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 1.52 1999/02/04 03:27:43 msmith 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/socket.h>
47#include <sys/socketvar.h>
48#include <sys/protosw.h>
49
50#include <vm/vm_zone.h>
51
52#include <net/route.h>
53#include <net/if.h>
54
55#define _IP_VHL
56#include <netinet/in.h>
57#include <netinet/in_systm.h>
58#include <netinet/ip.h>
59#include <netinet/in_pcb.h>
60#include <netinet/in_var.h>
61#include <netinet/ip_var.h>
62#include <netinet/tcp.h>
63#include <netinet/tcp_fsm.h>
64#include <netinet/tcp_seq.h>
65#include <netinet/tcp_timer.h>
66#include <netinet/tcp_var.h>
67#include <netinet/tcpip.h>
68#ifdef TCPDEBUG
69#include <netinet/tcp_debug.h>
70#endif
71
72int tcp_mssdflt = TCP_MSS;
73SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
74 CTLFLAG_RW, &tcp_mssdflt , 0, "");
75
76static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
77SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt,
78 CTLFLAG_RW, &tcp_rttdflt , 0, "");
79
80static int tcp_do_rfc1323 = 1;
81SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
82 CTLFLAG_RW, &tcp_do_rfc1323 , 0, "");
83
84static int tcp_do_rfc1644 = 0;
85SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644,
86 CTLFLAG_RW, &tcp_do_rfc1644 , 0, "");
87
88SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, &tcbinfo.ipi_count,
89 0, "Number of active PCBs");
90
91static void tcp_cleartaocache __P((void));
92static void tcp_notify __P((struct inpcb *, int));
93
94/*
95 * Target size of TCP PCB hash tables. Must be a power of two.
96 *
97 * Note that this can be overridden by the kernel environment
98 * variable net.inet.tcp.tcbhashsize
99 */
100#ifndef TCBHASHSIZE
101#define TCBHASHSIZE 512
102#endif
103
104/*
105 * This is the actual shape of what we allocate using the zone
106 * allocator. Doing it this way allows us to protect both structures
107 * using the same generation count, and also eliminates the overhead
108 * of allocating tcpcbs separately. By hiding the structure here,
109 * we avoid changing most of the rest of the code (although it needs
110 * to be changed, eventually, for greater efficiency).
111 */
112#define ALIGNMENT 32
113#define ALIGNM1 (ALIGNMENT - 1)
114struct inp_tp {
115 union {
116 struct inpcb inp;
117 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
118 } inp_tp_u;
119 struct tcpcb tcb;
120};
121#undef ALIGNMENT
122#undef ALIGNM1
123
124/*
125 * Tcp initialization
126 */
127void
128tcp_init()
129{
130 int hashsize;
131
132 tcp_iss = random(); /* wrong, but better than a constant */
133 tcp_ccgen = 1;
134 tcp_cleartaocache();
135 LIST_INIT(&tcb);
136 tcbinfo.listhead = &tcb;
137 if (!(getenv_int("net.inet.tcp.tcbhashsize", &hashsize)))
138 hashsize = TCBHASHSIZE;
139 if (!powerof2(hashsize)) {
140 printf("WARNING: TCB hash size not a power of 2\n");
141 hashsize = 512; /* safe default */
142 }
143 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
144 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
145 &tcbinfo.porthashmask);
146 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
147 ZONE_INTERRUPT, 0);
148 if (max_protohdr < sizeof(struct tcpiphdr))
149 max_protohdr = sizeof(struct tcpiphdr);
150 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
151 panic("tcp_init");
152}
153
154/*
155 * Create template to be used to send tcp packets on a connection.
156 * Call after host entry created, allocates an mbuf and fills
157 * in a skeletal tcp/ip header, minimizing the amount of work
158 * necessary when the connection is used.
159 */
160struct tcpiphdr *
161tcp_template(tp)
162 struct tcpcb *tp;
163{
164 register struct inpcb *inp = tp->t_inpcb;
165 register struct mbuf *m;
166 register struct tcpiphdr *n;
167
168 if ((n = tp->t_template) == 0) {
169 m = m_get(M_DONTWAIT, MT_HEADER);
170 if (m == NULL)
171 return (0);
172 m->m_len = sizeof (struct tcpiphdr);
173 n = mtod(m, struct tcpiphdr *);
174 }
175 bzero(n->ti_x1, sizeof(n->ti_x1));
176 n->ti_pr = IPPROTO_TCP;
177 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
178 n->ti_src = inp->inp_laddr;
179 n->ti_dst = inp->inp_faddr;
180 n->ti_sport = inp->inp_lport;
181 n->ti_dport = inp->inp_fport;
182 n->ti_seq = 0;
183 n->ti_ack = 0;
184 n->ti_x2 = 0;
185 n->ti_off = 5;
186 n->ti_flags = 0;
187 n->ti_win = 0;
188 n->ti_sum = 0;
189 n->ti_urp = 0;
190 return (n);
191}
192
193/*
194 * Send a single message to the TCP at address specified by
195 * the given TCP/IP header. If m == 0, then we make a copy
196 * of the tcpiphdr at ti and send directly to the addressed host.
197 * This is used to force keep alive messages out using the TCP
198 * template for a connection tp->t_template. If flags are given
199 * then we send a message back to the TCP which originated the
200 * segment ti, and discard the mbuf containing it and any other
201 * attached mbufs.
202 *
203 * In any case the ack and sequence number of the transmitted
204 * segment are as specified by the parameters.
205 *
206 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
207 */
208void
209tcp_respond(tp, ti, m, ack, seq, flags)
210 struct tcpcb *tp;
211 register struct tcpiphdr *ti;
212 register struct mbuf *m;
213 tcp_seq ack, seq;
214 int flags;
215{
216 register int tlen;
217 int win = 0;
218 struct route *ro = 0;
219 struct route sro;
220
221 if (tp) {
222 if (!(flags & TH_RST))
223 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
224 ro = &tp->t_inpcb->inp_route;
225 } else {
226 ro = &sro;
227 bzero(ro, sizeof *ro);
228 }
229 if (m == 0) {
230 m = m_gethdr(M_DONTWAIT, MT_HEADER);
231 if (m == NULL)
232 return;
233#ifdef TCP_COMPAT_42
234 tlen = 1;
235#else
236 tlen = 0;
237#endif
238 m->m_data += max_linkhdr;
239 *mtod(m, struct tcpiphdr *) = *ti;
240 ti = mtod(m, struct tcpiphdr *);
241 flags = TH_ACK;
242 } else {
243 m_freem(m->m_next);
244 m->m_next = 0;
245 m->m_data = (caddr_t)ti;
246 m->m_len = sizeof (struct tcpiphdr);
247 tlen = 0;
248#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
249 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long);
250 xchg(ti->ti_dport, ti->ti_sport, n_short);
251#undef xchg
252 }
253 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
254 tlen += sizeof (struct tcpiphdr);
255 m->m_len = tlen;
256 m->m_pkthdr.len = tlen;
257 m->m_pkthdr.rcvif = (struct ifnet *) 0;
258 bzero(ti->ti_x1, sizeof(ti->ti_x1));
259 ti->ti_seq = htonl(seq);
260 ti->ti_ack = htonl(ack);
261 ti->ti_x2 = 0;
262 ti->ti_off = sizeof (struct tcphdr) >> 2;
263 ti->ti_flags = flags;
264 if (tp)
265 ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
266 else
267 ti->ti_win = htons((u_short)win);
268 ti->ti_urp = 0;
269 ti->ti_sum = 0;
270 ti->ti_sum = in_cksum(m, tlen);
271 ((struct ip *)ti)->ip_len = tlen;
272 ((struct ip *)ti)->ip_ttl = ip_defttl;
273#ifdef TCPDEBUG
274 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
275 tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
276#endif
277 (void) ip_output(m, NULL, ro, 0, NULL);
278 if (ro == &sro && ro->ro_rt) {
279 RTFREE(ro->ro_rt);
280 }
281}
282
283/*
284 * Create a new TCP control block, making an
285 * empty reassembly queue and hooking it to the argument
286 * protocol control block. The `inp' parameter must have
287 * come from the zone allocator set up in tcp_init().
288 */
289struct tcpcb *
290tcp_newtcpcb(inp)
291 struct inpcb *inp;
292{
293 struct inp_tp *it;
294 register struct tcpcb *tp;
295
296 it = (struct inp_tp *)inp;
297 tp = &it->tcb;
298 bzero((char *) tp, sizeof(struct tcpcb));
299 tp->t_segq = NULL;
300 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
301
302 if (tcp_do_rfc1323)
303 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
304 if (tcp_do_rfc1644)
305 tp->t_flags |= TF_REQ_CC;
306 tp->t_inpcb = inp; /* XXX */
307 /*
308 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
309 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
310 * reasonable initial retransmit time.
311 */
312 tp->t_srtt = TCPTV_SRTTBASE;
313 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
314 tp->t_rttmin = TCPTV_MIN;
315 tp->t_rxtcur = TCPTV_RTOBASE;
316 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
317 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
318 inp->inp_ip_ttl = ip_defttl;
319 inp->inp_ppcb = (caddr_t)tp;
320 return (tp); /* XXX */
321}
322
323/*
324 * Drop a TCP connection, reporting
325 * the specified error. If connection is synchronized,
326 * then send a RST to peer.
327 */
328struct tcpcb *
329tcp_drop(tp, errno)
330 register struct tcpcb *tp;
331 int errno;
332{
333 struct socket *so = tp->t_inpcb->inp_socket;
334
335 if (TCPS_HAVERCVDSYN(tp->t_state)) {
336 tp->t_state = TCPS_CLOSED;
337 (void) tcp_output(tp);
338 tcpstat.tcps_drops++;
339 } else
340 tcpstat.tcps_conndrops++;
341 if (errno == ETIMEDOUT && tp->t_softerror)
342 errno = tp->t_softerror;
343 so->so_error = errno;
344 return (tcp_close(tp));
345}
346
347/*
348 * Close a TCP control block:
349 * discard all space held by the tcp
350 * discard internet protocol block
351 * wake up any sleepers
352 */
353struct tcpcb *
354tcp_close(tp)
355 register struct tcpcb *tp;
356{
357 register struct mbuf *q;
358 register struct mbuf *nq;
359 struct inpcb *inp = tp->t_inpcb;
360 struct socket *so = inp->inp_socket;
361 register struct rtentry *rt;
362 int dosavessthresh;
363
364 /*
365 * If we got enough samples through the srtt filter,
366 * save the rtt and rttvar in the routing entry.
367 * 'Enough' is arbitrarily defined as the 16 samples.
368 * 16 samples is enough for the srtt filter to converge
369 * to within 5% of the correct value; fewer samples and
370 * we could save a very bogus rtt.
371 *
372 * Don't update the default route's characteristics and don't
373 * update anything that the user "locked".
374 */
375 if (tp->t_rttupdated >= 16 &&
376 (rt = inp->inp_route.ro_rt) &&
377 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
378 register u_long i = 0;
379
380 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
381 i = tp->t_srtt *
382 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
383 if (rt->rt_rmx.rmx_rtt && i)
384 /*
385 * filter this update to half the old & half
386 * the new values, converting scale.
387 * See route.h and tcp_var.h for a
388 * description of the scaling constants.
389 */
390 rt->rt_rmx.rmx_rtt =
391 (rt->rt_rmx.rmx_rtt + i) / 2;
392 else
393 rt->rt_rmx.rmx_rtt = i;
394 tcpstat.tcps_cachedrtt++;
395 }
396 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
397 i = tp->t_rttvar *
398 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
399 if (rt->rt_rmx.rmx_rttvar && i)
400 rt->rt_rmx.rmx_rttvar =
401 (rt->rt_rmx.rmx_rttvar + i) / 2;
402 else
403 rt->rt_rmx.rmx_rttvar = i;
404 tcpstat.tcps_cachedrttvar++;
405 }
406 /*
407 * The old comment here said:
408 * update the pipelimit (ssthresh) if it has been updated
409 * already or if a pipesize was specified & the threshhold
410 * got below half the pipesize. I.e., wait for bad news
411 * before we start updating, then update on both good
412 * and bad news.
413 *
414 * But we want to save the ssthresh even if no pipesize is
415 * specified explicitly in the route, because such
416 * connections still have an implicit pipesize specified
417 * by the global tcp_sendspace. In the absence of a reliable
418 * way to calculate the pipesize, it will have to do.
419 */
420 i = tp->snd_ssthresh;
421 if (rt->rt_rmx.rmx_sendpipe != 0)
422 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
423 else
424 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
425 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
426 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
427 || dosavessthresh) {
428 /*
429 * convert the limit from user data bytes to
430 * packets then to packet data bytes.
431 */
432 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
433 if (i < 2)
434 i = 2;
435 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
436 if (rt->rt_rmx.rmx_ssthresh)
437 rt->rt_rmx.rmx_ssthresh =
438 (rt->rt_rmx.rmx_ssthresh + i) / 2;
439 else
440 rt->rt_rmx.rmx_ssthresh = i;
441 tcpstat.tcps_cachedssthresh++;
442 }
443 }
444 /* free the reassembly queue, if any */
445 for (q = tp->t_segq; q; q = nq) {
446 nq = q->m_nextpkt;
447 tp->t_segq = nq;
448 m_freem(q);
449 }
450 if (tp->t_template)
451 (void) m_free(dtom(tp->t_template));
452 inp->inp_ppcb = NULL;
453 soisdisconnected(so);
454 in_pcbdetach(inp);
455 tcpstat.tcps_closed++;
456 return ((struct tcpcb *)0);
457}
458
459void
460tcp_drain()
461{
462
463}
464
465/*
466 * Notify a tcp user of an asynchronous error;
467 * store error as soft error, but wake up user
468 * (for now, won't do anything until can select for soft error).
469 */
470static void
471tcp_notify(inp, error)
472 struct inpcb *inp;
473 int error;
474{
475 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
476 register struct socket *so = inp->inp_socket;
477
478 /*
479 * Ignore some errors if we are hooked up.
480 * If connection hasn't completed, has retransmitted several times,
481 * and receives a second error, give up now. This is better
482 * than waiting a long time to establish a connection that
483 * can never complete.
484 */
485 if (tp->t_state == TCPS_ESTABLISHED &&
486 (error == EHOSTUNREACH || error == ENETUNREACH ||
487 error == EHOSTDOWN)) {
488 return;
489 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
490 tp->t_softerror)
491 so->so_error = error;
492 else
493 tp->t_softerror = error;
494 wakeup((caddr_t) &so->so_timeo);
495 sorwakeup(so);
496 sowwakeup(so);
497}
498
499static int
500tcp_pcblist SYSCTL_HANDLER_ARGS
501{
502 int error, i, n, s;
503 struct inpcb *inp, **inp_list;
504 inp_gen_t gencnt;
505 struct xinpgen xig;
506
507 /*
508 * The process of preparing the TCB list is too time-consuming and
509 * resource-intensive to repeat twice on every request.
510 */
511 if (req->oldptr == 0) {
512 n = tcbinfo.ipi_count;
513 req->oldidx = 2 * (sizeof xig)
514 + (n + n/8) * sizeof(struct xtcpcb);
515 return 0;
516 }
517
518 if (req->newptr != 0)
519 return EPERM;
520
521 /*
522 * OK, now we're committed to doing something.
523 */
524 s = splnet();
525 gencnt = tcbinfo.ipi_gencnt;
526 n = tcbinfo.ipi_count;
527 splx(s);
528
529 xig.xig_len = sizeof xig;
530 xig.xig_count = n;
531 xig.xig_gen = gencnt;
532 xig.xig_sogen = so_gencnt;
533 error = SYSCTL_OUT(req, &xig, sizeof xig);
534 if (error)
535 return error;
536
537 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
538 if (inp_list == 0)
539 return ENOMEM;
540
541 s = splnet();
542 for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
543 inp = inp->inp_list.le_next) {
544 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
545 inp_list[i++] = inp;
546 }
547 splx(s);
548 n = i;
549
550 error = 0;
551 for (i = 0; i < n; i++) {
552 inp = inp_list[i];
553 if (inp->inp_gencnt <= gencnt) {
554 struct xtcpcb xt;
555 xt.xt_len = sizeof xt;
556 /* XXX should avoid extra copy */
557 bcopy(inp, &xt.xt_inp, sizeof *inp);
558 bcopy(inp->inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
559 if (inp->inp_socket)
560 sotoxsocket(inp->inp_socket, &xt.xt_socket);
561 error = SYSCTL_OUT(req, &xt, sizeof xt);
562 }
563 }
564 if (!error) {
565 /*
566 * Give the user an updated idea of our state.
567 * If the generation differs from what we told
568 * her before, she knows that something happened
569 * while we were processing this request, and it
570 * might be necessary to retry.
571 */
572 s = splnet();
573 xig.xig_gen = tcbinfo.ipi_gencnt;
574 xig.xig_sogen = so_gencnt;
575 xig.xig_count = tcbinfo.ipi_count;
576 splx(s);
577 error = SYSCTL_OUT(req, &xig, sizeof xig);
578 }
579 free(inp_list, M_TEMP);
580 return error;
581}
582
583SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
584 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
585
586void
587tcp_ctlinput(cmd, sa, vip)
588 int cmd;
589 struct sockaddr *sa;
590 void *vip;
591{
592 register struct ip *ip = vip;
593 register struct tcphdr *th;
594 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
595
596 if (cmd == PRC_QUENCH)
597 notify = tcp_quench;
598 else if (cmd == PRC_MSGSIZE)
599 notify = tcp_mtudisc;
600 else if (!PRC_IS_REDIRECT(cmd) &&
601 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
602 return;
603 if (ip) {
604 th = (struct tcphdr *)((caddr_t)ip
605 + (IP_VHL_HL(ip->ip_vhl) << 2));
606 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
607 cmd, notify);
608 } else
609 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
610}
611
612/*
613 * When a source quench is received, close congestion window
614 * to one segment. We will gradually open it again as we proceed.
615 */
616void
617tcp_quench(inp, errno)
618 struct inpcb *inp;
619 int errno;
620{
621 struct tcpcb *tp = intotcpcb(inp);
622
623 if (tp)
624 tp->snd_cwnd = tp->t_maxseg;
625}
626
627/*
628 * When `need fragmentation' ICMP is received, update our idea of the MSS
629 * based on the new value in the route. Also nudge TCP to send something,
630 * since we know the packet we just sent was dropped.
631 * This duplicates some code in the tcp_mss() function in tcp_input.c.
632 */
633void
634tcp_mtudisc(inp, errno)
635 struct inpcb *inp;
636 int errno;
637{
638 struct tcpcb *tp = intotcpcb(inp);
639 struct rtentry *rt;
640 struct rmxp_tao *taop;
641 struct socket *so = inp->inp_socket;
642 int offered;
643 int mss;
644
645 if (tp) {
646 rt = tcp_rtlookup(inp);
647 if (!rt || !rt->rt_rmx.rmx_mtu) {
648 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
649 return;
650 }
651 taop = rmx_taop(rt->rt_rmx);
652 offered = taop->tao_mssopt;
653 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
654 if (offered)
655 mss = min(mss, offered);
656 /*
657 * XXX - The above conditional probably violates the TCP
658 * spec. The problem is that, since we don't know the
659 * other end's MSS, we are supposed to use a conservative
660 * default. But, if we do that, then MTU discovery will
661 * never actually take place, because the conservative
662 * default is much less than the MTUs typically seen
663 * on the Internet today. For the moment, we'll sweep
664 * this under the carpet.
665 *
666 * The conservative default might not actually be a problem
667 * if the only case this occurs is when sending an initial
668 * SYN with options and data to a host we've never talked
669 * to before. Then, they will reply with an MSS value which
670 * will get recorded and the new parameters should get
671 * recomputed. For Further Study.
672 */
673 if (tp->t_maxopd <= mss)
674 return;
675 tp->t_maxopd = mss;
676
677 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
678 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
679 mss -= TCPOLEN_TSTAMP_APPA;
680 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
681 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
682 mss -= TCPOLEN_CC_APPA;
683#if (MCLBYTES & (MCLBYTES - 1)) == 0
684 if (mss > MCLBYTES)
685 mss &= ~(MCLBYTES-1);
686#else
687 if (mss > MCLBYTES)
688 mss = mss / MCLBYTES * MCLBYTES;
689#endif
690 if (so->so_snd.sb_hiwat < mss)
691 mss = so->so_snd.sb_hiwat;
692
693 tp->t_maxseg = mss;
694
695 tcpstat.tcps_mturesent++;
696 tp->t_rtt = 0;
697 tp->snd_nxt = tp->snd_una;
698 tcp_output(tp);
699 }
700}
701
702/*
703 * Look-up the routing entry to the peer of this inpcb. If no route
704 * is found and it cannot be allocated the return NULL. This routine
705 * is called by TCP routines that access the rmx structure and by tcp_mss
706 * to get the interface MTU.
707 */
708struct rtentry *
709tcp_rtlookup(inp)
710 struct inpcb *inp;
711{
712 struct route *ro;
713 struct rtentry *rt;
714
715 ro = &inp->inp_route;
716 rt = ro->ro_rt;
717 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
718 /* No route yet, so try to acquire one */
719 if (inp->inp_faddr.s_addr != INADDR_ANY) {
720 ro->ro_dst.sa_family = AF_INET;
721 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
722 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
723 inp->inp_faddr;
724 rtalloc(ro);
725 rt = ro->ro_rt;
726 }
727 }
728 return rt;
729}
730
731/*
732 * Return a pointer to the cached information about the remote host.
733 * The cached information is stored in the protocol specific part of
734 * the route metrics.
735 */
736struct rmxp_tao *
737tcp_gettaocache(inp)
738 struct inpcb *inp;
739{
740 struct rtentry *rt = tcp_rtlookup(inp);
741
742 /* Make sure this is a host route and is up. */
743 if (rt == NULL ||
744 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
745 return NULL;
746
747 return rmx_taop(rt->rt_rmx);
748}
749
750/*
751 * Clear all the TAO cache entries, called from tcp_init.
752 *
753 * XXX
754 * This routine is just an empty one, because we assume that the routing
755 * routing tables are initialized at the same time when TCP, so there is
756 * nothing in the cache left over.
757 */
758static void
759tcp_cleartaocache()
760{
761}
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 1.52 1999/02/04 03:27:43 msmith 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/socket.h>
47#include <sys/socketvar.h>
48#include <sys/protosw.h>
49
50#include <vm/vm_zone.h>
51
52#include <net/route.h>
53#include <net/if.h>
54
55#define _IP_VHL
56#include <netinet/in.h>
57#include <netinet/in_systm.h>
58#include <netinet/ip.h>
59#include <netinet/in_pcb.h>
60#include <netinet/in_var.h>
61#include <netinet/ip_var.h>
62#include <netinet/tcp.h>
63#include <netinet/tcp_fsm.h>
64#include <netinet/tcp_seq.h>
65#include <netinet/tcp_timer.h>
66#include <netinet/tcp_var.h>
67#include <netinet/tcpip.h>
68#ifdef TCPDEBUG
69#include <netinet/tcp_debug.h>
70#endif
71
72int tcp_mssdflt = TCP_MSS;
73SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
74 CTLFLAG_RW, &tcp_mssdflt , 0, "");
75
76static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
77SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt,
78 CTLFLAG_RW, &tcp_rttdflt , 0, "");
79
80static int tcp_do_rfc1323 = 1;
81SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
82 CTLFLAG_RW, &tcp_do_rfc1323 , 0, "");
83
84static int tcp_do_rfc1644 = 0;
85SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644,
86 CTLFLAG_RW, &tcp_do_rfc1644 , 0, "");
87
88SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, &tcbinfo.ipi_count,
89 0, "Number of active PCBs");
90
91static void tcp_cleartaocache __P((void));
92static void tcp_notify __P((struct inpcb *, int));
93
94/*
95 * Target size of TCP PCB hash tables. Must be a power of two.
96 *
97 * Note that this can be overridden by the kernel environment
98 * variable net.inet.tcp.tcbhashsize
99 */
100#ifndef TCBHASHSIZE
101#define TCBHASHSIZE 512
102#endif
103
104/*
105 * This is the actual shape of what we allocate using the zone
106 * allocator. Doing it this way allows us to protect both structures
107 * using the same generation count, and also eliminates the overhead
108 * of allocating tcpcbs separately. By hiding the structure here,
109 * we avoid changing most of the rest of the code (although it needs
110 * to be changed, eventually, for greater efficiency).
111 */
112#define ALIGNMENT 32
113#define ALIGNM1 (ALIGNMENT - 1)
114struct inp_tp {
115 union {
116 struct inpcb inp;
117 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
118 } inp_tp_u;
119 struct tcpcb tcb;
120};
121#undef ALIGNMENT
122#undef ALIGNM1
123
124/*
125 * Tcp initialization
126 */
127void
128tcp_init()
129{
130 int hashsize;
131
132 tcp_iss = random(); /* wrong, but better than a constant */
133 tcp_ccgen = 1;
134 tcp_cleartaocache();
135 LIST_INIT(&tcb);
136 tcbinfo.listhead = &tcb;
137 if (!(getenv_int("net.inet.tcp.tcbhashsize", &hashsize)))
138 hashsize = TCBHASHSIZE;
139 if (!powerof2(hashsize)) {
140 printf("WARNING: TCB hash size not a power of 2\n");
141 hashsize = 512; /* safe default */
142 }
143 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
144 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
145 &tcbinfo.porthashmask);
146 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
147 ZONE_INTERRUPT, 0);
148 if (max_protohdr < sizeof(struct tcpiphdr))
149 max_protohdr = sizeof(struct tcpiphdr);
150 if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
151 panic("tcp_init");
152}
153
154/*
155 * Create template to be used to send tcp packets on a connection.
156 * Call after host entry created, allocates an mbuf and fills
157 * in a skeletal tcp/ip header, minimizing the amount of work
158 * necessary when the connection is used.
159 */
160struct tcpiphdr *
161tcp_template(tp)
162 struct tcpcb *tp;
163{
164 register struct inpcb *inp = tp->t_inpcb;
165 register struct mbuf *m;
166 register struct tcpiphdr *n;
167
168 if ((n = tp->t_template) == 0) {
169 m = m_get(M_DONTWAIT, MT_HEADER);
170 if (m == NULL)
171 return (0);
172 m->m_len = sizeof (struct tcpiphdr);
173 n = mtod(m, struct tcpiphdr *);
174 }
175 bzero(n->ti_x1, sizeof(n->ti_x1));
176 n->ti_pr = IPPROTO_TCP;
177 n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
178 n->ti_src = inp->inp_laddr;
179 n->ti_dst = inp->inp_faddr;
180 n->ti_sport = inp->inp_lport;
181 n->ti_dport = inp->inp_fport;
182 n->ti_seq = 0;
183 n->ti_ack = 0;
184 n->ti_x2 = 0;
185 n->ti_off = 5;
186 n->ti_flags = 0;
187 n->ti_win = 0;
188 n->ti_sum = 0;
189 n->ti_urp = 0;
190 return (n);
191}
192
193/*
194 * Send a single message to the TCP at address specified by
195 * the given TCP/IP header. If m == 0, then we make a copy
196 * of the tcpiphdr at ti and send directly to the addressed host.
197 * This is used to force keep alive messages out using the TCP
198 * template for a connection tp->t_template. If flags are given
199 * then we send a message back to the TCP which originated the
200 * segment ti, and discard the mbuf containing it and any other
201 * attached mbufs.
202 *
203 * In any case the ack and sequence number of the transmitted
204 * segment are as specified by the parameters.
205 *
206 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
207 */
208void
209tcp_respond(tp, ti, m, ack, seq, flags)
210 struct tcpcb *tp;
211 register struct tcpiphdr *ti;
212 register struct mbuf *m;
213 tcp_seq ack, seq;
214 int flags;
215{
216 register int tlen;
217 int win = 0;
218 struct route *ro = 0;
219 struct route sro;
220
221 if (tp) {
222 if (!(flags & TH_RST))
223 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
224 ro = &tp->t_inpcb->inp_route;
225 } else {
226 ro = &sro;
227 bzero(ro, sizeof *ro);
228 }
229 if (m == 0) {
230 m = m_gethdr(M_DONTWAIT, MT_HEADER);
231 if (m == NULL)
232 return;
233#ifdef TCP_COMPAT_42
234 tlen = 1;
235#else
236 tlen = 0;
237#endif
238 m->m_data += max_linkhdr;
239 *mtod(m, struct tcpiphdr *) = *ti;
240 ti = mtod(m, struct tcpiphdr *);
241 flags = TH_ACK;
242 } else {
243 m_freem(m->m_next);
244 m->m_next = 0;
245 m->m_data = (caddr_t)ti;
246 m->m_len = sizeof (struct tcpiphdr);
247 tlen = 0;
248#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
249 xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, n_long);
250 xchg(ti->ti_dport, ti->ti_sport, n_short);
251#undef xchg
252 }
253 ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
254 tlen += sizeof (struct tcpiphdr);
255 m->m_len = tlen;
256 m->m_pkthdr.len = tlen;
257 m->m_pkthdr.rcvif = (struct ifnet *) 0;
258 bzero(ti->ti_x1, sizeof(ti->ti_x1));
259 ti->ti_seq = htonl(seq);
260 ti->ti_ack = htonl(ack);
261 ti->ti_x2 = 0;
262 ti->ti_off = sizeof (struct tcphdr) >> 2;
263 ti->ti_flags = flags;
264 if (tp)
265 ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
266 else
267 ti->ti_win = htons((u_short)win);
268 ti->ti_urp = 0;
269 ti->ti_sum = 0;
270 ti->ti_sum = in_cksum(m, tlen);
271 ((struct ip *)ti)->ip_len = tlen;
272 ((struct ip *)ti)->ip_ttl = ip_defttl;
273#ifdef TCPDEBUG
274 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
275 tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
276#endif
277 (void) ip_output(m, NULL, ro, 0, NULL);
278 if (ro == &sro && ro->ro_rt) {
279 RTFREE(ro->ro_rt);
280 }
281}
282
283/*
284 * Create a new TCP control block, making an
285 * empty reassembly queue and hooking it to the argument
286 * protocol control block. The `inp' parameter must have
287 * come from the zone allocator set up in tcp_init().
288 */
289struct tcpcb *
290tcp_newtcpcb(inp)
291 struct inpcb *inp;
292{
293 struct inp_tp *it;
294 register struct tcpcb *tp;
295
296 it = (struct inp_tp *)inp;
297 tp = &it->tcb;
298 bzero((char *) tp, sizeof(struct tcpcb));
299 tp->t_segq = NULL;
300 tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
301
302 if (tcp_do_rfc1323)
303 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
304 if (tcp_do_rfc1644)
305 tp->t_flags |= TF_REQ_CC;
306 tp->t_inpcb = inp; /* XXX */
307 /*
308 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
309 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
310 * reasonable initial retransmit time.
311 */
312 tp->t_srtt = TCPTV_SRTTBASE;
313 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
314 tp->t_rttmin = TCPTV_MIN;
315 tp->t_rxtcur = TCPTV_RTOBASE;
316 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
317 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
318 inp->inp_ip_ttl = ip_defttl;
319 inp->inp_ppcb = (caddr_t)tp;
320 return (tp); /* XXX */
321}
322
323/*
324 * Drop a TCP connection, reporting
325 * the specified error. If connection is synchronized,
326 * then send a RST to peer.
327 */
328struct tcpcb *
329tcp_drop(tp, errno)
330 register struct tcpcb *tp;
331 int errno;
332{
333 struct socket *so = tp->t_inpcb->inp_socket;
334
335 if (TCPS_HAVERCVDSYN(tp->t_state)) {
336 tp->t_state = TCPS_CLOSED;
337 (void) tcp_output(tp);
338 tcpstat.tcps_drops++;
339 } else
340 tcpstat.tcps_conndrops++;
341 if (errno == ETIMEDOUT && tp->t_softerror)
342 errno = tp->t_softerror;
343 so->so_error = errno;
344 return (tcp_close(tp));
345}
346
347/*
348 * Close a TCP control block:
349 * discard all space held by the tcp
350 * discard internet protocol block
351 * wake up any sleepers
352 */
353struct tcpcb *
354tcp_close(tp)
355 register struct tcpcb *tp;
356{
357 register struct mbuf *q;
358 register struct mbuf *nq;
359 struct inpcb *inp = tp->t_inpcb;
360 struct socket *so = inp->inp_socket;
361 register struct rtentry *rt;
362 int dosavessthresh;
363
364 /*
365 * If we got enough samples through the srtt filter,
366 * save the rtt and rttvar in the routing entry.
367 * 'Enough' is arbitrarily defined as the 16 samples.
368 * 16 samples is enough for the srtt filter to converge
369 * to within 5% of the correct value; fewer samples and
370 * we could save a very bogus rtt.
371 *
372 * Don't update the default route's characteristics and don't
373 * update anything that the user "locked".
374 */
375 if (tp->t_rttupdated >= 16 &&
376 (rt = inp->inp_route.ro_rt) &&
377 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
378 register u_long i = 0;
379
380 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
381 i = tp->t_srtt *
382 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
383 if (rt->rt_rmx.rmx_rtt && i)
384 /*
385 * filter this update to half the old & half
386 * the new values, converting scale.
387 * See route.h and tcp_var.h for a
388 * description of the scaling constants.
389 */
390 rt->rt_rmx.rmx_rtt =
391 (rt->rt_rmx.rmx_rtt + i) / 2;
392 else
393 rt->rt_rmx.rmx_rtt = i;
394 tcpstat.tcps_cachedrtt++;
395 }
396 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
397 i = tp->t_rttvar *
398 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
399 if (rt->rt_rmx.rmx_rttvar && i)
400 rt->rt_rmx.rmx_rttvar =
401 (rt->rt_rmx.rmx_rttvar + i) / 2;
402 else
403 rt->rt_rmx.rmx_rttvar = i;
404 tcpstat.tcps_cachedrttvar++;
405 }
406 /*
407 * The old comment here said:
408 * update the pipelimit (ssthresh) if it has been updated
409 * already or if a pipesize was specified & the threshhold
410 * got below half the pipesize. I.e., wait for bad news
411 * before we start updating, then update on both good
412 * and bad news.
413 *
414 * But we want to save the ssthresh even if no pipesize is
415 * specified explicitly in the route, because such
416 * connections still have an implicit pipesize specified
417 * by the global tcp_sendspace. In the absence of a reliable
418 * way to calculate the pipesize, it will have to do.
419 */
420 i = tp->snd_ssthresh;
421 if (rt->rt_rmx.rmx_sendpipe != 0)
422 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
423 else
424 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
425 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
426 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
427 || dosavessthresh) {
428 /*
429 * convert the limit from user data bytes to
430 * packets then to packet data bytes.
431 */
432 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
433 if (i < 2)
434 i = 2;
435 i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
436 if (rt->rt_rmx.rmx_ssthresh)
437 rt->rt_rmx.rmx_ssthresh =
438 (rt->rt_rmx.rmx_ssthresh + i) / 2;
439 else
440 rt->rt_rmx.rmx_ssthresh = i;
441 tcpstat.tcps_cachedssthresh++;
442 }
443 }
444 /* free the reassembly queue, if any */
445 for (q = tp->t_segq; q; q = nq) {
446 nq = q->m_nextpkt;
447 tp->t_segq = nq;
448 m_freem(q);
449 }
450 if (tp->t_template)
451 (void) m_free(dtom(tp->t_template));
452 inp->inp_ppcb = NULL;
453 soisdisconnected(so);
454 in_pcbdetach(inp);
455 tcpstat.tcps_closed++;
456 return ((struct tcpcb *)0);
457}
458
459void
460tcp_drain()
461{
462
463}
464
465/*
466 * Notify a tcp user of an asynchronous error;
467 * store error as soft error, but wake up user
468 * (for now, won't do anything until can select for soft error).
469 */
470static void
471tcp_notify(inp, error)
472 struct inpcb *inp;
473 int error;
474{
475 register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
476 register struct socket *so = inp->inp_socket;
477
478 /*
479 * Ignore some errors if we are hooked up.
480 * If connection hasn't completed, has retransmitted several times,
481 * and receives a second error, give up now. This is better
482 * than waiting a long time to establish a connection that
483 * can never complete.
484 */
485 if (tp->t_state == TCPS_ESTABLISHED &&
486 (error == EHOSTUNREACH || error == ENETUNREACH ||
487 error == EHOSTDOWN)) {
488 return;
489 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
490 tp->t_softerror)
491 so->so_error = error;
492 else
493 tp->t_softerror = error;
494 wakeup((caddr_t) &so->so_timeo);
495 sorwakeup(so);
496 sowwakeup(so);
497}
498
499static int
500tcp_pcblist SYSCTL_HANDLER_ARGS
501{
502 int error, i, n, s;
503 struct inpcb *inp, **inp_list;
504 inp_gen_t gencnt;
505 struct xinpgen xig;
506
507 /*
508 * The process of preparing the TCB list is too time-consuming and
509 * resource-intensive to repeat twice on every request.
510 */
511 if (req->oldptr == 0) {
512 n = tcbinfo.ipi_count;
513 req->oldidx = 2 * (sizeof xig)
514 + (n + n/8) * sizeof(struct xtcpcb);
515 return 0;
516 }
517
518 if (req->newptr != 0)
519 return EPERM;
520
521 /*
522 * OK, now we're committed to doing something.
523 */
524 s = splnet();
525 gencnt = tcbinfo.ipi_gencnt;
526 n = tcbinfo.ipi_count;
527 splx(s);
528
529 xig.xig_len = sizeof xig;
530 xig.xig_count = n;
531 xig.xig_gen = gencnt;
532 xig.xig_sogen = so_gencnt;
533 error = SYSCTL_OUT(req, &xig, sizeof xig);
534 if (error)
535 return error;
536
537 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
538 if (inp_list == 0)
539 return ENOMEM;
540
541 s = splnet();
542 for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
543 inp = inp->inp_list.le_next) {
544 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
545 inp_list[i++] = inp;
546 }
547 splx(s);
548 n = i;
549
550 error = 0;
551 for (i = 0; i < n; i++) {
552 inp = inp_list[i];
553 if (inp->inp_gencnt <= gencnt) {
554 struct xtcpcb xt;
555 xt.xt_len = sizeof xt;
556 /* XXX should avoid extra copy */
557 bcopy(inp, &xt.xt_inp, sizeof *inp);
558 bcopy(inp->inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
559 if (inp->inp_socket)
560 sotoxsocket(inp->inp_socket, &xt.xt_socket);
561 error = SYSCTL_OUT(req, &xt, sizeof xt);
562 }
563 }
564 if (!error) {
565 /*
566 * Give the user an updated idea of our state.
567 * If the generation differs from what we told
568 * her before, she knows that something happened
569 * while we were processing this request, and it
570 * might be necessary to retry.
571 */
572 s = splnet();
573 xig.xig_gen = tcbinfo.ipi_gencnt;
574 xig.xig_sogen = so_gencnt;
575 xig.xig_count = tcbinfo.ipi_count;
576 splx(s);
577 error = SYSCTL_OUT(req, &xig, sizeof xig);
578 }
579 free(inp_list, M_TEMP);
580 return error;
581}
582
583SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
584 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
585
586void
587tcp_ctlinput(cmd, sa, vip)
588 int cmd;
589 struct sockaddr *sa;
590 void *vip;
591{
592 register struct ip *ip = vip;
593 register struct tcphdr *th;
594 void (*notify) __P((struct inpcb *, int)) = tcp_notify;
595
596 if (cmd == PRC_QUENCH)
597 notify = tcp_quench;
598 else if (cmd == PRC_MSGSIZE)
599 notify = tcp_mtudisc;
600 else if (!PRC_IS_REDIRECT(cmd) &&
601 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
602 return;
603 if (ip) {
604 th = (struct tcphdr *)((caddr_t)ip
605 + (IP_VHL_HL(ip->ip_vhl) << 2));
606 in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
607 cmd, notify);
608 } else
609 in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
610}
611
612/*
613 * When a source quench is received, close congestion window
614 * to one segment. We will gradually open it again as we proceed.
615 */
616void
617tcp_quench(inp, errno)
618 struct inpcb *inp;
619 int errno;
620{
621 struct tcpcb *tp = intotcpcb(inp);
622
623 if (tp)
624 tp->snd_cwnd = tp->t_maxseg;
625}
626
627/*
628 * When `need fragmentation' ICMP is received, update our idea of the MSS
629 * based on the new value in the route. Also nudge TCP to send something,
630 * since we know the packet we just sent was dropped.
631 * This duplicates some code in the tcp_mss() function in tcp_input.c.
632 */
633void
634tcp_mtudisc(inp, errno)
635 struct inpcb *inp;
636 int errno;
637{
638 struct tcpcb *tp = intotcpcb(inp);
639 struct rtentry *rt;
640 struct rmxp_tao *taop;
641 struct socket *so = inp->inp_socket;
642 int offered;
643 int mss;
644
645 if (tp) {
646 rt = tcp_rtlookup(inp);
647 if (!rt || !rt->rt_rmx.rmx_mtu) {
648 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
649 return;
650 }
651 taop = rmx_taop(rt->rt_rmx);
652 offered = taop->tao_mssopt;
653 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
654 if (offered)
655 mss = min(mss, offered);
656 /*
657 * XXX - The above conditional probably violates the TCP
658 * spec. The problem is that, since we don't know the
659 * other end's MSS, we are supposed to use a conservative
660 * default. But, if we do that, then MTU discovery will
661 * never actually take place, because the conservative
662 * default is much less than the MTUs typically seen
663 * on the Internet today. For the moment, we'll sweep
664 * this under the carpet.
665 *
666 * The conservative default might not actually be a problem
667 * if the only case this occurs is when sending an initial
668 * SYN with options and data to a host we've never talked
669 * to before. Then, they will reply with an MSS value which
670 * will get recorded and the new parameters should get
671 * recomputed. For Further Study.
672 */
673 if (tp->t_maxopd <= mss)
674 return;
675 tp->t_maxopd = mss;
676
677 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
678 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
679 mss -= TCPOLEN_TSTAMP_APPA;
680 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
681 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
682 mss -= TCPOLEN_CC_APPA;
683#if (MCLBYTES & (MCLBYTES - 1)) == 0
684 if (mss > MCLBYTES)
685 mss &= ~(MCLBYTES-1);
686#else
687 if (mss > MCLBYTES)
688 mss = mss / MCLBYTES * MCLBYTES;
689#endif
690 if (so->so_snd.sb_hiwat < mss)
691 mss = so->so_snd.sb_hiwat;
692
693 tp->t_maxseg = mss;
694
695 tcpstat.tcps_mturesent++;
696 tp->t_rtt = 0;
697 tp->snd_nxt = tp->snd_una;
698 tcp_output(tp);
699 }
700}
701
702/*
703 * Look-up the routing entry to the peer of this inpcb. If no route
704 * is found and it cannot be allocated the return NULL. This routine
705 * is called by TCP routines that access the rmx structure and by tcp_mss
706 * to get the interface MTU.
707 */
708struct rtentry *
709tcp_rtlookup(inp)
710 struct inpcb *inp;
711{
712 struct route *ro;
713 struct rtentry *rt;
714
715 ro = &inp->inp_route;
716 rt = ro->ro_rt;
717 if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
718 /* No route yet, so try to acquire one */
719 if (inp->inp_faddr.s_addr != INADDR_ANY) {
720 ro->ro_dst.sa_family = AF_INET;
721 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
722 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
723 inp->inp_faddr;
724 rtalloc(ro);
725 rt = ro->ro_rt;
726 }
727 }
728 return rt;
729}
730
731/*
732 * Return a pointer to the cached information about the remote host.
733 * The cached information is stored in the protocol specific part of
734 * the route metrics.
735 */
736struct rmxp_tao *
737tcp_gettaocache(inp)
738 struct inpcb *inp;
739{
740 struct rtentry *rt = tcp_rtlookup(inp);
741
742 /* Make sure this is a host route and is up. */
743 if (rt == NULL ||
744 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
745 return NULL;
746
747 return rmx_taop(rt->rt_rmx);
748}
749
750/*
751 * Clear all the TAO cache entries, called from tcp_init.
752 *
753 * XXX
754 * This routine is just an empty one, because we assume that the routing
755 * routing tables are initialized at the same time when TCP, so there is
756 * nothing in the cache left over.
757 */
758static void
759tcp_cleartaocache()
760{
761}