xref: /freebsd/sys/netinet/tcp_usrreq.c (revision b6a05070fa77edc7ce6e60b61623fd806e807be6)
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/cc.h>
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 #include <netinet/tcp_fsm.h>
86 #include <netinet/tcp_seq.h>
87 #include <netinet/tcp_timer.h>
88 #include <netinet/tcp_var.h>
89 #include <netinet/tcpip.h>
90 #ifdef TCPPCAP
91 #include <netinet/tcp_pcap.h>
92 #endif
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #ifdef TCP_OFFLOAD
97 #include <netinet/tcp_offload.h>
98 #endif
99 
100 /*
101  * TCP protocol interface to socket abstraction.
102  */
103 static int	tcp_attach(struct socket *);
104 #ifdef INET
105 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
106 		    struct thread *td);
107 #endif /* INET */
108 #ifdef INET6
109 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
110 		    struct thread *td);
111 #endif /* INET6 */
112 static void	tcp_disconnect(struct tcpcb *);
113 static void	tcp_usrclosed(struct tcpcb *);
114 static void	tcp_fill_info(struct tcpcb *, struct tcp_info *);
115 
116 #ifdef TCPDEBUG
117 #define	TCPDEBUG0	int ostate = 0
118 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
119 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
120 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
121 #else
122 #define	TCPDEBUG0
123 #define	TCPDEBUG1()
124 #define	TCPDEBUG2(req)
125 #endif
126 
127 /*
128  * TCP attaches to socket via pru_attach(), reserving space,
129  * and an internet control block.
130  */
131 static int
132 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
133 {
134 	struct inpcb *inp;
135 	struct tcpcb *tp = NULL;
136 	int error;
137 	TCPDEBUG0;
138 
139 	inp = sotoinpcb(so);
140 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
141 	TCPDEBUG1();
142 
143 	error = tcp_attach(so);
144 	if (error)
145 		goto out;
146 
147 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
148 		so->so_linger = TCP_LINGERTIME;
149 
150 	inp = sotoinpcb(so);
151 	tp = intotcpcb(inp);
152 out:
153 	TCPDEBUG2(PRU_ATTACH);
154 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
155 	return error;
156 }
157 
158 /*
159  * tcp_detach is called when the socket layer loses its final reference
160  * to the socket, be it a file descriptor reference, a reference from TCP,
161  * etc.  At this point, there is only one case in which we will keep around
162  * inpcb state: time wait.
163  *
164  * This function can probably be re-absorbed back into tcp_usr_detach() now
165  * that there is a single detach path.
166  */
167 static void
168 tcp_detach(struct socket *so, struct inpcb *inp)
169 {
170 	struct tcpcb *tp;
171 
172 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
173 	INP_WLOCK_ASSERT(inp);
174 
175 	KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
176 	KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
177 
178 	tp = intotcpcb(inp);
179 
180 	if (inp->inp_flags & INP_TIMEWAIT) {
181 		/*
182 		 * There are two cases to handle: one in which the time wait
183 		 * state is being discarded (INP_DROPPED), and one in which
184 		 * this connection will remain in timewait.  In the former,
185 		 * it is time to discard all state (except tcptw, which has
186 		 * already been discarded by the timewait close code, which
187 		 * should be further up the call stack somewhere).  In the
188 		 * latter case, we detach from the socket, but leave the pcb
189 		 * present until timewait ends.
190 		 *
191 		 * XXXRW: Would it be cleaner to free the tcptw here?
192 		 *
193 		 * Astute question indeed, from twtcp perspective there are
194 		 * three cases to consider:
195 		 *
196 		 * #1 tcp_detach is called at tcptw creation time by
197 		 *  tcp_twstart, then do not discard the newly created tcptw
198 		 *  and leave inpcb present until timewait ends
199 		 * #2 tcp_detach is called at timewait end (or reuse) by
200 		 *  tcp_twclose, then the tcptw has already been discarded
201 		 *  (or reused) and inpcb is freed here
202 		 * #3 tcp_detach is called() after timewait ends (or reuse)
203 		 *  (e.g. by soclose), then tcptw has already been discarded
204 		 *  (or reused) and inpcb is freed here
205 		 *
206 		 *  In all three cases the tcptw should not be freed here.
207 		 */
208 		if (inp->inp_flags & INP_DROPPED) {
209 			KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
210 			    "INP_DROPPED && tp != NULL"));
211 			in_pcbdetach(inp);
212 			in_pcbfree(inp);
213 		} else {
214 			in_pcbdetach(inp);
215 			INP_WUNLOCK(inp);
216 		}
217 	} else {
218 		/*
219 		 * If the connection is not in timewait, we consider two
220 		 * two conditions: one in which no further processing is
221 		 * necessary (dropped || embryonic), and one in which TCP is
222 		 * not yet done, but no longer requires the socket, so the
223 		 * pcb will persist for the time being.
224 		 *
225 		 * XXXRW: Does the second case still occur?
226 		 */
227 		if (inp->inp_flags & INP_DROPPED ||
228 		    tp->t_state < TCPS_SYN_SENT) {
229 			tcp_discardcb(tp);
230 			in_pcbdetach(inp);
231 			in_pcbfree(inp);
232 		} else {
233 			in_pcbdetach(inp);
234 			INP_WUNLOCK(inp);
235 		}
236 	}
237 }
238 
239 /*
240  * pru_detach() detaches the TCP protocol from the socket.
241  * If the protocol state is non-embryonic, then can't
242  * do this directly: have to initiate a pru_disconnect(),
243  * which may finish later; embryonic TCB's can just
244  * be discarded here.
245  */
246 static void
247 tcp_usr_detach(struct socket *so)
248 {
249 	struct inpcb *inp;
250 	int rlock = 0;
251 
252 	inp = sotoinpcb(so);
253 	KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
254 	if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
255 		INP_INFO_RLOCK(&V_tcbinfo);
256 		rlock = 1;
257 	}
258 	INP_WLOCK(inp);
259 	KASSERT(inp->inp_socket != NULL,
260 	    ("tcp_usr_detach: inp_socket == NULL"));
261 	tcp_detach(so, inp);
262 	if (rlock)
263 		INP_INFO_RUNLOCK(&V_tcbinfo);
264 }
265 
266 #ifdef INET
267 /*
268  * Give the socket an address.
269  */
270 static int
271 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
272 {
273 	int error = 0;
274 	struct inpcb *inp;
275 	struct tcpcb *tp = NULL;
276 	struct sockaddr_in *sinp;
277 
278 	sinp = (struct sockaddr_in *)nam;
279 	if (nam->sa_len != sizeof (*sinp))
280 		return (EINVAL);
281 	/*
282 	 * Must check for multicast addresses and disallow binding
283 	 * to them.
284 	 */
285 	if (sinp->sin_family == AF_INET &&
286 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
287 		return (EAFNOSUPPORT);
288 
289 	TCPDEBUG0;
290 	inp = sotoinpcb(so);
291 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
292 	INP_WLOCK(inp);
293 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
294 		error = EINVAL;
295 		goto out;
296 	}
297 	tp = intotcpcb(inp);
298 	TCPDEBUG1();
299 	INP_HASH_WLOCK(&V_tcbinfo);
300 	error = in_pcbbind(inp, nam, td->td_ucred);
301 	INP_HASH_WUNLOCK(&V_tcbinfo);
302 out:
303 	TCPDEBUG2(PRU_BIND);
304 	TCP_PROBE2(debug__user, tp, PRU_BIND);
305 	INP_WUNLOCK(inp);
306 
307 	return (error);
308 }
309 #endif /* INET */
310 
311 #ifdef INET6
312 static int
313 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
314 {
315 	int error = 0;
316 	struct inpcb *inp;
317 	struct tcpcb *tp = NULL;
318 	struct sockaddr_in6 *sin6p;
319 
320 	sin6p = (struct sockaddr_in6 *)nam;
321 	if (nam->sa_len != sizeof (*sin6p))
322 		return (EINVAL);
323 	/*
324 	 * Must check for multicast addresses and disallow binding
325 	 * to them.
326 	 */
327 	if (sin6p->sin6_family == AF_INET6 &&
328 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
329 		return (EAFNOSUPPORT);
330 
331 	TCPDEBUG0;
332 	inp = sotoinpcb(so);
333 	KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
334 	INP_WLOCK(inp);
335 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
336 		error = EINVAL;
337 		goto out;
338 	}
339 	tp = intotcpcb(inp);
340 	TCPDEBUG1();
341 	INP_HASH_WLOCK(&V_tcbinfo);
342 	inp->inp_vflag &= ~INP_IPV4;
343 	inp->inp_vflag |= INP_IPV6;
344 #ifdef INET
345 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
346 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
347 			inp->inp_vflag |= INP_IPV4;
348 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
349 			struct sockaddr_in sin;
350 
351 			in6_sin6_2_sin(&sin, sin6p);
352 			inp->inp_vflag |= INP_IPV4;
353 			inp->inp_vflag &= ~INP_IPV6;
354 			error = in_pcbbind(inp, (struct sockaddr *)&sin,
355 			    td->td_ucred);
356 			INP_HASH_WUNLOCK(&V_tcbinfo);
357 			goto out;
358 		}
359 	}
360 #endif
361 	error = in6_pcbbind(inp, nam, td->td_ucred);
362 	INP_HASH_WUNLOCK(&V_tcbinfo);
363 out:
364 	TCPDEBUG2(PRU_BIND);
365 	TCP_PROBE2(debug__user, tp, PRU_BIND);
366 	INP_WUNLOCK(inp);
367 	return (error);
368 }
369 #endif /* INET6 */
370 
371 #ifdef INET
372 /*
373  * Prepare to accept connections.
374  */
375 static int
376 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
377 {
378 	int error = 0;
379 	struct inpcb *inp;
380 	struct tcpcb *tp = NULL;
381 
382 	TCPDEBUG0;
383 	inp = sotoinpcb(so);
384 	KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
385 	INP_WLOCK(inp);
386 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
387 		error = EINVAL;
388 		goto out;
389 	}
390 	tp = intotcpcb(inp);
391 	TCPDEBUG1();
392 	SOCK_LOCK(so);
393 	error = solisten_proto_check(so);
394 	INP_HASH_WLOCK(&V_tcbinfo);
395 	if (error == 0 && inp->inp_lport == 0)
396 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
397 	INP_HASH_WUNLOCK(&V_tcbinfo);
398 	if (error == 0) {
399 		tcp_state_change(tp, TCPS_LISTEN);
400 		solisten_proto(so, backlog);
401 #ifdef TCP_OFFLOAD
402 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
403 			tcp_offload_listen_start(tp);
404 #endif
405 	}
406 	SOCK_UNLOCK(so);
407 
408 out:
409 	TCPDEBUG2(PRU_LISTEN);
410 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
411 	INP_WUNLOCK(inp);
412 	return (error);
413 }
414 #endif /* INET */
415 
416 #ifdef INET6
417 static int
418 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
419 {
420 	int error = 0;
421 	struct inpcb *inp;
422 	struct tcpcb *tp = NULL;
423 
424 	TCPDEBUG0;
425 	inp = sotoinpcb(so);
426 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
427 	INP_WLOCK(inp);
428 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
429 		error = EINVAL;
430 		goto out;
431 	}
432 	tp = intotcpcb(inp);
433 	TCPDEBUG1();
434 	SOCK_LOCK(so);
435 	error = solisten_proto_check(so);
436 	INP_HASH_WLOCK(&V_tcbinfo);
437 	if (error == 0 && inp->inp_lport == 0) {
438 		inp->inp_vflag &= ~INP_IPV4;
439 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
440 			inp->inp_vflag |= INP_IPV4;
441 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
442 	}
443 	INP_HASH_WUNLOCK(&V_tcbinfo);
444 	if (error == 0) {
445 		tcp_state_change(tp, TCPS_LISTEN);
446 		solisten_proto(so, backlog);
447 #ifdef TCP_OFFLOAD
448 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
449 			tcp_offload_listen_start(tp);
450 #endif
451 	}
452 	SOCK_UNLOCK(so);
453 
454 out:
455 	TCPDEBUG2(PRU_LISTEN);
456 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
457 	INP_WUNLOCK(inp);
458 	return (error);
459 }
460 #endif /* INET6 */
461 
462 #ifdef INET
463 /*
464  * Initiate connection to peer.
465  * Create a template for use in transmissions on this connection.
466  * Enter SYN_SENT state, and mark socket as connecting.
467  * Start keep-alive timer, and seed output sequence space.
468  * Send initial segment on connection.
469  */
470 static int
471 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
472 {
473 	int error = 0;
474 	struct inpcb *inp;
475 	struct tcpcb *tp = NULL;
476 	struct sockaddr_in *sinp;
477 
478 	sinp = (struct sockaddr_in *)nam;
479 	if (nam->sa_len != sizeof (*sinp))
480 		return (EINVAL);
481 	/*
482 	 * Must disallow TCP ``connections'' to multicast addresses.
483 	 */
484 	if (sinp->sin_family == AF_INET
485 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
486 		return (EAFNOSUPPORT);
487 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
488 		return (error);
489 
490 	TCPDEBUG0;
491 	inp = sotoinpcb(so);
492 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
493 	INP_WLOCK(inp);
494 	if (inp->inp_flags & INP_TIMEWAIT) {
495 		error = EADDRINUSE;
496 		goto out;
497 	}
498 	if (inp->inp_flags & INP_DROPPED) {
499 		error = ECONNREFUSED;
500 		goto out;
501 	}
502 	tp = intotcpcb(inp);
503 	TCPDEBUG1();
504 	if ((error = tcp_connect(tp, nam, td)) != 0)
505 		goto out;
506 #ifdef TCP_OFFLOAD
507 	if (registered_toedevs > 0 &&
508 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
509 	    (error = tcp_offload_connect(so, nam)) == 0)
510 		goto out;
511 #endif
512 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
513 	error = tp->t_fb->tfb_tcp_output(tp);
514 out:
515 	TCPDEBUG2(PRU_CONNECT);
516 	INP_WUNLOCK(inp);
517 	return (error);
518 }
519 #endif /* INET */
520 
521 #ifdef INET6
522 static int
523 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
524 {
525 	int error = 0;
526 	struct inpcb *inp;
527 	struct tcpcb *tp = NULL;
528 	struct sockaddr_in6 *sin6p;
529 
530 	TCPDEBUG0;
531 
532 	sin6p = (struct sockaddr_in6 *)nam;
533 	if (nam->sa_len != sizeof (*sin6p))
534 		return (EINVAL);
535 	/*
536 	 * Must disallow TCP ``connections'' to multicast addresses.
537 	 */
538 	if (sin6p->sin6_family == AF_INET6
539 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
540 		return (EAFNOSUPPORT);
541 
542 	inp = sotoinpcb(so);
543 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
544 	INP_WLOCK(inp);
545 	if (inp->inp_flags & INP_TIMEWAIT) {
546 		error = EADDRINUSE;
547 		goto out;
548 	}
549 	if (inp->inp_flags & INP_DROPPED) {
550 		error = ECONNREFUSED;
551 		goto out;
552 	}
553 	tp = intotcpcb(inp);
554 	TCPDEBUG1();
555 #ifdef INET
556 	/*
557 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
558 	 * therefore probably require the hash lock, which isn't held here.
559 	 * Is this a significant problem?
560 	 */
561 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
562 		struct sockaddr_in sin;
563 
564 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
565 			error = EINVAL;
566 			goto out;
567 		}
568 
569 		in6_sin6_2_sin(&sin, sin6p);
570 		inp->inp_vflag |= INP_IPV4;
571 		inp->inp_vflag &= ~INP_IPV6;
572 		if ((error = prison_remote_ip4(td->td_ucred,
573 		    &sin.sin_addr)) != 0)
574 			goto out;
575 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
576 			goto out;
577 #ifdef TCP_OFFLOAD
578 		if (registered_toedevs > 0 &&
579 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
580 		    (error = tcp_offload_connect(so, nam)) == 0)
581 			goto out;
582 #endif
583 		error = tp->t_fb->tfb_tcp_output(tp);
584 		goto out;
585 	}
586 #endif
587 	inp->inp_vflag &= ~INP_IPV4;
588 	inp->inp_vflag |= INP_IPV6;
589 	inp->inp_inc.inc_flags |= INC_ISIPV6;
590 	if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
591 		goto out;
592 	if ((error = tcp6_connect(tp, nam, td)) != 0)
593 		goto out;
594 #ifdef TCP_OFFLOAD
595 	if (registered_toedevs > 0 &&
596 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
597 	    (error = tcp_offload_connect(so, nam)) == 0)
598 		goto out;
599 #endif
600 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
601 	error = tp->t_fb->tfb_tcp_output(tp);
602 
603 out:
604 	TCPDEBUG2(PRU_CONNECT);
605 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
606 	INP_WUNLOCK(inp);
607 	return (error);
608 }
609 #endif /* INET6 */
610 
611 /*
612  * Initiate disconnect from peer.
613  * If connection never passed embryonic stage, just drop;
614  * else if don't need to let data drain, then can just drop anyways,
615  * else have to begin TCP shutdown process: mark socket disconnecting,
616  * drain unread data, state switch to reflect user close, and
617  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
618  * when peer sends FIN and acks ours.
619  *
620  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
621  */
622 static int
623 tcp_usr_disconnect(struct socket *so)
624 {
625 	struct inpcb *inp;
626 	struct tcpcb *tp = NULL;
627 	int error = 0;
628 
629 	TCPDEBUG0;
630 	INP_INFO_RLOCK(&V_tcbinfo);
631 	inp = sotoinpcb(so);
632 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
633 	INP_WLOCK(inp);
634 	if (inp->inp_flags & INP_TIMEWAIT)
635 		goto out;
636 	if (inp->inp_flags & INP_DROPPED) {
637 		error = ECONNRESET;
638 		goto out;
639 	}
640 	tp = intotcpcb(inp);
641 	TCPDEBUG1();
642 	tcp_disconnect(tp);
643 out:
644 	TCPDEBUG2(PRU_DISCONNECT);
645 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
646 	INP_WUNLOCK(inp);
647 	INP_INFO_RUNLOCK(&V_tcbinfo);
648 	return (error);
649 }
650 
651 #ifdef INET
652 /*
653  * Accept a connection.  Essentially all the work is done at higher levels;
654  * just return the address of the peer, storing through addr.
655  */
656 static int
657 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
658 {
659 	int error = 0;
660 	struct inpcb *inp = NULL;
661 	struct tcpcb *tp = NULL;
662 	struct in_addr addr;
663 	in_port_t port = 0;
664 	TCPDEBUG0;
665 
666 	if (so->so_state & SS_ISDISCONNECTED)
667 		return (ECONNABORTED);
668 
669 	inp = sotoinpcb(so);
670 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
671 	INP_WLOCK(inp);
672 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
673 		error = ECONNABORTED;
674 		goto out;
675 	}
676 	tp = intotcpcb(inp);
677 	TCPDEBUG1();
678 
679 	/*
680 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
681 	 * copy the data of interest and defer the malloc until after we
682 	 * release the lock.
683 	 */
684 	port = inp->inp_fport;
685 	addr = inp->inp_faddr;
686 
687 out:
688 	TCPDEBUG2(PRU_ACCEPT);
689 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
690 	INP_WUNLOCK(inp);
691 	if (error == 0)
692 		*nam = in_sockaddr(port, &addr);
693 	return error;
694 }
695 #endif /* INET */
696 
697 #ifdef INET6
698 static int
699 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
700 {
701 	struct inpcb *inp = NULL;
702 	int error = 0;
703 	struct tcpcb *tp = NULL;
704 	struct in_addr addr;
705 	struct in6_addr addr6;
706 	in_port_t port = 0;
707 	int v4 = 0;
708 	TCPDEBUG0;
709 
710 	if (so->so_state & SS_ISDISCONNECTED)
711 		return (ECONNABORTED);
712 
713 	inp = sotoinpcb(so);
714 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
715 	INP_INFO_RLOCK(&V_tcbinfo);
716 	INP_WLOCK(inp);
717 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
718 		error = ECONNABORTED;
719 		goto out;
720 	}
721 	tp = intotcpcb(inp);
722 	TCPDEBUG1();
723 
724 	/*
725 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
726 	 * copy the data of interest and defer the malloc until after we
727 	 * release the lock.
728 	 */
729 	if (inp->inp_vflag & INP_IPV4) {
730 		v4 = 1;
731 		port = inp->inp_fport;
732 		addr = inp->inp_faddr;
733 	} else {
734 		port = inp->inp_fport;
735 		addr6 = inp->in6p_faddr;
736 	}
737 
738 out:
739 	TCPDEBUG2(PRU_ACCEPT);
740 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
741 	INP_WUNLOCK(inp);
742 	INP_INFO_RUNLOCK(&V_tcbinfo);
743 	if (error == 0) {
744 		if (v4)
745 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
746 		else
747 			*nam = in6_sockaddr(port, &addr6);
748 	}
749 	return error;
750 }
751 #endif /* INET6 */
752 
753 /*
754  * Mark the connection as being incapable of further output.
755  */
756 static int
757 tcp_usr_shutdown(struct socket *so)
758 {
759 	int error = 0;
760 	struct inpcb *inp;
761 	struct tcpcb *tp = NULL;
762 
763 	TCPDEBUG0;
764 	INP_INFO_RLOCK(&V_tcbinfo);
765 	inp = sotoinpcb(so);
766 	KASSERT(inp != NULL, ("inp == NULL"));
767 	INP_WLOCK(inp);
768 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
769 		error = ECONNRESET;
770 		goto out;
771 	}
772 	tp = intotcpcb(inp);
773 	TCPDEBUG1();
774 	socantsendmore(so);
775 	tcp_usrclosed(tp);
776 	if (!(inp->inp_flags & INP_DROPPED))
777 		error = tp->t_fb->tfb_tcp_output(tp);
778 
779 out:
780 	TCPDEBUG2(PRU_SHUTDOWN);
781 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
782 	INP_WUNLOCK(inp);
783 	INP_INFO_RUNLOCK(&V_tcbinfo);
784 
785 	return (error);
786 }
787 
788 /*
789  * After a receive, possibly send window update to peer.
790  */
791 static int
792 tcp_usr_rcvd(struct socket *so, int flags)
793 {
794 	struct inpcb *inp;
795 	struct tcpcb *tp = NULL;
796 	int error = 0;
797 
798 	TCPDEBUG0;
799 	inp = sotoinpcb(so);
800 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
801 	INP_WLOCK(inp);
802 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
803 		error = ECONNRESET;
804 		goto out;
805 	}
806 	tp = intotcpcb(inp);
807 	TCPDEBUG1();
808 #ifdef TCP_OFFLOAD
809 	if (tp->t_flags & TF_TOE)
810 		tcp_offload_rcvd(tp);
811 	else
812 #endif
813 	tp->t_fb->tfb_tcp_output(tp);
814 
815 out:
816 	TCPDEBUG2(PRU_RCVD);
817 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
818 	INP_WUNLOCK(inp);
819 	return (error);
820 }
821 
822 /*
823  * Do a send by putting data in output queue and updating urgent
824  * marker if URG set.  Possibly send more data.  Unlike the other
825  * pru_*() routines, the mbuf chains are our responsibility.  We
826  * must either enqueue them or free them.  The other pru_* routines
827  * generally are caller-frees.
828  */
829 static int
830 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
831     struct sockaddr *nam, struct mbuf *control, struct thread *td)
832 {
833 	int error = 0;
834 	struct inpcb *inp;
835 	struct tcpcb *tp = NULL;
836 #ifdef INET6
837 	int isipv6;
838 #endif
839 	TCPDEBUG0;
840 
841 	/*
842 	 * We require the pcbinfo lock if we will close the socket as part of
843 	 * this call.
844 	 */
845 	if (flags & PRUS_EOF)
846 		INP_INFO_RLOCK(&V_tcbinfo);
847 	inp = sotoinpcb(so);
848 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
849 	INP_WLOCK(inp);
850 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
851 		if (control)
852 			m_freem(control);
853 		/*
854 		 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
855 		 * for freeing memory.
856 		 */
857 		if (m && (flags & PRUS_NOTREADY) == 0)
858 			m_freem(m);
859 		error = ECONNRESET;
860 		goto out;
861 	}
862 #ifdef INET6
863 	isipv6 = nam && nam->sa_family == AF_INET6;
864 #endif /* INET6 */
865 	tp = intotcpcb(inp);
866 	TCPDEBUG1();
867 	if (control) {
868 		/* TCP doesn't do control messages (rights, creds, etc) */
869 		if (control->m_len) {
870 			m_freem(control);
871 			if (m)
872 				m_freem(m);
873 			error = EINVAL;
874 			goto out;
875 		}
876 		m_freem(control);	/* empty control, just free it */
877 	}
878 	if (!(flags & PRUS_OOB)) {
879 		sbappendstream(&so->so_snd, m, flags);
880 		if (nam && tp->t_state < TCPS_SYN_SENT) {
881 			/*
882 			 * Do implied connect if not yet connected,
883 			 * initialize window to default value, and
884 			 * initialize maxseg/maxopd using peer's cached
885 			 * MSS.
886 			 */
887 #ifdef INET6
888 			if (isipv6)
889 				error = tcp6_connect(tp, nam, td);
890 #endif /* INET6 */
891 #if defined(INET6) && defined(INET)
892 			else
893 #endif
894 #ifdef INET
895 				error = tcp_connect(tp, nam, td);
896 #endif
897 			if (error)
898 				goto out;
899 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
900 			tcp_mss(tp, -1);
901 		}
902 		if (flags & PRUS_EOF) {
903 			/*
904 			 * Close the send side of the connection after
905 			 * the data is sent.
906 			 */
907 			INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
908 			socantsendmore(so);
909 			tcp_usrclosed(tp);
910 		}
911 		if (!(inp->inp_flags & INP_DROPPED) &&
912 		    !(flags & PRUS_NOTREADY)) {
913 			if (flags & PRUS_MORETOCOME)
914 				tp->t_flags |= TF_MORETOCOME;
915 			error = tp->t_fb->tfb_tcp_output(tp);
916 			if (flags & PRUS_MORETOCOME)
917 				tp->t_flags &= ~TF_MORETOCOME;
918 		}
919 	} else {
920 		/*
921 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
922 		 */
923 		SOCKBUF_LOCK(&so->so_snd);
924 		if (sbspace(&so->so_snd) < -512) {
925 			SOCKBUF_UNLOCK(&so->so_snd);
926 			m_freem(m);
927 			error = ENOBUFS;
928 			goto out;
929 		}
930 		/*
931 		 * According to RFC961 (Assigned Protocols),
932 		 * the urgent pointer points to the last octet
933 		 * of urgent data.  We continue, however,
934 		 * to consider it to indicate the first octet
935 		 * of data past the urgent section.
936 		 * Otherwise, snd_up should be one lower.
937 		 */
938 		sbappendstream_locked(&so->so_snd, m, flags);
939 		SOCKBUF_UNLOCK(&so->so_snd);
940 		if (nam && tp->t_state < TCPS_SYN_SENT) {
941 			/*
942 			 * Do implied connect if not yet connected,
943 			 * initialize window to default value, and
944 			 * initialize maxseg/maxopd using peer's cached
945 			 * MSS.
946 			 */
947 #ifdef INET6
948 			if (isipv6)
949 				error = tcp6_connect(tp, nam, td);
950 #endif /* INET6 */
951 #if defined(INET6) && defined(INET)
952 			else
953 #endif
954 #ifdef INET
955 				error = tcp_connect(tp, nam, td);
956 #endif
957 			if (error)
958 				goto out;
959 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
960 			tcp_mss(tp, -1);
961 		}
962 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
963 		if (!(flags & PRUS_NOTREADY)) {
964 			tp->t_flags |= TF_FORCEDATA;
965 			error = tp->t_fb->tfb_tcp_output(tp);
966 			tp->t_flags &= ~TF_FORCEDATA;
967 		}
968 	}
969 out:
970 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
971 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
972 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
973 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
974 	INP_WUNLOCK(inp);
975 	if (flags & PRUS_EOF)
976 		INP_INFO_RUNLOCK(&V_tcbinfo);
977 	return (error);
978 }
979 
980 static int
981 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
982 {
983 	struct inpcb *inp;
984 	struct tcpcb *tp;
985 	int error;
986 
987 	inp = sotoinpcb(so);
988 	INP_WLOCK(inp);
989 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
990 		INP_WUNLOCK(inp);
991 		for (int i = 0; i < count; i++)
992 			m = m_free(m);
993 		return (ECONNRESET);
994 	}
995 	tp = intotcpcb(inp);
996 
997 	SOCKBUF_LOCK(&so->so_snd);
998 	error = sbready(&so->so_snd, m, count);
999 	SOCKBUF_UNLOCK(&so->so_snd);
1000 	if (error == 0)
1001 		error = tp->t_fb->tfb_tcp_output(tp);
1002 	INP_WUNLOCK(inp);
1003 
1004 	return (error);
1005 }
1006 
1007 /*
1008  * Abort the TCP.  Drop the connection abruptly.
1009  */
1010 static void
1011 tcp_usr_abort(struct socket *so)
1012 {
1013 	struct inpcb *inp;
1014 	struct tcpcb *tp = NULL;
1015 	TCPDEBUG0;
1016 
1017 	inp = sotoinpcb(so);
1018 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1019 
1020 	INP_INFO_RLOCK(&V_tcbinfo);
1021 	INP_WLOCK(inp);
1022 	KASSERT(inp->inp_socket != NULL,
1023 	    ("tcp_usr_abort: inp_socket == NULL"));
1024 
1025 	/*
1026 	 * If we still have full TCP state, and we're not dropped, drop.
1027 	 */
1028 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1029 	    !(inp->inp_flags & INP_DROPPED)) {
1030 		tp = intotcpcb(inp);
1031 		TCPDEBUG1();
1032 		tcp_drop(tp, ECONNABORTED);
1033 		TCPDEBUG2(PRU_ABORT);
1034 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1035 	}
1036 	if (!(inp->inp_flags & INP_DROPPED)) {
1037 		SOCK_LOCK(so);
1038 		so->so_state |= SS_PROTOREF;
1039 		SOCK_UNLOCK(so);
1040 		inp->inp_flags |= INP_SOCKREF;
1041 	}
1042 	INP_WUNLOCK(inp);
1043 	INP_INFO_RUNLOCK(&V_tcbinfo);
1044 }
1045 
1046 /*
1047  * TCP socket is closed.  Start friendly disconnect.
1048  */
1049 static void
1050 tcp_usr_close(struct socket *so)
1051 {
1052 	struct inpcb *inp;
1053 	struct tcpcb *tp = NULL;
1054 	TCPDEBUG0;
1055 
1056 	inp = sotoinpcb(so);
1057 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1058 
1059 	INP_INFO_RLOCK(&V_tcbinfo);
1060 	INP_WLOCK(inp);
1061 	KASSERT(inp->inp_socket != NULL,
1062 	    ("tcp_usr_close: inp_socket == NULL"));
1063 
1064 	/*
1065 	 * If we still have full TCP state, and we're not dropped, initiate
1066 	 * a disconnect.
1067 	 */
1068 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1069 	    !(inp->inp_flags & INP_DROPPED)) {
1070 		tp = intotcpcb(inp);
1071 		TCPDEBUG1();
1072 		tcp_disconnect(tp);
1073 		TCPDEBUG2(PRU_CLOSE);
1074 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
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  * Receive out-of-band data.
1088  */
1089 static int
1090 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1091 {
1092 	int error = 0;
1093 	struct inpcb *inp;
1094 	struct tcpcb *tp = NULL;
1095 
1096 	TCPDEBUG0;
1097 	inp = sotoinpcb(so);
1098 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1099 	INP_WLOCK(inp);
1100 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1101 		error = ECONNRESET;
1102 		goto out;
1103 	}
1104 	tp = intotcpcb(inp);
1105 	TCPDEBUG1();
1106 	if ((so->so_oobmark == 0 &&
1107 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1108 	    so->so_options & SO_OOBINLINE ||
1109 	    tp->t_oobflags & TCPOOB_HADDATA) {
1110 		error = EINVAL;
1111 		goto out;
1112 	}
1113 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1114 		error = EWOULDBLOCK;
1115 		goto out;
1116 	}
1117 	m->m_len = 1;
1118 	*mtod(m, caddr_t) = tp->t_iobc;
1119 	if ((flags & MSG_PEEK) == 0)
1120 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1121 
1122 out:
1123 	TCPDEBUG2(PRU_RCVOOB);
1124 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1125 	INP_WUNLOCK(inp);
1126 	return (error);
1127 }
1128 
1129 #ifdef INET
1130 struct pr_usrreqs tcp_usrreqs = {
1131 	.pru_abort =		tcp_usr_abort,
1132 	.pru_accept =		tcp_usr_accept,
1133 	.pru_attach =		tcp_usr_attach,
1134 	.pru_bind =		tcp_usr_bind,
1135 	.pru_connect =		tcp_usr_connect,
1136 	.pru_control =		in_control,
1137 	.pru_detach =		tcp_usr_detach,
1138 	.pru_disconnect =	tcp_usr_disconnect,
1139 	.pru_listen =		tcp_usr_listen,
1140 	.pru_peeraddr =		in_getpeeraddr,
1141 	.pru_rcvd =		tcp_usr_rcvd,
1142 	.pru_rcvoob =		tcp_usr_rcvoob,
1143 	.pru_send =		tcp_usr_send,
1144 	.pru_ready =		tcp_usr_ready,
1145 	.pru_shutdown =		tcp_usr_shutdown,
1146 	.pru_sockaddr =		in_getsockaddr,
1147 	.pru_sosetlabel =	in_pcbsosetlabel,
1148 	.pru_close =		tcp_usr_close,
1149 };
1150 #endif /* INET */
1151 
1152 #ifdef INET6
1153 struct pr_usrreqs tcp6_usrreqs = {
1154 	.pru_abort =		tcp_usr_abort,
1155 	.pru_accept =		tcp6_usr_accept,
1156 	.pru_attach =		tcp_usr_attach,
1157 	.pru_bind =		tcp6_usr_bind,
1158 	.pru_connect =		tcp6_usr_connect,
1159 	.pru_control =		in6_control,
1160 	.pru_detach =		tcp_usr_detach,
1161 	.pru_disconnect =	tcp_usr_disconnect,
1162 	.pru_listen =		tcp6_usr_listen,
1163 	.pru_peeraddr =		in6_mapped_peeraddr,
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 =		in6_mapped_sockaddr,
1170 	.pru_sosetlabel =	in_pcbsosetlabel,
1171 	.pru_close =		tcp_usr_close,
1172 };
1173 #endif /* INET6 */
1174 
1175 #ifdef INET
1176 /*
1177  * Common subroutine to open a TCP connection to remote host specified
1178  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1179  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1180  * to choose a local host address (interface).  If there is an existing
1181  * incarnation of the same connection in TIME-WAIT state and if the remote
1182  * host was sending CC options and if the connection duration was < MSL, then
1183  * truncate the previous TIME-WAIT state and proceed.
1184  * Initialize connection parameters and enter SYN-SENT state.
1185  */
1186 static int
1187 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1188 {
1189 	struct inpcb *inp = tp->t_inpcb, *oinp;
1190 	struct socket *so = inp->inp_socket;
1191 	struct in_addr laddr;
1192 	u_short lport;
1193 	int error;
1194 
1195 	INP_WLOCK_ASSERT(inp);
1196 	INP_HASH_WLOCK(&V_tcbinfo);
1197 
1198 	if (inp->inp_lport == 0) {
1199 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1200 		if (error)
1201 			goto out;
1202 	}
1203 
1204 	/*
1205 	 * Cannot simply call in_pcbconnect, because there might be an
1206 	 * earlier incarnation of this same connection still in
1207 	 * TIME_WAIT state, creating an ADDRINUSE error.
1208 	 */
1209 	laddr = inp->inp_laddr;
1210 	lport = inp->inp_lport;
1211 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1212 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1213 	if (error && oinp == NULL)
1214 		goto out;
1215 	if (oinp) {
1216 		error = EADDRINUSE;
1217 		goto out;
1218 	}
1219 	inp->inp_laddr = laddr;
1220 	in_pcbrehash(inp);
1221 	INP_HASH_WUNLOCK(&V_tcbinfo);
1222 
1223 	/*
1224 	 * Compute window scaling to request:
1225 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1226 	 * XXX: This should move to tcp_output().
1227 	 */
1228 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1229 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1230 		tp->request_r_scale++;
1231 
1232 	soisconnecting(so);
1233 	TCPSTAT_INC(tcps_connattempt);
1234 	tcp_state_change(tp, TCPS_SYN_SENT);
1235 	tp->iss = tcp_new_isn(tp);
1236 	tcp_sendseqinit(tp);
1237 
1238 	return 0;
1239 
1240 out:
1241 	INP_HASH_WUNLOCK(&V_tcbinfo);
1242 	return (error);
1243 }
1244 #endif /* INET */
1245 
1246 #ifdef INET6
1247 static int
1248 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1249 {
1250 	struct inpcb *inp = tp->t_inpcb;
1251 	int error;
1252 
1253 	INP_WLOCK_ASSERT(inp);
1254 	INP_HASH_WLOCK(&V_tcbinfo);
1255 
1256 	if (inp->inp_lport == 0) {
1257 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1258 		if (error)
1259 			goto out;
1260 	}
1261 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1262 	if (error != 0)
1263 		goto out;
1264 	INP_HASH_WUNLOCK(&V_tcbinfo);
1265 
1266 	/* Compute window scaling to request.  */
1267 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1268 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1269 		tp->request_r_scale++;
1270 
1271 	soisconnecting(inp->inp_socket);
1272 	TCPSTAT_INC(tcps_connattempt);
1273 	tcp_state_change(tp, TCPS_SYN_SENT);
1274 	tp->iss = tcp_new_isn(tp);
1275 	tcp_sendseqinit(tp);
1276 
1277 	return 0;
1278 
1279 out:
1280 	INP_HASH_WUNLOCK(&V_tcbinfo);
1281 	return error;
1282 }
1283 #endif /* INET6 */
1284 
1285 /*
1286  * Export TCP internal state information via a struct tcp_info, based on the
1287  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1288  * (TCP state machine, etc).  We export all information using FreeBSD-native
1289  * constants -- for example, the numeric values for tcpi_state will differ
1290  * from Linux.
1291  */
1292 static void
1293 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1294 {
1295 
1296 	INP_WLOCK_ASSERT(tp->t_inpcb);
1297 	bzero(ti, sizeof(*ti));
1298 
1299 	ti->tcpi_state = tp->t_state;
1300 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1301 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1302 	if (tp->t_flags & TF_SACK_PERMIT)
1303 		ti->tcpi_options |= TCPI_OPT_SACK;
1304 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1305 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1306 		ti->tcpi_snd_wscale = tp->snd_scale;
1307 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1308 	}
1309 
1310 	ti->tcpi_rto = tp->t_rxtcur * tick;
1311 	ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1312 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1313 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1314 
1315 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1316 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1317 
1318 	/*
1319 	 * FreeBSD-specific extension fields for tcp_info.
1320 	 */
1321 	ti->tcpi_rcv_space = tp->rcv_wnd;
1322 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1323 	ti->tcpi_snd_wnd = tp->snd_wnd;
1324 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1325 	ti->tcpi_snd_nxt = tp->snd_nxt;
1326 	ti->tcpi_snd_mss = tp->t_maxseg;
1327 	ti->tcpi_rcv_mss = tp->t_maxseg;
1328 	if (tp->t_flags & TF_TOE)
1329 		ti->tcpi_options |= TCPI_OPT_TOE;
1330 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1331 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1332 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1333 }
1334 
1335 /*
1336  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1337  * socket option arguments.  When it re-acquires the lock after the copy, it
1338  * has to revalidate that the connection is still valid for the socket
1339  * option.
1340  */
1341 #define INP_WLOCK_RECHECK(inp) do {					\
1342 	INP_WLOCK(inp);							\
1343 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1344 		INP_WUNLOCK(inp);					\
1345 		return (ECONNRESET);					\
1346 	}								\
1347 	tp = intotcpcb(inp);						\
1348 } while(0)
1349 
1350 int
1351 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1352 {
1353 	int	error;
1354 	struct	inpcb *inp;
1355 	struct	tcpcb *tp;
1356 	struct tcp_function_block *blk;
1357 	struct tcp_function_set fsn;
1358 
1359 	error = 0;
1360 	inp = sotoinpcb(so);
1361 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1362 	INP_WLOCK(inp);
1363 	if (sopt->sopt_level != IPPROTO_TCP) {
1364 #ifdef INET6
1365 		if (inp->inp_vflag & INP_IPV6PROTO) {
1366 			INP_WUNLOCK(inp);
1367 			error = ip6_ctloutput(so, sopt);
1368 		}
1369 #endif /* INET6 */
1370 #if defined(INET6) && defined(INET)
1371 		else
1372 #endif
1373 #ifdef INET
1374 		{
1375 			INP_WUNLOCK(inp);
1376 			error = ip_ctloutput(so, sopt);
1377 		}
1378 #endif
1379 		return (error);
1380 	}
1381 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1382 		INP_WUNLOCK(inp);
1383 		return (ECONNRESET);
1384 	}
1385 	tp = intotcpcb(inp);
1386 	/*
1387 	 * Protect the TCP option TCP_FUNCTION_BLK so
1388 	 * that a sub-function can *never* overwrite this.
1389 	 */
1390 	if ((sopt->sopt_dir == SOPT_SET) &&
1391 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1392 		INP_WUNLOCK(inp);
1393 		error = sooptcopyin(sopt, &fsn, sizeof fsn,
1394 		    sizeof fsn);
1395 		if (error)
1396 			return (error);
1397 		INP_WLOCK_RECHECK(inp);
1398 		if (tp->t_state != TCPS_CLOSED) {
1399 			/*
1400 			 * The user has advanced the state
1401 			 * past the initial point, we can't
1402 			 * switch since we are down the road
1403 			 * and a new set of functions may
1404 			 * not be compatibile.
1405 			 */
1406 			INP_WUNLOCK(inp);
1407 			return(EINVAL);
1408 		}
1409 		blk = find_and_ref_tcp_functions(&fsn);
1410 		if (blk == NULL) {
1411 			INP_WUNLOCK(inp);
1412 			return (ENOENT);
1413 		}
1414 		if (tp->t_fb != blk) {
1415 			if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1416 				refcount_release(&blk->tfb_refcnt);
1417 				INP_WUNLOCK(inp);
1418 				return (ENOENT);
1419 			}
1420 			/*
1421 			 * Release the old refcnt, the
1422 			 * lookup acquires a ref on the
1423 			 * new one.
1424 			 */
1425 			if (tp->t_fb->tfb_tcp_fb_fini)
1426 				(*tp->t_fb->tfb_tcp_fb_fini)(tp);
1427 			refcount_release(&tp->t_fb->tfb_refcnt);
1428 			tp->t_fb = blk;
1429 			if (tp->t_fb->tfb_tcp_fb_init) {
1430 				(*tp->t_fb->tfb_tcp_fb_init)(tp);
1431 			}
1432 		}
1433 #ifdef TCP_OFFLOAD
1434 		if (tp->t_flags & TF_TOE) {
1435 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1436 			     sopt->sopt_name);
1437 		}
1438 #endif
1439 		INP_WUNLOCK(inp);
1440 		return (error);
1441 	} else if ((sopt->sopt_dir == SOPT_GET) &&
1442 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1443 		strcpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name);
1444 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1445 		INP_WUNLOCK(inp);
1446 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1447 		return (error);
1448 	}
1449 	/* Pass in the INP locked, called must unlock it */
1450 	return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1451 }
1452 
1453 int
1454 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1455 {
1456 	int	error, opt, optval;
1457 	u_int	ui;
1458 	struct	tcp_info ti;
1459 	struct cc_algo *algo;
1460 	char buf[TCP_CA_NAME_MAX];
1461 
1462 	switch (sopt->sopt_dir) {
1463 	case SOPT_SET:
1464 		switch (sopt->sopt_name) {
1465 #ifdef TCP_SIGNATURE
1466 		case TCP_MD5SIG:
1467 			INP_WUNLOCK(inp);
1468 			error = sooptcopyin(sopt, &optval, sizeof optval,
1469 			    sizeof optval);
1470 			if (error)
1471 				return (error);
1472 
1473 			INP_WLOCK_RECHECK(inp);
1474 			if (optval > 0)
1475 				tp->t_flags |= TF_SIGNATURE;
1476 			else
1477 				tp->t_flags &= ~TF_SIGNATURE;
1478 			goto unlock_and_done;
1479 #endif /* TCP_SIGNATURE */
1480 
1481 		case TCP_NODELAY:
1482 		case TCP_NOOPT:
1483 			INP_WUNLOCK(inp);
1484 			error = sooptcopyin(sopt, &optval, sizeof optval,
1485 			    sizeof optval);
1486 			if (error)
1487 				return (error);
1488 
1489 			INP_WLOCK_RECHECK(inp);
1490 			switch (sopt->sopt_name) {
1491 			case TCP_NODELAY:
1492 				opt = TF_NODELAY;
1493 				break;
1494 			case TCP_NOOPT:
1495 				opt = TF_NOOPT;
1496 				break;
1497 			default:
1498 				opt = 0; /* dead code to fool gcc */
1499 				break;
1500 			}
1501 
1502 			if (optval)
1503 				tp->t_flags |= opt;
1504 			else
1505 				tp->t_flags &= ~opt;
1506 unlock_and_done:
1507 #ifdef TCP_OFFLOAD
1508 			if (tp->t_flags & TF_TOE) {
1509 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
1510 				    sopt->sopt_name);
1511 			}
1512 #endif
1513 			INP_WUNLOCK(inp);
1514 			break;
1515 
1516 		case TCP_NOPUSH:
1517 			INP_WUNLOCK(inp);
1518 			error = sooptcopyin(sopt, &optval, sizeof optval,
1519 			    sizeof optval);
1520 			if (error)
1521 				return (error);
1522 
1523 			INP_WLOCK_RECHECK(inp);
1524 			if (optval)
1525 				tp->t_flags |= TF_NOPUSH;
1526 			else if (tp->t_flags & TF_NOPUSH) {
1527 				tp->t_flags &= ~TF_NOPUSH;
1528 				if (TCPS_HAVEESTABLISHED(tp->t_state))
1529 					error = tp->t_fb->tfb_tcp_output(tp);
1530 			}
1531 			goto unlock_and_done;
1532 
1533 		case TCP_MAXSEG:
1534 			INP_WUNLOCK(inp);
1535 			error = sooptcopyin(sopt, &optval, sizeof optval,
1536 			    sizeof optval);
1537 			if (error)
1538 				return (error);
1539 
1540 			INP_WLOCK_RECHECK(inp);
1541 			if (optval > 0 && optval <= tp->t_maxseg &&
1542 			    optval + 40 >= V_tcp_minmss)
1543 				tp->t_maxseg = optval;
1544 			else
1545 				error = EINVAL;
1546 			goto unlock_and_done;
1547 
1548 		case TCP_INFO:
1549 			INP_WUNLOCK(inp);
1550 			error = EINVAL;
1551 			break;
1552 
1553 		case TCP_CONGESTION:
1554 			INP_WUNLOCK(inp);
1555 			bzero(buf, sizeof(buf));
1556 			error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1557 			if (error)
1558 				break;
1559 			INP_WLOCK_RECHECK(inp);
1560 			/*
1561 			 * Return EINVAL if we can't find the requested cc algo.
1562 			 */
1563 			error = EINVAL;
1564 			CC_LIST_RLOCK();
1565 			STAILQ_FOREACH(algo, &cc_list, entries) {
1566 				if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1567 				    == 0) {
1568 					/* We've found the requested algo. */
1569 					error = 0;
1570 					/*
1571 					 * We hold a write lock over the tcb
1572 					 * so it's safe to do these things
1573 					 * without ordering concerns.
1574 					 */
1575 					if (CC_ALGO(tp)->cb_destroy != NULL)
1576 						CC_ALGO(tp)->cb_destroy(tp->ccv);
1577 					CC_ALGO(tp) = algo;
1578 					/*
1579 					 * If something goes pear shaped
1580 					 * initialising the new algo,
1581 					 * fall back to newreno (which
1582 					 * does not require initialisation).
1583 					 */
1584 					if (algo->cb_init != NULL)
1585 						if (algo->cb_init(tp->ccv) > 0) {
1586 							CC_ALGO(tp) = &newreno_cc_algo;
1587 							/*
1588 							 * The only reason init
1589 							 * should fail is
1590 							 * because of malloc.
1591 							 */
1592 							error = ENOMEM;
1593 						}
1594 					break; /* Break the STAILQ_FOREACH. */
1595 				}
1596 			}
1597 			CC_LIST_RUNLOCK();
1598 			goto unlock_and_done;
1599 
1600 		case TCP_KEEPIDLE:
1601 		case TCP_KEEPINTVL:
1602 		case TCP_KEEPINIT:
1603 			INP_WUNLOCK(inp);
1604 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1605 			if (error)
1606 				return (error);
1607 
1608 			if (ui > (UINT_MAX / hz)) {
1609 				error = EINVAL;
1610 				break;
1611 			}
1612 			ui *= hz;
1613 
1614 			INP_WLOCK_RECHECK(inp);
1615 			switch (sopt->sopt_name) {
1616 			case TCP_KEEPIDLE:
1617 				tp->t_keepidle = ui;
1618 				/*
1619 				 * XXX: better check current remaining
1620 				 * timeout and "merge" it with new value.
1621 				 */
1622 				if ((tp->t_state > TCPS_LISTEN) &&
1623 				    (tp->t_state <= TCPS_CLOSING))
1624 					tcp_timer_activate(tp, TT_KEEP,
1625 					    TP_KEEPIDLE(tp));
1626 				break;
1627 			case TCP_KEEPINTVL:
1628 				tp->t_keepintvl = ui;
1629 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1630 				    (TP_MAXIDLE(tp) > 0))
1631 					tcp_timer_activate(tp, TT_2MSL,
1632 					    TP_MAXIDLE(tp));
1633 				break;
1634 			case TCP_KEEPINIT:
1635 				tp->t_keepinit = ui;
1636 				if (tp->t_state == TCPS_SYN_RECEIVED ||
1637 				    tp->t_state == TCPS_SYN_SENT)
1638 					tcp_timer_activate(tp, TT_KEEP,
1639 					    TP_KEEPINIT(tp));
1640 				break;
1641 			}
1642 			goto unlock_and_done;
1643 
1644 		case TCP_KEEPCNT:
1645 			INP_WUNLOCK(inp);
1646 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1647 			if (error)
1648 				return (error);
1649 
1650 			INP_WLOCK_RECHECK(inp);
1651 			tp->t_keepcnt = ui;
1652 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1653 			    (TP_MAXIDLE(tp) > 0))
1654 				tcp_timer_activate(tp, TT_2MSL,
1655 				    TP_MAXIDLE(tp));
1656 			goto unlock_and_done;
1657 
1658 #ifdef TCPPCAP
1659 		case TCP_PCAP_OUT:
1660 		case TCP_PCAP_IN:
1661 			INP_WUNLOCK(inp);
1662 			error = sooptcopyin(sopt, &optval, sizeof optval,
1663 			    sizeof optval);
1664 			if (error)
1665 				return (error);
1666 
1667 			INP_WLOCK_RECHECK(inp);
1668 			if (optval >= 0)
1669 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
1670 					&(tp->t_outpkts) : &(tp->t_inpkts),
1671 					optval);
1672 			else
1673 				error = EINVAL;
1674 			goto unlock_and_done;
1675 #endif
1676 
1677 		default:
1678 			INP_WUNLOCK(inp);
1679 			error = ENOPROTOOPT;
1680 			break;
1681 		}
1682 		break;
1683 
1684 	case SOPT_GET:
1685 		tp = intotcpcb(inp);
1686 		switch (sopt->sopt_name) {
1687 #ifdef TCP_SIGNATURE
1688 		case TCP_MD5SIG:
1689 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1690 			INP_WUNLOCK(inp);
1691 			error = sooptcopyout(sopt, &optval, sizeof optval);
1692 			break;
1693 #endif
1694 
1695 		case TCP_NODELAY:
1696 			optval = tp->t_flags & TF_NODELAY;
1697 			INP_WUNLOCK(inp);
1698 			error = sooptcopyout(sopt, &optval, sizeof optval);
1699 			break;
1700 		case TCP_MAXSEG:
1701 			optval = tp->t_maxseg;
1702 			INP_WUNLOCK(inp);
1703 			error = sooptcopyout(sopt, &optval, sizeof optval);
1704 			break;
1705 		case TCP_NOOPT:
1706 			optval = tp->t_flags & TF_NOOPT;
1707 			INP_WUNLOCK(inp);
1708 			error = sooptcopyout(sopt, &optval, sizeof optval);
1709 			break;
1710 		case TCP_NOPUSH:
1711 			optval = tp->t_flags & TF_NOPUSH;
1712 			INP_WUNLOCK(inp);
1713 			error = sooptcopyout(sopt, &optval, sizeof optval);
1714 			break;
1715 		case TCP_INFO:
1716 			tcp_fill_info(tp, &ti);
1717 			INP_WUNLOCK(inp);
1718 			error = sooptcopyout(sopt, &ti, sizeof ti);
1719 			break;
1720 		case TCP_CONGESTION:
1721 			bzero(buf, sizeof(buf));
1722 			strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1723 			INP_WUNLOCK(inp);
1724 			error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1725 			break;
1726 		case TCP_KEEPIDLE:
1727 		case TCP_KEEPINTVL:
1728 		case TCP_KEEPINIT:
1729 		case TCP_KEEPCNT:
1730 			switch (sopt->sopt_name) {
1731 			case TCP_KEEPIDLE:
1732 				ui = tp->t_keepidle / hz;
1733 				break;
1734 			case TCP_KEEPINTVL:
1735 				ui = tp->t_keepintvl / hz;
1736 				break;
1737 			case TCP_KEEPINIT:
1738 				ui = tp->t_keepinit / hz;
1739 				break;
1740 			case TCP_KEEPCNT:
1741 				ui = tp->t_keepcnt;
1742 				break;
1743 			}
1744 			INP_WUNLOCK(inp);
1745 			error = sooptcopyout(sopt, &ui, sizeof(ui));
1746 			break;
1747 #ifdef TCPPCAP
1748 		case TCP_PCAP_OUT:
1749 		case TCP_PCAP_IN:
1750 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
1751 					&(tp->t_outpkts) : &(tp->t_inpkts));
1752 			INP_WUNLOCK(inp);
1753 			error = sooptcopyout(sopt, &optval, sizeof optval);
1754 			break;
1755 #endif
1756 		default:
1757 			INP_WUNLOCK(inp);
1758 			error = ENOPROTOOPT;
1759 			break;
1760 		}
1761 		break;
1762 	}
1763 	return (error);
1764 }
1765 #undef INP_WLOCK_RECHECK
1766 
1767 /*
1768  * Attach TCP protocol to socket, allocating
1769  * internet protocol control block, tcp control block,
1770  * bufer space, and entering LISTEN state if to accept connections.
1771  */
1772 static int
1773 tcp_attach(struct socket *so)
1774 {
1775 	struct tcpcb *tp;
1776 	struct inpcb *inp;
1777 	int error;
1778 
1779 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1780 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1781 		if (error)
1782 			return (error);
1783 	}
1784 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
1785 	so->so_snd.sb_flags |= SB_AUTOSIZE;
1786 	INP_INFO_RLOCK(&V_tcbinfo);
1787 	error = in_pcballoc(so, &V_tcbinfo);
1788 	if (error) {
1789 		INP_INFO_RUNLOCK(&V_tcbinfo);
1790 		return (error);
1791 	}
1792 	inp = sotoinpcb(so);
1793 #ifdef INET6
1794 	if (inp->inp_vflag & INP_IPV6PROTO) {
1795 		inp->inp_vflag |= INP_IPV6;
1796 		inp->in6p_hops = -1;	/* use kernel default */
1797 	}
1798 	else
1799 #endif
1800 	inp->inp_vflag |= INP_IPV4;
1801 	tp = tcp_newtcpcb(inp);
1802 	if (tp == NULL) {
1803 		in_pcbdetach(inp);
1804 		in_pcbfree(inp);
1805 		INP_INFO_RUNLOCK(&V_tcbinfo);
1806 		return (ENOBUFS);
1807 	}
1808 	tp->t_state = TCPS_CLOSED;
1809 	INP_WUNLOCK(inp);
1810 	INP_INFO_RUNLOCK(&V_tcbinfo);
1811 	return (0);
1812 }
1813 
1814 /*
1815  * Initiate (or continue) disconnect.
1816  * If embryonic state, just send reset (once).
1817  * If in ``let data drain'' option and linger null, just drop.
1818  * Otherwise (hard), mark socket disconnecting and drop
1819  * current input data; switch states based on user close, and
1820  * send segment to peer (with FIN).
1821  */
1822 static void
1823 tcp_disconnect(struct tcpcb *tp)
1824 {
1825 	struct inpcb *inp = tp->t_inpcb;
1826 	struct socket *so = inp->inp_socket;
1827 
1828 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1829 	INP_WLOCK_ASSERT(inp);
1830 
1831 	/*
1832 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1833 	 * socket is still open.
1834 	 */
1835 	if (tp->t_state < TCPS_ESTABLISHED) {
1836 		tp = tcp_close(tp);
1837 		KASSERT(tp != NULL,
1838 		    ("tcp_disconnect: tcp_close() returned NULL"));
1839 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1840 		tp = tcp_drop(tp, 0);
1841 		KASSERT(tp != NULL,
1842 		    ("tcp_disconnect: tcp_drop() returned NULL"));
1843 	} else {
1844 		soisdisconnecting(so);
1845 		sbflush(&so->so_rcv);
1846 		tcp_usrclosed(tp);
1847 		if (!(inp->inp_flags & INP_DROPPED))
1848 			tp->t_fb->tfb_tcp_output(tp);
1849 	}
1850 }
1851 
1852 /*
1853  * User issued close, and wish to trail through shutdown states:
1854  * if never received SYN, just forget it.  If got a SYN from peer,
1855  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1856  * If already got a FIN from peer, then almost done; go to LAST_ACK
1857  * state.  In all other cases, have already sent FIN to peer (e.g.
1858  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1859  * for peer to send FIN or not respond to keep-alives, etc.
1860  * We can let the user exit from the close as soon as the FIN is acked.
1861  */
1862 static void
1863 tcp_usrclosed(struct tcpcb *tp)
1864 {
1865 
1866 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1867 	INP_WLOCK_ASSERT(tp->t_inpcb);
1868 
1869 	switch (tp->t_state) {
1870 	case TCPS_LISTEN:
1871 #ifdef TCP_OFFLOAD
1872 		tcp_offload_listen_stop(tp);
1873 #endif
1874 		tcp_state_change(tp, TCPS_CLOSED);
1875 		/* FALLTHROUGH */
1876 	case TCPS_CLOSED:
1877 		tp = tcp_close(tp);
1878 		/*
1879 		 * tcp_close() should never return NULL here as the socket is
1880 		 * still open.
1881 		 */
1882 		KASSERT(tp != NULL,
1883 		    ("tcp_usrclosed: tcp_close() returned NULL"));
1884 		break;
1885 
1886 	case TCPS_SYN_SENT:
1887 	case TCPS_SYN_RECEIVED:
1888 		tp->t_flags |= TF_NEEDFIN;
1889 		break;
1890 
1891 	case TCPS_ESTABLISHED:
1892 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
1893 		break;
1894 
1895 	case TCPS_CLOSE_WAIT:
1896 		tcp_state_change(tp, TCPS_LAST_ACK);
1897 		break;
1898 	}
1899 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
1900 		soisdisconnected(tp->t_inpcb->inp_socket);
1901 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
1902 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1903 			int timeout;
1904 
1905 			timeout = (tcp_fast_finwait2_recycle) ?
1906 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
1907 			tcp_timer_activate(tp, TT_2MSL, timeout);
1908 		}
1909 	}
1910 }
1911 
1912 #ifdef DDB
1913 static void
1914 db_print_indent(int indent)
1915 {
1916 	int i;
1917 
1918 	for (i = 0; i < indent; i++)
1919 		db_printf(" ");
1920 }
1921 
1922 static void
1923 db_print_tstate(int t_state)
1924 {
1925 
1926 	switch (t_state) {
1927 	case TCPS_CLOSED:
1928 		db_printf("TCPS_CLOSED");
1929 		return;
1930 
1931 	case TCPS_LISTEN:
1932 		db_printf("TCPS_LISTEN");
1933 		return;
1934 
1935 	case TCPS_SYN_SENT:
1936 		db_printf("TCPS_SYN_SENT");
1937 		return;
1938 
1939 	case TCPS_SYN_RECEIVED:
1940 		db_printf("TCPS_SYN_RECEIVED");
1941 		return;
1942 
1943 	case TCPS_ESTABLISHED:
1944 		db_printf("TCPS_ESTABLISHED");
1945 		return;
1946 
1947 	case TCPS_CLOSE_WAIT:
1948 		db_printf("TCPS_CLOSE_WAIT");
1949 		return;
1950 
1951 	case TCPS_FIN_WAIT_1:
1952 		db_printf("TCPS_FIN_WAIT_1");
1953 		return;
1954 
1955 	case TCPS_CLOSING:
1956 		db_printf("TCPS_CLOSING");
1957 		return;
1958 
1959 	case TCPS_LAST_ACK:
1960 		db_printf("TCPS_LAST_ACK");
1961 		return;
1962 
1963 	case TCPS_FIN_WAIT_2:
1964 		db_printf("TCPS_FIN_WAIT_2");
1965 		return;
1966 
1967 	case TCPS_TIME_WAIT:
1968 		db_printf("TCPS_TIME_WAIT");
1969 		return;
1970 
1971 	default:
1972 		db_printf("unknown");
1973 		return;
1974 	}
1975 }
1976 
1977 static void
1978 db_print_tflags(u_int t_flags)
1979 {
1980 	int comma;
1981 
1982 	comma = 0;
1983 	if (t_flags & TF_ACKNOW) {
1984 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
1985 		comma = 1;
1986 	}
1987 	if (t_flags & TF_DELACK) {
1988 		db_printf("%sTF_DELACK", comma ? ", " : "");
1989 		comma = 1;
1990 	}
1991 	if (t_flags & TF_NODELAY) {
1992 		db_printf("%sTF_NODELAY", comma ? ", " : "");
1993 		comma = 1;
1994 	}
1995 	if (t_flags & TF_NOOPT) {
1996 		db_printf("%sTF_NOOPT", comma ? ", " : "");
1997 		comma = 1;
1998 	}
1999 	if (t_flags & TF_SENTFIN) {
2000 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2001 		comma = 1;
2002 	}
2003 	if (t_flags & TF_REQ_SCALE) {
2004 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2005 		comma = 1;
2006 	}
2007 	if (t_flags & TF_RCVD_SCALE) {
2008 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2009 		comma = 1;
2010 	}
2011 	if (t_flags & TF_REQ_TSTMP) {
2012 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2013 		comma = 1;
2014 	}
2015 	if (t_flags & TF_RCVD_TSTMP) {
2016 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2017 		comma = 1;
2018 	}
2019 	if (t_flags & TF_SACK_PERMIT) {
2020 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2021 		comma = 1;
2022 	}
2023 	if (t_flags & TF_NEEDSYN) {
2024 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2025 		comma = 1;
2026 	}
2027 	if (t_flags & TF_NEEDFIN) {
2028 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2029 		comma = 1;
2030 	}
2031 	if (t_flags & TF_NOPUSH) {
2032 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2033 		comma = 1;
2034 	}
2035 	if (t_flags & TF_MORETOCOME) {
2036 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2037 		comma = 1;
2038 	}
2039 	if (t_flags & TF_LQ_OVERFLOW) {
2040 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2041 		comma = 1;
2042 	}
2043 	if (t_flags & TF_LASTIDLE) {
2044 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2045 		comma = 1;
2046 	}
2047 	if (t_flags & TF_RXWIN0SENT) {
2048 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2049 		comma = 1;
2050 	}
2051 	if (t_flags & TF_FASTRECOVERY) {
2052 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2053 		comma = 1;
2054 	}
2055 	if (t_flags & TF_CONGRECOVERY) {
2056 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2057 		comma = 1;
2058 	}
2059 	if (t_flags & TF_WASFRECOVERY) {
2060 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2061 		comma = 1;
2062 	}
2063 	if (t_flags & TF_SIGNATURE) {
2064 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2065 		comma = 1;
2066 	}
2067 	if (t_flags & TF_FORCEDATA) {
2068 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2069 		comma = 1;
2070 	}
2071 	if (t_flags & TF_TSO) {
2072 		db_printf("%sTF_TSO", comma ? ", " : "");
2073 		comma = 1;
2074 	}
2075 	if (t_flags & TF_ECN_PERMIT) {
2076 		db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
2077 		comma = 1;
2078 	}
2079 }
2080 
2081 static void
2082 db_print_toobflags(char t_oobflags)
2083 {
2084 	int comma;
2085 
2086 	comma = 0;
2087 	if (t_oobflags & TCPOOB_HAVEDATA) {
2088 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2089 		comma = 1;
2090 	}
2091 	if (t_oobflags & TCPOOB_HADDATA) {
2092 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2093 		comma = 1;
2094 	}
2095 }
2096 
2097 static void
2098 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2099 {
2100 
2101 	db_print_indent(indent);
2102 	db_printf("%s at %p\n", name, tp);
2103 
2104 	indent += 2;
2105 
2106 	db_print_indent(indent);
2107 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2108 	   LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2109 
2110 	db_print_indent(indent);
2111 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2112 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2113 
2114 	db_print_indent(indent);
2115 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2116 	    &tp->t_timers->tt_delack, tp->t_inpcb);
2117 
2118 	db_print_indent(indent);
2119 	db_printf("t_state: %d (", tp->t_state);
2120 	db_print_tstate(tp->t_state);
2121 	db_printf(")\n");
2122 
2123 	db_print_indent(indent);
2124 	db_printf("t_flags: 0x%x (", tp->t_flags);
2125 	db_print_tflags(tp->t_flags);
2126 	db_printf(")\n");
2127 
2128 	db_print_indent(indent);
2129 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2130 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2131 
2132 	db_print_indent(indent);
2133 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2134 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2135 
2136 	db_print_indent(indent);
2137 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2138 	    tp->iss, tp->irs, tp->rcv_nxt);
2139 
2140 	db_print_indent(indent);
2141 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %lu   rcv_up: 0x%08x\n",
2142 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2143 
2144 	db_print_indent(indent);
2145 	db_printf("snd_wnd: %lu   snd_cwnd: %lu\n",
2146 	   tp->snd_wnd, tp->snd_cwnd);
2147 
2148 	db_print_indent(indent);
2149 	db_printf("snd_ssthresh: %lu   snd_recover: "
2150 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2151 
2152 	db_print_indent(indent);
2153 	db_printf("t_maxopd: %u   t_rcvtime: %u   t_startime: %u\n",
2154 	    tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
2155 
2156 	db_print_indent(indent);
2157 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2158 	    tp->t_rtttime, tp->t_rtseq);
2159 
2160 	db_print_indent(indent);
2161 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2162 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2163 
2164 	db_print_indent(indent);
2165 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2166 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2167 	    tp->t_rttbest);
2168 
2169 	db_print_indent(indent);
2170 	db_printf("t_rttupdated: %lu   max_sndwnd: %lu   t_softerror: %d\n",
2171 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2172 
2173 	db_print_indent(indent);
2174 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2175 	db_print_toobflags(tp->t_oobflags);
2176 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2177 
2178 	db_print_indent(indent);
2179 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2180 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2181 
2182 	db_print_indent(indent);
2183 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2184 	    tp->ts_recent, tp->ts_recent_age);
2185 
2186 	db_print_indent(indent);
2187 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2188 	    "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2189 
2190 	db_print_indent(indent);
2191 	db_printf("snd_ssthresh_prev: %lu   snd_recover_prev: 0x%08x   "
2192 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2193 	    tp->snd_recover_prev, tp->t_badrxtwin);
2194 
2195 	db_print_indent(indent);
2196 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2197 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2198 
2199 	db_print_indent(indent);
2200 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2201 	    "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2202 
2203 	/* Skip sackblks, sackhint. */
2204 
2205 	db_print_indent(indent);
2206 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2207 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2208 }
2209 
2210 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2211 {
2212 	struct tcpcb *tp;
2213 
2214 	if (!have_addr) {
2215 		db_printf("usage: show tcpcb <addr>\n");
2216 		return;
2217 	}
2218 	tp = (struct tcpcb *)addr;
2219 
2220 	db_print_tcpcb(tp, "tcpcb", 0);
2221 }
2222 #endif
2223