xref: /freebsd-13.1/sys/netinet/tcp_usrreq.c (revision 00a4311a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
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.
8  * All rights reserved.
9  *
10  * Portions of this software were developed by Robert N. M. Watson under
11  * contract to Juniper Networks, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
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.
24  *
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
35  * SUCH DAMAGE.
36  *
37  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_ddb.h"
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_kern_tls.h"
48 #include "opt_tcpdebug.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/arb.h>
53 #include <sys/limits.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
56 #include <sys/kernel.h>
57 #include <sys/ktls.h>
58 #include <sys/qmath.h>
59 #include <sys/sysctl.h>
60 #include <sys/mbuf.h>
61 #ifdef INET6
62 #include <sys/domain.h>
63 #endif /* INET6 */
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/proc.h>
68 #include <sys/jail.h>
69 #include <sys/syslog.h>
70 #include <sys/stats.h>
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #include <net/if.h>
77 #include <net/if_var.h>
78 #include <net/route.h>
79 #include <net/vnet.h>
80 
81 #include <netinet/in.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip_var.h>
87 #ifdef INET6
88 #include <netinet/ip6.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #endif
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>
103 #ifdef TCPPCAP
104 #include <netinet/tcp_pcap.h>
105 #endif
106 #ifdef TCPDEBUG
107 #include <netinet/tcp_debug.h>
108 #endif
109 #ifdef TCP_OFFLOAD
110 #include <netinet/tcp_offload.h>
111 #endif
112 #include <netipsec/ipsec_support.h>
113 
114 #include <vm/vm.h>
115 #include <vm/vm_param.h>
116 #include <vm/pmap.h>
117 #include <vm/vm_extern.h>
118 #include <vm/vm_map.h>
119 #include <vm/vm_page.h>
120 
121 /*
122  * TCP protocol interface to socket abstraction.
123  */
124 #ifdef INET
125 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
126 		    struct thread *td);
127 #endif /* INET */
128 #ifdef INET6
129 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
130 		    struct thread *td);
131 #endif /* INET6 */
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 *);
135 
136 static int	tcp_pru_options_support(struct tcpcb *tp, int flags);
137 
138 #ifdef TCPDEBUG
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)
143 #else
144 #define	TCPDEBUG0
145 #define	TCPDEBUG1()
146 #define	TCPDEBUG2(req)
147 #endif
148 
149 /*
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.
152  */
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)
158 
159 /*
160  * TCP attaches to socket via pru_attach(), reserving space,
161  * and an internet control block.
162  */
163 static int
164 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
165 {
166 	struct inpcb *inp;
167 	struct tcpcb *tp = NULL;
168 	int error;
169 	TCPDEBUG0;
170 
171 	inp = sotoinpcb(so);
172 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
173 	TCPDEBUG1();
174 
175 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
176 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
177 		if (error)
178 			goto out;
179 	}
180 
181 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
182 	so->so_snd.sb_flags |= SB_AUTOSIZE;
183 	error = in_pcballoc(so, &V_tcbinfo);
184 	if (error)
185 		goto out;
186 	inp = sotoinpcb(so);
187 #ifdef INET6
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 */
193 	}
194 	else
195 #endif
196 		inp->inp_vflag |= INP_IPV4;
197 	tp = tcp_newtcpcb(inp);
198 	if (tp == NULL) {
199 		error = ENOBUFS;
200 		in_pcbdetach(inp);
201 		in_pcbfree(inp);
202 		goto out;
203 	}
204 	tp->t_state = TCPS_CLOSED;
205 	INP_WUNLOCK(inp);
206 	TCPSTATES_INC(TCPS_CLOSED);
207 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
208 		so->so_linger = TCP_LINGERTIME;
209 out:
210 	TCPDEBUG2(PRU_ATTACH);
211 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
212 	return (error);
213 }
214 
215 /*
216  * tcp_usr_detach is called when the socket layer loses its final reference
217  * to the socket, be it a file descriptor reference, a reference from TCP,
218  * etc.  At this point, there is only one case in which we will keep around
219  * inpcb state: time wait.
220  */
221 static void
222 tcp_usr_detach(struct socket *so)
223 {
224 	struct inpcb *inp;
225 	struct tcpcb *tp;
226 
227 	inp = sotoinpcb(so);
228 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
229 	INP_WLOCK(inp);
230 	KASSERT(so->so_pcb == inp && inp->inp_socket == so,
231 		("%s: socket %p inp %p mismatch", __func__, so, inp));
232 
233 	tp = intotcpcb(inp);
234 
235 	if (inp->inp_flags & INP_TIMEWAIT) {
236 		/*
237 		 * There are two cases to handle: one in which the time wait
238 		 * state is being discarded (INP_DROPPED), and one in which
239 		 * this connection will remain in timewait.  In the former,
240 		 * it is time to discard all state (except tcptw, which has
241 		 * already been discarded by the timewait close code, which
242 		 * should be further up the call stack somewhere).  In the
243 		 * latter case, we detach from the socket, but leave the pcb
244 		 * present until timewait ends.
245 		 *
246 		 * XXXRW: Would it be cleaner to free the tcptw here?
247 		 *
248 		 * Astute question indeed, from twtcp perspective there are
249 		 * four cases to consider:
250 		 *
251 		 * #1 tcp_usr_detach is called at tcptw creation time by
252 		 *  tcp_twstart, then do not discard the newly created tcptw
253 		 *  and leave inpcb present until timewait ends
254 		 * #2 tcp_usr_detach is called at tcptw creation time by
255 		 *  tcp_twstart, but connection is local and tw will be
256 		 *  discarded immediately
257 		 * #3 tcp_usr_detach is called at timewait end (or reuse) by
258 		 *  tcp_twclose, then the tcptw has already been discarded
259 		 *  (or reused) and inpcb is freed here
260 		 * #4 tcp_usr_detach is called() after timewait ends (or reuse)
261 		 *  (e.g. by soclose), then tcptw has already been discarded
262 		 *  (or reused) and inpcb is freed here
263 		 *
264 		 *  In all three cases the tcptw should not be freed here.
265 		 */
266 		if (inp->inp_flags & INP_DROPPED) {
267 			in_pcbdetach(inp);
268 			if (__predict_true(tp == NULL)) {
269 				in_pcbfree(inp);
270 			} else {
271 				/*
272 				 * This case should not happen as in TIMEWAIT
273 				 * state the inp should not be destroyed before
274 				 * its tcptw.  If INVARIANTS is defined, panic.
275 				 */
276 #ifdef INVARIANTS
277 				panic("%s: Panic before an inp double-free: "
278 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
279 				    , __func__);
280 #else
281 				log(LOG_ERR, "%s: Avoid an inp double-free: "
282 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
283 				    , __func__);
284 #endif
285 				INP_WUNLOCK(inp);
286 			}
287 		} else {
288 			in_pcbdetach(inp);
289 			INP_WUNLOCK(inp);
290 		}
291 	} else {
292 		/*
293 		 * If the connection is not in timewait, we consider two
294 		 * two conditions: one in which no further processing is
295 		 * necessary (dropped || embryonic), and one in which TCP is
296 		 * not yet done, but no longer requires the socket, so the
297 		 * pcb will persist for the time being.
298 		 *
299 		 * XXXRW: Does the second case still occur?
300 		 */
301 		if (inp->inp_flags & INP_DROPPED ||
302 		    tp->t_state < TCPS_SYN_SENT) {
303 			tcp_discardcb(tp);
304 			in_pcbdetach(inp);
305 			in_pcbfree(inp);
306 		} else {
307 			in_pcbdetach(inp);
308 			INP_WUNLOCK(inp);
309 		}
310 	}
311 }
312 
313 #ifdef INET
314 /*
315  * Give the socket an address.
316  */
317 static int
318 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
319 {
320 	int error = 0;
321 	struct inpcb *inp;
322 	struct tcpcb *tp = NULL;
323 	struct sockaddr_in *sinp;
324 
325 	sinp = (struct sockaddr_in *)nam;
326 	if (nam->sa_len != sizeof (*sinp))
327 		return (EINVAL);
328 	/*
329 	 * Must check for multicast addresses and disallow binding
330 	 * to them.
331 	 */
332 	if (sinp->sin_family == AF_INET &&
333 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
334 		return (EAFNOSUPPORT);
335 
336 	TCPDEBUG0;
337 	inp = sotoinpcb(so);
338 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
339 	INP_WLOCK(inp);
340 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
341 		error = EINVAL;
342 		goto out;
343 	}
344 	tp = intotcpcb(inp);
345 	TCPDEBUG1();
346 	INP_HASH_WLOCK(&V_tcbinfo);
347 	error = in_pcbbind(inp, nam, td->td_ucred);
348 	INP_HASH_WUNLOCK(&V_tcbinfo);
349 out:
350 	TCPDEBUG2(PRU_BIND);
351 	TCP_PROBE2(debug__user, tp, PRU_BIND);
352 	INP_WUNLOCK(inp);
353 
354 	return (error);
355 }
356 #endif /* INET */
357 
358 #ifdef INET6
359 static int
360 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
361 {
362 	int error = 0;
363 	struct inpcb *inp;
364 	struct tcpcb *tp = NULL;
365 	struct sockaddr_in6 *sin6;
366 	u_char vflagsav;
367 
368 	sin6 = (struct sockaddr_in6 *)nam;
369 	if (nam->sa_len != sizeof (*sin6))
370 		return (EINVAL);
371 	/*
372 	 * Must check for multicast addresses and disallow binding
373 	 * to them.
374 	 */
375 	if (sin6->sin6_family == AF_INET6 &&
376 	    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
377 		return (EAFNOSUPPORT);
378 
379 	TCPDEBUG0;
380 	inp = sotoinpcb(so);
381 	KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
382 	INP_WLOCK(inp);
383 	vflagsav = inp->inp_vflag;
384 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
385 		error = EINVAL;
386 		goto out;
387 	}
388 	tp = intotcpcb(inp);
389 	TCPDEBUG1();
390 	INP_HASH_WLOCK(&V_tcbinfo);
391 	inp->inp_vflag &= ~INP_IPV4;
392 	inp->inp_vflag |= INP_IPV6;
393 #ifdef INET
394 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
395 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
396 			inp->inp_vflag |= INP_IPV4;
397 		else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
398 			struct sockaddr_in sin;
399 
400 			in6_sin6_2_sin(&sin, sin6);
401 			if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
402 				error = EAFNOSUPPORT;
403 				INP_HASH_WUNLOCK(&V_tcbinfo);
404 				goto out;
405 			}
406 			inp->inp_vflag |= INP_IPV4;
407 			inp->inp_vflag &= ~INP_IPV6;
408 			error = in_pcbbind(inp, (struct sockaddr *)&sin,
409 			    td->td_ucred);
410 			INP_HASH_WUNLOCK(&V_tcbinfo);
411 			goto out;
412 		}
413 	}
414 #endif
415 	error = in6_pcbbind(inp, nam, td->td_ucred);
416 	INP_HASH_WUNLOCK(&V_tcbinfo);
417 out:
418 	if (error != 0)
419 		inp->inp_vflag = vflagsav;
420 	TCPDEBUG2(PRU_BIND);
421 	TCP_PROBE2(debug__user, tp, PRU_BIND);
422 	INP_WUNLOCK(inp);
423 	return (error);
424 }
425 #endif /* INET6 */
426 
427 #ifdef INET
428 /*
429  * Prepare to accept connections.
430  */
431 static int
432 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
433 {
434 	int error = 0;
435 	struct inpcb *inp;
436 	struct tcpcb *tp = NULL;
437 
438 	TCPDEBUG0;
439 	inp = sotoinpcb(so);
440 	KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
441 	INP_WLOCK(inp);
442 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
443 		error = EINVAL;
444 		goto out;
445 	}
446 	tp = intotcpcb(inp);
447 	TCPDEBUG1();
448 	SOCK_LOCK(so);
449 	error = solisten_proto_check(so);
450 	INP_HASH_WLOCK(&V_tcbinfo);
451 	if (error == 0 && inp->inp_lport == 0)
452 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
453 	INP_HASH_WUNLOCK(&V_tcbinfo);
454 	if (error == 0) {
455 		tcp_state_change(tp, TCPS_LISTEN);
456 		solisten_proto(so, backlog);
457 #ifdef TCP_OFFLOAD
458 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
459 			tcp_offload_listen_start(tp);
460 #endif
461 	}
462 	SOCK_UNLOCK(so);
463 
464 	if (IS_FASTOPEN(tp->t_flags))
465 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
466 
467 out:
468 	TCPDEBUG2(PRU_LISTEN);
469 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
470 	INP_WUNLOCK(inp);
471 	return (error);
472 }
473 #endif /* INET */
474 
475 #ifdef INET6
476 static int
477 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
478 {
479 	int error = 0;
480 	struct inpcb *inp;
481 	struct tcpcb *tp = NULL;
482 	u_char vflagsav;
483 
484 	TCPDEBUG0;
485 	inp = sotoinpcb(so);
486 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
487 	INP_WLOCK(inp);
488 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
489 		error = EINVAL;
490 		goto out;
491 	}
492 	vflagsav = inp->inp_vflag;
493 	tp = intotcpcb(inp);
494 	TCPDEBUG1();
495 	SOCK_LOCK(so);
496 	error = solisten_proto_check(so);
497 	INP_HASH_WLOCK(&V_tcbinfo);
498 	if (error == 0 && inp->inp_lport == 0) {
499 		inp->inp_vflag &= ~INP_IPV4;
500 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
501 			inp->inp_vflag |= INP_IPV4;
502 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
503 	}
504 	INP_HASH_WUNLOCK(&V_tcbinfo);
505 	if (error == 0) {
506 		tcp_state_change(tp, TCPS_LISTEN);
507 		solisten_proto(so, backlog);
508 #ifdef TCP_OFFLOAD
509 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
510 			tcp_offload_listen_start(tp);
511 #endif
512 	}
513 	SOCK_UNLOCK(so);
514 
515 	if (IS_FASTOPEN(tp->t_flags))
516 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
517 
518 	if (error != 0)
519 		inp->inp_vflag = vflagsav;
520 
521 out:
522 	TCPDEBUG2(PRU_LISTEN);
523 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
524 	INP_WUNLOCK(inp);
525 	return (error);
526 }
527 #endif /* INET6 */
528 
529 #ifdef INET
530 /*
531  * Initiate connection to peer.
532  * Create a template for use in transmissions on this connection.
533  * Enter SYN_SENT state, and mark socket as connecting.
534  * Start keep-alive timer, and seed output sequence space.
535  * Send initial segment on connection.
536  */
537 static int
538 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
539 {
540 	struct epoch_tracker et;
541 	int error = 0;
542 	struct inpcb *inp;
543 	struct tcpcb *tp = NULL;
544 	struct sockaddr_in *sinp;
545 
546 	sinp = (struct sockaddr_in *)nam;
547 	if (nam->sa_len != sizeof (*sinp))
548 		return (EINVAL);
549 	/*
550 	 * Must disallow TCP ``connections'' to multicast addresses.
551 	 */
552 	if (sinp->sin_family == AF_INET
553 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
554 		return (EAFNOSUPPORT);
555 	if ((sinp->sin_family == AF_INET) &&
556 	    ((ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) ||
557 	     (sinp->sin_addr.s_addr == INADDR_ANY)))
558 		return(EAFNOSUPPORT);
559 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
560 		return (error);
561 
562 	TCPDEBUG0;
563 	inp = sotoinpcb(so);
564 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
565 	INP_WLOCK(inp);
566 	if (inp->inp_flags & INP_TIMEWAIT) {
567 		error = EADDRINUSE;
568 		goto out;
569 	}
570 	if (inp->inp_flags & INP_DROPPED) {
571 		error = ECONNREFUSED;
572 		goto out;
573 	}
574 	tp = intotcpcb(inp);
575 	TCPDEBUG1();
576 	NET_EPOCH_ENTER(et);
577 	if ((error = tcp_connect(tp, nam, td)) != 0)
578 		goto out_in_epoch;
579 #ifdef TCP_OFFLOAD
580 	if (registered_toedevs > 0 &&
581 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
582 	    (error = tcp_offload_connect(so, nam)) == 0)
583 		goto out_in_epoch;
584 #endif
585 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
586 	error = tp->t_fb->tfb_tcp_output(tp);
587 out_in_epoch:
588 	NET_EPOCH_EXIT(et);
589 out:
590 	TCPDEBUG2(PRU_CONNECT);
591 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
592 	INP_WUNLOCK(inp);
593 	return (error);
594 }
595 #endif /* INET */
596 
597 #ifdef INET6
598 static int
599 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
600 {
601 	struct epoch_tracker et;
602 	int error = 0;
603 	struct inpcb *inp;
604 	struct tcpcb *tp = NULL;
605 	struct sockaddr_in6 *sin6;
606 	u_int8_t incflagsav;
607 	u_char vflagsav;
608 
609 	TCPDEBUG0;
610 
611 	sin6 = (struct sockaddr_in6 *)nam;
612 	if (nam->sa_len != sizeof (*sin6))
613 		return (EINVAL);
614 	/*
615 	 * Must disallow TCP ``connections'' to multicast addresses.
616 	 */
617 	if (sin6->sin6_family == AF_INET6
618 	    && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
619 		return (EAFNOSUPPORT);
620 
621 	inp = sotoinpcb(so);
622 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
623 	INP_WLOCK(inp);
624 	vflagsav = inp->inp_vflag;
625 	incflagsav = inp->inp_inc.inc_flags;
626 	if (inp->inp_flags & INP_TIMEWAIT) {
627 		error = EADDRINUSE;
628 		goto out;
629 	}
630 	if (inp->inp_flags & INP_DROPPED) {
631 		error = ECONNREFUSED;
632 		goto out;
633 	}
634 	tp = intotcpcb(inp);
635 	TCPDEBUG1();
636 #ifdef INET
637 	/*
638 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
639 	 * therefore probably require the hash lock, which isn't held here.
640 	 * Is this a significant problem?
641 	 */
642 	if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
643 		struct sockaddr_in sin;
644 
645 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
646 			error = EINVAL;
647 			goto out;
648 		}
649 		if ((inp->inp_vflag & INP_IPV4) == 0) {
650 			error = EAFNOSUPPORT;
651 			goto out;
652 		}
653 
654 		in6_sin6_2_sin(&sin, sin6);
655 		if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
656 			error = EAFNOSUPPORT;
657 			goto out;
658 		}
659 		if ((ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) ||
660 		    (sin.sin_addr.s_addr == INADDR_ANY)) {
661 			error = EAFNOSUPPORT;
662 			goto out;
663 		}
664 		if ((error = prison_remote_ip4(td->td_ucred,
665 		    &sin.sin_addr)) != 0)
666 			goto out;
667 		inp->inp_vflag |= INP_IPV4;
668 		inp->inp_vflag &= ~INP_IPV6;
669 		NET_EPOCH_ENTER(et);
670 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
671 			goto out_in_epoch;
672 #ifdef TCP_OFFLOAD
673 		if (registered_toedevs > 0 &&
674 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
675 		    (error = tcp_offload_connect(so, nam)) == 0)
676 			goto out_in_epoch;
677 #endif
678 		error = tp->t_fb->tfb_tcp_output(tp);
679 		goto out_in_epoch;
680 	} else {
681 		if ((inp->inp_vflag & INP_IPV6) == 0) {
682 			error = EAFNOSUPPORT;
683 			goto out;
684 		}
685 	}
686 #endif
687 	if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
688 		goto out;
689 	inp->inp_vflag &= ~INP_IPV4;
690 	inp->inp_vflag |= INP_IPV6;
691 	inp->inp_inc.inc_flags |= INC_ISIPV6;
692 	if ((error = tcp6_connect(tp, nam, td)) != 0)
693 		goto out;
694 #ifdef TCP_OFFLOAD
695 	if (registered_toedevs > 0 &&
696 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
697 	    (error = tcp_offload_connect(so, nam)) == 0)
698 		goto out;
699 #endif
700 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
701 	NET_EPOCH_ENTER(et);
702 	error = tp->t_fb->tfb_tcp_output(tp);
703 #ifdef INET
704 out_in_epoch:
705 #endif
706 	NET_EPOCH_EXIT(et);
707 out:
708 	/*
709 	 * If the implicit bind in the connect call fails, restore
710 	 * the flags we modified.
711 	 */
712 	if (error != 0 && inp->inp_lport == 0) {
713 		inp->inp_vflag = vflagsav;
714 		inp->inp_inc.inc_flags = incflagsav;
715 	}
716 
717 	TCPDEBUG2(PRU_CONNECT);
718 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
719 	INP_WUNLOCK(inp);
720 	return (error);
721 }
722 #endif /* INET6 */
723 
724 /*
725  * Initiate disconnect from peer.
726  * If connection never passed embryonic stage, just drop;
727  * else if don't need to let data drain, then can just drop anyways,
728  * else have to begin TCP shutdown process: mark socket disconnecting,
729  * drain unread data, state switch to reflect user close, and
730  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
731  * when peer sends FIN and acks ours.
732  *
733  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
734  */
735 static int
736 tcp_usr_disconnect(struct socket *so)
737 {
738 	struct inpcb *inp;
739 	struct tcpcb *tp = NULL;
740 	struct epoch_tracker et;
741 	int error = 0;
742 
743 	TCPDEBUG0;
744 	NET_EPOCH_ENTER(et);
745 	inp = sotoinpcb(so);
746 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
747 	INP_WLOCK(inp);
748 	if (inp->inp_flags & INP_TIMEWAIT)
749 		goto out;
750 	if (inp->inp_flags & INP_DROPPED) {
751 		error = ECONNRESET;
752 		goto out;
753 	}
754 	tp = intotcpcb(inp);
755 	TCPDEBUG1();
756 	tcp_disconnect(tp);
757 out:
758 	TCPDEBUG2(PRU_DISCONNECT);
759 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
760 	INP_WUNLOCK(inp);
761 	NET_EPOCH_EXIT(et);
762 	return (error);
763 }
764 
765 #ifdef INET
766 /*
767  * Accept a connection.  Essentially all the work is done at higher levels;
768  * just return the address of the peer, storing through addr.
769  */
770 static int
771 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
772 {
773 	int error = 0;
774 	struct inpcb *inp = NULL;
775 	struct tcpcb *tp = NULL;
776 	struct in_addr addr;
777 	in_port_t port = 0;
778 	TCPDEBUG0;
779 
780 	if (so->so_state & SS_ISDISCONNECTED)
781 		return (ECONNABORTED);
782 
783 	inp = sotoinpcb(so);
784 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
785 	INP_WLOCK(inp);
786 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
787 		error = ECONNABORTED;
788 		goto out;
789 	}
790 	tp = intotcpcb(inp);
791 	TCPDEBUG1();
792 
793 	/*
794 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
795 	 * copy the data of interest and defer the malloc until after we
796 	 * release the lock.
797 	 */
798 	port = inp->inp_fport;
799 	addr = inp->inp_faddr;
800 
801 out:
802 	TCPDEBUG2(PRU_ACCEPT);
803 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
804 	INP_WUNLOCK(inp);
805 	if (error == 0)
806 		*nam = in_sockaddr(port, &addr);
807 	return error;
808 }
809 #endif /* INET */
810 
811 #ifdef INET6
812 static int
813 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
814 {
815 	struct inpcb *inp = NULL;
816 	int error = 0;
817 	struct tcpcb *tp = NULL;
818 	struct in_addr addr;
819 	struct in6_addr addr6;
820 	struct epoch_tracker et;
821 	in_port_t port = 0;
822 	int v4 = 0;
823 	TCPDEBUG0;
824 
825 	if (so->so_state & SS_ISDISCONNECTED)
826 		return (ECONNABORTED);
827 
828 	inp = sotoinpcb(so);
829 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
830 	NET_EPOCH_ENTER(et);
831 	INP_WLOCK(inp);
832 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
833 		error = ECONNABORTED;
834 		goto out;
835 	}
836 	tp = intotcpcb(inp);
837 	TCPDEBUG1();
838 
839 	/*
840 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
841 	 * copy the data of interest and defer the malloc until after we
842 	 * release the lock.
843 	 */
844 	if (inp->inp_vflag & INP_IPV4) {
845 		v4 = 1;
846 		port = inp->inp_fport;
847 		addr = inp->inp_faddr;
848 	} else {
849 		port = inp->inp_fport;
850 		addr6 = inp->in6p_faddr;
851 	}
852 
853 out:
854 	TCPDEBUG2(PRU_ACCEPT);
855 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
856 	INP_WUNLOCK(inp);
857 	NET_EPOCH_EXIT(et);
858 	if (error == 0) {
859 		if (v4)
860 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
861 		else
862 			*nam = in6_sockaddr(port, &addr6);
863 	}
864 	return error;
865 }
866 #endif /* INET6 */
867 
868 /*
869  * Mark the connection as being incapable of further output.
870  */
871 static int
872 tcp_usr_shutdown(struct socket *so)
873 {
874 	int error = 0;
875 	struct inpcb *inp;
876 	struct tcpcb *tp = NULL;
877 	struct epoch_tracker et;
878 
879 	TCPDEBUG0;
880 	NET_EPOCH_ENTER(et);
881 	inp = sotoinpcb(so);
882 	KASSERT(inp != NULL, ("inp == NULL"));
883 	INP_WLOCK(inp);
884 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
885 		error = ECONNRESET;
886 		goto out;
887 	}
888 	tp = intotcpcb(inp);
889 	TCPDEBUG1();
890 	socantsendmore(so);
891 	tcp_usrclosed(tp);
892 	if (!(inp->inp_flags & INP_DROPPED))
893 		error = tp->t_fb->tfb_tcp_output(tp);
894 
895 out:
896 	TCPDEBUG2(PRU_SHUTDOWN);
897 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
898 	INP_WUNLOCK(inp);
899 	NET_EPOCH_EXIT(et);
900 
901 	return (error);
902 }
903 
904 /*
905  * After a receive, possibly send window update to peer.
906  */
907 static int
908 tcp_usr_rcvd(struct socket *so, int flags)
909 {
910 	struct epoch_tracker et;
911 	struct inpcb *inp;
912 	struct tcpcb *tp = NULL;
913 	int error = 0;
914 
915 	TCPDEBUG0;
916 	inp = sotoinpcb(so);
917 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
918 	INP_WLOCK(inp);
919 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
920 		error = ECONNRESET;
921 		goto out;
922 	}
923 	tp = intotcpcb(inp);
924 	TCPDEBUG1();
925 	/*
926 	 * For passively-created TFO connections, don't attempt a window
927 	 * update while still in SYN_RECEIVED as this may trigger an early
928 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
929 	 * application response data, or failing that, when the DELACK timer
930 	 * expires.
931 	 */
932 	if (IS_FASTOPEN(tp->t_flags) &&
933 	    (tp->t_state == TCPS_SYN_RECEIVED))
934 		goto out;
935 	NET_EPOCH_ENTER(et);
936 #ifdef TCP_OFFLOAD
937 	if (tp->t_flags & TF_TOE)
938 		tcp_offload_rcvd(tp);
939 	else
940 #endif
941 	tp->t_fb->tfb_tcp_output(tp);
942 	NET_EPOCH_EXIT(et);
943 out:
944 	TCPDEBUG2(PRU_RCVD);
945 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
946 	INP_WUNLOCK(inp);
947 	return (error);
948 }
949 
950 /*
951  * Do a send by putting data in output queue and updating urgent
952  * marker if URG set.  Possibly send more data.  Unlike the other
953  * pru_*() routines, the mbuf chains are our responsibility.  We
954  * must either enqueue them or free them.  The other pru_* routines
955  * generally are caller-frees.
956  */
957 static int
958 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
959     struct sockaddr *nam, struct mbuf *control, struct thread *td)
960 {
961 	struct epoch_tracker et;
962 	int error = 0;
963 	struct inpcb *inp;
964 	struct tcpcb *tp = NULL;
965 #ifdef INET
966 #ifdef INET6
967 	struct sockaddr_in sin;
968 #endif
969 	struct sockaddr_in *sinp;
970 #endif
971 #ifdef INET6
972 	int isipv6;
973 #endif
974 	u_int8_t incflagsav;
975 	u_char vflagsav;
976 	bool restoreflags;
977 	TCPDEBUG0;
978 
979 	/*
980 	 * We require the pcbinfo "read lock" if we will close the socket
981 	 * as part of this call.
982 	 */
983 	NET_EPOCH_ENTER(et);
984 	inp = sotoinpcb(so);
985 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
986 	INP_WLOCK(inp);
987 	vflagsav = inp->inp_vflag;
988 	incflagsav = inp->inp_inc.inc_flags;
989 	restoreflags = false;
990 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
991 		if (control)
992 			m_freem(control);
993 		/*
994 		 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
995 		 * for freeing memory.
996 		 */
997 		if (m && (flags & PRUS_NOTREADY) == 0)
998 			m_freem(m);
999 		error = ECONNRESET;
1000 		goto out;
1001 	}
1002 	tp = intotcpcb(inp);
1003 	if (flags & PRUS_OOB) {
1004 		if ((error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) {
1005 			if (control)
1006 				m_freem(control);
1007 			if (m && (flags & PRUS_NOTREADY) == 0)
1008 				m_freem(m);
1009 			goto out;
1010 		}
1011 	}
1012 	TCPDEBUG1();
1013 	if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1014 		switch (nam->sa_family) {
1015 #ifdef INET
1016 		case AF_INET:
1017 			sinp = (struct sockaddr_in *)nam;
1018 			if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1019 				if (m)
1020 					m_freem(m);
1021 				error = EINVAL;
1022 				goto out;
1023 			}
1024 			if ((inp->inp_vflag & INP_IPV6) != 0) {
1025 				if (m)
1026 					m_freem(m);
1027 				error = EAFNOSUPPORT;
1028 				goto out;
1029 			}
1030 			if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1031 				if (m)
1032 					m_freem(m);
1033 				error = EAFNOSUPPORT;
1034 				goto out;
1035 			}
1036 			if ((ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) ||
1037 			    (sinp->sin_addr.s_addr == INADDR_ANY)) {
1038 				if (m)
1039 					m_freem(m);
1040 				error = EAFNOSUPPORT;
1041 				goto out;
1042 			}
1043 			if ((error = prison_remote_ip4(td->td_ucred,
1044 			    &sinp->sin_addr))) {
1045 				if (m)
1046 					m_freem(m);
1047 				goto out;
1048 			}
1049 #ifdef INET6
1050 			isipv6 = 0;
1051 #endif
1052 			break;
1053 #endif /* INET */
1054 #ifdef INET6
1055 		case AF_INET6:
1056 		{
1057 			struct sockaddr_in6 *sin6;
1058 
1059 			sin6 = (struct sockaddr_in6 *)nam;
1060 			if (sin6->sin6_len != sizeof(*sin6)) {
1061 				if (m)
1062 					m_freem(m);
1063 				error = EINVAL;
1064 				goto out;
1065 			}
1066 			if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1067 				if (m != NULL)
1068 					m_freem(m);
1069 				error = EAFNOSUPPORT;
1070 				goto out;
1071 			}
1072 			if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1073 				if (m)
1074 					m_freem(m);
1075 				error = EAFNOSUPPORT;
1076 				goto out;
1077 			}
1078 			if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1079 #ifdef INET
1080 				if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1081 					error = EINVAL;
1082 					if (m)
1083 						m_freem(m);
1084 					goto out;
1085 				}
1086 				if ((inp->inp_vflag & INP_IPV4) == 0) {
1087 					error = EAFNOSUPPORT;
1088 					if (m)
1089 						m_freem(m);
1090 					goto out;
1091 				}
1092 				restoreflags = true;
1093 				inp->inp_vflag &= ~INP_IPV6;
1094 				sinp = &sin;
1095 				in6_sin6_2_sin(sinp, sin6);
1096 				if (IN_MULTICAST(
1097 				    ntohl(sinp->sin_addr.s_addr))) {
1098 					error = EAFNOSUPPORT;
1099 					if (m)
1100 						m_freem(m);
1101 					goto out;
1102 				}
1103 				if ((error = prison_remote_ip4(td->td_ucred,
1104 				    &sinp->sin_addr))) {
1105 					if (m)
1106 						m_freem(m);
1107 					goto out;
1108 				}
1109 				isipv6 = 0;
1110 #else /* !INET */
1111 				error = EAFNOSUPPORT;
1112 				if (m)
1113 					m_freem(m);
1114 				goto out;
1115 #endif /* INET */
1116 			} else {
1117 				if ((inp->inp_vflag & INP_IPV6) == 0) {
1118 					if (m)
1119 						m_freem(m);
1120 					error = EAFNOSUPPORT;
1121 					goto out;
1122 				}
1123 				restoreflags = true;
1124 				inp->inp_vflag &= ~INP_IPV4;
1125 				inp->inp_inc.inc_flags |= INC_ISIPV6;
1126 				if ((error = prison_remote_ip6(td->td_ucred,
1127 				    &sin6->sin6_addr))) {
1128 					if (m)
1129 						m_freem(m);
1130 					goto out;
1131 				}
1132 				isipv6 = 1;
1133 			}
1134 			break;
1135 		}
1136 #endif /* INET6 */
1137 		default:
1138 			if (m)
1139 				m_freem(m);
1140 			error = EAFNOSUPPORT;
1141 			goto out;
1142 		}
1143 	}
1144 	if (control) {
1145 		/* TCP doesn't do control messages (rights, creds, etc) */
1146 		if (control->m_len) {
1147 			m_freem(control);
1148 			if (m)
1149 				m_freem(m);
1150 			error = EINVAL;
1151 			goto out;
1152 		}
1153 		m_freem(control);	/* empty control, just free it */
1154 	}
1155 	if (!(flags & PRUS_OOB)) {
1156 		sbappendstream(&so->so_snd, m, flags);
1157 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1158 			/*
1159 			 * Do implied connect if not yet connected,
1160 			 * initialize window to default value, and
1161 			 * initialize maxseg using peer's cached MSS.
1162 			 */
1163 #ifdef INET6
1164 			if (isipv6)
1165 				error = tcp6_connect(tp, nam, td);
1166 #endif /* INET6 */
1167 #if defined(INET6) && defined(INET)
1168 			else
1169 #endif
1170 #ifdef INET
1171 				error = tcp_connect(tp,
1172 				    (struct sockaddr *)sinp, td);
1173 #endif
1174 			/*
1175 			 * The bind operation in tcp_connect succeeded. We
1176 			 * no longer want to restore the flags if later
1177 			 * operations fail.
1178 			 */
1179 			if (error == 0 || inp->inp_lport != 0)
1180 				restoreflags = false;
1181 
1182 			if (error)
1183 				goto out;
1184 			if (IS_FASTOPEN(tp->t_flags))
1185 				tcp_fastopen_connect(tp);
1186 			else {
1187 				tp->snd_wnd = TTCP_CLIENT_SND_WND;
1188 				tcp_mss(tp, -1);
1189 			}
1190 		}
1191 		if (flags & PRUS_EOF) {
1192 			/*
1193 			 * Close the send side of the connection after
1194 			 * the data is sent.
1195 			 */
1196 			socantsendmore(so);
1197 			tcp_usrclosed(tp);
1198 		}
1199 		if (!(inp->inp_flags & INP_DROPPED) &&
1200 		    !(flags & PRUS_NOTREADY)) {
1201 			if (flags & PRUS_MORETOCOME)
1202 				tp->t_flags |= TF_MORETOCOME;
1203 			error = tp->t_fb->tfb_tcp_output(tp);
1204 			if (flags & PRUS_MORETOCOME)
1205 				tp->t_flags &= ~TF_MORETOCOME;
1206 		}
1207 	} else {
1208 		/*
1209 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1210 		 */
1211 		SOCKBUF_LOCK(&so->so_snd);
1212 		if (sbspace(&so->so_snd) < -512) {
1213 			SOCKBUF_UNLOCK(&so->so_snd);
1214 			m_freem(m);
1215 			error = ENOBUFS;
1216 			goto out;
1217 		}
1218 		/*
1219 		 * According to RFC961 (Assigned Protocols),
1220 		 * the urgent pointer points to the last octet
1221 		 * of urgent data.  We continue, however,
1222 		 * to consider it to indicate the first octet
1223 		 * of data past the urgent section.
1224 		 * Otherwise, snd_up should be one lower.
1225 		 */
1226 		sbappendstream_locked(&so->so_snd, m, flags);
1227 		SOCKBUF_UNLOCK(&so->so_snd);
1228 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1229 			/*
1230 			 * Do implied connect if not yet connected,
1231 			 * initialize window to default value, and
1232 			 * initialize maxseg using peer's cached MSS.
1233 			 */
1234 
1235 			/*
1236 			 * Not going to contemplate SYN|URG
1237 			 */
1238 			if (IS_FASTOPEN(tp->t_flags))
1239 				tp->t_flags &= ~TF_FASTOPEN;
1240 #ifdef INET6
1241 			if (isipv6)
1242 				error = tcp6_connect(tp, nam, td);
1243 #endif /* INET6 */
1244 #if defined(INET6) && defined(INET)
1245 			else
1246 #endif
1247 #ifdef INET
1248 				error = tcp_connect(tp,
1249 				    (struct sockaddr *)sinp, td);
1250 #endif
1251 			/*
1252 			 * The bind operation in tcp_connect succeeded. We
1253 			 * no longer want to restore the flags if later
1254 			 * operations fail.
1255 			 */
1256 			if (error == 0 || inp->inp_lport != 0)
1257 				restoreflags = false;
1258 
1259 			if (error)
1260 				goto out;
1261 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
1262 			tcp_mss(tp, -1);
1263 		}
1264 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1265 		if (!(flags & PRUS_NOTREADY)) {
1266 			tp->t_flags |= TF_FORCEDATA;
1267 			error = tp->t_fb->tfb_tcp_output(tp);
1268 			tp->t_flags &= ~TF_FORCEDATA;
1269 		}
1270 	}
1271 	TCP_LOG_EVENT(tp, NULL,
1272 	    &inp->inp_socket->so_rcv,
1273 	    &inp->inp_socket->so_snd,
1274 	    TCP_LOG_USERSEND, error,
1275 	    0, NULL, false);
1276 out:
1277 	/*
1278 	 * If the request was unsuccessful and we changed flags,
1279 	 * restore the original flags.
1280 	 */
1281 	if (error != 0 && restoreflags) {
1282 		inp->inp_vflag = vflagsav;
1283 		inp->inp_inc.inc_flags = incflagsav;
1284 	}
1285 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1286 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1287 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1288 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1289 	INP_WUNLOCK(inp);
1290 	NET_EPOCH_EXIT(et);
1291 	return (error);
1292 }
1293 
1294 static int
1295 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1296 {
1297 	struct epoch_tracker et;
1298 	struct inpcb *inp;
1299 	struct tcpcb *tp;
1300 	int error;
1301 
1302 	inp = sotoinpcb(so);
1303 	INP_WLOCK(inp);
1304 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1305 		INP_WUNLOCK(inp);
1306 		mb_free_notready(m, count);
1307 		return (ECONNRESET);
1308 	}
1309 	tp = intotcpcb(inp);
1310 
1311 	SOCKBUF_LOCK(&so->so_snd);
1312 	error = sbready(&so->so_snd, m, count);
1313 	SOCKBUF_UNLOCK(&so->so_snd);
1314 	if (error == 0) {
1315 		NET_EPOCH_ENTER(et);
1316 		error = tp->t_fb->tfb_tcp_output(tp);
1317 		NET_EPOCH_EXIT(et);
1318 	}
1319 	INP_WUNLOCK(inp);
1320 
1321 	return (error);
1322 }
1323 
1324 /*
1325  * Abort the TCP.  Drop the connection abruptly.
1326  */
1327 static void
1328 tcp_usr_abort(struct socket *so)
1329 {
1330 	struct inpcb *inp;
1331 	struct tcpcb *tp = NULL;
1332 	struct epoch_tracker et;
1333 	TCPDEBUG0;
1334 
1335 	inp = sotoinpcb(so);
1336 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1337 
1338 	NET_EPOCH_ENTER(et);
1339 	INP_WLOCK(inp);
1340 	KASSERT(inp->inp_socket != NULL,
1341 	    ("tcp_usr_abort: inp_socket == NULL"));
1342 
1343 	/*
1344 	 * If we still have full TCP state, and we're not dropped, drop.
1345 	 */
1346 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1347 	    !(inp->inp_flags & INP_DROPPED)) {
1348 		tp = intotcpcb(inp);
1349 		TCPDEBUG1();
1350 		tp = tcp_drop(tp, ECONNABORTED);
1351 		if (tp == NULL)
1352 			goto dropped;
1353 		TCPDEBUG2(PRU_ABORT);
1354 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1355 	}
1356 	if (!(inp->inp_flags & INP_DROPPED)) {
1357 		SOCK_LOCK(so);
1358 		so->so_state |= SS_PROTOREF;
1359 		SOCK_UNLOCK(so);
1360 		inp->inp_flags |= INP_SOCKREF;
1361 	}
1362 	INP_WUNLOCK(inp);
1363 dropped:
1364 	NET_EPOCH_EXIT(et);
1365 }
1366 
1367 /*
1368  * TCP socket is closed.  Start friendly disconnect.
1369  */
1370 static void
1371 tcp_usr_close(struct socket *so)
1372 {
1373 	struct inpcb *inp;
1374 	struct tcpcb *tp = NULL;
1375 	struct epoch_tracker et;
1376 	TCPDEBUG0;
1377 
1378 	inp = sotoinpcb(so);
1379 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1380 
1381 	NET_EPOCH_ENTER(et);
1382 	INP_WLOCK(inp);
1383 	KASSERT(inp->inp_socket != NULL,
1384 	    ("tcp_usr_close: inp_socket == NULL"));
1385 
1386 	/*
1387 	 * If we still have full TCP state, and we're not dropped, initiate
1388 	 * a disconnect.
1389 	 */
1390 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1391 	    !(inp->inp_flags & INP_DROPPED)) {
1392 		tp = intotcpcb(inp);
1393 		TCPDEBUG1();
1394 		tcp_disconnect(tp);
1395 		TCPDEBUG2(PRU_CLOSE);
1396 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1397 	}
1398 	if (!(inp->inp_flags & INP_DROPPED)) {
1399 		SOCK_LOCK(so);
1400 		so->so_state |= SS_PROTOREF;
1401 		SOCK_UNLOCK(so);
1402 		inp->inp_flags |= INP_SOCKREF;
1403 	}
1404 	INP_WUNLOCK(inp);
1405 	NET_EPOCH_EXIT(et);
1406 }
1407 
1408 static int
1409 tcp_pru_options_support(struct tcpcb *tp, int flags)
1410 {
1411 	/*
1412 	 * If the specific TCP stack has a pru_options
1413 	 * specified then it does not always support
1414 	 * all the PRU_XX options and we must ask it.
1415 	 * If the function is not specified then all
1416 	 * of the PRU_XX options are supported.
1417 	 */
1418 	int ret = 0;
1419 
1420 	if (tp->t_fb->tfb_pru_options) {
1421 		ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1422 	}
1423 	return (ret);
1424 }
1425 
1426 /*
1427  * Receive out-of-band data.
1428  */
1429 static int
1430 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1431 {
1432 	int error = 0;
1433 	struct inpcb *inp;
1434 	struct tcpcb *tp = NULL;
1435 
1436 	TCPDEBUG0;
1437 	inp = sotoinpcb(so);
1438 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1439 	INP_WLOCK(inp);
1440 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1441 		error = ECONNRESET;
1442 		goto out;
1443 	}
1444 	tp = intotcpcb(inp);
1445 	error = tcp_pru_options_support(tp, PRUS_OOB);
1446 	if (error) {
1447 		goto out;
1448 	}
1449 	TCPDEBUG1();
1450 	if ((so->so_oobmark == 0 &&
1451 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1452 	    so->so_options & SO_OOBINLINE ||
1453 	    tp->t_oobflags & TCPOOB_HADDATA) {
1454 		error = EINVAL;
1455 		goto out;
1456 	}
1457 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1458 		error = EWOULDBLOCK;
1459 		goto out;
1460 	}
1461 	m->m_len = 1;
1462 	*mtod(m, caddr_t) = tp->t_iobc;
1463 	if ((flags & MSG_PEEK) == 0)
1464 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1465 
1466 out:
1467 	TCPDEBUG2(PRU_RCVOOB);
1468 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1469 	INP_WUNLOCK(inp);
1470 	return (error);
1471 }
1472 
1473 #ifdef INET
1474 struct pr_usrreqs tcp_usrreqs = {
1475 	.pru_abort =		tcp_usr_abort,
1476 	.pru_accept =		tcp_usr_accept,
1477 	.pru_attach =		tcp_usr_attach,
1478 	.pru_bind =		tcp_usr_bind,
1479 	.pru_connect =		tcp_usr_connect,
1480 	.pru_control =		in_control,
1481 	.pru_detach =		tcp_usr_detach,
1482 	.pru_disconnect =	tcp_usr_disconnect,
1483 	.pru_listen =		tcp_usr_listen,
1484 	.pru_peeraddr =		in_getpeeraddr,
1485 	.pru_rcvd =		tcp_usr_rcvd,
1486 	.pru_rcvoob =		tcp_usr_rcvoob,
1487 	.pru_send =		tcp_usr_send,
1488 	.pru_ready =		tcp_usr_ready,
1489 	.pru_shutdown =		tcp_usr_shutdown,
1490 	.pru_sockaddr =		in_getsockaddr,
1491 	.pru_sosetlabel =	in_pcbsosetlabel,
1492 	.pru_close =		tcp_usr_close,
1493 };
1494 #endif /* INET */
1495 
1496 #ifdef INET6
1497 struct pr_usrreqs tcp6_usrreqs = {
1498 	.pru_abort =		tcp_usr_abort,
1499 	.pru_accept =		tcp6_usr_accept,
1500 	.pru_attach =		tcp_usr_attach,
1501 	.pru_bind =		tcp6_usr_bind,
1502 	.pru_connect =		tcp6_usr_connect,
1503 	.pru_control =		in6_control,
1504 	.pru_detach =		tcp_usr_detach,
1505 	.pru_disconnect =	tcp_usr_disconnect,
1506 	.pru_listen =		tcp6_usr_listen,
1507 	.pru_peeraddr =		in6_mapped_peeraddr,
1508 	.pru_rcvd =		tcp_usr_rcvd,
1509 	.pru_rcvoob =		tcp_usr_rcvoob,
1510 	.pru_send =		tcp_usr_send,
1511 	.pru_ready =		tcp_usr_ready,
1512 	.pru_shutdown =		tcp_usr_shutdown,
1513 	.pru_sockaddr =		in6_mapped_sockaddr,
1514 	.pru_sosetlabel =	in_pcbsosetlabel,
1515 	.pru_close =		tcp_usr_close,
1516 };
1517 #endif /* INET6 */
1518 
1519 #ifdef INET
1520 /*
1521  * Common subroutine to open a TCP connection to remote host specified
1522  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1523  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1524  * to choose a local host address (interface).  If there is an existing
1525  * incarnation of the same connection in TIME-WAIT state and if the remote
1526  * host was sending CC options and if the connection duration was < MSL, then
1527  * truncate the previous TIME-WAIT state and proceed.
1528  * Initialize connection parameters and enter SYN-SENT state.
1529  */
1530 static int
1531 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1532 {
1533 	struct inpcb *inp = tp->t_inpcb, *oinp;
1534 	struct socket *so = inp->inp_socket;
1535 	struct in_addr laddr;
1536 	u_short lport;
1537 	int error;
1538 
1539 	NET_EPOCH_ASSERT();
1540 	INP_WLOCK_ASSERT(inp);
1541 	INP_HASH_WLOCK(&V_tcbinfo);
1542 
1543 	if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1544 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1545 		if (error)
1546 			goto out;
1547 	}
1548 
1549 	/*
1550 	 * Cannot simply call in_pcbconnect, because there might be an
1551 	 * earlier incarnation of this same connection still in
1552 	 * TIME_WAIT state, creating an ADDRINUSE error.
1553 	 */
1554 	laddr = inp->inp_laddr;
1555 	lport = inp->inp_lport;
1556 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1557 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1558 	if (error && oinp == NULL)
1559 		goto out;
1560 	if (oinp) {
1561 		error = EADDRINUSE;
1562 		goto out;
1563 	}
1564 	/* Handle initial bind if it hadn't been done in advance. */
1565 	if (inp->inp_lport == 0) {
1566 		inp->inp_lport = lport;
1567 		if (in_pcbinshash(inp) != 0) {
1568 			inp->inp_lport = 0;
1569 			error = EAGAIN;
1570 			goto out;
1571 		}
1572 	}
1573 	inp->inp_laddr = laddr;
1574 	in_pcbrehash(inp);
1575 	INP_HASH_WUNLOCK(&V_tcbinfo);
1576 
1577 	/*
1578 	 * Compute window scaling to request:
1579 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1580 	 * XXX: This should move to tcp_output().
1581 	 */
1582 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1583 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1584 		tp->request_r_scale++;
1585 
1586 	soisconnecting(so);
1587 	TCPSTAT_INC(tcps_connattempt);
1588 	tcp_state_change(tp, TCPS_SYN_SENT);
1589 	tp->iss = tcp_new_isn(&inp->inp_inc);
1590 	if (tp->t_flags & TF_REQ_TSTMP)
1591 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1592 	tcp_sendseqinit(tp);
1593 
1594 	return 0;
1595 
1596 out:
1597 	INP_HASH_WUNLOCK(&V_tcbinfo);
1598 	return (error);
1599 }
1600 #endif /* INET */
1601 
1602 #ifdef INET6
1603 static int
1604 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1605 {
1606 	struct inpcb *inp = tp->t_inpcb;
1607 	int error;
1608 
1609 	INP_WLOCK_ASSERT(inp);
1610 	INP_HASH_WLOCK(&V_tcbinfo);
1611 
1612 	if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1613 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1614 		if (error)
1615 			goto out;
1616 	}
1617 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1618 	if (error != 0)
1619 		goto out;
1620 	INP_HASH_WUNLOCK(&V_tcbinfo);
1621 
1622 	/* Compute window scaling to request.  */
1623 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1624 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1625 		tp->request_r_scale++;
1626 
1627 	soisconnecting(inp->inp_socket);
1628 	TCPSTAT_INC(tcps_connattempt);
1629 	tcp_state_change(tp, TCPS_SYN_SENT);
1630 	tp->iss = tcp_new_isn(&inp->inp_inc);
1631 	if (tp->t_flags & TF_REQ_TSTMP)
1632 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1633 	tcp_sendseqinit(tp);
1634 
1635 	return 0;
1636 
1637 out:
1638 	INP_HASH_WUNLOCK(&V_tcbinfo);
1639 	return error;
1640 }
1641 #endif /* INET6 */
1642 
1643 /*
1644  * Export TCP internal state information via a struct tcp_info, based on the
1645  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1646  * (TCP state machine, etc).  We export all information using FreeBSD-native
1647  * constants -- for example, the numeric values for tcpi_state will differ
1648  * from Linux.
1649  */
1650 static void
1651 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1652 {
1653 
1654 	INP_WLOCK_ASSERT(tp->t_inpcb);
1655 	bzero(ti, sizeof(*ti));
1656 
1657 	ti->tcpi_state = tp->t_state;
1658 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1659 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1660 	if (tp->t_flags & TF_SACK_PERMIT)
1661 		ti->tcpi_options |= TCPI_OPT_SACK;
1662 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1663 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1664 		ti->tcpi_snd_wscale = tp->snd_scale;
1665 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1666 	}
1667 	if (tp->t_flags2 & TF2_ECN_PERMIT)
1668 		ti->tcpi_options |= TCPI_OPT_ECN;
1669 
1670 	ti->tcpi_rto = tp->t_rxtcur * tick;
1671 	ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1672 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1673 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1674 
1675 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1676 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1677 
1678 	/*
1679 	 * FreeBSD-specific extension fields for tcp_info.
1680 	 */
1681 	ti->tcpi_rcv_space = tp->rcv_wnd;
1682 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1683 	ti->tcpi_snd_wnd = tp->snd_wnd;
1684 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1685 	ti->tcpi_snd_nxt = tp->snd_nxt;
1686 	ti->tcpi_snd_mss = tp->t_maxseg;
1687 	ti->tcpi_rcv_mss = tp->t_maxseg;
1688 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1689 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1690 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1691 #ifdef TCP_OFFLOAD
1692 	if (tp->t_flags & TF_TOE) {
1693 		ti->tcpi_options |= TCPI_OPT_TOE;
1694 		tcp_offload_tcp_info(tp, ti);
1695 	}
1696 #endif
1697 }
1698 
1699 /*
1700  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1701  * socket option arguments.  When it re-acquires the lock after the copy, it
1702  * has to revalidate that the connection is still valid for the socket
1703  * option.
1704  */
1705 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1706 	INP_WLOCK(inp);							\
1707 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1708 		INP_WUNLOCK(inp);					\
1709 		cleanup;						\
1710 		return (ECONNRESET);					\
1711 	}								\
1712 	tp = intotcpcb(inp);						\
1713 } while(0)
1714 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1715 
1716 int
1717 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1718 {
1719 	int	error;
1720 	struct	inpcb *inp;
1721 	struct	tcpcb *tp;
1722 	struct tcp_function_block *blk;
1723 	struct tcp_function_set fsn;
1724 
1725 	error = 0;
1726 	inp = sotoinpcb(so);
1727 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1728 	if (sopt->sopt_level != IPPROTO_TCP) {
1729 #ifdef INET6
1730 		if (inp->inp_vflag & INP_IPV6PROTO) {
1731 			error = ip6_ctloutput(so, sopt);
1732 			/*
1733 			 * In case of the IPV6_USE_MIN_MTU socket option,
1734 			 * the INC_IPV6MINMTU flag to announce a corresponding
1735 			 * MSS during the initial handshake.
1736 			 * If the TCP connection is not in the front states,
1737 			 * just reduce the MSS being used.
1738 			 * This avoids the sending of TCP segments which will
1739 			 * be fragmented at the IPv6 layer.
1740 			 */
1741 			if ((error == 0) &&
1742 			    (sopt->sopt_dir == SOPT_SET) &&
1743 			    (sopt->sopt_level == IPPROTO_IPV6) &&
1744 			    (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1745 				INP_WLOCK(inp);
1746 				if ((inp->inp_flags &
1747 				    (INP_TIMEWAIT | INP_DROPPED))) {
1748 					INP_WUNLOCK(inp);
1749 					return (ECONNRESET);
1750 				}
1751 				inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1752 				tp = intotcpcb(inp);
1753 				if ((tp->t_state >= TCPS_SYN_SENT) &&
1754 				    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1755 					struct ip6_pktopts *opt;
1756 
1757 					opt = inp->in6p_outputopts;
1758 					if ((opt != NULL) &&
1759 					    (opt->ip6po_minmtu ==
1760 					    IP6PO_MINMTU_ALL)) {
1761 						if (tp->t_maxseg > TCP6_MSS) {
1762 							tp->t_maxseg = TCP6_MSS;
1763 						}
1764 					}
1765 				}
1766 				INP_WUNLOCK(inp);
1767 			}
1768 		}
1769 #endif /* INET6 */
1770 #if defined(INET6) && defined(INET)
1771 		else
1772 #endif
1773 #ifdef INET
1774 		{
1775 			error = ip_ctloutput(so, sopt);
1776 		}
1777 #endif
1778 		return (error);
1779 	}
1780 	INP_WLOCK(inp);
1781 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1782 		INP_WUNLOCK(inp);
1783 		return (ECONNRESET);
1784 	}
1785 	tp = intotcpcb(inp);
1786 	/*
1787 	 * Protect the TCP option TCP_FUNCTION_BLK so
1788 	 * that a sub-function can *never* overwrite this.
1789 	 */
1790 	if ((sopt->sopt_dir == SOPT_SET) &&
1791 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1792 		INP_WUNLOCK(inp);
1793 		error = sooptcopyin(sopt, &fsn, sizeof fsn,
1794 		    sizeof fsn);
1795 		if (error)
1796 			return (error);
1797 		INP_WLOCK_RECHECK(inp);
1798 		blk = find_and_ref_tcp_functions(&fsn);
1799 		if (blk == NULL) {
1800 			INP_WUNLOCK(inp);
1801 			return (ENOENT);
1802 		}
1803 		if (tp->t_fb == blk) {
1804 			/* You already have this */
1805 			refcount_release(&blk->tfb_refcnt);
1806 			INP_WUNLOCK(inp);
1807 			return (0);
1808 		}
1809 		if (tp->t_state != TCPS_CLOSED) {
1810 			/*
1811 			 * The user has advanced the state
1812 			 * past the initial point, we may not
1813 			 * be able to switch.
1814 			 */
1815 			if (blk->tfb_tcp_handoff_ok != NULL) {
1816 				/*
1817 				 * Does the stack provide a
1818 				 * query mechanism, if so it may
1819 				 * still be possible?
1820 				 */
1821 				error = (*blk->tfb_tcp_handoff_ok)(tp);
1822 			} else
1823 				error = EINVAL;
1824 			if (error) {
1825 				refcount_release(&blk->tfb_refcnt);
1826 				INP_WUNLOCK(inp);
1827 				return(error);
1828 			}
1829 		}
1830 		if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1831 			refcount_release(&blk->tfb_refcnt);
1832 			INP_WUNLOCK(inp);
1833 			return (ENOENT);
1834 		}
1835 		/*
1836 		 * Release the old refcnt, the
1837 		 * lookup acquired a ref on the
1838 		 * new one already.
1839 		 */
1840 		if (tp->t_fb->tfb_tcp_fb_fini) {
1841 			/*
1842 			 * Tell the stack to cleanup with 0 i.e.
1843 			 * the tcb is not going away.
1844 			 */
1845 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1846 		}
1847 #ifdef TCPHPTS
1848 		/* Assure that we are not on any hpts */
1849 		tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1850 #endif
1851 		if (blk->tfb_tcp_fb_init) {
1852 			error = (*blk->tfb_tcp_fb_init)(tp);
1853 			if (error) {
1854 				refcount_release(&blk->tfb_refcnt);
1855 				if (tp->t_fb->tfb_tcp_fb_init) {
1856 					if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1857 						/* Fall back failed, drop the connection */
1858 						INP_WUNLOCK(inp);
1859 						soabort(so);
1860 						return(error);
1861 					}
1862 				}
1863 				goto err_out;
1864 			}
1865 		}
1866 		refcount_release(&tp->t_fb->tfb_refcnt);
1867 		tp->t_fb = blk;
1868 #ifdef TCP_OFFLOAD
1869 		if (tp->t_flags & TF_TOE) {
1870 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1871 			     sopt->sopt_name);
1872 		}
1873 #endif
1874 err_out:
1875 		INP_WUNLOCK(inp);
1876 		return (error);
1877 	} else if ((sopt->sopt_dir == SOPT_GET) &&
1878 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1879 		strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1880 		    TCP_FUNCTION_NAME_LEN_MAX);
1881 		fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1882 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1883 		INP_WUNLOCK(inp);
1884 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1885 		return (error);
1886 	}
1887 	/* Pass in the INP locked, called must unlock it */
1888 	return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1889 }
1890 
1891 /*
1892  * If this assert becomes untrue, we need to change the size of the buf
1893  * variable in tcp_default_ctloutput().
1894  */
1895 #ifdef CTASSERT
1896 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1897 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1898 #endif
1899 
1900 #ifdef KERN_TLS
1901 static int
1902 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1903 {
1904 	struct tls_enable_v0 tls_v0;
1905 	int error;
1906 
1907 	if (sopt->sopt_valsize == sizeof(tls_v0)) {
1908 		error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1909 		    sizeof(tls_v0));
1910 		if (error)
1911 			return (error);
1912 		memset(tls, 0, sizeof(*tls));
1913 		tls->cipher_key = tls_v0.cipher_key;
1914 		tls->iv = tls_v0.iv;
1915 		tls->auth_key = tls_v0.auth_key;
1916 		tls->cipher_algorithm = tls_v0.cipher_algorithm;
1917 		tls->cipher_key_len = tls_v0.cipher_key_len;
1918 		tls->iv_len = tls_v0.iv_len;
1919 		tls->auth_algorithm = tls_v0.auth_algorithm;
1920 		tls->auth_key_len = tls_v0.auth_key_len;
1921 		tls->flags = tls_v0.flags;
1922 		tls->tls_vmajor = tls_v0.tls_vmajor;
1923 		tls->tls_vminor = tls_v0.tls_vminor;
1924 		return (0);
1925 	}
1926 
1927 	return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
1928 }
1929 #endif
1930 
1931 int
1932 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1933 {
1934 	int	error, opt, optval;
1935 	u_int	ui;
1936 	struct	tcp_info ti;
1937 #ifdef KERN_TLS
1938 	struct tls_enable tls;
1939 #endif
1940 	struct cc_algo *algo;
1941 	char	*pbuf, buf[TCP_LOG_ID_LEN];
1942 #ifdef STATS
1943 	struct statsblob *sbp;
1944 #endif
1945 	size_t	len;
1946 
1947 	/*
1948 	 * For TCP_CCALGOOPT forward the control to CC module, for both
1949 	 * SOPT_SET and SOPT_GET.
1950 	 */
1951 	switch (sopt->sopt_name) {
1952 	case TCP_CCALGOOPT:
1953 		INP_WUNLOCK(inp);
1954 		if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1955 			return (EINVAL);
1956 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1957 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1958 		    sopt->sopt_valsize);
1959 		if (error) {
1960 			free(pbuf, M_TEMP);
1961 			return (error);
1962 		}
1963 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1964 		if (CC_ALGO(tp)->ctl_output != NULL)
1965 			error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1966 		else
1967 			error = ENOENT;
1968 		INP_WUNLOCK(inp);
1969 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
1970 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1971 		free(pbuf, M_TEMP);
1972 		return (error);
1973 	}
1974 
1975 	switch (sopt->sopt_dir) {
1976 	case SOPT_SET:
1977 		switch (sopt->sopt_name) {
1978 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1979 		case TCP_MD5SIG:
1980 			if (!TCPMD5_ENABLED()) {
1981 				INP_WUNLOCK(inp);
1982 				return (ENOPROTOOPT);
1983 			}
1984 			error = TCPMD5_PCBCTL(inp, sopt);
1985 			if (error)
1986 				return (error);
1987 			goto unlock_and_done;
1988 #endif /* IPSEC */
1989 
1990 		case TCP_NODELAY:
1991 		case TCP_NOOPT:
1992 			INP_WUNLOCK(inp);
1993 			error = sooptcopyin(sopt, &optval, sizeof optval,
1994 			    sizeof optval);
1995 			if (error)
1996 				return (error);
1997 
1998 			INP_WLOCK_RECHECK(inp);
1999 			switch (sopt->sopt_name) {
2000 			case TCP_NODELAY:
2001 				opt = TF_NODELAY;
2002 				break;
2003 			case TCP_NOOPT:
2004 				opt = TF_NOOPT;
2005 				break;
2006 			default:
2007 				opt = 0; /* dead code to fool gcc */
2008 				break;
2009 			}
2010 
2011 			if (optval)
2012 				tp->t_flags |= opt;
2013 			else
2014 				tp->t_flags &= ~opt;
2015 unlock_and_done:
2016 #ifdef TCP_OFFLOAD
2017 			if (tp->t_flags & TF_TOE) {
2018 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
2019 				    sopt->sopt_name);
2020 			}
2021 #endif
2022 			INP_WUNLOCK(inp);
2023 			break;
2024 
2025 		case TCP_NOPUSH:
2026 			INP_WUNLOCK(inp);
2027 			error = sooptcopyin(sopt, &optval, sizeof optval,
2028 			    sizeof optval);
2029 			if (error)
2030 				return (error);
2031 
2032 			INP_WLOCK_RECHECK(inp);
2033 			if (optval)
2034 				tp->t_flags |= TF_NOPUSH;
2035 			else if (tp->t_flags & TF_NOPUSH) {
2036 				tp->t_flags &= ~TF_NOPUSH;
2037 				if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2038 					struct epoch_tracker et;
2039 
2040 					NET_EPOCH_ENTER(et);
2041 					error = tp->t_fb->tfb_tcp_output(tp);
2042 					NET_EPOCH_EXIT(et);
2043 				}
2044 			}
2045 			goto unlock_and_done;
2046 
2047 		case TCP_MAXSEG:
2048 			INP_WUNLOCK(inp);
2049 			error = sooptcopyin(sopt, &optval, sizeof optval,
2050 			    sizeof optval);
2051 			if (error)
2052 				return (error);
2053 
2054 			INP_WLOCK_RECHECK(inp);
2055 			if (optval > 0 && optval <= tp->t_maxseg &&
2056 			    optval + 40 >= V_tcp_minmss)
2057 				tp->t_maxseg = optval;
2058 			else
2059 				error = EINVAL;
2060 			goto unlock_and_done;
2061 
2062 		case TCP_INFO:
2063 			INP_WUNLOCK(inp);
2064 			error = EINVAL;
2065 			break;
2066 
2067 		case TCP_STATS:
2068 			INP_WUNLOCK(inp);
2069 #ifdef STATS
2070 			error = sooptcopyin(sopt, &optval, sizeof optval,
2071 			    sizeof optval);
2072 			if (error)
2073 				return (error);
2074 
2075 			if (optval > 0)
2076 				sbp = stats_blob_alloc(
2077 				    V_tcp_perconn_stats_dflt_tpl, 0);
2078 			else
2079 				sbp = NULL;
2080 
2081 			INP_WLOCK_RECHECK(inp);
2082 			if ((tp->t_stats != NULL && sbp == NULL) ||
2083 			    (tp->t_stats == NULL && sbp != NULL)) {
2084 				struct statsblob *t = tp->t_stats;
2085 				tp->t_stats = sbp;
2086 				sbp = t;
2087 			}
2088 			INP_WUNLOCK(inp);
2089 
2090 			stats_blob_destroy(sbp);
2091 #else
2092 			return (EOPNOTSUPP);
2093 #endif /* !STATS */
2094 			break;
2095 
2096 		case TCP_CONGESTION:
2097 			INP_WUNLOCK(inp);
2098 			error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2099 			if (error)
2100 				break;
2101 			buf[sopt->sopt_valsize] = '\0';
2102 			INP_WLOCK_RECHECK(inp);
2103 			CC_LIST_RLOCK();
2104 			STAILQ_FOREACH(algo, &cc_list, entries)
2105 				if (strncmp(buf, algo->name,
2106 				    TCP_CA_NAME_MAX) == 0)
2107 					break;
2108 			CC_LIST_RUNLOCK();
2109 			if (algo == NULL) {
2110 				INP_WUNLOCK(inp);
2111 				error = EINVAL;
2112 				break;
2113 			}
2114 			/*
2115 			 * We hold a write lock over the tcb so it's safe to
2116 			 * do these things without ordering concerns.
2117 			 */
2118 			if (CC_ALGO(tp)->cb_destroy != NULL)
2119 				CC_ALGO(tp)->cb_destroy(tp->ccv);
2120 			CC_DATA(tp) = NULL;
2121 			CC_ALGO(tp) = algo;
2122 			/*
2123 			 * If something goes pear shaped initialising the new
2124 			 * algo, fall back to newreno (which does not
2125 			 * require initialisation).
2126 			 */
2127 			if (algo->cb_init != NULL &&
2128 			    algo->cb_init(tp->ccv) != 0) {
2129 				CC_ALGO(tp) = &newreno_cc_algo;
2130 				/*
2131 				 * The only reason init should fail is
2132 				 * because of malloc.
2133 				 */
2134 				error = ENOMEM;
2135 			}
2136 			INP_WUNLOCK(inp);
2137 			break;
2138 
2139 #ifdef KERN_TLS
2140 		case TCP_TXTLS_ENABLE:
2141 			INP_WUNLOCK(inp);
2142 			error = copyin_tls_enable(sopt, &tls);
2143 			if (error)
2144 				break;
2145 			error = ktls_enable_tx(so, &tls);
2146 			break;
2147 		case TCP_TXTLS_MODE:
2148 			INP_WUNLOCK(inp);
2149 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2150 			if (error)
2151 				return (error);
2152 
2153 			INP_WLOCK_RECHECK(inp);
2154 			error = ktls_set_tx_mode(so, ui);
2155 			INP_WUNLOCK(inp);
2156 			break;
2157 		case TCP_RXTLS_ENABLE:
2158 			INP_WUNLOCK(inp);
2159 			error = sooptcopyin(sopt, &tls, sizeof(tls),
2160 			    sizeof(tls));
2161 			if (error)
2162 				break;
2163 			error = ktls_enable_rx(so, &tls);
2164 			break;
2165 #endif
2166 
2167 		case TCP_KEEPIDLE:
2168 		case TCP_KEEPINTVL:
2169 		case TCP_KEEPINIT:
2170 			INP_WUNLOCK(inp);
2171 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2172 			if (error)
2173 				return (error);
2174 
2175 			if (ui > (UINT_MAX / hz)) {
2176 				error = EINVAL;
2177 				break;
2178 			}
2179 			ui *= hz;
2180 
2181 			INP_WLOCK_RECHECK(inp);
2182 			switch (sopt->sopt_name) {
2183 			case TCP_KEEPIDLE:
2184 				tp->t_keepidle = ui;
2185 				/*
2186 				 * XXX: better check current remaining
2187 				 * timeout and "merge" it with new value.
2188 				 */
2189 				if ((tp->t_state > TCPS_LISTEN) &&
2190 				    (tp->t_state <= TCPS_CLOSING))
2191 					tcp_timer_activate(tp, TT_KEEP,
2192 					    TP_KEEPIDLE(tp));
2193 				break;
2194 			case TCP_KEEPINTVL:
2195 				tp->t_keepintvl = ui;
2196 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2197 				    (TP_MAXIDLE(tp) > 0))
2198 					tcp_timer_activate(tp, TT_2MSL,
2199 					    TP_MAXIDLE(tp));
2200 				break;
2201 			case TCP_KEEPINIT:
2202 				tp->t_keepinit = ui;
2203 				if (tp->t_state == TCPS_SYN_RECEIVED ||
2204 				    tp->t_state == TCPS_SYN_SENT)
2205 					tcp_timer_activate(tp, TT_KEEP,
2206 					    TP_KEEPINIT(tp));
2207 				break;
2208 			}
2209 			goto unlock_and_done;
2210 
2211 		case TCP_KEEPCNT:
2212 			INP_WUNLOCK(inp);
2213 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2214 			if (error)
2215 				return (error);
2216 
2217 			INP_WLOCK_RECHECK(inp);
2218 			tp->t_keepcnt = ui;
2219 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2220 			    (TP_MAXIDLE(tp) > 0))
2221 				tcp_timer_activate(tp, TT_2MSL,
2222 				    TP_MAXIDLE(tp));
2223 			goto unlock_and_done;
2224 
2225 #ifdef TCPPCAP
2226 		case TCP_PCAP_OUT:
2227 		case TCP_PCAP_IN:
2228 			INP_WUNLOCK(inp);
2229 			error = sooptcopyin(sopt, &optval, sizeof optval,
2230 			    sizeof optval);
2231 			if (error)
2232 				return (error);
2233 
2234 			INP_WLOCK_RECHECK(inp);
2235 			if (optval >= 0)
2236 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2237 					&(tp->t_outpkts) : &(tp->t_inpkts),
2238 					optval);
2239 			else
2240 				error = EINVAL;
2241 			goto unlock_and_done;
2242 #endif
2243 
2244 		case TCP_FASTOPEN: {
2245 			struct tcp_fastopen tfo_optval;
2246 
2247 			INP_WUNLOCK(inp);
2248 			if (!V_tcp_fastopen_client_enable &&
2249 			    !V_tcp_fastopen_server_enable)
2250 				return (EPERM);
2251 
2252 			error = sooptcopyin(sopt, &tfo_optval,
2253 				    sizeof(tfo_optval), sizeof(int));
2254 			if (error)
2255 				return (error);
2256 
2257 			INP_WLOCK_RECHECK(inp);
2258 			if ((tp->t_state != TCPS_CLOSED) &&
2259 			    (tp->t_state != TCPS_LISTEN)) {
2260 				error = EINVAL;
2261 				goto unlock_and_done;
2262 			}
2263 			if (tfo_optval.enable) {
2264 				if (tp->t_state == TCPS_LISTEN) {
2265 					if (!V_tcp_fastopen_server_enable) {
2266 						error = EPERM;
2267 						goto unlock_and_done;
2268 					}
2269 
2270 					if (tp->t_tfo_pending == NULL)
2271 						tp->t_tfo_pending =
2272 						    tcp_fastopen_alloc_counter();
2273 				} else {
2274 					/*
2275 					 * If a pre-shared key was provided,
2276 					 * stash it in the client cookie
2277 					 * field of the tcpcb for use during
2278 					 * connect.
2279 					 */
2280 					if (sopt->sopt_valsize ==
2281 					    sizeof(tfo_optval)) {
2282 						memcpy(tp->t_tfo_cookie.client,
2283 						       tfo_optval.psk,
2284 						       TCP_FASTOPEN_PSK_LEN);
2285 						tp->t_tfo_client_cookie_len =
2286 						    TCP_FASTOPEN_PSK_LEN;
2287 					}
2288 				}
2289 				tp->t_flags |= TF_FASTOPEN;
2290 			} else
2291 				tp->t_flags &= ~TF_FASTOPEN;
2292 			goto unlock_and_done;
2293 		}
2294 
2295 #ifdef TCP_BLACKBOX
2296 		case TCP_LOG:
2297 			INP_WUNLOCK(inp);
2298 			error = sooptcopyin(sopt, &optval, sizeof optval,
2299 			    sizeof optval);
2300 			if (error)
2301 				return (error);
2302 
2303 			INP_WLOCK_RECHECK(inp);
2304 			error = tcp_log_state_change(tp, optval);
2305 			goto unlock_and_done;
2306 
2307 		case TCP_LOGBUF:
2308 			INP_WUNLOCK(inp);
2309 			error = EINVAL;
2310 			break;
2311 
2312 		case TCP_LOGID:
2313 			INP_WUNLOCK(inp);
2314 			error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2315 			if (error)
2316 				break;
2317 			buf[sopt->sopt_valsize] = '\0';
2318 			INP_WLOCK_RECHECK(inp);
2319 			error = tcp_log_set_id(tp, buf);
2320 			/* tcp_log_set_id() unlocks the INP. */
2321 			break;
2322 
2323 		case TCP_LOGDUMP:
2324 		case TCP_LOGDUMPID:
2325 			INP_WUNLOCK(inp);
2326 			error =
2327 			    sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2328 			if (error)
2329 				break;
2330 			buf[sopt->sopt_valsize] = '\0';
2331 			INP_WLOCK_RECHECK(inp);
2332 			if (sopt->sopt_name == TCP_LOGDUMP) {
2333 				error = tcp_log_dump_tp_logbuf(tp, buf,
2334 				    M_WAITOK, true);
2335 				INP_WUNLOCK(inp);
2336 			} else {
2337 				tcp_log_dump_tp_bucket_logbufs(tp, buf);
2338 				/*
2339 				 * tcp_log_dump_tp_bucket_logbufs() drops the
2340 				 * INP lock.
2341 				 */
2342 			}
2343 			break;
2344 #endif
2345 
2346 		default:
2347 			INP_WUNLOCK(inp);
2348 			error = ENOPROTOOPT;
2349 			break;
2350 		}
2351 		break;
2352 
2353 	case SOPT_GET:
2354 		tp = intotcpcb(inp);
2355 		switch (sopt->sopt_name) {
2356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2357 		case TCP_MD5SIG:
2358 			if (!TCPMD5_ENABLED()) {
2359 				INP_WUNLOCK(inp);
2360 				return (ENOPROTOOPT);
2361 			}
2362 			error = TCPMD5_PCBCTL(inp, sopt);
2363 			break;
2364 #endif
2365 
2366 		case TCP_NODELAY:
2367 			optval = tp->t_flags & TF_NODELAY;
2368 			INP_WUNLOCK(inp);
2369 			error = sooptcopyout(sopt, &optval, sizeof optval);
2370 			break;
2371 		case TCP_MAXSEG:
2372 			optval = tp->t_maxseg;
2373 			INP_WUNLOCK(inp);
2374 			error = sooptcopyout(sopt, &optval, sizeof optval);
2375 			break;
2376 		case TCP_NOOPT:
2377 			optval = tp->t_flags & TF_NOOPT;
2378 			INP_WUNLOCK(inp);
2379 			error = sooptcopyout(sopt, &optval, sizeof optval);
2380 			break;
2381 		case TCP_NOPUSH:
2382 			optval = tp->t_flags & TF_NOPUSH;
2383 			INP_WUNLOCK(inp);
2384 			error = sooptcopyout(sopt, &optval, sizeof optval);
2385 			break;
2386 		case TCP_INFO:
2387 			tcp_fill_info(tp, &ti);
2388 			INP_WUNLOCK(inp);
2389 			error = sooptcopyout(sopt, &ti, sizeof ti);
2390 			break;
2391 		case TCP_STATS:
2392 			{
2393 #ifdef STATS
2394 			int nheld;
2395 			TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2396 
2397 			error = 0;
2398 			socklen_t outsbsz = sopt->sopt_valsize;
2399 			if (tp->t_stats == NULL)
2400 				error = ENOENT;
2401 			else if (outsbsz >= tp->t_stats->cursz)
2402 				outsbsz = tp->t_stats->cursz;
2403 			else if (outsbsz >= sizeof(struct statsblob))
2404 				outsbsz = sizeof(struct statsblob);
2405 			else
2406 				error = EINVAL;
2407 			INP_WUNLOCK(inp);
2408 			if (error)
2409 				break;
2410 
2411 			sbp = sopt->sopt_val;
2412 			nheld = atop(round_page(((vm_offset_t)sbp) +
2413 			    (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2414 			vm_page_t ma[nheld];
2415 			if (vm_fault_quick_hold_pages(
2416 			    &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2417 			    outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2418 			    nheld) < 0) {
2419 				error = EFAULT;
2420 				break;
2421 			}
2422 
2423 			if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2424 			    SIZEOF_MEMBER(struct statsblob, flags))))
2425 				goto unhold;
2426 
2427 			INP_WLOCK_RECHECK(inp);
2428 			error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2429 			    sbflags | SB_CLONE_USRDSTNOFAULT);
2430 			INP_WUNLOCK(inp);
2431 			sopt->sopt_valsize = outsbsz;
2432 unhold:
2433 			vm_page_unhold_pages(ma, nheld);
2434 #else
2435 			INP_WUNLOCK(inp);
2436 			error = EOPNOTSUPP;
2437 #endif /* !STATS */
2438 			break;
2439 			}
2440 		case TCP_CONGESTION:
2441 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2442 			INP_WUNLOCK(inp);
2443 			error = sooptcopyout(sopt, buf, len + 1);
2444 			break;
2445 		case TCP_KEEPIDLE:
2446 		case TCP_KEEPINTVL:
2447 		case TCP_KEEPINIT:
2448 		case TCP_KEEPCNT:
2449 			switch (sopt->sopt_name) {
2450 			case TCP_KEEPIDLE:
2451 				ui = TP_KEEPIDLE(tp) / hz;
2452 				break;
2453 			case TCP_KEEPINTVL:
2454 				ui = TP_KEEPINTVL(tp) / hz;
2455 				break;
2456 			case TCP_KEEPINIT:
2457 				ui = TP_KEEPINIT(tp) / hz;
2458 				break;
2459 			case TCP_KEEPCNT:
2460 				ui = TP_KEEPCNT(tp);
2461 				break;
2462 			}
2463 			INP_WUNLOCK(inp);
2464 			error = sooptcopyout(sopt, &ui, sizeof(ui));
2465 			break;
2466 #ifdef TCPPCAP
2467 		case TCP_PCAP_OUT:
2468 		case TCP_PCAP_IN:
2469 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2470 					&(tp->t_outpkts) : &(tp->t_inpkts));
2471 			INP_WUNLOCK(inp);
2472 			error = sooptcopyout(sopt, &optval, sizeof optval);
2473 			break;
2474 #endif
2475 		case TCP_FASTOPEN:
2476 			optval = tp->t_flags & TF_FASTOPEN;
2477 			INP_WUNLOCK(inp);
2478 			error = sooptcopyout(sopt, &optval, sizeof optval);
2479 			break;
2480 #ifdef TCP_BLACKBOX
2481 		case TCP_LOG:
2482 			optval = tp->t_logstate;
2483 			INP_WUNLOCK(inp);
2484 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2485 			break;
2486 		case TCP_LOGBUF:
2487 			/* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2488 			error = tcp_log_getlogbuf(sopt, tp);
2489 			break;
2490 		case TCP_LOGID:
2491 			len = tcp_log_get_id(tp, buf);
2492 			INP_WUNLOCK(inp);
2493 			error = sooptcopyout(sopt, buf, len + 1);
2494 			break;
2495 		case TCP_LOGDUMP:
2496 		case TCP_LOGDUMPID:
2497 			INP_WUNLOCK(inp);
2498 			error = EINVAL;
2499 			break;
2500 #endif
2501 #ifdef KERN_TLS
2502 		case TCP_TXTLS_MODE:
2503 			optval = ktls_get_tx_mode(so);
2504 			INP_WUNLOCK(inp);
2505 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2506 			break;
2507 		case TCP_RXTLS_MODE:
2508 			optval = ktls_get_rx_mode(so);
2509 			INP_WUNLOCK(inp);
2510 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2511 			break;
2512 #endif
2513 		default:
2514 			INP_WUNLOCK(inp);
2515 			error = ENOPROTOOPT;
2516 			break;
2517 		}
2518 		break;
2519 	}
2520 	return (error);
2521 }
2522 #undef INP_WLOCK_RECHECK
2523 #undef INP_WLOCK_RECHECK_CLEANUP
2524 
2525 /*
2526  * Initiate (or continue) disconnect.
2527  * If embryonic state, just send reset (once).
2528  * If in ``let data drain'' option and linger null, just drop.
2529  * Otherwise (hard), mark socket disconnecting and drop
2530  * current input data; switch states based on user close, and
2531  * send segment to peer (with FIN).
2532  */
2533 static void
2534 tcp_disconnect(struct tcpcb *tp)
2535 {
2536 	struct inpcb *inp = tp->t_inpcb;
2537 	struct socket *so = inp->inp_socket;
2538 
2539 	NET_EPOCH_ASSERT();
2540 	INP_WLOCK_ASSERT(inp);
2541 
2542 	/*
2543 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2544 	 * socket is still open.
2545 	 */
2546 	if (tp->t_state < TCPS_ESTABLISHED &&
2547 	    !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2548 		tp = tcp_close(tp);
2549 		KASSERT(tp != NULL,
2550 		    ("tcp_disconnect: tcp_close() returned NULL"));
2551 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2552 		tp = tcp_drop(tp, 0);
2553 		KASSERT(tp != NULL,
2554 		    ("tcp_disconnect: tcp_drop() returned NULL"));
2555 	} else {
2556 		soisdisconnecting(so);
2557 		sbflush(&so->so_rcv);
2558 		tcp_usrclosed(tp);
2559 		if (!(inp->inp_flags & INP_DROPPED))
2560 			tp->t_fb->tfb_tcp_output(tp);
2561 	}
2562 }
2563 
2564 /*
2565  * User issued close, and wish to trail through shutdown states:
2566  * if never received SYN, just forget it.  If got a SYN from peer,
2567  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2568  * If already got a FIN from peer, then almost done; go to LAST_ACK
2569  * state.  In all other cases, have already sent FIN to peer (e.g.
2570  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2571  * for peer to send FIN or not respond to keep-alives, etc.
2572  * We can let the user exit from the close as soon as the FIN is acked.
2573  */
2574 static void
2575 tcp_usrclosed(struct tcpcb *tp)
2576 {
2577 
2578 	NET_EPOCH_ASSERT();
2579 	INP_WLOCK_ASSERT(tp->t_inpcb);
2580 
2581 	switch (tp->t_state) {
2582 	case TCPS_LISTEN:
2583 #ifdef TCP_OFFLOAD
2584 		tcp_offload_listen_stop(tp);
2585 #endif
2586 		tcp_state_change(tp, TCPS_CLOSED);
2587 		/* FALLTHROUGH */
2588 	case TCPS_CLOSED:
2589 		tp = tcp_close(tp);
2590 		/*
2591 		 * tcp_close() should never return NULL here as the socket is
2592 		 * still open.
2593 		 */
2594 		KASSERT(tp != NULL,
2595 		    ("tcp_usrclosed: tcp_close() returned NULL"));
2596 		break;
2597 
2598 	case TCPS_SYN_SENT:
2599 	case TCPS_SYN_RECEIVED:
2600 		tp->t_flags |= TF_NEEDFIN;
2601 		break;
2602 
2603 	case TCPS_ESTABLISHED:
2604 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
2605 		break;
2606 
2607 	case TCPS_CLOSE_WAIT:
2608 		tcp_state_change(tp, TCPS_LAST_ACK);
2609 		break;
2610 	}
2611 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
2612 		soisdisconnected(tp->t_inpcb->inp_socket);
2613 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
2614 		if (tp->t_state == TCPS_FIN_WAIT_2) {
2615 			int timeout;
2616 
2617 			timeout = (tcp_fast_finwait2_recycle) ?
2618 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
2619 			tcp_timer_activate(tp, TT_2MSL, timeout);
2620 		}
2621 	}
2622 }
2623 
2624 #ifdef DDB
2625 static void
2626 db_print_indent(int indent)
2627 {
2628 	int i;
2629 
2630 	for (i = 0; i < indent; i++)
2631 		db_printf(" ");
2632 }
2633 
2634 static void
2635 db_print_tstate(int t_state)
2636 {
2637 
2638 	switch (t_state) {
2639 	case TCPS_CLOSED:
2640 		db_printf("TCPS_CLOSED");
2641 		return;
2642 
2643 	case TCPS_LISTEN:
2644 		db_printf("TCPS_LISTEN");
2645 		return;
2646 
2647 	case TCPS_SYN_SENT:
2648 		db_printf("TCPS_SYN_SENT");
2649 		return;
2650 
2651 	case TCPS_SYN_RECEIVED:
2652 		db_printf("TCPS_SYN_RECEIVED");
2653 		return;
2654 
2655 	case TCPS_ESTABLISHED:
2656 		db_printf("TCPS_ESTABLISHED");
2657 		return;
2658 
2659 	case TCPS_CLOSE_WAIT:
2660 		db_printf("TCPS_CLOSE_WAIT");
2661 		return;
2662 
2663 	case TCPS_FIN_WAIT_1:
2664 		db_printf("TCPS_FIN_WAIT_1");
2665 		return;
2666 
2667 	case TCPS_CLOSING:
2668 		db_printf("TCPS_CLOSING");
2669 		return;
2670 
2671 	case TCPS_LAST_ACK:
2672 		db_printf("TCPS_LAST_ACK");
2673 		return;
2674 
2675 	case TCPS_FIN_WAIT_2:
2676 		db_printf("TCPS_FIN_WAIT_2");
2677 		return;
2678 
2679 	case TCPS_TIME_WAIT:
2680 		db_printf("TCPS_TIME_WAIT");
2681 		return;
2682 
2683 	default:
2684 		db_printf("unknown");
2685 		return;
2686 	}
2687 }
2688 
2689 static void
2690 db_print_tflags(u_int t_flags)
2691 {
2692 	int comma;
2693 
2694 	comma = 0;
2695 	if (t_flags & TF_ACKNOW) {
2696 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
2697 		comma = 1;
2698 	}
2699 	if (t_flags & TF_DELACK) {
2700 		db_printf("%sTF_DELACK", comma ? ", " : "");
2701 		comma = 1;
2702 	}
2703 	if (t_flags & TF_NODELAY) {
2704 		db_printf("%sTF_NODELAY", comma ? ", " : "");
2705 		comma = 1;
2706 	}
2707 	if (t_flags & TF_NOOPT) {
2708 		db_printf("%sTF_NOOPT", comma ? ", " : "");
2709 		comma = 1;
2710 	}
2711 	if (t_flags & TF_SENTFIN) {
2712 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2713 		comma = 1;
2714 	}
2715 	if (t_flags & TF_REQ_SCALE) {
2716 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2717 		comma = 1;
2718 	}
2719 	if (t_flags & TF_RCVD_SCALE) {
2720 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2721 		comma = 1;
2722 	}
2723 	if (t_flags & TF_REQ_TSTMP) {
2724 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2725 		comma = 1;
2726 	}
2727 	if (t_flags & TF_RCVD_TSTMP) {
2728 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2729 		comma = 1;
2730 	}
2731 	if (t_flags & TF_SACK_PERMIT) {
2732 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2733 		comma = 1;
2734 	}
2735 	if (t_flags & TF_NEEDSYN) {
2736 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2737 		comma = 1;
2738 	}
2739 	if (t_flags & TF_NEEDFIN) {
2740 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2741 		comma = 1;
2742 	}
2743 	if (t_flags & TF_NOPUSH) {
2744 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2745 		comma = 1;
2746 	}
2747 	if (t_flags & TF_MORETOCOME) {
2748 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2749 		comma = 1;
2750 	}
2751 	if (t_flags & TF_LQ_OVERFLOW) {
2752 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2753 		comma = 1;
2754 	}
2755 	if (t_flags & TF_LASTIDLE) {
2756 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2757 		comma = 1;
2758 	}
2759 	if (t_flags & TF_RXWIN0SENT) {
2760 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2761 		comma = 1;
2762 	}
2763 	if (t_flags & TF_FASTRECOVERY) {
2764 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2765 		comma = 1;
2766 	}
2767 	if (t_flags & TF_CONGRECOVERY) {
2768 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2769 		comma = 1;
2770 	}
2771 	if (t_flags & TF_WASFRECOVERY) {
2772 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2773 		comma = 1;
2774 	}
2775 	if (t_flags & TF_SIGNATURE) {
2776 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2777 		comma = 1;
2778 	}
2779 	if (t_flags & TF_FORCEDATA) {
2780 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2781 		comma = 1;
2782 	}
2783 	if (t_flags & TF_TSO) {
2784 		db_printf("%sTF_TSO", comma ? ", " : "");
2785 		comma = 1;
2786 	}
2787 	if (t_flags & TF_FASTOPEN) {
2788 		db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2789 		comma = 1;
2790 	}
2791 }
2792 
2793 static void
2794 db_print_tflags2(u_int t_flags2)
2795 {
2796 	int comma;
2797 
2798 	comma = 0;
2799 	if (t_flags2 & TF2_ECN_PERMIT) {
2800 		db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2801 		comma = 1;
2802 	}
2803 }
2804 
2805 
2806 static void
2807 db_print_toobflags(char t_oobflags)
2808 {
2809 	int comma;
2810 
2811 	comma = 0;
2812 	if (t_oobflags & TCPOOB_HAVEDATA) {
2813 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2814 		comma = 1;
2815 	}
2816 	if (t_oobflags & TCPOOB_HADDATA) {
2817 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2818 		comma = 1;
2819 	}
2820 }
2821 
2822 static void
2823 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2824 {
2825 
2826 	db_print_indent(indent);
2827 	db_printf("%s at %p\n", name, tp);
2828 
2829 	indent += 2;
2830 
2831 	db_print_indent(indent);
2832 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2833 	   TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2834 
2835 	db_print_indent(indent);
2836 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2837 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2838 
2839 	db_print_indent(indent);
2840 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2841 	    &tp->t_timers->tt_delack, tp->t_inpcb);
2842 
2843 	db_print_indent(indent);
2844 	db_printf("t_state: %d (", tp->t_state);
2845 	db_print_tstate(tp->t_state);
2846 	db_printf(")\n");
2847 
2848 	db_print_indent(indent);
2849 	db_printf("t_flags: 0x%x (", tp->t_flags);
2850 	db_print_tflags(tp->t_flags);
2851 	db_printf(")\n");
2852 
2853 	db_print_indent(indent);
2854 	db_printf("t_flags2: 0x%x (", tp->t_flags2);
2855 	db_print_tflags2(tp->t_flags2);
2856 	db_printf(")\n");
2857 
2858 	db_print_indent(indent);
2859 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2860 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2861 
2862 	db_print_indent(indent);
2863 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2864 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2865 
2866 	db_print_indent(indent);
2867 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2868 	    tp->iss, tp->irs, tp->rcv_nxt);
2869 
2870 	db_print_indent(indent);
2871 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2872 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2873 
2874 	db_print_indent(indent);
2875 	db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2876 	   tp->snd_wnd, tp->snd_cwnd);
2877 
2878 	db_print_indent(indent);
2879 	db_printf("snd_ssthresh: %u   snd_recover: "
2880 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2881 
2882 	db_print_indent(indent);
2883 	db_printf("t_rcvtime: %u   t_startime: %u\n",
2884 	    tp->t_rcvtime, tp->t_starttime);
2885 
2886 	db_print_indent(indent);
2887 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2888 	    tp->t_rtttime, tp->t_rtseq);
2889 
2890 	db_print_indent(indent);
2891 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2892 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2893 
2894 	db_print_indent(indent);
2895 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2896 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2897 	    tp->t_rttbest);
2898 
2899 	db_print_indent(indent);
2900 	db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2901 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2902 
2903 	db_print_indent(indent);
2904 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2905 	db_print_toobflags(tp->t_oobflags);
2906 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2907 
2908 	db_print_indent(indent);
2909 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2910 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2911 
2912 	db_print_indent(indent);
2913 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2914 	    tp->ts_recent, tp->ts_recent_age);
2915 
2916 	db_print_indent(indent);
2917 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2918 	    "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2919 
2920 	db_print_indent(indent);
2921 	db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2922 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2923 	    tp->snd_recover_prev, tp->t_badrxtwin);
2924 
2925 	db_print_indent(indent);
2926 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2927 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2928 
2929 	db_print_indent(indent);
2930 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
2931 	    tp->snd_fack, tp->rcv_numsacks);
2932 
2933 	/* Skip sackblks, sackhint. */
2934 
2935 	db_print_indent(indent);
2936 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2937 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2938 }
2939 
2940 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2941 {
2942 	struct tcpcb *tp;
2943 
2944 	if (!have_addr) {
2945 		db_printf("usage: show tcpcb <addr>\n");
2946 		return;
2947 	}
2948 	tp = (struct tcpcb *)addr;
2949 
2950 	db_print_tcpcb(tp, "tcpcb", 0);
2951 }
2952 #endif
2953