xref: /freebsd/sys/netinet/tcp_timewait.c (revision 7f3dea244c40159a41ab22da77a434d7c5b5e85a)
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  *	$Id: tcp_subr.c,v 1.57 1999/07/11 18:32:45 green Exp $
35  */
36 
37 #include "opt_compat.h"
38 #include "opt_tcpdebug.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 
51 #include <vm/vm_zone.h>
52 
53 #include <net/route.h>
54 #include <net/if.h>
55 
56 #define _IP_VHL
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_var.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_fsm.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/tcp_timer.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet/tcpip.h>
69 #ifdef TCPDEBUG
70 #include <netinet/tcp_debug.h>
71 #endif
72 
73 int 	tcp_mssdflt = TCP_MSS;
74 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
75     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
76 
77 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
78 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
79     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
80 
81 static int	tcp_do_rfc1323 = 1;
82 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
83     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
84 
85 static int	tcp_do_rfc1644 = 0;
86 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
87     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
88 
89 static int	tcp_tcbhashsize = 0;
90 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
91      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
92 
93 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
94     &tcbinfo.ipi_count, 0, "Number of active PCBs");
95 
96 static void	tcp_cleartaocache __P((void));
97 static void	tcp_notify __P((struct inpcb *, int));
98 
99 /*
100  * Target size of TCP PCB hash tables. Must be a power of two.
101  *
102  * Note that this can be overridden by the kernel environment
103  * variable net.inet.tcp.tcbhashsize
104  */
105 #ifndef TCBHASHSIZE
106 #define TCBHASHSIZE	512
107 #endif
108 
109 /*
110  * This is the actual shape of what we allocate using the zone
111  * allocator.  Doing it this way allows us to protect both structures
112  * using the same generation count, and also eliminates the overhead
113  * of allocating tcpcbs separately.  By hiding the structure here,
114  * we avoid changing most of the rest of the code (although it needs
115  * to be changed, eventually, for greater efficiency).
116  */
117 #define	ALIGNMENT	32
118 #define	ALIGNM1		(ALIGNMENT - 1)
119 struct	inp_tp {
120 	union {
121 		struct	inpcb inp;
122 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
123 	} inp_tp_u;
124 	struct	tcpcb tcb;
125 };
126 #undef ALIGNMENT
127 #undef ALIGNM1
128 
129 /*
130  * Tcp initialization
131  */
132 void
133 tcp_init()
134 {
135 	int hashsize;
136 
137 	tcp_iss = random();	/* wrong, but better than a constant */
138 	tcp_ccgen = 1;
139 	tcp_cleartaocache();
140 	LIST_INIT(&tcb);
141 	tcbinfo.listhead = &tcb;
142 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
143 	if (!powerof2(hashsize)) {
144 		printf("WARNING: TCB hash size not a power of 2\n");
145 		hashsize = 512; /* safe default */
146 	}
147 	tcp_tcbhashsize = hashsize;
148 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
149 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
150 					&tcbinfo.porthashmask);
151 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
152 				 ZONE_INTERRUPT, 0);
153 	if (max_protohdr < sizeof(struct tcpiphdr))
154 		max_protohdr = sizeof(struct tcpiphdr);
155 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
156 		panic("tcp_init");
157 }
158 
159 /*
160  * Create template to be used to send tcp packets on a connection.
161  * Call after host entry created, allocates an mbuf and fills
162  * in a skeletal tcp/ip header, minimizing the amount of work
163  * necessary when the connection is used.
164  */
165 struct tcpiphdr *
166 tcp_template(tp)
167 	struct tcpcb *tp;
168 {
169 	register struct inpcb *inp = tp->t_inpcb;
170 	register struct mbuf *m;
171 	register struct tcpiphdr *n;
172 
173 	if ((n = tp->t_template) == 0) {
174 		m = m_get(M_DONTWAIT, MT_HEADER);
175 		if (m == NULL)
176 			return (0);
177 		m->m_len = sizeof (struct tcpiphdr);
178 		n = mtod(m, struct tcpiphdr *);
179 	}
180 	bzero(n->ti_x1, sizeof(n->ti_x1));
181 	n->ti_pr = IPPROTO_TCP;
182 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
183 	n->ti_src = inp->inp_laddr;
184 	n->ti_dst = inp->inp_faddr;
185 	n->ti_sport = inp->inp_lport;
186 	n->ti_dport = inp->inp_fport;
187 	n->ti_seq = 0;
188 	n->ti_ack = 0;
189 	n->ti_x2 = 0;
190 	n->ti_off = 5;
191 	n->ti_flags = 0;
192 	n->ti_win = 0;
193 	n->ti_sum = 0;
194 	n->ti_urp = 0;
195 	return (n);
196 }
197 
198 /*
199  * Send a single message to the TCP at address specified by
200  * the given TCP/IP header.  If m == 0, then we make a copy
201  * of the tcpiphdr at ti and send directly to the addressed host.
202  * This is used to force keep alive messages out using the TCP
203  * template for a connection tp->t_template.  If flags are given
204  * then we send a message back to the TCP which originated the
205  * segment ti, and discard the mbuf containing it and any other
206  * attached mbufs.
207  *
208  * In any case the ack and sequence number of the transmitted
209  * segment are as specified by the parameters.
210  *
211  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
212  */
213 void
214 tcp_respond(tp, ti, m, ack, seq, flags)
215 	struct tcpcb *tp;
216 	register struct tcpiphdr *ti;
217 	register struct mbuf *m;
218 	tcp_seq ack, seq;
219 	int flags;
220 {
221 	register int tlen;
222 	int win = 0;
223 	struct route *ro = 0;
224 	struct route sro;
225 
226 	if (tp) {
227 		if (!(flags & TH_RST))
228 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
229 		ro = &tp->t_inpcb->inp_route;
230 	} else {
231 		ro = &sro;
232 		bzero(ro, sizeof *ro);
233 	}
234 	if (m == 0) {
235 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
236 		if (m == NULL)
237 			return;
238 #ifdef TCP_COMPAT_42
239 		tlen = 1;
240 #else
241 		tlen = 0;
242 #endif
243 		m->m_data += max_linkhdr;
244 		*mtod(m, struct tcpiphdr *) = *ti;
245 		ti = mtod(m, struct tcpiphdr *);
246 		flags = TH_ACK;
247 	} else {
248 		m_freem(m->m_next);
249 		m->m_next = 0;
250 		m->m_data = (caddr_t)ti;
251 		m->m_len = sizeof (struct tcpiphdr);
252 		tlen = 0;
253 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
254 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long);
255 		xchg(ti->ti_dport, ti->ti_sport, n_short);
256 #undef xchg
257 	}
258 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
259 	tlen += sizeof (struct tcpiphdr);
260 	m->m_len = tlen;
261 	m->m_pkthdr.len = tlen;
262 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
263 	bzero(ti->ti_x1, sizeof(ti->ti_x1));
264 	ti->ti_seq = htonl(seq);
265 	ti->ti_ack = htonl(ack);
266 	ti->ti_x2 = 0;
267 	ti->ti_off = sizeof (struct tcphdr) >> 2;
268 	ti->ti_flags = flags;
269 	if (tp)
270 		ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
271 	else
272 		ti->ti_win = htons((u_short)win);
273 	ti->ti_urp = 0;
274 	ti->ti_sum = 0;
275 	ti->ti_sum = in_cksum(m, tlen);
276 	((struct ip *)ti)->ip_len = tlen;
277 	((struct ip *)ti)->ip_ttl = ip_defttl;
278 #ifdef TCPDEBUG
279 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
280 		tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
281 #endif
282 	(void) ip_output(m, NULL, ro, 0, NULL);
283 	if (ro == &sro && ro->ro_rt) {
284 		RTFREE(ro->ro_rt);
285 	}
286 }
287 
288 /*
289  * Create a new TCP control block, making an
290  * empty reassembly queue and hooking it to the argument
291  * protocol control block.  The `inp' parameter must have
292  * come from the zone allocator set up in tcp_init().
293  */
294 struct tcpcb *
295 tcp_newtcpcb(inp)
296 	struct inpcb *inp;
297 {
298 	struct inp_tp *it;
299 	register struct tcpcb *tp;
300 
301 	it = (struct inp_tp *)inp;
302 	tp = &it->tcb;
303 	bzero((char *) tp, sizeof(struct tcpcb));
304 	tp->t_segq = NULL;
305 	tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
306 
307 	if (tcp_do_rfc1323)
308 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
309 	if (tcp_do_rfc1644)
310 		tp->t_flags |= TF_REQ_CC;
311 	tp->t_inpcb = inp;	/* XXX */
312 	/*
313 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
314 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
315 	 * reasonable initial retransmit time.
316 	 */
317 	tp->t_srtt = TCPTV_SRTTBASE;
318 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
319 	tp->t_rttmin = TCPTV_MIN;
320 	tp->t_rxtcur = TCPTV_RTOBASE;
321 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
322 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
323 	inp->inp_ip_ttl = ip_defttl;
324 	inp->inp_ppcb = (caddr_t)tp;
325 	return (tp);		/* XXX */
326 }
327 
328 /*
329  * Drop a TCP connection, reporting
330  * the specified error.  If connection is synchronized,
331  * then send a RST to peer.
332  */
333 struct tcpcb *
334 tcp_drop(tp, errno)
335 	register struct tcpcb *tp;
336 	int errno;
337 {
338 	struct socket *so = tp->t_inpcb->inp_socket;
339 
340 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
341 		tp->t_state = TCPS_CLOSED;
342 		(void) tcp_output(tp);
343 		tcpstat.tcps_drops++;
344 	} else
345 		tcpstat.tcps_conndrops++;
346 	if (errno == ETIMEDOUT && tp->t_softerror)
347 		errno = tp->t_softerror;
348 	so->so_error = errno;
349 	return (tcp_close(tp));
350 }
351 
352 /*
353  * Close a TCP control block:
354  *	discard all space held by the tcp
355  *	discard internet protocol block
356  *	wake up any sleepers
357  */
358 struct tcpcb *
359 tcp_close(tp)
360 	register struct tcpcb *tp;
361 {
362 	register struct mbuf *q;
363 	register struct mbuf *nq;
364 	struct inpcb *inp = tp->t_inpcb;
365 	struct socket *so = inp->inp_socket;
366 	register struct rtentry *rt;
367 	int dosavessthresh;
368 
369 	/*
370 	 * If we got enough samples through the srtt filter,
371 	 * save the rtt and rttvar in the routing entry.
372 	 * 'Enough' is arbitrarily defined as the 16 samples.
373 	 * 16 samples is enough for the srtt filter to converge
374 	 * to within 5% of the correct value; fewer samples and
375 	 * we could save a very bogus rtt.
376 	 *
377 	 * Don't update the default route's characteristics and don't
378 	 * update anything that the user "locked".
379 	 */
380 	if (tp->t_rttupdated >= 16 &&
381 	    (rt = inp->inp_route.ro_rt) &&
382 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
383 		register u_long i = 0;
384 
385 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
386 			i = tp->t_srtt *
387 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
388 			if (rt->rt_rmx.rmx_rtt && i)
389 				/*
390 				 * filter this update to half the old & half
391 				 * the new values, converting scale.
392 				 * See route.h and tcp_var.h for a
393 				 * description of the scaling constants.
394 				 */
395 				rt->rt_rmx.rmx_rtt =
396 				    (rt->rt_rmx.rmx_rtt + i) / 2;
397 			else
398 				rt->rt_rmx.rmx_rtt = i;
399 			tcpstat.tcps_cachedrtt++;
400 		}
401 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
402 			i = tp->t_rttvar *
403 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
404 			if (rt->rt_rmx.rmx_rttvar && i)
405 				rt->rt_rmx.rmx_rttvar =
406 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
407 			else
408 				rt->rt_rmx.rmx_rttvar = i;
409 			tcpstat.tcps_cachedrttvar++;
410 		}
411 		/*
412 		 * The old comment here said:
413 		 * update the pipelimit (ssthresh) if it has been updated
414 		 * already or if a pipesize was specified & the threshhold
415 		 * got below half the pipesize.  I.e., wait for bad news
416 		 * before we start updating, then update on both good
417 		 * and bad news.
418 		 *
419 		 * But we want to save the ssthresh even if no pipesize is
420 		 * specified explicitly in the route, because such
421 		 * connections still have an implicit pipesize specified
422 		 * by the global tcp_sendspace.  In the absence of a reliable
423 		 * way to calculate the pipesize, it will have to do.
424 		 */
425 		i = tp->snd_ssthresh;
426 		if (rt->rt_rmx.rmx_sendpipe != 0)
427 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
428 		else
429 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
430 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
431 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
432 		    || dosavessthresh) {
433 			/*
434 			 * convert the limit from user data bytes to
435 			 * packets then to packet data bytes.
436 			 */
437 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
438 			if (i < 2)
439 				i = 2;
440 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
441 			if (rt->rt_rmx.rmx_ssthresh)
442 				rt->rt_rmx.rmx_ssthresh =
443 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
444 			else
445 				rt->rt_rmx.rmx_ssthresh = i;
446 			tcpstat.tcps_cachedssthresh++;
447 		}
448 	}
449 	/* free the reassembly queue, if any */
450 	for (q = tp->t_segq; q; q = nq) {
451 		nq = q->m_nextpkt;
452 		tp->t_segq = nq;
453 		m_freem(q);
454 	}
455 	if (tp->t_template)
456 		(void) m_free(dtom(tp->t_template));
457 	inp->inp_ppcb = NULL;
458 	soisdisconnected(so);
459 	in_pcbdetach(inp);
460 	tcpstat.tcps_closed++;
461 	return ((struct tcpcb *)0);
462 }
463 
464 void
465 tcp_drain()
466 {
467 
468 }
469 
470 /*
471  * Notify a tcp user of an asynchronous error;
472  * store error as soft error, but wake up user
473  * (for now, won't do anything until can select for soft error).
474  */
475 static void
476 tcp_notify(inp, error)
477 	struct inpcb *inp;
478 	int error;
479 {
480 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
481 	register struct socket *so = inp->inp_socket;
482 
483 	/*
484 	 * Ignore some errors if we are hooked up.
485 	 * If connection hasn't completed, has retransmitted several times,
486 	 * and receives a second error, give up now.  This is better
487 	 * than waiting a long time to establish a connection that
488 	 * can never complete.
489 	 */
490 	if (tp->t_state == TCPS_ESTABLISHED &&
491 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
492 	      error == EHOSTDOWN)) {
493 		return;
494 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
495 	    tp->t_softerror)
496 		so->so_error = error;
497 	else
498 		tp->t_softerror = error;
499 	wakeup((caddr_t) &so->so_timeo);
500 	sorwakeup(so);
501 	sowwakeup(so);
502 }
503 
504 static int
505 tcp_pcblist SYSCTL_HANDLER_ARGS
506 {
507 	int error, i, n, s;
508 	struct inpcb *inp, **inp_list;
509 	inp_gen_t gencnt;
510 	struct xinpgen xig;
511 
512 	/*
513 	 * The process of preparing the TCB list is too time-consuming and
514 	 * resource-intensive to repeat twice on every request.
515 	 */
516 	if (req->oldptr == 0) {
517 		n = tcbinfo.ipi_count;
518 		req->oldidx = 2 * (sizeof xig)
519 			+ (n + n/8) * sizeof(struct xtcpcb);
520 		return 0;
521 	}
522 
523 	if (req->newptr != 0)
524 		return EPERM;
525 
526 	/*
527 	 * OK, now we're committed to doing something.
528 	 */
529 	s = splnet();
530 	gencnt = tcbinfo.ipi_gencnt;
531 	n = tcbinfo.ipi_count;
532 	splx(s);
533 
534 	xig.xig_len = sizeof xig;
535 	xig.xig_count = n;
536 	xig.xig_gen = gencnt;
537 	xig.xig_sogen = so_gencnt;
538 	error = SYSCTL_OUT(req, &xig, sizeof xig);
539 	if (error)
540 		return error;
541 
542 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
543 	if (inp_list == 0)
544 		return ENOMEM;
545 
546 	s = splnet();
547 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
548 	     inp = inp->inp_list.le_next) {
549 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
550 			inp_list[i++] = inp;
551 	}
552 	splx(s);
553 	n = i;
554 
555 	error = 0;
556 	for (i = 0; i < n; i++) {
557 		inp = inp_list[i];
558 		if (inp->inp_gencnt <= gencnt) {
559 			struct xtcpcb xt;
560 			caddr_t inp_ppcb;
561 			xt.xt_len = sizeof xt;
562 			/* XXX should avoid extra copy */
563 			bcopy(inp, &xt.xt_inp, sizeof *inp);
564 			inp_ppcb = inp->inp_ppcb;
565 			if (inp_ppcb != NULL)
566 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
567 			else
568 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
569 			if (inp->inp_socket)
570 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
571 			error = SYSCTL_OUT(req, &xt, sizeof xt);
572 		}
573 	}
574 	if (!error) {
575 		/*
576 		 * Give the user an updated idea of our state.
577 		 * If the generation differs from what we told
578 		 * her before, she knows that something happened
579 		 * while we were processing this request, and it
580 		 * might be necessary to retry.
581 		 */
582 		s = splnet();
583 		xig.xig_gen = tcbinfo.ipi_gencnt;
584 		xig.xig_sogen = so_gencnt;
585 		xig.xig_count = tcbinfo.ipi_count;
586 		splx(s);
587 		error = SYSCTL_OUT(req, &xig, sizeof xig);
588 	}
589 	free(inp_list, M_TEMP);
590 	return error;
591 }
592 
593 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
594 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
595 
596 static int
597 tcp_getcred SYSCTL_HANDLER_ARGS
598 {
599 	struct sockaddr_in addrs[2];
600 	struct inpcb *inp;
601 	int error, s;
602 
603 	error = suser(req->p);
604 	if (error)
605 		return (error);
606 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
607 	if (error)
608 		return (error);
609 	s = splnet();
610 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
611 	    addrs[0].sin_addr, addrs[0].sin_port, 0);
612 	if (inp == NULL || inp->inp_socket == NULL ||
613 	    inp->inp_socket->so_cred == NULL) {
614 		error = ENOENT;
615 		goto out;
616 	}
617 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred->pc_ucred,
618 	    sizeof(struct ucred));
619 out:
620 	splx(s);
621 	return (error);
622 }
623 
624 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
625     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
626 
627 void
628 tcp_ctlinput(cmd, sa, vip)
629 	int cmd;
630 	struct sockaddr *sa;
631 	void *vip;
632 {
633 	register struct ip *ip = vip;
634 	register struct tcphdr *th;
635 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
636 
637 	if (cmd == PRC_QUENCH)
638 		notify = tcp_quench;
639 	else if (cmd == PRC_MSGSIZE)
640 		notify = tcp_mtudisc;
641 	else if (!PRC_IS_REDIRECT(cmd) &&
642 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
643 		return;
644 	if (ip) {
645 		th = (struct tcphdr *)((caddr_t)ip
646 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
647 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
648 			cmd, notify);
649 	} else
650 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
651 }
652 
653 /*
654  * When a source quench is received, close congestion window
655  * to one segment.  We will gradually open it again as we proceed.
656  */
657 void
658 tcp_quench(inp, errno)
659 	struct inpcb *inp;
660 	int errno;
661 {
662 	struct tcpcb *tp = intotcpcb(inp);
663 
664 	if (tp)
665 		tp->snd_cwnd = tp->t_maxseg;
666 }
667 
668 /*
669  * When `need fragmentation' ICMP is received, update our idea of the MSS
670  * based on the new value in the route.  Also nudge TCP to send something,
671  * since we know the packet we just sent was dropped.
672  * This duplicates some code in the tcp_mss() function in tcp_input.c.
673  */
674 void
675 tcp_mtudisc(inp, errno)
676 	struct inpcb *inp;
677 	int errno;
678 {
679 	struct tcpcb *tp = intotcpcb(inp);
680 	struct rtentry *rt;
681 	struct rmxp_tao *taop;
682 	struct socket *so = inp->inp_socket;
683 	int offered;
684 	int mss;
685 
686 	if (tp) {
687 		rt = tcp_rtlookup(inp);
688 		if (!rt || !rt->rt_rmx.rmx_mtu) {
689 			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
690 			return;
691 		}
692 		taop = rmx_taop(rt->rt_rmx);
693 		offered = taop->tao_mssopt;
694 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
695 		if (offered)
696 			mss = min(mss, offered);
697 		/*
698 		 * XXX - The above conditional probably violates the TCP
699 		 * spec.  The problem is that, since we don't know the
700 		 * other end's MSS, we are supposed to use a conservative
701 		 * default.  But, if we do that, then MTU discovery will
702 		 * never actually take place, because the conservative
703 		 * default is much less than the MTUs typically seen
704 		 * on the Internet today.  For the moment, we'll sweep
705 		 * this under the carpet.
706 		 *
707 		 * The conservative default might not actually be a problem
708 		 * if the only case this occurs is when sending an initial
709 		 * SYN with options and data to a host we've never talked
710 		 * to before.  Then, they will reply with an MSS value which
711 		 * will get recorded and the new parameters should get
712 		 * recomputed.  For Further Study.
713 		 */
714 		if (tp->t_maxopd <= mss)
715 			return;
716 		tp->t_maxopd = mss;
717 
718 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
719 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
720 			mss -= TCPOLEN_TSTAMP_APPA;
721 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
722 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
723 			mss -= TCPOLEN_CC_APPA;
724 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
725 		if (mss > MCLBYTES)
726 			mss &= ~(MCLBYTES-1);
727 #else
728 		if (mss > MCLBYTES)
729 			mss = mss / MCLBYTES * MCLBYTES;
730 #endif
731 		if (so->so_snd.sb_hiwat < mss)
732 			mss = so->so_snd.sb_hiwat;
733 
734 		tp->t_maxseg = mss;
735 
736 		tcpstat.tcps_mturesent++;
737 		tp->t_rtt = 0;
738 		tp->snd_nxt = tp->snd_una;
739 		tcp_output(tp);
740 	}
741 }
742 
743 /*
744  * Look-up the routing entry to the peer of this inpcb.  If no route
745  * is found and it cannot be allocated the return NULL.  This routine
746  * is called by TCP routines that access the rmx structure and by tcp_mss
747  * to get the interface MTU.
748  */
749 struct rtentry *
750 tcp_rtlookup(inp)
751 	struct inpcb *inp;
752 {
753 	struct route *ro;
754 	struct rtentry *rt;
755 
756 	ro = &inp->inp_route;
757 	rt = ro->ro_rt;
758 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
759 		/* No route yet, so try to acquire one */
760 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
761 			ro->ro_dst.sa_family = AF_INET;
762 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
763 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
764 				inp->inp_faddr;
765 			rtalloc(ro);
766 			rt = ro->ro_rt;
767 		}
768 	}
769 	return rt;
770 }
771 
772 /*
773  * Return a pointer to the cached information about the remote host.
774  * The cached information is stored in the protocol specific part of
775  * the route metrics.
776  */
777 struct rmxp_tao *
778 tcp_gettaocache(inp)
779 	struct inpcb *inp;
780 {
781 	struct rtentry *rt = tcp_rtlookup(inp);
782 
783 	/* Make sure this is a host route and is up. */
784 	if (rt == NULL ||
785 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
786 		return NULL;
787 
788 	return rmx_taop(rt->rt_rmx);
789 }
790 
791 /*
792  * Clear all the TAO cache entries, called from tcp_init.
793  *
794  * XXX
795  * This routine is just an empty one, because we assume that the routing
796  * routing tables are initialized at the same time when TCP, so there is
797  * nothing in the cache left over.
798  */
799 static void
800 tcp_cleartaocache()
801 {
802 }
803