xref: /freebsd/sys/netinet/tcp_subr.c (revision a14a0223ae1b172e96dd2a1d849e22026a98b692)
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_tcpdebug.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 
52 #include <vm/vm_zone.h>
53 
54 #include <net/route.h>
55 #include <net/if.h>
56 
57 #define _IP_VHL
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip_var.h>
64 #include <netinet/tcp.h>
65 #include <netinet/tcp_fsm.h>
66 #include <netinet/tcp_seq.h>
67 #include <netinet/tcp_timer.h>
68 #include <netinet/tcp_var.h>
69 #include <netinet/tcpip.h>
70 #ifdef TCPDEBUG
71 #include <netinet/tcp_debug.h>
72 #endif
73 
74 int 	tcp_mssdflt = TCP_MSS;
75 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
76     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
77 
78 #ifdef INET6
79 int	tcp_v6mssdflt = TCP6_MSS;
80 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
81 	CTLFLAG_RW, &tcp_v6mssdflt , 0, "");
82 #endif
83 
84 #if 0
85 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
86 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
87     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
88 #endif
89 
90 static int	tcp_do_rfc1323 = 1;
91 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
92     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
93 
94 static int	tcp_do_rfc1644 = 0;
95 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
96     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
97 
98 static int	tcp_tcbhashsize = 0;
99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
100      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
101 
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
103     &tcbinfo.ipi_count, 0, "Number of active PCBs");
104 
105 static void	tcp_cleartaocache __P((void));
106 static void	tcp_notify __P((struct inpcb *, int));
107 
108 /*
109  * Target size of TCP PCB hash tables. Must be a power of two.
110  *
111  * Note that this can be overridden by the kernel environment
112  * variable net.inet.tcp.tcbhashsize
113  */
114 #ifndef TCBHASHSIZE
115 #define TCBHASHSIZE	512
116 #endif
117 
118 /*
119  * This is the actual shape of what we allocate using the zone
120  * allocator.  Doing it this way allows us to protect both structures
121  * using the same generation count, and also eliminates the overhead
122  * of allocating tcpcbs separately.  By hiding the structure here,
123  * we avoid changing most of the rest of the code (although it needs
124  * to be changed, eventually, for greater efficiency).
125  */
126 #define	ALIGNMENT	32
127 #define	ALIGNM1		(ALIGNMENT - 1)
128 struct	inp_tp {
129 	union {
130 		struct	inpcb inp;
131 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
132 	} inp_tp_u;
133 	struct	tcpcb tcb;
134 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
135 	struct	callout inp_tp_delack;
136 };
137 #undef ALIGNMENT
138 #undef ALIGNM1
139 
140 /*
141  * Tcp initialization
142  */
143 void
144 tcp_init()
145 {
146 	int hashsize;
147 
148 	tcp_iss = random();	/* wrong, but better than a constant */
149 	tcp_ccgen = 1;
150 	tcp_cleartaocache();
151 
152 	tcp_delacktime = TCPTV_DELACK;
153 	tcp_keepinit = TCPTV_KEEP_INIT;
154 	tcp_keepidle = TCPTV_KEEP_IDLE;
155 	tcp_keepintvl = TCPTV_KEEPINTVL;
156 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
157 	tcp_msl = TCPTV_MSL;
158 
159 	LIST_INIT(&tcb);
160 	tcbinfo.listhead = &tcb;
161 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
162 	if (!powerof2(hashsize)) {
163 		printf("WARNING: TCB hash size not a power of 2\n");
164 		hashsize = 512; /* safe default */
165 	}
166 	tcp_tcbhashsize = hashsize;
167 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
168 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
169 					&tcbinfo.porthashmask);
170 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
171 				 ZONE_INTERRUPT, 0);
172 
173 	if (max_protohdr < sizeof(struct tcpiphdr))
174 		max_protohdr = sizeof(struct tcpiphdr);
175 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
176 		panic("tcp_init");
177 }
178 
179 /*
180  * Create template to be used to send tcp packets on a connection.
181  * Call after host entry created, allocates an mbuf and fills
182  * in a skeletal tcp/ip header, minimizing the amount of work
183  * necessary when the connection is used.
184  */
185 struct tcpiphdr *
186 tcp_template(tp)
187 	struct tcpcb *tp;
188 {
189 	register struct inpcb *inp = tp->t_inpcb;
190 	register struct mbuf *m;
191 	register struct tcpiphdr *n;
192 
193 	if ((n = tp->t_template) == 0) {
194 		m = m_get(M_DONTWAIT, MT_HEADER);
195 		if (m == NULL)
196 			return (0);
197 		m->m_len = sizeof (struct tcpiphdr);
198 		n = mtod(m, struct tcpiphdr *);
199 	}
200 	bzero(n->ti_x1, sizeof(n->ti_x1));
201 	n->ti_pr = IPPROTO_TCP;
202 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
203 	n->ti_src = inp->inp_laddr;
204 	n->ti_dst = inp->inp_faddr;
205 	n->ti_sport = inp->inp_lport;
206 	n->ti_dport = inp->inp_fport;
207 	n->ti_seq = 0;
208 	n->ti_ack = 0;
209 	n->ti_x2 = 0;
210 	n->ti_off = 5;
211 	n->ti_flags = 0;
212 	n->ti_win = 0;
213 	n->ti_sum = 0;
214 	n->ti_urp = 0;
215 	return (n);
216 }
217 
218 /*
219  * Send a single message to the TCP at address specified by
220  * the given TCP/IP header.  If m == 0, then we make a copy
221  * of the tcpiphdr at ti and send directly to the addressed host.
222  * This is used to force keep alive messages out using the TCP
223  * template for a connection tp->t_template.  If flags are given
224  * then we send a message back to the TCP which originated the
225  * segment ti, and discard the mbuf containing it and any other
226  * attached mbufs.
227  *
228  * In any case the ack and sequence number of the transmitted
229  * segment are as specified by the parameters.
230  *
231  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
232  */
233 void
234 tcp_respond(tp, ti, m, ack, seq, flags)
235 	struct tcpcb *tp;
236 	register struct tcpiphdr *ti;
237 	register struct mbuf *m;
238 	tcp_seq ack, seq;
239 	int flags;
240 {
241 	register int tlen;
242 	int win = 0;
243 	struct route *ro = 0;
244 	struct route sro;
245 
246 	if (tp) {
247 		if (!(flags & TH_RST))
248 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
249 		ro = &tp->t_inpcb->inp_route;
250 	} else {
251 		ro = &sro;
252 		bzero(ro, sizeof *ro);
253 	}
254 	if (m == 0) {
255 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
256 		if (m == NULL)
257 			return;
258 #ifdef TCP_COMPAT_42
259 		tlen = 1;
260 #else
261 		tlen = 0;
262 #endif
263 		m->m_data += max_linkhdr;
264 		*mtod(m, struct tcpiphdr *) = *ti;
265 		ti = mtod(m, struct tcpiphdr *);
266 		flags = TH_ACK;
267 	} else {
268 		m_freem(m->m_next);
269 		m->m_next = 0;
270 		m->m_data = (caddr_t)ti;
271 		m->m_len = sizeof (struct tcpiphdr);
272 		tlen = 0;
273 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
274 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long);
275 		xchg(ti->ti_dport, ti->ti_sport, n_short);
276 #undef xchg
277 	}
278 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
279 	tlen += sizeof (struct tcpiphdr);
280 	m->m_len = tlen;
281 	m->m_pkthdr.len = tlen;
282 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
283 	bzero(ti->ti_x1, sizeof(ti->ti_x1));
284 	ti->ti_seq = htonl(seq);
285 	ti->ti_ack = htonl(ack);
286 	ti->ti_x2 = 0;
287 	ti->ti_off = sizeof (struct tcphdr) >> 2;
288 	ti->ti_flags = flags;
289 	if (tp)
290 		ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
291 	else
292 		ti->ti_win = htons((u_short)win);
293 	ti->ti_urp = 0;
294 	ti->ti_sum = 0;
295 	ti->ti_sum = in_cksum(m, tlen);
296 	((struct ip *)ti)->ip_len = tlen;
297 	((struct ip *)ti)->ip_ttl = ip_defttl;
298 #ifdef TCPDEBUG
299 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
300 		tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
301 #endif
302 	(void) ip_output(m, NULL, ro, 0, NULL);
303 	if (ro == &sro && ro->ro_rt) {
304 		RTFREE(ro->ro_rt);
305 	}
306 }
307 
308 /*
309  * Create a new TCP control block, making an
310  * empty reassembly queue and hooking it to the argument
311  * protocol control block.  The `inp' parameter must have
312  * come from the zone allocator set up in tcp_init().
313  */
314 struct tcpcb *
315 tcp_newtcpcb(inp)
316 	struct inpcb *inp;
317 {
318 	struct inp_tp *it;
319 	register struct tcpcb *tp;
320 
321 	it = (struct inp_tp *)inp;
322 	tp = &it->tcb;
323 	bzero((char *) tp, sizeof(struct tcpcb));
324 	tp->t_segq = NULL;
325 	tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
326 
327 	/* Set up our timeouts. */
328 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
329 	callout_init(tp->tt_persist = &it->inp_tp_persist);
330 	callout_init(tp->tt_keep = &it->inp_tp_keep);
331 	callout_init(tp->tt_2msl = &it->inp_tp_2msl);
332 	callout_init(tp->tt_delack = &it->inp_tp_delack);
333 
334 	if (tcp_do_rfc1323)
335 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
336 	if (tcp_do_rfc1644)
337 		tp->t_flags |= TF_REQ_CC;
338 	tp->t_inpcb = inp;	/* XXX */
339 	/*
340 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
341 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
342 	 * reasonable initial retransmit time.
343 	 */
344 	tp->t_srtt = TCPTV_SRTTBASE;
345 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
346 	tp->t_rttmin = TCPTV_MIN;
347 	tp->t_rxtcur = TCPTV_RTOBASE;
348 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
349 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
350 	tp->t_rcvtime = ticks;
351 	inp->inp_ip_ttl = ip_defttl;
352 	inp->inp_ppcb = (caddr_t)tp;
353 	return (tp);		/* XXX */
354 }
355 
356 /*
357  * Drop a TCP connection, reporting
358  * the specified error.  If connection is synchronized,
359  * then send a RST to peer.
360  */
361 struct tcpcb *
362 tcp_drop(tp, errno)
363 	register struct tcpcb *tp;
364 	int errno;
365 {
366 	struct socket *so = tp->t_inpcb->inp_socket;
367 
368 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
369 		tp->t_state = TCPS_CLOSED;
370 		(void) tcp_output(tp);
371 		tcpstat.tcps_drops++;
372 	} else
373 		tcpstat.tcps_conndrops++;
374 	if (errno == ETIMEDOUT && tp->t_softerror)
375 		errno = tp->t_softerror;
376 	so->so_error = errno;
377 	return (tcp_close(tp));
378 }
379 
380 /*
381  * Close a TCP control block:
382  *	discard all space held by the tcp
383  *	discard internet protocol block
384  *	wake up any sleepers
385  */
386 struct tcpcb *
387 tcp_close(tp)
388 	register struct tcpcb *tp;
389 {
390 	register struct mbuf *q;
391 	register struct mbuf *nq;
392 	struct inpcb *inp = tp->t_inpcb;
393 	struct socket *so = inp->inp_socket;
394 	register struct rtentry *rt;
395 	int dosavessthresh;
396 
397 	/*
398 	 * Make sure that all of our timers are stopped before we
399 	 * delete the PCB.
400 	 */
401 	callout_stop(tp->tt_rexmt);
402 	callout_stop(tp->tt_persist);
403 	callout_stop(tp->tt_keep);
404 	callout_stop(tp->tt_2msl);
405 	callout_stop(tp->tt_delack);
406 
407 	/*
408 	 * If we got enough samples through the srtt filter,
409 	 * save the rtt and rttvar in the routing entry.
410 	 * 'Enough' is arbitrarily defined as the 16 samples.
411 	 * 16 samples is enough for the srtt filter to converge
412 	 * to within 5% of the correct value; fewer samples and
413 	 * we could save a very bogus rtt.
414 	 *
415 	 * Don't update the default route's characteristics and don't
416 	 * update anything that the user "locked".
417 	 */
418 	if (tp->t_rttupdated >= 16 &&
419 	    (rt = inp->inp_route.ro_rt) &&
420 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
421 		register u_long i = 0;
422 
423 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
424 			i = tp->t_srtt *
425 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
426 			if (rt->rt_rmx.rmx_rtt && i)
427 				/*
428 				 * filter this update to half the old & half
429 				 * the new values, converting scale.
430 				 * See route.h and tcp_var.h for a
431 				 * description of the scaling constants.
432 				 */
433 				rt->rt_rmx.rmx_rtt =
434 				    (rt->rt_rmx.rmx_rtt + i) / 2;
435 			else
436 				rt->rt_rmx.rmx_rtt = i;
437 			tcpstat.tcps_cachedrtt++;
438 		}
439 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
440 			i = tp->t_rttvar *
441 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
442 			if (rt->rt_rmx.rmx_rttvar && i)
443 				rt->rt_rmx.rmx_rttvar =
444 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
445 			else
446 				rt->rt_rmx.rmx_rttvar = i;
447 			tcpstat.tcps_cachedrttvar++;
448 		}
449 		/*
450 		 * The old comment here said:
451 		 * update the pipelimit (ssthresh) if it has been updated
452 		 * already or if a pipesize was specified & the threshhold
453 		 * got below half the pipesize.  I.e., wait for bad news
454 		 * before we start updating, then update on both good
455 		 * and bad news.
456 		 *
457 		 * But we want to save the ssthresh even if no pipesize is
458 		 * specified explicitly in the route, because such
459 		 * connections still have an implicit pipesize specified
460 		 * by the global tcp_sendspace.  In the absence of a reliable
461 		 * way to calculate the pipesize, it will have to do.
462 		 */
463 		i = tp->snd_ssthresh;
464 		if (rt->rt_rmx.rmx_sendpipe != 0)
465 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
466 		else
467 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
468 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
469 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
470 		    || dosavessthresh) {
471 			/*
472 			 * convert the limit from user data bytes to
473 			 * packets then to packet data bytes.
474 			 */
475 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
476 			if (i < 2)
477 				i = 2;
478 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
479 			if (rt->rt_rmx.rmx_ssthresh)
480 				rt->rt_rmx.rmx_ssthresh =
481 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
482 			else
483 				rt->rt_rmx.rmx_ssthresh = i;
484 			tcpstat.tcps_cachedssthresh++;
485 		}
486 	}
487 	/* free the reassembly queue, if any */
488 	for (q = tp->t_segq; q; q = nq) {
489 		nq = q->m_nextpkt;
490 		tp->t_segq = nq;
491 		m_freem(q);
492 	}
493 	if (tp->t_template)
494 		(void) m_free(dtom(tp->t_template));
495 	inp->inp_ppcb = NULL;
496 	soisdisconnected(so);
497 	in_pcbdetach(inp);
498 	tcpstat.tcps_closed++;
499 	return ((struct tcpcb *)0);
500 }
501 
502 void
503 tcp_drain()
504 {
505 
506 }
507 
508 /*
509  * Notify a tcp user of an asynchronous error;
510  * store error as soft error, but wake up user
511  * (for now, won't do anything until can select for soft error).
512  */
513 static void
514 tcp_notify(inp, error)
515 	struct inpcb *inp;
516 	int error;
517 {
518 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
519 	register struct socket *so = inp->inp_socket;
520 
521 	/*
522 	 * Ignore some errors if we are hooked up.
523 	 * If connection hasn't completed, has retransmitted several times,
524 	 * and receives a second error, give up now.  This is better
525 	 * than waiting a long time to establish a connection that
526 	 * can never complete.
527 	 */
528 	if (tp->t_state == TCPS_ESTABLISHED &&
529 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
530 	      error == EHOSTDOWN)) {
531 		return;
532 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
533 	    tp->t_softerror)
534 		so->so_error = error;
535 	else
536 		tp->t_softerror = error;
537 	wakeup((caddr_t) &so->so_timeo);
538 	sorwakeup(so);
539 	sowwakeup(so);
540 }
541 
542 static int
543 tcp_pcblist SYSCTL_HANDLER_ARGS
544 {
545 	int error, i, n, s;
546 	struct inpcb *inp, **inp_list;
547 	inp_gen_t gencnt;
548 	struct xinpgen xig;
549 
550 	/*
551 	 * The process of preparing the TCB list is too time-consuming and
552 	 * resource-intensive to repeat twice on every request.
553 	 */
554 	if (req->oldptr == 0) {
555 		n = tcbinfo.ipi_count;
556 		req->oldidx = 2 * (sizeof xig)
557 			+ (n + n/8) * sizeof(struct xtcpcb);
558 		return 0;
559 	}
560 
561 	if (req->newptr != 0)
562 		return EPERM;
563 
564 	/*
565 	 * OK, now we're committed to doing something.
566 	 */
567 	s = splnet();
568 	gencnt = tcbinfo.ipi_gencnt;
569 	n = tcbinfo.ipi_count;
570 	splx(s);
571 
572 	xig.xig_len = sizeof xig;
573 	xig.xig_count = n;
574 	xig.xig_gen = gencnt;
575 	xig.xig_sogen = so_gencnt;
576 	error = SYSCTL_OUT(req, &xig, sizeof xig);
577 	if (error)
578 		return error;
579 
580 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
581 	if (inp_list == 0)
582 		return ENOMEM;
583 
584 	s = splnet();
585 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
586 	     inp = inp->inp_list.le_next) {
587 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
588 			inp_list[i++] = inp;
589 	}
590 	splx(s);
591 	n = i;
592 
593 	error = 0;
594 	for (i = 0; i < n; i++) {
595 		inp = inp_list[i];
596 		if (inp->inp_gencnt <= gencnt) {
597 			struct xtcpcb xt;
598 			caddr_t inp_ppcb;
599 			xt.xt_len = sizeof xt;
600 			/* XXX should avoid extra copy */
601 			bcopy(inp, &xt.xt_inp, sizeof *inp);
602 			inp_ppcb = inp->inp_ppcb;
603 			if (inp_ppcb != NULL)
604 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
605 			else
606 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
607 			if (inp->inp_socket)
608 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
609 			error = SYSCTL_OUT(req, &xt, sizeof xt);
610 		}
611 	}
612 	if (!error) {
613 		/*
614 		 * Give the user an updated idea of our state.
615 		 * If the generation differs from what we told
616 		 * her before, she knows that something happened
617 		 * while we were processing this request, and it
618 		 * might be necessary to retry.
619 		 */
620 		s = splnet();
621 		xig.xig_gen = tcbinfo.ipi_gencnt;
622 		xig.xig_sogen = so_gencnt;
623 		xig.xig_count = tcbinfo.ipi_count;
624 		splx(s);
625 		error = SYSCTL_OUT(req, &xig, sizeof xig);
626 	}
627 	free(inp_list, M_TEMP);
628 	return error;
629 }
630 
631 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
632 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
633 
634 static int
635 tcp_getcred SYSCTL_HANDLER_ARGS
636 {
637 	struct sockaddr_in addrs[2];
638 	struct inpcb *inp;
639 	int error, s;
640 
641 	error = suser(req->p);
642 	if (error)
643 		return (error);
644 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
645 	if (error)
646 		return (error);
647 	s = splnet();
648 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
649 	    addrs[0].sin_addr, addrs[0].sin_port, 0);
650 	if (inp == NULL || inp->inp_socket == NULL) {
651 		error = ENOENT;
652 		goto out;
653 	}
654 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
655 out:
656 	splx(s);
657 	return (error);
658 }
659 
660 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
661     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
662 
663 void
664 tcp_ctlinput(cmd, sa, vip)
665 	int cmd;
666 	struct sockaddr *sa;
667 	void *vip;
668 {
669 	register struct ip *ip = vip;
670 	register struct tcphdr *th;
671 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
672 
673 	if (cmd == PRC_QUENCH)
674 		notify = tcp_quench;
675 	else if (cmd == PRC_MSGSIZE)
676 		notify = tcp_mtudisc;
677 	else if (!PRC_IS_REDIRECT(cmd) &&
678 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
679 		return;
680 	if (ip) {
681 		th = (struct tcphdr *)((caddr_t)ip
682 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
683 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
684 			cmd, notify);
685 	} else
686 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
687 }
688 
689 /*
690  * When a source quench is received, close congestion window
691  * to one segment.  We will gradually open it again as we proceed.
692  */
693 void
694 tcp_quench(inp, errno)
695 	struct inpcb *inp;
696 	int errno;
697 {
698 	struct tcpcb *tp = intotcpcb(inp);
699 
700 	if (tp)
701 		tp->snd_cwnd = tp->t_maxseg;
702 }
703 
704 /*
705  * When `need fragmentation' ICMP is received, update our idea of the MSS
706  * based on the new value in the route.  Also nudge TCP to send something,
707  * since we know the packet we just sent was dropped.
708  * This duplicates some code in the tcp_mss() function in tcp_input.c.
709  */
710 void
711 tcp_mtudisc(inp, errno)
712 	struct inpcb *inp;
713 	int errno;
714 {
715 	struct tcpcb *tp = intotcpcb(inp);
716 	struct rtentry *rt;
717 	struct rmxp_tao *taop;
718 	struct socket *so = inp->inp_socket;
719 	int offered;
720 	int mss;
721 
722 	if (tp) {
723 		rt = tcp_rtlookup(inp);
724 		if (!rt || !rt->rt_rmx.rmx_mtu) {
725 			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
726 			return;
727 		}
728 		taop = rmx_taop(rt->rt_rmx);
729 		offered = taop->tao_mssopt;
730 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
731 		if (offered)
732 			mss = min(mss, offered);
733 		/*
734 		 * XXX - The above conditional probably violates the TCP
735 		 * spec.  The problem is that, since we don't know the
736 		 * other end's MSS, we are supposed to use a conservative
737 		 * default.  But, if we do that, then MTU discovery will
738 		 * never actually take place, because the conservative
739 		 * default is much less than the MTUs typically seen
740 		 * on the Internet today.  For the moment, we'll sweep
741 		 * this under the carpet.
742 		 *
743 		 * The conservative default might not actually be a problem
744 		 * if the only case this occurs is when sending an initial
745 		 * SYN with options and data to a host we've never talked
746 		 * to before.  Then, they will reply with an MSS value which
747 		 * will get recorded and the new parameters should get
748 		 * recomputed.  For Further Study.
749 		 */
750 		if (tp->t_maxopd <= mss)
751 			return;
752 		tp->t_maxopd = mss;
753 
754 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
755 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
756 			mss -= TCPOLEN_TSTAMP_APPA;
757 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
758 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
759 			mss -= TCPOLEN_CC_APPA;
760 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
761 		if (mss > MCLBYTES)
762 			mss &= ~(MCLBYTES-1);
763 #else
764 		if (mss > MCLBYTES)
765 			mss = mss / MCLBYTES * MCLBYTES;
766 #endif
767 		if (so->so_snd.sb_hiwat < mss)
768 			mss = so->so_snd.sb_hiwat;
769 
770 		tp->t_maxseg = mss;
771 
772 		tcpstat.tcps_mturesent++;
773 		tp->t_rtttime = 0;
774 		tp->snd_nxt = tp->snd_una;
775 		tcp_output(tp);
776 	}
777 }
778 
779 /*
780  * Look-up the routing entry to the peer of this inpcb.  If no route
781  * is found and it cannot be allocated the return NULL.  This routine
782  * is called by TCP routines that access the rmx structure and by tcp_mss
783  * to get the interface MTU.
784  */
785 struct rtentry *
786 tcp_rtlookup(inp)
787 	struct inpcb *inp;
788 {
789 	struct route *ro;
790 	struct rtentry *rt;
791 
792 	ro = &inp->inp_route;
793 	rt = ro->ro_rt;
794 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
795 		/* No route yet, so try to acquire one */
796 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
797 			ro->ro_dst.sa_family = AF_INET;
798 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
799 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
800 				inp->inp_faddr;
801 			rtalloc(ro);
802 			rt = ro->ro_rt;
803 		}
804 	}
805 	return rt;
806 }
807 
808 /*
809  * Return a pointer to the cached information about the remote host.
810  * The cached information is stored in the protocol specific part of
811  * the route metrics.
812  */
813 struct rmxp_tao *
814 tcp_gettaocache(inp)
815 	struct inpcb *inp;
816 {
817 	struct rtentry *rt = tcp_rtlookup(inp);
818 
819 	/* Make sure this is a host route and is up. */
820 	if (rt == NULL ||
821 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
822 		return NULL;
823 
824 	return rmx_taop(rt->rt_rmx);
825 }
826 
827 /*
828  * Clear all the TAO cache entries, called from tcp_init.
829  *
830  * XXX
831  * This routine is just an empty one, because we assume that the routing
832  * routing tables are initialized at the same time when TCP, so there is
833  * nothing in the cache left over.
834  */
835 static void
836 tcp_cleartaocache()
837 {
838 }
839