xref: /openbsd-src/sys/netinet/tcp_usrreq.c (revision 53555c846a0a6f917dbd0a191f826da995ab1c42)
1 /*	$OpenBSD: tcp_usrreq.c,v 1.239 2025/01/09 16:47:24 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 <netinet6/ip6_var.h>
94 #include <netinet/tcp.h>
95 #include <netinet/tcp_fsm.h>
96 #include <netinet/tcp_seq.h>
97 #include <netinet/tcp_timer.h>
98 #include <netinet/tcp_var.h>
99 #include <netinet/tcp_debug.h>
100 
101 #ifdef INET6
102 #include <netinet6/in6_var.h>
103 #endif
104 
105 /*
106  * Locks used to protect global variables in this file:
107  *	I	immutable after creation
108  */
109 
110 #ifndef TCP_SENDSPACE
111 #define	TCP_SENDSPACE	1024*16
112 #endif
113 u_int	tcp_sendspace = TCP_SENDSPACE;		/* [I] */
114 #ifndef TCP_RECVSPACE
115 #define	TCP_RECVSPACE	1024*16
116 #endif
117 u_int	tcp_recvspace = TCP_RECVSPACE;		/* [I] */
118 u_int	tcp_autorcvbuf_inc = 16 * 1024;		/* [I] */
119 
120 const struct pr_usrreqs tcp_usrreqs = {
121 	.pru_attach	= tcp_attach,
122 	.pru_detach	= tcp_detach,
123 	.pru_bind	= tcp_bind,
124 	.pru_listen	= tcp_listen,
125 	.pru_connect	= tcp_connect,
126 	.pru_accept	= tcp_accept,
127 	.pru_disconnect	= tcp_disconnect,
128 	.pru_shutdown	= tcp_shutdown,
129 	.pru_rcvd	= tcp_rcvd,
130 	.pru_send	= tcp_send,
131 	.pru_abort	= tcp_abort,
132 	.pru_sense	= tcp_sense,
133 	.pru_rcvoob	= tcp_rcvoob,
134 	.pru_sendoob	= tcp_sendoob,
135 	.pru_control	= in_control,
136 	.pru_sockaddr	= tcp_sockaddr,
137 	.pru_peeraddr	= tcp_peeraddr,
138 };
139 
140 #ifdef INET6
141 const struct pr_usrreqs tcp6_usrreqs = {
142 	.pru_attach	= tcp_attach,
143 	.pru_detach	= tcp_detach,
144 	.pru_bind	= tcp_bind,
145 	.pru_listen	= tcp_listen,
146 	.pru_connect	= tcp_connect,
147 	.pru_accept	= tcp_accept,
148 	.pru_disconnect	= tcp_disconnect,
149 	.pru_shutdown	= tcp_shutdown,
150 	.pru_rcvd	= tcp_rcvd,
151 	.pru_send	= tcp_send,
152 	.pru_abort	= tcp_abort,
153 	.pru_sense	= tcp_sense,
154 	.pru_rcvoob	= tcp_rcvoob,
155 	.pru_sendoob	= tcp_sendoob,
156 	.pru_control	= in6_control,
157 	.pru_sockaddr	= tcp_sockaddr,
158 	.pru_peeraddr	= tcp_peeraddr,
159 };
160 #endif
161 
162 const struct sysctl_bounded_args tcpctl_vars[] = {
163 	{ TCPCTL_RFC1323, &tcp_do_rfc1323, 0, 1 },
164 	{ TCPCTL_SACK, &tcp_do_sack, 0, 1 },
165 	{ TCPCTL_MSSDFLT, &tcp_mssdflt, TCP_MSS, 65535 },
166 	{ TCPCTL_RSTPPSLIMIT, &tcp_rst_ppslim, 1, 1000 * 1000 },
167 	{ TCPCTL_ACK_ON_PUSH, &tcp_ack_on_push, 0, 1 },
168 #ifdef TCP_ECN
169 	{ TCPCTL_ECN, &tcp_do_ecn, 0, 1 },
170 #endif
171 	{ TCPCTL_SYN_CACHE_LIMIT, &tcp_syn_cache_limit, 1, 1000 * 1000 },
172 	{ TCPCTL_SYN_BUCKET_LIMIT, &tcp_syn_bucket_limit, 1, INT_MAX },
173 	{ TCPCTL_RFC3390, &tcp_do_rfc3390, 0, 2 },
174 	{ TCPCTL_ALWAYS_KEEPALIVE, &tcp_always_keepalive, 0, 1 },
175 	{ TCPCTL_TSO, &tcp_do_tso, 0, 1 },
176 };
177 
178 struct	inpcbtable tcbtable;
179 #ifdef INET6
180 struct	inpcbtable tcb6table;
181 #endif
182 
183 int	tcp_fill_info(struct tcpcb *, struct socket *, struct mbuf *);
184 int	tcp_ident(void *, size_t *, void *, size_t, int);
185 
186 static inline int tcp_sogetpcb(struct socket *, struct inpcb **,
187 		    struct tcpcb **);
188 
189 static inline int
190 tcp_sogetpcb(struct socket *so, struct inpcb **rinp, struct tcpcb **rtp)
191 {
192 	struct inpcb *inp;
193 	struct tcpcb *tp;
194 
195 	/*
196 	 * When a TCP is attached to a socket, then there will be
197 	 * a (struct inpcb) pointed at by the socket, and this
198 	 * structure will point at a subsidiary (struct tcpcb).
199 	 */
200 	if ((inp = sotoinpcb(so)) == NULL || (tp = intotcpcb(inp)) == NULL) {
201 		int error;
202 
203 		if ((error = READ_ONCE(so->so_error)))
204 			return error;
205 		return EINVAL;
206 	}
207 
208 	*rinp = inp;
209 	*rtp = tp;
210 
211 	return 0;
212 }
213 
214 /*
215  * Export internal TCP state information via a struct tcp_info without
216  * leaking any sensitive information. Sequence numbers are reported
217  * relative to the initial sequence number.
218  */
219 int
220 tcp_fill_info(struct tcpcb *tp, struct socket *so, struct mbuf *m)
221 {
222 	struct proc *p = curproc;
223 	struct tcp_info *ti;
224 	u_int t = 1000;		/* msec => usec */
225 	uint64_t now;
226 
227 	if (sizeof(*ti) > MLEN) {
228 		MCLGETL(m, M_WAITOK, sizeof(*ti));
229 		if (!ISSET(m->m_flags, M_EXT))
230 			return ENOMEM;
231 	}
232 	ti = mtod(m, struct tcp_info *);
233 	m->m_len = sizeof(*ti);
234 	memset(ti, 0, sizeof(*ti));
235 	now = tcp_now();
236 
237 	ti->tcpi_state = tp->t_state;
238 	if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
239 		ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
240 	if (tp->t_flags & TF_SACK_PERMIT)
241 		ti->tcpi_options |= TCPI_OPT_SACK;
242 	if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
243 		ti->tcpi_options |= TCPI_OPT_WSCALE;
244 		ti->tcpi_snd_wscale = tp->snd_scale;
245 		ti->tcpi_rcv_wscale = tp->rcv_scale;
246 	}
247 #ifdef TCP_ECN
248 	if (tp->t_flags & TF_ECN_PERMIT)
249 		ti->tcpi_options |= TCPI_OPT_ECN;
250 #endif
251 
252 	ti->tcpi_rto = tp->t_rxtcur * t;
253 	ti->tcpi_snd_mss = tp->t_maxseg;
254 	ti->tcpi_rcv_mss = tp->t_peermss;
255 
256 	ti->tcpi_last_data_sent = (now - tp->t_sndtime) * t;
257 	ti->tcpi_last_ack_sent = (now - tp->t_sndacktime) * t;
258 	ti->tcpi_last_data_recv = (now - tp->t_rcvtime) * t;
259 	ti->tcpi_last_ack_recv = (now - tp->t_rcvacktime) * t;
260 
261 	ti->tcpi_rtt = ((uint64_t)tp->t_srtt * t) >>
262 	    (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
263 	ti->tcpi_rttvar = ((uint64_t)tp->t_rttvar * t) >>
264 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT);
265 	ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
266 	ti->tcpi_snd_cwnd = tp->snd_cwnd;
267 
268 	ti->tcpi_rcv_space = tp->rcv_wnd;
269 
270 	/*
271 	 * Provide only minimal information for unprivileged processes.
272 	 */
273 	if (suser(p) != 0)
274 		return 0;
275 
276 	/* FreeBSD-specific extension fields for tcp_info.  */
277 	ti->tcpi_snd_wnd = tp->snd_wnd;
278 	ti->tcpi_snd_nxt = tp->snd_nxt - tp->iss;
279 	ti->tcpi_rcv_nxt = tp->rcv_nxt - tp->irs;
280 	/* missing tcpi_toe_tid */
281 	ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
282 	ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
283 	ti->tcpi_snd_zerowin = tp->t_sndzerowin;
284 
285 	/* OpenBSD extensions */
286 	ti->tcpi_rttmin = tp->t_rttmin * t;
287 	ti->tcpi_max_sndwnd = tp->max_sndwnd;
288 	ti->tcpi_rcv_adv = tp->rcv_adv - tp->irs;
289 	ti->tcpi_rcv_up = tp->rcv_up - tp->irs;
290 	ti->tcpi_snd_una = tp->snd_una - tp->iss;
291 	ti->tcpi_snd_up = tp->snd_up - tp->iss;
292 	ti->tcpi_snd_wl1 = tp->snd_wl1 - tp->iss;
293 	ti->tcpi_snd_wl2 = tp->snd_wl2 - tp->iss;
294 	ti->tcpi_snd_max = tp->snd_max - tp->iss;
295 
296 	ti->tcpi_ts_recent = tp->ts_recent; /* XXX value from the wire */
297 	ti->tcpi_ts_recent_age = (now - tp->ts_recent_age) * t;
298 	ti->tcpi_rfbuf_cnt = tp->rfbuf_cnt;
299 	ti->tcpi_rfbuf_ts = (now - tp->rfbuf_ts) * t;
300 
301 	mtx_enter(&so->so_rcv.sb_mtx);
302 	ti->tcpi_so_rcv_sb_cc = so->so_rcv.sb_cc;
303 	ti->tcpi_so_rcv_sb_hiwat = so->so_rcv.sb_hiwat;
304 	ti->tcpi_so_rcv_sb_lowat = so->so_rcv.sb_lowat;
305 	ti->tcpi_so_rcv_sb_wat = so->so_rcv.sb_wat;
306 	mtx_leave(&so->so_rcv.sb_mtx);
307 	mtx_enter(&so->so_snd.sb_mtx);
308 	ti->tcpi_so_snd_sb_cc = so->so_snd.sb_cc;
309 	ti->tcpi_so_snd_sb_hiwat = so->so_snd.sb_hiwat;
310 	ti->tcpi_so_snd_sb_lowat = so->so_snd.sb_lowat;
311 	ti->tcpi_so_snd_sb_wat = so->so_snd.sb_wat;
312 	mtx_leave(&so->so_snd.sb_mtx);
313 
314 	return 0;
315 }
316 
317 int
318 tcp_ctloutput(int op, struct socket *so, int level, int optname,
319     struct mbuf *m)
320 {
321 	int error = 0;
322 	struct inpcb *inp;
323 	struct tcpcb *tp;
324 	int i;
325 
326 	inp = sotoinpcb(so);
327 	if (inp == NULL)
328 		return (ECONNRESET);
329 	if (level != IPPROTO_TCP) {
330 #ifdef INET6
331 		if (ISSET(inp->inp_flags, INP_IPV6))
332 			error = ip6_ctloutput(op, so, level, optname, m);
333 		else
334 #endif
335 			error = ip_ctloutput(op, so, level, optname, m);
336 		return (error);
337 	}
338 	tp = intotcpcb(inp);
339 
340 	switch (op) {
341 
342 	case PRCO_SETOPT:
343 		switch (optname) {
344 
345 		case TCP_NODELAY:
346 			if (m == NULL || m->m_len < sizeof (int))
347 				error = EINVAL;
348 			else if (*mtod(m, int *))
349 				tp->t_flags |= TF_NODELAY;
350 			else
351 				tp->t_flags &= ~TF_NODELAY;
352 			break;
353 
354 		case TCP_NOPUSH:
355 			if (m == NULL || m->m_len < sizeof (int))
356 				error = EINVAL;
357 			else if (*mtod(m, int *))
358 				tp->t_flags |= TF_NOPUSH;
359 			else if (tp->t_flags & TF_NOPUSH) {
360 				tp->t_flags &= ~TF_NOPUSH;
361 				if (TCPS_HAVEESTABLISHED(tp->t_state))
362 					error = tcp_output(tp);
363 			}
364 			break;
365 
366 		case TCP_MAXSEG:
367 			if (m == NULL || m->m_len < sizeof (int)) {
368 				error = EINVAL;
369 				break;
370 			}
371 
372 			i = *mtod(m, int *);
373 			if (i > 0 && i <= tp->t_maxseg)
374 				tp->t_maxseg = i;
375 			else
376 				error = EINVAL;
377 			break;
378 
379 		case TCP_SACK_ENABLE:
380 			if (m == NULL || m->m_len < sizeof (int)) {
381 				error = EINVAL;
382 				break;
383 			}
384 
385 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
386 				error = EPERM;
387 				break;
388 			}
389 
390 			if (tp->t_flags & TF_SIGNATURE) {
391 				error = EPERM;
392 				break;
393 			}
394 
395 			if (*mtod(m, int *))
396 				tp->sack_enable = 1;
397 			else
398 				tp->sack_enable = 0;
399 			break;
400 #ifdef TCP_SIGNATURE
401 		case TCP_MD5SIG:
402 			if (m == NULL || m->m_len < sizeof (int)) {
403 				error = EINVAL;
404 				break;
405 			}
406 
407 			if (TCPS_HAVEESTABLISHED(tp->t_state)) {
408 				error = EPERM;
409 				break;
410 			}
411 
412 			if (*mtod(m, int *)) {
413 				tp->t_flags |= TF_SIGNATURE;
414 				tp->sack_enable = 0;
415 			} else
416 				tp->t_flags &= ~TF_SIGNATURE;
417 			break;
418 #endif /* TCP_SIGNATURE */
419 		default:
420 			error = ENOPROTOOPT;
421 			break;
422 		}
423 		break;
424 
425 	case PRCO_GETOPT:
426 		switch (optname) {
427 		case TCP_NODELAY:
428 			m->m_len = sizeof(int);
429 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
430 			break;
431 		case TCP_NOPUSH:
432 			m->m_len = sizeof(int);
433 			*mtod(m, int *) = tp->t_flags & TF_NOPUSH;
434 			break;
435 		case TCP_MAXSEG:
436 			m->m_len = sizeof(int);
437 			*mtod(m, int *) = tp->t_maxseg;
438 			break;
439 		case TCP_SACK_ENABLE:
440 			m->m_len = sizeof(int);
441 			*mtod(m, int *) = tp->sack_enable;
442 			break;
443 		case TCP_INFO:
444 			error = tcp_fill_info(tp, so, m);
445 			break;
446 #ifdef TCP_SIGNATURE
447 		case TCP_MD5SIG:
448 			m->m_len = sizeof(int);
449 			*mtod(m, int *) = tp->t_flags & TF_SIGNATURE;
450 			break;
451 #endif
452 		default:
453 			error = ENOPROTOOPT;
454 			break;
455 		}
456 		break;
457 	}
458 	return (error);
459 }
460 
461 /*
462  * Attach TCP protocol to socket, allocating
463  * internet protocol control block, tcp control block,
464  * buffer space, and entering LISTEN state to accept connections.
465  */
466 int
467 tcp_attach(struct socket *so, int proto, int wait)
468 {
469 	struct inpcbtable *table;
470 	struct tcpcb *tp;
471 	struct inpcb *inp;
472 	int error;
473 
474 	if (so->so_pcb)
475 		return EISCONN;
476 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0 ||
477 	    sbcheckreserve(so->so_snd.sb_wat, tcp_sendspace) ||
478 	    sbcheckreserve(so->so_rcv.sb_wat, tcp_recvspace)) {
479 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
480 		if (error)
481 			return (error);
482 	}
483 
484 	NET_ASSERT_LOCKED();
485 #ifdef INET6
486 	if (so->so_proto->pr_domain->dom_family == PF_INET6)
487 		table = &tcb6table;
488 	else
489 #endif
490 		table = &tcbtable;
491 	error = in_pcballoc(so, table, wait);
492 	if (error)
493 		return (error);
494 	inp = sotoinpcb(so);
495 	tp = tcp_newtcpcb(inp, wait);
496 	if (tp == NULL) {
497 		unsigned int nofd = so->so_state & SS_NOFDREF;	/* XXX */
498 
499 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
500 		in_pcbdetach(inp);
501 		so->so_state |= nofd;
502 		return (ENOBUFS);
503 	}
504 	tp->t_state = TCPS_CLOSED;
505 #ifdef INET6
506 	if (ISSET(inp->inp_flags, INP_IPV6))
507 		tp->pf = PF_INET6;
508 	else
509 #endif
510 		tp->pf = PF_INET;
511 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
512 		so->so_linger = TCP_LINGERTIME;
513 
514 	if (so->so_options & SO_DEBUG)
515 		tcp_trace(TA_USER, TCPS_CLOSED, tp, tp, NULL, PRU_ATTACH, 0);
516 	return (0);
517 }
518 
519 int
520 tcp_detach(struct socket *so)
521 {
522 	struct inpcb *inp;
523 	struct tcpcb *otp = NULL, *tp;
524 	int error;
525 	short ostate;
526 
527 	soassertlocked(so);
528 
529 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
530 		return (error);
531 
532 	if (so->so_options & SO_DEBUG) {
533 		otp = tp;
534 		ostate = tp->t_state;
535 	}
536 
537 	/*
538 	 * Detach the TCP protocol from the socket.
539 	 * If the protocol state is non-embryonic, then can't
540 	 * do this directly: have to initiate a PRU_DISCONNECT,
541 	 * which may finish later; embryonic TCB's can just
542 	 * be discarded here.
543 	 */
544 	tp = tcp_dodisconnect(tp);
545 
546 	if (otp)
547 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_DETACH, 0);
548 	return (0);
549 }
550 
551 /*
552  * Give the socket an address.
553  */
554 int
555 tcp_bind(struct socket *so, struct mbuf *nam, struct proc *p)
556 {
557 	struct inpcb *inp;
558 	struct tcpcb *tp;
559 	int error;
560 	short ostate;
561 
562 	soassertlocked(so);
563 
564 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
565 		return (error);
566 
567 	if (so->so_options & SO_DEBUG)
568 		ostate = tp->t_state;
569 
570 	error = in_pcbbind(inp, nam, p);
571 
572 	if (so->so_options & SO_DEBUG)
573 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_BIND, 0);
574 	return (error);
575 }
576 
577 /*
578  * Prepare to accept connections.
579  */
580 int
581 tcp_listen(struct socket *so)
582 {
583 	struct inpcb *inp;
584 	struct tcpcb *tp, *otp = NULL;
585 	int error;
586 	short ostate;
587 
588 	soassertlocked(so);
589 
590 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
591 		return (error);
592 
593 	if (so->so_options & SO_DEBUG) {
594 		otp = tp;
595 		ostate = tp->t_state;
596 	}
597 
598 	if (inp->inp_lport == 0)
599 		if ((error = in_pcbbind(inp, NULL, curproc)))
600 			goto out;
601 
602 	/*
603 	 * If the in_pcbbind() above is called, the tp->pf
604 	 * should still be whatever it was before.
605 	 */
606 	tp->t_state = TCPS_LISTEN;
607 
608 out:
609 	if (otp)
610 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_LISTEN, 0);
611 	return (error);
612 }
613 
614 /*
615  * Initiate connection to peer.
616  * Create a template for use in transmissions on this connection.
617  * Enter SYN_SENT state, and mark socket as connecting.
618  * Start keep-alive timer, and seed output sequence space.
619  * Send initial segment on connection.
620  */
621 int
622 tcp_connect(struct socket *so, struct mbuf *nam)
623 {
624 	struct inpcb *inp;
625 	struct tcpcb *tp, *otp = NULL;
626 	int error;
627 	short ostate;
628 
629 	soassertlocked(so);
630 
631 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
632 		return (error);
633 
634 	if (so->so_options & SO_DEBUG) {
635 		otp = tp;
636 		ostate = tp->t_state;
637 	}
638 
639 #ifdef INET6
640 	if (ISSET(inp->inp_flags, INP_IPV6)) {
641 		struct sockaddr_in6 *sin6;
642 
643 		if ((error = in6_nam2sin6(nam, &sin6)))
644 			goto out;
645 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
646 		    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
647 			error = EINVAL;
648 			goto out;
649 		}
650 	} else
651 #endif
652 	{
653 		struct sockaddr_in *sin;
654 
655 		if ((error = in_nam2sin(nam, &sin)))
656 			goto out;
657 		if ((sin->sin_addr.s_addr == INADDR_ANY) ||
658 		    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
659 		    IN_MULTICAST(sin->sin_addr.s_addr) ||
660 		    in_broadcast(sin->sin_addr, inp->inp_rtableid)) {
661 			error = EINVAL;
662 			goto out;
663 		}
664 	}
665 	error = in_pcbconnect(inp, nam);
666 	if (error)
667 		goto out;
668 
669 	tp->t_template = tcp_template(tp);
670 	if (tp->t_template == 0) {
671 		in_pcbunset_faddr(inp);
672 		in_pcbdisconnect(inp);
673 		error = ENOBUFS;
674 		goto out;
675 	}
676 
677 	so->so_state |= SS_CONNECTOUT;
678 
679 	/* Compute window scaling to request.  */
680 	tcp_rscale(tp, sb_max);
681 
682 	soisconnecting(so);
683 	tcpstat_inc(tcps_connattempt);
684 	tp->t_state = TCPS_SYN_SENT;
685 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
686 	tcp_set_iss_tsm(tp);
687 	tcp_sendseqinit(tp);
688 	tp->snd_last = tp->snd_una;
689 	error = tcp_output(tp);
690 
691 out:
692 	if (otp)
693 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_CONNECT, 0);
694 	return (error);
695 }
696 
697 /*
698  * Accept a connection.  Essentially all the work is done at higher
699  * levels; just return the address of the peer, storing through addr.
700  */
701 int
702 tcp_accept(struct socket *so, struct mbuf *nam)
703 {
704 	struct inpcb *inp;
705 	struct tcpcb *tp;
706 	int error;
707 
708 	soassertlocked(so);
709 
710 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
711 		return (error);
712 
713 	in_setpeeraddr(inp, nam);
714 
715 	if (so->so_options & SO_DEBUG)
716 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_ACCEPT, 0);
717 	return (0);
718 }
719 
720 /*
721  * Initiate disconnect from peer.
722  * If connection never passed embryonic stage, just drop;
723  * else if don't need to let data drain, then can just drop anyways,
724  * else have to begin TCP shutdown process: mark socket disconnecting,
725  * drain unread data, state switch to reflect user close, and
726  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
727  * when peer sends FIN and acks ours.
728  *
729  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
730  */
731 int
732 tcp_disconnect(struct socket *so)
733 {
734 	struct inpcb *inp;
735 	struct tcpcb *tp, *otp = NULL;
736 	int error;
737 	short ostate;
738 
739 	soassertlocked(so);
740 
741 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
742 		return (error);
743 
744 	if (so->so_options & SO_DEBUG) {
745 		otp = tp;
746 		ostate = tp->t_state;
747 	}
748 
749 	tp = tcp_dodisconnect(tp);
750 
751 	if (otp)
752 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_DISCONNECT, 0);
753 	return (0);
754 }
755 
756 /*
757  * Mark the connection as being incapable of further output.
758  */
759 int
760 tcp_shutdown(struct socket *so)
761 {
762 	struct inpcb *inp;
763 	struct tcpcb *tp, *otp = NULL;
764 	int error;
765 	short ostate;
766 
767 	soassertlocked(so);
768 
769 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
770 		return (error);
771 
772 	if (so->so_options & SO_DEBUG) {
773 		otp = tp;
774 		ostate = tp->t_state;
775 	}
776 
777 	if (so->so_snd.sb_state & SS_CANTSENDMORE)
778 		goto out;
779 
780 	socantsendmore(so);
781 	tp = tcp_usrclosed(tp);
782 	if (tp)
783 		error = tcp_output(tp);
784 
785 out:
786 	if (otp)
787 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_SHUTDOWN, 0);
788 	return (error);
789 }
790 
791 /*
792  * After a receive, possibly send window update to peer.
793  */
794 void
795 tcp_rcvd(struct socket *so)
796 {
797 	struct inpcb *inp;
798 	struct tcpcb *tp;
799 	short ostate;
800 
801 	soassertlocked(so);
802 
803 	if (tcp_sogetpcb(so, &inp, &tp))
804 		return;
805 
806 	if (so->so_options & SO_DEBUG)
807 		ostate = tp->t_state;
808 
809 	/*
810 	 * soreceive() calls this function when a user receives
811 	 * ancillary data on a listening socket. We don't call
812 	 * tcp_output in such a case, since there is no header
813 	 * template for a listening socket and hence the kernel
814 	 * will panic.
815 	 */
816 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) != 0)
817 		(void) tcp_output(tp);
818 
819 	if (so->so_options & SO_DEBUG)
820 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_RCVD, 0);
821 }
822 
823 /*
824  * Do a send by putting data in output queue and updating urgent
825  * marker if URG set.  Possibly send more data.
826  */
827 int
828 tcp_send(struct socket *so, struct mbuf *m, struct mbuf *nam,
829     struct mbuf *control)
830 {
831 	struct inpcb *inp;
832 	struct tcpcb *tp;
833 	int error;
834 	short ostate;
835 
836 	soassertlocked(so);
837 
838 	if (control && control->m_len) {
839 		error = EINVAL;
840 		goto out;
841 	}
842 
843 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
844 		goto out;
845 
846 	if (so->so_options & SO_DEBUG)
847 		ostate = tp->t_state;
848 
849 	mtx_enter(&so->so_snd.sb_mtx);
850 	sbappendstream(so, &so->so_snd, m);
851 	mtx_leave(&so->so_snd.sb_mtx);
852 	m = NULL;
853 
854 	error = tcp_output(tp);
855 
856 	if (so->so_options & SO_DEBUG)
857 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_SEND, 0);
858 
859 out:
860 	m_freem(control);
861 	m_freem(m);
862 
863 	return (error);
864 }
865 
866 /*
867  * Abort the TCP.
868  */
869 void
870 tcp_abort(struct socket *so)
871 {
872 	struct inpcb *inp;
873 	struct tcpcb *tp, *otp = NULL;
874 	short ostate;
875 
876 	soassertlocked(so);
877 
878 	if (tcp_sogetpcb(so, &inp, &tp))
879 		return;
880 
881 	if (so->so_options & SO_DEBUG) {
882 		otp = tp;
883 		ostate = tp->t_state;
884 	}
885 
886 	tp = tcp_drop(tp, ECONNABORTED);
887 
888 	if (otp)
889 		tcp_trace(TA_USER, ostate, tp, otp, NULL, PRU_ABORT, 0);
890 }
891 
892 int
893 tcp_sense(struct socket *so, struct stat *ub)
894 {
895 	struct inpcb *inp;
896 	struct tcpcb *tp;
897 	int error;
898 
899 	soassertlocked(so);
900 
901 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
902 		return (error);
903 
904 	mtx_enter(&so->so_snd.sb_mtx);
905 	ub->st_blksize = so->so_snd.sb_hiwat;
906 	mtx_leave(&so->so_snd.sb_mtx);
907 
908 	if (so->so_options & SO_DEBUG)
909 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_SENSE, 0);
910 	return (0);
911 }
912 
913 int
914 tcp_rcvoob(struct socket *so, struct mbuf *m, int flags)
915 {
916 	struct inpcb *inp;
917 	struct tcpcb *tp;
918 	int error;
919 
920 	soassertlocked(so);
921 
922 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
923 		return (error);
924 
925 	if ((so->so_oobmark == 0 &&
926 	    (so->so_rcv.sb_state & SS_RCVATMARK) == 0) ||
927 	    so->so_options & SO_OOBINLINE ||
928 	    tp->t_oobflags & TCPOOB_HADDATA) {
929 		error = EINVAL;
930 		goto out;
931 	}
932 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
933 		error = EWOULDBLOCK;
934 		goto out;
935 	}
936 	m->m_len = 1;
937 	*mtod(m, caddr_t) = tp->t_iobc;
938 	if ((flags & MSG_PEEK) == 0)
939 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
940 out:
941 	if (so->so_options & SO_DEBUG)
942 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_RCVOOB, 0);
943 	return (error);
944 }
945 
946 int
947 tcp_sendoob(struct socket *so, struct mbuf *m, struct mbuf *nam,
948     struct mbuf *control)
949 {
950 	struct inpcb *inp;
951 	struct tcpcb *tp;
952 	int error;
953 	short ostate;
954 
955 	soassertlocked(so);
956 
957 	if (control && control->m_len) {
958 		error = EINVAL;
959 		goto release;
960 	}
961 
962 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
963 		goto release;
964 
965 	if (so->so_options & SO_DEBUG)
966 		ostate = tp->t_state;
967 
968 	if (sbspace(so, &so->so_snd) < -512) {
969 		error = ENOBUFS;
970 		goto out;
971 	}
972 
973 	/*
974 	 * According to RFC961 (Assigned Protocols),
975 	 * the urgent pointer points to the last octet
976 	 * of urgent data.  We continue, however,
977 	 * to consider it to indicate the first octet
978 	 * of data past the urgent section.
979 	 * Otherwise, snd_up should be one lower.
980 	 */
981 	mtx_enter(&so->so_snd.sb_mtx);
982 	sbappendstream(so, &so->so_snd, m);
983 	mtx_leave(&so->so_snd.sb_mtx);
984 	m = NULL;
985 	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
986 	tp->t_force = 1;
987 	error = tcp_output(tp);
988 	tp->t_force = 0;
989 
990 out:
991 	if (so->so_options & SO_DEBUG)
992 		tcp_trace(TA_USER, ostate, tp, tp, NULL, PRU_SENDOOB, 0);
993 
994 release:
995 	m_freem(control);
996 	m_freem(m);
997 
998 	return (error);
999 }
1000 
1001 int
1002 tcp_sockaddr(struct socket *so, struct mbuf *nam)
1003 {
1004 	struct inpcb *inp;
1005 	struct tcpcb *tp;
1006 	int error;
1007 
1008 	soassertlocked(so);
1009 
1010 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
1011 		return (error);
1012 
1013 	in_setsockaddr(inp, nam);
1014 
1015 	if (so->so_options & SO_DEBUG)
1016 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL,
1017 		    PRU_SOCKADDR, 0);
1018 	return (0);
1019 }
1020 
1021 int
1022 tcp_peeraddr(struct socket *so, struct mbuf *nam)
1023 {
1024 	struct inpcb *inp;
1025 	struct tcpcb *tp;
1026 	int error;
1027 
1028 	soassertlocked(so);
1029 
1030 	if ((error = tcp_sogetpcb(so, &inp, &tp)))
1031 		return (error);
1032 
1033 	in_setpeeraddr(inp, nam);
1034 
1035 	if (so->so_options & SO_DEBUG)
1036 		tcp_trace(TA_USER, tp->t_state, tp, tp, NULL, PRU_PEERADDR, 0);
1037 	return (0);
1038 }
1039 
1040 /*
1041  * Initiate (or continue) disconnect.
1042  * If embryonic state, just send reset (once).
1043  * If in ``let data drain'' option and linger null, just drop.
1044  * Otherwise (hard), mark socket disconnecting and drop
1045  * current input data; switch states based on user close, and
1046  * send segment to peer (with FIN).
1047  */
1048 struct tcpcb *
1049 tcp_dodisconnect(struct tcpcb *tp)
1050 {
1051 	struct socket *so = tp->t_inpcb->inp_socket;
1052 
1053 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
1054 		tp = tcp_close(tp);
1055 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1056 		tp = tcp_drop(tp, 0);
1057 	else {
1058 		soisdisconnecting(so);
1059 		mtx_enter(&so->so_rcv.sb_mtx);
1060 		sbflush(so, &so->so_rcv);
1061 		mtx_leave(&so->so_rcv.sb_mtx);
1062 		tp = tcp_usrclosed(tp);
1063 		if (tp)
1064 			(void) tcp_output(tp);
1065 	}
1066 	return (tp);
1067 }
1068 
1069 /*
1070  * User issued close, and wish to trail through shutdown states:
1071  * if never received SYN, just forget it.  If got a SYN from peer,
1072  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1073  * If already got a FIN from peer, then almost done; go to LAST_ACK
1074  * state.  In all other cases, have already sent FIN to peer (e.g.
1075  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1076  * for peer to send FIN or not respond to keep-alives, etc.
1077  * We can let the user exit from the close as soon as the FIN is acked.
1078  */
1079 struct tcpcb *
1080 tcp_usrclosed(struct tcpcb *tp)
1081 {
1082 
1083 	switch (tp->t_state) {
1084 
1085 	case TCPS_CLOSED:
1086 	case TCPS_LISTEN:
1087 	case TCPS_SYN_SENT:
1088 		tp->t_state = TCPS_CLOSED;
1089 		tp = tcp_close(tp);
1090 		break;
1091 
1092 	case TCPS_SYN_RECEIVED:
1093 	case TCPS_ESTABLISHED:
1094 		tp->t_state = TCPS_FIN_WAIT_1;
1095 		break;
1096 
1097 	case TCPS_CLOSE_WAIT:
1098 		tp->t_state = TCPS_LAST_ACK;
1099 		break;
1100 	}
1101 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1102 		soisdisconnected(tp->t_inpcb->inp_socket);
1103 		/*
1104 		 * If we are in FIN_WAIT_2, we arrived here because the
1105 		 * application did a shutdown of the send side.  Like the
1106 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
1107 		 * a full close, we start a timer to make sure sockets are
1108 		 * not left in FIN_WAIT_2 forever.
1109 		 */
1110 		if (tp->t_state == TCPS_FIN_WAIT_2)
1111 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1112 	}
1113 	return (tp);
1114 }
1115 
1116 /*
1117  * Look up a socket for ident or tcpdrop, ...
1118  */
1119 int
1120 tcp_ident(void *oldp, size_t *oldlenp, void *newp, size_t newlen, int dodrop)
1121 {
1122 	int error = 0;
1123 	struct tcp_ident_mapping tir;
1124 	struct inpcb *inp;
1125 	struct socket *so = NULL;
1126 	struct sockaddr_in *fin, *lin;
1127 #ifdef INET6
1128 	struct sockaddr_in6 *fin6, *lin6;
1129 	struct in6_addr f6, l6;
1130 #endif
1131 
1132 	if (dodrop) {
1133 		if (oldp != NULL || *oldlenp != 0)
1134 			return (EINVAL);
1135 		if (newp == NULL)
1136 			return (EPERM);
1137 		if (newlen < sizeof(tir))
1138 			return (ENOMEM);
1139 		if ((error = copyin(newp, &tir, sizeof (tir))) != 0 )
1140 			return (error);
1141 	} else {
1142 		if (oldp == NULL)
1143 			return (EINVAL);
1144 		if (*oldlenp < sizeof(tir))
1145 			return (ENOMEM);
1146 		if (newp != NULL || newlen != 0)
1147 			return (EINVAL);
1148 		if ((error = copyin(oldp, &tir, sizeof (tir))) != 0 )
1149 			return (error);
1150 	}
1151 
1152 	NET_LOCK_SHARED();
1153 
1154 	switch (tir.faddr.ss_family) {
1155 #ifdef INET6
1156 	case AF_INET6:
1157 		if (tir.laddr.ss_family != AF_INET6) {
1158 			NET_UNLOCK_SHARED();
1159 			return (EAFNOSUPPORT);
1160 		}
1161 		fin6 = (struct sockaddr_in6 *)&tir.faddr;
1162 		error = in6_embedscope(&f6, fin6, NULL, NULL);
1163 		if (error) {
1164 			NET_UNLOCK_SHARED();
1165 			return EINVAL;	/*?*/
1166 		}
1167 		lin6 = (struct sockaddr_in6 *)&tir.laddr;
1168 		error = in6_embedscope(&l6, lin6, NULL, NULL);
1169 		if (error) {
1170 			NET_UNLOCK_SHARED();
1171 			return EINVAL;	/*?*/
1172 		}
1173 		break;
1174 #endif
1175 	case AF_INET:
1176 		if (tir.laddr.ss_family != AF_INET) {
1177 			NET_UNLOCK_SHARED();
1178 			return (EAFNOSUPPORT);
1179 		}
1180 		fin = (struct sockaddr_in *)&tir.faddr;
1181 		lin = (struct sockaddr_in *)&tir.laddr;
1182 		break;
1183 	default:
1184 		NET_UNLOCK_SHARED();
1185 		return (EAFNOSUPPORT);
1186 	}
1187 
1188 	switch (tir.faddr.ss_family) {
1189 #ifdef INET6
1190 	case AF_INET6:
1191 		inp = in6_pcblookup(&tcb6table, &f6,
1192 		    fin6->sin6_port, &l6, lin6->sin6_port, tir.rdomain);
1193 		break;
1194 #endif
1195 	case AF_INET:
1196 		inp = in_pcblookup(&tcbtable, fin->sin_addr,
1197 		    fin->sin_port, lin->sin_addr, lin->sin_port, tir.rdomain);
1198 		break;
1199 	default:
1200 		unhandled_af(tir.faddr.ss_family);
1201 	}
1202 
1203 	if (dodrop) {
1204 		struct tcpcb *tp = NULL;
1205 
1206 		if (inp != NULL) {
1207 			so = in_pcbsolock_ref(inp);
1208 			if (so != NULL)
1209 				tp = intotcpcb(inp);
1210 		}
1211 		if (tp != NULL && !ISSET(so->so_options, SO_ACCEPTCONN))
1212 			tp = tcp_drop(tp, ECONNABORTED);
1213 		else
1214 			error = ESRCH;
1215 
1216 		in_pcbsounlock_rele(inp, so);
1217 		NET_UNLOCK_SHARED();
1218 		in_pcbunref(inp);
1219 		return (error);
1220 	}
1221 
1222 	if (inp == NULL) {
1223 		tcpstat_inc(tcps_pcbhashmiss);
1224 		switch (tir.faddr.ss_family) {
1225 #ifdef INET6
1226 		case AF_INET6:
1227 			inp = in6_pcblookup_listen(&tcb6table,
1228 			    &l6, lin6->sin6_port, NULL, tir.rdomain);
1229 			break;
1230 #endif
1231 		case AF_INET:
1232 			inp = in_pcblookup_listen(&tcbtable,
1233 			    lin->sin_addr, lin->sin_port, NULL, tir.rdomain);
1234 			break;
1235 		}
1236 	}
1237 
1238 	if (inp != NULL)
1239 		so = in_pcbsolock_ref(inp);
1240 
1241 	if (so != NULL && ISSET(so->so_state, SS_CONNECTOUT)) {
1242 		tir.ruid = so->so_ruid;
1243 		tir.euid = so->so_euid;
1244 	} else {
1245 		tir.ruid = -1;
1246 		tir.euid = -1;
1247 	}
1248 
1249 	in_pcbsounlock_rele(inp, so);
1250 	NET_UNLOCK_SHARED();
1251 	in_pcbunref(inp);
1252 
1253 	*oldlenp = sizeof(tir);
1254 	return copyout(&tir, oldp, sizeof(tir));
1255 }
1256 
1257 int
1258 tcp_sysctl_tcpstat(void *oldp, size_t *oldlenp, void *newp)
1259 {
1260 	uint64_t counters[tcps_ncounters];
1261 	struct tcpstat tcpstat;
1262 	struct syn_cache_set *set;
1263 	int i = 0;
1264 
1265 #define ASSIGN(field)	do { tcpstat.field = counters[i++]; } while (0)
1266 
1267 	memset(&tcpstat, 0, sizeof tcpstat);
1268 	counters_read(tcpcounters, counters, nitems(counters), NULL);
1269 	ASSIGN(tcps_connattempt);
1270 	ASSIGN(tcps_accepts);
1271 	ASSIGN(tcps_connects);
1272 	ASSIGN(tcps_drops);
1273 	ASSIGN(tcps_conndrops);
1274 	ASSIGN(tcps_closed);
1275 	ASSIGN(tcps_segstimed);
1276 	ASSIGN(tcps_rttupdated);
1277 	ASSIGN(tcps_delack);
1278 	ASSIGN(tcps_timeoutdrop);
1279 	ASSIGN(tcps_rexmttimeo);
1280 	ASSIGN(tcps_persisttimeo);
1281 	ASSIGN(tcps_persistdrop);
1282 	ASSIGN(tcps_keeptimeo);
1283 	ASSIGN(tcps_keepprobe);
1284 	ASSIGN(tcps_keepdrops);
1285 	ASSIGN(tcps_sndtotal);
1286 	ASSIGN(tcps_sndpack);
1287 	ASSIGN(tcps_sndbyte);
1288 	ASSIGN(tcps_sndrexmitpack);
1289 	ASSIGN(tcps_sndrexmitbyte);
1290 	ASSIGN(tcps_sndrexmitfast);
1291 	ASSIGN(tcps_sndacks);
1292 	ASSIGN(tcps_sndprobe);
1293 	ASSIGN(tcps_sndurg);
1294 	ASSIGN(tcps_sndwinup);
1295 	ASSIGN(tcps_sndctrl);
1296 	ASSIGN(tcps_rcvtotal);
1297 	ASSIGN(tcps_rcvpack);
1298 	ASSIGN(tcps_rcvbyte);
1299 	ASSIGN(tcps_rcvbadsum);
1300 	ASSIGN(tcps_rcvbadoff);
1301 	ASSIGN(tcps_rcvmemdrop);
1302 	ASSIGN(tcps_rcvnosec);
1303 	ASSIGN(tcps_rcvshort);
1304 	ASSIGN(tcps_rcvduppack);
1305 	ASSIGN(tcps_rcvdupbyte);
1306 	ASSIGN(tcps_rcvpartduppack);
1307 	ASSIGN(tcps_rcvpartdupbyte);
1308 	ASSIGN(tcps_rcvoopack);
1309 	ASSIGN(tcps_rcvoobyte);
1310 	ASSIGN(tcps_rcvpackafterwin);
1311 	ASSIGN(tcps_rcvbyteafterwin);
1312 	ASSIGN(tcps_rcvafterclose);
1313 	ASSIGN(tcps_rcvwinprobe);
1314 	ASSIGN(tcps_rcvdupack);
1315 	ASSIGN(tcps_rcvacktoomuch);
1316 	ASSIGN(tcps_rcvacktooold);
1317 	ASSIGN(tcps_rcvackpack);
1318 	ASSIGN(tcps_rcvackbyte);
1319 	ASSIGN(tcps_rcvwinupd);
1320 	ASSIGN(tcps_pawsdrop);
1321 	ASSIGN(tcps_predack);
1322 	ASSIGN(tcps_preddat);
1323 	ASSIGN(tcps_pcbhashmiss);
1324 	ASSIGN(tcps_noport);
1325 	ASSIGN(tcps_badsyn);
1326 	ASSIGN(tcps_dropsyn);
1327 	ASSIGN(tcps_rcvbadsig);
1328 	ASSIGN(tcps_rcvgoodsig);
1329 	ASSIGN(tcps_inswcsum);
1330 	ASSIGN(tcps_outswcsum);
1331 	ASSIGN(tcps_ecn_accepts);
1332 	ASSIGN(tcps_ecn_rcvece);
1333 	ASSIGN(tcps_ecn_rcvcwr);
1334 	ASSIGN(tcps_ecn_rcvce);
1335 	ASSIGN(tcps_ecn_sndect);
1336 	ASSIGN(tcps_ecn_sndece);
1337 	ASSIGN(tcps_ecn_sndcwr);
1338 	ASSIGN(tcps_cwr_ecn);
1339 	ASSIGN(tcps_cwr_frecovery);
1340 	ASSIGN(tcps_cwr_timeout);
1341 	ASSIGN(tcps_sc_added);
1342 	ASSIGN(tcps_sc_completed);
1343 	ASSIGN(tcps_sc_timed_out);
1344 	ASSIGN(tcps_sc_overflowed);
1345 	ASSIGN(tcps_sc_reset);
1346 	ASSIGN(tcps_sc_unreach);
1347 	ASSIGN(tcps_sc_bucketoverflow);
1348 	ASSIGN(tcps_sc_aborted);
1349 	ASSIGN(tcps_sc_dupesyn);
1350 	ASSIGN(tcps_sc_dropped);
1351 	ASSIGN(tcps_sc_collisions);
1352 	ASSIGN(tcps_sc_retransmitted);
1353 	ASSIGN(tcps_sc_seedrandom);
1354 	ASSIGN(tcps_sc_hash_size);
1355 	ASSIGN(tcps_sc_entry_count);
1356 	ASSIGN(tcps_sc_entry_limit);
1357 	ASSIGN(tcps_sc_bucket_maxlen);
1358 	ASSIGN(tcps_sc_bucket_limit);
1359 	ASSIGN(tcps_sc_uses_left);
1360 	ASSIGN(tcps_conndrained);
1361 	ASSIGN(tcps_sack_recovery_episode);
1362 	ASSIGN(tcps_sack_rexmits);
1363 	ASSIGN(tcps_sack_rexmit_bytes);
1364 	ASSIGN(tcps_sack_rcv_opts);
1365 	ASSIGN(tcps_sack_snd_opts);
1366 	ASSIGN(tcps_sack_drop_opts);
1367 	ASSIGN(tcps_outswtso);
1368 	ASSIGN(tcps_outhwtso);
1369 	ASSIGN(tcps_outpkttso);
1370 	ASSIGN(tcps_outbadtso);
1371 	ASSIGN(tcps_inswlro);
1372 	ASSIGN(tcps_inhwlro);
1373 	ASSIGN(tcps_inpktlro);
1374 	ASSIGN(tcps_inbadlro);
1375 
1376 #undef ASSIGN
1377 
1378 	mtx_enter(&syn_cache_mtx);
1379 	set = &tcp_syn_cache[tcp_syn_cache_active];
1380 	tcpstat.tcps_sc_hash_size = set->scs_size;
1381 	tcpstat.tcps_sc_entry_count = set->scs_count;
1382 	tcpstat.tcps_sc_entry_limit = atomic_load_int(&tcp_syn_cache_limit);
1383 	tcpstat.tcps_sc_bucket_maxlen = 0;
1384 	for (i = 0; i < set->scs_size; i++) {
1385 		if (tcpstat.tcps_sc_bucket_maxlen <
1386 		    set->scs_buckethead[i].sch_length)
1387 			tcpstat.tcps_sc_bucket_maxlen =
1388 				set->scs_buckethead[i].sch_length;
1389 	}
1390 	tcpstat.tcps_sc_bucket_limit = atomic_load_int(&tcp_syn_bucket_limit);
1391 	tcpstat.tcps_sc_uses_left = set->scs_use;
1392 	mtx_leave(&syn_cache_mtx);
1393 
1394 	return (sysctl_rdstruct(oldp, oldlenp, newp,
1395 	    &tcpstat, sizeof(tcpstat)));
1396 }
1397 
1398 /*
1399  * Sysctl for tcp variables.
1400  */
1401 int
1402 tcp_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1403     size_t newlen)
1404 {
1405 	int error, oval, nval;
1406 
1407 	/* All sysctl names at this level are terminal. */
1408 	if (namelen != 1)
1409 		return (ENOTDIR);
1410 
1411 	switch (name[0]) {
1412 	case TCPCTL_KEEPINITTIME:
1413 		NET_LOCK();
1414 		nval = tcptv_keep_init / TCP_TIME(1);
1415 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1416 		    1, 3 * (TCPTV_KEEP_INIT / TCP_TIME(1)));
1417 		if (!error)
1418 			tcptv_keep_init = TCP_TIME(nval);
1419 		NET_UNLOCK();
1420 		return (error);
1421 
1422 	case TCPCTL_KEEPIDLE:
1423 		NET_LOCK();
1424 		nval = tcp_keepidle / TCP_TIME(1);
1425 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1426 		    1, 5 * (TCPTV_KEEP_IDLE / TCP_TIME(1)));
1427 		if (!error)
1428 			tcp_keepidle = TCP_TIME(nval);
1429 		NET_UNLOCK();
1430 		return (error);
1431 
1432 	case TCPCTL_KEEPINTVL:
1433 		NET_LOCK();
1434 		nval = tcp_keepintvl / TCP_TIME(1);
1435 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen, &nval,
1436 		    1, 3 * (TCPTV_KEEPINTVL / TCP_TIME(1)));
1437 		if (!error)
1438 			tcp_keepintvl = TCP_TIME(nval);
1439 		NET_UNLOCK();
1440 		return (error);
1441 
1442 	case TCPCTL_BADDYNAMIC:
1443 		NET_LOCK();
1444 		error = sysctl_struct(oldp, oldlenp, newp, newlen,
1445 		    baddynamicports.tcp, sizeof(baddynamicports.tcp));
1446 		NET_UNLOCK();
1447 		return (error);
1448 
1449 	case TCPCTL_ROOTONLY:
1450 		if (newp && securelevel > 0)
1451 			return (EPERM);
1452 		NET_LOCK();
1453 		error = sysctl_struct(oldp, oldlenp, newp, newlen,
1454 		    rootonlyports.tcp, sizeof(rootonlyports.tcp));
1455 		NET_UNLOCK();
1456 		return (error);
1457 
1458 	case TCPCTL_IDENT:
1459 		return tcp_ident(oldp, oldlenp, newp, newlen, 0);
1460 
1461 	case TCPCTL_DROP:
1462 		return tcp_ident(oldp, oldlenp, newp, newlen, 1);
1463 
1464 	case TCPCTL_REASS_LIMIT:
1465 		NET_LOCK();
1466 		nval = tcp_reass_limit;
1467 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1468 		if (!error && nval != tcp_reass_limit) {
1469 			error = pool_sethardlimit(&tcpqe_pool, nval, NULL, 0);
1470 			if (!error)
1471 				tcp_reass_limit = nval;
1472 		}
1473 		NET_UNLOCK();
1474 		return (error);
1475 
1476 	case TCPCTL_SACKHOLE_LIMIT:
1477 		NET_LOCK();
1478 		nval = tcp_sackhole_limit;
1479 		error = sysctl_int(oldp, oldlenp, newp, newlen, &nval);
1480 		if (!error && nval != tcp_sackhole_limit) {
1481 			error = pool_sethardlimit(&sackhl_pool, nval, NULL, 0);
1482 			if (!error)
1483 				tcp_sackhole_limit = nval;
1484 		}
1485 		NET_UNLOCK();
1486 		return (error);
1487 
1488 	case TCPCTL_STATS:
1489 		return (tcp_sysctl_tcpstat(oldp, oldlenp, newp));
1490 
1491 	case TCPCTL_SYN_USE_LIMIT:
1492 		oval = nval = atomic_load_int(&tcp_syn_use_limit);
1493 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
1494 		    &nval, 0, INT_MAX);
1495 		if (!error && oval != nval) {
1496 			/*
1497 			 * Global tcp_syn_use_limit is used when reseeding a
1498 			 * new cache.  Also update the value in active cache.
1499 			 */
1500 			mtx_enter(&syn_cache_mtx);
1501 			if (tcp_syn_cache[0].scs_use > nval)
1502 				tcp_syn_cache[0].scs_use = nval;
1503 			if (tcp_syn_cache[1].scs_use > nval)
1504 				tcp_syn_cache[1].scs_use = nval;
1505 			tcp_syn_use_limit = nval;
1506 			mtx_leave(&syn_cache_mtx);
1507 		}
1508 		return (error);
1509 
1510 	case TCPCTL_SYN_HASH_SIZE:
1511 		oval = nval = atomic_load_int(&tcp_syn_hash_size);
1512 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
1513 		    &nval, 1, 100000);
1514 		if (!error && oval != nval) {
1515 			/*
1516 			 * If global hash size has been changed,
1517 			 * switch sets as soon as possible.  Then
1518 			 * the actual hash array will be reallocated.
1519 			 */
1520 			mtx_enter(&syn_cache_mtx);
1521 			if (tcp_syn_cache[0].scs_size != nval)
1522 				tcp_syn_cache[0].scs_use = 0;
1523 			if (tcp_syn_cache[1].scs_size != nval)
1524 				tcp_syn_cache[1].scs_use = 0;
1525 			tcp_syn_hash_size = nval;
1526 			mtx_leave(&syn_cache_mtx);
1527 		}
1528 		return (error);
1529 
1530 	default:
1531 		return sysctl_bounded_arr(tcpctl_vars, nitems(tcpctl_vars),
1532 		    name, namelen, oldp, oldlenp, newp, newlen);
1533 	}
1534 	/* NOTREACHED */
1535 }
1536 
1537 /*
1538  * Scale the send buffer so that inflight data is not accounted against
1539  * the limit. The buffer will scale with the congestion window, if the
1540  * the receiver stops acking data the window will shrink and therefore
1541  * the buffer size will shrink as well.
1542  * In low memory situation try to shrink the buffer to the initial size
1543  * disabling the send buffer scaling as long as the situation persists.
1544  */
1545 void
1546 tcp_update_sndspace(struct tcpcb *tp)
1547 {
1548 	struct socket *so = tp->t_inpcb->inp_socket;
1549 	u_long nmax;
1550 
1551 	mtx_enter(&so->so_snd.sb_mtx);
1552 
1553 	nmax = so->so_snd.sb_hiwat;
1554 
1555 	if (sbchecklowmem()) {
1556 		/* low on memory try to get rid of some */
1557 		if (tcp_sendspace < nmax)
1558 			nmax = tcp_sendspace;
1559 	} else if (so->so_snd.sb_wat != tcp_sendspace) {
1560 		/* user requested buffer size, auto-scaling disabled */
1561 		nmax = so->so_snd.sb_wat;
1562 	} else {
1563 		/* automatic buffer scaling */
1564 		nmax = MIN(sb_max, so->so_snd.sb_wat + tp->snd_max -
1565 		    tp->snd_una);
1566 	}
1567 
1568 	/* a writable socket must be preserved because of poll(2) semantics */
1569 	if (sbspace_locked(so, &so->so_snd) >= so->so_snd.sb_lowat) {
1570 		if (nmax < so->so_snd.sb_cc + so->so_snd.sb_lowat)
1571 			nmax = so->so_snd.sb_cc + so->so_snd.sb_lowat;
1572 		/* keep in sync with sbreserve() calculation */
1573 		if (nmax * 8 < so->so_snd.sb_mbcnt + so->so_snd.sb_lowat)
1574 			nmax = (so->so_snd.sb_mbcnt+so->so_snd.sb_lowat+7) / 8;
1575 	}
1576 
1577 	/* round to MSS boundary */
1578 	nmax = roundup(nmax, tp->t_maxseg);
1579 
1580 	if (nmax != so->so_snd.sb_hiwat)
1581 		sbreserve(so, &so->so_snd, nmax);
1582 
1583 	mtx_leave(&so->so_snd.sb_mtx);
1584 }
1585 
1586 /*
1587  * Scale the recv buffer by looking at how much data was transferred in
1588  * one approximated RTT. If more than a big part of the recv buffer was
1589  * transferred during that time we increase the buffer by a constant.
1590  * In low memory situation try to shrink the buffer to the initial size.
1591  */
1592 void
1593 tcp_update_rcvspace(struct tcpcb *tp)
1594 {
1595 	struct socket *so = tp->t_inpcb->inp_socket;
1596 	u_long nmax;
1597 
1598 	mtx_enter(&so->so_rcv.sb_mtx);
1599 
1600 	nmax = so->so_rcv.sb_hiwat;
1601 
1602 	if (sbchecklowmem()) {
1603 		/* low on memory try to get rid of some */
1604 		if (tcp_recvspace < nmax)
1605 			nmax = tcp_recvspace;
1606 	} else if (so->so_rcv.sb_wat != tcp_recvspace) {
1607 		/* user requested buffer size, auto-scaling disabled */
1608 		nmax = so->so_rcv.sb_wat;
1609 	} else {
1610 		/* automatic buffer scaling */
1611 		if (tp->rfbuf_cnt > so->so_rcv.sb_hiwat / 8 * 7)
1612 			nmax = MIN(sb_max, so->so_rcv.sb_hiwat +
1613 			    tcp_autorcvbuf_inc);
1614 	}
1615 
1616 	/* a readable socket must be preserved because of poll(2) semantics */
1617 	mtx_enter(&so->so_snd.sb_mtx);
1618 	if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat &&
1619 	    nmax < so->so_snd.sb_lowat)
1620 		nmax = so->so_snd.sb_lowat;
1621 	mtx_leave(&so->so_snd.sb_mtx);
1622 
1623 	if (nmax != so->so_rcv.sb_hiwat) {
1624 		/* round to MSS boundary */
1625 		nmax = roundup(nmax, tp->t_maxseg);
1626 		sbreserve(so, &so->so_rcv, nmax);
1627 	}
1628 
1629 	mtx_leave(&so->so_rcv.sb_mtx);
1630 }
1631