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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
30 * $FreeBSD: releng/5.3/sys/netinet/tcp_input.c 147670 2005-06-29 21:46:15Z simon $
31 */
32
33 #include "opt_ipfw.h" /* for ipfw_fwd */
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 #include "opt_tcp_sack.h"
41
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/mac.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h> /* for proc0 declaration */
48 #include <sys/protosw.h>
49 #include <sys/signalvar.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/systm.h>
55
56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
57
58 #include <vm/uma.h>
59
60 #include <net/if.h>
61 #include <net/route.h>
62
63 #include <netinet/in.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
70 #include <netinet/ip_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
83 #ifdef TCPDEBUG
84 #include <netinet/tcp_debug.h>
85 #endif /* TCPDEBUG */
86
87 #ifdef FAST_IPSEC
88 #include <netipsec/ipsec.h>
89 #include <netipsec/ipsec6.h>
90 #endif /*FAST_IPSEC*/
91
92 #ifdef IPSEC
93 #include <netinet6/ipsec.h>
94 #include <netinet6/ipsec6.h>
95 #include <netkey/key.h>
96 #endif /*IPSEC*/
97
98 #include <machine/in_cksum.h>
99
100 static const int tcprexmtthresh = 3;
101 tcp_cc tcp_ccgen;
102
103 struct tcpstat tcpstat;
104 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
105 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
106
107 static int log_in_vain = 0;
108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
109 &log_in_vain, 0, "Log all incoming TCP connections");
110
111 static int blackhole = 0;
112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
113 &blackhole, 0, "Do not send RST when dropping refused connections");
114
115 int tcp_delack_enabled = 1;
116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
117 &tcp_delack_enabled, 0,
118 "Delay ACK to try and piggyback it onto a data packet");
119
120 #ifdef TCP_DROP_SYNFIN
121 static int drop_synfin = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
123 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
124 #endif
125
126 static int tcp_do_rfc3042 = 1;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
128 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
129
130 static int tcp_do_rfc3390 = 1;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
132 &tcp_do_rfc3390, 0,
133 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
134
135 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
136 "TCP Segment Reassembly Queue");
137
138 static int tcp_reass_maxseg = 0;
139 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
140 &tcp_reass_maxseg, 0,
141 "Global maximum number of TCP Segments in Reassembly Queue");
142
143 int tcp_reass_qsize = 0;
144 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
145 &tcp_reass_qsize, 0,
146 "Global number of TCP Segments currently in Reassembly Queue");
147
148 static int tcp_reass_maxqlen = 48;
149 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
150 &tcp_reass_maxqlen, 0,
151 "Maximum number of TCP Segments per individual Reassembly Queue");
152
153 static int tcp_reass_overflows = 0;
154 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
155 &tcp_reass_overflows, 0,
156 "Global number of TCP Segment Reassembly Queue Overflows");
157
158 struct inpcbhead tcb;
159 #define tcb6 tcb /* for KAME src sync over BSD*'s */
160 struct inpcbinfo tcbinfo;
161 struct mtx *tcbinfo_mtx;
162
163 static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *,
164 int, int, struct tcphdr *);
165
166 static void tcp_pulloutofband(struct socket *,
167 struct tcphdr *, struct mbuf *, int);
168 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
169 struct mbuf *);
170 static void tcp_xmit_timer(struct tcpcb *, int);
171 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
172 static int tcp_timewait(struct tcptw *, struct tcpopt *,
173 struct tcphdr *, struct mbuf *, int);
174
175 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
176 #ifdef INET6
177 #define ND6_HINT(tp) \
178 do { \
179 if ((tp) && (tp)->t_inpcb && \
180 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
181 nd6_nud_hint(NULL, NULL, 0); \
182 } while (0)
183 #else
184 #define ND6_HINT(tp)
185 #endif
186
187 /*
188 * Indicate whether this ack should be delayed. We can delay the ack if
189 * - there is no delayed ack timer in progress and
190 * - our last ack wasn't a 0-sized window. We never want to delay
191 * the ack that opens up a 0-sized window and
192 * - delayed acks are enabled or
193 * - this is a half-synchronized T/TCP connection.
194 */
195 #define DELAY_ACK(tp) \
196 ((!callout_active(tp->tt_delack) && \
197 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
198 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
199
200 /* Initialize TCP reassembly queue */
201 uma_zone_t tcp_reass_zone;
202 void
203 tcp_reass_init()
204 {
205 tcp_reass_maxseg = nmbclusters / 16;
206 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
207 &tcp_reass_maxseg);
208 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
210 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
211 }
212
213 static int
214 tcp_reass(tp, th, tlenp, m)
215 register struct tcpcb *tp;
216 register struct tcphdr *th;
217 int *tlenp;
218 struct mbuf *m;
219 {
220 struct tseg_qent *q;
221 struct tseg_qent *p = NULL;
222 struct tseg_qent *nq;
223 struct tseg_qent *te = NULL;
224 struct socket *so = tp->t_inpcb->inp_socket;
225 int flags;
226
227 /*
228 * XXX: tcp_reass() is rather inefficient with its data structures
229 * and should be rewritten (see NetBSD for optimizations). While
230 * doing that it should move to its own file tcp_reass.c.
231 */
232
233 /*
234 * Call with th==0 after become established to
235 * force pre-ESTABLISHED data up to user socket.
236 */
237 if (th == 0)
238 goto present;
239
240 /*
241 * Limit the number of segments in the reassembly queue to prevent
242 * holding on to too many segments (and thus running out of mbufs).
243 * Make sure to let the missing segment through which caused this
244 * queue. Always keep one global queue entry spare to be able to
245 * process the missing segment.
246 */
247 if (th->th_seq != tp->rcv_nxt &&
248 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
249 tp->t_segqlen >= tcp_reass_maxqlen)) {
250 tcp_reass_overflows++;
251 tcpstat.tcps_rcvmemdrop++;
252 m_freem(m);
253 return (0);
254 }
255
256 /*
257 * Allocate a new queue entry. If we can't, or hit the zone limit
258 * just drop the pkt.
259 */
260 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
261 if (te == NULL) {
262 tcpstat.tcps_rcvmemdrop++;
263 m_freem(m);
264 return (0);
265 }
266 tp->t_segqlen++;
267 tcp_reass_qsize++;
268
269 /*
270 * Find a segment which begins after this one does.
271 */
272 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
273 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
274 break;
275 p = q;
276 }
277
278 /*
279 * If there is a preceding segment, it may provide some of
280 * our data already. If so, drop the data from the incoming
281 * segment. If it provides all of our data, drop us.
282 */
283 if (p != NULL) {
284 register int i;
285 /* conversion to int (in i) handles seq wraparound */
286 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
287 if (i > 0) {
288 if (i >= *tlenp) {
289 tcpstat.tcps_rcvduppack++;
290 tcpstat.tcps_rcvdupbyte += *tlenp;
291 m_freem(m);
292 uma_zfree(tcp_reass_zone, te);
293 tp->t_segqlen--;
294 tcp_reass_qsize--;
295 /*
296 * Try to present any queued data
297 * at the left window edge to the user.
298 * This is needed after the 3-WHS
299 * completes.
300 */
301 goto present; /* ??? */
302 }
303 m_adj(m, i);
304 *tlenp -= i;
305 th->th_seq += i;
306 }
307 }
308 tcpstat.tcps_rcvoopack++;
309 tcpstat.tcps_rcvoobyte += *tlenp;
310
311 /*
312 * While we overlap succeeding segments trim them or,
313 * if they are completely covered, dequeue them.
314 */
315 while (q) {
316 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
317 if (i <= 0)
318 break;
319 if (i < q->tqe_len) {
320 q->tqe_th->th_seq += i;
321 q->tqe_len -= i;
322 m_adj(q->tqe_m, i);
323 break;
324 }
325
326 nq = LIST_NEXT(q, tqe_q);
327 LIST_REMOVE(q, tqe_q);
328 m_freem(q->tqe_m);
329 uma_zfree(tcp_reass_zone, q);
330 tp->t_segqlen--;
331 tcp_reass_qsize--;
332 q = nq;
333 }
334
335 /* Insert the new segment queue entry into place. */
336 te->tqe_m = m;
337 te->tqe_th = th;
338 te->tqe_len = *tlenp;
339
340 if (p == NULL) {
341 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
342 } else {
343 LIST_INSERT_AFTER(p, te, tqe_q);
344 }
345
346 present:
347 /*
348 * Present data to user, advancing rcv_nxt through
349 * completed sequence space.
350 */
351 if (!TCPS_HAVEESTABLISHED(tp->t_state))
352 return (0);
353 q = LIST_FIRST(&tp->t_segq);
354 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
355 return (0);
356 SOCKBUF_LOCK(&so->so_rcv);
357 do {
358 tp->rcv_nxt += q->tqe_len;
359 flags = q->tqe_th->th_flags & TH_FIN;
360 nq = LIST_NEXT(q, tqe_q);
361 LIST_REMOVE(q, tqe_q);
362 /* Unlocked read. */
363 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
364 m_freem(q->tqe_m);
365 else
366 sbappendstream_locked(&so->so_rcv, q->tqe_m);
367 uma_zfree(tcp_reass_zone, q);
368 tp->t_segqlen--;
369 tcp_reass_qsize--;
370 q = nq;
371 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
372 ND6_HINT(tp);
373 sorwakeup_locked(so);
374 return (flags);
375 }
376
377 /*
378 * TCP input routine, follows pages 65-76 of the
379 * protocol specification dated September, 1981 very closely.
380 */
381 #ifdef INET6
382 int
383 tcp6_input(mp, offp, proto)
384 struct mbuf **mp;
385 int *offp, proto;
386 {
387 register struct mbuf *m = *mp;
388 struct in6_ifaddr *ia6;
389
390 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
391
392 /*
393 * draft-itojun-ipv6-tcp-to-anycast
394 * better place to put this in?
395 */
396 ia6 = ip6_getdstifaddr(m);
397 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
398 struct ip6_hdr *ip6;
399
400 ip6 = mtod(m, struct ip6_hdr *);
401 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
402 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
403 return IPPROTO_DONE;
404 }
405
406 tcp_input(m, *offp);
407 return IPPROTO_DONE;
408 }
409 #endif
410
411 void
412 tcp_input(m, off0)
413 register struct mbuf *m;
414 int off0;
415 {
416 register struct tcphdr *th;
417 register struct ip *ip = NULL;
418 register struct ipovly *ipov;
419 register struct inpcb *inp = NULL;
420 u_char *optp = NULL;
421 int optlen = 0;
422 int len, tlen, off;
423 int drop_hdrlen;
424 register struct tcpcb *tp = 0;
425 register int thflags;
426 struct socket *so = 0;
427 int todrop, acked, ourfinisacked, needoutput = 0;
428 u_long tiwin;
429 struct tcpopt to; /* options in this segment */
430 struct rmxp_tao tao; /* our TAO cache entry */
431 int headlocked = 0;
432 #ifdef IPFIREWALL_FORWARD
433 struct m_tag *fwd_tag;
434 #endif
435 int rstreason; /* For badport_bandlim accounting purposes */
436
437 struct ip6_hdr *ip6 = NULL;
438 #ifdef INET6
439 int isipv6;
440 #else
441 const int isipv6 = 0;
442 #endif
443
444 #ifdef TCPDEBUG
445 /*
446 * The size of tcp_saveipgen must be the size of the max ip header,
447 * now IPv6.
448 */
449 u_char tcp_saveipgen[40];
450 struct tcphdr tcp_savetcp;
451 short ostate = 0;
452 #endif
453
454 #ifdef INET6
455 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
456 #endif
457 bzero(&tao, sizeof(tao));
458 bzero((char *)&to, sizeof(to));
459
460 tcpstat.tcps_rcvtotal++;
461
462 if (isipv6) {
463 #ifdef INET6
464 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
465 ip6 = mtod(m, struct ip6_hdr *);
466 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
467 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
468 tcpstat.tcps_rcvbadsum++;
469 goto drop;
470 }
471 th = (struct tcphdr *)((caddr_t)ip6 + off0);
472
473 /*
474 * Be proactive about unspecified IPv6 address in source.
475 * As we use all-zero to indicate unbounded/unconnected pcb,
476 * unspecified IPv6 address can be used to confuse us.
477 *
478 * Note that packets with unspecified IPv6 destination is
479 * already dropped in ip6_input.
480 */
481 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
482 /* XXX stat */
483 goto drop;
484 }
485 #else
486 th = NULL; /* XXX: avoid compiler warning */
487 #endif
488 } else {
489 /*
490 * Get IP and TCP header together in first mbuf.
491 * Note: IP leaves IP header in first mbuf.
492 */
493 if (off0 > sizeof (struct ip)) {
494 ip_stripoptions(m, (struct mbuf *)0);
495 off0 = sizeof(struct ip);
496 }
497 if (m->m_len < sizeof (struct tcpiphdr)) {
498 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
499 tcpstat.tcps_rcvshort++;
500 return;
501 }
502 }
503 ip = mtod(m, struct ip *);
504 ipov = (struct ipovly *)ip;
505 th = (struct tcphdr *)((caddr_t)ip + off0);
506 tlen = ip->ip_len;
507
508 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
509 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
510 th->th_sum = m->m_pkthdr.csum_data;
511 else
512 th->th_sum = in_pseudo(ip->ip_src.s_addr,
513 ip->ip_dst.s_addr,
514 htonl(m->m_pkthdr.csum_data +
515 ip->ip_len +
516 IPPROTO_TCP));
517 th->th_sum ^= 0xffff;
518 #ifdef TCPDEBUG
519 ipov->ih_len = (u_short)tlen;
520 ipov->ih_len = htons(ipov->ih_len);
521 #endif
522 } else {
523 /*
524 * Checksum extended TCP header and data.
525 */
526 len = sizeof (struct ip) + tlen;
527 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
528 ipov->ih_len = (u_short)tlen;
529 ipov->ih_len = htons(ipov->ih_len);
530 th->th_sum = in_cksum(m, len);
531 }
532 if (th->th_sum) {
533 tcpstat.tcps_rcvbadsum++;
534 goto drop;
535 }
536 #ifdef INET6
537 /* Re-initialization for later version check */
538 ip->ip_v = IPVERSION;
539 #endif
540 }
541
542 /*
543 * Check that TCP offset makes sense,
544 * pull out TCP options and adjust length. XXX
545 */
546 off = th->th_off << 2;
547 if (off < sizeof (struct tcphdr) || off > tlen) {
548 tcpstat.tcps_rcvbadoff++;
549 goto drop;
550 }
551 tlen -= off; /* tlen is used instead of ti->ti_len */
552 if (off > sizeof (struct tcphdr)) {
553 if (isipv6) {
554 #ifdef INET6
555 IP6_EXTHDR_CHECK(m, off0, off, );
556 ip6 = mtod(m, struct ip6_hdr *);
557 th = (struct tcphdr *)((caddr_t)ip6 + off0);
558 #endif
559 } else {
560 if (m->m_len < sizeof(struct ip) + off) {
561 if ((m = m_pullup(m, sizeof (struct ip) + off))
562 == 0) {
563 tcpstat.tcps_rcvshort++;
564 return;
565 }
566 ip = mtod(m, struct ip *);
567 ipov = (struct ipovly *)ip;
568 th = (struct tcphdr *)((caddr_t)ip + off0);
569 }
570 }
571 optlen = off - sizeof (struct tcphdr);
572 optp = (u_char *)(th + 1);
573 }
574 thflags = th->th_flags;
575
576 #ifdef TCP_DROP_SYNFIN
577 /*
578 * If the drop_synfin option is enabled, drop all packets with
579 * both the SYN and FIN bits set. This prevents e.g. nmap from
580 * identifying the TCP/IP stack.
581 *
582 * This is a violation of the TCP specification.
583 */
584 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
585 goto drop;
586 #endif
587
588 /*
589 * Convert TCP protocol specific fields to host format.
590 */
591 th->th_seq = ntohl(th->th_seq);
592 th->th_ack = ntohl(th->th_ack);
593 th->th_win = ntohs(th->th_win);
594 th->th_urp = ntohs(th->th_urp);
595
596 /*
597 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
598 * until after ip6_savecontrol() is called and before other functions
599 * which don't want those proto headers.
600 * Because ip6_savecontrol() is going to parse the mbuf to
601 * search for data to be passed up to user-land, it wants mbuf
602 * parameters to be unchanged.
603 * XXX: the call of ip6_savecontrol() has been obsoleted based on
604 * latest version of the advanced API (20020110).
605 */
606 drop_hdrlen = off0 + off;
607
608 /*
609 * Locate pcb for segment.
610 */
611 INP_INFO_WLOCK(&tcbinfo);
612 headlocked = 1;
613 findpcb:
614 #ifdef IPFIREWALL_FORWARD
615 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
616 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
617
618 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
619 struct sockaddr_in *next_hop;
620
621 next_hop = (struct sockaddr_in *)(fwd_tag+1);
622 /*
623 * Transparently forwarded. Pretend to be the destination.
624 * already got one like this?
625 */
626 inp = in_pcblookup_hash(&tcbinfo,
627 ip->ip_src, th->th_sport,
628 ip->ip_dst, th->th_dport,
629 0, m->m_pkthdr.rcvif);
630 if (!inp) {
631 /* It's new. Try to find the ambushing socket. */
632 inp = in_pcblookup_hash(&tcbinfo,
633 ip->ip_src, th->th_sport,
634 next_hop->sin_addr,
635 next_hop->sin_port ?
636 ntohs(next_hop->sin_port) :
637 th->th_dport,
638 1, m->m_pkthdr.rcvif);
639 }
640 /* Remove the tag from the packet. We don't need it anymore. */
641 m_tag_delete(m, fwd_tag);
642 } else {
643 #endif /* IPFIREWALL_FORWARD */
644 if (isipv6) {
645 #ifdef INET6
646 inp = in6_pcblookup_hash(&tcbinfo,
647 &ip6->ip6_src, th->th_sport,
648 &ip6->ip6_dst, th->th_dport,
649 1, m->m_pkthdr.rcvif);
650 #endif
651 } else
652 inp = in_pcblookup_hash(&tcbinfo,
653 ip->ip_src, th->th_sport,
654 ip->ip_dst, th->th_dport,
655 1, m->m_pkthdr.rcvif);
656 #ifdef IPFIREWALL_FORWARD
657 }
658 #endif /* IPFIREWALL_FORWARD */
659
660 #if defined(IPSEC) || defined(FAST_IPSEC)
661 #ifdef INET6
662 if (isipv6) {
663 if (inp != NULL && ipsec6_in_reject(m, inp)) {
664 #ifdef IPSEC
665 ipsec6stat.in_polvio++;
666 #endif
667 goto drop;
668 }
669 } else
670 #endif /* INET6 */
671 if (inp != NULL && ipsec4_in_reject(m, inp)) {
672 #ifdef IPSEC
673 ipsecstat.in_polvio++;
674 #endif
675 goto drop;
676 }
677 #endif /*IPSEC || FAST_IPSEC*/
678
679 /*
680 * If the state is CLOSED (i.e., TCB does not exist) then
681 * all data in the incoming segment is discarded.
682 * If the TCB exists but is in CLOSED state, it is embryonic,
683 * but should either do a listen or a connect soon.
684 */
685 if (inp == NULL) {
686 if (log_in_vain) {
687 #ifdef INET6
688 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
689 #else
690 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
691 #endif
692
693 if (isipv6) {
694 #ifdef INET6
695 strcpy(dbuf, "[");
696 strcpy(sbuf, "[");
697 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
698 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
699 strcat(dbuf, "]");
700 strcat(sbuf, "]");
701 #endif
702 } else {
703 strcpy(dbuf, inet_ntoa(ip->ip_dst));
704 strcpy(sbuf, inet_ntoa(ip->ip_src));
705 }
706 switch (log_in_vain) {
707 case 1:
708 if ((thflags & TH_SYN) == 0)
709 break;
710 /* FALLTHROUGH */
711 case 2:
712 log(LOG_INFO,
713 "Connection attempt to TCP %s:%d "
714 "from %s:%d flags:0x%02x\n",
715 dbuf, ntohs(th->th_dport), sbuf,
716 ntohs(th->th_sport), thflags);
717 break;
718 default:
719 break;
720 }
721 }
722 if (blackhole) {
723 switch (blackhole) {
724 case 1:
725 if (thflags & TH_SYN)
726 goto drop;
727 break;
728 case 2:
729 goto drop;
730 default:
731 goto drop;
732 }
733 }
734 rstreason = BANDLIM_RST_CLOSEDPORT;
735 goto dropwithreset;
736 }
737 INP_LOCK(inp);
738 if (inp->inp_vflag & INP_TIMEWAIT) {
739 /*
740 * The only option of relevance is TOF_CC, and only if
741 * present in a SYN segment. See tcp_timewait().
742 */
743 if (thflags & TH_SYN)
744 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th);
745 if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
746 &to, th, m, tlen))
747 goto findpcb;
748 /*
749 * tcp_timewait unlocks inp.
750 */
751 INP_INFO_WUNLOCK(&tcbinfo);
752 return;
753 }
754 tp = intotcpcb(inp);
755 if (tp == 0) {
756 INP_UNLOCK(inp);
757 rstreason = BANDLIM_RST_CLOSEDPORT;
758 goto dropwithreset;
759 }
760 if (tp->t_state == TCPS_CLOSED)
761 goto drop;
762
763 /* Unscale the window into a 32-bit value. */
764 if ((thflags & TH_SYN) == 0)
765 tiwin = th->th_win << tp->snd_scale;
766 else
767 tiwin = th->th_win;
768
769 #ifdef MAC
770 INP_LOCK_ASSERT(inp);
771 if (mac_check_inpcb_deliver(inp, m))
772 goto drop;
773 #endif
774 so = inp->inp_socket;
775 #ifdef TCPDEBUG
776 if (so->so_options & SO_DEBUG) {
777 ostate = tp->t_state;
778 if (isipv6)
779 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
780 else
781 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
782 tcp_savetcp = *th;
783 }
784 #endif
785 if (so->so_options & SO_ACCEPTCONN) {
786 struct in_conninfo inc;
787
788 #ifdef INET6
789 inc.inc_isipv6 = isipv6;
790 #endif
791 if (isipv6) {
792 inc.inc6_faddr = ip6->ip6_src;
793 inc.inc6_laddr = ip6->ip6_dst;
794 } else {
795 inc.inc_faddr = ip->ip_src;
796 inc.inc_laddr = ip->ip_dst;
797 }
798 inc.inc_fport = th->th_sport;
799 inc.inc_lport = th->th_dport;
800
801 /*
802 * If the state is LISTEN then ignore segment if it contains
803 * a RST. If the segment contains an ACK then it is bad and
804 * send a RST. If it does not contain a SYN then it is not
805 * interesting; drop it.
806 *
807 * If the state is SYN_RECEIVED (syncache) and seg contains
808 * an ACK, but not for our SYN/ACK, send a RST. If the seg
809 * contains a RST, check the sequence number to see if it
810 * is a valid reset segment.
811 */
812 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
813 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
814 if (!syncache_expand(&inc, th, &so, m)) {
815 /*
816 * No syncache entry, or ACK was not
817 * for our SYN/ACK. Send a RST.
818 */
819 tcpstat.tcps_badsyn++;
820 rstreason = BANDLIM_RST_OPENPORT;
821 goto dropwithreset;
822 }
823 if (so == NULL) {
824 /*
825 * Could not complete 3-way handshake,
826 * connection is being closed down, and
827 * syncache will free mbuf.
828 */
829 INP_UNLOCK(inp);
830 INP_INFO_WUNLOCK(&tcbinfo);
831 return;
832 }
833 /*
834 * Socket is created in state SYN_RECEIVED.
835 * Continue processing segment.
836 */
837 INP_UNLOCK(inp);
838 inp = sotoinpcb(so);
839 INP_LOCK(inp);
840 tp = intotcpcb(inp);
841 /*
842 * This is what would have happened in
843 * tcp_output() when the SYN,ACK was sent.
844 */
845 tp->snd_up = tp->snd_una;
846 tp->snd_max = tp->snd_nxt = tp->iss + 1;
847 tp->last_ack_sent = tp->rcv_nxt;
848 /*
849 * RFC1323: The window in SYN & SYN/ACK
850 * segments is never scaled.
851 */
852 tp->snd_wnd = tiwin; /* unscaled */
853 goto after_listen;
854 }
855 if (thflags & TH_RST) {
856 syncache_chkrst(&inc, th);
857 goto drop;
858 }
859 if (thflags & TH_ACK) {
860 syncache_badack(&inc);
861 tcpstat.tcps_badsyn++;
862 rstreason = BANDLIM_RST_OPENPORT;
863 goto dropwithreset;
864 }
865 goto drop;
866 }
867
868 /*
869 * Segment's flags are (SYN) or (SYN|FIN).
870 */
871 #ifdef INET6
872 /*
873 * If deprecated address is forbidden,
874 * we do not accept SYN to deprecated interface
875 * address to prevent any new inbound connection from
876 * getting established.
877 * When we do not accept SYN, we send a TCP RST,
878 * with deprecated source address (instead of dropping
879 * it). We compromise it as it is much better for peer
880 * to send a RST, and RST will be the final packet
881 * for the exchange.
882 *
883 * If we do not forbid deprecated addresses, we accept
884 * the SYN packet. RFC2462 does not suggest dropping
885 * SYN in this case.
886 * If we decipher RFC2462 5.5.4, it says like this:
887 * 1. use of deprecated addr with existing
888 * communication is okay - "SHOULD continue to be
889 * used"
890 * 2. use of it with new communication:
891 * (2a) "SHOULD NOT be used if alternate address
892 * with sufficient scope is available"
893 * (2b) nothing mentioned otherwise.
894 * Here we fall into (2b) case as we have no choice in
895 * our source address selection - we must obey the peer.
896 *
897 * The wording in RFC2462 is confusing, and there are
898 * multiple description text for deprecated address
899 * handling - worse, they are not exactly the same.
900 * I believe 5.5.4 is the best one, so we follow 5.5.4.
901 */
902 if (isipv6 && !ip6_use_deprecated) {
903 struct in6_ifaddr *ia6;
904
905 if ((ia6 = ip6_getdstifaddr(m)) &&
906 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
907 INP_UNLOCK(inp);
908 tp = NULL;
909 rstreason = BANDLIM_RST_OPENPORT;
910 goto dropwithreset;
911 }
912 }
913 #endif
914 /*
915 * If it is from this socket, drop it, it must be forged.
916 * Don't bother responding if the destination was a broadcast.
917 */
918 if (th->th_dport == th->th_sport) {
919 if (isipv6) {
920 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
921 &ip6->ip6_src))
922 goto drop;
923 } else {
924 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
925 goto drop;
926 }
927 }
928 /*
929 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
930 *
931 * Note that it is quite possible to receive unicast
932 * link-layer packets with a broadcast IP address. Use
933 * in_broadcast() to find them.
934 */
935 if (m->m_flags & (M_BCAST|M_MCAST))
936 goto drop;
937 if (isipv6) {
938 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
939 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
940 goto drop;
941 } else {
942 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
943 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
944 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
945 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
946 goto drop;
947 }
948 /*
949 * SYN appears to be valid; create compressed TCP state
950 * for syncache, or perform t/tcp connection.
951 */
952 if (so->so_qlen <= so->so_qlimit) {
953 #ifdef TCPDEBUG
954 if (so->so_options & SO_DEBUG)
955 tcp_trace(TA_INPUT, ostate, tp,
956 (void *)tcp_saveipgen, &tcp_savetcp, 0);
957 #endif
958 tcp_dooptions(tp, &to, optp, optlen, 1, th);
959 if (!syncache_add(&inc, &to, th, &so, m))
960 goto drop;
961 if (so == NULL) {
962 /*
963 * Entry added to syncache, mbuf used to
964 * send SYN,ACK packet.
965 */
966 KASSERT(headlocked, ("headlocked"));
967 INP_UNLOCK(inp);
968 INP_INFO_WUNLOCK(&tcbinfo);
969 return;
970 }
971 /*
972 * Segment passed TAO tests.
973 */
974 INP_UNLOCK(inp);
975 inp = sotoinpcb(so);
976 INP_LOCK(inp);
977 tp = intotcpcb(inp);
978 tp->snd_wnd = tiwin;
979 tp->t_starttime = ticks;
980 tp->t_state = TCPS_ESTABLISHED;
981
982 /*
983 * T/TCP logic:
984 * If there is a FIN or if there is data, then
985 * delay SYN,ACK(SYN) in the hope of piggy-backing
986 * it on a response segment. Otherwise must send
987 * ACK now in case the other side is slow starting.
988 */
989 if (thflags & TH_FIN || tlen != 0)
990 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
991 else
992 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
993 tcpstat.tcps_connects++;
994 soisconnected(so);
995 goto trimthenstep6;
996 }
997 goto drop;
998 }
999 after_listen:
1000 KASSERT(headlocked, ("tcp_input(): after_listen head is not locked"));
1001 INP_LOCK_ASSERT(inp);
1002
1003 /* XXX temp debugging */
1004 /* should not happen - syncache should pick up these connections */
1005 if (tp->t_state == TCPS_LISTEN)
1006 panic("tcp_input: TCPS_LISTEN");
1007
1008 /*
1009 * This is the second part of the MSS DoS prevention code (after
1010 * minmss on the sending side) and it deals with too many too small
1011 * tcp packets in a too short timeframe (1 second).
1012 *
1013 * For every full second we count the number of received packets
1014 * and bytes. If we get a lot of packets per second for this connection
1015 * (tcp_minmssoverload) we take a closer look at it and compute the
1016 * average packet size for the past second. If that is less than
1017 * tcp_minmss we get too many packets with very small payload which
1018 * is not good and burdens our system (and every packet generates
1019 * a wakeup to the process connected to our socket). We can reasonable
1020 * expect this to be small packet DoS attack to exhaust our CPU
1021 * cycles.
1022 *
1023 * Care has to be taken for the minimum packet overload value. This
1024 * value defines the minimum number of packets per second before we
1025 * start to worry. This must not be too low to avoid killing for
1026 * example interactive connections with many small packets like
1027 * telnet or SSH.
1028 *
1029 * Setting either tcp_minmssoverload or tcp_minmss to "" disables
1030 * this check.
1031 *
1032 * Account for packet if payload packet, skip over ACK, etc.
1033 */
1034 if (tcp_minmss && tcp_minmssoverload &&
1035 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1036 if (tp->rcv_second > ticks) {
1037 tp->rcv_pps++;
1038 tp->rcv_byps += tlen + off;
1039 if (tp->rcv_pps > tcp_minmssoverload) {
1040 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1041 printf("too many small tcp packets from "
1042 "%s:%u, av. %lubyte/packet, "
1043 "dropping connection\n",
1044 #ifdef INET6
1045 isipv6 ?
1046 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1047 #endif
1048 inet_ntoa(inp->inp_inc.inc_faddr),
1049 inp->inp_inc.inc_fport,
1050 tp->rcv_byps / tp->rcv_pps);
1051 tp = tcp_drop(tp, ECONNRESET);
1052 tcpstat.tcps_minmssdrops++;
1053 goto drop;
1054 }
1055 }
1056 } else {
1057 tp->rcv_second = ticks + hz;
1058 tp->rcv_pps = 1;
1059 tp->rcv_byps = tlen + off;
1060 }
1061 }
1062
1063 /*
1064 * Segment received on connection.
1065 * Reset idle time and keep-alive timer.
1066 */
1067 tp->t_rcvtime = ticks;
1068 if (TCPS_HAVEESTABLISHED(tp->t_state))
1069 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1070
1071 /*
1072 * Process options only when we get SYN/ACK back. The SYN case
1073 * for incoming connections is handled in tcp_syncache.
1074 * XXX this is traditional behavior, may need to be cleaned up.
1075 */
1076 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th);
1077 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1078 if (to.to_flags & TOF_SCALE) {
1079 tp->t_flags |= TF_RCVD_SCALE;
1080 tp->requested_s_scale = to.to_requested_s_scale;
1081 }
1082 if (to.to_flags & TOF_TS) {
1083 tp->t_flags |= TF_RCVD_TSTMP;
1084 tp->ts_recent = to.to_tsval;
1085 tp->ts_recent_age = ticks;
1086 }
1087 if (to.to_flags & (TOF_CC|TOF_CCNEW))
1088 tp->t_flags |= TF_RCVD_CC;
1089 if (to.to_flags & TOF_MSS)
1090 tcp_mss(tp, to.to_mss);
1091 if (tp->sack_enable) {
1092 if (!(to.to_flags & TOF_SACK))
1093 tp->sack_enable = 0;
1094 else
1095 tp->t_flags |= TF_SACK_PERMIT;
1096 }
1097
1098 }
1099
1100 if (tp->sack_enable) {
1101 /* Delete stale (cumulatively acked) SACK holes */
1102 tcp_del_sackholes(tp, th);
1103 tp->rcv_laststart = th->th_seq; /* last recv'd segment*/
1104 tp->rcv_lastend = th->th_seq + tlen;
1105 }
1106
1107 /*
1108 * Header prediction: check for the two common cases
1109 * of a uni-directional data xfer. If the packet has
1110 * no control flags, is in-sequence, the window didn't
1111 * change and we're not retransmitting, it's a
1112 * candidate. If the length is zero and the ack moved
1113 * forward, we're the sender side of the xfer. Just
1114 * free the data acked & wake any higher level process
1115 * that was blocked waiting for space. If the length
1116 * is non-zero and the ack didn't move, we're the
1117 * receiver side. If we're getting packets in-order
1118 * (the reassembly queue is empty), add the data to
1119 * the socket buffer and note that we need a delayed ack.
1120 * Make sure that the hidden state-flags are also off.
1121 * Since we check for TCPS_ESTABLISHED above, it can only
1122 * be TH_NEEDSYN.
1123 */
1124 if (tp->t_state == TCPS_ESTABLISHED &&
1125 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1126 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1127 ((to.to_flags & TOF_TS) == 0 ||
1128 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1129 /*
1130 * Using the CC option is compulsory if once started:
1131 * the segment is OK if no T/TCP was negotiated or
1132 * if the segment has a CC option equal to CCrecv
1133 */
1134 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1135 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1136 th->th_seq == tp->rcv_nxt &&
1137 tiwin && tiwin == tp->snd_wnd &&
1138 tp->snd_nxt == tp->snd_max) {
1139
1140 /*
1141 * If last ACK falls within this segment's sequence numbers,
1142 * record the timestamp.
1143 * NOTE that the test is modified according to the latest
1144 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1145 */
1146 if ((to.to_flags & TOF_TS) != 0 &&
1147 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1148 tp->ts_recent_age = ticks;
1149 tp->ts_recent = to.to_tsval;
1150 }
1151
1152 if (tlen == 0) {
1153 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1154 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1155 tp->snd_cwnd >= tp->snd_wnd &&
1156 ((!tcp_do_newreno && !tp->sack_enable &&
1157 tp->t_dupacks < tcprexmtthresh) ||
1158 ((tcp_do_newreno || tp->sack_enable) &&
1159 !IN_FASTRECOVERY(tp)))) {
1160 KASSERT(headlocked, ("headlocked"));
1161 INP_INFO_WUNLOCK(&tcbinfo);
1162 /*
1163 * this is a pure ack for outstanding data.
1164 */
1165 ++tcpstat.tcps_predack;
1166 /*
1167 * "bad retransmit" recovery
1168 */
1169 if (tp->t_rxtshift == 1 &&
1170 ticks < tp->t_badrxtwin) {
1171 ++tcpstat.tcps_sndrexmitbad;
1172 tp->snd_cwnd = tp->snd_cwnd_prev;
1173 tp->snd_ssthresh =
1174 tp->snd_ssthresh_prev;
1175 tp->snd_recover = tp->snd_recover_prev;
1176 if (tp->t_flags & TF_WASFRECOVERY)
1177 ENTER_FASTRECOVERY(tp);
1178 tp->snd_nxt = tp->snd_max;
1179 tp->t_badrxtwin = 0;
1180 }
1181
1182 /*
1183 * Recalculate the transmit timer / rtt.
1184 *
1185 * Some boxes send broken timestamp replies
1186 * during the SYN+ACK phase, ignore
1187 * timestamps of 0 or we could calculate a
1188 * huge RTT and blow up the retransmit timer.
1189 */
1190 if ((to.to_flags & TOF_TS) != 0 &&
1191 to.to_tsecr) {
1192 tcp_xmit_timer(tp,
1193 ticks - to.to_tsecr + 1);
1194 } else if (tp->t_rtttime &&
1195 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1196 tcp_xmit_timer(tp,
1197 ticks - tp->t_rtttime);
1198 }
1199 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1200 acked = th->th_ack - tp->snd_una;
1201 tcpstat.tcps_rcvackpack++;
1202 tcpstat.tcps_rcvackbyte += acked;
1203 sbdrop(&so->so_snd, acked);
1204 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1205 SEQ_LEQ(th->th_ack, tp->snd_recover))
1206 tp->snd_recover = th->th_ack - 1;
1207 tp->snd_una = th->th_ack;
1208 /*
1209 * pull snd_wl2 up to prevent seq wrap relative
1210 * to th_ack.
1211 */
1212 tp->snd_wl2 = th->th_ack;
1213 tp->t_dupacks = 0;
1214 m_freem(m);
1215 ND6_HINT(tp); /* some progress has been done */
1216
1217 /*
1218 * If all outstanding data are acked, stop
1219 * retransmit timer, otherwise restart timer
1220 * using current (possibly backed-off) value.
1221 * If process is waiting for space,
1222 * wakeup/selwakeup/signal. If data
1223 * are ready to send, let tcp_output
1224 * decide between more output or persist.
1225
1226 #ifdef TCPDEBUG
1227 if (so->so_options & SO_DEBUG)
1228 tcp_trace(TA_INPUT, ostate, tp,
1229 (void *)tcp_saveipgen,
1230 &tcp_savetcp, 0);
1231 #endif
1232 */
1233 if (tp->snd_una == tp->snd_max)
1234 callout_stop(tp->tt_rexmt);
1235 else if (!callout_active(tp->tt_persist))
1236 callout_reset(tp->tt_rexmt,
1237 tp->t_rxtcur,
1238 tcp_timer_rexmt, tp);
1239
1240 sowwakeup(so);
1241 if (so->so_snd.sb_cc)
1242 (void) tcp_output(tp);
1243 goto check_delack;
1244 }
1245 } else if (th->th_ack == tp->snd_una &&
1246 LIST_EMPTY(&tp->t_segq) &&
1247 tlen <= sbspace(&so->so_rcv)) {
1248 KASSERT(headlocked, ("headlocked"));
1249 INP_INFO_WUNLOCK(&tcbinfo);
1250 /*
1251 * this is a pure, in-sequence data packet
1252 * with nothing on the reassembly queue and
1253 * we have enough buffer space to take it.
1254 */
1255 /* Clean receiver SACK report if present */
1256 if (tp->sack_enable && tp->rcv_numsacks)
1257 tcp_clean_sackreport(tp);
1258 ++tcpstat.tcps_preddat;
1259 tp->rcv_nxt += tlen;
1260 /*
1261 * Pull snd_wl1 up to prevent seq wrap relative to
1262 * th_seq.
1263 */
1264 tp->snd_wl1 = th->th_seq;
1265 /*
1266 * Pull rcv_up up to prevent seq wrap relative to
1267 * rcv_nxt.
1268 */
1269 tp->rcv_up = tp->rcv_nxt;
1270 tcpstat.tcps_rcvpack++;
1271 tcpstat.tcps_rcvbyte += tlen;
1272 ND6_HINT(tp); /* some progress has been done */
1273 /*
1274 #ifdef TCPDEBUG
1275 if (so->so_options & SO_DEBUG)
1276 tcp_trace(TA_INPUT, ostate, tp,
1277 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1278 #endif
1279 * Add data to socket buffer.
1280 */
1281 /* Unlocked read. */
1282 SOCKBUF_LOCK(&so->so_rcv);
1283 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1284 m_freem(m);
1285 } else {
1286 m_adj(m, drop_hdrlen); /* delayed header drop */
1287 sbappendstream_locked(&so->so_rcv, m);
1288 }
1289 sorwakeup_locked(so);
1290 if (DELAY_ACK(tp)) {
1291 tp->t_flags |= TF_DELACK;
1292 } else {
1293 tp->t_flags |= TF_ACKNOW;
1294 tcp_output(tp);
1295 }
1296 goto check_delack;
1297 }
1298 }
1299
1300 /*
1301 * Calculate amount of space in receive window,
1302 * and then do TCP input processing.
1303 * Receive window is amount of space in rcv queue,
1304 * but not less than advertised window.
1305 */
1306 { int win;
1307
1308 win = sbspace(&so->so_rcv);
1309 if (win < 0)
1310 win = 0;
1311 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1312 }
1313
1314 switch (tp->t_state) {
1315
1316 /*
1317 * If the state is SYN_RECEIVED:
1318 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1319 */
1320 case TCPS_SYN_RECEIVED:
1321 if ((thflags & TH_ACK) &&
1322 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1323 SEQ_GT(th->th_ack, tp->snd_max))) {
1324 rstreason = BANDLIM_RST_OPENPORT;
1325 goto dropwithreset;
1326 }
1327 break;
1328
1329 /*
1330 * If the state is SYN_SENT:
1331 * if seg contains an ACK, but not for our SYN, drop the input.
1332 * if seg contains a RST, then drop the connection.
1333 * if seg does not contain SYN, then drop it.
1334 * Otherwise this is an acceptable SYN segment
1335 * initialize tp->rcv_nxt and tp->irs
1336 * if seg contains ack then advance tp->snd_una
1337 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1338 * arrange for segment to be acked (eventually)
1339 * continue processing rest of data/controls, beginning with URG
1340 */
1341 case TCPS_SYN_SENT:
1342 if (tcp_do_rfc1644)
1343 tcp_hc_gettao(&inp->inp_inc, &tao);
1344
1345 if ((thflags & TH_ACK) &&
1346 (SEQ_LEQ(th->th_ack, tp->iss) ||
1347 SEQ_GT(th->th_ack, tp->snd_max))) {
1348 /*
1349 * If we have a cached CCsent for the remote host,
1350 * hence we haven't just crashed and restarted,
1351 * do not send a RST. This may be a retransmission
1352 * from the other side after our earlier ACK was lost.
1353 * Our new SYN, when it arrives, will serve as the
1354 * needed ACK.
1355 */
1356 if (tao.tao_ccsent != 0)
1357 goto drop;
1358 else {
1359 rstreason = BANDLIM_UNLIMITED;
1360 goto dropwithreset;
1361 }
1362 }
1363 if (thflags & TH_RST) {
1364 if (thflags & TH_ACK)
1365 tp = tcp_drop(tp, ECONNREFUSED);
1366 goto drop;
1367 }
1368 if ((thflags & TH_SYN) == 0)
1369 goto drop;
1370 tp->snd_wnd = th->th_win; /* initial send window */
1371 tp->cc_recv = to.to_cc; /* foreign CC */
1372
1373 tp->irs = th->th_seq;
1374 tcp_rcvseqinit(tp);
1375 if (thflags & TH_ACK) {
1376 /*
1377 * Our SYN was acked. If segment contains CC.ECHO
1378 * option, check it to make sure this segment really
1379 * matches our SYN. If not, just drop it as old
1380 * duplicate, but send an RST if we're still playing
1381 * by the old rules. If no CC.ECHO option, make sure
1382 * we don't get fooled into using T/TCP.
1383 */
1384 if (to.to_flags & TOF_CCECHO) {
1385 if (tp->cc_send != to.to_ccecho) {
1386 if (tao.tao_ccsent != 0)
1387 goto drop;
1388 else {
1389 rstreason = BANDLIM_UNLIMITED;
1390 goto dropwithreset;
1391 }
1392 }
1393 } else
1394 tp->t_flags &= ~TF_RCVD_CC;
1395 tcpstat.tcps_connects++;
1396 soisconnected(so);
1397 #ifdef MAC
1398 SOCK_LOCK(so);
1399 mac_set_socket_peer_from_mbuf(m, so);
1400 SOCK_UNLOCK(so);
1401 #endif
1402 /* Do window scaling on this connection? */
1403 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1404 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1405 tp->snd_scale = tp->requested_s_scale;
1406 tp->rcv_scale = tp->request_r_scale;
1407 }
1408 /* Segment is acceptable, update cache if undefined. */
1409 if (tao.tao_ccsent == 0 && tcp_do_rfc1644)
1410 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0);
1411
1412 tp->rcv_adv += tp->rcv_wnd;
1413 tp->snd_una++; /* SYN is acked */
1414 /*
1415 * If there's data, delay ACK; if there's also a FIN
1416 * ACKNOW will be turned on later.
1417 */
1418 if (DELAY_ACK(tp) && tlen != 0)
1419 callout_reset(tp->tt_delack, tcp_delacktime,
1420 tcp_timer_delack, tp);
1421 else
1422 tp->t_flags |= TF_ACKNOW;
1423 /*
1424 * Received <SYN,ACK> in SYN_SENT[*] state.
1425 * Transitions:
1426 * SYN_SENT --> ESTABLISHED
1427 * SYN_SENT* --> FIN_WAIT_1
1428 */
1429 tp->t_starttime = ticks;
1430 if (tp->t_flags & TF_NEEDFIN) {
1431 tp->t_state = TCPS_FIN_WAIT_1;
1432 tp->t_flags &= ~TF_NEEDFIN;
1433 thflags &= ~TH_SYN;
1434 } else {
1435 tp->t_state = TCPS_ESTABLISHED;
1436 callout_reset(tp->tt_keep, tcp_keepidle,
1437 tcp_timer_keep, tp);
1438 }
1439 } else {
1440 /*
1441 * Received initial SYN in SYN-SENT[*] state =>
1442 * simultaneous open. If segment contains CC option
1443 * and there is a cached CC, apply TAO test.
1444 * If it succeeds, connection is * half-synchronized.
1445 * Otherwise, do 3-way handshake:
1446 * SYN-SENT -> SYN-RECEIVED
1447 * SYN-SENT* -> SYN-RECEIVED*
1448 * If there was no CC option, clear cached CC value.
1449 */
1450 tp->t_flags |= TF_ACKNOW;
1451 callout_stop(tp->tt_rexmt);
1452 if (to.to_flags & TOF_CC) {
1453 if (tao.tao_cc != 0 &&
1454 CC_GT(to.to_cc, tao.tao_cc)) {
1455 /*
1456 * update cache and make transition:
1457 * SYN-SENT -> ESTABLISHED*
1458 * SYN-SENT* -> FIN-WAIT-1*
1459 */
1460 tao.tao_cc = to.to_cc;
1461 tcp_hc_updatetao(&inp->inp_inc,
1462 TCP_HC_TAO_CC, to.to_cc, 0);
1463 tp->t_starttime = ticks;
1464 if (tp->t_flags & TF_NEEDFIN) {
1465 tp->t_state = TCPS_FIN_WAIT_1;
1466 tp->t_flags &= ~TF_NEEDFIN;
1467 } else {
1468 tp->t_state = TCPS_ESTABLISHED;
1469 callout_reset(tp->tt_keep,
1470 tcp_keepidle,
1471 tcp_timer_keep,
1472 tp);
1473 }
1474 tp->t_flags |= TF_NEEDSYN;
1475 } else
1476 tp->t_state = TCPS_SYN_RECEIVED;
1477 } else {
1478 if (tcp_do_rfc1644) {
1479 /* CC.NEW or no option => invalidate cache */
1480 tao.tao_cc = 0;
1481 tcp_hc_updatetao(&inp->inp_inc,
1482 TCP_HC_TAO_CC, to.to_cc, 0);
1483 }
1484 tp->t_state = TCPS_SYN_RECEIVED;
1485 }
1486 }
1487
1488 trimthenstep6:
1489 KASSERT(headlocked,
1490 ("tcp_input(): trimthenstep6 head is not locked"));
1491 INP_LOCK_ASSERT(inp);
1492
1493 /*
1494 * Advance th->th_seq to correspond to first data byte.
1495 * If data, trim to stay within window,
1496 * dropping FIN if necessary.
1497 */
1498 th->th_seq++;
1499 if (tlen > tp->rcv_wnd) {
1500 todrop = tlen - tp->rcv_wnd;
1501 m_adj(m, -todrop);
1502 tlen = tp->rcv_wnd;
1503 thflags &= ~TH_FIN;
1504 tcpstat.tcps_rcvpackafterwin++;
1505 tcpstat.tcps_rcvbyteafterwin += todrop;
1506 }
1507 tp->snd_wl1 = th->th_seq - 1;
1508 tp->rcv_up = th->th_seq;
1509 /*
1510 * Client side of transaction: already sent SYN and data.
1511 * If the remote host used T/TCP to validate the SYN,
1512 * our data will be ACK'd; if so, enter normal data segment
1513 * processing in the middle of step 5, ack processing.
1514 * Otherwise, goto step 6.
1515 */
1516 if (thflags & TH_ACK)
1517 goto process_ACK;
1518
1519 goto step6;
1520
1521 /*
1522 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1523 * if segment contains a SYN and CC [not CC.NEW] option:
1524 * if state == TIME_WAIT and connection duration > MSL,
1525 * drop packet and send RST;
1526 *
1527 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1528 * ack the FIN (and data) in retransmission queue.
1529 * Complete close and delete TCPCB. Then reprocess
1530 * segment, hoping to find new TCPCB in LISTEN state;
1531 *
1532 * else must be old SYN; drop it.
1533 * else do normal processing.
1534 */
1535 case TCPS_LAST_ACK:
1536 case TCPS_CLOSING:
1537 case TCPS_TIME_WAIT:
1538 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1539 if ((thflags & TH_SYN) &&
1540 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1541 if (tp->t_state == TCPS_TIME_WAIT &&
1542 (ticks - tp->t_starttime) > tcp_msl) {
1543 rstreason = BANDLIM_UNLIMITED;
1544 goto dropwithreset;
1545 }
1546 if (CC_GT(to.to_cc, tp->cc_recv)) {
1547 tp = tcp_close(tp);
1548 goto findpcb;
1549 }
1550 else
1551 goto drop;
1552 }
1553 break; /* continue normal processing */
1554 }
1555
1556 /*
1557 * States other than LISTEN or SYN_SENT.
1558 * First check the RST flag and sequence number since reset segments
1559 * are exempt from the timestamp and connection count tests. This
1560 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1561 * below which allowed reset segments in half the sequence space
1562 * to fall though and be processed (which gives forged reset
1563 * segments with a random sequence number a 50 percent chance of
1564 * killing a connection).
1565 * Then check timestamp, if present.
1566 * Then check the connection count, if present.
1567 * Then check that at least some bytes of segment are within
1568 * receive window. If segment begins before rcv_nxt,
1569 * drop leading data (and SYN); if nothing left, just ack.
1570 *
1571 *
1572 * If the RST bit is set, check the sequence number to see
1573 * if this is a valid reset segment.
1574 * RFC 793 page 37:
1575 * In all states except SYN-SENT, all reset (RST) segments
1576 * are validated by checking their SEQ-fields. A reset is
1577 * valid if its sequence number is in the window.
1578 * Note: this does not take into account delayed ACKs, so
1579 * we should test against last_ack_sent instead of rcv_nxt.
1580 * The sequence number in the reset segment is normally an
1581 * echo of our outgoing acknowlegement numbers, but some hosts
1582 * send a reset with the sequence number at the rightmost edge
1583 * of our receive window, and we have to handle this case.
1584 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1585 * that brute force RST attacks are possible. To combat this,
1586 * we use a much stricter check while in the ESTABLISHED state,
1587 * only accepting RSTs where the sequence number is equal to
1588 * last_ack_sent. In all other states (the states in which a
1589 * RST is more likely), the more permissive check is used.
1590 * If we have multiple segments in flight, the intial reset
1591 * segment sequence numbers will be to the left of last_ack_sent,
1592 * but they will eventually catch up.
1593 * In any case, it never made sense to trim reset segments to
1594 * fit the receive window since RFC 1122 says:
1595 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1596 *
1597 * A TCP SHOULD allow a received RST segment to include data.
1598 *
1599 * DISCUSSION
1600 * It has been suggested that a RST segment could contain
1601 * ASCII text that encoded and explained the cause of the
1602 * RST. No standard has yet been established for such
1603 * data.
1604 *
1605 * If the reset segment passes the sequence number test examine
1606 * the state:
1607 * SYN_RECEIVED STATE:
1608 * If passive open, return to LISTEN state.
1609 * If active open, inform user that connection was refused.
1610 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1611 * Inform user that connection was reset, and close tcb.
1612 * CLOSING, LAST_ACK STATES:
1613 * Close the tcb.
1614 * TIME_WAIT STATE:
1615 * Drop the segment - see Stevens, vol. 2, p. 964 and
1616 * RFC 1337.
1617 */
1618 if (thflags & TH_RST) {
1619 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1620 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1621 switch (tp->t_state) {
1622
1623 case TCPS_SYN_RECEIVED:
1624 so->so_error = ECONNREFUSED;
1625 goto close;
1626
1627 case TCPS_ESTABLISHED:
1628 if (tp->last_ack_sent != th->th_seq) {
1629 tcpstat.tcps_badrst++;
1630 goto drop;
1631 }
1632 case TCPS_FIN_WAIT_1:
1633 case TCPS_FIN_WAIT_2:
1634 case TCPS_CLOSE_WAIT:
1635 so->so_error = ECONNRESET;
1636 close:
1637 tp->t_state = TCPS_CLOSED;
1638 tcpstat.tcps_drops++;
1639 tp = tcp_close(tp);
1640 break;
1641
1642 case TCPS_CLOSING:
1643 case TCPS_LAST_ACK:
1644 tp = tcp_close(tp);
1645 break;
1646
1647 case TCPS_TIME_WAIT:
1648 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1649 ("timewait"));
1650 break;
1651 }
1652 }
1653 goto drop;
1654 }
1655
1656 /*
1657 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1658 * and it's less than ts_recent, drop it.
1659 */
1660 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1661 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1662
1663 /* Check to see if ts_recent is over 24 days old. */
1664 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1665 /*
1666 * Invalidate ts_recent. If this segment updates
1667 * ts_recent, the age will be reset later and ts_recent
1668 * will get a valid value. If it does not, setting
1669 * ts_recent to zero will at least satisfy the
1670 * requirement that zero be placed in the timestamp
1671 * echo reply when ts_recent isn't valid. The
1672 * age isn't reset until we get a valid ts_recent
1673 * because we don't want out-of-order segments to be
1674 * dropped when ts_recent is old.
1675 */
1676 tp->ts_recent = 0;
1677 } else {
1678 tcpstat.tcps_rcvduppack++;
1679 tcpstat.tcps_rcvdupbyte += tlen;
1680 tcpstat.tcps_pawsdrop++;
1681 if (tlen)
1682 goto dropafterack;
1683 goto drop;
1684 }
1685 }
1686
1687 /*
1688 * T/TCP mechanism
1689 * If T/TCP was negotiated and the segment doesn't have CC,
1690 * or if its CC is wrong then drop the segment.
1691 * RST segments do not have to comply with this.
1692 */
1693 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1694 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1695 goto dropafterack;
1696
1697 /*
1698 * In the SYN-RECEIVED state, validate that the packet belongs to
1699 * this connection before trimming the data to fit the receive
1700 * window. Check the sequence number versus IRS since we know
1701 * the sequence numbers haven't wrapped. This is a partial fix
1702 * for the "LAND" DoS attack.
1703 */
1704 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1705 rstreason = BANDLIM_RST_OPENPORT;
1706 goto dropwithreset;
1707 }
1708
1709 todrop = tp->rcv_nxt - th->th_seq;
1710 if (todrop > 0) {
1711 if (thflags & TH_SYN) {
1712 thflags &= ~TH_SYN;
1713 th->th_seq++;
1714 if (th->th_urp > 1)
1715 th->th_urp--;
1716 else
1717 thflags &= ~TH_URG;
1718 todrop--;
1719 }
1720 /*
1721 * Following if statement from Stevens, vol. 2, p. 960.
1722 */
1723 if (todrop > tlen
1724 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1725 /*
1726 * Any valid FIN must be to the left of the window.
1727 * At this point the FIN must be a duplicate or out
1728 * of sequence; drop it.
1729 */
1730 thflags &= ~TH_FIN;
1731
1732 /*
1733 * Send an ACK to resynchronize and drop any data.
1734 * But keep on processing for RST or ACK.
1735 */
1736 tp->t_flags |= TF_ACKNOW;
1737 todrop = tlen;
1738 tcpstat.tcps_rcvduppack++;
1739 tcpstat.tcps_rcvdupbyte += todrop;
1740 } else {
1741 tcpstat.tcps_rcvpartduppack++;
1742 tcpstat.tcps_rcvpartdupbyte += todrop;
1743 }
1744 drop_hdrlen += todrop; /* drop from the top afterwards */
1745 th->th_seq += todrop;
1746 tlen -= todrop;
1747 if (th->th_urp > todrop)
1748 th->th_urp -= todrop;
1749 else {
1750 thflags &= ~TH_URG;
1751 th->th_urp = 0;
1752 }
1753 }
1754
1755 /*
1756 * If new data are received on a connection after the
1757 * user processes are gone, then RST the other end.
1758 */
1759 if ((so->so_state & SS_NOFDREF) &&
1760 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1761 tp = tcp_close(tp);
1762 tcpstat.tcps_rcvafterclose++;
1763 rstreason = BANDLIM_UNLIMITED;
1764 goto dropwithreset;
1765 }
1766
1767 /*
1768 * If segment ends after window, drop trailing data
1769 * (and PUSH and FIN); if nothing left, just ACK.
1770 */
1771 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1772 if (todrop > 0) {
1773 tcpstat.tcps_rcvpackafterwin++;
1774 if (todrop >= tlen) {
1775 tcpstat.tcps_rcvbyteafterwin += tlen;
1776 /*
1777 * If a new connection request is received
1778 * while in TIME_WAIT, drop the old connection
1779 * and start over if the sequence numbers
1780 * are above the previous ones.
1781 */
1782 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1783 if (thflags & TH_SYN &&
1784 tp->t_state == TCPS_TIME_WAIT &&
1785 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1786 tp = tcp_close(tp);
1787 goto findpcb;
1788 }
1789 /*
1790 * If window is closed can only take segments at
1791 * window edge, and have to drop data and PUSH from
1792 * incoming segments. Continue processing, but
1793 * remember to ack. Otherwise, drop segment
1794 * and ack.
1795 */
1796 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1797 tp->t_flags |= TF_ACKNOW;
1798 tcpstat.tcps_rcvwinprobe++;
1799 } else
1800 goto dropafterack;
1801 } else
1802 tcpstat.tcps_rcvbyteafterwin += todrop;
1803 m_adj(m, -todrop);
1804 tlen -= todrop;
1805 thflags &= ~(TH_PUSH|TH_FIN);
1806 }
1807
1808 /*
1809 * If last ACK falls within this segment's sequence numbers,
1810 * record its timestamp.
1811 * NOTE:
1812 * 1) That the test incorporates suggestions from the latest
1813 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1814 * 2) That updating only on newer timestamps interferes with
1815 * our earlier PAWS tests, so this check should be solely
1816 * predicated on the sequence space of this segment.
1817 * 3) That we modify the segment boundary check to be
1818 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1819 * instead of RFC1323's
1820 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1821 * This modified check allows us to overcome RFC1323's
1822 * limitations as described in Stevens TCP/IP Illustrated
1823 * Vol. 2 p.869. In such cases, we can still calculate the
1824 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1825 */
1826 if ((to.to_flags & TOF_TS) != 0 &&
1827 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1828 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1829 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1830 tp->ts_recent_age = ticks;
1831 tp->ts_recent = to.to_tsval;
1832 }
1833
1834 /*
1835 * If a SYN is in the window, then this is an
1836 * error and we send an RST and drop the connection.
1837 */
1838 if (thflags & TH_SYN) {
1839 tp = tcp_drop(tp, ECONNRESET);
1840 rstreason = BANDLIM_UNLIMITED;
1841 goto drop;
1842 }
1843
1844 /*
1845 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1846 * flag is on (half-synchronized state), then queue data for
1847 * later processing; else drop segment and return.
1848 */
1849 if ((thflags & TH_ACK) == 0) {
1850 if (tp->t_state == TCPS_SYN_RECEIVED ||
1851 (tp->t_flags & TF_NEEDSYN))
1852 goto step6;
1853 else
1854 goto drop;
1855 }
1856
1857 /*
1858 * Ack processing.
1859 */
1860 switch (tp->t_state) {
1861
1862 /*
1863 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1864 * ESTABLISHED state and continue processing.
1865 * The ACK was checked above.
1866 */
1867 case TCPS_SYN_RECEIVED:
1868
1869 tcpstat.tcps_connects++;
1870 soisconnected(so);
1871 /* Do window scaling? */
1872 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1873 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1874 tp->snd_scale = tp->requested_s_scale;
1875 tp->rcv_scale = tp->request_r_scale;
1876 }
1877 /*
1878 * Upon successful completion of 3-way handshake,
1879 * update cache.CC, pass any queued data to the user,
1880 * and advance state appropriately.
1881 */
1882 if (tcp_do_rfc1644) {
1883 tao.tao_cc = tp->cc_recv;
1884 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC,
1885 tp->cc_recv, 0);
1886 }
1887 /*
1888 * Make transitions:
1889 * SYN-RECEIVED -> ESTABLISHED
1890 * SYN-RECEIVED* -> FIN-WAIT-1
1891 */
1892 tp->t_starttime = ticks;
1893 if (tp->t_flags & TF_NEEDFIN) {
1894 tp->t_state = TCPS_FIN_WAIT_1;
1895 tp->t_flags &= ~TF_NEEDFIN;
1896 } else {
1897 tp->t_state = TCPS_ESTABLISHED;
1898 callout_reset(tp->tt_keep, tcp_keepidle,
1899 tcp_timer_keep, tp);
1900 }
1901 /*
1902 * If segment contains data or ACK, will call tcp_reass()
1903 * later; if not, do so now to pass queued data to user.
1904 */
1905 if (tlen == 0 && (thflags & TH_FIN) == 0)
1906 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1907 (struct mbuf *)0);
1908 tp->snd_wl1 = th->th_seq - 1;
1909 /* FALLTHROUGH */
1910
1911 /*
1912 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1913 * ACKs. If the ack is in the range
1914 * tp->snd_una < th->th_ack <= tp->snd_max
1915 * then advance tp->snd_una to th->th_ack and drop
1916 * data from the retransmission queue. If this ACK reflects
1917 * more up to date window information we update our window information.
1918 */
1919 case TCPS_ESTABLISHED:
1920 case TCPS_FIN_WAIT_1:
1921 case TCPS_FIN_WAIT_2:
1922 case TCPS_CLOSE_WAIT:
1923 case TCPS_CLOSING:
1924 case TCPS_LAST_ACK:
1925 case TCPS_TIME_WAIT:
1926 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1927 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1928 if (tlen == 0 && tiwin == tp->snd_wnd) {
1929 tcpstat.tcps_rcvdupack++;
1930 /*
1931 * If we have outstanding data (other than
1932 * a window probe), this is a completely
1933 * duplicate ack (ie, window info didn't
1934 * change), the ack is the biggest we've
1935 * seen and we've seen exactly our rexmt
1936 * threshhold of them, assume a packet
1937 * has been dropped and retransmit it.
1938 * Kludge snd_nxt & the congestion
1939 * window so we send only this one
1940 * packet.
1941 *
1942 * We know we're losing at the current
1943 * window size so do congestion avoidance
1944 * (set ssthresh to half the current window
1945 * and pull our congestion window back to
1946 * the new ssthresh).
1947 *
1948 * Dup acks mean that packets have left the
1949 * network (they're now cached at the receiver)
1950 * so bump cwnd by the amount in the receiver
1951 * to keep a constant cwnd packets in the
1952 * network.
1953 */
1954 if (!callout_active(tp->tt_rexmt) ||
1955 th->th_ack != tp->snd_una)
1956 tp->t_dupacks = 0;
1957 else if (++tp->t_dupacks > tcprexmtthresh ||
1958 ((tcp_do_newreno || tp->sack_enable) &&
1959 IN_FASTRECOVERY(tp))) {
1960 tp->snd_cwnd += tp->t_maxseg;
1961 (void) tcp_output(tp);
1962 goto drop;
1963 } else if (tp->t_dupacks == tcprexmtthresh) {
1964 tcp_seq onxt = tp->snd_nxt;
1965 u_int win;
1966
1967 /*
1968 * If we're doing sack, check to
1969 * see if we're already in sack
1970 * recovery. If we're not doing sack,
1971 * check to see if we're in newreno
1972 * recovery.
1973 */
1974 if (tp->sack_enable) {
1975 if (IN_FASTRECOVERY(tp)) {
1976 tp->t_dupacks = 0;
1977 break;
1978 }
1979 } else if (tcp_do_newreno) {
1980 if (SEQ_LEQ(th->th_ack,
1981 tp->snd_recover)) {
1982 tp->t_dupacks = 0;
1983 break;
1984 }
1985 }
1986 win = min(tp->snd_wnd, tp->snd_cwnd) /
1987 2 / tp->t_maxseg;
1988 if (win < 2)
1989 win = 2;
1990 tp->snd_ssthresh = win * tp->t_maxseg;
1991 ENTER_FASTRECOVERY(tp);
1992 tp->snd_recover = tp->snd_max;
1993 callout_stop(tp->tt_rexmt);
1994 tp->t_rtttime = 0;
1995 if (tp->sack_enable) {
1996 tcpstat.tcps_sack_recovery_episode++;
1997 tp->snd_cwnd =
1998 tp->t_maxseg *
1999 tp->t_dupacks;
2000 (void) tcp_output(tp);
2001 tp->snd_cwnd +=
2002 tp->snd_ssthresh;
2003 goto drop;
2004 }
2005
2006 tp->snd_nxt = th->th_ack;
2007 tp->snd_cwnd = tp->t_maxseg;
2008 (void) tcp_output(tp);
2009 KASSERT(tp->snd_limited <= 2,
2010 ("tp->snd_limited too big"));
2011 tp->snd_cwnd = tp->snd_ssthresh +
2012 tp->t_maxseg *
2013 (tp->t_dupacks - tp->snd_limited);
2014 if (SEQ_GT(onxt, tp->snd_nxt))
2015 tp->snd_nxt = onxt;
2016 goto drop;
2017 } else if (tcp_do_rfc3042) {
2018 u_long oldcwnd = tp->snd_cwnd;
2019 tcp_seq oldsndmax = tp->snd_max;
2020 u_int sent;
2021
2022 KASSERT(tp->t_dupacks == 1 ||
2023 tp->t_dupacks == 2,
2024 ("dupacks not 1 or 2"));
2025 if (tp->t_dupacks == 1)
2026 tp->snd_limited = 0;
2027 tp->snd_cwnd =
2028 (tp->snd_nxt - tp->snd_una) +
2029 (tp->t_dupacks - tp->snd_limited) *
2030 tp->t_maxseg;
2031 (void) tcp_output(tp);
2032 sent = tp->snd_max - oldsndmax;
2033 if (sent > tp->t_maxseg) {
2034 KASSERT((tp->t_dupacks == 2 &&
2035 tp->snd_limited == 0) ||
2036 (sent == tp->t_maxseg + 1 &&
2037 tp->t_flags & TF_SENTFIN),
2038 ("sent too much"));
2039 tp->snd_limited = 2;
2040 } else if (sent > 0)
2041 ++tp->snd_limited;
2042 tp->snd_cwnd = oldcwnd;
2043 goto drop;
2044 }
2045 } else
2046 tp->t_dupacks = 0;
2047 break;
2048 }
2049
2050 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2051
2052 /*
2053 * If the congestion window was inflated to account
2054 * for the other side's cached packets, retract it.
2055 */
2056 if (tcp_do_newreno || tp->sack_enable) {
2057 if (IN_FASTRECOVERY(tp)) {
2058 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2059 if (tp->sack_enable)
2060 tcp_sack_partialack(tp, th);
2061 else
2062 tcp_newreno_partial_ack(tp, th);
2063 } else {
2064 /*
2065 * Out of fast recovery.
2066 * Window inflation should have left us
2067 * with approximately snd_ssthresh
2068 * outstanding data.
2069 * But in case we would be inclined to
2070 * send a burst, better to do it via
2071 * the slow start mechanism.
2072 */
2073 if (SEQ_GT(th->th_ack +
2074 tp->snd_ssthresh,
2075 tp->snd_max))
2076 tp->snd_cwnd = tp->snd_max -
2077 th->th_ack +
2078 tp->t_maxseg;
2079 else
2080 tp->snd_cwnd = tp->snd_ssthresh;
2081 }
2082 }
2083 } else {
2084 if (tp->t_dupacks >= tcprexmtthresh &&
2085 tp->snd_cwnd > tp->snd_ssthresh)
2086 tp->snd_cwnd = tp->snd_ssthresh;
2087 }
2088 tp->t_dupacks = 0;
2089 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2090 tcpstat.tcps_rcvacktoomuch++;
2091 goto dropafterack;
2092 }
2093 /*
2094 * If we reach this point, ACK is not a duplicate,
2095 * i.e., it ACKs something we sent.
2096 */
2097 if (tp->t_flags & TF_NEEDSYN) {
2098 /*
2099 * T/TCP: Connection was half-synchronized, and our
2100 * SYN has been ACK'd (so connection is now fully
2101 * synchronized). Go to non-starred state,
2102 * increment snd_una for ACK of SYN, and check if
2103 * we can do window scaling.
2104 */
2105 tp->t_flags &= ~TF_NEEDSYN;
2106 tp->snd_una++;
2107 /* Do window scaling? */
2108 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2109 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2110 tp->snd_scale = tp->requested_s_scale;
2111 tp->rcv_scale = tp->request_r_scale;
2112 }
2113 }
2114
2115 process_ACK:
2116 KASSERT(headlocked,
2117 ("tcp_input(): process_ACK head is not locked"));
2118 INP_LOCK_ASSERT(inp);
2119
2120 acked = th->th_ack - tp->snd_una;
2121 tcpstat.tcps_rcvackpack++;
2122 tcpstat.tcps_rcvackbyte += acked;
2123
2124 /*
2125 * If we just performed our first retransmit, and the ACK
2126 * arrives within our recovery window, then it was a mistake
2127 * to do the retransmit in the first place. Recover our
2128 * original cwnd and ssthresh, and proceed to transmit where
2129 * we left off.
2130 */
2131 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2132 ++tcpstat.tcps_sndrexmitbad;
2133 tp->snd_cwnd = tp->snd_cwnd_prev;
2134 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2135 tp->snd_recover = tp->snd_recover_prev;
2136 if (tp->t_flags & TF_WASFRECOVERY)
2137 ENTER_FASTRECOVERY(tp);
2138 tp->snd_nxt = tp->snd_max;
2139 tp->t_badrxtwin = 0; /* XXX probably not required */
2140 }
2141
2142 /*
2143 * If we have a timestamp reply, update smoothed
2144 * round trip time. If no timestamp is present but
2145 * transmit timer is running and timed sequence
2146 * number was acked, update smoothed round trip time.
2147 * Since we now have an rtt measurement, cancel the
2148 * timer backoff (cf., Phil Karn's retransmit alg.).
2149 * Recompute the initial retransmit timer.
2150 *
2151 * Some boxes send broken timestamp replies
2152 * during the SYN+ACK phase, ignore
2153 * timestamps of 0 or we could calculate a
2154 * huge RTT and blow up the retransmit timer.
2155 */
2156 if ((to.to_flags & TOF_TS) != 0 &&
2157 to.to_tsecr) {
2158 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2159 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2160 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2161 }
2162 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2163
2164 /*
2165 * If all outstanding data is acked, stop retransmit
2166 * timer and remember to restart (more output or persist).
2167 * If there is more data to be acked, restart retransmit
2168 * timer, using current (possibly backed-off) value.
2169 */
2170 if (th->th_ack == tp->snd_max) {
2171 callout_stop(tp->tt_rexmt);
2172 needoutput = 1;
2173 } else if (!callout_active(tp->tt_persist))
2174 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2175 tcp_timer_rexmt, tp);
2176
2177 /*
2178 * If no data (only SYN) was ACK'd,
2179 * skip rest of ACK processing.
2180 */
2181 if (acked == 0)
2182 goto step6;
2183
2184 /*
2185 * When new data is acked, open the congestion window.
2186 * If the window gives us less than ssthresh packets
2187 * in flight, open exponentially (maxseg per packet).
2188 * Otherwise open linearly: maxseg per window
2189 * (maxseg^2 / cwnd per packet).
2190 */
2191 if ((!tcp_do_newreno && !tp->sack_enable) ||
2192 !IN_FASTRECOVERY(tp)) {
2193 register u_int cw = tp->snd_cwnd;
2194 register u_int incr = tp->t_maxseg;
2195 if (cw > tp->snd_ssthresh)
2196 incr = incr * incr / cw;
2197 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2198 }
2199 SOCKBUF_LOCK(&so->so_snd);
2200 if (acked > so->so_snd.sb_cc) {
2201 tp->snd_wnd -= so->so_snd.sb_cc;
2202 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2203 ourfinisacked = 1;
2204 } else {
2205 sbdrop_locked(&so->so_snd, acked);
2206 tp->snd_wnd -= acked;
2207 ourfinisacked = 0;
2208 }
2209 sowwakeup_locked(so);
2210 /* detect una wraparound */
2211 if ((tcp_do_newreno || tp->sack_enable) &&
2212 !IN_FASTRECOVERY(tp) &&
2213 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2214 SEQ_LEQ(th->th_ack, tp->snd_recover))
2215 tp->snd_recover = th->th_ack - 1;
2216 if ((tcp_do_newreno || tp->sack_enable) &&
2217 IN_FASTRECOVERY(tp) &&
2218 SEQ_GEQ(th->th_ack, tp->snd_recover))
2219 EXIT_FASTRECOVERY(tp);
2220 tp->snd_una = th->th_ack;
2221 if (tp->sack_enable) {
2222 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2223 tp->snd_recover = tp->snd_una;
2224 }
2225 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2226 tp->snd_nxt = tp->snd_una;
2227
2228 switch (tp->t_state) {
2229
2230 /*
2231 * In FIN_WAIT_1 STATE in addition to the processing
2232 * for the ESTABLISHED state if our FIN is now acknowledged
2233 * then enter FIN_WAIT_2.
2234 */
2235 case TCPS_FIN_WAIT_1:
2236 if (ourfinisacked) {
2237 /*
2238 * If we can't receive any more
2239 * data, then closing user can proceed.
2240 * Starting the timer is contrary to the
2241 * specification, but if we don't get a FIN
2242 * we'll hang forever.
2243 */
2244 /* XXXjl
2245 * we should release the tp also, and use a
2246 * compressed state.
2247 */
2248 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2249 soisdisconnected(so);
2250 callout_reset(tp->tt_2msl, tcp_maxidle,
2251 tcp_timer_2msl, tp);
2252 }
2253 tp->t_state = TCPS_FIN_WAIT_2;
2254 }
2255 break;
2256
2257 /*
2258 * In CLOSING STATE in addition to the processing for
2259 * the ESTABLISHED state if the ACK acknowledges our FIN
2260 * then enter the TIME-WAIT state, otherwise ignore
2261 * the segment.
2262 */
2263 case TCPS_CLOSING:
2264 if (ourfinisacked) {
2265 KASSERT(headlocked, ("headlocked"));
2266 tcp_twstart(tp);
2267 INP_INFO_WUNLOCK(&tcbinfo);
2268 m_freem(m);
2269 return;
2270 }
2271 break;
2272
2273 /*
2274 * In LAST_ACK, we may still be waiting for data to drain
2275 * and/or to be acked, as well as for the ack of our FIN.
2276 * If our FIN is now acknowledged, delete the TCB,
2277 * enter the closed state and return.
2278 */
2279 case TCPS_LAST_ACK:
2280 if (ourfinisacked) {
2281 tp = tcp_close(tp);
2282 goto drop;
2283 }
2284 break;
2285
2286 /*
2287 * In TIME_WAIT state the only thing that should arrive
2288 * is a retransmission of the remote FIN. Acknowledge
2289 * it and restart the finack timer.
2290 */
2291 case TCPS_TIME_WAIT:
2292 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2293 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2294 tcp_timer_2msl, tp);
2295 goto dropafterack;
2296 }
2297 }
2298
2299 step6:
2300 KASSERT(headlocked, ("tcp_input(): step6 head is not locked"));
2301 INP_LOCK_ASSERT(inp);
2302
2303 /*
2304 * Update window information.
2305 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2306 */
2307 if ((thflags & TH_ACK) &&
2308 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2309 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2310 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2311 /* keep track of pure window updates */
2312 if (tlen == 0 &&
2313 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2314 tcpstat.tcps_rcvwinupd++;
2315 tp->snd_wnd = tiwin;
2316 tp->snd_wl1 = th->th_seq;
2317 tp->snd_wl2 = th->th_ack;
2318 if (tp->snd_wnd > tp->max_sndwnd)
2319 tp->max_sndwnd = tp->snd_wnd;
2320 needoutput = 1;
2321 }
2322
2323 /*
2324 * Process segments with URG.
2325 */
2326 if ((thflags & TH_URG) && th->th_urp &&
2327 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2328 /*
2329 * This is a kludge, but if we receive and accept
2330 * random urgent pointers, we'll crash in
2331 * soreceive. It's hard to imagine someone
2332 * actually wanting to send this much urgent data.
2333 */
2334 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2335 th->th_urp = 0; /* XXX */
2336 thflags &= ~TH_URG; /* XXX */
2337 goto dodata; /* XXX */
2338 }
2339 /*
2340 * If this segment advances the known urgent pointer,
2341 * then mark the data stream. This should not happen
2342 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2343 * a FIN has been received from the remote side.
2344 * In these states we ignore the URG.
2345 *
2346 * According to RFC961 (Assigned Protocols),
2347 * the urgent pointer points to the last octet
2348 * of urgent data. We continue, however,
2349 * to consider it to indicate the first octet
2350 * of data past the urgent section as the original
2351 * spec states (in one of two places).
2352 */
2353 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2354 tp->rcv_up = th->th_seq + th->th_urp;
2355 SOCKBUF_LOCK(&so->so_rcv);
2356 so->so_oobmark = so->so_rcv.sb_cc +
2357 (tp->rcv_up - tp->rcv_nxt) - 1;
2358 if (so->so_oobmark == 0)
2359 so->so_rcv.sb_state |= SBS_RCVATMARK;
2360 SOCKBUF_UNLOCK(&so->so_rcv);
2361 sohasoutofband(so);
2362 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2363 }
2364 /*
2365 * Remove out of band data so doesn't get presented to user.
2366 * This can happen independent of advancing the URG pointer,
2367 * but if two URG's are pending at once, some out-of-band
2368 * data may creep in... ick.
2369 */
2370 if (th->th_urp <= (u_long)tlen &&
2371 !(so->so_options & SO_OOBINLINE)) {
2372 /* hdr drop is delayed */
2373 tcp_pulloutofband(so, th, m, drop_hdrlen);
2374 }
2375 } else {
2376 /*
2377 * If no out of band data is expected,
2378 * pull receive urgent pointer along
2379 * with the receive window.
2380 */
2381 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2382 tp->rcv_up = tp->rcv_nxt;
2383 }
2384 dodata: /* XXX */
2385 KASSERT(headlocked, ("tcp_input(): dodata head is not locked"));
2386 INP_LOCK_ASSERT(inp);
2387
2388 /*
2389 * Process the segment text, merging it into the TCP sequencing queue,
2390 * and arranging for acknowledgment of receipt if necessary.
2391 * This process logically involves adjusting tp->rcv_wnd as data
2392 * is presented to the user (this happens in tcp_usrreq.c,
2393 * case PRU_RCVD). If a FIN has already been received on this
2394 * connection then we just ignore the text.
2395 */
2396 if ((tlen || (thflags & TH_FIN)) &&
2397 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2398 m_adj(m, drop_hdrlen); /* delayed header drop */
2399 /*
2400 * Insert segment which includes th into TCP reassembly queue
2401 * with control block tp. Set thflags to whether reassembly now
2402 * includes a segment with FIN. This handles the common case
2403 * inline (segment is the next to be received on an established
2404 * connection, and the queue is empty), avoiding linkage into
2405 * and removal from the queue and repetition of various
2406 * conversions.
2407 * Set DELACK for segments received in order, but ack
2408 * immediately when segments are out of order (so
2409 * fast retransmit can work).
2410 */
2411 if (th->th_seq == tp->rcv_nxt &&
2412 LIST_EMPTY(&tp->t_segq) &&
2413 TCPS_HAVEESTABLISHED(tp->t_state)) {
2414 if (DELAY_ACK(tp))
2415 tp->t_flags |= TF_DELACK;
2416 else
2417 tp->t_flags |= TF_ACKNOW;
2418 tp->rcv_nxt += tlen;
2419 thflags = th->th_flags & TH_FIN;
2420 tcpstat.tcps_rcvpack++;
2421 tcpstat.tcps_rcvbyte += tlen;
2422 ND6_HINT(tp);
2423 /* Unlocked read. */
2424 SOCKBUF_LOCK(&so->so_rcv);
2425 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2426 m_freem(m);
2427 else
2428 sbappendstream_locked(&so->so_rcv, m);
2429 sorwakeup_locked(so);
2430 } else {
2431 thflags = tcp_reass(tp, th, &tlen, m);
2432 tp->t_flags |= TF_ACKNOW;
2433 }
2434 if (tp->sack_enable)
2435 tcp_update_sack_list(tp);
2436 /*
2437 * Note the amount of data that peer has sent into
2438 * our window, in order to estimate the sender's
2439 * buffer size.
2440 */
2441 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2442 } else {
2443 m_freem(m);
2444 thflags &= ~TH_FIN;
2445 }
2446
2447 /*
2448 * If FIN is received ACK the FIN and let the user know
2449 * that the connection is closing.
2450 */
2451 if (thflags & TH_FIN) {
2452 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2453 socantrcvmore(so);
2454 /*
2455 * If connection is half-synchronized
2456 * (ie NEEDSYN flag on) then delay ACK,
2457 * so it may be piggybacked when SYN is sent.
2458 * Otherwise, since we received a FIN then no
2459 * more input can be expected, send ACK now.
2460 */
2461 if (tp->t_flags & TF_NEEDSYN)
2462 tp->t_flags |= TF_DELACK;
2463 else
2464 tp->t_flags |= TF_ACKNOW;
2465 tp->rcv_nxt++;
2466 }
2467 switch (tp->t_state) {
2468
2469 /*
2470 * In SYN_RECEIVED and ESTABLISHED STATES
2471 * enter the CLOSE_WAIT state.
2472 */
2473 case TCPS_SYN_RECEIVED:
2474 tp->t_starttime = ticks;
2475 /*FALLTHROUGH*/
2476 case TCPS_ESTABLISHED:
2477 tp->t_state = TCPS_CLOSE_WAIT;
2478 break;
2479
2480 /*
2481 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2482 * enter the CLOSING state.
2483 */
2484 case TCPS_FIN_WAIT_1:
2485 tp->t_state = TCPS_CLOSING;
2486 break;
2487
2488 /*
2489 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2490 * starting the time-wait timer, turning off the other
2491 * standard timers.
2492 */
2493 case TCPS_FIN_WAIT_2:
2494 KASSERT(headlocked == 1, ("headlocked should be 1"));
2495 tcp_twstart(tp);
2496 INP_INFO_WUNLOCK(&tcbinfo);
2497 return;
2498
2499 /*
2500 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2501 */
2502 case TCPS_TIME_WAIT:
2503 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2504 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2505 tcp_timer_2msl, tp);
2506 break;
2507 }
2508 }
2509 INP_INFO_WUNLOCK(&tcbinfo);
2510 #ifdef TCPDEBUG
2511 if (so->so_options & SO_DEBUG)
2512 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2513 &tcp_savetcp, 0);
2514 #endif
2515
2516 /*
2517 * Return any desired output.
2518 */
2519 if (needoutput || (tp->t_flags & TF_ACKNOW))
2520 (void) tcp_output(tp);
2521
2522 check_delack:
2523 INP_LOCK_ASSERT(inp);
2524 if (tp->t_flags & TF_DELACK) {
2525 tp->t_flags &= ~TF_DELACK;
2526 callout_reset(tp->tt_delack, tcp_delacktime,
2527 tcp_timer_delack, tp);
2528 }
2529 INP_UNLOCK(inp);
2530 return;
2531
2532 dropafterack:
2533 /*
2534 * Generate an ACK dropping incoming segment if it occupies
2535 * sequence space, where the ACK reflects our state.
2536 *
2537 * We can now skip the test for the RST flag since all
2538 * paths to this code happen after packets containing
2539 * RST have been dropped.
2540 *
2541 * In the SYN-RECEIVED state, don't send an ACK unless the
2542 * segment we received passes the SYN-RECEIVED ACK test.
2543 * If it fails send a RST. This breaks the loop in the
2544 * "LAND" DoS attack, and also prevents an ACK storm
2545 * between two listening ports that have been sent forged
2546 * SYN segments, each with the source address of the other.
2547 */
2548 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2549 (SEQ_GT(tp->snd_una, th->th_ack) ||
2550 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2551 rstreason = BANDLIM_RST_OPENPORT;
2552 goto dropwithreset;
2553 }
2554 #ifdef TCPDEBUG
2555 if (so->so_options & SO_DEBUG)
2556 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2557 &tcp_savetcp, 0);
2558 #endif
2559 KASSERT(headlocked, ("headlocked should be 1"));
2560 INP_INFO_WUNLOCK(&tcbinfo);
2561 m_freem(m);
2562 tp->t_flags |= TF_ACKNOW;
2563 (void) tcp_output(tp);
2564 INP_UNLOCK(inp);
2565 return;
2566
2567 dropwithreset:
2568 /*
2569 * Generate a RST, dropping incoming segment.
2570 * Make ACK acceptable to originator of segment.
2571 * Don't bother to respond if destination was broadcast/multicast.
2572 */
2573 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2574 goto drop;
2575 if (isipv6) {
2576 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2577 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2578 goto drop;
2579 } else {
2580 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2581 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2582 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2583 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2584 goto drop;
2585 }
2586 /* IPv6 anycast check is done at tcp6_input() */
2587
2588 /*
2589 * Perform bandwidth limiting.
2590 */
2591 if (badport_bandlim(rstreason) < 0)
2592 goto drop;
2593
2594 #ifdef TCPDEBUG
2595 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2596 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2597 &tcp_savetcp, 0);
2598 #endif
2599
2600 if (thflags & TH_ACK)
2601 /* mtod() below is safe as long as hdr dropping is delayed */
2602 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2603 TH_RST);
2604 else {
2605 if (thflags & TH_SYN)
2606 tlen++;
2607 /* mtod() below is safe as long as hdr dropping is delayed */
2608 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2609 (tcp_seq)0, TH_RST|TH_ACK);
2610 }
2611
2612 if (tp)
2613 INP_UNLOCK(inp);
2614 if (headlocked)
2615 INP_INFO_WUNLOCK(&tcbinfo);
2616 return;
2617
2618 drop:
2619 /*
2620 * Drop space held by incoming segment and return.
2621 */
2622 #ifdef TCPDEBUG
2623 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2624 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2625 &tcp_savetcp, 0);
2626 #endif
2627 if (tp)
2628 INP_UNLOCK(inp);
2629 m_freem(m);
2630 if (headlocked)
2631 INP_INFO_WUNLOCK(&tcbinfo);
2632 return;
2633 }
2634
2635 /*
2636 * Parse TCP options and place in tcpopt.
2637 */
2638 static void
2639 tcp_dooptions(tp, to, cp, cnt, is_syn, th)
2640 struct tcpcb *tp;
2641 struct tcpopt *to;
2642 u_char *cp;
2643 int cnt;
2644 int is_syn;
2645 struct tcphdr *th;
2646 {
2647 int opt, optlen;
2648
2649 to->to_flags = 0;
2650 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2651 opt = cp[0];
2652 if (opt == TCPOPT_EOL)
2653 break;
2654 if (opt == TCPOPT_NOP)
2655 optlen = 1;
2656 else {
2657 if (cnt < 2)
2658 break;
2659 optlen = cp[1];
2660 if (optlen < 2 || optlen > cnt)
2661 break;
2662 }
2663 switch (opt) {
2664 case TCPOPT_MAXSEG:
2665 if (optlen != TCPOLEN_MAXSEG)
2666 continue;
2667 if (!is_syn)
2668 continue;
2669 to->to_flags |= TOF_MSS;
2670 bcopy((char *)cp + 2,
2671 (char *)&to->to_mss, sizeof(to->to_mss));
2672 to->to_mss = ntohs(to->to_mss);
2673 break;
2674 case TCPOPT_WINDOW:
2675 if (optlen != TCPOLEN_WINDOW)
2676 continue;
2677 if (! is_syn)
2678 continue;
2679 to->to_flags |= TOF_SCALE;
2680 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2681 break;
2682 case TCPOPT_TIMESTAMP:
2683 if (optlen != TCPOLEN_TIMESTAMP)
2684 continue;
2685 to->to_flags |= TOF_TS;
2686 bcopy((char *)cp + 2,
2687 (char *)&to->to_tsval, sizeof(to->to_tsval));
2688 to->to_tsval = ntohl(to->to_tsval);
2689 bcopy((char *)cp + 6,
2690 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2691 to->to_tsecr = ntohl(to->to_tsecr);
2692 /*
2693 * If echoed timestamp is later than the current time,
2694 * fall back to non RFC1323 RTT calculation.
2695 */
2696 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2697 to->to_tsecr = 0;
2698 break;
2699 case TCPOPT_CC:
2700 if (optlen != TCPOLEN_CC)
2701 continue;
2702 to->to_flags |= TOF_CC;
2703 bcopy((char *)cp + 2,
2704 (char *)&to->to_cc, sizeof(to->to_cc));
2705 to->to_cc = ntohl(to->to_cc);
2706 break;
2707 case TCPOPT_CCNEW:
2708 if (optlen != TCPOLEN_CC)
2709 continue;
2710 if (!is_syn)
2711 continue;
2712 to->to_flags |= TOF_CCNEW;
2713 bcopy((char *)cp + 2,
2714 (char *)&to->to_cc, sizeof(to->to_cc));
2715 to->to_cc = ntohl(to->to_cc);
2716 break;
2717 case TCPOPT_CCECHO:
2718 if (optlen != TCPOLEN_CC)
2719 continue;
2720 if (!is_syn)
2721 continue;
2722 to->to_flags |= TOF_CCECHO;
2723 bcopy((char *)cp + 2,
2724 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2725 to->to_ccecho = ntohl(to->to_ccecho);
2726 break;
2727 #ifdef TCP_SIGNATURE
2728 /*
2729 * XXX In order to reply to a host which has set the
2730 * TCP_SIGNATURE option in its initial SYN, we have to
2731 * record the fact that the option was observed here
2732 * for the syncache code to perform the correct response.
2733 */
2734 case TCPOPT_SIGNATURE:
2735 if (optlen != TCPOLEN_SIGNATURE)
2736 continue;
2737 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2738 break;
2739 #endif
2740 case TCPOPT_SACK_PERMITTED:
2741 if (!tcp_do_sack ||
2742 optlen != TCPOLEN_SACK_PERMITTED)
2743 continue;
2744 if (is_syn) {
2745 /* MUST only be set on SYN */
2746 to->to_flags |= TOF_SACK;
2747 }
2748 break;
2749
2750 case TCPOPT_SACK:
2751 if (!tp || tcp_sack_option(tp, th, cp, optlen))
2752 continue;
2753 break;
2754 default:
2755 continue;
2756 }
2757 }
2758 }
2759
2760 /*
2761 * Pull out of band byte out of a segment so
2762 * it doesn't appear in the user's data queue.
2763 * It is still reflected in the segment length for
2764 * sequencing purposes.
2765 */
2766 static void
2767 tcp_pulloutofband(so, th, m, off)
2768 struct socket *so;
2769 struct tcphdr *th;
2770 register struct mbuf *m;
2771 int off; /* delayed to be droped hdrlen */
2772 {
2773 int cnt = off + th->th_urp - 1;
2774
2775 while (cnt >= 0) {
2776 if (m->m_len > cnt) {
2777 char *cp = mtod(m, caddr_t) + cnt;
2778 struct tcpcb *tp = sototcpcb(so);
2779
2780 tp->t_iobc = *cp;
2781 tp->t_oobflags |= TCPOOB_HAVEDATA;
2782 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2783 m->m_len--;
2784 if (m->m_flags & M_PKTHDR)
2785 m->m_pkthdr.len--;
2786 return;
2787 }
2788 cnt -= m->m_len;
2789 m = m->m_next;
2790 if (m == 0)
2791 break;
2792 }
2793 panic("tcp_pulloutofband");
2794 }
2795
2796 /*
2797 * Collect new round-trip time estimate
2798 * and update averages and current timeout.
2799 */
2800 static void
2801 tcp_xmit_timer(tp, rtt)
2802 register struct tcpcb *tp;
2803 int rtt;
2804 {
2805 register int delta;
2806
2807 tcpstat.tcps_rttupdated++;
2808 tp->t_rttupdated++;
2809 if (tp->t_srtt != 0) {
2810 /*
2811 * srtt is stored as fixed point with 5 bits after the
2812 * binary point (i.e., scaled by 8). The following magic
2813 * is equivalent to the smoothing algorithm in rfc793 with
2814 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2815 * point). Adjust rtt to origin 0.
2816 */
2817 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2818 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2819
2820 if ((tp->t_srtt += delta) <= 0)
2821 tp->t_srtt = 1;
2822
2823 /*
2824 * We accumulate a smoothed rtt variance (actually, a
2825 * smoothed mean difference), then set the retransmit
2826 * timer to smoothed rtt + 4 times the smoothed variance.
2827 * rttvar is stored as fixed point with 4 bits after the
2828 * binary point (scaled by 16). The following is
2829 * equivalent to rfc793 smoothing with an alpha of .75
2830 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2831 * rfc793's wired-in beta.
2832 */
2833 if (delta < 0)
2834 delta = -delta;
2835 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2836 if ((tp->t_rttvar += delta) <= 0)
2837 tp->t_rttvar = 1;
2838 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2839 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2840 } else {
2841 /*
2842 * No rtt measurement yet - use the unsmoothed rtt.
2843 * Set the variance to half the rtt (so our first
2844 * retransmit happens at 3*rtt).
2845 */
2846 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2847 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2848 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2849 }
2850 tp->t_rtttime = 0;
2851 tp->t_rxtshift = 0;
2852
2853 /*
2854 * the retransmit should happen at rtt + 4 * rttvar.
2855 * Because of the way we do the smoothing, srtt and rttvar
2856 * will each average +1/2 tick of bias. When we compute
2857 * the retransmit timer, we want 1/2 tick of rounding and
2858 * 1 extra tick because of +-1/2 tick uncertainty in the
2859 * firing of the timer. The bias will give us exactly the
2860 * 1.5 tick we need. But, because the bias is
2861 * statistical, we have to test that we don't drop below
2862 * the minimum feasible timer (which is 2 ticks).
2863 */
2864 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2865 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2866
2867 /*
2868 * We received an ack for a packet that wasn't retransmitted;
2869 * it is probably safe to discard any error indications we've
2870 * received recently. This isn't quite right, but close enough
2871 * for now (a route might have failed after we sent a segment,
2872 * and the return path might not be symmetrical).
2873 */
2874 tp->t_softerror = 0;
2875 }
2876
2877 /*
2878 * Determine a reasonable value for maxseg size.
2879 * If the route is known, check route for mtu.
2880 * If none, use an mss that can be handled on the outgoing
2881 * interface without forcing IP to fragment; if bigger than
2882 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2883 * to utilize large mbufs. If no route is found, route has no mtu,
2884 * or the destination isn't local, use a default, hopefully conservative
2885 * size (usually 512 or the default IP max size, but no more than the mtu
2886 * of the interface), as we can't discover anything about intervening
2887 * gateways or networks. We also initialize the congestion/slow start
2888 * window to be a single segment if the destination isn't local.
2889 * While looking at the routing entry, we also initialize other path-dependent
2890 * parameters from pre-set or cached values in the routing entry.
2891 *
2892 * Also take into account the space needed for options that we
2893 * send regularly. Make maxseg shorter by that amount to assure
2894 * that we can send maxseg amount of data even when the options
2895 * are present. Store the upper limit of the length of options plus
2896 * data in maxopd.
2897 *
2898 *
2899 * In case of T/TCP, we call this routine during implicit connection
2900 * setup as well (offer = -1), to initialize maxseg from the cached
2901 * MSS of our peer.
2902 *
2903 * NOTE that this routine is only called when we process an incoming
2904 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2905 */
2906 void
2907 tcp_mss(tp, offer)
2908 struct tcpcb *tp;
2909 int offer;
2910 {
2911 int rtt, mss;
2912 u_long bufsize;
2913 u_long maxmtu;
2914 struct inpcb *inp = tp->t_inpcb;
2915 struct socket *so;
2916 struct hc_metrics_lite metrics;
2917 struct rmxp_tao tao;
2918 int origoffer = offer;
2919 #ifdef INET6
2920 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2921 size_t min_protoh = isipv6 ?
2922 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2923 sizeof (struct tcpiphdr);
2924 #else
2925 const size_t min_protoh = sizeof(struct tcpiphdr);
2926 #endif
2927 bzero(&tao, sizeof(tao));
2928
2929 /* initialize */
2930 #ifdef INET6
2931 if (isipv6) {
2932 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2933 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2934 } else
2935 #endif
2936 {
2937 maxmtu = tcp_maxmtu(&inp->inp_inc);
2938 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2939 }
2940 so = inp->inp_socket;
2941
2942 /*
2943 * no route to sender, stay with default mss and return
2944 */
2945 if (maxmtu == 0)
2946 return;
2947
2948 /* what have we got? */
2949 switch (offer) {
2950 case 0:
2951 /*
2952 * Offer == 0 means that there was no MSS on the SYN
2953 * segment, in this case we use tcp_mssdflt.
2954 */
2955 offer =
2956 #ifdef INET6
2957 isipv6 ? tcp_v6mssdflt :
2958 #endif
2959 tcp_mssdflt;
2960 break;
2961
2962 case -1:
2963 /*
2964 * Offer == -1 means that we didn't receive SYN yet,
2965 * use cached value in that case;
2966 */
2967 if (tcp_do_rfc1644)
2968 tcp_hc_gettao(&inp->inp_inc, &tao);
2969 if (tao.tao_mssopt != 0)
2970 offer = tao.tao_mssopt;
2971 /* FALLTHROUGH */
2972
2973 default:
2974 /*
2975 * Prevent DoS attack with too small MSS. Round up
2976 * to at least minmss.
2977 */
2978 offer = max(offer, tcp_minmss);
2979 /*
2980 * Sanity check: make sure that maxopd will be large
2981 * enough to allow some data on segments even if the
2982 * all the option space is used (40bytes). Otherwise
2983 * funny things may happen in tcp_output.
2984 */
2985 offer = max(offer, 64);
2986 if (tcp_do_rfc1644)
2987 tcp_hc_updatetao(&inp->inp_inc,
2988 TCP_HC_TAO_MSSOPT, 0, offer);
2989 }
2990
2991 /*
2992 * rmx information is now retrieved from tcp_hostcache
2993 */
2994 tcp_hc_get(&inp->inp_inc, &metrics);
2995
2996 /*
2997 * if there's a discovered mtu int tcp hostcache, use it
2998 * else, use the link mtu.
2999 */
3000 if (metrics.rmx_mtu)
3001 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3002 else {
3003 #ifdef INET6
3004 if (isipv6) {
3005 mss = maxmtu - min_protoh;
3006 if (!path_mtu_discovery &&
3007 !in6_localaddr(&inp->in6p_faddr))
3008 mss = min(mss, tcp_v6mssdflt);
3009 } else
3010 #endif
3011 {
3012 mss = maxmtu - min_protoh;
3013 if (!path_mtu_discovery &&
3014 !in_localaddr(inp->inp_faddr))
3015 mss = min(mss, tcp_mssdflt);
3016 }
3017 }
3018 mss = min(mss, offer);
3019
3020 /*
3021 * maxopd stores the maximum length of data AND options
3022 * in a segment; maxseg is the amount of data in a normal
3023 * segment. We need to store this value (maxopd) apart
3024 * from maxseg, because now every segment carries options
3025 * and thus we normally have somewhat less data in segments.
3026 */
3027 tp->t_maxopd = mss;
3028
3029 /*
3030 * In case of T/TCP, origoffer==-1 indicates, that no segments
3031 * were received yet. In this case we just guess, otherwise
3032 * we do the same as before T/TCP.
3033 */
3034 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3035 (origoffer == -1 ||
3036 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3037 mss -= TCPOLEN_TSTAMP_APPA;
3038 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
3039 (origoffer == -1 ||
3040 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3041 mss -= TCPOLEN_CC_APPA;
3042 tp->t_maxseg = mss;
3043
3044 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3045 if (mss > MCLBYTES)
3046 mss &= ~(MCLBYTES-1);
3047 #else
3048 if (mss > MCLBYTES)
3049 mss = mss / MCLBYTES * MCLBYTES;
3050 #endif
3051 tp->t_maxseg = mss;
3052
3053 /*
3054 * If there's a pipesize, change the socket buffer to that size,
3055 * don't change if sb_hiwat is different than default (then it
3056 * has been changed on purpose with setsockopt).
3057 * Make the socket buffers an integral number of mss units;
3058 * if the mss is larger than the socket buffer, decrease the mss.
3059 */
3060 SOCKBUF_LOCK(&so->so_snd);
3061 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3062 bufsize = metrics.rmx_sendpipe;
3063 else
3064 bufsize = so->so_snd.sb_hiwat;
3065 if (bufsize < mss)
3066 mss = bufsize;
3067 else {
3068 bufsize = roundup(bufsize, mss);
3069 if (bufsize > sb_max)
3070 bufsize = sb_max;
3071 if (bufsize > so->so_snd.sb_hiwat)
3072 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3073 }
3074 SOCKBUF_UNLOCK(&so->so_snd);
3075 tp->t_maxseg = mss;
3076
3077 SOCKBUF_LOCK(&so->so_rcv);
3078 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3079 bufsize = metrics.rmx_recvpipe;
3080 else
3081 bufsize = so->so_rcv.sb_hiwat;
3082 if (bufsize > mss) {
3083 bufsize = roundup(bufsize, mss);
3084 if (bufsize > sb_max)
3085 bufsize = sb_max;
3086 if (bufsize > so->so_rcv.sb_hiwat)
3087 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3088 }
3089 SOCKBUF_UNLOCK(&so->so_rcv);
3090 /*
3091 * While we're here, check the others too
3092 */
3093 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3094 tp->t_srtt = rtt;
3095 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3096 tcpstat.tcps_usedrtt++;
3097 if (metrics.rmx_rttvar) {
3098 tp->t_rttvar = metrics.rmx_rttvar;
3099 tcpstat.tcps_usedrttvar++;
3100 } else {
3101 /* default variation is +- 1 rtt */
3102 tp->t_rttvar =
3103 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3104 }
3105 TCPT_RANGESET(tp->t_rxtcur,
3106 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3107 tp->t_rttmin, TCPTV_REXMTMAX);
3108 }
3109 if (metrics.rmx_ssthresh) {
3110 /*
3111 * There's some sort of gateway or interface
3112 * buffer limit on the path. Use this to set
3113 * the slow start threshhold, but set the
3114 * threshold to no less than 2*mss.
3115 */
3116 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3117 tcpstat.tcps_usedssthresh++;
3118 }
3119 if (metrics.rmx_bandwidth)
3120 tp->snd_bandwidth = metrics.rmx_bandwidth;
3121
3122 /*
3123 * Set the slow-start flight size depending on whether this
3124 * is a local network or not.
3125 *
3126 * Extend this so we cache the cwnd too and retrieve it here.
3127 * Make cwnd even bigger than RFC3390 suggests but only if we
3128 * have previous experience with the remote host. Be careful
3129 * not make cwnd bigger than remote receive window or our own
3130 * send socket buffer. Maybe put some additional upper bound
3131 * on the retrieved cwnd. Should do incremental updates to
3132 * hostcache when cwnd collapses so next connection doesn't
3133 * overloads the path again.
3134 *
3135 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3136 * We currently check only in syncache_socket for that.
3137 */
3138 #define TCP_METRICS_CWND
3139 #ifdef TCP_METRICS_CWND
3140 if (metrics.rmx_cwnd)
3141 tp->snd_cwnd = max(mss,
3142 min(metrics.rmx_cwnd / 2,
3143 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3144 else
3145 #endif
3146 if (tcp_do_rfc3390)
3147 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3148 #ifdef INET6
3149 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3150 (!isipv6 && in_localaddr(inp->inp_faddr)))
3151 #else
3152 else if (in_localaddr(inp->inp_faddr))
3153 #endif
3154 tp->snd_cwnd = mss * ss_fltsz_local;
3155 else
3156 tp->snd_cwnd = mss * ss_fltsz;
3157 }
3158
3159 /*
3160 * Determine the MSS option to send on an outgoing SYN.
3161 */
3162 int
3163 tcp_mssopt(inc)
3164 struct in_conninfo *inc;
3165 {
3166 int mss = 0;
3167 u_long maxmtu = 0;
3168 u_long thcmtu = 0;
3169 size_t min_protoh;
3170 #ifdef INET6
3171 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3172 #endif
3173
3174 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3175
3176 #ifdef INET6
3177 if (isipv6) {
3178 mss = tcp_v6mssdflt;
3179 maxmtu = tcp_maxmtu6(inc);
3180 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3181 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3182 } else
3183 #endif
3184 {
3185 mss = tcp_mssdflt;
3186 maxmtu = tcp_maxmtu(inc);
3187 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3188 min_protoh = sizeof(struct tcpiphdr);
3189 }
3190 if (maxmtu && thcmtu)
3191 mss = min(maxmtu, thcmtu) - min_protoh;
3192 else if (maxmtu || thcmtu)
3193 mss = max(maxmtu, thcmtu) - min_protoh;
3194
3195 return (mss);
3196 }
3197
3198
3199 /*
3200 * On a partial ack arrives, force the retransmission of the
3201 * next unacknowledged segment. Do not clear tp->t_dupacks.
3202 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3203 * be started again.
3204 */
3205 static void
3206 tcp_newreno_partial_ack(tp, th)
3207 struct tcpcb *tp;
3208 struct tcphdr *th;
3209 {
3210 tcp_seq onxt = tp->snd_nxt;
3211 u_long ocwnd = tp->snd_cwnd;
3212
3213 callout_stop(tp->tt_rexmt);
3214 tp->t_rtttime = 0;
3215 tp->snd_nxt = th->th_ack;
3216 /*
3217 * Set snd_cwnd to one segment beyond acknowledged offset.
3218 * (tp->snd_una has not yet been updated when this function is called.)
3219 */
3220 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3221 tp->t_flags |= TF_ACKNOW;
3222 (void) tcp_output(tp);
3223 tp->snd_cwnd = ocwnd;
3224 if (SEQ_GT(onxt, tp->snd_nxt))
3225 tp->snd_nxt = onxt;
3226 /*
3227 * Partial window deflation. Relies on fact that tp->snd_una
3228 * not updated yet.
3229 */
3230 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3231 }
3232
3233 /*
3234 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3235 * looking for a pcb in the listen state. Returns 0 otherwise.
3236 */
3237 static int
3238 tcp_timewait(tw, to, th, m, tlen)
3239 struct tcptw *tw;
3240 struct tcpopt *to;
3241 struct tcphdr *th;
3242 struct mbuf *m;
3243 int tlen;
3244 {
3245 int thflags;
3246 tcp_seq seq;
3247 #ifdef INET6
3248 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3249 #else
3250 const int isipv6 = 0;
3251 #endif
3252
3253 thflags = th->th_flags;
3254
3255 /*
3256 * NOTE: for FIN_WAIT_2 (to be added later),
3257 * must validate sequence number before accepting RST
3258 */
3259
3260 /*
3261 * If the segment contains RST:
3262 * Drop the segment - see Stevens, vol. 2, p. 964 and
3263 * RFC 1337.
3264 */
3265 if (thflags & TH_RST)
3266 goto drop;
3267
3268 /*
3269 * If segment contains a SYN and CC [not CC.NEW] option:
3270 * if connection duration > MSL, drop packet and send RST;
3271 *
3272 * if SEG.CC > CCrecv then is new SYN.
3273 * Complete close and delete TCPCB. Then reprocess
3274 * segment, hoping to find new TCPCB in LISTEN state;
3275 *
3276 * else must be old SYN; drop it.
3277 * else do normal processing.
3278 */
3279 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) {
3280 if ((ticks - tw->t_starttime) > tcp_msl)
3281 goto reset;
3282 if (CC_GT(to->to_cc, tw->cc_recv)) {
3283 (void) tcp_twclose(tw, 0);
3284 return (1);
3285 }
3286 goto drop;
3287 }
3288
3289 #if 0
3290 /* PAWS not needed at the moment */
3291 /*
3292 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3293 * and it's less than ts_recent, drop it.
3294 */
3295 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3296 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3297 if ((thflags & TH_ACK) == 0)
3298 goto drop;
3299 goto ack;
3300 }
3301 /*
3302 * ts_recent is never updated because we never accept new segments.
3303 */
3304 #endif
3305
3306 /*
3307 * If a new connection request is received
3308 * while in TIME_WAIT, drop the old connection
3309 * and start over if the sequence numbers
3310 * are above the previous ones.
3311 */
3312 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3313 (void) tcp_twclose(tw, 0);
3314 return (1);
3315 }
3316
3317 /*
3318 * Drop the the segment if it does not contain an ACK.
3319 */
3320 if ((thflags & TH_ACK) == 0)
3321 goto drop;
3322
3323 /*
3324 * Reset the 2MSL timer if this is a duplicate FIN.
3325 */
3326 if (thflags & TH_FIN) {
3327 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3328 if (seq + 1 == tw->rcv_nxt)
3329 tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3330 }
3331
3332 /*
3333 * Acknowledge the segment if it has data or is not a duplicate ACK.
3334 */
3335 if (thflags != TH_ACK || tlen != 0 ||
3336 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3337 tcp_twrespond(tw, TH_ACK);
3338 goto drop;
3339
3340 reset:
3341 /*
3342 * Generate a RST, dropping incoming segment.
3343 * Make ACK acceptable to originator of segment.
3344 * Don't bother to respond if destination was broadcast/multicast.
3345 */
3346 if (m->m_flags & (M_BCAST|M_MCAST))
3347 goto drop;
3348 if (isipv6) {
3349 struct ip6_hdr *ip6;
3350
3351 /* IPv6 anycast check is done at tcp6_input() */
3352 ip6 = mtod(m, struct ip6_hdr *);
3353 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3354 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3355 goto drop;
3356 } else {
3357 struct ip *ip;
3358
3359 ip = mtod(m, struct ip *);
3360 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3361 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3362 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3363 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3364 goto drop;
3365 }
3366 if (thflags & TH_ACK) {
3367 tcp_respond(NULL,
3368 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3369 } else {
3370 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3371 tcp_respond(NULL,
3372 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3373 }
3374 INP_UNLOCK(tw->tw_inpcb);
3375 return (0);
3376
3377 drop:
3378 INP_UNLOCK(tw->tw_inpcb);
3379 m_freem(m);
3380 return (0);
3381 }
Cache object: 3d1f3e8d4757fc7b2c026e50ee367e51
|