xref: /freebsd/sys/netinet/tcp_usrreq.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
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 
63 #ifdef DDB
64 #include <ddb/ddb.h>
65 #endif
66 
67 #include <net/if.h>
68 #include <net/if_var.h>
69 #include <net/route.h>
70 #include <net/vnet.h>
71 
72 #include <netinet/in.h>
73 #include <netinet/in_kdtrace.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/in_var.h>
77 #include <netinet/ip_var.h>
78 #ifdef INET6
79 #include <netinet/ip6.h>
80 #include <netinet6/in6_pcb.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #endif
84 #ifdef TCP_RFC7413
85 #include <netinet/tcp_fastopen.h>
86 #endif
87 #include <netinet/tcp.h>
88 #include <netinet/tcp_fsm.h>
89 #include <netinet/tcp_seq.h>
90 #include <netinet/tcp_timer.h>
91 #include <netinet/tcp_var.h>
92 #include <netinet/tcpip.h>
93 #include <netinet/cc/cc.h>
94 #ifdef TCPPCAP
95 #include <netinet/tcp_pcap.h>
96 #endif
97 #ifdef TCPDEBUG
98 #include <netinet/tcp_debug.h>
99 #endif
100 #ifdef TCP_OFFLOAD
101 #include <netinet/tcp_offload.h>
102 #endif
103 
104 /*
105  * TCP protocol interface to socket abstraction.
106  */
107 static int	tcp_attach(struct socket *);
108 #ifdef INET
109 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
110 		    struct thread *td);
111 #endif /* INET */
112 #ifdef INET6
113 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
114 		    struct thread *td);
115 #endif /* INET6 */
116 static void	tcp_disconnect(struct tcpcb *);
117 static void	tcp_usrclosed(struct tcpcb *);
118 static void	tcp_fill_info(struct tcpcb *, struct tcp_info *);
119 
120 #ifdef TCPDEBUG
121 #define	TCPDEBUG0	int ostate = 0
122 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
123 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
124 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
125 #else
126 #define	TCPDEBUG0
127 #define	TCPDEBUG1()
128 #define	TCPDEBUG2(req)
129 #endif
130 
131 /*
132  * TCP attaches to socket via pru_attach(), reserving space,
133  * and an internet control block.
134  */
135 static int
136 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
137 {
138 	struct inpcb *inp;
139 	struct tcpcb *tp = NULL;
140 	int error;
141 	TCPDEBUG0;
142 
143 	inp = sotoinpcb(so);
144 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
145 	TCPDEBUG1();
146 
147 	error = tcp_attach(so);
148 	if (error)
149 		goto out;
150 
151 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
152 		so->so_linger = TCP_LINGERTIME;
153 
154 	inp = sotoinpcb(so);
155 	tp = intotcpcb(inp);
156 out:
157 	TCPDEBUG2(PRU_ATTACH);
158 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
159 	return error;
160 }
161 
162 /*
163  * tcp_detach is called when the socket layer loses its final reference
164  * to the socket, be it a file descriptor reference, a reference from TCP,
165  * etc.  At this point, there is only one case in which we will keep around
166  * inpcb state: time wait.
167  *
168  * This function can probably be re-absorbed back into tcp_usr_detach() now
169  * that there is a single detach path.
170  */
171 static void
172 tcp_detach(struct socket *so, struct inpcb *inp)
173 {
174 	struct tcpcb *tp;
175 
176 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
177 	INP_WLOCK_ASSERT(inp);
178 
179 	KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
180 	KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
181 
182 	tp = intotcpcb(inp);
183 
184 	if (inp->inp_flags & INP_TIMEWAIT) {
185 		/*
186 		 * There are two cases to handle: one in which the time wait
187 		 * state is being discarded (INP_DROPPED), and one in which
188 		 * this connection will remain in timewait.  In the former,
189 		 * it is time to discard all state (except tcptw, which has
190 		 * already been discarded by the timewait close code, which
191 		 * should be further up the call stack somewhere).  In the
192 		 * latter case, we detach from the socket, but leave the pcb
193 		 * present until timewait ends.
194 		 *
195 		 * XXXRW: Would it be cleaner to free the tcptw here?
196 		 *
197 		 * Astute question indeed, from twtcp perspective there are
198 		 * three cases to consider:
199 		 *
200 		 * #1 tcp_detach is called at tcptw creation time by
201 		 *  tcp_twstart, then do not discard the newly created tcptw
202 		 *  and leave inpcb present until timewait ends
203 		 * #2 tcp_detach is called at timewait end (or reuse) by
204 		 *  tcp_twclose, then the tcptw has already been discarded
205 		 *  (or reused) and inpcb is freed here
206 		 * #3 tcp_detach is called() after timewait ends (or reuse)
207 		 *  (e.g. by soclose), then tcptw has already been discarded
208 		 *  (or reused) and inpcb is freed here
209 		 *
210 		 *  In all three cases the tcptw should not be freed here.
211 		 */
212 		if (inp->inp_flags & INP_DROPPED) {
213 			KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
214 			    "INP_DROPPED && tp != NULL"));
215 			in_pcbdetach(inp);
216 			in_pcbfree(inp);
217 		} else {
218 			in_pcbdetach(inp);
219 			INP_WUNLOCK(inp);
220 		}
221 	} else {
222 		/*
223 		 * If the connection is not in timewait, we consider two
224 		 * two conditions: one in which no further processing is
225 		 * necessary (dropped || embryonic), and one in which TCP is
226 		 * not yet done, but no longer requires the socket, so the
227 		 * pcb will persist for the time being.
228 		 *
229 		 * XXXRW: Does the second case still occur?
230 		 */
231 		if (inp->inp_flags & INP_DROPPED ||
232 		    tp->t_state < TCPS_SYN_SENT) {
233 			tcp_discardcb(tp);
234 			in_pcbdetach(inp);
235 			in_pcbfree(inp);
236 		} else {
237 			in_pcbdetach(inp);
238 			INP_WUNLOCK(inp);
239 		}
240 	}
241 }
242 
243 /*
244  * pru_detach() detaches the TCP protocol from the socket.
245  * If the protocol state is non-embryonic, then can't
246  * do this directly: have to initiate a pru_disconnect(),
247  * which may finish later; embryonic TCB's can just
248  * be discarded here.
249  */
250 static void
251 tcp_usr_detach(struct socket *so)
252 {
253 	struct inpcb *inp;
254 	int rlock = 0;
255 
256 	inp = sotoinpcb(so);
257 	KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
258 	if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
259 		INP_INFO_RLOCK(&V_tcbinfo);
260 		rlock = 1;
261 	}
262 	INP_WLOCK(inp);
263 	KASSERT(inp->inp_socket != NULL,
264 	    ("tcp_usr_detach: inp_socket == NULL"));
265 	tcp_detach(so, inp);
266 	if (rlock)
267 		INP_INFO_RUNLOCK(&V_tcbinfo);
268 }
269 
270 #ifdef INET
271 /*
272  * Give the socket an address.
273  */
274 static int
275 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
276 {
277 	int error = 0;
278 	struct inpcb *inp;
279 	struct tcpcb *tp = NULL;
280 	struct sockaddr_in *sinp;
281 
282 	sinp = (struct sockaddr_in *)nam;
283 	if (nam->sa_len != sizeof (*sinp))
284 		return (EINVAL);
285 	/*
286 	 * Must check for multicast addresses and disallow binding
287 	 * to them.
288 	 */
289 	if (sinp->sin_family == AF_INET &&
290 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
291 		return (EAFNOSUPPORT);
292 
293 	TCPDEBUG0;
294 	inp = sotoinpcb(so);
295 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
296 	INP_WLOCK(inp);
297 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
298 		error = EINVAL;
299 		goto out;
300 	}
301 	tp = intotcpcb(inp);
302 	TCPDEBUG1();
303 	INP_HASH_WLOCK(&V_tcbinfo);
304 	error = in_pcbbind(inp, nam, td->td_ucred);
305 	INP_HASH_WUNLOCK(&V_tcbinfo);
306 out:
307 	TCPDEBUG2(PRU_BIND);
308 	TCP_PROBE2(debug__user, tp, PRU_BIND);
309 	INP_WUNLOCK(inp);
310 
311 	return (error);
312 }
313 #endif /* INET */
314 
315 #ifdef INET6
316 static int
317 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
318 {
319 	int error = 0;
320 	struct inpcb *inp;
321 	struct tcpcb *tp = NULL;
322 	struct sockaddr_in6 *sin6p;
323 
324 	sin6p = (struct sockaddr_in6 *)nam;
325 	if (nam->sa_len != sizeof (*sin6p))
326 		return (EINVAL);
327 	/*
328 	 * Must check for multicast addresses and disallow binding
329 	 * to them.
330 	 */
331 	if (sin6p->sin6_family == AF_INET6 &&
332 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
333 		return (EAFNOSUPPORT);
334 
335 	TCPDEBUG0;
336 	inp = sotoinpcb(so);
337 	KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
338 	INP_WLOCK(inp);
339 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
340 		error = EINVAL;
341 		goto out;
342 	}
343 	tp = intotcpcb(inp);
344 	TCPDEBUG1();
345 	INP_HASH_WLOCK(&V_tcbinfo);
346 	inp->inp_vflag &= ~INP_IPV4;
347 	inp->inp_vflag |= INP_IPV6;
348 #ifdef INET
349 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
350 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
351 			inp->inp_vflag |= INP_IPV4;
352 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
353 			struct sockaddr_in sin;
354 
355 			in6_sin6_2_sin(&sin, sin6p);
356 			inp->inp_vflag |= INP_IPV4;
357 			inp->inp_vflag &= ~INP_IPV6;
358 			error = in_pcbbind(inp, (struct sockaddr *)&sin,
359 			    td->td_ucred);
360 			INP_HASH_WUNLOCK(&V_tcbinfo);
361 			goto out;
362 		}
363 	}
364 #endif
365 	error = in6_pcbbind(inp, nam, td->td_ucred);
366 	INP_HASH_WUNLOCK(&V_tcbinfo);
367 out:
368 	TCPDEBUG2(PRU_BIND);
369 	TCP_PROBE2(debug__user, tp, PRU_BIND);
370 	INP_WUNLOCK(inp);
371 	return (error);
372 }
373 #endif /* INET6 */
374 
375 #ifdef INET
376 /*
377  * Prepare to accept connections.
378  */
379 static int
380 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
381 {
382 	int error = 0;
383 	struct inpcb *inp;
384 	struct tcpcb *tp = NULL;
385 
386 	TCPDEBUG0;
387 	inp = sotoinpcb(so);
388 	KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
389 	INP_WLOCK(inp);
390 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
391 		error = EINVAL;
392 		goto out;
393 	}
394 	tp = intotcpcb(inp);
395 	TCPDEBUG1();
396 	SOCK_LOCK(so);
397 	error = solisten_proto_check(so);
398 	INP_HASH_WLOCK(&V_tcbinfo);
399 	if (error == 0 && inp->inp_lport == 0)
400 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
401 	INP_HASH_WUNLOCK(&V_tcbinfo);
402 	if (error == 0) {
403 		tcp_state_change(tp, TCPS_LISTEN);
404 		solisten_proto(so, backlog);
405 #ifdef TCP_OFFLOAD
406 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
407 			tcp_offload_listen_start(tp);
408 #endif
409 	}
410 	SOCK_UNLOCK(so);
411 
412 #ifdef TCP_RFC7413
413 	if (tp->t_flags & TF_FASTOPEN)
414 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
415 #endif
416 out:
417 	TCPDEBUG2(PRU_LISTEN);
418 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
419 	INP_WUNLOCK(inp);
420 	return (error);
421 }
422 #endif /* INET */
423 
424 #ifdef INET6
425 static int
426 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
427 {
428 	int error = 0;
429 	struct inpcb *inp;
430 	struct tcpcb *tp = NULL;
431 
432 	TCPDEBUG0;
433 	inp = sotoinpcb(so);
434 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
435 	INP_WLOCK(inp);
436 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
437 		error = EINVAL;
438 		goto out;
439 	}
440 	tp = intotcpcb(inp);
441 	TCPDEBUG1();
442 	SOCK_LOCK(so);
443 	error = solisten_proto_check(so);
444 	INP_HASH_WLOCK(&V_tcbinfo);
445 	if (error == 0 && inp->inp_lport == 0) {
446 		inp->inp_vflag &= ~INP_IPV4;
447 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
448 			inp->inp_vflag |= INP_IPV4;
449 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
450 	}
451 	INP_HASH_WUNLOCK(&V_tcbinfo);
452 	if (error == 0) {
453 		tcp_state_change(tp, TCPS_LISTEN);
454 		solisten_proto(so, backlog);
455 #ifdef TCP_OFFLOAD
456 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
457 			tcp_offload_listen_start(tp);
458 #endif
459 	}
460 	SOCK_UNLOCK(so);
461 
462 #ifdef TCP_RFC7413
463 	if (tp->t_flags & TF_FASTOPEN)
464 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
465 #endif
466 out:
467 	TCPDEBUG2(PRU_LISTEN);
468 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
469 	INP_WUNLOCK(inp);
470 	return (error);
471 }
472 #endif /* INET6 */
473 
474 #ifdef INET
475 /*
476  * Initiate connection to peer.
477  * Create a template for use in transmissions on this connection.
478  * Enter SYN_SENT state, and mark socket as connecting.
479  * Start keep-alive timer, and seed output sequence space.
480  * Send initial segment on connection.
481  */
482 static int
483 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
484 {
485 	int error = 0;
486 	struct inpcb *inp;
487 	struct tcpcb *tp = NULL;
488 	struct sockaddr_in *sinp;
489 
490 	sinp = (struct sockaddr_in *)nam;
491 	if (nam->sa_len != sizeof (*sinp))
492 		return (EINVAL);
493 	/*
494 	 * Must disallow TCP ``connections'' to multicast addresses.
495 	 */
496 	if (sinp->sin_family == AF_INET
497 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
498 		return (EAFNOSUPPORT);
499 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
500 		return (error);
501 
502 	TCPDEBUG0;
503 	inp = sotoinpcb(so);
504 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
505 	INP_WLOCK(inp);
506 	if (inp->inp_flags & INP_TIMEWAIT) {
507 		error = EADDRINUSE;
508 		goto out;
509 	}
510 	if (inp->inp_flags & INP_DROPPED) {
511 		error = ECONNREFUSED;
512 		goto out;
513 	}
514 	tp = intotcpcb(inp);
515 	TCPDEBUG1();
516 	if ((error = tcp_connect(tp, nam, td)) != 0)
517 		goto out;
518 #ifdef TCP_OFFLOAD
519 	if (registered_toedevs > 0 &&
520 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
521 	    (error = tcp_offload_connect(so, nam)) == 0)
522 		goto out;
523 #endif
524 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
525 	error = tp->t_fb->tfb_tcp_output(tp);
526 out:
527 	TCPDEBUG2(PRU_CONNECT);
528 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
529 	INP_WUNLOCK(inp);
530 	return (error);
531 }
532 #endif /* INET */
533 
534 #ifdef INET6
535 static int
536 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
537 {
538 	int error = 0;
539 	struct inpcb *inp;
540 	struct tcpcb *tp = NULL;
541 	struct sockaddr_in6 *sin6p;
542 
543 	TCPDEBUG0;
544 
545 	sin6p = (struct sockaddr_in6 *)nam;
546 	if (nam->sa_len != sizeof (*sin6p))
547 		return (EINVAL);
548 	/*
549 	 * Must disallow TCP ``connections'' to multicast addresses.
550 	 */
551 	if (sin6p->sin6_family == AF_INET6
552 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
553 		return (EAFNOSUPPORT);
554 
555 	inp = sotoinpcb(so);
556 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
557 	INP_WLOCK(inp);
558 	if (inp->inp_flags & INP_TIMEWAIT) {
559 		error = EADDRINUSE;
560 		goto out;
561 	}
562 	if (inp->inp_flags & INP_DROPPED) {
563 		error = ECONNREFUSED;
564 		goto out;
565 	}
566 	tp = intotcpcb(inp);
567 	TCPDEBUG1();
568 #ifdef INET
569 	/*
570 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
571 	 * therefore probably require the hash lock, which isn't held here.
572 	 * Is this a significant problem?
573 	 */
574 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
575 		struct sockaddr_in sin;
576 
577 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
578 			error = EINVAL;
579 			goto out;
580 		}
581 
582 		in6_sin6_2_sin(&sin, sin6p);
583 		inp->inp_vflag |= INP_IPV4;
584 		inp->inp_vflag &= ~INP_IPV6;
585 		if ((error = prison_remote_ip4(td->td_ucred,
586 		    &sin.sin_addr)) != 0)
587 			goto out;
588 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
589 			goto out;
590 #ifdef TCP_OFFLOAD
591 		if (registered_toedevs > 0 &&
592 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
593 		    (error = tcp_offload_connect(so, nam)) == 0)
594 			goto out;
595 #endif
596 		error = tp->t_fb->tfb_tcp_output(tp);
597 		goto out;
598 	}
599 #endif
600 	inp->inp_vflag &= ~INP_IPV4;
601 	inp->inp_vflag |= INP_IPV6;
602 	inp->inp_inc.inc_flags |= INC_ISIPV6;
603 	if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
604 		goto out;
605 	if ((error = tcp6_connect(tp, nam, 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 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
614 	error = tp->t_fb->tfb_tcp_output(tp);
615 
616 out:
617 	TCPDEBUG2(PRU_CONNECT);
618 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
619 	INP_WUNLOCK(inp);
620 	return (error);
621 }
622 #endif /* INET6 */
623 
624 /*
625  * Initiate disconnect from peer.
626  * If connection never passed embryonic stage, just drop;
627  * else if don't need to let data drain, then can just drop anyways,
628  * else have to begin TCP shutdown process: mark socket disconnecting,
629  * drain unread data, state switch to reflect user close, and
630  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
631  * when peer sends FIN and acks ours.
632  *
633  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
634  */
635 static int
636 tcp_usr_disconnect(struct socket *so)
637 {
638 	struct inpcb *inp;
639 	struct tcpcb *tp = NULL;
640 	int error = 0;
641 
642 	TCPDEBUG0;
643 	INP_INFO_RLOCK(&V_tcbinfo);
644 	inp = sotoinpcb(so);
645 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
646 	INP_WLOCK(inp);
647 	if (inp->inp_flags & INP_TIMEWAIT)
648 		goto out;
649 	if (inp->inp_flags & INP_DROPPED) {
650 		error = ECONNRESET;
651 		goto out;
652 	}
653 	tp = intotcpcb(inp);
654 	TCPDEBUG1();
655 	tcp_disconnect(tp);
656 out:
657 	TCPDEBUG2(PRU_DISCONNECT);
658 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
659 	INP_WUNLOCK(inp);
660 	INP_INFO_RUNLOCK(&V_tcbinfo);
661 	return (error);
662 }
663 
664 #ifdef INET
665 /*
666  * Accept a connection.  Essentially all the work is done at higher levels;
667  * just return the address of the peer, storing through addr.
668  */
669 static int
670 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
671 {
672 	int error = 0;
673 	struct inpcb *inp = NULL;
674 	struct tcpcb *tp = NULL;
675 	struct in_addr addr;
676 	in_port_t port = 0;
677 	TCPDEBUG0;
678 
679 	if (so->so_state & SS_ISDISCONNECTED)
680 		return (ECONNABORTED);
681 
682 	inp = sotoinpcb(so);
683 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
684 	INP_WLOCK(inp);
685 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
686 		error = ECONNABORTED;
687 		goto out;
688 	}
689 	tp = intotcpcb(inp);
690 	TCPDEBUG1();
691 
692 	/*
693 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
694 	 * copy the data of interest and defer the malloc until after we
695 	 * release the lock.
696 	 */
697 	port = inp->inp_fport;
698 	addr = inp->inp_faddr;
699 
700 out:
701 	TCPDEBUG2(PRU_ACCEPT);
702 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
703 	INP_WUNLOCK(inp);
704 	if (error == 0)
705 		*nam = in_sockaddr(port, &addr);
706 	return error;
707 }
708 #endif /* INET */
709 
710 #ifdef INET6
711 static int
712 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
713 {
714 	struct inpcb *inp = NULL;
715 	int error = 0;
716 	struct tcpcb *tp = NULL;
717 	struct in_addr addr;
718 	struct in6_addr addr6;
719 	in_port_t port = 0;
720 	int v4 = 0;
721 	TCPDEBUG0;
722 
723 	if (so->so_state & SS_ISDISCONNECTED)
724 		return (ECONNABORTED);
725 
726 	inp = sotoinpcb(so);
727 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
728 	INP_INFO_RLOCK(&V_tcbinfo);
729 	INP_WLOCK(inp);
730 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
731 		error = ECONNABORTED;
732 		goto out;
733 	}
734 	tp = intotcpcb(inp);
735 	TCPDEBUG1();
736 
737 	/*
738 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
739 	 * copy the data of interest and defer the malloc until after we
740 	 * release the lock.
741 	 */
742 	if (inp->inp_vflag & INP_IPV4) {
743 		v4 = 1;
744 		port = inp->inp_fport;
745 		addr = inp->inp_faddr;
746 	} else {
747 		port = inp->inp_fport;
748 		addr6 = inp->in6p_faddr;
749 	}
750 
751 out:
752 	TCPDEBUG2(PRU_ACCEPT);
753 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
754 	INP_WUNLOCK(inp);
755 	INP_INFO_RUNLOCK(&V_tcbinfo);
756 	if (error == 0) {
757 		if (v4)
758 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
759 		else
760 			*nam = in6_sockaddr(port, &addr6);
761 	}
762 	return error;
763 }
764 #endif /* INET6 */
765 
766 /*
767  * Mark the connection as being incapable of further output.
768  */
769 static int
770 tcp_usr_shutdown(struct socket *so)
771 {
772 	int error = 0;
773 	struct inpcb *inp;
774 	struct tcpcb *tp = NULL;
775 
776 	TCPDEBUG0;
777 	INP_INFO_RLOCK(&V_tcbinfo);
778 	inp = sotoinpcb(so);
779 	KASSERT(inp != NULL, ("inp == NULL"));
780 	INP_WLOCK(inp);
781 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
782 		error = ECONNRESET;
783 		goto out;
784 	}
785 	tp = intotcpcb(inp);
786 	TCPDEBUG1();
787 	socantsendmore(so);
788 	tcp_usrclosed(tp);
789 	if (!(inp->inp_flags & INP_DROPPED))
790 		error = tp->t_fb->tfb_tcp_output(tp);
791 
792 out:
793 	TCPDEBUG2(PRU_SHUTDOWN);
794 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
795 	INP_WUNLOCK(inp);
796 	INP_INFO_RUNLOCK(&V_tcbinfo);
797 
798 	return (error);
799 }
800 
801 /*
802  * After a receive, possibly send window update to peer.
803  */
804 static int
805 tcp_usr_rcvd(struct socket *so, int flags)
806 {
807 	struct inpcb *inp;
808 	struct tcpcb *tp = NULL;
809 	int error = 0;
810 
811 	TCPDEBUG0;
812 	inp = sotoinpcb(so);
813 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
814 	INP_WLOCK(inp);
815 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
816 		error = ECONNRESET;
817 		goto out;
818 	}
819 	tp = intotcpcb(inp);
820 	TCPDEBUG1();
821 #ifdef TCP_RFC7413
822 	/*
823 	 * For passively-created TFO connections, don't attempt a window
824 	 * update while still in SYN_RECEIVED as this may trigger an early
825 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
826 	 * application response data, or failing that, when the DELACK timer
827 	 * expires.
828 	 */
829 	if ((tp->t_flags & TF_FASTOPEN) &&
830 	    (tp->t_state == TCPS_SYN_RECEIVED))
831 		goto out;
832 #endif
833 #ifdef TCP_OFFLOAD
834 	if (tp->t_flags & TF_TOE)
835 		tcp_offload_rcvd(tp);
836 	else
837 #endif
838 	tp->t_fb->tfb_tcp_output(tp);
839 
840 out:
841 	TCPDEBUG2(PRU_RCVD);
842 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
843 	INP_WUNLOCK(inp);
844 	return (error);
845 }
846 
847 /*
848  * Do a send by putting data in output queue and updating urgent
849  * marker if URG set.  Possibly send more data.  Unlike the other
850  * pru_*() routines, the mbuf chains are our responsibility.  We
851  * must either enqueue them or free them.  The other pru_* routines
852  * generally are caller-frees.
853  */
854 static int
855 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
856     struct sockaddr *nam, struct mbuf *control, struct thread *td)
857 {
858 	int error = 0;
859 	struct inpcb *inp;
860 	struct tcpcb *tp = NULL;
861 #ifdef INET6
862 	int isipv6;
863 #endif
864 	TCPDEBUG0;
865 
866 	/*
867 	 * We require the pcbinfo lock if we will close the socket as part of
868 	 * this call.
869 	 */
870 	if (flags & PRUS_EOF)
871 		INP_INFO_RLOCK(&V_tcbinfo);
872 	inp = sotoinpcb(so);
873 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
874 	INP_WLOCK(inp);
875 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
876 		if (control)
877 			m_freem(control);
878 		/*
879 		 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
880 		 * for freeing memory.
881 		 */
882 		if (m && (flags & PRUS_NOTREADY) == 0)
883 			m_freem(m);
884 		error = ECONNRESET;
885 		goto out;
886 	}
887 #ifdef INET6
888 	isipv6 = nam && nam->sa_family == AF_INET6;
889 #endif /* INET6 */
890 	tp = intotcpcb(inp);
891 	TCPDEBUG1();
892 	if (control) {
893 		/* TCP doesn't do control messages (rights, creds, etc) */
894 		if (control->m_len) {
895 			m_freem(control);
896 			if (m)
897 				m_freem(m);
898 			error = EINVAL;
899 			goto out;
900 		}
901 		m_freem(control);	/* empty control, just free it */
902 	}
903 	if (!(flags & PRUS_OOB)) {
904 		sbappendstream(&so->so_snd, m, flags);
905 		if (nam && tp->t_state < TCPS_SYN_SENT) {
906 			/*
907 			 * Do implied connect if not yet connected,
908 			 * initialize window to default value, and
909 			 * initialize maxseg using peer's cached MSS.
910 			 */
911 #ifdef INET6
912 			if (isipv6)
913 				error = tcp6_connect(tp, nam, td);
914 #endif /* INET6 */
915 #if defined(INET6) && defined(INET)
916 			else
917 #endif
918 #ifdef INET
919 				error = tcp_connect(tp, nam, td);
920 #endif
921 			if (error)
922 				goto out;
923 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
924 			tcp_mss(tp, -1);
925 		}
926 		if (flags & PRUS_EOF) {
927 			/*
928 			 * Close the send side of the connection after
929 			 * the data is sent.
930 			 */
931 			INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
932 			socantsendmore(so);
933 			tcp_usrclosed(tp);
934 		}
935 		if (!(inp->inp_flags & INP_DROPPED) &&
936 		    !(flags & PRUS_NOTREADY)) {
937 			if (flags & PRUS_MORETOCOME)
938 				tp->t_flags |= TF_MORETOCOME;
939 			error = tp->t_fb->tfb_tcp_output(tp);
940 			if (flags & PRUS_MORETOCOME)
941 				tp->t_flags &= ~TF_MORETOCOME;
942 		}
943 	} else {
944 		/*
945 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
946 		 */
947 		SOCKBUF_LOCK(&so->so_snd);
948 		if (sbspace(&so->so_snd) < -512) {
949 			SOCKBUF_UNLOCK(&so->so_snd);
950 			m_freem(m);
951 			error = ENOBUFS;
952 			goto out;
953 		}
954 		/*
955 		 * According to RFC961 (Assigned Protocols),
956 		 * the urgent pointer points to the last octet
957 		 * of urgent data.  We continue, however,
958 		 * to consider it to indicate the first octet
959 		 * of data past the urgent section.
960 		 * Otherwise, snd_up should be one lower.
961 		 */
962 		sbappendstream_locked(&so->so_snd, m, flags);
963 		SOCKBUF_UNLOCK(&so->so_snd);
964 		if (nam && tp->t_state < TCPS_SYN_SENT) {
965 			/*
966 			 * Do implied connect if not yet connected,
967 			 * initialize window to default value, and
968 			 * initialize maxseg using peer's cached MSS.
969 			 */
970 #ifdef INET6
971 			if (isipv6)
972 				error = tcp6_connect(tp, nam, td);
973 #endif /* INET6 */
974 #if defined(INET6) && defined(INET)
975 			else
976 #endif
977 #ifdef INET
978 				error = tcp_connect(tp, nam, td);
979 #endif
980 			if (error)
981 				goto out;
982 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
983 			tcp_mss(tp, -1);
984 		}
985 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
986 		if (!(flags & PRUS_NOTREADY)) {
987 			tp->t_flags |= TF_FORCEDATA;
988 			error = tp->t_fb->tfb_tcp_output(tp);
989 			tp->t_flags &= ~TF_FORCEDATA;
990 		}
991 	}
992 out:
993 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
994 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
995 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
996 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
997 	INP_WUNLOCK(inp);
998 	if (flags & PRUS_EOF)
999 		INP_INFO_RUNLOCK(&V_tcbinfo);
1000 	return (error);
1001 }
1002 
1003 static int
1004 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1005 {
1006 	struct inpcb *inp;
1007 	struct tcpcb *tp;
1008 	int error;
1009 
1010 	inp = sotoinpcb(so);
1011 	INP_WLOCK(inp);
1012 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1013 		INP_WUNLOCK(inp);
1014 		for (int i = 0; i < count; i++)
1015 			m = m_free(m);
1016 		return (ECONNRESET);
1017 	}
1018 	tp = intotcpcb(inp);
1019 
1020 	SOCKBUF_LOCK(&so->so_snd);
1021 	error = sbready(&so->so_snd, m, count);
1022 	SOCKBUF_UNLOCK(&so->so_snd);
1023 	if (error == 0)
1024 		error = tp->t_fb->tfb_tcp_output(tp);
1025 	INP_WUNLOCK(inp);
1026 
1027 	return (error);
1028 }
1029 
1030 /*
1031  * Abort the TCP.  Drop the connection abruptly.
1032  */
1033 static void
1034 tcp_usr_abort(struct socket *so)
1035 {
1036 	struct inpcb *inp;
1037 	struct tcpcb *tp = NULL;
1038 	TCPDEBUG0;
1039 
1040 	inp = sotoinpcb(so);
1041 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1042 
1043 	INP_INFO_RLOCK(&V_tcbinfo);
1044 	INP_WLOCK(inp);
1045 	KASSERT(inp->inp_socket != NULL,
1046 	    ("tcp_usr_abort: inp_socket == NULL"));
1047 
1048 	/*
1049 	 * If we still have full TCP state, and we're not dropped, drop.
1050 	 */
1051 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1052 	    !(inp->inp_flags & INP_DROPPED)) {
1053 		tp = intotcpcb(inp);
1054 		TCPDEBUG1();
1055 		tcp_drop(tp, ECONNABORTED);
1056 		TCPDEBUG2(PRU_ABORT);
1057 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1058 	}
1059 	if (!(inp->inp_flags & INP_DROPPED)) {
1060 		SOCK_LOCK(so);
1061 		so->so_state |= SS_PROTOREF;
1062 		SOCK_UNLOCK(so);
1063 		inp->inp_flags |= INP_SOCKREF;
1064 	}
1065 	INP_WUNLOCK(inp);
1066 	INP_INFO_RUNLOCK(&V_tcbinfo);
1067 }
1068 
1069 /*
1070  * TCP socket is closed.  Start friendly disconnect.
1071  */
1072 static void
1073 tcp_usr_close(struct socket *so)
1074 {
1075 	struct inpcb *inp;
1076 	struct tcpcb *tp = NULL;
1077 	TCPDEBUG0;
1078 
1079 	inp = sotoinpcb(so);
1080 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1081 
1082 	INP_INFO_RLOCK(&V_tcbinfo);
1083 	INP_WLOCK(inp);
1084 	KASSERT(inp->inp_socket != NULL,
1085 	    ("tcp_usr_close: inp_socket == NULL"));
1086 
1087 	/*
1088 	 * If we still have full TCP state, and we're not dropped, initiate
1089 	 * a disconnect.
1090 	 */
1091 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1092 	    !(inp->inp_flags & INP_DROPPED)) {
1093 		tp = intotcpcb(inp);
1094 		TCPDEBUG1();
1095 		tcp_disconnect(tp);
1096 		TCPDEBUG2(PRU_CLOSE);
1097 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1098 	}
1099 	if (!(inp->inp_flags & INP_DROPPED)) {
1100 		SOCK_LOCK(so);
1101 		so->so_state |= SS_PROTOREF;
1102 		SOCK_UNLOCK(so);
1103 		inp->inp_flags |= INP_SOCKREF;
1104 	}
1105 	INP_WUNLOCK(inp);
1106 	INP_INFO_RUNLOCK(&V_tcbinfo);
1107 }
1108 
1109 /*
1110  * Receive out-of-band data.
1111  */
1112 static int
1113 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1114 {
1115 	int error = 0;
1116 	struct inpcb *inp;
1117 	struct tcpcb *tp = NULL;
1118 
1119 	TCPDEBUG0;
1120 	inp = sotoinpcb(so);
1121 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1122 	INP_WLOCK(inp);
1123 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1124 		error = ECONNRESET;
1125 		goto out;
1126 	}
1127 	tp = intotcpcb(inp);
1128 	TCPDEBUG1();
1129 	if ((so->so_oobmark == 0 &&
1130 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1131 	    so->so_options & SO_OOBINLINE ||
1132 	    tp->t_oobflags & TCPOOB_HADDATA) {
1133 		error = EINVAL;
1134 		goto out;
1135 	}
1136 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1137 		error = EWOULDBLOCK;
1138 		goto out;
1139 	}
1140 	m->m_len = 1;
1141 	*mtod(m, caddr_t) = tp->t_iobc;
1142 	if ((flags & MSG_PEEK) == 0)
1143 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1144 
1145 out:
1146 	TCPDEBUG2(PRU_RCVOOB);
1147 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1148 	INP_WUNLOCK(inp);
1149 	return (error);
1150 }
1151 
1152 #ifdef INET
1153 struct pr_usrreqs tcp_usrreqs = {
1154 	.pru_abort =		tcp_usr_abort,
1155 	.pru_accept =		tcp_usr_accept,
1156 	.pru_attach =		tcp_usr_attach,
1157 	.pru_bind =		tcp_usr_bind,
1158 	.pru_connect =		tcp_usr_connect,
1159 	.pru_control =		in_control,
1160 	.pru_detach =		tcp_usr_detach,
1161 	.pru_disconnect =	tcp_usr_disconnect,
1162 	.pru_listen =		tcp_usr_listen,
1163 	.pru_peeraddr =		in_getpeeraddr,
1164 	.pru_rcvd =		tcp_usr_rcvd,
1165 	.pru_rcvoob =		tcp_usr_rcvoob,
1166 	.pru_send =		tcp_usr_send,
1167 	.pru_ready =		tcp_usr_ready,
1168 	.pru_shutdown =		tcp_usr_shutdown,
1169 	.pru_sockaddr =		in_getsockaddr,
1170 	.pru_sosetlabel =	in_pcbsosetlabel,
1171 	.pru_close =		tcp_usr_close,
1172 };
1173 #endif /* INET */
1174 
1175 #ifdef INET6
1176 struct pr_usrreqs tcp6_usrreqs = {
1177 	.pru_abort =		tcp_usr_abort,
1178 	.pru_accept =		tcp6_usr_accept,
1179 	.pru_attach =		tcp_usr_attach,
1180 	.pru_bind =		tcp6_usr_bind,
1181 	.pru_connect =		tcp6_usr_connect,
1182 	.pru_control =		in6_control,
1183 	.pru_detach =		tcp_usr_detach,
1184 	.pru_disconnect =	tcp_usr_disconnect,
1185 	.pru_listen =		tcp6_usr_listen,
1186 	.pru_peeraddr =		in6_mapped_peeraddr,
1187 	.pru_rcvd =		tcp_usr_rcvd,
1188 	.pru_rcvoob =		tcp_usr_rcvoob,
1189 	.pru_send =		tcp_usr_send,
1190 	.pru_ready =		tcp_usr_ready,
1191 	.pru_shutdown =		tcp_usr_shutdown,
1192 	.pru_sockaddr =		in6_mapped_sockaddr,
1193 	.pru_sosetlabel =	in_pcbsosetlabel,
1194 	.pru_close =		tcp_usr_close,
1195 };
1196 #endif /* INET6 */
1197 
1198 #ifdef INET
1199 /*
1200  * Common subroutine to open a TCP connection to remote host specified
1201  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1202  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1203  * to choose a local host address (interface).  If there is an existing
1204  * incarnation of the same connection in TIME-WAIT state and if the remote
1205  * host was sending CC options and if the connection duration was < MSL, then
1206  * truncate the previous TIME-WAIT state and proceed.
1207  * Initialize connection parameters and enter SYN-SENT state.
1208  */
1209 static int
1210 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1211 {
1212 	struct inpcb *inp = tp->t_inpcb, *oinp;
1213 	struct socket *so = inp->inp_socket;
1214 	struct in_addr laddr;
1215 	u_short lport;
1216 	int error;
1217 
1218 	INP_WLOCK_ASSERT(inp);
1219 	INP_HASH_WLOCK(&V_tcbinfo);
1220 
1221 	if (inp->inp_lport == 0) {
1222 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1223 		if (error)
1224 			goto out;
1225 	}
1226 
1227 	/*
1228 	 * Cannot simply call in_pcbconnect, because there might be an
1229 	 * earlier incarnation of this same connection still in
1230 	 * TIME_WAIT state, creating an ADDRINUSE error.
1231 	 */
1232 	laddr = inp->inp_laddr;
1233 	lport = inp->inp_lport;
1234 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1235 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1236 	if (error && oinp == NULL)
1237 		goto out;
1238 	if (oinp) {
1239 		error = EADDRINUSE;
1240 		goto out;
1241 	}
1242 	inp->inp_laddr = laddr;
1243 	in_pcbrehash(inp);
1244 	INP_HASH_WUNLOCK(&V_tcbinfo);
1245 
1246 	/*
1247 	 * Compute window scaling to request:
1248 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1249 	 * XXX: This should move to tcp_output().
1250 	 */
1251 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1252 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1253 		tp->request_r_scale++;
1254 
1255 	soisconnecting(so);
1256 	TCPSTAT_INC(tcps_connattempt);
1257 	tcp_state_change(tp, TCPS_SYN_SENT);
1258 	tp->iss = tcp_new_isn(tp);
1259 	tcp_sendseqinit(tp);
1260 
1261 	return 0;
1262 
1263 out:
1264 	INP_HASH_WUNLOCK(&V_tcbinfo);
1265 	return (error);
1266 }
1267 #endif /* INET */
1268 
1269 #ifdef INET6
1270 static int
1271 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1272 {
1273 	struct inpcb *inp = tp->t_inpcb;
1274 	int error;
1275 
1276 	INP_WLOCK_ASSERT(inp);
1277 	INP_HASH_WLOCK(&V_tcbinfo);
1278 
1279 	if (inp->inp_lport == 0) {
1280 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1281 		if (error)
1282 			goto out;
1283 	}
1284 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1285 	if (error != 0)
1286 		goto out;
1287 	INP_HASH_WUNLOCK(&V_tcbinfo);
1288 
1289 	/* Compute window scaling to request.  */
1290 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1291 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1292 		tp->request_r_scale++;
1293 
1294 	soisconnecting(inp->inp_socket);
1295 	TCPSTAT_INC(tcps_connattempt);
1296 	tcp_state_change(tp, TCPS_SYN_SENT);
1297 	tp->iss = tcp_new_isn(tp);
1298 	tcp_sendseqinit(tp);
1299 
1300 	return 0;
1301 
1302 out:
1303 	INP_HASH_WUNLOCK(&V_tcbinfo);
1304 	return error;
1305 }
1306 #endif /* INET6 */
1307 
1308 /*
1309  * Export TCP internal state information via a struct tcp_info, based on the
1310  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1311  * (TCP state machine, etc).  We export all information using FreeBSD-native
1312  * constants -- for example, the numeric values for tcpi_state will differ
1313  * from Linux.
1314  */
1315 static void
1316 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1317 {
1318 
1319 	INP_WLOCK_ASSERT(tp->t_inpcb);
1320 	bzero(ti, sizeof(*ti));
1321 
1322 	ti->tcpi_state = tp->t_state;
1323 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1324 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1325 	if (tp->t_flags & TF_SACK_PERMIT)
1326 		ti->tcpi_options |= TCPI_OPT_SACK;
1327 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1328 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1329 		ti->tcpi_snd_wscale = tp->snd_scale;
1330 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1331 	}
1332 
1333 	ti->tcpi_rto = tp->t_rxtcur * tick;
1334 	ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1335 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1336 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1337 
1338 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1339 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1340 
1341 	/*
1342 	 * FreeBSD-specific extension fields for tcp_info.
1343 	 */
1344 	ti->tcpi_rcv_space = tp->rcv_wnd;
1345 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1346 	ti->tcpi_snd_wnd = tp->snd_wnd;
1347 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1348 	ti->tcpi_snd_nxt = tp->snd_nxt;
1349 	ti->tcpi_snd_mss = tp->t_maxseg;
1350 	ti->tcpi_rcv_mss = tp->t_maxseg;
1351 	if (tp->t_flags & TF_TOE)
1352 		ti->tcpi_options |= TCPI_OPT_TOE;
1353 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1354 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1355 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1356 }
1357 
1358 /*
1359  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1360  * socket option arguments.  When it re-acquires the lock after the copy, it
1361  * has to revalidate that the connection is still valid for the socket
1362  * option.
1363  */
1364 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1365 	INP_WLOCK(inp);							\
1366 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1367 		INP_WUNLOCK(inp);					\
1368 		cleanup;						\
1369 		return (ECONNRESET);					\
1370 	}								\
1371 	tp = intotcpcb(inp);						\
1372 } while(0)
1373 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1374 
1375 int
1376 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1377 {
1378 	int	error;
1379 	struct	inpcb *inp;
1380 	struct	tcpcb *tp;
1381 	struct tcp_function_block *blk;
1382 	struct tcp_function_set fsn;
1383 
1384 	error = 0;
1385 	inp = sotoinpcb(so);
1386 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1387 	INP_WLOCK(inp);
1388 	if (sopt->sopt_level != IPPROTO_TCP) {
1389 #ifdef INET6
1390 		if (inp->inp_vflag & INP_IPV6PROTO) {
1391 			INP_WUNLOCK(inp);
1392 			error = ip6_ctloutput(so, sopt);
1393 		}
1394 #endif /* INET6 */
1395 #if defined(INET6) && defined(INET)
1396 		else
1397 #endif
1398 #ifdef INET
1399 		{
1400 			INP_WUNLOCK(inp);
1401 			error = ip_ctloutput(so, sopt);
1402 		}
1403 #endif
1404 		return (error);
1405 	}
1406 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1407 		INP_WUNLOCK(inp);
1408 		return (ECONNRESET);
1409 	}
1410 	tp = intotcpcb(inp);
1411 	/*
1412 	 * Protect the TCP option TCP_FUNCTION_BLK so
1413 	 * that a sub-function can *never* overwrite this.
1414 	 */
1415 	if ((sopt->sopt_dir == SOPT_SET) &&
1416 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1417 		INP_WUNLOCK(inp);
1418 		error = sooptcopyin(sopt, &fsn, sizeof fsn,
1419 		    sizeof fsn);
1420 		if (error)
1421 			return (error);
1422 		INP_WLOCK_RECHECK(inp);
1423 		if (tp->t_state != TCPS_CLOSED) {
1424 			/*
1425 			 * The user has advanced the state
1426 			 * past the initial point, we can't
1427 			 * switch since we are down the road
1428 			 * and a new set of functions may
1429 			 * not be compatibile.
1430 			 */
1431 			INP_WUNLOCK(inp);
1432 			return(EINVAL);
1433 		}
1434 		blk = find_and_ref_tcp_functions(&fsn);
1435 		if (blk == NULL) {
1436 			INP_WUNLOCK(inp);
1437 			return (ENOENT);
1438 		}
1439 		if (tp->t_fb != blk) {
1440 			if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1441 				refcount_release(&blk->tfb_refcnt);
1442 				INP_WUNLOCK(inp);
1443 				return (ENOENT);
1444 			}
1445 			/*
1446 			 * Release the old refcnt, the
1447 			 * lookup acquires a ref on the
1448 			 * new one.
1449 			 */
1450 			if (tp->t_fb->tfb_tcp_fb_fini)
1451 				(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1452 			refcount_release(&tp->t_fb->tfb_refcnt);
1453 			tp->t_fb = blk;
1454 			if (tp->t_fb->tfb_tcp_fb_init) {
1455 				(*tp->t_fb->tfb_tcp_fb_init)(tp);
1456 			}
1457 		}
1458 #ifdef TCP_OFFLOAD
1459 		if (tp->t_flags & TF_TOE) {
1460 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1461 			     sopt->sopt_name);
1462 		}
1463 #endif
1464 		INP_WUNLOCK(inp);
1465 		return (error);
1466 	} else if ((sopt->sopt_dir == SOPT_GET) &&
1467 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1468 		strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1469 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1470 		INP_WUNLOCK(inp);
1471 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1472 		return (error);
1473 	}
1474 	/* Pass in the INP locked, called must unlock it */
1475 	return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1476 }
1477 
1478 int
1479 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1480 {
1481 	int	error, opt, optval;
1482 	u_int	ui;
1483 	struct	tcp_info ti;
1484 	struct cc_algo *algo;
1485 	char	*pbuf, buf[TCP_CA_NAME_MAX];
1486 	size_t	len;
1487 
1488 	/*
1489 	 * For TCP_CCALGOOPT forward the control to CC module, for both
1490 	 * SOPT_SET and SOPT_GET.
1491 	 */
1492 	switch (sopt->sopt_name) {
1493 	case TCP_CCALGOOPT:
1494 		INP_WUNLOCK(inp);
1495 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1496 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1497 		    sopt->sopt_valsize);
1498 		if (error) {
1499 			free(pbuf, M_TEMP);
1500 			return (error);
1501 		}
1502 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1503 		if (CC_ALGO(tp)->ctl_output != NULL)
1504 			error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1505 		else
1506 			error = ENOENT;
1507 		INP_WUNLOCK(inp);
1508 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
1509 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1510 		free(pbuf, M_TEMP);
1511 		return (error);
1512 	}
1513 
1514 	switch (sopt->sopt_dir) {
1515 	case SOPT_SET:
1516 		switch (sopt->sopt_name) {
1517 #ifdef TCP_SIGNATURE
1518 		case TCP_MD5SIG:
1519 			INP_WUNLOCK(inp);
1520 			error = sooptcopyin(sopt, &optval, sizeof optval,
1521 			    sizeof optval);
1522 			if (error)
1523 				return (error);
1524 
1525 			INP_WLOCK_RECHECK(inp);
1526 			if (optval > 0)
1527 				tp->t_flags |= TF_SIGNATURE;
1528 			else
1529 				tp->t_flags &= ~TF_SIGNATURE;
1530 			goto unlock_and_done;
1531 #endif /* TCP_SIGNATURE */
1532 
1533 		case TCP_NODELAY:
1534 		case TCP_NOOPT:
1535 			INP_WUNLOCK(inp);
1536 			error = sooptcopyin(sopt, &optval, sizeof optval,
1537 			    sizeof optval);
1538 			if (error)
1539 				return (error);
1540 
1541 			INP_WLOCK_RECHECK(inp);
1542 			switch (sopt->sopt_name) {
1543 			case TCP_NODELAY:
1544 				opt = TF_NODELAY;
1545 				break;
1546 			case TCP_NOOPT:
1547 				opt = TF_NOOPT;
1548 				break;
1549 			default:
1550 				opt = 0; /* dead code to fool gcc */
1551 				break;
1552 			}
1553 
1554 			if (optval)
1555 				tp->t_flags |= opt;
1556 			else
1557 				tp->t_flags &= ~opt;
1558 unlock_and_done:
1559 #ifdef TCP_OFFLOAD
1560 			if (tp->t_flags & TF_TOE) {
1561 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
1562 				    sopt->sopt_name);
1563 			}
1564 #endif
1565 			INP_WUNLOCK(inp);
1566 			break;
1567 
1568 		case TCP_NOPUSH:
1569 			INP_WUNLOCK(inp);
1570 			error = sooptcopyin(sopt, &optval, sizeof optval,
1571 			    sizeof optval);
1572 			if (error)
1573 				return (error);
1574 
1575 			INP_WLOCK_RECHECK(inp);
1576 			if (optval)
1577 				tp->t_flags |= TF_NOPUSH;
1578 			else if (tp->t_flags & TF_NOPUSH) {
1579 				tp->t_flags &= ~TF_NOPUSH;
1580 				if (TCPS_HAVEESTABLISHED(tp->t_state))
1581 					error = tp->t_fb->tfb_tcp_output(tp);
1582 			}
1583 			goto unlock_and_done;
1584 
1585 		case TCP_MAXSEG:
1586 			INP_WUNLOCK(inp);
1587 			error = sooptcopyin(sopt, &optval, sizeof optval,
1588 			    sizeof optval);
1589 			if (error)
1590 				return (error);
1591 
1592 			INP_WLOCK_RECHECK(inp);
1593 			if (optval > 0 && optval <= tp->t_maxseg &&
1594 			    optval + 40 >= V_tcp_minmss)
1595 				tp->t_maxseg = optval;
1596 			else
1597 				error = EINVAL;
1598 			goto unlock_and_done;
1599 
1600 		case TCP_INFO:
1601 			INP_WUNLOCK(inp);
1602 			error = EINVAL;
1603 			break;
1604 
1605 		case TCP_CONGESTION:
1606 			INP_WUNLOCK(inp);
1607 			error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1608 			if (error)
1609 				break;
1610 			buf[sopt->sopt_valsize] = '\0';
1611 			INP_WLOCK_RECHECK(inp);
1612 			CC_LIST_RLOCK();
1613 			STAILQ_FOREACH(algo, &cc_list, entries)
1614 				if (strncmp(buf, algo->name,
1615 				    TCP_CA_NAME_MAX) == 0)
1616 					break;
1617 			CC_LIST_RUNLOCK();
1618 			if (algo == NULL) {
1619 				INP_WUNLOCK(inp);
1620 				error = EINVAL;
1621 				break;
1622 			}
1623 			/*
1624 			 * We hold a write lock over the tcb so it's safe to
1625 			 * do these things without ordering concerns.
1626 			 */
1627 			if (CC_ALGO(tp)->cb_destroy != NULL)
1628 				CC_ALGO(tp)->cb_destroy(tp->ccv);
1629 			CC_ALGO(tp) = algo;
1630 			/*
1631 			 * If something goes pear shaped initialising the new
1632 			 * algo, fall back to newreno (which does not
1633 			 * require initialisation).
1634 			 */
1635 			if (algo->cb_init != NULL &&
1636 			    algo->cb_init(tp->ccv) != 0) {
1637 				CC_ALGO(tp) = &newreno_cc_algo;
1638 				/*
1639 				 * The only reason init should fail is
1640 				 * because of malloc.
1641 				 */
1642 				error = ENOMEM;
1643 			}
1644 			INP_WUNLOCK(inp);
1645 			break;
1646 
1647 		case TCP_KEEPIDLE:
1648 		case TCP_KEEPINTVL:
1649 		case TCP_KEEPINIT:
1650 			INP_WUNLOCK(inp);
1651 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1652 			if (error)
1653 				return (error);
1654 
1655 			if (ui > (UINT_MAX / hz)) {
1656 				error = EINVAL;
1657 				break;
1658 			}
1659 			ui *= hz;
1660 
1661 			INP_WLOCK_RECHECK(inp);
1662 			switch (sopt->sopt_name) {
1663 			case TCP_KEEPIDLE:
1664 				tp->t_keepidle = ui;
1665 				/*
1666 				 * XXX: better check current remaining
1667 				 * timeout and "merge" it with new value.
1668 				 */
1669 				if ((tp->t_state > TCPS_LISTEN) &&
1670 				    (tp->t_state <= TCPS_CLOSING))
1671 					tcp_timer_activate(tp, TT_KEEP,
1672 					    TP_KEEPIDLE(tp));
1673 				break;
1674 			case TCP_KEEPINTVL:
1675 				tp->t_keepintvl = ui;
1676 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1677 				    (TP_MAXIDLE(tp) > 0))
1678 					tcp_timer_activate(tp, TT_2MSL,
1679 					    TP_MAXIDLE(tp));
1680 				break;
1681 			case TCP_KEEPINIT:
1682 				tp->t_keepinit = ui;
1683 				if (tp->t_state == TCPS_SYN_RECEIVED ||
1684 				    tp->t_state == TCPS_SYN_SENT)
1685 					tcp_timer_activate(tp, TT_KEEP,
1686 					    TP_KEEPINIT(tp));
1687 				break;
1688 			}
1689 			goto unlock_and_done;
1690 
1691 		case TCP_KEEPCNT:
1692 			INP_WUNLOCK(inp);
1693 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1694 			if (error)
1695 				return (error);
1696 
1697 			INP_WLOCK_RECHECK(inp);
1698 			tp->t_keepcnt = ui;
1699 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1700 			    (TP_MAXIDLE(tp) > 0))
1701 				tcp_timer_activate(tp, TT_2MSL,
1702 				    TP_MAXIDLE(tp));
1703 			goto unlock_and_done;
1704 
1705 #ifdef TCPPCAP
1706 		case TCP_PCAP_OUT:
1707 		case TCP_PCAP_IN:
1708 			INP_WUNLOCK(inp);
1709 			error = sooptcopyin(sopt, &optval, sizeof optval,
1710 			    sizeof optval);
1711 			if (error)
1712 				return (error);
1713 
1714 			INP_WLOCK_RECHECK(inp);
1715 			if (optval >= 0)
1716 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1717 					&(tp->t_outpkts) : &(tp->t_inpkts),
1718 					optval);
1719 			else
1720 				error = EINVAL;
1721 			goto unlock_and_done;
1722 #endif
1723 
1724 #ifdef TCP_RFC7413
1725 		case TCP_FASTOPEN:
1726 			INP_WUNLOCK(inp);
1727 			if (!V_tcp_fastopen_enabled)
1728 				return (EPERM);
1729 
1730 			error = sooptcopyin(sopt, &optval, sizeof optval,
1731 			    sizeof optval);
1732 			if (error)
1733 				return (error);
1734 
1735 			INP_WLOCK_RECHECK(inp);
1736 			if (optval) {
1737 				tp->t_flags |= TF_FASTOPEN;
1738 				if ((tp->t_state == TCPS_LISTEN) &&
1739 				    (tp->t_tfo_pending == NULL))
1740 					tp->t_tfo_pending =
1741 					    tcp_fastopen_alloc_counter();
1742 			} else
1743 				tp->t_flags &= ~TF_FASTOPEN;
1744 			goto unlock_and_done;
1745 #endif
1746 
1747 		default:
1748 			INP_WUNLOCK(inp);
1749 			error = ENOPROTOOPT;
1750 			break;
1751 		}
1752 		break;
1753 
1754 	case SOPT_GET:
1755 		tp = intotcpcb(inp);
1756 		switch (sopt->sopt_name) {
1757 #ifdef TCP_SIGNATURE
1758 		case TCP_MD5SIG:
1759 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1760 			INP_WUNLOCK(inp);
1761 			error = sooptcopyout(sopt, &optval, sizeof optval);
1762 			break;
1763 #endif
1764 
1765 		case TCP_NODELAY:
1766 			optval = tp->t_flags & TF_NODELAY;
1767 			INP_WUNLOCK(inp);
1768 			error = sooptcopyout(sopt, &optval, sizeof optval);
1769 			break;
1770 		case TCP_MAXSEG:
1771 			optval = tp->t_maxseg;
1772 			INP_WUNLOCK(inp);
1773 			error = sooptcopyout(sopt, &optval, sizeof optval);
1774 			break;
1775 		case TCP_NOOPT:
1776 			optval = tp->t_flags & TF_NOOPT;
1777 			INP_WUNLOCK(inp);
1778 			error = sooptcopyout(sopt, &optval, sizeof optval);
1779 			break;
1780 		case TCP_NOPUSH:
1781 			optval = tp->t_flags & TF_NOPUSH;
1782 			INP_WUNLOCK(inp);
1783 			error = sooptcopyout(sopt, &optval, sizeof optval);
1784 			break;
1785 		case TCP_INFO:
1786 			tcp_fill_info(tp, &ti);
1787 			INP_WUNLOCK(inp);
1788 			error = sooptcopyout(sopt, &ti, sizeof ti);
1789 			break;
1790 		case TCP_CONGESTION:
1791 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1792 			INP_WUNLOCK(inp);
1793 			error = sooptcopyout(sopt, buf, len + 1);
1794 			break;
1795 		case TCP_KEEPIDLE:
1796 		case TCP_KEEPINTVL:
1797 		case TCP_KEEPINIT:
1798 		case TCP_KEEPCNT:
1799 			switch (sopt->sopt_name) {
1800 			case TCP_KEEPIDLE:
1801 				ui = tp->t_keepidle / hz;
1802 				break;
1803 			case TCP_KEEPINTVL:
1804 				ui = tp->t_keepintvl / hz;
1805 				break;
1806 			case TCP_KEEPINIT:
1807 				ui = tp->t_keepinit / hz;
1808 				break;
1809 			case TCP_KEEPCNT:
1810 				ui = tp->t_keepcnt;
1811 				break;
1812 			}
1813 			INP_WUNLOCK(inp);
1814 			error = sooptcopyout(sopt, &ui, sizeof(ui));
1815 			break;
1816 #ifdef TCPPCAP
1817 		case TCP_PCAP_OUT:
1818 		case TCP_PCAP_IN:
1819 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1820 					&(tp->t_outpkts) : &(tp->t_inpkts));
1821 			INP_WUNLOCK(inp);
1822 			error = sooptcopyout(sopt, &optval, sizeof optval);
1823 			break;
1824 #endif
1825 
1826 #ifdef TCP_RFC7413
1827 		case TCP_FASTOPEN:
1828 			optval = tp->t_flags & TF_FASTOPEN;
1829 			INP_WUNLOCK(inp);
1830 			error = sooptcopyout(sopt, &optval, sizeof optval);
1831 			break;
1832 #endif
1833 		default:
1834 			INP_WUNLOCK(inp);
1835 			error = ENOPROTOOPT;
1836 			break;
1837 		}
1838 		break;
1839 	}
1840 	return (error);
1841 }
1842 #undef INP_WLOCK_RECHECK
1843 #undef INP_WLOCK_RECHECK_CLEANUP
1844 
1845 /*
1846  * Attach TCP protocol to socket, allocating
1847  * internet protocol control block, tcp control block,
1848  * bufer space, and entering LISTEN state if to accept connections.
1849  */
1850 static int
1851 tcp_attach(struct socket *so)
1852 {
1853 	struct tcpcb *tp;
1854 	struct inpcb *inp;
1855 	int error;
1856 
1857 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1858 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1859 		if (error)
1860 			return (error);
1861 	}
1862 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
1863 	so->so_snd.sb_flags |= SB_AUTOSIZE;
1864 	INP_INFO_RLOCK(&V_tcbinfo);
1865 	error = in_pcballoc(so, &V_tcbinfo);
1866 	if (error) {
1867 		INP_INFO_RUNLOCK(&V_tcbinfo);
1868 		return (error);
1869 	}
1870 	inp = sotoinpcb(so);
1871 #ifdef INET6
1872 	if (inp->inp_vflag & INP_IPV6PROTO) {
1873 		inp->inp_vflag |= INP_IPV6;
1874 		inp->in6p_hops = -1;	/* use kernel default */
1875 	}
1876 	else
1877 #endif
1878 	inp->inp_vflag |= INP_IPV4;
1879 	tp = tcp_newtcpcb(inp);
1880 	if (tp == NULL) {
1881 		in_pcbdetach(inp);
1882 		in_pcbfree(inp);
1883 		INP_INFO_RUNLOCK(&V_tcbinfo);
1884 		return (ENOBUFS);
1885 	}
1886 	tp->t_state = TCPS_CLOSED;
1887 	INP_WUNLOCK(inp);
1888 	INP_INFO_RUNLOCK(&V_tcbinfo);
1889 	TCPSTATES_INC(TCPS_CLOSED);
1890 	return (0);
1891 }
1892 
1893 /*
1894  * Initiate (or continue) disconnect.
1895  * If embryonic state, just send reset (once).
1896  * If in ``let data drain'' option and linger null, just drop.
1897  * Otherwise (hard), mark socket disconnecting and drop
1898  * current input data; switch states based on user close, and
1899  * send segment to peer (with FIN).
1900  */
1901 static void
1902 tcp_disconnect(struct tcpcb *tp)
1903 {
1904 	struct inpcb *inp = tp->t_inpcb;
1905 	struct socket *so = inp->inp_socket;
1906 
1907 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1908 	INP_WLOCK_ASSERT(inp);
1909 
1910 	/*
1911 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1912 	 * socket is still open.
1913 	 */
1914 	if (tp->t_state < TCPS_ESTABLISHED) {
1915 		tp = tcp_close(tp);
1916 		KASSERT(tp != NULL,
1917 		    ("tcp_disconnect: tcp_close() returned NULL"));
1918 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1919 		tp = tcp_drop(tp, 0);
1920 		KASSERT(tp != NULL,
1921 		    ("tcp_disconnect: tcp_drop() returned NULL"));
1922 	} else {
1923 		soisdisconnecting(so);
1924 		sbflush(&so->so_rcv);
1925 		tcp_usrclosed(tp);
1926 		if (!(inp->inp_flags & INP_DROPPED))
1927 			tp->t_fb->tfb_tcp_output(tp);
1928 	}
1929 }
1930 
1931 /*
1932  * User issued close, and wish to trail through shutdown states:
1933  * if never received SYN, just forget it.  If got a SYN from peer,
1934  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1935  * If already got a FIN from peer, then almost done; go to LAST_ACK
1936  * state.  In all other cases, have already sent FIN to peer (e.g.
1937  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1938  * for peer to send FIN or not respond to keep-alives, etc.
1939  * We can let the user exit from the close as soon as the FIN is acked.
1940  */
1941 static void
1942 tcp_usrclosed(struct tcpcb *tp)
1943 {
1944 
1945 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1946 	INP_WLOCK_ASSERT(tp->t_inpcb);
1947 
1948 	switch (tp->t_state) {
1949 	case TCPS_LISTEN:
1950 #ifdef TCP_OFFLOAD
1951 		tcp_offload_listen_stop(tp);
1952 #endif
1953 		tcp_state_change(tp, TCPS_CLOSED);
1954 		/* FALLTHROUGH */
1955 	case TCPS_CLOSED:
1956 		tp = tcp_close(tp);
1957 		/*
1958 		 * tcp_close() should never return NULL here as the socket is
1959 		 * still open.
1960 		 */
1961 		KASSERT(tp != NULL,
1962 		    ("tcp_usrclosed: tcp_close() returned NULL"));
1963 		break;
1964 
1965 	case TCPS_SYN_SENT:
1966 	case TCPS_SYN_RECEIVED:
1967 		tp->t_flags |= TF_NEEDFIN;
1968 		break;
1969 
1970 	case TCPS_ESTABLISHED:
1971 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
1972 		break;
1973 
1974 	case TCPS_CLOSE_WAIT:
1975 		tcp_state_change(tp, TCPS_LAST_ACK);
1976 		break;
1977 	}
1978 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
1979 		soisdisconnected(tp->t_inpcb->inp_socket);
1980 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
1981 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1982 			int timeout;
1983 
1984 			timeout = (tcp_fast_finwait2_recycle) ?
1985 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
1986 			tcp_timer_activate(tp, TT_2MSL, timeout);
1987 		}
1988 	}
1989 }
1990 
1991 #ifdef DDB
1992 static void
1993 db_print_indent(int indent)
1994 {
1995 	int i;
1996 
1997 	for (i = 0; i < indent; i++)
1998 		db_printf(" ");
1999 }
2000 
2001 static void
2002 db_print_tstate(int t_state)
2003 {
2004 
2005 	switch (t_state) {
2006 	case TCPS_CLOSED:
2007 		db_printf("TCPS_CLOSED");
2008 		return;
2009 
2010 	case TCPS_LISTEN:
2011 		db_printf("TCPS_LISTEN");
2012 		return;
2013 
2014 	case TCPS_SYN_SENT:
2015 		db_printf("TCPS_SYN_SENT");
2016 		return;
2017 
2018 	case TCPS_SYN_RECEIVED:
2019 		db_printf("TCPS_SYN_RECEIVED");
2020 		return;
2021 
2022 	case TCPS_ESTABLISHED:
2023 		db_printf("TCPS_ESTABLISHED");
2024 		return;
2025 
2026 	case TCPS_CLOSE_WAIT:
2027 		db_printf("TCPS_CLOSE_WAIT");
2028 		return;
2029 
2030 	case TCPS_FIN_WAIT_1:
2031 		db_printf("TCPS_FIN_WAIT_1");
2032 		return;
2033 
2034 	case TCPS_CLOSING:
2035 		db_printf("TCPS_CLOSING");
2036 		return;
2037 
2038 	case TCPS_LAST_ACK:
2039 		db_printf("TCPS_LAST_ACK");
2040 		return;
2041 
2042 	case TCPS_FIN_WAIT_2:
2043 		db_printf("TCPS_FIN_WAIT_2");
2044 		return;
2045 
2046 	case TCPS_TIME_WAIT:
2047 		db_printf("TCPS_TIME_WAIT");
2048 		return;
2049 
2050 	default:
2051 		db_printf("unknown");
2052 		return;
2053 	}
2054 }
2055 
2056 static void
2057 db_print_tflags(u_int t_flags)
2058 {
2059 	int comma;
2060 
2061 	comma = 0;
2062 	if (t_flags & TF_ACKNOW) {
2063 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
2064 		comma = 1;
2065 	}
2066 	if (t_flags & TF_DELACK) {
2067 		db_printf("%sTF_DELACK", comma ? ", " : "");
2068 		comma = 1;
2069 	}
2070 	if (t_flags & TF_NODELAY) {
2071 		db_printf("%sTF_NODELAY", comma ? ", " : "");
2072 		comma = 1;
2073 	}
2074 	if (t_flags & TF_NOOPT) {
2075 		db_printf("%sTF_NOOPT", comma ? ", " : "");
2076 		comma = 1;
2077 	}
2078 	if (t_flags & TF_SENTFIN) {
2079 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2080 		comma = 1;
2081 	}
2082 	if (t_flags & TF_REQ_SCALE) {
2083 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2084 		comma = 1;
2085 	}
2086 	if (t_flags & TF_RCVD_SCALE) {
2087 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2088 		comma = 1;
2089 	}
2090 	if (t_flags & TF_REQ_TSTMP) {
2091 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2092 		comma = 1;
2093 	}
2094 	if (t_flags & TF_RCVD_TSTMP) {
2095 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2096 		comma = 1;
2097 	}
2098 	if (t_flags & TF_SACK_PERMIT) {
2099 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2100 		comma = 1;
2101 	}
2102 	if (t_flags & TF_NEEDSYN) {
2103 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2104 		comma = 1;
2105 	}
2106 	if (t_flags & TF_NEEDFIN) {
2107 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2108 		comma = 1;
2109 	}
2110 	if (t_flags & TF_NOPUSH) {
2111 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2112 		comma = 1;
2113 	}
2114 	if (t_flags & TF_MORETOCOME) {
2115 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2116 		comma = 1;
2117 	}
2118 	if (t_flags & TF_LQ_OVERFLOW) {
2119 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2120 		comma = 1;
2121 	}
2122 	if (t_flags & TF_LASTIDLE) {
2123 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2124 		comma = 1;
2125 	}
2126 	if (t_flags & TF_RXWIN0SENT) {
2127 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2128 		comma = 1;
2129 	}
2130 	if (t_flags & TF_FASTRECOVERY) {
2131 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2132 		comma = 1;
2133 	}
2134 	if (t_flags & TF_CONGRECOVERY) {
2135 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2136 		comma = 1;
2137 	}
2138 	if (t_flags & TF_WASFRECOVERY) {
2139 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2140 		comma = 1;
2141 	}
2142 	if (t_flags & TF_SIGNATURE) {
2143 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2144 		comma = 1;
2145 	}
2146 	if (t_flags & TF_FORCEDATA) {
2147 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2148 		comma = 1;
2149 	}
2150 	if (t_flags & TF_TSO) {
2151 		db_printf("%sTF_TSO", comma ? ", " : "");
2152 		comma = 1;
2153 	}
2154 	if (t_flags & TF_ECN_PERMIT) {
2155 		db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2156 		comma = 1;
2157 	}
2158 	if (t_flags & TF_FASTOPEN) {
2159 		db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2160 		comma = 1;
2161 	}
2162 }
2163 
2164 static void
2165 db_print_toobflags(char t_oobflags)
2166 {
2167 	int comma;
2168 
2169 	comma = 0;
2170 	if (t_oobflags & TCPOOB_HAVEDATA) {
2171 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2172 		comma = 1;
2173 	}
2174 	if (t_oobflags & TCPOOB_HADDATA) {
2175 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2176 		comma = 1;
2177 	}
2178 }
2179 
2180 static void
2181 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2182 {
2183 
2184 	db_print_indent(indent);
2185 	db_printf("%s at %p\n", name, tp);
2186 
2187 	indent += 2;
2188 
2189 	db_print_indent(indent);
2190 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2191 	   LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2192 
2193 	db_print_indent(indent);
2194 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2195 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2196 
2197 	db_print_indent(indent);
2198 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2199 	    &tp->t_timers->tt_delack, tp->t_inpcb);
2200 
2201 	db_print_indent(indent);
2202 	db_printf("t_state: %d (", tp->t_state);
2203 	db_print_tstate(tp->t_state);
2204 	db_printf(")\n");
2205 
2206 	db_print_indent(indent);
2207 	db_printf("t_flags: 0x%x (", tp->t_flags);
2208 	db_print_tflags(tp->t_flags);
2209 	db_printf(")\n");
2210 
2211 	db_print_indent(indent);
2212 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2213 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2214 
2215 	db_print_indent(indent);
2216 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2217 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2218 
2219 	db_print_indent(indent);
2220 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2221 	    tp->iss, tp->irs, tp->rcv_nxt);
2222 
2223 	db_print_indent(indent);
2224 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2225 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2226 
2227 	db_print_indent(indent);
2228 	db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2229 	   tp->snd_wnd, tp->snd_cwnd);
2230 
2231 	db_print_indent(indent);
2232 	db_printf("snd_ssthresh: %lu   snd_recover: "
2233 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2234 
2235 	db_print_indent(indent);
2236 	db_printf("t_rcvtime: %u   t_startime: %u\n",
2237 	    tp->t_rcvtime, tp->t_starttime);
2238 
2239 	db_print_indent(indent);
2240 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2241 	    tp->t_rtttime, tp->t_rtseq);
2242 
2243 	db_print_indent(indent);
2244 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2245 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2246 
2247 	db_print_indent(indent);
2248 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2249 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2250 	    tp->t_rttbest);
2251 
2252 	db_print_indent(indent);
2253 	db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2254 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2255 
2256 	db_print_indent(indent);
2257 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2258 	db_print_toobflags(tp->t_oobflags);
2259 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2260 
2261 	db_print_indent(indent);
2262 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2263 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2264 
2265 	db_print_indent(indent);
2266 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2267 	    tp->ts_recent, tp->ts_recent_age);
2268 
2269 	db_print_indent(indent);
2270 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2271 	    "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2272 
2273 	db_print_indent(indent);
2274 	db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2275 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2276 	    tp->snd_recover_prev, tp->t_badrxtwin);
2277 
2278 	db_print_indent(indent);
2279 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2280 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2281 
2282 	db_print_indent(indent);
2283 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2284 	    "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2285 
2286 	/* Skip sackblks, sackhint. */
2287 
2288 	db_print_indent(indent);
2289 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2290 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2291 }
2292 
2293 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2294 {
2295 	struct tcpcb *tp;
2296 
2297 	if (!have_addr) {
2298 		db_printf("usage: show tcpcb <addr>\n");
2299 		return;
2300 	}
2301 	tp = (struct tcpcb *)addr;
2302 
2303 	db_print_tcpcb(tp, "tcpcb", 0);
2304 }
2305 #endif
2306