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 1.22 1995/11/14 20:34:41 phk Exp $
35 */
36
37#include <sys/param.h>
38#include <sys/proc.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/sysctl.h>
42#include <sys/malloc.h>
43#include <sys/mbuf.h>
44#include <sys/socket.h>
45#include <sys/socketvar.h>
46#include <sys/protosw.h>
47#include <sys/errno.h>
48#include <sys/queue.h>
49
50#include <net/route.h>
51#include <net/if.h>
52
53#include <netinet/in.h>
54#include <netinet/in_systm.h>
55#include <netinet/ip.h>
56#include <netinet/in_pcb.h>
57#include <netinet/in_var.h>
58#include <netinet/ip_var.h>
59#include <netinet/ip_icmp.h>
60#include <netinet/tcp.h>
61#include <netinet/tcp_fsm.h>
62#include <netinet/tcp_seq.h>
63#include <netinet/tcp_timer.h>
64#include <netinet/tcp_var.h>
65#include <netinet/tcpip.h>
66#ifdef TCPDEBUG
67#include <netinet/tcp_debug.h>
68#endif
69
70int tcp_mssdflt = TCP_MSS;
71SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
72 CTLFLAG_RW, &tcp_mssdflt , 0, "");
73
74static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
75SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt,
76 CTLFLAG_RW, &tcp_rttdflt , 0, "");
77
78static int tcp_do_rfc1323 = 1;
79SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
80 CTLFLAG_RW, &tcp_do_rfc1323 , 0, "");
81
82static int tcp_do_rfc1644 = 1;
83SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644,
84 CTLFLAG_RW, &tcp_do_rfc1644 , 0, "");
85
86static void tcp_cleartaocache(void);
87static void tcp_notify __P((struct inpcb *, int));
88
89/*
90 * Target size of TCP PCB hash table. Will be rounded down to a prime
91 * number.
92 */
93#ifndef TCBHASHSIZE
94#define TCBHASHSIZE 128
95#endif
96
97/*
98 * Tcp initialization
99 */
100void
101tcp_init()
102{
103
104 tcp_iss = random(); /* wrong, but better than a constant */
105 tcp_ccgen = 1;
106 tcp_cleartaocache();
107 LIST_INIT(&tcb);
108 tcbinfo.listhead = &tcb;
109 tcbinfo.hashbase = phashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashsize);
110 if (max_protohdr < sizeof(struct tcpiphdr))
111 max_protohdr = sizeof(struct tcpiphdr);
112 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
113 panic("tcp_init");
114}
115
116/*
117 * Create template to be used to send tcp packets on a connection.
118 * Call after host entry created, allocates an mbuf and fills
119 * in a skeletal tcp/ip header, minimizing the amount of work
120 * necessary when the connection is used.
121 */
122struct tcpiphdr *
123tcp_template(tp)
124 struct tcpcb *tp;
125{
126 register struct inpcb *inp = tp->t_inpcb;
127 register struct mbuf *m;
128 register struct tcpiphdr *n;
129
130 if ((n = tp->t_template) == 0) {
131 m = m_get(M_DONTWAIT, MT_HEADER);
132 if (m == NULL)
133 return (0);
134 m->m_len = sizeof (struct tcpiphdr);
135 n = mtod(m, struct tcpiphdr *);
136 }
137 n->ti_next = n->ti_prev = 0;
138 n->ti_x1 = 0;
139 n->ti_pr = IPPROTO_TCP;
140 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
141 n->ti_src = inp->inp_laddr;
142 n->ti_dst = inp->inp_faddr;
143 n->ti_sport = inp->inp_lport;
144 n->ti_dport = inp->inp_fport;
145 n->ti_seq = 0;
146 n->ti_ack = 0;
147 n->ti_x2 = 0;
148 n->ti_off = 5;
149 n->ti_flags = 0;
150 n->ti_win = 0;
151 n->ti_sum = 0;
152 n->ti_urp = 0;
153 return (n);
154}
155
156/*
157 * Send a single message to the TCP at address specified by
158 * the given TCP/IP header. If m == 0, then we make a copy
159 * of the tcpiphdr at ti and send directly to the addressed host.
160 * This is used to force keep alive messages out using the TCP
161 * template for a connection tp->t_template. If flags are given
162 * then we send a message back to the TCP which originated the
163 * segment ti, and discard the mbuf containing it and any other
164 * attached mbufs.
165 *
166 * In any case the ack and sequence number of the transmitted
167 * segment are as specified by the parameters.
168 */
169void
170tcp_respond(tp, ti, m, ack, seq, flags)
171 struct tcpcb *tp;
172 register struct tcpiphdr *ti;
173 register struct mbuf *m;
174 tcp_seq ack, seq;
175 int flags;
176{
177 register int tlen;
178 int win = 0;
179 struct route *ro = 0;
180
181 if (tp) {
182 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
183 ro = &tp->t_inpcb->inp_route;
184 }
185 if (m == 0) {
186 m = m_gethdr(M_DONTWAIT, MT_HEADER);
187 if (m == NULL)
188 return;
189#ifdef TCP_COMPAT_42
190 tlen = 1;
191#else
192 tlen = 0;
193#endif
194 m->m_data += max_linkhdr;
195 *mtod(m, struct tcpiphdr *) = *ti;
196 ti = mtod(m, struct tcpiphdr *);
197 flags = TH_ACK;
198 } else {
199 m_freem(m->m_next);
200 m->m_next = 0;
201 m->m_data = (caddr_t)ti;
202 m->m_len = sizeof (struct tcpiphdr);
203 tlen = 0;
204#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
205 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
206 xchg(ti->ti_dport, ti->ti_sport, u_short);
207#undef xchg
208 }
209 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
210 tlen += sizeof (struct tcpiphdr);
211 m->m_len = tlen;
212 m->m_pkthdr.len = tlen;
213 m->m_pkthdr.rcvif = (struct ifnet *) 0;
214 ti->ti_next = ti->ti_prev = 0;
215 ti->ti_x1 = 0;
216 ti->ti_seq = htonl(seq);
217 ti->ti_ack = htonl(ack);
218 ti->ti_x2 = 0;
219 ti->ti_off = sizeof (struct tcphdr) >> 2;
220 ti->ti_flags = flags;
221 if (tp)
222 ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
223 else
224 ti->ti_win = htons((u_short)win);
225 ti->ti_urp = 0;
226 ti->ti_sum = 0;
227 ti->ti_sum = in_cksum(m, tlen);
228 ((struct ip *)ti)->ip_len = tlen;
229 ((struct ip *)ti)->ip_ttl = ip_defttl;
230#ifdef TCPDEBUG
231 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
232 tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
233#endif
234 (void) ip_output(m, NULL, ro, 0, NULL);
235}
236
237/*
238 * Create a new TCP control block, making an
239 * empty reassembly queue and hooking it to the argument
240 * protocol control block.
241 */
242struct tcpcb *
243tcp_newtcpcb(inp)
244 struct inpcb *inp;
245{
246 register struct tcpcb *tp;
247
248 tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
249 if (tp == NULL)
250 return ((struct tcpcb *)0);
251 bzero((char *) tp, sizeof(struct tcpcb));
252 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
253 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
254
255 if (tcp_do_rfc1323)
256 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
257 if (tcp_do_rfc1644)
258 tp->t_flags |= TF_REQ_CC;
259 tp->t_inpcb = inp;
260 /*
261 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
262 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
263 * reasonable initial retransmit time.
264 */
265 tp->t_srtt = TCPTV_SRTTBASE;
266 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
267 tp->t_rttmin = TCPTV_MIN;
268 TCPT_RANGESET(tp->t_rxtcur,
269 ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
270 TCPTV_MIN, TCPTV_REXMTMAX);
271 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
272 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
273 inp->inp_ip.ip_ttl = ip_defttl;
274 inp->inp_ppcb = (caddr_t)tp;
275 return (tp);
276}
277
278/*
279 * Drop a TCP connection, reporting
280 * the specified error. If connection is synchronized,
281 * then send a RST to peer.
282 */
283struct tcpcb *
284tcp_drop(tp, errno)
285 register struct tcpcb *tp;
286 int errno;
287{
288 struct socket *so = tp->t_inpcb->inp_socket;
289
290 if (TCPS_HAVERCVDSYN(tp->t_state)) {
291 tp->t_state = TCPS_CLOSED;
292 (void) tcp_output(tp);
293 tcpstat.tcps_drops++;
294 } else
295 tcpstat.tcps_conndrops++;
296 if (errno == ETIMEDOUT && tp->t_softerror)
297 errno = tp->t_softerror;
298 so->so_error = errno;
299 return (tcp_close(tp));
300}
301
302/*
303 * Close a TCP control block:
304 * discard all space held by the tcp
305 * discard internet protocol block
306 * wake up any sleepers
307 */
308struct tcpcb *
309tcp_close(tp)
310 register struct tcpcb *tp;
311{
312 register struct tcpiphdr *t;
313 struct inpcb *inp = tp->t_inpcb;
314 struct socket *so = inp->inp_socket;
315 register struct mbuf *m;
316#ifdef RTV_RTT
317 register struct rtentry *rt;
318
319 /*
320 * If we got enough samples through the srtt filter,
321 * save the rtt and rttvar in the routing entry.
322 * 'Enough' is arbitrarily defined as the 16 samples.
323 * 16 samples is enough for the srtt filter to converge
324 * to within 5% of the correct value; fewer samples and
325 * we could save a very bogus rtt.
326 *
327 * Don't update the default route's characteristics and don't
328 * update anything that the user "locked".
329 */
330 if (tp->t_rttupdated >= 16 &&
331 (rt = inp->inp_route.ro_rt) &&
332 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
333 register u_long i = 0;
334
335 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
336 i = tp->t_srtt *
337 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
338 if (rt->rt_rmx.rmx_rtt && i)
339 /*
340 * filter this update to half the old & half
341 * the new values, converting scale.
342 * See route.h and tcp_var.h for a
343 * description of the scaling constants.
344 */
345 rt->rt_rmx.rmx_rtt =
346 (rt->rt_rmx.rmx_rtt + i) / 2;
347 else
348 rt->rt_rmx.rmx_rtt = i;
349 tcpstat.tcps_cachedrtt++;
350 }
351 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
352 i = tp->t_rttvar *
353 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
354 if (rt->rt_rmx.rmx_rttvar && i)
355 rt->rt_rmx.rmx_rttvar =
356 (rt->rt_rmx.rmx_rttvar + i) / 2;
357 else
358 rt->rt_rmx.rmx_rttvar = i;
359 tcpstat.tcps_cachedrttvar++;
360 }
361 /*
362 * update the pipelimit (ssthresh) if it has been updated
363 * already or if a pipesize was specified & the threshhold
364 * got below half the pipesize. I.e., wait for bad news
365 * before we start updating, then update on both good
366 * and bad news.
367 */
368 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
369 ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) ||
370 i < (rt->rt_rmx.rmx_sendpipe / 2)) {
371 /*
372 * convert the limit from user data bytes to
373 * packets then to packet data bytes.
374 */
375 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
376 if (i < 2)
377 i = 2;
378 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
379 if (rt->rt_rmx.rmx_ssthresh)
380 rt->rt_rmx.rmx_ssthresh =
381 (rt->rt_rmx.rmx_ssthresh + i) / 2;
382 else
383 rt->rt_rmx.rmx_ssthresh = i;
384 tcpstat.tcps_cachedssthresh++;
385 }
386 }
387#endif /* RTV_RTT */
388 /* free the reassembly queue, if any */
389 t = tp->seg_next;
390 while (t != (struct tcpiphdr *)tp) {
391 t = (struct tcpiphdr *)t->ti_next;
392 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
393 remque(t->ti_prev);
394 m_freem(m);
395 }
396 if (tp->t_template)
397 (void) m_free(dtom(tp->t_template));
398 free(tp, M_PCB);
399 inp->inp_ppcb = 0;
400 soisdisconnected(so);
401 in_pcbdetach(inp);
402 tcpstat.tcps_closed++;
403 return ((struct tcpcb *)0);
404}
405
406void
407tcp_drain()
408{
409
410}
411
412/*
413 * Notify a tcp user of an asynchronous error;
414 * store error as soft error, but wake up user
415 * (for now, won't do anything until can select for soft error).
416 */
417static void
418tcp_notify(inp, error)
419 struct inpcb *inp;
420 int error;
421{
422 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
423 register struct socket *so = inp->inp_socket;
424
425 /*
426 * Ignore some errors if we are hooked up.
427 * If connection hasn't completed, has retransmitted several times,
428 * and receives a second error, give up now. This is better
429 * than waiting a long time to establish a connection that
430 * can never complete.
431 */
432 if (tp->t_state == TCPS_ESTABLISHED &&
433 (error == EHOSTUNREACH || error == ENETUNREACH ||
434 error == EHOSTDOWN)) {
435 return;
436 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
437 tp->t_softerror)
438 so->so_error = error;
439 else
440 tp->t_softerror = error;
441 wakeup((caddr_t) &so->so_timeo);
442 sorwakeup(so);
443 sowwakeup(so);
444}
445
446void
447tcp_ctlinput(cmd, sa, ip)
448 int cmd;
449 struct sockaddr *sa;
450 register struct ip *ip;
451{
452 register struct tcphdr *th;
453 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
454
455 if (cmd == PRC_QUENCH)
456 notify = tcp_quench;
457#if 1
458 else if (cmd == PRC_MSGSIZE)
459 notify = tcp_mtudisc;
460#endif
461 else if (!PRC_IS_REDIRECT(cmd) &&
462 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
463 return;
464 if (ip) {
465 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
466 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
467 cmd, notify);
468 } else
469 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
470}
471
472/*
473 * When a source quench is received, close congestion window
474 * to one segment. We will gradually open it again as we proceed.
475 */
476void
477tcp_quench(inp, errno)
478 struct inpcb *inp;
479 int errno;
480{
481 struct tcpcb *tp = intotcpcb(inp);
482
483 if (tp)
484 tp->snd_cwnd = tp->t_maxseg;
485}
486
487#if 1
488/*
489 * When `need fragmentation' ICMP is received, update our idea of the MSS
490 * based on the new value in the route. Also nudge TCP to send something,
491 * since we know the packet we just sent was dropped.
492 * This duplicates some code in the tcp_mss() function in tcp_input.c.
493 */
494void
495tcp_mtudisc(inp, errno)
496 struct inpcb *inp;
497 int errno;
498{
499 struct tcpcb *tp = intotcpcb(inp);
500 struct rtentry *rt;
501 struct rmxp_tao *taop;
502 struct socket *so = inp->inp_socket;
503 int offered;
504 int mss;
505
506 if (tp) {
507 rt = tcp_rtlookup(inp);
508 if (!rt || !rt->rt_rmx.rmx_mtu) {
509 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
510 return;
511 }
512 taop = rmx_taop(rt->rt_rmx);
513 offered = taop->tao_mssopt;
514 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
515 mss = min(mss, offered);
516 if (tp->t_maxopd <= mss)
517 return;
518 tp->t_maxopd = mss;
519
520 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
521 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
522 mss -= TCPOLEN_TSTAMP_APPA;
523 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
524 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
525 mss -= TCPOLEN_CC_APPA;
526#if (MCLBYTES & (MCLBYTES - 1)) == 0
527 if (mss > MCLBYTES)
528 mss &= ~(MCLBYTES-1);
529#else
530 if (mss > MCLBYTES)
531 mss = mss / MCLBYTES * MCLBYTES;
532#endif
533 if (so->so_snd.sb_hiwat < mss)
534 mss = so->so_snd.sb_hiwat;
535
536 tp->t_maxseg = mss;
537
538 tcpstat.tcps_mturesent++;
539 tp->t_rtt = 0;
540 tp->snd_nxt = tp->snd_una;
541 tcp_output(tp);
542 }
543}
544#endif
545
546/*
547 * Look-up the routing entry to the peer of this inpcb. If no route
548 * is found and it cannot be allocated the return NULL. This routine
549 * is called by TCP routines that access the rmx structure and by tcp_mss
550 * to get the interface MTU.
551 */
552struct rtentry *
553tcp_rtlookup(inp)
554 struct inpcb *inp;
555{
556 struct route *ro;
557 struct rtentry *rt;
558
559 ro = &inp->inp_route;
560 rt = ro->ro_rt;
561 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
562 /* No route yet, so try to acquire one */
563 if (inp->inp_faddr.s_addr != INADDR_ANY) {
564 ro->ro_dst.sa_family = AF_INET;
565 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
566 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
567 inp->inp_faddr;
568 rtalloc(ro);
569 rt = ro->ro_rt;
570 }
571 }
572 return rt;
573}
574
575/*
576 * Return a pointer to the cached information about the remote host.
577 * The cached information is stored in the protocol specific part of
578 * the route metrics.
579 */
580struct rmxp_tao *
581tcp_gettaocache(inp)
582 struct inpcb *inp;
583{
584 struct rtentry *rt = tcp_rtlookup(inp);
585
586 /* Make sure this is a host route and is up. */
587 if (rt == NULL ||
588 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
589 return NULL;
590
591 return rmx_taop(rt->rt_rmx);
592}
593
594/*
595 * Clear all the TAO cache entries, called from tcp_init.
596 *
597 * XXX
598 * This routine is just an empty one, because we assume that the routing
599 * routing tables are initialized at the same time when TCP, so there is
600 * nothing in the cache left over.
601 */
602static void
603tcp_cleartaocache(void)
604{ }
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 1.22 1995/11/14 20:34:41 phk Exp $
35 */
36
37#include <sys/param.h>
38#include <sys/proc.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/sysctl.h>
42#include <sys/malloc.h>
43#include <sys/mbuf.h>
44#include <sys/socket.h>
45#include <sys/socketvar.h>
46#include <sys/protosw.h>
47#include <sys/errno.h>
48#include <sys/queue.h>
49
50#include <net/route.h>
51#include <net/if.h>
52
53#include <netinet/in.h>
54#include <netinet/in_systm.h>
55#include <netinet/ip.h>
56#include <netinet/in_pcb.h>
57#include <netinet/in_var.h>
58#include <netinet/ip_var.h>
59#include <netinet/ip_icmp.h>
60#include <netinet/tcp.h>
61#include <netinet/tcp_fsm.h>
62#include <netinet/tcp_seq.h>
63#include <netinet/tcp_timer.h>
64#include <netinet/tcp_var.h>
65#include <netinet/tcpip.h>
66#ifdef TCPDEBUG
67#include <netinet/tcp_debug.h>
68#endif
69
70int tcp_mssdflt = TCP_MSS;
71SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
72 CTLFLAG_RW, &tcp_mssdflt , 0, "");
73
74static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
75SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt,
76 CTLFLAG_RW, &tcp_rttdflt , 0, "");
77
78static int tcp_do_rfc1323 = 1;
79SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
80 CTLFLAG_RW, &tcp_do_rfc1323 , 0, "");
81
82static int tcp_do_rfc1644 = 1;
83SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644,
84 CTLFLAG_RW, &tcp_do_rfc1644 , 0, "");
85
86static void tcp_cleartaocache(void);
87static void tcp_notify __P((struct inpcb *, int));
88
89/*
90 * Target size of TCP PCB hash table. Will be rounded down to a prime
91 * number.
92 */
93#ifndef TCBHASHSIZE
94#define TCBHASHSIZE 128
95#endif
96
97/*
98 * Tcp initialization
99 */
100void
101tcp_init()
102{
103
104 tcp_iss = random(); /* wrong, but better than a constant */
105 tcp_ccgen = 1;
106 tcp_cleartaocache();
107 LIST_INIT(&tcb);
108 tcbinfo.listhead = &tcb;
109 tcbinfo.hashbase = phashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashsize);
110 if (max_protohdr < sizeof(struct tcpiphdr))
111 max_protohdr = sizeof(struct tcpiphdr);
112 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
113 panic("tcp_init");
114}
115
116/*
117 * Create template to be used to send tcp packets on a connection.
118 * Call after host entry created, allocates an mbuf and fills
119 * in a skeletal tcp/ip header, minimizing the amount of work
120 * necessary when the connection is used.
121 */
122struct tcpiphdr *
123tcp_template(tp)
124 struct tcpcb *tp;
125{
126 register struct inpcb *inp = tp->t_inpcb;
127 register struct mbuf *m;
128 register struct tcpiphdr *n;
129
130 if ((n = tp->t_template) == 0) {
131 m = m_get(M_DONTWAIT, MT_HEADER);
132 if (m == NULL)
133 return (0);
134 m->m_len = sizeof (struct tcpiphdr);
135 n = mtod(m, struct tcpiphdr *);
136 }
137 n->ti_next = n->ti_prev = 0;
138 n->ti_x1 = 0;
139 n->ti_pr = IPPROTO_TCP;
140 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
141 n->ti_src = inp->inp_laddr;
142 n->ti_dst = inp->inp_faddr;
143 n->ti_sport = inp->inp_lport;
144 n->ti_dport = inp->inp_fport;
145 n->ti_seq = 0;
146 n->ti_ack = 0;
147 n->ti_x2 = 0;
148 n->ti_off = 5;
149 n->ti_flags = 0;
150 n->ti_win = 0;
151 n->ti_sum = 0;
152 n->ti_urp = 0;
153 return (n);
154}
155
156/*
157 * Send a single message to the TCP at address specified by
158 * the given TCP/IP header. If m == 0, then we make a copy
159 * of the tcpiphdr at ti and send directly to the addressed host.
160 * This is used to force keep alive messages out using the TCP
161 * template for a connection tp->t_template. If flags are given
162 * then we send a message back to the TCP which originated the
163 * segment ti, and discard the mbuf containing it and any other
164 * attached mbufs.
165 *
166 * In any case the ack and sequence number of the transmitted
167 * segment are as specified by the parameters.
168 */
169void
170tcp_respond(tp, ti, m, ack, seq, flags)
171 struct tcpcb *tp;
172 register struct tcpiphdr *ti;
173 register struct mbuf *m;
174 tcp_seq ack, seq;
175 int flags;
176{
177 register int tlen;
178 int win = 0;
179 struct route *ro = 0;
180
181 if (tp) {
182 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
183 ro = &tp->t_inpcb->inp_route;
184 }
185 if (m == 0) {
186 m = m_gethdr(M_DONTWAIT, MT_HEADER);
187 if (m == NULL)
188 return;
189#ifdef TCP_COMPAT_42
190 tlen = 1;
191#else
192 tlen = 0;
193#endif
194 m->m_data += max_linkhdr;
195 *mtod(m, struct tcpiphdr *) = *ti;
196 ti = mtod(m, struct tcpiphdr *);
197 flags = TH_ACK;
198 } else {
199 m_freem(m->m_next);
200 m->m_next = 0;
201 m->m_data = (caddr_t)ti;
202 m->m_len = sizeof (struct tcpiphdr);
203 tlen = 0;
204#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
205 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
206 xchg(ti->ti_dport, ti->ti_sport, u_short);
207#undef xchg
208 }
209 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
210 tlen += sizeof (struct tcpiphdr);
211 m->m_len = tlen;
212 m->m_pkthdr.len = tlen;
213 m->m_pkthdr.rcvif = (struct ifnet *) 0;
214 ti->ti_next = ti->ti_prev = 0;
215 ti->ti_x1 = 0;
216 ti->ti_seq = htonl(seq);
217 ti->ti_ack = htonl(ack);
218 ti->ti_x2 = 0;
219 ti->ti_off = sizeof (struct tcphdr) >> 2;
220 ti->ti_flags = flags;
221 if (tp)
222 ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
223 else
224 ti->ti_win = htons((u_short)win);
225 ti->ti_urp = 0;
226 ti->ti_sum = 0;
227 ti->ti_sum = in_cksum(m, tlen);
228 ((struct ip *)ti)->ip_len = tlen;
229 ((struct ip *)ti)->ip_ttl = ip_defttl;
230#ifdef TCPDEBUG
231 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
232 tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
233#endif
234 (void) ip_output(m, NULL, ro, 0, NULL);
235}
236
237/*
238 * Create a new TCP control block, making an
239 * empty reassembly queue and hooking it to the argument
240 * protocol control block.
241 */
242struct tcpcb *
243tcp_newtcpcb(inp)
244 struct inpcb *inp;
245{
246 register struct tcpcb *tp;
247
248 tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
249 if (tp == NULL)
250 return ((struct tcpcb *)0);
251 bzero((char *) tp, sizeof(struct tcpcb));
252 tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
253 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
254
255 if (tcp_do_rfc1323)
256 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
257 if (tcp_do_rfc1644)
258 tp->t_flags |= TF_REQ_CC;
259 tp->t_inpcb = inp;
260 /*
261 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
262 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
263 * reasonable initial retransmit time.
264 */
265 tp->t_srtt = TCPTV_SRTTBASE;
266 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
267 tp->t_rttmin = TCPTV_MIN;
268 TCPT_RANGESET(tp->t_rxtcur,
269 ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
270 TCPTV_MIN, TCPTV_REXMTMAX);
271 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
272 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
273 inp->inp_ip.ip_ttl = ip_defttl;
274 inp->inp_ppcb = (caddr_t)tp;
275 return (tp);
276}
277
278/*
279 * Drop a TCP connection, reporting
280 * the specified error. If connection is synchronized,
281 * then send a RST to peer.
282 */
283struct tcpcb *
284tcp_drop(tp, errno)
285 register struct tcpcb *tp;
286 int errno;
287{
288 struct socket *so = tp->t_inpcb->inp_socket;
289
290 if (TCPS_HAVERCVDSYN(tp->t_state)) {
291 tp->t_state = TCPS_CLOSED;
292 (void) tcp_output(tp);
293 tcpstat.tcps_drops++;
294 } else
295 tcpstat.tcps_conndrops++;
296 if (errno == ETIMEDOUT && tp->t_softerror)
297 errno = tp->t_softerror;
298 so->so_error = errno;
299 return (tcp_close(tp));
300}
301
302/*
303 * Close a TCP control block:
304 * discard all space held by the tcp
305 * discard internet protocol block
306 * wake up any sleepers
307 */
308struct tcpcb *
309tcp_close(tp)
310 register struct tcpcb *tp;
311{
312 register struct tcpiphdr *t;
313 struct inpcb *inp = tp->t_inpcb;
314 struct socket *so = inp->inp_socket;
315 register struct mbuf *m;
316#ifdef RTV_RTT
317 register struct rtentry *rt;
318
319 /*
320 * If we got enough samples through the srtt filter,
321 * save the rtt and rttvar in the routing entry.
322 * 'Enough' is arbitrarily defined as the 16 samples.
323 * 16 samples is enough for the srtt filter to converge
324 * to within 5% of the correct value; fewer samples and
325 * we could save a very bogus rtt.
326 *
327 * Don't update the default route's characteristics and don't
328 * update anything that the user "locked".
329 */
330 if (tp->t_rttupdated >= 16 &&
331 (rt = inp->inp_route.ro_rt) &&
332 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
333 register u_long i = 0;
334
335 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
336 i = tp->t_srtt *
337 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
338 if (rt->rt_rmx.rmx_rtt && i)
339 /*
340 * filter this update to half the old & half
341 * the new values, converting scale.
342 * See route.h and tcp_var.h for a
343 * description of the scaling constants.
344 */
345 rt->rt_rmx.rmx_rtt =
346 (rt->rt_rmx.rmx_rtt + i) / 2;
347 else
348 rt->rt_rmx.rmx_rtt = i;
349 tcpstat.tcps_cachedrtt++;
350 }
351 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
352 i = tp->t_rttvar *
353 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
354 if (rt->rt_rmx.rmx_rttvar && i)
355 rt->rt_rmx.rmx_rttvar =
356 (rt->rt_rmx.rmx_rttvar + i) / 2;
357 else
358 rt->rt_rmx.rmx_rttvar = i;
359 tcpstat.tcps_cachedrttvar++;
360 }
361 /*
362 * update the pipelimit (ssthresh) if it has been updated
363 * already or if a pipesize was specified & the threshhold
364 * got below half the pipesize. I.e., wait for bad news
365 * before we start updating, then update on both good
366 * and bad news.
367 */
368 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
369 ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) ||
370 i < (rt->rt_rmx.rmx_sendpipe / 2)) {
371 /*
372 * convert the limit from user data bytes to
373 * packets then to packet data bytes.
374 */
375 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
376 if (i < 2)
377 i = 2;
378 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
379 if (rt->rt_rmx.rmx_ssthresh)
380 rt->rt_rmx.rmx_ssthresh =
381 (rt->rt_rmx.rmx_ssthresh + i) / 2;
382 else
383 rt->rt_rmx.rmx_ssthresh = i;
384 tcpstat.tcps_cachedssthresh++;
385 }
386 }
387#endif /* RTV_RTT */
388 /* free the reassembly queue, if any */
389 t = tp->seg_next;
390 while (t != (struct tcpiphdr *)tp) {
391 t = (struct tcpiphdr *)t->ti_next;
392 m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
393 remque(t->ti_prev);
394 m_freem(m);
395 }
396 if (tp->t_template)
397 (void) m_free(dtom(tp->t_template));
398 free(tp, M_PCB);
399 inp->inp_ppcb = 0;
400 soisdisconnected(so);
401 in_pcbdetach(inp);
402 tcpstat.tcps_closed++;
403 return ((struct tcpcb *)0);
404}
405
406void
407tcp_drain()
408{
409
410}
411
412/*
413 * Notify a tcp user of an asynchronous error;
414 * store error as soft error, but wake up user
415 * (for now, won't do anything until can select for soft error).
416 */
417static void
418tcp_notify(inp, error)
419 struct inpcb *inp;
420 int error;
421{
422 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
423 register struct socket *so = inp->inp_socket;
424
425 /*
426 * Ignore some errors if we are hooked up.
427 * If connection hasn't completed, has retransmitted several times,
428 * and receives a second error, give up now. This is better
429 * than waiting a long time to establish a connection that
430 * can never complete.
431 */
432 if (tp->t_state == TCPS_ESTABLISHED &&
433 (error == EHOSTUNREACH || error == ENETUNREACH ||
434 error == EHOSTDOWN)) {
435 return;
436 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
437 tp->t_softerror)
438 so->so_error = error;
439 else
440 tp->t_softerror = error;
441 wakeup((caddr_t) &so->so_timeo);
442 sorwakeup(so);
443 sowwakeup(so);
444}
445
446void
447tcp_ctlinput(cmd, sa, ip)
448 int cmd;
449 struct sockaddr *sa;
450 register struct ip *ip;
451{
452 register struct tcphdr *th;
453 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
454
455 if (cmd == PRC_QUENCH)
456 notify = tcp_quench;
457#if 1
458 else if (cmd == PRC_MSGSIZE)
459 notify = tcp_mtudisc;
460#endif
461 else if (!PRC_IS_REDIRECT(cmd) &&
462 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
463 return;
464 if (ip) {
465 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
466 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
467 cmd, notify);
468 } else
469 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
470}
471
472/*
473 * When a source quench is received, close congestion window
474 * to one segment. We will gradually open it again as we proceed.
475 */
476void
477tcp_quench(inp, errno)
478 struct inpcb *inp;
479 int errno;
480{
481 struct tcpcb *tp = intotcpcb(inp);
482
483 if (tp)
484 tp->snd_cwnd = tp->t_maxseg;
485}
486
487#if 1
488/*
489 * When `need fragmentation' ICMP is received, update our idea of the MSS
490 * based on the new value in the route. Also nudge TCP to send something,
491 * since we know the packet we just sent was dropped.
492 * This duplicates some code in the tcp_mss() function in tcp_input.c.
493 */
494void
495tcp_mtudisc(inp, errno)
496 struct inpcb *inp;
497 int errno;
498{
499 struct tcpcb *tp = intotcpcb(inp);
500 struct rtentry *rt;
501 struct rmxp_tao *taop;
502 struct socket *so = inp->inp_socket;
503 int offered;
504 int mss;
505
506 if (tp) {
507 rt = tcp_rtlookup(inp);
508 if (!rt || !rt->rt_rmx.rmx_mtu) {
509 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
510 return;
511 }
512 taop = rmx_taop(rt->rt_rmx);
513 offered = taop->tao_mssopt;
514 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
515 mss = min(mss, offered);
516 if (tp->t_maxopd <= mss)
517 return;
518 tp->t_maxopd = mss;
519
520 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
521 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
522 mss -= TCPOLEN_TSTAMP_APPA;
523 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
524 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
525 mss -= TCPOLEN_CC_APPA;
526#if (MCLBYTES & (MCLBYTES - 1)) == 0
527 if (mss > MCLBYTES)
528 mss &= ~(MCLBYTES-1);
529#else
530 if (mss > MCLBYTES)
531 mss = mss / MCLBYTES * MCLBYTES;
532#endif
533 if (so->so_snd.sb_hiwat < mss)
534 mss = so->so_snd.sb_hiwat;
535
536 tp->t_maxseg = mss;
537
538 tcpstat.tcps_mturesent++;
539 tp->t_rtt = 0;
540 tp->snd_nxt = tp->snd_una;
541 tcp_output(tp);
542 }
543}
544#endif
545
546/*
547 * Look-up the routing entry to the peer of this inpcb. If no route
548 * is found and it cannot be allocated the return NULL. This routine
549 * is called by TCP routines that access the rmx structure and by tcp_mss
550 * to get the interface MTU.
551 */
552struct rtentry *
553tcp_rtlookup(inp)
554 struct inpcb *inp;
555{
556 struct route *ro;
557 struct rtentry *rt;
558
559 ro = &inp->inp_route;
560 rt = ro->ro_rt;
561 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
562 /* No route yet, so try to acquire one */
563 if (inp->inp_faddr.s_addr != INADDR_ANY) {
564 ro->ro_dst.sa_family = AF_INET;
565 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
566 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
567 inp->inp_faddr;
568 rtalloc(ro);
569 rt = ro->ro_rt;
570 }
571 }
572 return rt;
573}
574
575/*
576 * Return a pointer to the cached information about the remote host.
577 * The cached information is stored in the protocol specific part of
578 * the route metrics.
579 */
580struct rmxp_tao *
581tcp_gettaocache(inp)
582 struct inpcb *inp;
583{
584 struct rtentry *rt = tcp_rtlookup(inp);
585
586 /* Make sure this is a host route and is up. */
587 if (rt == NULL ||
588 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
589 return NULL;
590
591 return rmx_taop(rt->rt_rmx);
592}
593
594/*
595 * Clear all the TAO cache entries, called from tcp_init.
596 *
597 * XXX
598 * This routine is just an empty one, because we assume that the routing
599 * routing tables are initialized at the same time when TCP, so there is
600 * nothing in the cache left over.
601 */
602static void
603tcp_cleartaocache(void)
604{ }