xref: /freebsd/sys/netinet/tcp_timewait.c (revision 5129159789cc9d7bc514e4546b88e3427695002d)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
34  * $FreeBSD$
35  */
36 
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/protosw.h>
52 
53 #include <vm/vm_zone.h>
54 
55 #include <net/route.h>
56 #include <net/if.h>
57 
58 #define _IP_VHL
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcpip.h>
71 #ifdef TCPDEBUG
72 #include <netinet/tcp_debug.h>
73 #endif
74 
75 int 	tcp_mssdflt = TCP_MSS;
76 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
77     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
78 
79 #ifdef INET6
80 int	tcp_v6mssdflt = TCP6_MSS;
81 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
82 	CTLFLAG_RW, &tcp_v6mssdflt , 0, "");
83 #endif
84 
85 #if 0
86 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
87 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
88     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
89 #endif
90 
91 static int	tcp_do_rfc1323 = 1;
92 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
93     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
94 
95 static int	tcp_do_rfc1644 = 0;
96 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
97     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
98 
99 static int	tcp_tcbhashsize = 0;
100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
101      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
102 
103 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
104     &tcbinfo.ipi_count, 0, "Number of active PCBs");
105 
106 static void	tcp_cleartaocache __P((void));
107 static void	tcp_notify __P((struct inpcb *, int));
108 
109 /*
110  * Target size of TCP PCB hash tables. Must be a power of two.
111  *
112  * Note that this can be overridden by the kernel environment
113  * variable net.inet.tcp.tcbhashsize
114  */
115 #ifndef TCBHASHSIZE
116 #define TCBHASHSIZE	512
117 #endif
118 
119 /*
120  * This is the actual shape of what we allocate using the zone
121  * allocator.  Doing it this way allows us to protect both structures
122  * using the same generation count, and also eliminates the overhead
123  * of allocating tcpcbs separately.  By hiding the structure here,
124  * we avoid changing most of the rest of the code (although it needs
125  * to be changed, eventually, for greater efficiency).
126  */
127 #define	ALIGNMENT	32
128 #define	ALIGNM1		(ALIGNMENT - 1)
129 struct	inp_tp {
130 	union {
131 		struct	inpcb inp;
132 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
133 	} inp_tp_u;
134 	struct	tcpcb tcb;
135 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
136 	struct	callout inp_tp_delack;
137 };
138 #undef ALIGNMENT
139 #undef ALIGNM1
140 
141 /*
142  * Tcp initialization
143  */
144 void
145 tcp_init()
146 {
147 	int hashsize;
148 
149 	tcp_iss = random();	/* wrong, but better than a constant */
150 	tcp_ccgen = 1;
151 	tcp_cleartaocache();
152 
153 	tcp_delacktime = TCPTV_DELACK;
154 	tcp_keepinit = TCPTV_KEEP_INIT;
155 	tcp_keepidle = TCPTV_KEEP_IDLE;
156 	tcp_keepintvl = TCPTV_KEEPINTVL;
157 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
158 	tcp_msl = TCPTV_MSL;
159 
160 	LIST_INIT(&tcb);
161 	tcbinfo.listhead = &tcb;
162 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
163 	if (!powerof2(hashsize)) {
164 		printf("WARNING: TCB hash size not a power of 2\n");
165 		hashsize = 512; /* safe default */
166 	}
167 	tcp_tcbhashsize = hashsize;
168 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
169 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
170 					&tcbinfo.porthashmask);
171 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
172 				 ZONE_INTERRUPT, 0);
173 
174 	if (max_protohdr < sizeof(struct tcpiphdr))
175 		max_protohdr = sizeof(struct tcpiphdr);
176 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
177 		panic("tcp_init");
178 }
179 
180 /*
181  * Create template to be used to send tcp packets on a connection.
182  * Call after host entry created, allocates an mbuf and fills
183  * in a skeletal tcp/ip header, minimizing the amount of work
184  * necessary when the connection is used.
185  */
186 struct tcpiphdr *
187 tcp_template(tp)
188 	struct tcpcb *tp;
189 {
190 	register struct inpcb *inp = tp->t_inpcb;
191 	register struct mbuf *m;
192 	register struct tcpiphdr *n;
193 
194 	if ((n = tp->t_template) == 0) {
195 		m = m_get(M_DONTWAIT, MT_HEADER);
196 		if (m == NULL)
197 			return (0);
198 		m->m_len = sizeof (struct tcpiphdr);
199 		n = mtod(m, struct tcpiphdr *);
200 	}
201 	bzero(n->ti_x1, sizeof(n->ti_x1));
202 	n->ti_pr = IPPROTO_TCP;
203 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
204 	n->ti_src = inp->inp_laddr;
205 	n->ti_dst = inp->inp_faddr;
206 	n->ti_sport = inp->inp_lport;
207 	n->ti_dport = inp->inp_fport;
208 	n->ti_seq = 0;
209 	n->ti_ack = 0;
210 	n->ti_x2 = 0;
211 	n->ti_off = 5;
212 	n->ti_flags = 0;
213 	n->ti_win = 0;
214 	n->ti_sum = 0;
215 	n->ti_urp = 0;
216 	return (n);
217 }
218 
219 /*
220  * Send a single message to the TCP at address specified by
221  * the given TCP/IP header.  If m == 0, then we make a copy
222  * of the tcpiphdr at ti and send directly to the addressed host.
223  * This is used to force keep alive messages out using the TCP
224  * template for a connection tp->t_template.  If flags are given
225  * then we send a message back to the TCP which originated the
226  * segment ti, and discard the mbuf containing it and any other
227  * attached mbufs.
228  *
229  * In any case the ack and sequence number of the transmitted
230  * segment are as specified by the parameters.
231  *
232  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
233  */
234 void
235 tcp_respond(tp, ti, m, ack, seq, flags)
236 	struct tcpcb *tp;
237 	register struct tcpiphdr *ti;
238 	register struct mbuf *m;
239 	tcp_seq ack, seq;
240 	int flags;
241 {
242 	register int tlen;
243 	int win = 0;
244 	struct route *ro = 0;
245 	struct route sro;
246 
247 	if (tp) {
248 		if (!(flags & TH_RST))
249 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
250 		ro = &tp->t_inpcb->inp_route;
251 	} else {
252 		ro = &sro;
253 		bzero(ro, sizeof *ro);
254 	}
255 	if (m == 0) {
256 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
257 		if (m == NULL)
258 			return;
259 #ifdef TCP_COMPAT_42
260 		tlen = 1;
261 #else
262 		tlen = 0;
263 #endif
264 		m->m_data += max_linkhdr;
265 		*mtod(m, struct tcpiphdr *) = *ti;
266 		ti = mtod(m, struct tcpiphdr *);
267 		flags = TH_ACK;
268 	} else {
269 		m_freem(m->m_next);
270 		m->m_next = 0;
271 		m->m_data = (caddr_t)ti;
272 		m->m_len = sizeof (struct tcpiphdr);
273 		tlen = 0;
274 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
275 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long);
276 		xchg(ti->ti_dport, ti->ti_sport, n_short);
277 #undef xchg
278 	}
279 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
280 	tlen += sizeof (struct tcpiphdr);
281 	m->m_len = tlen;
282 	m->m_pkthdr.len = tlen;
283 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
284 	bzero(ti->ti_x1, sizeof(ti->ti_x1));
285 	ti->ti_seq = htonl(seq);
286 	ti->ti_ack = htonl(ack);
287 	ti->ti_x2 = 0;
288 	ti->ti_off = sizeof (struct tcphdr) >> 2;
289 	ti->ti_flags = flags;
290 	if (tp)
291 		ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
292 	else
293 		ti->ti_win = htons((u_short)win);
294 	ti->ti_urp = 0;
295 	ti->ti_sum = 0;
296 	ti->ti_sum = in_cksum(m, tlen);
297 	((struct ip *)ti)->ip_len = tlen;
298 	((struct ip *)ti)->ip_ttl = ip_defttl;
299 #ifdef TCPDEBUG
300 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
301 		tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
302 #endif
303 	(void) ip_output(m, NULL, ro, 0, NULL);
304 	if (ro == &sro && ro->ro_rt) {
305 		RTFREE(ro->ro_rt);
306 	}
307 }
308 
309 /*
310  * Create a new TCP control block, making an
311  * empty reassembly queue and hooking it to the argument
312  * protocol control block.  The `inp' parameter must have
313  * come from the zone allocator set up in tcp_init().
314  */
315 struct tcpcb *
316 tcp_newtcpcb(inp)
317 	struct inpcb *inp;
318 {
319 	struct inp_tp *it;
320 	register struct tcpcb *tp;
321 
322 	it = (struct inp_tp *)inp;
323 	tp = &it->tcb;
324 	bzero((char *) tp, sizeof(struct tcpcb));
325 	tp->t_segq = NULL;
326 	tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
327 
328 	/* Set up our timeouts. */
329 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
330 	callout_init(tp->tt_persist = &it->inp_tp_persist);
331 	callout_init(tp->tt_keep = &it->inp_tp_keep);
332 	callout_init(tp->tt_2msl = &it->inp_tp_2msl);
333 	callout_init(tp->tt_delack = &it->inp_tp_delack);
334 
335 	if (tcp_do_rfc1323)
336 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
337 	if (tcp_do_rfc1644)
338 		tp->t_flags |= TF_REQ_CC;
339 	tp->t_inpcb = inp;	/* XXX */
340 	/*
341 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
342 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
343 	 * reasonable initial retransmit time.
344 	 */
345 	tp->t_srtt = TCPTV_SRTTBASE;
346 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
347 	tp->t_rttmin = TCPTV_MIN;
348 	tp->t_rxtcur = TCPTV_RTOBASE;
349 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
350 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
351 	tp->t_rcvtime = ticks;
352 	inp->inp_ip_ttl = ip_defttl;
353 	inp->inp_ppcb = (caddr_t)tp;
354 	return (tp);		/* XXX */
355 }
356 
357 /*
358  * Drop a TCP connection, reporting
359  * the specified error.  If connection is synchronized,
360  * then send a RST to peer.
361  */
362 struct tcpcb *
363 tcp_drop(tp, errno)
364 	register struct tcpcb *tp;
365 	int errno;
366 {
367 	struct socket *so = tp->t_inpcb->inp_socket;
368 
369 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
370 		tp->t_state = TCPS_CLOSED;
371 		(void) tcp_output(tp);
372 		tcpstat.tcps_drops++;
373 	} else
374 		tcpstat.tcps_conndrops++;
375 	if (errno == ETIMEDOUT && tp->t_softerror)
376 		errno = tp->t_softerror;
377 	so->so_error = errno;
378 	return (tcp_close(tp));
379 }
380 
381 /*
382  * Close a TCP control block:
383  *	discard all space held by the tcp
384  *	discard internet protocol block
385  *	wake up any sleepers
386  */
387 struct tcpcb *
388 tcp_close(tp)
389 	register struct tcpcb *tp;
390 {
391 	register struct mbuf *q;
392 	register struct mbuf *nq;
393 	struct inpcb *inp = tp->t_inpcb;
394 	struct socket *so = inp->inp_socket;
395 	register struct rtentry *rt;
396 	int dosavessthresh;
397 
398 	/*
399 	 * Make sure that all of our timers are stopped before we
400 	 * delete the PCB.
401 	 */
402 	callout_stop(tp->tt_rexmt);
403 	callout_stop(tp->tt_persist);
404 	callout_stop(tp->tt_keep);
405 	callout_stop(tp->tt_2msl);
406 	callout_stop(tp->tt_delack);
407 
408 	/*
409 	 * If we got enough samples through the srtt filter,
410 	 * save the rtt and rttvar in the routing entry.
411 	 * 'Enough' is arbitrarily defined as the 16 samples.
412 	 * 16 samples is enough for the srtt filter to converge
413 	 * to within 5% of the correct value; fewer samples and
414 	 * we could save a very bogus rtt.
415 	 *
416 	 * Don't update the default route's characteristics and don't
417 	 * update anything that the user "locked".
418 	 */
419 	if (tp->t_rttupdated >= 16 &&
420 	    (rt = inp->inp_route.ro_rt) &&
421 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
422 		register u_long i = 0;
423 
424 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
425 			i = tp->t_srtt *
426 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
427 			if (rt->rt_rmx.rmx_rtt && i)
428 				/*
429 				 * filter this update to half the old & half
430 				 * the new values, converting scale.
431 				 * See route.h and tcp_var.h for a
432 				 * description of the scaling constants.
433 				 */
434 				rt->rt_rmx.rmx_rtt =
435 				    (rt->rt_rmx.rmx_rtt + i) / 2;
436 			else
437 				rt->rt_rmx.rmx_rtt = i;
438 			tcpstat.tcps_cachedrtt++;
439 		}
440 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
441 			i = tp->t_rttvar *
442 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
443 			if (rt->rt_rmx.rmx_rttvar && i)
444 				rt->rt_rmx.rmx_rttvar =
445 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
446 			else
447 				rt->rt_rmx.rmx_rttvar = i;
448 			tcpstat.tcps_cachedrttvar++;
449 		}
450 		/*
451 		 * The old comment here said:
452 		 * update the pipelimit (ssthresh) if it has been updated
453 		 * already or if a pipesize was specified & the threshhold
454 		 * got below half the pipesize.  I.e., wait for bad news
455 		 * before we start updating, then update on both good
456 		 * and bad news.
457 		 *
458 		 * But we want to save the ssthresh even if no pipesize is
459 		 * specified explicitly in the route, because such
460 		 * connections still have an implicit pipesize specified
461 		 * by the global tcp_sendspace.  In the absence of a reliable
462 		 * way to calculate the pipesize, it will have to do.
463 		 */
464 		i = tp->snd_ssthresh;
465 		if (rt->rt_rmx.rmx_sendpipe != 0)
466 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
467 		else
468 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
469 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
470 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
471 		    || dosavessthresh) {
472 			/*
473 			 * convert the limit from user data bytes to
474 			 * packets then to packet data bytes.
475 			 */
476 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
477 			if (i < 2)
478 				i = 2;
479 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
480 			if (rt->rt_rmx.rmx_ssthresh)
481 				rt->rt_rmx.rmx_ssthresh =
482 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
483 			else
484 				rt->rt_rmx.rmx_ssthresh = i;
485 			tcpstat.tcps_cachedssthresh++;
486 		}
487 	}
488 	/* free the reassembly queue, if any */
489 	for (q = tp->t_segq; q; q = nq) {
490 		nq = q->m_nextpkt;
491 		tp->t_segq = nq;
492 		m_freem(q);
493 	}
494 	if (tp->t_template)
495 		(void) m_free(dtom(tp->t_template));
496 	inp->inp_ppcb = NULL;
497 	soisdisconnected(so);
498 	in_pcbdetach(inp);
499 	tcpstat.tcps_closed++;
500 	return ((struct tcpcb *)0);
501 }
502 
503 void
504 tcp_drain()
505 {
506 
507 }
508 
509 /*
510  * Notify a tcp user of an asynchronous error;
511  * store error as soft error, but wake up user
512  * (for now, won't do anything until can select for soft error).
513  */
514 static void
515 tcp_notify(inp, error)
516 	struct inpcb *inp;
517 	int error;
518 {
519 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
520 	register struct socket *so = inp->inp_socket;
521 
522 	/*
523 	 * Ignore some errors if we are hooked up.
524 	 * If connection hasn't completed, has retransmitted several times,
525 	 * and receives a second error, give up now.  This is better
526 	 * than waiting a long time to establish a connection that
527 	 * can never complete.
528 	 */
529 	if (tp->t_state == TCPS_ESTABLISHED &&
530 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
531 	      error == EHOSTDOWN)) {
532 		return;
533 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
534 	    tp->t_softerror)
535 		so->so_error = error;
536 	else
537 		tp->t_softerror = error;
538 	wakeup((caddr_t) &so->so_timeo);
539 	sorwakeup(so);
540 	sowwakeup(so);
541 }
542 
543 static int
544 tcp_pcblist SYSCTL_HANDLER_ARGS
545 {
546 	int error, i, n, s;
547 	struct inpcb *inp, **inp_list;
548 	inp_gen_t gencnt;
549 	struct xinpgen xig;
550 
551 	/*
552 	 * The process of preparing the TCB list is too time-consuming and
553 	 * resource-intensive to repeat twice on every request.
554 	 */
555 	if (req->oldptr == 0) {
556 		n = tcbinfo.ipi_count;
557 		req->oldidx = 2 * (sizeof xig)
558 			+ (n + n/8) * sizeof(struct xtcpcb);
559 		return 0;
560 	}
561 
562 	if (req->newptr != 0)
563 		return EPERM;
564 
565 	/*
566 	 * OK, now we're committed to doing something.
567 	 */
568 	s = splnet();
569 	gencnt = tcbinfo.ipi_gencnt;
570 	n = tcbinfo.ipi_count;
571 	splx(s);
572 
573 	xig.xig_len = sizeof xig;
574 	xig.xig_count = n;
575 	xig.xig_gen = gencnt;
576 	xig.xig_sogen = so_gencnt;
577 	error = SYSCTL_OUT(req, &xig, sizeof xig);
578 	if (error)
579 		return error;
580 
581 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
582 	if (inp_list == 0)
583 		return ENOMEM;
584 
585 	s = splnet();
586 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
587 	     inp = inp->inp_list.le_next) {
588 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
589 			inp_list[i++] = inp;
590 	}
591 	splx(s);
592 	n = i;
593 
594 	error = 0;
595 	for (i = 0; i < n; i++) {
596 		inp = inp_list[i];
597 		if (inp->inp_gencnt <= gencnt) {
598 			struct xtcpcb xt;
599 			caddr_t inp_ppcb;
600 			xt.xt_len = sizeof xt;
601 			/* XXX should avoid extra copy */
602 			bcopy(inp, &xt.xt_inp, sizeof *inp);
603 			inp_ppcb = inp->inp_ppcb;
604 			if (inp_ppcb != NULL)
605 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
606 			else
607 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
608 			if (inp->inp_socket)
609 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
610 			error = SYSCTL_OUT(req, &xt, sizeof xt);
611 		}
612 	}
613 	if (!error) {
614 		/*
615 		 * Give the user an updated idea of our state.
616 		 * If the generation differs from what we told
617 		 * her before, she knows that something happened
618 		 * while we were processing this request, and it
619 		 * might be necessary to retry.
620 		 */
621 		s = splnet();
622 		xig.xig_gen = tcbinfo.ipi_gencnt;
623 		xig.xig_sogen = so_gencnt;
624 		xig.xig_count = tcbinfo.ipi_count;
625 		splx(s);
626 		error = SYSCTL_OUT(req, &xig, sizeof xig);
627 	}
628 	free(inp_list, M_TEMP);
629 	return error;
630 }
631 
632 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
633 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
634 
635 static int
636 tcp_getcred SYSCTL_HANDLER_ARGS
637 {
638 	struct sockaddr_in addrs[2];
639 	struct inpcb *inp;
640 	int error, s;
641 
642 	error = suser(req->p);
643 	if (error)
644 		return (error);
645 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
646 	if (error)
647 		return (error);
648 	s = splnet();
649 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
650 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
651 	if (inp == NULL || inp->inp_socket == NULL) {
652 		error = ENOENT;
653 		goto out;
654 	}
655 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
656 out:
657 	splx(s);
658 	return (error);
659 }
660 
661 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
662     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
663 
664 void
665 tcp_ctlinput(cmd, sa, vip)
666 	int cmd;
667 	struct sockaddr *sa;
668 	void *vip;
669 {
670 	register struct ip *ip = vip;
671 	register struct tcphdr *th;
672 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
673 
674 	if (cmd == PRC_QUENCH)
675 		notify = tcp_quench;
676 	else if (cmd == PRC_MSGSIZE)
677 		notify = tcp_mtudisc;
678 	else if (!PRC_IS_REDIRECT(cmd) &&
679 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
680 		return;
681 	if (ip) {
682 		th = (struct tcphdr *)((caddr_t)ip
683 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
684 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
685 			cmd, notify);
686 	} else
687 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
688 }
689 
690 /*
691  * When a source quench is received, close congestion window
692  * to one segment.  We will gradually open it again as we proceed.
693  */
694 void
695 tcp_quench(inp, errno)
696 	struct inpcb *inp;
697 	int errno;
698 {
699 	struct tcpcb *tp = intotcpcb(inp);
700 
701 	if (tp)
702 		tp->snd_cwnd = tp->t_maxseg;
703 }
704 
705 /*
706  * When `need fragmentation' ICMP is received, update our idea of the MSS
707  * based on the new value in the route.  Also nudge TCP to send something,
708  * since we know the packet we just sent was dropped.
709  * This duplicates some code in the tcp_mss() function in tcp_input.c.
710  */
711 void
712 tcp_mtudisc(inp, errno)
713 	struct inpcb *inp;
714 	int errno;
715 {
716 	struct tcpcb *tp = intotcpcb(inp);
717 	struct rtentry *rt;
718 	struct rmxp_tao *taop;
719 	struct socket *so = inp->inp_socket;
720 	int offered;
721 	int mss;
722 
723 	if (tp) {
724 		rt = tcp_rtlookup(inp);
725 		if (!rt || !rt->rt_rmx.rmx_mtu) {
726 			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
727 			return;
728 		}
729 		taop = rmx_taop(rt->rt_rmx);
730 		offered = taop->tao_mssopt;
731 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
732 		if (offered)
733 			mss = min(mss, offered);
734 		/*
735 		 * XXX - The above conditional probably violates the TCP
736 		 * spec.  The problem is that, since we don't know the
737 		 * other end's MSS, we are supposed to use a conservative
738 		 * default.  But, if we do that, then MTU discovery will
739 		 * never actually take place, because the conservative
740 		 * default is much less than the MTUs typically seen
741 		 * on the Internet today.  For the moment, we'll sweep
742 		 * this under the carpet.
743 		 *
744 		 * The conservative default might not actually be a problem
745 		 * if the only case this occurs is when sending an initial
746 		 * SYN with options and data to a host we've never talked
747 		 * to before.  Then, they will reply with an MSS value which
748 		 * will get recorded and the new parameters should get
749 		 * recomputed.  For Further Study.
750 		 */
751 		if (tp->t_maxopd <= mss)
752 			return;
753 		tp->t_maxopd = mss;
754 
755 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
756 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
757 			mss -= TCPOLEN_TSTAMP_APPA;
758 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
759 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
760 			mss -= TCPOLEN_CC_APPA;
761 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
762 		if (mss > MCLBYTES)
763 			mss &= ~(MCLBYTES-1);
764 #else
765 		if (mss > MCLBYTES)
766 			mss = mss / MCLBYTES * MCLBYTES;
767 #endif
768 		if (so->so_snd.sb_hiwat < mss)
769 			mss = so->so_snd.sb_hiwat;
770 
771 		tp->t_maxseg = mss;
772 
773 		tcpstat.tcps_mturesent++;
774 		tp->t_rtttime = 0;
775 		tp->snd_nxt = tp->snd_una;
776 		tcp_output(tp);
777 	}
778 }
779 
780 /*
781  * Look-up the routing entry to the peer of this inpcb.  If no route
782  * is found and it cannot be allocated the return NULL.  This routine
783  * is called by TCP routines that access the rmx structure and by tcp_mss
784  * to get the interface MTU.
785  */
786 struct rtentry *
787 tcp_rtlookup(inp)
788 	struct inpcb *inp;
789 {
790 	struct route *ro;
791 	struct rtentry *rt;
792 
793 	ro = &inp->inp_route;
794 	rt = ro->ro_rt;
795 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
796 		/* No route yet, so try to acquire one */
797 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
798 			ro->ro_dst.sa_family = AF_INET;
799 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
800 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
801 				inp->inp_faddr;
802 			rtalloc(ro);
803 			rt = ro->ro_rt;
804 		}
805 	}
806 	return rt;
807 }
808 
809 /*
810  * Return a pointer to the cached information about the remote host.
811  * The cached information is stored in the protocol specific part of
812  * the route metrics.
813  */
814 struct rmxp_tao *
815 tcp_gettaocache(inp)
816 	struct inpcb *inp;
817 {
818 	struct rtentry *rt = tcp_rtlookup(inp);
819 
820 	/* Make sure this is a host route and is up. */
821 	if (rt == NULL ||
822 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
823 		return NULL;
824 
825 	return rmx_taop(rt->rt_rmx);
826 }
827 
828 /*
829  * Clear all the TAO cache entries, called from tcp_init.
830  *
831  * XXX
832  * This routine is just an empty one, because we assume that the routing
833  * routing tables are initialized at the same time when TCP, so there is
834  * nothing in the cache left over.
835  */
836 static void
837 tcp_cleartaocache()
838 {
839 }
840