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 * $FreeBSD$
35 */
36
37 #include "opt_ipfw.h" /* for ipfw_fwd */
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_ipsec.h"
41 #include "opt_tcpdebug.h"
42 #include "opt_tcp_input.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/proc.h> /* for proc0 declaration */
51 #include <sys/protosw.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/syslog.h>
55
56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
57
58 #include <net/if.h>
59 #include <net/route.h>
60
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
65 #include <netinet/in_var.h>
66 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet6/tcp6_var.h>
80 #include <netinet/tcpip.h>
81 #ifdef TCPDEBUG
82 #include <netinet/tcp_debug.h>
83
84 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
85 struct tcphdr tcp_savetcp;
86 #endif /* TCPDEBUG */
87
88 #ifdef FAST_IPSEC
89 #include <netipsec/ipsec.h>
90 #include <netipsec/ipsec6.h>
91 #endif
92
93 #ifdef IPSEC
94 #include <netinet6/ipsec.h>
95 #include <netinet6/ipsec6.h>
96 #include <netkey/key.h>
97 #endif /*IPSEC*/
98
99 #include <machine/in_cksum.h>
100
101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
102
103 static const int tcprexmtthresh = 3;
104 tcp_cc tcp_ccgen;
105
106 struct tcpstat tcpstat;
107 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
109
110 static int log_in_vain = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &log_in_vain, 0, "Log all incoming TCP connections");
113
114 static int blackhole = 0;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
116 &blackhole, 0, "Do not send RST when dropping refused connections");
117
118 int tcp_delack_enabled = 1;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
120 &tcp_delack_enabled, 0,
121 "Delay ACK to try and piggyback it onto a data packet");
122
123 #ifdef TCP_DROP_SYNFIN
124 static int drop_synfin = 0;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
127 #endif
128
129 static int tcp_insecure_rst = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
131 &tcp_insecure_rst, 0,
132 "Follow the old (insecure) criteria for accepting RST packets.");
133
134 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
135 "TCP Segment Reassembly Queue");
136
137 int tcp_reass_maxseg = 0;
138 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
139 &tcp_reass_maxseg, 0,
140 "Global maximum number of TCP Segments in Reassembly Queue");
141
142 int tcp_reass_qsize = 0;
143 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
144 &tcp_reass_qsize, 0,
145 "Global number of TCP Segments currently in Reassembly Queue");
146
147 static int tcp_reass_overflows = 0;
148 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
149 &tcp_reass_overflows, 0,
150 "Global number of TCP Segment Reassembly Queue Overflows");
151
152 struct inpcbhead tcb;
153 #define tcb6 tcb /* for KAME src sync over BSD*'s */
154 struct inpcbinfo tcbinfo;
155
156 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
157 static void tcp_pulloutofband(struct socket *,
158 struct tcphdr *, struct mbuf *, int);
159 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
160 struct mbuf *);
161 static void tcp_xmit_timer(struct tcpcb *, int);
162 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
163
164 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
165 #ifdef INET6
166 #define ND6_HINT(tp) \
167 do { \
168 if ((tp) && (tp)->t_inpcb && \
169 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
170 (tp)->t_inpcb->in6p_route.ro_rt) \
171 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
172 } while (0)
173 #else
174 #define ND6_HINT(tp)
175 #endif
176
177 /*
178 * Indicate whether this ack should be delayed. We can delay the ack if
179 * - delayed acks are enabled and
180 * - there is no delayed ack timer in progress and
181 * - our last ack wasn't a 0-sized window. We never want to delay
182 * the ack that opens up a 0-sized window.
183 */
184 #define DELAY_ACK(tp) \
185 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
186 (tp->t_flags & TF_RXWIN0SENT) == 0)
187
188 static int
189 tcp_reass(tp, th, tlenp, m)
190 register struct tcpcb *tp;
191 register struct tcphdr *th;
192 int *tlenp;
193 struct mbuf *m;
194 {
195 struct tseg_qent *q;
196 struct tseg_qent *p = NULL;
197 struct tseg_qent *nq;
198 struct tseg_qent *te;
199 struct socket *so = tp->t_inpcb->inp_socket;
200 int flags;
201
202 /*
203 * Call with th==0 after become established to
204 * force pre-ESTABLISHED data up to user socket.
205 */
206 if (th == 0)
207 goto present;
208
209 /*
210 * Limit the number of segments in the reassembly queue to prevent
211 * holding on to too many segments (and thus running out of mbufs).
212 * Make sure to let the missing segment through which caused this
213 * queue. Always keep one global queue entry spare to be able to
214 * process the missing segment.
215 */
216 if (th->th_seq != tp->rcv_nxt &&
217 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
218 tcp_reass_overflows++;
219 tcpstat.tcps_rcvmemdrop++;
220 m_freem(m);
221 return (0);
222 }
223
224 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
225 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
226 M_NOWAIT);
227 if (te == NULL) {
228 tcpstat.tcps_rcvmemdrop++;
229 m_freem(m);
230 return (0);
231 }
232 tcp_reass_qsize++;
233
234 /*
235 * Find a segment which begins after this one does.
236 */
237 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
238 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
239 break;
240 p = q;
241 }
242
243 /*
244 * If there is a preceding segment, it may provide some of
245 * our data already. If so, drop the data from the incoming
246 * segment. If it provides all of our data, drop us.
247 */
248 if (p != NULL) {
249 register int i;
250 /* conversion to int (in i) handles seq wraparound */
251 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
252 if (i > 0) {
253 if (i >= *tlenp) {
254 tcpstat.tcps_rcvduppack++;
255 tcpstat.tcps_rcvdupbyte += *tlenp;
256 m_freem(m);
257 free(te, M_TSEGQ);
258 tcp_reass_qsize--;
259 /*
260 * Try to present any queued data
261 * at the left window edge to the user.
262 * This is needed after the 3-WHS
263 * completes.
264 */
265 goto present; /* ??? */
266 }
267 m_adj(m, i);
268 *tlenp -= i;
269 th->th_seq += i;
270 }
271 }
272 tcpstat.tcps_rcvoopack++;
273 tcpstat.tcps_rcvoobyte += *tlenp;
274
275 /*
276 * While we overlap succeeding segments trim them or,
277 * if they are completely covered, dequeue them.
278 */
279 while (q) {
280 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
281 if (i <= 0)
282 break;
283 if (i < q->tqe_len) {
284 q->tqe_th->th_seq += i;
285 q->tqe_len -= i;
286 m_adj(q->tqe_m, i);
287 break;
288 }
289
290 nq = LIST_NEXT(q, tqe_q);
291 LIST_REMOVE(q, tqe_q);
292 m_freem(q->tqe_m);
293 free(q, M_TSEGQ);
294 tcp_reass_qsize--;
295 q = nq;
296 }
297
298 /* Insert the new segment queue entry into place. */
299 te->tqe_m = m;
300 te->tqe_th = th;
301 te->tqe_len = *tlenp;
302
303 if (p == NULL) {
304 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
305 } else {
306 LIST_INSERT_AFTER(p, te, tqe_q);
307 }
308
309 present:
310 /*
311 * Present data to user, advancing rcv_nxt through
312 * completed sequence space.
313 */
314 if (!TCPS_HAVEESTABLISHED(tp->t_state))
315 return (0);
316 q = LIST_FIRST(&tp->t_segq);
317 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
318 return (0);
319 do {
320 tp->rcv_nxt += q->tqe_len;
321 flags = q->tqe_th->th_flags & TH_FIN;
322 nq = LIST_NEXT(q, tqe_q);
323 LIST_REMOVE(q, tqe_q);
324 if (so->so_state & SS_CANTRCVMORE)
325 m_freem(q->tqe_m);
326 else
327 sbappendstream(&so->so_rcv, q->tqe_m);
328 free(q, M_TSEGQ);
329 tcp_reass_qsize--;
330 q = nq;
331 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
332 ND6_HINT(tp);
333 sorwakeup(so);
334 return (flags);
335 }
336
337 /*
338 * TCP input routine, follows pages 65-76 of the
339 * protocol specification dated September, 1981 very closely.
340 */
341 #ifdef INET6
342 int
343 tcp6_input(mp, offp, proto)
344 struct mbuf **mp;
345 int *offp, proto;
346 {
347 register struct mbuf *m = *mp;
348 struct in6_ifaddr *ia6;
349
350 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
351
352 /*
353 * draft-itojun-ipv6-tcp-to-anycast
354 * better place to put this in?
355 */
356 ia6 = ip6_getdstifaddr(m);
357 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
358 struct ip6_hdr *ip6;
359
360 ip6 = mtod(m, struct ip6_hdr *);
361 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
362 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
363 return IPPROTO_DONE;
364 }
365
366 tcp_input(m, *offp, proto);
367 return IPPROTO_DONE;
368 }
369 #endif
370
371 void
372 tcp_input(m, off0, proto)
373 register struct mbuf *m;
374 int off0, proto;
375 {
376 register struct tcphdr *th;
377 register struct ip *ip = NULL;
378 register struct ipovly *ipov;
379 register struct inpcb *inp = NULL;
380 u_char *optp = NULL;
381 int optlen = 0;
382 int len, tlen, off;
383 int drop_hdrlen;
384 register struct tcpcb *tp = NULL;
385 register int thflags;
386 struct socket *so = 0;
387 int todrop, acked, ourfinisacked, needoutput = 0;
388 u_long tiwin;
389 struct tcpopt to; /* options in this segment */
390 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
391 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
392 struct sockaddr_in *next_hop = NULL;
393 int rstreason; /* For badport_bandlim accounting purposes */
394 struct ip6_hdr *ip6 = NULL;
395 #ifdef INET6
396 int isipv6;
397 #else
398 const int isipv6 = 0;
399 #endif
400 #ifdef TCPDEBUG
401 short ostate = 0;
402 #endif
403
404 /* Grab info from MT_TAG mbufs prepended to the chain. */
405 for (;m && m->m_type == MT_TAG; m = m->m_next) {
406 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
407 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
408 }
409 #ifdef INET6
410 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
411 #endif
412 bzero((char *)&to, sizeof(to));
413
414 tcpstat.tcps_rcvtotal++;
415
416 if (isipv6) {
417 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
418 ip6 = mtod(m, struct ip6_hdr *);
419 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
420 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
421 tcpstat.tcps_rcvbadsum++;
422 goto drop;
423 }
424 th = (struct tcphdr *)((caddr_t)ip6 + off0);
425
426 /*
427 * Be proactive about unspecified IPv6 address in source.
428 * As we use all-zero to indicate unbounded/unconnected pcb,
429 * unspecified IPv6 address can be used to confuse us.
430 *
431 * Note that packets with unspecified IPv6 destination is
432 * already dropped in ip6_input.
433 */
434 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
435 /* XXX stat */
436 goto drop;
437 }
438 } else {
439 /*
440 * Get IP and TCP header together in first mbuf.
441 * Note: IP leaves IP header in first mbuf.
442 */
443 if (off0 > sizeof(struct ip)) {
444 ip_stripoptions(m, (struct mbuf *)0);
445 off0 = sizeof(struct ip);
446 }
447 if (m->m_len < sizeof(struct tcpiphdr)) {
448 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
449 tcpstat.tcps_rcvshort++;
450 return;
451 }
452 }
453 ip = mtod(m, struct ip *);
454 ipov = (struct ipovly *)ip;
455 th = (struct tcphdr *)((caddr_t)ip + off0);
456 tlen = ip->ip_len;
457
458 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
459 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
460 th->th_sum = m->m_pkthdr.csum_data;
461 else
462 th->th_sum = in_pseudo(ip->ip_src.s_addr,
463 ip->ip_dst.s_addr,
464 htonl(m->m_pkthdr.csum_data +
465 ip->ip_len +
466 IPPROTO_TCP));
467 th->th_sum ^= 0xffff;
468 } else {
469 /*
470 * Checksum extended TCP header and data.
471 */
472 len = sizeof(struct ip) + tlen;
473 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
474 ipov->ih_len = (u_short)tlen;
475 ipov->ih_len = htons(ipov->ih_len);
476 th->th_sum = in_cksum(m, len);
477 }
478 if (th->th_sum) {
479 tcpstat.tcps_rcvbadsum++;
480 goto drop;
481 }
482 #ifdef INET6
483 /* Re-initialization for later version check */
484 ip->ip_v = IPVERSION;
485 #endif
486 }
487
488 /*
489 * Check that TCP offset makes sense,
490 * pull out TCP options and adjust length. XXX
491 */
492 off = th->th_off << 2;
493 if (off < sizeof(struct tcphdr) || off > tlen) {
494 tcpstat.tcps_rcvbadoff++;
495 goto drop;
496 }
497 tlen -= off; /* tlen is used instead of ti->ti_len */
498 if (off > sizeof(struct tcphdr)) {
499 if (isipv6) {
500 IP6_EXTHDR_CHECK(m, off0, off, );
501 ip6 = mtod(m, struct ip6_hdr *);
502 th = (struct tcphdr *)((caddr_t)ip6 + off0);
503 } else {
504 if (m->m_len < sizeof(struct ip) + off) {
505 if ((m = m_pullup(m, sizeof(struct ip) + off))
506 == 0) {
507 tcpstat.tcps_rcvshort++;
508 return;
509 }
510 ip = mtod(m, struct ip *);
511 ipov = (struct ipovly *)ip;
512 th = (struct tcphdr *)((caddr_t)ip + off0);
513 }
514 }
515 optlen = off - sizeof(struct tcphdr);
516 optp = (u_char *)(th + 1);
517 }
518 thflags = th->th_flags;
519
520 #ifdef TCP_DROP_SYNFIN
521 /*
522 * If the drop_synfin option is enabled, drop all packets with
523 * both the SYN and FIN bits set. This prevents e.g. nmap from
524 * identifying the TCP/IP stack.
525 *
526 * This is a violation of the TCP specification.
527 */
528 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
529 goto drop;
530 #endif
531
532 /*
533 * Convert TCP protocol specific fields to host format.
534 */
535 th->th_seq = ntohl(th->th_seq);
536 th->th_ack = ntohl(th->th_ack);
537 th->th_win = ntohs(th->th_win);
538 th->th_urp = ntohs(th->th_urp);
539
540 /*
541 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
542 * until after ip6_savecontrol() is called and before other functions
543 * which don't want those proto headers.
544 * Because ip6_savecontrol() is going to parse the mbuf to
545 * search for data to be passed up to user-land, it wants mbuf
546 * parameters to be unchanged.
547 * XXX: the call of ip6_savecontrol() has been obsoleted based on
548 * latest version of the advanced API (20020110).
549 */
550 drop_hdrlen = off0 + off;
551
552 /*
553 * Locate pcb for segment.
554 */
555 findpcb:
556 /* IPFIREWALL_FORWARD section */
557 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
558 /*
559 * Transparently forwarded. Pretend to be the destination.
560 * already got one like this?
561 */
562 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
563 ip->ip_dst, th->th_dport,
564 0, m->m_pkthdr.rcvif);
565 if (!inp) {
566 /* It's new. Try find the ambushing socket. */
567 inp = in_pcblookup_hash(&tcbinfo,
568 ip->ip_src, th->th_sport,
569 next_hop->sin_addr,
570 next_hop->sin_port ?
571 ntohs(next_hop->sin_port) :
572 th->th_dport,
573 1, m->m_pkthdr.rcvif);
574 }
575 } else {
576 if (isipv6)
577 inp = in6_pcblookup_hash(&tcbinfo,
578 &ip6->ip6_src, th->th_sport,
579 &ip6->ip6_dst, th->th_dport,
580 1, m->m_pkthdr.rcvif);
581 else
582 inp = in_pcblookup_hash(&tcbinfo,
583 ip->ip_src, th->th_sport,
584 ip->ip_dst, th->th_dport,
585 1, m->m_pkthdr.rcvif);
586 }
587
588 #ifdef IPSEC
589 if (isipv6) {
590 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
591 ipsec6stat.in_polvio++;
592 goto drop;
593 }
594 } else {
595 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
596 ipsecstat.in_polvio++;
597 goto drop;
598 }
599 }
600 #endif
601 #ifdef FAST_IPSEC
602 if (isipv6) {
603 if (inp != NULL && ipsec6_in_reject(m, inp)) {
604 goto drop;
605 }
606 } else {
607 if (inp != NULL && ipsec4_in_reject(m, inp)) {
608 goto drop;
609 }
610 }
611 #endif
612
613 /*
614 * If the state is CLOSED (i.e., TCB does not exist) then
615 * all data in the incoming segment is discarded.
616 * If the TCB exists but is in CLOSED state, it is embryonic,
617 * but should either do a listen or a connect soon.
618 */
619 if (inp == NULL) {
620 if (log_in_vain) {
621 #ifdef INET6
622 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
623 #else
624 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
625 #endif
626 if (isipv6) {
627 strcpy(dbuf, "[");
628 strcpy(sbuf, "[");
629 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
630 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
631 strcat(dbuf, "]");
632 strcat(sbuf, "]");
633 } else {
634 strcpy(dbuf, inet_ntoa(ip->ip_dst));
635 strcpy(sbuf, inet_ntoa(ip->ip_src));
636 }
637 switch (log_in_vain) {
638 case 1:
639 if ((thflags & TH_SYN) == 0)
640 break;
641 case 2:
642 log(LOG_INFO,
643 "Connection attempt to TCP %s:%d "
644 "from %s:%d flags:0x%02x\n",
645 dbuf, ntohs(th->th_dport), sbuf,
646 ntohs(th->th_sport), thflags);
647 break;
648 default:
649 break;
650 }
651 }
652 if (blackhole) {
653 switch (blackhole) {
654 case 1:
655 if (thflags & TH_SYN)
656 goto drop;
657 break;
658 case 2:
659 goto drop;
660 default:
661 goto drop;
662 }
663 }
664 rstreason = BANDLIM_RST_CLOSEDPORT;
665 goto dropwithreset;
666 }
667 tp = intotcpcb(inp);
668 if (tp == NULL) {
669 rstreason = BANDLIM_RST_CLOSEDPORT;
670 goto dropwithreset;
671 }
672 if (tp->t_state == TCPS_CLOSED)
673 goto drop;
674
675 /* Unscale the window into a 32-bit value. */
676 if ((thflags & TH_SYN) == 0)
677 tiwin = th->th_win << tp->snd_scale;
678 else
679 tiwin = th->th_win;
680
681 so = inp->inp_socket;
682 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
683 struct in_conninfo inc;
684 #ifdef TCPDEBUG
685 if (so->so_options & SO_DEBUG) {
686 ostate = tp->t_state;
687 if (isipv6)
688 bcopy((char *)ip6, (char *)tcp_saveipgen,
689 sizeof(*ip6));
690 else
691 bcopy((char *)ip, (char *)tcp_saveipgen,
692 sizeof(*ip));
693 tcp_savetcp = *th;
694 }
695 #endif
696 /* skip if this isn't a listen socket */
697 if ((so->so_options & SO_ACCEPTCONN) == 0)
698 goto after_listen;
699 #ifdef INET6
700 inc.inc_isipv6 = isipv6;
701 #endif
702 if (isipv6) {
703 inc.inc6_faddr = ip6->ip6_src;
704 inc.inc6_laddr = ip6->ip6_dst;
705 inc.inc6_route.ro_rt = NULL; /* XXX */
706 } else {
707 inc.inc_faddr = ip->ip_src;
708 inc.inc_laddr = ip->ip_dst;
709 inc.inc_route.ro_rt = NULL; /* XXX */
710 }
711 inc.inc_fport = th->th_sport;
712 inc.inc_lport = th->th_dport;
713
714 /*
715 * If the state is LISTEN then ignore segment if it contains
716 * a RST. If the segment contains an ACK then it is bad and
717 * send a RST. If it does not contain a SYN then it is not
718 * interesting; drop it.
719 *
720 * If the state is SYN_RECEIVED (syncache) and seg contains
721 * an ACK, but not for our SYN/ACK, send a RST. If the seg
722 * contains a RST, check the sequence number to see if it
723 * is a valid reset segment.
724 */
725 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
726 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
727 if (!syncache_expand(&inc, th, &so, m)) {
728 /*
729 * No syncache entry, or ACK was not
730 * for our SYN/ACK. Send a RST.
731 */
732 tcpstat.tcps_badsyn++;
733 rstreason = BANDLIM_RST_OPENPORT;
734 goto dropwithreset;
735 }
736 if (so == NULL)
737 /*
738 * Could not complete 3-way handshake,
739 * connection is being closed down, and
740 * syncache will free mbuf.
741 */
742 return;
743 /*
744 * Socket is created in state SYN_RECEIVED.
745 * Continue processing segment.
746 */
747 inp = sotoinpcb(so);
748 tp = intotcpcb(inp);
749 /*
750 * This is what would have happened in
751 * tcp_output() when the SYN,ACK was sent.
752 */
753 tp->snd_up = tp->snd_una;
754 tp->snd_max = tp->snd_nxt = tp->iss + 1;
755 tp->last_ack_sent = tp->rcv_nxt;
756 /*
757 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
758 * until the _second_ ACK is received:
759 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
760 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
761 * move to ESTAB, set snd_wnd to tiwin.
762 */
763 tp->snd_wnd = tiwin; /* unscaled */
764 goto after_listen;
765 }
766 if (thflags & TH_RST) {
767 syncache_chkrst(&inc, th);
768 goto drop;
769 }
770 if (thflags & TH_ACK) {
771 syncache_badack(&inc);
772 tcpstat.tcps_badsyn++;
773 rstreason = BANDLIM_RST_OPENPORT;
774 goto dropwithreset;
775 }
776 goto drop;
777 }
778
779 /*
780 * Segment's flags are (SYN) or (SYN|FIN).
781 */
782 #ifdef INET6
783 /*
784 * If deprecated address is forbidden,
785 * we do not accept SYN to deprecated interface
786 * address to prevent any new inbound connection from
787 * getting established.
788 * When we do not accept SYN, we send a TCP RST,
789 * with deprecated source address (instead of dropping
790 * it). We compromise it as it is much better for peer
791 * to send a RST, and RST will be the final packet
792 * for the exchange.
793 *
794 * If we do not forbid deprecated addresses, we accept
795 * the SYN packet. RFC2462 does not suggest dropping
796 * SYN in this case.
797 * If we decipher RFC2462 5.5.4, it says like this:
798 * 1. use of deprecated addr with existing
799 * communication is okay - "SHOULD continue to be
800 * used"
801 * 2. use of it with new communication:
802 * (2a) "SHOULD NOT be used if alternate address
803 * with sufficient scope is available"
804 * (2b) nothing mentioned otherwise.
805 * Here we fall into (2b) case as we have no choice in
806 * our source address selection - we must obey the peer.
807 *
808 * The wording in RFC2462 is confusing, and there are
809 * multiple description text for deprecated address
810 * handling - worse, they are not exactly the same.
811 * I believe 5.5.4 is the best one, so we follow 5.5.4.
812 */
813 if (isipv6 && !ip6_use_deprecated) {
814 struct in6_ifaddr *ia6;
815
816 if ((ia6 = ip6_getdstifaddr(m)) &&
817 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
818 tp = NULL;
819 rstreason = BANDLIM_RST_OPENPORT;
820 goto dropwithreset;
821 }
822 }
823 #endif
824 /*
825 * If it is from this socket, drop it, it must be forged.
826 * Don't bother responding if the destination was a broadcast.
827 */
828 if (th->th_dport == th->th_sport) {
829 if (isipv6) {
830 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
831 &ip6->ip6_src))
832 goto drop;
833 } else {
834 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
835 goto drop;
836 }
837 }
838 /*
839 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
840 *
841 * Note that it is quite possible to receive unicast
842 * link-layer packets with a broadcast IP address. Use
843 * in_broadcast() to find them.
844 */
845 if (m->m_flags & (M_BCAST|M_MCAST))
846 goto drop;
847 if (isipv6) {
848 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
849 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
850 goto drop;
851 } else {
852 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
853 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
854 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
855 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
856 goto drop;
857 }
858 /*
859 * SYN appears to be valid; create compressed TCP state
860 * for syncache, or perform t/tcp connection.
861 */
862 if (so->so_qlen <= so->so_qlimit) {
863 tcp_dooptions(&to, optp, optlen, 1);
864 if (!syncache_add(&inc, &to, th, &so, m))
865 goto drop;
866 if (so == NULL)
867 /*
868 * Entry added to syncache, mbuf used to
869 * send SYN,ACK packet.
870 */
871 return;
872 /*
873 * Segment passed TAO tests.
874 */
875 inp = sotoinpcb(so);
876 tp = intotcpcb(inp);
877 tp->snd_wnd = tiwin;
878 tp->t_starttime = ticks;
879 tp->t_state = TCPS_ESTABLISHED;
880
881 /*
882 * If there is a FIN, or if there is data and the
883 * connection is local, then delay SYN,ACK(SYN) in
884 * the hope of piggy-backing it on a response
885 * segment. Otherwise must send ACK now in case
886 * the other side is slow starting.
887 */
888 if (DELAY_ACK(tp) &&
889 ((thflags & TH_FIN) ||
890 (tlen != 0 &&
891 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
892 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
893 callout_reset(tp->tt_delack, tcp_delacktime,
894 tcp_timer_delack, tp);
895 tp->t_flags |= TF_NEEDSYN;
896 } else
897 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
898
899 tcpstat.tcps_connects++;
900 soisconnected(so);
901 goto trimthenstep6;
902 }
903 goto drop;
904 }
905 after_listen:
906
907 /* XXX temp debugging */
908 /* should not happen - syncache should pick up these connections */
909 if (tp->t_state == TCPS_LISTEN)
910 panic("tcp_input: TCPS_LISTEN");
911
912 /*
913 * Segment received on connection.
914 * Reset idle time and keep-alive timer.
915 */
916 tp->t_rcvtime = ticks;
917 if (TCPS_HAVEESTABLISHED(tp->t_state))
918 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
919
920 /*
921 * Process options.
922 * XXX this is tradtitional behavior, may need to be cleaned up.
923 */
924 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
925 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
926 if (to.to_flags & TOF_SCALE) {
927 tp->t_flags |= TF_RCVD_SCALE;
928 tp->requested_s_scale = to.to_requested_s_scale;
929 }
930 if (to.to_flags & TOF_TS) {
931 tp->t_flags |= TF_RCVD_TSTMP;
932 tp->ts_recent = to.to_tsval;
933 tp->ts_recent_age = ticks;
934 }
935 if (to.to_flags & (TOF_CC|TOF_CCNEW))
936 tp->t_flags |= TF_RCVD_CC;
937 if (to.to_flags & TOF_MSS)
938 tcp_mss(tp, to.to_mss);
939 }
940
941 /*
942 * Header prediction: check for the two common cases
943 * of a uni-directional data xfer. If the packet has
944 * no control flags, is in-sequence, the window didn't
945 * change and we're not retransmitting, it's a
946 * candidate. If the length is zero and the ack moved
947 * forward, we're the sender side of the xfer. Just
948 * free the data acked & wake any higher level process
949 * that was blocked waiting for space. If the length
950 * is non-zero and the ack didn't move, we're the
951 * receiver side. If we're getting packets in-order
952 * (the reassembly queue is empty), add the data to
953 * the socket buffer and note that we need a delayed ack.
954 * Make sure that the hidden state-flags are also off.
955 * Since we check for TCPS_ESTABLISHED above, it can only
956 * be TH_NEEDSYN.
957 */
958 if (tp->t_state == TCPS_ESTABLISHED &&
959 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
960 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
961 ((to.to_flags & TOF_TS) == 0 ||
962 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
963 /*
964 * Using the CC option is compulsory if once started:
965 * the segment is OK if no T/TCP was negotiated or
966 * if the segment has a CC option equal to CCrecv
967 */
968 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
969 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
970 th->th_seq == tp->rcv_nxt &&
971 tiwin && tiwin == tp->snd_wnd &&
972 tp->snd_nxt == tp->snd_max) {
973
974 /*
975 * If last ACK falls within this segment's sequence numbers,
976 * record the timestamp.
977 * NOTE that the test is modified according to the latest
978 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
979 */
980 if ((to.to_flags & TOF_TS) != 0 &&
981 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
982 tp->ts_recent_age = ticks;
983 tp->ts_recent = to.to_tsval;
984 }
985
986 if (tlen == 0) {
987 if (SEQ_GT(th->th_ack, tp->snd_una) &&
988 SEQ_LEQ(th->th_ack, tp->snd_max) &&
989 tp->snd_cwnd >= tp->snd_wnd &&
990 ((!tcp_do_newreno &&
991 tp->t_dupacks < tcprexmtthresh) ||
992 (tcp_do_newreno &&
993 !SEQ_LT(tp->snd_una, tp->snd_recover)))) {
994 /*
995 * this is a pure ack for outstanding data.
996 */
997 ++tcpstat.tcps_predack;
998 /*
999 * "bad retransmit" recovery
1000 */
1001 if (tp->t_rxtshift == 1 &&
1002 ticks < tp->t_badrxtwin) {
1003 tp->snd_cwnd = tp->snd_cwnd_prev;
1004 tp->snd_ssthresh =
1005 tp->snd_ssthresh_prev;
1006 tp->snd_high = tp->snd_high_prev;
1007 tp->snd_nxt = tp->snd_max;
1008 tp->t_badrxtwin = 0;
1009 }
1010 /*
1011 * Recalculate the retransmit timer / rtt.
1012 *
1013 * Some machines (certain windows boxes)
1014 * send broken timestamp replies during the
1015 * SYN+ACK phase, ignore timestamps of 0.
1016 */
1017 if ((to.to_flags & TOF_TS) != 0 &&
1018 to.to_tsecr) {
1019 tcp_xmit_timer(tp,
1020 ticks - to.to_tsecr + 1);
1021 } else if (tp->t_rtttime &&
1022 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1023 tcp_xmit_timer(tp,
1024 ticks - tp->t_rtttime);
1025 }
1026 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1027 acked = th->th_ack - tp->snd_una;
1028 tcpstat.tcps_rcvackpack++;
1029 tcpstat.tcps_rcvackbyte += acked;
1030 sbdrop(&so->so_snd, acked);
1031 if (SEQ_GT(tp->snd_una, tp->snd_high) &&
1032 SEQ_LEQ(th->th_ack, tp->snd_high))
1033 tp->snd_high = th->th_ack - 1;
1034 tp->snd_una = tp->snd_recover = th->th_ack;
1035 tp->t_dupacks = 0;
1036 m_freem(m);
1037 ND6_HINT(tp); /* some progress has been done */
1038
1039 /*
1040 * If all outstanding data are acked, stop
1041 * retransmit timer, otherwise restart timer
1042 * using current (possibly backed-off) value.
1043 * If process is waiting for space,
1044 * wakeup/selwakeup/signal. If data
1045 * are ready to send, let tcp_output
1046 * decide between more output or persist.
1047 */
1048 if (tp->snd_una == tp->snd_max)
1049 callout_stop(tp->tt_rexmt);
1050 else if (!callout_active(tp->tt_persist))
1051 callout_reset(tp->tt_rexmt,
1052 tp->t_rxtcur,
1053 tcp_timer_rexmt, tp);
1054
1055 sowwakeup(so);
1056 if (so->so_snd.sb_cc)
1057 (void) tcp_output(tp);
1058 return;
1059 }
1060 } else if (th->th_ack == tp->snd_una &&
1061 LIST_EMPTY(&tp->t_segq) &&
1062 tlen <= sbspace(&so->so_rcv)) {
1063 /*
1064 * this is a pure, in-sequence data packet
1065 * with nothing on the reassembly queue and
1066 * we have enough buffer space to take it.
1067 */
1068 ++tcpstat.tcps_preddat;
1069 tp->rcv_nxt += tlen;
1070 tcpstat.tcps_rcvpack++;
1071 tcpstat.tcps_rcvbyte += tlen;
1072 ND6_HINT(tp); /* some progress has been done */
1073 /*
1074 * Add data to socket buffer.
1075 */
1076 if (so->so_state & SS_CANTRCVMORE) {
1077 m_freem(m);
1078 } else {
1079 m_adj(m, drop_hdrlen); /* delayed header drop */
1080 sbappendstream(&so->so_rcv, m);
1081 }
1082 sorwakeup(so);
1083 if (DELAY_ACK(tp)) {
1084 callout_reset(tp->tt_delack, tcp_delacktime,
1085 tcp_timer_delack, tp);
1086 } else {
1087 tp->t_flags |= TF_ACKNOW;
1088 tcp_output(tp);
1089 }
1090 return;
1091 }
1092 }
1093
1094 /*
1095 * Calculate amount of space in receive window,
1096 * and then do TCP input processing.
1097 * Receive window is amount of space in rcv queue,
1098 * but not less than advertised window.
1099 */
1100 { int win;
1101
1102 win = sbspace(&so->so_rcv);
1103 if (win < 0)
1104 win = 0;
1105 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1106 }
1107
1108 switch (tp->t_state) {
1109
1110 /*
1111 * If the state is SYN_RECEIVED:
1112 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1113 */
1114 case TCPS_SYN_RECEIVED:
1115 if ((thflags & TH_ACK) &&
1116 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1117 SEQ_GT(th->th_ack, tp->snd_max))) {
1118 rstreason = BANDLIM_RST_OPENPORT;
1119 goto dropwithreset;
1120 }
1121 break;
1122
1123 /*
1124 * If the state is SYN_SENT:
1125 * if seg contains an ACK, but not for our SYN, drop the input.
1126 * if seg contains a RST, then drop the connection.
1127 * if seg does not contain SYN, then drop it.
1128 * Otherwise this is an acceptable SYN segment
1129 * initialize tp->rcv_nxt and tp->irs
1130 * if seg contains ack then advance tp->snd_una
1131 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1132 * arrange for segment to be acked (eventually)
1133 * continue processing rest of data/controls, beginning with URG
1134 */
1135 case TCPS_SYN_SENT:
1136 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1137 taop = &tao_noncached;
1138 bzero(taop, sizeof(*taop));
1139 }
1140
1141 if ((thflags & TH_ACK) &&
1142 (SEQ_LEQ(th->th_ack, tp->iss) ||
1143 SEQ_GT(th->th_ack, tp->snd_max))) {
1144 /*
1145 * If we have a cached CCsent for the remote host,
1146 * hence we haven't just crashed and restarted,
1147 * do not send a RST. This may be a retransmission
1148 * from the other side after our earlier ACK was lost.
1149 * Our new SYN, when it arrives, will serve as the
1150 * needed ACK.
1151 */
1152 if (taop->tao_ccsent != 0)
1153 goto drop;
1154 else {
1155 rstreason = BANDLIM_UNLIMITED;
1156 goto dropwithreset;
1157 }
1158 }
1159 if (thflags & TH_RST) {
1160 if (thflags & TH_ACK)
1161 tp = tcp_drop(tp, ECONNREFUSED);
1162 goto drop;
1163 }
1164 if ((thflags & TH_SYN) == 0)
1165 goto drop;
1166 tp->snd_wnd = th->th_win; /* initial send window */
1167 tp->cc_recv = to.to_cc; /* foreign CC */
1168
1169 tp->irs = th->th_seq;
1170 tcp_rcvseqinit(tp);
1171 if (thflags & TH_ACK) {
1172 /*
1173 * Our SYN was acked. If segment contains CC.ECHO
1174 * option, check it to make sure this segment really
1175 * matches our SYN. If not, just drop it as old
1176 * duplicate, but send an RST if we're still playing
1177 * by the old rules. If no CC.ECHO option, make sure
1178 * we don't get fooled into using T/TCP.
1179 */
1180 if (to.to_flags & TOF_CCECHO) {
1181 if (tp->cc_send != to.to_ccecho) {
1182 if (taop->tao_ccsent != 0)
1183 goto drop;
1184 else {
1185 rstreason = BANDLIM_UNLIMITED;
1186 goto dropwithreset;
1187 }
1188 }
1189 } else
1190 tp->t_flags &= ~TF_RCVD_CC;
1191 tcpstat.tcps_connects++;
1192 soisconnected(so);
1193 /* Do window scaling on this connection? */
1194 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1195 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1196 tp->snd_scale = tp->requested_s_scale;
1197 tp->rcv_scale = tp->request_r_scale;
1198 }
1199 /* Segment is acceptable, update cache if undefined. */
1200 if (taop->tao_ccsent == 0)
1201 taop->tao_ccsent = to.to_ccecho;
1202
1203 tp->rcv_adv += tp->rcv_wnd;
1204 tp->snd_una++; /* SYN is acked */
1205 /*
1206 * If there's data, delay ACK; if there's also a FIN
1207 * ACKNOW will be turned on later.
1208 */
1209 if (DELAY_ACK(tp) && tlen != 0)
1210 callout_reset(tp->tt_delack, tcp_delacktime,
1211 tcp_timer_delack, tp);
1212 else
1213 tp->t_flags |= TF_ACKNOW;
1214 /*
1215 * Received <SYN,ACK> in SYN_SENT[*] state.
1216 * Transitions:
1217 * SYN_SENT --> ESTABLISHED
1218 * SYN_SENT* --> FIN_WAIT_1
1219 */
1220 tp->t_starttime = ticks;
1221 if (tp->t_flags & TF_NEEDFIN) {
1222 tp->t_state = TCPS_FIN_WAIT_1;
1223 tp->t_flags &= ~TF_NEEDFIN;
1224 thflags &= ~TH_SYN;
1225 } else {
1226 tp->t_state = TCPS_ESTABLISHED;
1227 callout_reset(tp->tt_keep, tcp_keepidle,
1228 tcp_timer_keep, tp);
1229 }
1230 } else {
1231 /*
1232 * Received initial SYN in SYN-SENT[*] state =>
1233 * simultaneous open. If segment contains CC option
1234 * and there is a cached CC, apply TAO test.
1235 * If it succeeds, connection is * half-synchronized.
1236 * Otherwise, do 3-way handshake:
1237 * SYN-SENT -> SYN-RECEIVED
1238 * SYN-SENT* -> SYN-RECEIVED*
1239 * If there was no CC option, clear cached CC value.
1240 */
1241 tp->t_flags |= TF_ACKNOW;
1242 callout_stop(tp->tt_rexmt);
1243 if (to.to_flags & TOF_CC) {
1244 if (taop->tao_cc != 0 &&
1245 CC_GT(to.to_cc, taop->tao_cc)) {
1246 /*
1247 * update cache and make transition:
1248 * SYN-SENT -> ESTABLISHED*
1249 * SYN-SENT* -> FIN-WAIT-1*
1250 */
1251 taop->tao_cc = to.to_cc;
1252 tp->t_starttime = ticks;
1253 if (tp->t_flags & TF_NEEDFIN) {
1254 tp->t_state = TCPS_FIN_WAIT_1;
1255 tp->t_flags &= ~TF_NEEDFIN;
1256 } else {
1257 tp->t_state = TCPS_ESTABLISHED;
1258 callout_reset(tp->tt_keep,
1259 tcp_keepidle,
1260 tcp_timer_keep,
1261 tp);
1262 }
1263 tp->t_flags |= TF_NEEDSYN;
1264 } else
1265 tp->t_state = TCPS_SYN_RECEIVED;
1266 } else {
1267 /* CC.NEW or no option => invalidate cache */
1268 taop->tao_cc = 0;
1269 tp->t_state = TCPS_SYN_RECEIVED;
1270 }
1271 }
1272
1273 trimthenstep6:
1274 /*
1275 * Advance th->th_seq to correspond to first data byte.
1276 * If data, trim to stay within window,
1277 * dropping FIN if necessary.
1278 */
1279 th->th_seq++;
1280 if (tlen > tp->rcv_wnd) {
1281 todrop = tlen - tp->rcv_wnd;
1282 m_adj(m, -todrop);
1283 tlen = tp->rcv_wnd;
1284 thflags &= ~TH_FIN;
1285 tcpstat.tcps_rcvpackafterwin++;
1286 tcpstat.tcps_rcvbyteafterwin += todrop;
1287 }
1288 tp->snd_wl1 = th->th_seq - 1;
1289 tp->rcv_up = th->th_seq;
1290 /*
1291 * Client side of transaction: already sent SYN and data.
1292 * If the remote host used T/TCP to validate the SYN,
1293 * our data will be ACK'd; if so, enter normal data segment
1294 * processing in the middle of step 5, ack processing.
1295 * Otherwise, goto step 6.
1296 */
1297 if (thflags & TH_ACK)
1298 goto process_ACK;
1299
1300 goto step6;
1301
1302 /*
1303 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1304 * if segment contains a SYN and CC [not CC.NEW] option:
1305 * if state == TIME_WAIT and connection duration > MSL,
1306 * drop packet and send RST;
1307 *
1308 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1309 * ack the FIN (and data) in retransmission queue.
1310 * Complete close and delete TCPCB. Then reprocess
1311 * segment, hoping to find new TCPCB in LISTEN state;
1312 *
1313 * else must be old SYN; drop it.
1314 * else do normal processing.
1315 */
1316 case TCPS_LAST_ACK:
1317 case TCPS_CLOSING:
1318 case TCPS_TIME_WAIT:
1319 if ((thflags & TH_SYN) &&
1320 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1321 if (tp->t_state == TCPS_TIME_WAIT &&
1322 (ticks - tp->t_starttime) > tcp_msl) {
1323 rstreason = BANDLIM_UNLIMITED;
1324 goto dropwithreset;
1325 }
1326 if (CC_GT(to.to_cc, tp->cc_recv)) {
1327 tp = tcp_close(tp);
1328 goto findpcb;
1329 }
1330 else
1331 goto drop;
1332 }
1333 break; /* continue normal processing */
1334 }
1335
1336 /*
1337 * States other than LISTEN or SYN_SENT.
1338 * First check the RST flag and sequence number since reset segments
1339 * are exempt from the timestamp and connection count tests. This
1340 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1341 * below which allowed reset segments in half the sequence space
1342 * to fall though and be processed (which gives forged reset
1343 * segments with a random sequence number a 50 percent chance of
1344 * killing a connection).
1345 * Then check timestamp, if present.
1346 * Then check the connection count, if present.
1347 * Then check that at least some bytes of segment are within
1348 * receive window. If segment begins before rcv_nxt,
1349 * drop leading data (and SYN); if nothing left, just ack.
1350 *
1351 *
1352 * If the RST bit is set, check the sequence number to see
1353 * if this is a valid reset segment.
1354 * RFC 793 page 37:
1355 * In all states except SYN-SENT, all reset (RST) segments
1356 * are validated by checking their SEQ-fields. A reset is
1357 * valid if its sequence number is in the window.
1358 * Note: this does not take into account delayed ACKs, so
1359 * we should test against last_ack_sent instead of rcv_nxt.
1360 * The sequence number in the reset segment is normally an
1361 * echo of our outgoing acknowlegement numbers, but some hosts
1362 * send a reset with the sequence number at the rightmost edge
1363 * of our receive window, and we have to handle this case.
1364 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1365 * that brute force RST attacks are possible. To combat this,
1366 * we use a much stricter check while in the ESTABLISHED state,
1367 * only accepting RSTs where the sequence number is equal to
1368 * last_ack_sent. In all other states (the states in which a
1369 * RST is more likely), the more permissive check is used.
1370 * If we have multiple segments in flight, the intial reset
1371 * segment sequence numbers will be to the left of last_ack_sent,
1372 * but they will eventually catch up.
1373 * In any case, it never made sense to trim reset segments to
1374 * fit the receive window since RFC 1122 says:
1375 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1376 *
1377 * A TCP SHOULD allow a received RST segment to include data.
1378 *
1379 * DISCUSSION
1380 * It has been suggested that a RST segment could contain
1381 * ASCII text that encoded and explained the cause of the
1382 * RST. No standard has yet been established for such
1383 * data.
1384 *
1385 * If the reset segment passes the sequence number test examine
1386 * the state:
1387 * SYN_RECEIVED STATE:
1388 * If passive open, return to LISTEN state.
1389 * If active open, inform user that connection was refused.
1390 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1391 * Inform user that connection was reset, and close tcb.
1392 * CLOSING, LAST_ACK STATES:
1393 * Close the tcb.
1394 * TIME_WAIT STATE:
1395 * Drop the segment - see Stevens, vol. 2, p. 964 and
1396 * RFC 1337.
1397 */
1398 if (thflags & TH_RST) {
1399 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1400 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1401 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1402 switch (tp->t_state) {
1403
1404 case TCPS_SYN_RECEIVED:
1405 so->so_error = ECONNREFUSED;
1406 goto close;
1407
1408 case TCPS_ESTABLISHED:
1409 if (tp->last_ack_sent != th->th_seq &&
1410 tcp_insecure_rst == 0) {
1411 /* tcpstat.tcps_badrst++; */
1412 goto drop;
1413 }
1414 case TCPS_FIN_WAIT_1:
1415 case TCPS_FIN_WAIT_2:
1416 case TCPS_CLOSE_WAIT:
1417 so->so_error = ECONNRESET;
1418 close:
1419 tp->t_state = TCPS_CLOSED;
1420 tcpstat.tcps_drops++;
1421 tp = tcp_close(tp);
1422 break;
1423
1424 case TCPS_CLOSING:
1425 case TCPS_LAST_ACK:
1426 tp = tcp_close(tp);
1427 break;
1428
1429 case TCPS_TIME_WAIT:
1430 break;
1431 }
1432 }
1433 goto drop;
1434 }
1435
1436 /*
1437 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1438 * and it's less than ts_recent, drop it.
1439 */
1440 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1441 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1442
1443 /* Check to see if ts_recent is over 24 days old. */
1444 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1445 /*
1446 * Invalidate ts_recent. If this segment updates
1447 * ts_recent, the age will be reset later and ts_recent
1448 * will get a valid value. If it does not, setting
1449 * ts_recent to zero will at least satisfy the
1450 * requirement that zero be placed in the timestamp
1451 * echo reply when ts_recent isn't valid. The
1452 * age isn't reset until we get a valid ts_recent
1453 * because we don't want out-of-order segments to be
1454 * dropped when ts_recent is old.
1455 */
1456 tp->ts_recent = 0;
1457 } else {
1458 tcpstat.tcps_rcvduppack++;
1459 tcpstat.tcps_rcvdupbyte += tlen;
1460 tcpstat.tcps_pawsdrop++;
1461 if (tlen)
1462 goto dropafterack;
1463 goto drop;
1464 }
1465 }
1466
1467 /*
1468 * T/TCP mechanism
1469 * If T/TCP was negotiated and the segment doesn't have CC,
1470 * or if its CC is wrong then drop the segment.
1471 * RST segments do not have to comply with this.
1472 */
1473 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1474 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1475 goto dropafterack;
1476
1477 /*
1478 * In the SYN-RECEIVED state, validate that the packet belongs to
1479 * this connection before trimming the data to fit the receive
1480 * window. Check the sequence number versus IRS since we know
1481 * the sequence numbers haven't wrapped. This is a partial fix
1482 * for the "LAND" DoS attack.
1483 */
1484 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1485 rstreason = BANDLIM_RST_OPENPORT;
1486 goto dropwithreset;
1487 }
1488
1489 todrop = tp->rcv_nxt - th->th_seq;
1490 if (todrop > 0) {
1491 if (thflags & TH_SYN) {
1492 thflags &= ~TH_SYN;
1493 th->th_seq++;
1494 if (th->th_urp > 1)
1495 th->th_urp--;
1496 else
1497 thflags &= ~TH_URG;
1498 todrop--;
1499 }
1500 /*
1501 * Following if statement from Stevens, vol. 2, p. 960.
1502 */
1503 if (todrop > tlen
1504 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1505 /*
1506 * Any valid FIN must be to the left of the window.
1507 * At this point the FIN must be a duplicate or out
1508 * of sequence; drop it.
1509 */
1510 thflags &= ~TH_FIN;
1511
1512 /*
1513 * Send an ACK to resynchronize and drop any data.
1514 * But keep on processing for RST or ACK.
1515 */
1516 tp->t_flags |= TF_ACKNOW;
1517 todrop = tlen;
1518 tcpstat.tcps_rcvduppack++;
1519 tcpstat.tcps_rcvdupbyte += todrop;
1520 } else {
1521 tcpstat.tcps_rcvpartduppack++;
1522 tcpstat.tcps_rcvpartdupbyte += todrop;
1523 }
1524 drop_hdrlen += todrop; /* drop from the top afterwards */
1525 th->th_seq += todrop;
1526 tlen -= todrop;
1527 if (th->th_urp > todrop)
1528 th->th_urp -= todrop;
1529 else {
1530 thflags &= ~TH_URG;
1531 th->th_urp = 0;
1532 }
1533 }
1534
1535 /*
1536 * If new data are received on a connection after the
1537 * user processes are gone, then RST the other end.
1538 */
1539 if ((so->so_state & SS_NOFDREF) &&
1540 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1541 tp = tcp_close(tp);
1542 tcpstat.tcps_rcvafterclose++;
1543 rstreason = BANDLIM_UNLIMITED;
1544 goto dropwithreset;
1545 }
1546
1547 /*
1548 * If segment ends after window, drop trailing data
1549 * (and PUSH and FIN); if nothing left, just ACK.
1550 */
1551 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1552 if (todrop > 0) {
1553 tcpstat.tcps_rcvpackafterwin++;
1554 if (todrop >= tlen) {
1555 tcpstat.tcps_rcvbyteafterwin += tlen;
1556 /*
1557 * If a new connection request is received
1558 * while in TIME_WAIT, drop the old connection
1559 * and start over if the sequence numbers
1560 * are above the previous ones.
1561 */
1562 if (thflags & TH_SYN &&
1563 tp->t_state == TCPS_TIME_WAIT &&
1564 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1565 tp = tcp_close(tp);
1566 goto findpcb;
1567 }
1568 /*
1569 * If window is closed can only take segments at
1570 * window edge, and have to drop data and PUSH from
1571 * incoming segments. Continue processing, but
1572 * remember to ack. Otherwise, drop segment
1573 * and ack.
1574 */
1575 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1576 tp->t_flags |= TF_ACKNOW;
1577 tcpstat.tcps_rcvwinprobe++;
1578 } else
1579 goto dropafterack;
1580 } else
1581 tcpstat.tcps_rcvbyteafterwin += todrop;
1582 m_adj(m, -todrop);
1583 tlen -= todrop;
1584 thflags &= ~(TH_PUSH|TH_FIN);
1585 }
1586
1587 /*
1588 * If last ACK falls within this segment's sequence numbers,
1589 * record its timestamp.
1590 * NOTE:
1591 * 1) That the test incorporates suggestions from the latest
1592 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1593 * 2) That updating only on newer timestamps interferes with
1594 * our earlier PAWS tests, so this check should be solely
1595 * predicated on the sequence space of this segment.
1596 * 3) That we modify the segment boundary check to be
1597 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1598 * instead of RFC1323's
1599 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1600 * This modified check allows us to overcome RFC1323's
1601 * limitations as described in Stevens TCP/IP Illustrated
1602 * Vol. 2 p.869. In such cases, we can still calculate the
1603 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1604 */
1605 if ((to.to_flags & TOF_TS) != 0 &&
1606 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1607 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1608 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1609 tp->ts_recent_age = ticks;
1610 tp->ts_recent = to.to_tsval;
1611 }
1612
1613 /*
1614 * If a SYN is in the window, then this is an
1615 * error and we send an RST and drop the connection.
1616 */
1617 if (thflags & TH_SYN) {
1618 tp = tcp_drop(tp, ECONNRESET);
1619 rstreason = BANDLIM_UNLIMITED;
1620 goto dropwithreset;
1621 }
1622
1623 /*
1624 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1625 * flag is on (half-synchronized state), then queue data for
1626 * later processing; else drop segment and return.
1627 */
1628 if ((thflags & TH_ACK) == 0) {
1629 if (tp->t_state == TCPS_SYN_RECEIVED ||
1630 (tp->t_flags & TF_NEEDSYN))
1631 goto step6;
1632 else if (tp->t_flags & TF_ACKNOW)
1633 goto dropafterack;
1634 else
1635 goto drop;
1636 }
1637
1638 /*
1639 * Ack processing.
1640 */
1641 switch (tp->t_state) {
1642
1643 /*
1644 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1645 * ESTABLISHED state and continue processing.
1646 * The ACK was checked above.
1647 */
1648 case TCPS_SYN_RECEIVED:
1649
1650 tcpstat.tcps_connects++;
1651 soisconnected(so);
1652 /* Do window scaling? */
1653 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1654 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1655 tp->snd_scale = tp->requested_s_scale;
1656 tp->rcv_scale = tp->request_r_scale;
1657 }
1658 /*
1659 * Upon successful completion of 3-way handshake,
1660 * update cache.CC if it was undefined, pass any queued
1661 * data to the user, and advance state appropriately.
1662 */
1663 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1664 taop->tao_cc == 0)
1665 taop->tao_cc = tp->cc_recv;
1666
1667 /*
1668 * Make transitions:
1669 * SYN-RECEIVED -> ESTABLISHED
1670 * SYN-RECEIVED* -> FIN-WAIT-1
1671 */
1672 tp->t_starttime = ticks;
1673 if (tp->t_flags & TF_NEEDFIN) {
1674 tp->t_state = TCPS_FIN_WAIT_1;
1675 tp->t_flags &= ~TF_NEEDFIN;
1676 } else {
1677 tp->t_state = TCPS_ESTABLISHED;
1678 callout_reset(tp->tt_keep, tcp_keepidle,
1679 tcp_timer_keep, tp);
1680 }
1681 /*
1682 * If segment contains data or ACK, will call tcp_reass()
1683 * later; if not, do so now to pass queued data to user.
1684 */
1685 if (tlen == 0 && (thflags & TH_FIN) == 0)
1686 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1687 (struct mbuf *)0);
1688 tp->snd_wl1 = th->th_seq - 1;
1689 /* fall into ... */
1690
1691 /*
1692 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1693 * ACKs. If the ack is in the range
1694 * tp->snd_una < th->th_ack <= tp->snd_max
1695 * then advance tp->snd_una to th->th_ack and drop
1696 * data from the retransmission queue. If this ACK reflects
1697 * more up to date window information we update our window information.
1698 */
1699 case TCPS_ESTABLISHED:
1700 case TCPS_FIN_WAIT_1:
1701 case TCPS_FIN_WAIT_2:
1702 case TCPS_CLOSE_WAIT:
1703 case TCPS_CLOSING:
1704 case TCPS_LAST_ACK:
1705 case TCPS_TIME_WAIT:
1706
1707 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1708 if (tlen == 0 && tiwin == tp->snd_wnd) {
1709 tcpstat.tcps_rcvdupack++;
1710 /*
1711 * If we have outstanding data (other than
1712 * a window probe), this is a completely
1713 * duplicate ack (ie, window info didn't
1714 * change), the ack is the biggest we've
1715 * seen and we've seen exactly our rexmt
1716 * threshhold of them, assume a packet
1717 * has been dropped and retransmit it.
1718 * Kludge snd_nxt & the congestion
1719 * window so we send only this one
1720 * packet.
1721 *
1722 * We know we're losing at the current
1723 * window size so do congestion avoidance
1724 * (set ssthresh to half the current window
1725 * and pull our congestion window back to
1726 * the new ssthresh).
1727 *
1728 * Dup acks mean that packets have left the
1729 * network (they're now cached at the receiver)
1730 * so bump cwnd by the amount in the receiver
1731 * to keep a constant cwnd packets in the
1732 * network.
1733 */
1734 if (!callout_active(tp->tt_rexmt) ||
1735 th->th_ack != tp->snd_una)
1736 tp->t_dupacks = 0;
1737 else if (++tp->t_dupacks > tcprexmtthresh ||
1738 (tcp_do_newreno &&
1739 SEQ_LT(tp->snd_una,
1740 tp->snd_recover))) {
1741 tp->snd_cwnd += tp->t_maxseg;
1742 (void) tcp_output(tp);
1743 goto drop;
1744 } else if (tp->t_dupacks == tcprexmtthresh) {
1745 tcp_seq onxt = tp->snd_nxt;
1746 u_int win;
1747 if (tcp_do_newreno &&
1748 SEQ_LEQ(th->th_ack, tp->snd_high)) {
1749 tp->t_dupacks = 0;
1750 break;
1751 }
1752 win = min(tp->snd_wnd, tp->snd_cwnd) /
1753 2 / tp->t_maxseg;
1754 if (win < 2)
1755 win = 2;
1756 tp->snd_ssthresh = win * tp->t_maxseg;
1757 tp->snd_recover = tp->snd_max;
1758 callout_stop(tp->tt_rexmt);
1759 tp->t_rtttime = 0;
1760 tp->snd_nxt = th->th_ack;
1761 tp->snd_cwnd = tp->t_maxseg;
1762 (void) tcp_output(tp);
1763 tp->snd_cwnd = tp->snd_ssthresh +
1764 tp->t_maxseg * tp->t_dupacks;
1765 if (SEQ_GT(onxt, tp->snd_nxt))
1766 tp->snd_nxt = onxt;
1767 goto drop;
1768 }
1769 } else
1770 tp->t_dupacks = 0;
1771 break;
1772 }
1773
1774 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1775
1776 /*
1777 * If the congestion window was inflated to account
1778 * for the other side's cached packets, retract it.
1779 */
1780 if (tcp_do_newreno) {
1781 if (SEQ_LT(tp->snd_una, tp->snd_recover)) {
1782 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1783 tcp_newreno_partial_ack(tp, th);
1784 } else {
1785 /*
1786 * Window inflation should have left us
1787 * with approximately snd_ssthresh
1788 * outstanding data.
1789 * But in case we would be inclined to
1790 * send a burst, better to do it via
1791 * the slow start mechanism.
1792 */
1793 if (SEQ_GT(th->th_ack +
1794 tp->snd_ssthresh,
1795 tp->snd_max))
1796 tp->snd_cwnd = tp->snd_max -
1797 th->th_ack +
1798 tp->t_maxseg;
1799 else
1800 tp->snd_cwnd = tp->snd_ssthresh;
1801 }
1802 }
1803 } else {
1804 if (tp->t_dupacks >= tcprexmtthresh &&
1805 tp->snd_cwnd > tp->snd_ssthresh)
1806 tp->snd_cwnd = tp->snd_ssthresh;
1807 }
1808 tp->t_dupacks = 0;
1809 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1810 tcpstat.tcps_rcvacktoomuch++;
1811 goto dropafterack;
1812 }
1813 /*
1814 * If we reach this point, ACK is not a duplicate,
1815 * i.e., it ACKs something we sent.
1816 */
1817 if (tp->t_flags & TF_NEEDSYN) {
1818 /*
1819 * T/TCP: Connection was half-synchronized, and our
1820 * SYN has been ACK'd (so connection is now fully
1821 * synchronized). Go to non-starred state,
1822 * increment snd_una for ACK of SYN, and check if
1823 * we can do window scaling.
1824 */
1825 tp->t_flags &= ~TF_NEEDSYN;
1826 tp->snd_una++;
1827 /* Do window scaling? */
1828 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1829 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1830 tp->snd_scale = tp->requested_s_scale;
1831 tp->rcv_scale = tp->request_r_scale;
1832 }
1833 }
1834
1835 process_ACK:
1836 acked = th->th_ack - tp->snd_una;
1837 tcpstat.tcps_rcvackpack++;
1838 tcpstat.tcps_rcvackbyte += acked;
1839
1840 /*
1841 * If we just performed our first retransmit, and the ACK
1842 * arrives within our recovery window, then it was a mistake
1843 * to do the retransmit in the first place. Recover our
1844 * original cwnd and ssthresh, and proceed to transmit where
1845 * we left off.
1846 */
1847 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1848 tp->snd_cwnd = tp->snd_cwnd_prev;
1849 tp->snd_ssthresh = tp->snd_ssthresh_prev;
1850 tp->snd_high = tp->snd_high_prev;
1851 tp->snd_nxt = tp->snd_max;
1852 tp->t_badrxtwin = 0; /* XXX probably not required */
1853 }
1854
1855 /*
1856 * If we have a timestamp reply, update smoothed
1857 * round trip time. If no timestamp is present but
1858 * transmit timer is running and timed sequence
1859 * number was acked, update smoothed round trip time.
1860 * Since we now have an rtt measurement, cancel the
1861 * timer backoff (cf., Phil Karn's retransmit alg.).
1862 * Recompute the initial retransmit timer.
1863 *
1864 * Some machines (certain windows boxes) send broken
1865 * timestamp replies during the SYN+ACK phase, ignore
1866 * timestamps of 0.
1867 */
1868 if ((to.to_flags & TOF_TS) != 0 &&
1869 to.to_tsecr) {
1870 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1871 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1872 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1873 }
1874 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1875
1876 /*
1877 * If all outstanding data is acked, stop retransmit
1878 * timer and remember to restart (more output or persist).
1879 * If there is more data to be acked, restart retransmit
1880 * timer, using current (possibly backed-off) value.
1881 */
1882 if (th->th_ack == tp->snd_max) {
1883 callout_stop(tp->tt_rexmt);
1884 needoutput = 1;
1885 } else if (!callout_active(tp->tt_persist))
1886 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1887 tcp_timer_rexmt, tp);
1888
1889 /*
1890 * If no data (only SYN) was ACK'd,
1891 * skip rest of ACK processing.
1892 */
1893 if (acked == 0)
1894 goto step6;
1895
1896 /*
1897 * When new data is acked, open the congestion window.
1898 * If the window gives us less than ssthresh packets
1899 * in flight, open exponentially (maxseg per packet).
1900 * Otherwise open linearly: maxseg per window
1901 * (maxseg^2 / cwnd per packet).
1902 */
1903 if (!tcp_do_newreno || SEQ_GEQ(tp->snd_una, tp->snd_recover)) {
1904 register u_int cw = tp->snd_cwnd;
1905 register u_int incr = tp->t_maxseg;
1906 if (cw > tp->snd_ssthresh)
1907 incr = incr * incr / cw;
1908 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1909 }
1910 if (acked > so->so_snd.sb_cc) {
1911 tp->snd_wnd -= so->so_snd.sb_cc;
1912 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1913 ourfinisacked = 1;
1914 } else {
1915 sbdrop(&so->so_snd, acked);
1916 tp->snd_wnd -= acked;
1917 ourfinisacked = 0;
1918 }
1919 sowwakeup(so);
1920 /* detect una wraparound */
1921 if (SEQ_GEQ(tp->snd_una, tp->snd_recover) &&
1922 SEQ_LT(th->th_ack, tp->snd_recover))
1923 tp->snd_recover = th->th_ack;
1924 if (SEQ_GT(tp->snd_una, tp->snd_high) &&
1925 SEQ_LEQ(th->th_ack, tp->snd_high))
1926 tp->snd_high = th->th_ack - 1;
1927 tp->snd_una = th->th_ack;
1928 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1929 tp->snd_nxt = tp->snd_una;
1930
1931 switch (tp->t_state) {
1932
1933 /*
1934 * In FIN_WAIT_1 STATE in addition to the processing
1935 * for the ESTABLISHED state if our FIN is now acknowledged
1936 * then enter FIN_WAIT_2.
1937 */
1938 case TCPS_FIN_WAIT_1:
1939 if (ourfinisacked) {
1940 /*
1941 * If we can't receive any more
1942 * data, then closing user can proceed.
1943 * Starting the timer is contrary to the
1944 * specification, but if we don't get a FIN
1945 * we'll hang forever.
1946 */
1947 if (so->so_state & SS_CANTRCVMORE) {
1948 soisdisconnected(so);
1949 callout_reset(tp->tt_2msl, tcp_maxidle,
1950 tcp_timer_2msl, tp);
1951 }
1952 tp->t_state = TCPS_FIN_WAIT_2;
1953 }
1954 break;
1955
1956 /*
1957 * In CLOSING STATE in addition to the processing for
1958 * the ESTABLISHED state if the ACK acknowledges our FIN
1959 * then enter the TIME-WAIT state, otherwise ignore
1960 * the segment.
1961 */
1962 case TCPS_CLOSING:
1963 if (ourfinisacked) {
1964 tp->t_state = TCPS_TIME_WAIT;
1965 tcp_canceltimers(tp);
1966 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1967 if (tp->cc_recv != 0 &&
1968 (ticks - tp->t_starttime) < tcp_msl)
1969 callout_reset(tp->tt_2msl,
1970 tp->t_rxtcur *
1971 TCPTV_TWTRUNC,
1972 tcp_timer_2msl, tp);
1973 else
1974 callout_reset(tp->tt_2msl, 2 * tcp_msl,
1975 tcp_timer_2msl, tp);
1976 soisdisconnected(so);
1977 }
1978 break;
1979
1980 /*
1981 * In LAST_ACK, we may still be waiting for data to drain
1982 * and/or to be acked, as well as for the ack of our FIN.
1983 * If our FIN is now acknowledged, delete the TCB,
1984 * enter the closed state and return.
1985 */
1986 case TCPS_LAST_ACK:
1987 if (ourfinisacked) {
1988 tp = tcp_close(tp);
1989 goto drop;
1990 }
1991 break;
1992
1993 /*
1994 * In TIME_WAIT state the only thing that should arrive
1995 * is a retransmission of the remote FIN. Acknowledge
1996 * it and restart the finack timer.
1997 */
1998 case TCPS_TIME_WAIT:
1999 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2000 tcp_timer_2msl, tp);
2001 goto dropafterack;
2002 }
2003 }
2004
2005 step6:
2006 /*
2007 * Update window information.
2008 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2009 */
2010 if ((thflags & TH_ACK) &&
2011 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2012 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2013 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2014 /* keep track of pure window updates */
2015 if (tlen == 0 &&
2016 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2017 tcpstat.tcps_rcvwinupd++;
2018 tp->snd_wnd = tiwin;
2019 tp->snd_wl1 = th->th_seq;
2020 tp->snd_wl2 = th->th_ack;
2021 if (tp->snd_wnd > tp->max_sndwnd)
2022 tp->max_sndwnd = tp->snd_wnd;
2023 needoutput = 1;
2024 }
2025
2026 /*
2027 * Process segments with URG.
2028 */
2029 if ((thflags & TH_URG) && th->th_urp &&
2030 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2031 /*
2032 * This is a kludge, but if we receive and accept
2033 * random urgent pointers, we'll crash in
2034 * soreceive. It's hard to imagine someone
2035 * actually wanting to send this much urgent data.
2036 */
2037 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2038 th->th_urp = 0; /* XXX */
2039 thflags &= ~TH_URG; /* XXX */
2040 goto dodata; /* XXX */
2041 }
2042 /*
2043 * If this segment advances the known urgent pointer,
2044 * then mark the data stream. This should not happen
2045 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2046 * a FIN has been received from the remote side.
2047 * In these states we ignore the URG.
2048 *
2049 * According to RFC961 (Assigned Protocols),
2050 * the urgent pointer points to the last octet
2051 * of urgent data. We continue, however,
2052 * to consider it to indicate the first octet
2053 * of data past the urgent section as the original
2054 * spec states (in one of two places).
2055 */
2056 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2057 tp->rcv_up = th->th_seq + th->th_urp;
2058 so->so_oobmark = so->so_rcv.sb_cc +
2059 (tp->rcv_up - tp->rcv_nxt) - 1;
2060 if (so->so_oobmark == 0)
2061 so->so_state |= SS_RCVATMARK;
2062 sohasoutofband(so);
2063 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2064 }
2065 /*
2066 * Remove out of band data so doesn't get presented to user.
2067 * This can happen independent of advancing the URG pointer,
2068 * but if two URG's are pending at once, some out-of-band
2069 * data may creep in... ick.
2070 */
2071 if (th->th_urp <= (u_long)tlen
2072 #ifdef SO_OOBINLINE
2073 && (so->so_options & SO_OOBINLINE) == 0
2074 #endif
2075 )
2076 tcp_pulloutofband(so, th, m,
2077 drop_hdrlen); /* hdr drop is delayed */
2078 } else {
2079 /*
2080 * If no out of band data is expected,
2081 * pull receive urgent pointer along
2082 * with the receive window.
2083 */
2084 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2085 tp->rcv_up = tp->rcv_nxt;
2086 }
2087 dodata: /* XXX */
2088
2089 /*
2090 * Process the segment text, merging it into the TCP sequencing queue,
2091 * and arranging for acknowledgment of receipt if necessary.
2092 * This process logically involves adjusting tp->rcv_wnd as data
2093 * is presented to the user (this happens in tcp_usrreq.c,
2094 * case PRU_RCVD). If a FIN has already been received on this
2095 * connection then we just ignore the text.
2096 */
2097 if ((tlen || (thflags & TH_FIN)) &&
2098 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2099 m_adj(m, drop_hdrlen); /* delayed header drop */
2100 /*
2101 * Insert segment which includes th into TCP reassembly queue
2102 * with control block tp. Set thflags to whether reassembly now
2103 * includes a segment with FIN. This handles the common case
2104 * inline (segment is the next to be received on an established
2105 * connection, and the queue is empty), avoiding linkage into
2106 * and removal from the queue and repetition of various
2107 * conversions.
2108 * Set DELACK for segments received in order, but ack
2109 * immediately when segments are out of order (so
2110 * fast retransmit can work).
2111 */
2112 if (th->th_seq == tp->rcv_nxt &&
2113 LIST_EMPTY(&tp->t_segq) &&
2114 TCPS_HAVEESTABLISHED(tp->t_state)) {
2115 if (DELAY_ACK(tp))
2116 callout_reset(tp->tt_delack, tcp_delacktime,
2117 tcp_timer_delack, tp);
2118 else
2119 tp->t_flags |= TF_ACKNOW;
2120 tp->rcv_nxt += tlen;
2121 thflags = th->th_flags & TH_FIN;
2122 tcpstat.tcps_rcvpack++;
2123 tcpstat.tcps_rcvbyte += tlen;
2124 ND6_HINT(tp);
2125 if (so->so_state & SS_CANTRCVMORE)
2126 m_freem(m);
2127 else
2128 sbappendstream(&so->so_rcv, m);
2129 sorwakeup(so);
2130 } else {
2131 thflags = tcp_reass(tp, th, &tlen, m);
2132 tp->t_flags |= TF_ACKNOW;
2133 }
2134
2135 /*
2136 * Note the amount of data that peer has sent into
2137 * our window, in order to estimate the sender's
2138 * buffer size.
2139 */
2140 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2141 } else {
2142 m_freem(m);
2143 thflags &= ~TH_FIN;
2144 }
2145
2146 /*
2147 * If FIN is received ACK the FIN and let the user know
2148 * that the connection is closing.
2149 */
2150 if (thflags & TH_FIN) {
2151 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2152 socantrcvmore(so);
2153 /*
2154 * If connection is half-synchronized
2155 * (ie NEEDSYN flag on) then delay ACK,
2156 * so it may be piggybacked when SYN is sent.
2157 * Otherwise, since we received a FIN then no
2158 * more input can be expected, send ACK now.
2159 */
2160 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2161 callout_reset(tp->tt_delack, tcp_delacktime,
2162 tcp_timer_delack, tp);
2163 else
2164 tp->t_flags |= TF_ACKNOW;
2165 tp->rcv_nxt++;
2166 }
2167 switch (tp->t_state) {
2168
2169 /*
2170 * In SYN_RECEIVED and ESTABLISHED STATES
2171 * enter the CLOSE_WAIT state.
2172 */
2173 case TCPS_SYN_RECEIVED:
2174 tp->t_starttime = ticks;
2175 /*FALLTHROUGH*/
2176 case TCPS_ESTABLISHED:
2177 tp->t_state = TCPS_CLOSE_WAIT;
2178 break;
2179
2180 /*
2181 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2182 * enter the CLOSING state.
2183 */
2184 case TCPS_FIN_WAIT_1:
2185 tp->t_state = TCPS_CLOSING;
2186 break;
2187
2188 /*
2189 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2190 * starting the time-wait timer, turning off the other
2191 * standard timers.
2192 */
2193 case TCPS_FIN_WAIT_2:
2194 tp->t_state = TCPS_TIME_WAIT;
2195 tcp_canceltimers(tp);
2196 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2197 if (tp->cc_recv != 0 &&
2198 (ticks - tp->t_starttime) < tcp_msl) {
2199 callout_reset(tp->tt_2msl,
2200 tp->t_rxtcur * TCPTV_TWTRUNC,
2201 tcp_timer_2msl, tp);
2202 /* For transaction client, force ACK now. */
2203 tp->t_flags |= TF_ACKNOW;
2204 }
2205 else
2206 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2207 tcp_timer_2msl, tp);
2208 soisdisconnected(so);
2209 break;
2210
2211 /*
2212 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2213 */
2214 case TCPS_TIME_WAIT:
2215 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2216 tcp_timer_2msl, tp);
2217 break;
2218 }
2219 }
2220 #ifdef TCPDEBUG
2221 if (so->so_options & SO_DEBUG)
2222 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2223 &tcp_savetcp, 0);
2224 #endif
2225
2226 /*
2227 * Return any desired output.
2228 */
2229 if (needoutput || (tp->t_flags & TF_ACKNOW))
2230 (void) tcp_output(tp);
2231 return;
2232
2233 dropafterack:
2234 /*
2235 * Generate an ACK dropping incoming segment if it occupies
2236 * sequence space, where the ACK reflects our state.
2237 *
2238 * We can now skip the test for the RST flag since all
2239 * paths to this code happen after packets containing
2240 * RST have been dropped.
2241 *
2242 * In the SYN-RECEIVED state, don't send an ACK unless the
2243 * segment we received passes the SYN-RECEIVED ACK test.
2244 * If it fails send a RST. This breaks the loop in the
2245 * "LAND" DoS attack, and also prevents an ACK storm
2246 * between two listening ports that have been sent forged
2247 * SYN segments, each with the source address of the other.
2248 */
2249 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2250 (SEQ_GT(tp->snd_una, th->th_ack) ||
2251 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2252 rstreason = BANDLIM_RST_OPENPORT;
2253 goto dropwithreset;
2254 }
2255 #ifdef TCPDEBUG
2256 if (so->so_options & SO_DEBUG)
2257 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2258 &tcp_savetcp, 0);
2259 #endif
2260 m_freem(m);
2261 tp->t_flags |= TF_ACKNOW;
2262 (void) tcp_output(tp);
2263 return;
2264
2265 dropwithreset:
2266 /*
2267 * Generate a RST, dropping incoming segment.
2268 * Make ACK acceptable to originator of segment.
2269 * Don't bother to respond if destination was broadcast/multicast.
2270 */
2271 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2272 goto drop;
2273 if (isipv6) {
2274 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2275 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2276 goto drop;
2277 } else {
2278 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2279 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2280 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2281 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2282 goto drop;
2283 }
2284 /* IPv6 anycast check is done at tcp6_input() */
2285
2286 /*
2287 * Perform bandwidth limiting.
2288 */
2289 #ifdef ICMP_BANDLIM
2290 if (badport_bandlim(rstreason) < 0)
2291 goto drop;
2292 #endif
2293
2294 #ifdef TCPDEBUG
2295 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2296 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2297 &tcp_savetcp, 0);
2298 #endif
2299 if (thflags & TH_ACK)
2300 /* mtod() below is safe as long as hdr dropping is delayed */
2301 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2302 TH_RST);
2303 else {
2304 if (thflags & TH_SYN)
2305 tlen++;
2306 /* mtod() below is safe as long as hdr dropping is delayed */
2307 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2308 (tcp_seq)0, TH_RST|TH_ACK);
2309 }
2310 return;
2311
2312 drop:
2313 /*
2314 * Drop space held by incoming segment and return.
2315 */
2316 #ifdef TCPDEBUG
2317 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2318 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2319 &tcp_savetcp, 0);
2320 #endif
2321 m_freem(m);
2322 return;
2323 }
2324
2325 /*
2326 * Parse TCP options and place in tcpopt.
2327 */
2328 static void
2329 tcp_dooptions(to, cp, cnt, is_syn)
2330 struct tcpopt *to;
2331 u_char *cp;
2332 int cnt;
2333 {
2334 int opt, optlen;
2335
2336 to->to_flags = 0;
2337 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2338 opt = cp[0];
2339 if (opt == TCPOPT_EOL)
2340 break;
2341 if (opt == TCPOPT_NOP)
2342 optlen = 1;
2343 else {
2344 if (cnt < 2)
2345 break;
2346 optlen = cp[1];
2347 if (optlen < 2 || optlen > cnt)
2348 break;
2349 }
2350 switch (opt) {
2351 case TCPOPT_MAXSEG:
2352 if (optlen != TCPOLEN_MAXSEG)
2353 continue;
2354 if (!is_syn)
2355 continue;
2356 to->to_flags |= TOF_MSS;
2357 bcopy((char *)cp + 2,
2358 (char *)&to->to_mss, sizeof(to->to_mss));
2359 to->to_mss = ntohs(to->to_mss);
2360 break;
2361 case TCPOPT_WINDOW:
2362 if (optlen != TCPOLEN_WINDOW)
2363 continue;
2364 if (! is_syn)
2365 continue;
2366 to->to_flags |= TOF_SCALE;
2367 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2368 break;
2369 case TCPOPT_TIMESTAMP:
2370 if (optlen != TCPOLEN_TIMESTAMP)
2371 continue;
2372 to->to_flags |= TOF_TS;
2373 bcopy((char *)cp + 2,
2374 (char *)&to->to_tsval, sizeof(to->to_tsval));
2375 to->to_tsval = ntohl(to->to_tsval);
2376 bcopy((char *)cp + 6,
2377 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2378 to->to_tsecr = ntohl(to->to_tsecr);
2379 /*
2380 * If echoed timestamp is later than the current time,
2381 * fall back to non RFC1323 RTT calculation.
2382 */
2383 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2384 to->to_tsecr = 0;
2385 break;
2386 case TCPOPT_CC:
2387 if (optlen != TCPOLEN_CC)
2388 continue;
2389 to->to_flags |= TOF_CC;
2390 bcopy((char *)cp + 2,
2391 (char *)&to->to_cc, sizeof(to->to_cc));
2392 to->to_cc = ntohl(to->to_cc);
2393 break;
2394 case TCPOPT_CCNEW:
2395 if (optlen != TCPOLEN_CC)
2396 continue;
2397 if (!is_syn)
2398 continue;
2399 to->to_flags |= TOF_CCNEW;
2400 bcopy((char *)cp + 2,
2401 (char *)&to->to_cc, sizeof(to->to_cc));
2402 to->to_cc = ntohl(to->to_cc);
2403 break;
2404 case TCPOPT_CCECHO:
2405 if (optlen != TCPOLEN_CC)
2406 continue;
2407 if (!is_syn)
2408 continue;
2409 to->to_flags |= TOF_CCECHO;
2410 bcopy((char *)cp + 2,
2411 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2412 to->to_ccecho = ntohl(to->to_ccecho);
2413 break;
2414 #ifdef TCP_SIGNATURE
2415 /*
2416 * XXX In order to reply to a host which has set the
2417 * TCP_SIGNATURE option in its initial SYN, we have to
2418 * record the fact that the option was observed here
2419 * for the syncache code to perform the correct response.
2420 */
2421 case TCPOPT_SIGNATURE:
2422 if (optlen != TCPOLEN_SIGNATURE)
2423 continue;
2424 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2425 break;
2426 #endif
2427 default:
2428 continue;
2429 }
2430 }
2431 }
2432
2433 /*
2434 * Pull out of band byte out of a segment so
2435 * it doesn't appear in the user's data queue.
2436 * It is still reflected in the segment length for
2437 * sequencing purposes.
2438 */
2439 static void
2440 tcp_pulloutofband(so, th, m, off)
2441 struct socket *so;
2442 struct tcphdr *th;
2443 register struct mbuf *m;
2444 int off; /* delayed to be droped hdrlen */
2445 {
2446 int cnt = off + th->th_urp - 1;
2447
2448 while (cnt >= 0) {
2449 if (m->m_len > cnt) {
2450 char *cp = mtod(m, caddr_t) + cnt;
2451 struct tcpcb *tp = sototcpcb(so);
2452
2453 tp->t_iobc = *cp;
2454 tp->t_oobflags |= TCPOOB_HAVEDATA;
2455 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2456 m->m_len--;
2457 if (m->m_flags & M_PKTHDR)
2458 m->m_pkthdr.len--;
2459 return;
2460 }
2461 cnt -= m->m_len;
2462 m = m->m_next;
2463 if (m == 0)
2464 break;
2465 }
2466 panic("tcp_pulloutofband");
2467 }
2468
2469 /*
2470 * Collect new round-trip time estimate
2471 * and update averages and current timeout.
2472 */
2473 static void
2474 tcp_xmit_timer(tp, rtt)
2475 register struct tcpcb *tp;
2476 int rtt;
2477 {
2478 register int delta;
2479
2480 tcpstat.tcps_rttupdated++;
2481 tp->t_rttupdated++;
2482 if (tp->t_srtt != 0) {
2483 /*
2484 * srtt is stored as fixed point with 5 bits after the
2485 * binary point (i.e., scaled by 8). The following magic
2486 * is equivalent to the smoothing algorithm in rfc793 with
2487 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2488 * point). Adjust rtt to origin 0.
2489 */
2490 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2491 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2492
2493 if ((tp->t_srtt += delta) <= 0)
2494 tp->t_srtt = 1;
2495
2496 /*
2497 * We accumulate a smoothed rtt variance (actually, a
2498 * smoothed mean difference), then set the retransmit
2499 * timer to smoothed rtt + 4 times the smoothed variance.
2500 * rttvar is stored as fixed point with 4 bits after the
2501 * binary point (scaled by 16). The following is
2502 * equivalent to rfc793 smoothing with an alpha of .75
2503 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2504 * rfc793's wired-in beta.
2505 */
2506 if (delta < 0)
2507 delta = -delta;
2508 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2509 if ((tp->t_rttvar += delta) <= 0)
2510 tp->t_rttvar = 1;
2511 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2512 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2513 } else {
2514 /*
2515 * No rtt measurement yet - use the unsmoothed rtt.
2516 * Set the variance to half the rtt (so our first
2517 * retransmit happens at 3*rtt).
2518 */
2519 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2520 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2521 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2522 }
2523 tp->t_rtttime = 0;
2524 tp->t_rxtshift = 0;
2525
2526 /*
2527 * the retransmit should happen at rtt + 4 * rttvar.
2528 * Because of the way we do the smoothing, srtt and rttvar
2529 * will each average +1/2 tick of bias. When we compute
2530 * the retransmit timer, we want 1/2 tick of rounding and
2531 * 1 extra tick because of +-1/2 tick uncertainty in the
2532 * firing of the timer. The bias will give us exactly the
2533 * 1.5 tick we need. But, because the bias is
2534 * statistical, we have to test that we don't drop below
2535 * the minimum feasible timer (which is 2 ticks).
2536 */
2537 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2538 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2539
2540 /*
2541 * We received an ack for a packet that wasn't retransmitted;
2542 * it is probably safe to discard any error indications we've
2543 * received recently. This isn't quite right, but close enough
2544 * for now (a route might have failed after we sent a segment,
2545 * and the return path might not be symmetrical).
2546 */
2547 tp->t_softerror = 0;
2548 }
2549
2550 /*
2551 * Determine a reasonable value for maxseg size.
2552 * If the route is known, check route for mtu.
2553 * If none, use an mss that can be handled on the outgoing
2554 * interface without forcing IP to fragment; if bigger than
2555 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2556 * to utilize large mbufs. If no route is found, route has no mtu,
2557 * or the destination isn't local, use a default, hopefully conservative
2558 * size (usually 512 or the default IP max size, but no more than the mtu
2559 * of the interface), as we can't discover anything about intervening
2560 * gateways or networks. We also initialize the congestion/slow start
2561 * window to be a single segment if the destination isn't local.
2562 * While looking at the routing entry, we also initialize other path-dependent
2563 * parameters from pre-set or cached values in the routing entry.
2564 *
2565 * Also take into account the space needed for options that we
2566 * send regularly. Make maxseg shorter by that amount to assure
2567 * that we can send maxseg amount of data even when the options
2568 * are present. Store the upper limit of the length of options plus
2569 * data in maxopd.
2570 *
2571 * NOTE that this routine is only called when we process an incoming
2572 * segment, for outgoing segments only tcp_mssopt is called.
2573 *
2574 * In case of T/TCP, we call this routine during implicit connection
2575 * setup as well (offer = -1), to initialize maxseg from the cached
2576 * MSS of our peer.
2577 */
2578 void
2579 tcp_mss(tp, offer)
2580 struct tcpcb *tp;
2581 int offer;
2582 {
2583 register struct rtentry *rt;
2584 struct ifnet *ifp;
2585 register int rtt, mss;
2586 u_long bufsize;
2587 struct inpcb *inp = tp->t_inpcb;
2588 struct socket *so;
2589 struct rmxp_tao *taop;
2590 int origoffer = offer;
2591 #ifdef INET6
2592 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2593 size_t min_protoh = isipv6 ?
2594 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2595 sizeof(struct tcpiphdr);
2596 #else
2597 const int isipv6 = 0;
2598 const size_t min_protoh = sizeof(struct tcpiphdr);
2599 #endif
2600
2601 if (isipv6)
2602 rt = tcp_rtlookup6(&inp->inp_inc);
2603 else
2604 rt = tcp_rtlookup(&inp->inp_inc);
2605 if (rt == NULL) {
2606 tp->t_maxopd = tp->t_maxseg =
2607 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2608 return;
2609 }
2610 ifp = rt->rt_ifp;
2611 so = inp->inp_socket;
2612
2613 taop = rmx_taop(rt->rt_rmx);
2614 /*
2615 * Offer == -1 means that we didn't receive SYN yet,
2616 * use cached value in that case;
2617 */
2618 if (offer == -1)
2619 offer = taop->tao_mssopt;
2620 /*
2621 * Offer == 0 means that there was no MSS on the SYN segment,
2622 * in this case we use tcp_mssdflt.
2623 */
2624 if (offer == 0)
2625 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2626 else
2627 /*
2628 * Sanity check: make sure that maxopd will be large
2629 * enough to allow some data on segments even is the
2630 * all the option space is used (40bytes). Otherwise
2631 * funny things may happen in tcp_output.
2632 */
2633 offer = max(offer, 64);
2634 taop->tao_mssopt = offer;
2635
2636 /*
2637 * While we're here, check if there's an initial rtt
2638 * or rttvar. Convert from the route-table units
2639 * to scaled multiples of the slow timeout timer.
2640 */
2641 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2642 /*
2643 * XXX the lock bit for RTT indicates that the value
2644 * is also a minimum value; this is subject to time.
2645 */
2646 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2647 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2648 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2649 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2650 tcpstat.tcps_usedrtt++;
2651 if (rt->rt_rmx.rmx_rttvar) {
2652 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2653 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2654 tcpstat.tcps_usedrttvar++;
2655 } else {
2656 /* default variation is +- 1 rtt */
2657 tp->t_rttvar =
2658 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2659 }
2660 TCPT_RANGESET(tp->t_rxtcur,
2661 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2662 tp->t_rttmin, TCPTV_REXMTMAX);
2663 }
2664 /*
2665 * if there's an mtu associated with the route, use it
2666 * else, use the link mtu.
2667 */
2668 if (rt->rt_rmx.rmx_mtu)
2669 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2670 else {
2671 if (isipv6) {
2672 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2673 min_protoh;
2674 if (!in6_localaddr(&inp->in6p_faddr))
2675 mss = min(mss, tcp_v6mssdflt);
2676 } else {
2677 mss = ifp->if_mtu - min_protoh;
2678 if (!in_localaddr(inp->inp_faddr))
2679 mss = min(mss, tcp_mssdflt);
2680 }
2681 }
2682 mss = min(mss, offer);
2683 /*
2684 * maxopd stores the maximum length of data AND options
2685 * in a segment; maxseg is the amount of data in a normal
2686 * segment. We need to store this value (maxopd) apart
2687 * from maxseg, because now every segment carries options
2688 * and thus we normally have somewhat less data in segments.
2689 */
2690 tp->t_maxopd = mss;
2691
2692 /*
2693 * In case of T/TCP, origoffer==-1 indicates, that no segments
2694 * were received yet. In this case we just guess, otherwise
2695 * we do the same as before T/TCP.
2696 */
2697 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2698 (origoffer == -1 ||
2699 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2700 mss -= TCPOLEN_TSTAMP_APPA;
2701 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2702 (origoffer == -1 ||
2703 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2704 mss -= TCPOLEN_CC_APPA;
2705
2706 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2707 if (mss > MCLBYTES)
2708 mss &= ~(MCLBYTES-1);
2709 #else
2710 if (mss > MCLBYTES)
2711 mss = mss / MCLBYTES * MCLBYTES;
2712 #endif
2713 /*
2714 * If there's a pipesize, change the socket buffer
2715 * to that size. Make the socket buffers an integral
2716 * number of mss units; if the mss is larger than
2717 * the socket buffer, decrease the mss.
2718 */
2719 #ifdef RTV_SPIPE
2720 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2721 #endif
2722 bufsize = so->so_snd.sb_hiwat;
2723 if (bufsize < mss)
2724 mss = bufsize;
2725 else {
2726 bufsize = roundup(bufsize, mss);
2727 if (bufsize > sb_max)
2728 bufsize = sb_max;
2729 if (bufsize > so->so_snd.sb_hiwat)
2730 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2731 }
2732 tp->t_maxseg = mss;
2733
2734 #ifdef RTV_RPIPE
2735 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2736 #endif
2737 bufsize = so->so_rcv.sb_hiwat;
2738 if (bufsize > mss) {
2739 bufsize = roundup(bufsize, mss);
2740 if (bufsize > sb_max)
2741 bufsize = sb_max;
2742 if (bufsize > so->so_rcv.sb_hiwat)
2743 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2744 }
2745
2746 /*
2747 * Set the slow-start flight size depending on whether this
2748 * is a local network or not.
2749 */
2750 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2751 (!isipv6 && in_localaddr(inp->inp_faddr)))
2752 tp->snd_cwnd = mss * ss_fltsz_local;
2753 else
2754 tp->snd_cwnd = mss * ss_fltsz;
2755
2756 if (rt->rt_rmx.rmx_ssthresh) {
2757 /*
2758 * There's some sort of gateway or interface
2759 * buffer limit on the path. Use this to set
2760 * the slow start threshhold, but set the
2761 * threshold to no less than 2*mss.
2762 */
2763 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2764 tcpstat.tcps_usedssthresh++;
2765 }
2766 }
2767
2768 /*
2769 * Determine the MSS option to send on an outgoing SYN.
2770 */
2771 int
2772 tcp_mssopt(tp)
2773 struct tcpcb *tp;
2774 {
2775 struct rtentry *rt;
2776 #ifdef INET6
2777 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2778 int min_protoh = isipv6 ?
2779 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2780 sizeof(struct tcpiphdr);
2781 #else
2782 const int isipv6 = 0;
2783 const size_t min_protoh = sizeof(struct tcpiphdr);
2784 #endif
2785
2786 if (isipv6)
2787 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2788 else
2789 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2790 if (rt == NULL)
2791 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2792
2793 return (rt->rt_ifp->if_mtu - min_protoh);
2794 }
2795
2796
2797 /*
2798 * When a partial ack arrives, force the retransmission of the
2799 * next unacknowledged segment. Do not clear tp->t_dupacks.
2800 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2801 * be started again.
2802 */
2803 static void
2804 tcp_newreno_partial_ack(tp, th)
2805 struct tcpcb *tp;
2806 struct tcphdr *th;
2807 {
2808 tcp_seq onxt = tp->snd_nxt;
2809 u_long ocwnd = tp->snd_cwnd;
2810
2811 callout_stop(tp->tt_rexmt);
2812 tp->t_rtttime = 0;
2813 tp->snd_nxt = th->th_ack;
2814 /*
2815 * Set snd_cwnd to one segment beyond acknowledged offset
2816 * (tp->snd_una has not yet been updated when this function is called.)
2817 */
2818 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2819 tp->t_flags |= TF_ACKNOW;
2820 (void) tcp_output(tp);
2821 tp->snd_cwnd = ocwnd;
2822 if (SEQ_GT(onxt, tp->snd_nxt))
2823 tp->snd_nxt = onxt;
2824 /*
2825 * Partial window deflation. Relies on fact that tp->snd_una
2826 * not updated yet.
2827 */
2828 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
2829 }
Cache object: 52f57cbdcd28579bb9c0b200f569a93a
|