1 /*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
34 * $FreeBSD: releng/5.2/sys/netinet/tcp_input.c 127031 2004-03-15 18:17:10Z fjoe $
35 */
36
37 #include "opt_ipfw.h" /* for ipfw_fwd */
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_mac.h"
41 #include "opt_tcpdebug.h"
42 #include "opt_tcp_input.h"
43
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/mac.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/proc.h> /* for proc0 declaration */
50 #include <sys/protosw.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 #include <sys/systm.h>
57
58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
59
60 #include <vm/uma.h>
61
62 #include <net/if.h>
63 #include <net/route.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
71 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
72 #include <netinet/ip_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet/icmp6.h>
75 #include <netinet6/in6_pcb.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet6/tcp6_var.h>
84 #include <netinet/tcpip.h>
85 #ifdef TCPDEBUG
86 #include <netinet/tcp_debug.h>
87 #endif /* TCPDEBUG */
88
89 #ifdef FAST_IPSEC
90 #include <netipsec/ipsec.h>
91 #include <netipsec/ipsec6.h>
92 #endif /*FAST_IPSEC*/
93
94 #ifdef IPSEC
95 #include <netinet6/ipsec.h>
96 #include <netinet6/ipsec6.h>
97 #include <netkey/key.h>
98 #endif /*IPSEC*/
99
100 #include <machine/in_cksum.h>
101
102 static const int tcprexmtthresh = 3;
103 tcp_cc tcp_ccgen;
104
105 struct tcpstat tcpstat;
106 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
107 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
108
109 static int log_in_vain = 0;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
111 &log_in_vain, 0, "Log all incoming TCP connections");
112
113 static int blackhole = 0;
114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
115 &blackhole, 0, "Do not send RST when dropping refused connections");
116
117 int tcp_delack_enabled = 1;
118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
119 &tcp_delack_enabled, 0,
120 "Delay ACK to try and piggyback it onto a data packet");
121
122 #ifdef TCP_DROP_SYNFIN
123 static int drop_synfin = 0;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
125 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
126 #endif
127
128 static int tcp_do_rfc3042 = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
130 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
131
132 static int tcp_do_rfc3390 = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
134 &tcp_do_rfc3390, 0,
135 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
136
137 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
138 "TCP Segment Reassembly Queue");
139
140 static int tcp_reass_maxseg = 0;
141 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
142 &tcp_reass_maxseg, 0,
143 "Global maximum number of TCP Segments in Reassembly Queue");
144
145 int tcp_reass_qsize = 0;
146 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
147 &tcp_reass_qsize, 0,
148 "Global number of TCP Segments currently in Reassembly Queue");
149
150 static int tcp_reass_overflows = 0;
151 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
152 &tcp_reass_overflows, 0,
153 "Global number of TCP Segment Reassembly Queue Overflows");
154
155 struct inpcbhead tcb;
156 #define tcb6 tcb /* for KAME src sync over BSD*'s */
157 struct inpcbinfo tcbinfo;
158 struct mtx *tcbinfo_mtx;
159
160 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
161 static void tcp_pulloutofband(struct socket *,
162 struct tcphdr *, struct mbuf *, int);
163 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
164 struct mbuf *);
165 static void tcp_xmit_timer(struct tcpcb *, int);
166 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
167 static int tcp_timewait(struct tcptw *, struct tcpopt *,
168 struct tcphdr *, struct mbuf *, int);
169
170 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
171 #ifdef INET6
172 #define ND6_HINT(tp) \
173 do { \
174 if ((tp) && (tp)->t_inpcb && \
175 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
176 nd6_nud_hint(NULL, NULL, 0); \
177 } while (0)
178 #else
179 #define ND6_HINT(tp)
180 #endif
181
182 /*
183 * Indicate whether this ack should be delayed. We can delay the ack if
184 * - there is no delayed ack timer in progress and
185 * - our last ack wasn't a 0-sized window. We never want to delay
186 * the ack that opens up a 0-sized window and
187 * - delayed acks are enabled or
188 * - this is a half-synchronized T/TCP connection.
189 */
190 #define DELAY_ACK(tp) \
191 ((!callout_active(tp->tt_delack) && \
192 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
193 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
194
195 /* Initialize TCP reassembly queue */
196 uma_zone_t tcp_reass_zone;
197 void
198 tcp_reass_init()
199 {
200 tcp_reass_maxseg = nmbclusters / 16;
201 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
202 &tcp_reass_maxseg);
203 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
205 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
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 = NULL;
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 /*
230 * Limit the number of segments in the reassembly queue to prevent
231 * holding on to too many segments (and thus running out of mbufs).
232 * Make sure to let the missing segment through which caused this
233 * queue. Always keep one global queue entry spare to be able to
234 * process the missing segment.
235 */
236 if (th->th_seq != tp->rcv_nxt &&
237 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
238 tcp_reass_overflows++;
239 tcpstat.tcps_rcvmemdrop++;
240 m_freem(m);
241 return (0);
242 }
243
244 /*
245 * Allocate a new queue entry. If we can't, or hit the zone limit
246 * just drop the pkt.
247 */
248 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
249 if (te == NULL) {
250 tcpstat.tcps_rcvmemdrop++;
251 m_freem(m);
252 return (0);
253 }
254 tcp_reass_qsize++;
255
256 /*
257 * Find a segment which begins after this one does.
258 */
259 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
260 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
261 break;
262 p = q;
263 }
264
265 /*
266 * If there is a preceding segment, it may provide some of
267 * our data already. If so, drop the data from the incoming
268 * segment. If it provides all of our data, drop us.
269 */
270 if (p != NULL) {
271 register int i;
272 /* conversion to int (in i) handles seq wraparound */
273 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
274 if (i > 0) {
275 if (i >= *tlenp) {
276 tcpstat.tcps_rcvduppack++;
277 tcpstat.tcps_rcvdupbyte += *tlenp;
278 m_freem(m);
279 uma_zfree(tcp_reass_zone, te);
280 tcp_reass_qsize--;
281 /*
282 * Try to present any queued data
283 * at the left window edge to the user.
284 * This is needed after the 3-WHS
285 * completes.
286 */
287 goto present; /* ??? */
288 }
289 m_adj(m, i);
290 *tlenp -= i;
291 th->th_seq += i;
292 }
293 }
294 tcpstat.tcps_rcvoopack++;
295 tcpstat.tcps_rcvoobyte += *tlenp;
296
297 /*
298 * While we overlap succeeding segments trim them or,
299 * if they are completely covered, dequeue them.
300 */
301 while (q) {
302 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
303 if (i <= 0)
304 break;
305 if (i < q->tqe_len) {
306 q->tqe_th->th_seq += i;
307 q->tqe_len -= i;
308 m_adj(q->tqe_m, i);
309 break;
310 }
311
312 nq = LIST_NEXT(q, tqe_q);
313 LIST_REMOVE(q, tqe_q);
314 m_freem(q->tqe_m);
315 uma_zfree(tcp_reass_zone, q);
316 tcp_reass_qsize--;
317 q = nq;
318 }
319
320 /* Insert the new segment queue entry into place. */
321 te->tqe_m = m;
322 te->tqe_th = th;
323 te->tqe_len = *tlenp;
324
325 if (p == NULL) {
326 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
327 } else {
328 LIST_INSERT_AFTER(p, te, tqe_q);
329 }
330
331 present:
332 /*
333 * Present data to user, advancing rcv_nxt through
334 * completed sequence space.
335 */
336 if (!TCPS_HAVEESTABLISHED(tp->t_state))
337 return (0);
338 q = LIST_FIRST(&tp->t_segq);
339 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
340 return (0);
341 do {
342 tp->rcv_nxt += q->tqe_len;
343 flags = q->tqe_th->th_flags & TH_FIN;
344 nq = LIST_NEXT(q, tqe_q);
345 LIST_REMOVE(q, tqe_q);
346 if (so->so_state & SS_CANTRCVMORE)
347 m_freem(q->tqe_m);
348 else
349 sbappendstream(&so->so_rcv, q->tqe_m);
350 uma_zfree(tcp_reass_zone, q);
351 tcp_reass_qsize--;
352 q = nq;
353 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
354 ND6_HINT(tp);
355 sorwakeup(so);
356 return (flags);
357 }
358
359 /*
360 * TCP input routine, follows pages 65-76 of the
361 * protocol specification dated September, 1981 very closely.
362 */
363 #ifdef 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 register 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 = NULL;
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 u_long tiwin;
411 struct tcpopt to; /* options in this segment */
412 struct rmxp_tao tao; /* our TAO cache entry */
413 int headlocked = 0;
414 struct sockaddr_in *next_hop = NULL;
415 int rstreason; /* For badport_bandlim accounting purposes */
416
417 struct ip6_hdr *ip6 = NULL;
418 #ifdef INET6
419 int isipv6;
420 #else
421 const int isipv6 = 0;
422 #endif
423
424 #ifdef TCPDEBUG
425 /*
426 * The size of tcp_saveipgen must be the size of the max ip header,
427 * now IPv6.
428 */
429 u_char tcp_saveipgen[40];
430 struct tcphdr tcp_savetcp;
431 short ostate = 0;
432 #endif
433
434 /* Grab info from MT_TAG mbufs prepended to the chain. */
435 for (;m && m->m_type == MT_TAG; m = m->m_next) {
436 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
437 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
438 }
439 #ifdef INET6
440 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
441 #endif
442 bzero(&tao, sizeof(tao));
443 bzero((char *)&to, sizeof(to));
444
445 tcpstat.tcps_rcvtotal++;
446
447 if (isipv6) {
448 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
449 ip6 = mtod(m, struct ip6_hdr *);
450 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
451 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
452 tcpstat.tcps_rcvbadsum++;
453 goto drop;
454 }
455 th = (struct tcphdr *)((caddr_t)ip6 + off0);
456
457 /*
458 * Be proactive about unspecified IPv6 address in source.
459 * As we use all-zero to indicate unbounded/unconnected pcb,
460 * unspecified IPv6 address can be used to confuse us.
461 *
462 * Note that packets with unspecified IPv6 destination is
463 * already dropped in ip6_input.
464 */
465 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
466 /* XXX stat */
467 goto drop;
468 }
469 } else {
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 }
478 if (m->m_len < sizeof (struct tcpiphdr)) {
479 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
480 tcpstat.tcps_rcvshort++;
481 return;
482 }
483 }
484 ip = mtod(m, struct ip *);
485 ipov = (struct ipovly *)ip;
486 th = (struct tcphdr *)((caddr_t)ip + off0);
487 tlen = ip->ip_len;
488
489 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
490 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
491 th->th_sum = m->m_pkthdr.csum_data;
492 else
493 th->th_sum = in_pseudo(ip->ip_src.s_addr,
494 ip->ip_dst.s_addr,
495 htonl(m->m_pkthdr.csum_data +
496 ip->ip_len +
497 IPPROTO_TCP));
498 th->th_sum ^= 0xffff;
499 #ifdef TCPDEBUG
500 ipov->ih_len = (u_short)tlen;
501 ipov->ih_len = htons(ipov->ih_len);
502 #endif
503 } else {
504 /*
505 * Checksum extended TCP header and data.
506 */
507 len = sizeof (struct ip) + tlen;
508 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
509 ipov->ih_len = (u_short)tlen;
510 ipov->ih_len = htons(ipov->ih_len);
511 th->th_sum = in_cksum(m, len);
512 }
513 if (th->th_sum) {
514 tcpstat.tcps_rcvbadsum++;
515 goto drop;
516 }
517 #ifdef INET6
518 /* Re-initialization for later version check */
519 ip->ip_v = IPVERSION;
520 #endif
521 }
522
523 /*
524 * Check that TCP offset makes sense,
525 * pull out TCP options and adjust length. XXX
526 */
527 off = th->th_off << 2;
528 if (off < sizeof (struct tcphdr) || off > tlen) {
529 tcpstat.tcps_rcvbadoff++;
530 goto drop;
531 }
532 tlen -= off; /* tlen is used instead of ti->ti_len */
533 if (off > sizeof (struct tcphdr)) {
534 if (isipv6) {
535 IP6_EXTHDR_CHECK(m, off0, off, );
536 ip6 = mtod(m, struct ip6_hdr *);
537 th = (struct tcphdr *)((caddr_t)ip6 + off0);
538 } else {
539 if (m->m_len < sizeof(struct ip) + off) {
540 if ((m = m_pullup(m, sizeof (struct ip) + off))
541 == 0) {
542 tcpstat.tcps_rcvshort++;
543 return;
544 }
545 ip = mtod(m, struct ip *);
546 ipov = (struct ipovly *)ip;
547 th = (struct tcphdr *)((caddr_t)ip + off0);
548 }
549 }
550 optlen = off - sizeof (struct tcphdr);
551 optp = (u_char *)(th + 1);
552 }
553 thflags = th->th_flags;
554
555 #ifdef TCP_DROP_SYNFIN
556 /*
557 * If the drop_synfin option is enabled, drop all packets with
558 * both the SYN and FIN bits set. This prevents e.g. nmap from
559 * identifying the TCP/IP stack.
560 *
561 * This is a violation of the TCP specification.
562 */
563 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
564 goto drop;
565 #endif
566
567 /*
568 * Convert TCP protocol specific fields to host format.
569 */
570 th->th_seq = ntohl(th->th_seq);
571 th->th_ack = ntohl(th->th_ack);
572 th->th_win = ntohs(th->th_win);
573 th->th_urp = ntohs(th->th_urp);
574
575 /*
576 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
577 * until after ip6_savecontrol() is called and before other functions
578 * which don't want those proto headers.
579 * Because ip6_savecontrol() is going to parse the mbuf to
580 * search for data to be passed up to user-land, it wants mbuf
581 * parameters to be unchanged.
582 * XXX: the call of ip6_savecontrol() has been obsoleted based on
583 * latest version of the advanced API (20020110).
584 */
585 drop_hdrlen = off0 + off;
586
587 /*
588 * Locate pcb for segment.
589 */
590 INP_INFO_WLOCK(&tcbinfo);
591 headlocked = 1;
592 findpcb:
593 /* IPFIREWALL_FORWARD section */
594 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
595 /*
596 * Transparently forwarded. Pretend to be the destination.
597 * already got one like this?
598 */
599 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
600 ip->ip_dst, th->th_dport,
601 0, m->m_pkthdr.rcvif);
602 if (!inp) {
603 /* It's new. Try find the ambushing socket. */
604 inp = in_pcblookup_hash(&tcbinfo,
605 ip->ip_src, th->th_sport,
606 next_hop->sin_addr,
607 next_hop->sin_port ?
608 ntohs(next_hop->sin_port) :
609 th->th_dport,
610 1, m->m_pkthdr.rcvif);
611 }
612 } else {
613 if (isipv6)
614 inp = in6_pcblookup_hash(&tcbinfo,
615 &ip6->ip6_src, th->th_sport,
616 &ip6->ip6_dst, th->th_dport,
617 1, m->m_pkthdr.rcvif);
618 else
619 inp = in_pcblookup_hash(&tcbinfo,
620 ip->ip_src, th->th_sport,
621 ip->ip_dst, th->th_dport,
622 1, m->m_pkthdr.rcvif);
623 }
624
625 #ifdef IPSEC
626 if (isipv6) {
627 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
628 ipsec6stat.in_polvio++;
629 goto drop;
630 }
631 } else {
632 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
633 ipsecstat.in_polvio++;
634 goto drop;
635 }
636 }
637 #endif
638 #ifdef FAST_IPSEC
639 if (isipv6) {
640 if (inp != NULL && ipsec6_in_reject(m, inp)) {
641 goto drop;
642 }
643 } else
644 if (inp != NULL && ipsec4_in_reject(m, inp)) {
645 goto drop;
646 }
647 #endif /*FAST_IPSEC*/
648
649 /*
650 * If the state is CLOSED (i.e., TCB does not exist) then
651 * all data in the incoming segment is discarded.
652 * If the TCB exists but is in CLOSED state, it is embryonic,
653 * but should either do a listen or a connect soon.
654 */
655 if (inp == NULL) {
656 if (log_in_vain) {
657 #ifdef INET6
658 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
659 #else
660 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
661 #endif
662
663 if (isipv6) {
664 strcpy(dbuf, "[");
665 strcpy(sbuf, "[");
666 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
667 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
668 strcat(dbuf, "]");
669 strcat(sbuf, "]");
670 } else {
671 strcpy(dbuf, inet_ntoa(ip->ip_dst));
672 strcpy(sbuf, inet_ntoa(ip->ip_src));
673 }
674 switch (log_in_vain) {
675 case 1:
676 if ((thflags & TH_SYN) == 0)
677 break;
678 /* FALLTHROUGH */
679 case 2:
680 log(LOG_INFO,
681 "Connection attempt to TCP %s:%d "
682 "from %s:%d flags:0x%02x\n",
683 dbuf, ntohs(th->th_dport), sbuf,
684 ntohs(th->th_sport), thflags);
685 break;
686 default:
687 break;
688 }
689 }
690 if (blackhole) {
691 switch (blackhole) {
692 case 1:
693 if (thflags & TH_SYN)
694 goto drop;
695 break;
696 case 2:
697 goto drop;
698 default:
699 goto drop;
700 }
701 }
702 rstreason = BANDLIM_RST_CLOSEDPORT;
703 goto dropwithreset;
704 }
705 INP_LOCK(inp);
706 if (inp->inp_vflag & INP_TIMEWAIT) {
707 /*
708 * The only option of relevance is TOF_CC, and only if
709 * present in a SYN segment. See tcp_timewait().
710 */
711 if (thflags & TH_SYN)
712 tcp_dooptions(&to, optp, optlen, 1);
713 if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
714 &to, th, m, tlen))
715 goto findpcb;
716 /*
717 * tcp_timewait unlocks inp.
718 */
719 INP_INFO_WUNLOCK(&tcbinfo);
720 return;
721 }
722 tp = intotcpcb(inp);
723 if (tp == 0) {
724 INP_UNLOCK(inp);
725 rstreason = BANDLIM_RST_CLOSEDPORT;
726 goto dropwithreset;
727 }
728 if (tp->t_state == TCPS_CLOSED)
729 goto drop;
730
731 /* Unscale the window into a 32-bit value. */
732 if ((thflags & TH_SYN) == 0)
733 tiwin = th->th_win << tp->snd_scale;
734 else
735 tiwin = th->th_win;
736
737 #ifdef MAC
738 if (mac_check_inpcb_deliver(inp, m))
739 goto drop;
740 #endif
741 so = inp->inp_socket;
742 #ifdef TCPDEBUG
743 if (so->so_options & SO_DEBUG) {
744 ostate = tp->t_state;
745 if (isipv6)
746 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
747 else
748 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
749 tcp_savetcp = *th;
750 }
751 #endif
752 if (so->so_options & SO_ACCEPTCONN) {
753 struct in_conninfo inc;
754
755 #ifdef INET6
756 inc.inc_isipv6 = isipv6;
757 #endif
758 if (isipv6) {
759 inc.inc6_faddr = ip6->ip6_src;
760 inc.inc6_laddr = ip6->ip6_dst;
761 } else {
762 inc.inc_faddr = ip->ip_src;
763 inc.inc_laddr = ip->ip_dst;
764 }
765 inc.inc_fport = th->th_sport;
766 inc.inc_lport = th->th_dport;
767
768 /*
769 * If the state is LISTEN then ignore segment if it contains
770 * a RST. If the segment contains an ACK then it is bad and
771 * send a RST. If it does not contain a SYN then it is not
772 * interesting; drop it.
773 *
774 * If the state is SYN_RECEIVED (syncache) and seg contains
775 * an ACK, but not for our SYN/ACK, send a RST. If the seg
776 * contains a RST, check the sequence number to see if it
777 * is a valid reset segment.
778 */
779 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
780 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
781 if (!syncache_expand(&inc, th, &so, m)) {
782 /*
783 * No syncache entry, or ACK was not
784 * for our SYN/ACK. Send a RST.
785 */
786 tcpstat.tcps_badsyn++;
787 rstreason = BANDLIM_RST_OPENPORT;
788 goto dropwithreset;
789 }
790 if (so == NULL) {
791 /*
792 * Could not complete 3-way handshake,
793 * connection is being closed down, and
794 * syncache will free mbuf.
795 */
796 INP_UNLOCK(inp);
797 INP_INFO_WUNLOCK(&tcbinfo);
798 return;
799 }
800 /*
801 * Socket is created in state SYN_RECEIVED.
802 * Continue processing segment.
803 */
804 INP_UNLOCK(inp);
805 inp = sotoinpcb(so);
806 INP_LOCK(inp);
807 tp = intotcpcb(inp);
808 /*
809 * This is what would have happened in
810 * tcp_output() when the SYN,ACK was sent.
811 */
812 tp->snd_up = tp->snd_una;
813 tp->snd_max = tp->snd_nxt = tp->iss + 1;
814 tp->last_ack_sent = tp->rcv_nxt;
815 /*
816 * RFC1323: The window in SYN & SYN/ACK
817 * segments is never scaled.
818 */
819 tp->snd_wnd = tiwin; /* unscaled */
820 goto after_listen;
821 }
822 if (thflags & TH_RST) {
823 syncache_chkrst(&inc, th);
824 goto drop;
825 }
826 if (thflags & TH_ACK) {
827 syncache_badack(&inc);
828 tcpstat.tcps_badsyn++;
829 rstreason = BANDLIM_RST_OPENPORT;
830 goto dropwithreset;
831 }
832 goto drop;
833 }
834
835 /*
836 * Segment's flags are (SYN) or (SYN|FIN).
837 */
838 #ifdef INET6
839 /*
840 * If deprecated address is forbidden,
841 * we do not accept SYN to deprecated interface
842 * address to prevent any new inbound connection from
843 * getting established.
844 * When we do not accept SYN, we send a TCP RST,
845 * with deprecated source address (instead of dropping
846 * it). We compromise it as it is much better for peer
847 * to send a RST, and RST will be the final packet
848 * for the exchange.
849 *
850 * If we do not forbid deprecated addresses, we accept
851 * the SYN packet. RFC2462 does not suggest dropping
852 * SYN in this case.
853 * If we decipher RFC2462 5.5.4, it says like this:
854 * 1. use of deprecated addr with existing
855 * communication is okay - "SHOULD continue to be
856 * used"
857 * 2. use of it with new communication:
858 * (2a) "SHOULD NOT be used if alternate address
859 * with sufficient scope is available"
860 * (2b) nothing mentioned otherwise.
861 * Here we fall into (2b) case as we have no choice in
862 * our source address selection - we must obey the peer.
863 *
864 * The wording in RFC2462 is confusing, and there are
865 * multiple description text for deprecated address
866 * handling - worse, they are not exactly the same.
867 * I believe 5.5.4 is the best one, so we follow 5.5.4.
868 */
869 if (isipv6 && !ip6_use_deprecated) {
870 struct in6_ifaddr *ia6;
871
872 if ((ia6 = ip6_getdstifaddr(m)) &&
873 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
874 INP_UNLOCK(inp);
875 tp = NULL;
876 rstreason = BANDLIM_RST_OPENPORT;
877 goto dropwithreset;
878 }
879 }
880 #endif
881 /*
882 * If it is from this socket, drop it, it must be forged.
883 * Don't bother responding if the destination was a broadcast.
884 */
885 if (th->th_dport == th->th_sport) {
886 if (isipv6) {
887 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
888 &ip6->ip6_src))
889 goto drop;
890 } else {
891 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
892 goto drop;
893 }
894 }
895 /*
896 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
897 *
898 * Note that it is quite possible to receive unicast
899 * link-layer packets with a broadcast IP address. Use
900 * in_broadcast() to find them.
901 */
902 if (m->m_flags & (M_BCAST|M_MCAST))
903 goto drop;
904 if (isipv6) {
905 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
906 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
907 goto drop;
908 } else {
909 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
910 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
911 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
912 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
913 goto drop;
914 }
915 /*
916 * SYN appears to be valid; create compressed TCP state
917 * for syncache, or perform t/tcp connection.
918 */
919 if (so->so_qlen <= so->so_qlimit) {
920 #ifdef TCPDEBUG
921 if (so->so_options & SO_DEBUG)
922 tcp_trace(TA_INPUT, ostate, tp,
923 (void *)tcp_saveipgen, &tcp_savetcp, 0);
924 #endif
925 tcp_dooptions(&to, optp, optlen, 1);
926 if (!syncache_add(&inc, &to, th, &so, m))
927 goto drop;
928 if (so == NULL) {
929 /*
930 * Entry added to syncache, mbuf used to
931 * send SYN,ACK packet.
932 */
933 KASSERT(headlocked, ("headlocked"));
934 INP_UNLOCK(inp);
935 INP_INFO_WUNLOCK(&tcbinfo);
936 return;
937 }
938 /*
939 * Segment passed TAO tests.
940 */
941 INP_UNLOCK(inp);
942 inp = sotoinpcb(so);
943 INP_LOCK(inp);
944 tp = intotcpcb(inp);
945 tp->snd_wnd = tiwin;
946 tp->t_starttime = ticks;
947 tp->t_state = TCPS_ESTABLISHED;
948
949 /*
950 * T/TCP logic:
951 * If there is a FIN or if there is data, then
952 * delay SYN,ACK(SYN) in the hope of piggy-backing
953 * it on a response segment. Otherwise must send
954 * ACK now in case the other side is slow starting.
955 */
956 if (thflags & TH_FIN || tlen != 0)
957 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
958 else
959 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
960 tcpstat.tcps_connects++;
961 soisconnected(so);
962 goto trimthenstep6;
963 }
964 goto drop;
965 }
966 after_listen:
967
968 /* XXX temp debugging */
969 /* should not happen - syncache should pick up these connections */
970 if (tp->t_state == TCPS_LISTEN)
971 panic("tcp_input: TCPS_LISTEN");
972
973 /*
974 * This is the second part of the MSS DoS prevention code (after
975 * minmss on the sending side) and it deals with too many too small
976 * tcp packets in a too short timeframe (1 second).
977 *
978 * For every full second we count the number of received packets
979 * and bytes. If we get a lot of packets per second for this connection
980 * (tcp_minmssoverload) we take a closer look at it and compute the
981 * average packet size for the past second. If that is less than
982 * tcp_minmss we get too many packets with very small payload which
983 * is not good and burdens our system (and every packet generates
984 * a wakeup to the process connected to our socket). We can reasonable
985 * expect this to be small packet DoS attack to exhaust our CPU
986 * cycles.
987 *
988 * Care has to be taken for the minimum packet overload value. This
989 * value defines the minimum number of packets per second before we
990 * start to worry. This must not be too low to avoid killing for
991 * example interactive connections with many small packets like
992 * telnet or SSH.
993 *
994 * Setting either tcp_minmssoverload or tcp_minmss to "" disables
995 * this check.
996 *
997 * Account for packet if payload packet, skip over ACK, etc.
998 */
999 if (tcp_minmss && tcp_minmssoverload &&
1000 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1001 if (tp->rcv_second > ticks) {
1002 tp->rcv_pps++;
1003 tp->rcv_byps += tlen + off;
1004 if (tp->rcv_pps > tcp_minmssoverload) {
1005 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1006 printf("too many small tcp packets from "
1007 "%s:%u, av. %lubyte/packet, "
1008 "dropping connection\n",
1009 #ifdef INET6
1010 isipv6 ?
1011 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1012 #endif
1013 inet_ntoa(inp->inp_inc.inc_faddr),
1014 inp->inp_inc.inc_fport,
1015 tp->rcv_byps / tp->rcv_pps);
1016 tp = tcp_drop(tp, ECONNRESET);
1017 tcpstat.tcps_minmssdrops++;
1018 goto drop;
1019 }
1020 }
1021 } else {
1022 tp->rcv_second = ticks + hz;
1023 tp->rcv_pps = 1;
1024 tp->rcv_byps = tlen + off;
1025 }
1026 }
1027
1028 /*
1029 * Segment received on connection.
1030 * Reset idle time and keep-alive timer.
1031 */
1032 tp->t_rcvtime = ticks;
1033 if (TCPS_HAVEESTABLISHED(tp->t_state))
1034 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1035
1036 /*
1037 * Process options only when we get SYN/ACK back. The SYN case
1038 * for incoming connections is handled in tcp_syncache.
1039 * XXX this is traditional behavior, may need to be cleaned up.
1040 */
1041 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
1042 if (thflags & TH_SYN) {
1043 if (to.to_flags & TOF_SCALE) {
1044 tp->t_flags |= TF_RCVD_SCALE;
1045 tp->requested_s_scale = to.to_requested_s_scale;
1046 }
1047 if (to.to_flags & TOF_TS) {
1048 tp->t_flags |= TF_RCVD_TSTMP;
1049 tp->ts_recent = to.to_tsval;
1050 tp->ts_recent_age = ticks;
1051 }
1052 if (to.to_flags & (TOF_CC|TOF_CCNEW))
1053 tp->t_flags |= TF_RCVD_CC;
1054 if (to.to_flags & TOF_MSS)
1055 tcp_mss(tp, to.to_mss);
1056 }
1057
1058 /*
1059 * Header prediction: check for the two common cases
1060 * of a uni-directional data xfer. If the packet has
1061 * no control flags, is in-sequence, the window didn't
1062 * change and we're not retransmitting, it's a
1063 * candidate. If the length is zero and the ack moved
1064 * forward, we're the sender side of the xfer. Just
1065 * free the data acked & wake any higher level process
1066 * that was blocked waiting for space. If the length
1067 * is non-zero and the ack didn't move, we're the
1068 * receiver side. If we're getting packets in-order
1069 * (the reassembly queue is empty), add the data to
1070 * the socket buffer and note that we need a delayed ack.
1071 * Make sure that the hidden state-flags are also off.
1072 * Since we check for TCPS_ESTABLISHED above, it can only
1073 * be TH_NEEDSYN.
1074 */
1075 if (tp->t_state == TCPS_ESTABLISHED &&
1076 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1077 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1078 ((to.to_flags & TOF_TS) == 0 ||
1079 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1080 /*
1081 * Using the CC option is compulsory if once started:
1082 * the segment is OK if no T/TCP was negotiated or
1083 * if the segment has a CC option equal to CCrecv
1084 */
1085 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1086 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1087 th->th_seq == tp->rcv_nxt &&
1088 tiwin && tiwin == tp->snd_wnd &&
1089 tp->snd_nxt == tp->snd_max) {
1090
1091 /*
1092 * If last ACK falls within this segment's sequence numbers,
1093 * record the timestamp.
1094 * NOTE that the test is modified according to the latest
1095 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1096 */
1097 if ((to.to_flags & TOF_TS) != 0 &&
1098 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1099 tp->ts_recent_age = ticks;
1100 tp->ts_recent = to.to_tsval;
1101 }
1102
1103 if (tlen == 0) {
1104 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1105 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1106 tp->snd_cwnd >= tp->snd_wnd &&
1107 ((!tcp_do_newreno &&
1108 tp->t_dupacks < tcprexmtthresh) ||
1109 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
1110 KASSERT(headlocked, ("headlocked"));
1111 INP_INFO_WUNLOCK(&tcbinfo);
1112 /*
1113 * this is a pure ack for outstanding data.
1114 */
1115 ++tcpstat.tcps_predack;
1116 /*
1117 * "bad retransmit" recovery
1118 */
1119 if (tp->t_rxtshift == 1 &&
1120 ticks < tp->t_badrxtwin) {
1121 ++tcpstat.tcps_sndrexmitbad;
1122 tp->snd_cwnd = tp->snd_cwnd_prev;
1123 tp->snd_ssthresh =
1124 tp->snd_ssthresh_prev;
1125 tp->snd_recover = tp->snd_recover_prev;
1126 if (tp->t_flags & TF_WASFRECOVERY)
1127 ENTER_FASTRECOVERY(tp);
1128 tp->snd_nxt = tp->snd_max;
1129 tp->t_badrxtwin = 0;
1130 }
1131
1132 /*
1133 * Recalculate the transmit timer / rtt.
1134 *
1135 * Some boxes send broken timestamp replies
1136 * during the SYN+ACK phase, ignore
1137 * timestamps of 0 or we could calculate a
1138 * huge RTT and blow up the retransmit timer.
1139 */
1140 if ((to.to_flags & TOF_TS) != 0 &&
1141 to.to_tsecr) {
1142 tcp_xmit_timer(tp,
1143 ticks - to.to_tsecr + 1);
1144 } else if (tp->t_rtttime &&
1145 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1146 tcp_xmit_timer(tp,
1147 ticks - tp->t_rtttime);
1148 }
1149 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1150 acked = th->th_ack - tp->snd_una;
1151 tcpstat.tcps_rcvackpack++;
1152 tcpstat.tcps_rcvackbyte += acked;
1153 sbdrop(&so->so_snd, acked);
1154 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1155 SEQ_LEQ(th->th_ack, tp->snd_recover))
1156 tp->snd_recover = th->th_ack - 1;
1157 tp->snd_una = th->th_ack;
1158 /*
1159 * pull snd_wl2 up to prevent seq wrap relative
1160 * to th_ack.
1161 */
1162 tp->snd_wl2 = th->th_ack;
1163 tp->t_dupacks = 0;
1164 m_freem(m);
1165 ND6_HINT(tp); /* some progress has been done */
1166
1167 /*
1168 * If all outstanding data are acked, stop
1169 * retransmit timer, otherwise restart timer
1170 * using current (possibly backed-off) value.
1171 * If process is waiting for space,
1172 * wakeup/selwakeup/signal. If data
1173 * are ready to send, let tcp_output
1174 * decide between more output or persist.
1175
1176 #ifdef TCPDEBUG
1177 if (so->so_options & SO_DEBUG)
1178 tcp_trace(TA_INPUT, ostate, tp,
1179 (void *)tcp_saveipgen,
1180 &tcp_savetcp, 0);
1181 #endif
1182 */
1183 if (tp->snd_una == tp->snd_max)
1184 callout_stop(tp->tt_rexmt);
1185 else if (!callout_active(tp->tt_persist))
1186 callout_reset(tp->tt_rexmt,
1187 tp->t_rxtcur,
1188 tcp_timer_rexmt, tp);
1189
1190 sowwakeup(so);
1191 if (so->so_snd.sb_cc)
1192 (void) tcp_output(tp);
1193 goto check_delack;
1194 }
1195 } else if (th->th_ack == tp->snd_una &&
1196 LIST_EMPTY(&tp->t_segq) &&
1197 tlen <= sbspace(&so->so_rcv)) {
1198 KASSERT(headlocked, ("headlocked"));
1199 INP_INFO_WUNLOCK(&tcbinfo);
1200 /*
1201 * this is a pure, in-sequence data packet
1202 * with nothing on the reassembly queue and
1203 * we have enough buffer space to take it.
1204 */
1205 ++tcpstat.tcps_preddat;
1206 tp->rcv_nxt += tlen;
1207 /*
1208 * Pull snd_wl1 up to prevent seq wrap relative to
1209 * th_seq.
1210 */
1211 tp->snd_wl1 = th->th_seq;
1212 /*
1213 * Pull rcv_up up to prevent seq wrap relative to
1214 * rcv_nxt.
1215 */
1216 tp->rcv_up = tp->rcv_nxt;
1217 tcpstat.tcps_rcvpack++;
1218 tcpstat.tcps_rcvbyte += tlen;
1219 ND6_HINT(tp); /* some progress has been done */
1220 /*
1221 #ifdef TCPDEBUG
1222 if (so->so_options & SO_DEBUG)
1223 tcp_trace(TA_INPUT, ostate, tp,
1224 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1225 #endif
1226 * Add data to socket buffer.
1227 */
1228 if (so->so_state & SS_CANTRCVMORE) {
1229 m_freem(m);
1230 } else {
1231 m_adj(m, drop_hdrlen); /* delayed header drop */
1232 sbappendstream(&so->so_rcv, m);
1233 }
1234 sorwakeup(so);
1235 if (DELAY_ACK(tp)) {
1236 tp->t_flags |= TF_DELACK;
1237 } else {
1238 tp->t_flags |= TF_ACKNOW;
1239 tcp_output(tp);
1240 }
1241 goto check_delack;
1242 }
1243 }
1244
1245 /*
1246 * Calculate amount of space in receive window,
1247 * and then do TCP input processing.
1248 * Receive window is amount of space in rcv queue,
1249 * but not less than advertised window.
1250 */
1251 { int win;
1252
1253 win = sbspace(&so->so_rcv);
1254 if (win < 0)
1255 win = 0;
1256 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1257 }
1258
1259 switch (tp->t_state) {
1260
1261 /*
1262 * If the state is SYN_RECEIVED:
1263 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1264 */
1265 case TCPS_SYN_RECEIVED:
1266 if ((thflags & TH_ACK) &&
1267 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1268 SEQ_GT(th->th_ack, tp->snd_max))) {
1269 rstreason = BANDLIM_RST_OPENPORT;
1270 goto dropwithreset;
1271 }
1272 break;
1273
1274 /*
1275 * If the state is SYN_SENT:
1276 * if seg contains an ACK, but not for our SYN, drop the input.
1277 * if seg contains a RST, then drop the connection.
1278 * if seg does not contain SYN, then drop it.
1279 * Otherwise this is an acceptable SYN segment
1280 * initialize tp->rcv_nxt and tp->irs
1281 * if seg contains ack then advance tp->snd_una
1282 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1283 * arrange for segment to be acked (eventually)
1284 * continue processing rest of data/controls, beginning with URG
1285 */
1286 case TCPS_SYN_SENT:
1287 if (tcp_do_rfc1644)
1288 tcp_hc_gettao(&inp->inp_inc, &tao);
1289
1290 if ((thflags & TH_ACK) &&
1291 (SEQ_LEQ(th->th_ack, tp->iss) ||
1292 SEQ_GT(th->th_ack, tp->snd_max))) {
1293 /*
1294 * If we have a cached CCsent for the remote host,
1295 * hence we haven't just crashed and restarted,
1296 * do not send a RST. This may be a retransmission
1297 * from the other side after our earlier ACK was lost.
1298 * Our new SYN, when it arrives, will serve as the
1299 * needed ACK.
1300 */
1301 if (tao.tao_ccsent != 0)
1302 goto drop;
1303 else {
1304 rstreason = BANDLIM_UNLIMITED;
1305 goto dropwithreset;
1306 }
1307 }
1308 if (thflags & TH_RST) {
1309 if (thflags & TH_ACK)
1310 tp = tcp_drop(tp, ECONNREFUSED);
1311 goto drop;
1312 }
1313 if ((thflags & TH_SYN) == 0)
1314 goto drop;
1315 tp->snd_wnd = th->th_win; /* initial send window */
1316 tp->cc_recv = to.to_cc; /* foreign CC */
1317
1318 tp->irs = th->th_seq;
1319 tcp_rcvseqinit(tp);
1320 if (thflags & TH_ACK) {
1321 /*
1322 * Our SYN was acked. If segment contains CC.ECHO
1323 * option, check it to make sure this segment really
1324 * matches our SYN. If not, just drop it as old
1325 * duplicate, but send an RST if we're still playing
1326 * by the old rules. If no CC.ECHO option, make sure
1327 * we don't get fooled into using T/TCP.
1328 */
1329 if (to.to_flags & TOF_CCECHO) {
1330 if (tp->cc_send != to.to_ccecho) {
1331 if (tao.tao_ccsent != 0)
1332 goto drop;
1333 else {
1334 rstreason = BANDLIM_UNLIMITED;
1335 goto dropwithreset;
1336 }
1337 }
1338 } else
1339 tp->t_flags &= ~TF_RCVD_CC;
1340 tcpstat.tcps_connects++;
1341 soisconnected(so);
1342 #ifdef MAC
1343 mac_set_socket_peer_from_mbuf(m, so);
1344 #endif
1345 /* Do window scaling on this connection? */
1346 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1347 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1348 tp->snd_scale = tp->requested_s_scale;
1349 tp->rcv_scale = tp->request_r_scale;
1350 }
1351 /* Segment is acceptable, update cache if undefined. */
1352 if (tao.tao_ccsent == 0 && tcp_do_rfc1644)
1353 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0);
1354
1355 tp->rcv_adv += tp->rcv_wnd;
1356 tp->snd_una++; /* SYN is acked */
1357 /*
1358 * If there's data, delay ACK; if there's also a FIN
1359 * ACKNOW will be turned on later.
1360 */
1361 if (DELAY_ACK(tp) && tlen != 0)
1362 callout_reset(tp->tt_delack, tcp_delacktime,
1363 tcp_timer_delack, tp);
1364 else
1365 tp->t_flags |= TF_ACKNOW;
1366 /*
1367 * Received <SYN,ACK> in SYN_SENT[*] state.
1368 * Transitions:
1369 * SYN_SENT --> ESTABLISHED
1370 * SYN_SENT* --> FIN_WAIT_1
1371 */
1372 tp->t_starttime = ticks;
1373 if (tp->t_flags & TF_NEEDFIN) {
1374 tp->t_state = TCPS_FIN_WAIT_1;
1375 tp->t_flags &= ~TF_NEEDFIN;
1376 thflags &= ~TH_SYN;
1377 } else {
1378 tp->t_state = TCPS_ESTABLISHED;
1379 callout_reset(tp->tt_keep, tcp_keepidle,
1380 tcp_timer_keep, tp);
1381 }
1382 } else {
1383 /*
1384 * Received initial SYN in SYN-SENT[*] state =>
1385 * simultaneous open. If segment contains CC option
1386 * and there is a cached CC, apply TAO test.
1387 * If it succeeds, connection is * half-synchronized.
1388 * Otherwise, do 3-way handshake:
1389 * SYN-SENT -> SYN-RECEIVED
1390 * SYN-SENT* -> SYN-RECEIVED*
1391 * If there was no CC option, clear cached CC value.
1392 */
1393 tp->t_flags |= TF_ACKNOW;
1394 callout_stop(tp->tt_rexmt);
1395 if (to.to_flags & TOF_CC) {
1396 if (tao.tao_cc != 0 &&
1397 CC_GT(to.to_cc, tao.tao_cc)) {
1398 /*
1399 * update cache and make transition:
1400 * SYN-SENT -> ESTABLISHED*
1401 * SYN-SENT* -> FIN-WAIT-1*
1402 */
1403 tao.tao_cc = to.to_cc;
1404 tcp_hc_updatetao(&inp->inp_inc,
1405 TCP_HC_TAO_CC, to.to_cc, 0);
1406 tp->t_starttime = ticks;
1407 if (tp->t_flags & TF_NEEDFIN) {
1408 tp->t_state = TCPS_FIN_WAIT_1;
1409 tp->t_flags &= ~TF_NEEDFIN;
1410 } else {
1411 tp->t_state = TCPS_ESTABLISHED;
1412 callout_reset(tp->tt_keep,
1413 tcp_keepidle,
1414 tcp_timer_keep,
1415 tp);
1416 }
1417 tp->t_flags |= TF_NEEDSYN;
1418 } else
1419 tp->t_state = TCPS_SYN_RECEIVED;
1420 } else {
1421 if (tcp_do_rfc1644) {
1422 /* CC.NEW or no option => invalidate cache */
1423 tao.tao_cc = 0;
1424 tcp_hc_updatetao(&inp->inp_inc,
1425 TCP_HC_TAO_CC, to.to_cc, 0);
1426 }
1427 tp->t_state = TCPS_SYN_RECEIVED;
1428 }
1429 }
1430
1431 trimthenstep6:
1432 /*
1433 * Advance th->th_seq to correspond to first data byte.
1434 * If data, trim to stay within window,
1435 * dropping FIN if necessary.
1436 */
1437 th->th_seq++;
1438 if (tlen > tp->rcv_wnd) {
1439 todrop = tlen - tp->rcv_wnd;
1440 m_adj(m, -todrop);
1441 tlen = tp->rcv_wnd;
1442 thflags &= ~TH_FIN;
1443 tcpstat.tcps_rcvpackafterwin++;
1444 tcpstat.tcps_rcvbyteafterwin += todrop;
1445 }
1446 tp->snd_wl1 = th->th_seq - 1;
1447 tp->rcv_up = th->th_seq;
1448 /*
1449 * Client side of transaction: already sent SYN and data.
1450 * If the remote host used T/TCP to validate the SYN,
1451 * our data will be ACK'd; if so, enter normal data segment
1452 * processing in the middle of step 5, ack processing.
1453 * Otherwise, goto step 6.
1454 */
1455 if (thflags & TH_ACK)
1456 goto process_ACK;
1457
1458 goto step6;
1459
1460 /*
1461 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1462 * if segment contains a SYN and CC [not CC.NEW] option:
1463 * if state == TIME_WAIT and connection duration > MSL,
1464 * drop packet and send RST;
1465 *
1466 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1467 * ack the FIN (and data) in retransmission queue.
1468 * Complete close and delete TCPCB. Then reprocess
1469 * segment, hoping to find new TCPCB in LISTEN state;
1470 *
1471 * else must be old SYN; drop it.
1472 * else do normal processing.
1473 */
1474 case TCPS_LAST_ACK:
1475 case TCPS_CLOSING:
1476 case TCPS_TIME_WAIT:
1477 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1478 if ((thflags & TH_SYN) &&
1479 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1480 if (tp->t_state == TCPS_TIME_WAIT &&
1481 (ticks - tp->t_starttime) > tcp_msl) {
1482 rstreason = BANDLIM_UNLIMITED;
1483 goto dropwithreset;
1484 }
1485 if (CC_GT(to.to_cc, tp->cc_recv)) {
1486 tp = tcp_close(tp);
1487 goto findpcb;
1488 }
1489 else
1490 goto drop;
1491 }
1492 break; /* continue normal processing */
1493 }
1494
1495 /*
1496 * States other than LISTEN or SYN_SENT.
1497 * First check the RST flag and sequence number since reset segments
1498 * are exempt from the timestamp and connection count tests. This
1499 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1500 * below which allowed reset segments in half the sequence space
1501 * to fall though and be processed (which gives forged reset
1502 * segments with a random sequence number a 50 percent chance of
1503 * killing a connection).
1504 * Then check timestamp, if present.
1505 * Then check the connection count, if present.
1506 * Then check that at least some bytes of segment are within
1507 * receive window. If segment begins before rcv_nxt,
1508 * drop leading data (and SYN); if nothing left, just ack.
1509 *
1510 *
1511 * If the RST bit is set, check the sequence number to see
1512 * if this is a valid reset segment.
1513 * RFC 793 page 37:
1514 * In all states except SYN-SENT, all reset (RST) segments
1515 * are validated by checking their SEQ-fields. A reset is
1516 * valid if its sequence number is in the window.
1517 * Note: this does not take into account delayed ACKs, so
1518 * we should test against last_ack_sent instead of rcv_nxt.
1519 * The sequence number in the reset segment is normally an
1520 * echo of our outgoing acknowlegement numbers, but some hosts
1521 * send a reset with the sequence number at the rightmost edge
1522 * of our receive window, and we have to handle this case.
1523 * If we have multiple segments in flight, the intial reset
1524 * segment sequence numbers will be to the left of last_ack_sent,
1525 * but they will eventually catch up.
1526 * In any case, it never made sense to trim reset segments to
1527 * fit the receive window since RFC 1122 says:
1528 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1529 *
1530 * A TCP SHOULD allow a received RST segment to include data.
1531 *
1532 * DISCUSSION
1533 * It has been suggested that a RST segment could contain
1534 * ASCII text that encoded and explained the cause of the
1535 * RST. No standard has yet been established for such
1536 * data.
1537 *
1538 * If the reset segment passes the sequence number test examine
1539 * the state:
1540 * SYN_RECEIVED STATE:
1541 * If passive open, return to LISTEN state.
1542 * If active open, inform user that connection was refused.
1543 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1544 * Inform user that connection was reset, and close tcb.
1545 * CLOSING, LAST_ACK STATES:
1546 * Close the tcb.
1547 * TIME_WAIT STATE:
1548 * Drop the segment - see Stevens, vol. 2, p. 964 and
1549 * RFC 1337.
1550 */
1551 if (thflags & TH_RST) {
1552 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1553 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1554 switch (tp->t_state) {
1555
1556 case TCPS_SYN_RECEIVED:
1557 so->so_error = ECONNREFUSED;
1558 goto close;
1559
1560 case TCPS_ESTABLISHED:
1561 case TCPS_FIN_WAIT_1:
1562 case TCPS_FIN_WAIT_2:
1563 case TCPS_CLOSE_WAIT:
1564 so->so_error = ECONNRESET;
1565 close:
1566 tp->t_state = TCPS_CLOSED;
1567 tcpstat.tcps_drops++;
1568 tp = tcp_close(tp);
1569 break;
1570
1571 case TCPS_CLOSING:
1572 case TCPS_LAST_ACK:
1573 tp = tcp_close(tp);
1574 break;
1575
1576 case TCPS_TIME_WAIT:
1577 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1578 ("timewait"));
1579 break;
1580 }
1581 }
1582 goto drop;
1583 }
1584
1585 /*
1586 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1587 * and it's less than ts_recent, drop it.
1588 */
1589 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1590 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1591
1592 /* Check to see if ts_recent is over 24 days old. */
1593 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1594 /*
1595 * Invalidate ts_recent. If this segment updates
1596 * ts_recent, the age will be reset later and ts_recent
1597 * will get a valid value. If it does not, setting
1598 * ts_recent to zero will at least satisfy the
1599 * requirement that zero be placed in the timestamp
1600 * echo reply when ts_recent isn't valid. The
1601 * age isn't reset until we get a valid ts_recent
1602 * because we don't want out-of-order segments to be
1603 * dropped when ts_recent is old.
1604 */
1605 tp->ts_recent = 0;
1606 } else {
1607 tcpstat.tcps_rcvduppack++;
1608 tcpstat.tcps_rcvdupbyte += tlen;
1609 tcpstat.tcps_pawsdrop++;
1610 if (tlen)
1611 goto dropafterack;
1612 goto drop;
1613 }
1614 }
1615
1616 /*
1617 * T/TCP mechanism
1618 * If T/TCP was negotiated and the segment doesn't have CC,
1619 * or if its CC is wrong then drop the segment.
1620 * RST segments do not have to comply with this.
1621 */
1622 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1623 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1624 goto dropafterack;
1625
1626 /*
1627 * In the SYN-RECEIVED state, validate that the packet belongs to
1628 * this connection before trimming the data to fit the receive
1629 * window. Check the sequence number versus IRS since we know
1630 * the sequence numbers haven't wrapped. This is a partial fix
1631 * for the "LAND" DoS attack.
1632 */
1633 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1634 rstreason = BANDLIM_RST_OPENPORT;
1635 goto dropwithreset;
1636 }
1637
1638 todrop = tp->rcv_nxt - th->th_seq;
1639 if (todrop > 0) {
1640 if (thflags & TH_SYN) {
1641 thflags &= ~TH_SYN;
1642 th->th_seq++;
1643 if (th->th_urp > 1)
1644 th->th_urp--;
1645 else
1646 thflags &= ~TH_URG;
1647 todrop--;
1648 }
1649 /*
1650 * Following if statement from Stevens, vol. 2, p. 960.
1651 */
1652 if (todrop > tlen
1653 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1654 /*
1655 * Any valid FIN must be to the left of the window.
1656 * At this point the FIN must be a duplicate or out
1657 * of sequence; drop it.
1658 */
1659 thflags &= ~TH_FIN;
1660
1661 /*
1662 * Send an ACK to resynchronize and drop any data.
1663 * But keep on processing for RST or ACK.
1664 */
1665 tp->t_flags |= TF_ACKNOW;
1666 todrop = tlen;
1667 tcpstat.tcps_rcvduppack++;
1668 tcpstat.tcps_rcvdupbyte += todrop;
1669 } else {
1670 tcpstat.tcps_rcvpartduppack++;
1671 tcpstat.tcps_rcvpartdupbyte += todrop;
1672 }
1673 drop_hdrlen += todrop; /* drop from the top afterwards */
1674 th->th_seq += todrop;
1675 tlen -= todrop;
1676 if (th->th_urp > todrop)
1677 th->th_urp -= todrop;
1678 else {
1679 thflags &= ~TH_URG;
1680 th->th_urp = 0;
1681 }
1682 }
1683
1684 /*
1685 * If new data are received on a connection after the
1686 * user processes are gone, then RST the other end.
1687 */
1688 if ((so->so_state & SS_NOFDREF) &&
1689 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1690 tp = tcp_close(tp);
1691 tcpstat.tcps_rcvafterclose++;
1692 rstreason = BANDLIM_UNLIMITED;
1693 goto dropwithreset;
1694 }
1695
1696 /*
1697 * If segment ends after window, drop trailing data
1698 * (and PUSH and FIN); if nothing left, just ACK.
1699 */
1700 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1701 if (todrop > 0) {
1702 tcpstat.tcps_rcvpackafterwin++;
1703 if (todrop >= tlen) {
1704 tcpstat.tcps_rcvbyteafterwin += tlen;
1705 /*
1706 * If a new connection request is received
1707 * while in TIME_WAIT, drop the old connection
1708 * and start over if the sequence numbers
1709 * are above the previous ones.
1710 */
1711 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1712 if (thflags & TH_SYN &&
1713 tp->t_state == TCPS_TIME_WAIT &&
1714 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1715 tp = tcp_close(tp);
1716 goto findpcb;
1717 }
1718 /*
1719 * If window is closed can only take segments at
1720 * window edge, and have to drop data and PUSH from
1721 * incoming segments. Continue processing, but
1722 * remember to ack. Otherwise, drop segment
1723 * and ack.
1724 */
1725 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1726 tp->t_flags |= TF_ACKNOW;
1727 tcpstat.tcps_rcvwinprobe++;
1728 } else
1729 goto dropafterack;
1730 } else
1731 tcpstat.tcps_rcvbyteafterwin += todrop;
1732 m_adj(m, -todrop);
1733 tlen -= todrop;
1734 thflags &= ~(TH_PUSH|TH_FIN);
1735 }
1736
1737 /*
1738 * If last ACK falls within this segment's sequence numbers,
1739 * record its timestamp.
1740 * NOTE that the test is modified according to the latest
1741 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1742 */
1743 if ((to.to_flags & TOF_TS) != 0 &&
1744 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1745 tp->ts_recent_age = ticks;
1746 tp->ts_recent = to.to_tsval;
1747 }
1748
1749 /*
1750 * If a SYN is in the window, then this is an
1751 * error and we send an RST and drop the connection.
1752 */
1753 if (thflags & TH_SYN) {
1754 tp = tcp_drop(tp, ECONNRESET);
1755 rstreason = BANDLIM_UNLIMITED;
1756 goto drop;
1757 }
1758
1759 /*
1760 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1761 * flag is on (half-synchronized state), then queue data for
1762 * later processing; else drop segment and return.
1763 */
1764 if ((thflags & TH_ACK) == 0) {
1765 if (tp->t_state == TCPS_SYN_RECEIVED ||
1766 (tp->t_flags & TF_NEEDSYN))
1767 goto step6;
1768 else
1769 goto drop;
1770 }
1771
1772 /*
1773 * Ack processing.
1774 */
1775 switch (tp->t_state) {
1776
1777 /*
1778 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1779 * ESTABLISHED state and continue processing.
1780 * The ACK was checked above.
1781 */
1782 case TCPS_SYN_RECEIVED:
1783
1784 tcpstat.tcps_connects++;
1785 soisconnected(so);
1786 /* Do window scaling? */
1787 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1788 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1789 tp->snd_scale = tp->requested_s_scale;
1790 tp->rcv_scale = tp->request_r_scale;
1791 }
1792 /*
1793 * Upon successful completion of 3-way handshake,
1794 * update cache.CC, pass any queued data to the user,
1795 * and advance state appropriately.
1796 */
1797 if (tcp_do_rfc1644) {
1798 tao.tao_cc = tp->cc_recv;
1799 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC,
1800 tp->cc_recv, 0);
1801 }
1802 /*
1803 * Make transitions:
1804 * SYN-RECEIVED -> ESTABLISHED
1805 * SYN-RECEIVED* -> FIN-WAIT-1
1806 */
1807 tp->t_starttime = ticks;
1808 if (tp->t_flags & TF_NEEDFIN) {
1809 tp->t_state = TCPS_FIN_WAIT_1;
1810 tp->t_flags &= ~TF_NEEDFIN;
1811 } else {
1812 tp->t_state = TCPS_ESTABLISHED;
1813 callout_reset(tp->tt_keep, tcp_keepidle,
1814 tcp_timer_keep, tp);
1815 }
1816 /*
1817 * If segment contains data or ACK, will call tcp_reass()
1818 * later; if not, do so now to pass queued data to user.
1819 */
1820 if (tlen == 0 && (thflags & TH_FIN) == 0)
1821 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1822 (struct mbuf *)0);
1823 tp->snd_wl1 = th->th_seq - 1;
1824 /* FALLTHROUGH */
1825
1826 /*
1827 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1828 * ACKs. If the ack is in the range
1829 * tp->snd_una < th->th_ack <= tp->snd_max
1830 * then advance tp->snd_una to th->th_ack and drop
1831 * data from the retransmission queue. If this ACK reflects
1832 * more up to date window information we update our window information.
1833 */
1834 case TCPS_ESTABLISHED:
1835 case TCPS_FIN_WAIT_1:
1836 case TCPS_FIN_WAIT_2:
1837 case TCPS_CLOSE_WAIT:
1838 case TCPS_CLOSING:
1839 case TCPS_LAST_ACK:
1840 case TCPS_TIME_WAIT:
1841 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1842 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1843 if (tlen == 0 && tiwin == tp->snd_wnd) {
1844 tcpstat.tcps_rcvdupack++;
1845 /*
1846 * If we have outstanding data (other than
1847 * a window probe), this is a completely
1848 * duplicate ack (ie, window info didn't
1849 * change), the ack is the biggest we've
1850 * seen and we've seen exactly our rexmt
1851 * threshhold of them, assume a packet
1852 * has been dropped and retransmit it.
1853 * Kludge snd_nxt & the congestion
1854 * window so we send only this one
1855 * packet.
1856 *
1857 * We know we're losing at the current
1858 * window size so do congestion avoidance
1859 * (set ssthresh to half the current window
1860 * and pull our congestion window back to
1861 * the new ssthresh).
1862 *
1863 * Dup acks mean that packets have left the
1864 * network (they're now cached at the receiver)
1865 * so bump cwnd by the amount in the receiver
1866 * to keep a constant cwnd packets in the
1867 * network.
1868 */
1869 if (!callout_active(tp->tt_rexmt) ||
1870 th->th_ack != tp->snd_una)
1871 tp->t_dupacks = 0;
1872 else if (++tp->t_dupacks > tcprexmtthresh ||
1873 (tcp_do_newreno &&
1874 IN_FASTRECOVERY(tp))) {
1875 tp->snd_cwnd += tp->t_maxseg;
1876 (void) tcp_output(tp);
1877 goto drop;
1878 } else if (tp->t_dupacks == tcprexmtthresh) {
1879 tcp_seq onxt = tp->snd_nxt;
1880 u_int win;
1881 if (tcp_do_newreno &&
1882 SEQ_LEQ(th->th_ack,
1883 tp->snd_recover)) {
1884 tp->t_dupacks = 0;
1885 break;
1886 }
1887 win = min(tp->snd_wnd, tp->snd_cwnd) /
1888 2 / tp->t_maxseg;
1889 if (win < 2)
1890 win = 2;
1891 tp->snd_ssthresh = win * tp->t_maxseg;
1892 ENTER_FASTRECOVERY(tp);
1893 tp->snd_recover = tp->snd_max;
1894 callout_stop(tp->tt_rexmt);
1895 tp->t_rtttime = 0;
1896 tp->snd_nxt = th->th_ack;
1897 tp->snd_cwnd = tp->t_maxseg;
1898 (void) tcp_output(tp);
1899 KASSERT(tp->snd_limited <= 2,
1900 ("tp->snd_limited too big"));
1901 tp->snd_cwnd = tp->snd_ssthresh +
1902 tp->t_maxseg *
1903 (tp->t_dupacks - tp->snd_limited);
1904 if (SEQ_GT(onxt, tp->snd_nxt))
1905 tp->snd_nxt = onxt;
1906 goto drop;
1907 } else if (tcp_do_rfc3042) {
1908 u_long oldcwnd = tp->snd_cwnd;
1909 tcp_seq oldsndmax = tp->snd_max;
1910 u_int sent;
1911 KASSERT(tp->t_dupacks == 1 ||
1912 tp->t_dupacks == 2,
1913 ("dupacks not 1 or 2"));
1914 if (tp->t_dupacks == 1) {
1915 tp->snd_limited = 0;
1916 tp->snd_cwnd += tp->t_maxseg;
1917 } else {
1918 tp->snd_cwnd +=
1919 tp->t_maxseg * 2;
1920 }
1921 (void) tcp_output(tp);
1922 sent = tp->snd_max - oldsndmax;
1923 if (sent > tp->t_maxseg) {
1924 KASSERT(tp->snd_limited == 0 &&
1925 tp->t_dupacks == 2,
1926 ("sent too much"));
1927 tp->snd_limited = 2;
1928 } else if (sent > 0)
1929 ++tp->snd_limited;
1930 tp->snd_cwnd = oldcwnd;
1931 goto drop;
1932 }
1933 } else
1934 tp->t_dupacks = 0;
1935 break;
1936 }
1937
1938 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1939
1940 /*
1941 * If the congestion window was inflated to account
1942 * for the other side's cached packets, retract it.
1943 */
1944 if (tcp_do_newreno) {
1945 if (IN_FASTRECOVERY(tp)) {
1946 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1947 tcp_newreno_partial_ack(tp, th);
1948 } else {
1949 /*
1950 * Window inflation should have left us
1951 * with approximately snd_ssthresh
1952 * outstanding data.
1953 * But in case we would be inclined to
1954 * send a burst, better to do it via
1955 * the slow start mechanism.
1956 */
1957 if (SEQ_GT(th->th_ack +
1958 tp->snd_ssthresh,
1959 tp->snd_max))
1960 tp->snd_cwnd = tp->snd_max -
1961 th->th_ack +
1962 tp->t_maxseg;
1963 else
1964 tp->snd_cwnd = tp->snd_ssthresh;
1965 }
1966 }
1967 } else {
1968 if (tp->t_dupacks >= tcprexmtthresh &&
1969 tp->snd_cwnd > tp->snd_ssthresh)
1970 tp->snd_cwnd = tp->snd_ssthresh;
1971 }
1972 tp->t_dupacks = 0;
1973 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1974 tcpstat.tcps_rcvacktoomuch++;
1975 goto dropafterack;
1976 }
1977 /*
1978 * If we reach this point, ACK is not a duplicate,
1979 * i.e., it ACKs something we sent.
1980 */
1981 if (tp->t_flags & TF_NEEDSYN) {
1982 /*
1983 * T/TCP: Connection was half-synchronized, and our
1984 * SYN has been ACK'd (so connection is now fully
1985 * synchronized). Go to non-starred state,
1986 * increment snd_una for ACK of SYN, and check if
1987 * we can do window scaling.
1988 */
1989 tp->t_flags &= ~TF_NEEDSYN;
1990 tp->snd_una++;
1991 /* Do window scaling? */
1992 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1993 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1994 tp->snd_scale = tp->requested_s_scale;
1995 tp->rcv_scale = tp->request_r_scale;
1996 }
1997 }
1998
1999 process_ACK:
2000 acked = th->th_ack - tp->snd_una;
2001 tcpstat.tcps_rcvackpack++;
2002 tcpstat.tcps_rcvackbyte += acked;
2003
2004 /*
2005 * If we just performed our first retransmit, and the ACK
2006 * arrives within our recovery window, then it was a mistake
2007 * to do the retransmit in the first place. Recover our
2008 * original cwnd and ssthresh, and proceed to transmit where
2009 * we left off.
2010 */
2011 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2012 ++tcpstat.tcps_sndrexmitbad;
2013 tp->snd_cwnd = tp->snd_cwnd_prev;
2014 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2015 tp->snd_recover = tp->snd_recover_prev;
2016 if (tp->t_flags & TF_WASFRECOVERY)
2017 ENTER_FASTRECOVERY(tp);
2018 tp->snd_nxt = tp->snd_max;
2019 tp->t_badrxtwin = 0; /* XXX probably not required */
2020 }
2021
2022 /*
2023 * If we have a timestamp reply, update smoothed
2024 * round trip time. If no timestamp is present but
2025 * transmit timer is running and timed sequence
2026 * number was acked, update smoothed round trip time.
2027 * Since we now have an rtt measurement, cancel the
2028 * timer backoff (cf., Phil Karn's retransmit alg.).
2029 * Recompute the initial retransmit timer.
2030 *
2031 * Some boxes send broken timestamp replies
2032 * during the SYN+ACK phase, ignore
2033 * timestamps of 0 or we could calculate a
2034 * huge RTT and blow up the retransmit timer.
2035 */
2036 if ((to.to_flags & TOF_TS) != 0 &&
2037 to.to_tsecr) {
2038 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2039 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2040 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2041 }
2042 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2043
2044 /*
2045 * If all outstanding data is acked, stop retransmit
2046 * timer and remember to restart (more output or persist).
2047 * If there is more data to be acked, restart retransmit
2048 * timer, using current (possibly backed-off) value.
2049 */
2050 if (th->th_ack == tp->snd_max) {
2051 callout_stop(tp->tt_rexmt);
2052 needoutput = 1;
2053 } else if (!callout_active(tp->tt_persist))
2054 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2055 tcp_timer_rexmt, tp);
2056
2057 /*
2058 * If no data (only SYN) was ACK'd,
2059 * skip rest of ACK processing.
2060 */
2061 if (acked == 0)
2062 goto step6;
2063
2064 /*
2065 * When new data is acked, open the congestion window.
2066 * If the window gives us less than ssthresh packets
2067 * in flight, open exponentially (maxseg per packet).
2068 * Otherwise open linearly: maxseg per window
2069 * (maxseg^2 / cwnd per packet).
2070 */
2071 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
2072 register u_int cw = tp->snd_cwnd;
2073 register u_int incr = tp->t_maxseg;
2074 if (cw > tp->snd_ssthresh)
2075 incr = incr * incr / cw;
2076 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2077 }
2078 if (acked > so->so_snd.sb_cc) {
2079 tp->snd_wnd -= so->so_snd.sb_cc;
2080 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2081 ourfinisacked = 1;
2082 } else {
2083 sbdrop(&so->so_snd, acked);
2084 tp->snd_wnd -= acked;
2085 ourfinisacked = 0;
2086 }
2087 sowwakeup(so);
2088 /* detect una wraparound */
2089 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
2090 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2091 SEQ_LEQ(th->th_ack, tp->snd_recover))
2092 tp->snd_recover = th->th_ack - 1;
2093 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
2094 SEQ_GEQ(th->th_ack, tp->snd_recover))
2095 EXIT_FASTRECOVERY(tp);
2096 tp->snd_una = th->th_ack;
2097 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2098 tp->snd_nxt = tp->snd_una;
2099
2100 switch (tp->t_state) {
2101
2102 /*
2103 * In FIN_WAIT_1 STATE in addition to the processing
2104 * for the ESTABLISHED state if our FIN is now acknowledged
2105 * then enter FIN_WAIT_2.
2106 */
2107 case TCPS_FIN_WAIT_1:
2108 if (ourfinisacked) {
2109 /*
2110 * If we can't receive any more
2111 * data, then closing user can proceed.
2112 * Starting the timer is contrary to the
2113 * specification, but if we don't get a FIN
2114 * we'll hang forever.
2115 */
2116 /* XXXjl
2117 * we should release the tp also, and use a
2118 * compressed state.
2119 */
2120 if (so->so_state & SS_CANTRCVMORE) {
2121 soisdisconnected(so);
2122 callout_reset(tp->tt_2msl, tcp_maxidle,
2123 tcp_timer_2msl, tp);
2124 }
2125 tp->t_state = TCPS_FIN_WAIT_2;
2126 }
2127 break;
2128
2129 /*
2130 * In CLOSING STATE in addition to the processing for
2131 * the ESTABLISHED state if the ACK acknowledges our FIN
2132 * then enter the TIME-WAIT state, otherwise ignore
2133 * the segment.
2134 */
2135 case TCPS_CLOSING:
2136 if (ourfinisacked) {
2137 KASSERT(headlocked, ("headlocked"));
2138 tcp_twstart(tp);
2139 INP_INFO_WUNLOCK(&tcbinfo);
2140 m_freem(m);
2141 return;
2142 }
2143 break;
2144
2145 /*
2146 * In LAST_ACK, we may still be waiting for data to drain
2147 * and/or to be acked, as well as for the ack of our FIN.
2148 * If our FIN is now acknowledged, delete the TCB,
2149 * enter the closed state and return.
2150 */
2151 case TCPS_LAST_ACK:
2152 if (ourfinisacked) {
2153 tp = tcp_close(tp);
2154 goto drop;
2155 }
2156 break;
2157
2158 /*
2159 * In TIME_WAIT state the only thing that should arrive
2160 * is a retransmission of the remote FIN. Acknowledge
2161 * it and restart the finack timer.
2162 */
2163 case TCPS_TIME_WAIT:
2164 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2165 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2166 tcp_timer_2msl, tp);
2167 goto dropafterack;
2168 }
2169 }
2170
2171 step6:
2172 /*
2173 * Update window information.
2174 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2175 */
2176 if ((thflags & TH_ACK) &&
2177 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2178 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2179 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2180 /* keep track of pure window updates */
2181 if (tlen == 0 &&
2182 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2183 tcpstat.tcps_rcvwinupd++;
2184 tp->snd_wnd = tiwin;
2185 tp->snd_wl1 = th->th_seq;
2186 tp->snd_wl2 = th->th_ack;
2187 if (tp->snd_wnd > tp->max_sndwnd)
2188 tp->max_sndwnd = tp->snd_wnd;
2189 needoutput = 1;
2190 }
2191
2192 /*
2193 * Process segments with URG.
2194 */
2195 if ((thflags & TH_URG) && th->th_urp &&
2196 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2197 /*
2198 * This is a kludge, but if we receive and accept
2199 * random urgent pointers, we'll crash in
2200 * soreceive. It's hard to imagine someone
2201 * actually wanting to send this much urgent data.
2202 */
2203 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2204 th->th_urp = 0; /* XXX */
2205 thflags &= ~TH_URG; /* XXX */
2206 goto dodata; /* XXX */
2207 }
2208 /*
2209 * If this segment advances the known urgent pointer,
2210 * then mark the data stream. This should not happen
2211 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2212 * a FIN has been received from the remote side.
2213 * In these states we ignore the URG.
2214 *
2215 * According to RFC961 (Assigned Protocols),
2216 * the urgent pointer points to the last octet
2217 * of urgent data. We continue, however,
2218 * to consider it to indicate the first octet
2219 * of data past the urgent section as the original
2220 * spec states (in one of two places).
2221 */
2222 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2223 tp->rcv_up = th->th_seq + th->th_urp;
2224 so->so_oobmark = so->so_rcv.sb_cc +
2225 (tp->rcv_up - tp->rcv_nxt) - 1;
2226 if (so->so_oobmark == 0)
2227 so->so_state |= SS_RCVATMARK;
2228 sohasoutofband(so);
2229 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2230 }
2231 /*
2232 * Remove out of band data so doesn't get presented to user.
2233 * This can happen independent of advancing the URG pointer,
2234 * but if two URG's are pending at once, some out-of-band
2235 * data may creep in... ick.
2236 */
2237 if (th->th_urp <= (u_long)tlen &&
2238 !(so->so_options & SO_OOBINLINE)) {
2239 /* hdr drop is delayed */
2240 tcp_pulloutofband(so, th, m, drop_hdrlen);
2241 }
2242 } else {
2243 /*
2244 * If no out of band data is expected,
2245 * pull receive urgent pointer along
2246 * with the receive window.
2247 */
2248 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2249 tp->rcv_up = tp->rcv_nxt;
2250 }
2251 dodata: /* XXX */
2252 KASSERT(headlocked, ("headlocked"));
2253 /*
2254 * Process the segment text, merging it into the TCP sequencing queue,
2255 * and arranging for acknowledgment of receipt if necessary.
2256 * This process logically involves adjusting tp->rcv_wnd as data
2257 * is presented to the user (this happens in tcp_usrreq.c,
2258 * case PRU_RCVD). If a FIN has already been received on this
2259 * connection then we just ignore the text.
2260 */
2261 if ((tlen || (thflags & TH_FIN)) &&
2262 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2263 m_adj(m, drop_hdrlen); /* delayed header drop */
2264 /*
2265 * Insert segment which includes th into TCP reassembly queue
2266 * with control block tp. Set thflags to whether reassembly now
2267 * includes a segment with FIN. This handles the common case
2268 * inline (segment is the next to be received on an established
2269 * connection, and the queue is empty), avoiding linkage into
2270 * and removal from the queue and repetition of various
2271 * conversions.
2272 * Set DELACK for segments received in order, but ack
2273 * immediately when segments are out of order (so
2274 * fast retransmit can work).
2275 */
2276 if (th->th_seq == tp->rcv_nxt &&
2277 LIST_EMPTY(&tp->t_segq) &&
2278 TCPS_HAVEESTABLISHED(tp->t_state)) {
2279 if (DELAY_ACK(tp))
2280 tp->t_flags |= TF_DELACK;
2281 else
2282 tp->t_flags |= TF_ACKNOW;
2283 tp->rcv_nxt += tlen;
2284 thflags = th->th_flags & TH_FIN;
2285 tcpstat.tcps_rcvpack++;
2286 tcpstat.tcps_rcvbyte += tlen;
2287 ND6_HINT(tp);
2288 if (so->so_state & SS_CANTRCVMORE)
2289 m_freem(m);
2290 else
2291 sbappendstream(&so->so_rcv, m);
2292 sorwakeup(so);
2293 } else {
2294 thflags = tcp_reass(tp, th, &tlen, m);
2295 tp->t_flags |= TF_ACKNOW;
2296 }
2297
2298 /*
2299 * Note the amount of data that peer has sent into
2300 * our window, in order to estimate the sender's
2301 * buffer size.
2302 */
2303 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2304 } else {
2305 m_freem(m);
2306 thflags &= ~TH_FIN;
2307 }
2308
2309 /*
2310 * If FIN is received ACK the FIN and let the user know
2311 * that the connection is closing.
2312 */
2313 if (thflags & TH_FIN) {
2314 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2315 socantrcvmore(so);
2316 /*
2317 * If connection is half-synchronized
2318 * (ie NEEDSYN flag on) then delay ACK,
2319 * so it may be piggybacked when SYN is sent.
2320 * Otherwise, since we received a FIN then no
2321 * more input can be expected, send ACK now.
2322 */
2323 if (tp->t_flags & TF_NEEDSYN)
2324 tp->t_flags |= TF_DELACK;
2325 else
2326 tp->t_flags |= TF_ACKNOW;
2327 tp->rcv_nxt++;
2328 }
2329 switch (tp->t_state) {
2330
2331 /*
2332 * In SYN_RECEIVED and ESTABLISHED STATES
2333 * enter the CLOSE_WAIT state.
2334 */
2335 case TCPS_SYN_RECEIVED:
2336 tp->t_starttime = ticks;
2337 /*FALLTHROUGH*/
2338 case TCPS_ESTABLISHED:
2339 tp->t_state = TCPS_CLOSE_WAIT;
2340 break;
2341
2342 /*
2343 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2344 * enter the CLOSING state.
2345 */
2346 case TCPS_FIN_WAIT_1:
2347 tp->t_state = TCPS_CLOSING;
2348 break;
2349
2350 /*
2351 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2352 * starting the time-wait timer, turning off the other
2353 * standard timers.
2354 */
2355 case TCPS_FIN_WAIT_2:
2356 KASSERT(headlocked == 1, ("headlocked should be 1"));
2357 tcp_twstart(tp);
2358 INP_INFO_WUNLOCK(&tcbinfo);
2359 return;
2360
2361 /*
2362 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2363 */
2364 case TCPS_TIME_WAIT:
2365 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2366 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2367 tcp_timer_2msl, tp);
2368 break;
2369 }
2370 }
2371 INP_INFO_WUNLOCK(&tcbinfo);
2372 #ifdef TCPDEBUG
2373 if (so->so_options & SO_DEBUG)
2374 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2375 &tcp_savetcp, 0);
2376 #endif
2377
2378 /*
2379 * Return any desired output.
2380 */
2381 if (needoutput || (tp->t_flags & TF_ACKNOW))
2382 (void) tcp_output(tp);
2383
2384 check_delack:
2385 if (tp->t_flags & TF_DELACK) {
2386 tp->t_flags &= ~TF_DELACK;
2387 callout_reset(tp->tt_delack, tcp_delacktime,
2388 tcp_timer_delack, tp);
2389 }
2390 INP_UNLOCK(inp);
2391 return;
2392
2393 dropafterack:
2394 /*
2395 * Generate an ACK dropping incoming segment if it occupies
2396 * sequence space, where the ACK reflects our state.
2397 *
2398 * We can now skip the test for the RST flag since all
2399 * paths to this code happen after packets containing
2400 * RST have been dropped.
2401 *
2402 * In the SYN-RECEIVED state, don't send an ACK unless the
2403 * segment we received passes the SYN-RECEIVED ACK test.
2404 * If it fails send a RST. This breaks the loop in the
2405 * "LAND" DoS attack, and also prevents an ACK storm
2406 * between two listening ports that have been sent forged
2407 * SYN segments, each with the source address of the other.
2408 */
2409 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2410 (SEQ_GT(tp->snd_una, th->th_ack) ||
2411 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2412 rstreason = BANDLIM_RST_OPENPORT;
2413 goto dropwithreset;
2414 }
2415 #ifdef TCPDEBUG
2416 if (so->so_options & SO_DEBUG)
2417 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2418 &tcp_savetcp, 0);
2419 #endif
2420 KASSERT(headlocked, ("headlocked should be 1"));
2421 INP_INFO_WUNLOCK(&tcbinfo);
2422 m_freem(m);
2423 tp->t_flags |= TF_ACKNOW;
2424 (void) tcp_output(tp);
2425 INP_UNLOCK(inp);
2426 return;
2427
2428 dropwithreset:
2429 /*
2430 * Generate a RST, dropping incoming segment.
2431 * Make ACK acceptable to originator of segment.
2432 * Don't bother to respond if destination was broadcast/multicast.
2433 */
2434 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2435 goto drop;
2436 if (isipv6) {
2437 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2438 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2439 goto drop;
2440 } else {
2441 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2442 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2443 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2444 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2445 goto drop;
2446 }
2447 /* IPv6 anycast check is done at tcp6_input() */
2448
2449 /*
2450 * Perform bandwidth limiting.
2451 */
2452 if (badport_bandlim(rstreason) < 0)
2453 goto drop;
2454
2455 #ifdef TCPDEBUG
2456 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2457 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2458 &tcp_savetcp, 0);
2459 #endif
2460
2461 if (thflags & TH_ACK)
2462 /* mtod() below is safe as long as hdr dropping is delayed */
2463 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2464 TH_RST);
2465 else {
2466 if (thflags & TH_SYN)
2467 tlen++;
2468 /* mtod() below is safe as long as hdr dropping is delayed */
2469 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2470 (tcp_seq)0, TH_RST|TH_ACK);
2471 }
2472
2473 if (tp)
2474 INP_UNLOCK(inp);
2475 if (headlocked)
2476 INP_INFO_WUNLOCK(&tcbinfo);
2477 return;
2478
2479 drop:
2480 /*
2481 * Drop space held by incoming segment and return.
2482 */
2483 #ifdef TCPDEBUG
2484 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2485 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2486 &tcp_savetcp, 0);
2487 #endif
2488 if (tp)
2489 INP_UNLOCK(inp);
2490 m_freem(m);
2491 if (headlocked)
2492 INP_INFO_WUNLOCK(&tcbinfo);
2493 return;
2494 }
2495
2496 /*
2497 * Parse TCP options and place in tcpopt.
2498 */
2499 static void
2500 tcp_dooptions(to, cp, cnt, is_syn)
2501 struct tcpopt *to;
2502 u_char *cp;
2503 int cnt;
2504 int is_syn;
2505 {
2506 int opt, optlen;
2507
2508 to->to_flags = 0;
2509 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2510 opt = cp[0];
2511 if (opt == TCPOPT_EOL)
2512 break;
2513 if (opt == TCPOPT_NOP)
2514 optlen = 1;
2515 else {
2516 if (cnt < 2)
2517 break;
2518 optlen = cp[1];
2519 if (optlen < 2 || optlen > cnt)
2520 break;
2521 }
2522 switch (opt) {
2523 case TCPOPT_MAXSEG:
2524 if (optlen != TCPOLEN_MAXSEG)
2525 continue;
2526 if (!is_syn)
2527 continue;
2528 to->to_flags |= TOF_MSS;
2529 bcopy((char *)cp + 2,
2530 (char *)&to->to_mss, sizeof(to->to_mss));
2531 to->to_mss = ntohs(to->to_mss);
2532 break;
2533 case TCPOPT_WINDOW:
2534 if (optlen != TCPOLEN_WINDOW)
2535 continue;
2536 if (! is_syn)
2537 continue;
2538 to->to_flags |= TOF_SCALE;
2539 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2540 break;
2541 case TCPOPT_TIMESTAMP:
2542 if (optlen != TCPOLEN_TIMESTAMP)
2543 continue;
2544 to->to_flags |= TOF_TS;
2545 bcopy((char *)cp + 2,
2546 (char *)&to->to_tsval, sizeof(to->to_tsval));
2547 to->to_tsval = ntohl(to->to_tsval);
2548 bcopy((char *)cp + 6,
2549 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2550 to->to_tsecr = ntohl(to->to_tsecr);
2551 break;
2552 case TCPOPT_CC:
2553 if (optlen != TCPOLEN_CC)
2554 continue;
2555 to->to_flags |= TOF_CC;
2556 bcopy((char *)cp + 2,
2557 (char *)&to->to_cc, sizeof(to->to_cc));
2558 to->to_cc = ntohl(to->to_cc);
2559 break;
2560 case TCPOPT_CCNEW:
2561 if (optlen != TCPOLEN_CC)
2562 continue;
2563 if (!is_syn)
2564 continue;
2565 to->to_flags |= TOF_CCNEW;
2566 bcopy((char *)cp + 2,
2567 (char *)&to->to_cc, sizeof(to->to_cc));
2568 to->to_cc = ntohl(to->to_cc);
2569 break;
2570 case TCPOPT_CCECHO:
2571 if (optlen != TCPOLEN_CC)
2572 continue;
2573 if (!is_syn)
2574 continue;
2575 to->to_flags |= TOF_CCECHO;
2576 bcopy((char *)cp + 2,
2577 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2578 to->to_ccecho = ntohl(to->to_ccecho);
2579 break;
2580 default:
2581 continue;
2582 }
2583 }
2584 }
2585
2586 /*
2587 * Pull out of band byte out of a segment so
2588 * it doesn't appear in the user's data queue.
2589 * It is still reflected in the segment length for
2590 * sequencing purposes.
2591 */
2592 static void
2593 tcp_pulloutofband(so, th, m, off)
2594 struct socket *so;
2595 struct tcphdr *th;
2596 register struct mbuf *m;
2597 int off; /* delayed to be droped hdrlen */
2598 {
2599 int cnt = off + th->th_urp - 1;
2600
2601 while (cnt >= 0) {
2602 if (m->m_len > cnt) {
2603 char *cp = mtod(m, caddr_t) + cnt;
2604 struct tcpcb *tp = sototcpcb(so);
2605
2606 tp->t_iobc = *cp;
2607 tp->t_oobflags |= TCPOOB_HAVEDATA;
2608 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2609 m->m_len--;
2610 if (m->m_flags & M_PKTHDR)
2611 m->m_pkthdr.len--;
2612 return;
2613 }
2614 cnt -= m->m_len;
2615 m = m->m_next;
2616 if (m == 0)
2617 break;
2618 }
2619 panic("tcp_pulloutofband");
2620 }
2621
2622 /*
2623 * Collect new round-trip time estimate
2624 * and update averages and current timeout.
2625 */
2626 static void
2627 tcp_xmit_timer(tp, rtt)
2628 register struct tcpcb *tp;
2629 int rtt;
2630 {
2631 register int delta;
2632
2633 tcpstat.tcps_rttupdated++;
2634 tp->t_rttupdated++;
2635 if (tp->t_srtt != 0) {
2636 /*
2637 * srtt is stored as fixed point with 5 bits after the
2638 * binary point (i.e., scaled by 8). The following magic
2639 * is equivalent to the smoothing algorithm in rfc793 with
2640 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2641 * point). Adjust rtt to origin 0.
2642 */
2643 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2644 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2645
2646 if ((tp->t_srtt += delta) <= 0)
2647 tp->t_srtt = 1;
2648
2649 /*
2650 * We accumulate a smoothed rtt variance (actually, a
2651 * smoothed mean difference), then set the retransmit
2652 * timer to smoothed rtt + 4 times the smoothed variance.
2653 * rttvar is stored as fixed point with 4 bits after the
2654 * binary point (scaled by 16). The following is
2655 * equivalent to rfc793 smoothing with an alpha of .75
2656 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2657 * rfc793's wired-in beta.
2658 */
2659 if (delta < 0)
2660 delta = -delta;
2661 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2662 if ((tp->t_rttvar += delta) <= 0)
2663 tp->t_rttvar = 1;
2664 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2665 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2666 } else {
2667 /*
2668 * No rtt measurement yet - use the unsmoothed rtt.
2669 * Set the variance to half the rtt (so our first
2670 * retransmit happens at 3*rtt).
2671 */
2672 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2673 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2674 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2675 }
2676 tp->t_rtttime = 0;
2677 tp->t_rxtshift = 0;
2678
2679 /*
2680 * the retransmit should happen at rtt + 4 * rttvar.
2681 * Because of the way we do the smoothing, srtt and rttvar
2682 * will each average +1/2 tick of bias. When we compute
2683 * the retransmit timer, we want 1/2 tick of rounding and
2684 * 1 extra tick because of +-1/2 tick uncertainty in the
2685 * firing of the timer. The bias will give us exactly the
2686 * 1.5 tick we need. But, because the bias is
2687 * statistical, we have to test that we don't drop below
2688 * the minimum feasible timer (which is 2 ticks).
2689 */
2690 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2691 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2692
2693 /*
2694 * We received an ack for a packet that wasn't retransmitted;
2695 * it is probably safe to discard any error indications we've
2696 * received recently. This isn't quite right, but close enough
2697 * for now (a route might have failed after we sent a segment,
2698 * and the return path might not be symmetrical).
2699 */
2700 tp->t_softerror = 0;
2701 }
2702
2703 /*
2704 * Determine a reasonable value for maxseg size.
2705 * If the route is known, check route for mtu.
2706 * If none, use an mss that can be handled on the outgoing
2707 * interface without forcing IP to fragment; if bigger than
2708 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2709 * to utilize large mbufs. If no route is found, route has no mtu,
2710 * or the destination isn't local, use a default, hopefully conservative
2711 * size (usually 512 or the default IP max size, but no more than the mtu
2712 * of the interface), as we can't discover anything about intervening
2713 * gateways or networks. We also initialize the congestion/slow start
2714 * window to be a single segment if the destination isn't local.
2715 * While looking at the routing entry, we also initialize other path-dependent
2716 * parameters from pre-set or cached values in the routing entry.
2717 *
2718 * Also take into account the space needed for options that we
2719 * send regularly. Make maxseg shorter by that amount to assure
2720 * that we can send maxseg amount of data even when the options
2721 * are present. Store the upper limit of the length of options plus
2722 * data in maxopd.
2723 *
2724 *
2725 * In case of T/TCP, we call this routine during implicit connection
2726 * setup as well (offer = -1), to initialize maxseg from the cached
2727 * MSS of our peer.
2728 *
2729 * NOTE that this routine is only called when we process an incoming
2730 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2731 */
2732 void
2733 tcp_mss(tp, offer)
2734 struct tcpcb *tp;
2735 int offer;
2736 {
2737 int rtt, mss;
2738 u_long bufsize;
2739 u_long maxmtu;
2740 struct inpcb *inp = tp->t_inpcb;
2741 struct socket *so;
2742 struct hc_metrics_lite metrics;
2743 struct rmxp_tao tao;
2744 int origoffer = offer;
2745 #ifdef INET6
2746 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2747 size_t min_protoh = isipv6 ?
2748 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2749 sizeof (struct tcpiphdr);
2750 #else
2751 const size_t min_protoh = sizeof(struct tcpiphdr);
2752 #endif
2753 bzero(&tao, sizeof(tao));
2754
2755 /* initialize */
2756 #ifdef INET6
2757 if (isipv6) {
2758 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2759 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2760 } else
2761 #endif
2762 {
2763 maxmtu = tcp_maxmtu(&inp->inp_inc);
2764 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2765 }
2766 so = inp->inp_socket;
2767
2768 /*
2769 * no route to sender, take default mss and return
2770 */
2771 if (maxmtu == 0)
2772 return;
2773
2774 /* what have we got? */
2775 switch (offer) {
2776 case 0:
2777 /*
2778 * Offer == 0 means that there was no MSS on the SYN
2779 * segment, in this case we use tcp_mssdflt.
2780 */
2781 offer =
2782 #ifdef INET6
2783 isipv6 ? tcp_v6mssdflt :
2784 #endif
2785 tcp_mssdflt;
2786 break;
2787
2788 case -1:
2789 /*
2790 * Offer == -1 means that we didn't receive SYN yet,
2791 * use cached value in that case;
2792 */
2793 if (tcp_do_rfc1644)
2794 tcp_hc_gettao(&inp->inp_inc, &tao);
2795 if (tao.tao_mssopt != 0)
2796 offer = tao.tao_mssopt;
2797 /* FALLTHROUGH */
2798
2799 default:
2800 /*
2801 * Prevent DoS attack with too small MSS. Round up
2802 * to at least minmss.
2803 */
2804 offer = max(offer, tcp_minmss);
2805 /*
2806 * Sanity check: make sure that maxopd will be large
2807 * enough to allow some data on segments even if the
2808 * all the option space is used (40bytes). Otherwise
2809 * funny things may happen in tcp_output.
2810 */
2811 offer = max(offer, 64);
2812 if (tcp_do_rfc1644)
2813 tcp_hc_updatetao(&inp->inp_inc,
2814 TCP_HC_TAO_MSSOPT, 0, offer);
2815 }
2816
2817 /*
2818 * rmx information is now retrieved from tcp_hostcache
2819 */
2820 tcp_hc_get(&inp->inp_inc, &metrics);
2821
2822 /*
2823 * if there's a discovered mtu int tcp hostcache, use it
2824 * else, use the link mtu.
2825 */
2826 if (metrics.rmx_mtu)
2827 mss = metrics.rmx_mtu - min_protoh;
2828 else {
2829 #ifdef INET6
2830 if (isipv6) {
2831 mss = maxmtu - min_protoh;
2832 if (!path_mtu_discovery &&
2833 !in6_localaddr(&inp->in6p_faddr))
2834 mss = min(mss, tcp_v6mssdflt);
2835 } else
2836 #endif
2837 {
2838 mss = maxmtu - min_protoh;
2839 if (!path_mtu_discovery &&
2840 !in_localaddr(inp->inp_faddr))
2841 mss = min(mss, tcp_mssdflt);
2842 }
2843 }
2844 mss = min(mss, offer);
2845
2846 /*
2847 * maxopd stores the maximum length of data AND options
2848 * in a segment; maxseg is the amount of data in a normal
2849 * segment. We need to store this value (maxopd) apart
2850 * from maxseg, because now every segment carries options
2851 * and thus we normally have somewhat less data in segments.
2852 */
2853 tp->t_maxopd = mss;
2854
2855 /*
2856 * In case of T/TCP, origoffer==-1 indicates, that no segments
2857 * were received yet. In this case we just guess, otherwise
2858 * we do the same as before T/TCP.
2859 */
2860 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2861 (origoffer == -1 ||
2862 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2863 mss -= TCPOLEN_TSTAMP_APPA;
2864 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2865 (origoffer == -1 ||
2866 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2867 mss -= TCPOLEN_CC_APPA;
2868 tp->t_maxseg = mss;
2869
2870 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2871 if (mss > MCLBYTES)
2872 mss &= ~(MCLBYTES-1);
2873 #else
2874 if (mss > MCLBYTES)
2875 mss = mss / MCLBYTES * MCLBYTES;
2876 #endif
2877 tp->t_maxseg = mss;
2878
2879 /*
2880 * If there's a pipesize, change the socket buffer to that size,
2881 * don't change if sb_hiwat is different than default (then it
2882 * has been changed on purpose with setsockopt).
2883 * Make the socket buffers an integral number of mss units;
2884 * if the mss is larger than the socket buffer, decrease the mss.
2885 */
2886 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2887 bufsize = metrics.rmx_sendpipe;
2888 else
2889 bufsize = so->so_snd.sb_hiwat;
2890 if (bufsize < mss)
2891 mss = bufsize;
2892 else {
2893 bufsize = roundup(bufsize, mss);
2894 if (bufsize > sb_max)
2895 bufsize = sb_max;
2896 if (bufsize > so->so_snd.sb_hiwat)
2897 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2898 }
2899 tp->t_maxseg = mss;
2900
2901 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2902 bufsize = metrics.rmx_recvpipe;
2903 else
2904 bufsize = so->so_rcv.sb_hiwat;
2905 if (bufsize > mss) {
2906 bufsize = roundup(bufsize, mss);
2907 if (bufsize > sb_max)
2908 bufsize = sb_max;
2909 if (bufsize > so->so_rcv.sb_hiwat)
2910 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2911 }
2912 /*
2913 * While we're here, check the others too
2914 */
2915 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2916 tp->t_srtt = rtt;
2917 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2918 tcpstat.tcps_usedrtt++;
2919 if (metrics.rmx_rttvar) {
2920 tp->t_rttvar = metrics.rmx_rttvar;
2921 tcpstat.tcps_usedrttvar++;
2922 } else {
2923 /* default variation is +- 1 rtt */
2924 tp->t_rttvar =
2925 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2926 }
2927 TCPT_RANGESET(tp->t_rxtcur,
2928 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2929 tp->t_rttmin, TCPTV_REXMTMAX);
2930 }
2931 if (metrics.rmx_ssthresh) {
2932 /*
2933 * There's some sort of gateway or interface
2934 * buffer limit on the path. Use this to set
2935 * the slow start threshhold, but set the
2936 * threshold to no less than 2*mss.
2937 */
2938 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
2939 tcpstat.tcps_usedssthresh++;
2940 }
2941 if (metrics.rmx_bandwidth)
2942 tp->snd_bandwidth = metrics.rmx_bandwidth;
2943
2944 /*
2945 * Set the slow-start flight size depending on whether this
2946 * is a local network or not.
2947 *
2948 * Extend this so we cache the cwnd too and retrieve it here.
2949 * Make cwnd even bigger than RFC3390 suggests but only if we
2950 * have previous experience with the remote host. Be careful
2951 * not make cwnd bigger than remote receive window or our own
2952 * send socket buffer. Maybe put some additional upper bound
2953 * on the retrieved cwnd. Should do incremental updates to
2954 * hostcache when cwnd collapses so next connection doesn't
2955 * overloads the path again.
2956 *
2957 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
2958 * We currently check only in syncache_socket for that.
2959 */
2960 #define TCP_METRICS_CWND
2961 #ifdef TCP_METRICS_CWND
2962 if (metrics.rmx_cwnd)
2963 tp->snd_cwnd = max(mss,
2964 min(metrics.rmx_cwnd / 2,
2965 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
2966 else
2967 #endif
2968 if (tcp_do_rfc3390)
2969 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2970 #ifdef INET6
2971 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2972 (!isipv6 && in_localaddr(inp->inp_faddr)))
2973 #else
2974 else if (in_localaddr(inp->inp_faddr))
2975 #endif
2976 tp->snd_cwnd = mss * ss_fltsz_local;
2977 else
2978 tp->snd_cwnd = mss * ss_fltsz;
2979 }
2980
2981 /*
2982 * Determine the MSS option to send on an outgoing SYN.
2983 */
2984 int
2985 tcp_mssopt(inc)
2986 struct in_conninfo *inc;
2987 {
2988 int mss = 0;
2989 u_long maxmtu = 0;
2990 u_long thcmtu = 0;
2991 size_t min_protoh;
2992 #ifdef INET6
2993 int isipv6 = inc->inc_isipv6 ? 1 : 0;
2994 #endif
2995
2996 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
2997
2998 #ifdef INET6
2999 if (isipv6) {
3000 mss = tcp_v6mssdflt;
3001 maxmtu = tcp_maxmtu6(inc);
3002 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3003 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3004 } else
3005 #endif
3006 {
3007 mss = tcp_mssdflt;
3008 maxmtu = tcp_maxmtu(inc);
3009 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3010 min_protoh = sizeof(struct tcpiphdr);
3011 }
3012 if (maxmtu && thcmtu)
3013 mss = min(maxmtu, thcmtu) - min_protoh;
3014 else if (maxmtu || thcmtu)
3015 mss = max(maxmtu, thcmtu) - min_protoh;
3016
3017 return (mss);
3018 }
3019
3020
3021 /*
3022 * On a partial ack arrives, force the retransmission of the
3023 * next unacknowledged segment. Do not clear tp->t_dupacks.
3024 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3025 * be started again.
3026 */
3027 static void
3028 tcp_newreno_partial_ack(tp, th)
3029 struct tcpcb *tp;
3030 struct tcphdr *th;
3031 {
3032 tcp_seq onxt = tp->snd_nxt;
3033 u_long ocwnd = tp->snd_cwnd;
3034
3035 callout_stop(tp->tt_rexmt);
3036 tp->t_rtttime = 0;
3037 tp->snd_nxt = th->th_ack;
3038 /*
3039 * Set snd_cwnd to one segment beyond acknowledged offset.
3040 * (tp->snd_una has not yet been updated when this function is called.)
3041 */
3042 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3043 tp->t_flags |= TF_ACKNOW;
3044 (void) tcp_output(tp);
3045 tp->snd_cwnd = ocwnd;
3046 if (SEQ_GT(onxt, tp->snd_nxt))
3047 tp->snd_nxt = onxt;
3048 /*
3049 * Partial window deflation. Relies on fact that tp->snd_una
3050 * not updated yet.
3051 */
3052 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3053 }
3054
3055 /*
3056 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3057 * looking for a pcb in the listen state. Returns 0 otherwise.
3058 */
3059 static int
3060 tcp_timewait(tw, to, th, m, tlen)
3061 struct tcptw *tw;
3062 struct tcpopt *to;
3063 struct tcphdr *th;
3064 struct mbuf *m;
3065 int tlen;
3066 {
3067 int thflags;
3068 tcp_seq seq;
3069 #ifdef INET6
3070 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3071 #else
3072 const int isipv6 = 0;
3073 #endif
3074
3075 thflags = th->th_flags;
3076
3077 /*
3078 * NOTE: for FIN_WAIT_2 (to be added later),
3079 * must validate sequence number before accepting RST
3080 */
3081
3082 /*
3083 * If the segment contains RST:
3084 * Drop the segment - see Stevens, vol. 2, p. 964 and
3085 * RFC 1337.
3086 */
3087 if (thflags & TH_RST)
3088 goto drop;
3089
3090 /*
3091 * If segment contains a SYN and CC [not CC.NEW] option:
3092 * if connection duration > MSL, drop packet and send RST;
3093 *
3094 * if SEG.CC > CCrecv then is new SYN.
3095 * Complete close and delete TCPCB. Then reprocess
3096 * segment, hoping to find new TCPCB in LISTEN state;
3097 *
3098 * else must be old SYN; drop it.
3099 * else do normal processing.
3100 */
3101 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) {
3102 if ((ticks - tw->t_starttime) > tcp_msl)
3103 goto reset;
3104 if (CC_GT(to->to_cc, tw->cc_recv)) {
3105 (void) tcp_twclose(tw, 0);
3106 return (1);
3107 }
3108 goto drop;
3109 }
3110
3111 #if 0
3112 /* PAWS not needed at the moment */
3113 /*
3114 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3115 * and it's less than ts_recent, drop it.
3116 */
3117 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3118 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3119 if ((thflags & TH_ACK) == 0)
3120 goto drop;
3121 goto ack;
3122 }
3123 /*
3124 * ts_recent is never updated because we never accept new segments.
3125 */
3126 #endif
3127
3128 /*
3129 * If a new connection request is received
3130 * while in TIME_WAIT, drop the old connection
3131 * and start over if the sequence numbers
3132 * are above the previous ones.
3133 */
3134 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3135 (void) tcp_twclose(tw, 0);
3136 return (1);
3137 }
3138
3139 /*
3140 * Drop the the segment if it does not contain an ACK.
3141 */
3142 if ((thflags & TH_ACK) == 0)
3143 goto drop;
3144
3145 /*
3146 * Reset the 2MSL timer if this is a duplicate FIN.
3147 */
3148 if (thflags & TH_FIN) {
3149 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3150 if (seq + 1 == tw->rcv_nxt)
3151 tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3152 }
3153
3154 /*
3155 * Acknowledge the segment if it has data or is not a duplicate ACK.
3156 */
3157 if (thflags != TH_ACK || tlen != 0 ||
3158 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3159 tcp_twrespond(tw, NULL, m, TH_ACK);
3160 goto drop;
3161
3162 reset:
3163 /*
3164 * Generate a RST, dropping incoming segment.
3165 * Make ACK acceptable to originator of segment.
3166 * Don't bother to respond if destination was broadcast/multicast.
3167 */
3168 if (m->m_flags & (M_BCAST|M_MCAST))
3169 goto drop;
3170 if (isipv6) {
3171 struct ip6_hdr *ip6;
3172
3173 /* IPv6 anycast check is done at tcp6_input() */
3174 ip6 = mtod(m, struct ip6_hdr *);
3175 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3176 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3177 goto drop;
3178 } else {
3179 struct ip *ip;
3180
3181 ip = mtod(m, struct ip *);
3182 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3183 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3184 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3185 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3186 goto drop;
3187 }
3188 if (thflags & TH_ACK) {
3189 tcp_respond(NULL,
3190 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3191 } else {
3192 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3193 tcp_respond(NULL,
3194 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3195 }
3196 INP_UNLOCK(tw->tw_inpcb);
3197 return (0);
3198
3199 drop:
3200 INP_UNLOCK(tw->tw_inpcb);
3201 m_freem(m);
3202 return (0);
3203 }
Cache object: 5513d11e845fa595b9be55ab5d95ba15
|