xref: /freebsd/sys/netinet/tcp_usrreq.c (revision def7fe87e9b28032572ca6f820a260677fd0c2d5)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2006-2007 Robert N. M. Watson
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * All rights reserved.
7  *
8  * Portions of this software were developed by Robert N. M. Watson under
9  * contract to Juniper Networks, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ddb.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_tcpdebug.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/limits.h>
49 #include <sys/malloc.h>
50 #include <sys/refcount.h>
51 #include <sys/kernel.h>
52 #include <sys/sysctl.h>
53 #include <sys/mbuf.h>
54 #ifdef INET6
55 #include <sys/domain.h>
56 #endif /* INET6 */
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/proc.h>
61 #include <sys/jail.h>
62 #include <sys/syslog.h>
63 
64 #ifdef DDB
65 #include <ddb/ddb.h>
66 #endif
67 
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/route.h>
71 #include <net/vnet.h>
72 
73 #include <netinet/in.h>
74 #include <netinet/in_kdtrace.h>
75 #include <netinet/in_pcb.h>
76 #include <netinet/in_systm.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip_var.h>
79 #ifdef INET6
80 #include <netinet/ip6.h>
81 #include <netinet6/in6_pcb.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #endif
85 #ifdef TCP_RFC7413
86 #include <netinet/tcp_fastopen.h>
87 #endif
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/tcpip.h>
94 #include <netinet/cc/cc.h>
95 #ifdef TCPPCAP
96 #include <netinet/tcp_pcap.h>
97 #endif
98 #ifdef TCPDEBUG
99 #include <netinet/tcp_debug.h>
100 #endif
101 #ifdef TCP_OFFLOAD
102 #include <netinet/tcp_offload.h>
103 #endif
104 
105 /*
106  * TCP protocol interface to socket abstraction.
107  */
108 static int	tcp_attach(struct socket *);
109 #ifdef INET
110 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
111 		    struct thread *td);
112 #endif /* INET */
113 #ifdef INET6
114 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
115 		    struct thread *td);
116 #endif /* INET6 */
117 static void	tcp_disconnect(struct tcpcb *);
118 static void	tcp_usrclosed(struct tcpcb *);
119 static void	tcp_fill_info(struct tcpcb *, struct tcp_info *);
120 
121 #ifdef TCPDEBUG
122 #define	TCPDEBUG0	int ostate = 0
123 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
124 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
125 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
126 #else
127 #define	TCPDEBUG0
128 #define	TCPDEBUG1()
129 #define	TCPDEBUG2(req)
130 #endif
131 
132 /*
133  * TCP attaches to socket via pru_attach(), reserving space,
134  * and an internet control block.
135  */
136 static int
137 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
138 {
139 	struct inpcb *inp;
140 	struct tcpcb *tp = NULL;
141 	int error;
142 	TCPDEBUG0;
143 
144 	inp = sotoinpcb(so);
145 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
146 	TCPDEBUG1();
147 
148 	error = tcp_attach(so);
149 	if (error)
150 		goto out;
151 
152 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
153 		so->so_linger = TCP_LINGERTIME;
154 
155 	inp = sotoinpcb(so);
156 	tp = intotcpcb(inp);
157 out:
158 	TCPDEBUG2(PRU_ATTACH);
159 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
160 	return error;
161 }
162 
163 /*
164  * tcp_detach is called when the socket layer loses its final reference
165  * to the socket, be it a file descriptor reference, a reference from TCP,
166  * etc.  At this point, there is only one case in which we will keep around
167  * inpcb state: time wait.
168  *
169  * This function can probably be re-absorbed back into tcp_usr_detach() now
170  * that there is a single detach path.
171  */
172 static void
173 tcp_detach(struct socket *so, struct inpcb *inp)
174 {
175 	struct tcpcb *tp;
176 
177 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
178 	INP_WLOCK_ASSERT(inp);
179 
180 	KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
181 	KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
182 
183 	tp = intotcpcb(inp);
184 
185 	if (inp->inp_flags & INP_TIMEWAIT) {
186 		/*
187 		 * There are two cases to handle: one in which the time wait
188 		 * state is being discarded (INP_DROPPED), and one in which
189 		 * this connection will remain in timewait.  In the former,
190 		 * it is time to discard all state (except tcptw, which has
191 		 * already been discarded by the timewait close code, which
192 		 * should be further up the call stack somewhere).  In the
193 		 * latter case, we detach from the socket, but leave the pcb
194 		 * present until timewait ends.
195 		 *
196 		 * XXXRW: Would it be cleaner to free the tcptw here?
197 		 *
198 		 * Astute question indeed, from twtcp perspective there are
199 		 * three cases to consider:
200 		 *
201 		 * #1 tcp_detach is called at tcptw creation time by
202 		 *  tcp_twstart, then do not discard the newly created tcptw
203 		 *  and leave inpcb present until timewait ends
204 		 * #2 tcp_detach is called at timewait end (or reuse) by
205 		 *  tcp_twclose, then the tcptw has already been discarded
206 		 *  (or reused) and inpcb is freed here
207 		 * #3 tcp_detach is called() after timewait ends (or reuse)
208 		 *  (e.g. by soclose), then tcptw has already been discarded
209 		 *  (or reused) and inpcb is freed here
210 		 *
211 		 *  In all three cases the tcptw should not be freed here.
212 		 */
213 		if (inp->inp_flags & INP_DROPPED) {
214 			in_pcbdetach(inp);
215 			if (__predict_true(tp == NULL)) {
216 				in_pcbfree(inp);
217 			} else {
218 				/*
219 				 * This case should not happen as in TIMEWAIT
220 				 * state the inp should not be destroyed before
221 				 * its tcptw.  If INVARIANTS is defined, panic.
222 				 */
223 #ifdef INVARIANTS
224 				panic("%s: Panic before an inp double-free: "
225 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
226 				    , __func__);
227 #else
228 				log(LOG_ERR, "%s: Avoid an inp double-free: "
229 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
230 				    , __func__);
231 #endif
232 				INP_WUNLOCK(inp);
233 			}
234 		} else {
235 			in_pcbdetach(inp);
236 			INP_WUNLOCK(inp);
237 		}
238 	} else {
239 		/*
240 		 * If the connection is not in timewait, we consider two
241 		 * two conditions: one in which no further processing is
242 		 * necessary (dropped || embryonic), and one in which TCP is
243 		 * not yet done, but no longer requires the socket, so the
244 		 * pcb will persist for the time being.
245 		 *
246 		 * XXXRW: Does the second case still occur?
247 		 */
248 		if (inp->inp_flags & INP_DROPPED ||
249 		    tp->t_state < TCPS_SYN_SENT) {
250 			tcp_discardcb(tp);
251 			in_pcbdetach(inp);
252 			in_pcbfree(inp);
253 		} else {
254 			in_pcbdetach(inp);
255 			INP_WUNLOCK(inp);
256 		}
257 	}
258 }
259 
260 /*
261  * pru_detach() detaches the TCP protocol from the socket.
262  * If the protocol state is non-embryonic, then can't
263  * do this directly: have to initiate a pru_disconnect(),
264  * which may finish later; embryonic TCB's can just
265  * be discarded here.
266  */
267 static void
268 tcp_usr_detach(struct socket *so)
269 {
270 	struct inpcb *inp;
271 	int rlock = 0;
272 
273 	inp = sotoinpcb(so);
274 	KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
275 	if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
276 		INP_INFO_RLOCK(&V_tcbinfo);
277 		rlock = 1;
278 	}
279 	INP_WLOCK(inp);
280 	KASSERT(inp->inp_socket != NULL,
281 	    ("tcp_usr_detach: inp_socket == NULL"));
282 	tcp_detach(so, inp);
283 	if (rlock)
284 		INP_INFO_RUNLOCK(&V_tcbinfo);
285 }
286 
287 #ifdef INET
288 /*
289  * Give the socket an address.
290  */
291 static int
292 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
293 {
294 	int error = 0;
295 	struct inpcb *inp;
296 	struct tcpcb *tp = NULL;
297 	struct sockaddr_in *sinp;
298 
299 	sinp = (struct sockaddr_in *)nam;
300 	if (nam->sa_len != sizeof (*sinp))
301 		return (EINVAL);
302 	/*
303 	 * Must check for multicast addresses and disallow binding
304 	 * to them.
305 	 */
306 	if (sinp->sin_family == AF_INET &&
307 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
308 		return (EAFNOSUPPORT);
309 
310 	TCPDEBUG0;
311 	inp = sotoinpcb(so);
312 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
313 	INP_WLOCK(inp);
314 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
315 		error = EINVAL;
316 		goto out;
317 	}
318 	tp = intotcpcb(inp);
319 	TCPDEBUG1();
320 	INP_HASH_WLOCK(&V_tcbinfo);
321 	error = in_pcbbind(inp, nam, td->td_ucred);
322 	INP_HASH_WUNLOCK(&V_tcbinfo);
323 out:
324 	TCPDEBUG2(PRU_BIND);
325 	TCP_PROBE2(debug__user, tp, PRU_BIND);
326 	INP_WUNLOCK(inp);
327 
328 	return (error);
329 }
330 #endif /* INET */
331 
332 #ifdef INET6
333 static int
334 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
335 {
336 	int error = 0;
337 	struct inpcb *inp;
338 	struct tcpcb *tp = NULL;
339 	struct sockaddr_in6 *sin6p;
340 
341 	sin6p = (struct sockaddr_in6 *)nam;
342 	if (nam->sa_len != sizeof (*sin6p))
343 		return (EINVAL);
344 	/*
345 	 * Must check for multicast addresses and disallow binding
346 	 * to them.
347 	 */
348 	if (sin6p->sin6_family == AF_INET6 &&
349 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
350 		return (EAFNOSUPPORT);
351 
352 	TCPDEBUG0;
353 	inp = sotoinpcb(so);
354 	KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
355 	INP_WLOCK(inp);
356 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
357 		error = EINVAL;
358 		goto out;
359 	}
360 	tp = intotcpcb(inp);
361 	TCPDEBUG1();
362 	INP_HASH_WLOCK(&V_tcbinfo);
363 	inp->inp_vflag &= ~INP_IPV4;
364 	inp->inp_vflag |= INP_IPV6;
365 #ifdef INET
366 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
367 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
368 			inp->inp_vflag |= INP_IPV4;
369 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
370 			struct sockaddr_in sin;
371 
372 			in6_sin6_2_sin(&sin, sin6p);
373 			inp->inp_vflag |= INP_IPV4;
374 			inp->inp_vflag &= ~INP_IPV6;
375 			error = in_pcbbind(inp, (struct sockaddr *)&sin,
376 			    td->td_ucred);
377 			INP_HASH_WUNLOCK(&V_tcbinfo);
378 			goto out;
379 		}
380 	}
381 #endif
382 	error = in6_pcbbind(inp, nam, td->td_ucred);
383 	INP_HASH_WUNLOCK(&V_tcbinfo);
384 out:
385 	TCPDEBUG2(PRU_BIND);
386 	TCP_PROBE2(debug__user, tp, PRU_BIND);
387 	INP_WUNLOCK(inp);
388 	return (error);
389 }
390 #endif /* INET6 */
391 
392 #ifdef INET
393 /*
394  * Prepare to accept connections.
395  */
396 static int
397 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
398 {
399 	int error = 0;
400 	struct inpcb *inp;
401 	struct tcpcb *tp = NULL;
402 
403 	TCPDEBUG0;
404 	inp = sotoinpcb(so);
405 	KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
406 	INP_WLOCK(inp);
407 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
408 		error = EINVAL;
409 		goto out;
410 	}
411 	tp = intotcpcb(inp);
412 	TCPDEBUG1();
413 	SOCK_LOCK(so);
414 	error = solisten_proto_check(so);
415 	INP_HASH_WLOCK(&V_tcbinfo);
416 	if (error == 0 && inp->inp_lport == 0)
417 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
418 	INP_HASH_WUNLOCK(&V_tcbinfo);
419 	if (error == 0) {
420 		tcp_state_change(tp, TCPS_LISTEN);
421 		solisten_proto(so, backlog);
422 #ifdef TCP_OFFLOAD
423 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
424 			tcp_offload_listen_start(tp);
425 #endif
426 	}
427 	SOCK_UNLOCK(so);
428 
429 #ifdef TCP_RFC7413
430 	if (IS_FASTOPEN(tp->t_flags))
431 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
432 #endif
433 out:
434 	TCPDEBUG2(PRU_LISTEN);
435 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
436 	INP_WUNLOCK(inp);
437 	return (error);
438 }
439 #endif /* INET */
440 
441 #ifdef INET6
442 static int
443 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
444 {
445 	int error = 0;
446 	struct inpcb *inp;
447 	struct tcpcb *tp = NULL;
448 
449 	TCPDEBUG0;
450 	inp = sotoinpcb(so);
451 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
452 	INP_WLOCK(inp);
453 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
454 		error = EINVAL;
455 		goto out;
456 	}
457 	tp = intotcpcb(inp);
458 	TCPDEBUG1();
459 	SOCK_LOCK(so);
460 	error = solisten_proto_check(so);
461 	INP_HASH_WLOCK(&V_tcbinfo);
462 	if (error == 0 && inp->inp_lport == 0) {
463 		inp->inp_vflag &= ~INP_IPV4;
464 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
465 			inp->inp_vflag |= INP_IPV4;
466 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
467 	}
468 	INP_HASH_WUNLOCK(&V_tcbinfo);
469 	if (error == 0) {
470 		tcp_state_change(tp, TCPS_LISTEN);
471 		solisten_proto(so, backlog);
472 #ifdef TCP_OFFLOAD
473 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
474 			tcp_offload_listen_start(tp);
475 #endif
476 	}
477 	SOCK_UNLOCK(so);
478 
479 #ifdef TCP_RFC7413
480 	if (IS_FASTOPEN(tp->t_flags))
481 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
482 #endif
483 out:
484 	TCPDEBUG2(PRU_LISTEN);
485 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
486 	INP_WUNLOCK(inp);
487 	return (error);
488 }
489 #endif /* INET6 */
490 
491 #ifdef INET
492 /*
493  * Initiate connection to peer.
494  * Create a template for use in transmissions on this connection.
495  * Enter SYN_SENT state, and mark socket as connecting.
496  * Start keep-alive timer, and seed output sequence space.
497  * Send initial segment on connection.
498  */
499 static int
500 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
501 {
502 	int error = 0;
503 	struct inpcb *inp;
504 	struct tcpcb *tp = NULL;
505 	struct sockaddr_in *sinp;
506 
507 	sinp = (struct sockaddr_in *)nam;
508 	if (nam->sa_len != sizeof (*sinp))
509 		return (EINVAL);
510 	/*
511 	 * Must disallow TCP ``connections'' to multicast addresses.
512 	 */
513 	if (sinp->sin_family == AF_INET
514 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
515 		return (EAFNOSUPPORT);
516 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
517 		return (error);
518 
519 	TCPDEBUG0;
520 	inp = sotoinpcb(so);
521 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
522 	INP_WLOCK(inp);
523 	if (inp->inp_flags & INP_TIMEWAIT) {
524 		error = EADDRINUSE;
525 		goto out;
526 	}
527 	if (inp->inp_flags & INP_DROPPED) {
528 		error = ECONNREFUSED;
529 		goto out;
530 	}
531 	tp = intotcpcb(inp);
532 	TCPDEBUG1();
533 	if ((error = tcp_connect(tp, nam, td)) != 0)
534 		goto out;
535 #ifdef TCP_OFFLOAD
536 	if (registered_toedevs > 0 &&
537 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
538 	    (error = tcp_offload_connect(so, nam)) == 0)
539 		goto out;
540 #endif
541 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
542 	error = tp->t_fb->tfb_tcp_output(tp);
543 out:
544 	TCPDEBUG2(PRU_CONNECT);
545 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
546 	INP_WUNLOCK(inp);
547 	return (error);
548 }
549 #endif /* INET */
550 
551 #ifdef INET6
552 static int
553 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
554 {
555 	int error = 0;
556 	struct inpcb *inp;
557 	struct tcpcb *tp = NULL;
558 	struct sockaddr_in6 *sin6p;
559 
560 	TCPDEBUG0;
561 
562 	sin6p = (struct sockaddr_in6 *)nam;
563 	if (nam->sa_len != sizeof (*sin6p))
564 		return (EINVAL);
565 	/*
566 	 * Must disallow TCP ``connections'' to multicast addresses.
567 	 */
568 	if (sin6p->sin6_family == AF_INET6
569 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
570 		return (EAFNOSUPPORT);
571 
572 	inp = sotoinpcb(so);
573 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
574 	INP_WLOCK(inp);
575 	if (inp->inp_flags & INP_TIMEWAIT) {
576 		error = EADDRINUSE;
577 		goto out;
578 	}
579 	if (inp->inp_flags & INP_DROPPED) {
580 		error = ECONNREFUSED;
581 		goto out;
582 	}
583 	tp = intotcpcb(inp);
584 	TCPDEBUG1();
585 #ifdef INET
586 	/*
587 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
588 	 * therefore probably require the hash lock, which isn't held here.
589 	 * Is this a significant problem?
590 	 */
591 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
592 		struct sockaddr_in sin;
593 
594 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
595 			error = EINVAL;
596 			goto out;
597 		}
598 
599 		in6_sin6_2_sin(&sin, sin6p);
600 		inp->inp_vflag |= INP_IPV4;
601 		inp->inp_vflag &= ~INP_IPV6;
602 		if ((error = prison_remote_ip4(td->td_ucred,
603 		    &sin.sin_addr)) != 0)
604 			goto out;
605 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
606 			goto out;
607 #ifdef TCP_OFFLOAD
608 		if (registered_toedevs > 0 &&
609 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
610 		    (error = tcp_offload_connect(so, nam)) == 0)
611 			goto out;
612 #endif
613 		error = tp->t_fb->tfb_tcp_output(tp);
614 		goto out;
615 	}
616 #endif
617 	inp->inp_vflag &= ~INP_IPV4;
618 	inp->inp_vflag |= INP_IPV6;
619 	inp->inp_inc.inc_flags |= INC_ISIPV6;
620 	if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
621 		goto out;
622 	if ((error = tcp6_connect(tp, nam, td)) != 0)
623 		goto out;
624 #ifdef TCP_OFFLOAD
625 	if (registered_toedevs > 0 &&
626 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
627 	    (error = tcp_offload_connect(so, nam)) == 0)
628 		goto out;
629 #endif
630 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
631 	error = tp->t_fb->tfb_tcp_output(tp);
632 
633 out:
634 	TCPDEBUG2(PRU_CONNECT);
635 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
636 	INP_WUNLOCK(inp);
637 	return (error);
638 }
639 #endif /* INET6 */
640 
641 /*
642  * Initiate disconnect from peer.
643  * If connection never passed embryonic stage, just drop;
644  * else if don't need to let data drain, then can just drop anyways,
645  * else have to begin TCP shutdown process: mark socket disconnecting,
646  * drain unread data, state switch to reflect user close, and
647  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
648  * when peer sends FIN and acks ours.
649  *
650  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
651  */
652 static int
653 tcp_usr_disconnect(struct socket *so)
654 {
655 	struct inpcb *inp;
656 	struct tcpcb *tp = NULL;
657 	int error = 0;
658 
659 	TCPDEBUG0;
660 	INP_INFO_RLOCK(&V_tcbinfo);
661 	inp = sotoinpcb(so);
662 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
663 	INP_WLOCK(inp);
664 	if (inp->inp_flags & INP_TIMEWAIT)
665 		goto out;
666 	if (inp->inp_flags & INP_DROPPED) {
667 		error = ECONNRESET;
668 		goto out;
669 	}
670 	tp = intotcpcb(inp);
671 	TCPDEBUG1();
672 	tcp_disconnect(tp);
673 out:
674 	TCPDEBUG2(PRU_DISCONNECT);
675 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
676 	INP_WUNLOCK(inp);
677 	INP_INFO_RUNLOCK(&V_tcbinfo);
678 	return (error);
679 }
680 
681 #ifdef INET
682 /*
683  * Accept a connection.  Essentially all the work is done at higher levels;
684  * just return the address of the peer, storing through addr.
685  */
686 static int
687 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
688 {
689 	int error = 0;
690 	struct inpcb *inp = NULL;
691 	struct tcpcb *tp = NULL;
692 	struct in_addr addr;
693 	in_port_t port = 0;
694 	TCPDEBUG0;
695 
696 	if (so->so_state & SS_ISDISCONNECTED)
697 		return (ECONNABORTED);
698 
699 	inp = sotoinpcb(so);
700 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
701 	INP_WLOCK(inp);
702 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
703 		error = ECONNABORTED;
704 		goto out;
705 	}
706 	tp = intotcpcb(inp);
707 	TCPDEBUG1();
708 
709 	/*
710 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
711 	 * copy the data of interest and defer the malloc until after we
712 	 * release the lock.
713 	 */
714 	port = inp->inp_fport;
715 	addr = inp->inp_faddr;
716 
717 out:
718 	TCPDEBUG2(PRU_ACCEPT);
719 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
720 	INP_WUNLOCK(inp);
721 	if (error == 0)
722 		*nam = in_sockaddr(port, &addr);
723 	return error;
724 }
725 #endif /* INET */
726 
727 #ifdef INET6
728 static int
729 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
730 {
731 	struct inpcb *inp = NULL;
732 	int error = 0;
733 	struct tcpcb *tp = NULL;
734 	struct in_addr addr;
735 	struct in6_addr addr6;
736 	in_port_t port = 0;
737 	int v4 = 0;
738 	TCPDEBUG0;
739 
740 	if (so->so_state & SS_ISDISCONNECTED)
741 		return (ECONNABORTED);
742 
743 	inp = sotoinpcb(so);
744 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
745 	INP_INFO_RLOCK(&V_tcbinfo);
746 	INP_WLOCK(inp);
747 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
748 		error = ECONNABORTED;
749 		goto out;
750 	}
751 	tp = intotcpcb(inp);
752 	TCPDEBUG1();
753 
754 	/*
755 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
756 	 * copy the data of interest and defer the malloc until after we
757 	 * release the lock.
758 	 */
759 	if (inp->inp_vflag & INP_IPV4) {
760 		v4 = 1;
761 		port = inp->inp_fport;
762 		addr = inp->inp_faddr;
763 	} else {
764 		port = inp->inp_fport;
765 		addr6 = inp->in6p_faddr;
766 	}
767 
768 out:
769 	TCPDEBUG2(PRU_ACCEPT);
770 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
771 	INP_WUNLOCK(inp);
772 	INP_INFO_RUNLOCK(&V_tcbinfo);
773 	if (error == 0) {
774 		if (v4)
775 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
776 		else
777 			*nam = in6_sockaddr(port, &addr6);
778 	}
779 	return error;
780 }
781 #endif /* INET6 */
782 
783 /*
784  * Mark the connection as being incapable of further output.
785  */
786 static int
787 tcp_usr_shutdown(struct socket *so)
788 {
789 	int error = 0;
790 	struct inpcb *inp;
791 	struct tcpcb *tp = NULL;
792 
793 	TCPDEBUG0;
794 	INP_INFO_RLOCK(&V_tcbinfo);
795 	inp = sotoinpcb(so);
796 	KASSERT(inp != NULL, ("inp == NULL"));
797 	INP_WLOCK(inp);
798 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
799 		error = ECONNRESET;
800 		goto out;
801 	}
802 	tp = intotcpcb(inp);
803 	TCPDEBUG1();
804 	socantsendmore(so);
805 	tcp_usrclosed(tp);
806 	if (!(inp->inp_flags & INP_DROPPED))
807 		error = tp->t_fb->tfb_tcp_output(tp);
808 
809 out:
810 	TCPDEBUG2(PRU_SHUTDOWN);
811 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
812 	INP_WUNLOCK(inp);
813 	INP_INFO_RUNLOCK(&V_tcbinfo);
814 
815 	return (error);
816 }
817 
818 /*
819  * After a receive, possibly send window update to peer.
820  */
821 static int
822 tcp_usr_rcvd(struct socket *so, int flags)
823 {
824 	struct inpcb *inp;
825 	struct tcpcb *tp = NULL;
826 	int error = 0;
827 
828 	TCPDEBUG0;
829 	inp = sotoinpcb(so);
830 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
831 	INP_WLOCK(inp);
832 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
833 		error = ECONNRESET;
834 		goto out;
835 	}
836 	tp = intotcpcb(inp);
837 	TCPDEBUG1();
838 #ifdef TCP_RFC7413
839 	/*
840 	 * For passively-created TFO connections, don't attempt a window
841 	 * update while still in SYN_RECEIVED as this may trigger an early
842 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
843 	 * application response data, or failing that, when the DELACK timer
844 	 * expires.
845 	 */
846 	if (IS_FASTOPEN(tp->t_flags) &&
847 	    (tp->t_state == TCPS_SYN_RECEIVED))
848 		goto out;
849 #endif
850 #ifdef TCP_OFFLOAD
851 	if (tp->t_flags & TF_TOE)
852 		tcp_offload_rcvd(tp);
853 	else
854 #endif
855 	tp->t_fb->tfb_tcp_output(tp);
856 
857 out:
858 	TCPDEBUG2(PRU_RCVD);
859 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
860 	INP_WUNLOCK(inp);
861 	return (error);
862 }
863 
864 /*
865  * Do a send by putting data in output queue and updating urgent
866  * marker if URG set.  Possibly send more data.  Unlike the other
867  * pru_*() routines, the mbuf chains are our responsibility.  We
868  * must either enqueue them or free them.  The other pru_* routines
869  * generally are caller-frees.
870  */
871 static int
872 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
873     struct sockaddr *nam, struct mbuf *control, struct thread *td)
874 {
875 	int error = 0;
876 	struct inpcb *inp;
877 	struct tcpcb *tp = NULL;
878 #ifdef INET6
879 	int isipv6;
880 #endif
881 	TCPDEBUG0;
882 
883 	/*
884 	 * We require the pcbinfo lock if we will close the socket as part of
885 	 * this call.
886 	 */
887 	if (flags & PRUS_EOF)
888 		INP_INFO_RLOCK(&V_tcbinfo);
889 	inp = sotoinpcb(so);
890 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
891 	INP_WLOCK(inp);
892 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
893 		if (control)
894 			m_freem(control);
895 		/*
896 		 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
897 		 * for freeing memory.
898 		 */
899 		if (m && (flags & PRUS_NOTREADY) == 0)
900 			m_freem(m);
901 		error = ECONNRESET;
902 		goto out;
903 	}
904 #ifdef INET6
905 	isipv6 = nam && nam->sa_family == AF_INET6;
906 #endif /* INET6 */
907 	tp = intotcpcb(inp);
908 	TCPDEBUG1();
909 	if (control) {
910 		/* TCP doesn't do control messages (rights, creds, etc) */
911 		if (control->m_len) {
912 			m_freem(control);
913 			if (m)
914 				m_freem(m);
915 			error = EINVAL;
916 			goto out;
917 		}
918 		m_freem(control);	/* empty control, just free it */
919 	}
920 	if (!(flags & PRUS_OOB)) {
921 		sbappendstream(&so->so_snd, m, flags);
922 		if (nam && tp->t_state < TCPS_SYN_SENT) {
923 			/*
924 			 * Do implied connect if not yet connected,
925 			 * initialize window to default value, and
926 			 * initialize maxseg using peer's cached MSS.
927 			 */
928 #ifdef INET6
929 			if (isipv6)
930 				error = tcp6_connect(tp, nam, td);
931 #endif /* INET6 */
932 #if defined(INET6) && defined(INET)
933 			else
934 #endif
935 #ifdef INET
936 				error = tcp_connect(tp, nam, td);
937 #endif
938 			if (error)
939 				goto out;
940 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
941 			tcp_mss(tp, -1);
942 		}
943 		if (flags & PRUS_EOF) {
944 			/*
945 			 * Close the send side of the connection after
946 			 * the data is sent.
947 			 */
948 			INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
949 			socantsendmore(so);
950 			tcp_usrclosed(tp);
951 		}
952 		if (!(inp->inp_flags & INP_DROPPED) &&
953 		    !(flags & PRUS_NOTREADY)) {
954 			if (flags & PRUS_MORETOCOME)
955 				tp->t_flags |= TF_MORETOCOME;
956 			error = tp->t_fb->tfb_tcp_output(tp);
957 			if (flags & PRUS_MORETOCOME)
958 				tp->t_flags &= ~TF_MORETOCOME;
959 		}
960 	} else {
961 		/*
962 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
963 		 */
964 		SOCKBUF_LOCK(&so->so_snd);
965 		if (sbspace(&so->so_snd) < -512) {
966 			SOCKBUF_UNLOCK(&so->so_snd);
967 			m_freem(m);
968 			error = ENOBUFS;
969 			goto out;
970 		}
971 		/*
972 		 * According to RFC961 (Assigned Protocols),
973 		 * the urgent pointer points to the last octet
974 		 * of urgent data.  We continue, however,
975 		 * to consider it to indicate the first octet
976 		 * of data past the urgent section.
977 		 * Otherwise, snd_up should be one lower.
978 		 */
979 		sbappendstream_locked(&so->so_snd, m, flags);
980 		SOCKBUF_UNLOCK(&so->so_snd);
981 		if (nam && tp->t_state < TCPS_SYN_SENT) {
982 			/*
983 			 * Do implied connect if not yet connected,
984 			 * initialize window to default value, and
985 			 * initialize maxseg using peer's cached MSS.
986 			 */
987 #ifdef INET6
988 			if (isipv6)
989 				error = tcp6_connect(tp, nam, td);
990 #endif /* INET6 */
991 #if defined(INET6) && defined(INET)
992 			else
993 #endif
994 #ifdef INET
995 				error = tcp_connect(tp, nam, td);
996 #endif
997 			if (error)
998 				goto out;
999 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
1000 			tcp_mss(tp, -1);
1001 		}
1002 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1003 		if (!(flags & PRUS_NOTREADY)) {
1004 			tp->t_flags |= TF_FORCEDATA;
1005 			error = tp->t_fb->tfb_tcp_output(tp);
1006 			tp->t_flags &= ~TF_FORCEDATA;
1007 		}
1008 	}
1009 out:
1010 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1011 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1012 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1013 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1014 	INP_WUNLOCK(inp);
1015 	if (flags & PRUS_EOF)
1016 		INP_INFO_RUNLOCK(&V_tcbinfo);
1017 	return (error);
1018 }
1019 
1020 static int
1021 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1022 {
1023 	struct inpcb *inp;
1024 	struct tcpcb *tp;
1025 	int error;
1026 
1027 	inp = sotoinpcb(so);
1028 	INP_WLOCK(inp);
1029 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1030 		INP_WUNLOCK(inp);
1031 		for (int i = 0; i < count; i++)
1032 			m = m_free(m);
1033 		return (ECONNRESET);
1034 	}
1035 	tp = intotcpcb(inp);
1036 
1037 	SOCKBUF_LOCK(&so->so_snd);
1038 	error = sbready(&so->so_snd, m, count);
1039 	SOCKBUF_UNLOCK(&so->so_snd);
1040 	if (error == 0)
1041 		error = tp->t_fb->tfb_tcp_output(tp);
1042 	INP_WUNLOCK(inp);
1043 
1044 	return (error);
1045 }
1046 
1047 /*
1048  * Abort the TCP.  Drop the connection abruptly.
1049  */
1050 static void
1051 tcp_usr_abort(struct socket *so)
1052 {
1053 	struct inpcb *inp;
1054 	struct tcpcb *tp = NULL;
1055 	TCPDEBUG0;
1056 
1057 	inp = sotoinpcb(so);
1058 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1059 
1060 	INP_INFO_RLOCK(&V_tcbinfo);
1061 	INP_WLOCK(inp);
1062 	KASSERT(inp->inp_socket != NULL,
1063 	    ("tcp_usr_abort: inp_socket == NULL"));
1064 
1065 	/*
1066 	 * If we still have full TCP state, and we're not dropped, drop.
1067 	 */
1068 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1069 	    !(inp->inp_flags & INP_DROPPED)) {
1070 		tp = intotcpcb(inp);
1071 		TCPDEBUG1();
1072 		tcp_drop(tp, ECONNABORTED);
1073 		TCPDEBUG2(PRU_ABORT);
1074 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1075 	}
1076 	if (!(inp->inp_flags & INP_DROPPED)) {
1077 		SOCK_LOCK(so);
1078 		so->so_state |= SS_PROTOREF;
1079 		SOCK_UNLOCK(so);
1080 		inp->inp_flags |= INP_SOCKREF;
1081 	}
1082 	INP_WUNLOCK(inp);
1083 	INP_INFO_RUNLOCK(&V_tcbinfo);
1084 }
1085 
1086 /*
1087  * TCP socket is closed.  Start friendly disconnect.
1088  */
1089 static void
1090 tcp_usr_close(struct socket *so)
1091 {
1092 	struct inpcb *inp;
1093 	struct tcpcb *tp = NULL;
1094 	TCPDEBUG0;
1095 
1096 	inp = sotoinpcb(so);
1097 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1098 
1099 	INP_INFO_RLOCK(&V_tcbinfo);
1100 	INP_WLOCK(inp);
1101 	KASSERT(inp->inp_socket != NULL,
1102 	    ("tcp_usr_close: inp_socket == NULL"));
1103 
1104 	/*
1105 	 * If we still have full TCP state, and we're not dropped, initiate
1106 	 * a disconnect.
1107 	 */
1108 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1109 	    !(inp->inp_flags & INP_DROPPED)) {
1110 		tp = intotcpcb(inp);
1111 		TCPDEBUG1();
1112 		tcp_disconnect(tp);
1113 		TCPDEBUG2(PRU_CLOSE);
1114 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1115 	}
1116 	if (!(inp->inp_flags & INP_DROPPED)) {
1117 		SOCK_LOCK(so);
1118 		so->so_state |= SS_PROTOREF;
1119 		SOCK_UNLOCK(so);
1120 		inp->inp_flags |= INP_SOCKREF;
1121 	}
1122 	INP_WUNLOCK(inp);
1123 	INP_INFO_RUNLOCK(&V_tcbinfo);
1124 }
1125 
1126 /*
1127  * Receive out-of-band data.
1128  */
1129 static int
1130 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1131 {
1132 	int error = 0;
1133 	struct inpcb *inp;
1134 	struct tcpcb *tp = NULL;
1135 
1136 	TCPDEBUG0;
1137 	inp = sotoinpcb(so);
1138 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1139 	INP_WLOCK(inp);
1140 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1141 		error = ECONNRESET;
1142 		goto out;
1143 	}
1144 	tp = intotcpcb(inp);
1145 	TCPDEBUG1();
1146 	if ((so->so_oobmark == 0 &&
1147 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1148 	    so->so_options & SO_OOBINLINE ||
1149 	    tp->t_oobflags & TCPOOB_HADDATA) {
1150 		error = EINVAL;
1151 		goto out;
1152 	}
1153 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1154 		error = EWOULDBLOCK;
1155 		goto out;
1156 	}
1157 	m->m_len = 1;
1158 	*mtod(m, caddr_t) = tp->t_iobc;
1159 	if ((flags & MSG_PEEK) == 0)
1160 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1161 
1162 out:
1163 	TCPDEBUG2(PRU_RCVOOB);
1164 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1165 	INP_WUNLOCK(inp);
1166 	return (error);
1167 }
1168 
1169 #ifdef INET
1170 struct pr_usrreqs tcp_usrreqs = {
1171 	.pru_abort =		tcp_usr_abort,
1172 	.pru_accept =		tcp_usr_accept,
1173 	.pru_attach =		tcp_usr_attach,
1174 	.pru_bind =		tcp_usr_bind,
1175 	.pru_connect =		tcp_usr_connect,
1176 	.pru_control =		in_control,
1177 	.pru_detach =		tcp_usr_detach,
1178 	.pru_disconnect =	tcp_usr_disconnect,
1179 	.pru_listen =		tcp_usr_listen,
1180 	.pru_peeraddr =		in_getpeeraddr,
1181 	.pru_rcvd =		tcp_usr_rcvd,
1182 	.pru_rcvoob =		tcp_usr_rcvoob,
1183 	.pru_send =		tcp_usr_send,
1184 	.pru_ready =		tcp_usr_ready,
1185 	.pru_shutdown =		tcp_usr_shutdown,
1186 	.pru_sockaddr =		in_getsockaddr,
1187 	.pru_sosetlabel =	in_pcbsosetlabel,
1188 	.pru_close =		tcp_usr_close,
1189 };
1190 #endif /* INET */
1191 
1192 #ifdef INET6
1193 struct pr_usrreqs tcp6_usrreqs = {
1194 	.pru_abort =		tcp_usr_abort,
1195 	.pru_accept =		tcp6_usr_accept,
1196 	.pru_attach =		tcp_usr_attach,
1197 	.pru_bind =		tcp6_usr_bind,
1198 	.pru_connect =		tcp6_usr_connect,
1199 	.pru_control =		in6_control,
1200 	.pru_detach =		tcp_usr_detach,
1201 	.pru_disconnect =	tcp_usr_disconnect,
1202 	.pru_listen =		tcp6_usr_listen,
1203 	.pru_peeraddr =		in6_mapped_peeraddr,
1204 	.pru_rcvd =		tcp_usr_rcvd,
1205 	.pru_rcvoob =		tcp_usr_rcvoob,
1206 	.pru_send =		tcp_usr_send,
1207 	.pru_ready =		tcp_usr_ready,
1208 	.pru_shutdown =		tcp_usr_shutdown,
1209 	.pru_sockaddr =		in6_mapped_sockaddr,
1210 	.pru_sosetlabel =	in_pcbsosetlabel,
1211 	.pru_close =		tcp_usr_close,
1212 };
1213 #endif /* INET6 */
1214 
1215 #ifdef INET
1216 /*
1217  * Common subroutine to open a TCP connection to remote host specified
1218  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1219  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1220  * to choose a local host address (interface).  If there is an existing
1221  * incarnation of the same connection in TIME-WAIT state and if the remote
1222  * host was sending CC options and if the connection duration was < MSL, then
1223  * truncate the previous TIME-WAIT state and proceed.
1224  * Initialize connection parameters and enter SYN-SENT state.
1225  */
1226 static int
1227 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1228 {
1229 	struct inpcb *inp = tp->t_inpcb, *oinp;
1230 	struct socket *so = inp->inp_socket;
1231 	struct in_addr laddr;
1232 	u_short lport;
1233 	int error;
1234 
1235 	INP_WLOCK_ASSERT(inp);
1236 	INP_HASH_WLOCK(&V_tcbinfo);
1237 
1238 	if (inp->inp_lport == 0) {
1239 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1240 		if (error)
1241 			goto out;
1242 	}
1243 
1244 	/*
1245 	 * Cannot simply call in_pcbconnect, because there might be an
1246 	 * earlier incarnation of this same connection still in
1247 	 * TIME_WAIT state, creating an ADDRINUSE error.
1248 	 */
1249 	laddr = inp->inp_laddr;
1250 	lport = inp->inp_lport;
1251 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1252 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1253 	if (error && oinp == NULL)
1254 		goto out;
1255 	if (oinp) {
1256 		error = EADDRINUSE;
1257 		goto out;
1258 	}
1259 	inp->inp_laddr = laddr;
1260 	in_pcbrehash(inp);
1261 	INP_HASH_WUNLOCK(&V_tcbinfo);
1262 
1263 	/*
1264 	 * Compute window scaling to request:
1265 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1266 	 * XXX: This should move to tcp_output().
1267 	 */
1268 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1269 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1270 		tp->request_r_scale++;
1271 
1272 	soisconnecting(so);
1273 	TCPSTAT_INC(tcps_connattempt);
1274 	tcp_state_change(tp, TCPS_SYN_SENT);
1275 	tp->iss = tcp_new_isn(tp);
1276 	tcp_sendseqinit(tp);
1277 
1278 	return 0;
1279 
1280 out:
1281 	INP_HASH_WUNLOCK(&V_tcbinfo);
1282 	return (error);
1283 }
1284 #endif /* INET */
1285 
1286 #ifdef INET6
1287 static int
1288 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1289 {
1290 	struct inpcb *inp = tp->t_inpcb;
1291 	int error;
1292 
1293 	INP_WLOCK_ASSERT(inp);
1294 	INP_HASH_WLOCK(&V_tcbinfo);
1295 
1296 	if (inp->inp_lport == 0) {
1297 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1298 		if (error)
1299 			goto out;
1300 	}
1301 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1302 	if (error != 0)
1303 		goto out;
1304 	INP_HASH_WUNLOCK(&V_tcbinfo);
1305 
1306 	/* Compute window scaling to request.  */
1307 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1308 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1309 		tp->request_r_scale++;
1310 
1311 	soisconnecting(inp->inp_socket);
1312 	TCPSTAT_INC(tcps_connattempt);
1313 	tcp_state_change(tp, TCPS_SYN_SENT);
1314 	tp->iss = tcp_new_isn(tp);
1315 	tcp_sendseqinit(tp);
1316 
1317 	return 0;
1318 
1319 out:
1320 	INP_HASH_WUNLOCK(&V_tcbinfo);
1321 	return error;
1322 }
1323 #endif /* INET6 */
1324 
1325 /*
1326  * Export TCP internal state information via a struct tcp_info, based on the
1327  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1328  * (TCP state machine, etc).  We export all information using FreeBSD-native
1329  * constants -- for example, the numeric values for tcpi_state will differ
1330  * from Linux.
1331  */
1332 static void
1333 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1334 {
1335 
1336 	INP_WLOCK_ASSERT(tp->t_inpcb);
1337 	bzero(ti, sizeof(*ti));
1338 
1339 	ti->tcpi_state = tp->t_state;
1340 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1341 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1342 	if (tp->t_flags & TF_SACK_PERMIT)
1343 		ti->tcpi_options |= TCPI_OPT_SACK;
1344 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1345 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1346 		ti->tcpi_snd_wscale = tp->snd_scale;
1347 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1348 	}
1349 	if (tp->t_flags & TF_ECN_PERMIT)
1350 		ti->tcpi_options |= TCPI_OPT_ECN;
1351 
1352 	ti->tcpi_rto = tp->t_rxtcur * tick;
1353 	ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1354 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1355 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1356 
1357 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1358 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1359 
1360 	/*
1361 	 * FreeBSD-specific extension fields for tcp_info.
1362 	 */
1363 	ti->tcpi_rcv_space = tp->rcv_wnd;
1364 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1365 	ti->tcpi_snd_wnd = tp->snd_wnd;
1366 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1367 	ti->tcpi_snd_nxt = tp->snd_nxt;
1368 	ti->tcpi_snd_mss = tp->t_maxseg;
1369 	ti->tcpi_rcv_mss = tp->t_maxseg;
1370 	if (tp->t_flags & TF_TOE)
1371 		ti->tcpi_options |= TCPI_OPT_TOE;
1372 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1373 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1374 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1375 }
1376 
1377 /*
1378  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1379  * socket option arguments.  When it re-acquires the lock after the copy, it
1380  * has to revalidate that the connection is still valid for the socket
1381  * option.
1382  */
1383 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1384 	INP_WLOCK(inp);							\
1385 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1386 		INP_WUNLOCK(inp);					\
1387 		cleanup;						\
1388 		return (ECONNRESET);					\
1389 	}								\
1390 	tp = intotcpcb(inp);						\
1391 } while(0)
1392 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1393 
1394 int
1395 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1396 {
1397 	int	error;
1398 	struct	inpcb *inp;
1399 	struct	tcpcb *tp;
1400 	struct tcp_function_block *blk;
1401 	struct tcp_function_set fsn;
1402 
1403 	error = 0;
1404 	inp = sotoinpcb(so);
1405 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1406 	INP_WLOCK(inp);
1407 	if (sopt->sopt_level != IPPROTO_TCP) {
1408 #ifdef INET6
1409 		if (inp->inp_vflag & INP_IPV6PROTO) {
1410 			INP_WUNLOCK(inp);
1411 			error = ip6_ctloutput(so, sopt);
1412 		}
1413 #endif /* INET6 */
1414 #if defined(INET6) && defined(INET)
1415 		else
1416 #endif
1417 #ifdef INET
1418 		{
1419 			INP_WUNLOCK(inp);
1420 			error = ip_ctloutput(so, sopt);
1421 		}
1422 #endif
1423 		return (error);
1424 	}
1425 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1426 		INP_WUNLOCK(inp);
1427 		return (ECONNRESET);
1428 	}
1429 	tp = intotcpcb(inp);
1430 	/*
1431 	 * Protect the TCP option TCP_FUNCTION_BLK so
1432 	 * that a sub-function can *never* overwrite this.
1433 	 */
1434 	if ((sopt->sopt_dir == SOPT_SET) &&
1435 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1436 		INP_WUNLOCK(inp);
1437 		error = sooptcopyin(sopt, &fsn, sizeof fsn,
1438 		    sizeof fsn);
1439 		if (error)
1440 			return (error);
1441 		INP_WLOCK_RECHECK(inp);
1442 		blk = find_and_ref_tcp_functions(&fsn);
1443 		if (blk == NULL) {
1444 			INP_WUNLOCK(inp);
1445 			return (ENOENT);
1446 		}
1447 		if (tp->t_fb == blk) {
1448 			/* You already have this */
1449 			refcount_release(&blk->tfb_refcnt);
1450 			INP_WUNLOCK(inp);
1451 			return (0);
1452 		}
1453 		if (tp->t_state != TCPS_CLOSED) {
1454 			int error=EINVAL;
1455 			/*
1456 			 * The user has advanced the state
1457 			 * past the initial point, we may not
1458 			 * be able to switch.
1459 			 */
1460 			if (blk->tfb_tcp_handoff_ok != NULL) {
1461 				/*
1462 				 * Does the stack provide a
1463 				 * query mechanism, if so it may
1464 				 * still be possible?
1465 				 */
1466 				error = (*blk->tfb_tcp_handoff_ok)(tp);
1467 			}
1468 			if (error) {
1469 				refcount_release(&blk->tfb_refcnt);
1470 				INP_WUNLOCK(inp);
1471 				return(error);
1472 			}
1473 		}
1474 		if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1475 			refcount_release(&blk->tfb_refcnt);
1476 			INP_WUNLOCK(inp);
1477 			return (ENOENT);
1478 		}
1479 		/*
1480 		 * Release the old refcnt, the
1481 		 * lookup acquired a ref on the
1482 		 * new one already.
1483 		 */
1484 		if (tp->t_fb->tfb_tcp_fb_fini) {
1485 			/*
1486 			 * Tell the stack to cleanup with 0 i.e.
1487 			 * the tcb is not going away.
1488 			 */
1489 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1490 		}
1491 		refcount_release(&tp->t_fb->tfb_refcnt);
1492 		tp->t_fb = blk;
1493 		if (tp->t_fb->tfb_tcp_fb_init) {
1494 			(*tp->t_fb->tfb_tcp_fb_init)(tp);
1495 		}
1496 #ifdef TCP_OFFLOAD
1497 		if (tp->t_flags & TF_TOE) {
1498 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1499 			     sopt->sopt_name);
1500 		}
1501 #endif
1502 		INP_WUNLOCK(inp);
1503 		return (error);
1504 	} else if ((sopt->sopt_dir == SOPT_GET) &&
1505 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1506 		strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1507 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1508 		INP_WUNLOCK(inp);
1509 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1510 		return (error);
1511 	}
1512 	/* Pass in the INP locked, called must unlock it */
1513 	return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1514 }
1515 
1516 int
1517 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1518 {
1519 	int	error, opt, optval;
1520 	u_int	ui;
1521 	struct	tcp_info ti;
1522 	struct cc_algo *algo;
1523 	char	*pbuf, buf[TCP_CA_NAME_MAX];
1524 	size_t	len;
1525 
1526 	/*
1527 	 * For TCP_CCALGOOPT forward the control to CC module, for both
1528 	 * SOPT_SET and SOPT_GET.
1529 	 */
1530 	switch (sopt->sopt_name) {
1531 	case TCP_CCALGOOPT:
1532 		INP_WUNLOCK(inp);
1533 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1534 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1535 		    sopt->sopt_valsize);
1536 		if (error) {
1537 			free(pbuf, M_TEMP);
1538 			return (error);
1539 		}
1540 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1541 		if (CC_ALGO(tp)->ctl_output != NULL)
1542 			error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1543 		else
1544 			error = ENOENT;
1545 		INP_WUNLOCK(inp);
1546 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
1547 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1548 		free(pbuf, M_TEMP);
1549 		return (error);
1550 	}
1551 
1552 	switch (sopt->sopt_dir) {
1553 	case SOPT_SET:
1554 		switch (sopt->sopt_name) {
1555 #ifdef TCP_SIGNATURE
1556 		case TCP_MD5SIG:
1557 			INP_WUNLOCK(inp);
1558 			error = sooptcopyin(sopt, &optval, sizeof optval,
1559 			    sizeof optval);
1560 			if (error)
1561 				return (error);
1562 
1563 			INP_WLOCK_RECHECK(inp);
1564 			if (optval > 0)
1565 				tp->t_flags |= TF_SIGNATURE;
1566 			else
1567 				tp->t_flags &= ~TF_SIGNATURE;
1568 			goto unlock_and_done;
1569 #endif /* TCP_SIGNATURE */
1570 
1571 		case TCP_NODELAY:
1572 		case TCP_NOOPT:
1573 			INP_WUNLOCK(inp);
1574 			error = sooptcopyin(sopt, &optval, sizeof optval,
1575 			    sizeof optval);
1576 			if (error)
1577 				return (error);
1578 
1579 			INP_WLOCK_RECHECK(inp);
1580 			switch (sopt->sopt_name) {
1581 			case TCP_NODELAY:
1582 				opt = TF_NODELAY;
1583 				break;
1584 			case TCP_NOOPT:
1585 				opt = TF_NOOPT;
1586 				break;
1587 			default:
1588 				opt = 0; /* dead code to fool gcc */
1589 				break;
1590 			}
1591 
1592 			if (optval)
1593 				tp->t_flags |= opt;
1594 			else
1595 				tp->t_flags &= ~opt;
1596 unlock_and_done:
1597 #ifdef TCP_OFFLOAD
1598 			if (tp->t_flags & TF_TOE) {
1599 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
1600 				    sopt->sopt_name);
1601 			}
1602 #endif
1603 			INP_WUNLOCK(inp);
1604 			break;
1605 
1606 		case TCP_NOPUSH:
1607 			INP_WUNLOCK(inp);
1608 			error = sooptcopyin(sopt, &optval, sizeof optval,
1609 			    sizeof optval);
1610 			if (error)
1611 				return (error);
1612 
1613 			INP_WLOCK_RECHECK(inp);
1614 			if (optval)
1615 				tp->t_flags |= TF_NOPUSH;
1616 			else if (tp->t_flags & TF_NOPUSH) {
1617 				tp->t_flags &= ~TF_NOPUSH;
1618 				if (TCPS_HAVEESTABLISHED(tp->t_state))
1619 					error = tp->t_fb->tfb_tcp_output(tp);
1620 			}
1621 			goto unlock_and_done;
1622 
1623 		case TCP_MAXSEG:
1624 			INP_WUNLOCK(inp);
1625 			error = sooptcopyin(sopt, &optval, sizeof optval,
1626 			    sizeof optval);
1627 			if (error)
1628 				return (error);
1629 
1630 			INP_WLOCK_RECHECK(inp);
1631 			if (optval > 0 && optval <= tp->t_maxseg &&
1632 			    optval + 40 >= V_tcp_minmss)
1633 				tp->t_maxseg = optval;
1634 			else
1635 				error = EINVAL;
1636 			goto unlock_and_done;
1637 
1638 		case TCP_INFO:
1639 			INP_WUNLOCK(inp);
1640 			error = EINVAL;
1641 			break;
1642 
1643 		case TCP_CONGESTION:
1644 			INP_WUNLOCK(inp);
1645 			error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1646 			if (error)
1647 				break;
1648 			buf[sopt->sopt_valsize] = '\0';
1649 			INP_WLOCK_RECHECK(inp);
1650 			CC_LIST_RLOCK();
1651 			STAILQ_FOREACH(algo, &cc_list, entries)
1652 				if (strncmp(buf, algo->name,
1653 				    TCP_CA_NAME_MAX) == 0)
1654 					break;
1655 			CC_LIST_RUNLOCK();
1656 			if (algo == NULL) {
1657 				INP_WUNLOCK(inp);
1658 				error = EINVAL;
1659 				break;
1660 			}
1661 			/*
1662 			 * We hold a write lock over the tcb so it's safe to
1663 			 * do these things without ordering concerns.
1664 			 */
1665 			if (CC_ALGO(tp)->cb_destroy != NULL)
1666 				CC_ALGO(tp)->cb_destroy(tp->ccv);
1667 			CC_ALGO(tp) = algo;
1668 			/*
1669 			 * If something goes pear shaped initialising the new
1670 			 * algo, fall back to newreno (which does not
1671 			 * require initialisation).
1672 			 */
1673 			if (algo->cb_init != NULL &&
1674 			    algo->cb_init(tp->ccv) != 0) {
1675 				CC_ALGO(tp) = &newreno_cc_algo;
1676 				/*
1677 				 * The only reason init should fail is
1678 				 * because of malloc.
1679 				 */
1680 				error = ENOMEM;
1681 			}
1682 			INP_WUNLOCK(inp);
1683 			break;
1684 
1685 		case TCP_KEEPIDLE:
1686 		case TCP_KEEPINTVL:
1687 		case TCP_KEEPINIT:
1688 			INP_WUNLOCK(inp);
1689 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1690 			if (error)
1691 				return (error);
1692 
1693 			if (ui > (UINT_MAX / hz)) {
1694 				error = EINVAL;
1695 				break;
1696 			}
1697 			ui *= hz;
1698 
1699 			INP_WLOCK_RECHECK(inp);
1700 			switch (sopt->sopt_name) {
1701 			case TCP_KEEPIDLE:
1702 				tp->t_keepidle = ui;
1703 				/*
1704 				 * XXX: better check current remaining
1705 				 * timeout and "merge" it with new value.
1706 				 */
1707 				if ((tp->t_state > TCPS_LISTEN) &&
1708 				    (tp->t_state <= TCPS_CLOSING))
1709 					tcp_timer_activate(tp, TT_KEEP,
1710 					    TP_KEEPIDLE(tp));
1711 				break;
1712 			case TCP_KEEPINTVL:
1713 				tp->t_keepintvl = ui;
1714 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1715 				    (TP_MAXIDLE(tp) > 0))
1716 					tcp_timer_activate(tp, TT_2MSL,
1717 					    TP_MAXIDLE(tp));
1718 				break;
1719 			case TCP_KEEPINIT:
1720 				tp->t_keepinit = ui;
1721 				if (tp->t_state == TCPS_SYN_RECEIVED ||
1722 				    tp->t_state == TCPS_SYN_SENT)
1723 					tcp_timer_activate(tp, TT_KEEP,
1724 					    TP_KEEPINIT(tp));
1725 				break;
1726 			}
1727 			goto unlock_and_done;
1728 
1729 		case TCP_KEEPCNT:
1730 			INP_WUNLOCK(inp);
1731 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1732 			if (error)
1733 				return (error);
1734 
1735 			INP_WLOCK_RECHECK(inp);
1736 			tp->t_keepcnt = ui;
1737 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1738 			    (TP_MAXIDLE(tp) > 0))
1739 				tcp_timer_activate(tp, TT_2MSL,
1740 				    TP_MAXIDLE(tp));
1741 			goto unlock_and_done;
1742 
1743 #ifdef TCPPCAP
1744 		case TCP_PCAP_OUT:
1745 		case TCP_PCAP_IN:
1746 			INP_WUNLOCK(inp);
1747 			error = sooptcopyin(sopt, &optval, sizeof optval,
1748 			    sizeof optval);
1749 			if (error)
1750 				return (error);
1751 
1752 			INP_WLOCK_RECHECK(inp);
1753 			if (optval >= 0)
1754 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1755 					&(tp->t_outpkts) : &(tp->t_inpkts),
1756 					optval);
1757 			else
1758 				error = EINVAL;
1759 			goto unlock_and_done;
1760 #endif
1761 
1762 #ifdef TCP_RFC7413
1763 		case TCP_FASTOPEN:
1764 			INP_WUNLOCK(inp);
1765 			if (!V_tcp_fastopen_enabled)
1766 				return (EPERM);
1767 
1768 			error = sooptcopyin(sopt, &optval, sizeof optval,
1769 			    sizeof optval);
1770 			if (error)
1771 				return (error);
1772 
1773 			INP_WLOCK_RECHECK(inp);
1774 			if (optval) {
1775 				tp->t_flags |= TF_FASTOPEN;
1776 				if ((tp->t_state == TCPS_LISTEN) &&
1777 				    (tp->t_tfo_pending == NULL))
1778 					tp->t_tfo_pending =
1779 					    tcp_fastopen_alloc_counter();
1780 			} else
1781 				tp->t_flags &= ~TF_FASTOPEN;
1782 			goto unlock_and_done;
1783 #endif
1784 
1785 		default:
1786 			INP_WUNLOCK(inp);
1787 			error = ENOPROTOOPT;
1788 			break;
1789 		}
1790 		break;
1791 
1792 	case SOPT_GET:
1793 		tp = intotcpcb(inp);
1794 		switch (sopt->sopt_name) {
1795 #ifdef TCP_SIGNATURE
1796 		case TCP_MD5SIG:
1797 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1798 			INP_WUNLOCK(inp);
1799 			error = sooptcopyout(sopt, &optval, sizeof optval);
1800 			break;
1801 #endif
1802 
1803 		case TCP_NODELAY:
1804 			optval = tp->t_flags & TF_NODELAY;
1805 			INP_WUNLOCK(inp);
1806 			error = sooptcopyout(sopt, &optval, sizeof optval);
1807 			break;
1808 		case TCP_MAXSEG:
1809 			optval = tp->t_maxseg;
1810 			INP_WUNLOCK(inp);
1811 			error = sooptcopyout(sopt, &optval, sizeof optval);
1812 			break;
1813 		case TCP_NOOPT:
1814 			optval = tp->t_flags & TF_NOOPT;
1815 			INP_WUNLOCK(inp);
1816 			error = sooptcopyout(sopt, &optval, sizeof optval);
1817 			break;
1818 		case TCP_NOPUSH:
1819 			optval = tp->t_flags & TF_NOPUSH;
1820 			INP_WUNLOCK(inp);
1821 			error = sooptcopyout(sopt, &optval, sizeof optval);
1822 			break;
1823 		case TCP_INFO:
1824 			tcp_fill_info(tp, &ti);
1825 			INP_WUNLOCK(inp);
1826 			error = sooptcopyout(sopt, &ti, sizeof ti);
1827 			break;
1828 		case TCP_CONGESTION:
1829 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1830 			INP_WUNLOCK(inp);
1831 			error = sooptcopyout(sopt, buf, len + 1);
1832 			break;
1833 		case TCP_KEEPIDLE:
1834 		case TCP_KEEPINTVL:
1835 		case TCP_KEEPINIT:
1836 		case TCP_KEEPCNT:
1837 			switch (sopt->sopt_name) {
1838 			case TCP_KEEPIDLE:
1839 				ui = TP_KEEPIDLE(tp) / hz;
1840 				break;
1841 			case TCP_KEEPINTVL:
1842 				ui = TP_KEEPINTVL(tp) / hz;
1843 				break;
1844 			case TCP_KEEPINIT:
1845 				ui = TP_KEEPINIT(tp) / hz;
1846 				break;
1847 			case TCP_KEEPCNT:
1848 				ui = TP_KEEPCNT(tp);
1849 				break;
1850 			}
1851 			INP_WUNLOCK(inp);
1852 			error = sooptcopyout(sopt, &ui, sizeof(ui));
1853 			break;
1854 #ifdef TCPPCAP
1855 		case TCP_PCAP_OUT:
1856 		case TCP_PCAP_IN:
1857 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1858 					&(tp->t_outpkts) : &(tp->t_inpkts));
1859 			INP_WUNLOCK(inp);
1860 			error = sooptcopyout(sopt, &optval, sizeof optval);
1861 			break;
1862 #endif
1863 
1864 #ifdef TCP_RFC7413
1865 		case TCP_FASTOPEN:
1866 			optval = tp->t_flags & TF_FASTOPEN;
1867 			INP_WUNLOCK(inp);
1868 			error = sooptcopyout(sopt, &optval, sizeof optval);
1869 			break;
1870 #endif
1871 		default:
1872 			INP_WUNLOCK(inp);
1873 			error = ENOPROTOOPT;
1874 			break;
1875 		}
1876 		break;
1877 	}
1878 	return (error);
1879 }
1880 #undef INP_WLOCK_RECHECK
1881 #undef INP_WLOCK_RECHECK_CLEANUP
1882 
1883 /*
1884  * Attach TCP protocol to socket, allocating
1885  * internet protocol control block, tcp control block,
1886  * bufer space, and entering LISTEN state if to accept connections.
1887  */
1888 static int
1889 tcp_attach(struct socket *so)
1890 {
1891 	struct tcpcb *tp;
1892 	struct inpcb *inp;
1893 	int error;
1894 
1895 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1896 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1897 		if (error)
1898 			return (error);
1899 	}
1900 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
1901 	so->so_snd.sb_flags |= SB_AUTOSIZE;
1902 	INP_INFO_RLOCK(&V_tcbinfo);
1903 	error = in_pcballoc(so, &V_tcbinfo);
1904 	if (error) {
1905 		INP_INFO_RUNLOCK(&V_tcbinfo);
1906 		return (error);
1907 	}
1908 	inp = sotoinpcb(so);
1909 #ifdef INET6
1910 	if (inp->inp_vflag & INP_IPV6PROTO) {
1911 		inp->inp_vflag |= INP_IPV6;
1912 		inp->in6p_hops = -1;	/* use kernel default */
1913 	}
1914 	else
1915 #endif
1916 	inp->inp_vflag |= INP_IPV4;
1917 	tp = tcp_newtcpcb(inp);
1918 	if (tp == NULL) {
1919 		in_pcbdetach(inp);
1920 		in_pcbfree(inp);
1921 		INP_INFO_RUNLOCK(&V_tcbinfo);
1922 		return (ENOBUFS);
1923 	}
1924 	tp->t_state = TCPS_CLOSED;
1925 	INP_WUNLOCK(inp);
1926 	INP_INFO_RUNLOCK(&V_tcbinfo);
1927 	TCPSTATES_INC(TCPS_CLOSED);
1928 	return (0);
1929 }
1930 
1931 /*
1932  * Initiate (or continue) disconnect.
1933  * If embryonic state, just send reset (once).
1934  * If in ``let data drain'' option and linger null, just drop.
1935  * Otherwise (hard), mark socket disconnecting and drop
1936  * current input data; switch states based on user close, and
1937  * send segment to peer (with FIN).
1938  */
1939 static void
1940 tcp_disconnect(struct tcpcb *tp)
1941 {
1942 	struct inpcb *inp = tp->t_inpcb;
1943 	struct socket *so = inp->inp_socket;
1944 
1945 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1946 	INP_WLOCK_ASSERT(inp);
1947 
1948 	/*
1949 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1950 	 * socket is still open.
1951 	 */
1952 	if (tp->t_state < TCPS_ESTABLISHED) {
1953 		tp = tcp_close(tp);
1954 		KASSERT(tp != NULL,
1955 		    ("tcp_disconnect: tcp_close() returned NULL"));
1956 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1957 		tp = tcp_drop(tp, 0);
1958 		KASSERT(tp != NULL,
1959 		    ("tcp_disconnect: tcp_drop() returned NULL"));
1960 	} else {
1961 		soisdisconnecting(so);
1962 		sbflush(&so->so_rcv);
1963 		tcp_usrclosed(tp);
1964 		if (!(inp->inp_flags & INP_DROPPED))
1965 			tp->t_fb->tfb_tcp_output(tp);
1966 	}
1967 }
1968 
1969 /*
1970  * User issued close, and wish to trail through shutdown states:
1971  * if never received SYN, just forget it.  If got a SYN from peer,
1972  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1973  * If already got a FIN from peer, then almost done; go to LAST_ACK
1974  * state.  In all other cases, have already sent FIN to peer (e.g.
1975  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1976  * for peer to send FIN or not respond to keep-alives, etc.
1977  * We can let the user exit from the close as soon as the FIN is acked.
1978  */
1979 static void
1980 tcp_usrclosed(struct tcpcb *tp)
1981 {
1982 
1983 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1984 	INP_WLOCK_ASSERT(tp->t_inpcb);
1985 
1986 	switch (tp->t_state) {
1987 	case TCPS_LISTEN:
1988 #ifdef TCP_OFFLOAD
1989 		tcp_offload_listen_stop(tp);
1990 #endif
1991 		tcp_state_change(tp, TCPS_CLOSED);
1992 		/* FALLTHROUGH */
1993 	case TCPS_CLOSED:
1994 		tp = tcp_close(tp);
1995 		/*
1996 		 * tcp_close() should never return NULL here as the socket is
1997 		 * still open.
1998 		 */
1999 		KASSERT(tp != NULL,
2000 		    ("tcp_usrclosed: tcp_close() returned NULL"));
2001 		break;
2002 
2003 	case TCPS_SYN_SENT:
2004 	case TCPS_SYN_RECEIVED:
2005 		tp->t_flags |= TF_NEEDFIN;
2006 		break;
2007 
2008 	case TCPS_ESTABLISHED:
2009 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
2010 		break;
2011 
2012 	case TCPS_CLOSE_WAIT:
2013 		tcp_state_change(tp, TCPS_LAST_ACK);
2014 		break;
2015 	}
2016 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
2017 		soisdisconnected(tp->t_inpcb->inp_socket);
2018 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
2019 		if (tp->t_state == TCPS_FIN_WAIT_2) {
2020 			int timeout;
2021 
2022 			timeout = (tcp_fast_finwait2_recycle) ?
2023 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
2024 			tcp_timer_activate(tp, TT_2MSL, timeout);
2025 		}
2026 	}
2027 }
2028 
2029 #ifdef DDB
2030 static void
2031 db_print_indent(int indent)
2032 {
2033 	int i;
2034 
2035 	for (i = 0; i < indent; i++)
2036 		db_printf(" ");
2037 }
2038 
2039 static void
2040 db_print_tstate(int t_state)
2041 {
2042 
2043 	switch (t_state) {
2044 	case TCPS_CLOSED:
2045 		db_printf("TCPS_CLOSED");
2046 		return;
2047 
2048 	case TCPS_LISTEN:
2049 		db_printf("TCPS_LISTEN");
2050 		return;
2051 
2052 	case TCPS_SYN_SENT:
2053 		db_printf("TCPS_SYN_SENT");
2054 		return;
2055 
2056 	case TCPS_SYN_RECEIVED:
2057 		db_printf("TCPS_SYN_RECEIVED");
2058 		return;
2059 
2060 	case TCPS_ESTABLISHED:
2061 		db_printf("TCPS_ESTABLISHED");
2062 		return;
2063 
2064 	case TCPS_CLOSE_WAIT:
2065 		db_printf("TCPS_CLOSE_WAIT");
2066 		return;
2067 
2068 	case TCPS_FIN_WAIT_1:
2069 		db_printf("TCPS_FIN_WAIT_1");
2070 		return;
2071 
2072 	case TCPS_CLOSING:
2073 		db_printf("TCPS_CLOSING");
2074 		return;
2075 
2076 	case TCPS_LAST_ACK:
2077 		db_printf("TCPS_LAST_ACK");
2078 		return;
2079 
2080 	case TCPS_FIN_WAIT_2:
2081 		db_printf("TCPS_FIN_WAIT_2");
2082 		return;
2083 
2084 	case TCPS_TIME_WAIT:
2085 		db_printf("TCPS_TIME_WAIT");
2086 		return;
2087 
2088 	default:
2089 		db_printf("unknown");
2090 		return;
2091 	}
2092 }
2093 
2094 static void
2095 db_print_tflags(u_int t_flags)
2096 {
2097 	int comma;
2098 
2099 	comma = 0;
2100 	if (t_flags & TF_ACKNOW) {
2101 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
2102 		comma = 1;
2103 	}
2104 	if (t_flags & TF_DELACK) {
2105 		db_printf("%sTF_DELACK", comma ? ", " : "");
2106 		comma = 1;
2107 	}
2108 	if (t_flags & TF_NODELAY) {
2109 		db_printf("%sTF_NODELAY", comma ? ", " : "");
2110 		comma = 1;
2111 	}
2112 	if (t_flags & TF_NOOPT) {
2113 		db_printf("%sTF_NOOPT", comma ? ", " : "");
2114 		comma = 1;
2115 	}
2116 	if (t_flags & TF_SENTFIN) {
2117 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2118 		comma = 1;
2119 	}
2120 	if (t_flags & TF_REQ_SCALE) {
2121 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2122 		comma = 1;
2123 	}
2124 	if (t_flags & TF_RCVD_SCALE) {
2125 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2126 		comma = 1;
2127 	}
2128 	if (t_flags & TF_REQ_TSTMP) {
2129 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2130 		comma = 1;
2131 	}
2132 	if (t_flags & TF_RCVD_TSTMP) {
2133 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2134 		comma = 1;
2135 	}
2136 	if (t_flags & TF_SACK_PERMIT) {
2137 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2138 		comma = 1;
2139 	}
2140 	if (t_flags & TF_NEEDSYN) {
2141 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2142 		comma = 1;
2143 	}
2144 	if (t_flags & TF_NEEDFIN) {
2145 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2146 		comma = 1;
2147 	}
2148 	if (t_flags & TF_NOPUSH) {
2149 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2150 		comma = 1;
2151 	}
2152 	if (t_flags & TF_MORETOCOME) {
2153 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2154 		comma = 1;
2155 	}
2156 	if (t_flags & TF_LQ_OVERFLOW) {
2157 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2158 		comma = 1;
2159 	}
2160 	if (t_flags & TF_LASTIDLE) {
2161 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2162 		comma = 1;
2163 	}
2164 	if (t_flags & TF_RXWIN0SENT) {
2165 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2166 		comma = 1;
2167 	}
2168 	if (t_flags & TF_FASTRECOVERY) {
2169 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2170 		comma = 1;
2171 	}
2172 	if (t_flags & TF_CONGRECOVERY) {
2173 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2174 		comma = 1;
2175 	}
2176 	if (t_flags & TF_WASFRECOVERY) {
2177 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2178 		comma = 1;
2179 	}
2180 	if (t_flags & TF_SIGNATURE) {
2181 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2182 		comma = 1;
2183 	}
2184 	if (t_flags & TF_FORCEDATA) {
2185 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2186 		comma = 1;
2187 	}
2188 	if (t_flags & TF_TSO) {
2189 		db_printf("%sTF_TSO", comma ? ", " : "");
2190 		comma = 1;
2191 	}
2192 	if (t_flags & TF_ECN_PERMIT) {
2193 		db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2194 		comma = 1;
2195 	}
2196 	if (t_flags & TF_FASTOPEN) {
2197 		db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2198 		comma = 1;
2199 	}
2200 }
2201 
2202 static void
2203 db_print_toobflags(char t_oobflags)
2204 {
2205 	int comma;
2206 
2207 	comma = 0;
2208 	if (t_oobflags & TCPOOB_HAVEDATA) {
2209 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2210 		comma = 1;
2211 	}
2212 	if (t_oobflags & TCPOOB_HADDATA) {
2213 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2214 		comma = 1;
2215 	}
2216 }
2217 
2218 static void
2219 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2220 {
2221 
2222 	db_print_indent(indent);
2223 	db_printf("%s at %p\n", name, tp);
2224 
2225 	indent += 2;
2226 
2227 	db_print_indent(indent);
2228 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2229 	   LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2230 
2231 	db_print_indent(indent);
2232 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2233 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2234 
2235 	db_print_indent(indent);
2236 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2237 	    &tp->t_timers->tt_delack, tp->t_inpcb);
2238 
2239 	db_print_indent(indent);
2240 	db_printf("t_state: %d (", tp->t_state);
2241 	db_print_tstate(tp->t_state);
2242 	db_printf(")\n");
2243 
2244 	db_print_indent(indent);
2245 	db_printf("t_flags: 0x%x (", tp->t_flags);
2246 	db_print_tflags(tp->t_flags);
2247 	db_printf(")\n");
2248 
2249 	db_print_indent(indent);
2250 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2251 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2252 
2253 	db_print_indent(indent);
2254 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2255 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2256 
2257 	db_print_indent(indent);
2258 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2259 	    tp->iss, tp->irs, tp->rcv_nxt);
2260 
2261 	db_print_indent(indent);
2262 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2263 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2264 
2265 	db_print_indent(indent);
2266 	db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2267 	   tp->snd_wnd, tp->snd_cwnd);
2268 
2269 	db_print_indent(indent);
2270 	db_printf("snd_ssthresh: %u   snd_recover: "
2271 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2272 
2273 	db_print_indent(indent);
2274 	db_printf("t_rcvtime: %u   t_startime: %u\n",
2275 	    tp->t_rcvtime, tp->t_starttime);
2276 
2277 	db_print_indent(indent);
2278 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2279 	    tp->t_rtttime, tp->t_rtseq);
2280 
2281 	db_print_indent(indent);
2282 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2283 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2284 
2285 	db_print_indent(indent);
2286 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2287 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2288 	    tp->t_rttbest);
2289 
2290 	db_print_indent(indent);
2291 	db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2292 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2293 
2294 	db_print_indent(indent);
2295 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2296 	db_print_toobflags(tp->t_oobflags);
2297 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2298 
2299 	db_print_indent(indent);
2300 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2301 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2302 
2303 	db_print_indent(indent);
2304 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2305 	    tp->ts_recent, tp->ts_recent_age);
2306 
2307 	db_print_indent(indent);
2308 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2309 	    "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2310 
2311 	db_print_indent(indent);
2312 	db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2313 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2314 	    tp->snd_recover_prev, tp->t_badrxtwin);
2315 
2316 	db_print_indent(indent);
2317 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2318 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2319 
2320 	db_print_indent(indent);
2321 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2322 	    "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2323 
2324 	/* Skip sackblks, sackhint. */
2325 
2326 	db_print_indent(indent);
2327 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2328 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2329 }
2330 
2331 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2332 {
2333 	struct tcpcb *tp;
2334 
2335 	if (!have_addr) {
2336 		db_printf("usage: show tcpcb <addr>\n");
2337 		return;
2338 	}
2339 	tp = (struct tcpcb *)addr;
2340 
2341 	db_print_tcpcb(tp, "tcpcb", 0);
2342 }
2343 #endif
2344