xref: /freebsd/sys/netinet/tcp_usrreq.c (revision a3cefe7f2b4df0f70ff92d4570ce18e517af43ec)
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 
38 #include "opt_ddb.h"
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
42 #include "opt_kern_tls.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/arb.h>
47 #include <sys/limits.h>
48 #include <sys/malloc.h>
49 #include <sys/refcount.h>
50 #include <sys/kernel.h>
51 #include <sys/ktls.h>
52 #include <sys/qmath.h>
53 #include <sys/sysctl.h>
54 #include <sys/mbuf.h>
55 #ifdef INET6
56 #include <sys/domain.h>
57 #endif /* INET6 */
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/protosw.h>
61 #include <sys/proc.h>
62 #include <sys/jail.h>
63 #include <sys/stats.h>
64 
65 #ifdef DDB
66 #include <ddb/ddb.h>
67 #endif
68 
69 #include <net/if.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 #include <net/vnet.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/in_rss.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #include <netinet/ip_var.h>
82 #ifdef INET6
83 #include <netinet/ip6.h>
84 #include <netinet6/in6_pcb.h>
85 #include <netinet6/in6_rss.h>
86 #include <netinet6/ip6_var.h>
87 #include <netinet6/scope6_var.h>
88 #endif
89 #include <netinet/tcp.h>
90 #include <netinet/tcp_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/tcp_log_buf.h>
95 #include <netinet/tcpip.h>
96 #include <netinet/cc/cc.h>
97 #include <netinet/tcp_fastopen.h>
98 #include <netinet/tcp_hpts.h>
99 #ifdef TCP_OFFLOAD
100 #include <netinet/tcp_offload.h>
101 #endif
102 #include <netipsec/ipsec_support.h>
103 
104 #include <vm/vm.h>
105 #include <vm/vm_param.h>
106 #include <vm/pmap.h>
107 #include <vm/vm_extern.h>
108 #include <vm/vm_map.h>
109 #include <vm/vm_page.h>
110 
111 /*
112  * TCP protocol interface to socket abstraction.
113  */
114 #ifdef INET
115 static int	tcp_connect(struct tcpcb *, struct sockaddr_in *,
116 		    struct thread *td);
117 #endif /* INET */
118 #ifdef INET6
119 static int	tcp6_connect(struct tcpcb *, struct sockaddr_in6 *,
120 		    struct thread *td);
121 #endif /* INET6 */
122 static void	tcp_disconnect(struct tcpcb *);
123 static void	tcp_usrclosed(struct tcpcb *);
124 static void	tcp_fill_info(const struct tcpcb *, struct tcp_info *);
125 
126 static int	tcp_pru_options_support(struct tcpcb *tp, int flags);
127 
128 static void
129 tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error)
130 {
131 	struct tcp_log_buffer *lgb;
132 
133 	KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL"));
134 	INP_WLOCK_ASSERT(tptoinpcb(tp));
135 	if (tcp_bblogging_on(tp)) {
136 		lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error,
137 		    0, NULL, false, NULL, NULL, 0, NULL);
138 	} else {
139 		lgb = NULL;
140 	}
141 	if (lgb != NULL) {
142 		if (error >= 0) {
143 			lgb->tlb_errno = (uint32_t)error;
144 		}
145 		lgb->tlb_flex1 = pru;
146 	}
147 }
148 
149 /*
150  * TCP attaches to socket via pr_attach(), reserving space,
151  * and an internet control block.
152  */
153 static int
154 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
155 {
156 	struct inpcb *inp;
157 	struct tcpcb *tp = NULL;
158 	int error;
159 
160 	inp = sotoinpcb(so);
161 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
162 
163 	error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
164 	if (error)
165 		goto out;
166 
167 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
168 	so->so_snd.sb_flags |= (SB_AUTOLOWAT | SB_AUTOSIZE);
169 	error = in_pcballoc(so, &V_tcbinfo);
170 	if (error)
171 		goto out;
172 	inp = sotoinpcb(so);
173 	tp = tcp_newtcpcb(inp, NULL);
174 	if (tp == NULL) {
175 		error = ENOBUFS;
176 		in_pcbfree(inp);
177 		goto out;
178 	}
179 	tp->t_state = TCPS_CLOSED;
180 	tcp_bblog_pru(tp, PRU_ATTACH, error);
181 	INP_WUNLOCK(inp);
182 	TCPSTATES_INC(TCPS_CLOSED);
183 out:
184 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
185 	return (error);
186 }
187 
188 /*
189  * tcp_usr_detach is called when the socket layer loses its final reference
190  * to the socket, be it a file descriptor reference, a reference from TCP,
191  * etc.  At this point, there is only one case in which we will keep around
192  * inpcb state: time wait.
193  */
194 static void
195 tcp_usr_detach(struct socket *so)
196 {
197 	struct inpcb *inp;
198 	struct tcpcb *tp;
199 
200 	inp = sotoinpcb(so);
201 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
202 	INP_WLOCK(inp);
203 	KASSERT(so->so_pcb == inp && inp->inp_socket == so,
204 		("%s: socket %p inp %p mismatch", __func__, so, inp));
205 
206 	tp = intotcpcb(inp);
207 
208 	KASSERT(tp->t_flags & TF_DISCONNECTED ||
209 	    tp->t_state < TCPS_SYN_SENT,
210 	    ("%s: inp %p not disconnected or embryonic", __func__, inp));
211 
212 	tcp_discardcb(tp);
213 	in_pcbfree(inp);
214 }
215 
216 #ifdef INET
217 /*
218  * Give the socket an address.
219  */
220 static int
221 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
222 {
223 	struct inpcb *inp = sotoinpcb(so);
224 	struct tcpcb *tp = intotcpcb(inp);
225 	struct sockaddr_in *sinp;
226 	int error = 0;
227 
228 	INP_WLOCK(inp);
229 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
230 		INP_WUNLOCK(inp);
231 		return (EINVAL);
232 	}
233 
234 	sinp = (struct sockaddr_in *)nam;
235 	if (nam->sa_family != AF_INET) {
236 		/*
237 		 * Preserve compatibility with old programs.
238 		 */
239 		if (nam->sa_family != AF_UNSPEC ||
240 		    nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
241 		    sinp->sin_addr.s_addr != INADDR_ANY) {
242 			error = EAFNOSUPPORT;
243 			goto out;
244 		}
245 		nam->sa_family = AF_INET;
246 	}
247 	if (nam->sa_len != sizeof(*sinp)) {
248 		error = EINVAL;
249 		goto out;
250 	}
251 	/*
252 	 * Must check for multicast addresses and disallow binding
253 	 * to them.
254 	 */
255 	if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
256 		error = EAFNOSUPPORT;
257 		goto out;
258 	}
259 	error = in_pcbbind(inp, sinp, V_tcp_bind_all_fibs ? 0 : INPBIND_FIB,
260 	    td->td_ucred);
261 out:
262 	tcp_bblog_pru(tp, PRU_BIND, error);
263 	TCP_PROBE2(debug__user, tp, PRU_BIND);
264 	INP_WUNLOCK(inp);
265 
266 	return (error);
267 }
268 #endif /* INET */
269 
270 #ifdef INET6
271 static int
272 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
273 {
274 	struct inpcb *inp = sotoinpcb(so);
275 	struct tcpcb *tp = intotcpcb(inp);
276 	struct sockaddr_in6 *sin6;
277 	int error = 0;
278 	u_char vflagsav;
279 
280 	INP_WLOCK(inp);
281 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
282 		INP_WUNLOCK(inp);
283 		return (EINVAL);
284 	}
285 
286 	vflagsav = inp->inp_vflag;
287 
288 	sin6 = (struct sockaddr_in6 *)nam;
289 	if (nam->sa_family != AF_INET6) {
290 		error = EAFNOSUPPORT;
291 		goto out;
292 	}
293 	if (nam->sa_len != sizeof(*sin6)) {
294 		error = EINVAL;
295 		goto out;
296 	}
297 	/*
298 	 * Must check for multicast addresses and disallow binding
299 	 * to them.
300 	 */
301 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
302 		error = EAFNOSUPPORT;
303 		goto out;
304 	}
305 
306 	inp->inp_vflag &= ~INP_IPV4;
307 	inp->inp_vflag |= INP_IPV6;
308 #ifdef INET
309 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
310 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
311 			inp->inp_vflag |= INP_IPV4;
312 		else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
313 			struct sockaddr_in sin;
314 
315 			in6_sin6_2_sin(&sin, sin6);
316 			if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
317 				error = EAFNOSUPPORT;
318 				goto out;
319 			}
320 			inp->inp_vflag |= INP_IPV4;
321 			inp->inp_vflag &= ~INP_IPV6;
322 			error = in_pcbbind(inp, &sin, 0, td->td_ucred);
323 			goto out;
324 		}
325 	}
326 #endif
327 	error = in6_pcbbind(inp, sin6, V_tcp_bind_all_fibs ? 0 : INPBIND_FIB,
328 	    td->td_ucred);
329 out:
330 	if (error != 0)
331 		inp->inp_vflag = vflagsav;
332 	tcp_bblog_pru(tp, PRU_BIND, error);
333 	TCP_PROBE2(debug__user, tp, PRU_BIND);
334 	INP_WUNLOCK(inp);
335 	return (error);
336 }
337 #endif /* INET6 */
338 
339 #ifdef INET
340 /*
341  * Prepare to accept connections.
342  */
343 static int
344 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
345 {
346 	struct inpcb *inp = sotoinpcb(so);
347 	struct tcpcb *tp = intotcpcb(inp);
348 	int error = 0;
349 	bool already_listening;
350 
351 	INP_WLOCK(inp);
352 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
353 		INP_WUNLOCK(inp);
354 		return (EINVAL);
355 	}
356 
357 	SOCK_LOCK(so);
358 	already_listening = SOLISTENING(so);
359 	error = solisten_proto_check(so);
360 	if (error != 0) {
361 		SOCK_UNLOCK(so);
362 		goto out;
363 	}
364 	if (inp->inp_lport == 0) {
365 		error = in_pcbbind(inp, NULL,
366 		    V_tcp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred);
367 	}
368 	if (error == 0) {
369 		tcp_state_change(tp, TCPS_LISTEN);
370 		solisten_proto(so, backlog);
371 #ifdef TCP_OFFLOAD
372 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
373 			tcp_offload_listen_start(tp);
374 #endif
375 	} else {
376 		solisten_proto_abort(so);
377 	}
378 	SOCK_UNLOCK(so);
379 	if (already_listening)
380 		goto out;
381 
382 	if (error == 0) {
383 		in_pcblisten(inp);
384 		if (tp->t_flags & TF_FASTOPEN)
385 			tp->t_tfo_pending = tcp_fastopen_alloc_counter();
386 	}
387 
388 out:
389 	tcp_bblog_pru(tp, PRU_LISTEN, error);
390 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
391 	INP_WUNLOCK(inp);
392 	return (error);
393 }
394 #endif /* INET */
395 
396 #ifdef INET6
397 static int
398 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
399 {
400 	struct inpcb *inp = sotoinpcb(so);
401 	struct tcpcb *tp = intotcpcb(inp);
402 	u_char vflagsav;
403 	int error = 0;
404 	bool already_listening;
405 
406 	INP_WLOCK(inp);
407 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
408 		INP_WUNLOCK(inp);
409 		return (EINVAL);
410 	}
411 
412 	vflagsav = inp->inp_vflag;
413 
414 	SOCK_LOCK(so);
415 	already_listening = SOLISTENING(so);
416 	error = solisten_proto_check(so);
417 	if (error != 0) {
418 		SOCK_UNLOCK(so);
419 		goto out;
420 	}
421 	if (inp->inp_lport == 0) {
422 		inp->inp_vflag &= ~INP_IPV4;
423 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
424 			inp->inp_vflag |= INP_IPV4;
425 		error = in6_pcbbind(inp, NULL,
426 		    V_tcp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred);
427 	}
428 	if (error == 0) {
429 		tcp_state_change(tp, TCPS_LISTEN);
430 		solisten_proto(so, backlog);
431 #ifdef TCP_OFFLOAD
432 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
433 			tcp_offload_listen_start(tp);
434 #endif
435 	} else {
436 		solisten_proto_abort(so);
437 	}
438 	SOCK_UNLOCK(so);
439 	if (already_listening)
440 		goto out;
441 
442 	if (error == 0) {
443 		in_pcblisten(inp);
444 		if (tp->t_flags & TF_FASTOPEN)
445 			tp->t_tfo_pending = tcp_fastopen_alloc_counter();
446 	} else
447 		inp->inp_vflag = vflagsav;
448 
449 out:
450 	tcp_bblog_pru(tp, PRU_LISTEN, error);
451 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
452 	INP_WUNLOCK(inp);
453 	return (error);
454 }
455 #endif /* INET6 */
456 
457 #ifdef INET
458 /*
459  * Initiate connection to peer.
460  * Create a template for use in transmissions on this connection.
461  * Enter SYN_SENT state, and mark socket as connecting.
462  * Start keep-alive timer, and seed output sequence space.
463  * Send initial segment on connection.
464  */
465 static int
466 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
467 {
468 	struct epoch_tracker et;
469 	struct inpcb *inp = sotoinpcb(so);
470 	struct tcpcb *tp = intotcpcb(inp);
471 	struct sockaddr_in *sinp;
472 	int error = 0;
473 
474 	INP_WLOCK(inp);
475 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
476 		INP_WUNLOCK(inp);
477 		return (ECONNREFUSED);
478 	}
479 
480 	sinp = (struct sockaddr_in *)nam;
481 	if (nam->sa_family != AF_INET) {
482 		error = EAFNOSUPPORT;
483 		goto out;
484 	}
485 	if (nam->sa_len != sizeof (*sinp)) {
486 		error = EINVAL;
487 		goto out;
488 	}
489 	/*
490 	 * Must disallow TCP ``connections'' to multicast addresses.
491 	 */
492 	if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
493 		error = EAFNOSUPPORT;
494 		goto out;
495 	}
496 	if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
497 		error = EACCES;
498 		goto out;
499 	}
500 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
501 		goto out;
502 	if (SOLISTENING(so)) {
503 		error = EOPNOTSUPP;
504 		goto out;
505 	}
506 	NET_EPOCH_ENTER(et);
507 	if ((error = tcp_connect(tp, sinp, td)) != 0)
508 		goto out_in_epoch;
509 #ifdef TCP_OFFLOAD
510 	if (registered_toedevs > 0 &&
511 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
512 	    (error = tcp_offload_connect(so, nam)) == 0)
513 		goto out_in_epoch;
514 #endif
515 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
516 	error = tcp_output(tp);
517 	KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
518 	    ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
519 out_in_epoch:
520 	NET_EPOCH_EXIT(et);
521 out:
522 	tcp_bblog_pru(tp, PRU_CONNECT, error);
523 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
524 	INP_WUNLOCK(inp);
525 	return (error);
526 }
527 #endif /* INET */
528 
529 #ifdef INET6
530 static int
531 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
532 {
533 	struct epoch_tracker et;
534 	struct inpcb *inp = sotoinpcb(so);
535 	struct tcpcb *tp = intotcpcb(inp);
536 	struct sockaddr_in6 *sin6;
537 	int error = 0;
538 	u_int8_t incflagsav;
539 	u_char vflagsav;
540 
541 	INP_WLOCK(inp);
542 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
543 		INP_WUNLOCK(inp);
544 		return (ECONNREFUSED);
545 	}
546 
547 	vflagsav = inp->inp_vflag;
548 	incflagsav = inp->inp_inc.inc_flags;
549 
550 	sin6 = (struct sockaddr_in6 *)nam;
551 	if (nam->sa_family != AF_INET6) {
552 		error = EAFNOSUPPORT;
553 		goto out;
554 	}
555 	if (nam->sa_len != sizeof (*sin6)) {
556 		error = EINVAL;
557 		goto out;
558 	}
559 	/*
560 	 * Must disallow TCP ``connections'' to multicast addresses.
561 	 */
562 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
563 		error = EAFNOSUPPORT;
564 		goto out;
565 	}
566 	if (SOLISTENING(so)) {
567 		error = EOPNOTSUPP;
568 		goto out;
569 	}
570 #ifdef INET
571 	/*
572 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
573 	 * therefore probably require the hash lock, which isn't held here.
574 	 * Is this a significant problem?
575 	 */
576 	if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
577 		struct sockaddr_in sin;
578 
579 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
580 			error = EINVAL;
581 			goto out;
582 		}
583 		if ((inp->inp_vflag & INP_IPV4) == 0) {
584 			error = EAFNOSUPPORT;
585 			goto out;
586 		}
587 
588 		in6_sin6_2_sin(&sin, sin6);
589 		if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
590 			error = EAFNOSUPPORT;
591 			goto out;
592 		}
593 		if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
594 			error = EACCES;
595 			goto out;
596 		}
597 		if ((error = prison_remote_ip4(td->td_ucred,
598 		    &sin.sin_addr)) != 0)
599 			goto out;
600 		inp->inp_vflag |= INP_IPV4;
601 		inp->inp_vflag &= ~INP_IPV6;
602 		NET_EPOCH_ENTER(et);
603 		if ((error = tcp_connect(tp, &sin, td)) != 0)
604 			goto out_in_epoch;
605 #ifdef TCP_OFFLOAD
606 		if (registered_toedevs > 0 &&
607 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
608 		    (error = tcp_offload_connect(so, nam)) == 0)
609 			goto out_in_epoch;
610 #endif
611 		error = tcp_output(tp);
612 		goto out_in_epoch;
613 	} else {
614 		if ((inp->inp_vflag & INP_IPV6) == 0) {
615 			error = EAFNOSUPPORT;
616 			goto out;
617 		}
618 	}
619 #endif
620 	if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
621 		goto out;
622 	inp->inp_vflag &= ~INP_IPV4;
623 	inp->inp_vflag |= INP_IPV6;
624 	inp->inp_inc.inc_flags |= INC_ISIPV6;
625 	NET_EPOCH_ENTER(et);
626 	if ((error = tcp6_connect(tp, sin6, td)) != 0)
627 		goto out_in_epoch;
628 #ifdef TCP_OFFLOAD
629 	if (registered_toedevs > 0 &&
630 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
631 	    (error = tcp_offload_connect(so, nam)) == 0)
632 		goto out_in_epoch;
633 #endif
634 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
635 	error = tcp_output(tp);
636 out_in_epoch:
637 	NET_EPOCH_EXIT(et);
638 out:
639 	KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
640 	    ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
641 	/*
642 	 * If the implicit bind in the connect call fails, restore
643 	 * the flags we modified.
644 	 */
645 	if (error != 0 && inp->inp_lport == 0) {
646 		inp->inp_vflag = vflagsav;
647 		inp->inp_inc.inc_flags = incflagsav;
648 	}
649 
650 	tcp_bblog_pru(tp, PRU_CONNECT, error);
651 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
652 	INP_WUNLOCK(inp);
653 	return (error);
654 }
655 #endif /* INET6 */
656 
657 /*
658  * Initiate disconnect from peer.
659  * If connection never passed embryonic stage, just drop;
660  * else if don't need to let data drain, then can just drop anyways,
661  * else have to begin TCP shutdown process: mark socket disconnecting,
662  * drain unread data, state switch to reflect user close, and
663  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
664  * when peer sends FIN and acks ours.
665  *
666  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
667  */
668 static int
669 tcp_usr_disconnect(struct socket *so)
670 {
671 	struct inpcb *inp;
672 	struct tcpcb *tp = NULL;
673 	struct epoch_tracker et;
674 
675 	NET_EPOCH_ENTER(et);
676 	inp = sotoinpcb(so);
677 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
678 	INP_WLOCK(inp);
679 	tp = intotcpcb(inp);
680 
681 	if (tp->t_state == TCPS_TIME_WAIT)
682 		goto out;
683 	tcp_disconnect(tp);
684 out:
685 	tcp_bblog_pru(tp, PRU_DISCONNECT, 0);
686 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
687 	INP_WUNLOCK(inp);
688 	NET_EPOCH_EXIT(et);
689 	return (0);
690 }
691 
692 #ifdef INET
693 /*
694  * Accept a connection.  Essentially all the work is done at higher levels;
695  * just return the address of the peer, storing through addr.
696  */
697 static int
698 tcp_usr_accept(struct socket *so, struct sockaddr *sa)
699 {
700 	struct inpcb *inp = sotoinpcb(so);
701 	struct tcpcb *tp = intotcpcb(inp);
702 	int error = 0;
703 
704 	INP_WLOCK(inp);
705 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
706 		INP_WUNLOCK(inp);
707 		return (ECONNABORTED);
708 	}
709 
710 	if (so->so_state & SS_ISDISCONNECTED)
711 		error = ECONNABORTED;
712 	else
713 		*(struct sockaddr_in *)sa = (struct sockaddr_in ){
714 			.sin_family = AF_INET,
715 			.sin_len = sizeof(struct sockaddr_in),
716 			.sin_port = inp->inp_fport,
717 			.sin_addr = inp->inp_faddr,
718 		};
719 	tcp_bblog_pru(tp, PRU_ACCEPT, error);
720 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
721 	INP_WUNLOCK(inp);
722 
723 	return (error);
724 }
725 #endif /* INET */
726 
727 #ifdef INET6
728 static int
729 tcp6_usr_accept(struct socket *so, struct sockaddr *sa)
730 {
731 	struct inpcb *inp = sotoinpcb(so);
732 	struct tcpcb *tp = intotcpcb(inp);
733 	int error = 0;
734 
735 	INP_WLOCK(inp);
736 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
737 		INP_WUNLOCK(inp);
738 		return (ECONNABORTED);
739 	}
740 
741 	if (so->so_state & SS_ISDISCONNECTED) {
742 		error = ECONNABORTED;
743 	} else {
744 		if (inp->inp_vflag & INP_IPV4) {
745 			struct sockaddr_in sin = {
746 				.sin_family = AF_INET,
747 				.sin_len = sizeof(struct sockaddr_in),
748 				.sin_port = inp->inp_fport,
749 				.sin_addr = inp->inp_faddr,
750 			};
751 			in6_sin_2_v4mapsin6(&sin, (struct sockaddr_in6 *)sa);
752 		} else {
753 			*(struct sockaddr_in6 *)sa = (struct sockaddr_in6 ){
754 				.sin6_family = AF_INET6,
755 				.sin6_len = sizeof(struct sockaddr_in6),
756 				.sin6_port = inp->inp_fport,
757 				.sin6_addr = inp->in6p_faddr,
758 			};
759 			/* XXX: should catch errors */
760 			(void)sa6_recoverscope((struct sockaddr_in6 *)sa);
761 		}
762 	}
763 
764 	tcp_bblog_pru(tp, PRU_ACCEPT, error);
765 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
766 	INP_WUNLOCK(inp);
767 
768 	return (error);
769 }
770 #endif /* INET6 */
771 
772 /*
773  * Mark the connection as being incapable of further output.
774  */
775 static int
776 tcp_usr_shutdown(struct socket *so, enum shutdown_how how)
777 {
778 	struct epoch_tracker et;
779 	struct inpcb *inp = sotoinpcb(so);
780 	struct tcpcb *tp = intotcpcb(inp);
781 	int error = 0;
782 
783 	SOCK_LOCK(so);
784 	if (SOLISTENING(so)) {
785 		if (how != SHUT_WR) {
786 			so->so_error = ECONNABORTED;
787 			solisten_wakeup(so);	/* unlocks so */
788 		} else
789 			SOCK_UNLOCK(so);
790 		return (ENOTCONN);
791 	} else if ((so->so_state &
792 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
793 		SOCK_UNLOCK(so);
794 		return (ENOTCONN);
795 	}
796 	SOCK_UNLOCK(so);
797 
798 	switch (how) {
799 	case SHUT_RD:
800 		sorflush(so);
801 		break;
802 	case SHUT_RDWR:
803 		sorflush(so);
804 		/* FALLTHROUGH */
805 	case SHUT_WR:
806 		/*
807 		 * XXXGL: mimicing old soshutdown() here. But shouldn't we
808 		 * return ECONNRESEST for SHUT_RD as well?
809 		 */
810 		INP_WLOCK(inp);
811 		if (tp->t_flags & TF_DISCONNECTED) {
812 			INP_WUNLOCK(inp);
813 			return (ECONNRESET);
814 		}
815 
816 		socantsendmore(so);
817 		NET_EPOCH_ENTER(et);
818 		tcp_usrclosed(tp);
819 		error = tcp_output_nodrop(tp);
820 		tcp_bblog_pru(tp, PRU_SHUTDOWN, error);
821 		TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
822 		error = tcp_unlock_or_drop(tp, error);
823 		NET_EPOCH_EXIT(et);
824 	}
825 	wakeup(&so->so_timeo);
826 
827 	return (error);
828 }
829 
830 /*
831  * After a receive, possibly send window update to peer.
832  */
833 static int
834 tcp_usr_rcvd(struct socket *so, int flags)
835 {
836 	struct epoch_tracker et;
837 	struct inpcb *inp = sotoinpcb(so);
838 	struct tcpcb *tp = intotcpcb(inp);
839 	int outrv = 0, error = 0;
840 
841 	INP_WLOCK(inp);
842 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
843 		/* XXXGL: how could this happen?! */
844 		INP_WUNLOCK(inp);
845 		return (ECONNRESET);
846 	}
847 
848 	NET_EPOCH_ENTER(et);
849 	/*
850 	 * For passively-created TFO connections, don't attempt a window
851 	 * update while still in SYN_RECEIVED as this may trigger an early
852 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
853 	 * application response data, or failing that, when the DELACK timer
854 	 * expires.
855 	 */
856 	if ((tp->t_flags & TF_FASTOPEN) && (tp->t_state == TCPS_SYN_RECEIVED))
857 		goto out;
858 #ifdef TCP_OFFLOAD
859 	if (tp->t_flags & TF_TOE)
860 		tcp_offload_rcvd(tp);
861 	else
862 #endif
863 		outrv = tcp_output_nodrop(tp);
864 out:
865 	tcp_bblog_pru(tp, PRU_RCVD, error);
866 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
867 	(void) tcp_unlock_or_drop(tp, outrv);
868 	NET_EPOCH_EXIT(et);
869 	return (error);
870 }
871 
872 /*
873  * Do a send by putting data in output queue and updating urgent
874  * marker if URG set.  Possibly send more data.  Unlike the other
875  * pr_*() routines, the mbuf chains are our responsibility.  We
876  * must either enqueue them or free them.  The other pr_*() routines
877  * generally are caller-frees.
878  */
879 static int
880 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
881     struct sockaddr *nam, struct mbuf *control, struct thread *td)
882 {
883 	struct epoch_tracker et;
884 	struct inpcb *inp = sotoinpcb(so);
885 	struct tcpcb *tp = intotcpcb(inp);
886 #ifdef INET
887 #ifdef INET6
888 	struct sockaddr_in sin;
889 #endif
890 	struct sockaddr_in *sinp;
891 #endif
892 #ifdef INET6
893 	struct sockaddr_in6 *sin6;
894 	int isipv6;
895 #endif
896 	int error = 0;
897 	u_int8_t incflagsav;
898 	u_char vflagsav;
899 	bool restoreflags;
900 
901 	INP_WLOCK(inp);
902 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
903 		if (m != NULL && (flags & PRUS_NOTREADY) == 0)
904 			m_freem(m);
905 		INP_WUNLOCK(inp);
906 		return (ECONNRESET);
907 	}
908 
909 	vflagsav = inp->inp_vflag;
910 	incflagsav = inp->inp_inc.inc_flags;
911 	restoreflags = false;
912 
913 	NET_EPOCH_ENTER(et);
914 	if (control != NULL) {
915 		/* TCP doesn't do control messages (rights, creds, etc) */
916 		if (control->m_len > 0) {
917 			m_freem(control);
918 			error = EINVAL;
919 			goto out;
920 		}
921 		m_freem(control);	/* empty control, just free it */
922 	}
923 
924 	if ((flags & PRUS_OOB) != 0 &&
925 	    (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
926 		goto out;
927 
928 	if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
929 		if (tp->t_state == TCPS_LISTEN) {
930 			error = EINVAL;
931 			goto out;
932 		}
933 		switch (nam->sa_family) {
934 #ifdef INET
935 		case AF_INET:
936 			sinp = (struct sockaddr_in *)nam;
937 			if (sinp->sin_len != sizeof(struct sockaddr_in)) {
938 				error = EINVAL;
939 				goto out;
940 			}
941 			if ((inp->inp_vflag & INP_IPV6) != 0) {
942 				error = EAFNOSUPPORT;
943 				goto out;
944 			}
945 			if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
946 				error = EAFNOSUPPORT;
947 				goto out;
948 			}
949 			if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
950 				error = EACCES;
951 				goto out;
952 			}
953 			if ((error = prison_remote_ip4(td->td_ucred,
954 			    &sinp->sin_addr)))
955 				goto out;
956 #ifdef INET6
957 			isipv6 = 0;
958 #endif
959 			break;
960 #endif /* INET */
961 #ifdef INET6
962 		case AF_INET6:
963 			sin6 = (struct sockaddr_in6 *)nam;
964 			if (sin6->sin6_len != sizeof(*sin6)) {
965 				error = EINVAL;
966 				goto out;
967 			}
968 			if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
969 				error = EAFNOSUPPORT;
970 				goto out;
971 			}
972 			if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
973 				error = EAFNOSUPPORT;
974 				goto out;
975 			}
976 			if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
977 #ifdef INET
978 				if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
979 					error = EINVAL;
980 					goto out;
981 				}
982 				if ((inp->inp_vflag & INP_IPV4) == 0) {
983 					error = EAFNOSUPPORT;
984 					goto out;
985 				}
986 				restoreflags = true;
987 				inp->inp_vflag &= ~INP_IPV6;
988 				sinp = &sin;
989 				in6_sin6_2_sin(sinp, sin6);
990 				if (IN_MULTICAST(
991 				    ntohl(sinp->sin_addr.s_addr))) {
992 					error = EAFNOSUPPORT;
993 					goto out;
994 				}
995 				if ((error = prison_remote_ip4(td->td_ucred,
996 				    &sinp->sin_addr)))
997 					goto out;
998 				isipv6 = 0;
999 #else /* !INET */
1000 				error = EAFNOSUPPORT;
1001 				goto out;
1002 #endif /* INET */
1003 			} else {
1004 				if ((inp->inp_vflag & INP_IPV6) == 0) {
1005 					error = EAFNOSUPPORT;
1006 					goto out;
1007 				}
1008 				restoreflags = true;
1009 				inp->inp_vflag &= ~INP_IPV4;
1010 				inp->inp_inc.inc_flags |= INC_ISIPV6;
1011 				if ((error = prison_remote_ip6(td->td_ucred,
1012 				    &sin6->sin6_addr)))
1013 					goto out;
1014 				isipv6 = 1;
1015 			}
1016 			break;
1017 #endif /* INET6 */
1018 		default:
1019 			error = EAFNOSUPPORT;
1020 			goto out;
1021 		}
1022 	}
1023 	if (!(flags & PRUS_OOB)) {
1024 		if (tp->t_acktime == 0)
1025 			tp->t_acktime = ticks;
1026 		sbappendstream(&so->so_snd, m, flags);
1027 		m = NULL;
1028 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1029 			KASSERT(tp->t_state == TCPS_CLOSED,
1030 			    ("%s: tp %p is listening", __func__, tp));
1031 
1032 			/*
1033 			 * Do implied connect if not yet connected,
1034 			 * initialize window to default value, and
1035 			 * initialize maxseg using peer's cached MSS.
1036 			 */
1037 #ifdef INET6
1038 			if (isipv6)
1039 				error = tcp6_connect(tp, sin6, td);
1040 #endif /* INET6 */
1041 #if defined(INET6) && defined(INET)
1042 			else
1043 #endif
1044 #ifdef INET
1045 				error = tcp_connect(tp, sinp, td);
1046 #endif
1047 			/*
1048 			 * The bind operation in tcp_connect succeeded. We
1049 			 * no longer want to restore the flags if later
1050 			 * operations fail.
1051 			 */
1052 			if (error == 0 || inp->inp_lport != 0)
1053 				restoreflags = false;
1054 
1055 			if (error) {
1056 				/* m is freed if PRUS_NOTREADY is unset. */
1057 				sbflush(&so->so_snd);
1058 				goto out;
1059 			}
1060 			if (tp->t_flags & TF_FASTOPEN)
1061 				tcp_fastopen_connect(tp);
1062 			else {
1063 				tp->snd_wnd = TTCP_CLIENT_SND_WND;
1064 				tcp_mss(tp, -1);
1065 			}
1066 		}
1067 		if (flags & PRUS_EOF) {
1068 			/*
1069 			 * Close the send side of the connection after
1070 			 * the data is sent.
1071 			 */
1072 			socantsendmore(so);
1073 			tcp_usrclosed(tp);
1074 		}
1075 		if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1076 		    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1077 		    (tp->t_fbyte_out == 0) &&
1078 		    (so->so_snd.sb_ccc > 0)) {
1079 			tp->t_fbyte_out = ticks;
1080 			if (tp->t_fbyte_out == 0)
1081 				tp->t_fbyte_out = 1;
1082 			if (tp->t_fbyte_out && tp->t_fbyte_in)
1083 				tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1084 		}
1085 		if (!(flags & PRUS_NOTREADY)) {
1086 			if (flags & PRUS_MORETOCOME)
1087 				tp->t_flags |= TF_MORETOCOME;
1088 			error = tcp_output_nodrop(tp);
1089 			if (flags & PRUS_MORETOCOME)
1090 				tp->t_flags &= ~TF_MORETOCOME;
1091 		}
1092 	} else {
1093 		/*
1094 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1095 		 */
1096 		SOCK_SENDBUF_LOCK(so);
1097 		if (sbspace(&so->so_snd) < -512) {
1098 			SOCK_SENDBUF_UNLOCK(so);
1099 			error = ENOBUFS;
1100 			goto out;
1101 		}
1102 		/*
1103 		 * According to RFC961 (Assigned Protocols),
1104 		 * the urgent pointer points to the last octet
1105 		 * of urgent data.  We continue, however,
1106 		 * to consider it to indicate the first octet
1107 		 * of data past the urgent section.
1108 		 * Otherwise, snd_up should be one lower.
1109 		 */
1110 		if (tp->t_acktime == 0)
1111 			tp->t_acktime = ticks;
1112 		sbappendstream_locked(&so->so_snd, m, flags);
1113 		SOCK_SENDBUF_UNLOCK(so);
1114 		m = NULL;
1115 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1116 			/*
1117 			 * Do implied connect if not yet connected,
1118 			 * initialize window to default value, and
1119 			 * initialize maxseg using peer's cached MSS.
1120 			 */
1121 
1122 			/*
1123 			 * Not going to contemplate SYN|URG
1124 			 */
1125 			if (tp->t_flags & TF_FASTOPEN)
1126 				tp->t_flags &= ~TF_FASTOPEN;
1127 #ifdef INET6
1128 			if (isipv6)
1129 				error = tcp6_connect(tp, sin6, td);
1130 #endif /* INET6 */
1131 #if defined(INET6) && defined(INET)
1132 			else
1133 #endif
1134 #ifdef INET
1135 				error = tcp_connect(tp, sinp, td);
1136 #endif
1137 			/*
1138 			 * The bind operation in tcp_connect succeeded. We
1139 			 * no longer want to restore the flags if later
1140 			 * operations fail.
1141 			 */
1142 			if (error == 0 || inp->inp_lport != 0)
1143 				restoreflags = false;
1144 
1145 			if (error != 0) {
1146 				/* m is freed if PRUS_NOTREADY is unset. */
1147 				sbflush(&so->so_snd);
1148 				goto out;
1149 			}
1150 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
1151 			tcp_mss(tp, -1);
1152 		}
1153 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1154 		if ((flags & PRUS_NOTREADY) == 0) {
1155 			tp->t_flags |= TF_FORCEDATA;
1156 			error = tcp_output_nodrop(tp);
1157 			tp->t_flags &= ~TF_FORCEDATA;
1158 		}
1159 	}
1160 	TCP_LOG_EVENT(tp, NULL,
1161 	    &inp->inp_socket->so_rcv,
1162 	    &inp->inp_socket->so_snd,
1163 	    TCP_LOG_USERSEND, error,
1164 	    0, NULL, false);
1165 
1166 out:
1167 	/*
1168 	 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1169 	 * responsible for freeing memory.
1170 	 */
1171 	if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1172 		m_freem(m);
1173 
1174 	/*
1175 	 * If the request was unsuccessful and we changed flags,
1176 	 * restore the original flags.
1177 	 */
1178 	if (error != 0 && restoreflags) {
1179 		inp->inp_vflag = vflagsav;
1180 		inp->inp_inc.inc_flags = incflagsav;
1181 	}
1182 	tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1183 		      ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error);
1184 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1185 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1186 	error = tcp_unlock_or_drop(tp, error);
1187 	NET_EPOCH_EXIT(et);
1188 	return (error);
1189 }
1190 
1191 static int
1192 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1193 {
1194 	struct epoch_tracker et;
1195 	struct inpcb *inp = sotoinpcb(so);
1196 	struct tcpcb *tp = intotcpcb(inp);
1197 	int error;
1198 
1199 	INP_WLOCK(inp);
1200 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
1201 		INP_WUNLOCK(inp);
1202 		mb_free_notready(m, count);
1203 		return (ECONNRESET);
1204 	}
1205 
1206 	SOCK_SENDBUF_LOCK(so);
1207 	error = sbready(&so->so_snd, m, count);
1208 	SOCK_SENDBUF_UNLOCK(so);
1209 	if (error) {
1210 		INP_WUNLOCK(inp);
1211 		return (error);
1212 	}
1213 	NET_EPOCH_ENTER(et);
1214 	error = tcp_output_unlock(tp);
1215 	NET_EPOCH_EXIT(et);
1216 
1217 	return (error);
1218 }
1219 
1220 /*
1221  * Abort the TCP.  Drop the connection abruptly.
1222  */
1223 static void
1224 tcp_usr_abort(struct socket *so)
1225 {
1226 	struct epoch_tracker et;
1227 	struct inpcb *inp = sotoinpcb(so);
1228 	struct tcpcb *tp = intotcpcb(inp);
1229 
1230 	/*
1231 	 * If we still have full TCP state, and we're not dropped, drop.
1232 	 */
1233 	NET_EPOCH_ENTER(et);
1234 	INP_WLOCK(inp);
1235 	if (!(tp->t_flags & TF_DISCONNECTED)) {
1236 		tp = tcp_drop(tp, ECONNABORTED);
1237 		if (tp == NULL)
1238 			goto dropped;
1239 		tcp_bblog_pru(tp, PRU_ABORT, 0);
1240 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1241 	}
1242 	if (!(tp->t_flags & TF_DISCONNECTED)) {
1243 		soref(so);
1244 		inp->inp_flags |= INP_SOCKREF;
1245 	}
1246 	INP_WUNLOCK(inp);
1247 dropped:
1248 	NET_EPOCH_EXIT(et);
1249 }
1250 
1251 /*
1252  * TCP socket is closed.  Start friendly disconnect.
1253  */
1254 static void
1255 tcp_usr_close(struct socket *so)
1256 {
1257 	struct epoch_tracker et;
1258 	struct inpcb *inp = sotoinpcb(so);
1259 	struct tcpcb *tp = intotcpcb(inp);
1260 
1261 	/*
1262 	 * If we are still connected and we're not dropped, initiate
1263 	 * a disconnect.
1264 	 */
1265 	NET_EPOCH_ENTER(et);
1266 	INP_WLOCK(inp);
1267 	if (!(tp->t_flags & TF_DISCONNECTED)) {
1268 		if (tp->t_state != TCPS_TIME_WAIT) {
1269 			tp->t_flags |= TF_CLOSED;
1270 			tcp_disconnect(tp);
1271 			tcp_bblog_pru(tp, PRU_CLOSE, 0);
1272 			TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1273 		}
1274 	}
1275 	if (!(tp->t_flags & TF_DISCONNECTED)) {
1276 		soref(so);
1277 		inp->inp_flags |= INP_SOCKREF;
1278 	}
1279 	INP_WUNLOCK(inp);
1280 	NET_EPOCH_EXIT(et);
1281 }
1282 
1283 static int
1284 tcp_pru_options_support(struct tcpcb *tp, int flags)
1285 {
1286 	/*
1287 	 * If the specific TCP stack has a pru_options
1288 	 * specified then it does not always support
1289 	 * all the PRU_XX options and we must ask it.
1290 	 * If the function is not specified then all
1291 	 * of the PRU_XX options are supported.
1292 	 */
1293 	int ret = 0;
1294 
1295 	if (tp->t_fb->tfb_pru_options) {
1296 		ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1297 	}
1298 	return (ret);
1299 }
1300 
1301 /*
1302  * Receive out-of-band data.
1303  */
1304 static int
1305 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1306 {
1307 	struct inpcb *inp = sotoinpcb(so);
1308 	struct tcpcb *tp = intotcpcb(inp);
1309 	int error = 0;
1310 
1311 	INP_WLOCK(inp);
1312 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
1313 		/* XXXGL: how could this happen?! */
1314 		INP_WUNLOCK(inp);
1315 		return (ECONNRESET);
1316 	}
1317 
1318 	error = tcp_pru_options_support(tp, PRUS_OOB);
1319 	if (error) {
1320 		goto out;
1321 	}
1322 	if ((so->so_oobmark == 0 &&
1323 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1324 	    so->so_options & SO_OOBINLINE ||
1325 	    tp->t_oobflags & TCPOOB_HADDATA) {
1326 		error = EINVAL;
1327 		goto out;
1328 	}
1329 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1330 		error = EWOULDBLOCK;
1331 		goto out;
1332 	}
1333 	m->m_len = 1;
1334 	*mtod(m, caddr_t) = tp->t_iobc;
1335 	if ((flags & MSG_PEEK) == 0)
1336 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1337 
1338 out:
1339 	tcp_bblog_pru(tp, PRU_RCVOOB, error);
1340 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1341 	INP_WUNLOCK(inp);
1342 	return (error);
1343 }
1344 
1345 #ifdef INET
1346 struct protosw tcp_protosw = {
1347 	.pr_type =		SOCK_STREAM,
1348 	.pr_protocol =		IPPROTO_TCP,
1349 	.pr_flags =		PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD |
1350 				    PR_CAPATTACH,
1351 	.pr_ctloutput =		tcp_ctloutput,
1352 	.pr_abort =		tcp_usr_abort,
1353 	.pr_accept =		tcp_usr_accept,
1354 	.pr_attach =		tcp_usr_attach,
1355 	.pr_bind =		tcp_usr_bind,
1356 	.pr_connect =		tcp_usr_connect,
1357 	.pr_control =		in_control,
1358 	.pr_detach =		tcp_usr_detach,
1359 	.pr_disconnect =	tcp_usr_disconnect,
1360 	.pr_listen =		tcp_usr_listen,
1361 	.pr_peeraddr =		in_getpeeraddr,
1362 	.pr_rcvd =		tcp_usr_rcvd,
1363 	.pr_rcvoob =		tcp_usr_rcvoob,
1364 	.pr_send =		tcp_usr_send,
1365 	.pr_sendfile_wait =	sendfile_wait_generic,
1366 	.pr_ready =		tcp_usr_ready,
1367 	.pr_shutdown =		tcp_usr_shutdown,
1368 	.pr_sockaddr =		in_getsockaddr,
1369 	.pr_sosetlabel =	in_pcbsosetlabel,
1370 	.pr_close =		tcp_usr_close,
1371 };
1372 #endif /* INET */
1373 
1374 #ifdef INET6
1375 struct protosw tcp6_protosw = {
1376 	.pr_type =		SOCK_STREAM,
1377 	.pr_protocol =		IPPROTO_TCP,
1378 	.pr_flags =		PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD |
1379 				    PR_CAPATTACH,
1380 	.pr_ctloutput =		tcp_ctloutput,
1381 	.pr_abort =		tcp_usr_abort,
1382 	.pr_accept =		tcp6_usr_accept,
1383 	.pr_attach =		tcp_usr_attach,
1384 	.pr_bind =		tcp6_usr_bind,
1385 	.pr_connect =		tcp6_usr_connect,
1386 	.pr_control =		in6_control,
1387 	.pr_detach =		tcp_usr_detach,
1388 	.pr_disconnect =	tcp_usr_disconnect,
1389 	.pr_listen =		tcp6_usr_listen,
1390 	.pr_peeraddr =		in6_mapped_peeraddr,
1391 	.pr_rcvd =		tcp_usr_rcvd,
1392 	.pr_rcvoob =		tcp_usr_rcvoob,
1393 	.pr_send =		tcp_usr_send,
1394 	.pr_sendfile_wait =	sendfile_wait_generic,
1395 	.pr_ready =		tcp_usr_ready,
1396 	.pr_shutdown =		tcp_usr_shutdown,
1397 	.pr_sockaddr =		in6_mapped_sockaddr,
1398 	.pr_sosetlabel =	in_pcbsosetlabel,
1399 	.pr_close =		tcp_usr_close,
1400 };
1401 #endif /* INET6 */
1402 
1403 #ifdef INET
1404 /*
1405  * Common subroutine to open a TCP connection to remote host specified
1406  * by struct sockaddr_in.  Call in_pcbconnect() to choose local host address
1407  * and assign a local port number and install the inpcb into the hash.
1408  * Initialize connection parameters and enter SYN-SENT state.
1409  */
1410 static int
1411 tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td)
1412 {
1413 	struct inpcb *inp = tptoinpcb(tp);
1414 	struct socket *so = tptosocket(tp);
1415 	int error;
1416 
1417 	NET_EPOCH_ASSERT();
1418 	INP_WLOCK_ASSERT(inp);
1419 
1420 	if (__predict_false((so->so_state &
1421 	    (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING |
1422 	    SS_ISDISCONNECTED)) != 0))
1423 		return (EISCONN);
1424 	if (__predict_false((so->so_options & SO_REUSEPORT_LB) != 0))
1425 		return (EOPNOTSUPP);
1426 
1427 	error = in_pcbconnect(inp, sin, td->td_ucred);
1428 	if (error != 0)
1429 		return (error);
1430 
1431 	/* set the hash on the connection */
1432 	rss_proto_software_hash_v4(inp->inp_faddr, inp->inp_laddr,
1433 	    inp->inp_fport, inp->inp_lport, IPPROTO_TCP,
1434 	    &inp->inp_flowid, &inp->inp_flowtype);
1435 	/*
1436 	 * Compute window scaling to request:
1437 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1438 	 * XXX: This should move to tcp_output().
1439 	 */
1440 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1441 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1442 		tp->request_r_scale++;
1443 
1444 	soisconnecting(so);
1445 	TCPSTAT_INC(tcps_connattempt);
1446 	tcp_state_change(tp, TCPS_SYN_SENT);
1447 	tp->iss = tcp_new_isn(&inp->inp_inc);
1448 	if (tp->t_flags & TF_REQ_TSTMP)
1449 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1450 	tcp_sendseqinit(tp);
1451 
1452 	return (0);
1453 }
1454 #endif /* INET */
1455 
1456 #ifdef INET6
1457 static int
1458 tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td)
1459 {
1460 	struct inpcb *inp = tptoinpcb(tp);
1461 	struct socket *so = tptosocket(tp);
1462 	int error;
1463 
1464 	NET_EPOCH_ASSERT();
1465 	INP_WLOCK_ASSERT(inp);
1466 
1467 	if (__predict_false((so->so_state &
1468 	    (SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING |
1469 	    SS_ISDISCONNECTED)) != 0))
1470 		return (EISCONN);
1471 	if (__predict_false((so->so_options & SO_REUSEPORT_LB) != 0))
1472 		return (EOPNOTSUPP);
1473 
1474 	error = in6_pcbconnect(inp, sin6, td->td_ucred, true);
1475 	if (error != 0)
1476 		return (error);
1477 
1478 	/* set the hash on the connection */
1479 	rss_proto_software_hash_v6(&inp->in6p_faddr,
1480 	    &inp->in6p_laddr, inp->inp_fport, inp->inp_lport, IPPROTO_TCP,
1481 	    &inp->inp_flowid, &inp->inp_flowtype);
1482 	/* Compute window scaling to request.  */
1483 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1484 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1485 		tp->request_r_scale++;
1486 
1487 	soisconnecting(so);
1488 	TCPSTAT_INC(tcps_connattempt);
1489 	tcp_state_change(tp, TCPS_SYN_SENT);
1490 	tp->iss = tcp_new_isn(&inp->inp_inc);
1491 	if (tp->t_flags & TF_REQ_TSTMP)
1492 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1493 	tcp_sendseqinit(tp);
1494 
1495 	return (0);
1496 }
1497 #endif /* INET6 */
1498 
1499 /*
1500  * Export TCP internal state information via a struct tcp_info, based on the
1501  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1502  * (TCP state machine, etc).  We export all information using FreeBSD-native
1503  * constants -- for example, the numeric values for tcpi_state will differ
1504  * from Linux.
1505  */
1506 void
1507 tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti)
1508 {
1509 
1510 	INP_LOCK_ASSERT(tptoinpcb(tp));
1511 	bzero(ti, sizeof(*ti));
1512 
1513 	ti->tcpi_state = tp->t_state;
1514 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1515 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1516 	if (tp->t_flags & TF_SACK_PERMIT)
1517 		ti->tcpi_options |= TCPI_OPT_SACK;
1518 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1519 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1520 		ti->tcpi_snd_wscale = tp->snd_scale;
1521 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1522 	}
1523 	switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) {
1524 		case TF2_ECN_PERMIT:
1525 			ti->tcpi_options |= TCPI_OPT_ECN;
1526 			break;
1527 		case TF2_ACE_PERMIT:
1528 			/* FALLTHROUGH */
1529 		case TF2_ECN_PERMIT | TF2_ACE_PERMIT:
1530 			ti->tcpi_options |= TCPI_OPT_ACE;
1531 			break;
1532 		default:
1533 			break;
1534 	}
1535 	if (tp->t_flags & TF_FASTOPEN)
1536 		ti->tcpi_options |= TCPI_OPT_TFO;
1537 
1538 	ti->tcpi_rto = tp->t_rxtcur * tick;
1539 	ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1540 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1541 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1542 
1543 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1544 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1545 
1546 	/*
1547 	 * FreeBSD-specific extension fields for tcp_info.
1548 	 */
1549 	ti->tcpi_rcv_space = tp->rcv_wnd;
1550 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1551 	ti->tcpi_snd_wnd = tp->snd_wnd;
1552 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1553 	ti->tcpi_snd_nxt = tp->snd_nxt;
1554 	ti->tcpi_snd_mss = tp->t_maxseg;
1555 	ti->tcpi_rcv_mss = tp->t_maxseg;
1556 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1557 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1558 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1559 	ti->tcpi_snd_una = tp->snd_una;
1560 	ti->tcpi_snd_max = tp->snd_max;
1561 	ti->tcpi_rcv_numsacks = tp->rcv_numsacks;
1562 	ti->tcpi_rcv_adv = tp->rcv_adv;
1563 	ti->tcpi_dupacks = tp->t_dupacks;
1564 	ti->tcpi_rttmin = tp->t_rttlow;
1565 #ifdef TCP_OFFLOAD
1566 	if (tp->t_flags & TF_TOE) {
1567 		ti->tcpi_options |= TCPI_OPT_TOE;
1568 		tcp_offload_tcp_info(tp, ti);
1569 	}
1570 #endif
1571 	/*
1572 	 * AccECN related counters.
1573 	 */
1574 	if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) ==
1575 	    (TF2_ECN_PERMIT | TF2_ACE_PERMIT))
1576 		/*
1577 		 * Internal counter starts at 5 for AccECN
1578 		 * but 0 for RFC3168 ECN.
1579 		 */
1580 		ti->tcpi_delivered_ce = tp->t_scep - 5;
1581 	else
1582 		ti->tcpi_delivered_ce = tp->t_scep;
1583 	ti->tcpi_received_ce = tp->t_rcep;
1584 }
1585 
1586 /*
1587  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1588  * socket option arguments.  When it re-acquires the lock after the copy, it
1589  * has to revalidate that the connection is still valid for the socket
1590  * option.
1591  * XXXGL: review if this is really needed
1592  */
1593 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1594 	INP_WLOCK(inp);							\
1595 	tp = intotcpcb(inp);						\
1596 	if (tp->t_flags & TF_DISCONNECTED) {				\
1597 		INP_WUNLOCK(inp);					\
1598 		cleanup;						\
1599 		return (ECONNRESET);					\
1600 	}								\
1601 } while(0)
1602 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1603 
1604 int
1605 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
1606 {
1607 	struct socket *so = inp->inp_socket;
1608 	struct tcpcb *tp = intotcpcb(inp);
1609 	int error = 0;
1610 
1611 	MPASS(sopt->sopt_dir == SOPT_SET);
1612 	INP_WLOCK_ASSERT(inp);
1613 	KASSERT((tp->t_flags & TF_DISCONNECTED) == 0,
1614 	    ("tp_flags == %x", tp->t_flags));
1615 	KASSERT(so != NULL, ("inp_socket == NULL"));
1616 
1617 	if (sopt->sopt_level != IPPROTO_TCP) {
1618 		INP_WUNLOCK(inp);
1619 #ifdef INET6
1620 		if (inp->inp_vflag & INP_IPV6PROTO)
1621 			error = ip6_ctloutput(so, sopt);
1622 #endif
1623 #if defined(INET6) && defined(INET)
1624 		else
1625 #endif
1626 #ifdef INET
1627 			error = ip_ctloutput(so, sopt);
1628 #endif
1629 		/*
1630 		 * When an IP-level socket option affects TCP, pass control
1631 		 * down to stack tfb_tcp_ctloutput, otherwise return what
1632 		 * IP level returned.
1633 		 */
1634 		switch (sopt->sopt_level) {
1635 #ifdef INET6
1636 		case IPPROTO_IPV6:
1637 			if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
1638 				return (error);
1639 			switch (sopt->sopt_name) {
1640 			case IPV6_TCLASS:
1641 				/* Notify tcp stacks that care (e.g. RACK). */
1642 				break;
1643 			case IPV6_USE_MIN_MTU:
1644 				/* Update t_maxseg accordingly. */
1645 				break;
1646 			default:
1647 				return (error);
1648 			}
1649 			break;
1650 #endif
1651 #ifdef INET
1652 		case IPPROTO_IP:
1653 			switch (sopt->sopt_name) {
1654 			case IP_TOS:
1655 				inp->inp_ip_tos &= ~IPTOS_ECN_MASK;
1656 				break;
1657 			case IP_TTL:
1658 				/* Notify tcp stacks that care (e.g. RACK). */
1659 				break;
1660 			default:
1661 				return (error);
1662 			}
1663 			break;
1664 #endif
1665 		default:
1666 			return (error);
1667 		}
1668 		INP_WLOCK_RECHECK(inp);
1669 	} else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
1670 		/*
1671 		 * Protect the TCP option TCP_FUNCTION_BLK so
1672 		 * that a sub-function can *never* overwrite this.
1673 		 */
1674 		struct tcp_function_set fsn;
1675 		struct tcp_function_block *blk;
1676 		void *ptr = NULL;
1677 
1678 		INP_WUNLOCK(inp);
1679 		error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
1680 		if (error)
1681 			return (error);
1682 
1683 		INP_WLOCK_RECHECK(inp);
1684 
1685 		blk = find_and_ref_tcp_functions(&fsn);
1686 		if (blk == NULL) {
1687 			INP_WUNLOCK(inp);
1688 			return (ENOENT);
1689 		}
1690 		if (tp->t_fb == blk) {
1691 			/* You already have this */
1692 			refcount_release(&blk->tfb_refcnt);
1693 			INP_WUNLOCK(inp);
1694 			return (0);
1695 		}
1696 		if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1697 			refcount_release(&blk->tfb_refcnt);
1698 			INP_WUNLOCK(inp);
1699 			return (ENOENT);
1700 		}
1701 		error = (*blk->tfb_tcp_handoff_ok)(tp);
1702 		if (error) {
1703 			refcount_release(&blk->tfb_refcnt);
1704 			INP_WUNLOCK(inp);
1705 			return (error);
1706 		}
1707 		/*
1708 		 * Ensure the new stack takes ownership with a
1709 		 * clean slate on peak rate threshold.
1710 		 */
1711 		if (tp->t_fb->tfb_tcp_timer_stop_all != NULL)
1712 			tp->t_fb->tfb_tcp_timer_stop_all(tp);
1713 		if (blk->tfb_tcp_fb_init) {
1714 			error = (*blk->tfb_tcp_fb_init)(tp, &ptr);
1715 			if (error) {
1716 				/*
1717 				 * Release the ref count the lookup
1718 				 * acquired.
1719 				 */
1720 				refcount_release(&blk->tfb_refcnt);
1721 				/*
1722 				 * Now there is a chance that the
1723 				 * init() function mucked with some
1724 				 * things before it failed, such as
1725 				 * hpts or inp_flags2 or timer granularity.
1726 				 * It should not of, but lets give the old
1727 				 * stack a chance to reset to a known good state.
1728 				 */
1729 				if (tp->t_fb->tfb_switch_failed) {
1730 					(*tp->t_fb->tfb_switch_failed)(tp);
1731 				}
1732 			 	goto err_out;
1733 			}
1734 		}
1735 		if (tp->t_fb->tfb_tcp_fb_fini) {
1736 			struct epoch_tracker et;
1737 			/*
1738 			 * Tell the stack to cleanup with 0 i.e.
1739 			 * the tcb is not going away.
1740 			 */
1741 			NET_EPOCH_ENTER(et);
1742 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1743 			NET_EPOCH_EXIT(et);
1744 		}
1745 		/*
1746 		 * Release the old refcnt, the
1747 		 * lookup acquired a ref on the
1748 		 * new one already.
1749 		 */
1750 		refcount_release(&tp->t_fb->tfb_refcnt);
1751 		/*
1752 		 * Set in the new stack.
1753 		 */
1754 		tp->t_fb = blk;
1755 		tp->t_fb_ptr = ptr;
1756 #ifdef TCP_OFFLOAD
1757 		if (tp->t_flags & TF_TOE) {
1758 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1759 			     sopt->sopt_name);
1760 		}
1761 #endif
1762 err_out:
1763 		INP_WUNLOCK(inp);
1764 		return (error);
1765 
1766 	}
1767 
1768 	/* Pass in the INP locked, callee must unlock it. */
1769 	return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt));
1770 }
1771 
1772 static int
1773 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
1774 {
1775 	struct socket *so = inp->inp_socket;
1776 	struct tcpcb *tp = intotcpcb(inp);
1777 	int error = 0;
1778 
1779 	MPASS(sopt->sopt_dir == SOPT_GET);
1780 	INP_WLOCK_ASSERT(inp);
1781 	KASSERT((tp->t_flags & TF_DISCONNECTED) == 0,
1782 	    ("tp_flags == %x", tp->t_flags));
1783 	KASSERT(so != NULL, ("inp_socket == NULL"));
1784 
1785 	if (sopt->sopt_level != IPPROTO_TCP) {
1786 		INP_WUNLOCK(inp);
1787 #ifdef INET6
1788 		if (inp->inp_vflag & INP_IPV6PROTO)
1789 			error = ip6_ctloutput(so, sopt);
1790 #endif /* INET6 */
1791 #if defined(INET6) && defined(INET)
1792 		else
1793 #endif
1794 #ifdef INET
1795 			error = ip_ctloutput(so, sopt);
1796 #endif
1797 		return (error);
1798 	}
1799 	if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
1800 	     (sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
1801 		struct tcp_function_set fsn;
1802 
1803 		if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
1804 			memset(&fsn, 0, sizeof(fsn));
1805 			find_tcp_function_alias(tp->t_fb, &fsn);
1806 		} else {
1807 			strncpy(fsn.function_set_name,
1808 			    tp->t_fb->tfb_tcp_block_name,
1809 			    TCP_FUNCTION_NAME_LEN_MAX);
1810 			fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1811 		}
1812 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1813 		INP_WUNLOCK(inp);
1814 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1815 		return (error);
1816 	}
1817 
1818 	/* Pass in the INP locked, callee must unlock it. */
1819 	return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt));
1820 }
1821 
1822 int
1823 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1824 {
1825 	struct inpcb *inp = sotoinpcb(so);
1826 	struct tcpcb *tp = intotcpcb(inp);
1827 
1828 	INP_WLOCK(inp);
1829 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
1830 		INP_WUNLOCK(inp);
1831 		return (ECONNRESET);
1832 	}
1833 	if (sopt->sopt_dir == SOPT_SET)
1834 		return (tcp_ctloutput_set(inp, sopt));
1835 	else if (sopt->sopt_dir == SOPT_GET)
1836 		return (tcp_ctloutput_get(inp, sopt));
1837 	else
1838 		panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
1839 }
1840 
1841 /*
1842  * If this assert becomes untrue, we need to change the size of the buf
1843  * variable in tcp_default_ctloutput().
1844  */
1845 #ifdef CTASSERT
1846 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1847 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1848 #endif
1849 
1850 extern struct cc_algo newreno_cc_algo;
1851 
1852 static int
1853 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt)
1854 {
1855 	struct cc_algo *algo;
1856 	void *ptr = NULL;
1857 	struct tcpcb *tp = intotcpcb(inp);
1858 	struct cc_var cc_mem;
1859 	char	buf[TCP_CA_NAME_MAX];
1860 	size_t mem_sz;
1861 	int error;
1862 
1863 	INP_WUNLOCK(inp);
1864 	error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1865 	if (error)
1866 		return(error);
1867 	buf[sopt->sopt_valsize] = '\0';
1868 	CC_LIST_RLOCK();
1869 	STAILQ_FOREACH(algo, &cc_list, entries) {
1870 		if (strncmp(buf, algo->name,
1871 			    TCP_CA_NAME_MAX) == 0) {
1872 			if (algo->flags & CC_MODULE_BEING_REMOVED) {
1873 				/* We can't "see" modules being unloaded */
1874 				continue;
1875 			}
1876 			break;
1877 		}
1878 	}
1879 	if (algo == NULL) {
1880 		CC_LIST_RUNLOCK();
1881 		return(ESRCH);
1882 	}
1883 	/*
1884 	 * With a reference the algorithm cannot be removed
1885 	 * so we hold a reference through the change process.
1886 	 */
1887 	cc_refer(algo);
1888 	CC_LIST_RUNLOCK();
1889 	if (algo->cb_init != NULL) {
1890 		/* We can now pre-get the memory for the CC */
1891 		mem_sz = (*algo->cc_data_sz)();
1892 		if (mem_sz == 0) {
1893 			goto no_mem_needed;
1894 		}
1895 		ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
1896 	} else {
1897 no_mem_needed:
1898 		mem_sz = 0;
1899 		ptr = NULL;
1900 	}
1901 	/*
1902 	 * Make sure its all clean and zero and also get
1903 	 * back the inplock.
1904 	 */
1905 	memset(&cc_mem, 0, sizeof(cc_mem));
1906 	INP_WLOCK(inp);
1907 	if (__predict_false(tp->t_flags & TF_DISCONNECTED)) {
1908 		INP_WUNLOCK(inp);
1909 		if (ptr)
1910 			free(ptr, M_CC_MEM);
1911 		/* Release our temp reference */
1912 		CC_LIST_RLOCK();
1913 		cc_release(algo);
1914 		CC_LIST_RUNLOCK();
1915 		return (ECONNRESET);
1916 	}
1917 	if (ptr != NULL)
1918 		memset(ptr, 0, mem_sz);
1919 	cc_mem.tp = tp;
1920 	/*
1921 	 * We once again hold a write lock over the tcb so it's
1922 	 * safe to do these things without ordering concerns.
1923 	 * Note here we init into stack memory.
1924 	 */
1925 	if (algo->cb_init != NULL)
1926 		error = algo->cb_init(&cc_mem, ptr);
1927 	else
1928 		error = 0;
1929 	/*
1930 	 * The CC algorithms, when given their memory
1931 	 * should not fail we could in theory have a
1932 	 * KASSERT here.
1933 	 */
1934 	if (error == 0) {
1935 		/*
1936 		 * Touchdown, lets go ahead and move the
1937 		 * connection to the new CC module by
1938 		 * copying in the cc_mem after we call
1939 		 * the old ones cleanup (if any).
1940 		 */
1941 		if (CC_ALGO(tp)->cb_destroy != NULL)
1942 			CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
1943 		/* Detach the old CC from the tcpcb  */
1944 		cc_detach(tp);
1945 		/* Copy in our temp memory that was inited */
1946 		memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var));
1947 		/* Now attach the new, which takes a reference */
1948 		cc_attach(tp, algo);
1949 		/* Ok now are we where we have gotten past any conn_init? */
1950 		if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
1951 			/* Yep run the connection init for the new CC */
1952 			CC_ALGO(tp)->conn_init(&tp->t_ccv);
1953 		}
1954 	} else if (ptr)
1955 		free(ptr, M_CC_MEM);
1956 	INP_WUNLOCK(inp);
1957 	/* Now lets release our temp reference */
1958 	CC_LIST_RLOCK();
1959 	cc_release(algo);
1960 	CC_LIST_RUNLOCK();
1961 	return (error);
1962 }
1963 
1964 int
1965 tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt)
1966 {
1967 	struct inpcb *inp = tptoinpcb(tp);
1968 	int	error, opt, optval;
1969 	u_int	ui;
1970 	struct	tcp_info ti;
1971 #ifdef KERN_TLS
1972 	struct tls_enable tls;
1973 	struct socket *so = inp->inp_socket;
1974 #endif
1975 	char	*pbuf, buf[TCP_LOG_ID_LEN];
1976 #ifdef STATS
1977 	struct statsblob *sbp;
1978 #endif
1979 	size_t	len;
1980 
1981 	INP_WLOCK_ASSERT(inp);
1982 	KASSERT((tp->t_flags & TF_DISCONNECTED) == 0,
1983 	    ("tp_flags == %x", tp->t_flags));
1984 	KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL"));
1985 
1986 	switch (sopt->sopt_level) {
1987 #ifdef INET6
1988 	case IPPROTO_IPV6:
1989 		MPASS(inp->inp_vflag & INP_IPV6PROTO);
1990 		switch (sopt->sopt_name) {
1991 		case IPV6_USE_MIN_MTU:
1992 			tcp6_use_min_mtu(tp);
1993 			/* FALLTHROUGH */
1994 		}
1995 		INP_WUNLOCK(inp);
1996 		return (0);
1997 #endif
1998 #ifdef INET
1999 	case IPPROTO_IP:
2000 		INP_WUNLOCK(inp);
2001 		return (0);
2002 #endif
2003 	}
2004 
2005 	/*
2006 	 * For TCP_CCALGOOPT forward the control to CC module, for both
2007 	 * SOPT_SET and SOPT_GET.
2008 	 */
2009 	switch (sopt->sopt_name) {
2010 	case TCP_CCALGOOPT:
2011 		INP_WUNLOCK(inp);
2012 		if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2013 			return (EINVAL);
2014 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2015 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2016 		    sopt->sopt_valsize);
2017 		if (error) {
2018 			free(pbuf, M_TEMP);
2019 			return (error);
2020 		}
2021 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2022 		if (CC_ALGO(tp)->ctl_output != NULL)
2023 			error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf);
2024 		else
2025 			error = ENOENT;
2026 		INP_WUNLOCK(inp);
2027 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
2028 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2029 		free(pbuf, M_TEMP);
2030 		return (error);
2031 	}
2032 
2033 	switch (sopt->sopt_dir) {
2034 	case SOPT_SET:
2035 		switch (sopt->sopt_name) {
2036 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2037 		case TCP_MD5SIG:
2038 			INP_WUNLOCK(inp);
2039 			if (!TCPMD5_ENABLED())
2040 				return (ENOPROTOOPT);
2041 			error = TCPMD5_PCBCTL(inp, sopt);
2042 			if (error)
2043 				return (error);
2044 			INP_WLOCK_RECHECK(inp);
2045 			goto unlock_and_done;
2046 #endif /* IPSEC */
2047 
2048 		case TCP_NODELAY:
2049 		case TCP_NOOPT:
2050 			INP_WUNLOCK(inp);
2051 			error = sooptcopyin(sopt, &optval, sizeof optval,
2052 			    sizeof optval);
2053 			if (error)
2054 				return (error);
2055 
2056 			INP_WLOCK_RECHECK(inp);
2057 			switch (sopt->sopt_name) {
2058 			case TCP_NODELAY:
2059 				opt = TF_NODELAY;
2060 				break;
2061 			case TCP_NOOPT:
2062 				opt = TF_NOOPT;
2063 				break;
2064 			default:
2065 				opt = 0; /* dead code to fool gcc */
2066 				break;
2067 			}
2068 
2069 			if (optval)
2070 				tp->t_flags |= opt;
2071 			else
2072 				tp->t_flags &= ~opt;
2073 unlock_and_done:
2074 #ifdef TCP_OFFLOAD
2075 			if (tp->t_flags & TF_TOE) {
2076 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
2077 				    sopt->sopt_name);
2078 			}
2079 #endif
2080 			INP_WUNLOCK(inp);
2081 			break;
2082 
2083 		case TCP_NOPUSH:
2084 			INP_WUNLOCK(inp);
2085 			error = sooptcopyin(sopt, &optval, sizeof optval,
2086 			    sizeof optval);
2087 			if (error)
2088 				return (error);
2089 
2090 			INP_WLOCK_RECHECK(inp);
2091 			if (optval)
2092 				tp->t_flags |= TF_NOPUSH;
2093 			else if (tp->t_flags & TF_NOPUSH) {
2094 				tp->t_flags &= ~TF_NOPUSH;
2095 				if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2096 					struct epoch_tracker et;
2097 
2098 					NET_EPOCH_ENTER(et);
2099 					error = tcp_output_nodrop(tp);
2100 					NET_EPOCH_EXIT(et);
2101 				}
2102 			}
2103 			goto unlock_and_done;
2104 
2105 		case TCP_REMOTE_UDP_ENCAPS_PORT:
2106 			INP_WUNLOCK(inp);
2107 			error = sooptcopyin(sopt, &optval, sizeof optval,
2108 			    sizeof optval);
2109 			if (error)
2110 				return (error);
2111 			if ((optval < TCP_TUNNELING_PORT_MIN) ||
2112 			    (optval > TCP_TUNNELING_PORT_MAX)) {
2113 				/* Its got to be in range */
2114 				return (EINVAL);
2115 			}
2116 			if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2117 				/* You have to have enabled a UDP tunneling port first */
2118 				return (EINVAL);
2119 			}
2120 			INP_WLOCK_RECHECK(inp);
2121 			if (tp->t_state != TCPS_CLOSED) {
2122 				/* You can't change after you are connected */
2123 				error = EINVAL;
2124 			} else {
2125 				/* Ok we are all good set the port */
2126 				tp->t_port = htons(optval);
2127 			}
2128 			goto unlock_and_done;
2129 
2130 		case TCP_MAXSEG:
2131 			INP_WUNLOCK(inp);
2132 			error = sooptcopyin(sopt, &optval, sizeof optval,
2133 			    sizeof optval);
2134 			if (error)
2135 				return (error);
2136 
2137 			INP_WLOCK_RECHECK(inp);
2138 			if (optval > 0 && optval <= tp->t_maxseg &&
2139 			    optval + 40 >= V_tcp_minmss) {
2140 				tp->t_maxseg = optval;
2141 				if (tp->t_maxseg < V_tcp_mssdflt) {
2142 					/*
2143 					 * The MSS is so small we should not process incoming
2144 					 * SACK's since we are subject to attack in such a
2145 					 * case.
2146 					 */
2147 					tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
2148 				} else {
2149 					tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
2150 				}
2151 			} else
2152 				error = EINVAL;
2153 			goto unlock_and_done;
2154 
2155 		case TCP_INFO:
2156 			INP_WUNLOCK(inp);
2157 			error = EINVAL;
2158 			break;
2159 
2160 		case TCP_STATS:
2161 			INP_WUNLOCK(inp);
2162 #ifdef STATS
2163 			error = sooptcopyin(sopt, &optval, sizeof optval,
2164 			    sizeof optval);
2165 			if (error)
2166 				return (error);
2167 
2168 			if (optval > 0)
2169 				sbp = stats_blob_alloc(
2170 				    V_tcp_perconn_stats_dflt_tpl, 0);
2171 			else
2172 				sbp = NULL;
2173 
2174 			INP_WLOCK_RECHECK(inp);
2175 			if ((tp->t_stats != NULL && sbp == NULL) ||
2176 			    (tp->t_stats == NULL && sbp != NULL)) {
2177 				struct statsblob *t = tp->t_stats;
2178 				tp->t_stats = sbp;
2179 				sbp = t;
2180 			}
2181 			INP_WUNLOCK(inp);
2182 
2183 			stats_blob_destroy(sbp);
2184 #else
2185 			return (EOPNOTSUPP);
2186 #endif /* !STATS */
2187 			break;
2188 
2189 		case TCP_CONGESTION:
2190 			error = tcp_set_cc_mod(inp, sopt);
2191 			break;
2192 
2193 		case TCP_REUSPORT_LB_NUMA:
2194 			INP_WUNLOCK(inp);
2195 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2196 			    sizeof(optval));
2197 			INP_WLOCK_RECHECK(inp);
2198 			if (!error)
2199 				error = in_pcblbgroup_numa(inp, optval);
2200 			INP_WUNLOCK(inp);
2201 			break;
2202 
2203 #ifdef KERN_TLS
2204 		case TCP_TXTLS_ENABLE:
2205 			INP_WUNLOCK(inp);
2206 			error = ktls_copyin_tls_enable(sopt, &tls);
2207 			if (error != 0)
2208 				break;
2209 			error = ktls_enable_tx(so, &tls);
2210 			ktls_cleanup_tls_enable(&tls);
2211 			break;
2212 		case TCP_TXTLS_MODE:
2213 			INP_WUNLOCK(inp);
2214 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2215 			if (error != 0)
2216 				return (error);
2217 
2218 			INP_WLOCK_RECHECK(inp);
2219 			error = ktls_set_tx_mode(so, ui);
2220 			INP_WUNLOCK(inp);
2221 			break;
2222 		case TCP_RXTLS_ENABLE:
2223 			INP_WUNLOCK(inp);
2224 			error = ktls_copyin_tls_enable(sopt, &tls);
2225 			if (error != 0)
2226 				break;
2227 			error = ktls_enable_rx(so, &tls);
2228 			ktls_cleanup_tls_enable(&tls);
2229 			break;
2230 #endif
2231 		case TCP_MAXUNACKTIME:
2232 		case TCP_KEEPIDLE:
2233 		case TCP_KEEPINTVL:
2234 		case TCP_KEEPINIT:
2235 			INP_WUNLOCK(inp);
2236 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2237 			if (error)
2238 				return (error);
2239 
2240 			if (ui > (UINT_MAX / hz)) {
2241 				error = EINVAL;
2242 				break;
2243 			}
2244 			ui *= hz;
2245 
2246 			INP_WLOCK_RECHECK(inp);
2247 			switch (sopt->sopt_name) {
2248 			case TCP_MAXUNACKTIME:
2249 				tp->t_maxunacktime = ui;
2250 				break;
2251 
2252 			case TCP_KEEPIDLE:
2253 				tp->t_keepidle = ui;
2254 				/*
2255 				 * XXX: better check current remaining
2256 				 * timeout and "merge" it with new value.
2257 				 */
2258 				if ((tp->t_state > TCPS_LISTEN) &&
2259 				    (tp->t_state <= TCPS_CLOSING))
2260 					tcp_timer_activate(tp, TT_KEEP,
2261 					    TP_KEEPIDLE(tp));
2262 				break;
2263 			case TCP_KEEPINTVL:
2264 				tp->t_keepintvl = ui;
2265 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2266 				    (TP_MAXIDLE(tp) > 0))
2267 					tcp_timer_activate(tp, TT_2MSL,
2268 					    TP_MAXIDLE(tp));
2269 				break;
2270 			case TCP_KEEPINIT:
2271 				tp->t_keepinit = ui;
2272 				if (tp->t_state == TCPS_SYN_RECEIVED ||
2273 				    tp->t_state == TCPS_SYN_SENT)
2274 					tcp_timer_activate(tp, TT_KEEP,
2275 					    TP_KEEPINIT(tp));
2276 				break;
2277 			}
2278 			goto unlock_and_done;
2279 
2280 		case TCP_KEEPCNT:
2281 			INP_WUNLOCK(inp);
2282 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2283 			if (error)
2284 				return (error);
2285 
2286 			INP_WLOCK_RECHECK(inp);
2287 			tp->t_keepcnt = ui;
2288 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2289 			    (TP_MAXIDLE(tp) > 0))
2290 				tcp_timer_activate(tp, TT_2MSL,
2291 				    TP_MAXIDLE(tp));
2292 			goto unlock_and_done;
2293 
2294 		case TCP_FASTOPEN: {
2295 			struct tcp_fastopen tfo_optval;
2296 
2297 			INP_WUNLOCK(inp);
2298 			if (!V_tcp_fastopen_client_enable &&
2299 			    !V_tcp_fastopen_server_enable)
2300 				return (EPERM);
2301 
2302 			error = sooptcopyin(sopt, &tfo_optval,
2303 				    sizeof(tfo_optval), sizeof(int));
2304 			if (error)
2305 				return (error);
2306 
2307 			INP_WLOCK_RECHECK(inp);
2308 			if ((tp->t_state != TCPS_CLOSED) &&
2309 			    (tp->t_state != TCPS_LISTEN)) {
2310 				error = EINVAL;
2311 				goto unlock_and_done;
2312 			}
2313 			if (tfo_optval.enable) {
2314 				if (tp->t_state == TCPS_LISTEN) {
2315 					if (!V_tcp_fastopen_server_enable) {
2316 						error = EPERM;
2317 						goto unlock_and_done;
2318 					}
2319 
2320 					if (tp->t_tfo_pending == NULL)
2321 						tp->t_tfo_pending =
2322 						    tcp_fastopen_alloc_counter();
2323 				} else {
2324 					/*
2325 					 * If a pre-shared key was provided,
2326 					 * stash it in the client cookie
2327 					 * field of the tcpcb for use during
2328 					 * connect.
2329 					 */
2330 					if (sopt->sopt_valsize ==
2331 					    sizeof(tfo_optval)) {
2332 						memcpy(tp->t_tfo_cookie.client,
2333 						       tfo_optval.psk,
2334 						       TCP_FASTOPEN_PSK_LEN);
2335 						tp->t_tfo_client_cookie_len =
2336 						    TCP_FASTOPEN_PSK_LEN;
2337 					}
2338 				}
2339 				tp->t_flags |= TF_FASTOPEN;
2340 			} else
2341 				tp->t_flags &= ~TF_FASTOPEN;
2342 			goto unlock_and_done;
2343 		}
2344 
2345 #ifdef TCP_BLACKBOX
2346 		case TCP_LOG:
2347 			INP_WUNLOCK(inp);
2348 			error = sooptcopyin(sopt, &optval, sizeof optval,
2349 			    sizeof optval);
2350 			if (error)
2351 				return (error);
2352 
2353 			INP_WLOCK_RECHECK(inp);
2354 			error = tcp_log_state_change(tp, optval);
2355 			goto unlock_and_done;
2356 
2357 		case TCP_LOGBUF:
2358 			INP_WUNLOCK(inp);
2359 			error = EINVAL;
2360 			break;
2361 
2362 		case TCP_LOGID:
2363 			INP_WUNLOCK(inp);
2364 			error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2365 			if (error)
2366 				break;
2367 			buf[sopt->sopt_valsize] = '\0';
2368 			INP_WLOCK_RECHECK(inp);
2369 			error = tcp_log_set_id(tp, buf);
2370 			/* tcp_log_set_id() unlocks the INP. */
2371 			break;
2372 
2373 		case TCP_LOGDUMP:
2374 		case TCP_LOGDUMPID:
2375 			INP_WUNLOCK(inp);
2376 			error =
2377 			    sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2378 			if (error)
2379 				break;
2380 			buf[sopt->sopt_valsize] = '\0';
2381 			INP_WLOCK_RECHECK(inp);
2382 			if (sopt->sopt_name == TCP_LOGDUMP) {
2383 				error = tcp_log_dump_tp_logbuf(tp, buf,
2384 				    M_WAITOK, true);
2385 				INP_WUNLOCK(inp);
2386 			} else {
2387 				tcp_log_dump_tp_bucket_logbufs(tp, buf);
2388 				/*
2389 				 * tcp_log_dump_tp_bucket_logbufs() drops the
2390 				 * INP lock.
2391 				 */
2392 			}
2393 			break;
2394 #endif
2395 
2396 		default:
2397 			INP_WUNLOCK(inp);
2398 			error = ENOPROTOOPT;
2399 			break;
2400 		}
2401 		break;
2402 
2403 	case SOPT_GET:
2404 		tp = intotcpcb(inp);
2405 		switch (sopt->sopt_name) {
2406 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2407 		case TCP_MD5SIG:
2408 			INP_WUNLOCK(inp);
2409 			if (!TCPMD5_ENABLED())
2410 				return (ENOPROTOOPT);
2411 			error = TCPMD5_PCBCTL(inp, sopt);
2412 			break;
2413 #endif
2414 
2415 		case TCP_NODELAY:
2416 			optval = tp->t_flags & TF_NODELAY;
2417 			INP_WUNLOCK(inp);
2418 			error = sooptcopyout(sopt, &optval, sizeof optval);
2419 			break;
2420 		case TCP_MAXSEG:
2421 			optval = tp->t_maxseg;
2422 			INP_WUNLOCK(inp);
2423 			error = sooptcopyout(sopt, &optval, sizeof optval);
2424 			break;
2425 		case TCP_REMOTE_UDP_ENCAPS_PORT:
2426 			optval = ntohs(tp->t_port);
2427 			INP_WUNLOCK(inp);
2428 			error = sooptcopyout(sopt, &optval, sizeof optval);
2429 			break;
2430 		case TCP_NOOPT:
2431 			optval = tp->t_flags & TF_NOOPT;
2432 			INP_WUNLOCK(inp);
2433 			error = sooptcopyout(sopt, &optval, sizeof optval);
2434 			break;
2435 		case TCP_NOPUSH:
2436 			optval = tp->t_flags & TF_NOPUSH;
2437 			INP_WUNLOCK(inp);
2438 			error = sooptcopyout(sopt, &optval, sizeof optval);
2439 			break;
2440 		case TCP_INFO:
2441 			tcp_fill_info(tp, &ti);
2442 			INP_WUNLOCK(inp);
2443 			error = sooptcopyout(sopt, &ti, sizeof ti);
2444 			break;
2445 		case TCP_STATS:
2446 			{
2447 #ifdef STATS
2448 			int nheld;
2449 			TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2450 
2451 			error = 0;
2452 			socklen_t outsbsz = sopt->sopt_valsize;
2453 			if (tp->t_stats == NULL)
2454 				error = ENOENT;
2455 			else if (outsbsz >= tp->t_stats->cursz)
2456 				outsbsz = tp->t_stats->cursz;
2457 			else if (outsbsz >= sizeof(struct statsblob))
2458 				outsbsz = sizeof(struct statsblob);
2459 			else
2460 				error = EINVAL;
2461 			INP_WUNLOCK(inp);
2462 			if (error)
2463 				break;
2464 
2465 			sbp = sopt->sopt_val;
2466 			nheld = atop(round_page(((vm_offset_t)sbp) +
2467 			    (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2468 			vm_page_t ma[nheld];
2469 			if (vm_fault_quick_hold_pages(
2470 			    &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2471 			    outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2472 			    nheld) < 0) {
2473 				error = EFAULT;
2474 				break;
2475 			}
2476 
2477 			if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2478 			    SIZEOF_MEMBER(struct statsblob, flags))))
2479 				goto unhold;
2480 
2481 			INP_WLOCK_RECHECK(inp);
2482 			error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2483 			    sbflags | SB_CLONE_USRDSTNOFAULT);
2484 			INP_WUNLOCK(inp);
2485 			sopt->sopt_valsize = outsbsz;
2486 unhold:
2487 			vm_page_unhold_pages(ma, nheld);
2488 #else
2489 			INP_WUNLOCK(inp);
2490 			error = EOPNOTSUPP;
2491 #endif /* !STATS */
2492 			break;
2493 			}
2494 		case TCP_CONGESTION:
2495 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2496 			INP_WUNLOCK(inp);
2497 			error = sooptcopyout(sopt, buf, len + 1);
2498 			break;
2499 		case TCP_MAXUNACKTIME:
2500 		case TCP_KEEPIDLE:
2501 		case TCP_KEEPINTVL:
2502 		case TCP_KEEPINIT:
2503 		case TCP_KEEPCNT:
2504 			switch (sopt->sopt_name) {
2505 			case TCP_MAXUNACKTIME:
2506 				ui = TP_MAXUNACKTIME(tp) / hz;
2507 				break;
2508 			case TCP_KEEPIDLE:
2509 				ui = TP_KEEPIDLE(tp) / hz;
2510 				break;
2511 			case TCP_KEEPINTVL:
2512 				ui = TP_KEEPINTVL(tp) / hz;
2513 				break;
2514 			case TCP_KEEPINIT:
2515 				ui = TP_KEEPINIT(tp) / hz;
2516 				break;
2517 			case TCP_KEEPCNT:
2518 				ui = TP_KEEPCNT(tp);
2519 				break;
2520 			}
2521 			INP_WUNLOCK(inp);
2522 			error = sooptcopyout(sopt, &ui, sizeof(ui));
2523 			break;
2524 		case TCP_FASTOPEN:
2525 			optval = tp->t_flags & TF_FASTOPEN;
2526 			INP_WUNLOCK(inp);
2527 			error = sooptcopyout(sopt, &optval, sizeof optval);
2528 			break;
2529 #ifdef TCP_BLACKBOX
2530 		case TCP_LOG:
2531 			optval = tcp_get_bblog_state(tp);
2532 			INP_WUNLOCK(inp);
2533 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2534 			break;
2535 		case TCP_LOGBUF:
2536 			/* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2537 			error = tcp_log_getlogbuf(sopt, tp);
2538 			break;
2539 		case TCP_LOGID:
2540 			len = tcp_log_get_id(tp, buf);
2541 			INP_WUNLOCK(inp);
2542 			error = sooptcopyout(sopt, buf, len + 1);
2543 			break;
2544 		case TCP_LOGDUMP:
2545 		case TCP_LOGDUMPID:
2546 			INP_WUNLOCK(inp);
2547 			error = EINVAL;
2548 			break;
2549 #endif
2550 #ifdef KERN_TLS
2551 		case TCP_TXTLS_MODE:
2552 			error = ktls_get_tx_mode(so, &optval);
2553 			INP_WUNLOCK(inp);
2554 			if (error == 0)
2555 				error = sooptcopyout(sopt, &optval,
2556 				    sizeof(optval));
2557 			break;
2558 		case TCP_RXTLS_MODE:
2559 			error = ktls_get_rx_mode(so, &optval);
2560 			INP_WUNLOCK(inp);
2561 			if (error == 0)
2562 				error = sooptcopyout(sopt, &optval,
2563 				    sizeof(optval));
2564 			break;
2565 #endif
2566 		default:
2567 			INP_WUNLOCK(inp);
2568 			error = ENOPROTOOPT;
2569 			break;
2570 		}
2571 		break;
2572 	}
2573 	return (error);
2574 }
2575 #undef INP_WLOCK_RECHECK
2576 #undef INP_WLOCK_RECHECK_CLEANUP
2577 
2578 /*
2579  * Initiate (or continue) disconnect.
2580  * If embryonic state, just send reset (once).
2581  * If in ``let data drain'' option and linger null, just drop.
2582  * Otherwise (hard), mark socket disconnecting and drop
2583  * current input data; switch states based on user close, and
2584  * send segment to peer (with FIN).
2585  */
2586 static void
2587 tcp_disconnect(struct tcpcb *tp)
2588 {
2589 	struct socket *so = tptosocket(tp);
2590 
2591 	NET_EPOCH_ASSERT();
2592 	INP_WLOCK_ASSERT(tptoinpcb(tp));
2593 
2594 	/*
2595 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2596 	 * socket is still open.
2597 	 */
2598 	if (tp->t_state < TCPS_ESTABLISHED &&
2599 	    !(tp->t_state > TCPS_LISTEN && (tp->t_flags & TF_FASTOPEN))) {
2600 		tp = tcp_close(tp);
2601 		KASSERT(tp != NULL,
2602 		    ("tcp_disconnect: tcp_close() returned NULL"));
2603 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2604 		tp = tcp_drop(tp, 0);
2605 		KASSERT(tp != NULL,
2606 		    ("tcp_disconnect: tcp_drop() returned NULL"));
2607 	} else {
2608 		soisdisconnecting(so);
2609 		sbflush(&so->so_rcv);
2610 		tcp_usrclosed(tp);
2611 		if (!(tp->t_flags & TF_DISCONNECTED))
2612 			/* Ignore stack's drop request, we already at it. */
2613 			(void)tcp_output_nodrop(tp);
2614 	}
2615 }
2616 
2617 /*
2618  * User issued close, and wish to trail through shutdown states:
2619  * if never received SYN, just forget it.  If got a SYN from peer,
2620  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2621  * If already got a FIN from peer, then almost done; go to LAST_ACK
2622  * state.  In all other cases, have already sent FIN to peer (e.g.
2623  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2624  * for peer to send FIN or not respond to keep-alives, etc.
2625  * We can let the user exit from the close as soon as the FIN is acked.
2626  */
2627 static void
2628 tcp_usrclosed(struct tcpcb *tp)
2629 {
2630 
2631 	NET_EPOCH_ASSERT();
2632 	INP_WLOCK_ASSERT(tptoinpcb(tp));
2633 
2634 	switch (tp->t_state) {
2635 	case TCPS_LISTEN:
2636 	case TCPS_CLOSED:
2637 		tp = tcp_close(tp);
2638 		/*
2639 		 * tcp_close() should never return NULL here as the socket is
2640 		 * still open.
2641 		 */
2642 		KASSERT(tp != NULL,
2643 		    ("tcp_usrclosed: tcp_close() returned NULL"));
2644 		break;
2645 
2646 	case TCPS_SYN_SENT:
2647 	case TCPS_SYN_RECEIVED:
2648 		tp->t_flags |= TF_NEEDFIN;
2649 		break;
2650 
2651 	case TCPS_ESTABLISHED:
2652 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
2653 		break;
2654 
2655 	case TCPS_CLOSE_WAIT:
2656 		tcp_state_change(tp, TCPS_LAST_ACK);
2657 		break;
2658 	}
2659 	if (tp->t_acktime == 0)
2660 		tp->t_acktime = ticks;
2661 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
2662 		tcp_free_sackholes(tp);
2663 		soisdisconnected(tptosocket(tp));
2664 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
2665 		if (tp->t_state == TCPS_FIN_WAIT_2) {
2666 			int timeout;
2667 
2668 			timeout = (tcp_fast_finwait2_recycle) ?
2669 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
2670 			tcp_timer_activate(tp, TT_2MSL, timeout);
2671 		}
2672 	}
2673 }
2674 
2675 #ifdef DDB
2676 static void
2677 db_print_indent(int indent)
2678 {
2679 	int i;
2680 
2681 	for (i = 0; i < indent; i++)
2682 		db_printf(" ");
2683 }
2684 
2685 static void
2686 db_print_tstate(int t_state)
2687 {
2688 
2689 	switch (t_state) {
2690 	case TCPS_CLOSED:
2691 		db_printf("TCPS_CLOSED");
2692 		return;
2693 
2694 	case TCPS_LISTEN:
2695 		db_printf("TCPS_LISTEN");
2696 		return;
2697 
2698 	case TCPS_SYN_SENT:
2699 		db_printf("TCPS_SYN_SENT");
2700 		return;
2701 
2702 	case TCPS_SYN_RECEIVED:
2703 		db_printf("TCPS_SYN_RECEIVED");
2704 		return;
2705 
2706 	case TCPS_ESTABLISHED:
2707 		db_printf("TCPS_ESTABLISHED");
2708 		return;
2709 
2710 	case TCPS_CLOSE_WAIT:
2711 		db_printf("TCPS_CLOSE_WAIT");
2712 		return;
2713 
2714 	case TCPS_FIN_WAIT_1:
2715 		db_printf("TCPS_FIN_WAIT_1");
2716 		return;
2717 
2718 	case TCPS_CLOSING:
2719 		db_printf("TCPS_CLOSING");
2720 		return;
2721 
2722 	case TCPS_LAST_ACK:
2723 		db_printf("TCPS_LAST_ACK");
2724 		return;
2725 
2726 	case TCPS_FIN_WAIT_2:
2727 		db_printf("TCPS_FIN_WAIT_2");
2728 		return;
2729 
2730 	case TCPS_TIME_WAIT:
2731 		db_printf("TCPS_TIME_WAIT");
2732 		return;
2733 
2734 	default:
2735 		db_printf("unknown");
2736 		return;
2737 	}
2738 }
2739 
2740 static void
2741 db_print_bblog_state(int state)
2742 {
2743 	switch (state) {
2744 	case TCP_LOG_STATE_RATIO_OFF:
2745 		db_printf("TCP_LOG_STATE_RATIO_OFF");
2746 		break;
2747 	case TCP_LOG_STATE_CLEAR:
2748 		db_printf("TCP_LOG_STATE_CLEAR");
2749 		break;
2750 	case TCP_LOG_STATE_OFF:
2751 		db_printf("TCP_LOG_STATE_OFF");
2752 		break;
2753 	case TCP_LOG_STATE_TAIL:
2754 		db_printf("TCP_LOG_STATE_TAIL");
2755 		break;
2756 	case TCP_LOG_STATE_HEAD:
2757 		db_printf("TCP_LOG_STATE_HEAD");
2758 		break;
2759 	case TCP_LOG_STATE_HEAD_AUTO:
2760 		db_printf("TCP_LOG_STATE_HEAD_AUTO");
2761 		break;
2762 	case TCP_LOG_STATE_CONTINUAL:
2763 		db_printf("TCP_LOG_STATE_CONTINUAL");
2764 		break;
2765 	case TCP_LOG_STATE_TAIL_AUTO:
2766 		db_printf("TCP_LOG_STATE_TAIL_AUTO");
2767 		break;
2768 	case TCP_LOG_VIA_BBPOINTS:
2769 		db_printf("TCP_LOG_STATE_BBPOINTS");
2770 		break;
2771 	default:
2772 		db_printf("UNKNOWN(%d)", state);
2773 		break;
2774 	}
2775 }
2776 
2777 static void
2778 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent, bool show_bblog,
2779     bool show_inpcb, bool only_locked)
2780 {
2781 
2782 	if (only_locked && tp->t_inpcb.inp_lock.rw_lock == RW_UNLOCKED)
2783 		return;
2784 
2785 	db_print_indent(indent);
2786 	db_printf("%s at %p\n", name, tp);
2787 
2788 	indent += 2;
2789 
2790 	if (show_inpcb)
2791 		db_print_inpcb(tptoinpcb(tp), "t_inpcb", indent);
2792 
2793 	db_print_indent(indent);
2794 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2795 	   TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2796 
2797 	db_print_indent(indent);
2798 	db_printf("t_callout: %p   t_timers: %p\n",
2799 	    &tp->t_callout, &tp->t_timers);
2800 
2801 	db_print_indent(indent);
2802 	db_printf("t_state: %d (", tp->t_state);
2803 	db_print_tstate(tp->t_state);
2804 	db_printf(")\n");
2805 
2806 	db_print_indent(indent);
2807 	db_printf("t_flags: 0x%b\n", tp->t_flags, TF_BITS);
2808 
2809 	db_print_indent(indent);
2810 	db_printf("t_flags2: 0x%b\n", tp->t_flags2, TF2_BITS);
2811 
2812 	db_print_indent(indent);
2813 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: 0x%08x\n",
2814 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2815 
2816 	db_print_indent(indent);
2817 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2818 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2819 
2820 	db_print_indent(indent);
2821 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2822 	    tp->iss, tp->irs, tp->rcv_nxt);
2823 
2824 	db_print_indent(indent);
2825 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2826 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2827 
2828 	db_print_indent(indent);
2829 	db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2830 	   tp->snd_wnd, tp->snd_cwnd);
2831 
2832 	db_print_indent(indent);
2833 	db_printf("snd_ssthresh: %u   snd_recover: "
2834 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2835 
2836 	db_print_indent(indent);
2837 	db_printf("t_rcvtime: %u   t_startime: %u\n",
2838 	    tp->t_rcvtime, tp->t_starttime);
2839 
2840 	db_print_indent(indent);
2841 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2842 	    tp->t_rtttime, tp->t_rtseq);
2843 
2844 	db_print_indent(indent);
2845 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2846 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2847 
2848 	db_print_indent(indent);
2849 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u\n",
2850 	    tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin);
2851 
2852 	db_print_indent(indent);
2853 	db_printf("t_rttupdated: %u   max_sndwnd: %u   t_softerror: %d\n",
2854 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2855 
2856 	db_print_indent(indent);
2857 	db_printf("t_oobflags: 0x%b   t_iobc: 0x%02x\n", tp->t_oobflags,
2858 	    TCPOOB_BITS, tp->t_iobc);
2859 
2860 	db_print_indent(indent);
2861 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2862 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2863 
2864 	db_print_indent(indent);
2865 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2866 	    tp->ts_recent, tp->ts_recent_age);
2867 
2868 	db_print_indent(indent);
2869 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2870 	    "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2871 
2872 	db_print_indent(indent);
2873 	db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2874 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2875 	    tp->snd_recover_prev, tp->t_badrxtwin);
2876 
2877 	db_print_indent(indent);
2878 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2879 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2880 
2881 	db_print_indent(indent);
2882 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
2883 	    tp->snd_fack, tp->rcv_numsacks);
2884 
2885 	/* Skip sackblks, sackhint. */
2886 
2887 	db_print_indent(indent);
2888 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2889 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2890 
2891 	db_print_indent(indent);
2892 	db_printf("t_fb.tfb_tcp_block_name: %s\n", tp->t_fb->tfb_tcp_block_name);
2893 
2894 	db_print_indent(indent);
2895 	db_printf("t_cc.name: %s\n", tp->t_cc->name);
2896 
2897 	db_print_indent(indent);
2898 	db_printf("_t_logstate: %d (", tp->_t_logstate);
2899 	db_print_bblog_state(tp->_t_logstate);
2900 	db_printf(")\n");
2901 
2902 	db_print_indent(indent);
2903 	db_printf("t_lognum: %d   t_loglimit: %d   t_logsn: %u\n",
2904 	    tp->t_lognum, tp->t_loglimit, tp->t_logsn);
2905 
2906 	if (show_bblog) {
2907 #ifdef TCP_BLACKBOX
2908 		db_print_bblog_entries(&tp->t_logs, indent);
2909 #else
2910 		db_print_indent(indent);
2911 		db_printf("BBLog not supported\n");
2912 #endif
2913 	}
2914 }
2915 
2916 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2917 {
2918 	struct tcpcb *tp;
2919 	bool show_bblog, show_inpcb;
2920 
2921 	if (!have_addr) {
2922 		db_printf("usage: show tcpcb[/bi] <addr>\n");
2923 		return;
2924 	}
2925 	show_bblog = strchr(modif, 'b') != NULL;
2926 	show_inpcb = strchr(modif, 'i') != NULL;
2927 	tp = (struct tcpcb *)addr;
2928 	db_print_tcpcb(tp, "tcpcb", 0, show_bblog, show_inpcb, false);
2929 }
2930 
2931 DB_SHOW_ALL_COMMAND(tcpcbs, db_show_all_tcpcbs)
2932 {
2933 	VNET_ITERATOR_DECL(vnet_iter);
2934 	struct inpcb *inp;
2935 	bool only_locked, show_bblog, show_inpcb;
2936 
2937 	only_locked = strchr(modif, 'l') != NULL;
2938 	show_bblog = strchr(modif, 'b') != NULL;
2939 	show_inpcb = strchr(modif, 'i') != NULL;
2940 	VNET_FOREACH(vnet_iter) {
2941 		CURVNET_SET(vnet_iter);
2942 		for (u_int i = 0; i <= V_tcbinfo.ipi_porthashmask; i++)
2943 			CK_LIST_FOREACH(inp, &V_tcbinfo.ipi_porthashbase[i],
2944 			    inp_portlist) {
2945 				db_print_tcpcb(intotcpcb(inp), "tcpcb", 0,
2946 				    show_bblog, show_inpcb, only_locked);
2947 				if (db_pager_quit)
2948 					goto break_hash;
2949 			}
2950 break_hash:
2951 		CK_LIST_FOREACH(inp, &V_tcbinfo.ipi_list_unconn,
2952 		    inp_unconn_list) {
2953 			db_print_tcpcb(intotcpcb(inp), "tcpcb", 0,
2954 			    show_bblog, show_inpcb, only_locked);
2955 			if (db_pager_quit)
2956 				break;
2957 		}
2958 		CURVNET_RESTORE();
2959 	}
2960 }
2961 #endif
2962