xref: /freebsd/sys/netinet/tcp_usrreq.c (revision 1323ec571215a77ddd21294f0871979d5ad6b992)
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/stats.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif
74 
75 #include <net/if.h>
76 #include <net/if_var.h>
77 #include <net/route.h>
78 #include <net/vnet.h>
79 
80 #include <netinet/in.h>
81 #include <netinet/in_kdtrace.h>
82 #include <netinet/in_pcb.h>
83 #include <netinet/in_systm.h>
84 #include <netinet/in_var.h>
85 #include <netinet/ip.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 #ifdef INET
125 static int	tcp_connect(struct tcpcb *, struct sockaddr *,
126 		    struct thread *td);
127 #endif /* INET */
128 #ifdef INET6
129 static int	tcp6_connect(struct tcpcb *, struct sockaddr *,
130 		    struct thread *td);
131 #endif /* INET6 */
132 static void	tcp_disconnect(struct tcpcb *);
133 static void	tcp_usrclosed(struct tcpcb *);
134 static void	tcp_fill_info(struct tcpcb *, struct tcp_info *);
135 
136 static int	tcp_pru_options_support(struct tcpcb *tp, int flags);
137 
138 #ifdef TCPDEBUG
139 #define	TCPDEBUG0	int ostate = 0
140 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
141 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
142 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
143 #else
144 #define	TCPDEBUG0
145 #define	TCPDEBUG1()
146 #define	TCPDEBUG2(req)
147 #endif
148 
149 /*
150  * tcp_require_unique port requires a globally-unique source port for each
151  * outgoing connection.  The default is to require the 4-tuple to be unique.
152  */
153 VNET_DEFINE(int, tcp_require_unique_port) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port,
155     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0,
156     "Require globally-unique ephemeral port for outgoing connections");
157 #define	V_tcp_require_unique_port	VNET(tcp_require_unique_port)
158 
159 /*
160  * TCP attaches to socket via pru_attach(), reserving space,
161  * and an internet control block.
162  */
163 static int
164 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
165 {
166 	struct inpcb *inp;
167 	struct tcpcb *tp = NULL;
168 	int error;
169 	TCPDEBUG0;
170 
171 	inp = sotoinpcb(so);
172 	KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
173 	TCPDEBUG1();
174 
175 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
176 		error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
177 		if (error)
178 			goto out;
179 	}
180 
181 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
182 	so->so_snd.sb_flags |= SB_AUTOSIZE;
183 	error = in_pcballoc(so, &V_tcbinfo);
184 	if (error)
185 		goto out;
186 	inp = sotoinpcb(so);
187 #ifdef INET6
188 	if (inp->inp_vflag & INP_IPV6PROTO) {
189 		inp->inp_vflag |= INP_IPV6;
190 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
191 			inp->inp_vflag |= INP_IPV4;
192 		inp->in6p_hops = -1;	/* use kernel default */
193 	}
194 	else
195 #endif
196 		inp->inp_vflag |= INP_IPV4;
197 	tp = tcp_newtcpcb(inp);
198 	if (tp == NULL) {
199 		error = ENOBUFS;
200 		in_pcbdetach(inp);
201 		in_pcbfree(inp);
202 		goto out;
203 	}
204 	tp->t_state = TCPS_CLOSED;
205 	INP_WUNLOCK(inp);
206 	TCPSTATES_INC(TCPS_CLOSED);
207 out:
208 	TCPDEBUG2(PRU_ATTACH);
209 	TCP_PROBE2(debug__user, tp, PRU_ATTACH);
210 	return (error);
211 }
212 
213 /*
214  * tcp_usr_detach is called when the socket layer loses its final reference
215  * to the socket, be it a file descriptor reference, a reference from TCP,
216  * etc.  At this point, there is only one case in which we will keep around
217  * inpcb state: time wait.
218  */
219 static void
220 tcp_usr_detach(struct socket *so)
221 {
222 	struct inpcb *inp;
223 	struct tcpcb *tp;
224 
225 	inp = sotoinpcb(so);
226 	KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
227 	INP_WLOCK(inp);
228 	KASSERT(so->so_pcb == inp && inp->inp_socket == so,
229 		("%s: socket %p inp %p mismatch", __func__, so, inp));
230 
231 	tp = intotcpcb(inp);
232 
233 	if (inp->inp_flags & INP_TIMEWAIT) {
234 		/*
235 		 * There are two cases to handle: one in which the time wait
236 		 * state is being discarded (INP_DROPPED), and one in which
237 		 * this connection will remain in timewait.  In the former,
238 		 * it is time to discard all state (except tcptw, which has
239 		 * already been discarded by the timewait close code, which
240 		 * should be further up the call stack somewhere).  In the
241 		 * latter case, we detach from the socket, but leave the pcb
242 		 * present until timewait ends.
243 		 *
244 		 * XXXRW: Would it be cleaner to free the tcptw here?
245 		 *
246 		 * Astute question indeed, from twtcp perspective there are
247 		 * four cases to consider:
248 		 *
249 		 * #1 tcp_usr_detach is called at tcptw creation time by
250 		 *  tcp_twstart, then do not discard the newly created tcptw
251 		 *  and leave inpcb present until timewait ends
252 		 * #2 tcp_usr_detach is called at tcptw creation time by
253 		 *  tcp_twstart, but connection is local and tw will be
254 		 *  discarded immediately
255 		 * #3 tcp_usr_detach is called at timewait end (or reuse) by
256 		 *  tcp_twclose, then the tcptw has already been discarded
257 		 *  (or reused) and inpcb is freed here
258 		 * #4 tcp_usr_detach is called() after timewait ends (or reuse)
259 		 *  (e.g. by soclose), then tcptw has already been discarded
260 		 *  (or reused) and inpcb is freed here
261 		 *
262 		 *  In all three cases the tcptw should not be freed here.
263 		 */
264 		if (inp->inp_flags & INP_DROPPED) {
265 			KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
266 			    "INP_DROPPED && tp != NULL"));
267 			in_pcbdetach(inp);
268 			in_pcbfree(inp);
269 		} else {
270 			in_pcbdetach(inp);
271 			INP_WUNLOCK(inp);
272 		}
273 	} else {
274 		/*
275 		 * If the connection is not in timewait, we consider two
276 		 * two conditions: one in which no further processing is
277 		 * necessary (dropped || embryonic), and one in which TCP is
278 		 * not yet done, but no longer requires the socket, so the
279 		 * pcb will persist for the time being.
280 		 *
281 		 * XXXRW: Does the second case still occur?
282 		 */
283 		if (inp->inp_flags & INP_DROPPED ||
284 		    tp->t_state < TCPS_SYN_SENT) {
285 			tcp_discardcb(tp);
286 			in_pcbdetach(inp);
287 			in_pcbfree(inp);
288 		} else {
289 			in_pcbdetach(inp);
290 			INP_WUNLOCK(inp);
291 		}
292 	}
293 }
294 
295 #ifdef INET
296 /*
297  * Give the socket an address.
298  */
299 static int
300 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
301 {
302 	int error = 0;
303 	struct inpcb *inp;
304 	struct tcpcb *tp = NULL;
305 	struct sockaddr_in *sinp;
306 
307 	sinp = (struct sockaddr_in *)nam;
308 	if (nam->sa_family != AF_INET) {
309 		/*
310 		 * Preserve compatibility with old programs.
311 		 */
312 		if (nam->sa_family != AF_UNSPEC ||
313 		    nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
314 		    sinp->sin_addr.s_addr != INADDR_ANY)
315 			return (EAFNOSUPPORT);
316 		nam->sa_family = AF_INET;
317 	}
318 	if (nam->sa_len != sizeof(*sinp))
319 		return (EINVAL);
320 
321 	/*
322 	 * Must check for multicast addresses and disallow binding
323 	 * to them.
324 	 */
325 	if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
326 		return (EAFNOSUPPORT);
327 
328 	TCPDEBUG0;
329 	inp = sotoinpcb(so);
330 	KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
331 	INP_WLOCK(inp);
332 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
333 		error = EINVAL;
334 		goto out;
335 	}
336 	tp = intotcpcb(inp);
337 	TCPDEBUG1();
338 	INP_HASH_WLOCK(&V_tcbinfo);
339 	error = in_pcbbind(inp, nam, td->td_ucred);
340 	INP_HASH_WUNLOCK(&V_tcbinfo);
341 out:
342 	TCPDEBUG2(PRU_BIND);
343 	TCP_PROBE2(debug__user, tp, PRU_BIND);
344 	INP_WUNLOCK(inp);
345 
346 	return (error);
347 }
348 #endif /* INET */
349 
350 #ifdef INET6
351 static int
352 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
353 {
354 	int error = 0;
355 	struct inpcb *inp;
356 	struct tcpcb *tp = NULL;
357 	struct sockaddr_in6 *sin6;
358 	u_char vflagsav;
359 
360 	sin6 = (struct sockaddr_in6 *)nam;
361 	if (nam->sa_family != AF_INET6)
362 		return (EAFNOSUPPORT);
363 	if (nam->sa_len != sizeof(*sin6))
364 		return (EINVAL);
365 
366 	/*
367 	 * Must check for multicast addresses and disallow binding
368 	 * to them.
369 	 */
370 	if (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 	if (error != 0) {
445 		SOCK_UNLOCK(so);
446 		goto out;
447 	}
448 	if (inp->inp_lport == 0) {
449 		INP_HASH_WLOCK(&V_tcbinfo);
450 		error = in_pcbbind(inp, NULL, td->td_ucred);
451 		INP_HASH_WUNLOCK(&V_tcbinfo);
452 	}
453 	if (error == 0) {
454 		tcp_state_change(tp, TCPS_LISTEN);
455 		solisten_proto(so, backlog);
456 #ifdef TCP_OFFLOAD
457 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
458 			tcp_offload_listen_start(tp);
459 #endif
460 	} else {
461 		solisten_proto_abort(so);
462 	}
463 	SOCK_UNLOCK(so);
464 
465 	if (IS_FASTOPEN(tp->t_flags))
466 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
467 
468 out:
469 	TCPDEBUG2(PRU_LISTEN);
470 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
471 	INP_WUNLOCK(inp);
472 	return (error);
473 }
474 #endif /* INET */
475 
476 #ifdef INET6
477 static int
478 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
479 {
480 	int error = 0;
481 	struct inpcb *inp;
482 	struct tcpcb *tp = NULL;
483 	u_char vflagsav;
484 
485 	TCPDEBUG0;
486 	inp = sotoinpcb(so);
487 	KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
488 	INP_WLOCK(inp);
489 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
490 		error = EINVAL;
491 		goto out;
492 	}
493 	vflagsav = inp->inp_vflag;
494 	tp = intotcpcb(inp);
495 	TCPDEBUG1();
496 	SOCK_LOCK(so);
497 	error = solisten_proto_check(so);
498 	if (error != 0) {
499 		SOCK_UNLOCK(so);
500 		goto out;
501 	}
502 	INP_HASH_WLOCK(&V_tcbinfo);
503 	if (inp->inp_lport == 0) {
504 		inp->inp_vflag &= ~INP_IPV4;
505 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
506 			inp->inp_vflag |= INP_IPV4;
507 		error = in6_pcbbind(inp, NULL, td->td_ucred);
508 	}
509 	INP_HASH_WUNLOCK(&V_tcbinfo);
510 	if (error == 0) {
511 		tcp_state_change(tp, TCPS_LISTEN);
512 		solisten_proto(so, backlog);
513 #ifdef TCP_OFFLOAD
514 		if ((so->so_options & SO_NO_OFFLOAD) == 0)
515 			tcp_offload_listen_start(tp);
516 #endif
517 	} else {
518 		solisten_proto_abort(so);
519 	}
520 	SOCK_UNLOCK(so);
521 
522 	if (IS_FASTOPEN(tp->t_flags))
523 		tp->t_tfo_pending = tcp_fastopen_alloc_counter();
524 
525 	if (error != 0)
526 		inp->inp_vflag = vflagsav;
527 
528 out:
529 	TCPDEBUG2(PRU_LISTEN);
530 	TCP_PROBE2(debug__user, tp, PRU_LISTEN);
531 	INP_WUNLOCK(inp);
532 	return (error);
533 }
534 #endif /* INET6 */
535 
536 #ifdef INET
537 /*
538  * Initiate connection to peer.
539  * Create a template for use in transmissions on this connection.
540  * Enter SYN_SENT state, and mark socket as connecting.
541  * Start keep-alive timer, and seed output sequence space.
542  * Send initial segment on connection.
543  */
544 static int
545 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
546 {
547 	struct epoch_tracker et;
548 	int error = 0;
549 	struct inpcb *inp;
550 	struct tcpcb *tp = NULL;
551 	struct sockaddr_in *sinp;
552 
553 	sinp = (struct sockaddr_in *)nam;
554 	if (nam->sa_family != AF_INET)
555 		return (EAFNOSUPPORT);
556 	if (nam->sa_len != sizeof (*sinp))
557 		return (EINVAL);
558 
559 	/*
560 	 * Must disallow TCP ``connections'' to multicast addresses.
561 	 */
562 	if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
563 		return (EAFNOSUPPORT);
564 	if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST)
565 		return (EACCES);
566 	if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
567 		return (error);
568 
569 	TCPDEBUG0;
570 	inp = sotoinpcb(so);
571 	KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
572 	INP_WLOCK(inp);
573 	if (inp->inp_flags & INP_TIMEWAIT) {
574 		error = EADDRINUSE;
575 		goto out;
576 	}
577 	if (inp->inp_flags & INP_DROPPED) {
578 		error = ECONNREFUSED;
579 		goto out;
580 	}
581 	if (SOLISTENING(so)) {
582 		error = EOPNOTSUPP;
583 		goto out;
584 	}
585 	tp = intotcpcb(inp);
586 	TCPDEBUG1();
587 	NET_EPOCH_ENTER(et);
588 	if ((error = tcp_connect(tp, nam, td)) != 0)
589 		goto out_in_epoch;
590 #ifdef TCP_OFFLOAD
591 	if (registered_toedevs > 0 &&
592 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
593 	    (error = tcp_offload_connect(so, nam)) == 0)
594 		goto out_in_epoch;
595 #endif
596 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
597 	error = tcp_output(tp);
598 	KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
599 	    ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
600 out_in_epoch:
601 	NET_EPOCH_EXIT(et);
602 out:
603 	TCPDEBUG2(PRU_CONNECT);
604 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
605 	INP_WUNLOCK(inp);
606 	return (error);
607 }
608 #endif /* INET */
609 
610 #ifdef INET6
611 static int
612 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
613 {
614 	struct epoch_tracker et;
615 	int error = 0;
616 	struct inpcb *inp;
617 	struct tcpcb *tp = NULL;
618 	struct sockaddr_in6 *sin6;
619 	u_int8_t incflagsav;
620 	u_char vflagsav;
621 
622 	TCPDEBUG0;
623 
624 	sin6 = (struct sockaddr_in6 *)nam;
625 	if (nam->sa_family != AF_INET6)
626 		return (EAFNOSUPPORT);
627 	if (nam->sa_len != sizeof (*sin6))
628 		return (EINVAL);
629 
630 	/*
631 	 * Must disallow TCP ``connections'' to multicast addresses.
632 	 */
633 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
634 		return (EAFNOSUPPORT);
635 
636 	inp = sotoinpcb(so);
637 	KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
638 	INP_WLOCK(inp);
639 	vflagsav = inp->inp_vflag;
640 	incflagsav = inp->inp_inc.inc_flags;
641 	if (inp->inp_flags & INP_TIMEWAIT) {
642 		error = EADDRINUSE;
643 		goto out;
644 	}
645 	if (inp->inp_flags & INP_DROPPED) {
646 		error = ECONNREFUSED;
647 		goto out;
648 	}
649 	if (SOLISTENING(so)) {
650 		error = EINVAL;
651 		goto out;
652 	}
653 	tp = intotcpcb(inp);
654 	TCPDEBUG1();
655 #ifdef INET
656 	/*
657 	 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
658 	 * therefore probably require the hash lock, which isn't held here.
659 	 * Is this a significant problem?
660 	 */
661 	if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
662 		struct sockaddr_in sin;
663 
664 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
665 			error = EINVAL;
666 			goto out;
667 		}
668 		if ((inp->inp_vflag & INP_IPV4) == 0) {
669 			error = EAFNOSUPPORT;
670 			goto out;
671 		}
672 
673 		in6_sin6_2_sin(&sin, sin6);
674 		if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
675 			error = EAFNOSUPPORT;
676 			goto out;
677 		}
678 		if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
679 			error = EACCES;
680 			goto out;
681 		}
682 		if ((error = prison_remote_ip4(td->td_ucred,
683 		    &sin.sin_addr)) != 0)
684 			goto out;
685 		inp->inp_vflag |= INP_IPV4;
686 		inp->inp_vflag &= ~INP_IPV6;
687 		NET_EPOCH_ENTER(et);
688 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
689 			goto out_in_epoch;
690 #ifdef TCP_OFFLOAD
691 		if (registered_toedevs > 0 &&
692 		    (so->so_options & SO_NO_OFFLOAD) == 0 &&
693 		    (error = tcp_offload_connect(so, nam)) == 0)
694 			goto out_in_epoch;
695 #endif
696 		error = tcp_output(tp);
697 		goto out_in_epoch;
698 	} else {
699 		if ((inp->inp_vflag & INP_IPV6) == 0) {
700 			error = EAFNOSUPPORT;
701 			goto out;
702 		}
703 	}
704 #endif
705 	if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
706 		goto out;
707 	inp->inp_vflag &= ~INP_IPV4;
708 	inp->inp_vflag |= INP_IPV6;
709 	inp->inp_inc.inc_flags |= INC_ISIPV6;
710 	if ((error = tcp6_connect(tp, nam, td)) != 0)
711 		goto out;
712 #ifdef TCP_OFFLOAD
713 	if (registered_toedevs > 0 &&
714 	    (so->so_options & SO_NO_OFFLOAD) == 0 &&
715 	    (error = tcp_offload_connect(so, nam)) == 0)
716 		goto out;
717 #endif
718 	tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
719 	NET_EPOCH_ENTER(et);
720 	error = tcp_output(tp);
721 #ifdef INET
722 out_in_epoch:
723 #endif
724 	NET_EPOCH_EXIT(et);
725 out:
726 	KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
727 	    ", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
728 	/*
729 	 * If the implicit bind in the connect call fails, restore
730 	 * the flags we modified.
731 	 */
732 	if (error != 0 && inp->inp_lport == 0) {
733 		inp->inp_vflag = vflagsav;
734 		inp->inp_inc.inc_flags = incflagsav;
735 	}
736 
737 	TCPDEBUG2(PRU_CONNECT);
738 	TCP_PROBE2(debug__user, tp, PRU_CONNECT);
739 	INP_WUNLOCK(inp);
740 	return (error);
741 }
742 #endif /* INET6 */
743 
744 /*
745  * Initiate disconnect from peer.
746  * If connection never passed embryonic stage, just drop;
747  * else if don't need to let data drain, then can just drop anyways,
748  * else have to begin TCP shutdown process: mark socket disconnecting,
749  * drain unread data, state switch to reflect user close, and
750  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
751  * when peer sends FIN and acks ours.
752  *
753  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
754  */
755 static int
756 tcp_usr_disconnect(struct socket *so)
757 {
758 	struct inpcb *inp;
759 	struct tcpcb *tp = NULL;
760 	struct epoch_tracker et;
761 	int error = 0;
762 
763 	TCPDEBUG0;
764 	NET_EPOCH_ENTER(et);
765 	inp = sotoinpcb(so);
766 	KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
767 	INP_WLOCK(inp);
768 	if (inp->inp_flags & INP_TIMEWAIT)
769 		goto out;
770 	if (inp->inp_flags & INP_DROPPED) {
771 		error = ECONNRESET;
772 		goto out;
773 	}
774 	tp = intotcpcb(inp);
775 	TCPDEBUG1();
776 	tcp_disconnect(tp);
777 out:
778 	TCPDEBUG2(PRU_DISCONNECT);
779 	TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
780 	INP_WUNLOCK(inp);
781 	NET_EPOCH_EXIT(et);
782 	return (error);
783 }
784 
785 #ifdef INET
786 /*
787  * Accept a connection.  Essentially all the work is done at higher levels;
788  * just return the address of the peer, storing through addr.
789  */
790 static int
791 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
792 {
793 	int error = 0;
794 	struct inpcb *inp = NULL;
795 	struct tcpcb *tp = NULL;
796 	struct in_addr addr;
797 	in_port_t port = 0;
798 	TCPDEBUG0;
799 
800 	if (so->so_state & SS_ISDISCONNECTED)
801 		return (ECONNABORTED);
802 
803 	inp = sotoinpcb(so);
804 	KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
805 	INP_WLOCK(inp);
806 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
807 		error = ECONNABORTED;
808 		goto out;
809 	}
810 	tp = intotcpcb(inp);
811 	TCPDEBUG1();
812 
813 	/*
814 	 * We inline in_getpeeraddr and COMMON_END here, so that we can
815 	 * copy the data of interest and defer the malloc until after we
816 	 * release the lock.
817 	 */
818 	port = inp->inp_fport;
819 	addr = inp->inp_faddr;
820 
821 out:
822 	TCPDEBUG2(PRU_ACCEPT);
823 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
824 	INP_WUNLOCK(inp);
825 	if (error == 0)
826 		*nam = in_sockaddr(port, &addr);
827 	return error;
828 }
829 #endif /* INET */
830 
831 #ifdef INET6
832 static int
833 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
834 {
835 	struct inpcb *inp = NULL;
836 	int error = 0;
837 	struct tcpcb *tp = NULL;
838 	struct in_addr addr;
839 	struct in6_addr addr6;
840 	struct epoch_tracker et;
841 	in_port_t port = 0;
842 	int v4 = 0;
843 	TCPDEBUG0;
844 
845 	if (so->so_state & SS_ISDISCONNECTED)
846 		return (ECONNABORTED);
847 
848 	inp = sotoinpcb(so);
849 	KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
850 	NET_EPOCH_ENTER(et);
851 	INP_WLOCK(inp);
852 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
853 		error = ECONNABORTED;
854 		goto out;
855 	}
856 	tp = intotcpcb(inp);
857 	TCPDEBUG1();
858 
859 	/*
860 	 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
861 	 * copy the data of interest and defer the malloc until after we
862 	 * release the lock.
863 	 */
864 	if (inp->inp_vflag & INP_IPV4) {
865 		v4 = 1;
866 		port = inp->inp_fport;
867 		addr = inp->inp_faddr;
868 	} else {
869 		port = inp->inp_fport;
870 		addr6 = inp->in6p_faddr;
871 	}
872 
873 out:
874 	TCPDEBUG2(PRU_ACCEPT);
875 	TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
876 	INP_WUNLOCK(inp);
877 	NET_EPOCH_EXIT(et);
878 	if (error == 0) {
879 		if (v4)
880 			*nam = in6_v4mapsin6_sockaddr(port, &addr);
881 		else
882 			*nam = in6_sockaddr(port, &addr6);
883 	}
884 	return error;
885 }
886 #endif /* INET6 */
887 
888 /*
889  * Mark the connection as being incapable of further output.
890  */
891 static int
892 tcp_usr_shutdown(struct socket *so)
893 {
894 	int error = 0;
895 	struct inpcb *inp;
896 	struct tcpcb *tp = NULL;
897 	struct epoch_tracker et;
898 
899 	TCPDEBUG0;
900 	inp = sotoinpcb(so);
901 	KASSERT(inp != NULL, ("inp == NULL"));
902 	INP_WLOCK(inp);
903 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
904 		INP_WUNLOCK(inp);
905 		return (ECONNRESET);
906 	}
907 	tp = intotcpcb(inp);
908 	NET_EPOCH_ENTER(et);
909 	TCPDEBUG1();
910 	socantsendmore(so);
911 	tcp_usrclosed(tp);
912 	if (!(inp->inp_flags & INP_DROPPED))
913 		error = tcp_output_nodrop(tp);
914 	TCPDEBUG2(PRU_SHUTDOWN);
915 	TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
916 	error = tcp_unlock_or_drop(tp, error);
917 	NET_EPOCH_EXIT(et);
918 
919 	return (error);
920 }
921 
922 /*
923  * After a receive, possibly send window update to peer.
924  */
925 static int
926 tcp_usr_rcvd(struct socket *so, int flags)
927 {
928 	struct epoch_tracker et;
929 	struct inpcb *inp;
930 	struct tcpcb *tp = NULL;
931 	int outrv = 0, error = 0;
932 
933 	TCPDEBUG0;
934 	inp = sotoinpcb(so);
935 	KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
936 	INP_WLOCK(inp);
937 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
938 		INP_WUNLOCK(inp);
939 		return (ECONNRESET);
940 	}
941 	tp = intotcpcb(inp);
942 	NET_EPOCH_ENTER(et);
943 	TCPDEBUG1();
944 	/*
945 	 * For passively-created TFO connections, don't attempt a window
946 	 * update while still in SYN_RECEIVED as this may trigger an early
947 	 * SYN|ACK.  It is preferable to have the SYN|ACK be sent along with
948 	 * application response data, or failing that, when the DELACK timer
949 	 * expires.
950 	 */
951 	if (IS_FASTOPEN(tp->t_flags) &&
952 	    (tp->t_state == TCPS_SYN_RECEIVED))
953 		goto out;
954 #ifdef TCP_OFFLOAD
955 	if (tp->t_flags & TF_TOE)
956 		tcp_offload_rcvd(tp);
957 	else
958 #endif
959 		outrv = tcp_output_nodrop(tp);
960 out:
961 	TCPDEBUG2(PRU_RCVD);
962 	TCP_PROBE2(debug__user, tp, PRU_RCVD);
963 	(void) tcp_unlock_or_drop(tp, outrv);
964 	NET_EPOCH_EXIT(et);
965 	return (error);
966 }
967 
968 /*
969  * Do a send by putting data in output queue and updating urgent
970  * marker if URG set.  Possibly send more data.  Unlike the other
971  * pru_*() routines, the mbuf chains are our responsibility.  We
972  * must either enqueue them or free them.  The other pru_* routines
973  * generally are caller-frees.
974  */
975 static int
976 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
977     struct sockaddr *nam, struct mbuf *control, struct thread *td)
978 {
979 	struct epoch_tracker et;
980 	int error = 0;
981 	struct inpcb *inp;
982 	struct tcpcb *tp = NULL;
983 #ifdef INET
984 #ifdef INET6
985 	struct sockaddr_in sin;
986 #endif
987 	struct sockaddr_in *sinp;
988 #endif
989 #ifdef INET6
990 	int isipv6;
991 #endif
992 	u_int8_t incflagsav;
993 	u_char vflagsav;
994 	bool restoreflags;
995 	TCPDEBUG0;
996 
997 	if (control != NULL) {
998 		/* TCP doesn't do control messages (rights, creds, etc) */
999 		if (control->m_len) {
1000 			m_freem(control);
1001 			return (EINVAL);
1002 		}
1003 		m_freem(control);	/* empty control, just free it */
1004 	}
1005 
1006 	inp = sotoinpcb(so);
1007 	KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
1008 	INP_WLOCK(inp);
1009 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1010 		if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1011 			m_freem(m);
1012 		INP_WUNLOCK(inp);
1013 		return (ECONNRESET);
1014 	}
1015 
1016 	vflagsav = inp->inp_vflag;
1017 	incflagsav = inp->inp_inc.inc_flags;
1018 	restoreflags = false;
1019 	tp = intotcpcb(inp);
1020 
1021 	NET_EPOCH_ENTER(et);
1022 	if ((flags & PRUS_OOB) != 0 &&
1023 	    (error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
1024 		goto out;
1025 
1026 	TCPDEBUG1();
1027 	if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1028 		if (tp->t_state == TCPS_LISTEN) {
1029 			error = EINVAL;
1030 			goto out;
1031 		}
1032 		switch (nam->sa_family) {
1033 #ifdef INET
1034 		case AF_INET:
1035 			sinp = (struct sockaddr_in *)nam;
1036 			if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1037 				error = EINVAL;
1038 				goto out;
1039 			}
1040 			if ((inp->inp_vflag & INP_IPV6) != 0) {
1041 				error = EAFNOSUPPORT;
1042 				goto out;
1043 			}
1044 			if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1045 				error = EAFNOSUPPORT;
1046 				goto out;
1047 			}
1048 			if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1049 				error = EACCES;
1050 				goto out;
1051 			}
1052 			if ((error = prison_remote_ip4(td->td_ucred,
1053 			    &sinp->sin_addr)))
1054 				goto out;
1055 #ifdef INET6
1056 			isipv6 = 0;
1057 #endif
1058 			break;
1059 #endif /* INET */
1060 #ifdef INET6
1061 		case AF_INET6:
1062 		{
1063 			struct sockaddr_in6 *sin6;
1064 
1065 			sin6 = (struct sockaddr_in6 *)nam;
1066 			if (sin6->sin6_len != sizeof(*sin6)) {
1067 				error = EINVAL;
1068 				goto out;
1069 			}
1070 			if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1071 				error = EAFNOSUPPORT;
1072 				goto out;
1073 			}
1074 			if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1075 				error = EAFNOSUPPORT;
1076 				goto out;
1077 			}
1078 			if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1079 #ifdef INET
1080 				if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1081 					error = EINVAL;
1082 					goto out;
1083 				}
1084 				if ((inp->inp_vflag & INP_IPV4) == 0) {
1085 					error = EAFNOSUPPORT;
1086 					goto out;
1087 				}
1088 				restoreflags = true;
1089 				inp->inp_vflag &= ~INP_IPV6;
1090 				sinp = &sin;
1091 				in6_sin6_2_sin(sinp, sin6);
1092 				if (IN_MULTICAST(
1093 				    ntohl(sinp->sin_addr.s_addr))) {
1094 					error = EAFNOSUPPORT;
1095 					goto out;
1096 				}
1097 				if ((error = prison_remote_ip4(td->td_ucred,
1098 				    &sinp->sin_addr)))
1099 					goto out;
1100 				isipv6 = 0;
1101 #else /* !INET */
1102 				error = EAFNOSUPPORT;
1103 				goto out;
1104 #endif /* INET */
1105 			} else {
1106 				if ((inp->inp_vflag & INP_IPV6) == 0) {
1107 					error = EAFNOSUPPORT;
1108 					goto out;
1109 				}
1110 				restoreflags = true;
1111 				inp->inp_vflag &= ~INP_IPV4;
1112 				inp->inp_inc.inc_flags |= INC_ISIPV6;
1113 				if ((error = prison_remote_ip6(td->td_ucred,
1114 				    &sin6->sin6_addr)))
1115 					goto out;
1116 				isipv6 = 1;
1117 			}
1118 			break;
1119 		}
1120 #endif /* INET6 */
1121 		default:
1122 			error = EAFNOSUPPORT;
1123 			goto out;
1124 		}
1125 	}
1126 	if (!(flags & PRUS_OOB)) {
1127 		sbappendstream(&so->so_snd, m, flags);
1128 		m = NULL;
1129 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1130 			KASSERT(tp->t_state == TCPS_CLOSED,
1131 			    ("%s: tp %p is listening", __func__, tp));
1132 
1133 			/*
1134 			 * Do implied connect if not yet connected,
1135 			 * initialize window to default value, and
1136 			 * initialize maxseg using peer's cached MSS.
1137 			 */
1138 #ifdef INET6
1139 			if (isipv6)
1140 				error = tcp6_connect(tp, nam, td);
1141 #endif /* INET6 */
1142 #if defined(INET6) && defined(INET)
1143 			else
1144 #endif
1145 #ifdef INET
1146 				error = tcp_connect(tp,
1147 				    (struct sockaddr *)sinp, td);
1148 #endif
1149 			/*
1150 			 * The bind operation in tcp_connect succeeded. We
1151 			 * no longer want to restore the flags if later
1152 			 * operations fail.
1153 			 */
1154 			if (error == 0 || inp->inp_lport != 0)
1155 				restoreflags = false;
1156 
1157 			if (error) {
1158 				/* m is freed if PRUS_NOTREADY is unset. */
1159 				sbflush(&so->so_snd);
1160 				goto out;
1161 			}
1162 			if (IS_FASTOPEN(tp->t_flags))
1163 				tcp_fastopen_connect(tp);
1164 			else {
1165 				tp->snd_wnd = TTCP_CLIENT_SND_WND;
1166 				tcp_mss(tp, -1);
1167 			}
1168 		}
1169 		if (flags & PRUS_EOF) {
1170 			/*
1171 			 * Close the send side of the connection after
1172 			 * the data is sent.
1173 			 */
1174 			socantsendmore(so);
1175 			tcp_usrclosed(tp);
1176 		}
1177 		if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1178 		    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1179 		    (tp->t_fbyte_out == 0) &&
1180 		    (so->so_snd.sb_ccc > 0)) {
1181 			tp->t_fbyte_out = ticks;
1182 			if (tp->t_fbyte_out == 0)
1183 				tp->t_fbyte_out = 1;
1184 			if (tp->t_fbyte_out && tp->t_fbyte_in)
1185 				tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1186 		}
1187 		if (!(inp->inp_flags & INP_DROPPED) &&
1188 		    !(flags & PRUS_NOTREADY)) {
1189 			if (flags & PRUS_MORETOCOME)
1190 				tp->t_flags |= TF_MORETOCOME;
1191 			error = tcp_output_nodrop(tp);
1192 			if (flags & PRUS_MORETOCOME)
1193 				tp->t_flags &= ~TF_MORETOCOME;
1194 		}
1195 	} else {
1196 		/*
1197 		 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1198 		 */
1199 		SOCKBUF_LOCK(&so->so_snd);
1200 		if (sbspace(&so->so_snd) < -512) {
1201 			SOCKBUF_UNLOCK(&so->so_snd);
1202 			error = ENOBUFS;
1203 			goto out;
1204 		}
1205 		/*
1206 		 * According to RFC961 (Assigned Protocols),
1207 		 * the urgent pointer points to the last octet
1208 		 * of urgent data.  We continue, however,
1209 		 * to consider it to indicate the first octet
1210 		 * of data past the urgent section.
1211 		 * Otherwise, snd_up should be one lower.
1212 		 */
1213 		sbappendstream_locked(&so->so_snd, m, flags);
1214 		SOCKBUF_UNLOCK(&so->so_snd);
1215 		m = NULL;
1216 		if (nam && tp->t_state < TCPS_SYN_SENT) {
1217 			/*
1218 			 * Do implied connect if not yet connected,
1219 			 * initialize window to default value, and
1220 			 * initialize maxseg using peer's cached MSS.
1221 			 */
1222 
1223 			/*
1224 			 * Not going to contemplate SYN|URG
1225 			 */
1226 			if (IS_FASTOPEN(tp->t_flags))
1227 				tp->t_flags &= ~TF_FASTOPEN;
1228 #ifdef INET6
1229 			if (isipv6)
1230 				error = tcp6_connect(tp, nam, td);
1231 #endif /* INET6 */
1232 #if defined(INET6) && defined(INET)
1233 			else
1234 #endif
1235 #ifdef INET
1236 				error = tcp_connect(tp,
1237 				    (struct sockaddr *)sinp, td);
1238 #endif
1239 			/*
1240 			 * The bind operation in tcp_connect succeeded. We
1241 			 * no longer want to restore the flags if later
1242 			 * operations fail.
1243 			 */
1244 			if (error == 0 || inp->inp_lport != 0)
1245 				restoreflags = false;
1246 
1247 			if (error != 0) {
1248 				/* m is freed if PRUS_NOTREADY is unset. */
1249 				sbflush(&so->so_snd);
1250 				goto out;
1251 			}
1252 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
1253 			tcp_mss(tp, -1);
1254 		}
1255 		tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1256 		if ((flags & PRUS_NOTREADY) == 0) {
1257 			tp->t_flags |= TF_FORCEDATA;
1258 			error = tcp_output_nodrop(tp);
1259 			tp->t_flags &= ~TF_FORCEDATA;
1260 		}
1261 	}
1262 	TCP_LOG_EVENT(tp, NULL,
1263 	    &inp->inp_socket->so_rcv,
1264 	    &inp->inp_socket->so_snd,
1265 	    TCP_LOG_USERSEND, error,
1266 	    0, NULL, false);
1267 
1268 out:
1269 	/*
1270 	 * In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
1271 	 * responsible for freeing memory.
1272 	 */
1273 	if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1274 		m_freem(m);
1275 
1276 	/*
1277 	 * If the request was unsuccessful and we changed flags,
1278 	 * restore the original flags.
1279 	 */
1280 	if (error != 0 && restoreflags) {
1281 		inp->inp_vflag = vflagsav;
1282 		inp->inp_inc.inc_flags = incflagsav;
1283 	}
1284 	TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1285 		  ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1286 	TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1287 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1288 	error = tcp_unlock_or_drop(tp, error);
1289 	NET_EPOCH_EXIT(et);
1290 	return (error);
1291 }
1292 
1293 static int
1294 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1295 {
1296 	struct epoch_tracker et;
1297 	struct inpcb *inp;
1298 	struct tcpcb *tp;
1299 	int error;
1300 
1301 	inp = sotoinpcb(so);
1302 	INP_WLOCK(inp);
1303 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1304 		INP_WUNLOCK(inp);
1305 		mb_free_notready(m, count);
1306 		return (ECONNRESET);
1307 	}
1308 	tp = intotcpcb(inp);
1309 
1310 	SOCKBUF_LOCK(&so->so_snd);
1311 	error = sbready(&so->so_snd, m, count);
1312 	SOCKBUF_UNLOCK(&so->so_snd);
1313 	if (error) {
1314 		INP_WUNLOCK(inp);
1315 		return (error);
1316 	}
1317 	NET_EPOCH_ENTER(et);
1318 	error = tcp_output_unlock(tp);
1319 	NET_EPOCH_EXIT(et);
1320 
1321 	return (error);
1322 }
1323 
1324 /*
1325  * Abort the TCP.  Drop the connection abruptly.
1326  */
1327 static void
1328 tcp_usr_abort(struct socket *so)
1329 {
1330 	struct inpcb *inp;
1331 	struct tcpcb *tp = NULL;
1332 	struct epoch_tracker et;
1333 	TCPDEBUG0;
1334 
1335 	inp = sotoinpcb(so);
1336 	KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1337 
1338 	NET_EPOCH_ENTER(et);
1339 	INP_WLOCK(inp);
1340 	KASSERT(inp->inp_socket != NULL,
1341 	    ("tcp_usr_abort: inp_socket == NULL"));
1342 
1343 	/*
1344 	 * If we still have full TCP state, and we're not dropped, drop.
1345 	 */
1346 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1347 	    !(inp->inp_flags & INP_DROPPED)) {
1348 		tp = intotcpcb(inp);
1349 		TCPDEBUG1();
1350 		tp = tcp_drop(tp, ECONNABORTED);
1351 		if (tp == NULL)
1352 			goto dropped;
1353 		TCPDEBUG2(PRU_ABORT);
1354 		TCP_PROBE2(debug__user, tp, PRU_ABORT);
1355 	}
1356 	if (!(inp->inp_flags & INP_DROPPED)) {
1357 		SOCK_LOCK(so);
1358 		so->so_state |= SS_PROTOREF;
1359 		SOCK_UNLOCK(so);
1360 		inp->inp_flags |= INP_SOCKREF;
1361 	}
1362 	INP_WUNLOCK(inp);
1363 dropped:
1364 	NET_EPOCH_EXIT(et);
1365 }
1366 
1367 /*
1368  * TCP socket is closed.  Start friendly disconnect.
1369  */
1370 static void
1371 tcp_usr_close(struct socket *so)
1372 {
1373 	struct inpcb *inp;
1374 	struct tcpcb *tp = NULL;
1375 	struct epoch_tracker et;
1376 	TCPDEBUG0;
1377 
1378 	inp = sotoinpcb(so);
1379 	KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1380 
1381 	NET_EPOCH_ENTER(et);
1382 	INP_WLOCK(inp);
1383 	KASSERT(inp->inp_socket != NULL,
1384 	    ("tcp_usr_close: inp_socket == NULL"));
1385 
1386 	/*
1387 	 * If we still have full TCP state, and we're not dropped, initiate
1388 	 * a disconnect.
1389 	 */
1390 	if (!(inp->inp_flags & INP_TIMEWAIT) &&
1391 	    !(inp->inp_flags & INP_DROPPED)) {
1392 		tp = intotcpcb(inp);
1393 		TCPDEBUG1();
1394 		tcp_disconnect(tp);
1395 		TCPDEBUG2(PRU_CLOSE);
1396 		TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1397 	}
1398 	if (!(inp->inp_flags & INP_DROPPED)) {
1399 		SOCK_LOCK(so);
1400 		so->so_state |= SS_PROTOREF;
1401 		SOCK_UNLOCK(so);
1402 		inp->inp_flags |= INP_SOCKREF;
1403 	}
1404 	INP_WUNLOCK(inp);
1405 	NET_EPOCH_EXIT(et);
1406 }
1407 
1408 static int
1409 tcp_pru_options_support(struct tcpcb *tp, int flags)
1410 {
1411 	/*
1412 	 * If the specific TCP stack has a pru_options
1413 	 * specified then it does not always support
1414 	 * all the PRU_XX options and we must ask it.
1415 	 * If the function is not specified then all
1416 	 * of the PRU_XX options are supported.
1417 	 */
1418 	int ret = 0;
1419 
1420 	if (tp->t_fb->tfb_pru_options) {
1421 		ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1422 	}
1423 	return (ret);
1424 }
1425 
1426 /*
1427  * Receive out-of-band data.
1428  */
1429 static int
1430 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1431 {
1432 	int error = 0;
1433 	struct inpcb *inp;
1434 	struct tcpcb *tp = NULL;
1435 
1436 	TCPDEBUG0;
1437 	inp = sotoinpcb(so);
1438 	KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1439 	INP_WLOCK(inp);
1440 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1441 		error = ECONNRESET;
1442 		goto out;
1443 	}
1444 	tp = intotcpcb(inp);
1445 	error = tcp_pru_options_support(tp, PRUS_OOB);
1446 	if (error) {
1447 		goto out;
1448 	}
1449 	TCPDEBUG1();
1450 	if ((so->so_oobmark == 0 &&
1451 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1452 	    so->so_options & SO_OOBINLINE ||
1453 	    tp->t_oobflags & TCPOOB_HADDATA) {
1454 		error = EINVAL;
1455 		goto out;
1456 	}
1457 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1458 		error = EWOULDBLOCK;
1459 		goto out;
1460 	}
1461 	m->m_len = 1;
1462 	*mtod(m, caddr_t) = tp->t_iobc;
1463 	if ((flags & MSG_PEEK) == 0)
1464 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1465 
1466 out:
1467 	TCPDEBUG2(PRU_RCVOOB);
1468 	TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1469 	INP_WUNLOCK(inp);
1470 	return (error);
1471 }
1472 
1473 #ifdef INET
1474 struct pr_usrreqs tcp_usrreqs = {
1475 	.pru_abort =		tcp_usr_abort,
1476 	.pru_accept =		tcp_usr_accept,
1477 	.pru_attach =		tcp_usr_attach,
1478 	.pru_bind =		tcp_usr_bind,
1479 	.pru_connect =		tcp_usr_connect,
1480 	.pru_control =		in_control,
1481 	.pru_detach =		tcp_usr_detach,
1482 	.pru_disconnect =	tcp_usr_disconnect,
1483 	.pru_listen =		tcp_usr_listen,
1484 	.pru_peeraddr =		in_getpeeraddr,
1485 	.pru_rcvd =		tcp_usr_rcvd,
1486 	.pru_rcvoob =		tcp_usr_rcvoob,
1487 	.pru_send =		tcp_usr_send,
1488 	.pru_ready =		tcp_usr_ready,
1489 	.pru_shutdown =		tcp_usr_shutdown,
1490 	.pru_sockaddr =		in_getsockaddr,
1491 	.pru_sosetlabel =	in_pcbsosetlabel,
1492 	.pru_close =		tcp_usr_close,
1493 };
1494 #endif /* INET */
1495 
1496 #ifdef INET6
1497 struct pr_usrreqs tcp6_usrreqs = {
1498 	.pru_abort =		tcp_usr_abort,
1499 	.pru_accept =		tcp6_usr_accept,
1500 	.pru_attach =		tcp_usr_attach,
1501 	.pru_bind =		tcp6_usr_bind,
1502 	.pru_connect =		tcp6_usr_connect,
1503 	.pru_control =		in6_control,
1504 	.pru_detach =		tcp_usr_detach,
1505 	.pru_disconnect =	tcp_usr_disconnect,
1506 	.pru_listen =		tcp6_usr_listen,
1507 	.pru_peeraddr =		in6_mapped_peeraddr,
1508 	.pru_rcvd =		tcp_usr_rcvd,
1509 	.pru_rcvoob =		tcp_usr_rcvoob,
1510 	.pru_send =		tcp_usr_send,
1511 	.pru_ready =		tcp_usr_ready,
1512 	.pru_shutdown =		tcp_usr_shutdown,
1513 	.pru_sockaddr =		in6_mapped_sockaddr,
1514 	.pru_sosetlabel =	in_pcbsosetlabel,
1515 	.pru_close =		tcp_usr_close,
1516 };
1517 #endif /* INET6 */
1518 
1519 #ifdef INET
1520 /*
1521  * Common subroutine to open a TCP connection to remote host specified
1522  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1523  * port number if needed.  Call in_pcbconnect_setup to do the routing and
1524  * to choose a local host address (interface).  If there is an existing
1525  * incarnation of the same connection in TIME-WAIT state and if the remote
1526  * host was sending CC options and if the connection duration was < MSL, then
1527  * truncate the previous TIME-WAIT state and proceed.
1528  * Initialize connection parameters and enter SYN-SENT state.
1529  */
1530 static int
1531 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1532 {
1533 	struct inpcb *inp = tp->t_inpcb, *oinp;
1534 	struct socket *so = inp->inp_socket;
1535 	struct in_addr laddr;
1536 	u_short lport;
1537 	int error;
1538 
1539 	NET_EPOCH_ASSERT();
1540 	INP_WLOCK_ASSERT(inp);
1541 	INP_HASH_WLOCK(&V_tcbinfo);
1542 
1543 	if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1544 		error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1545 		if (error)
1546 			goto out;
1547 	}
1548 
1549 	/*
1550 	 * Cannot simply call in_pcbconnect, because there might be an
1551 	 * earlier incarnation of this same connection still in
1552 	 * TIME_WAIT state, creating an ADDRINUSE error.
1553 	 */
1554 	laddr = inp->inp_laddr;
1555 	lport = inp->inp_lport;
1556 	error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1557 	    &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1558 	if (error && oinp == NULL)
1559 		goto out;
1560 	if (oinp) {
1561 		error = EADDRINUSE;
1562 		goto out;
1563 	}
1564 	/* Handle initial bind if it hadn't been done in advance. */
1565 	if (inp->inp_lport == 0) {
1566 		inp->inp_lport = lport;
1567 		if (in_pcbinshash(inp) != 0) {
1568 			inp->inp_lport = 0;
1569 			error = EAGAIN;
1570 			goto out;
1571 		}
1572 	}
1573 	inp->inp_laddr = laddr;
1574 	in_pcbrehash(inp);
1575 	INP_HASH_WUNLOCK(&V_tcbinfo);
1576 
1577 	/*
1578 	 * Compute window scaling to request:
1579 	 * Scale to fit into sweet spot.  See tcp_syncache.c.
1580 	 * XXX: This should move to tcp_output().
1581 	 */
1582 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1583 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1584 		tp->request_r_scale++;
1585 
1586 	soisconnecting(so);
1587 	TCPSTAT_INC(tcps_connattempt);
1588 	tcp_state_change(tp, TCPS_SYN_SENT);
1589 	tp->iss = tcp_new_isn(&inp->inp_inc);
1590 	if (tp->t_flags & TF_REQ_TSTMP)
1591 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1592 	tcp_sendseqinit(tp);
1593 
1594 	return 0;
1595 
1596 out:
1597 	INP_HASH_WUNLOCK(&V_tcbinfo);
1598 	return (error);
1599 }
1600 #endif /* INET */
1601 
1602 #ifdef INET6
1603 static int
1604 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1605 {
1606 	struct inpcb *inp = tp->t_inpcb;
1607 	int error;
1608 
1609 	INP_WLOCK_ASSERT(inp);
1610 	INP_HASH_WLOCK(&V_tcbinfo);
1611 
1612 	if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1613 		error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1614 		if (error)
1615 			goto out;
1616 	}
1617 	error = in6_pcbconnect(inp, nam, td->td_ucred);
1618 	if (error != 0)
1619 		goto out;
1620 	INP_HASH_WUNLOCK(&V_tcbinfo);
1621 
1622 	/* Compute window scaling to request.  */
1623 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1624 	    (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1625 		tp->request_r_scale++;
1626 
1627 	soisconnecting(inp->inp_socket);
1628 	TCPSTAT_INC(tcps_connattempt);
1629 	tcp_state_change(tp, TCPS_SYN_SENT);
1630 	tp->iss = tcp_new_isn(&inp->inp_inc);
1631 	if (tp->t_flags & TF_REQ_TSTMP)
1632 		tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1633 	tcp_sendseqinit(tp);
1634 
1635 	return 0;
1636 
1637 out:
1638 	INP_HASH_WUNLOCK(&V_tcbinfo);
1639 	return error;
1640 }
1641 #endif /* INET6 */
1642 
1643 /*
1644  * Export TCP internal state information via a struct tcp_info, based on the
1645  * Linux 2.6 API.  Not ABI compatible as our constants are mapped differently
1646  * (TCP state machine, etc).  We export all information using FreeBSD-native
1647  * constants -- for example, the numeric values for tcpi_state will differ
1648  * from Linux.
1649  */
1650 static void
1651 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1652 {
1653 
1654 	INP_WLOCK_ASSERT(tp->t_inpcb);
1655 	bzero(ti, sizeof(*ti));
1656 
1657 	ti->tcpi_state = tp->t_state;
1658 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1659 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1660 	if (tp->t_flags & TF_SACK_PERMIT)
1661 		ti->tcpi_options |= TCPI_OPT_SACK;
1662 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1663 		ti->tcpi_options |= TCPI_OPT_WSCALE;
1664 		ti->tcpi_snd_wscale = tp->snd_scale;
1665 		ti->tcpi_rcv_wscale = tp->rcv_scale;
1666 	}
1667 	if (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT))
1668 		ti->tcpi_options |= TCPI_OPT_ECN;
1669 
1670 	ti->tcpi_rto = tp->t_rxtcur * tick;
1671 	ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1672 	ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1673 	ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1674 
1675 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1676 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
1677 
1678 	/*
1679 	 * FreeBSD-specific extension fields for tcp_info.
1680 	 */
1681 	ti->tcpi_rcv_space = tp->rcv_wnd;
1682 	ti->tcpi_rcv_nxt = tp->rcv_nxt;
1683 	ti->tcpi_snd_wnd = tp->snd_wnd;
1684 	ti->tcpi_snd_bwnd = 0;		/* Unused, kept for compat. */
1685 	ti->tcpi_snd_nxt = tp->snd_nxt;
1686 	ti->tcpi_snd_mss = tp->t_maxseg;
1687 	ti->tcpi_rcv_mss = tp->t_maxseg;
1688 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1689 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1690 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1691 #ifdef TCP_OFFLOAD
1692 	if (tp->t_flags & TF_TOE) {
1693 		ti->tcpi_options |= TCPI_OPT_TOE;
1694 		tcp_offload_tcp_info(tp, ti);
1695 	}
1696 #endif
1697 }
1698 
1699 /*
1700  * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1701  * socket option arguments.  When it re-acquires the lock after the copy, it
1702  * has to revalidate that the connection is still valid for the socket
1703  * option.
1704  */
1705 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do {			\
1706 	INP_WLOCK(inp);							\
1707 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {		\
1708 		INP_WUNLOCK(inp);					\
1709 		cleanup;						\
1710 		return (ECONNRESET);					\
1711 	}								\
1712 	tp = intotcpcb(inp);						\
1713 } while(0)
1714 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1715 
1716 int
1717 tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
1718 {
1719 	struct socket *so = inp->inp_socket;
1720 	struct tcpcb *tp = intotcpcb(inp);
1721 	int error = 0;
1722 
1723 	MPASS(sopt->sopt_dir == SOPT_SET);
1724 	INP_WLOCK_ASSERT(inp);
1725 	KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1726 	    ("inp_flags == %x", inp->inp_flags));
1727 	KASSERT(so != NULL, ("inp_socket == NULL"));
1728 
1729 	if (sopt->sopt_level != IPPROTO_TCP) {
1730 		INP_WUNLOCK(inp);
1731 #ifdef INET6
1732 		if (inp->inp_vflag & INP_IPV6PROTO)
1733 			error = ip6_ctloutput(so, sopt);
1734 #endif
1735 #if defined(INET6) && defined(INET)
1736 		else
1737 #endif
1738 #ifdef INET
1739 			error = ip_ctloutput(so, sopt);
1740 #endif
1741 		/*
1742 		 * When an IP-level socket option affects TCP, pass control
1743 		 * down to stack tfb_tcp_ctloutput, otherwise return what
1744 		 * IP level returned.
1745 		 */
1746 		switch (sopt->sopt_level) {
1747 #ifdef INET6
1748 		case IPPROTO_IPV6:
1749 			if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
1750 				return (error);
1751 			switch (sopt->sopt_name) {
1752 			case IPV6_TCLASS:
1753 				/* Notify tcp stacks that care (e.g. RACK). */
1754 				break;
1755 			case IPV6_USE_MIN_MTU:
1756 				/* Update t_maxseg accordingly. */
1757 				break;
1758 			default:
1759 				return (error);
1760 			}
1761 			break;
1762 #endif
1763 #ifdef INET
1764 		case IPPROTO_IP:
1765 			switch (sopt->sopt_name) {
1766 			case IP_TOS:
1767 				inp->inp_ip_tos &= ~IPTOS_ECN_MASK;
1768 				break;
1769 			case IP_TTL:
1770 				/* Notify tcp stacks that care (e.g. RACK). */
1771 				break;
1772 			default:
1773 				return (error);
1774 			}
1775 			break;
1776 #endif
1777 		default:
1778 			return (error);
1779 		}
1780 		INP_WLOCK(inp);
1781 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1782 			INP_WUNLOCK(inp);
1783 			return (ECONNRESET);
1784 		}
1785 	} else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
1786 		/*
1787 		 * Protect the TCP option TCP_FUNCTION_BLK so
1788 		 * that a sub-function can *never* overwrite this.
1789 		 */
1790 		struct tcp_function_set fsn;
1791 		struct tcp_function_block *blk;
1792 
1793 		INP_WUNLOCK(inp);
1794 		error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
1795 		if (error)
1796 			return (error);
1797 
1798 		INP_WLOCK(inp);
1799 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1800 			INP_WUNLOCK(inp);
1801 			return (ECONNRESET);
1802 		}
1803 		tp = intotcpcb(inp);
1804 
1805 		blk = find_and_ref_tcp_functions(&fsn);
1806 		if (blk == NULL) {
1807 			INP_WUNLOCK(inp);
1808 			return (ENOENT);
1809 		}
1810 		if (tp->t_fb == blk) {
1811 			/* You already have this */
1812 			refcount_release(&blk->tfb_refcnt);
1813 			INP_WUNLOCK(inp);
1814 			return (0);
1815 		}
1816 		if (tp->t_state != TCPS_CLOSED) {
1817 			/*
1818 			 * The user has advanced the state
1819 			 * past the initial point, we may not
1820 			 * be able to switch.
1821 			 */
1822 			if (blk->tfb_tcp_handoff_ok != NULL) {
1823 				/*
1824 				 * Does the stack provide a
1825 				 * query mechanism, if so it may
1826 				 * still be possible?
1827 				 */
1828 				error = (*blk->tfb_tcp_handoff_ok)(tp);
1829 			} else
1830 				error = EINVAL;
1831 			if (error) {
1832 				refcount_release(&blk->tfb_refcnt);
1833 				INP_WUNLOCK(inp);
1834 				return(error);
1835 			}
1836 		}
1837 		if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1838 			refcount_release(&blk->tfb_refcnt);
1839 			INP_WUNLOCK(inp);
1840 			return (ENOENT);
1841 		}
1842 		/*
1843 		 * Release the old refcnt, the
1844 		 * lookup acquired a ref on the
1845 		 * new one already.
1846 		 */
1847 		if (tp->t_fb->tfb_tcp_fb_fini) {
1848 			struct epoch_tracker et;
1849 			/*
1850 			 * Tell the stack to cleanup with 0 i.e.
1851 			 * the tcb is not going away.
1852 			 */
1853 			NET_EPOCH_ENTER(et);
1854 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1855 			NET_EPOCH_EXIT(et);
1856 		}
1857 #ifdef TCPHPTS
1858 		/* Assure that we are not on any hpts */
1859 		tcp_hpts_remove(tp->t_inpcb);
1860 #endif
1861 		if (blk->tfb_tcp_fb_init) {
1862 			error = (*blk->tfb_tcp_fb_init)(tp);
1863 			if (error) {
1864 				refcount_release(&blk->tfb_refcnt);
1865 				if (tp->t_fb->tfb_tcp_fb_init) {
1866 					if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0)  {
1867 						/* Fall back failed, drop the connection */
1868 						INP_WUNLOCK(inp);
1869 						soabort(so);
1870 						return (error);
1871 					}
1872 				}
1873 				goto err_out;
1874 			}
1875 		}
1876 		refcount_release(&tp->t_fb->tfb_refcnt);
1877 		tp->t_fb = blk;
1878 #ifdef TCP_OFFLOAD
1879 		if (tp->t_flags & TF_TOE) {
1880 			tcp_offload_ctloutput(tp, sopt->sopt_dir,
1881 			     sopt->sopt_name);
1882 		}
1883 #endif
1884 err_out:
1885 		INP_WUNLOCK(inp);
1886 		return (error);
1887 	}
1888 
1889 	/* Pass in the INP locked, callee must unlock it. */
1890 	return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1891 }
1892 
1893 static int
1894 tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
1895 {
1896 	struct socket *so = inp->inp_socket;
1897 	struct tcpcb *tp = intotcpcb(inp);
1898 	int error = 0;
1899 
1900 	MPASS(sopt->sopt_dir == SOPT_GET);
1901 	INP_WLOCK_ASSERT(inp);
1902 	KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
1903 	    ("inp_flags == %x", inp->inp_flags));
1904 	KASSERT(so != NULL, ("inp_socket == NULL"));
1905 
1906 	if (sopt->sopt_level != IPPROTO_TCP) {
1907 		INP_WUNLOCK(inp);
1908 #ifdef INET6
1909 		if (inp->inp_vflag & INP_IPV6PROTO)
1910 			error = ip6_ctloutput(so, sopt);
1911 #endif /* INET6 */
1912 #if defined(INET6) && defined(INET)
1913 		else
1914 #endif
1915 #ifdef INET
1916 			error = ip_ctloutput(so, sopt);
1917 #endif
1918 		return (error);
1919 	}
1920 	if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
1921 	     (sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
1922 		struct tcp_function_set fsn;
1923 
1924 		if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
1925 			memset(&fsn, 0, sizeof(fsn));
1926 			find_tcp_function_alias(tp->t_fb, &fsn);
1927 		} else {
1928 			strncpy(fsn.function_set_name,
1929 			    tp->t_fb->tfb_tcp_block_name,
1930 			    TCP_FUNCTION_NAME_LEN_MAX);
1931 			fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1932 		}
1933 		fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1934 		INP_WUNLOCK(inp);
1935 		error = sooptcopyout(sopt, &fsn, sizeof fsn);
1936 		return (error);
1937 	}
1938 
1939 	/* Pass in the INP locked, callee must unlock it. */
1940 	return (tp->t_fb->tfb_tcp_ctloutput(inp, sopt));
1941 }
1942 
1943 int
1944 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1945 {
1946 	struct	inpcb *inp;
1947 
1948 	inp = sotoinpcb(so);
1949 	KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1950 
1951 	INP_WLOCK(inp);
1952 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1953 		INP_WUNLOCK(inp);
1954 		return (ECONNRESET);
1955 	}
1956 	if (sopt->sopt_dir == SOPT_SET)
1957 		return (tcp_ctloutput_set(inp, sopt));
1958 	else if (sopt->sopt_dir == SOPT_GET)
1959 		return (tcp_ctloutput_get(inp, sopt));
1960 	else
1961 		panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
1962 }
1963 
1964 /*
1965  * If this assert becomes untrue, we need to change the size of the buf
1966  * variable in tcp_default_ctloutput().
1967  */
1968 #ifdef CTASSERT
1969 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
1970 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
1971 #endif
1972 
1973 #ifdef KERN_TLS
1974 static int
1975 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
1976 {
1977 	struct tls_enable_v0 tls_v0;
1978 	int error;
1979 
1980 	if (sopt->sopt_valsize == sizeof(tls_v0)) {
1981 		error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
1982 		    sizeof(tls_v0));
1983 		if (error)
1984 			return (error);
1985 		memset(tls, 0, sizeof(*tls));
1986 		tls->cipher_key = tls_v0.cipher_key;
1987 		tls->iv = tls_v0.iv;
1988 		tls->auth_key = tls_v0.auth_key;
1989 		tls->cipher_algorithm = tls_v0.cipher_algorithm;
1990 		tls->cipher_key_len = tls_v0.cipher_key_len;
1991 		tls->iv_len = tls_v0.iv_len;
1992 		tls->auth_algorithm = tls_v0.auth_algorithm;
1993 		tls->auth_key_len = tls_v0.auth_key_len;
1994 		tls->flags = tls_v0.flags;
1995 		tls->tls_vmajor = tls_v0.tls_vmajor;
1996 		tls->tls_vminor = tls_v0.tls_vminor;
1997 		return (0);
1998 	}
1999 
2000 	return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
2001 }
2002 #endif
2003 
2004 extern struct cc_algo newreno_cc_algo;
2005 
2006 static int
2007 tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt)
2008 {
2009 	struct cc_algo *algo;
2010 	void *ptr = NULL;
2011 	struct tcpcb *tp;
2012 	struct cc_var cc_mem;
2013 	char	buf[TCP_CA_NAME_MAX];
2014 	size_t mem_sz;
2015 	int error;
2016 
2017 	INP_WUNLOCK(inp);
2018 	error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2019 	if (error)
2020 		return(error);
2021 	buf[sopt->sopt_valsize] = '\0';
2022 	CC_LIST_RLOCK();
2023 	STAILQ_FOREACH(algo, &cc_list, entries) {
2024 		if (strncmp(buf, algo->name,
2025 			    TCP_CA_NAME_MAX) == 0) {
2026 			if (algo->flags & CC_MODULE_BEING_REMOVED) {
2027 				/* We can't "see" modules being unloaded */
2028 				continue;
2029 			}
2030 			break;
2031 		}
2032 	}
2033 	if (algo == NULL) {
2034 		CC_LIST_RUNLOCK();
2035 		return(ESRCH);
2036 	}
2037 	/*
2038 	 * With a reference the algorithm cannot be removed
2039 	 * so we hold a reference through the change process.
2040 	 */
2041 	cc_refer(algo);
2042 	CC_LIST_RUNLOCK();
2043 	if (algo->cb_init != NULL) {
2044 		/* We can now pre-get the memory for the CC */
2045 		mem_sz = (*algo->cc_data_sz)();
2046 		if (mem_sz == 0) {
2047 			goto no_mem_needed;
2048 		}
2049 		ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
2050 	} else {
2051 no_mem_needed:
2052 		mem_sz = 0;
2053 		ptr = NULL;
2054 	}
2055 	/*
2056 	 * Make sure its all clean and zero and also get
2057 	 * back the inplock.
2058 	 */
2059 	memset(&cc_mem, 0, sizeof(cc_mem));
2060 	INP_WLOCK(inp);
2061 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2062 		INP_WUNLOCK(inp);
2063 		if (ptr)
2064 			free(ptr, M_CC_MEM);
2065 		/* Release our temp reference */
2066 		CC_LIST_RLOCK();
2067 		cc_release(algo);
2068 		CC_LIST_RUNLOCK();
2069 		return (ECONNRESET);
2070 	}
2071 	tp = intotcpcb(inp);
2072 	if (ptr != NULL)
2073 		memset(ptr, 0, mem_sz);
2074 	cc_mem.ccvc.tcp = tp;
2075 	/*
2076 	 * We once again hold a write lock over the tcb so it's
2077 	 * safe to do these things without ordering concerns.
2078 	 * Note here we init into stack memory.
2079 	 */
2080 	if (algo->cb_init != NULL)
2081 		error = algo->cb_init(&cc_mem, ptr);
2082 	else
2083 		error = 0;
2084 	/*
2085 	 * The CC algorithms, when given their memory
2086 	 * should not fail we could in theory have a
2087 	 * KASSERT here.
2088 	 */
2089 	if (error == 0) {
2090 		/*
2091 		 * Touchdown, lets go ahead and move the
2092 		 * connection to the new CC module by
2093 		 * copying in the cc_mem after we call
2094 		 * the old ones cleanup (if any).
2095 		 */
2096 		if (CC_ALGO(tp)->cb_destroy != NULL)
2097 			CC_ALGO(tp)->cb_destroy(tp->ccv);
2098 		/* Detach the old CC from the tcpcb  */
2099 		cc_detach(tp);
2100 		/* Copy in our temp memory that was inited */
2101 		memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2102 		/* Now attach the new, which takes a reference */
2103 		cc_attach(tp, algo);
2104 		/* Ok now are we where we have gotten past any conn_init? */
2105 		if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
2106 			/* Yep run the connection init for the new CC */
2107 			CC_ALGO(tp)->conn_init(tp->ccv);
2108 		}
2109 	} else if (ptr)
2110 		free(ptr, M_CC_MEM);
2111 	INP_WUNLOCK(inp);
2112 	/* Now lets release our temp reference */
2113 	CC_LIST_RLOCK();
2114 	cc_release(algo);
2115 	CC_LIST_RUNLOCK();
2116 	return (error);
2117 }
2118 
2119 int
2120 tcp_default_ctloutput(struct inpcb *inp, struct sockopt *sopt)
2121 {
2122 	struct tcpcb *tp = intotcpcb(inp);
2123 	int	error, opt, optval;
2124 	u_int	ui;
2125 	struct	tcp_info ti;
2126 #ifdef KERN_TLS
2127 	struct tls_enable tls;
2128 	struct socket *so = inp->inp_socket;
2129 #endif
2130 	char	*pbuf, buf[TCP_LOG_ID_LEN];
2131 #ifdef STATS
2132 	struct statsblob *sbp;
2133 #endif
2134 	size_t	len;
2135 
2136 	INP_WLOCK_ASSERT(inp);
2137 	KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
2138 	    ("inp_flags == %x", inp->inp_flags));
2139 	KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL"));
2140 
2141 	switch (sopt->sopt_level) {
2142 #ifdef INET6
2143 	case IPPROTO_IPV6:
2144 		MPASS(inp->inp_vflag & INP_IPV6PROTO);
2145 		switch (sopt->sopt_name) {
2146 		case IPV6_USE_MIN_MTU:
2147 			tcp6_use_min_mtu(tp);
2148 			/* FALLTHROUGH */
2149 		}
2150 		INP_WUNLOCK(inp);
2151 		return (0);
2152 #endif
2153 #ifdef INET
2154 	case IPPROTO_IP:
2155 		INP_WUNLOCK(inp);
2156 		return (0);
2157 #endif
2158 	}
2159 
2160 	/*
2161 	 * For TCP_CCALGOOPT forward the control to CC module, for both
2162 	 * SOPT_SET and SOPT_GET.
2163 	 */
2164 	switch (sopt->sopt_name) {
2165 	case TCP_CCALGOOPT:
2166 		INP_WUNLOCK(inp);
2167 		if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2168 			return (EINVAL);
2169 		pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2170 		error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2171 		    sopt->sopt_valsize);
2172 		if (error) {
2173 			free(pbuf, M_TEMP);
2174 			return (error);
2175 		}
2176 		INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2177 		if (CC_ALGO(tp)->ctl_output != NULL)
2178 			error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
2179 		else
2180 			error = ENOENT;
2181 		INP_WUNLOCK(inp);
2182 		if (error == 0 && sopt->sopt_dir == SOPT_GET)
2183 			error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2184 		free(pbuf, M_TEMP);
2185 		return (error);
2186 	}
2187 
2188 	switch (sopt->sopt_dir) {
2189 	case SOPT_SET:
2190 		switch (sopt->sopt_name) {
2191 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2192 		case TCP_MD5SIG:
2193 			if (!TCPMD5_ENABLED()) {
2194 				INP_WUNLOCK(inp);
2195 				return (ENOPROTOOPT);
2196 			}
2197 			error = TCPMD5_PCBCTL(inp, sopt);
2198 			if (error)
2199 				return (error);
2200 			goto unlock_and_done;
2201 #endif /* IPSEC */
2202 
2203 		case TCP_NODELAY:
2204 		case TCP_NOOPT:
2205 		case TCP_LRD:
2206 			INP_WUNLOCK(inp);
2207 			error = sooptcopyin(sopt, &optval, sizeof optval,
2208 			    sizeof optval);
2209 			if (error)
2210 				return (error);
2211 
2212 			INP_WLOCK_RECHECK(inp);
2213 			switch (sopt->sopt_name) {
2214 			case TCP_NODELAY:
2215 				opt = TF_NODELAY;
2216 				break;
2217 			case TCP_NOOPT:
2218 				opt = TF_NOOPT;
2219 				break;
2220 			case TCP_LRD:
2221 				opt = TF_LRD;
2222 				break;
2223 			default:
2224 				opt = 0; /* dead code to fool gcc */
2225 				break;
2226 			}
2227 
2228 			if (optval)
2229 				tp->t_flags |= opt;
2230 			else
2231 				tp->t_flags &= ~opt;
2232 unlock_and_done:
2233 #ifdef TCP_OFFLOAD
2234 			if (tp->t_flags & TF_TOE) {
2235 				tcp_offload_ctloutput(tp, sopt->sopt_dir,
2236 				    sopt->sopt_name);
2237 			}
2238 #endif
2239 			INP_WUNLOCK(inp);
2240 			break;
2241 
2242 		case TCP_NOPUSH:
2243 			INP_WUNLOCK(inp);
2244 			error = sooptcopyin(sopt, &optval, sizeof optval,
2245 			    sizeof optval);
2246 			if (error)
2247 				return (error);
2248 
2249 			INP_WLOCK_RECHECK(inp);
2250 			if (optval)
2251 				tp->t_flags |= TF_NOPUSH;
2252 			else if (tp->t_flags & TF_NOPUSH) {
2253 				tp->t_flags &= ~TF_NOPUSH;
2254 				if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2255 					struct epoch_tracker et;
2256 
2257 					NET_EPOCH_ENTER(et);
2258 					error = tcp_output_nodrop(tp);
2259 					NET_EPOCH_EXIT(et);
2260 				}
2261 			}
2262 			goto unlock_and_done;
2263 
2264 		case TCP_REMOTE_UDP_ENCAPS_PORT:
2265 			INP_WUNLOCK(inp);
2266 			error = sooptcopyin(sopt, &optval, sizeof optval,
2267 			    sizeof optval);
2268 			if (error)
2269 				return (error);
2270 			if ((optval < TCP_TUNNELING_PORT_MIN) ||
2271 			    (optval > TCP_TUNNELING_PORT_MAX)) {
2272 				/* Its got to be in range */
2273 				return (EINVAL);
2274 			}
2275 			if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
2276 				/* You have to have enabled a UDP tunneling port first */
2277 				return (EINVAL);
2278 			}
2279 			INP_WLOCK_RECHECK(inp);
2280 			if (tp->t_state != TCPS_CLOSED) {
2281 				/* You can't change after you are connected */
2282 				error = EINVAL;
2283 			} else {
2284 				/* Ok we are all good set the port */
2285 				tp->t_port = htons(optval);
2286 			}
2287 			goto unlock_and_done;
2288 
2289 		case TCP_MAXSEG:
2290 			INP_WUNLOCK(inp);
2291 			error = sooptcopyin(sopt, &optval, sizeof optval,
2292 			    sizeof optval);
2293 			if (error)
2294 				return (error);
2295 
2296 			INP_WLOCK_RECHECK(inp);
2297 			if (optval > 0 && optval <= tp->t_maxseg &&
2298 			    optval + 40 >= V_tcp_minmss)
2299 				tp->t_maxseg = optval;
2300 			else
2301 				error = EINVAL;
2302 			goto unlock_and_done;
2303 
2304 		case TCP_INFO:
2305 			INP_WUNLOCK(inp);
2306 			error = EINVAL;
2307 			break;
2308 
2309 		case TCP_STATS:
2310 			INP_WUNLOCK(inp);
2311 #ifdef STATS
2312 			error = sooptcopyin(sopt, &optval, sizeof optval,
2313 			    sizeof optval);
2314 			if (error)
2315 				return (error);
2316 
2317 			if (optval > 0)
2318 				sbp = stats_blob_alloc(
2319 				    V_tcp_perconn_stats_dflt_tpl, 0);
2320 			else
2321 				sbp = NULL;
2322 
2323 			INP_WLOCK_RECHECK(inp);
2324 			if ((tp->t_stats != NULL && sbp == NULL) ||
2325 			    (tp->t_stats == NULL && sbp != NULL)) {
2326 				struct statsblob *t = tp->t_stats;
2327 				tp->t_stats = sbp;
2328 				sbp = t;
2329 			}
2330 			INP_WUNLOCK(inp);
2331 
2332 			stats_blob_destroy(sbp);
2333 #else
2334 			return (EOPNOTSUPP);
2335 #endif /* !STATS */
2336 			break;
2337 
2338 		case TCP_CONGESTION:
2339 			error = tcp_set_cc_mod(inp, sopt);
2340 			break;
2341 
2342 		case TCP_REUSPORT_LB_NUMA:
2343 			INP_WUNLOCK(inp);
2344 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2345 			    sizeof(optval));
2346 			INP_WLOCK_RECHECK(inp);
2347 			if (!error)
2348 				error = in_pcblbgroup_numa(inp, optval);
2349 			INP_WUNLOCK(inp);
2350 			break;
2351 
2352 #ifdef KERN_TLS
2353 		case TCP_TXTLS_ENABLE:
2354 			INP_WUNLOCK(inp);
2355 			error = copyin_tls_enable(sopt, &tls);
2356 			if (error)
2357 				break;
2358 			error = ktls_enable_tx(so, &tls);
2359 			break;
2360 		case TCP_TXTLS_MODE:
2361 			INP_WUNLOCK(inp);
2362 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2363 			if (error)
2364 				return (error);
2365 
2366 			INP_WLOCK_RECHECK(inp);
2367 			error = ktls_set_tx_mode(so, ui);
2368 			INP_WUNLOCK(inp);
2369 			break;
2370 		case TCP_RXTLS_ENABLE:
2371 			INP_WUNLOCK(inp);
2372 			error = sooptcopyin(sopt, &tls, sizeof(tls),
2373 			    sizeof(tls));
2374 			if (error)
2375 				break;
2376 			error = ktls_enable_rx(so, &tls);
2377 			break;
2378 #endif
2379 
2380 		case TCP_KEEPIDLE:
2381 		case TCP_KEEPINTVL:
2382 		case TCP_KEEPINIT:
2383 			INP_WUNLOCK(inp);
2384 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2385 			if (error)
2386 				return (error);
2387 
2388 			if (ui > (UINT_MAX / hz)) {
2389 				error = EINVAL;
2390 				break;
2391 			}
2392 			ui *= hz;
2393 
2394 			INP_WLOCK_RECHECK(inp);
2395 			switch (sopt->sopt_name) {
2396 			case TCP_KEEPIDLE:
2397 				tp->t_keepidle = ui;
2398 				/*
2399 				 * XXX: better check current remaining
2400 				 * timeout and "merge" it with new value.
2401 				 */
2402 				if ((tp->t_state > TCPS_LISTEN) &&
2403 				    (tp->t_state <= TCPS_CLOSING))
2404 					tcp_timer_activate(tp, TT_KEEP,
2405 					    TP_KEEPIDLE(tp));
2406 				break;
2407 			case TCP_KEEPINTVL:
2408 				tp->t_keepintvl = ui;
2409 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2410 				    (TP_MAXIDLE(tp) > 0))
2411 					tcp_timer_activate(tp, TT_2MSL,
2412 					    TP_MAXIDLE(tp));
2413 				break;
2414 			case TCP_KEEPINIT:
2415 				tp->t_keepinit = ui;
2416 				if (tp->t_state == TCPS_SYN_RECEIVED ||
2417 				    tp->t_state == TCPS_SYN_SENT)
2418 					tcp_timer_activate(tp, TT_KEEP,
2419 					    TP_KEEPINIT(tp));
2420 				break;
2421 			}
2422 			goto unlock_and_done;
2423 
2424 		case TCP_KEEPCNT:
2425 			INP_WUNLOCK(inp);
2426 			error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2427 			if (error)
2428 				return (error);
2429 
2430 			INP_WLOCK_RECHECK(inp);
2431 			tp->t_keepcnt = ui;
2432 			if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2433 			    (TP_MAXIDLE(tp) > 0))
2434 				tcp_timer_activate(tp, TT_2MSL,
2435 				    TP_MAXIDLE(tp));
2436 			goto unlock_and_done;
2437 
2438 #ifdef TCPPCAP
2439 		case TCP_PCAP_OUT:
2440 		case TCP_PCAP_IN:
2441 			INP_WUNLOCK(inp);
2442 			error = sooptcopyin(sopt, &optval, sizeof optval,
2443 			    sizeof optval);
2444 			if (error)
2445 				return (error);
2446 
2447 			INP_WLOCK_RECHECK(inp);
2448 			if (optval >= 0)
2449 				tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2450 					&(tp->t_outpkts) : &(tp->t_inpkts),
2451 					optval);
2452 			else
2453 				error = EINVAL;
2454 			goto unlock_and_done;
2455 #endif
2456 
2457 		case TCP_FASTOPEN: {
2458 			struct tcp_fastopen tfo_optval;
2459 
2460 			INP_WUNLOCK(inp);
2461 			if (!V_tcp_fastopen_client_enable &&
2462 			    !V_tcp_fastopen_server_enable)
2463 				return (EPERM);
2464 
2465 			error = sooptcopyin(sopt, &tfo_optval,
2466 				    sizeof(tfo_optval), sizeof(int));
2467 			if (error)
2468 				return (error);
2469 
2470 			INP_WLOCK_RECHECK(inp);
2471 			if ((tp->t_state != TCPS_CLOSED) &&
2472 			    (tp->t_state != TCPS_LISTEN)) {
2473 				error = EINVAL;
2474 				goto unlock_and_done;
2475 			}
2476 			if (tfo_optval.enable) {
2477 				if (tp->t_state == TCPS_LISTEN) {
2478 					if (!V_tcp_fastopen_server_enable) {
2479 						error = EPERM;
2480 						goto unlock_and_done;
2481 					}
2482 
2483 					if (tp->t_tfo_pending == NULL)
2484 						tp->t_tfo_pending =
2485 						    tcp_fastopen_alloc_counter();
2486 				} else {
2487 					/*
2488 					 * If a pre-shared key was provided,
2489 					 * stash it in the client cookie
2490 					 * field of the tcpcb for use during
2491 					 * connect.
2492 					 */
2493 					if (sopt->sopt_valsize ==
2494 					    sizeof(tfo_optval)) {
2495 						memcpy(tp->t_tfo_cookie.client,
2496 						       tfo_optval.psk,
2497 						       TCP_FASTOPEN_PSK_LEN);
2498 						tp->t_tfo_client_cookie_len =
2499 						    TCP_FASTOPEN_PSK_LEN;
2500 					}
2501 				}
2502 				tp->t_flags |= TF_FASTOPEN;
2503 			} else
2504 				tp->t_flags &= ~TF_FASTOPEN;
2505 			goto unlock_and_done;
2506 		}
2507 
2508 #ifdef TCP_BLACKBOX
2509 		case TCP_LOG:
2510 			INP_WUNLOCK(inp);
2511 			error = sooptcopyin(sopt, &optval, sizeof optval,
2512 			    sizeof optval);
2513 			if (error)
2514 				return (error);
2515 
2516 			INP_WLOCK_RECHECK(inp);
2517 			error = tcp_log_state_change(tp, optval);
2518 			goto unlock_and_done;
2519 
2520 		case TCP_LOGBUF:
2521 			INP_WUNLOCK(inp);
2522 			error = EINVAL;
2523 			break;
2524 
2525 		case TCP_LOGID:
2526 			INP_WUNLOCK(inp);
2527 			error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2528 			if (error)
2529 				break;
2530 			buf[sopt->sopt_valsize] = '\0';
2531 			INP_WLOCK_RECHECK(inp);
2532 			error = tcp_log_set_id(tp, buf);
2533 			/* tcp_log_set_id() unlocks the INP. */
2534 			break;
2535 
2536 		case TCP_LOGDUMP:
2537 		case TCP_LOGDUMPID:
2538 			INP_WUNLOCK(inp);
2539 			error =
2540 			    sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2541 			if (error)
2542 				break;
2543 			buf[sopt->sopt_valsize] = '\0';
2544 			INP_WLOCK_RECHECK(inp);
2545 			if (sopt->sopt_name == TCP_LOGDUMP) {
2546 				error = tcp_log_dump_tp_logbuf(tp, buf,
2547 				    M_WAITOK, true);
2548 				INP_WUNLOCK(inp);
2549 			} else {
2550 				tcp_log_dump_tp_bucket_logbufs(tp, buf);
2551 				/*
2552 				 * tcp_log_dump_tp_bucket_logbufs() drops the
2553 				 * INP lock.
2554 				 */
2555 			}
2556 			break;
2557 #endif
2558 
2559 		default:
2560 			INP_WUNLOCK(inp);
2561 			error = ENOPROTOOPT;
2562 			break;
2563 		}
2564 		break;
2565 
2566 	case SOPT_GET:
2567 		tp = intotcpcb(inp);
2568 		switch (sopt->sopt_name) {
2569 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2570 		case TCP_MD5SIG:
2571 			if (!TCPMD5_ENABLED()) {
2572 				INP_WUNLOCK(inp);
2573 				return (ENOPROTOOPT);
2574 			}
2575 			error = TCPMD5_PCBCTL(inp, sopt);
2576 			break;
2577 #endif
2578 
2579 		case TCP_NODELAY:
2580 			optval = tp->t_flags & TF_NODELAY;
2581 			INP_WUNLOCK(inp);
2582 			error = sooptcopyout(sopt, &optval, sizeof optval);
2583 			break;
2584 		case TCP_MAXSEG:
2585 			optval = tp->t_maxseg;
2586 			INP_WUNLOCK(inp);
2587 			error = sooptcopyout(sopt, &optval, sizeof optval);
2588 			break;
2589 		case TCP_REMOTE_UDP_ENCAPS_PORT:
2590 			optval = ntohs(tp->t_port);
2591 			INP_WUNLOCK(inp);
2592 			error = sooptcopyout(sopt, &optval, sizeof optval);
2593 			break;
2594 		case TCP_NOOPT:
2595 			optval = tp->t_flags & TF_NOOPT;
2596 			INP_WUNLOCK(inp);
2597 			error = sooptcopyout(sopt, &optval, sizeof optval);
2598 			break;
2599 		case TCP_NOPUSH:
2600 			optval = tp->t_flags & TF_NOPUSH;
2601 			INP_WUNLOCK(inp);
2602 			error = sooptcopyout(sopt, &optval, sizeof optval);
2603 			break;
2604 		case TCP_INFO:
2605 			tcp_fill_info(tp, &ti);
2606 			INP_WUNLOCK(inp);
2607 			error = sooptcopyout(sopt, &ti, sizeof ti);
2608 			break;
2609 		case TCP_STATS:
2610 			{
2611 #ifdef STATS
2612 			int nheld;
2613 			TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2614 
2615 			error = 0;
2616 			socklen_t outsbsz = sopt->sopt_valsize;
2617 			if (tp->t_stats == NULL)
2618 				error = ENOENT;
2619 			else if (outsbsz >= tp->t_stats->cursz)
2620 				outsbsz = tp->t_stats->cursz;
2621 			else if (outsbsz >= sizeof(struct statsblob))
2622 				outsbsz = sizeof(struct statsblob);
2623 			else
2624 				error = EINVAL;
2625 			INP_WUNLOCK(inp);
2626 			if (error)
2627 				break;
2628 
2629 			sbp = sopt->sopt_val;
2630 			nheld = atop(round_page(((vm_offset_t)sbp) +
2631 			    (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2632 			vm_page_t ma[nheld];
2633 			if (vm_fault_quick_hold_pages(
2634 			    &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2635 			    outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2636 			    nheld) < 0) {
2637 				error = EFAULT;
2638 				break;
2639 			}
2640 
2641 			if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2642 			    SIZEOF_MEMBER(struct statsblob, flags))))
2643 				goto unhold;
2644 
2645 			INP_WLOCK_RECHECK(inp);
2646 			error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2647 			    sbflags | SB_CLONE_USRDSTNOFAULT);
2648 			INP_WUNLOCK(inp);
2649 			sopt->sopt_valsize = outsbsz;
2650 unhold:
2651 			vm_page_unhold_pages(ma, nheld);
2652 #else
2653 			INP_WUNLOCK(inp);
2654 			error = EOPNOTSUPP;
2655 #endif /* !STATS */
2656 			break;
2657 			}
2658 		case TCP_CONGESTION:
2659 			len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2660 			INP_WUNLOCK(inp);
2661 			error = sooptcopyout(sopt, buf, len + 1);
2662 			break;
2663 		case TCP_KEEPIDLE:
2664 		case TCP_KEEPINTVL:
2665 		case TCP_KEEPINIT:
2666 		case TCP_KEEPCNT:
2667 			switch (sopt->sopt_name) {
2668 			case TCP_KEEPIDLE:
2669 				ui = TP_KEEPIDLE(tp) / hz;
2670 				break;
2671 			case TCP_KEEPINTVL:
2672 				ui = TP_KEEPINTVL(tp) / hz;
2673 				break;
2674 			case TCP_KEEPINIT:
2675 				ui = TP_KEEPINIT(tp) / hz;
2676 				break;
2677 			case TCP_KEEPCNT:
2678 				ui = TP_KEEPCNT(tp);
2679 				break;
2680 			}
2681 			INP_WUNLOCK(inp);
2682 			error = sooptcopyout(sopt, &ui, sizeof(ui));
2683 			break;
2684 #ifdef TCPPCAP
2685 		case TCP_PCAP_OUT:
2686 		case TCP_PCAP_IN:
2687 			optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2688 					&(tp->t_outpkts) : &(tp->t_inpkts));
2689 			INP_WUNLOCK(inp);
2690 			error = sooptcopyout(sopt, &optval, sizeof optval);
2691 			break;
2692 #endif
2693 		case TCP_FASTOPEN:
2694 			optval = tp->t_flags & TF_FASTOPEN;
2695 			INP_WUNLOCK(inp);
2696 			error = sooptcopyout(sopt, &optval, sizeof optval);
2697 			break;
2698 #ifdef TCP_BLACKBOX
2699 		case TCP_LOG:
2700 			optval = tp->t_logstate;
2701 			INP_WUNLOCK(inp);
2702 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2703 			break;
2704 		case TCP_LOGBUF:
2705 			/* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2706 			error = tcp_log_getlogbuf(sopt, tp);
2707 			break;
2708 		case TCP_LOGID:
2709 			len = tcp_log_get_id(tp, buf);
2710 			INP_WUNLOCK(inp);
2711 			error = sooptcopyout(sopt, buf, len + 1);
2712 			break;
2713 		case TCP_LOGDUMP:
2714 		case TCP_LOGDUMPID:
2715 			INP_WUNLOCK(inp);
2716 			error = EINVAL;
2717 			break;
2718 #endif
2719 #ifdef KERN_TLS
2720 		case TCP_TXTLS_MODE:
2721 			error = ktls_get_tx_mode(so, &optval);
2722 			INP_WUNLOCK(inp);
2723 			if (error == 0)
2724 				error = sooptcopyout(sopt, &optval,
2725 				    sizeof(optval));
2726 			break;
2727 		case TCP_RXTLS_MODE:
2728 			error = ktls_get_rx_mode(so, &optval);
2729 			INP_WUNLOCK(inp);
2730 			if (error == 0)
2731 				error = sooptcopyout(sopt, &optval,
2732 				    sizeof(optval));
2733 			break;
2734 #endif
2735 		case TCP_LRD:
2736 			optval = tp->t_flags & TF_LRD;
2737 			INP_WUNLOCK(inp);
2738 			error = sooptcopyout(sopt, &optval, sizeof optval);
2739 			break;
2740 		default:
2741 			INP_WUNLOCK(inp);
2742 			error = ENOPROTOOPT;
2743 			break;
2744 		}
2745 		break;
2746 	}
2747 	return (error);
2748 }
2749 #undef INP_WLOCK_RECHECK
2750 #undef INP_WLOCK_RECHECK_CLEANUP
2751 
2752 /*
2753  * Initiate (or continue) disconnect.
2754  * If embryonic state, just send reset (once).
2755  * If in ``let data drain'' option and linger null, just drop.
2756  * Otherwise (hard), mark socket disconnecting and drop
2757  * current input data; switch states based on user close, and
2758  * send segment to peer (with FIN).
2759  */
2760 static void
2761 tcp_disconnect(struct tcpcb *tp)
2762 {
2763 	struct inpcb *inp = tp->t_inpcb;
2764 	struct socket *so = inp->inp_socket;
2765 
2766 	NET_EPOCH_ASSERT();
2767 	INP_WLOCK_ASSERT(inp);
2768 
2769 	/*
2770 	 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2771 	 * socket is still open.
2772 	 */
2773 	if (tp->t_state < TCPS_ESTABLISHED &&
2774 	    !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2775 		tp = tcp_close(tp);
2776 		KASSERT(tp != NULL,
2777 		    ("tcp_disconnect: tcp_close() returned NULL"));
2778 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2779 		tp = tcp_drop(tp, 0);
2780 		KASSERT(tp != NULL,
2781 		    ("tcp_disconnect: tcp_drop() returned NULL"));
2782 	} else {
2783 		soisdisconnecting(so);
2784 		sbflush(&so->so_rcv);
2785 		tcp_usrclosed(tp);
2786 		if (!(inp->inp_flags & INP_DROPPED))
2787 			/* Ignore stack's drop request, we already at it. */
2788 			(void)tcp_output_nodrop(tp);
2789 	}
2790 }
2791 
2792 /*
2793  * User issued close, and wish to trail through shutdown states:
2794  * if never received SYN, just forget it.  If got a SYN from peer,
2795  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2796  * If already got a FIN from peer, then almost done; go to LAST_ACK
2797  * state.  In all other cases, have already sent FIN to peer (e.g.
2798  * after PRU_SHUTDOWN), and just have to play tedious game waiting
2799  * for peer to send FIN or not respond to keep-alives, etc.
2800  * We can let the user exit from the close as soon as the FIN is acked.
2801  */
2802 static void
2803 tcp_usrclosed(struct tcpcb *tp)
2804 {
2805 
2806 	NET_EPOCH_ASSERT();
2807 	INP_WLOCK_ASSERT(tp->t_inpcb);
2808 
2809 	switch (tp->t_state) {
2810 	case TCPS_LISTEN:
2811 #ifdef TCP_OFFLOAD
2812 		tcp_offload_listen_stop(tp);
2813 #endif
2814 		tcp_state_change(tp, TCPS_CLOSED);
2815 		/* FALLTHROUGH */
2816 	case TCPS_CLOSED:
2817 		tp = tcp_close(tp);
2818 		/*
2819 		 * tcp_close() should never return NULL here as the socket is
2820 		 * still open.
2821 		 */
2822 		KASSERT(tp != NULL,
2823 		    ("tcp_usrclosed: tcp_close() returned NULL"));
2824 		break;
2825 
2826 	case TCPS_SYN_SENT:
2827 	case TCPS_SYN_RECEIVED:
2828 		tp->t_flags |= TF_NEEDFIN;
2829 		break;
2830 
2831 	case TCPS_ESTABLISHED:
2832 		tcp_state_change(tp, TCPS_FIN_WAIT_1);
2833 		break;
2834 
2835 	case TCPS_CLOSE_WAIT:
2836 		tcp_state_change(tp, TCPS_LAST_ACK);
2837 		break;
2838 	}
2839 	if (tp->t_state >= TCPS_FIN_WAIT_2) {
2840 		soisdisconnected(tp->t_inpcb->inp_socket);
2841 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
2842 		if (tp->t_state == TCPS_FIN_WAIT_2) {
2843 			int timeout;
2844 
2845 			timeout = (tcp_fast_finwait2_recycle) ?
2846 			    tcp_finwait2_timeout : TP_MAXIDLE(tp);
2847 			tcp_timer_activate(tp, TT_2MSL, timeout);
2848 		}
2849 	}
2850 }
2851 
2852 #ifdef DDB
2853 static void
2854 db_print_indent(int indent)
2855 {
2856 	int i;
2857 
2858 	for (i = 0; i < indent; i++)
2859 		db_printf(" ");
2860 }
2861 
2862 static void
2863 db_print_tstate(int t_state)
2864 {
2865 
2866 	switch (t_state) {
2867 	case TCPS_CLOSED:
2868 		db_printf("TCPS_CLOSED");
2869 		return;
2870 
2871 	case TCPS_LISTEN:
2872 		db_printf("TCPS_LISTEN");
2873 		return;
2874 
2875 	case TCPS_SYN_SENT:
2876 		db_printf("TCPS_SYN_SENT");
2877 		return;
2878 
2879 	case TCPS_SYN_RECEIVED:
2880 		db_printf("TCPS_SYN_RECEIVED");
2881 		return;
2882 
2883 	case TCPS_ESTABLISHED:
2884 		db_printf("TCPS_ESTABLISHED");
2885 		return;
2886 
2887 	case TCPS_CLOSE_WAIT:
2888 		db_printf("TCPS_CLOSE_WAIT");
2889 		return;
2890 
2891 	case TCPS_FIN_WAIT_1:
2892 		db_printf("TCPS_FIN_WAIT_1");
2893 		return;
2894 
2895 	case TCPS_CLOSING:
2896 		db_printf("TCPS_CLOSING");
2897 		return;
2898 
2899 	case TCPS_LAST_ACK:
2900 		db_printf("TCPS_LAST_ACK");
2901 		return;
2902 
2903 	case TCPS_FIN_WAIT_2:
2904 		db_printf("TCPS_FIN_WAIT_2");
2905 		return;
2906 
2907 	case TCPS_TIME_WAIT:
2908 		db_printf("TCPS_TIME_WAIT");
2909 		return;
2910 
2911 	default:
2912 		db_printf("unknown");
2913 		return;
2914 	}
2915 }
2916 
2917 static void
2918 db_print_tflags(u_int t_flags)
2919 {
2920 	int comma;
2921 
2922 	comma = 0;
2923 	if (t_flags & TF_ACKNOW) {
2924 		db_printf("%sTF_ACKNOW", comma ? ", " : "");
2925 		comma = 1;
2926 	}
2927 	if (t_flags & TF_DELACK) {
2928 		db_printf("%sTF_DELACK", comma ? ", " : "");
2929 		comma = 1;
2930 	}
2931 	if (t_flags & TF_NODELAY) {
2932 		db_printf("%sTF_NODELAY", comma ? ", " : "");
2933 		comma = 1;
2934 	}
2935 	if (t_flags & TF_NOOPT) {
2936 		db_printf("%sTF_NOOPT", comma ? ", " : "");
2937 		comma = 1;
2938 	}
2939 	if (t_flags & TF_SENTFIN) {
2940 		db_printf("%sTF_SENTFIN", comma ? ", " : "");
2941 		comma = 1;
2942 	}
2943 	if (t_flags & TF_REQ_SCALE) {
2944 		db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2945 		comma = 1;
2946 	}
2947 	if (t_flags & TF_RCVD_SCALE) {
2948 		db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2949 		comma = 1;
2950 	}
2951 	if (t_flags & TF_REQ_TSTMP) {
2952 		db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2953 		comma = 1;
2954 	}
2955 	if (t_flags & TF_RCVD_TSTMP) {
2956 		db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2957 		comma = 1;
2958 	}
2959 	if (t_flags & TF_SACK_PERMIT) {
2960 		db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2961 		comma = 1;
2962 	}
2963 	if (t_flags & TF_NEEDSYN) {
2964 		db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2965 		comma = 1;
2966 	}
2967 	if (t_flags & TF_NEEDFIN) {
2968 		db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2969 		comma = 1;
2970 	}
2971 	if (t_flags & TF_NOPUSH) {
2972 		db_printf("%sTF_NOPUSH", comma ? ", " : "");
2973 		comma = 1;
2974 	}
2975 	if (t_flags & TF_PREVVALID) {
2976 		db_printf("%sTF_PREVVALID", comma ? ", " : "");
2977 		comma = 1;
2978 	}
2979 	if (t_flags & TF_MORETOCOME) {
2980 		db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2981 		comma = 1;
2982 	}
2983 	if (t_flags & TF_LQ_OVERFLOW) {
2984 		db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2985 		comma = 1;
2986 	}
2987 	if (t_flags & TF_LASTIDLE) {
2988 		db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2989 		comma = 1;
2990 	}
2991 	if (t_flags & TF_RXWIN0SENT) {
2992 		db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2993 		comma = 1;
2994 	}
2995 	if (t_flags & TF_FASTRECOVERY) {
2996 		db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2997 		comma = 1;
2998 	}
2999 	if (t_flags & TF_CONGRECOVERY) {
3000 		db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
3001 		comma = 1;
3002 	}
3003 	if (t_flags & TF_WASFRECOVERY) {
3004 		db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
3005 		comma = 1;
3006 	}
3007 	if (t_flags & TF_WASCRECOVERY) {
3008 		db_printf("%sTF_WASCRECOVERY", comma ? ", " : "");
3009 		comma = 1;
3010 	}
3011 	if (t_flags & TF_SIGNATURE) {
3012 		db_printf("%sTF_SIGNATURE", comma ? ", " : "");
3013 		comma = 1;
3014 	}
3015 	if (t_flags & TF_FORCEDATA) {
3016 		db_printf("%sTF_FORCEDATA", comma ? ", " : "");
3017 		comma = 1;
3018 	}
3019 	if (t_flags & TF_TSO) {
3020 		db_printf("%sTF_TSO", comma ? ", " : "");
3021 		comma = 1;
3022 	}
3023 	if (t_flags & TF_FASTOPEN) {
3024 		db_printf("%sTF_FASTOPEN", comma ? ", " : "");
3025 		comma = 1;
3026 	}
3027 }
3028 
3029 static void
3030 db_print_tflags2(u_int t_flags2)
3031 {
3032 	int comma;
3033 
3034 	comma = 0;
3035 	if (t_flags2 & TF2_PLPMTU_BLACKHOLE) {
3036 		db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : "");
3037 		comma = 1;
3038 	}
3039 	if (t_flags2 & TF2_PLPMTU_PMTUD) {
3040 		db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : "");
3041 		comma = 1;
3042 	}
3043 	if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) {
3044 		db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : "");
3045 		comma = 1;
3046 	}
3047 	if (t_flags2 & TF2_LOG_AUTO) {
3048 		db_printf("%sTF2_LOG_AUTO", comma ? ", " : "");
3049 		comma = 1;
3050 	}
3051 	if (t_flags2 & TF2_DROP_AF_DATA) {
3052 		db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : "");
3053 		comma = 1;
3054 	}
3055 	if (t_flags2 & TF2_ECN_PERMIT) {
3056 		db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
3057 		comma = 1;
3058 	}
3059 	if (t_flags2 & TF2_ECN_SND_CWR) {
3060 		db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : "");
3061 		comma = 1;
3062 	}
3063 	if (t_flags2 & TF2_ECN_SND_ECE) {
3064 		db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : "");
3065 		comma = 1;
3066 	}
3067 	if (t_flags2 & TF2_ACE_PERMIT) {
3068 		db_printf("%sTF2_ACE_PERMIT", comma ? ", " : "");
3069 		comma = 1;
3070 	}
3071 	if (t_flags2 & TF2_FBYTES_COMPLETE) {
3072 		db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : "");
3073 		comma = 1;
3074 	}
3075 }
3076 
3077 static void
3078 db_print_toobflags(char t_oobflags)
3079 {
3080 	int comma;
3081 
3082 	comma = 0;
3083 	if (t_oobflags & TCPOOB_HAVEDATA) {
3084 		db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
3085 		comma = 1;
3086 	}
3087 	if (t_oobflags & TCPOOB_HADDATA) {
3088 		db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
3089 		comma = 1;
3090 	}
3091 }
3092 
3093 static void
3094 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
3095 {
3096 
3097 	db_print_indent(indent);
3098 	db_printf("%s at %p\n", name, tp);
3099 
3100 	indent += 2;
3101 
3102 	db_print_indent(indent);
3103 	db_printf("t_segq first: %p   t_segqlen: %d   t_dupacks: %d\n",
3104 	   TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
3105 
3106 	db_print_indent(indent);
3107 	db_printf("tt_rexmt: %p   tt_persist: %p   tt_keep: %p\n",
3108 	    &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
3109 
3110 	db_print_indent(indent);
3111 	db_printf("tt_2msl: %p   tt_delack: %p   t_inpcb: %p\n", &tp->t_timers->tt_2msl,
3112 	    &tp->t_timers->tt_delack, tp->t_inpcb);
3113 
3114 	db_print_indent(indent);
3115 	db_printf("t_state: %d (", tp->t_state);
3116 	db_print_tstate(tp->t_state);
3117 	db_printf(")\n");
3118 
3119 	db_print_indent(indent);
3120 	db_printf("t_flags: 0x%x (", tp->t_flags);
3121 	db_print_tflags(tp->t_flags);
3122 	db_printf(")\n");
3123 
3124 	db_print_indent(indent);
3125 	db_printf("t_flags2: 0x%x (", tp->t_flags2);
3126 	db_print_tflags2(tp->t_flags2);
3127 	db_printf(")\n");
3128 
3129 	db_print_indent(indent);
3130 	db_printf("snd_una: 0x%08x   snd_max: 0x%08x   snd_nxt: x0%08x\n",
3131 	    tp->snd_una, tp->snd_max, tp->snd_nxt);
3132 
3133 	db_print_indent(indent);
3134 	db_printf("snd_up: 0x%08x   snd_wl1: 0x%08x   snd_wl2: 0x%08x\n",
3135 	   tp->snd_up, tp->snd_wl1, tp->snd_wl2);
3136 
3137 	db_print_indent(indent);
3138 	db_printf("iss: 0x%08x   irs: 0x%08x   rcv_nxt: 0x%08x\n",
3139 	    tp->iss, tp->irs, tp->rcv_nxt);
3140 
3141 	db_print_indent(indent);
3142 	db_printf("rcv_adv: 0x%08x   rcv_wnd: %u   rcv_up: 0x%08x\n",
3143 	    tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
3144 
3145 	db_print_indent(indent);
3146 	db_printf("snd_wnd: %u   snd_cwnd: %u\n",
3147 	   tp->snd_wnd, tp->snd_cwnd);
3148 
3149 	db_print_indent(indent);
3150 	db_printf("snd_ssthresh: %u   snd_recover: "
3151 	    "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
3152 
3153 	db_print_indent(indent);
3154 	db_printf("t_rcvtime: %u   t_startime: %u\n",
3155 	    tp->t_rcvtime, tp->t_starttime);
3156 
3157 	db_print_indent(indent);
3158 	db_printf("t_rttime: %u   t_rtsq: 0x%08x\n",
3159 	    tp->t_rtttime, tp->t_rtseq);
3160 
3161 	db_print_indent(indent);
3162 	db_printf("t_rxtcur: %d   t_maxseg: %u   t_srtt: %d\n",
3163 	    tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
3164 
3165 	db_print_indent(indent);
3166 	db_printf("t_rttvar: %d   t_rxtshift: %d   t_rttmin: %u   "
3167 	    "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
3168 	    tp->t_rttbest);
3169 
3170 	db_print_indent(indent);
3171 	db_printf("t_rttupdated: %lu   max_sndwnd: %u   t_softerror: %d\n",
3172 	    tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
3173 
3174 	db_print_indent(indent);
3175 	db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
3176 	db_print_toobflags(tp->t_oobflags);
3177 	db_printf(")   t_iobc: 0x%02x\n", tp->t_iobc);
3178 
3179 	db_print_indent(indent);
3180 	db_printf("snd_scale: %u   rcv_scale: %u   request_r_scale: %u\n",
3181 	    tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
3182 
3183 	db_print_indent(indent);
3184 	db_printf("ts_recent: %u   ts_recent_age: %u\n",
3185 	    tp->ts_recent, tp->ts_recent_age);
3186 
3187 	db_print_indent(indent);
3188 	db_printf("ts_offset: %u   last_ack_sent: 0x%08x   snd_cwnd_prev: "
3189 	    "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
3190 
3191 	db_print_indent(indent);
3192 	db_printf("snd_ssthresh_prev: %u   snd_recover_prev: 0x%08x   "
3193 	    "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
3194 	    tp->snd_recover_prev, tp->t_badrxtwin);
3195 
3196 	db_print_indent(indent);
3197 	db_printf("snd_numholes: %d  snd_holes first: %p\n",
3198 	    tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
3199 
3200 	db_print_indent(indent);
3201 	db_printf("snd_fack: 0x%08x   rcv_numsacks: %d\n",
3202 	    tp->snd_fack, tp->rcv_numsacks);
3203 
3204 	/* Skip sackblks, sackhint. */
3205 
3206 	db_print_indent(indent);
3207 	db_printf("t_rttlow: %d   rfbuf_ts: %u   rfbuf_cnt: %d\n",
3208 	    tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
3209 }
3210 
3211 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
3212 {
3213 	struct tcpcb *tp;
3214 
3215 	if (!have_addr) {
3216 		db_printf("usage: show tcpcb <addr>\n");
3217 		return;
3218 	}
3219 	tp = (struct tcpcb *)addr;
3220 
3221 	db_print_tcpcb(tp, "tcpcb", 0);
3222 }
3223 #endif
3224