xref: /freebsd/sys/netinet/tcp_usrreq.c (revision 312809fe7fefbc8d5caa2b59089a5d9266378057)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2006-2007 Robert N. M. Watson
7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
8  * All rights reserved.
9  *
10  * Portions of this software were developed by Robert N. M. Watson under
11  * contract to Juniper Networks, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_ddb.h"
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_kern_tls.h"
48 #include "opt_tcpdebug.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/arb.h>
53 #include <sys/limits.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
56 #include <sys/kernel.h>
57 #include <sys/ktls.h>
58 #include <sys/qmath.h>
59 #include <sys/sysctl.h>
60 #include <sys/mbuf.h>
61 #ifdef INET6
62 #include <sys/domain.h>
63 #endif /* INET6 */
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/proc.h>
68 #include <sys/jail.h>
69 #include <sys/syslog.h>
70 #include <sys/stats.h>
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #include <net/if.h>
77 #include <net/if_var.h>
78 #include <net/route.h>
79 #include <net/vnet.h>
80 
81 #include <netinet/in.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip_var.h>
87 #ifdef INET6
88 #include <netinet/ip6.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #endif
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_log_buf.h>
99 #include <netinet/tcpip.h>
100 #include <netinet/cc/cc.h>
101 #include <netinet/tcp_fastopen.h>
102 #include <netinet/tcp_hpts.h>
103 #ifdef TCPPCAP
104 #include <netinet/tcp_pcap.h>
105 #endif
106 #ifdef TCPDEBUG
107 #include <netinet/tcp_debug.h>
108 #endif
109 #ifdef TCP_OFFLOAD
110 #include <netinet/tcp_offload.h>
111 #endif
112 #include <netipsec/ipsec_support.h>
113 
114 #include <vm/vm.h>
115 #include <vm/vm_param.h>
116 #include <vm/pmap.h>
117 #include <vm/vm_extern.h>
118 #include <vm/vm_map.h>
119 #include <vm/vm_page.h>
120 
121 /*
122  * TCP protocol interface to socket abstraction.
123  */
124 static int	tcp_attach(struct socket *);
125 #ifdef INET
126 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
127 		    struct thread *td);
128 #endif /* INET */
129 #ifdef INET6
130 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
131 		    struct thread *td);
132 #endif /* INET6 */
133 static void	tcp_disconnect(struct tcpcb *);
134 static void	tcp_usrclosed(struct tcpcb *);
135 static void	tcp_fill_info(struct tcpcb *, struct tcp_info *);
136 
137 #ifdef TCPDEBUG
138 #define	TCPDEBUG0	int ostate = 0
139 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
140 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
141 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
142 #else
143 #define	TCPDEBUG0
144 #define	TCPDEBUG1()
145 #define	TCPDEBUG2(req)
146 #endif
147 
148 /*
149  * TCP attaches to socket via pru_attach(), reserving space,
150  * and an internet control block.
151  */
152 static int
153 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
154 {
155 	struct inpcb *inp;
156 	struct tcpcb *tp = NULL;
157 	int error;
158 	TCPDEBUG0;
159 
160 	inp = sotoinpcb(so);
161 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
162 	TCPDEBUG1();
163 
164 	error = tcp_attach(so);
165 	if (error)
166 		goto out;
167 
168 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
169 		so->so_linger = TCP_LINGERTIME;
170 
171 	inp = sotoinpcb(so);
172 	tp = intotcpcb(inp);
173 out:
174 	TCPDEBUG2(PRU_ATTACH);
175 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
176 	return error;
177 }
178 
179 /*
180  * tcp_detach is called when the socket layer loses its final reference
181  * to the socket, be it a file descriptor reference, a reference from TCP,
182  * etc.  At this point, there is only one case in which we will keep around
183  * inpcb state: time wait.
184  *
185  * This function can probably be re-absorbed back into tcp_usr_detach() now
186  * that there is a single detach path.
187  */
188 static void
189 tcp_detach(struct socket *so, struct inpcb *inp)
190 {
191 	struct tcpcb *tp;
192 
193 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
194 	INP_WLOCK_ASSERT(inp);
195 
196 	KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
197 	KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
198 
199 	tp = intotcpcb(inp);
200 
201 	if (inp->inp_flags & INP_TIMEWAIT) {
202 		/*
203 		 * There are two cases to handle: one in which the time wait
204 		 * state is being discarded (INP_DROPPED), and one in which
205 		 * this connection will remain in timewait.  In the former,
206 		 * it is time to discard all state (except tcptw, which has
207 		 * already been discarded by the timewait close code, which
208 		 * should be further up the call stack somewhere).  In the
209 		 * latter case, we detach from the socket, but leave the pcb
210 		 * present until timewait ends.
211 		 *
212 		 * XXXRW: Would it be cleaner to free the tcptw here?
213 		 *
214 		 * Astute question indeed, from twtcp perspective there are
215 		 * four cases to consider:
216 		 *
217 		 * #1 tcp_detach is called at tcptw creation time by
218 		 *  tcp_twstart, then do not discard the newly created tcptw
219 		 *  and leave inpcb present until timewait ends
220 		 * #2 tcp_detach is called at tcptw creation time by
221 		 *  tcp_twstart, but connection is local and tw will be
222 		 *  discarded immediately
223 		 * #3 tcp_detach is called at timewait end (or reuse) by
224 		 *  tcp_twclose, then the tcptw has already been discarded
225 		 *  (or reused) and inpcb is freed here
226 		 * #4 tcp_detach is called() after timewait ends (or reuse)
227 		 *  (e.g. by soclose), then tcptw has already been discarded
228 		 *  (or reused) and inpcb is freed here
229 		 *
230 		 *  In all three cases the tcptw should not be freed here.
231 		 */
232 		if (inp->inp_flags & INP_DROPPED) {
233 			in_pcbdetach(inp);
234 			if (__predict_true(tp == NULL)) {
235 				in_pcbfree(inp);
236 			} else {
237 				/*
238 				 * This case should not happen as in TIMEWAIT
239 				 * state the inp should not be destroyed before
240 				 * its tcptw.  If INVARIANTS is defined, panic.
241 				 */
242 #ifdef INVARIANTS
243 				panic("%s: Panic before an inp double-free: "
244 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
245 				    , __func__);
246 #else
247 				log(LOG_ERR, "%s: Avoid an inp double-free: "
248 				    "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
249 				    , __func__);
250 #endif
251 				INP_WUNLOCK(inp);
252 			}
253 		} else {
254 			in_pcbdetach(inp);
255 			INP_WUNLOCK(inp);
256 		}
257 	} else {
258 		/*
259 		 * If the connection is not in timewait, we consider two
260 		 * two conditions: one in which no further processing is
261 		 * necessary (dropped || embryonic), and one in which TCP is
262 		 * not yet done, but no longer requires the socket, so the
263 		 * pcb will persist for the time being.
264 		 *
265 		 * XXXRW: Does the second case still occur?
266 		 */
267 		if (inp->inp_flags & INP_DROPPED ||
268 		    tp->t_state < TCPS_SYN_SENT) {
269 			tcp_discardcb(tp);
270 			in_pcbdetach(inp);
271 			in_pcbfree(inp);
272 		} else {
273 			in_pcbdetach(inp);
274 			INP_WUNLOCK(inp);
275 		}
276 	}
277 }
278 
279 /*
280  * pru_detach() detaches the TCP protocol from the socket.
281  * If the protocol state is non-embryonic, then can't
282  * do this directly: have to initiate a pru_disconnect(),
283  * which may finish later; embryonic TCB's can just
284  * be discarded here.
285  */
286 static void
287 tcp_usr_detach(struct socket *so)
288 {
289 	struct inpcb *inp;
290 	int rlock = 0;
291 	struct epoch_tracker et;
292 
293 	inp = sotoinpcb(so);
294 	KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
295 	if (!INP_INFO_WLOCKED(&V_tcbinfo)) {
296 		NET_EPOCH_ENTER(et);
297 		rlock = 1;
298 	}
299 	INP_WLOCK(inp);
300 	KASSERT(inp->inp_socket != NULL,
301 	    ("tcp_usr_detach: inp_socket == NULL"));
302 	tcp_detach(so, inp);
303 	if (rlock)
304 		NET_EPOCH_EXIT(et);
305 }
306 
307 #ifdef INET
308 /*
309  * Give the socket an address.
310  */
311 static int
312 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
313 {
314 	int error = 0;
315 	struct inpcb *inp;
316 	struct tcpcb *tp = NULL;
317 	struct sockaddr_in *sinp;
318 
319 	sinp = (struct sockaddr_in *)nam;
320 	if (nam->sa_len != sizeof (*sinp))
321 		return (EINVAL);
322 	/*
323 	 * Must check for multicast addresses and disallow binding
324 	 * to them.
325 	 */
326 	if (sinp->sin_family == AF_INET &&
327 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
328 		return (EAFNOSUPPORT);
329 
330 	TCPDEBUG0;
331 	inp = sotoinpcb(so);
332 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
333 	INP_WLOCK(inp);
334 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
335 		error = EINVAL;
336 		goto out;
337 	}
338 	tp = intotcpcb(inp);
339 	TCPDEBUG1();
340 	INP_HASH_WLOCK(&V_tcbinfo);
341 	error = in_pcbbind(inp, nam, td->td_ucred);
342 	INP_HASH_WUNLOCK(&V_tcbinfo);
343 out:
344 	TCPDEBUG2(PRU_BIND);
345 	TCP_PROBE2(debug__user, tp, PRU_BIND);
346 	INP_WUNLOCK(inp);
347 
348 	return (error);
349 }
350 #endif /* INET */
351 
352 #ifdef INET6
353 static int
354 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
355 {
356 	int error = 0;
357 	struct inpcb *inp;
358 	struct tcpcb *tp = NULL;
359 	struct sockaddr_in6 *sin6;
360 	u_char vflagsav;
361 
362 	sin6 = (struct sockaddr_in6 *)nam;
363 	if (nam->sa_len != sizeof (*sin6))
364 		return (EINVAL);
365 	/*
366 	 * Must check for multicast addresses and disallow binding
367 	 * to them.
368 	 */
369 	if (sin6->sin6_family == AF_INET6 &&
370 	    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
371 		return (EAFNOSUPPORT);
372 
373 	TCPDEBUG0;
374 	inp = sotoinpcb(so);
375 	KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
376 	INP_WLOCK(inp);
377 	vflagsav = inp->inp_vflag;
378 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
379 		error = EINVAL;
380 		goto out;
381 	}
382 	tp = intotcpcb(inp);
383 	TCPDEBUG1();
384 	INP_HASH_WLOCK(&V_tcbinfo);
385 	inp->inp_vflag &= ~INP_IPV4;
386 	inp->inp_vflag |= INP_IPV6;
387 #ifdef INET
388 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
389 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
390 			inp->inp_vflag |= INP_IPV4;
391 		else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
392 			struct sockaddr_in sin;
393 
394 			in6_sin6_2_sin(&sin, sin6);
395 			if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
396 				error = EAFNOSUPPORT;
397 				INP_HASH_WUNLOCK(&V_tcbinfo);
398 				goto out;
399 			}
400 			inp->inp_vflag |= INP_IPV4;
401 			inp->inp_vflag &= ~INP_IPV6;
402 			error = in_pcbbind(inp, (struct sockaddr *)&sin,
403 			    td->td_ucred);
404 			INP_HASH_WUNLOCK(&V_tcbinfo);
405 			goto out;
406 		}
407 	}
408 #endif
409 	error = in6_pcbbind(inp, nam, td->td_ucred);
410 	INP_HASH_WUNLOCK(&V_tcbinfo);
411 out:
412 	if (error != 0)
413 		inp->inp_vflag = vflagsav;
414 	TCPDEBUG2(PRU_BIND);
415 	TCP_PROBE2(debug__user, tp, PRU_BIND);
416 	INP_WUNLOCK(inp);
417 	return (error);
418 }
419 #endif /* INET6 */
420 
421 #ifdef INET
422 /*
423  * Prepare to accept connections.
424  */
425 static int
426 tcp_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, ("tcp_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 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
447 	INP_HASH_WUNLOCK(&V_tcbinfo);
448 	if (error == 0) {
449 		tcp_state_change(tp, TCPS_LISTEN);
450 		solisten_proto(so, backlog);
451 #ifdef TCP_OFFLOAD
452 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
453 			tcp_offload_listen_start(tp);
454 #endif
455 	}
456 	SOCK_UNLOCK(so);
457 
458 	if (IS_FASTOPEN(tp->t_flags))
459 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
460 
461 out:
462 	TCPDEBUG2(PRU_LISTEN);
463 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
464 	INP_WUNLOCK(inp);
465 	return (error);
466 }
467 #endif /* INET */
468 
469 #ifdef INET6
470 static int
471 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
472 {
473 	int error = 0;
474 	struct inpcb *inp;
475 	struct tcpcb *tp = NULL;
476 	u_char vflagsav;
477 
478 	TCPDEBUG0;
479 	inp = sotoinpcb(so);
480 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
481 	INP_WLOCK(inp);
482 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
483 		error = EINVAL;
484 		goto out;
485 	}
486 	vflagsav = inp->inp_vflag;
487 	tp = intotcpcb(inp);
488 	TCPDEBUG1();
489 	SOCK_LOCK(so);
490 	error = solisten_proto_check(so);
491 	INP_HASH_WLOCK(&V_tcbinfo);
492 	if (error == 0 && inp->inp_lport == 0) {
493 		inp->inp_vflag &= ~INP_IPV4;
494 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
495 			inp->inp_vflag |= INP_IPV4;
496 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
497 	}
498 	INP_HASH_WUNLOCK(&V_tcbinfo);
499 	if (error == 0) {
500 		tcp_state_change(tp, TCPS_LISTEN);
501 		solisten_proto(so, backlog);
502 #ifdef TCP_OFFLOAD
503 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
504 			tcp_offload_listen_start(tp);
505 #endif
506 	}
507 	SOCK_UNLOCK(so);
508 
509 	if (IS_FASTOPEN(tp->t_flags))
510 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
511 
512 	if (error != 0)
513 		inp->inp_vflag = vflagsav;
514 
515 out:
516 	TCPDEBUG2(PRU_LISTEN);
517 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
518 	INP_WUNLOCK(inp);
519 	return (error);
520 }
521 #endif /* INET6 */
522 
523 #ifdef INET
524 /*
525  * Initiate connection to peer.
526  * Create a template for use in transmissions on this connection.
527  * Enter SYN_SENT state, and mark socket as connecting.
528  * Start keep-alive timer, and seed output sequence space.
529  * Send initial segment on connection.
530  */
531 static int
532 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
533 {
534 	int error = 0;
535 	struct inpcb *inp;
536 	struct tcpcb *tp = NULL;
537 	struct sockaddr_in *sinp;
538 
539 	sinp = (struct sockaddr_in *)nam;
540 	if (nam->sa_len != sizeof (*sinp))
541 		return (EINVAL);
542 	/*
543 	 * Must disallow TCP ``connections'' to multicast addresses.
544 	 */
545 	if (sinp->sin_family == AF_INET
546 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
547 		return (EAFNOSUPPORT);
548 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
549 		return (error);
550 
551 	TCPDEBUG0;
552 	inp = sotoinpcb(so);
553 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
554 	INP_WLOCK(inp);
555 	if (inp->inp_flags & INP_TIMEWAIT) {
556 		error = EADDRINUSE;
557 		goto out;
558 	}
559 	if (inp->inp_flags & INP_DROPPED) {
560 		error = ECONNREFUSED;
561 		goto out;
562 	}
563 	tp = intotcpcb(inp);
564 	TCPDEBUG1();
565 	if ((error = tcp_connect(tp, nam, td)) != 0)
566 		goto out;
567 #ifdef TCP_OFFLOAD
568 	if (registered_toedevs > 0 &&
569 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
570 	    (error = tcp_offload_connect(so, nam)) == 0)
571 		goto out;
572 #endif
573 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
574 	error = tp->t_fb->tfb_tcp_output(tp);
575 out:
576 	TCPDEBUG2(PRU_CONNECT);
577 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
578 	INP_WUNLOCK(inp);
579 	return (error);
580 }
581 #endif /* INET */
582 
583 #ifdef INET6
584 static int
585 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
586 {
587 	int error = 0;
588 	struct inpcb *inp;
589 	struct tcpcb *tp = NULL;
590 	struct sockaddr_in6 *sin6;
591 	u_int8_t incflagsav;
592 	u_char vflagsav;
593 
594 	TCPDEBUG0;
595 
596 	sin6 = (struct sockaddr_in6 *)nam;
597 	if (nam->sa_len != sizeof (*sin6))
598 		return (EINVAL);
599 	/*
600 	 * Must disallow TCP ``connections'' to multicast addresses.
601 	 */
602 	if (sin6->sin6_family == AF_INET6
603 	    && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
604 		return (EAFNOSUPPORT);
605 
606 	inp = sotoinpcb(so);
607 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
608 	INP_WLOCK(inp);
609 	vflagsav = inp->inp_vflag;
610 	incflagsav = inp->inp_inc.inc_flags;
611 	if (inp->inp_flags & INP_TIMEWAIT) {
612 		error = EADDRINUSE;
613 		goto out;
614 	}
615 	if (inp->inp_flags & INP_DROPPED) {
616 		error = ECONNREFUSED;
617 		goto out;
618 	}
619 	tp = intotcpcb(inp);
620 	TCPDEBUG1();
621 #ifdef INET
622 	/*
623 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
624 	 * therefore probably require the hash lock, which isn't held here.
625 	 * Is this a significant problem?
626 	 */
627 	if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
628 		struct sockaddr_in sin;
629 
630 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
631 			error = EINVAL;
632 			goto out;
633 		}
634 		if ((inp->inp_vflag & INP_IPV4) == 0) {
635 			error = EAFNOSUPPORT;
636 			goto out;
637 		}
638 
639 		in6_sin6_2_sin(&sin, sin6);
640 		if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
641 			error = EAFNOSUPPORT;
642 			goto out;
643 		}
644 		if ((error = prison_remote_ip4(td->td_ucred,
645 		    &sin.sin_addr)) != 0)
646 			goto out;
647 		inp->inp_vflag |= INP_IPV4;
648 		inp->inp_vflag &= ~INP_IPV6;
649 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
650 			goto out;
651 #ifdef TCP_OFFLOAD
652 		if (registered_toedevs > 0 &&
653 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
654 		    (error = tcp_offload_connect(so, nam)) == 0)
655 			goto out;
656 #endif
657 		error = tp->t_fb->tfb_tcp_output(tp);
658 		goto out;
659 	} else {
660 		if ((inp->inp_vflag & INP_IPV6) == 0) {
661 			error = EAFNOSUPPORT;
662 			goto out;
663 		}
664 	}
665 #endif
666 	if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
667 		goto out;
668 	inp->inp_vflag &= ~INP_IPV4;
669 	inp->inp_vflag |= INP_IPV6;
670 	inp->inp_inc.inc_flags |= INC_ISIPV6;
671 	if ((error = tcp6_connect(tp, nam, td)) != 0)
672 		goto out;
673 #ifdef TCP_OFFLOAD
674 	if (registered_toedevs > 0 &&
675 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
676 	    (error = tcp_offload_connect(so, nam)) == 0)
677 		goto out;
678 #endif
679 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
680 	error = tp->t_fb->tfb_tcp_output(tp);
681 
682 out:
683 	/*
684 	 * If the implicit bind in the connect call fails, restore
685 	 * the flags we modified.
686 	 */
687 	if (error != 0 && inp->inp_lport == 0) {
688 		inp->inp_vflag = vflagsav;
689 		inp->inp_inc.inc_flags = incflagsav;
690 	}
691 
692 	TCPDEBUG2(PRU_CONNECT);
693 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
694 	INP_WUNLOCK(inp);
695 	return (error);
696 }
697 #endif /* INET6 */
698 
699 /*
700  * Initiate disconnect from peer.
701  * If connection never passed embryonic stage, just drop;
702  * else if don't need to let data drain, then can just drop anyways,
703  * else have to begin TCP shutdown process: mark socket disconnecting,
704  * drain unread data, state switch to reflect user close, and
705  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
706  * when peer sends FIN and acks ours.
707  *
708  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
709  */
710 static int
711 tcp_usr_disconnect(struct socket *so)
712 {
713 	struct inpcb *inp;
714 	struct tcpcb *tp = NULL;
715 	struct epoch_tracker et;
716 	int error = 0;
717 
718 	TCPDEBUG0;
719 	NET_EPOCH_ENTER(et);
720 	inp = sotoinpcb(so);
721 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
722 	INP_WLOCK(inp);
723 	if (inp->inp_flags & INP_TIMEWAIT)
724 		goto out;
725 	if (inp->inp_flags & INP_DROPPED) {
726 		error = ECONNRESET;
727 		goto out;
728 	}
729 	tp = intotcpcb(inp);
730 	TCPDEBUG1();
731 	tcp_disconnect(tp);
732 out:
733 	TCPDEBUG2(PRU_DISCONNECT);
734 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
735 	INP_WUNLOCK(inp);
736 	NET_EPOCH_EXIT(et);
737 	return (error);
738 }
739 
740 #ifdef INET
741 /*
742  * Accept a connection.  Essentially all the work is done at higher levels;
743  * just return the address of the peer, storing through addr.
744  */
745 static int
746 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
747 {
748 	int error = 0;
749 	struct inpcb *inp = NULL;
750 	struct tcpcb *tp = NULL;
751 	struct in_addr addr;
752 	in_port_t port = 0;
753 	TCPDEBUG0;
754 
755 	if (so->so_state & SS_ISDISCONNECTED)
756 		return (ECONNABORTED);
757 
758 	inp = sotoinpcb(so);
759 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
760 	INP_WLOCK(inp);
761 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
762 		error = ECONNABORTED;
763 		goto out;
764 	}
765 	tp = intotcpcb(inp);
766 	TCPDEBUG1();
767 
768 	/*
769 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
770 	 * copy the data of interest and defer the malloc until after we
771 	 * release the lock.
772 	 */
773 	port = inp->inp_fport;
774 	addr = inp->inp_faddr;
775 
776 out:
777 	TCPDEBUG2(PRU_ACCEPT);
778 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
779 	INP_WUNLOCK(inp);
780 	if (error == 0)
781 		*nam = in_sockaddr(port, &addr);
782 	return error;
783 }
784 #endif /* INET */
785 
786 #ifdef INET6
787 static int
788 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
789 {
790 	struct inpcb *inp = NULL;
791 	int error = 0;
792 	struct tcpcb *tp = NULL;
793 	struct in_addr addr;
794 	struct in6_addr addr6;
795 	struct epoch_tracker et;
796 	in_port_t port = 0;
797 	int v4 = 0;
798 	TCPDEBUG0;
799 
800 	if (so->so_state & SS_ISDISCONNECTED)
801 		return (ECONNABORTED);
802 
803 	inp = sotoinpcb(so);
804 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
805 	NET_EPOCH_ENTER(et);
806 	INP_WLOCK(inp);
807 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
808 		error = ECONNABORTED;
809 		goto out;
810 	}
811 	tp = intotcpcb(inp);
812 	TCPDEBUG1();
813 
814 	/*
815 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
816 	 * copy the data of interest and defer the malloc until after we
817 	 * release the lock.
818 	 */
819 	if (inp->inp_vflag & INP_IPV4) {
820 		v4 = 1;
821 		port = inp->inp_fport;
822 		addr = inp->inp_faddr;
823 	} else {
824 		port = inp->inp_fport;
825 		addr6 = inp->in6p_faddr;
826 	}
827 
828 out:
829 	TCPDEBUG2(PRU_ACCEPT);
830 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
831 	INP_WUNLOCK(inp);
832 	NET_EPOCH_EXIT(et);
833 	if (error == 0) {
834 		if (v4)
835 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
836 		else
837 			*nam = in6_sockaddr(port, &addr6);
838 	}
839 	return error;
840 }
841 #endif /* INET6 */
842 
843 /*
844  * Mark the connection as being incapable of further output.
845  */
846 static int
847 tcp_usr_shutdown(struct socket *so)
848 {
849 	int error = 0;
850 	struct inpcb *inp;
851 	struct tcpcb *tp = NULL;
852 	struct epoch_tracker et;
853 
854 	TCPDEBUG0;
855 	NET_EPOCH_ENTER(et);
856 	inp = sotoinpcb(so);
857 	KASSERT(inp != NULL, ("inp == NULL"));
858 	INP_WLOCK(inp);
859 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
860 		error = ECONNRESET;
861 		goto out;
862 	}
863 	tp = intotcpcb(inp);
864 	TCPDEBUG1();
865 	socantsendmore(so);
866 	tcp_usrclosed(tp);
867 	if (!(inp->inp_flags & INP_DROPPED))
868 		error = tp->t_fb->tfb_tcp_output(tp);
869 
870 out:
871 	TCPDEBUG2(PRU_SHUTDOWN);
872 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
873 	INP_WUNLOCK(inp);
874 	NET_EPOCH_EXIT(et);
875 
876 	return (error);
877 }
878 
879 /*
880  * After a receive, possibly send window update to peer.
881  */
882 static int
883 tcp_usr_rcvd(struct socket *so, int flags)
884 {
885 	struct inpcb *inp;
886 	struct tcpcb *tp = NULL;
887 	int error = 0;
888 
889 	TCPDEBUG0;
890 	inp = sotoinpcb(so);
891 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
892 	INP_WLOCK(inp);
893 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
894 		error = ECONNRESET;
895 		goto out;
896 	}
897 	tp = intotcpcb(inp);
898 	TCPDEBUG1();
899 	/*
900 	 * For passively-created TFO connections, don't attempt a window
901 	 * update while still in SYN_RECEIVED as this may trigger an early
902 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
903 	 * application response data, or failing that, when the DELACK timer
904 	 * expires.
905 	 */
906 	if (IS_FASTOPEN(tp->t_flags) &&
907 	    (tp->t_state == TCPS_SYN_RECEIVED))
908 		goto out;
909 #ifdef TCP_OFFLOAD
910 	if (tp->t_flags & TF_TOE)
911 		tcp_offload_rcvd(tp);
912 	else
913 #endif
914 	tp->t_fb->tfb_tcp_output(tp);
915 
916 out:
917 	TCPDEBUG2(PRU_RCVD);
918 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
919 	INP_WUNLOCK(inp);
920 	return (error);
921 }
922 
923 /*
924  * Do a send by putting data in output queue and updating urgent
925  * marker if URG set.  Possibly send more data.  Unlike the other
926  * pru_*() routines, the mbuf chains are our responsibility.  We
927  * must either enqueue them or free them.  The other pru_* routines
928  * generally are caller-frees.
929  */
930 static int
931 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
932     struct sockaddr *nam, struct mbuf *control, struct thread *td)
933 {
934 	struct epoch_tracker et;
935 	int error = 0;
936 	struct inpcb *inp;
937 	struct tcpcb *tp = NULL;
938 #ifdef INET
939 #ifdef INET6
940 	struct sockaddr_in sin;
941 #endif
942 	struct sockaddr_in *sinp;
943 #endif
944 #ifdef INET6
945 	int isipv6;
946 #endif
947 	u_int8_t incflagsav;
948 	u_char vflagsav;
949 	bool restoreflags;
950 	TCPDEBUG0;
951 
952 	/*
953 	 * We require the pcbinfo "read lock" if we will close the socket
954 	 * as part of this call.
955 	 */
956 	if (flags & PRUS_EOF)
957 		NET_EPOCH_ENTER(et);
958 	inp = sotoinpcb(so);
959 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
960 	INP_WLOCK(inp);
961 	vflagsav = inp->inp_vflag;
962 	incflagsav = inp->inp_inc.inc_flags;
963 	restoreflags = false;
964 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
965 		if (control)
966 			m_freem(control);
967 		/*
968 		 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
969 		 * for freeing memory.
970 		 */
971 		if (m && (flags & PRUS_NOTREADY) == 0)
972 			m_freem(m);
973 		error = ECONNRESET;
974 		goto out;
975 	}
976 	tp = intotcpcb(inp);
977 	TCPDEBUG1();
978 	if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
979 		switch (nam->sa_family) {
980 #ifdef INET
981 		case AF_INET:
982 			sinp = (struct sockaddr_in *)nam;
983 			if (sinp->sin_len != sizeof(struct sockaddr_in)) {
984 				if (m)
985 					m_freem(m);
986 				error = EINVAL;
987 				goto out;
988 			}
989 			if ((inp->inp_vflag & INP_IPV6) != 0) {
990 				if (m)
991 					m_freem(m);
992 				error = EAFNOSUPPORT;
993 				goto out;
994 			}
995 			if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
996 				if (m)
997 					m_freem(m);
998 				error = EAFNOSUPPORT;
999 				goto out;
1000 			}
1001 			if ((error = prison_remote_ip4(td->td_ucred,
1002 			    &sinp->sin_addr))) {
1003 				if (m)
1004 					m_freem(m);
1005 				goto out;
1006 			}
1007 #ifdef INET6
1008 			isipv6 = 0;
1009 #endif
1010 			break;
1011 #endif /* INET */
1012 #ifdef INET6
1013 		case AF_INET6:
1014 		{
1015 			struct sockaddr_in6 *sin6;
1016 
1017 			sin6 = (struct sockaddr_in6 *)nam;
1018 			if (sin6->sin6_len != sizeof(*sin6)) {
1019 				if (m)
1020 					m_freem(m);
1021 				error = EINVAL;
1022 				goto out;
1023 			}
1024 			if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1025 				if (m)
1026 					m_freem(m);
1027 				error = EAFNOSUPPORT;
1028 				goto out;
1029 			}
1030 			if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1031 #ifdef INET
1032 				if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1033 					error = EINVAL;
1034 					if (m)
1035 						m_freem(m);
1036 					goto out;
1037 				}
1038 				if ((inp->inp_vflag & INP_IPV4) == 0) {
1039 					error = EAFNOSUPPORT;
1040 					if (m)
1041 						m_freem(m);
1042 					goto out;
1043 				}
1044 				restoreflags = true;
1045 				inp->inp_vflag &= ~INP_IPV6;
1046 				sinp = &sin;
1047 				in6_sin6_2_sin(sinp, sin6);
1048 				if (IN_MULTICAST(
1049 				    ntohl(sinp->sin_addr.s_addr))) {
1050 					error = EAFNOSUPPORT;
1051 					if (m)
1052 						m_freem(m);
1053 					goto out;
1054 				}
1055 				if ((error = prison_remote_ip4(td->td_ucred,
1056 				    &sinp->sin_addr))) {
1057 					if (m)
1058 						m_freem(m);
1059 					goto out;
1060 				}
1061 				isipv6 = 0;
1062 #else /* !INET */
1063 				error = EAFNOSUPPORT;
1064 				if (m)
1065 					m_freem(m);
1066 				goto out;
1067 #endif /* INET */
1068 			} else {
1069 				if ((inp->inp_vflag & INP_IPV6) == 0) {
1070 					if (m)
1071 						m_freem(m);
1072 					error = EAFNOSUPPORT;
1073 					goto out;
1074 				}
1075 				restoreflags = true;
1076 				inp->inp_vflag &= ~INP_IPV4;
1077 				inp->inp_inc.inc_flags |= INC_ISIPV6;
1078 				if ((error = prison_remote_ip6(td->td_ucred,
1079 				    &sin6->sin6_addr))) {
1080 					if (m)
1081 						m_freem(m);
1082 					goto out;
1083 				}
1084 				isipv6 = 1;
1085 			}
1086 			break;
1087 		}
1088 #endif /* INET6 */
1089 		default:
1090 			if (m)
1091 				m_freem(m);
1092 			error = EAFNOSUPPORT;
1093 			goto out;
1094 		}
1095 	}
1096 	if (control) {
1097 		/* TCP doesn't do control messages (rights, creds, etc) */
1098 		if (control->m_len) {
1099 			m_freem(control);
1100 			if (m)
1101 				m_freem(m);
1102 			error = EINVAL;
1103 			goto out;
1104 		}
1105 		m_freem(control);	/* empty control, just free it */
1106 	}
1107 	if (!(flags & PRUS_OOB)) {
1108 		sbappendstream(&so->so_snd, m, flags);
1109 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1110 			/*
1111 			 * Do implied connect if not yet connected,
1112 			 * initialize window to default value, and
1113 			 * initialize maxseg using peer's cached MSS.
1114 			 */
1115 #ifdef INET6
1116 			if (isipv6)
1117 				error = tcp6_connect(tp, nam, td);
1118 #endif /* INET6 */
1119 #if defined(INET6) && defined(INET)
1120 			else
1121 #endif
1122 #ifdef INET
1123 				error = tcp_connect(tp,
1124 				    (struct sockaddr *)sinp, td);
1125 #endif
1126 			/*
1127 			 * The bind operation in tcp_connect succeeded. We
1128 			 * no longer want to restore the flags if later
1129 			 * operations fail.
1130 			 */
1131 			if (error == 0 || inp->inp_lport != 0)
1132 				restoreflags = false;
1133 
1134 			if (error)
1135 				goto out;
1136 			if (IS_FASTOPEN(tp->t_flags))
1137 				tcp_fastopen_connect(tp);
1138 			else {
1139 				tp->snd_wnd = TTCP_CLIENT_SND_WND;
1140 				tcp_mss(tp, -1);
1141 			}
1142 		}
1143 		if (flags & PRUS_EOF) {
1144 			/*
1145 			 * Close the send side of the connection after
1146 			 * the data is sent.
1147 			 */
1148 			NET_EPOCH_ASSERT();
1149 			socantsendmore(so);
1150 			tcp_usrclosed(tp);
1151 		}
1152 		if (!(inp->inp_flags & INP_DROPPED) &&
1153 		    !(flags & PRUS_NOTREADY)) {
1154 			if (flags & PRUS_MORETOCOME)
1155 				tp->t_flags |= TF_MORETOCOME;
1156 			error = tp->t_fb->tfb_tcp_output(tp);
1157 			if (flags & PRUS_MORETOCOME)
1158 				tp->t_flags &= ~TF_MORETOCOME;
1159 		}
1160 	} else {
1161 		/*
1162 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1163 		 */
1164 		SOCKBUF_LOCK(&so->so_snd);
1165 		if (sbspace(&so->so_snd) < -512) {
1166 			SOCKBUF_UNLOCK(&so->so_snd);
1167 			m_freem(m);
1168 			error = ENOBUFS;
1169 			goto out;
1170 		}
1171 		/*
1172 		 * According to RFC961 (Assigned Protocols),
1173 		 * the urgent pointer points to the last octet
1174 		 * of urgent data.  We continue, however,
1175 		 * to consider it to indicate the first octet
1176 		 * of data past the urgent section.
1177 		 * Otherwise, snd_up should be one lower.
1178 		 */
1179 		sbappendstream_locked(&so->so_snd, m, flags);
1180 		SOCKBUF_UNLOCK(&so->so_snd);
1181 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1182 			/*
1183 			 * Do implied connect if not yet connected,
1184 			 * initialize window to default value, and
1185 			 * initialize maxseg using peer's cached MSS.
1186 			 */
1187 
1188 			/*
1189 			 * Not going to contemplate SYN|URG
1190 			 */
1191 			if (IS_FASTOPEN(tp->t_flags))
1192 				tp->t_flags &= ~TF_FASTOPEN;
1193 #ifdef INET6
1194 			if (isipv6)
1195 				error = tcp6_connect(tp, nam, td);
1196 #endif /* INET6 */
1197 #if defined(INET6) && defined(INET)
1198 			else
1199 #endif
1200 #ifdef INET
1201 				error = tcp_connect(tp,
1202 				    (struct sockaddr *)sinp, td);
1203 #endif
1204 			/*
1205 			 * The bind operation in tcp_connect succeeded. We
1206 			 * no longer want to restore the flags if later
1207 			 * operations fail.
1208 			 */
1209 			if (error == 0 || inp->inp_lport != 0)
1210 				restoreflags = false;
1211 
1212 			if (error)
1213 				goto out;
1214 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
1215 			tcp_mss(tp, -1);
1216 		}
1217 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1218 		if (!(flags & PRUS_NOTREADY)) {
1219 			tp->t_flags |= TF_FORCEDATA;
1220 			error = tp->t_fb->tfb_tcp_output(tp);
1221 			tp->t_flags &= ~TF_FORCEDATA;
1222 		}
1223 	}
1224 	TCP_LOG_EVENT(tp, NULL,
1225 	    &inp->inp_socket->so_rcv,
1226 	    &inp->inp_socket->so_snd,
1227 	    TCP_LOG_USERSEND, error,
1228 	    0, NULL, false);
1229 out:
1230 	/*
1231 	 * If the request was unsuccessful and we changed flags,
1232 	 * restore the original flags.
1233 	 */
1234 	if (error != 0 && restoreflags) {
1235 		inp->inp_vflag = vflagsav;
1236 		inp->inp_inc.inc_flags = incflagsav;
1237 	}
1238 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1239 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1240 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1241 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1242 	INP_WUNLOCK(inp);
1243 	if (flags & PRUS_EOF)
1244 		NET_EPOCH_EXIT(et);
1245 	return (error);
1246 }
1247 
1248 static int
1249 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1250 {
1251 	struct inpcb *inp;
1252 	struct tcpcb *tp;
1253 	int error;
1254 
1255 	inp = sotoinpcb(so);
1256 	INP_WLOCK(inp);
1257 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1258 		INP_WUNLOCK(inp);
1259 		mb_free_notready(m, count);
1260 		return (ECONNRESET);
1261 	}
1262 	tp = intotcpcb(inp);
1263 
1264 	SOCKBUF_LOCK(&so->so_snd);
1265 	error = sbready(&so->so_snd, m, count);
1266 	SOCKBUF_UNLOCK(&so->so_snd);
1267 	if (error == 0)
1268 		error = tp->t_fb->tfb_tcp_output(tp);
1269 	INP_WUNLOCK(inp);
1270 
1271 	return (error);
1272 }
1273 
1274 /*
1275  * Abort the TCP.  Drop the connection abruptly.
1276  */
1277 static void
1278 tcp_usr_abort(struct socket *so)
1279 {
1280 	struct inpcb *inp;
1281 	struct tcpcb *tp = NULL;
1282 	struct epoch_tracker et;
1283 	TCPDEBUG0;
1284 
1285 	inp = sotoinpcb(so);
1286 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1287 
1288 	NET_EPOCH_ENTER(et);
1289 	INP_WLOCK(inp);
1290 	KASSERT(inp->inp_socket != NULL,
1291 	    ("tcp_usr_abort: inp_socket == NULL"));
1292 
1293 	/*
1294 	 * If we still have full TCP state, and we're not dropped, drop.
1295 	 */
1296 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1297 	    !(inp->inp_flags & INP_DROPPED)) {
1298 		tp = intotcpcb(inp);
1299 		TCPDEBUG1();
1300 		tp = tcp_drop(tp, ECONNABORTED);
1301 		if (tp == NULL)
1302 			goto dropped;
1303 		TCPDEBUG2(PRU_ABORT);
1304 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1305 	}
1306 	if (!(inp->inp_flags & INP_DROPPED)) {
1307 		SOCK_LOCK(so);
1308 		so->so_state |= SS_PROTOREF;
1309 		SOCK_UNLOCK(so);
1310 		inp->inp_flags |= INP_SOCKREF;
1311 	}
1312 	INP_WUNLOCK(inp);
1313 dropped:
1314 	NET_EPOCH_EXIT(et);
1315 }
1316 
1317 /*
1318  * TCP socket is closed.  Start friendly disconnect.
1319  */
1320 static void
1321 tcp_usr_close(struct socket *so)
1322 {
1323 	struct inpcb *inp;
1324 	struct tcpcb *tp = NULL;
1325 	struct epoch_tracker et;
1326 	TCPDEBUG0;
1327 
1328 	inp = sotoinpcb(so);
1329 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1330 
1331 	NET_EPOCH_ENTER(et);
1332 	INP_WLOCK(inp);
1333 	KASSERT(inp->inp_socket != NULL,
1334 	    ("tcp_usr_close: inp_socket == NULL"));
1335 
1336 	/*
1337 	 * If we still have full TCP state, and we're not dropped, initiate
1338 	 * a disconnect.
1339 	 */
1340 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1341 	    !(inp->inp_flags & INP_DROPPED)) {
1342 		tp = intotcpcb(inp);
1343 		TCPDEBUG1();
1344 		tcp_disconnect(tp);
1345 		TCPDEBUG2(PRU_CLOSE);
1346 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1347 	}
1348 	if (!(inp->inp_flags & INP_DROPPED)) {
1349 		SOCK_LOCK(so);
1350 		so->so_state |= SS_PROTOREF;
1351 		SOCK_UNLOCK(so);
1352 		inp->inp_flags |= INP_SOCKREF;
1353 	}
1354 	INP_WUNLOCK(inp);
1355 	NET_EPOCH_EXIT(et);
1356 }
1357 
1358 /*
1359  * Receive out-of-band data.
1360  */
1361 static int
1362 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1363 {
1364 	int error = 0;
1365 	struct inpcb *inp;
1366 	struct tcpcb *tp = NULL;
1367 
1368 	TCPDEBUG0;
1369 	inp = sotoinpcb(so);
1370 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1371 	INP_WLOCK(inp);
1372 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1373 		error = ECONNRESET;
1374 		goto out;
1375 	}
1376 	tp = intotcpcb(inp);
1377 	TCPDEBUG1();
1378 	if ((so->so_oobmark == 0 &&
1379 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1380 	    so->so_options & SO_OOBINLINE ||
1381 	    tp->t_oobflags & TCPOOB_HADDATA) {
1382 		error = EINVAL;
1383 		goto out;
1384 	}
1385 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1386 		error = EWOULDBLOCK;
1387 		goto out;
1388 	}
1389 	m->m_len = 1;
1390 	*mtod(m, caddr_t) = tp->t_iobc;
1391 	if ((flags & MSG_PEEK) == 0)
1392 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1393 
1394 out:
1395 	TCPDEBUG2(PRU_RCVOOB);
1396 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1397 	INP_WUNLOCK(inp);
1398 	return (error);
1399 }
1400 
1401 #ifdef INET
1402 struct pr_usrreqs tcp_usrreqs = {
1403 	.pru_abort =		tcp_usr_abort,
1404 	.pru_accept =		tcp_usr_accept,
1405 	.pru_attach =		tcp_usr_attach,
1406 	.pru_bind =		tcp_usr_bind,
1407 	.pru_connect =		tcp_usr_connect,
1408 	.pru_control =		in_control,
1409 	.pru_detach =		tcp_usr_detach,
1410 	.pru_disconnect =	tcp_usr_disconnect,
1411 	.pru_listen =		tcp_usr_listen,
1412 	.pru_peeraddr =		in_getpeeraddr,
1413 	.pru_rcvd =		tcp_usr_rcvd,
1414 	.pru_rcvoob =		tcp_usr_rcvoob,
1415 	.pru_send =		tcp_usr_send,
1416 	.pru_ready =		tcp_usr_ready,
1417 	.pru_shutdown =		tcp_usr_shutdown,
1418 	.pru_sockaddr =		in_getsockaddr,
1419 	.pru_sosetlabel =	in_pcbsosetlabel,
1420 	.pru_close =		tcp_usr_close,
1421 };
1422 #endif /* INET */
1423 
1424 #ifdef INET6
1425 struct pr_usrreqs tcp6_usrreqs = {
1426 	.pru_abort =		tcp_usr_abort,
1427 	.pru_accept =		tcp6_usr_accept,
1428 	.pru_attach =		tcp_usr_attach,
1429 	.pru_bind =		tcp6_usr_bind,
1430 	.pru_connect =		tcp6_usr_connect,
1431 	.pru_control =		in6_control,
1432 	.pru_detach =		tcp_usr_detach,
1433 	.pru_disconnect =	tcp_usr_disconnect,
1434 	.pru_listen =		tcp6_usr_listen,
1435 	.pru_peeraddr =		in6_mapped_peeraddr,
1436 	.pru_rcvd =		tcp_usr_rcvd,
1437 	.pru_rcvoob =		tcp_usr_rcvoob,
1438 	.pru_send =		tcp_usr_send,
1439 	.pru_ready =		tcp_usr_ready,
1440 	.pru_shutdown =		tcp_usr_shutdown,
1441 	.pru_sockaddr =		in6_mapped_sockaddr,
1442 	.pru_sosetlabel =	in_pcbsosetlabel,
1443 	.pru_close =		tcp_usr_close,
1444 };
1445 #endif /* INET6 */
1446 
1447 #ifdef INET
1448 /*
1449  * Common subroutine to open a TCP connection to remote host specified
1450  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1451  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1452  * to choose a local host address (interface).  If there is an existing
1453  * incarnation of the same connection in TIME-WAIT state and if the remote
1454  * host was sending CC options and if the connection duration was < MSL, then
1455  * truncate the previous TIME-WAIT state and proceed.
1456  * Initialize connection parameters and enter SYN-SENT state.
1457  */
1458 static int
1459 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1460 {
1461 	struct inpcb *inp = tp->t_inpcb, *oinp;
1462 	struct socket *so = inp->inp_socket;
1463 	struct in_addr laddr;
1464 	u_short lport;
1465 	int error;
1466 
1467 	INP_WLOCK_ASSERT(inp);
1468 	INP_HASH_WLOCK(&V_tcbinfo);
1469 
1470 	if (inp->inp_lport == 0) {
1471 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1472 		if (error)
1473 			goto out;
1474 	}
1475 
1476 	/*
1477 	 * Cannot simply call in_pcbconnect, because there might be an
1478 	 * earlier incarnation of this same connection still in
1479 	 * TIME_WAIT state, creating an ADDRINUSE error.
1480 	 */
1481 	laddr = inp->inp_laddr;
1482 	lport = inp->inp_lport;
1483 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1484 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1485 	if (error && oinp == NULL)
1486 		goto out;
1487 	if (oinp) {
1488 		error = EADDRINUSE;
1489 		goto out;
1490 	}
1491 	inp->inp_laddr = laddr;
1492 	in_pcbrehash(inp);
1493 	INP_HASH_WUNLOCK(&V_tcbinfo);
1494 
1495 	/*
1496 	 * Compute window scaling to request:
1497 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1498 	 * XXX: This should move to tcp_output().
1499 	 */
1500 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1501 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1502 		tp->request_r_scale++;
1503 
1504 	soisconnecting(so);
1505 	TCPSTAT_INC(tcps_connattempt);
1506 	tcp_state_change(tp, TCPS_SYN_SENT);
1507 	tp->iss = tcp_new_isn(&inp->inp_inc);
1508 	if (tp->t_flags & TF_REQ_TSTMP)
1509 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1510 	tcp_sendseqinit(tp);
1511 
1512 	return 0;
1513 
1514 out:
1515 	INP_HASH_WUNLOCK(&V_tcbinfo);
1516 	return (error);
1517 }
1518 #endif /* INET */
1519 
1520 #ifdef INET6
1521 static int
1522 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1523 {
1524 	struct inpcb *inp = tp->t_inpcb;
1525 	int error;
1526 
1527 	INP_WLOCK_ASSERT(inp);
1528 	INP_HASH_WLOCK(&V_tcbinfo);
1529 
1530 	if (inp->inp_lport == 0) {
1531 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1532 		if (error)
1533 			goto out;
1534 	}
1535 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1536 	if (error != 0)
1537 		goto out;
1538 	INP_HASH_WUNLOCK(&V_tcbinfo);
1539 
1540 	/* Compute window scaling to request.  */
1541 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1542 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1543 		tp->request_r_scale++;
1544 
1545 	soisconnecting(inp->inp_socket);
1546 	TCPSTAT_INC(tcps_connattempt);
1547 	tcp_state_change(tp, TCPS_SYN_SENT);
1548 	tp->iss = tcp_new_isn(&inp->inp_inc);
1549 	if (tp->t_flags & TF_REQ_TSTMP)
1550 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1551 	tcp_sendseqinit(tp);
1552 
1553 	return 0;
1554 
1555 out:
1556 	INP_HASH_WUNLOCK(&V_tcbinfo);
1557 	return error;
1558 }
1559 #endif /* INET6 */
1560 
1561 /*
1562  * Export TCP internal state information via a struct tcp_info, based on the
1563  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1564  * (TCP state machine, etc).  We export all information using FreeBSD-native
1565  * constants -- for example, the numeric values for tcpi_state will differ
1566  * from Linux.
1567  */
1568 static void
1569 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1570 {
1571 
1572 	INP_WLOCK_ASSERT(tp->t_inpcb);
1573 	bzero(ti, sizeof(*ti));
1574 
1575 	ti->tcpi_state = tp->t_state;
1576 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1577 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1578 	if (tp->t_flags & TF_SACK_PERMIT)
1579 		ti->tcpi_options |= TCPI_OPT_SACK;
1580 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1581 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1582 		ti->tcpi_snd_wscale = tp->snd_scale;
1583 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1584 	}
1585 	if (tp->t_flags2 & TF2_ECN_PERMIT)
1586 		ti->tcpi_options |= TCPI_OPT_ECN;
1587 
1588 	ti->tcpi_rto = tp->t_rxtcur * tick;
1589 	ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1590 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1591 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1592 
1593 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1594 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1595 
1596 	/*
1597 	 * FreeBSD-specific extension fields for tcp_info.
1598 	 */
1599 	ti->tcpi_rcv_space = tp->rcv_wnd;
1600 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1601 	ti->tcpi_snd_wnd = tp->snd_wnd;
1602 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1603 	ti->tcpi_snd_nxt = tp->snd_nxt;
1604 	ti->tcpi_snd_mss = tp->t_maxseg;
1605 	ti->tcpi_rcv_mss = tp->t_maxseg;
1606 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1607 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1608 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1609 #ifdef TCP_OFFLOAD
1610 	if (tp->t_flags & TF_TOE) {
1611 		ti->tcpi_options |= TCPI_OPT_TOE;
1612 		tcp_offload_tcp_info(tp, ti);
1613 	}
1614 #endif
1615 }
1616 
1617 /*
1618  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1619  * socket option arguments.  When it re-acquires the lock after the copy, it
1620  * has to revalidate that the connection is still valid for the socket
1621  * option.
1622  */
1623 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1624 	INP_WLOCK(inp);							\
1625 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1626 		INP_WUNLOCK(inp);					\
1627 		cleanup;						\
1628 		return (ECONNRESET);					\
1629 	}								\
1630 	tp = intotcpcb(inp);						\
1631 } while(0)
1632 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1633 
1634 int
1635 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1636 {
1637 	int	error;
1638 	struct	inpcb *inp;
1639 	struct	tcpcb *tp;
1640 	struct tcp_function_block *blk;
1641 	struct tcp_function_set fsn;
1642 
1643 	error = 0;
1644 	inp = sotoinpcb(so);
1645 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1646 	if (sopt->sopt_level != IPPROTO_TCP) {
1647 #ifdef INET6
1648 		if (inp->inp_vflag & INP_IPV6PROTO) {
1649 			error = ip6_ctloutput(so, sopt);
1650 			/*
1651 			 * In case of the IPV6_USE_MIN_MTU socket option,
1652 			 * the INC_IPV6MINMTU flag to announce a corresponding
1653 			 * MSS during the initial handshake.
1654 			 * If the TCP connection is not in the front states,
1655 			 * just reduce the MSS being used.
1656 			 * This avoids the sending of TCP segments which will
1657 			 * be fragmented at the IPv6 layer.
1658 			 */
1659 			if ((error == 0) &&
1660 			    (sopt->sopt_dir == SOPT_SET) &&
1661 			    (sopt->sopt_level == IPPROTO_IPV6) &&
1662 			    (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1663 				INP_WLOCK(inp);
1664 				if ((inp->inp_flags &
1665 				    (INP_TIMEWAIT | INP_DROPPED))) {
1666 					INP_WUNLOCK(inp);
1667 					return (ECONNRESET);
1668 				}
1669 				inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1670 				tp = intotcpcb(inp);
1671 				if ((tp->t_state >= TCPS_SYN_SENT) &&
1672 				    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1673 					struct ip6_pktopts *opt;
1674 
1675 					opt = inp->in6p_outputopts;
1676 					if ((opt != NULL) &&
1677 					    (opt->ip6po_minmtu ==
1678 					    IP6PO_MINMTU_ALL)) {
1679 						if (tp->t_maxseg > TCP6_MSS) {
1680 							tp->t_maxseg = TCP6_MSS;
1681 						}
1682 					}
1683 				}
1684 				INP_WUNLOCK(inp);
1685 			}
1686 		}
1687 #endif /* INET6 */
1688 #if defined(INET6) && defined(INET)
1689 		else
1690 #endif
1691 #ifdef INET
1692 		{
1693 			error = ip_ctloutput(so, sopt);
1694 		}
1695 #endif
1696 		return (error);
1697 	}
1698 	INP_WLOCK(inp);
1699 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1700 		INP_WUNLOCK(inp);
1701 		return (ECONNRESET);
1702 	}
1703 	tp = intotcpcb(inp);
1704 	/*
1705 	 * Protect the TCP option TCP_FUNCTION_BLK so
1706 	 * that a sub-function can *never* overwrite this.
1707 	 */
1708 	if ((sopt->sopt_dir == SOPT_SET) &&
1709 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1710 		INP_WUNLOCK(inp);
1711 		error = sooptcopyin(sopt, &fsn, sizeof fsn,
1712 		    sizeof fsn);
1713 		if (error)
1714 			return (error);
1715 		INP_WLOCK_RECHECK(inp);
1716 		blk = find_and_ref_tcp_functions(&fsn);
1717 		if (blk == NULL) {
1718 			INP_WUNLOCK(inp);
1719 			return (ENOENT);
1720 		}
1721 		if (tp->t_fb == blk) {
1722 			/* You already have this */
1723 			refcount_release(&blk->tfb_refcnt);
1724 			INP_WUNLOCK(inp);
1725 			return (0);
1726 		}
1727 		if (tp->t_state != TCPS_CLOSED) {
1728 			/*
1729 			 * The user has advanced the state
1730 			 * past the initial point, we may not
1731 			 * be able to switch.
1732 			 */
1733 			if (blk->tfb_tcp_handoff_ok != NULL) {
1734 				/*
1735 				 * Does the stack provide a
1736 				 * query mechanism, if so it may
1737 				 * still be possible?
1738 				 */
1739 				error = (*blk->tfb_tcp_handoff_ok)(tp);
1740 			} else
1741 				error = EINVAL;
1742 			if (error) {
1743 				refcount_release(&blk->tfb_refcnt);
1744 				INP_WUNLOCK(inp);
1745 				return(error);
1746 			}
1747 		}
1748 		if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1749 			refcount_release(&blk->tfb_refcnt);
1750 			INP_WUNLOCK(inp);
1751 			return (ENOENT);
1752 		}
1753 		/*
1754 		 * Release the old refcnt, the
1755 		 * lookup acquired a ref on the
1756 		 * new one already.
1757 		 */
1758 		if (tp->t_fb->tfb_tcp_fb_fini) {
1759 			/*
1760 			 * Tell the stack to cleanup with 0 i.e.
1761 			 * the tcb is not going away.
1762 			 */
1763 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1764 		}
1765 #ifdef TCPHPTS
1766 		/* Assure that we are not on any hpts */
1767 		tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1768 #endif
1769 		if (blk->tfb_tcp_fb_init) {
1770 			error = (*blk->tfb_tcp_fb_init)(tp);
1771 			if (error) {
1772 				refcount_release(&blk->tfb_refcnt);
1773 				if (tp->t_fb->tfb_tcp_fb_init) {
1774 					if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1775 						/* Fall back failed, drop the connection */
1776 						INP_WUNLOCK(inp);
1777 						soabort(so);
1778 						return(error);
1779 					}
1780 				}
1781 				goto err_out;
1782 			}
1783 		}
1784 		refcount_release(&tp->t_fb->tfb_refcnt);
1785 		tp->t_fb = blk;
1786 #ifdef TCP_OFFLOAD
1787 		if (tp->t_flags & TF_TOE) {
1788 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1789 			     sopt->sopt_name);
1790 		}
1791 #endif
1792 err_out:
1793 		INP_WUNLOCK(inp);
1794 		return (error);
1795 	} else if ((sopt->sopt_dir == SOPT_GET) &&
1796 	    (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1797 		strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1798 		    TCP_FUNCTION_NAME_LEN_MAX);
1799 		fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1800 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1801 		INP_WUNLOCK(inp);
1802 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1803 		return (error);
1804 	}
1805 	/* Pass in the INP locked, called must unlock it */
1806 	return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1807 }
1808 
1809 /*
1810  * If this assert becomes untrue, we need to change the size of the buf
1811  * variable in tcp_default_ctloutput().
1812  */
1813 #ifdef CTASSERT
1814 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1815 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1816 #endif
1817 
1818 int
1819 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
1820 {
1821 	int	error, opt, optval;
1822 	u_int	ui;
1823 	struct	tcp_info ti;
1824 #ifdef KERN_TLS
1825 	struct tls_enable tls;
1826 #endif
1827 	struct cc_algo *algo;
1828 	char	*pbuf, buf[TCP_LOG_ID_LEN];
1829 #ifdef STATS
1830 	struct statsblob *sbp;
1831 #endif
1832 	size_t	len;
1833 
1834 	/*
1835 	 * For TCP_CCALGOOPT forward the control to CC module, for both
1836 	 * SOPT_SET and SOPT_GET.
1837 	 */
1838 	switch (sopt->sopt_name) {
1839 	case TCP_CCALGOOPT:
1840 		INP_WUNLOCK(inp);
1841 		if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
1842 			return (EINVAL);
1843 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
1844 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
1845 		    sopt->sopt_valsize);
1846 		if (error) {
1847 			free(pbuf, M_TEMP);
1848 			return (error);
1849 		}
1850 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
1851 		if (CC_ALGO(tp)->ctl_output != NULL)
1852 			error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
1853 		else
1854 			error = ENOENT;
1855 		INP_WUNLOCK(inp);
1856 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
1857 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
1858 		free(pbuf, M_TEMP);
1859 		return (error);
1860 	}
1861 
1862 	switch (sopt->sopt_dir) {
1863 	case SOPT_SET:
1864 		switch (sopt->sopt_name) {
1865 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1866 		case TCP_MD5SIG:
1867 			if (!TCPMD5_ENABLED()) {
1868 				INP_WUNLOCK(inp);
1869 				return (ENOPROTOOPT);
1870 			}
1871 			error = TCPMD5_PCBCTL(inp, sopt);
1872 			if (error)
1873 				return (error);
1874 			goto unlock_and_done;
1875 #endif /* IPSEC */
1876 
1877 		case TCP_NODELAY:
1878 		case TCP_NOOPT:
1879 			INP_WUNLOCK(inp);
1880 			error = sooptcopyin(sopt, &optval, sizeof optval,
1881 			    sizeof optval);
1882 			if (error)
1883 				return (error);
1884 
1885 			INP_WLOCK_RECHECK(inp);
1886 			switch (sopt->sopt_name) {
1887 			case TCP_NODELAY:
1888 				opt = TF_NODELAY;
1889 				break;
1890 			case TCP_NOOPT:
1891 				opt = TF_NOOPT;
1892 				break;
1893 			default:
1894 				opt = 0; /* dead code to fool gcc */
1895 				break;
1896 			}
1897 
1898 			if (optval)
1899 				tp->t_flags |= opt;
1900 			else
1901 				tp->t_flags &= ~opt;
1902 unlock_and_done:
1903 #ifdef TCP_OFFLOAD
1904 			if (tp->t_flags & TF_TOE) {
1905 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
1906 				    sopt->sopt_name);
1907 			}
1908 #endif
1909 			INP_WUNLOCK(inp);
1910 			break;
1911 
1912 		case TCP_NOPUSH:
1913 			INP_WUNLOCK(inp);
1914 			error = sooptcopyin(sopt, &optval, sizeof optval,
1915 			    sizeof optval);
1916 			if (error)
1917 				return (error);
1918 
1919 			INP_WLOCK_RECHECK(inp);
1920 			if (optval)
1921 				tp->t_flags |= TF_NOPUSH;
1922 			else if (tp->t_flags & TF_NOPUSH) {
1923 				tp->t_flags &= ~TF_NOPUSH;
1924 				if (TCPS_HAVEESTABLISHED(tp->t_state))
1925 					error = tp->t_fb->tfb_tcp_output(tp);
1926 			}
1927 			goto unlock_and_done;
1928 
1929 		case TCP_MAXSEG:
1930 			INP_WUNLOCK(inp);
1931 			error = sooptcopyin(sopt, &optval, sizeof optval,
1932 			    sizeof optval);
1933 			if (error)
1934 				return (error);
1935 
1936 			INP_WLOCK_RECHECK(inp);
1937 			if (optval > 0 && optval <= tp->t_maxseg &&
1938 			    optval + 40 >= V_tcp_minmss)
1939 				tp->t_maxseg = optval;
1940 			else
1941 				error = EINVAL;
1942 			goto unlock_and_done;
1943 
1944 		case TCP_INFO:
1945 			INP_WUNLOCK(inp);
1946 			error = EINVAL;
1947 			break;
1948 
1949 		case TCP_STATS:
1950 			INP_WUNLOCK(inp);
1951 #ifdef STATS
1952 			error = sooptcopyin(sopt, &optval, sizeof optval,
1953 			    sizeof optval);
1954 			if (error)
1955 				return (error);
1956 
1957 			if (optval > 0)
1958 				sbp = stats_blob_alloc(
1959 				    V_tcp_perconn_stats_dflt_tpl, 0);
1960 			else
1961 				sbp = NULL;
1962 
1963 			INP_WLOCK_RECHECK(inp);
1964 			if ((tp->t_stats != NULL && sbp == NULL) ||
1965 			    (tp->t_stats == NULL && sbp != NULL)) {
1966 				struct statsblob *t = tp->t_stats;
1967 				tp->t_stats = sbp;
1968 				sbp = t;
1969 			}
1970 			INP_WUNLOCK(inp);
1971 
1972 			stats_blob_destroy(sbp);
1973 #else
1974 			return (EOPNOTSUPP);
1975 #endif /* !STATS */
1976 			break;
1977 
1978 		case TCP_CONGESTION:
1979 			INP_WUNLOCK(inp);
1980 			error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
1981 			if (error)
1982 				break;
1983 			buf[sopt->sopt_valsize] = '\0';
1984 			INP_WLOCK_RECHECK(inp);
1985 			CC_LIST_RLOCK();
1986 			STAILQ_FOREACH(algo, &cc_list, entries)
1987 				if (strncmp(buf, algo->name,
1988 				    TCP_CA_NAME_MAX) == 0)
1989 					break;
1990 			CC_LIST_RUNLOCK();
1991 			if (algo == NULL) {
1992 				INP_WUNLOCK(inp);
1993 				error = EINVAL;
1994 				break;
1995 			}
1996 			/*
1997 			 * We hold a write lock over the tcb so it's safe to
1998 			 * do these things without ordering concerns.
1999 			 */
2000 			if (CC_ALGO(tp)->cb_destroy != NULL)
2001 				CC_ALGO(tp)->cb_destroy(tp->ccv);
2002 			CC_DATA(tp) = NULL;
2003 			CC_ALGO(tp) = algo;
2004 			/*
2005 			 * If something goes pear shaped initialising the new
2006 			 * algo, fall back to newreno (which does not
2007 			 * require initialisation).
2008 			 */
2009 			if (algo->cb_init != NULL &&
2010 			    algo->cb_init(tp->ccv) != 0) {
2011 				CC_ALGO(tp) = &newreno_cc_algo;
2012 				/*
2013 				 * The only reason init should fail is
2014 				 * because of malloc.
2015 				 */
2016 				error = ENOMEM;
2017 			}
2018 			INP_WUNLOCK(inp);
2019 			break;
2020 
2021 #ifdef KERN_TLS
2022 		case TCP_TXTLS_ENABLE:
2023 			INP_WUNLOCK(inp);
2024 			error = sooptcopyin(sopt, &tls, sizeof(tls),
2025 			    sizeof(tls));
2026 			if (error)
2027 				break;
2028 			error = ktls_enable_tx(so, &tls);
2029 			break;
2030 		case TCP_TXTLS_MODE:
2031 			INP_WUNLOCK(inp);
2032 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2033 			if (error)
2034 				return (error);
2035 
2036 			INP_WLOCK_RECHECK(inp);
2037 			error = ktls_set_tx_mode(so, ui);
2038 			INP_WUNLOCK(inp);
2039 			break;
2040 #endif
2041 
2042 		case TCP_KEEPIDLE:
2043 		case TCP_KEEPINTVL:
2044 		case TCP_KEEPINIT:
2045 			INP_WUNLOCK(inp);
2046 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2047 			if (error)
2048 				return (error);
2049 
2050 			if (ui > (UINT_MAX / hz)) {
2051 				error = EINVAL;
2052 				break;
2053 			}
2054 			ui *= hz;
2055 
2056 			INP_WLOCK_RECHECK(inp);
2057 			switch (sopt->sopt_name) {
2058 			case TCP_KEEPIDLE:
2059 				tp->t_keepidle = ui;
2060 				/*
2061 				 * XXX: better check current remaining
2062 				 * timeout and "merge" it with new value.
2063 				 */
2064 				if ((tp->t_state > TCPS_LISTEN) &&
2065 				    (tp->t_state <= TCPS_CLOSING))
2066 					tcp_timer_activate(tp, TT_KEEP,
2067 					    TP_KEEPIDLE(tp));
2068 				break;
2069 			case TCP_KEEPINTVL:
2070 				tp->t_keepintvl = ui;
2071 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2072 				    (TP_MAXIDLE(tp) > 0))
2073 					tcp_timer_activate(tp, TT_2MSL,
2074 					    TP_MAXIDLE(tp));
2075 				break;
2076 			case TCP_KEEPINIT:
2077 				tp->t_keepinit = ui;
2078 				if (tp->t_state == TCPS_SYN_RECEIVED ||
2079 				    tp->t_state == TCPS_SYN_SENT)
2080 					tcp_timer_activate(tp, TT_KEEP,
2081 					    TP_KEEPINIT(tp));
2082 				break;
2083 			}
2084 			goto unlock_and_done;
2085 
2086 		case TCP_KEEPCNT:
2087 			INP_WUNLOCK(inp);
2088 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2089 			if (error)
2090 				return (error);
2091 
2092 			INP_WLOCK_RECHECK(inp);
2093 			tp->t_keepcnt = ui;
2094 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2095 			    (TP_MAXIDLE(tp) > 0))
2096 				tcp_timer_activate(tp, TT_2MSL,
2097 				    TP_MAXIDLE(tp));
2098 			goto unlock_and_done;
2099 
2100 #ifdef TCPPCAP
2101 		case TCP_PCAP_OUT:
2102 		case TCP_PCAP_IN:
2103 			INP_WUNLOCK(inp);
2104 			error = sooptcopyin(sopt, &optval, sizeof optval,
2105 			    sizeof optval);
2106 			if (error)
2107 				return (error);
2108 
2109 			INP_WLOCK_RECHECK(inp);
2110 			if (optval >= 0)
2111 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2112 					&(tp->t_outpkts) : &(tp->t_inpkts),
2113 					optval);
2114 			else
2115 				error = EINVAL;
2116 			goto unlock_and_done;
2117 #endif
2118 
2119 		case TCP_FASTOPEN: {
2120 			struct tcp_fastopen tfo_optval;
2121 
2122 			INP_WUNLOCK(inp);
2123 			if (!V_tcp_fastopen_client_enable &&
2124 			    !V_tcp_fastopen_server_enable)
2125 				return (EPERM);
2126 
2127 			error = sooptcopyin(sopt, &tfo_optval,
2128 				    sizeof(tfo_optval), sizeof(int));
2129 			if (error)
2130 				return (error);
2131 
2132 			INP_WLOCK_RECHECK(inp);
2133 			if (tfo_optval.enable) {
2134 				if (tp->t_state == TCPS_LISTEN) {
2135 					if (!V_tcp_fastopen_server_enable) {
2136 						error = EPERM;
2137 						goto unlock_and_done;
2138 					}
2139 
2140 					tp->t_flags |= TF_FASTOPEN;
2141 					if (tp->t_tfo_pending == NULL)
2142 						tp->t_tfo_pending =
2143 						    tcp_fastopen_alloc_counter();
2144 				} else {
2145 					/*
2146 					 * If a pre-shared key was provided,
2147 					 * stash it in the client cookie
2148 					 * field of the tcpcb for use during
2149 					 * connect.
2150 					 */
2151 					if (sopt->sopt_valsize ==
2152 					    sizeof(tfo_optval)) {
2153 						memcpy(tp->t_tfo_cookie.client,
2154 						       tfo_optval.psk,
2155 						       TCP_FASTOPEN_PSK_LEN);
2156 						tp->t_tfo_client_cookie_len =
2157 						    TCP_FASTOPEN_PSK_LEN;
2158 					}
2159 					tp->t_flags |= TF_FASTOPEN;
2160 				}
2161 			} else
2162 				tp->t_flags &= ~TF_FASTOPEN;
2163 			goto unlock_and_done;
2164 		}
2165 
2166 #ifdef TCP_BLACKBOX
2167 		case TCP_LOG:
2168 			INP_WUNLOCK(inp);
2169 			error = sooptcopyin(sopt, &optval, sizeof optval,
2170 			    sizeof optval);
2171 			if (error)
2172 				return (error);
2173 
2174 			INP_WLOCK_RECHECK(inp);
2175 			error = tcp_log_state_change(tp, optval);
2176 			goto unlock_and_done;
2177 
2178 		case TCP_LOGBUF:
2179 			INP_WUNLOCK(inp);
2180 			error = EINVAL;
2181 			break;
2182 
2183 		case TCP_LOGID:
2184 			INP_WUNLOCK(inp);
2185 			error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2186 			if (error)
2187 				break;
2188 			buf[sopt->sopt_valsize] = '\0';
2189 			INP_WLOCK_RECHECK(inp);
2190 			error = tcp_log_set_id(tp, buf);
2191 			/* tcp_log_set_id() unlocks the INP. */
2192 			break;
2193 
2194 		case TCP_LOGDUMP:
2195 		case TCP_LOGDUMPID:
2196 			INP_WUNLOCK(inp);
2197 			error =
2198 			    sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2199 			if (error)
2200 				break;
2201 			buf[sopt->sopt_valsize] = '\0';
2202 			INP_WLOCK_RECHECK(inp);
2203 			if (sopt->sopt_name == TCP_LOGDUMP) {
2204 				error = tcp_log_dump_tp_logbuf(tp, buf,
2205 				    M_WAITOK, true);
2206 				INP_WUNLOCK(inp);
2207 			} else {
2208 				tcp_log_dump_tp_bucket_logbufs(tp, buf);
2209 				/*
2210 				 * tcp_log_dump_tp_bucket_logbufs() drops the
2211 				 * INP lock.
2212 				 */
2213 			}
2214 			break;
2215 #endif
2216 
2217 		default:
2218 			INP_WUNLOCK(inp);
2219 			error = ENOPROTOOPT;
2220 			break;
2221 		}
2222 		break;
2223 
2224 	case SOPT_GET:
2225 		tp = intotcpcb(inp);
2226 		switch (sopt->sopt_name) {
2227 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2228 		case TCP_MD5SIG:
2229 			if (!TCPMD5_ENABLED()) {
2230 				INP_WUNLOCK(inp);
2231 				return (ENOPROTOOPT);
2232 			}
2233 			error = TCPMD5_PCBCTL(inp, sopt);
2234 			break;
2235 #endif
2236 
2237 		case TCP_NODELAY:
2238 			optval = tp->t_flags & TF_NODELAY;
2239 			INP_WUNLOCK(inp);
2240 			error = sooptcopyout(sopt, &optval, sizeof optval);
2241 			break;
2242 		case TCP_MAXSEG:
2243 			optval = tp->t_maxseg;
2244 			INP_WUNLOCK(inp);
2245 			error = sooptcopyout(sopt, &optval, sizeof optval);
2246 			break;
2247 		case TCP_NOOPT:
2248 			optval = tp->t_flags & TF_NOOPT;
2249 			INP_WUNLOCK(inp);
2250 			error = sooptcopyout(sopt, &optval, sizeof optval);
2251 			break;
2252 		case TCP_NOPUSH:
2253 			optval = tp->t_flags & TF_NOPUSH;
2254 			INP_WUNLOCK(inp);
2255 			error = sooptcopyout(sopt, &optval, sizeof optval);
2256 			break;
2257 		case TCP_INFO:
2258 			tcp_fill_info(tp, &ti);
2259 			INP_WUNLOCK(inp);
2260 			error = sooptcopyout(sopt, &ti, sizeof ti);
2261 			break;
2262 		case TCP_STATS:
2263 			{
2264 #ifdef STATS
2265 			int nheld;
2266 			TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2267 
2268 			error = 0;
2269 			socklen_t outsbsz = sopt->sopt_valsize;
2270 			if (tp->t_stats == NULL)
2271 				error = ENOENT;
2272 			else if (outsbsz >= tp->t_stats->cursz)
2273 				outsbsz = tp->t_stats->cursz;
2274 			else if (outsbsz >= sizeof(struct statsblob))
2275 				outsbsz = sizeof(struct statsblob);
2276 			else
2277 				error = EINVAL;
2278 			INP_WUNLOCK(inp);
2279 			if (error)
2280 				break;
2281 
2282 			sbp = sopt->sopt_val;
2283 			nheld = atop(round_page(((vm_offset_t)sbp) +
2284 			    (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2285 			vm_page_t ma[nheld];
2286 			if (vm_fault_quick_hold_pages(
2287 			    &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2288 			    outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2289 			    nheld) < 0) {
2290 				error = EFAULT;
2291 				break;
2292 			}
2293 
2294 			if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2295 			    SIZEOF_MEMBER(struct statsblob, flags))))
2296 				goto unhold;
2297 
2298 			INP_WLOCK_RECHECK(inp);
2299 			error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2300 			    sbflags | SB_CLONE_USRDSTNOFAULT);
2301 			INP_WUNLOCK(inp);
2302 			sopt->sopt_valsize = outsbsz;
2303 unhold:
2304 			vm_page_unhold_pages(ma, nheld);
2305 #else
2306 			INP_WUNLOCK(inp);
2307 			error = EOPNOTSUPP;
2308 #endif /* !STATS */
2309 			break;
2310 			}
2311 		case TCP_CONGESTION:
2312 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2313 			INP_WUNLOCK(inp);
2314 			error = sooptcopyout(sopt, buf, len + 1);
2315 			break;
2316 		case TCP_KEEPIDLE:
2317 		case TCP_KEEPINTVL:
2318 		case TCP_KEEPINIT:
2319 		case TCP_KEEPCNT:
2320 			switch (sopt->sopt_name) {
2321 			case TCP_KEEPIDLE:
2322 				ui = TP_KEEPIDLE(tp) / hz;
2323 				break;
2324 			case TCP_KEEPINTVL:
2325 				ui = TP_KEEPINTVL(tp) / hz;
2326 				break;
2327 			case TCP_KEEPINIT:
2328 				ui = TP_KEEPINIT(tp) / hz;
2329 				break;
2330 			case TCP_KEEPCNT:
2331 				ui = TP_KEEPCNT(tp);
2332 				break;
2333 			}
2334 			INP_WUNLOCK(inp);
2335 			error = sooptcopyout(sopt, &ui, sizeof(ui));
2336 			break;
2337 #ifdef TCPPCAP
2338 		case TCP_PCAP_OUT:
2339 		case TCP_PCAP_IN:
2340 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2341 					&(tp->t_outpkts) : &(tp->t_inpkts));
2342 			INP_WUNLOCK(inp);
2343 			error = sooptcopyout(sopt, &optval, sizeof optval);
2344 			break;
2345 #endif
2346 		case TCP_FASTOPEN:
2347 			optval = tp->t_flags & TF_FASTOPEN;
2348 			INP_WUNLOCK(inp);
2349 			error = sooptcopyout(sopt, &optval, sizeof optval);
2350 			break;
2351 #ifdef TCP_BLACKBOX
2352 		case TCP_LOG:
2353 			optval = tp->t_logstate;
2354 			INP_WUNLOCK(inp);
2355 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2356 			break;
2357 		case TCP_LOGBUF:
2358 			/* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2359 			error = tcp_log_getlogbuf(sopt, tp);
2360 			break;
2361 		case TCP_LOGID:
2362 			len = tcp_log_get_id(tp, buf);
2363 			INP_WUNLOCK(inp);
2364 			error = sooptcopyout(sopt, buf, len + 1);
2365 			break;
2366 		case TCP_LOGDUMP:
2367 		case TCP_LOGDUMPID:
2368 			INP_WUNLOCK(inp);
2369 			error = EINVAL;
2370 			break;
2371 #endif
2372 #ifdef KERN_TLS
2373 		case TCP_TXTLS_MODE:
2374 			optval = ktls_get_tx_mode(so);
2375 			INP_WUNLOCK(inp);
2376 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2377 			break;
2378 #endif
2379 		default:
2380 			INP_WUNLOCK(inp);
2381 			error = ENOPROTOOPT;
2382 			break;
2383 		}
2384 		break;
2385 	}
2386 	return (error);
2387 }
2388 #undef INP_WLOCK_RECHECK
2389 #undef INP_WLOCK_RECHECK_CLEANUP
2390 
2391 /*
2392  * Attach TCP protocol to socket, allocating
2393  * internet protocol control block, tcp control block,
2394  * bufer space, and entering LISTEN state if to accept connections.
2395  */
2396 static int
2397 tcp_attach(struct socket *so)
2398 {
2399 	struct tcpcb *tp;
2400 	struct inpcb *inp;
2401 	struct epoch_tracker et;
2402 	int error;
2403 
2404 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2405 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
2406 		if (error)
2407 			return (error);
2408 	}
2409 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
2410 	so->so_snd.sb_flags |= SB_AUTOSIZE;
2411 	NET_EPOCH_ENTER(et);
2412 	error = in_pcballoc(so, &V_tcbinfo);
2413 	if (error) {
2414 		NET_EPOCH_EXIT(et);
2415 		return (error);
2416 	}
2417 	inp = sotoinpcb(so);
2418 #ifdef INET6
2419 	if (inp->inp_vflag & INP_IPV6PROTO) {
2420 		inp->inp_vflag |= INP_IPV6;
2421 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
2422 			inp->inp_vflag |= INP_IPV4;
2423 		inp->in6p_hops = -1;	/* use kernel default */
2424 	}
2425 	else
2426 #endif
2427 	inp->inp_vflag |= INP_IPV4;
2428 	tp = tcp_newtcpcb(inp);
2429 	if (tp == NULL) {
2430 		in_pcbdetach(inp);
2431 		in_pcbfree(inp);
2432 		NET_EPOCH_EXIT(et);
2433 		return (ENOBUFS);
2434 	}
2435 	tp->t_state = TCPS_CLOSED;
2436 	INP_WUNLOCK(inp);
2437 	NET_EPOCH_EXIT(et);
2438 	TCPSTATES_INC(TCPS_CLOSED);
2439 	return (0);
2440 }
2441 
2442 /*
2443  * Initiate (or continue) disconnect.
2444  * If embryonic state, just send reset (once).
2445  * If in ``let data drain'' option and linger null, just drop.
2446  * Otherwise (hard), mark socket disconnecting and drop
2447  * current input data; switch states based on user close, and
2448  * send segment to peer (with FIN).
2449  */
2450 static void
2451 tcp_disconnect(struct tcpcb *tp)
2452 {
2453 	struct inpcb *inp = tp->t_inpcb;
2454 	struct socket *so = inp->inp_socket;
2455 
2456 	NET_EPOCH_ASSERT();
2457 	INP_WLOCK_ASSERT(inp);
2458 
2459 	/*
2460 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2461 	 * socket is still open.
2462 	 */
2463 	if (tp->t_state < TCPS_ESTABLISHED &&
2464 	    !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2465 		tp = tcp_close(tp);
2466 		KASSERT(tp != NULL,
2467 		    ("tcp_disconnect: tcp_close() returned NULL"));
2468 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2469 		tp = tcp_drop(tp, 0);
2470 		KASSERT(tp != NULL,
2471 		    ("tcp_disconnect: tcp_drop() returned NULL"));
2472 	} else {
2473 		soisdisconnecting(so);
2474 		sbflush(&so->so_rcv);
2475 		tcp_usrclosed(tp);
2476 		if (!(inp->inp_flags & INP_DROPPED))
2477 			tp->t_fb->tfb_tcp_output(tp);
2478 	}
2479 }
2480 
2481 /*
2482  * User issued close, and wish to trail through shutdown states:
2483  * if never received SYN, just forget it.  If got a SYN from peer,
2484  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2485  * If already got a FIN from peer, then almost done; go to LAST_ACK
2486  * state.  In all other cases, have already sent FIN to peer (e.g.
2487  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2488  * for peer to send FIN or not respond to keep-alives, etc.
2489  * We can let the user exit from the close as soon as the FIN is acked.
2490  */
2491 static void
2492 tcp_usrclosed(struct tcpcb *tp)
2493 {
2494 
2495 	NET_EPOCH_ASSERT();
2496 	INP_WLOCK_ASSERT(tp->t_inpcb);
2497 
2498 	switch (tp->t_state) {
2499 	case TCPS_LISTEN:
2500 #ifdef TCP_OFFLOAD
2501 		tcp_offload_listen_stop(tp);
2502 #endif
2503 		tcp_state_change(tp, TCPS_CLOSED);
2504 		/* FALLTHROUGH */
2505 	case TCPS_CLOSED:
2506 		tp = tcp_close(tp);
2507 		/*
2508 		 * tcp_close() should never return NULL here as the socket is
2509 		 * still open.
2510 		 */
2511 		KASSERT(tp != NULL,
2512 		    ("tcp_usrclosed: tcp_close() returned NULL"));
2513 		break;
2514 
2515 	case TCPS_SYN_SENT:
2516 	case TCPS_SYN_RECEIVED:
2517 		tp->t_flags |= TF_NEEDFIN;
2518 		break;
2519 
2520 	case TCPS_ESTABLISHED:
2521 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
2522 		break;
2523 
2524 	case TCPS_CLOSE_WAIT:
2525 		tcp_state_change(tp, TCPS_LAST_ACK);
2526 		break;
2527 	}
2528 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
2529 		soisdisconnected(tp->t_inpcb->inp_socket);
2530 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
2531 		if (tp->t_state == TCPS_FIN_WAIT_2) {
2532 			int timeout;
2533 
2534 			timeout = (tcp_fast_finwait2_recycle) ?
2535 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
2536 			tcp_timer_activate(tp, TT_2MSL, timeout);
2537 		}
2538 	}
2539 }
2540 
2541 #ifdef DDB
2542 static void
2543 db_print_indent(int indent)
2544 {
2545 	int i;
2546 
2547 	for (i = 0; i < indent; i++)
2548 		db_printf(" ");
2549 }
2550 
2551 static void
2552 db_print_tstate(int t_state)
2553 {
2554 
2555 	switch (t_state) {
2556 	case TCPS_CLOSED:
2557 		db_printf("TCPS_CLOSED");
2558 		return;
2559 
2560 	case TCPS_LISTEN:
2561 		db_printf("TCPS_LISTEN");
2562 		return;
2563 
2564 	case TCPS_SYN_SENT:
2565 		db_printf("TCPS_SYN_SENT");
2566 		return;
2567 
2568 	case TCPS_SYN_RECEIVED:
2569 		db_printf("TCPS_SYN_RECEIVED");
2570 		return;
2571 
2572 	case TCPS_ESTABLISHED:
2573 		db_printf("TCPS_ESTABLISHED");
2574 		return;
2575 
2576 	case TCPS_CLOSE_WAIT:
2577 		db_printf("TCPS_CLOSE_WAIT");
2578 		return;
2579 
2580 	case TCPS_FIN_WAIT_1:
2581 		db_printf("TCPS_FIN_WAIT_1");
2582 		return;
2583 
2584 	case TCPS_CLOSING:
2585 		db_printf("TCPS_CLOSING");
2586 		return;
2587 
2588 	case TCPS_LAST_ACK:
2589 		db_printf("TCPS_LAST_ACK");
2590 		return;
2591 
2592 	case TCPS_FIN_WAIT_2:
2593 		db_printf("TCPS_FIN_WAIT_2");
2594 		return;
2595 
2596 	case TCPS_TIME_WAIT:
2597 		db_printf("TCPS_TIME_WAIT");
2598 		return;
2599 
2600 	default:
2601 		db_printf("unknown");
2602 		return;
2603 	}
2604 }
2605 
2606 static void
2607 db_print_tflags(u_int t_flags)
2608 {
2609 	int comma;
2610 
2611 	comma = 0;
2612 	if (t_flags & TF_ACKNOW) {
2613 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
2614 		comma = 1;
2615 	}
2616 	if (t_flags & TF_DELACK) {
2617 		db_printf("%sTF_DELACK", comma ? ", " : "");
2618 		comma = 1;
2619 	}
2620 	if (t_flags & TF_NODELAY) {
2621 		db_printf("%sTF_NODELAY", comma ? ", " : "");
2622 		comma = 1;
2623 	}
2624 	if (t_flags & TF_NOOPT) {
2625 		db_printf("%sTF_NOOPT", comma ? ", " : "");
2626 		comma = 1;
2627 	}
2628 	if (t_flags & TF_SENTFIN) {
2629 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2630 		comma = 1;
2631 	}
2632 	if (t_flags & TF_REQ_SCALE) {
2633 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2634 		comma = 1;
2635 	}
2636 	if (t_flags & TF_RCVD_SCALE) {
2637 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2638 		comma = 1;
2639 	}
2640 	if (t_flags & TF_REQ_TSTMP) {
2641 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2642 		comma = 1;
2643 	}
2644 	if (t_flags & TF_RCVD_TSTMP) {
2645 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2646 		comma = 1;
2647 	}
2648 	if (t_flags & TF_SACK_PERMIT) {
2649 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2650 		comma = 1;
2651 	}
2652 	if (t_flags & TF_NEEDSYN) {
2653 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2654 		comma = 1;
2655 	}
2656 	if (t_flags & TF_NEEDFIN) {
2657 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2658 		comma = 1;
2659 	}
2660 	if (t_flags & TF_NOPUSH) {
2661 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2662 		comma = 1;
2663 	}
2664 	if (t_flags & TF_MORETOCOME) {
2665 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2666 		comma = 1;
2667 	}
2668 	if (t_flags & TF_LQ_OVERFLOW) {
2669 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2670 		comma = 1;
2671 	}
2672 	if (t_flags & TF_LASTIDLE) {
2673 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2674 		comma = 1;
2675 	}
2676 	if (t_flags & TF_RXWIN0SENT) {
2677 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2678 		comma = 1;
2679 	}
2680 	if (t_flags & TF_FASTRECOVERY) {
2681 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2682 		comma = 1;
2683 	}
2684 	if (t_flags & TF_CONGRECOVERY) {
2685 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2686 		comma = 1;
2687 	}
2688 	if (t_flags & TF_WASFRECOVERY) {
2689 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2690 		comma = 1;
2691 	}
2692 	if (t_flags & TF_SIGNATURE) {
2693 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2694 		comma = 1;
2695 	}
2696 	if (t_flags & TF_FORCEDATA) {
2697 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2698 		comma = 1;
2699 	}
2700 	if (t_flags & TF_TSO) {
2701 		db_printf("%sTF_TSO", comma ? ", " : "");
2702 		comma = 1;
2703 	}
2704 	if (t_flags & TF_FASTOPEN) {
2705 		db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2706 		comma = 1;
2707 	}
2708 }
2709 
2710 static void
2711 db_print_tflags2(u_int t_flags2)
2712 {
2713 	int comma;
2714 
2715 	comma = 0;
2716 	if (t_flags2 & TF2_ECN_PERMIT) {
2717 		db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2718 		comma = 1;
2719 	}
2720 }
2721 
2722 
2723 static void
2724 db_print_toobflags(char t_oobflags)
2725 {
2726 	int comma;
2727 
2728 	comma = 0;
2729 	if (t_oobflags & TCPOOB_HAVEDATA) {
2730 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2731 		comma = 1;
2732 	}
2733 	if (t_oobflags & TCPOOB_HADDATA) {
2734 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2735 		comma = 1;
2736 	}
2737 }
2738 
2739 static void
2740 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2741 {
2742 
2743 	db_print_indent(indent);
2744 	db_printf("%s at %p\n", name, tp);
2745 
2746 	indent += 2;
2747 
2748 	db_print_indent(indent);
2749 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
2750 	   TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2751 
2752 	db_print_indent(indent);
2753 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
2754 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2755 
2756 	db_print_indent(indent);
2757 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2758 	    &tp->t_timers->tt_delack, tp->t_inpcb);
2759 
2760 	db_print_indent(indent);
2761 	db_printf("t_state: %d (", tp->t_state);
2762 	db_print_tstate(tp->t_state);
2763 	db_printf(")\n");
2764 
2765 	db_print_indent(indent);
2766 	db_printf("t_flags: 0x%x (", tp->t_flags);
2767 	db_print_tflags(tp->t_flags);
2768 	db_printf(")\n");
2769 
2770 	db_print_indent(indent);
2771 	db_printf("t_flags2: 0x%x (", tp->t_flags2);
2772 	db_print_tflags2(tp->t_flags2);
2773 	db_printf(")\n");
2774 
2775 	db_print_indent(indent);
2776 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
2777 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
2778 
2779 	db_print_indent(indent);
2780 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
2781 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2782 
2783 	db_print_indent(indent);
2784 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
2785 	    tp->iss, tp->irs, tp->rcv_nxt);
2786 
2787 	db_print_indent(indent);
2788 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
2789 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2790 
2791 	db_print_indent(indent);
2792 	db_printf("snd_wnd: %u   snd_cwnd: %u\n",
2793 	   tp->snd_wnd, tp->snd_cwnd);
2794 
2795 	db_print_indent(indent);
2796 	db_printf("snd_ssthresh: %u   snd_recover: "
2797 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2798 
2799 	db_print_indent(indent);
2800 	db_printf("t_rcvtime: %u   t_startime: %u\n",
2801 	    tp->t_rcvtime, tp->t_starttime);
2802 
2803 	db_print_indent(indent);
2804 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
2805 	    tp->t_rtttime, tp->t_rtseq);
2806 
2807 	db_print_indent(indent);
2808 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
2809 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
2810 
2811 	db_print_indent(indent);
2812 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
2813 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
2814 	    tp->t_rttbest);
2815 
2816 	db_print_indent(indent);
2817 	db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
2818 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2819 
2820 	db_print_indent(indent);
2821 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2822 	db_print_toobflags(tp->t_oobflags);
2823 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
2824 
2825 	db_print_indent(indent);
2826 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
2827 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2828 
2829 	db_print_indent(indent);
2830 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
2831 	    tp->ts_recent, tp->ts_recent_age);
2832 
2833 	db_print_indent(indent);
2834 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
2835 	    "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2836 
2837 	db_print_indent(indent);
2838 	db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
2839 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2840 	    tp->snd_recover_prev, tp->t_badrxtwin);
2841 
2842 	db_print_indent(indent);
2843 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
2844 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2845 
2846 	db_print_indent(indent);
2847 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d   sack_newdata: "
2848 	    "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2849 
2850 	/* Skip sackblks, sackhint. */
2851 
2852 	db_print_indent(indent);
2853 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
2854 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2855 }
2856 
2857 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2858 {
2859 	struct tcpcb *tp;
2860 
2861 	if (!have_addr) {
2862 		db_printf("usage: show tcpcb <addr>\n");
2863 		return;
2864 	}
2865 	tp = (struct tcpcb *)addr;
2866 
2867 	db_print_tcpcb(tp, "tcpcb", 0);
2868 }
2869 #endif
2870