xref: /openbsd-src/sys/netinet/tcp_usrreq.c (revision c7e8ea31cd41a963f06f0a8ba93948b06aa6b4a4)
1 /*	$OpenBSD: tcp_usrreq.c,v 1.153 2017/08/15 17:47:15 bluhm Exp $	*/
2 /*	$NetBSD: tcp_usrreq.c,v 1.20 1996/02/13 23:44:16 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)COPYRIGHT	1.1 (NRL) 17 January 1995
33  *
34  * NRL grants permission for redistribution and use in source and binary
35  * forms, with or without modification, of the software and documentation
36  * created at NRL provided that the following conditions are met:
37  *
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgements:
45  *	This product includes software developed by the University of
46  *	California, Berkeley and its contributors.
47  *	This product includes software developed at the Information
48  *	Technology Division, US Naval Research Laboratory.
49  * 4. Neither the name of the NRL nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
54  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
55  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
56  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
57  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
58  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
59  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
60  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
61  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
62  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
63  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
64  *
65  * The views and conclusions contained in the software and documentation
66  * are those of the authors and should not be interpreted as representing
67  * official policies, either expressed or implied, of the US Naval
68  * Research Laboratory (NRL).
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/mbuf.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/protosw.h>
77 #include <sys/stat.h>
78 #include <sys/sysctl.h>
79 #include <sys/domain.h>
80 #include <sys/kernel.h>
81 #include <sys/pool.h>
82 
83 #include <net/if.h>
84 #include <net/if_var.h>
85 #include <net/route.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/ip_var.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcp_debug.h>
98 
99 #ifdef INET6
100 #include <netinet6/in6_var.h>
101 #endif
102 
103 #ifndef TCP_SENDSPACE
104 #define	TCP_SENDSPACE	1024*16
105 #endif
106 u_int	tcp_sendspace = TCP_SENDSPACE;
107 #ifndef TCP_RECVSPACE
108 #define	TCP_RECVSPACE	1024*16
109 #endif
110 u_int	tcp_recvspace = TCP_RECVSPACE;
111 u_int	tcp_autorcvbuf_inc = 16 * 1024;
112 
113 int *tcpctl_vars[TCPCTL_MAXID] = TCPCTL_VARS;
114 
115 struct	inpcbtable tcbtable;
116 
117 int tcp_ident(void *, size_t *, void *, size_t, int);
118 
119 /*
120  * Process a TCP user request for TCP tb.  If this is a send request
121  * then m is the mbuf chain of send data.  If this is a timer expiration
122  * (called from the software clock routine), then timertype tells which timer.
123  */
124 /*ARGSUSED*/
125 int
126 tcp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
127     struct mbuf *control, struct proc *p)
128 {
129 	struct inpcb *inp;
130 	struct tcpcb *tp = NULL;
131 	int error = 0;
132 	short ostate;
133 
134 	NET_ASSERT_LOCKED();
135 
136 	if (req == PRU_CONTROL) {
137 #ifdef INET6
138 		if (sotopf(so) == PF_INET6)
139 			return in6_control(so, (u_long)m, (caddr_t)nam,
140 			    (struct ifnet *)control);
141 		else
142 #endif /* INET6 */
143 			return (in_control(so, (u_long)m, (caddr_t)nam,
144 			    (struct ifnet *)control));
145 	}
146 	if (control && control->m_len) {
147 		m_freem(control);
148 		m_freem(m);
149 		return (EINVAL);
150 	}
151 
152 	inp = sotoinpcb(so);
153 	/*
154 	 * When a TCP is attached to a socket, then there will be
155 	 * a (struct inpcb) pointed at by the socket, and this
156 	 * structure will point at a subsidiary (struct tcpcb).
157 	 */
158 	if (inp == NULL) {
159 		error = so->so_error;
160 		if (error == 0)
161 			error = EINVAL;
162 		/*
163 		 * The following corrects an mbuf leak under rare
164 		 * circumstances
165 		 */
166 		if (req == PRU_SEND || req == PRU_SENDOOB)
167 			m_freem(m);
168 		return (error);
169 	}
170 	if (inp) {
171 		tp = intotcpcb(inp);
172 		/* tp might get 0 when using socket splicing */
173 		if (tp == NULL) {
174 			return (0);
175 		}
176 #ifdef KPROF
177 		tcp_acounts[tp->t_state][req]++;
178 #endif
179 		ostate = tp->t_state;
180 	} else
181 		ostate = 0;
182 	switch (req) {
183 
184 	/*
185 	 * PRU_DETACH detaches the TCP protocol from the socket.
186 	 * If the protocol state is non-embryonic, then can't
187 	 * do this directly: have to initiate a PRU_DISCONNECT,
188 	 * which may finish later; embryonic TCB's can just
189 	 * be discarded here.
190 	 */
191 	case PRU_DETACH:
192 		tp = tcp_disconnect(tp);
193 		break;
194 
195 	/*
196 	 * Give the socket an address.
197 	 */
198 	case PRU_BIND:
199 		error = in_pcbbind(inp, nam, p);
200 		break;
201 
202 	/*
203 	 * Prepare to accept connections.
204 	 */
205 	case PRU_LISTEN:
206 		if (inp->inp_lport == 0)
207 			error = in_pcbbind(inp, NULL, p);
208 		/* If the in_pcbbind() above is called, the tp->pf
209 		   should still be whatever it was before. */
210 		if (error == 0)
211 			tp->t_state = TCPS_LISTEN;
212 		break;
213 
214 	/*
215 	 * Initiate connection to peer.
216 	 * Create a template for use in transmissions on this connection.
217 	 * Enter SYN_SENT state, and mark socket as connecting.
218 	 * Start keep-alive timer, and seed output sequence space.
219 	 * Send initial segment on connection.
220 	 */
221 	case PRU_CONNECT:
222 #ifdef INET6
223 		if (inp->inp_flags & INP_IPV6) {
224 			struct sockaddr_in6 *sin6;
225 
226 			if ((error = in6_nam2sin6(nam, &sin6)))
227 				break;
228 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
229 			    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
230 				error = EINVAL;
231 				break;
232 			}
233 			error = in6_pcbconnect(inp, nam);
234 		} else
235 #endif /* INET6 */
236 		{
237 			struct sockaddr_in *sin;
238 
239 			if ((error = in_nam2sin(nam, &sin)))
240 				break;
241 			if ((sin->sin_addr.s_addr == INADDR_ANY) ||
242 			    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
243 			    IN_MULTICAST(sin->sin_addr.s_addr) ||
244 			    in_broadcast(sin->sin_addr, inp->inp_rtableid)) {
245 				error = EINVAL;
246 				break;
247 			}
248 			error = in_pcbconnect(inp, nam);
249 		}
250 		if (error)
251 			break;
252 
253 		tp->t_template = tcp_template(tp);
254 		if (tp->t_template == 0) {
255 			in_pcbdisconnect(inp);
256 			error = ENOBUFS;
257 			break;
258 		}
259 
260 		so->so_state |= SS_CONNECTOUT;
261 
262 		/* Compute window scaling to request.  */
263 		tcp_rscale(tp, sb_max);
264 
265 		soisconnecting(so);
266 		tcpstat_inc(tcps_connattempt);
267 		tp->t_state = TCPS_SYN_SENT;
268 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
269 		tcp_set_iss_tsm(tp);
270 		tcp_sendseqinit(tp);
271 #if defined(TCP_SACK)
272 		tp->snd_last = tp->snd_una;
273 #endif
274 #if defined(TCP_SACK) && defined(TCP_FACK)
275 		tp->snd_fack = tp->snd_una;
276 		tp->retran_data = 0;
277 		tp->snd_awnd = 0;
278 #endif
279 		error = tcp_output(tp);
280 		break;
281 
282 	/*
283 	 * Create a TCP connection between two sockets.
284 	 */
285 	case PRU_CONNECT2:
286 		error = EOPNOTSUPP;
287 		break;
288 
289 	/*
290 	 * Initiate disconnect from peer.
291 	 * If connection never passed embryonic stage, just drop;
292 	 * else if don't need to let data drain, then can just drop anyways,
293 	 * else have to begin TCP shutdown process: mark socket disconnecting,
294 	 * drain unread data, state switch to reflect user close, and
295 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
296 	 * when peer sends FIN and acks ours.
297 	 *
298 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
299 	 */
300 	case PRU_DISCONNECT:
301 		tp = tcp_disconnect(tp);
302 		break;
303 
304 	/*
305 	 * Accept a connection.  Essentially all the work is
306 	 * done at higher levels; just return the address
307 	 * of the peer, storing through addr.
308 	 */
309 	case PRU_ACCEPT:
310 #ifdef INET6
311 		if (inp->inp_flags & INP_IPV6)
312 			in6_setpeeraddr(inp, nam);
313 		else
314 #endif
315 			in_setpeeraddr(inp, nam);
316 		break;
317 
318 	/*
319 	 * Mark the connection as being incapable of further output.
320 	 */
321 	case PRU_SHUTDOWN:
322 		if (so->so_state & SS_CANTSENDMORE)
323 			break;
324 		socantsendmore(so);
325 		tp = tcp_usrclosed(tp);
326 		if (tp)
327 			error = tcp_output(tp);
328 		break;
329 
330 	/*
331 	 * After a receive, possibly send window update to peer.
332 	 */
333 	case PRU_RCVD:
334 		/*
335 		 * soreceive() calls this function when a user receives
336 		 * ancillary data on a listening socket. We don't call
337 		 * tcp_output in such a case, since there is no header
338 		 * template for a listening socket and hence the kernel
339 		 * will panic.
340 		 */
341 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) != 0)
342 			(void) tcp_output(tp);
343 		break;
344 
345 	/*
346 	 * Do a send by putting data in output queue and updating urgent
347 	 * marker if URG set.  Possibly send more data.
348 	 */
349 	case PRU_SEND:
350 		sbappendstream(so, &so->so_snd, m);
351 		error = tcp_output(tp);
352 		break;
353 
354 	/*
355 	 * Abort the TCP.
356 	 */
357 	case PRU_ABORT:
358 		tp = tcp_drop(tp, ECONNABORTED);
359 		break;
360 
361 	case PRU_SENSE:
362 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
363 		return (0);
364 
365 	case PRU_RCVOOB:
366 		if ((so->so_oobmark == 0 &&
367 		    (so->so_state & SS_RCVATMARK) == 0) ||
368 		    so->so_options & SO_OOBINLINE ||
369 		    tp->t_oobflags & TCPOOB_HADDATA) {
370 			error = EINVAL;
371 			break;
372 		}
373 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
374 			error = EWOULDBLOCK;
375 			break;
376 		}
377 		m->m_len = 1;
378 		*mtod(m, caddr_t) = tp->t_iobc;
379 		if (((long)nam & MSG_PEEK) == 0)
380 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
381 		break;
382 
383 	case PRU_SENDOOB:
384 		if (sbspace(so, &so->so_snd) < -512) {
385 			m_freem(m);
386 			error = ENOBUFS;
387 			break;
388 		}
389 		/*
390 		 * According to RFC961 (Assigned Protocols),
391 		 * the urgent pointer points to the last octet
392 		 * of urgent data.  We continue, however,
393 		 * to consider it to indicate the first octet
394 		 * of data past the urgent section.
395 		 * Otherwise, snd_up should be one lower.
396 		 */
397 		sbappendstream(so, &so->so_snd, m);
398 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
399 		tp->t_force = 1;
400 		error = tcp_output(tp);
401 		tp->t_force = 0;
402 		break;
403 
404 	case PRU_SOCKADDR:
405 #ifdef INET6
406 		if (inp->inp_flags & INP_IPV6)
407 			in6_setsockaddr(inp, nam);
408 		else
409 #endif
410 			in_setsockaddr(inp, nam);
411 		break;
412 
413 	case PRU_PEERADDR:
414 #ifdef INET6
415 		if (inp->inp_flags & INP_IPV6)
416 			in6_setpeeraddr(inp, nam);
417 		else
418 #endif
419 			in_setpeeraddr(inp, nam);
420 		break;
421 
422 	default:
423 		panic("tcp_usrreq");
424 	}
425 	if (tp && (so->so_options & SO_DEBUG))
426 		tcp_trace(TA_USER, ostate, tp, (caddr_t)0, req, 0);
427 	return (error);
428 }
429 
430 int
431 tcp_ctloutput(int op, struct socket *so, int level, int optname,
432     struct mbuf *m)
433 {
434 	int error = 0;
435 	struct inpcb *inp;
436 	struct tcpcb *tp;
437 	int i;
438 
439 	inp = sotoinpcb(so);
440 	if (inp == NULL) {
441 		if (op == PRCO_SETOPT)
442 			(void) m_free(m);
443 		return (ECONNRESET);
444 	}
445 	if (level != IPPROTO_TCP) {
446 		switch (so->so_proto->pr_domain->dom_family) {
447 #ifdef INET6
448 		case PF_INET6:
449 			error = ip6_ctloutput(op, so, level, optname, m);
450 			break;
451 #endif /* INET6 */
452 		case PF_INET:
453 			error = ip_ctloutput(op, so, level, optname, m);
454 			break;
455 		default:
456 			error = EAFNOSUPPORT;	/*?*/
457 			break;
458 		}
459 		return (error);
460 	}
461 	tp = intotcpcb(inp);
462 
463 	switch (op) {
464 
465 	case PRCO_SETOPT:
466 		switch (optname) {
467 
468 		case TCP_NODELAY:
469 			if (m == NULL || m->m_len < sizeof (int))
470 				error = EINVAL;
471 			else if (*mtod(m, int *))
472 				tp->t_flags |= TF_NODELAY;
473 			else
474 				tp->t_flags &= ~TF_NODELAY;
475 			break;
476 
477 		case TCP_NOPUSH:
478 			if (m == NULL || m->m_len < sizeof (int))
479 				error = EINVAL;
480 			else if (*mtod(m, int *))
481 				tp->t_flags |= TF_NOPUSH;
482 			else if (tp->t_flags & TF_NOPUSH) {
483 				tp->t_flags &= ~TF_NOPUSH;
484 				if (TCPS_HAVEESTABLISHED(tp->t_state))
485 					error = tcp_output(tp);
486 			}
487 			break;
488 
489 		case TCP_MAXSEG:
490 			if (m == NULL || m->m_len < sizeof (int)) {
491 				error = EINVAL;
492 				break;
493 			}
494 
495 			i = *mtod(m, int *);
496 			if (i > 0 && i <= tp->t_maxseg)
497 				tp->t_maxseg = i;
498 			else
499 				error = EINVAL;
500 			break;
501 
502 #ifdef TCP_SACK
503 		case TCP_SACK_ENABLE:
504 			if (m == NULL || m->m_len < sizeof (int)) {
505 				error = EINVAL;
506 				break;
507 			}
508 
509 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
510 				error = EPERM;
511 				break;
512 			}
513 
514 			if (tp->t_flags & TF_SIGNATURE) {
515 				error = EPERM;
516 				break;
517 			}
518 
519 			if (*mtod(m, int *))
520 				tp->sack_enable = 1;
521 			else
522 				tp->sack_enable = 0;
523 			break;
524 #endif
525 #ifdef TCP_SIGNATURE
526 		case TCP_MD5SIG:
527 			if (m == NULL || m->m_len < sizeof (int)) {
528 				error = EINVAL;
529 				break;
530 			}
531 
532 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
533 				error = EPERM;
534 				break;
535 			}
536 
537 			if (*mtod(m, int *)) {
538 				tp->t_flags |= TF_SIGNATURE;
539 #ifdef TCP_SACK
540 				tp->sack_enable = 0;
541 #endif /* TCP_SACK */
542 			} else
543 				tp->t_flags &= ~TF_SIGNATURE;
544 			break;
545 #endif /* TCP_SIGNATURE */
546 		default:
547 			error = ENOPROTOOPT;
548 			break;
549 		}
550 		m_free(m);
551 		break;
552 
553 	case PRCO_GETOPT:
554 		m->m_len = sizeof(int);
555 
556 		switch (optname) {
557 		case TCP_NODELAY:
558 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
559 			break;
560 		case TCP_NOPUSH:
561 			*mtod(m, int *) = tp->t_flags & TF_NOPUSH;
562 			break;
563 		case TCP_MAXSEG:
564 			*mtod(m, int *) = tp->t_maxseg;
565 			break;
566 #ifdef TCP_SACK
567 		case TCP_SACK_ENABLE:
568 			*mtod(m, int *) = tp->sack_enable;
569 			break;
570 #endif
571 #ifdef TCP_SIGNATURE
572 		case TCP_MD5SIG:
573 			*mtod(m, int *) = tp->t_flags & TF_SIGNATURE;
574 			break;
575 #endif
576 		default:
577 			error = ENOPROTOOPT;
578 			break;
579 		}
580 		break;
581 	}
582 	return (error);
583 }
584 
585 /*
586  * Attach TCP protocol to socket, allocating
587  * internet protocol control block, tcp control block,
588  * bufer space, and entering LISTEN state if to accept connections.
589  */
590 int
591 tcp_attach(struct socket *so, int proto)
592 {
593 	struct tcpcb *tp;
594 	struct inpcb *inp;
595 	int error;
596 
597 	if (so->so_pcb)
598 		return EISCONN;
599 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0 ||
600 	    sbcheckreserve(so->so_snd.sb_wat, tcp_sendspace) ||
601 	    sbcheckreserve(so->so_rcv.sb_wat, tcp_recvspace)) {
602 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
603 		if (error)
604 			return (error);
605 	}
606 
607 	error = in_pcballoc(so, &tcbtable);
608 	if (error)
609 		return (error);
610 	inp = sotoinpcb(so);
611 	tp = tcp_newtcpcb(inp);
612 	if (tp == NULL) {
613 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
614 
615 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
616 		in_pcbdetach(inp);
617 		so->so_state |= nofd;
618 		return (ENOBUFS);
619 	}
620 	tp->t_state = TCPS_CLOSED;
621 #ifdef INET6
622 	/* we disallow IPv4 mapped address completely. */
623 	if (inp->inp_flags & INP_IPV6)
624 		tp->pf = PF_INET6;
625 	else
626 		tp->pf = PF_INET;
627 #else
628 	tp->pf = PF_INET;
629 #endif
630 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
631 		so->so_linger = TCP_LINGERTIME;
632 
633 	if (tp && (so->so_options & SO_DEBUG))
634 		tcp_trace(TA_USER, 0, tp, (caddr_t)0, 0 /* XXX */, 0);
635 	return (0);
636 }
637 
638 /*
639  * Initiate (or continue) disconnect.
640  * If embryonic state, just send reset (once).
641  * If in ``let data drain'' option and linger null, just drop.
642  * Otherwise (hard), mark socket disconnecting and drop
643  * current input data; switch states based on user close, and
644  * send segment to peer (with FIN).
645  */
646 struct tcpcb *
647 tcp_disconnect(struct tcpcb *tp)
648 {
649 	struct socket *so = tp->t_inpcb->inp_socket;
650 
651 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
652 		tp = tcp_close(tp);
653 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
654 		tp = tcp_drop(tp, 0);
655 	else {
656 		soisdisconnecting(so);
657 		sbflush(so, &so->so_rcv);
658 		tp = tcp_usrclosed(tp);
659 		if (tp)
660 			(void) tcp_output(tp);
661 	}
662 	return (tp);
663 }
664 
665 /*
666  * User issued close, and wish to trail through shutdown states:
667  * if never received SYN, just forget it.  If got a SYN from peer,
668  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
669  * If already got a FIN from peer, then almost done; go to LAST_ACK
670  * state.  In all other cases, have already sent FIN to peer (e.g.
671  * after PRU_SHUTDOWN), and just have to play tedious game waiting
672  * for peer to send FIN or not respond to keep-alives, etc.
673  * We can let the user exit from the close as soon as the FIN is acked.
674  */
675 struct tcpcb *
676 tcp_usrclosed(struct tcpcb *tp)
677 {
678 
679 	switch (tp->t_state) {
680 
681 	case TCPS_CLOSED:
682 	case TCPS_LISTEN:
683 	case TCPS_SYN_SENT:
684 		tp->t_state = TCPS_CLOSED;
685 		tp = tcp_close(tp);
686 		break;
687 
688 	case TCPS_SYN_RECEIVED:
689 	case TCPS_ESTABLISHED:
690 		tp->t_state = TCPS_FIN_WAIT_1;
691 		break;
692 
693 	case TCPS_CLOSE_WAIT:
694 		tp->t_state = TCPS_LAST_ACK;
695 		break;
696 	}
697 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
698 		soisdisconnected(tp->t_inpcb->inp_socket);
699 		/*
700 		 * If we are in FIN_WAIT_2, we arrived here because the
701 		 * application did a shutdown of the send side.  Like the
702 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
703 		 * a full close, we start a timer to make sure sockets are
704 		 * not left in FIN_WAIT_2 forever.
705 		 */
706 		if (tp->t_state == TCPS_FIN_WAIT_2)
707 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
708 	}
709 	return (tp);
710 }
711 
712 /*
713  * Look up a socket for ident or tcpdrop, ...
714  */
715 int
716 tcp_ident(void *oldp, size_t *oldlenp, void *newp, size_t newlen, int dodrop)
717 {
718 	int error = 0;
719 	struct tcp_ident_mapping tir;
720 	struct inpcb *inp;
721 	struct tcpcb *tp = NULL;
722 	struct sockaddr_in *fin, *lin;
723 #ifdef INET6
724 	struct sockaddr_in6 *fin6, *lin6;
725 	struct in6_addr f6, l6;
726 #endif
727 
728 	NET_ASSERT_LOCKED();
729 
730 	if (dodrop) {
731 		if (oldp != NULL || *oldlenp != 0)
732 			return (EINVAL);
733 		if (newp == NULL)
734 			return (EPERM);
735 		if (newlen < sizeof(tir))
736 			return (ENOMEM);
737 		if ((error = copyin(newp, &tir, sizeof (tir))) != 0 )
738 			return (error);
739 	} else {
740 		if (oldp == NULL)
741 			return (EINVAL);
742 		if (*oldlenp < sizeof(tir))
743 			return (ENOMEM);
744 		if (newp != NULL || newlen != 0)
745 			return (EINVAL);
746 		if ((error = copyin(oldp, &tir, sizeof (tir))) != 0 )
747 			return (error);
748 	}
749 	switch (tir.faddr.ss_family) {
750 #ifdef INET6
751 	case AF_INET6:
752 		fin6 = (struct sockaddr_in6 *)&tir.faddr;
753 		error = in6_embedscope(&f6, fin6, NULL);
754 		if (error)
755 			return EINVAL;	/*?*/
756 		lin6 = (struct sockaddr_in6 *)&tir.laddr;
757 		error = in6_embedscope(&l6, lin6, NULL);
758 		if (error)
759 			return EINVAL;	/*?*/
760 		break;
761 #endif
762 	case AF_INET:
763 		fin = (struct sockaddr_in *)&tir.faddr;
764 		lin = (struct sockaddr_in *)&tir.laddr;
765 		break;
766 	default:
767 		return (EINVAL);
768 	}
769 
770 	switch (tir.faddr.ss_family) {
771 #ifdef INET6
772 	case AF_INET6:
773 		inp = in6_pcbhashlookup(&tcbtable, &f6,
774 		    fin6->sin6_port, &l6, lin6->sin6_port, tir.rdomain);
775 		break;
776 #endif
777 	case AF_INET:
778 		inp = in_pcbhashlookup(&tcbtable, fin->sin_addr,
779 		    fin->sin_port, lin->sin_addr, lin->sin_port, tir.rdomain);
780 		break;
781 	default:
782 		unhandled_af(tir.faddr.ss_family);
783 	}
784 
785 	if (dodrop) {
786 		if (inp && (tp = intotcpcb(inp)) &&
787 		    ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0))
788 			tp = tcp_drop(tp, ECONNABORTED);
789 		else
790 			error = ESRCH;
791 		return (error);
792 	}
793 
794 	if (inp == NULL) {
795 		tcpstat_inc(tcps_pcbhashmiss);
796 		switch (tir.faddr.ss_family) {
797 #ifdef INET6
798 		case AF_INET6:
799 			inp = in6_pcblookup_listen(&tcbtable,
800 			    &l6, lin6->sin6_port, 0, NULL, tir.rdomain);
801 			break;
802 #endif
803 		case AF_INET:
804 			inp = in_pcblookup_listen(&tcbtable,
805 			    lin->sin_addr, lin->sin_port, 0, NULL, tir.rdomain);
806 			break;
807 		}
808 	}
809 
810 	if (inp != NULL && (inp->inp_socket->so_state & SS_CONNECTOUT)) {
811 		tir.ruid = inp->inp_socket->so_ruid;
812 		tir.euid = inp->inp_socket->so_euid;
813 	} else {
814 		tir.ruid = -1;
815 		tir.euid = -1;
816 	}
817 
818 	*oldlenp = sizeof (tir);
819 	error = copyout((void *)&tir, oldp, sizeof (tir));
820 	return (error);
821 }
822 
823 int
824 tcp_sysctl_tcpstat(void *oldp, size_t *oldlenp, void *newp)
825 {
826 	uint64_t counters[tcps_ncounters];
827 	struct tcpstat tcpstat;
828 	struct syn_cache_set *set;
829 	int i = 0;
830 
831 #define ASSIGN(field)	do { tcpstat.field = counters[i++]; } while (0)
832 
833 	memset(&tcpstat, 0, sizeof tcpstat);
834 	counters_read(tcpcounters, counters, nitems(counters));
835 	ASSIGN(tcps_connattempt);
836 	ASSIGN(tcps_accepts);
837 	ASSIGN(tcps_connects);
838 	ASSIGN(tcps_drops);
839 	ASSIGN(tcps_conndrops);
840 	ASSIGN(tcps_closed);
841 	ASSIGN(tcps_segstimed);
842 	ASSIGN(tcps_rttupdated);
843 	ASSIGN(tcps_delack);
844 	ASSIGN(tcps_timeoutdrop);
845 	ASSIGN(tcps_rexmttimeo);
846 	ASSIGN(tcps_persisttimeo);
847 	ASSIGN(tcps_persistdrop);
848 	ASSIGN(tcps_keeptimeo);
849 	ASSIGN(tcps_keepprobe);
850 	ASSIGN(tcps_keepdrops);
851 	ASSIGN(tcps_sndtotal);
852 	ASSIGN(tcps_sndpack);
853 	ASSIGN(tcps_sndbyte);
854 	ASSIGN(tcps_sndrexmitpack);
855 	ASSIGN(tcps_sndrexmitbyte);
856 	ASSIGN(tcps_sndrexmitfast);
857 	ASSIGN(tcps_sndacks);
858 	ASSIGN(tcps_sndprobe);
859 	ASSIGN(tcps_sndurg);
860 	ASSIGN(tcps_sndwinup);
861 	ASSIGN(tcps_sndctrl);
862 	ASSIGN(tcps_rcvtotal);
863 	ASSIGN(tcps_rcvpack);
864 	ASSIGN(tcps_rcvbyte);
865 	ASSIGN(tcps_rcvbadsum);
866 	ASSIGN(tcps_rcvbadoff);
867 	ASSIGN(tcps_rcvmemdrop);
868 	ASSIGN(tcps_rcvnosec);
869 	ASSIGN(tcps_rcvshort);
870 	ASSIGN(tcps_rcvduppack);
871 	ASSIGN(tcps_rcvdupbyte);
872 	ASSIGN(tcps_rcvpartduppack);
873 	ASSIGN(tcps_rcvpartdupbyte);
874 	ASSIGN(tcps_rcvoopack);
875 	ASSIGN(tcps_rcvoobyte);
876 	ASSIGN(tcps_rcvpackafterwin);
877 	ASSIGN(tcps_rcvbyteafterwin);
878 	ASSIGN(tcps_rcvafterclose);
879 	ASSIGN(tcps_rcvwinprobe);
880 	ASSIGN(tcps_rcvdupack);
881 	ASSIGN(tcps_rcvacktoomuch);
882 	ASSIGN(tcps_rcvacktooold);
883 	ASSIGN(tcps_rcvackpack);
884 	ASSIGN(tcps_rcvackbyte);
885 	ASSIGN(tcps_rcvwinupd);
886 	ASSIGN(tcps_pawsdrop);
887 	ASSIGN(tcps_predack);
888 	ASSIGN(tcps_preddat);
889 	ASSIGN(tcps_pcbhashmiss);
890 	ASSIGN(tcps_noport);
891 	ASSIGN(tcps_badsyn);
892 	ASSIGN(tcps_dropsyn);
893 	ASSIGN(tcps_rcvbadsig);
894 	ASSIGN(tcps_rcvgoodsig);
895 	ASSIGN(tcps_inswcsum);
896 	ASSIGN(tcps_outswcsum);
897 	ASSIGN(tcps_ecn_accepts);
898 	ASSIGN(tcps_ecn_rcvece);
899 	ASSIGN(tcps_ecn_rcvcwr);
900 	ASSIGN(tcps_ecn_rcvce);
901 	ASSIGN(tcps_ecn_sndect);
902 	ASSIGN(tcps_ecn_sndece);
903 	ASSIGN(tcps_ecn_sndcwr);
904 	ASSIGN(tcps_cwr_ecn);
905 	ASSIGN(tcps_cwr_frecovery);
906 	ASSIGN(tcps_cwr_timeout);
907 	ASSIGN(tcps_sc_added);
908 	ASSIGN(tcps_sc_completed);
909 	ASSIGN(tcps_sc_timed_out);
910 	ASSIGN(tcps_sc_overflowed);
911 	ASSIGN(tcps_sc_reset);
912 	ASSIGN(tcps_sc_unreach);
913 	ASSIGN(tcps_sc_bucketoverflow);
914 	ASSIGN(tcps_sc_aborted);
915 	ASSIGN(tcps_sc_dupesyn);
916 	ASSIGN(tcps_sc_dropped);
917 	ASSIGN(tcps_sc_collisions);
918 	ASSIGN(tcps_sc_retransmitted);
919 	ASSIGN(tcps_sc_seedrandom);
920 	ASSIGN(tcps_sc_hash_size);
921 	ASSIGN(tcps_sc_entry_count);
922 	ASSIGN(tcps_sc_entry_limit);
923 	ASSIGN(tcps_sc_bucket_maxlen);
924 	ASSIGN(tcps_sc_bucket_limit);
925 	ASSIGN(tcps_sc_uses_left);
926 	ASSIGN(tcps_conndrained);
927 	ASSIGN(tcps_sack_recovery_episode);
928 	ASSIGN(tcps_sack_rexmits);
929 	ASSIGN(tcps_sack_rexmit_bytes);
930 	ASSIGN(tcps_sack_rcv_opts);
931 	ASSIGN(tcps_sack_snd_opts);
932 
933 #undef ASSIGN
934 
935 	set = &tcp_syn_cache[tcp_syn_cache_active];
936 	tcpstat.tcps_sc_hash_size = set->scs_size;
937 	tcpstat.tcps_sc_entry_count = set->scs_count;
938 	tcpstat.tcps_sc_entry_limit = tcp_syn_cache_limit;
939 	tcpstat.tcps_sc_bucket_maxlen = 0;
940 	for (i = 0; i < set->scs_size; i++) {
941 		if (tcpstat.tcps_sc_bucket_maxlen <
942 		    set->scs_buckethead[i].sch_length)
943 			tcpstat.tcps_sc_bucket_maxlen =
944 				set->scs_buckethead[i].sch_length;
945 	}
946 	tcpstat.tcps_sc_bucket_limit = tcp_syn_bucket_limit;
947 	tcpstat.tcps_sc_uses_left = set->scs_use;
948 
949 	return (sysctl_rdstruct(oldp, oldlenp, newp,
950 	    &tcpstat, sizeof(tcpstat)));
951 }
952 
953 /*
954  * Sysctl for tcp variables.
955  */
956 int
957 tcp_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
958     size_t newlen)
959 {
960 	int error, nval;
961 
962 	NET_ASSERT_LOCKED();
963 
964 	/* All sysctl names at this level are terminal. */
965 	if (namelen != 1)
966 		return (ENOTDIR);
967 
968 	switch (name[0]) {
969 #ifdef TCP_SACK
970 	case TCPCTL_SACK:
971 		return (sysctl_int(oldp, oldlenp, newp, newlen,
972 		    &tcp_do_sack));
973 #endif
974 	case TCPCTL_SLOWHZ:
975 		return (sysctl_rdint(oldp, oldlenp, newp, PR_SLOWHZ));
976 
977 	case TCPCTL_BADDYNAMIC:
978 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
979 		    baddynamicports.tcp, sizeof(baddynamicports.tcp)));
980 
981 	case TCPCTL_ROOTONLY:
982 		if (newp && securelevel > 0)
983 			return (EPERM);
984 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
985 		    rootonlyports.tcp, sizeof(rootonlyports.tcp)));
986 
987 	case TCPCTL_IDENT:
988 		return (tcp_ident(oldp, oldlenp, newp, newlen, 0));
989 
990 	case TCPCTL_DROP:
991 		return (tcp_ident(oldp, oldlenp, newp, newlen, 1));
992 
993 	case TCPCTL_ALWAYS_KEEPALIVE:
994 		return (sysctl_int(oldp, oldlenp, newp, newlen,
995 		    &tcp_always_keepalive));
996 
997 #ifdef TCP_ECN
998 	case TCPCTL_ECN:
999 		return (sysctl_int(oldp, oldlenp, newp, newlen,
1000 		   &tcp_do_ecn));
1001 #endif
1002 	case TCPCTL_REASS_LIMIT:
1003 		nval = tcp_reass_limit;
1004 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1005 		if (error)
1006 			return (error);
1007 		if (nval != tcp_reass_limit) {
1008 			error = pool_sethardlimit(&tcpqe_pool, nval, NULL, 0);
1009 			if (error)
1010 				return (error);
1011 			tcp_reass_limit = nval;
1012 		}
1013 		return (0);
1014 #ifdef TCP_SACK
1015 	case TCPCTL_SACKHOLE_LIMIT:
1016 		nval = tcp_sackhole_limit;
1017 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1018 		if (error)
1019 			return (error);
1020 		if (nval != tcp_sackhole_limit) {
1021 			error = pool_sethardlimit(&sackhl_pool, nval, NULL, 0);
1022 			if (error)
1023 				return (error);
1024 			tcp_sackhole_limit = nval;
1025 		}
1026 		return (0);
1027 #endif
1028 
1029 	case TCPCTL_STATS:
1030 		return (tcp_sysctl_tcpstat(oldp, oldlenp, newp));
1031 
1032 	case TCPCTL_SYN_USE_LIMIT:
1033 		error = sysctl_int(oldp, oldlenp, newp, newlen,
1034 		    &tcp_syn_use_limit);
1035 		if (error)
1036 			return (error);
1037 		if (newp != NULL) {
1038 			/*
1039 			 * Global tcp_syn_use_limit is used when reseeding a
1040 			 * new cache.  Also update the value in active cache.
1041 			 */
1042 			if (tcp_syn_cache[0].scs_use > tcp_syn_use_limit)
1043 				tcp_syn_cache[0].scs_use = tcp_syn_use_limit;
1044 			if (tcp_syn_cache[1].scs_use > tcp_syn_use_limit)
1045 				tcp_syn_cache[1].scs_use = tcp_syn_use_limit;
1046 		}
1047 		return (0);
1048 
1049 	case TCPCTL_SYN_HASH_SIZE:
1050 		nval = tcp_syn_hash_size;
1051 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1052 		if (error)
1053 			return (error);
1054 		if (nval != tcp_syn_hash_size) {
1055 			if (nval < 1 || nval > 100000)
1056 				return (EINVAL);
1057 			/*
1058 			 * If global hash size has been changed, switch sets as
1059 			 * soon as possible.  Then the actual hash array will
1060 			 * be reallocated.
1061 			 */
1062 			if (tcp_syn_cache[0].scs_size != nval)
1063 				tcp_syn_cache[0].scs_use = 0;
1064 			if (tcp_syn_cache[1].scs_size != nval)
1065 				tcp_syn_cache[1].scs_use = 0;
1066 			tcp_syn_hash_size = nval;
1067 		}
1068 		return (0);
1069 
1070 	default:
1071 		if (name[0] < TCPCTL_MAXID)
1072 			return (sysctl_int_arr(tcpctl_vars, name, namelen,
1073 			    oldp, oldlenp, newp, newlen));
1074 		return (ENOPROTOOPT);
1075 	}
1076 	/* NOTREACHED */
1077 }
1078 
1079 /*
1080  * Scale the send buffer so that inflight data is not accounted against
1081  * the limit. The buffer will scale with the congestion window, if the
1082  * the receiver stops acking data the window will shrink and therefor
1083  * the buffer size will shrink as well.
1084  * In low memory situation try to shrink the buffer to the initial size
1085  * disabling the send buffer scaling as long as the situation persists.
1086  */
1087 void
1088 tcp_update_sndspace(struct tcpcb *tp)
1089 {
1090 	struct socket *so = tp->t_inpcb->inp_socket;
1091 	u_long nmax = so->so_snd.sb_hiwat;
1092 
1093 	if (sbchecklowmem()) {
1094 		/* low on memory try to get rid of some */
1095 		if (tcp_sendspace < nmax)
1096 			nmax = tcp_sendspace;
1097 	} else if (so->so_snd.sb_wat != tcp_sendspace)
1098 		/* user requested buffer size, auto-scaling disabled */
1099 		nmax = so->so_snd.sb_wat;
1100 	else
1101 		/* automatic buffer scaling */
1102 		nmax = MIN(sb_max, so->so_snd.sb_wat + tp->snd_max -
1103 		    tp->snd_una);
1104 
1105 	/* a writable socket must be preserved because of poll(2) semantics */
1106 	if (sbspace(so, &so->so_snd) >= so->so_snd.sb_lowat) {
1107 		if (nmax < so->so_snd.sb_cc + so->so_snd.sb_lowat)
1108 			nmax = so->so_snd.sb_cc + so->so_snd.sb_lowat;
1109 		if (nmax * 2 < so->so_snd.sb_mbcnt + so->so_snd.sb_lowat)
1110 			nmax = (so->so_snd.sb_mbcnt+so->so_snd.sb_lowat+1) / 2;
1111 	}
1112 
1113 	/* round to MSS boundary */
1114 	nmax = roundup(nmax, tp->t_maxseg);
1115 
1116 	if (nmax != so->so_snd.sb_hiwat)
1117 		sbreserve(so, &so->so_snd, nmax);
1118 }
1119 
1120 /*
1121  * Scale the recv buffer by looking at how much data was transferred in
1122  * on approximated RTT. If more than a big part of the recv buffer was
1123  * transferred during that time we increase the buffer by a constant.
1124  * In low memory situation try to shrink the buffer to the initial size.
1125  */
1126 void
1127 tcp_update_rcvspace(struct tcpcb *tp)
1128 {
1129 	struct socket *so = tp->t_inpcb->inp_socket;
1130 	u_long nmax = so->so_rcv.sb_hiwat;
1131 
1132 	if (sbchecklowmem()) {
1133 		/* low on memory try to get rid of some */
1134 		if (tcp_recvspace < nmax)
1135 			nmax = tcp_recvspace;
1136 	} else if (so->so_rcv.sb_wat != tcp_recvspace)
1137 		/* user requested buffer size, auto-scaling disabled */
1138 		nmax = so->so_rcv.sb_wat;
1139 	else {
1140 		/* automatic buffer scaling */
1141 		if (tp->rfbuf_cnt > so->so_rcv.sb_hiwat / 8 * 7)
1142 			nmax = MIN(sb_max, so->so_rcv.sb_hiwat +
1143 			    tcp_autorcvbuf_inc);
1144 	}
1145 
1146 	/* a readable socket must be preserved because of poll(2) semantics */
1147 	if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat &&
1148 	    nmax < so->so_snd.sb_lowat)
1149 		nmax = so->so_snd.sb_lowat;
1150 
1151 	if (nmax == so->so_rcv.sb_hiwat)
1152 		return;
1153 
1154 	/* round to MSS boundary */
1155 	nmax = roundup(nmax, tp->t_maxseg);
1156 	sbreserve(so, &so->so_rcv, nmax);
1157 }
1158