]> git.saurik.com Git - apple/xnu.git/blob - bsd/netinet/tcp_input.c
projects / apple / xnu.git / blob
? search:


74493fdb740ded65aa26b04a76e5748172ecf82c
[apple/xnu.git] / bsd / netinet / tcp_input.c
1 /*
2 * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
11 *
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
18 * under the License.
19 *
20 * @APPLE_LICENSE_HEADER_END@
21 */
22 /*
23 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
24 * The Regents of the University of California. All rights reserved.
25 *
26 * Redistribution and use in source and binary forms, with or without
27 * modification, are permitted provided that the following conditions
28 * are met:
29 * 1. Redistributions of source code must retain the above copyright
30 * notice, this list of conditions and the following disclaimer.
31 * 2. Redistributions in binary form must reproduce the above copyright
32 * notice, this list of conditions and the following disclaimer in the
33 * documentation and/or other materials provided with the distribution.
34 * 3. All advertising materials mentioning features or use of this software
35 * must display the following acknowledgement:
36 * This product includes software developed by the University of
37 * California, Berkeley and its contributors.
38 * 4. Neither the name of the University nor the names of its contributors
39 * may be used to endorse or promote products derived from this software
40 * without specific prior written permission.
41 *
42 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52 * SUCH DAMAGE.
53 *
54 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
55 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.16 2001/08/22 00:59:12 silby Exp $
56 */
57
58
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/sysctl.h>
63 #include <sys/malloc.h>
64 #include <sys/mbuf.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/syslog.h>
70
71 #include <kern/cpu_number.h> /* before tcp_seq.h, for tcp_random18() */
72
73 #include <net/if.h>
74 #include <net/if_types.h>
75 #include <net/route.h>
76
77 #include <netinet/in.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/ip.h>
80 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
81 #include <netinet/in_var.h>
82 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
83 #include <netinet/in_pcb.h>
84 #include <netinet/ip_var.h>
85 #if INET6
86 #include <netinet/ip6.h>
87 #include <netinet/icmp6.h>
88 #include <netinet6/nd6.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/in6_pcb.h>
91 #endif
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #if INET6
98 #include <netinet6/tcp6_var.h>
99 #endif
100 #include <netinet/tcpip.h>
101 #if TCPDEBUG
102 #include <netinet/tcp_debug.h>
103 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
104 struct tcphdr tcp_savetcp;
105 #endif /* TCPDEBUG */
106
107 #if IPSEC
108 #include <netinet6/ipsec.h>
109 #if INET6
110 #include <netinet6/ipsec6.h>
111 #endif
112 #include <netkey/key.h>
113 #endif /*IPSEC*/
114
115 #include <sys/kdebug.h>
116
117 #ifndef __APPLE__
118 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
119 #endif
120
121 #define DBG_LAYER_BEG NETDBG_CODE(DBG_NETTCP, 0)
122 #define DBG_LAYER_END NETDBG_CODE(DBG_NETTCP, 2)
123 #define DBG_FNC_TCP_INPUT NETDBG_CODE(DBG_NETTCP, (3 << 8))
124 #define DBG_FNC_TCP_NEWCONN NETDBG_CODE(DBG_NETTCP, (7 << 8))
125
126 static int tcprexmtthresh = 3;
127 tcp_cc tcp_ccgen;
128 extern int apple_hwcksum_rx;
129
130 #if IPSEC
131 extern int ipsec_bypass;
132 #endif
133
134 struct tcpstat tcpstat;
135 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD,
136 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
137
138 static int log_in_vain = 0;
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
140 &log_in_vain, 0, "Log all incoming TCP connections");
141
142 static int blackhole = 0;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
144 &blackhole, 0, "Do not send RST when dropping refused connections");
145
146 int tcp_delack_enabled = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
148 &tcp_delack_enabled, 0,
149 "Delay ACK to try and piggyback it onto a data packet");
150
151 int tcp_lq_overflow = 1;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW,
153 &tcp_lq_overflow, 0,
154 "Listen Queue Overflow");
155
156 #if TCP_DROP_SYNFIN
157 static int drop_synfin = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
159 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
160 #endif
161
162 __private_extern__ int slowlink_wsize = 8192;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowlink_wsize, CTLFLAG_RW,
164 &slowlink_wsize, 0, "Maximum advertised window size for slowlink");
165
166
167 u_long tcp_now;
168 struct inpcbhead tcb;
169 #define tcb6 tcb /* for KAME src sync over BSD*'s */
170 struct inpcbinfo tcbinfo;
171
172 static void tcp_dooptions __P((struct tcpcb *,
173 u_char *, int, struct tcphdr *, struct tcpopt *));
174 static void tcp_pulloutofband __P((struct socket *,
175 struct tcphdr *, struct mbuf *, int));
176 static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *,
177 struct mbuf *));
178 static void tcp_xmit_timer __P((struct tcpcb *, int));
179 static int tcp_newreno __P((struct tcpcb *, struct tcphdr *));
180
181 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
182 #if INET6
183 #define ND6_HINT(tp) \
184 do { \
185 if ((tp) && (tp)->t_inpcb && \
186 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
187 (tp)->t_inpcb->in6p_route.ro_rt) \
188 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
189 } while (0)
190 #else
191 #define ND6_HINT(tp)
192 #endif
193
194 extern u_long *delack_bitmask;
195
196 /*
197 * Indicate whether this ack should be delayed. We can delay the ack if
198 * - delayed acks are enabled and
199 * - there is no delayed ack timer in progress and
200 * - our last ack wasn't a 0-sized window. We never want to delay
201 * the ack that opens up a 0-sized window.
202 */
203 #define DELAY_ACK(tp) \
204 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
205 (tp->t_flags & TF_RXWIN0SENT) == 0)
206
207
208 static int
209 tcp_reass(tp, th, tlenp, m)
210 register struct tcpcb *tp;
211 register struct tcphdr *th;
212 int *tlenp;
213 struct mbuf *m;
214 {
215 struct tseg_qent *q;
216 struct tseg_qent *p = NULL;
217 struct tseg_qent *nq;
218 struct tseg_qent *te;
219 struct socket *so = tp->t_inpcb->inp_socket;
220 int flags;
221
222 /*
223 * Call with th==0 after become established to
224 * force pre-ESTABLISHED data up to user socket.
225 */
226 if (th == 0)
227 goto present;
228
229 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
230 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ,
231 M_NOWAIT);
232 if (te == NULL) {
233 tcpstat.tcps_rcvmemdrop++;
234 m_freem(m);
235 return (0);
236 }
237
238 /*
239 * Find a segment which begins after this one does.
240 */
241 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
242 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
243 break;
244 p = q;
245 }
246
247 /*
248 * If there is a preceding segment, it may provide some of
249 * our data already. If so, drop the data from the incoming
250 * segment. If it provides all of our data, drop us.
251 */
252 if (p != NULL) {
253 register int i;
254 /* conversion to int (in i) handles seq wraparound */
255 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
256 if (i > 0) {
257 if (i >= *tlenp) {
258 tcpstat.tcps_rcvduppack++;
259 tcpstat.tcps_rcvdupbyte += *tlenp;
260 m_freem(m);
261 FREE(te, M_TSEGQ);
262 /*
263 * Try to present any queued data
264 * at the left window edge to the user.
265 * This is needed after the 3-WHS
266 * completes.
267 */
268 goto present; /* ??? */
269 }
270 m_adj(m, i);
271 *tlenp -= i;
272 th->th_seq += i;
273 }
274 }
275 tcpstat.tcps_rcvoopack++;
276 tcpstat.tcps_rcvoobyte += *tlenp;
277
278 /*
279 * While we overlap succeeding segments trim them or,
280 * if they are completely covered, dequeue them.
281 */
282 while (q) {
283 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
284 if (i <= 0)
285 break;
286 if (i < q->tqe_len) {
287 q->tqe_th->th_seq += i;
288 q->tqe_len -= i;
289 m_adj(q->tqe_m, i);
290 break;
291 }
292
293 nq = LIST_NEXT(q, tqe_q);
294 LIST_REMOVE(q, tqe_q);
295 m_freem(q->tqe_m);
296 FREE(q, M_TSEGQ);
297 q = nq;
298 }
299
300 /* Insert the new segment queue entry into place. */
301 te->tqe_m = m;
302 te->tqe_th = th;
303 te->tqe_len = *tlenp;
304
305 if (p == NULL) {
306 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
307 } else {
308 LIST_INSERT_AFTER(p, te, tqe_q);
309 }
310
311 present:
312 /*
313 * Present data to user, advancing rcv_nxt through
314 * completed sequence space.
315 */
316 if (!TCPS_HAVEESTABLISHED(tp->t_state))
317 return (0);
318 q = LIST_FIRST(&tp->t_segq);
319 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
320 return (0);
321 do {
322 tp->rcv_nxt += q->tqe_len;
323 flags = q->tqe_th->th_flags & TH_FIN;
324 nq = LIST_NEXT(q, tqe_q);
325 LIST_REMOVE(q, tqe_q);
326 if (so->so_state & SS_CANTRCVMORE)
327 m_freem(q->tqe_m);
328 else
329 sbappend(&so->so_rcv, q->tqe_m);
330 FREE(q, M_TSEGQ);
331 q = nq;
332 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
333 ND6_HINT(tp);
334
335 #if INET6
336 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
337
338 KERNEL_DEBUG(DBG_LAYER_BEG,
339 ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
340 (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) |
341 (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)),
342 0,0,0);
343 }
344 else
345 #endif
346 {
347 KERNEL_DEBUG(DBG_LAYER_BEG,
348 ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
349 (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) |
350 (tp->t_inpcb->inp_faddr.s_addr & 0xffff)),
351 0,0,0);
352 }
353 sorwakeup(so);
354 return (flags);
355
356 }
357
358
359 /*
360 * TCP input routine, follows pages 65-76 of the
361 * protocol specification dated September, 1981 very closely.
362 */
363 #if INET6
364 int
365 tcp6_input(mp, offp, proto)
366 struct mbuf **mp;
367 int *offp, proto;
368 {
369 register struct mbuf *m = *mp;
370 struct in6_ifaddr *ia6;
371
372 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
373
374 /*
375 * draft-itojun-ipv6-tcp-to-anycast
376 * better place to put this in?
377 */
378 ia6 = ip6_getdstifaddr(m);
379 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
380 struct ip6_hdr *ip6;
381
382 ip6 = mtod(m, struct ip6_hdr *);
383 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
384 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
385 return IPPROTO_DONE;
386 }
387
388 tcp_input(m, *offp);
389 return IPPROTO_DONE;
390 }
391 #endif
392
393 void
394 tcp_input(m, off0)
395 struct mbuf *m;
396 int off0;
397 {
398 register struct tcphdr *th;
399 register struct ip *ip = NULL;
400 register struct ipovly *ipov;
401 register struct inpcb *inp;
402 u_char *optp = NULL;
403 int optlen = 0;
404 int len, tlen, off;
405 int drop_hdrlen;
406 register struct tcpcb *tp = 0;
407 register int thflags;
408 struct socket *so = 0;
409 int todrop, acked, ourfinisacked, needoutput = 0;
410 struct in_addr laddr;
411 #if INET6
412 struct in6_addr laddr6;
413 #endif
414 int dropsocket = 0;
415 int iss = 0;
416 u_long tiwin;
417 struct tcpopt to; /* options in this segment */
418 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
419 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
420 #if TCPDEBUG
421 short ostate = 0;
422 #endif
423 #if INET6
424 struct ip6_hdr *ip6 = NULL;
425 int isipv6;
426 #endif /* INET6 */
427 int rstreason; /* For badport_bandlim accounting purposes */
428 struct proc *proc0=current_proc();
429
430 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_START,0,0,0,0,0);
431
432 #if INET6
433 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
434 #endif
435 bzero((char *)&to, sizeof(to));
436
437 tcpstat.tcps_rcvtotal++;
438
439
440
441 #if INET6
442 if (isipv6) {
443 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
444 ip6 = mtod(m, struct ip6_hdr *);
445 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
446 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
447 tcpstat.tcps_rcvbadsum++;
448 goto drop;
449 }
450 th = (struct tcphdr *)((caddr_t)ip6 + off0);
451
452 KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport),
453 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
454 th->th_seq, th->th_ack, th->th_win);
455 /*
456 * Be proactive about unspecified IPv6 address in source.
457 * As we use all-zero to indicate unbounded/unconnected pcb,
458 * unspecified IPv6 address can be used to confuse us.
459 *
460 * Note that packets with unspecified IPv6 destination is
461 * already dropped in ip6_input.
462 */
463 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
464 /* XXX stat */
465 goto drop;
466 }
467 } else
468 #endif /* INET6 */
469 {
470 /*
471 * Get IP and TCP header together in first mbuf.
472 * Note: IP leaves IP header in first mbuf.
473 */
474 if (off0 > sizeof (struct ip)) {
475 ip_stripoptions(m, (struct mbuf *)0);
476 off0 = sizeof(struct ip);
477 if (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16)
478 m->m_pkthdr.csum_flags = 0; /* invalidate hwcksuming */
479
480 }
481 if (m->m_len < sizeof (struct tcpiphdr)) {
482 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
483 tcpstat.tcps_rcvshort++;
484 return;
485 }
486 }
487 ip = mtod(m, struct ip *);
488 ipov = (struct ipovly *)ip;
489 th = (struct tcphdr *)((caddr_t)ip + off0);
490 tlen = ip->ip_len;
491
492 KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport),
493 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
494 th->th_seq, th->th_ack, th->th_win);
495
496 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
497 if (apple_hwcksum_rx && (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16)) {
498 u_short pseudo;
499 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
500 ipov->ih_len = (u_short)tlen;
501 HTONS(ipov->ih_len);
502 pseudo = in_cksum(m, sizeof (struct ip));
503 th->th_sum = in_addword(pseudo, (m->m_pkthdr.csum_data & 0xFFFF));
504 } else {
505 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
506 th->th_sum = m->m_pkthdr.csum_data;
507 else
508 th->th_sum = in_pseudo(ip->ip_src.s_addr,
509 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data +
510 ip->ip_len + IPPROTO_TCP));
511 }
512 th->th_sum ^= 0xffff;
513 } else {
514 /*
515 * Checksum extended TCP header and data.
516 */
517 len = sizeof (struct ip) + tlen;
518 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
519 ipov->ih_len = (u_short)tlen;
520 HTONS(ipov->ih_len);
521 th->th_sum = in_cksum(m, len);
522 }
523 if (th->th_sum) {
524 tcpstat.tcps_rcvbadsum++;
525 goto drop;
526 }
527 #if INET6
528 /* Re-initialization for later version check */
529 ip->ip_v = IPVERSION;
530 #endif
531 }
532
533 /*
534 * Check that TCP offset makes sense,
535 * pull out TCP options and adjust length. XXX
536 */
537 off = th->th_off << 2;
538 if (off < sizeof (struct tcphdr) || off > tlen) {
539 tcpstat.tcps_rcvbadoff++;
540 goto drop;
541 }
542 tlen -= off; /* tlen is used instead of ti->ti_len */
543 if (off > sizeof (struct tcphdr)) {
544 #if INET6
545 if (isipv6) {
546 IP6_EXTHDR_CHECK(m, off0, off, );
547 ip6 = mtod(m, struct ip6_hdr *);
548 th = (struct tcphdr *)((caddr_t)ip6 + off0);
549 } else
550 #endif /* INET6 */
551 {
552 if (m->m_len < sizeof(struct ip) + off) {
553 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
554 tcpstat.tcps_rcvshort++;
555 return;
556 }
557 ip = mtod(m, struct ip *);
558 ipov = (struct ipovly *)ip;
559 th = (struct tcphdr *)((caddr_t)ip + off0);
560 }
561 }
562 optlen = off - sizeof (struct tcphdr);
563 optp = (u_char *)(th + 1);
564 /*
565 * Do quick retrieval of timestamp options ("options
566 * prediction?"). If timestamp is the only option and it's
567 * formatted as recommended in RFC 1323 appendix A, we
568 * quickly get the values now and not bother calling
569 * tcp_dooptions(), etc.
570 */
571 if ((optlen == TCPOLEN_TSTAMP_APPA ||
572 (optlen > TCPOLEN_TSTAMP_APPA &&
573 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
574 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
575 (th->th_flags & TH_SYN) == 0) {
576 to.to_flag |= TOF_TS;
577 to.to_tsval = ntohl(*(u_int32_t *)(optp + 4));
578 to.to_tsecr = ntohl(*(u_int32_t *)(optp + 8));
579 optp = NULL; /* we've parsed the options */
580 }
581 }
582 thflags = th->th_flags;
583
584 #if TCP_DROP_SYNFIN
585 /*
586 * If the drop_synfin option is enabled, drop all packets with
587 * both the SYN and FIN bits set. This prevents e.g. nmap from
588 * identifying the TCP/IP stack.
589 *
590 * This is incompatible with RFC1644 extensions (T/TCP).
591 */
592 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
593 goto drop;
594 #endif
595
596 /*
597 * Convert TCP protocol specific fields to host format.
598 */
599 NTOHL(th->th_seq);
600 NTOHL(th->th_ack);
601 NTOHS(th->th_win);
602 NTOHS(th->th_urp);
603
604 /*
605 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
606 * until after ip6_savecontrol() is called and before other functions
607 * which don't want those proto headers.
608 * Because ip6_savecontrol() is going to parse the mbuf to
609 * search for data to be passed up to user-land, it wants mbuf
610 * parameters to be unchanged.
611 */
612 drop_hdrlen = off0 + off;
613
614 /*
615 * Locate pcb for segment.
616 */
617 findpcb:
618 #if IPFIREWALL_FORWARD
619 if (ip_fw_fwd_addr != NULL
620 #if INET6
621 && isipv6 == NULL /* IPv6 support is not yet */
622 #endif /* INET6 */
623 ) {
624 /*
625 * Diverted. Pretend to be the destination.
626 * already got one like this?
627 */
628 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
629 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif);
630 if (!inp) {
631 /*
632 * No, then it's new. Try find the ambushing socket
633 */
634 if (!ip_fw_fwd_addr->sin_port) {
635 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src,
636 th->th_sport, ip_fw_fwd_addr->sin_addr,
637 th->th_dport, 1, m->m_pkthdr.rcvif);
638 } else {
639 inp = in_pcblookup_hash(&tcbinfo,
640 ip->ip_src, th->th_sport,
641 ip_fw_fwd_addr->sin_addr,
642 ntohs(ip_fw_fwd_addr->sin_port), 1,
643 m->m_pkthdr.rcvif);
644 }
645 }
646 ip_fw_fwd_addr = NULL;
647 } else
648 #endif /* IPFIREWALL_FORWARD */
649 {
650 #if INET6
651 if (isipv6)
652 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport,
653 &ip6->ip6_dst, th->th_dport, 1,
654 m->m_pkthdr.rcvif);
655 else
656 #endif /* INET6 */
657 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
658 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif);
659 }
660
661 #if IPSEC
662 #if INET6
663 if (isipv6) {
664 if (ipsec_bypass == 0 && inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
665 ipsec6stat.in_polvio++;
666 goto drop;
667 }
668 } else
669 #endif /* INET6 */
670 if (ipsec_bypass == 0 && inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
671 ipsecstat.in_polvio++;
672 goto drop;
673 }
674 #endif /*IPSEC*/
675
676 /*
677 * If the state is CLOSED (i.e., TCB does not exist) then
678 * all data in the incoming segment is discarded.
679 * If the TCB exists but is in CLOSED state, it is embryonic,
680 * but should either do a listen or a connect soon.
681 */
682 if (inp == NULL) {
683 if (log_in_vain) {
684 #if INET6
685 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN];
686 #else /* INET6 */
687 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
688 #endif /* INET6 */
689
690 #if INET6
691 if (isipv6) {
692 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst));
693 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src));
694 } else
695 #endif
696 {
697 strcpy(dbuf, inet_ntoa(ip->ip_dst));
698 strcpy(sbuf, inet_ntoa(ip->ip_src));
699 }
700 switch (log_in_vain) {
701 case 1:
702 if(thflags & TH_SYN)
703 log(LOG_INFO,
704 "Connection attempt to TCP %s:%d from %s:%d\n",
705 dbuf, ntohs(th->th_dport),
706 sbuf,
707 ntohs(th->th_sport));
708 break;
709 case 2:
710 log(LOG_INFO,
711 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n",
712 dbuf, ntohs(th->th_dport), sbuf,
713 ntohs(th->th_sport), thflags);
714 break;
715 default:
716 break;
717 }
718 }
719 if (blackhole) {
720 switch (blackhole) {
721 case 1:
722 if (thflags & TH_SYN)
723 goto drop;
724 break;
725 case 2:
726 goto drop;
727 default:
728 goto drop;
729 }
730 }
731 rstreason = BANDLIM_RST_CLOSEDPORT;
732 goto dropwithreset;
733 }
734 tp = intotcpcb(inp);
735 if (tp == 0) {
736 rstreason = BANDLIM_RST_CLOSEDPORT;
737 goto dropwithreset;
738 }
739 if (tp->t_state == TCPS_CLOSED)
740 goto drop;
741
742 #ifdef __APPLE__
743 /*
744 * Bogus state when listening port owned by SharedIP with loopback as the
745 * only configured interface: BlueBox does not filters loopback
746 */
747 if (tp->t_state == TCP_NSTATES)
748 goto drop;
749 #endif
750
751 /* Unscale the window into a 32-bit value. */
752 if ((thflags & TH_SYN) == 0)
753 tiwin = th->th_win << tp->snd_scale;
754 else
755 tiwin = th->th_win;
756
757 so = inp->inp_socket;
758 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
759 #if TCPDEBUG
760 if (so->so_options & SO_DEBUG) {
761 ostate = tp->t_state;
762 #if INET6
763 if (isipv6)
764 bcopy((char *)ip6, (char *)tcp_saveipgen,
765 sizeof(*ip6));
766 else
767 #endif /* INET6 */
768 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
769 tcp_savetcp = *th;
770 }
771 #endif
772 if (so->so_options & SO_ACCEPTCONN) {
773 register struct tcpcb *tp0 = tp;
774 struct socket *so2;
775 #if IPSEC
776 struct socket *oso;
777 #endif
778 #if INET6
779 struct inpcb *oinp = sotoinpcb(so);
780 #endif /* INET6 */
781
782 #if !IPSEC
783 /*
784 * Current IPsec implementation makes incorrect IPsec
785 * cache if this check is done here.
786 * So delay this until duplicated socket is created.
787 */
788 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
789 /*
790 * Note: dropwithreset makes sure we don't
791 * send a RST in response to a RST.
792 */
793 if (thflags & TH_ACK) {
794 tcpstat.tcps_badsyn++;
795 rstreason = BANDLIM_RST_OPENPORT;
796 goto dropwithreset;
797 }
798 goto drop;
799 }
800 #endif
801 KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_START,0,0,0,0,0);
802
803 #if INET6
804 /*
805 * If deprecated address is forbidden,
806 * we do not accept SYN to deprecated interface
807 * address to prevent any new inbound connection from
808 * getting established.
809 * When we do not accept SYN, we send a TCP RST,
810 * with deprecated source address (instead of dropping
811 * it). We compromise it as it is much better for peer
812 * to send a RST, and RST will be the final packet
813 * for the exchange.
814 *
815 * If we do not forbid deprecated addresses, we accept
816 * the SYN packet. RFC2462 does not suggest dropping
817 * SYN in this case.
818 * If we decipher RFC2462 5.5.4, it says like this:
819 * 1. use of deprecated addr with existing
820 * communication is okay - "SHOULD continue to be
821 * used"
822 * 2. use of it with new communication:
823 * (2a) "SHOULD NOT be used if alternate address
824 * with sufficient scope is available"
825 * (2b) nothing mentioned otherwise.
826 * Here we fall into (2b) case as we have no choice in
827 * our source address selection - we must obey the peer.
828 *
829 * The wording in RFC2462 is confusing, and there are
830 * multiple description text for deprecated address
831 * handling - worse, they are not exactly the same.
832 * I believe 5.5.4 is the best one, so we follow 5.5.4.
833 */
834 if (isipv6 && !ip6_use_deprecated) {
835 struct in6_ifaddr *ia6;
836
837 if ((ia6 = ip6_getdstifaddr(m)) &&
838 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
839 tp = NULL;
840 rstreason = BANDLIM_RST_OPENPORT;
841 goto dropwithreset;
842 }
843 }
844 #endif
845
846 so2 = sonewconn(so, 0);
847 if (so2 == 0) {
848 tcpstat.tcps_listendrop++;
849 so2 = sodropablereq(so);
850 if (so2) {
851 if (tcp_lq_overflow)
852 sototcpcb(so2)->t_flags |=
853 TF_LQ_OVERFLOW;
854 tcp_drop(sototcpcb(so2), ETIMEDOUT);
855 so2 = sonewconn(so, 0);
856 }
857 if (!so2)
858 goto drop;
859 }
860 #if IPSEC
861 oso = so;
862 #endif
863 so = so2;
864 /*
865 * This is ugly, but ....
866 *
867 * Mark socket as temporary until we're
868 * committed to keeping it. The code at
869 * ``drop'' and ``dropwithreset'' check the
870 * flag dropsocket to see if the temporary
871 * socket created here should be discarded.
872 * We mark the socket as discardable until
873 * we're committed to it below in TCPS_LISTEN.
874 */
875 dropsocket++;
876 inp = (struct inpcb *)so->so_pcb;
877 #if INET6
878 if (isipv6)
879 inp->in6p_laddr = ip6->ip6_dst;
880 else {
881 inp->inp_vflag &= ~INP_IPV6;
882 inp->inp_vflag |= INP_IPV4;
883 #endif /* INET6 */
884 inp->inp_laddr = ip->ip_dst;
885 #if INET6
886 }
887 #endif /* INET6 */
888 inp->inp_lport = th->th_dport;
889 if (in_pcbinshash(inp) != 0) {
890 /*
891 * Undo the assignments above if we failed to
892 * put the PCB on the hash lists.
893 */
894 #if INET6
895 if (isipv6)
896 inp->in6p_laddr = in6addr_any;
897 else
898 #endif /* INET6 */
899 inp->inp_laddr.s_addr = INADDR_ANY;
900 inp->inp_lport = 0;
901 goto drop;
902 }
903 #if IPSEC
904 /*
905 * To avoid creating incorrectly cached IPsec
906 * association, this is need to be done here.
907 *
908 * Subject: (KAME-snap 748)
909 * From: Wayne Knowles <w.knowles@niwa.cri.nz>
910 * ftp://ftp.kame.net/pub/mail-list/snap-users/748
911 */
912 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
913 /*
914 * Note: dropwithreset makes sure we don't
915 * send a RST in response to a RST.
916 */
917 if (thflags & TH_ACK) {
918 tcpstat.tcps_badsyn++;
919 rstreason = BANDLIM_RST_OPENPORT;
920 goto dropwithreset;
921 }
922 goto drop;
923 }
924 #endif
925 #if INET6
926 if (isipv6) {
927 /*
928 * Inherit socket options from the listening
929 * socket.
930 * Note that in6p_inputopts are not (even
931 * should not be) copied, since it stores
932 * previously received options and is used to
933 * detect if each new option is different than
934 * the previous one and hence should be passed
935 * to a user.
936 * If we copied in6p_inputopts, a user would
937 * not be able to receive options just after
938 * calling the accept system call.
939 */
940 inp->inp_flags |=
941 oinp->inp_flags & INP_CONTROLOPTS;
942 if (oinp->in6p_outputopts)
943 inp->in6p_outputopts =
944 ip6_copypktopts(oinp->in6p_outputopts,
945 M_NOWAIT);
946 } else
947 #endif /* INET6 */
948 inp->inp_options = ip_srcroute();
949 #if IPSEC
950 /* copy old policy into new socket's */
951 if (sotoinpcb(oso)->inp_sp)
952 {
953 int error = 0;
954 /* Is it a security hole here to silently fail to copy the policy? */
955 if (inp->inp_sp != NULL)
956 error = ipsec_init_policy(so, &inp->inp_sp);
957 if (error != 0 || ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
958 printf("tcp_input: could not copy policy\n");
959 }
960 #endif
961 tp = intotcpcb(inp);
962 tp->t_state = TCPS_LISTEN;
963 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT|TF_NODELAY);
964
965 /* Compute proper scaling value from buffer space */
966 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
967 TCP_MAXWIN << tp->request_r_scale <
968 so->so_rcv.sb_hiwat)
969 tp->request_r_scale++;
970
971 KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_END,0,0,0,0,0);
972 }
973 }
974
975 /*
976 * Segment received on connection.
977 * Reset idle time and keep-alive timer.
978 */
979 tp->t_rcvtime = 0;
980 if (TCPS_HAVEESTABLISHED(tp->t_state))
981 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
982
983 /*
984 * Process options if not in LISTEN state,
985 * else do it below (after getting remote address).
986 */
987 if (tp->t_state != TCPS_LISTEN && optp)
988 tcp_dooptions(tp, optp, optlen, th, &to);
989
990 /*
991 * Header prediction: check for the two common cases
992 * of a uni-directional data xfer. If the packet has
993 * no control flags, is in-sequence, the window didn't
994 * change and we're not retransmitting, it's a
995 * candidate. If the length is zero and the ack moved
996 * forward, we're the sender side of the xfer. Just
997 * free the data acked & wake any higher level process
998 * that was blocked waiting for space. If the length
999 * is non-zero and the ack didn't move, we're the
1000 * receiver side. If we're getting packets in-order
1001 * (the reassembly queue is empty), add the data to
1002 * the socket buffer and note that we need a delayed ack.
1003 * Make sure that the hidden state-flags are also off.
1004 * Since we check for TCPS_ESTABLISHED above, it can only
1005 * be TH_NEEDSYN.
1006 */
1007 if (tp->t_state == TCPS_ESTABLISHED &&
1008 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1009 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1010 ((to.to_flag & TOF_TS) == 0 ||
1011 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1012 /*
1013 * Using the CC option is compulsory if once started:
1014 * the segment is OK if no T/TCP was negotiated or
1015 * if the segment has a CC option equal to CCrecv
1016 */
1017 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1018 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1019 th->th_seq == tp->rcv_nxt &&
1020 tiwin && tiwin == tp->snd_wnd &&
1021 tp->snd_nxt == tp->snd_max) {
1022
1023 /*
1024 * If last ACK falls within this segment's sequence numbers,
1025 * record the timestamp.
1026 * NOTE that the test is modified according to the latest
1027 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1028 */
1029 if ((to.to_flag & TOF_TS) != 0 &&
1030 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1031 tp->ts_recent_age = tcp_now;
1032 tp->ts_recent = to.to_tsval;
1033 }
1034
1035 if (tlen == 0) {
1036 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1037 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1038 tp->snd_cwnd >= tp->snd_wnd &&
1039 tp->t_dupacks < tcprexmtthresh) {
1040 /*
1041 * this is a pure ack for outstanding data.
1042 */
1043 ++tcpstat.tcps_predack;
1044 /*
1045 * "bad retransmit" recovery
1046 */
1047 if (tp->t_rxtshift == 1 &&
1048 tcp_now < tp->t_badrxtwin) {
1049 tp->snd_cwnd = tp->snd_cwnd_prev;
1050 tp->snd_ssthresh =
1051 tp->snd_ssthresh_prev;
1052 tp->snd_nxt = tp->snd_max;
1053 tp->t_badrxtwin = 0;
1054 }
1055 if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0)) /* Makes sure we already have a TS */
1056 tcp_xmit_timer(tp,
1057 tcp_now - to.to_tsecr + 1);
1058 else if (tp->t_rtttime &&
1059 SEQ_GT(th->th_ack, tp->t_rtseq))
1060 tcp_xmit_timer(tp, tp->t_rtttime);
1061 acked = th->th_ack - tp->snd_una;
1062 tcpstat.tcps_rcvackpack++;
1063 tcpstat.tcps_rcvackbyte += acked;
1064 sbdrop(&so->so_snd, acked);
1065 tp->snd_una = th->th_ack;
1066 m_freem(m);
1067 ND6_HINT(tp); /* some progress has been done */
1068
1069 /*
1070 * If all outstanding data are acked, stop
1071 * retransmit timer, otherwise restart timer
1072 * using current (possibly backed-off) value.
1073 * If process is waiting for space,
1074 * wakeup/selwakeup/signal. If data
1075 * are ready to send, let tcp_output
1076 * decide between more output or persist.
1077 */
1078 if (tp->snd_una == tp->snd_max)
1079 tp->t_timer[TCPT_REXMT] = 0;
1080 else if (tp->t_timer[TCPT_PERSIST] == 0)
1081 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1082
1083 if (so->so_snd.sb_cc)
1084 (void) tcp_output(tp);
1085 sowwakeup(so);
1086 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
1087 return;
1088 }
1089 } else if (th->th_ack == tp->snd_una &&
1090 LIST_EMPTY(&tp->t_segq) &&
1091 tlen <= sbspace(&so->so_rcv)) {
1092 /*
1093 * this is a pure, in-sequence data packet
1094 * with nothing on the reassembly queue and
1095 * we have enough buffer space to take it.
1096 */
1097 ++tcpstat.tcps_preddat;
1098 tp->rcv_nxt += tlen;
1099 tcpstat.tcps_rcvpack++;
1100 tcpstat.tcps_rcvbyte += tlen;
1101 ND6_HINT(tp); /* some progress has been done */
1102 /*
1103 * Add data to socket buffer.
1104 */
1105 m_adj(m, drop_hdrlen); /* delayed header drop */
1106 sbappend(&so->so_rcv, m);
1107 #if INET6
1108 if (isipv6) {
1109 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
1110 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
1111 th->th_seq, th->th_ack, th->th_win);
1112 }
1113 else
1114 #endif
1115 {
1116 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
1117 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
1118 th->th_seq, th->th_ack, th->th_win);
1119 }
1120 if (tcp_delack_enabled) {
1121 TCP_DELACK_BITSET(tp->t_inpcb->hash_element);
1122 tp->t_flags |= TF_DELACK;
1123 } else {
1124 tp->t_flags |= TF_ACKNOW;
1125 tcp_output(tp);
1126 }
1127 sorwakeup(so);
1128 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
1129 return;
1130 }
1131 }
1132
1133 /*
1134 * Calculate amount of space in receive window,
1135 * and then do TCP input processing.
1136 * Receive window is amount of space in rcv queue,
1137 * but not less than advertised window.
1138 */
1139 { int win;
1140
1141 win = sbspace(&so->so_rcv);
1142 if (win < 0)
1143 win = 0;
1144 else { /* clip rcv window to 4K for modems */
1145 if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0)
1146 win = min(win, slowlink_wsize);
1147 }
1148 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1149 }
1150
1151 switch (tp->t_state) {
1152
1153 /*
1154 * If the state is LISTEN then ignore segment if it contains an RST.
1155 * If the segment contains an ACK then it is bad and send a RST.
1156 * If it does not contain a SYN then it is not interesting; drop it.
1157 * If it is from this socket, drop it, it must be forged.
1158 * Don't bother responding if the destination was a broadcast.
1159 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
1160 * tp->iss, and send a segment:
1161 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1162 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
1163 * Fill in remote peer address fields if not previously specified.
1164 * Enter SYN_RECEIVED state, and process any other fields of this
1165 * segment in this state.
1166 */
1167 case TCPS_LISTEN: {
1168 register struct sockaddr_in *sin;
1169 #if INET6
1170 register struct sockaddr_in6 *sin6;
1171 #endif
1172
1173 if (thflags & TH_RST)
1174 goto drop;
1175 if (thflags & TH_ACK) {
1176 rstreason = BANDLIM_RST_OPENPORT;
1177 goto dropwithreset;
1178 }
1179 if ((thflags & TH_SYN) == 0)
1180 goto drop;
1181 if (th->th_dport == th->th_sport) {
1182 #if INET6
1183 if (isipv6) {
1184 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1185 &ip6->ip6_src))
1186 goto drop;
1187 } else
1188 #endif /* INET6 */
1189 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1190 goto drop;
1191 }
1192 /*
1193 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1194 * in_broadcast() should never return true on a received
1195 * packet with M_BCAST not set.
1196 *
1197 * Packets with a multicast source address should also
1198 * be discarded.
1199 */
1200 if (m->m_flags & (M_BCAST|M_MCAST))
1201 goto drop;
1202 #if INET6
1203 if (isipv6) {
1204 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1205 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1206 goto drop;
1207 } else
1208 #endif
1209 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1210 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1211 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1212 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1213 goto drop;
1214 #if INET6
1215 if (isipv6) {
1216 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6,
1217 M_SONAME, M_NOWAIT);
1218 if (sin6 == NULL)
1219 goto drop;
1220 bzero(sin6, sizeof(*sin6));
1221 sin6->sin6_family = AF_INET6;
1222 sin6->sin6_len = sizeof(*sin6);
1223 sin6->sin6_addr = ip6->ip6_src;
1224 sin6->sin6_port = th->th_sport;
1225 laddr6 = inp->in6p_laddr;
1226 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1227 inp->in6p_laddr = ip6->ip6_dst;
1228 if (in6_pcbconnect(inp, (struct sockaddr *)sin6,
1229 proc0)) {
1230 inp->in6p_laddr = laddr6;
1231 FREE(sin6, M_SONAME);
1232 goto drop;
1233 }
1234 FREE(sin6, M_SONAME);
1235 } else
1236 #endif
1237 {
1238 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
1239 M_NOWAIT);
1240 if (sin == NULL)
1241 goto drop;
1242 sin->sin_family = AF_INET;
1243 sin->sin_len = sizeof(*sin);
1244 sin->sin_addr = ip->ip_src;
1245 sin->sin_port = th->th_sport;
1246 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
1247 laddr = inp->inp_laddr;
1248 if (inp->inp_laddr.s_addr == INADDR_ANY)
1249 inp->inp_laddr = ip->ip_dst;
1250 if (in_pcbconnect(inp, (struct sockaddr *)sin, proc0)) {
1251 inp->inp_laddr = laddr;
1252 FREE(sin, M_SONAME);
1253 goto drop;
1254 }
1255 FREE(sin, M_SONAME);
1256 }
1257 if ((taop = tcp_gettaocache(inp)) == NULL) {
1258 taop = &tao_noncached;
1259 bzero(taop, sizeof(*taop));
1260 }
1261 tcp_dooptions(tp, optp, optlen, th, &to);
1262 if (iss)
1263 tp->iss = iss;
1264 else {
1265 tp->iss = tcp_new_isn(tp);
1266 }
1267 tp->irs = th->th_seq;
1268 tcp_sendseqinit(tp);
1269 tcp_rcvseqinit(tp);
1270 tp->snd_recover = tp->snd_una;
1271 /*
1272 * Initialization of the tcpcb for transaction;
1273 * set SND.WND = SEG.WND,
1274 * initialize CCsend and CCrecv.
1275 */
1276 tp->snd_wnd = tiwin; /* initial send-window */
1277 tp->cc_send = CC_INC(tcp_ccgen);
1278 tp->cc_recv = to.to_cc;
1279 /*
1280 * Perform TAO test on incoming CC (SEG.CC) option, if any.
1281 * - compare SEG.CC against cached CC from the same host,
1282 * if any.
1283 * - if SEG.CC > chached value, SYN must be new and is accepted
1284 * immediately: save new CC in the cache, mark the socket
1285 * connected, enter ESTABLISHED state, turn on flag to
1286 * send a SYN in the next segment.
1287 * A virtual advertised window is set in rcv_adv to
1288 * initialize SWS prevention. Then enter normal segment
1289 * processing: drop SYN, process data and FIN.
1290 * - otherwise do a normal 3-way handshake.
1291 */
1292 if ((to.to_flag & TOF_CC) != 0) {
1293 if (((tp->t_flags & TF_NOPUSH) != 0) &&
1294 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
1295
1296 taop->tao_cc = to.to_cc;
1297
1298 tp->t_state = TCPS_ESTABLISHED;
1299
1300 /*
1301 * If there is a FIN, or if there is data and the
1302 * connection is local, then delay SYN,ACK(SYN) in
1303 * the hope of piggy-backing it on a response
1304 * segment. Otherwise must send ACK now in case
1305 * the other side is slow starting.
1306 */
1307 if (tcp_delack_enabled && ((thflags & TH_FIN) ||
1308 (tlen != 0 &&
1309 #if INET6
1310 (isipv6 && in6_localaddr(&inp->in6p_faddr))
1311 ||
1312 (!isipv6 &&
1313 #endif /* INET6 */
1314 in_localaddr(inp->inp_faddr)
1315 #if INET6
1316 )
1317 #endif /* INET6 */
1318 ))) {
1319 TCP_DELACK_BITSET(tp->t_inpcb->hash_element);
1320 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1321 }
1322 else
1323 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1324
1325 /*
1326 * Limit the `virtual advertised window' to TCP_MAXWIN
1327 * here. Even if we requested window scaling, it will
1328 * become effective only later when our SYN is acked.
1329 */
1330 if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) /* clip window size for for slow link */
1331 tp->rcv_adv += min(tp->rcv_wnd, slowlink_wsize);
1332 else
1333 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
1334 tcpstat.tcps_connects++;
1335 soisconnected(so);
1336 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
1337 dropsocket = 0; /* committed to socket */
1338 tcpstat.tcps_accepts++;
1339 goto trimthenstep6;
1340 }
1341 /* else do standard 3-way handshake */
1342 } else {
1343 /*
1344 * No CC option, but maybe CC.NEW:
1345 * invalidate cached value.
1346 */
1347 taop->tao_cc = 0;
1348 }
1349 /*
1350 * TAO test failed or there was no CC option,
1351 * do a standard 3-way handshake.
1352 */
1353 tp->t_flags |= TF_ACKNOW;
1354 tp->t_state = TCPS_SYN_RECEIVED;
1355 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
1356 dropsocket = 0; /* committed to socket */
1357 tcpstat.tcps_accepts++;
1358 goto trimthenstep6;
1359 }
1360
1361 /*
1362 * If the state is SYN_RECEIVED:
1363 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1364 */
1365 case TCPS_SYN_RECEIVED:
1366 if ((thflags & TH_ACK) &&
1367 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1368 SEQ_GT(th->th_ack, tp->snd_max))) {
1369 rstreason = BANDLIM_RST_OPENPORT;
1370 goto dropwithreset;
1371 }
1372 break;
1373
1374 /*
1375 * If the state is SYN_SENT:
1376 * if seg contains an ACK, but not for our SYN, drop the input.
1377 * if seg contains a RST, then drop the connection.
1378 * if seg does not contain SYN, then drop it.
1379 * Otherwise this is an acceptable SYN segment
1380 * initialize tp->rcv_nxt and tp->irs
1381 * if seg contains ack then advance tp->snd_una
1382 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1383 * arrange for segment to be acked (eventually)
1384 * continue processing rest of data/controls, beginning with URG
1385 */
1386 case TCPS_SYN_SENT:
1387 if ((taop = tcp_gettaocache(inp)) == NULL) {
1388 taop = &tao_noncached;
1389 bzero(taop, sizeof(*taop));
1390 }
1391
1392 if ((thflags & TH_ACK) &&
1393 (SEQ_LEQ(th->th_ack, tp->iss) ||
1394 SEQ_GT(th->th_ack, tp->snd_max))) {
1395 /*
1396 * If we have a cached CCsent for the remote host,
1397 * hence we haven't just crashed and restarted,
1398 * do not send a RST. This may be a retransmission
1399 * from the other side after our earlier ACK was lost.
1400 * Our new SYN, when it arrives, will serve as the
1401 * needed ACK.
1402 */
1403 if (taop->tao_ccsent != 0)
1404 goto drop;
1405 else {
1406 rstreason = BANDLIM_UNLIMITED;
1407 goto dropwithreset;
1408 }
1409 }
1410 if (thflags & TH_RST) {
1411 if (thflags & TH_ACK) {
1412 tp = tcp_drop(tp, ECONNREFUSED);
1413 postevent(so, 0, EV_RESET);
1414 }
1415 goto drop;
1416 }
1417 if ((thflags & TH_SYN) == 0)
1418 goto drop;
1419 tp->snd_wnd = th->th_win; /* initial send window */
1420 tp->cc_recv = to.to_cc; /* foreign CC */
1421
1422 tp->irs = th->th_seq;
1423 tcp_rcvseqinit(tp);
1424 if (thflags & TH_ACK) {
1425 /*
1426 * Our SYN was acked. If segment contains CC.ECHO
1427 * option, check it to make sure this segment really
1428 * matches our SYN. If not, just drop it as old
1429 * duplicate, but send an RST if we're still playing
1430 * by the old rules. If no CC.ECHO option, make sure
1431 * we don't get fooled into using T/TCP.
1432 */
1433 if (to.to_flag & TOF_CCECHO) {
1434 if (tp->cc_send != to.to_ccecho) {
1435 if (taop->tao_ccsent != 0)
1436 goto drop;
1437 else {
1438 rstreason = BANDLIM_UNLIMITED;
1439 goto dropwithreset;
1440 }
1441 }
1442 } else
1443 tp->t_flags &= ~TF_RCVD_CC;
1444 tcpstat.tcps_connects++;
1445 soisconnected(so);
1446 /* Do window scaling on this connection? */
1447 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1448 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1449 tp->snd_scale = tp->requested_s_scale;
1450 tp->rcv_scale = tp->request_r_scale;
1451 }
1452 /* Segment is acceptable, update cache if undefined. */
1453 if (taop->tao_ccsent == 0)
1454 taop->tao_ccsent = to.to_ccecho;
1455
1456 tp->rcv_adv += tp->rcv_wnd;
1457 tp->snd_una++; /* SYN is acked */
1458 /*
1459 * If there's data, delay ACK; if there's also a FIN
1460 * ACKNOW will be turned on later.
1461 */
1462 if (tcp_delack_enabled && tlen != 0) {
1463 TCP_DELACK_BITSET(tp->t_inpcb->hash_element);
1464 tp->t_flags |= TF_DELACK;
1465 }
1466 else
1467 tp->t_flags |= TF_ACKNOW;
1468 /*
1469 * Received <SYN,ACK> in SYN_SENT[*] state.
1470 * Transitions:
1471 * SYN_SENT --> ESTABLISHED
1472 * SYN_SENT* --> FIN_WAIT_1
1473 */
1474 if (tp->t_flags & TF_NEEDFIN) {
1475 tp->t_state = TCPS_FIN_WAIT_1;
1476 tp->t_flags &= ~TF_NEEDFIN;
1477 thflags &= ~TH_SYN;
1478 } else {
1479 tp->t_state = TCPS_ESTABLISHED;
1480 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1481 }
1482 } else {
1483 /*
1484 * Received initial SYN in SYN-SENT[*] state => simul-
1485 * taneous open. If segment contains CC option and there is
1486 * a cached CC, apply TAO test; if it succeeds, connection is
1487 * half-synchronized. Otherwise, do 3-way handshake:
1488 * SYN-SENT -> SYN-RECEIVED
1489 * SYN-SENT* -> SYN-RECEIVED*
1490 * If there was no CC option, clear cached CC value.
1491 */
1492 tp->t_flags |= TF_ACKNOW;
1493 tp->t_timer[TCPT_REXMT] = 0;
1494 if (to.to_flag & TOF_CC) {
1495 if (taop->tao_cc != 0 &&
1496 CC_GT(to.to_cc, taop->tao_cc)) {
1497 /*
1498 * update cache and make transition:
1499 * SYN-SENT -> ESTABLISHED*
1500 * SYN-SENT* -> FIN-WAIT-1*
1501 */
1502 taop->tao_cc = to.to_cc;
1503 if (tp->t_flags & TF_NEEDFIN) {
1504 tp->t_state = TCPS_FIN_WAIT_1;
1505 tp->t_flags &= ~TF_NEEDFIN;
1506 } else {
1507 tp->t_state = TCPS_ESTABLISHED;
1508 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1509 }
1510 tp->t_flags |= TF_NEEDSYN;
1511 } else
1512 tp->t_state = TCPS_SYN_RECEIVED;
1513 } else {
1514 /* CC.NEW or no option => invalidate cache */
1515 taop->tao_cc = 0;
1516 tp->t_state = TCPS_SYN_RECEIVED;
1517 }
1518 }
1519
1520 trimthenstep6:
1521 /*
1522 * Advance th->th_seq to correspond to first data byte.
1523 * If data, trim to stay within window,
1524 * dropping FIN if necessary.
1525 */
1526 th->th_seq++;
1527 if (tlen > tp->rcv_wnd) {
1528 todrop = tlen - tp->rcv_wnd;
1529 m_adj(m, -todrop);
1530 tlen = tp->rcv_wnd;
1531 thflags &= ~TH_FIN;
1532 tcpstat.tcps_rcvpackafterwin++;
1533 tcpstat.tcps_rcvbyteafterwin += todrop;
1534 }
1535 tp->snd_wl1 = th->th_seq - 1;
1536 tp->rcv_up = th->th_seq;
1537 /*
1538 * Client side of transaction: already sent SYN and data.
1539 * If the remote host used T/TCP to validate the SYN,
1540 * our data will be ACK'd; if so, enter normal data segment
1541 * processing in the middle of step 5, ack processing.
1542 * Otherwise, goto step 6.
1543 */
1544 if (thflags & TH_ACK)
1545 goto process_ACK;
1546 goto step6;
1547 /*
1548 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1549 * if segment contains a SYN and CC [not CC.NEW] option:
1550 * if state == TIME_WAIT and connection duration > MSL,
1551 * drop packet and send RST;
1552 *
1553 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1554 * ack the FIN (and data) in retransmission queue.
1555 * Complete close and delete TCPCB. Then reprocess
1556 * segment, hoping to find new TCPCB in LISTEN state;
1557 *
1558 * else must be old SYN; drop it.
1559 * else do normal processing.
1560 */
1561 case TCPS_LAST_ACK:
1562 case TCPS_CLOSING:
1563 case TCPS_TIME_WAIT:
1564 if ((thflags & TH_SYN) &&
1565 (to.to_flag & TOF_CC) && tp->cc_recv != 0) {
1566 if (tp->t_state == TCPS_TIME_WAIT &&
1567 tp->t_starttime > tcp_msl) {
1568 rstreason = BANDLIM_UNLIMITED;
1569 goto dropwithreset;
1570 }
1571 if (CC_GT(to.to_cc, tp->cc_recv)) {
1572 tp = tcp_close(tp);
1573 goto findpcb;
1574 }
1575 else
1576 goto drop;
1577 }
1578 break; /* continue normal processing */
1579 }
1580
1581 /*
1582 * States other than LISTEN or SYN_SENT.
1583 * First check the RST flag and sequence number since reset segments
1584 * are exempt from the timestamp and connection count tests. This
1585 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1586 * below which allowed reset segments in half the sequence space
1587 * to fall though and be processed (which gives forged reset
1588 * segments with a random sequence number a 50 percent chance of
1589 * killing a connection).
1590 * Then check timestamp, if present.
1591 * Then check the connection count, if present.
1592 * Then check that at least some bytes of segment are within
1593 * receive window. If segment begins before rcv_nxt,
1594 * drop leading data (and SYN); if nothing left, just ack.
1595 *
1596 *
1597 * If the RST bit is set, check the sequence number to see
1598 * if this is a valid reset segment.
1599 * RFC 793 page 37:
1600 * In all states except SYN-SENT, all reset (RST) segments
1601 * are validated by checking their SEQ-fields. A reset is
1602 * valid if its sequence number is in the window.
1603 * Note: this does not take into account delayed ACKs, so
1604 * we should test against last_ack_sent instead of rcv_nxt.
1605 * The sequence number in the reset segment is normally an
1606 * echo of our outgoing acknowlegement numbers, but some hosts
1607 * send a reset with the sequence number at the rightmost edge
1608 * of our receive window, and we have to handle this case.
1609 * If we have multiple segments in flight, the intial reset
1610 * segment sequence numbers will be to the left of last_ack_sent,
1611 * but they will eventually catch up.
1612 * In any case, it never made sense to trim reset segments to
1613 * fit the receive window since RFC 1122 says:
1614 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1615 *
1616 * A TCP SHOULD allow a received RST segment to include data.
1617 *
1618 * DISCUSSION
1619 * It has been suggested that a RST segment could contain
1620 * ASCII text that encoded and explained the cause of the
1621 * RST. No standard has yet been established for such
1622 * data.
1623 *
1624 * If the reset segment passes the sequence number test examine
1625 * the state:
1626 * SYN_RECEIVED STATE:
1627 * If passive open, return to LISTEN state.
1628 * If active open, inform user that connection was refused.
1629 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1630 * Inform user that connection was reset, and close tcb.
1631 * CLOSING, LAST_ACK STATES:
1632 * Close the tcb.
1633 * TIME_WAIT STATE:
1634 * Drop the segment - see Stevens, vol. 2, p. 964 and
1635 * RFC 1337.
1636 */
1637 if (thflags & TH_RST) {
1638 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1639 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1640 switch (tp->t_state) {
1641
1642 case TCPS_SYN_RECEIVED:
1643 so->so_error = ECONNREFUSED;
1644 goto close;
1645
1646 case TCPS_ESTABLISHED:
1647 case TCPS_FIN_WAIT_1:
1648 case TCPS_CLOSE_WAIT:
1649 /*
1650 Drop through ...
1651 */
1652 case TCPS_FIN_WAIT_2:
1653 so->so_error = ECONNRESET;
1654 close:
1655 postevent(so, 0, EV_RESET);
1656 tp->t_state = TCPS_CLOSED;
1657 tcpstat.tcps_drops++;
1658 tp = tcp_close(tp);
1659 break;
1660
1661 case TCPS_CLOSING:
1662 case TCPS_LAST_ACK:
1663 tp = tcp_close(tp);
1664 break;
1665
1666 case TCPS_TIME_WAIT:
1667 break;
1668 }
1669 }
1670 goto drop;
1671 }
1672
1673 /*
1674 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1675 * and it's less than ts_recent, drop it.
1676 */
1677 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent &&
1678 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1679
1680 /* Check to see if ts_recent is over 24 days old. */
1681 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1682 /*
1683 * Invalidate ts_recent. If this segment updates
1684 * ts_recent, the age will be reset later and ts_recent
1685 * will get a valid value. If it does not, setting
1686 * ts_recent to zero will at least satisfy the
1687 * requirement that zero be placed in the timestamp
1688 * echo reply when ts_recent isn't valid. The
1689 * age isn't reset until we get a valid ts_recent
1690 * because we don't want out-of-order segments to be
1691 * dropped when ts_recent is old.
1692 */
1693 tp->ts_recent = 0;
1694 } else {
1695 tcpstat.tcps_rcvduppack++;
1696 tcpstat.tcps_rcvdupbyte += tlen;
1697 tcpstat.tcps_pawsdrop++;
1698 goto dropafterack;
1699 }
1700 }
1701
1702 /*
1703 * T/TCP mechanism
1704 * If T/TCP was negotiated and the segment doesn't have CC,
1705 * or if its CC is wrong then drop the segment.
1706 * RST segments do not have to comply with this.
1707 */
1708 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1709 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1710 goto dropafterack;
1711
1712 /*
1713 * In the SYN-RECEIVED state, validate that the packet belongs to
1714 * this connection before trimming the data to fit the receive
1715 * window. Check the sequence number versus IRS since we know
1716 * the sequence numbers haven't wrapped. This is a partial fix
1717 * for the "LAND" DoS attack.
1718 */
1719 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1720 rstreason = BANDLIM_RST_OPENPORT;
1721 goto dropwithreset;
1722 }
1723
1724 todrop = tp->rcv_nxt - th->th_seq;
1725 if (todrop > 0) {
1726 if (thflags & TH_SYN) {
1727 thflags &= ~TH_SYN;
1728 th->th_seq++;
1729 if (th->th_urp > 1)
1730 th->th_urp--;
1731 else
1732 thflags &= ~TH_URG;
1733 todrop--;
1734 }
1735 /*
1736 * Following if statement from Stevens, vol. 2, p. 960.
1737 */
1738 if (todrop > tlen
1739 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1740 /*
1741 * Any valid FIN must be to the left of the window.
1742 * At this point the FIN must be a duplicate or out
1743 * of sequence; drop it.
1744 */
1745 thflags &= ~TH_FIN;
1746
1747 /*
1748 * Send an ACK to resynchronize and drop any data.
1749 * But keep on processing for RST or ACK.
1750 */
1751 tp->t_flags |= TF_ACKNOW;
1752 todrop = tlen;
1753 tcpstat.tcps_rcvduppack++;
1754 tcpstat.tcps_rcvdupbyte += todrop;
1755 } else {
1756 tcpstat.tcps_rcvpartduppack++;
1757 tcpstat.tcps_rcvpartdupbyte += todrop;
1758 }
1759 drop_hdrlen += todrop; /* drop from the top afterwards */
1760 th->th_seq += todrop;
1761 tlen -= todrop;
1762 if (th->th_urp > todrop)
1763 th->th_urp -= todrop;
1764 else {
1765 thflags &= ~TH_URG;
1766 th->th_urp = 0;
1767 }
1768 }
1769
1770 /*
1771 * If new data are received on a connection after the
1772 * user processes are gone, then RST the other end.
1773 */
1774 if ((so->so_state & SS_NOFDREF) &&
1775 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1776 tp = tcp_close(tp);
1777 tcpstat.tcps_rcvafterclose++;
1778 rstreason = BANDLIM_UNLIMITED;
1779 goto dropwithreset;
1780 }
1781
1782 /*
1783 * If segment ends after window, drop trailing data
1784 * (and PUSH and FIN); if nothing left, just ACK.
1785 */
1786 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1787 if (todrop > 0) {
1788 tcpstat.tcps_rcvpackafterwin++;
1789 if (todrop >= tlen) {
1790 tcpstat.tcps_rcvbyteafterwin += tlen;
1791 /*
1792 * If a new connection request is received
1793 * while in TIME_WAIT, drop the old connection
1794 * and start over if the sequence numbers
1795 * are above the previous ones.
1796 */
1797 if (thflags & TH_SYN &&
1798 tp->t_state == TCPS_TIME_WAIT &&
1799 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1800 iss = tcp_new_isn(tp);
1801 tp = tcp_close(tp);
1802 goto findpcb;
1803 }
1804 /*
1805 * If window is closed can only take segments at
1806 * window edge, and have to drop data and PUSH from
1807 * incoming segments. Continue processing, but
1808 * remember to ack. Otherwise, drop segment
1809 * and ack.
1810 */
1811 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1812 tp->t_flags |= TF_ACKNOW;
1813 tcpstat.tcps_rcvwinprobe++;
1814 } else
1815 goto dropafterack;
1816 } else
1817 tcpstat.tcps_rcvbyteafterwin += todrop;
1818 m_adj(m, -todrop);
1819 tlen -= todrop;
1820 thflags &= ~(TH_PUSH|TH_FIN);
1821 }
1822
1823 /*
1824 * If last ACK falls within this segment's sequence numbers,
1825 * record its timestamp.
1826 * NOTE that the test is modified according to the latest
1827 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1828 */
1829 if ((to.to_flag & TOF_TS) != 0 &&
1830 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1831 tp->ts_recent_age = tcp_now;
1832 tp->ts_recent = to.to_tsval;
1833 }
1834
1835 /*
1836 * If a SYN is in the window, then this is an
1837 * error and we send an RST and drop the connection.
1838 */
1839 if (thflags & TH_SYN) {
1840 tp = tcp_drop(tp, ECONNRESET);
1841 rstreason = BANDLIM_UNLIMITED;
1842 postevent(so, 0, EV_RESET);
1843 goto dropwithreset;
1844 }
1845
1846 /*
1847 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1848 * flag is on (half-synchronized state), then queue data for
1849 * later processing; else drop segment and return.
1850 */
1851 if ((thflags & TH_ACK) == 0) {
1852 if (tp->t_state == TCPS_SYN_RECEIVED ||
1853 (tp->t_flags & TF_NEEDSYN))
1854 goto step6;
1855 else
1856 goto drop;
1857 }
1858
1859 /*
1860 * Ack processing.
1861 */
1862 switch (tp->t_state) {
1863
1864 /*
1865 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1866 * ESTABLISHED state and continue processing.
1867 * The ACK was checked above.
1868 */
1869 case TCPS_SYN_RECEIVED:
1870
1871 tcpstat.tcps_connects++;
1872 soisconnected(so);
1873
1874 /* Do window scaling? */
1875 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1876 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1877 tp->snd_scale = tp->requested_s_scale;
1878 tp->rcv_scale = tp->request_r_scale;
1879 }
1880 /*
1881 * Upon successful completion of 3-way handshake,
1882 * update cache.CC if it was undefined, pass any queued
1883 * data to the user, and advance state appropriately.
1884 */
1885 if ((taop = tcp_gettaocache(inp)) != NULL &&
1886 taop->tao_cc == 0)
1887 taop->tao_cc = tp->cc_recv;
1888
1889 /*
1890 * Make transitions:
1891 * SYN-RECEIVED -> ESTABLISHED
1892 * SYN-RECEIVED* -> FIN-WAIT-1
1893 */
1894 if (tp->t_flags & TF_NEEDFIN) {
1895 tp->t_state = TCPS_FIN_WAIT_1;
1896 tp->t_flags &= ~TF_NEEDFIN;
1897 } else {
1898 tp->t_state = TCPS_ESTABLISHED;
1899 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1900 }
1901 /*
1902 * If segment contains data or ACK, will call tcp_reass()
1903 * later; if not, do so now to pass queued data to user.
1904 */
1905 if (tlen == 0 && (thflags & TH_FIN) == 0)
1906 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1907 (struct mbuf *)0);
1908 tp->snd_wl1 = th->th_seq - 1;
1909 /* fall into ... */
1910
1911 /*
1912 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1913 * ACKs. If the ack is in the range
1914 * tp->snd_una < th->th_ack <= tp->snd_max
1915 * then advance tp->snd_una to th->th_ack and drop
1916 * data from the retransmission queue. If this ACK reflects
1917 * more up to date window information we update our window information.
1918 */
1919 case TCPS_ESTABLISHED:
1920 case TCPS_FIN_WAIT_1:
1921 case TCPS_FIN_WAIT_2:
1922 case TCPS_CLOSE_WAIT:
1923 case TCPS_CLOSING:
1924 case TCPS_LAST_ACK:
1925 case TCPS_TIME_WAIT:
1926
1927 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1928 if (tlen == 0 && tiwin == tp->snd_wnd) {
1929 tcpstat.tcps_rcvdupack++;
1930 /*
1931 * If we have outstanding data (other than
1932 * a window probe), this is a completely
1933 * duplicate ack (ie, window info didn't
1934 * change), the ack is the biggest we've
1935 * seen and we've seen exactly our rexmt
1936 * threshhold of them, assume a packet
1937 * has been dropped and retransmit it.
1938 * Kludge snd_nxt & the congestion
1939 * window so we send only this one
1940 * packet.
1941 *
1942 * We know we're losing at the current
1943 * window size so do congestion avoidance
1944 * (set ssthresh to half the current window
1945 * and pull our congestion window back to
1946 * the new ssthresh).
1947 *
1948 * Dup acks mean that packets have left the
1949 * network (they're now cached at the receiver)
1950 * so bump cwnd by the amount in the receiver
1951 * to keep a constant cwnd packets in the
1952 * network.
1953 */
1954 if (tp->t_timer[TCPT_REXMT] == 0 ||
1955 th->th_ack != tp->snd_una)
1956 tp->t_dupacks = 0;
1957 else if (++tp->t_dupacks == tcprexmtthresh) {
1958 tcp_seq onxt = tp->snd_nxt;
1959 u_int win =
1960 min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1961 tp->t_maxseg;
1962 if (tcp_do_newreno && SEQ_LT(th->th_ack,
1963 tp->snd_recover)) {
1964 /* False retransmit, should not
1965 * cut window
1966 */
1967 tp->snd_cwnd += tp->t_maxseg;
1968 tp->t_dupacks = 0;
1969 (void) tcp_output(tp);
1970 goto drop;
1971 }
1972 if (win < 2)
1973 win = 2;
1974 tp->snd_ssthresh = win * tp->t_maxseg;
1975 tp->snd_recover = tp->snd_max;
1976 tp->t_timer[TCPT_REXMT] = 0;
1977 tp->t_rtttime = 0;
1978 tp->snd_nxt = th->th_ack;
1979 tp->snd_cwnd = tp->t_maxseg;
1980 (void) tcp_output(tp);
1981 tp->snd_cwnd = tp->snd_ssthresh +
1982 tp->t_maxseg * tp->t_dupacks;
1983 if (SEQ_GT(onxt, tp->snd_nxt))
1984 tp->snd_nxt = onxt;
1985 goto drop;
1986 } else if (tp->t_dupacks > tcprexmtthresh) {
1987 tp->snd_cwnd += tp->t_maxseg;
1988 (void) tcp_output(tp);
1989 goto drop;
1990 }
1991 } else
1992 tp->t_dupacks = 0;
1993 break;
1994 }
1995 /*
1996 * If the congestion window was inflated to account
1997 * for the other side's cached packets, retract it.
1998 */
1999 if (tcp_do_newreno == 0) {
2000 if (tp->t_dupacks >= tcprexmtthresh &&
2001 tp->snd_cwnd > tp->snd_ssthresh)
2002 tp->snd_cwnd = tp->snd_ssthresh;
2003 tp->t_dupacks = 0;
2004 } else if (tp->t_dupacks >= tcprexmtthresh &&
2005 !tcp_newreno(tp, th)) {
2006 /*
2007 * Window inflation should have left us with approx.
2008 * snd_ssthresh outstanding data. But in case we
2009 * would be inclined to send a burst, better to do
2010 * it via the slow start mechanism.
2011 */
2012 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max))
2013 tp->snd_cwnd =
2014 tp->snd_max - th->th_ack + tp->t_maxseg;
2015 else
2016 tp->snd_cwnd = tp->snd_ssthresh;
2017 tp->t_dupacks = 0;
2018 }
2019
2020 if (tp->t_dupacks < tcprexmtthresh)
2021 tp->t_dupacks = 0;
2022
2023 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2024 tcpstat.tcps_rcvacktoomuch++;
2025 goto dropafterack;
2026 }
2027 /*
2028 * If we reach this point, ACK is not a duplicate,
2029 * i.e., it ACKs something we sent.
2030 */
2031 if (tp->t_flags & TF_NEEDSYN) {
2032 /*
2033 * T/TCP: Connection was half-synchronized, and our
2034 * SYN has been ACK'd (so connection is now fully
2035 * synchronized). Go to non-starred state,
2036 * increment snd_una for ACK of SYN, and check if
2037 * we can do window scaling.
2038 */
2039 tp->t_flags &= ~TF_NEEDSYN;
2040 tp->snd_una++;
2041 /* Do window scaling? */
2042 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2043 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2044 tp->snd_scale = tp->requested_s_scale;
2045 tp->rcv_scale = tp->request_r_scale;
2046 }
2047 }
2048
2049 process_ACK:
2050 acked = th->th_ack - tp->snd_una;
2051 tcpstat.tcps_rcvackpack++;
2052 tcpstat.tcps_rcvackbyte += acked;
2053
2054 /*
2055 * If we just performed our first retransmit, and the ACK
2056 * arrives within our recovery window, then it was a mistake
2057 * to do the retransmit in the first place. Recover our
2058 * original cwnd and ssthresh, and proceed to transmit where
2059 * we left off.
2060 */
2061 if (tp->t_rxtshift == 1 && tcp_now < tp->t_badrxtwin) {
2062 tp->snd_cwnd = tp->snd_cwnd_prev;
2063 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2064 tp->snd_nxt = tp->snd_max;
2065 tp->t_badrxtwin = 0; /* XXX probably not required */
2066 }
2067
2068 /*
2069 * If we have a timestamp reply, update smoothed
2070 * round trip time. If no timestamp is present but
2071 * transmit timer is running and timed sequence
2072 * number was acked, update smoothed round trip time.
2073 * Since we now have an rtt measurement, cancel the
2074 * timer backoff (cf., Phil Karn's retransmit alg.).
2075 * Recompute the initial retransmit timer.
2076 * Also makes sure we have a valid time stamp in hand
2077 */
2078 if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0))
2079 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
2080 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2081 tcp_xmit_timer(tp, tp->t_rtttime);
2082
2083 /*
2084 * If all outstanding data is acked, stop retransmit
2085 * timer and remember to restart (more output or persist).
2086 * If there is more data to be acked, restart retransmit
2087 * timer, using current (possibly backed-off) value.
2088 */
2089 if (th->th_ack == tp->snd_max) {
2090 tp->t_timer[TCPT_REXMT] = 0;
2091 needoutput = 1;
2092 } else if (tp->t_timer[TCPT_PERSIST] == 0)
2093 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
2094
2095 /*
2096 * If no data (only SYN) was ACK'd,
2097 * skip rest of ACK processing.
2098 */
2099 if (acked == 0)
2100 goto step6;
2101
2102 /*
2103 * When new data is acked, open the congestion window.
2104 * If the window gives us less than ssthresh packets
2105 * in flight, open exponentially (maxseg per packet).
2106 * Otherwise open linearly: maxseg per window
2107 * (maxseg^2 / cwnd per packet).
2108 */
2109 {
2110 register u_int cw = tp->snd_cwnd;
2111 register u_int incr = tp->t_maxseg;
2112
2113 if (cw > tp->snd_ssthresh)
2114 incr = incr * incr / cw;
2115 /*
2116 * If t_dupacks != 0 here, it indicates that we are still
2117 * in NewReno fast recovery mode, so we leave the congestion
2118 * window alone.
2119 */
2120 if (tcp_do_newreno == 0 || tp->t_dupacks == 0)
2121 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale);
2122 }
2123 if (acked > so->so_snd.sb_cc) {
2124 tp->snd_wnd -= so->so_snd.sb_cc;
2125 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2126 ourfinisacked = 1;
2127 } else {
2128 sbdrop(&so->so_snd, acked);
2129 tp->snd_wnd -= acked;
2130 ourfinisacked = 0;
2131 }
2132 tp->snd_una = th->th_ack;
2133 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2134 tp->snd_nxt = tp->snd_una;
2135 sowwakeup(so);
2136
2137 switch (tp->t_state) {
2138
2139 /*
2140 * In FIN_WAIT_1 STATE in addition to the processing
2141 * for the ESTABLISHED state if our FIN is now acknowledged
2142 * then enter FIN_WAIT_2.
2143 */
2144 case TCPS_FIN_WAIT_1:
2145 if (ourfinisacked) {
2146 /*
2147 * If we can't receive any more
2148 * data, then closing user can proceed.
2149 * Starting the timer is contrary to the
2150 * specification, but if we don't get a FIN
2151 * we'll hang forever.
2152 */
2153 if (so->so_state & SS_CANTRCVMORE) {
2154 soisdisconnected(so);
2155 tp->t_timer[TCPT_2MSL] = tcp_maxidle;
2156 }
2157 add_to_time_wait(tp);
2158 tp->t_state = TCPS_FIN_WAIT_2;
2159 }
2160 break;
2161
2162 /*
2163 * In CLOSING STATE in addition to the processing for
2164 * the ESTABLISHED state if the ACK acknowledges our FIN
2165 * then enter the TIME-WAIT state, otherwise ignore
2166 * the segment.
2167 */
2168 case TCPS_CLOSING:
2169 if (ourfinisacked) {
2170 tp->t_state = TCPS_TIME_WAIT;
2171 tcp_canceltimers(tp);
2172 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2173 if (tp->cc_recv != 0 &&
2174 tp->t_starttime < tcp_msl)
2175 tp->t_timer[TCPT_2MSL] =
2176 tp->t_rxtcur * TCPTV_TWTRUNC;
2177 else
2178 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2179 add_to_time_wait(tp);
2180 soisdisconnected(so);
2181 }
2182 break;
2183
2184 /*
2185 * In LAST_ACK, we may still be waiting for data to drain
2186 * and/or to be acked, as well as for the ack of our FIN.
2187 * If our FIN is now acknowledged, delete the TCB,
2188 * enter the closed state and return.
2189 */
2190 case TCPS_LAST_ACK:
2191 if (ourfinisacked) {
2192 tp = tcp_close(tp);
2193 goto drop;
2194 }
2195 break;
2196
2197 /*
2198 * In TIME_WAIT state the only thing that should arrive
2199 * is a retransmission of the remote FIN. Acknowledge
2200 * it and restart the finack timer.
2201 */
2202 case TCPS_TIME_WAIT:
2203 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2204 add_to_time_wait(tp);
2205 goto dropafterack;
2206 }
2207 }
2208
2209 step6:
2210 /*
2211 * Update window information.
2212 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2213 */
2214 if ((thflags & TH_ACK) &&
2215 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2216 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2217 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2218 /* keep track of pure window updates */
2219 if (tlen == 0 &&
2220 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2221 tcpstat.tcps_rcvwinupd++;
2222 tp->snd_wnd = tiwin;
2223 tp->snd_wl1 = th->th_seq;
2224 tp->snd_wl2 = th->th_ack;
2225 if (tp->snd_wnd > tp->max_sndwnd)
2226 tp->max_sndwnd = tp->snd_wnd;
2227 needoutput = 1;
2228 }
2229
2230 /*
2231 * Process segments with URG.
2232 */
2233 if ((thflags & TH_URG) && th->th_urp &&
2234 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2235 /*
2236 * This is a kludge, but if we receive and accept
2237 * random urgent pointers, we'll crash in
2238 * soreceive. It's hard to imagine someone
2239 * actually wanting to send this much urgent data.
2240 */
2241 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2242 th->th_urp = 0; /* XXX */
2243 thflags &= ~TH_URG; /* XXX */
2244 goto dodata; /* XXX */
2245 }
2246 /*
2247 * If this segment advances the known urgent pointer,
2248 * then mark the data stream. This should not happen
2249 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2250 * a FIN has been received from the remote side.
2251 * In these states we ignore the URG.
2252 *
2253 * According to RFC961 (Assigned Protocols),
2254 * the urgent pointer points to the last octet
2255 * of urgent data. We continue, however,
2256 * to consider it to indicate the first octet
2257 * of data past the urgent section as the original
2258 * spec states (in one of two places).
2259 */
2260 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2261 tp->rcv_up = th->th_seq + th->th_urp;
2262 so->so_oobmark = so->so_rcv.sb_cc +
2263 (tp->rcv_up - tp->rcv_nxt) - 1;
2264 if (so->so_oobmark == 0) {
2265 so->so_state |= SS_RCVATMARK;
2266 postevent(so, 0, EV_OOB);
2267 }
2268 sohasoutofband(so);
2269 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2270 }
2271 /*
2272 * Remove out of band data so doesn't get presented to user.
2273 * This can happen independent of advancing the URG pointer,
2274 * but if two URG's are pending at once, some out-of-band
2275 * data may creep in... ick.
2276 */
2277 if (th->th_urp <= (u_long)tlen
2278 #if SO_OOBINLINE
2279 && (so->so_options & SO_OOBINLINE) == 0
2280 #endif
2281 )
2282 tcp_pulloutofband(so, th, m,
2283 drop_hdrlen); /* hdr drop is delayed */
2284 } else
2285 /*
2286 * If no out of band data is expected,
2287 * pull receive urgent pointer along
2288 * with the receive window.
2289 */
2290 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2291 tp->rcv_up = tp->rcv_nxt;
2292 dodata: /* XXX */
2293
2294 /*
2295 * Process the segment text, merging it into the TCP sequencing queue,
2296 * and arranging for acknowledgment of receipt if necessary.
2297 * This process logically involves adjusting tp->rcv_wnd as data
2298 * is presented to the user (this happens in tcp_usrreq.c,
2299 * case PRU_RCVD). If a FIN has already been received on this
2300 * connection then we just ignore the text.
2301 */
2302 if ((tlen || (thflags&TH_FIN)) &&
2303 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2304 m_adj(m, drop_hdrlen); /* delayed header drop */
2305 /*
2306 * Insert segment which inludes th into reassembly queue of tcp with
2307 * control block tp. Return TH_FIN if reassembly now includes
2308 * a segment with FIN. This handle the common case inline (segment
2309 * is the next to be received on an established connection, and the
2310 * queue is empty), avoiding linkage into and removal from the queue
2311 * and repetition of various conversions.
2312 * Set DELACK for segments received in order, but ack immediately
2313 * when segments are out of order (so fast retransmit can work).
2314 */
2315 if (th->th_seq == tp->rcv_nxt &&
2316 LIST_EMPTY(&tp->t_segq) &&
2317 TCPS_HAVEESTABLISHED(tp->t_state)) {
2318 #ifdef __APPLE__
2319 if (tcp_delack_enabled) {
2320 TCP_DELACK_BITSET(tp->t_inpcb->hash_element);
2321 tp->t_flags |= TF_DELACK;
2322 }
2323 #else
2324 if (DELAY_ACK(tp))
2325 callout_reset(tp->tt_delack, tcp_delacktime,
2326 tcp_timer_delack, tp);
2327 #endif
2328 else
2329 tp->t_flags |= TF_ACKNOW;
2330 tp->rcv_nxt += tlen;
2331 thflags = th->th_flags & TH_FIN;
2332 tcpstat.tcps_rcvpack++;
2333 tcpstat.tcps_rcvbyte += tlen;
2334 ND6_HINT(tp);
2335 sbappend(&so->so_rcv, m);
2336 sorwakeup(so);
2337 } else {
2338 thflags = tcp_reass(tp, th, &tlen, m);
2339 tp->t_flags |= TF_ACKNOW;
2340 }
2341
2342 if (tp->t_flags & TF_DELACK)
2343 {
2344 #if INET6
2345 if (isipv6) {
2346 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
2347 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
2348 th->th_seq, th->th_ack, th->th_win);
2349 }
2350 else
2351 #endif
2352 {
2353 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
2354 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
2355 th->th_seq, th->th_ack, th->th_win);
2356 }
2357
2358 }
2359 /*
2360 * Note the amount of data that peer has sent into
2361 * our window, in order to estimate the sender's
2362 * buffer size.
2363 */
2364 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2365 } else {
2366 m_freem(m);
2367 thflags &= ~TH_FIN;
2368 }
2369
2370 /*
2371 * If FIN is received ACK the FIN and let the user know
2372 * that the connection is closing.
2373 */
2374 if (thflags & TH_FIN) {
2375 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2376 socantrcvmore(so);
2377 postevent(so, 0, EV_FIN);
2378 /*
2379 * If connection is half-synchronized
2380 * (ie NEEDSYN flag on) then delay ACK,
2381 * so it may be piggybacked when SYN is sent.
2382 * Otherwise, since we received a FIN then no
2383 * more input can be expected, send ACK now.
2384 */
2385 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) {
2386 TCP_DELACK_BITSET(tp->t_inpcb->hash_element);
2387 tp->t_flags |= TF_DELACK;
2388 }
2389 else
2390 tp->t_flags |= TF_ACKNOW;
2391 tp->rcv_nxt++;
2392 }
2393 switch (tp->t_state) {
2394
2395 /*
2396 * In SYN_RECEIVED and ESTABLISHED STATES
2397 * enter the CLOSE_WAIT state.
2398 */
2399 case TCPS_SYN_RECEIVED:
2400 /*FALLTHROUGH*/
2401 case TCPS_ESTABLISHED:
2402 tp->t_state = TCPS_CLOSE_WAIT;
2403 break;
2404
2405 /*
2406 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2407 * enter the CLOSING state.
2408 */
2409 case TCPS_FIN_WAIT_1:
2410 tp->t_state = TCPS_CLOSING;
2411 break;
2412
2413 /*
2414 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2415 * starting the time-wait timer, turning off the other
2416 * standard timers.
2417 */
2418 case TCPS_FIN_WAIT_2:
2419 tp->t_state = TCPS_TIME_WAIT;
2420 tcp_canceltimers(tp);
2421 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2422 if (tp->cc_recv != 0 &&
2423 tp->t_starttime < tcp_msl) {
2424 tp->t_timer[TCPT_2MSL] =
2425 tp->t_rxtcur * TCPTV_TWTRUNC;
2426 /* For transaction client, force ACK now. */
2427 tp->t_flags |= TF_ACKNOW;
2428 }
2429 else
2430 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2431
2432 add_to_time_wait(tp);
2433 soisdisconnected(so);
2434 break;
2435
2436 /*
2437 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2438 */
2439 case TCPS_TIME_WAIT:
2440 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2441 add_to_time_wait(tp);
2442 break;
2443 }
2444 }
2445 #if TCPDEBUG
2446 if (so->so_options & SO_DEBUG)
2447 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2448 &tcp_savetcp, 0);
2449 #endif
2450
2451 /*
2452 * Return any desired output.
2453 */
2454 if (needoutput || (tp->t_flags & TF_ACKNOW))
2455 (void) tcp_output(tp);
2456 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2457 return;
2458
2459 dropafterack:
2460 /*
2461 * Generate an ACK dropping incoming segment if it occupies
2462 * sequence space, where the ACK reflects our state.
2463 *
2464 * We can now skip the test for the RST flag since all
2465 * paths to this code happen after packets containing
2466 * RST have been dropped.
2467 *
2468 * In the SYN-RECEIVED state, don't send an ACK unless the
2469 * segment we received passes the SYN-RECEIVED ACK test.
2470 * If it fails send a RST. This breaks the loop in the
2471 * "LAND" DoS attack, and also prevents an ACK storm
2472 * between two listening ports that have been sent forged
2473 * SYN segments, each with the source address of the other.
2474 */
2475 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2476 (SEQ_GT(tp->snd_una, th->th_ack) ||
2477 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2478 rstreason = BANDLIM_RST_OPENPORT;
2479 goto dropwithreset;
2480 }
2481 #if TCPDEBUG
2482 if (so->so_options & SO_DEBUG)
2483 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2484 &tcp_savetcp, 0);
2485 #endif
2486 m_freem(m);
2487 tp->t_flags |= TF_ACKNOW;
2488 (void) tcp_output(tp);
2489 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2490 return;
2491
2492 dropwithreset:
2493 /*
2494 * Generate a RST, dropping incoming segment.
2495 * Make ACK acceptable to originator of segment.
2496 * Don't bother to respond if destination was broadcast/multicast.
2497 */
2498 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2499 goto drop;
2500 #if INET6
2501 if (isipv6) {
2502 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2503 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2504 goto drop;
2505 } else
2506 #endif /* INET6 */
2507 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2508 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2509 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2510 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2511 goto drop;
2512 /* IPv6 anycast check is done at tcp6_input() */
2513
2514 /*
2515 * Perform bandwidth limiting.
2516 */
2517 #if ICMP_BANDLIM
2518 if (badport_bandlim(rstreason) < 0)
2519 goto drop;
2520 #endif
2521
2522 #if TCPDEBUG
2523 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2524 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2525 &tcp_savetcp, 0);
2526 #endif
2527 if (thflags & TH_ACK)
2528 /* mtod() below is safe as long as hdr dropping is delayed */
2529 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2530 TH_RST);
2531 else {
2532 if (thflags & TH_SYN)
2533 tlen++;
2534 /* mtod() below is safe as long as hdr dropping is delayed */
2535 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2536 (tcp_seq)0, TH_RST|TH_ACK);
2537 }
2538 /* destroy temporarily created socket */
2539 if (dropsocket)
2540 (void) soabort(so);
2541 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2542 return;
2543
2544 drop:
2545 /*
2546 * Drop space held by incoming segment and return.
2547 */
2548 #if TCPDEBUG
2549 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2550 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2551 &tcp_savetcp, 0);
2552 #endif
2553 m_freem(m);
2554 /* destroy temporarily created socket */
2555 if (dropsocket)
2556 (void) soabort(so);
2557 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2558 return;
2559 }
2560
2561 static void
2562 tcp_dooptions(tp, cp, cnt, th, to)
2563 struct tcpcb *tp;
2564 u_char *cp;
2565 int cnt;
2566 struct tcphdr *th;
2567 struct tcpopt *to;
2568 {
2569 u_short mss = 0;
2570 int opt, optlen;
2571
2572 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2573 opt = cp[0];
2574 if (opt == TCPOPT_EOL)
2575 break;
2576 if (opt == TCPOPT_NOP)
2577 optlen = 1;
2578 else {
2579 if (cnt < 2)
2580 break;
2581 optlen = cp[1];
2582 if (optlen < 2 || optlen > cnt)
2583 break;
2584 }
2585 switch (opt) {
2586
2587 default:
2588 continue;
2589
2590 case TCPOPT_MAXSEG:
2591 if (optlen != TCPOLEN_MAXSEG)
2592 continue;
2593 if (!(th->th_flags & TH_SYN))
2594 continue;
2595 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2596 NTOHS(mss);
2597 break;
2598
2599 case TCPOPT_WINDOW:
2600 if (optlen != TCPOLEN_WINDOW)
2601 continue;
2602 if (!(th->th_flags & TH_SYN))
2603 continue;
2604 tp->t_flags |= TF_RCVD_SCALE;
2605 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2606 break;
2607
2608 case TCPOPT_TIMESTAMP:
2609 if (optlen != TCPOLEN_TIMESTAMP)
2610 continue;
2611 to->to_flag |= TOF_TS;
2612 bcopy((char *)cp + 2,
2613 (char *)&to->to_tsval, sizeof(to->to_tsval));
2614 NTOHL(to->to_tsval);
2615 bcopy((char *)cp + 6,
2616 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2617 NTOHL(to->to_tsecr);
2618
2619 /*
2620 * A timestamp received in a SYN makes
2621 * it ok to send timestamp requests and replies.
2622 */
2623 if (th->th_flags & TH_SYN) {
2624 tp->t_flags |= TF_RCVD_TSTMP;
2625 tp->ts_recent = to->to_tsval;
2626 tp->ts_recent_age = tcp_now;
2627 }
2628 break;
2629 case TCPOPT_CC:
2630 if (optlen != TCPOLEN_CC)
2631 continue;
2632 to->to_flag |= TOF_CC;
2633 bcopy((char *)cp + 2,
2634 (char *)&to->to_cc, sizeof(to->to_cc));
2635 NTOHL(to->to_cc);
2636 /*
2637 * A CC or CC.new option received in a SYN makes
2638 * it ok to send CC in subsequent segments.
2639 */
2640 if (th->th_flags & TH_SYN)
2641 tp->t_flags |= TF_RCVD_CC;
2642 break;
2643 case TCPOPT_CCNEW:
2644 if (optlen != TCPOLEN_CC)
2645 continue;
2646 if (!(th->th_flags & TH_SYN))
2647 continue;
2648 to->to_flag |= TOF_CCNEW;
2649 bcopy((char *)cp + 2,
2650 (char *)&to->to_cc, sizeof(to->to_cc));
2651 NTOHL(to->to_cc);
2652 /*
2653 * A CC or CC.new option received in a SYN makes
2654 * it ok to send CC in subsequent segments.
2655 */
2656 tp->t_flags |= TF_RCVD_CC;
2657 break;
2658 case TCPOPT_CCECHO:
2659 if (optlen != TCPOLEN_CC)
2660 continue;
2661 if (!(th->th_flags & TH_SYN))
2662 continue;
2663 to->to_flag |= TOF_CCECHO;
2664 bcopy((char *)cp + 2,
2665 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2666 NTOHL(to->to_ccecho);
2667 break;
2668 }
2669 }
2670 if (th->th_flags & TH_SYN)
2671 tcp_mss(tp, mss); /* sets t_maxseg */
2672 }
2673
2674 /*
2675 * Pull out of band byte out of a segment so
2676 * it doesn't appear in the user's data queue.
2677 * It is still reflected in the segment length for
2678 * sequencing purposes.
2679 */
2680 static void
2681 tcp_pulloutofband(so, th, m, off)
2682 struct socket *so;
2683 struct tcphdr *th;
2684 register struct mbuf *m;
2685 int off; /* delayed to be droped hdrlen */
2686 {
2687 int cnt = off + th->th_urp - 1;
2688
2689 while (cnt >= 0) {
2690 if (m->m_len > cnt) {
2691 char *cp = mtod(m, caddr_t) + cnt;
2692 struct tcpcb *tp = sototcpcb(so);
2693
2694 tp->t_iobc = *cp;
2695 tp->t_oobflags |= TCPOOB_HAVEDATA;
2696 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2697 m->m_len--;
2698 if (m->m_flags & M_PKTHDR)
2699 m->m_pkthdr.len--;
2700 return;
2701 }
2702 cnt -= m->m_len;
2703 m = m->m_next;
2704 if (m == 0)
2705 break;
2706 }
2707 panic("tcp_pulloutofband");
2708 }
2709
2710 /*
2711 * Collect new round-trip time estimate
2712 * and update averages and current timeout.
2713 */
2714 static void
2715 tcp_xmit_timer(tp, rtt)
2716 register struct tcpcb *tp;
2717 int rtt;
2718 {
2719 register int delta;
2720
2721 tcpstat.tcps_rttupdated++;
2722 tp->t_rttupdated++;
2723 if (tp->t_srtt != 0) {
2724 /*
2725 * srtt is stored as fixed point with 5 bits after the
2726 * binary point (i.e., scaled by 8). The following magic
2727 * is equivalent to the smoothing algorithm in rfc793 with
2728 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2729 * point). Adjust rtt to origin 0.
2730 */
2731 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2732 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2733
2734 if ((tp->t_srtt += delta) <= 0)
2735 tp->t_srtt = 1;
2736
2737 /*
2738 * We accumulate a smoothed rtt variance (actually, a
2739 * smoothed mean difference), then set the retransmit
2740 * timer to smoothed rtt + 4 times the smoothed variance.
2741 * rttvar is stored as fixed point with 4 bits after the
2742 * binary point (scaled by 16). The following is
2743 * equivalent to rfc793 smoothing with an alpha of .75
2744 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2745 * rfc793's wired-in beta.
2746 */
2747 if (delta < 0)
2748 delta = -delta;
2749 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2750 if ((tp->t_rttvar += delta) <= 0)
2751 tp->t_rttvar = 1;
2752 } else {
2753 /*
2754 * No rtt measurement yet - use the unsmoothed rtt.
2755 * Set the variance to half the rtt (so our first
2756 * retransmit happens at 3*rtt).
2757 */
2758 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2759 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2760 }
2761 tp->t_rtttime = 0;
2762 tp->t_rxtshift = 0;
2763
2764 /*
2765 * the retransmit should happen at rtt + 4 * rttvar.
2766 * Because of the way we do the smoothing, srtt and rttvar
2767 * will each average +1/2 tick of bias. When we compute
2768 * the retransmit timer, we want 1/2 tick of rounding and
2769 * 1 extra tick because of +-1/2 tick uncertainty in the
2770 * firing of the timer. The bias will give us exactly the
2771 * 1.5 tick we need. But, because the bias is
2772 * statistical, we have to test that we don't drop below
2773 * the minimum feasible timer (which is 2 ticks).
2774 */
2775 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2776 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2777
2778 /*
2779 * We received an ack for a packet that wasn't retransmitted;
2780 * it is probably safe to discard any error indications we've
2781 * received recently. This isn't quite right, but close enough
2782 * for now (a route might have failed after we sent a segment,
2783 * and the return path might not be symmetrical).
2784 */
2785 tp->t_softerror = 0;
2786 }
2787
2788 /*
2789 * Determine a reasonable value for maxseg size.
2790 * If the route is known, check route for mtu.
2791 * If none, use an mss that can be handled on the outgoing
2792 * interface without forcing IP to fragment; if bigger than
2793 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2794 * to utilize large mbufs. If no route is found, route has no mtu,
2795 * or the destination isn't local, use a default, hopefully conservative
2796 * size (usually 512 or the default IP max size, but no more than the mtu
2797 * of the interface), as we can't discover anything about intervening
2798 * gateways or networks. We also initialize the congestion/slow start
2799 * window to be a single segment if the destination isn't local.
2800 * While looking at the routing entry, we also initialize other path-dependent
2801 * parameters from pre-set or cached values in the routing entry.
2802 *
2803 * Also take into account the space needed for options that we
2804 * send regularly. Make maxseg shorter by that amount to assure
2805 * that we can send maxseg amount of data even when the options
2806 * are present. Store the upper limit of the length of options plus
2807 * data in maxopd.
2808 *
2809 * NOTE that this routine is only called when we process an incoming
2810 * segment, for outgoing segments only tcp_mssopt is called.
2811 *
2812 * In case of T/TCP, we call this routine during implicit connection
2813 * setup as well (offer = -1), to initialize maxseg from the cached
2814 * MSS of our peer.
2815 */
2816 void
2817 tcp_mss(tp, offer)
2818 struct tcpcb *tp;
2819 int offer;
2820 {
2821 register struct rtentry *rt;
2822 struct ifnet *ifp;
2823 register int rtt, mss;
2824 u_long bufsize;
2825 struct inpcb *inp;
2826 struct socket *so;
2827 struct rmxp_tao *taop;
2828 int origoffer = offer;
2829 #if INET6
2830 int isipv6;
2831 int min_protoh;
2832 #endif
2833
2834 inp = tp->t_inpcb;
2835 #if INET6
2836 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2837 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr)
2838 : sizeof (struct tcpiphdr);
2839 #else
2840 #define min_protoh (sizeof (struct tcpiphdr))
2841 #endif
2842 #if INET6
2843 if (isipv6)
2844 rt = tcp_rtlookup6(inp);
2845 else
2846 #endif /* INET6 */
2847 rt = tcp_rtlookup(inp);
2848 if (rt == NULL) {
2849 tp->t_maxopd = tp->t_maxseg =
2850 #if INET6
2851 isipv6 ? tcp_v6mssdflt :
2852 #endif /* INET6 */
2853 tcp_mssdflt;
2854 return;
2855 }
2856 ifp = rt->rt_ifp;
2857 /*
2858 * Slower link window correction:
2859 * If a value is specificied for slowlink_wsize use it for PPP links
2860 * believed to be on a serial modem (speed <128Kbps). Excludes 9600bps as
2861 * it is the default value adversized by pseudo-devices over ppp.
2862 */
2863 if (ifp->if_type == IFT_PPP && slowlink_wsize > 0 &&
2864 ifp->if_baudrate > 9600 && ifp->if_baudrate <= 128000) {
2865 tp->t_flags |= TF_SLOWLINK;
2866 }
2867 so = inp->inp_socket;
2868
2869 taop = rmx_taop(rt->rt_rmx);
2870 /*
2871 * Offer == -1 means that we didn't receive SYN yet,
2872 * use cached value in that case;
2873 */
2874 if (offer == -1)
2875 offer = taop->tao_mssopt;
2876 /*
2877 * Offer == 0 means that there was no MSS on the SYN segment,
2878 * in this case we use tcp_mssdflt.
2879 */
2880 if (offer == 0)
2881 offer =
2882 #if INET6
2883 isipv6 ? tcp_v6mssdflt :
2884 #endif /* INET6 */
2885 tcp_mssdflt;
2886 else
2887 /*
2888 * Sanity check: make sure that maxopd will be large
2889 * enough to allow some data on segments even is the
2890 * all the option space is used (40bytes). Otherwise
2891 * funny things may happen in tcp_output.
2892 */
2893 offer = max(offer, 64);
2894 taop->tao_mssopt = offer;
2895
2896 /*
2897 * While we're here, check if there's an initial rtt
2898 * or rttvar. Convert from the route-table units
2899 * to scaled multiples of the slow timeout timer.
2900 */
2901 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2902 /*
2903 * XXX the lock bit for RTT indicates that the value
2904 * is also a minimum value; this is subject to time.
2905 */
2906 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2907 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
2908 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
2909 tcpstat.tcps_usedrtt++;
2910 if (rt->rt_rmx.rmx_rttvar) {
2911 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2912 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
2913 tcpstat.tcps_usedrttvar++;
2914 } else {
2915 /* default variation is +- 1 rtt */
2916 tp->t_rttvar =
2917 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2918 }
2919 TCPT_RANGESET(tp->t_rxtcur,
2920 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2921 tp->t_rttmin, TCPTV_REXMTMAX);
2922 }
2923 /*
2924 * if there's an mtu associated with the route, use it
2925 * else, use the link mtu.
2926 */
2927 if (rt->rt_rmx.rmx_mtu)
2928 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2929 else
2930 {
2931 mss =
2932 #if INET6
2933 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu :
2934 #endif
2935 ifp->if_mtu
2936 #if INET6
2937 )
2938 #endif
2939 - min_protoh;
2940 #if INET6
2941 if (isipv6) {
2942 if (!in6_localaddr(&inp->in6p_faddr))
2943 mss = min(mss, tcp_v6mssdflt);
2944 } else
2945 #endif /* INET6 */
2946 if (!in_localaddr(inp->inp_faddr))
2947 mss = min(mss, tcp_mssdflt);
2948 }
2949 mss = min(mss, offer);
2950 /*
2951 * maxopd stores the maximum length of data AND options
2952 * in a segment; maxseg is the amount of data in a normal
2953 * segment. We need to store this value (maxopd) apart
2954 * from maxseg, because now every segment carries options
2955 * and thus we normally have somewhat less data in segments.
2956 */
2957 tp->t_maxopd = mss;
2958
2959 /*
2960 * In case of T/TCP, origoffer==-1 indicates, that no segments
2961 * were received yet. In this case we just guess, otherwise
2962 * we do the same as before T/TCP.
2963 */
2964 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2965 (origoffer == -1 ||
2966 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2967 mss -= TCPOLEN_TSTAMP_APPA;
2968 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2969 (origoffer == -1 ||
2970 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2971 mss -= TCPOLEN_CC_APPA;
2972
2973 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2974 if (mss > MCLBYTES)
2975 mss &= ~(MCLBYTES-1);
2976 #else
2977 if (mss > MCLBYTES)
2978 mss = mss / MCLBYTES * MCLBYTES;
2979 #endif
2980 /*
2981 * If there's a pipesize, change the socket buffer
2982 * to that size. Make the socket buffers an integral
2983 * number of mss units; if the mss is larger than
2984 * the socket buffer, decrease the mss.
2985 */
2986 #if RTV_SPIPE
2987 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2988 #endif
2989 bufsize = so->so_snd.sb_hiwat;
2990 if (bufsize < mss)
2991 mss = bufsize;
2992 else {
2993 bufsize = roundup(bufsize, mss);
2994 if (bufsize > sb_max)
2995 bufsize = sb_max;
2996 (void)sbreserve(&so->so_snd, bufsize);
2997 }
2998 tp->t_maxseg = mss;
2999
3000 #if RTV_RPIPE
3001 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3002 #endif
3003 bufsize = so->so_rcv.sb_hiwat;
3004 if (bufsize > mss) {
3005 bufsize = roundup(bufsize, mss);
3006 if (bufsize > sb_max)
3007 bufsize = sb_max;
3008 (void)sbreserve(&so->so_rcv, bufsize);
3009 }
3010
3011 /*
3012 * Set the slow-start flight size depending on whether this
3013 * is a local network or not.
3014 */
3015 if (
3016 #if INET6
3017 (isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3018 (!isipv6 &&
3019 #endif
3020 in_localaddr(inp->inp_faddr)
3021 #if INET6
3022 )
3023 #endif
3024 )
3025 tp->snd_cwnd = mss * ss_fltsz_local;
3026 else
3027 tp->snd_cwnd = mss * ss_fltsz;
3028
3029 if (rt->rt_rmx.rmx_ssthresh) {
3030 /*
3031 * There's some sort of gateway or interface
3032 * buffer limit on the path. Use this to set
3033 * the slow start threshhold, but set the
3034 * threshold to no less than 2*mss.
3035 */
3036 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3037 tcpstat.tcps_usedssthresh++;
3038 }
3039 }
3040
3041 /*
3042 * Determine the MSS option to send on an outgoing SYN.
3043 */
3044 int
3045 tcp_mssopt(tp)
3046 struct tcpcb *tp;
3047 {
3048 struct rtentry *rt;
3049 #if INET6
3050 int isipv6;
3051 int min_protoh;
3052 #endif
3053
3054 #if INET6
3055 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3056 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr)
3057 : sizeof (struct tcpiphdr);
3058 #else
3059 #define min_protoh (sizeof (struct tcpiphdr))
3060 #endif
3061 #if INET6
3062 if (isipv6)
3063 rt = tcp_rtlookup6(tp->t_inpcb);
3064 else
3065 #endif /* INET6 */
3066 rt = tcp_rtlookup(tp->t_inpcb);
3067 if (rt == NULL)
3068 return
3069 #if INET6
3070 isipv6 ? tcp_v6mssdflt :
3071 #endif /* INET6 */
3072 tcp_mssdflt;
3073 /*
3074 * Slower link window correction:
3075 * If a value is specificied for slowlink_wsize use it for PPP links
3076 * believed to be on a serial modem (speed <128Kbps). Excludes 9600bps as
3077 * it is the default value adversized by pseudo-devices over ppp.
3078 */
3079 if (rt->rt_ifp->if_type == IFT_PPP && slowlink_wsize > 0 &&
3080 rt->rt_ifp->if_baudrate > 9600 && rt->rt_ifp->if_baudrate <= 128000) {
3081 tp->t_flags |= TF_SLOWLINK;
3082 }
3083
3084 return rt->rt_ifp->if_mtu - min_protoh;
3085 }
3086
3087
3088 /*
3089 * Checks for partial ack. If partial ack arrives, force the retransmission
3090 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3091 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to
3092 * be started again. If the ack advances at least to tp->snd_recover, return 0.
3093 */
3094 static int
3095 tcp_newreno(tp, th)
3096 struct tcpcb *tp;
3097 struct tcphdr *th;
3098 {
3099 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
3100 tcp_seq onxt = tp->snd_nxt;
3101 u_long ocwnd = tp->snd_cwnd;
3102 #ifdef __APPLE__
3103 tp->t_timer[TCPT_REXMT] = 0;
3104 #else
3105 callout_stop(tp->tt_rexmt);
3106 #endif
3107 tp->t_rtttime = 0;
3108 tp->snd_nxt = th->th_ack;
3109 /*
3110 * Set snd_cwnd to one segment beyond acknowledged offset
3111 * (tp->snd_una has not yet been updated when this function
3112 * is called)
3113 */
3114 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3115 (void) tcp_output(tp);
3116 tp->snd_cwnd = ocwnd;
3117 if (SEQ_GT(onxt, tp->snd_nxt))
3118 tp->snd_nxt = onxt;
3119 /*
3120 * Partial window deflation. Relies on fact that tp->snd_una
3121 * not updated yet.
3122 */
3123 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3124 return (1);
3125 }
3126 return (0);
3127 }
mirror of XNU