xref: /openbsd-src/sys/netinet/tcp_usrreq.c (revision 8dfe214903ce3625c937d5fad2469e8a0d1d4d71)
1 /*	$OpenBSD: tcp_usrreq.c,v 1.227 2023/12/03 20:24:17 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 #include <sys/proc.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/in_pcb.h>
92 #include <netinet/ip_var.h>
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_debug.h>
99 
100 #ifdef INET6
101 #include <netinet6/in6_var.h>
102 #endif
103 
104 #ifndef TCP_SENDSPACE
105 #define	TCP_SENDSPACE	1024*16
106 #endif
107 u_int	tcp_sendspace = TCP_SENDSPACE;
108 #ifndef TCP_RECVSPACE
109 #define	TCP_RECVSPACE	1024*16
110 #endif
111 u_int	tcp_recvspace = TCP_RECVSPACE;
112 u_int	tcp_autorcvbuf_inc = 16 * 1024;
113 
114 const struct pr_usrreqs tcp_usrreqs = {
115 	.pru_attach	= tcp_attach,
116 	.pru_detach	= tcp_detach,
117 	.pru_bind	= tcp_bind,
118 	.pru_listen	= tcp_listen,
119 	.pru_connect	= tcp_connect,
120 	.pru_accept	= tcp_accept,
121 	.pru_disconnect	= tcp_disconnect,
122 	.pru_shutdown	= tcp_shutdown,
123 	.pru_rcvd	= tcp_rcvd,
124 	.pru_send	= tcp_send,
125 	.pru_abort	= tcp_abort,
126 	.pru_sense	= tcp_sense,
127 	.pru_rcvoob	= tcp_rcvoob,
128 	.pru_sendoob	= tcp_sendoob,
129 	.pru_control	= in_control,
130 	.pru_sockaddr	= tcp_sockaddr,
131 	.pru_peeraddr	= tcp_peeraddr,
132 };
133 
134 #ifdef INET6
135 const struct pr_usrreqs tcp6_usrreqs = {
136 	.pru_attach	= tcp_attach,
137 	.pru_detach	= tcp_detach,
138 	.pru_bind	= tcp_bind,
139 	.pru_listen	= tcp_listen,
140 	.pru_connect	= tcp_connect,
141 	.pru_accept	= tcp_accept,
142 	.pru_disconnect	= tcp_disconnect,
143 	.pru_shutdown	= tcp_shutdown,
144 	.pru_rcvd	= tcp_rcvd,
145 	.pru_send	= tcp_send,
146 	.pru_abort	= tcp_abort,
147 	.pru_sense	= tcp_sense,
148 	.pru_rcvoob	= tcp_rcvoob,
149 	.pru_sendoob	= tcp_sendoob,
150 	.pru_control	= in6_control,
151 	.pru_sockaddr	= tcp_sockaddr,
152 	.pru_peeraddr	= tcp_peeraddr,
153 };
154 #endif
155 
156 const struct sysctl_bounded_args tcpctl_vars[] = {
157 	{ TCPCTL_RFC1323, &tcp_do_rfc1323, 0, 1 },
158 	{ TCPCTL_SACK, &tcp_do_sack, 0, 1 },
159 	{ TCPCTL_MSSDFLT, &tcp_mssdflt, TCP_MSS, 65535 },
160 	{ TCPCTL_RSTPPSLIMIT, &tcp_rst_ppslim, 1, 1000 * 1000 },
161 	{ TCPCTL_ACK_ON_PUSH, &tcp_ack_on_push, 0, 1 },
162 #ifdef TCP_ECN
163 	{ TCPCTL_ECN, &tcp_do_ecn, 0, 1 },
164 #endif
165 	{ TCPCTL_SYN_CACHE_LIMIT, &tcp_syn_cache_limit, 1, 1000 * 1000 },
166 	{ TCPCTL_SYN_BUCKET_LIMIT, &tcp_syn_bucket_limit, 1, INT_MAX },
167 	{ TCPCTL_RFC3390, &tcp_do_rfc3390, 0, 2 },
168 	{ TCPCTL_ALWAYS_KEEPALIVE, &tcp_always_keepalive, 0, 1 },
169 	{ TCPCTL_TSO, &tcp_do_tso, 0, 1 },
170 };
171 
172 struct	inpcbtable tcbtable;
173 
174 int	tcp_fill_info(struct tcpcb *, struct socket *, struct mbuf *);
175 int	tcp_ident(void *, size_t *, void *, size_t, int);
176 
177 static inline int tcp_sogetpcb(struct socket *, struct inpcb **,
178                       struct tcpcb **);
179 
180 static inline int
181 tcp_sogetpcb(struct socket *so, struct inpcb **rinp, struct tcpcb **rtp)
182 {
183 	struct inpcb *inp;
184 	struct tcpcb *tp;
185 
186 	/*
187 	 * When a TCP is attached to a socket, then there will be
188 	 * a (struct inpcb) pointed at by the socket, and this
189 	 * structure will point at a subsidiary (struct tcpcb).
190 	 */
191 	if ((inp = sotoinpcb(so)) == NULL || (tp = intotcpcb(inp)) == NULL) {
192 		if (so->so_error)
193 			return so->so_error;
194 		return EINVAL;
195 	}
196 
197 	*rinp = inp;
198 	*rtp = tp;
199 
200 	return 0;
201 }
202 
203 /*
204  * Export internal TCP state information via a struct tcp_info without
205  * leaking any sensitive information. Sequence numbers are reported
206  * relative to the initial sequence number.
207  */
208 int
209 tcp_fill_info(struct tcpcb *tp, struct socket *so, struct mbuf *m)
210 {
211 	struct proc *p = curproc;
212 	struct tcp_info *ti;
213 	u_int t = 1000;		/* msec => usec */
214 	uint64_t now;
215 
216 	if (sizeof(*ti) > MLEN) {
217 		MCLGETL(m, M_WAITOK, sizeof(*ti));
218 		if (!ISSET(m->m_flags, M_EXT))
219 			return ENOMEM;
220 	}
221 	ti = mtod(m, struct tcp_info *);
222 	m->m_len = sizeof(*ti);
223 	memset(ti, 0, sizeof(*ti));
224 	now = tcp_now();
225 
226 	ti->tcpi_state = tp->t_state;
227 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
228 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
229 	if (tp->t_flags & TF_SACK_PERMIT)
230 		ti->tcpi_options |= TCPI_OPT_SACK;
231 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
232 		ti->tcpi_options |= TCPI_OPT_WSCALE;
233 		ti->tcpi_snd_wscale = tp->snd_scale;
234 		ti->tcpi_rcv_wscale = tp->rcv_scale;
235 	}
236 #ifdef TCP_ECN
237 	if (tp->t_flags & TF_ECN_PERMIT)
238 		ti->tcpi_options |= TCPI_OPT_ECN;
239 #endif
240 
241 	ti->tcpi_rto = tp->t_rxtcur * t;
242 	ti->tcpi_snd_mss = tp->t_maxseg;
243 	ti->tcpi_rcv_mss = tp->t_peermss;
244 
245 	ti->tcpi_last_data_sent = (now - tp->t_sndtime) * t;
246 	ti->tcpi_last_ack_sent = (now - tp->t_sndacktime) * t;
247 	ti->tcpi_last_data_recv = (now - tp->t_rcvtime) * t;
248 	ti->tcpi_last_ack_recv = (now - tp->t_rcvacktime) * t;
249 
250 	ti->tcpi_rtt = ((uint64_t)tp->t_srtt * t) >>
251 	    (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
252 	ti->tcpi_rttvar = ((uint64_t)tp->t_rttvar * t) >>
253 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT);
254 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
255 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
256 
257 	ti->tcpi_rcv_space = tp->rcv_wnd;
258 
259 	/*
260 	 * Provide only minimal information for unprivileged processes.
261 	 */
262 	if (suser(p) != 0)
263 		return 0;
264 
265 	/* FreeBSD-specific extension fields for tcp_info.  */
266 	ti->tcpi_snd_wnd = tp->snd_wnd;
267 	ti->tcpi_snd_nxt = tp->snd_nxt - tp->iss;
268 	ti->tcpi_rcv_nxt = tp->rcv_nxt - tp->irs;
269 	/* missing tcpi_toe_tid */
270 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
271 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
272 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
273 
274 	/* OpenBSD extensions */
275 	ti->tcpi_rttmin = tp->t_rttmin * t;
276 	ti->tcpi_max_sndwnd = tp->max_sndwnd;
277 	ti->tcpi_rcv_adv = tp->rcv_adv - tp->irs;
278 	ti->tcpi_rcv_up = tp->rcv_up - tp->irs;
279 	ti->tcpi_snd_una = tp->snd_una - tp->iss;
280 	ti->tcpi_snd_up = tp->snd_up - tp->iss;
281 	ti->tcpi_snd_wl1 = tp->snd_wl1 - tp->iss;
282 	ti->tcpi_snd_wl2 = tp->snd_wl2 - tp->iss;
283 	ti->tcpi_snd_max = tp->snd_max - tp->iss;
284 
285 	ti->tcpi_ts_recent = tp->ts_recent; /* XXX value from the wire */
286 	ti->tcpi_ts_recent_age = (now - tp->ts_recent_age) * t;
287 	ti->tcpi_rfbuf_cnt = tp->rfbuf_cnt;
288 	ti->tcpi_rfbuf_ts = (now - tp->rfbuf_ts) * t;
289 
290 	ti->tcpi_so_rcv_sb_cc = so->so_rcv.sb_cc;
291 	ti->tcpi_so_rcv_sb_hiwat = so->so_rcv.sb_hiwat;
292 	ti->tcpi_so_rcv_sb_lowat = so->so_rcv.sb_lowat;
293 	ti->tcpi_so_rcv_sb_wat = so->so_rcv.sb_wat;
294 	ti->tcpi_so_snd_sb_cc = so->so_snd.sb_cc;
295 	ti->tcpi_so_snd_sb_hiwat = so->so_snd.sb_hiwat;
296 	ti->tcpi_so_snd_sb_lowat = so->so_snd.sb_lowat;
297 	ti->tcpi_so_snd_sb_wat = so->so_snd.sb_wat;
298 
299 	return 0;
300 }
301 
302 int
303 tcp_ctloutput(int op, struct socket *so, int level, int optname,
304     struct mbuf *m)
305 {
306 	int error = 0;
307 	struct inpcb *inp;
308 	struct tcpcb *tp;
309 	int i;
310 
311 	inp = sotoinpcb(so);
312 	if (inp == NULL)
313 		return (ECONNRESET);
314 	if (level != IPPROTO_TCP) {
315 #ifdef INET6
316 		if (ISSET(inp->inp_flags, INP_IPV6))
317 			error = ip6_ctloutput(op, so, level, optname, m);
318 		else
319 #endif /* INET6 */
320 			error = ip_ctloutput(op, so, level, optname, m);
321 		return (error);
322 	}
323 	tp = intotcpcb(inp);
324 
325 	switch (op) {
326 
327 	case PRCO_SETOPT:
328 		switch (optname) {
329 
330 		case TCP_NODELAY:
331 			if (m == NULL || m->m_len < sizeof (int))
332 				error = EINVAL;
333 			else if (*mtod(m, int *))
334 				tp->t_flags |= TF_NODELAY;
335 			else
336 				tp->t_flags &= ~TF_NODELAY;
337 			break;
338 
339 		case TCP_NOPUSH:
340 			if (m == NULL || m->m_len < sizeof (int))
341 				error = EINVAL;
342 			else if (*mtod(m, int *))
343 				tp->t_flags |= TF_NOPUSH;
344 			else if (tp->t_flags & TF_NOPUSH) {
345 				tp->t_flags &= ~TF_NOPUSH;
346 				if (TCPS_HAVEESTABLISHED(tp->t_state))
347 					error = tcp_output(tp);
348 			}
349 			break;
350 
351 		case TCP_MAXSEG:
352 			if (m == NULL || m->m_len < sizeof (int)) {
353 				error = EINVAL;
354 				break;
355 			}
356 
357 			i = *mtod(m, int *);
358 			if (i > 0 && i <= tp->t_maxseg)
359 				tp->t_maxseg = i;
360 			else
361 				error = EINVAL;
362 			break;
363 
364 		case TCP_SACK_ENABLE:
365 			if (m == NULL || m->m_len < sizeof (int)) {
366 				error = EINVAL;
367 				break;
368 			}
369 
370 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
371 				error = EPERM;
372 				break;
373 			}
374 
375 			if (tp->t_flags & TF_SIGNATURE) {
376 				error = EPERM;
377 				break;
378 			}
379 
380 			if (*mtod(m, int *))
381 				tp->sack_enable = 1;
382 			else
383 				tp->sack_enable = 0;
384 			break;
385 #ifdef TCP_SIGNATURE
386 		case TCP_MD5SIG:
387 			if (m == NULL || m->m_len < sizeof (int)) {
388 				error = EINVAL;
389 				break;
390 			}
391 
392 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
393 				error = EPERM;
394 				break;
395 			}
396 
397 			if (*mtod(m, int *)) {
398 				tp->t_flags |= TF_SIGNATURE;
399 				tp->sack_enable = 0;
400 			} else
401 				tp->t_flags &= ~TF_SIGNATURE;
402 			break;
403 #endif /* TCP_SIGNATURE */
404 		default:
405 			error = ENOPROTOOPT;
406 			break;
407 		}
408 		break;
409 
410 	case PRCO_GETOPT:
411 		switch (optname) {
412 		case TCP_NODELAY:
413 			m->m_len = sizeof(int);
414 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
415 			break;
416 		case TCP_NOPUSH:
417 			m->m_len = sizeof(int);
418 			*mtod(m, int *) = tp->t_flags & TF_NOPUSH;
419 			break;
420 		case TCP_MAXSEG:
421 			m->m_len = sizeof(int);
422 			*mtod(m, int *) = tp->t_maxseg;
423 			break;
424 		case TCP_SACK_ENABLE:
425 			m->m_len = sizeof(int);
426 			*mtod(m, int *) = tp->sack_enable;
427 			break;
428 		case TCP_INFO:
429 			error = tcp_fill_info(tp, so, m);
430 			break;
431 #ifdef TCP_SIGNATURE
432 		case TCP_MD5SIG:
433 			m->m_len = sizeof(int);
434 			*mtod(m, int *) = tp->t_flags & TF_SIGNATURE;
435 			break;
436 #endif
437 		default:
438 			error = ENOPROTOOPT;
439 			break;
440 		}
441 		break;
442 	}
443 	return (error);
444 }
445 
446 /*
447  * Attach TCP protocol to socket, allocating
448  * internet protocol control block, tcp control block,
449  * buffer space, and entering LISTEN state to accept connections.
450  */
451 int
452 tcp_attach(struct socket *so, int proto, int wait)
453 {
454 	struct tcpcb *tp;
455 	struct inpcb *inp;
456 	int error;
457 
458 	if (so->so_pcb)
459 		return EISCONN;
460 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0 ||
461 	    sbcheckreserve(so->so_snd.sb_wat, tcp_sendspace) ||
462 	    sbcheckreserve(so->so_rcv.sb_wat, tcp_recvspace)) {
463 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
464 		if (error)
465 			return (error);
466 	}
467 
468 	NET_ASSERT_LOCKED();
469 	error = in_pcballoc(so, &tcbtable, wait);
470 	if (error)
471 		return (error);
472 	inp = sotoinpcb(so);
473 	tp = tcp_newtcpcb(inp, wait);
474 	if (tp == NULL) {
475 		unsigned int nofd = so->so_state & SS_NOFDREF;	/* XXX */
476 
477 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
478 		in_pcbdetach(inp);
479 		so->so_state |= nofd;
480 		return (ENOBUFS);
481 	}
482 	tp->t_state = TCPS_CLOSED;
483 #ifdef INET6
484 	/* we disallow IPv4 mapped address completely. */
485 	if (inp->inp_flags & INP_IPV6)
486 		tp->pf = PF_INET6;
487 	else
488 		tp->pf = PF_INET;
489 #else
490 	tp->pf = PF_INET;
491 #endif
492 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
493 		so->so_linger = TCP_LINGERTIME;
494 
495 	if (so->so_options & SO_DEBUG)
496 		tcp_trace(TA_USER, TCPS_CLOSED, tp, tp, NULL, PRU_ATTACH, 0);
497 	return (0);
498 }
499 
500 int
501 tcp_detach(struct socket *so)
502 {
503 	struct inpcb *inp;
504 	struct tcpcb *otp = NULL, *tp;
505 	int error = 0;
506 	short ostate;
507 
508 	soassertlocked(so);
509 
510 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
511 		return (error);
512 
513 	if (so->so_options & SO_DEBUG) {
514 		otp = tp;
515 		ostate = tp->t_state;
516 	}
517 
518 	/*
519 	 * Detach the TCP protocol from the socket.
520 	 * If the protocol state is non-embryonic, then can't
521 	 * do this directly: have to initiate a PRU_DISCONNECT,
522 	 * which may finish later; embryonic TCB's can just
523 	 * be discarded here.
524 	 */
525 	tp = tcp_dodisconnect(tp);
526 
527 	if (otp)
528 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_DETACH, 0);
529 	return (error);
530 }
531 
532 /*
533  * Give the socket an address.
534  */
535 int
536 tcp_bind(struct socket *so, struct mbuf *nam, struct proc *p)
537 {
538 	struct inpcb *inp;
539 	struct tcpcb *tp;
540 	int error;
541 	short ostate;
542 
543 	soassertlocked(so);
544 
545 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
546 		return (error);
547 
548 	if (so->so_options & SO_DEBUG)
549 		ostate = tp->t_state;
550 
551 	error = in_pcbbind(inp, nam, p);
552 
553 	if (so->so_options & SO_DEBUG)
554 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_BIND, 0);
555 	return (error);
556 }
557 
558 /*
559  * Prepare to accept connections.
560  */
561 int
562 tcp_listen(struct socket *so)
563 {
564 	struct inpcb *inp;
565 	struct tcpcb *tp, *otp = NULL;
566 	int error;
567 	short ostate;
568 
569 	soassertlocked(so);
570 
571 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
572 		return (error);
573 
574 	if (so->so_options & SO_DEBUG) {
575 		otp = tp;
576 		ostate = tp->t_state;
577 	}
578 
579 	if (inp->inp_lport == 0)
580 		if ((error = in_pcbbind(inp, NULL, curproc)))
581 			goto out;
582 
583 	/*
584 	 * If the in_pcbbind() above is called, the tp->pf
585 	 * should still be whatever it was before.
586 	 */
587 	tp->t_state = TCPS_LISTEN;
588 
589 out:
590 	if (otp)
591 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_LISTEN, 0);
592 	return (error);
593 }
594 
595 /*
596  * Initiate connection to peer.
597  * Create a template for use in transmissions on this connection.
598  * Enter SYN_SENT state, and mark socket as connecting.
599  * Start keep-alive timer, and seed output sequence space.
600  * Send initial segment on connection.
601  */
602 int
603 tcp_connect(struct socket *so, struct mbuf *nam)
604 {
605 	struct inpcb *inp;
606 	struct tcpcb *tp, *otp = NULL;
607 	int error;
608 	short ostate;
609 
610 	soassertlocked(so);
611 
612 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
613 		return (error);
614 
615 	if (so->so_options & SO_DEBUG) {
616 		otp = tp;
617 		ostate = tp->t_state;
618 	}
619 
620 #ifdef INET6
621 	if (inp->inp_flags & INP_IPV6) {
622 		struct sockaddr_in6 *sin6;
623 
624 		if ((error = in6_nam2sin6(nam, &sin6)))
625 			goto out;
626 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
627 		    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
628 			error = EINVAL;
629 			goto out;
630 		}
631 	} else
632 #endif /* INET6 */
633 	{
634 		struct sockaddr_in *sin;
635 
636 		if ((error = in_nam2sin(nam, &sin)))
637 			goto out;
638 		if ((sin->sin_addr.s_addr == INADDR_ANY) ||
639 		    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
640 		    IN_MULTICAST(sin->sin_addr.s_addr) ||
641 		    in_broadcast(sin->sin_addr, inp->inp_rtableid)) {
642 			error = EINVAL;
643 			goto out;
644 		}
645 	}
646 	error = in_pcbconnect(inp, nam);
647 	if (error)
648 		goto out;
649 
650 	tp->t_template = tcp_template(tp);
651 	if (tp->t_template == 0) {
652 		in_pcbunset_faddr(inp);
653 		in_pcbdisconnect(inp);
654 		error = ENOBUFS;
655 		goto out;
656 	}
657 
658 	so->so_state |= SS_CONNECTOUT;
659 
660 	/* Compute window scaling to request.  */
661 	tcp_rscale(tp, sb_max);
662 
663 	soisconnecting(so);
664 	tcpstat_inc(tcps_connattempt);
665 	tp->t_state = TCPS_SYN_SENT;
666 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
667 	tcp_set_iss_tsm(tp);
668 	tcp_sendseqinit(tp);
669 	tp->snd_last = tp->snd_una;
670 	error = tcp_output(tp);
671 
672 out:
673 	if (otp)
674 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_CONNECT, 0);
675 	return (error);
676 }
677 
678 /*
679  * Accept a connection.  Essentially all the work is done at higher
680  * levels; just return the address of the peer, storing through addr.
681  */
682 int
683 tcp_accept(struct socket *so, struct mbuf *nam)
684 {
685 	struct inpcb *inp;
686 	struct tcpcb *tp;
687 	int error;
688 	short ostate;
689 
690 	soassertlocked(so);
691 
692 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
693 		return (error);
694 
695 	if (so->so_options & SO_DEBUG)
696 		ostate = tp->t_state;
697 
698 #ifdef INET6
699 	if (inp->inp_flags & INP_IPV6)
700 		in6_setpeeraddr(inp, nam);
701 	else
702 #endif
703 		in_setpeeraddr(inp, nam);
704 
705 	if (so->so_options & SO_DEBUG)
706 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_ACCEPT, 0);
707 	return (error);
708 }
709 
710 /*
711  * Initiate disconnect from peer.
712  * If connection never passed embryonic stage, just drop;
713  * else if don't need to let data drain, then can just drop anyways,
714  * else have to begin TCP shutdown process: mark socket disconnecting,
715  * drain unread data, state switch to reflect user close, and
716  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
717  * when peer sends FIN and acks ours.
718  *
719  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
720  */
721 int
722 tcp_disconnect(struct socket *so)
723 {
724 	struct inpcb *inp;
725 	struct tcpcb *tp, *otp = NULL;
726 	int error;
727 	short ostate;
728 
729 	soassertlocked(so);
730 
731 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
732 		return (error);
733 
734 	if (so->so_options & SO_DEBUG) {
735 		otp = tp;
736 		ostate = tp->t_state;
737 	}
738 
739 	tp = tcp_dodisconnect(tp);
740 
741 	if (otp)
742 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_DISCONNECT, 0);
743 	return (0);
744 }
745 
746 /*
747  * Mark the connection as being incapable of further output.
748  */
749 int
750 tcp_shutdown(struct socket *so)
751 {
752 	struct inpcb *inp;
753 	struct tcpcb *tp, *otp = NULL;
754 	int error;
755 	short ostate;
756 
757 	soassertlocked(so);
758 
759 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
760 		return (error);
761 
762 	if (so->so_options & SO_DEBUG) {
763 		otp = tp;
764 		ostate = tp->t_state;
765 	}
766 
767 	if (so->so_snd.sb_state & SS_CANTSENDMORE)
768 		goto out;
769 
770 	socantsendmore(so);
771 	tp = tcp_usrclosed(tp);
772 	if (tp)
773 		error = tcp_output(tp);
774 
775 out:
776 	if (otp)
777 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_SHUTDOWN, 0);
778 	return (error);
779 }
780 
781 /*
782  * After a receive, possibly send window update to peer.
783  */
784 void
785 tcp_rcvd(struct socket *so)
786 {
787 	struct inpcb *inp;
788 	struct tcpcb *tp;
789 	short ostate;
790 
791 	soassertlocked(so);
792 
793 	if (tcp_sogetpcb(so, &inp, &tp))
794 		return;
795 
796 	if (so->so_options & SO_DEBUG)
797 		ostate = tp->t_state;
798 
799 	/*
800 	 * soreceive() calls this function when a user receives
801 	 * ancillary data on a listening socket. We don't call
802 	 * tcp_output in such a case, since there is no header
803 	 * template for a listening socket and hence the kernel
804 	 * will panic.
805 	 */
806 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) != 0)
807 		(void) tcp_output(tp);
808 
809 	if (so->so_options & SO_DEBUG)
810 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_RCVD, 0);
811 }
812 
813 /*
814  * Do a send by putting data in output queue and updating urgent
815  * marker if URG set.  Possibly send more data.
816  */
817 int
818 tcp_send(struct socket *so, struct mbuf *m, struct mbuf *nam,
819     struct mbuf *control)
820 {
821 	struct inpcb *inp;
822 	struct tcpcb *tp;
823 	int error;
824 	short ostate;
825 
826 	soassertlocked(so);
827 
828 	if (control && control->m_len) {
829 		error = EINVAL;
830 		goto out;
831 	}
832 
833 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
834 		goto out;
835 
836 	if (so->so_options & SO_DEBUG)
837 		ostate = tp->t_state;
838 
839 	sbappendstream(so, &so->so_snd, m);
840 	m = NULL;
841 
842 	error = tcp_output(tp);
843 
844 	if (so->so_options & SO_DEBUG)
845 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_SEND, 0);
846 
847 out:
848 	m_freem(control);
849 	m_freem(m);
850 
851 	return (error);
852 }
853 
854 /*
855  * Abort the TCP.
856  */
857 void
858 tcp_abort(struct socket *so)
859 {
860 	struct inpcb *inp;
861 	struct tcpcb *tp, *otp = NULL;
862 	short ostate;
863 
864 	soassertlocked(so);
865 
866 	if (tcp_sogetpcb(so, &inp, &tp))
867 		return;
868 
869 	if (so->so_options & SO_DEBUG) {
870 		otp = tp;
871 		ostate = tp->t_state;
872 	}
873 
874 	tp = tcp_drop(tp, ECONNABORTED);
875 
876 	if (otp)
877 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_ABORT, 0);
878 }
879 
880 int
881 tcp_sense(struct socket *so, struct stat *ub)
882 {
883 	struct inpcb *inp;
884 	struct tcpcb *tp;
885 	int error;
886 
887 	soassertlocked(so);
888 
889 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
890 		return (error);
891 
892 	ub->st_blksize = so->so_snd.sb_hiwat;
893 
894 	if (so->so_options & SO_DEBUG)
895 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_SENSE, 0);
896 	return (0);
897 }
898 
899 int
900 tcp_rcvoob(struct socket *so, struct mbuf *m, int flags)
901 {
902 	struct inpcb *inp;
903 	struct tcpcb *tp;
904 	int error;
905 
906 	soassertlocked(so);
907 
908 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
909 		return (error);
910 
911 	if ((so->so_oobmark == 0 &&
912 	    (so->so_rcv.sb_state & SS_RCVATMARK) == 0) ||
913 	    so->so_options & SO_OOBINLINE ||
914 	    tp->t_oobflags & TCPOOB_HADDATA) {
915 		error = EINVAL;
916 		goto out;
917 	}
918 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
919 		error = EWOULDBLOCK;
920 		goto out;
921 	}
922 	m->m_len = 1;
923 	*mtod(m, caddr_t) = tp->t_iobc;
924 	if ((flags & MSG_PEEK) == 0)
925 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
926 out:
927 	if (so->so_options & SO_DEBUG)
928 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_RCVOOB, 0);
929 	return (error);
930 }
931 
932 int
933 tcp_sendoob(struct socket *so, struct mbuf *m, struct mbuf *nam,
934     struct mbuf *control)
935 {
936 	struct inpcb *inp;
937 	struct tcpcb *tp;
938 	int error;
939 	short ostate;
940 
941 	soassertlocked(so);
942 
943 	if (control && control->m_len) {
944 		error = EINVAL;
945 		goto release;
946 	}
947 
948 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
949 		goto release;
950 
951 	if (so->so_options & SO_DEBUG)
952 		ostate = tp->t_state;
953 
954 	if (sbspace(so, &so->so_snd) < -512) {
955 		error = ENOBUFS;
956 		goto out;
957 	}
958 
959 	/*
960 	 * According to RFC961 (Assigned Protocols),
961 	 * the urgent pointer points to the last octet
962 	 * of urgent data.  We continue, however,
963 	 * to consider it to indicate the first octet
964 	 * of data past the urgent section.
965 	 * Otherwise, snd_up should be one lower.
966 	 */
967 	sbappendstream(so, &so->so_snd, m);
968 	m = NULL;
969 	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
970 	tp->t_force = 1;
971 	error = tcp_output(tp);
972 	tp->t_force = 0;
973 
974 out:
975 	if (so->so_options & SO_DEBUG)
976 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_SENDOOB, 0);
977 
978 release:
979 	m_freem(control);
980 	m_freem(m);
981 
982 	return (error);
983 }
984 
985 int
986 tcp_sockaddr(struct socket *so, struct mbuf *nam)
987 {
988 	struct inpcb *inp;
989 	struct tcpcb *tp;
990 	int error;
991 
992 	soassertlocked(so);
993 
994 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
995 		return (error);
996 
997 #ifdef INET6
998 	if (inp->inp_flags & INP_IPV6)
999 		in6_setsockaddr(inp, nam);
1000 	else
1001 #endif
1002 		in_setsockaddr(inp, nam);
1003 
1004 	if (so->so_options & SO_DEBUG)
1005 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL,
1006 		    PRU_SOCKADDR, 0);
1007 	return (0);
1008 }
1009 
1010 int
1011 tcp_peeraddr(struct socket *so, struct mbuf *nam)
1012 {
1013 	struct inpcb *inp;
1014 	struct tcpcb *tp;
1015 	int error;
1016 
1017 	soassertlocked(so);
1018 
1019 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
1020 		return (error);
1021 
1022 #ifdef INET6
1023 	if (inp->inp_flags & INP_IPV6)
1024 		in6_setpeeraddr(inp, nam);
1025 	else
1026 #endif
1027 		in_setpeeraddr(inp, nam);
1028 
1029 	if (so->so_options & SO_DEBUG)
1030 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL,
1031 		    PRU_PEERADDR, 0);
1032 	return (0);
1033 }
1034 
1035 /*
1036  * Initiate (or continue) disconnect.
1037  * If embryonic state, just send reset (once).
1038  * If in ``let data drain'' option and linger null, just drop.
1039  * Otherwise (hard), mark socket disconnecting and drop
1040  * current input data; switch states based on user close, and
1041  * send segment to peer (with FIN).
1042  */
1043 struct tcpcb *
1044 tcp_dodisconnect(struct tcpcb *tp)
1045 {
1046 	struct socket *so = tp->t_inpcb->inp_socket;
1047 
1048 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
1049 		tp = tcp_close(tp);
1050 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1051 		tp = tcp_drop(tp, 0);
1052 	else {
1053 		soisdisconnecting(so);
1054 		sbflush(so, &so->so_rcv);
1055 		tp = tcp_usrclosed(tp);
1056 		if (tp)
1057 			(void) tcp_output(tp);
1058 	}
1059 	return (tp);
1060 }
1061 
1062 /*
1063  * User issued close, and wish to trail through shutdown states:
1064  * if never received SYN, just forget it.  If got a SYN from peer,
1065  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1066  * If already got a FIN from peer, then almost done; go to LAST_ACK
1067  * state.  In all other cases, have already sent FIN to peer (e.g.
1068  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1069  * for peer to send FIN or not respond to keep-alives, etc.
1070  * We can let the user exit from the close as soon as the FIN is acked.
1071  */
1072 struct tcpcb *
1073 tcp_usrclosed(struct tcpcb *tp)
1074 {
1075 
1076 	switch (tp->t_state) {
1077 
1078 	case TCPS_CLOSED:
1079 	case TCPS_LISTEN:
1080 	case TCPS_SYN_SENT:
1081 		tp->t_state = TCPS_CLOSED;
1082 		tp = tcp_close(tp);
1083 		break;
1084 
1085 	case TCPS_SYN_RECEIVED:
1086 	case TCPS_ESTABLISHED:
1087 		tp->t_state = TCPS_FIN_WAIT_1;
1088 		break;
1089 
1090 	case TCPS_CLOSE_WAIT:
1091 		tp->t_state = TCPS_LAST_ACK;
1092 		break;
1093 	}
1094 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1095 		soisdisconnected(tp->t_inpcb->inp_socket);
1096 		/*
1097 		 * If we are in FIN_WAIT_2, we arrived here because the
1098 		 * application did a shutdown of the send side.  Like the
1099 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
1100 		 * a full close, we start a timer to make sure sockets are
1101 		 * not left in FIN_WAIT_2 forever.
1102 		 */
1103 		if (tp->t_state == TCPS_FIN_WAIT_2)
1104 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1105 	}
1106 	return (tp);
1107 }
1108 
1109 /*
1110  * Look up a socket for ident or tcpdrop, ...
1111  */
1112 int
1113 tcp_ident(void *oldp, size_t *oldlenp, void *newp, size_t newlen, int dodrop)
1114 {
1115 	int error = 0;
1116 	struct tcp_ident_mapping tir;
1117 	struct inpcb *inp;
1118 	struct tcpcb *tp = NULL;
1119 	struct sockaddr_in *fin, *lin;
1120 #ifdef INET6
1121 	struct sockaddr_in6 *fin6, *lin6;
1122 	struct in6_addr f6, l6;
1123 #endif
1124 
1125 	NET_ASSERT_LOCKED();
1126 
1127 	if (dodrop) {
1128 		if (oldp != NULL || *oldlenp != 0)
1129 			return (EINVAL);
1130 		if (newp == NULL)
1131 			return (EPERM);
1132 		if (newlen < sizeof(tir))
1133 			return (ENOMEM);
1134 		if ((error = copyin(newp, &tir, sizeof (tir))) != 0 )
1135 			return (error);
1136 	} else {
1137 		if (oldp == NULL)
1138 			return (EINVAL);
1139 		if (*oldlenp < sizeof(tir))
1140 			return (ENOMEM);
1141 		if (newp != NULL || newlen != 0)
1142 			return (EINVAL);
1143 		if ((error = copyin(oldp, &tir, sizeof (tir))) != 0 )
1144 			return (error);
1145 	}
1146 	switch (tir.faddr.ss_family) {
1147 #ifdef INET6
1148 	case AF_INET6:
1149 		fin6 = (struct sockaddr_in6 *)&tir.faddr;
1150 		error = in6_embedscope(&f6, fin6, NULL, NULL);
1151 		if (error)
1152 			return EINVAL;	/*?*/
1153 		lin6 = (struct sockaddr_in6 *)&tir.laddr;
1154 		error = in6_embedscope(&l6, lin6, NULL, NULL);
1155 		if (error)
1156 			return EINVAL;	/*?*/
1157 		break;
1158 #endif
1159 	case AF_INET:
1160 		fin = (struct sockaddr_in *)&tir.faddr;
1161 		lin = (struct sockaddr_in *)&tir.laddr;
1162 		break;
1163 	default:
1164 		return (EINVAL);
1165 	}
1166 
1167 	switch (tir.faddr.ss_family) {
1168 #ifdef INET6
1169 	case AF_INET6:
1170 		inp = in6_pcblookup(&tcbtable, &f6,
1171 		    fin6->sin6_port, &l6, lin6->sin6_port, tir.rdomain);
1172 		break;
1173 #endif
1174 	case AF_INET:
1175 		inp = in_pcblookup(&tcbtable, fin->sin_addr,
1176 		    fin->sin_port, lin->sin_addr, lin->sin_port, tir.rdomain);
1177 		break;
1178 	default:
1179 		unhandled_af(tir.faddr.ss_family);
1180 	}
1181 
1182 	if (dodrop) {
1183 		if (inp && (tp = intotcpcb(inp)) &&
1184 		    ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0))
1185 			tp = tcp_drop(tp, ECONNABORTED);
1186 		else
1187 			error = ESRCH;
1188 		in_pcbunref(inp);
1189 		return (error);
1190 	}
1191 
1192 	if (inp == NULL) {
1193 		tcpstat_inc(tcps_pcbhashmiss);
1194 		switch (tir.faddr.ss_family) {
1195 #ifdef INET6
1196 		case AF_INET6:
1197 			inp = in6_pcblookup_listen(&tcbtable,
1198 			    &l6, lin6->sin6_port, NULL, tir.rdomain);
1199 			break;
1200 #endif
1201 		case AF_INET:
1202 			inp = in_pcblookup_listen(&tcbtable,
1203 			    lin->sin_addr, lin->sin_port, NULL, tir.rdomain);
1204 			break;
1205 		}
1206 	}
1207 
1208 	if (inp != NULL && (inp->inp_socket->so_state & SS_CONNECTOUT)) {
1209 		tir.ruid = inp->inp_socket->so_ruid;
1210 		tir.euid = inp->inp_socket->so_euid;
1211 	} else {
1212 		tir.ruid = -1;
1213 		tir.euid = -1;
1214 	}
1215 
1216 	*oldlenp = sizeof (tir);
1217 	error = copyout((void *)&tir, oldp, sizeof (tir));
1218 	in_pcbunref(inp);
1219 	return (error);
1220 }
1221 
1222 int
1223 tcp_sysctl_tcpstat(void *oldp, size_t *oldlenp, void *newp)
1224 {
1225 	uint64_t counters[tcps_ncounters];
1226 	struct tcpstat tcpstat;
1227 	struct syn_cache_set *set;
1228 	int i = 0;
1229 
1230 #define ASSIGN(field)	do { tcpstat.field = counters[i++]; } while (0)
1231 
1232 	memset(&tcpstat, 0, sizeof tcpstat);
1233 	counters_read(tcpcounters, counters, nitems(counters), NULL);
1234 	ASSIGN(tcps_connattempt);
1235 	ASSIGN(tcps_accepts);
1236 	ASSIGN(tcps_connects);
1237 	ASSIGN(tcps_drops);
1238 	ASSIGN(tcps_conndrops);
1239 	ASSIGN(tcps_closed);
1240 	ASSIGN(tcps_segstimed);
1241 	ASSIGN(tcps_rttupdated);
1242 	ASSIGN(tcps_delack);
1243 	ASSIGN(tcps_timeoutdrop);
1244 	ASSIGN(tcps_rexmttimeo);
1245 	ASSIGN(tcps_persisttimeo);
1246 	ASSIGN(tcps_persistdrop);
1247 	ASSIGN(tcps_keeptimeo);
1248 	ASSIGN(tcps_keepprobe);
1249 	ASSIGN(tcps_keepdrops);
1250 	ASSIGN(tcps_sndtotal);
1251 	ASSIGN(tcps_sndpack);
1252 	ASSIGN(tcps_sndbyte);
1253 	ASSIGN(tcps_sndrexmitpack);
1254 	ASSIGN(tcps_sndrexmitbyte);
1255 	ASSIGN(tcps_sndrexmitfast);
1256 	ASSIGN(tcps_sndacks);
1257 	ASSIGN(tcps_sndprobe);
1258 	ASSIGN(tcps_sndurg);
1259 	ASSIGN(tcps_sndwinup);
1260 	ASSIGN(tcps_sndctrl);
1261 	ASSIGN(tcps_rcvtotal);
1262 	ASSIGN(tcps_rcvpack);
1263 	ASSIGN(tcps_rcvbyte);
1264 	ASSIGN(tcps_rcvbadsum);
1265 	ASSIGN(tcps_rcvbadoff);
1266 	ASSIGN(tcps_rcvmemdrop);
1267 	ASSIGN(tcps_rcvnosec);
1268 	ASSIGN(tcps_rcvshort);
1269 	ASSIGN(tcps_rcvduppack);
1270 	ASSIGN(tcps_rcvdupbyte);
1271 	ASSIGN(tcps_rcvpartduppack);
1272 	ASSIGN(tcps_rcvpartdupbyte);
1273 	ASSIGN(tcps_rcvoopack);
1274 	ASSIGN(tcps_rcvoobyte);
1275 	ASSIGN(tcps_rcvpackafterwin);
1276 	ASSIGN(tcps_rcvbyteafterwin);
1277 	ASSIGN(tcps_rcvafterclose);
1278 	ASSIGN(tcps_rcvwinprobe);
1279 	ASSIGN(tcps_rcvdupack);
1280 	ASSIGN(tcps_rcvacktoomuch);
1281 	ASSIGN(tcps_rcvacktooold);
1282 	ASSIGN(tcps_rcvackpack);
1283 	ASSIGN(tcps_rcvackbyte);
1284 	ASSIGN(tcps_rcvwinupd);
1285 	ASSIGN(tcps_pawsdrop);
1286 	ASSIGN(tcps_predack);
1287 	ASSIGN(tcps_preddat);
1288 	ASSIGN(tcps_pcbhashmiss);
1289 	ASSIGN(tcps_noport);
1290 	ASSIGN(tcps_badsyn);
1291 	ASSIGN(tcps_dropsyn);
1292 	ASSIGN(tcps_rcvbadsig);
1293 	ASSIGN(tcps_rcvgoodsig);
1294 	ASSIGN(tcps_inswcsum);
1295 	ASSIGN(tcps_outswcsum);
1296 	ASSIGN(tcps_ecn_accepts);
1297 	ASSIGN(tcps_ecn_rcvece);
1298 	ASSIGN(tcps_ecn_rcvcwr);
1299 	ASSIGN(tcps_ecn_rcvce);
1300 	ASSIGN(tcps_ecn_sndect);
1301 	ASSIGN(tcps_ecn_sndece);
1302 	ASSIGN(tcps_ecn_sndcwr);
1303 	ASSIGN(tcps_cwr_ecn);
1304 	ASSIGN(tcps_cwr_frecovery);
1305 	ASSIGN(tcps_cwr_timeout);
1306 	ASSIGN(tcps_sc_added);
1307 	ASSIGN(tcps_sc_completed);
1308 	ASSIGN(tcps_sc_timed_out);
1309 	ASSIGN(tcps_sc_overflowed);
1310 	ASSIGN(tcps_sc_reset);
1311 	ASSIGN(tcps_sc_unreach);
1312 	ASSIGN(tcps_sc_bucketoverflow);
1313 	ASSIGN(tcps_sc_aborted);
1314 	ASSIGN(tcps_sc_dupesyn);
1315 	ASSIGN(tcps_sc_dropped);
1316 	ASSIGN(tcps_sc_collisions);
1317 	ASSIGN(tcps_sc_retransmitted);
1318 	ASSIGN(tcps_sc_seedrandom);
1319 	ASSIGN(tcps_sc_hash_size);
1320 	ASSIGN(tcps_sc_entry_count);
1321 	ASSIGN(tcps_sc_entry_limit);
1322 	ASSIGN(tcps_sc_bucket_maxlen);
1323 	ASSIGN(tcps_sc_bucket_limit);
1324 	ASSIGN(tcps_sc_uses_left);
1325 	ASSIGN(tcps_conndrained);
1326 	ASSIGN(tcps_sack_recovery_episode);
1327 	ASSIGN(tcps_sack_rexmits);
1328 	ASSIGN(tcps_sack_rexmit_bytes);
1329 	ASSIGN(tcps_sack_rcv_opts);
1330 	ASSIGN(tcps_sack_snd_opts);
1331 	ASSIGN(tcps_sack_drop_opts);
1332 	ASSIGN(tcps_outswtso);
1333 	ASSIGN(tcps_outhwtso);
1334 	ASSIGN(tcps_outpkttso);
1335 	ASSIGN(tcps_outbadtso);
1336 	ASSIGN(tcps_inswlro);
1337 	ASSIGN(tcps_inhwlro);
1338 	ASSIGN(tcps_inpktlro);
1339 	ASSIGN(tcps_inbadlro);
1340 
1341 #undef ASSIGN
1342 
1343 	mtx_enter(&syn_cache_mtx);
1344 	set = &tcp_syn_cache[tcp_syn_cache_active];
1345 	tcpstat.tcps_sc_hash_size = set->scs_size;
1346 	tcpstat.tcps_sc_entry_count = set->scs_count;
1347 	tcpstat.tcps_sc_entry_limit = tcp_syn_cache_limit;
1348 	tcpstat.tcps_sc_bucket_maxlen = 0;
1349 	for (i = 0; i < set->scs_size; i++) {
1350 		if (tcpstat.tcps_sc_bucket_maxlen <
1351 		    set->scs_buckethead[i].sch_length)
1352 			tcpstat.tcps_sc_bucket_maxlen =
1353 				set->scs_buckethead[i].sch_length;
1354 	}
1355 	tcpstat.tcps_sc_bucket_limit = tcp_syn_bucket_limit;
1356 	tcpstat.tcps_sc_uses_left = set->scs_use;
1357 	mtx_leave(&syn_cache_mtx);
1358 
1359 	return (sysctl_rdstruct(oldp, oldlenp, newp,
1360 	    &tcpstat, sizeof(tcpstat)));
1361 }
1362 
1363 /*
1364  * Sysctl for tcp variables.
1365  */
1366 int
1367 tcp_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1368     size_t newlen)
1369 {
1370 	int error, nval;
1371 
1372 	/* All sysctl names at this level are terminal. */
1373 	if (namelen != 1)
1374 		return (ENOTDIR);
1375 
1376 	switch (name[0]) {
1377 	case TCPCTL_KEEPINITTIME:
1378 		NET_LOCK();
1379 		nval = tcptv_keep_init / TCP_TIME(1);
1380 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1381 		    1, 3 * (TCPTV_KEEP_INIT / TCP_TIME(1)));
1382 		if (!error)
1383 			tcptv_keep_init = TCP_TIME(nval);
1384 		NET_UNLOCK();
1385 		return (error);
1386 
1387 	case TCPCTL_KEEPIDLE:
1388 		NET_LOCK();
1389 		nval = tcp_keepidle / TCP_TIME(1);
1390 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1391 		    1, 5 * (TCPTV_KEEP_IDLE / TCP_TIME(1)));
1392 		if (!error)
1393 			tcp_keepidle = TCP_TIME(nval);
1394 		NET_UNLOCK();
1395 		return (error);
1396 
1397 	case TCPCTL_KEEPINTVL:
1398 		NET_LOCK();
1399 		nval = tcp_keepintvl / TCP_TIME(1);
1400 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1401 		    1, 3 * (TCPTV_KEEPINTVL / TCP_TIME(1)));
1402 		if (!error)
1403 			tcp_keepintvl = TCP_TIME(nval);
1404 		NET_UNLOCK();
1405 		return (error);
1406 
1407 	case TCPCTL_BADDYNAMIC:
1408 		NET_LOCK();
1409 		error = sysctl_struct(oldp, oldlenp, newp, newlen,
1410 		    baddynamicports.tcp, sizeof(baddynamicports.tcp));
1411 		NET_UNLOCK();
1412 		return (error);
1413 
1414 	case TCPCTL_ROOTONLY:
1415 		if (newp && securelevel > 0)
1416 			return (EPERM);
1417 		NET_LOCK();
1418 		error = sysctl_struct(oldp, oldlenp, newp, newlen,
1419 		    rootonlyports.tcp, sizeof(rootonlyports.tcp));
1420 		NET_UNLOCK();
1421 		return (error);
1422 
1423 	case TCPCTL_IDENT:
1424 		NET_LOCK();
1425 		error = tcp_ident(oldp, oldlenp, newp, newlen, 0);
1426 		NET_UNLOCK();
1427 		return (error);
1428 
1429 	case TCPCTL_DROP:
1430 		NET_LOCK();
1431 		error = tcp_ident(oldp, oldlenp, newp, newlen, 1);
1432 		NET_UNLOCK();
1433 		return (error);
1434 
1435 	case TCPCTL_REASS_LIMIT:
1436 		NET_LOCK();
1437 		nval = tcp_reass_limit;
1438 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1439 		if (!error && nval != tcp_reass_limit) {
1440 			error = pool_sethardlimit(&tcpqe_pool, nval, NULL, 0);
1441 			if (!error)
1442 				tcp_reass_limit = nval;
1443 		}
1444 		NET_UNLOCK();
1445 		return (error);
1446 
1447 	case TCPCTL_SACKHOLE_LIMIT:
1448 		NET_LOCK();
1449 		nval = tcp_sackhole_limit;
1450 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1451 		if (!error && nval != tcp_sackhole_limit) {
1452 			error = pool_sethardlimit(&sackhl_pool, nval, NULL, 0);
1453 			if (!error)
1454 				tcp_sackhole_limit = nval;
1455 		}
1456 		NET_UNLOCK();
1457 		return (error);
1458 
1459 	case TCPCTL_STATS:
1460 		return (tcp_sysctl_tcpstat(oldp, oldlenp, newp));
1461 
1462 	case TCPCTL_SYN_USE_LIMIT:
1463 		NET_LOCK();
1464 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
1465 		    &tcp_syn_use_limit, 0, INT_MAX);
1466 		if (!error && newp != NULL) {
1467 			/*
1468 			 * Global tcp_syn_use_limit is used when reseeding a
1469 			 * new cache.  Also update the value in active cache.
1470 			 */
1471 			mtx_enter(&syn_cache_mtx);
1472 			if (tcp_syn_cache[0].scs_use > tcp_syn_use_limit)
1473 				tcp_syn_cache[0].scs_use = tcp_syn_use_limit;
1474 			if (tcp_syn_cache[1].scs_use > tcp_syn_use_limit)
1475 				tcp_syn_cache[1].scs_use = tcp_syn_use_limit;
1476 			mtx_leave(&syn_cache_mtx);
1477 		}
1478 		NET_UNLOCK();
1479 		return (error);
1480 
1481 	case TCPCTL_SYN_HASH_SIZE:
1482 		NET_LOCK();
1483 		nval = tcp_syn_hash_size;
1484 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
1485 		    &nval, 1, 100000);
1486 		if (!error && nval != tcp_syn_hash_size) {
1487 			/*
1488 			 * If global hash size has been changed,
1489 			 * switch sets as soon as possible.  Then
1490 			 * the actual hash array will be reallocated.
1491 			 */
1492 			mtx_enter(&syn_cache_mtx);
1493 			if (tcp_syn_cache[0].scs_size != nval)
1494 				tcp_syn_cache[0].scs_use = 0;
1495 			if (tcp_syn_cache[1].scs_size != nval)
1496 				tcp_syn_cache[1].scs_use = 0;
1497 			tcp_syn_hash_size = nval;
1498 			mtx_leave(&syn_cache_mtx);
1499 		}
1500 		NET_UNLOCK();
1501 		return (error);
1502 
1503 	default:
1504 		NET_LOCK();
1505 		error = sysctl_bounded_arr(tcpctl_vars, nitems(tcpctl_vars),
1506 		    name, namelen, oldp, oldlenp, newp, newlen);
1507 		NET_UNLOCK();
1508 		return (error);
1509 	}
1510 	/* NOTREACHED */
1511 }
1512 
1513 /*
1514  * Scale the send buffer so that inflight data is not accounted against
1515  * the limit. The buffer will scale with the congestion window, if the
1516  * the receiver stops acking data the window will shrink and therefore
1517  * the buffer size will shrink as well.
1518  * In low memory situation try to shrink the buffer to the initial size
1519  * disabling the send buffer scaling as long as the situation persists.
1520  */
1521 void
1522 tcp_update_sndspace(struct tcpcb *tp)
1523 {
1524 	struct socket *so = tp->t_inpcb->inp_socket;
1525 	u_long nmax = so->so_snd.sb_hiwat;
1526 
1527 	if (sbchecklowmem()) {
1528 		/* low on memory try to get rid of some */
1529 		if (tcp_sendspace < nmax)
1530 			nmax = tcp_sendspace;
1531 	} else if (so->so_snd.sb_wat != tcp_sendspace)
1532 		/* user requested buffer size, auto-scaling disabled */
1533 		nmax = so->so_snd.sb_wat;
1534 	else
1535 		/* automatic buffer scaling */
1536 		nmax = MIN(sb_max, so->so_snd.sb_wat + tp->snd_max -
1537 		    tp->snd_una);
1538 
1539 	/* a writable socket must be preserved because of poll(2) semantics */
1540 	if (sbspace(so, &so->so_snd) >= so->so_snd.sb_lowat) {
1541 		if (nmax < so->so_snd.sb_cc + so->so_snd.sb_lowat)
1542 			nmax = so->so_snd.sb_cc + so->so_snd.sb_lowat;
1543 		/* keep in sync with sbreserve() calculation */
1544 		if (nmax * 8 < so->so_snd.sb_mbcnt + so->so_snd.sb_lowat)
1545 			nmax = (so->so_snd.sb_mbcnt+so->so_snd.sb_lowat+7) / 8;
1546 	}
1547 
1548 	/* round to MSS boundary */
1549 	nmax = roundup(nmax, tp->t_maxseg);
1550 
1551 	if (nmax != so->so_snd.sb_hiwat)
1552 		sbreserve(so, &so->so_snd, nmax);
1553 }
1554 
1555 /*
1556  * Scale the recv buffer by looking at how much data was transferred in
1557  * one approximated RTT. If more than a big part of the recv buffer was
1558  * transferred during that time we increase the buffer by a constant.
1559  * In low memory situation try to shrink the buffer to the initial size.
1560  */
1561 void
1562 tcp_update_rcvspace(struct tcpcb *tp)
1563 {
1564 	struct socket *so = tp->t_inpcb->inp_socket;
1565 	u_long nmax = so->so_rcv.sb_hiwat;
1566 
1567 	if (sbchecklowmem()) {
1568 		/* low on memory try to get rid of some */
1569 		if (tcp_recvspace < nmax)
1570 			nmax = tcp_recvspace;
1571 	} else if (so->so_rcv.sb_wat != tcp_recvspace)
1572 		/* user requested buffer size, auto-scaling disabled */
1573 		nmax = so->so_rcv.sb_wat;
1574 	else {
1575 		/* automatic buffer scaling */
1576 		if (tp->rfbuf_cnt > so->so_rcv.sb_hiwat / 8 * 7)
1577 			nmax = MIN(sb_max, so->so_rcv.sb_hiwat +
1578 			    tcp_autorcvbuf_inc);
1579 	}
1580 
1581 	/* a readable socket must be preserved because of poll(2) semantics */
1582 	if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat &&
1583 	    nmax < so->so_snd.sb_lowat)
1584 		nmax = so->so_snd.sb_lowat;
1585 
1586 	if (nmax == so->so_rcv.sb_hiwat)
1587 		return;
1588 
1589 	/* round to MSS boundary */
1590 	nmax = roundup(nmax, tp->t_maxseg);
1591 	sbreserve(so, &so->so_rcv, nmax);
1592 }
1593