_usr_abort,
1498 .pru_accept = tcp6_usr_accept,
1499 .pru_attach = tcp_usr_attach,
1500 .pru_bind = tcp6_usr_bind,
1501 .pru_connect = tcp6_usr_connect,
1502 .pru_control = in6_control,
1503 .pru_detach = tcp_usr_detach,
1504 .pru_disconnect = tcp_usr_disconnect,
1505 .pru_listen = tcp6_usr_listen,
1506 .pru_peeraddr = in6_mapped_peeraddr,
1507 .pru_rcvd = tcp_usr_rcvd,
1508 .pru_rcvoob = tcp_usr_rcvoob,
1509 .pru_send = tcp_usr_send,
1510 .pru_ready = tcp_usr_ready,
1511 .pru_shutdown = tcp_usr_shutdown,
1512 .pru_sockaddr = in6_mapped_sockaddr,
1513 .pru_sosetlabel = in_pcbsosetlabel,
1514 .pru_close = tcp_usr_close,
1520 * Common subroutine to open a TCP connection to remote host specified
1521 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1522 * port number if needed. Call in_pcbconnect_setup to do the routing and
1523 * to choose a local host address (interface). If there is an existing
1524 * incarnation of the same connection in TIME-WAIT state and if the remote
1525 * host was sending CC options and if the connection duration was < MSL, then
1526 * truncate the previous TIME-WAIT state and proceed.
1527 * Initialize connection parameters and enter SYN-SENT state.
1530 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1532 struct inpcb *inp = tp->t_inpcb, *oinp;
1533 struct socket *so = inp->inp_socket;
1534 struct in_addr laddr;
1539 INP_WLOCK_ASSERT(inp);
1540 INP_HASH_WLOCK(&V_tcbinfo);
1542 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1543 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1549 * Cannot simply call in_pcbconnect, because there might be an
1550 * earlier incarnation of this same connection still in
1551 * TIME_WAIT state, creating an ADDRINUSE error.
1553 laddr = inp->inp_laddr;
1554 lport = inp->inp_lport;
1555 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1556 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1557 if (error && oinp == NULL)
1563 /* Handle initial bind if it hadn't been done in advance. */
1564 if (inp->inp_lport == 0) {
1565 inp->inp_lport = lport;
1566 if (in_pcbinshash(inp) != 0) {
1572 inp->inp_laddr = laddr;
1574 INP_HASH_WUNLOCK(&V_tcbinfo);
1577 * Compute window scaling to request:
1578 * Scale to fit into sweet spot. See tcp_syncache.c.
1579 * XXX: This should move to tcp_output().
1581 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1582 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1583 tp->request_r_scale++;
1586 TCPSTAT_INC(tcps_connattempt);
1587 tcp_state_change(tp, TCPS_SYN_SENT);
1588 tp->iss = tcp_new_isn(&inp->inp_inc);
1589 if (tp->t_flags & TF_REQ_TSTMP)
1590 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1591 tcp_sendseqinit(tp);
1596 INP_HASH_WUNLOCK(&V_tcbinfo);
1603 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1605 struct inpcb *inp = tp->t_inpcb;
1608 INP_WLOCK_ASSERT(inp);
1609 INP_HASH_WLOCK(&V_tcbinfo);
1611 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1612 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1616 error = in6_pcbconnect(inp, nam, td->td_ucred);
1619 INP_HASH_WUNLOCK(&V_tcbinfo);
1621 /* Compute window scaling to request. */
1622 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1623 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1624 tp->request_r_scale++;
1626 soisconnecting(inp->inp_socket);
1627 TCPSTAT_INC(tcps_connattempt);
1628 tcp_state_change(tp, TCPS_SYN_SENT);
1629 tp->iss = tcp_new_isn(&inp->inp_inc);
1630 if (tp->t_flags & TF_REQ_TSTMP)
1631 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1632 tcp_sendseqinit(tp);
1637 INP_HASH_WUNLOCK(&V_tcbinfo);
1643 * Export TCP internal state information via a struct tcp_info, based on the
1644 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1645 * (TCP state machine, etc). We export all information using FreeBSD-native
1646 * constants -- for example, the numeric values for tcpi_state will differ
1650 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1653 INP_WLOCK_ASSERT(tp->t_inpcb);
1654 bzero(ti, sizeof(*ti));
1656 ti->tcpi_state = tp->t_state;
1657 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1658 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1659 if (tp->t_flags & TF_SACK_PERMIT)
1660 ti->tcpi_options |= TCPI_OPT_SACK;
1661 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1662 ti->tcpi_options |= TCPI_OPT_WSCALE;
1663 ti->tcpi_snd_wscale = tp->snd_scale;
1664 ti->tcpi_rcv_wscale = tp->rcv_scale;
1666 if (tp->t_flags2 & TF2_ECN_PERMIT)
1667 ti->tcpi_options |= TCPI_OPT_ECN;
1669 ti->tcpi_rto = tp->t_rxtcur * tick;
1670 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1671 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1672 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1674 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1675 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1678 * FreeBSD-specific extension fields for tcp_info.
1680 ti->tcpi_rcv_space = tp->rcv_wnd;
1681 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1682 ti->tcpi_snd_wnd = tp->snd_wnd;
1683 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1684 ti->tcpi_snd_nxt = tp->snd_nxt;
1685 ti->tcpi_snd_mss = tp->t_maxseg;
1686 ti->tcpi_rcv_mss = tp->t_maxseg;
1687 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1688 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1689 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1691 if (tp->t_flags & TF_TOE) {
1692 ti->tcpi_options |= TCPI_OPT_TOE;
1693 tcp_offload_tcp_info(tp, ti);
1699 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1700 * socket option arguments. When it re-acquires the lock after the copy, it
1701 * has to revalidate that the connection is still valid for the socket
1704 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
1706 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1709 return (ECONNRESET); \
1711 tp = intotcpcb(inp); \
1713 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1716 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
1718 struct tcpcb *tp = intotcpcb(inp);
1721 MPASS(sopt->sopt_dir == SOPT_SET);
1723 if (sopt->sopt_level != IPPROTO_TCP) {
1725 if (inp->inp_vflag & INP_IPV6PROTO)
1726 error = ip6_ctloutput(inp->inp_socket, sopt);
1728 #if defined(INET6) && defined(INET)
1732 error = ip_ctloutput(inp->inp_socket, sopt);
1735 * When an IP-level socket option affects TCP, pass control
1736 * down to stack tfb_tcp_ctloutput, otherwise return what
1737 * IP level returned.
1739 switch (sopt->sopt_level) {
1742 if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
1744 switch (sopt->sopt_name) {
1746 /* Notify tcp stacks that care (e.g. RACK). */
1748 case IPV6_USE_MIN_MTU:
1749 /* Update t_maxseg accordingly. */
1758 switch (sopt->sopt_name) {
1761 /* Notify tcp stacks that care (e.g. RACK). */
1771 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
1773 * Protect the TCP option TCP_FUNCTION_BLK so
1774 * that a sub-function can *never* overwrite this.
1776 struct tcp_function_set fsn;
1777 struct tcp_function_block *blk;
1779 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
1784 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1786 return (ECONNRESET);
1788 tp = intotcpcb(inp);
1790 blk = find_and_ref_tcp_functions(&fsn);
1795 if (tp->t_fb == blk) {
1796 /* You already have this */
1797 refcount_release(&blk->tfb_refcnt);
1801 if (tp->t_state != TCPS_CLOSED) {
1803 * The user has advanced the state
1804 * past the initial point, we may not
1805 * be able to switch.
1807 if (blk->tfb_tcp_handoff_ok != NULL) {
1809 * Does the stack provide a
1810 * query mechanism, if so it may
1811 * still be possible?
1813 error = (*blk->tfb_tcp_handoff_ok)(tp);
1817 refcount_release(&blk->tfb_refcnt);
1822 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1823 refcount_release(&blk->tfb_refcnt);
1828 * Release the old refcnt, the
1829 * lookup acquired a ref on the
1832 if (tp->t_fb->tfb_tcp_fb_fini) {
1833 struct epoch_tracker et;
1835 * Tell the stack to cleanup with 0 i.e.
1836 * the tcb is not going away.
1838 NET_EPOCH_ENTER(et);
1839 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1843 /* Assure that we are not on any hpts */
1844 tcp_hpts_remove(tp->t_inpcb);
1846 if (blk->tfb_tcp_fb_init) {
1847 error = (*blk->tfb_tcp_fb_init)(tp);
1849 refcount_release(&blk->tfb_refcnt);
1850 if (tp->t_fb->tfb_tcp_fb_init) {
1851 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) {
1852 /* Fall back failed, drop the connection */
1854 soabort(inp->inp_socket);
1861 refcount_release(&tp->t_fb->tfb_refcnt);
1864 if (tp->t_flags & TF_TOE) {
1865 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1875 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1877 return (ECONNRESET);
1879 tp = intotcpcb(inp);
1881 /* Pass in the INP locked, caller must unlock it. */
1882 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp));
1886 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
1891 MPASS(sopt->sopt_dir == SOPT_GET);
1893 if (sopt->sopt_level != IPPROTO_TCP) {
1895 if (inp->inp_vflag & INP_IPV6PROTO)
1896 error = ip6_ctloutput(inp->inp_socket, sopt);
1898 #if defined(INET6) && defined(INET)
1902 error = ip_ctloutput(inp->inp_socket, sopt);
1907 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1909 return (ECONNRESET);
1911 tp = intotcpcb(inp);
1912 if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
1913 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
1914 struct tcp_function_set fsn;
1916 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
1917 memset(&fsn, 0, sizeof(fsn));
1918 find_tcp_function_alias(tp->t_fb, &fsn);
1920 strncpy(fsn.function_set_name,
1921 tp->t_fb->tfb_tcp_block_name,
1922 TCP_FUNCTION_NAME_LEN_MAX);
1923 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1925 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1927 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1931 /* Pass in the INP locked, caller must unlock it. */
1932 return (tp->t_fb->tfb_tcp_ctloutput(inp->inp_socket, sopt, inp, tp));
1936 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1942 inp = sotoinpcb(so);
1943 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1945 if (sopt->sopt_dir == SOPT_SET)
1946 return (tcp_ctloutput_set(inp, sopt));
1947 else if (sopt->sopt_dir == SOPT_GET)
1948 return (tcp_ctloutput_get(inp, sopt));
1950 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
1954 * If this assert becomes untrue, we need to change the size of the buf
1955 * variable in tcp_default_ctloutput().
1958 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1959 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1964 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1966 struct tls_enable_v0 tls_v0;
1969 if (sopt->sopt_valsize == sizeof(tls_v0)) {
1970 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1974 memset(tls, 0, sizeof(*tls));
1975 tls->cipher_key = tls_v0.cipher_key;
1976 tls->iv = tls_v0.iv;
1977 tls->auth_key = tls_v0.auth_key;
1978 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1979 tls->cipher_key_len = tls_v0.cipher_key_len;
1980 tls->iv_len = tls_v0.iv_len;
1981 tls->auth_algorithm = tls_v0.auth_algorithm;
1982 tls->auth_key_len = tls_v0.auth_key_len;
1983 tls->flags = tls_v0.flags;
1984 tls->tls_vmajor = tls_v0.tls_vmajor;
1985 tls->tls_vminor = tls_v0.tls_vminor;
1989 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
1993 extern struct cc_algo newreno_cc_algo;
1996 tcp_congestion(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1998 struct cc_algo *algo;
2000 struct cc_var cc_mem;
2001 char buf[TCP_CA_NAME_MAX];
2006 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2009 buf[sopt->sopt_valsize] = '\0';
2011 STAILQ_FOREACH(algo, &cc_list, entries)
2012 if (strncmp(buf, algo->name,
2013 TCP_CA_NAME_MAX) == 0) {
2014 if (algo->flags & CC_MODULE_BEING_REMOVED) {
2015 /* We can't "see" modules being unloaded */
2025 if (algo->cb_init != NULL) {
2026 /* We can now pre-get the memory for the CC */
2027 mem_sz = (*algo->cc_data_sz)();
2032 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
2034 STAILQ_FOREACH(algo, &cc_list, entries)
2035 if (strncmp(buf, algo->name,
2036 TCP_CA_NAME_MAX) == 0)
2040 free(ptr, M_CC_MEM);
2050 * Make sure its all clean and zero and also get
2053 memset(&cc_mem, 0, sizeof(cc_mem));
2054 if (mem_sz != (*algo->cc_data_sz)()) {
2056 free(ptr, M_CC_MEM);
2060 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2063 free(ptr, M_CC_MEM);
2064 return (ECONNRESET);
2066 tp = intotcpcb(inp);
2068 memset(ptr, 0, mem_sz);
2070 cc_mem.ccvc.tcp = tp;
2072 * We once again hold a write lock over the tcb so it's
2073 * safe to do these things without ordering concerns.
2074 * Note here we init into stack memory.
2076 if (algo->cb_init != NULL)
2077 error = algo->cb_init(&cc_mem, ptr);
2081 * The CC algorithms, when given their memory
2082 * should not fail we could in theory have a
2087 * Touchdown, lets go ahead and move the
2088 * connection to the new CC module by
2089 * copying in the cc_mem after we call
2090 * the old ones cleanup (if any).
2092 if (CC_ALGO(tp)->cb_destroy != NULL)
2093 CC_ALGO(tp)->cb_destroy(tp->ccv);
2094 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2096 /* Ok now are we where we have gotten past any conn_init? */
2097 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
2098 /* Yep run the connection init for the new CC */
2099 CC_ALGO(tp)->conn_init(tp->ccv);
2102 free(ptr, M_CC_MEM);
2108 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
2110 int error, opt, optval;
2114 struct tls_enable tls;
2116 char *pbuf, buf[TCP_LOG_ID_LEN];
2118 struct statsblob *sbp;
2122 INP_WLOCK_ASSERT(inp);
2124 switch (sopt->sopt_level) {
2127 MPASS(inp->inp_vflag & INP_IPV6PROTO);
2128 switch (sopt->sopt_name) {
2129 case IPV6_USE_MIN_MTU:
2130 tcp6_use_min_mtu(tp);
2144 * For TCP_CCALGOOPT forward the control to CC module, for both
2145 * SOPT_SET and SOPT_GET.
2147 switch (sopt->sopt_name) {
2150 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2152 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2153 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2154 sopt->sopt_valsize);
2159 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2160 if (CC_ALGO(tp)->ctl_output != NULL)
2161 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
2165 if (error == 0 && sopt->sopt_dir == SOPT_GET)
2166 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2171 switch (sopt->sopt_dir) {
2173 switch (sopt->sopt_name) {
2174 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2176 if (!TCPMD5_ENABLED()) {
2178 return (ENOPROTOOPT);
2180 error = TCPMD5_PCBCTL(inp, sopt);
2183 goto unlock_and_done;
2190 error = sooptcopyin(sopt, &optval, sizeof optval,
2195 INP_WLOCK_RECHECK(inp);
2196 switch (sopt->sopt_name) {
2207 opt = 0; /* dead code to fool gcc */
2214 tp->t_flags &= ~opt;
2217 if (tp->t_flags & TF_TOE) {
2218 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2227 error = sooptcopyin(sopt, &optval, sizeof optval,
2232 INP_WLOCK_RECHECK(inp);
2234 tp->t_flags |= TF_NOPUSH;
2235 else if (tp->t_flags & TF_NOPUSH) {
2236 tp->t_flags &= ~TF_NOPUSH;
2237 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2238 struct epoch_tracker et;
2240 NET_EPOCH_ENTER(et);
2241 error = tcp_output_nodrop(tp);
2245 goto unlock_and_done;
2247 case TCP_REMOTE_UDP_ENCAPS_PORT:
2249 error = sooptcopyin(sopt, &optval, sizeof optval,
2253 if ((optval < TCP_TUNNELING_PORT_MIN) ||
2254 (optval > TCP_TUNNELING_PORT_MAX)) {
2255 /* Its got to be in range */
2258 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2259 /* You have to have enabled a UDP tunneling port first */
2262 INP_WLOCK_RECHECK(inp);
2263 if (tp->t_state != TCPS_CLOSED) {
2264 /* You can't change after you are connected */
2267 /* Ok we are all good set the port */
2268 tp->t_port = htons(optval);
2270 goto unlock_and_done;
2274 error = sooptcopyin(sopt, &optval, sizeof optval,
2279 INP_WLOCK_RECHECK(inp);
2280 if (optval > 0 && optval <= tp->t_maxseg &&
2281 optval + 40 >= V_tcp_minmss)
2282 tp->t_maxseg = optval;
2285 goto unlock_and_done;
2295 error = sooptcopyin(sopt, &optval, sizeof optval,
2301 sbp = stats_blob_alloc(
2302 V_tcp_perconn_stats_dflt_tpl, 0);
2306 INP_WLOCK_RECHECK(inp);
2307 if ((tp->t_stats != NULL && sbp == NULL) ||
2308 (tp->t_stats == NULL && sbp != NULL)) {
2309 struct statsblob *t = tp->t_stats;
2315 stats_blob_destroy(sbp);
2317 return (EOPNOTSUPP);
2321 case TCP_CONGESTION:
2322 error = tcp_congestion(so, sopt, inp, tp);
2325 case TCP_REUSPORT_LB_NUMA:
2327 error = sooptcopyin(sopt, &optval, sizeof(optval),
2329 INP_WLOCK_RECHECK(inp);
2331 error = in_pcblbgroup_numa(inp, optval);
2336 case TCP_TXTLS_ENABLE:
2338 error = copyin_tls_enable(sopt, &tls);
2341 error = ktls_enable_tx(so, &tls);
2343 case TCP_TXTLS_MODE:
2345 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2349 INP_WLOCK_RECHECK(inp);
2350 error = ktls_set_tx_mode(so, ui);
2353 case TCP_RXTLS_ENABLE:
2355 error = sooptcopyin(sopt, &tls, sizeof(tls),
2359 error = ktls_enable_rx(so, &tls);
2367 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2371 if (ui > (UINT_MAX / hz)) {
2377 INP_WLOCK_RECHECK(inp);
2378 switch (sopt->sopt_name) {
2380 tp->t_keepidle = ui;
2382 * XXX: better check current remaining
2383 * timeout and "merge" it with new value.
2385 if ((tp->t_state > TCPS_LISTEN) &&
2386 (tp->t_state <= TCPS_CLOSING))
2387 tcp_timer_activate(tp, TT_KEEP,
2391 tp->t_keepintvl = ui;
2392 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2393 (TP_MAXIDLE(tp) > 0))
2394 tcp_timer_activate(tp, TT_2MSL,
2398 tp->t_keepinit = ui;
2399 if (tp->t_state == TCPS_SYN_RECEIVED ||
2400 tp->t_state == TCPS_SYN_SENT)
2401 tcp_timer_activate(tp, TT_KEEP,
2405 goto unlock_and_done;
2409 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2413 INP_WLOCK_RECHECK(inp);
2415 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2416 (TP_MAXIDLE(tp) > 0))
2417 tcp_timer_activate(tp, TT_2MSL,
2419 goto unlock_and_done;
2425 error = sooptcopyin(sopt, &optval, sizeof optval,
2430 INP_WLOCK_RECHECK(inp);
2432 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2433 &(tp->t_outpkts) : &(tp->t_inpkts),
2437 goto unlock_and_done;
2440 case TCP_FASTOPEN: {
2441 struct tcp_fastopen tfo_optval;
2444 if (!V_tcp_fastopen_client_enable &&
2445 !V_tcp_fastopen_server_enable)
2448 error = sooptcopyin(sopt, &tfo_optval,
2449 sizeof(tfo_optval), sizeof(int));
2453 INP_WLOCK_RECHECK(inp);
2454 if ((tp->t_state != TCPS_CLOSED) &&
2455 (tp->t_state != TCPS_LISTEN)) {
2457 goto unlock_and_done;
2459 if (tfo_optval.enable) {
2460 if (tp->t_state == TCPS_LISTEN) {
2461 if (!V_tcp_fastopen_server_enable) {
2463 goto unlock_and_done;
2466 if (tp->t_tfo_pending == NULL)
2468 tcp_fastopen_alloc_counter();
2471 * If a pre-shared key was provided,
2472 * stash it in the client cookie
2473 * field of the tcpcb for use during
2476 if (sopt->sopt_valsize ==
2477 sizeof(tfo_optval)) {
2478 memcpy(tp->t_tfo_cookie.client,
2480 TCP_FASTOPEN_PSK_LEN);
2481 tp->t_tfo_client_cookie_len =
2482 TCP_FASTOPEN_PSK_LEN;
2485 tp->t_flags |= TF_FASTOPEN;
2487 tp->t_flags &= ~TF_FASTOPEN;
2488 goto unlock_and_done;
2494 error = sooptcopyin(sopt, &optval, sizeof optval,
2499 INP_WLOCK_RECHECK(inp);
2500 error = tcp_log_state_change(tp, optval);
2501 goto unlock_and_done;
2510 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2513 buf[sopt->sopt_valsize] = '\0';
2514 INP_WLOCK_RECHECK(inp);
2515 error = tcp_log_set_id(tp, buf);
2516 /* tcp_log_set_id() unlocks the INP. */
2523 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2526 buf[sopt->sopt_valsize] = '\0';
2527 INP_WLOCK_RECHECK(inp);
2528 if (sopt->sopt_name == TCP_LOGDUMP) {
2529 error = tcp_log_dump_tp_logbuf(tp, buf,
2533 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2535 * tcp_log_dump_tp_bucket_logbufs() drops the
2544 error = ENOPROTOOPT;
2550 tp = intotcpcb(inp);
2551 switch (sopt->sopt_name) {
2552 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2554 if (!TCPMD5_ENABLED()) {
2556 return (ENOPROTOOPT);
2558 error = TCPMD5_PCBCTL(inp, sopt);
2563 optval = tp->t_flags & TF_NODELAY;
2565 error = sooptcopyout(sopt, &optval, sizeof optval);
2568 optval = tp->t_maxseg;
2570 error = sooptcopyout(sopt, &optval, sizeof optval);
2572 case TCP_REMOTE_UDP_ENCAPS_PORT:
2573 optval = ntohs(tp->t_port);
2575 error = sooptcopyout(sopt, &optval, sizeof optval);
2578 optval = tp->t_flags & TF_NOOPT;
2580 error = sooptcopyout(sopt, &optval, sizeof optval);
2583 optval = tp->t_flags & TF_NOPUSH;
2585 error = sooptcopyout(sopt, &optval, sizeof optval);
2588 tcp_fill_info(tp, &ti);
2590 error = sooptcopyout(sopt, &ti, sizeof ti);
2596 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2599 socklen_t outsbsz = sopt->sopt_valsize;
2600 if (tp->t_stats == NULL)
2602 else if (outsbsz >= tp->t_stats->cursz)
2603 outsbsz = tp->t_stats->cursz;
2604 else if (outsbsz >= sizeof(struct statsblob))
2605 outsbsz = sizeof(struct statsblob);
2612 sbp = sopt->sopt_val;
2613 nheld = atop(round_page(((vm_offset_t)sbp) +
2614 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2615 vm_page_t ma[nheld];
2616 if (vm_fault_quick_hold_pages(
2617 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2618 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2624 if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2625 SIZEOF_MEMBER(struct statsblob, flags))))
2628 INP_WLOCK_RECHECK(inp);
2629 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2630 sbflags | SB_CLONE_USRDSTNOFAULT);
2632 sopt->sopt_valsize = outsbsz;
2634 vm_page_unhold_pages(ma, nheld);
2641 case TCP_CONGESTION:
2642 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2644 error = sooptcopyout(sopt, buf, len + 1);
2650 switch (sopt->sopt_name) {
2652 ui = TP_KEEPIDLE(tp) / hz;
2655 ui = TP_KEEPINTVL(tp) / hz;
2658 ui = TP_KEEPINIT(tp) / hz;
2661 ui = TP_KEEPCNT(tp);
2665 error = sooptcopyout(sopt, &ui, sizeof(ui));
2670 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2671 &(tp->t_outpkts) : &(tp->t_inpkts));
2673 error = sooptcopyout(sopt, &optval, sizeof optval);
2677 optval = tp->t_flags & TF_FASTOPEN;
2679 error = sooptcopyout(sopt, &optval, sizeof optval);
2683 optval = tp->t_logstate;
2685 error = sooptcopyout(sopt, &optval, sizeof(optval));
2688 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2689 error = tcp_log_getlogbuf(sopt, tp);
2692 len = tcp_log_get_id(tp, buf);
2694 error = sooptcopyout(sopt, buf, len + 1);
2703 case TCP_TXTLS_MODE:
2704 error = ktls_get_tx_mode(so, &optval);
2707 error = sooptcopyout(sopt, &optval,
2710 case TCP_RXTLS_MODE:
2711 error = ktls_get_rx_mode(so, &optval);
2714 error = sooptcopyout(sopt, &optval,
2719 optval = tp->t_flags & TF_LRD;
2721 error = sooptcopyout(sopt, &optval, sizeof optval);
2725 error = ENOPROTOOPT;
2732 #undef INP_WLOCK_RECHECK
2733 #undef INP_WLOCK_RECHECK_CLEANUP
2736 * Initiate (or continue) disconnect.
2737 * If embryonic state, just send reset (once).
2738 * If in ``let data drain'' option and linger null, just drop.
2739 * Otherwise (hard), mark socket disconnecting and drop
2740 * current input data; switch states based on user close, and
2741 * send segment to peer (with FIN).
2744 tcp_disconnect(struct tcpcb *tp)
2746 struct inpcb *inp = tp->t_inpcb;
2747 struct socket *so = inp->inp_socket;
2750 INP_WLOCK_ASSERT(inp);
2753 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2754 * socket is still open.
2756 if (tp->t_state < TCPS_ESTABLISHED &&
2757 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2760 ("tcp_disconnect: tcp_close() returned NULL"));
2761 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2762 tp = tcp_drop(tp, 0);
2764 ("tcp_disconnect: tcp_drop() returned NULL"));
2766 soisdisconnecting(so);
2767 sbflush(&so->so_rcv);
2769 if (!(inp->inp_flags & INP_DROPPED))
2770 /* Ignore stack's drop request, we already at it. */
2771 (void)tcp_output_nodrop(tp);
2776 * User issued close, and wish to trail through shutdown states:
2777 * if never received SYN, just forget it. If got a SYN from peer,
2778 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2779 * If already got a FIN from peer, then almost done; go to LAST_ACK
2780 * state. In all other cases, have already sent FIN to peer (e.g.
2781 * after PRU_SHUTDOWN), and just have to play tedious game waiting
2782 * for peer to send FIN or not respond to keep-alives, etc.
2783 * We can let the user exit from the close as soon as the FIN is acked.
2786 tcp_usrclosed(struct tcpcb *tp)
2790 INP_WLOCK_ASSERT(tp->t_inpcb);
2792 switch (tp->t_state) {
2795 tcp_offload_listen_stop(tp);
2797 tcp_state_change(tp, TCPS_CLOSED);
2802 * tcp_close() should never return NULL here as the socket is
2806 ("tcp_usrclosed: tcp_close() returned NULL"));
2810 case TCPS_SYN_RECEIVED:
2811 tp->t_flags |= TF_NEEDFIN;
2814 case TCPS_ESTABLISHED:
2815 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2818 case TCPS_CLOSE_WAIT:
2819 tcp_state_change(tp, TCPS_LAST_ACK);
2822 if (tp->t_state >= TCPS_FIN_WAIT_2) {
2823 soisdisconnected(tp->t_inpcb->inp_socket);
2824 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2825 if (tp->t_state == TCPS_FIN_WAIT_2) {
2828 timeout = (tcp_fast_finwait2_recycle) ?
2829 tcp_finwait2_timeout : TP_MAXIDLE(tp);
2830 tcp_timer_activate(tp, TT_2MSL, timeout);
2837 db_print_indent(int indent)
2841 for (i = 0; i < indent; i++)
2846 db_print_tstate(int t_state)
2851 db_printf("TCPS_CLOSED");
2855 db_printf("TCPS_LISTEN");
2859 db_printf("TCPS_SYN_SENT");
2862 case TCPS_SYN_RECEIVED:
2863 db_printf("TCPS_SYN_RECEIVED");
2866 case TCPS_ESTABLISHED:
2867 db_printf("TCPS_ESTABLISHED");
2870 case TCPS_CLOSE_WAIT:
2871 db_printf("TCPS_CLOSE_WAIT");
2874 case TCPS_FIN_WAIT_1:
2875 db_printf("TCPS_FIN_WAIT_1");
2879 db_printf("TCPS_CLOSING");
2883 db_printf("TCPS_LAST_ACK");
2886 case TCPS_FIN_WAIT_2:
2887 db_printf("TCPS_FIN_WAIT_2");
2890 case TCPS_TIME_WAIT:
2891 db_printf("TCPS_TIME_WAIT");
2895 db_printf("unknown");
2901 db_print_tflags(u_int t_flags)
2906 if (t_flags & TF_ACKNOW) {
2907 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2910 if (t_flags & TF_DELACK) {
2911 db_printf("%sTF_DELACK", comma ? ", " : "");
2914 if (t_flags & TF_NODELAY) {
2915 db_printf("%sTF_NODELAY", comma ? ", " : "");
2918 if (t_flags & TF_NOOPT) {
2919 db_printf("%sTF_NOOPT", comma ? ", " : "");
2922 if (t_flags & TF_SENTFIN) {
2923 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2926 if (t_flags & TF_REQ_SCALE) {
2927 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2930 if (t_flags & TF_RCVD_SCALE) {
2931 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2934 if (t_flags & TF_REQ_TSTMP) {
2935 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2938 if (t_flags & TF_RCVD_TSTMP) {
2939 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2942 if (t_flags & TF_SACK_PERMIT) {
2943 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2946 if (t_flags & TF_NEEDSYN) {
2947 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2950 if (t_flags & TF_NEEDFIN) {
2951 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2954 if (t_flags & TF_NOPUSH) {
2955 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2958 if (t_flags & TF_MORETOCOME) {
2959 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2962 if (t_flags & TF_LQ_OVERFLOW) {
2963 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2966 if (t_flags & TF_LASTIDLE) {
2967 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2970 if (t_flags & TF_RXWIN0SENT) {
2971 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2974 if (t_flags & TF_FASTRECOVERY) {
2975 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2978 if (t_flags & TF_CONGRECOVERY) {
2979 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2982 if (t_flags & TF_WASFRECOVERY) {
2983 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2986 if (t_flags & TF_SIGNATURE) {
2987 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2990 if (t_flags & TF_FORCEDATA) {
2991 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2994 if (t_flags & TF_TSO) {
2995 db_printf("%sTF_TSO", comma ? ", " : "");
2998 if (t_flags & TF_FASTOPEN) {
2999 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
3005 db_print_tflags2(u_int t_flags2)
3010 if (t_flags2 & TF2_ECN_PERMIT) {
3011 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
3017 db_print_toobflags(char t_oobflags)
3022 if (t_oobflags & TCPOOB_HAVEDATA) {
3023 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
3026 if (t_oobflags & TCPOOB_HADDATA) {
3027 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
3033 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
3036 db_print_indent(indent);
3037 db_printf("%s at %p\n", name, tp);
3041 db_print_indent(indent);
3042 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
3043 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
3045 db_print_indent(indent);
3046 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
3047 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
3049 db_print_indent(indent);
3050 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
3051 &tp->t_timers->tt_delack, tp->t_inpcb);
3053 db_print_indent(indent);
3054 db_printf("t_state: %d (", tp->t_state);
3055 db_print_tstate(tp->t_state);
3058 db_print_indent(indent);
3059 db_printf("t_flags: 0x%x (", tp->t_flags);
3060 db_print_tflags(tp->t_flags);
3063 db_print_indent(indent);
3064 db_printf("t_flags2: 0x%x (", tp->t_flags2);
3065 db_print_tflags2(tp->t_flags2);
3068 db_print_indent(indent);
3069 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
3070 tp->snd_una, tp->snd_max, tp->snd_nxt);
3072 db_print_indent(indent);
3073 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
3074 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
3076 db_print_indent(indent);
3077 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
3078 tp->iss, tp->irs, tp->rcv_nxt);
3080 db_print_indent(indent);
3081 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
3082 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
3084 db_print_indent(indent);
3085 db_printf("snd_wnd: %u snd_cwnd: %u\n",
3086 tp->snd_wnd, tp->snd_cwnd);
3088 db_print_indent(indent);
3089 db_printf("snd_ssthresh: %u snd_recover: "
3090 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
3092 db_print_indent(indent);
3093 db_printf("t_rcvtime: %u t_startime: %u\n",
3094 tp->t_rcvtime, tp->t_starttime);
3096 db_print_indent(indent);
3097 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
3098 tp->t_rtttime, tp->t_rtseq);
3100 db_print_indent(indent);
3101 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
3102 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
3104 db_print_indent(indent);
3105 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
3106 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
3109 db_print_indent(indent);
3110 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n",
3111 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
3113 db_print_indent(indent);
3114 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
3115 db_print_toobflags(tp->t_oobflags);
3116 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
3118 db_print_indent(indent);
3119 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
3120 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
3122 db_print_indent(indent);
3123 db_printf("ts_recent: %u ts_recent_age: %u\n",
3124 tp->ts_recent, tp->ts_recent_age);
3126 db_print_indent(indent);
3127 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
3128 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
3130 db_print_indent(indent);
3131 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
3132 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
3133 tp->snd_recover_prev, tp->t_badrxtwin);
3135 db_print_indent(indent);
3136 db_printf("snd_numholes: %d snd_holes first: %p\n",
3137 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
3139 db_print_indent(indent);
3140 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n",
3141 tp->snd_fack, tp->rcv_numsacks);
3143 /* Skip sackblks, sackhint. */
3145 db_print_indent(indent);
3146 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
3147 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
3150 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
3155 db_printf("usage: show tcpcb <addr>\n");
3158 tp = (struct tcpcb *)addr;
3160 db_print_tcpcb(tp, "tcpcb", 0);