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, it must be either
276 * dropped or embryonic.
278 KASSERT(inp->inp_flags & INP_DROPPED ||
279 tp->t_state < TCPS_SYN_SENT,
280 ("%s: inp %p not dropped or embryonic", __func__, inp));
289 * Give the socket an address.
292 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
297 struct tcpcb *tp = NULL;
299 struct sockaddr_in *sinp;
301 sinp = (struct sockaddr_in *)nam;
302 if (nam->sa_family != AF_INET) {
304 * Preserve compatibility with old programs.
306 if (nam->sa_family != AF_UNSPEC ||
307 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
308 sinp->sin_addr.s_addr != INADDR_ANY)
309 return (EAFNOSUPPORT);
310 nam->sa_family = AF_INET;
312 if (nam->sa_len != sizeof(*sinp))
316 * Must check for multicast addresses and disallow binding
319 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
320 return (EAFNOSUPPORT);
324 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
326 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
334 INP_HASH_WLOCK(&V_tcbinfo);
335 error = in_pcbbind(inp, nam, td->td_ucred);
336 INP_HASH_WUNLOCK(&V_tcbinfo);
339 TCP_PROBE2(debug__user, tp, PRU_BIND);
348 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
353 struct tcpcb *tp = NULL;
355 struct sockaddr_in6 *sin6;
358 sin6 = (struct sockaddr_in6 *)nam;
359 if (nam->sa_family != AF_INET6)
360 return (EAFNOSUPPORT);
361 if (nam->sa_len != sizeof(*sin6))
365 * Must check for multicast addresses and disallow binding
368 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
369 return (EAFNOSUPPORT);
373 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
375 vflagsav = inp->inp_vflag;
376 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;
796 struct tcpcb *tp = NULL;
802 if (so->so_state & SS_ISDISCONNECTED)
803 return (ECONNABORTED);
806 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
808 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
809 error = ECONNABORTED;
818 * We inline in_getpeeraddr and COMMON_END here, so that we can
819 * copy the data of interest and defer the malloc until after we
822 port = inp->inp_fport;
823 addr = inp->inp_faddr;
826 TCPDEBUG2(PRU_ACCEPT);
827 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
830 *nam = in_sockaddr(port, &addr);
837 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
839 struct inpcb *inp = NULL;
842 struct tcpcb *tp = NULL;
845 struct in6_addr addr6;
846 struct epoch_tracker et;
851 if (so->so_state & SS_ISDISCONNECTED)
852 return (ECONNABORTED);
855 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
858 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
859 error = ECONNABORTED;
868 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
869 * copy the data of interest and defer the malloc until after we
872 if (inp->inp_vflag & INP_IPV4) {
874 port = inp->inp_fport;
875 addr = inp->inp_faddr;
877 port = inp->inp_fport;
878 addr6 = inp->in6p_faddr;
882 TCPDEBUG2(PRU_ACCEPT);
883 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
888 *nam = in6_v4mapsin6_sockaddr(port, &addr);
890 *nam = in6_sockaddr(port, &addr6);
897 * Mark the connection as being incapable of further output.
900 tcp_usr_shutdown(struct socket *so)
904 struct tcpcb *tp = NULL;
905 struct epoch_tracker et;
909 KASSERT(inp != NULL, ("inp == NULL"));
911 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
920 if (!(inp->inp_flags & INP_DROPPED))
921 error = tcp_output_nodrop(tp);
922 TCPDEBUG2(PRU_SHUTDOWN);
923 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
924 error = tcp_unlock_or_drop(tp, error);
931 * After a receive, possibly send window update to peer.
934 tcp_usr_rcvd(struct socket *so, int flags)
936 struct epoch_tracker et;
938 struct tcpcb *tp = NULL;
939 int outrv = 0, error = 0;
943 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
945 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
953 * For passively-created TFO connections, don't attempt a window
954 * update while still in SYN_RECEIVED as this may trigger an early
955 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with
956 * application response data, or failing that, when the DELACK timer
959 if (IS_FASTOPEN(tp->t_flags) &&
960 (tp->t_state == TCPS_SYN_RECEIVED))
963 if (tp->t_flags & TF_TOE)
964 tcp_offload_rcvd(tp);
967 outrv = tcp_output_nodrop(tp);
970 TCP_PROBE2(debug__user, tp, PRU_RCVD);
971 (void) tcp_unlock_or_drop(tp, outrv);
977 * Do a send by putting data in output queue and updating urgent
978 * marker if URG set. Possibly send more data. Unlike the other
979 * pru_*() routines, the mbuf chains are our responsibility. We
980 * must either enqueue them or free them. The other pru_* routines
981 * generally are caller-frees.
984 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
985 struct sockaddr *nam, struct mbuf *control, struct thread *td)
987 struct epoch_tracker et;
990 struct tcpcb *tp = NULL;
993 struct sockaddr_in sin;
995 struct sockaddr_in *sinp;
1000 u_int8_t incflagsav;
1005 if (control != NULL) {
1006 /* TCP doesn't do control messages (rights, creds, etc) */
1007 if (control->m_len) {
1011 m_freem(control); /* empty control, just free it */
1014 inp = sotoinpcb(so);
1015 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
1017 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1018 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1021 return (ECONNRESET);
1024 vflagsav = inp->inp_vflag;
1025 incflagsav = inp->inp_inc.inc_flags;
1026 restoreflags = false;
1027 tp = intotcpcb(inp);
1029 NET_EPOCH_ENTER(et);
1030 if ((flags & PRUS_OOB) != 0 &&
1031 (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1035 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1036 if (tp->t_state == TCPS_LISTEN) {
1040 switch (nam->sa_family) {
1043 sinp = (struct sockaddr_in *)nam;
1044 if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1048 if ((inp->inp_vflag & INP_IPV6) != 0) {
1049 error = EAFNOSUPPORT;
1052 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1053 error = EAFNOSUPPORT;
1056 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1060 if ((error = prison_remote_ip4(td->td_ucred,
1071 struct sockaddr_in6 *sin6;
1073 sin6 = (struct sockaddr_in6 *)nam;
1074 if (sin6->sin6_len != sizeof(*sin6)) {
1078 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1079 error = EAFNOSUPPORT;
1082 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1083 error = EAFNOSUPPORT;
1086 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1088 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1092 if ((inp->inp_vflag & INP_IPV4) == 0) {
1093 error = EAFNOSUPPORT;
1096 restoreflags = true;
1097 inp->inp_vflag &= ~INP_IPV6;
1099 in6_sin6_2_sin(sinp, sin6);
1101 ntohl(sinp->sin_addr.s_addr))) {
1102 error = EAFNOSUPPORT;
1105 if ((error = prison_remote_ip4(td->td_ucred,
1110 error = EAFNOSUPPORT;
1114 if ((inp->inp_vflag & INP_IPV6) == 0) {
1115 error = EAFNOSUPPORT;
1118 restoreflags = true;
1119 inp->inp_vflag &= ~INP_IPV4;
1120 inp->inp_inc.inc_flags |= INC_ISIPV6;
1121 if ((error = prison_remote_ip6(td->td_ucred,
1130 error = EAFNOSUPPORT;
1134 if (!(flags & PRUS_OOB)) {
1135 sbappendstream(&so->so_snd, m, flags);
1137 if (nam && tp->t_state < TCPS_SYN_SENT) {
1138 KASSERT(tp->t_state == TCPS_CLOSED,
1139 ("%s: tp %p is listening", __func__, tp));
1142 * Do implied connect if not yet connected,
1143 * initialize window to default value, and
1144 * initialize maxseg using peer's cached MSS.
1148 error = tcp6_connect(tp, nam, td);
1150 #if defined(INET6) && defined(INET)
1154 error = tcp_connect(tp,
1155 (struct sockaddr *)sinp, td);
1158 * The bind operation in tcp_connect succeeded. We
1159 * no longer want to restore the flags if later
1162 if (error == 0 || inp->inp_lport != 0)
1163 restoreflags = false;
1166 /* m is freed if PRUS_NOTREADY is unset. */
1167 sbflush(&so->so_snd);
1170 if (IS_FASTOPEN(tp->t_flags))
1171 tcp_fastopen_connect(tp);
1173 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1177 if (flags & PRUS_EOF) {
1179 * Close the send side of the connection after
1185 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1186 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1187 (tp->t_fbyte_out == 0) &&
1188 (so->so_snd.sb_ccc > 0)) {
1189 tp->t_fbyte_out = ticks;
1190 if (tp->t_fbyte_out == 0)
1191 tp->t_fbyte_out = 1;
1192 if (tp->t_fbyte_out && tp->t_fbyte_in)
1193 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1195 if (!(inp->inp_flags & INP_DROPPED) &&
1196 !(flags & PRUS_NOTREADY)) {
1197 if (flags & PRUS_MORETOCOME)
1198 tp->t_flags |= TF_MORETOCOME;
1199 error = tcp_output_nodrop(tp);
1200 if (flags & PRUS_MORETOCOME)
1201 tp->t_flags &= ~TF_MORETOCOME;
1205 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1207 SOCKBUF_LOCK(&so->so_snd);
1208 if (sbspace(&so->so_snd) < -512) {
1209 SOCKBUF_UNLOCK(&so->so_snd);
1214 * According to RFC961 (Assigned Protocols),
1215 * the urgent pointer points to the last octet
1216 * of urgent data. We continue, however,
1217 * to consider it to indicate the first octet
1218 * of data past the urgent section.
1219 * Otherwise, snd_up should be one lower.
1221 sbappendstream_locked(&so->so_snd, m, flags);
1222 SOCKBUF_UNLOCK(&so->so_snd);
1224 if (nam && tp->t_state < TCPS_SYN_SENT) {
1226 * Do implied connect if not yet connected,
1227 * initialize window to default value, and
1228 * initialize maxseg using peer's cached MSS.
1232 * Not going to contemplate SYN|URG
1234 if (IS_FASTOPEN(tp->t_flags))
1235 tp->t_flags &= ~TF_FASTOPEN;
1238 error = tcp6_connect(tp, nam, td);
1240 #if defined(INET6) && defined(INET)
1244 error = tcp_connect(tp,
1245 (struct sockaddr *)sinp, td);
1248 * The bind operation in tcp_connect succeeded. We
1249 * no longer want to restore the flags if later
1252 if (error == 0 || inp->inp_lport != 0)
1253 restoreflags = false;
1256 /* m is freed if PRUS_NOTREADY is unset. */
1257 sbflush(&so->so_snd);
1260 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1263 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1264 if ((flags & PRUS_NOTREADY) == 0) {
1265 tp->t_flags |= TF_FORCEDATA;
1266 error = tcp_output_nodrop(tp);
1267 tp->t_flags &= ~TF_FORCEDATA;
1270 TCP_LOG_EVENT(tp, NULL,
1271 &inp->inp_socket->so_rcv,
1272 &inp->inp_socket->so_snd,
1273 TCP_LOG_USERSEND, error,
1278 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1279 * responsible for freeing memory.
1281 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1285 * If the request was unsuccessful and we changed flags,
1286 * restore the original flags.
1288 if (error != 0 && restoreflags) {
1289 inp->inp_vflag = vflagsav;
1290 inp->inp_inc.inc_flags = incflagsav;
1292 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1293 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1294 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1295 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1296 error = tcp_unlock_or_drop(tp, error);
1302 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1304 struct epoch_tracker et;
1309 inp = sotoinpcb(so);
1311 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1313 mb_free_notready(m, count);
1314 return (ECONNRESET);
1316 tp = intotcpcb(inp);
1318 SOCKBUF_LOCK(&so->so_snd);
1319 error = sbready(&so->so_snd, m, count);
1320 SOCKBUF_UNLOCK(&so->so_snd);
1325 NET_EPOCH_ENTER(et);
1326 error = tcp_output_unlock(tp);
1333 * Abort the TCP. Drop the connection abruptly.
1336 tcp_usr_abort(struct socket *so)
1339 struct tcpcb *tp = NULL;
1340 struct epoch_tracker et;
1343 inp = sotoinpcb(so);
1344 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1346 NET_EPOCH_ENTER(et);
1348 KASSERT(inp->inp_socket != NULL,
1349 ("tcp_usr_abort: inp_socket == NULL"));
1352 * If we still have full TCP state, and we're not dropped, drop.
1354 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1355 !(inp->inp_flags & INP_DROPPED)) {
1356 tp = intotcpcb(inp);
1358 tp = tcp_drop(tp, ECONNABORTED);
1361 TCPDEBUG2(PRU_ABORT);
1362 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1364 if (!(inp->inp_flags & INP_DROPPED)) {
1366 inp->inp_flags |= INP_SOCKREF;
1374 * TCP socket is closed. Start friendly disconnect.
1377 tcp_usr_close(struct socket *so)
1380 struct tcpcb *tp = NULL;
1381 struct epoch_tracker et;
1384 inp = sotoinpcb(so);
1385 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1387 NET_EPOCH_ENTER(et);
1389 KASSERT(inp->inp_socket != NULL,
1390 ("tcp_usr_close: inp_socket == NULL"));
1393 * If we still have full TCP state, and we're not dropped, initiate
1396 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1397 !(inp->inp_flags & INP_DROPPED)) {
1398 tp = intotcpcb(inp);
1399 tp->t_flags |= TF_CLOSED;
1402 TCPDEBUG2(PRU_CLOSE);
1403 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1405 if (!(inp->inp_flags & INP_DROPPED)) {
1407 inp->inp_flags |= INP_SOCKREF;
1414 tcp_pru_options_support(struct tcpcb *tp, int flags)
1417 * If the specific TCP stack has a pru_options
1418 * specified then it does not always support
1419 * all the PRU_XX options and we must ask it.
1420 * If the function is not specified then all
1421 * of the PRU_XX options are supported.
1425 if (tp->t_fb->tfb_pru_options) {
1426 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1432 * Receive out-of-band data.
1435 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1439 struct tcpcb *tp = NULL;
1442 inp = sotoinpcb(so);
1443 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1445 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1449 tp = intotcpcb(inp);
1450 error = tcp_pru_options_support(tp, PRUS_OOB);
1455 if ((so->so_oobmark == 0 &&
1456 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1457 so->so_options & SO_OOBINLINE ||
1458 tp->t_oobflags & TCPOOB_HADDATA) {
1462 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1463 error = EWOULDBLOCK;
1467 *mtod(m, caddr_t) = tp->t_iobc;
1468 if ((flags & MSG_PEEK) == 0)
1469 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1472 TCPDEBUG2(PRU_RCVOOB);
1473 TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1479 struct pr_usrreqs tcp_usrreqs = {
1480 .pru_abort = tcp_usr_abort,
1481 .pru_accept = tcp_usr_accept,
1482 .pru_attach = tcp_usr_attach,
1483 .pru_bind = tcp_usr_bind,
1484 .pru_connect = tcp_usr_connect,
1485 .pru_control = in_control,
1486 .pru_detach = tcp_usr_detach,
1487 .pru_disconnect = tcp_usr_disconnect,
1488 .pru_listen = tcp_usr_listen,
1489 .pru_peeraddr = in_getpeeraddr,
1490 .pru_rcvd = tcp_usr_rcvd,
1491 .pru_rcvoob = tcp_usr_rcvoob,
1492 .pru_send = tcp_usr_send,
1493 .pru_ready = tcp_usr_ready,
1494 .pru_shutdown = tcp_usr_shutdown,
1495 .pru_sockaddr = in_getsockaddr,
1496 .pru_sosetlabel = in_pcbsosetlabel,
1497 .pru_close = tcp_usr_close,
1502 struct pr_usrreqs tcp6_usrreqs = {
1503 .pru_abort = tcp_usr_abort,
1504 .pru_accept = tcp6_usr_accept,
1505 .pru_attach = tcp_usr_attach,
1506 .pru_bind = tcp6_usr_bind,
1507 .pru_connect = tcp6_usr_connect,
1508 .pru_control = in6_control,
1509 .pru_detach = tcp_usr_detach,
1510 .pru_disconnect = tcp_usr_disconnect,
1511 .pru_listen = tcp6_usr_listen,
1512 .pru_peeraddr = in6_mapped_peeraddr,
1513 .pru_rcvd = tcp_usr_rcvd,
1514 .pru_rcvoob = tcp_usr_rcvoob,
1515 .pru_send = tcp_usr_send,
1516 .pru_ready = tcp_usr_ready,
1517 .pru_shutdown = tcp_usr_shutdown,
1518 .pru_sockaddr = in6_mapped_sockaddr,
1519 .pru_sosetlabel = in_pcbsosetlabel,
1520 .pru_close = tcp_usr_close,
1526 * Common subroutine to open a TCP connection to remote host specified
1527 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1528 * port number if needed. Call in_pcbconnect_setup to do the routing and
1529 * to choose a local host address (interface). If there is an existing
1530 * incarnation of the same connection in TIME-WAIT state and if the remote
1531 * host was sending CC options and if the connection duration was < MSL, then
1532 * truncate the previous TIME-WAIT state and proceed.
1533 * Initialize connection parameters and enter SYN-SENT state.
1536 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1538 struct inpcb *inp = tp->t_inpcb, *oinp;
1539 struct socket *so = inp->inp_socket;
1540 struct in_addr laddr;
1545 INP_WLOCK_ASSERT(inp);
1546 INP_HASH_WLOCK(&V_tcbinfo);
1548 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1549 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1555 * Cannot simply call in_pcbconnect, because there might be an
1556 * earlier incarnation of this same connection still in
1557 * TIME_WAIT state, creating an ADDRINUSE error.
1559 laddr = inp->inp_laddr;
1560 lport = inp->inp_lport;
1561 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1562 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1563 if (error && oinp == NULL)
1569 /* Handle initial bind if it hadn't been done in advance. */
1570 if (inp->inp_lport == 0) {
1571 inp->inp_lport = lport;
1572 if (in_pcbinshash(inp) != 0) {
1578 inp->inp_laddr = laddr;
1580 INP_HASH_WUNLOCK(&V_tcbinfo);
1583 * Compute window scaling to request:
1584 * Scale to fit into sweet spot. See tcp_syncache.c.
1585 * XXX: This should move to tcp_output().
1587 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1588 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1589 tp->request_r_scale++;
1592 TCPSTAT_INC(tcps_connattempt);
1593 tcp_state_change(tp, TCPS_SYN_SENT);
1594 tp->iss = tcp_new_isn(&inp->inp_inc);
1595 if (tp->t_flags & TF_REQ_TSTMP)
1596 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1597 tcp_sendseqinit(tp);
1602 INP_HASH_WUNLOCK(&V_tcbinfo);
1609 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1611 struct inpcb *inp = tp->t_inpcb;
1614 INP_WLOCK_ASSERT(inp);
1615 INP_HASH_WLOCK(&V_tcbinfo);
1617 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1618 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1622 error = in6_pcbconnect(inp, nam, td->td_ucred);
1625 INP_HASH_WUNLOCK(&V_tcbinfo);
1627 /* Compute window scaling to request. */
1628 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1629 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1630 tp->request_r_scale++;
1632 soisconnecting(inp->inp_socket);
1633 TCPSTAT_INC(tcps_connattempt);
1634 tcp_state_change(tp, TCPS_SYN_SENT);
1635 tp->iss = tcp_new_isn(&inp->inp_inc);
1636 if (tp->t_flags & TF_REQ_TSTMP)
1637 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1638 tcp_sendseqinit(tp);
1643 INP_HASH_WUNLOCK(&V_tcbinfo);
1649 * Export TCP internal state information via a struct tcp_info, based on the
1650 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1651 * (TCP state machine, etc). We export all information using FreeBSD-native
1652 * constants -- for example, the numeric values for tcpi_state will differ
1656 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1659 INP_WLOCK_ASSERT(tp->t_inpcb);
1660 bzero(ti, sizeof(*ti));
1662 ti->tcpi_state = tp->t_state;
1663 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1664 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1665 if (tp->t_flags & TF_SACK_PERMIT)
1666 ti->tcpi_options |= TCPI_OPT_SACK;
1667 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1668 ti->tcpi_options |= TCPI_OPT_WSCALE;
1669 ti->tcpi_snd_wscale = tp->snd_scale;
1670 ti->tcpi_rcv_wscale = tp->rcv_scale;
1672 if (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT))
1673 ti->tcpi_options |= TCPI_OPT_ECN;
1675 ti->tcpi_rto = tp->t_rxtcur * tick;
1676 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1677 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1678 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1680 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1681 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1684 * FreeBSD-specific extension fields for tcp_info.
1686 ti->tcpi_rcv_space = tp->rcv_wnd;
1687 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1688 ti->tcpi_snd_wnd = tp->snd_wnd;
1689 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1690 ti->tcpi_snd_nxt = tp->snd_nxt;
1691 ti->tcpi_snd_mss = tp->t_maxseg;
1692 ti->tcpi_rcv_mss = tp->t_maxseg;
1693 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1694 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1695 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1697 if (tp->t_flags & TF_TOE) {
1698 ti->tcpi_options |= TCPI_OPT_TOE;
1699 tcp_offload_tcp_info(tp, ti);
1705 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1706 * socket option arguments. When it re-acquires the lock after the copy, it
1707 * has to revalidate that the connection is still valid for the socket
1710 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
1712 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1715 return (ECONNRESET); \
1717 tp = intotcpcb(inp); \
1719 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1722 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
1724 struct socket *so = inp->inp_socket;
1725 struct tcpcb *tp = intotcpcb(inp);
1728 MPASS(sopt->sopt_dir == SOPT_SET);
1729 INP_WLOCK_ASSERT(inp);
1730 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1731 ("inp_flags == %x", inp->inp_flags));
1732 KASSERT(so != NULL, ("inp_socket == NULL"));
1734 if (sopt->sopt_level != IPPROTO_TCP) {
1737 if (inp->inp_vflag & INP_IPV6PROTO)
1738 error = ip6_ctloutput(so, sopt);
1740 #if defined(INET6) && defined(INET)
1744 error = ip_ctloutput(so, sopt);
1747 * When an IP-level socket option affects TCP, pass control
1748 * down to stack tfb_tcp_ctloutput, otherwise return what
1749 * IP level returned.
1751 switch (sopt->sopt_level) {
1754 if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
1756 switch (sopt->sopt_name) {
1758 /* Notify tcp stacks that care (e.g. RACK). */
1760 case IPV6_USE_MIN_MTU:
1761 /* Update t_maxseg accordingly. */
1770 switch (sopt->sopt_name) {
1772 inp->inp_ip_tos &= ~IPTOS_ECN_MASK;
1775 /* Notify tcp stacks that care (e.g. RACK). */
1786 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1788 return (ECONNRESET);
1790 } else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
1792 * Protect the TCP option TCP_FUNCTION_BLK so
1793 * that a sub-function can *never* overwrite this.
1795 struct tcp_function_set fsn;
1796 struct tcp_function_block *blk;
1799 error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
1804 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1806 return (ECONNRESET);
1808 tp = intotcpcb(inp);
1810 blk = find_and_ref_tcp_functions(&fsn);
1815 if (tp->t_fb == blk) {
1816 /* You already have this */
1817 refcount_release(&blk->tfb_refcnt);
1821 if (tp->t_state != TCPS_CLOSED) {
1823 * The user has advanced the state
1824 * past the initial point, we may not
1825 * be able to switch.
1827 if (blk->tfb_tcp_handoff_ok != NULL) {
1829 * Does the stack provide a
1830 * query mechanism, if so it may
1831 * still be possible?
1833 error = (*blk->tfb_tcp_handoff_ok)(tp);
1837 refcount_release(&blk->tfb_refcnt);
1842 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1843 refcount_release(&blk->tfb_refcnt);
1848 * Release the old refcnt, the
1849 * lookup acquired a ref on the
1852 if (tp->t_fb->tfb_tcp_fb_fini) {
1853 struct epoch_tracker et;
1855 * Tell the stack to cleanup with 0 i.e.
1856 * the tcb is not going away.
1858 NET_EPOCH_ENTER(et);
1859 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1863 /* Assure that we are not on any hpts */
1864 tcp_hpts_remove(tp->t_inpcb);
1866 if (blk->tfb_tcp_fb_init) {
1867 error = (*blk->tfb_tcp_fb_init)(tp);
1869 refcount_release(&blk->tfb_refcnt);
1870 if (tp->t_fb->tfb_tcp_fb_init) {
1871 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) {
1872 /* Fall back failed, drop the connection */
1881 refcount_release(&tp->t_fb->tfb_refcnt);
1884 if (tp->t_flags & TF_TOE) {
1885 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1894 /* Pass in the INP locked, callee must unlock it. */
1895 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1899 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
1901 struct socket *so = inp->inp_socket;
1902 struct tcpcb *tp = intotcpcb(inp);
1905 MPASS(sopt->sopt_dir == SOPT_GET);
1906 INP_WLOCK_ASSERT(inp);
1907 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1908 ("inp_flags == %x", inp->inp_flags));
1909 KASSERT(so != NULL, ("inp_socket == NULL"));
1911 if (sopt->sopt_level != IPPROTO_TCP) {
1914 if (inp->inp_vflag & INP_IPV6PROTO)
1915 error = ip6_ctloutput(so, sopt);
1917 #if defined(INET6) && defined(INET)
1921 error = ip_ctloutput(so, sopt);
1925 if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
1926 (sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
1927 struct tcp_function_set fsn;
1929 if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
1930 memset(&fsn, 0, sizeof(fsn));
1931 find_tcp_function_alias(tp->t_fb, &fsn);
1933 strncpy(fsn.function_set_name,
1934 tp->t_fb->tfb_tcp_block_name,
1935 TCP_FUNCTION_NAME_LEN_MAX);
1936 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1938 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1940 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1944 /* Pass in the INP locked, callee must unlock it. */
1945 return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1949 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1953 inp = sotoinpcb(so);
1954 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1957 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1959 return (ECONNRESET);
1961 if (sopt->sopt_dir == SOPT_SET)
1962 return (tcp_ctloutput_set(inp, sopt));
1963 else if (sopt->sopt_dir == SOPT_GET)
1964 return (tcp_ctloutput_get(inp, sopt));
1966 panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
1970 * If this assert becomes untrue, we need to change the size of the buf
1971 * variable in tcp_default_ctloutput().
1974 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1975 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1980 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1982 struct tls_enable_v0 tls_v0;
1985 if (sopt->sopt_valsize == sizeof(tls_v0)) {
1986 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1990 memset(tls, 0, sizeof(*tls));
1991 tls->cipher_key = tls_v0.cipher_key;
1992 tls->iv = tls_v0.iv;
1993 tls->auth_key = tls_v0.auth_key;
1994 tls->cipher_algorithm = tls_v0.cipher_algorithm;
1995 tls->cipher_key_len = tls_v0.cipher_key_len;
1996 tls->iv_len = tls_v0.iv_len;
1997 tls->auth_algorithm = tls_v0.auth_algorithm;
1998 tls->auth_key_len = tls_v0.auth_key_len;
1999 tls->flags = tls_v0.flags;
2000 tls->tls_vmajor = tls_v0.tls_vmajor;
2001 tls->tls_vminor = tls_v0.tls_vminor;
2005 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
2009 extern struct cc_algo newreno_cc_algo;
2012 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt)
2014 struct cc_algo *algo;
2017 struct cc_var cc_mem;
2018 char buf[TCP_CA_NAME_MAX];
2023 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2026 buf[sopt->sopt_valsize] = '\0';
2028 STAILQ_FOREACH(algo, &cc_list, entries) {
2029 if (strncmp(buf, algo->name,
2030 TCP_CA_NAME_MAX) == 0) {
2031 if (algo->flags & CC_MODULE_BEING_REMOVED) {
2032 /* We can't "see" modules being unloaded */
2043 * With a reference the algorithm cannot be removed
2044 * so we hold a reference through the change process.
2048 if (algo->cb_init != NULL) {
2049 /* We can now pre-get the memory for the CC */
2050 mem_sz = (*algo->cc_data_sz)();
2054 ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
2061 * Make sure its all clean and zero and also get
2064 memset(&cc_mem, 0, sizeof(cc_mem));
2066 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2069 free(ptr, M_CC_MEM);
2070 /* Release our temp reference */
2074 return (ECONNRESET);
2076 tp = intotcpcb(inp);
2078 memset(ptr, 0, mem_sz);
2079 cc_mem.ccvc.tcp = tp;
2081 * We once again hold a write lock over the tcb so it's
2082 * safe to do these things without ordering concerns.
2083 * Note here we init into stack memory.
2085 if (algo->cb_init != NULL)
2086 error = algo->cb_init(&cc_mem, ptr);
2090 * The CC algorithms, when given their memory
2091 * should not fail we could in theory have a
2096 * Touchdown, lets go ahead and move the
2097 * connection to the new CC module by
2098 * copying in the cc_mem after we call
2099 * the old ones cleanup (if any).
2101 if (CC_ALGO(tp)->cb_destroy != NULL)
2102 CC_ALGO(tp)->cb_destroy(tp->ccv);
2103 /* Detach the old CC from the tcpcb */
2105 /* Copy in our temp memory that was inited */
2106 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2107 /* Now attach the new, which takes a reference */
2108 cc_attach(tp, algo);
2109 /* Ok now are we where we have gotten past any conn_init? */
2110 if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
2111 /* Yep run the connection init for the new CC */
2112 CC_ALGO(tp)->conn_init(tp->ccv);
2115 free(ptr, M_CC_MEM);
2117 /* Now lets release our temp reference */
2125 tcp_default_ctloutput(struct inpcb *inp, struct sockopt *sopt)
2127 struct tcpcb *tp = intotcpcb(inp);
2128 int error, opt, optval;
2132 struct tls_enable tls;
2133 struct socket *so = inp->inp_socket;
2135 char *pbuf, buf[TCP_LOG_ID_LEN];
2137 struct statsblob *sbp;
2141 INP_WLOCK_ASSERT(inp);
2142 KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
2143 ("inp_flags == %x", inp->inp_flags));
2144 KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL"));
2146 switch (sopt->sopt_level) {
2149 MPASS(inp->inp_vflag & INP_IPV6PROTO);
2150 switch (sopt->sopt_name) {
2151 case IPV6_USE_MIN_MTU:
2152 tcp6_use_min_mtu(tp);
2166 * For TCP_CCALGOOPT forward the control to CC module, for both
2167 * SOPT_SET and SOPT_GET.
2169 switch (sopt->sopt_name) {
2172 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2174 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2175 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2176 sopt->sopt_valsize);
2181 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2182 if (CC_ALGO(tp)->ctl_output != NULL)
2183 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
2187 if (error == 0 && sopt->sopt_dir == SOPT_GET)
2188 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2193 switch (sopt->sopt_dir) {
2195 switch (sopt->sopt_name) {
2196 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2199 if (!TCPMD5_ENABLED())
2200 return (ENOPROTOOPT);
2201 error = TCPMD5_PCBCTL(inp, sopt);
2204 INP_WLOCK_RECHECK(inp);
2205 goto unlock_and_done;
2212 error = sooptcopyin(sopt, &optval, sizeof optval,
2217 INP_WLOCK_RECHECK(inp);
2218 switch (sopt->sopt_name) {
2229 opt = 0; /* dead code to fool gcc */
2236 tp->t_flags &= ~opt;
2239 if (tp->t_flags & TF_TOE) {
2240 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2249 error = sooptcopyin(sopt, &optval, sizeof optval,
2254 INP_WLOCK_RECHECK(inp);
2256 tp->t_flags |= TF_NOPUSH;
2257 else if (tp->t_flags & TF_NOPUSH) {
2258 tp->t_flags &= ~TF_NOPUSH;
2259 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2260 struct epoch_tracker et;
2262 NET_EPOCH_ENTER(et);
2263 error = tcp_output_nodrop(tp);
2267 goto unlock_and_done;
2269 case TCP_REMOTE_UDP_ENCAPS_PORT:
2271 error = sooptcopyin(sopt, &optval, sizeof optval,
2275 if ((optval < TCP_TUNNELING_PORT_MIN) ||
2276 (optval > TCP_TUNNELING_PORT_MAX)) {
2277 /* Its got to be in range */
2280 if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2281 /* You have to have enabled a UDP tunneling port first */
2284 INP_WLOCK_RECHECK(inp);
2285 if (tp->t_state != TCPS_CLOSED) {
2286 /* You can't change after you are connected */
2289 /* Ok we are all good set the port */
2290 tp->t_port = htons(optval);
2292 goto unlock_and_done;
2296 error = sooptcopyin(sopt, &optval, sizeof optval,
2301 INP_WLOCK_RECHECK(inp);
2302 if (optval > 0 && optval <= tp->t_maxseg &&
2303 optval + 40 >= V_tcp_minmss)
2304 tp->t_maxseg = optval;
2307 goto unlock_and_done;
2317 error = sooptcopyin(sopt, &optval, sizeof optval,
2323 sbp = stats_blob_alloc(
2324 V_tcp_perconn_stats_dflt_tpl, 0);
2328 INP_WLOCK_RECHECK(inp);
2329 if ((tp->t_stats != NULL && sbp == NULL) ||
2330 (tp->t_stats == NULL && sbp != NULL)) {
2331 struct statsblob *t = tp->t_stats;
2337 stats_blob_destroy(sbp);
2339 return (EOPNOTSUPP);
2343 case TCP_CONGESTION:
2344 error = tcp_set_cc_mod(inp, sopt);
2347 case TCP_REUSPORT_LB_NUMA:
2349 error = sooptcopyin(sopt, &optval, sizeof(optval),
2351 INP_WLOCK_RECHECK(inp);
2353 error = in_pcblbgroup_numa(inp, optval);
2358 case TCP_TXTLS_ENABLE:
2360 error = copyin_tls_enable(sopt, &tls);
2363 error = ktls_enable_tx(so, &tls);
2365 case TCP_TXTLS_MODE:
2367 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2371 INP_WLOCK_RECHECK(inp);
2372 error = ktls_set_tx_mode(so, ui);
2375 case TCP_RXTLS_ENABLE:
2377 error = sooptcopyin(sopt, &tls, sizeof(tls),
2381 error = ktls_enable_rx(so, &tls);
2389 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2393 if (ui > (UINT_MAX / hz)) {
2399 INP_WLOCK_RECHECK(inp);
2400 switch (sopt->sopt_name) {
2402 tp->t_keepidle = ui;
2404 * XXX: better check current remaining
2405 * timeout and "merge" it with new value.
2407 if ((tp->t_state > TCPS_LISTEN) &&
2408 (tp->t_state <= TCPS_CLOSING))
2409 tcp_timer_activate(tp, TT_KEEP,
2413 tp->t_keepintvl = ui;
2414 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2415 (TP_MAXIDLE(tp) > 0))
2416 tcp_timer_activate(tp, TT_2MSL,
2420 tp->t_keepinit = ui;
2421 if (tp->t_state == TCPS_SYN_RECEIVED ||
2422 tp->t_state == TCPS_SYN_SENT)
2423 tcp_timer_activate(tp, TT_KEEP,
2427 goto unlock_and_done;
2431 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2435 INP_WLOCK_RECHECK(inp);
2437 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2438 (TP_MAXIDLE(tp) > 0))
2439 tcp_timer_activate(tp, TT_2MSL,
2441 goto unlock_and_done;
2447 error = sooptcopyin(sopt, &optval, sizeof optval,
2452 INP_WLOCK_RECHECK(inp);
2454 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2455 &(tp->t_outpkts) : &(tp->t_inpkts),
2459 goto unlock_and_done;
2462 case TCP_FASTOPEN: {
2463 struct tcp_fastopen tfo_optval;
2466 if (!V_tcp_fastopen_client_enable &&
2467 !V_tcp_fastopen_server_enable)
2470 error = sooptcopyin(sopt, &tfo_optval,
2471 sizeof(tfo_optval), sizeof(int));
2475 INP_WLOCK_RECHECK(inp);
2476 if ((tp->t_state != TCPS_CLOSED) &&
2477 (tp->t_state != TCPS_LISTEN)) {
2479 goto unlock_and_done;
2481 if (tfo_optval.enable) {
2482 if (tp->t_state == TCPS_LISTEN) {
2483 if (!V_tcp_fastopen_server_enable) {
2485 goto unlock_and_done;
2488 if (tp->t_tfo_pending == NULL)
2490 tcp_fastopen_alloc_counter();
2493 * If a pre-shared key was provided,
2494 * stash it in the client cookie
2495 * field of the tcpcb for use during
2498 if (sopt->sopt_valsize ==
2499 sizeof(tfo_optval)) {
2500 memcpy(tp->t_tfo_cookie.client,
2502 TCP_FASTOPEN_PSK_LEN);
2503 tp->t_tfo_client_cookie_len =
2504 TCP_FASTOPEN_PSK_LEN;
2507 tp->t_flags |= TF_FASTOPEN;
2509 tp->t_flags &= ~TF_FASTOPEN;
2510 goto unlock_and_done;
2516 error = sooptcopyin(sopt, &optval, sizeof optval,
2521 INP_WLOCK_RECHECK(inp);
2522 error = tcp_log_state_change(tp, optval);
2523 goto unlock_and_done;
2532 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2535 buf[sopt->sopt_valsize] = '\0';
2536 INP_WLOCK_RECHECK(inp);
2537 error = tcp_log_set_id(tp, buf);
2538 /* tcp_log_set_id() unlocks the INP. */
2545 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2548 buf[sopt->sopt_valsize] = '\0';
2549 INP_WLOCK_RECHECK(inp);
2550 if (sopt->sopt_name == TCP_LOGDUMP) {
2551 error = tcp_log_dump_tp_logbuf(tp, buf,
2555 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2557 * tcp_log_dump_tp_bucket_logbufs() drops the
2566 error = ENOPROTOOPT;
2572 tp = intotcpcb(inp);
2573 switch (sopt->sopt_name) {
2574 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2577 if (!TCPMD5_ENABLED())
2578 return (ENOPROTOOPT);
2579 error = TCPMD5_PCBCTL(inp, sopt);
2584 optval = tp->t_flags & TF_NODELAY;
2586 error = sooptcopyout(sopt, &optval, sizeof optval);
2589 optval = tp->t_maxseg;
2591 error = sooptcopyout(sopt, &optval, sizeof optval);
2593 case TCP_REMOTE_UDP_ENCAPS_PORT:
2594 optval = ntohs(tp->t_port);
2596 error = sooptcopyout(sopt, &optval, sizeof optval);
2599 optval = tp->t_flags & TF_NOOPT;
2601 error = sooptcopyout(sopt, &optval, sizeof optval);
2604 optval = tp->t_flags & TF_NOPUSH;
2606 error = sooptcopyout(sopt, &optval, sizeof optval);
2609 tcp_fill_info(tp, &ti);
2611 error = sooptcopyout(sopt, &ti, sizeof ti);
2617 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2620 socklen_t outsbsz = sopt->sopt_valsize;
2621 if (tp->t_stats == NULL)
2623 else if (outsbsz >= tp->t_stats->cursz)
2624 outsbsz = tp->t_stats->cursz;
2625 else if (outsbsz >= sizeof(struct statsblob))
2626 outsbsz = sizeof(struct statsblob);
2633 sbp = sopt->sopt_val;
2634 nheld = atop(round_page(((vm_offset_t)sbp) +
2635 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2636 vm_page_t ma[nheld];
2637 if (vm_fault_quick_hold_pages(
2638 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2639 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2645 if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2646 SIZEOF_MEMBER(struct statsblob, flags))))
2649 INP_WLOCK_RECHECK(inp);
2650 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2651 sbflags | SB_CLONE_USRDSTNOFAULT);
2653 sopt->sopt_valsize = outsbsz;
2655 vm_page_unhold_pages(ma, nheld);
2662 case TCP_CONGESTION:
2663 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2665 error = sooptcopyout(sopt, buf, len + 1);
2671 switch (sopt->sopt_name) {
2673 ui = TP_KEEPIDLE(tp) / hz;
2676 ui = TP_KEEPINTVL(tp) / hz;
2679 ui = TP_KEEPINIT(tp) / hz;
2682 ui = TP_KEEPCNT(tp);
2686 error = sooptcopyout(sopt, &ui, sizeof(ui));
2691 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2692 &(tp->t_outpkts) : &(tp->t_inpkts));
2694 error = sooptcopyout(sopt, &optval, sizeof optval);
2698 optval = tp->t_flags & TF_FASTOPEN;
2700 error = sooptcopyout(sopt, &optval, sizeof optval);
2704 optval = tp->t_logstate;
2706 error = sooptcopyout(sopt, &optval, sizeof(optval));
2709 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2710 error = tcp_log_getlogbuf(sopt, tp);
2713 len = tcp_log_get_id(tp, buf);
2715 error = sooptcopyout(sopt, buf, len + 1);
2724 case TCP_TXTLS_MODE:
2725 error = ktls_get_tx_mode(so, &optval);
2728 error = sooptcopyout(sopt, &optval,
2731 case TCP_RXTLS_MODE:
2732 error = ktls_get_rx_mode(so, &optval);
2735 error = sooptcopyout(sopt, &optval,
2740 optval = tp->t_flags & TF_LRD;
2742 error = sooptcopyout(sopt, &optval, sizeof optval);
2746 error = ENOPROTOOPT;
2753 #undef INP_WLOCK_RECHECK
2754 #undef INP_WLOCK_RECHECK_CLEANUP
2757 * Initiate (or continue) disconnect.
2758 * If embryonic state, just send reset (once).
2759 * If in ``let data drain'' option and linger null, just drop.
2760 * Otherwise (hard), mark socket disconnecting and drop
2761 * current input data; switch states based on user close, and
2762 * send segment to peer (with FIN).
2765 tcp_disconnect(struct tcpcb *tp)
2767 struct inpcb *inp = tp->t_inpcb;
2768 struct socket *so = inp->inp_socket;
2771 INP_WLOCK_ASSERT(inp);
2774 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2775 * socket is still open.
2777 if (tp->t_state < TCPS_ESTABLISHED &&
2778 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2781 ("tcp_disconnect: tcp_close() returned NULL"));
2782 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2783 tp = tcp_drop(tp, 0);
2785 ("tcp_disconnect: tcp_drop() returned NULL"));
2787 soisdisconnecting(so);
2788 sbflush(&so->so_rcv);
2790 if (!(inp->inp_flags & INP_DROPPED))
2791 /* Ignore stack's drop request, we already at it. */
2792 (void)tcp_output_nodrop(tp);
2797 * User issued close, and wish to trail through shutdown states:
2798 * if never received SYN, just forget it. If got a SYN from peer,
2799 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2800 * If already got a FIN from peer, then almost done; go to LAST_ACK
2801 * state. In all other cases, have already sent FIN to peer (e.g.
2802 * after PRU_SHUTDOWN), and just have to play tedious game waiting
2803 * for peer to send FIN or not respond to keep-alives, etc.
2804 * We can let the user exit from the close as soon as the FIN is acked.
2807 tcp_usrclosed(struct tcpcb *tp)
2811 INP_WLOCK_ASSERT(tp->t_inpcb);
2813 switch (tp->t_state) {
2816 tcp_offload_listen_stop(tp);
2818 tcp_state_change(tp, TCPS_CLOSED);
2823 * tcp_close() should never return NULL here as the socket is
2827 ("tcp_usrclosed: tcp_close() returned NULL"));
2831 case TCPS_SYN_RECEIVED:
2832 tp->t_flags |= TF_NEEDFIN;
2835 case TCPS_ESTABLISHED:
2836 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2839 case TCPS_CLOSE_WAIT:
2840 tcp_state_change(tp, TCPS_LAST_ACK);
2843 if (tp->t_state >= TCPS_FIN_WAIT_2) {
2844 soisdisconnected(tp->t_inpcb->inp_socket);
2845 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2846 if (tp->t_state == TCPS_FIN_WAIT_2) {
2849 timeout = (tcp_fast_finwait2_recycle) ?
2850 tcp_finwait2_timeout : TP_MAXIDLE(tp);
2851 tcp_timer_activate(tp, TT_2MSL, timeout);
2858 db_print_indent(int indent)
2862 for (i = 0; i < indent; i++)
2867 db_print_tstate(int t_state)
2872 db_printf("TCPS_CLOSED");
2876 db_printf("TCPS_LISTEN");
2880 db_printf("TCPS_SYN_SENT");
2883 case TCPS_SYN_RECEIVED:
2884 db_printf("TCPS_SYN_RECEIVED");
2887 case TCPS_ESTABLISHED:
2888 db_printf("TCPS_ESTABLISHED");
2891 case TCPS_CLOSE_WAIT:
2892 db_printf("TCPS_CLOSE_WAIT");
2895 case TCPS_FIN_WAIT_1:
2896 db_printf("TCPS_FIN_WAIT_1");
2900 db_printf("TCPS_CLOSING");
2904 db_printf("TCPS_LAST_ACK");
2907 case TCPS_FIN_WAIT_2:
2908 db_printf("TCPS_FIN_WAIT_2");
2911 case TCPS_TIME_WAIT:
2912 db_printf("TCPS_TIME_WAIT");
2916 db_printf("unknown");
2922 db_print_tflags(u_int t_flags)
2927 if (t_flags & TF_ACKNOW) {
2928 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2931 if (t_flags & TF_DELACK) {
2932 db_printf("%sTF_DELACK", comma ? ", " : "");
2935 if (t_flags & TF_NODELAY) {
2936 db_printf("%sTF_NODELAY", comma ? ", " : "");
2939 if (t_flags & TF_NOOPT) {
2940 db_printf("%sTF_NOOPT", comma ? ", " : "");
2943 if (t_flags & TF_SENTFIN) {
2944 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2947 if (t_flags & TF_REQ_SCALE) {
2948 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2951 if (t_flags & TF_RCVD_SCALE) {
2952 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2955 if (t_flags & TF_REQ_TSTMP) {
2956 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2959 if (t_flags & TF_RCVD_TSTMP) {
2960 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2963 if (t_flags & TF_SACK_PERMIT) {
2964 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2967 if (t_flags & TF_NEEDSYN) {
2968 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2971 if (t_flags & TF_NEEDFIN) {
2972 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2975 if (t_flags & TF_NOPUSH) {
2976 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2979 if (t_flags & TF_PREVVALID) {
2980 db_printf("%sTF_PREVVALID", comma ? ", " : "");
2983 if (t_flags & TF_MORETOCOME) {
2984 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2987 if (t_flags & TF_LQ_OVERFLOW) {
2988 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2991 if (t_flags & TF_LASTIDLE) {
2992 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2995 if (t_flags & TF_RXWIN0SENT) {
2996 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2999 if (t_flags & TF_FASTRECOVERY) {
3000 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
3003 if (t_flags & TF_CONGRECOVERY) {
3004 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
3007 if (t_flags & TF_WASFRECOVERY) {
3008 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
3011 if (t_flags & TF_WASCRECOVERY) {
3012 db_printf("%sTF_WASCRECOVERY", comma ? ", " : "");
3015 if (t_flags & TF_SIGNATURE) {
3016 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
3019 if (t_flags & TF_FORCEDATA) {
3020 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
3023 if (t_flags & TF_TSO) {
3024 db_printf("%sTF_TSO", comma ? ", " : "");
3027 if (t_flags & TF_FASTOPEN) {
3028 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
3034 db_print_tflags2(u_int t_flags2)
3039 if (t_flags2 & TF2_PLPMTU_BLACKHOLE) {
3040 db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : "");
3043 if (t_flags2 & TF2_PLPMTU_PMTUD) {
3044 db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : "");
3047 if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) {
3048 db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : "");
3051 if (t_flags2 & TF2_LOG_AUTO) {
3052 db_printf("%sTF2_LOG_AUTO", comma ? ", " : "");
3055 if (t_flags2 & TF2_DROP_AF_DATA) {
3056 db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : "");
3059 if (t_flags2 & TF2_ECN_PERMIT) {
3060 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
3063 if (t_flags2 & TF2_ECN_SND_CWR) {
3064 db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : "");
3067 if (t_flags2 & TF2_ECN_SND_ECE) {
3068 db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : "");
3071 if (t_flags2 & TF2_ACE_PERMIT) {
3072 db_printf("%sTF2_ACE_PERMIT", comma ? ", " : "");
3075 if (t_flags2 & TF2_FBYTES_COMPLETE) {
3076 db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : "");
3082 db_print_toobflags(char t_oobflags)
3087 if (t_oobflags & TCPOOB_HAVEDATA) {
3088 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
3091 if (t_oobflags & TCPOOB_HADDATA) {
3092 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
3098 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
3101 db_print_indent(indent);
3102 db_printf("%s at %p\n", name, tp);
3106 db_print_indent(indent);
3107 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
3108 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
3110 db_print_indent(indent);
3111 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
3112 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
3114 db_print_indent(indent);
3115 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
3116 &tp->t_timers->tt_delack, tp->t_inpcb);
3118 db_print_indent(indent);
3119 db_printf("t_state: %d (", tp->t_state);
3120 db_print_tstate(tp->t_state);
3123 db_print_indent(indent);
3124 db_printf("t_flags: 0x%x (", tp->t_flags);
3125 db_print_tflags(tp->t_flags);
3128 db_print_indent(indent);
3129 db_printf("t_flags2: 0x%x (", tp->t_flags2);
3130 db_print_tflags2(tp->t_flags2);
3133 db_print_indent(indent);
3134 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
3135 tp->snd_una, tp->snd_max, tp->snd_nxt);
3137 db_print_indent(indent);
3138 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
3139 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
3141 db_print_indent(indent);
3142 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
3143 tp->iss, tp->irs, tp->rcv_nxt);
3145 db_print_indent(indent);
3146 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
3147 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
3149 db_print_indent(indent);
3150 db_printf("snd_wnd: %u snd_cwnd: %u\n",
3151 tp->snd_wnd, tp->snd_cwnd);
3153 db_print_indent(indent);
3154 db_printf("snd_ssthresh: %u snd_recover: "
3155 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
3157 db_print_indent(indent);
3158 db_printf("t_rcvtime: %u t_startime: %u\n",
3159 tp->t_rcvtime, tp->t_starttime);
3161 db_print_indent(indent);
3162 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
3163 tp->t_rtttime, tp->t_rtseq);
3165 db_print_indent(indent);
3166 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
3167 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
3169 db_print_indent(indent);
3170 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
3171 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
3174 db_print_indent(indent);
3175 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n",
3176 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
3178 db_print_indent(indent);
3179 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
3180 db_print_toobflags(tp->t_oobflags);
3181 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
3183 db_print_indent(indent);
3184 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
3185 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
3187 db_print_indent(indent);
3188 db_printf("ts_recent: %u ts_recent_age: %u\n",
3189 tp->ts_recent, tp->ts_recent_age);
3191 db_print_indent(indent);
3192 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
3193 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
3195 db_print_indent(indent);
3196 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
3197 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
3198 tp->snd_recover_prev, tp->t_badrxtwin);
3200 db_print_indent(indent);
3201 db_printf("snd_numholes: %d snd_holes first: %p\n",
3202 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
3204 db_print_indent(indent);
3205 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n",
3206 tp->snd_fack, tp->rcv_numsacks);
3208 /* Skip sackblks, sackhint. */
3210 db_print_indent(indent);
3211 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
3212 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
3215 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
3220 db_printf("usage: show tcpcb <addr>\n");
3223 tp = (struct tcpcb *)addr;
3225 db_print_tcpcb(tp, "tcpcb", 0);