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/5.4/sys/netinet/tcp_input.c 147668 2005-06-29 21:41:03Z 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 static int tcp_insecure_rst = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
137 &tcp_insecure_rst, 0,
138 "Follow the old (insecure) criteria for accepting RST packets.");
139
140 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
141 "TCP Segment Reassembly Queue");
142
143 static int tcp_reass_maxseg = 0;
144 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
145 &tcp_reass_maxseg, 0,
146 "Global maximum number of TCP Segments in Reassembly Queue");
147
148 int tcp_reass_qsize = 0;
149 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
150 &tcp_reass_qsize, 0,
151 "Global number of TCP Segments currently in Reassembly Queue");
152
153 static int tcp_reass_maxqlen = 48;
154 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
155 &tcp_reass_maxqlen, 0,
156 "Maximum number of TCP Segments per individual Reassembly Queue");
157
158 static int tcp_reass_overflows = 0;
159 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
160 &tcp_reass_overflows, 0,
161 "Global number of TCP Segment Reassembly Queue Overflows");
162
163 struct inpcbhead tcb;
164 #define tcb6 tcb /* for KAME src sync over BSD*'s */
165 struct inpcbinfo tcbinfo;
166 struct mtx *tcbinfo_mtx;
167
168 static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *,
169 int, int, struct tcphdr *);
170
171 static void tcp_pulloutofband(struct socket *,
172 struct tcphdr *, struct mbuf *, int);
173 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
174 struct mbuf *);
175 static void tcp_xmit_timer(struct tcpcb *, int);
176 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
177 static int tcp_timewait(struct tcptw *, struct tcpopt *,
178 struct tcphdr *, struct mbuf *, int);
179
180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
181 #ifdef INET6
182 #define ND6_HINT(tp) \
183 do { \
184 if ((tp) && (tp)->t_inpcb && \
185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
186 nd6_nud_hint(NULL, NULL, 0); \
187 } while (0)
188 #else
189 #define ND6_HINT(tp)
190 #endif
191
192 /*
193 * Indicate whether this ack should be delayed. We can delay the ack if
194 * - there is no delayed ack timer in progress and
195 * - our last ack wasn't a 0-sized window. We never want to delay
196 * the ack that opens up a 0-sized window and
197 * - delayed acks are enabled or
198 * - this is a half-synchronized T/TCP connection.
199 */
200 #define DELAY_ACK(tp) \
201 ((!callout_active(tp->tt_delack) && \
202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
204
205 /* Initialize TCP reassembly queue */
206 uma_zone_t tcp_reass_zone;
207 void
208 tcp_reass_init()
209 {
210 tcp_reass_maxseg = nmbclusters / 16;
211 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
212 &tcp_reass_maxseg);
213 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
215 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
216 }
217
218 static int
219 tcp_reass(tp, th, tlenp, m)
220 register struct tcpcb *tp;
221 register struct tcphdr *th;
222 int *tlenp;
223 struct mbuf *m;
224 {
225 struct tseg_qent *q;
226 struct tseg_qent *p = NULL;
227 struct tseg_qent *nq;
228 struct tseg_qent *te = NULL;
229 struct socket *so = tp->t_inpcb->inp_socket;
230 int flags;
231
232 INP_LOCK_ASSERT(tp->t_inpcb);
233
234 /*
235 * XXX: tcp_reass() is rather inefficient with its data structures
236 * and should be rewritten (see NetBSD for optimizations). While
237 * doing that it should move to its own file tcp_reass.c.
238 */
239
240 /*
241 * Call with th==NULL after become established to
242 * force pre-ESTABLISHED data up to user socket.
243 */
244 if (th == NULL)
245 goto present;
246
247 /*
248 * Limit the number of segments in the reassembly queue to prevent
249 * holding on to too many segments (and thus running out of mbufs).
250 * Make sure to let the missing segment through which caused this
251 * queue. Always keep one global queue entry spare to be able to
252 * process the missing segment.
253 */
254 if (th->th_seq != tp->rcv_nxt &&
255 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
256 tp->t_segqlen >= tcp_reass_maxqlen)) {
257 tcp_reass_overflows++;
258 tcpstat.tcps_rcvmemdrop++;
259 m_freem(m);
260 return (0);
261 }
262
263 /*
264 * Allocate a new queue entry. If we can't, or hit the zone limit
265 * just drop the pkt.
266 */
267 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
268 if (te == NULL) {
269 tcpstat.tcps_rcvmemdrop++;
270 m_freem(m);
271 return (0);
272 }
273 tp->t_segqlen++;
274 tcp_reass_qsize++;
275
276 /*
277 * Find a segment which begins after this one does.
278 */
279 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
280 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
281 break;
282 p = q;
283 }
284
285 /*
286 * If there is a preceding segment, it may provide some of
287 * our data already. If so, drop the data from the incoming
288 * segment. If it provides all of our data, drop us.
289 */
290 if (p != NULL) {
291 register int i;
292 /* conversion to int (in i) handles seq wraparound */
293 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
294 if (i > 0) {
295 if (i >= *tlenp) {
296 tcpstat.tcps_rcvduppack++;
297 tcpstat.tcps_rcvdupbyte += *tlenp;
298 m_freem(m);
299 uma_zfree(tcp_reass_zone, te);
300 tp->t_segqlen--;
301 tcp_reass_qsize--;
302 /*
303 * Try to present any queued data
304 * at the left window edge to the user.
305 * This is needed after the 3-WHS
306 * completes.
307 */
308 goto present; /* ??? */
309 }
310 m_adj(m, i);
311 *tlenp -= i;
312 th->th_seq += i;
313 }
314 }
315 tcpstat.tcps_rcvoopack++;
316 tcpstat.tcps_rcvoobyte += *tlenp;
317
318 /*
319 * While we overlap succeeding segments trim them or,
320 * if they are completely covered, dequeue them.
321 */
322 while (q) {
323 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
324 if (i <= 0)
325 break;
326 if (i < q->tqe_len) {
327 q->tqe_th->th_seq += i;
328 q->tqe_len -= i;
329 m_adj(q->tqe_m, i);
330 break;
331 }
332
333 nq = LIST_NEXT(q, tqe_q);
334 LIST_REMOVE(q, tqe_q);
335 m_freem(q->tqe_m);
336 uma_zfree(tcp_reass_zone, q);
337 tp->t_segqlen--;
338 tcp_reass_qsize--;
339 q = nq;
340 }
341
342 /* Insert the new segment queue entry into place. */
343 te->tqe_m = m;
344 te->tqe_th = th;
345 te->tqe_len = *tlenp;
346
347 if (p == NULL) {
348 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
349 } else {
350 LIST_INSERT_AFTER(p, te, tqe_q);
351 }
352
353 present:
354 /*
355 * Present data to user, advancing rcv_nxt through
356 * completed sequence space.
357 */
358 if (!TCPS_HAVEESTABLISHED(tp->t_state))
359 return (0);
360 q = LIST_FIRST(&tp->t_segq);
361 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
362 return (0);
363 SOCKBUF_LOCK(&so->so_rcv);
364 do {
365 tp->rcv_nxt += q->tqe_len;
366 flags = q->tqe_th->th_flags & TH_FIN;
367 nq = LIST_NEXT(q, tqe_q);
368 LIST_REMOVE(q, tqe_q);
369 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
370 m_freem(q->tqe_m);
371 else
372 sbappendstream_locked(&so->so_rcv, q->tqe_m);
373 uma_zfree(tcp_reass_zone, q);
374 tp->t_segqlen--;
375 tcp_reass_qsize--;
376 q = nq;
377 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
378 ND6_HINT(tp);
379 sorwakeup_locked(so);
380 return (flags);
381 }
382
383 /*
384 * TCP input routine, follows pages 65-76 of the
385 * protocol specification dated September, 1981 very closely.
386 */
387 #ifdef INET6
388 int
389 tcp6_input(mp, offp, proto)
390 struct mbuf **mp;
391 int *offp, proto;
392 {
393 register struct mbuf *m = *mp;
394 struct in6_ifaddr *ia6;
395
396 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
397
398 /*
399 * draft-itojun-ipv6-tcp-to-anycast
400 * better place to put this in?
401 */
402 ia6 = ip6_getdstifaddr(m);
403 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
404 struct ip6_hdr *ip6;
405
406 ip6 = mtod(m, struct ip6_hdr *);
407 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
408 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
409 return IPPROTO_DONE;
410 }
411
412 tcp_input(m, *offp);
413 return IPPROTO_DONE;
414 }
415 #endif
416
417 void
418 tcp_input(m, off0)
419 register struct mbuf *m;
420 int off0;
421 {
422 register struct tcphdr *th;
423 register struct ip *ip = NULL;
424 register struct ipovly *ipov;
425 register struct inpcb *inp = NULL;
426 u_char *optp = NULL;
427 int optlen = 0;
428 int len, tlen, off;
429 int drop_hdrlen;
430 register struct tcpcb *tp = 0;
431 register int thflags;
432 struct socket *so = 0;
433 int todrop, acked, ourfinisacked, needoutput = 0;
434 u_long tiwin;
435 struct tcpopt to; /* options in this segment */
436 struct rmxp_tao tao; /* our TAO cache entry */
437 int headlocked = 0;
438 #ifdef IPFIREWALL_FORWARD
439 struct m_tag *fwd_tag;
440 #endif
441 int rstreason; /* For badport_bandlim accounting purposes */
442
443 struct ip6_hdr *ip6 = NULL;
444 #ifdef INET6
445 int isipv6;
446 #else
447 const int isipv6 = 0;
448 #endif
449
450 #ifdef TCPDEBUG
451 /*
452 * The size of tcp_saveipgen must be the size of the max ip header,
453 * now IPv6.
454 */
455 u_char tcp_saveipgen[40];
456 struct tcphdr tcp_savetcp;
457 short ostate = 0;
458 #endif
459
460 #ifdef INET6
461 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
462 #endif
463 bzero(&tao, sizeof(tao));
464 bzero((char *)&to, sizeof(to));
465
466 tcpstat.tcps_rcvtotal++;
467
468 if (isipv6) {
469 #ifdef INET6
470 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
471 ip6 = mtod(m, struct ip6_hdr *);
472 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
473 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
474 tcpstat.tcps_rcvbadsum++;
475 goto drop;
476 }
477 th = (struct tcphdr *)((caddr_t)ip6 + off0);
478
479 /*
480 * Be proactive about unspecified IPv6 address in source.
481 * As we use all-zero to indicate unbounded/unconnected pcb,
482 * unspecified IPv6 address can be used to confuse us.
483 *
484 * Note that packets with unspecified IPv6 destination is
485 * already dropped in ip6_input.
486 */
487 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
488 /* XXX stat */
489 goto drop;
490 }
491 #else
492 th = NULL; /* XXX: avoid compiler warning */
493 #endif
494 } else {
495 /*
496 * Get IP and TCP header together in first mbuf.
497 * Note: IP leaves IP header in first mbuf.
498 */
499 if (off0 > sizeof (struct ip)) {
500 ip_stripoptions(m, (struct mbuf *)0);
501 off0 = sizeof(struct ip);
502 }
503 if (m->m_len < sizeof (struct tcpiphdr)) {
504 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
505 tcpstat.tcps_rcvshort++;
506 return;
507 }
508 }
509 ip = mtod(m, struct ip *);
510 ipov = (struct ipovly *)ip;
511 th = (struct tcphdr *)((caddr_t)ip + off0);
512 tlen = ip->ip_len;
513
514 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
515 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
516 th->th_sum = m->m_pkthdr.csum_data;
517 else
518 th->th_sum = in_pseudo(ip->ip_src.s_addr,
519 ip->ip_dst.s_addr,
520 htonl(m->m_pkthdr.csum_data +
521 ip->ip_len +
522 IPPROTO_TCP));
523 th->th_sum ^= 0xffff;
524 #ifdef TCPDEBUG
525 ipov->ih_len = (u_short)tlen;
526 ipov->ih_len = htons(ipov->ih_len);
527 #endif
528 } else {
529 /*
530 * Checksum extended TCP header and data.
531 */
532 len = sizeof (struct ip) + tlen;
533 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
534 ipov->ih_len = (u_short)tlen;
535 ipov->ih_len = htons(ipov->ih_len);
536 th->th_sum = in_cksum(m, len);
537 }
538 if (th->th_sum) {
539 tcpstat.tcps_rcvbadsum++;
540 goto drop;
541 }
542 #ifdef INET6
543 /* Re-initialization for later version check */
544 ip->ip_v = IPVERSION;
545 #endif
546 }
547
548 /*
549 * Check that TCP offset makes sense,
550 * pull out TCP options and adjust length. XXX
551 */
552 off = th->th_off << 2;
553 if (off < sizeof (struct tcphdr) || off > tlen) {
554 tcpstat.tcps_rcvbadoff++;
555 goto drop;
556 }
557 tlen -= off; /* tlen is used instead of ti->ti_len */
558 if (off > sizeof (struct tcphdr)) {
559 if (isipv6) {
560 #ifdef INET6
561 IP6_EXTHDR_CHECK(m, off0, off, );
562 ip6 = mtod(m, struct ip6_hdr *);
563 th = (struct tcphdr *)((caddr_t)ip6 + off0);
564 #endif
565 } else {
566 if (m->m_len < sizeof(struct ip) + off) {
567 if ((m = m_pullup(m, sizeof (struct ip) + off))
568 == 0) {
569 tcpstat.tcps_rcvshort++;
570 return;
571 }
572 ip = mtod(m, struct ip *);
573 ipov = (struct ipovly *)ip;
574 th = (struct tcphdr *)((caddr_t)ip + off0);
575 }
576 }
577 optlen = off - sizeof (struct tcphdr);
578 optp = (u_char *)(th + 1);
579 }
580 thflags = th->th_flags;
581
582 #ifdef TCP_DROP_SYNFIN
583 /*
584 * If the drop_synfin option is enabled, drop all packets with
585 * both the SYN and FIN bits set. This prevents e.g. nmap from
586 * identifying the TCP/IP stack.
587 *
588 * This is a violation of the TCP specification.
589 */
590 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
591 goto drop;
592 #endif
593
594 /*
595 * Convert TCP protocol specific fields to host format.
596 */
597 th->th_seq = ntohl(th->th_seq);
598 th->th_ack = ntohl(th->th_ack);
599 th->th_win = ntohs(th->th_win);
600 th->th_urp = ntohs(th->th_urp);
601
602 /*
603 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
604 * until after ip6_savecontrol() is called and before other functions
605 * which don't want those proto headers.
606 * Because ip6_savecontrol() is going to parse the mbuf to
607 * search for data to be passed up to user-land, it wants mbuf
608 * parameters to be unchanged.
609 * XXX: the call of ip6_savecontrol() has been obsoleted based on
610 * latest version of the advanced API (20020110).
611 */
612 drop_hdrlen = off0 + off;
613
614 /*
615 * Locate pcb for segment.
616 */
617 INP_INFO_WLOCK(&tcbinfo);
618 headlocked = 1;
619 findpcb:
620 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
621 #ifdef IPFIREWALL_FORWARD
622 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
623 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
624
625 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
626 struct sockaddr_in *next_hop;
627
628 next_hop = (struct sockaddr_in *)(fwd_tag+1);
629 /*
630 * Transparently forwarded. Pretend to be the destination.
631 * already got one like this?
632 */
633 inp = in_pcblookup_hash(&tcbinfo,
634 ip->ip_src, th->th_sport,
635 ip->ip_dst, th->th_dport,
636 0, m->m_pkthdr.rcvif);
637 if (!inp) {
638 /* It's new. Try to find the ambushing socket. */
639 inp = in_pcblookup_hash(&tcbinfo,
640 ip->ip_src, th->th_sport,
641 next_hop->sin_addr,
642 next_hop->sin_port ?
643 ntohs(next_hop->sin_port) :
644 th->th_dport,
645 1, m->m_pkthdr.rcvif);
646 }
647 /* Remove the tag from the packet. We don't need it anymore. */
648 m_tag_delete(m, fwd_tag);
649 } else {
650 #endif /* IPFIREWALL_FORWARD */
651 if (isipv6) {
652 #ifdef INET6
653 inp = in6_pcblookup_hash(&tcbinfo,
654 &ip6->ip6_src, th->th_sport,
655 &ip6->ip6_dst, th->th_dport,
656 1, m->m_pkthdr.rcvif);
657 #endif
658 } else
659 inp = in_pcblookup_hash(&tcbinfo,
660 ip->ip_src, th->th_sport,
661 ip->ip_dst, th->th_dport,
662 1, m->m_pkthdr.rcvif);
663 #ifdef IPFIREWALL_FORWARD
664 }
665 #endif /* IPFIREWALL_FORWARD */
666
667 #if defined(IPSEC) || defined(FAST_IPSEC)
668 #ifdef INET6
669 if (isipv6) {
670 if (inp != NULL && ipsec6_in_reject(m, inp)) {
671 #ifdef IPSEC
672 ipsec6stat.in_polvio++;
673 #endif
674 goto drop;
675 }
676 } else
677 #endif /* INET6 */
678 if (inp != NULL && ipsec4_in_reject(m, inp)) {
679 #ifdef IPSEC
680 ipsecstat.in_polvio++;
681 #endif
682 goto drop;
683 }
684 #endif /*IPSEC || FAST_IPSEC*/
685
686 /*
687 * If the state is CLOSED (i.e., TCB does not exist) then
688 * all data in the incoming segment is discarded.
689 * If the TCB exists but is in CLOSED state, it is embryonic,
690 * but should either do a listen or a connect soon.
691 */
692 if (inp == NULL) {
693 if (log_in_vain) {
694 #ifdef INET6
695 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
696 #else
697 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
698 #endif
699
700 if (isipv6) {
701 #ifdef INET6
702 strcpy(dbuf, "[");
703 strcpy(sbuf, "[");
704 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
705 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
706 strcat(dbuf, "]");
707 strcat(sbuf, "]");
708 #endif
709 } else {
710 strcpy(dbuf, inet_ntoa(ip->ip_dst));
711 strcpy(sbuf, inet_ntoa(ip->ip_src));
712 }
713 switch (log_in_vain) {
714 case 1:
715 if ((thflags & TH_SYN) == 0)
716 break;
717 /* FALLTHROUGH */
718 case 2:
719 log(LOG_INFO,
720 "Connection attempt to TCP %s:%d "
721 "from %s:%d flags:0x%02x\n",
722 dbuf, ntohs(th->th_dport), sbuf,
723 ntohs(th->th_sport), thflags);
724 break;
725 default:
726 break;
727 }
728 }
729 if (blackhole) {
730 switch (blackhole) {
731 case 1:
732 if (thflags & TH_SYN)
733 goto drop;
734 break;
735 case 2:
736 goto drop;
737 default:
738 goto drop;
739 }
740 }
741 rstreason = BANDLIM_RST_CLOSEDPORT;
742 goto dropwithreset;
743 }
744 INP_LOCK(inp);
745 if (inp->inp_vflag & INP_TIMEWAIT) {
746 /*
747 * The only option of relevance is TOF_CC, and only if
748 * present in a SYN segment. See tcp_timewait().
749 */
750 if (thflags & TH_SYN)
751 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th);
752 if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
753 &to, th, m, tlen))
754 goto findpcb;
755 /*
756 * tcp_timewait unlocks inp.
757 */
758 INP_INFO_WUNLOCK(&tcbinfo);
759 return;
760 }
761 tp = intotcpcb(inp);
762 if (tp == 0) {
763 INP_UNLOCK(inp);
764 rstreason = BANDLIM_RST_CLOSEDPORT;
765 goto dropwithreset;
766 }
767 if (tp->t_state == TCPS_CLOSED)
768 goto drop;
769
770 /* Unscale the window into a 32-bit value. */
771 if ((thflags & TH_SYN) == 0)
772 tiwin = th->th_win << tp->snd_scale;
773 else
774 tiwin = th->th_win;
775
776 #ifdef MAC
777 INP_LOCK_ASSERT(inp);
778 if (mac_check_inpcb_deliver(inp, m))
779 goto drop;
780 #endif
781 so = inp->inp_socket;
782 #ifdef TCPDEBUG
783 if (so->so_options & SO_DEBUG) {
784 ostate = tp->t_state;
785 if (isipv6)
786 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
787 else
788 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
789 tcp_savetcp = *th;
790 }
791 #endif
792 if (so->so_options & SO_ACCEPTCONN) {
793 struct in_conninfo inc;
794
795 #ifdef INET6
796 inc.inc_isipv6 = isipv6;
797 #endif
798 if (isipv6) {
799 inc.inc6_faddr = ip6->ip6_src;
800 inc.inc6_laddr = ip6->ip6_dst;
801 } else {
802 inc.inc_faddr = ip->ip_src;
803 inc.inc_laddr = ip->ip_dst;
804 }
805 inc.inc_fport = th->th_sport;
806 inc.inc_lport = th->th_dport;
807
808 /*
809 * If the state is LISTEN then ignore segment if it contains
810 * a RST. If the segment contains an ACK then it is bad and
811 * send a RST. If it does not contain a SYN then it is not
812 * interesting; drop it.
813 *
814 * If the state is SYN_RECEIVED (syncache) and seg contains
815 * an ACK, but not for our SYN/ACK, send a RST. If the seg
816 * contains a RST, check the sequence number to see if it
817 * is a valid reset segment.
818 */
819 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
820 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
821 if (!syncache_expand(&inc, th, &so, m)) {
822 /*
823 * No syncache entry, or ACK was not
824 * for our SYN/ACK. Send a RST.
825 */
826 tcpstat.tcps_badsyn++;
827 rstreason = BANDLIM_RST_OPENPORT;
828 goto dropwithreset;
829 }
830 if (so == NULL) {
831 /*
832 * Could not complete 3-way handshake,
833 * connection is being closed down, and
834 * syncache will free mbuf.
835 */
836 INP_UNLOCK(inp);
837 INP_INFO_WUNLOCK(&tcbinfo);
838 return;
839 }
840 /*
841 * Socket is created in state SYN_RECEIVED.
842 * Continue processing segment.
843 */
844 INP_UNLOCK(inp);
845 inp = sotoinpcb(so);
846 INP_LOCK(inp);
847 tp = intotcpcb(inp);
848 /*
849 * This is what would have happened in
850 * tcp_output() when the SYN,ACK was sent.
851 */
852 tp->snd_up = tp->snd_una;
853 tp->snd_max = tp->snd_nxt = tp->iss + 1;
854 tp->last_ack_sent = tp->rcv_nxt;
855 /*
856 * RFC1323: The window in SYN & SYN/ACK
857 * segments is never scaled.
858 */
859 tp->snd_wnd = tiwin; /* unscaled */
860 goto after_listen;
861 }
862 if (thflags & TH_RST) {
863 syncache_chkrst(&inc, th);
864 goto drop;
865 }
866 if (thflags & TH_ACK) {
867 syncache_badack(&inc);
868 tcpstat.tcps_badsyn++;
869 rstreason = BANDLIM_RST_OPENPORT;
870 goto dropwithreset;
871 }
872 goto drop;
873 }
874
875 /*
876 * Segment's flags are (SYN) or (SYN|FIN).
877 */
878 #ifdef INET6
879 /*
880 * If deprecated address is forbidden,
881 * we do not accept SYN to deprecated interface
882 * address to prevent any new inbound connection from
883 * getting established.
884 * When we do not accept SYN, we send a TCP RST,
885 * with deprecated source address (instead of dropping
886 * it). We compromise it as it is much better for peer
887 * to send a RST, and RST will be the final packet
888 * for the exchange.
889 *
890 * If we do not forbid deprecated addresses, we accept
891 * the SYN packet. RFC2462 does not suggest dropping
892 * SYN in this case.
893 * If we decipher RFC2462 5.5.4, it says like this:
894 * 1. use of deprecated addr with existing
895 * communication is okay - "SHOULD continue to be
896 * used"
897 * 2. use of it with new communication:
898 * (2a) "SHOULD NOT be used if alternate address
899 * with sufficient scope is available"
900 * (2b) nothing mentioned otherwise.
901 * Here we fall into (2b) case as we have no choice in
902 * our source address selection - we must obey the peer.
903 *
904 * The wording in RFC2462 is confusing, and there are
905 * multiple description text for deprecated address
906 * handling - worse, they are not exactly the same.
907 * I believe 5.5.4 is the best one, so we follow 5.5.4.
908 */
909 if (isipv6 && !ip6_use_deprecated) {
910 struct in6_ifaddr *ia6;
911
912 if ((ia6 = ip6_getdstifaddr(m)) &&
913 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
914 INP_UNLOCK(inp);
915 tp = NULL;
916 rstreason = BANDLIM_RST_OPENPORT;
917 goto dropwithreset;
918 }
919 }
920 #endif
921 /*
922 * If it is from this socket, drop it, it must be forged.
923 * Don't bother responding if the destination was a broadcast.
924 */
925 if (th->th_dport == th->th_sport) {
926 if (isipv6) {
927 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
928 &ip6->ip6_src))
929 goto drop;
930 } else {
931 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
932 goto drop;
933 }
934 }
935 /*
936 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
937 *
938 * Note that it is quite possible to receive unicast
939 * link-layer packets with a broadcast IP address. Use
940 * in_broadcast() to find them.
941 */
942 if (m->m_flags & (M_BCAST|M_MCAST))
943 goto drop;
944 if (isipv6) {
945 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
946 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
947 goto drop;
948 } else {
949 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
950 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
951 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
952 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
953 goto drop;
954 }
955 /*
956 * SYN appears to be valid; create compressed TCP state
957 * for syncache, or perform t/tcp connection.
958 */
959 if (so->so_qlen <= so->so_qlimit) {
960 #ifdef TCPDEBUG
961 if (so->so_options & SO_DEBUG)
962 tcp_trace(TA_INPUT, ostate, tp,
963 (void *)tcp_saveipgen, &tcp_savetcp, 0);
964 #endif
965 tcp_dooptions(tp, &to, optp, optlen, 1, th);
966 if (!syncache_add(&inc, &to, th, &so, m))
967 goto drop;
968 if (so == NULL) {
969 /*
970 * Entry added to syncache, mbuf used to
971 * send SYN,ACK packet.
972 */
973 KASSERT(headlocked, ("headlocked"));
974 INP_UNLOCK(inp);
975 INP_INFO_WUNLOCK(&tcbinfo);
976 return;
977 }
978 /*
979 * Segment passed TAO tests.
980 */
981 INP_UNLOCK(inp);
982 inp = sotoinpcb(so);
983 INP_LOCK(inp);
984 tp = intotcpcb(inp);
985 tp->snd_wnd = tiwin;
986 tp->t_starttime = ticks;
987 tp->t_state = TCPS_ESTABLISHED;
988
989 /*
990 * T/TCP logic:
991 * If there is a FIN or if there is data, then
992 * delay SYN,ACK(SYN) in the hope of piggy-backing
993 * it on a response segment. Otherwise must send
994 * ACK now in case the other side is slow starting.
995 */
996 if (thflags & TH_FIN || tlen != 0)
997 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
998 else
999 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1000 tcpstat.tcps_connects++;
1001 soisconnected(so);
1002 goto trimthenstep6;
1003 }
1004 goto drop;
1005 }
1006 after_listen:
1007 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1008 INP_LOCK_ASSERT(inp);
1009
1010 /* XXX temp debugging */
1011 /* should not happen - syncache should pick up these connections */
1012 if (tp->t_state == TCPS_LISTEN)
1013 panic("tcp_input: TCPS_LISTEN");
1014
1015 /*
1016 * This is the second part of the MSS DoS prevention code (after
1017 * minmss on the sending side) and it deals with too many too small
1018 * tcp packets in a too short timeframe (1 second).
1019 *
1020 * For every full second we count the number of received packets
1021 * and bytes. If we get a lot of packets per second for this connection
1022 * (tcp_minmssoverload) we take a closer look at it and compute the
1023 * average packet size for the past second. If that is less than
1024 * tcp_minmss we get too many packets with very small payload which
1025 * is not good and burdens our system (and every packet generates
1026 * a wakeup to the process connected to our socket). We can reasonable
1027 * expect this to be small packet DoS attack to exhaust our CPU
1028 * cycles.
1029 *
1030 * Care has to be taken for the minimum packet overload value. This
1031 * value defines the minimum number of packets per second before we
1032 * start to worry. This must not be too low to avoid killing for
1033 * example interactive connections with many small packets like
1034 * telnet or SSH.
1035 *
1036 * Setting either tcp_minmssoverload or tcp_minmss to "" disables
1037 * this check.
1038 *
1039 * Account for packet if payload packet, skip over ACK, etc.
1040 */
1041 if (tcp_minmss && tcp_minmssoverload &&
1042 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1043 if (tp->rcv_second > ticks) {
1044 tp->rcv_pps++;
1045 tp->rcv_byps += tlen + off;
1046 if (tp->rcv_pps > tcp_minmssoverload) {
1047 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1048 printf("too many small tcp packets from "
1049 "%s:%u, av. %lubyte/packet, "
1050 "dropping connection\n",
1051 #ifdef INET6
1052 isipv6 ?
1053 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1054 #endif
1055 inet_ntoa(inp->inp_inc.inc_faddr),
1056 inp->inp_inc.inc_fport,
1057 tp->rcv_byps / tp->rcv_pps);
1058 tp = tcp_drop(tp, ECONNRESET);
1059 tcpstat.tcps_minmssdrops++;
1060 goto drop;
1061 }
1062 }
1063 } else {
1064 tp->rcv_second = ticks + hz;
1065 tp->rcv_pps = 1;
1066 tp->rcv_byps = tlen + off;
1067 }
1068 }
1069
1070 /*
1071 * Segment received on connection.
1072 * Reset idle time and keep-alive timer.
1073 */
1074 tp->t_rcvtime = ticks;
1075 if (TCPS_HAVEESTABLISHED(tp->t_state))
1076 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1077
1078 /*
1079 * Process options only when we get SYN/ACK back. The SYN case
1080 * for incoming connections is handled in tcp_syncache.
1081 * XXX this is traditional behavior, may need to be cleaned up.
1082 */
1083 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th);
1084 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1085 if (to.to_flags & TOF_SCALE) {
1086 tp->t_flags |= TF_RCVD_SCALE;
1087 tp->requested_s_scale = to.to_requested_s_scale;
1088 }
1089 if (to.to_flags & TOF_TS) {
1090 tp->t_flags |= TF_RCVD_TSTMP;
1091 tp->ts_recent = to.to_tsval;
1092 tp->ts_recent_age = ticks;
1093 }
1094 if (to.to_flags & (TOF_CC|TOF_CCNEW))
1095 tp->t_flags |= TF_RCVD_CC;
1096 if (to.to_flags & TOF_MSS)
1097 tcp_mss(tp, to.to_mss);
1098 if (tp->sack_enable) {
1099 if (!(to.to_flags & TOF_SACK))
1100 tp->sack_enable = 0;
1101 else
1102 tp->t_flags |= TF_SACK_PERMIT;
1103 }
1104
1105 }
1106
1107 if (tp->sack_enable) {
1108 /* Delete stale (cumulatively acked) SACK holes */
1109 tcp_del_sackholes(tp, th);
1110 }
1111
1112 /*
1113 * Header prediction: check for the two common cases
1114 * of a uni-directional data xfer. If the packet has
1115 * no control flags, is in-sequence, the window didn't
1116 * change and we're not retransmitting, it's a
1117 * candidate. If the length is zero and the ack moved
1118 * forward, we're the sender side of the xfer. Just
1119 * free the data acked & wake any higher level process
1120 * that was blocked waiting for space. If the length
1121 * is non-zero and the ack didn't move, we're the
1122 * receiver side. If we're getting packets in-order
1123 * (the reassembly queue is empty), add the data to
1124 * the socket buffer and note that we need a delayed ack.
1125 * Make sure that the hidden state-flags are also off.
1126 * Since we check for TCPS_ESTABLISHED above, it can only
1127 * be TH_NEEDSYN.
1128 */
1129 if (tp->t_state == TCPS_ESTABLISHED &&
1130 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1131 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1132 ((to.to_flags & TOF_TS) == 0 ||
1133 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1134 /*
1135 * Using the CC option is compulsory if once started:
1136 * the segment is OK if no T/TCP was negotiated or
1137 * if the segment has a CC option equal to CCrecv
1138 */
1139 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1140 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1141 th->th_seq == tp->rcv_nxt &&
1142 tiwin && tiwin == tp->snd_wnd &&
1143 tp->snd_nxt == tp->snd_max) {
1144
1145 /*
1146 * If last ACK falls within this segment's sequence numbers,
1147 * record the timestamp.
1148 * NOTE that the test is modified according to the latest
1149 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1150 */
1151 if ((to.to_flags & TOF_TS) != 0 &&
1152 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1153 tp->ts_recent_age = ticks;
1154 tp->ts_recent = to.to_tsval;
1155 }
1156
1157 if (tlen == 0) {
1158 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1159 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1160 tp->snd_cwnd >= tp->snd_wnd &&
1161 ((!tcp_do_newreno && !tp->sack_enable &&
1162 tp->t_dupacks < tcprexmtthresh) ||
1163 ((tcp_do_newreno || tp->sack_enable) &&
1164 !IN_FASTRECOVERY(tp)))) {
1165 KASSERT(headlocked, ("headlocked"));
1166 INP_INFO_WUNLOCK(&tcbinfo);
1167 headlocked = 0;
1168 /*
1169 * this is a pure ack for outstanding data.
1170 */
1171 ++tcpstat.tcps_predack;
1172 /*
1173 * "bad retransmit" recovery
1174 */
1175 if (tp->t_rxtshift == 1 &&
1176 ticks < tp->t_badrxtwin) {
1177 ++tcpstat.tcps_sndrexmitbad;
1178 tp->snd_cwnd = tp->snd_cwnd_prev;
1179 tp->snd_ssthresh =
1180 tp->snd_ssthresh_prev;
1181 tp->snd_recover = tp->snd_recover_prev;
1182 if (tp->t_flags & TF_WASFRECOVERY)
1183 ENTER_FASTRECOVERY(tp);
1184 tp->snd_nxt = tp->snd_max;
1185 tp->t_badrxtwin = 0;
1186 }
1187
1188 /*
1189 * Recalculate the transmit timer / rtt.
1190 *
1191 * Some boxes send broken timestamp replies
1192 * during the SYN+ACK phase, ignore
1193 * timestamps of 0 or we could calculate a
1194 * huge RTT and blow up the retransmit timer.
1195 */
1196 if ((to.to_flags & TOF_TS) != 0 &&
1197 to.to_tsecr) {
1198 tcp_xmit_timer(tp,
1199 ticks - to.to_tsecr + 1);
1200 } else if (tp->t_rtttime &&
1201 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1202 tcp_xmit_timer(tp,
1203 ticks - tp->t_rtttime);
1204 }
1205 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1206 acked = th->th_ack - tp->snd_una;
1207 tcpstat.tcps_rcvackpack++;
1208 tcpstat.tcps_rcvackbyte += acked;
1209 sbdrop(&so->so_snd, acked);
1210 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1211 SEQ_LEQ(th->th_ack, tp->snd_recover))
1212 tp->snd_recover = th->th_ack - 1;
1213 tp->snd_una = th->th_ack;
1214 /*
1215 * pull snd_wl2 up to prevent seq wrap relative
1216 * to th_ack.
1217 */
1218 tp->snd_wl2 = th->th_ack;
1219 tp->t_dupacks = 0;
1220 m_freem(m);
1221 ND6_HINT(tp); /* some progress has been done */
1222
1223 /*
1224 * If all outstanding data are acked, stop
1225 * retransmit timer, otherwise restart timer
1226 * using current (possibly backed-off) value.
1227 * If process is waiting for space,
1228 * wakeup/selwakeup/signal. If data
1229 * are ready to send, let tcp_output
1230 * decide between more output or persist.
1231
1232 #ifdef TCPDEBUG
1233 if (so->so_options & SO_DEBUG)
1234 tcp_trace(TA_INPUT, ostate, tp,
1235 (void *)tcp_saveipgen,
1236 &tcp_savetcp, 0);
1237 #endif
1238 */
1239 if (tp->snd_una == tp->snd_max)
1240 callout_stop(tp->tt_rexmt);
1241 else if (!callout_active(tp->tt_persist))
1242 callout_reset(tp->tt_rexmt,
1243 tp->t_rxtcur,
1244 tcp_timer_rexmt, tp);
1245
1246 sowwakeup(so);
1247 if (so->so_snd.sb_cc)
1248 (void) tcp_output(tp);
1249 goto check_delack;
1250 }
1251 } else if (th->th_ack == tp->snd_una &&
1252 LIST_EMPTY(&tp->t_segq) &&
1253 tlen <= sbspace(&so->so_rcv)) {
1254 KASSERT(headlocked, ("headlocked"));
1255 INP_INFO_WUNLOCK(&tcbinfo);
1256 headlocked = 0;
1257 /*
1258 * this is a pure, in-sequence data packet
1259 * with nothing on the reassembly queue and
1260 * we have enough buffer space to take it.
1261 */
1262 /* Clean receiver SACK report if present */
1263 if (tp->sack_enable && tp->rcv_numsacks)
1264 tcp_clean_sackreport(tp);
1265 ++tcpstat.tcps_preddat;
1266 tp->rcv_nxt += tlen;
1267 /*
1268 * Pull snd_wl1 up to prevent seq wrap relative to
1269 * th_seq.
1270 */
1271 tp->snd_wl1 = th->th_seq;
1272 /*
1273 * Pull rcv_up up to prevent seq wrap relative to
1274 * rcv_nxt.
1275 */
1276 tp->rcv_up = tp->rcv_nxt;
1277 tcpstat.tcps_rcvpack++;
1278 tcpstat.tcps_rcvbyte += tlen;
1279 ND6_HINT(tp); /* some progress has been done */
1280 /*
1281 #ifdef TCPDEBUG
1282 if (so->so_options & SO_DEBUG)
1283 tcp_trace(TA_INPUT, ostate, tp,
1284 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1285 #endif
1286 * Add data to socket buffer.
1287 */
1288 SOCKBUF_LOCK(&so->so_rcv);
1289 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1290 m_freem(m);
1291 } else {
1292 m_adj(m, drop_hdrlen); /* delayed header drop */
1293 sbappendstream_locked(&so->so_rcv, m);
1294 }
1295 sorwakeup_locked(so);
1296 if (DELAY_ACK(tp)) {
1297 tp->t_flags |= TF_DELACK;
1298 } else {
1299 tp->t_flags |= TF_ACKNOW;
1300 tcp_output(tp);
1301 }
1302 goto check_delack;
1303 }
1304 }
1305
1306 /*
1307 * Calculate amount of space in receive window,
1308 * and then do TCP input processing.
1309 * Receive window is amount of space in rcv queue,
1310 * but not less than advertised window.
1311 */
1312 { int win;
1313
1314 win = sbspace(&so->so_rcv);
1315 if (win < 0)
1316 win = 0;
1317 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1318 }
1319
1320 switch (tp->t_state) {
1321
1322 /*
1323 * If the state is SYN_RECEIVED:
1324 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1325 */
1326 case TCPS_SYN_RECEIVED:
1327 if ((thflags & TH_ACK) &&
1328 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1329 SEQ_GT(th->th_ack, tp->snd_max))) {
1330 rstreason = BANDLIM_RST_OPENPORT;
1331 goto dropwithreset;
1332 }
1333 break;
1334
1335 /*
1336 * If the state is SYN_SENT:
1337 * if seg contains an ACK, but not for our SYN, drop the input.
1338 * if seg contains a RST, then drop the connection.
1339 * if seg does not contain SYN, then drop it.
1340 * Otherwise this is an acceptable SYN segment
1341 * initialize tp->rcv_nxt and tp->irs
1342 * if seg contains ack then advance tp->snd_una
1343 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1344 * arrange for segment to be acked (eventually)
1345 * continue processing rest of data/controls, beginning with URG
1346 */
1347 case TCPS_SYN_SENT:
1348 if (tcp_do_rfc1644)
1349 tcp_hc_gettao(&inp->inp_inc, &tao);
1350
1351 if ((thflags & TH_ACK) &&
1352 (SEQ_LEQ(th->th_ack, tp->iss) ||
1353 SEQ_GT(th->th_ack, tp->snd_max))) {
1354 /*
1355 * If we have a cached CCsent for the remote host,
1356 * hence we haven't just crashed and restarted,
1357 * do not send a RST. This may be a retransmission
1358 * from the other side after our earlier ACK was lost.
1359 * Our new SYN, when it arrives, will serve as the
1360 * needed ACK.
1361 */
1362 if (tao.tao_ccsent != 0)
1363 goto drop;
1364 else {
1365 rstreason = BANDLIM_UNLIMITED;
1366 goto dropwithreset;
1367 }
1368 }
1369 if (thflags & TH_RST) {
1370 if (thflags & TH_ACK)
1371 tp = tcp_drop(tp, ECONNREFUSED);
1372 goto drop;
1373 }
1374 if ((thflags & TH_SYN) == 0)
1375 goto drop;
1376 tp->snd_wnd = th->th_win; /* initial send window */
1377 tp->cc_recv = to.to_cc; /* foreign CC */
1378
1379 tp->irs = th->th_seq;
1380 tcp_rcvseqinit(tp);
1381 if (thflags & TH_ACK) {
1382 /*
1383 * Our SYN was acked. If segment contains CC.ECHO
1384 * option, check it to make sure this segment really
1385 * matches our SYN. If not, just drop it as old
1386 * duplicate, but send an RST if we're still playing
1387 * by the old rules. If no CC.ECHO option, make sure
1388 * we don't get fooled into using T/TCP.
1389 */
1390 if (to.to_flags & TOF_CCECHO) {
1391 if (tp->cc_send != to.to_ccecho) {
1392 if (tao.tao_ccsent != 0)
1393 goto drop;
1394 else {
1395 rstreason = BANDLIM_UNLIMITED;
1396 goto dropwithreset;
1397 }
1398 }
1399 } else
1400 tp->t_flags &= ~TF_RCVD_CC;
1401 tcpstat.tcps_connects++;
1402 soisconnected(so);
1403 #ifdef MAC
1404 SOCK_LOCK(so);
1405 mac_set_socket_peer_from_mbuf(m, so);
1406 SOCK_UNLOCK(so);
1407 #endif
1408 /* Do window scaling on this connection? */
1409 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1410 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1411 tp->snd_scale = tp->requested_s_scale;
1412 tp->rcv_scale = tp->request_r_scale;
1413 }
1414 /* Segment is acceptable, update cache if undefined. */
1415 if (tao.tao_ccsent == 0 && tcp_do_rfc1644)
1416 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0);
1417
1418 tp->rcv_adv += tp->rcv_wnd;
1419 tp->snd_una++; /* SYN is acked */
1420 /*
1421 * If there's data, delay ACK; if there's also a FIN
1422 * ACKNOW will be turned on later.
1423 */
1424 if (DELAY_ACK(tp) && tlen != 0)
1425 callout_reset(tp->tt_delack, tcp_delacktime,
1426 tcp_timer_delack, tp);
1427 else
1428 tp->t_flags |= TF_ACKNOW;
1429 /*
1430 * Received <SYN,ACK> in SYN_SENT[*] state.
1431 * Transitions:
1432 * SYN_SENT --> ESTABLISHED
1433 * SYN_SENT* --> FIN_WAIT_1
1434 */
1435 tp->t_starttime = ticks;
1436 if (tp->t_flags & TF_NEEDFIN) {
1437 tp->t_state = TCPS_FIN_WAIT_1;
1438 tp->t_flags &= ~TF_NEEDFIN;
1439 thflags &= ~TH_SYN;
1440 } else {
1441 tp->t_state = TCPS_ESTABLISHED;
1442 callout_reset(tp->tt_keep, tcp_keepidle,
1443 tcp_timer_keep, tp);
1444 }
1445 } else {
1446 /*
1447 * Received initial SYN in SYN-SENT[*] state =>
1448 * simultaneous open. If segment contains CC option
1449 * and there is a cached CC, apply TAO test.
1450 * If it succeeds, connection is * half-synchronized.
1451 * Otherwise, do 3-way handshake:
1452 * SYN-SENT -> SYN-RECEIVED
1453 * SYN-SENT* -> SYN-RECEIVED*
1454 * If there was no CC option, clear cached CC value.
1455 */
1456 tp->t_flags |= TF_ACKNOW;
1457 callout_stop(tp->tt_rexmt);
1458 if (to.to_flags & TOF_CC) {
1459 if (tao.tao_cc != 0 &&
1460 CC_GT(to.to_cc, tao.tao_cc)) {
1461 /*
1462 * update cache and make transition:
1463 * SYN-SENT -> ESTABLISHED*
1464 * SYN-SENT* -> FIN-WAIT-1*
1465 */
1466 tao.tao_cc = to.to_cc;
1467 tcp_hc_updatetao(&inp->inp_inc,
1468 TCP_HC_TAO_CC, to.to_cc, 0);
1469 tp->t_starttime = ticks;
1470 if (tp->t_flags & TF_NEEDFIN) {
1471 tp->t_state = TCPS_FIN_WAIT_1;
1472 tp->t_flags &= ~TF_NEEDFIN;
1473 } else {
1474 tp->t_state = TCPS_ESTABLISHED;
1475 callout_reset(tp->tt_keep,
1476 tcp_keepidle,
1477 tcp_timer_keep,
1478 tp);
1479 }
1480 tp->t_flags |= TF_NEEDSYN;
1481 } else
1482 tp->t_state = TCPS_SYN_RECEIVED;
1483 } else {
1484 if (tcp_do_rfc1644) {
1485 /* CC.NEW or no option => invalidate cache */
1486 tao.tao_cc = 0;
1487 tcp_hc_updatetao(&inp->inp_inc,
1488 TCP_HC_TAO_CC, to.to_cc, 0);
1489 }
1490 tp->t_state = TCPS_SYN_RECEIVED;
1491 }
1492 }
1493
1494 trimthenstep6:
1495 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1496 "locked"));
1497 INP_LOCK_ASSERT(inp);
1498
1499 /*
1500 * Advance th->th_seq to correspond to first data byte.
1501 * If data, trim to stay within window,
1502 * dropping FIN if necessary.
1503 */
1504 th->th_seq++;
1505 if (tlen > tp->rcv_wnd) {
1506 todrop = tlen - tp->rcv_wnd;
1507 m_adj(m, -todrop);
1508 tlen = tp->rcv_wnd;
1509 thflags &= ~TH_FIN;
1510 tcpstat.tcps_rcvpackafterwin++;
1511 tcpstat.tcps_rcvbyteafterwin += todrop;
1512 }
1513 tp->snd_wl1 = th->th_seq - 1;
1514 tp->rcv_up = th->th_seq;
1515 /*
1516 * Client side of transaction: already sent SYN and data.
1517 * If the remote host used T/TCP to validate the SYN,
1518 * our data will be ACK'd; if so, enter normal data segment
1519 * processing in the middle of step 5, ack processing.
1520 * Otherwise, goto step 6.
1521 */
1522 if (thflags & TH_ACK)
1523 goto process_ACK;
1524
1525 goto step6;
1526
1527 /*
1528 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1529 * if segment contains a SYN and CC [not CC.NEW] option:
1530 * if state == TIME_WAIT and connection duration > MSL,
1531 * drop packet and send RST;
1532 *
1533 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1534 * ack the FIN (and data) in retransmission queue.
1535 * Complete close and delete TCPCB. Then reprocess
1536 * segment, hoping to find new TCPCB in LISTEN state;
1537 *
1538 * else must be old SYN; drop it.
1539 * else do normal processing.
1540 */
1541 case TCPS_LAST_ACK:
1542 case TCPS_CLOSING:
1543 case TCPS_TIME_WAIT:
1544 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1545 if ((thflags & TH_SYN) &&
1546 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1547 if (tp->t_state == TCPS_TIME_WAIT &&
1548 (ticks - tp->t_starttime) > tcp_msl) {
1549 rstreason = BANDLIM_UNLIMITED;
1550 goto dropwithreset;
1551 }
1552 if (CC_GT(to.to_cc, tp->cc_recv)) {
1553 tp = tcp_close(tp);
1554 goto findpcb;
1555 }
1556 else
1557 goto drop;
1558 }
1559 break; /* continue normal processing */
1560 }
1561
1562 /*
1563 * States other than LISTEN or SYN_SENT.
1564 * First check the RST flag and sequence number since reset segments
1565 * are exempt from the timestamp and connection count tests. This
1566 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1567 * below which allowed reset segments in half the sequence space
1568 * to fall though and be processed (which gives forged reset
1569 * segments with a random sequence number a 50 percent chance of
1570 * killing a connection).
1571 * Then check timestamp, if present.
1572 * Then check the connection count, if present.
1573 * Then check that at least some bytes of segment are within
1574 * receive window. If segment begins before rcv_nxt,
1575 * drop leading data (and SYN); if nothing left, just ack.
1576 *
1577 *
1578 * If the RST bit is set, check the sequence number to see
1579 * if this is a valid reset segment.
1580 * RFC 793 page 37:
1581 * In all states except SYN-SENT, all reset (RST) segments
1582 * are validated by checking their SEQ-fields. A reset is
1583 * valid if its sequence number is in the window.
1584 * Note: this does not take into account delayed ACKs, so
1585 * we should test against last_ack_sent instead of rcv_nxt.
1586 * The sequence number in the reset segment is normally an
1587 * echo of our outgoing acknowlegement numbers, but some hosts
1588 * send a reset with the sequence number at the rightmost edge
1589 * of our receive window, and we have to handle this case.
1590 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1591 * that brute force RST attacks are possible. To combat this,
1592 * we use a much stricter check while in the ESTABLISHED state,
1593 * only accepting RSTs where the sequence number is equal to
1594 * last_ack_sent. In all other states (the states in which a
1595 * RST is more likely), the more permissive check is used.
1596 * If we have multiple segments in flight, the intial reset
1597 * segment sequence numbers will be to the left of last_ack_sent,
1598 * but they will eventually catch up.
1599 * In any case, it never made sense to trim reset segments to
1600 * fit the receive window since RFC 1122 says:
1601 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1602 *
1603 * A TCP SHOULD allow a received RST segment to include data.
1604 *
1605 * DISCUSSION
1606 * It has been suggested that a RST segment could contain
1607 * ASCII text that encoded and explained the cause of the
1608 * RST. No standard has yet been established for such
1609 * data.
1610 *
1611 * If the reset segment passes the sequence number test examine
1612 * the state:
1613 * SYN_RECEIVED STATE:
1614 * If passive open, return to LISTEN state.
1615 * If active open, inform user that connection was refused.
1616 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1617 * Inform user that connection was reset, and close tcb.
1618 * CLOSING, LAST_ACK STATES:
1619 * Close the tcb.
1620 * TIME_WAIT STATE:
1621 * Drop the segment - see Stevens, vol. 2, p. 964 and
1622 * RFC 1337.
1623 */
1624 if (thflags & TH_RST) {
1625 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1626 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1627 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1628 switch (tp->t_state) {
1629
1630 case TCPS_SYN_RECEIVED:
1631 so->so_error = ECONNREFUSED;
1632 goto close;
1633
1634 case TCPS_ESTABLISHED:
1635 if (tp->last_ack_sent != th->th_seq &&
1636 tcp_insecure_rst == 0) {
1637 tcpstat.tcps_badrst++;
1638 goto drop;
1639 }
1640 case TCPS_FIN_WAIT_1:
1641 case TCPS_FIN_WAIT_2:
1642 case TCPS_CLOSE_WAIT:
1643 so->so_error = ECONNRESET;
1644 close:
1645 tp->t_state = TCPS_CLOSED;
1646 tcpstat.tcps_drops++;
1647 tp = tcp_close(tp);
1648 break;
1649
1650 case TCPS_CLOSING:
1651 case TCPS_LAST_ACK:
1652 tp = tcp_close(tp);
1653 break;
1654
1655 case TCPS_TIME_WAIT:
1656 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1657 ("timewait"));
1658 break;
1659 }
1660 }
1661 goto drop;
1662 }
1663
1664 /*
1665 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1666 * and it's less than ts_recent, drop it.
1667 */
1668 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1669 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1670
1671 /* Check to see if ts_recent is over 24 days old. */
1672 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1673 /*
1674 * Invalidate ts_recent. If this segment updates
1675 * ts_recent, the age will be reset later and ts_recent
1676 * will get a valid value. If it does not, setting
1677 * ts_recent to zero will at least satisfy the
1678 * requirement that zero be placed in the timestamp
1679 * echo reply when ts_recent isn't valid. The
1680 * age isn't reset until we get a valid ts_recent
1681 * because we don't want out-of-order segments to be
1682 * dropped when ts_recent is old.
1683 */
1684 tp->ts_recent = 0;
1685 } else {
1686 tcpstat.tcps_rcvduppack++;
1687 tcpstat.tcps_rcvdupbyte += tlen;
1688 tcpstat.tcps_pawsdrop++;
1689 if (tlen)
1690 goto dropafterack;
1691 goto drop;
1692 }
1693 }
1694
1695 /*
1696 * T/TCP mechanism
1697 * If T/TCP was negotiated and the segment doesn't have CC,
1698 * or if its CC is wrong then drop the segment.
1699 * RST segments do not have to comply with this.
1700 */
1701 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1702 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1703 goto dropafterack;
1704
1705 /*
1706 * In the SYN-RECEIVED state, validate that the packet belongs to
1707 * this connection before trimming the data to fit the receive
1708 * window. Check the sequence number versus IRS since we know
1709 * the sequence numbers haven't wrapped. This is a partial fix
1710 * for the "LAND" DoS attack.
1711 */
1712 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1713 rstreason = BANDLIM_RST_OPENPORT;
1714 goto dropwithreset;
1715 }
1716
1717 todrop = tp->rcv_nxt - th->th_seq;
1718 if (todrop > 0) {
1719 if (thflags & TH_SYN) {
1720 thflags &= ~TH_SYN;
1721 th->th_seq++;
1722 if (th->th_urp > 1)
1723 th->th_urp--;
1724 else
1725 thflags &= ~TH_URG;
1726 todrop--;
1727 }
1728 /*
1729 * Following if statement from Stevens, vol. 2, p. 960.
1730 */
1731 if (todrop > tlen
1732 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1733 /*
1734 * Any valid FIN must be to the left of the window.
1735 * At this point the FIN must be a duplicate or out
1736 * of sequence; drop it.
1737 */
1738 thflags &= ~TH_FIN;
1739
1740 /*
1741 * Send an ACK to resynchronize and drop any data.
1742 * But keep on processing for RST or ACK.
1743 */
1744 tp->t_flags |= TF_ACKNOW;
1745 todrop = tlen;
1746 tcpstat.tcps_rcvduppack++;
1747 tcpstat.tcps_rcvdupbyte += todrop;
1748 } else {
1749 tcpstat.tcps_rcvpartduppack++;
1750 tcpstat.tcps_rcvpartdupbyte += todrop;
1751 }
1752 drop_hdrlen += todrop; /* drop from the top afterwards */
1753 th->th_seq += todrop;
1754 tlen -= todrop;
1755 if (th->th_urp > todrop)
1756 th->th_urp -= todrop;
1757 else {
1758 thflags &= ~TH_URG;
1759 th->th_urp = 0;
1760 }
1761 }
1762
1763 /*
1764 * If new data are received on a connection after the
1765 * user processes are gone, then RST the other end.
1766 */
1767 if ((so->so_state & SS_NOFDREF) &&
1768 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1769 tp = tcp_close(tp);
1770 tcpstat.tcps_rcvafterclose++;
1771 rstreason = BANDLIM_UNLIMITED;
1772 goto dropwithreset;
1773 }
1774
1775 /*
1776 * If segment ends after window, drop trailing data
1777 * (and PUSH and FIN); if nothing left, just ACK.
1778 */
1779 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1780 if (todrop > 0) {
1781 tcpstat.tcps_rcvpackafterwin++;
1782 if (todrop >= tlen) {
1783 tcpstat.tcps_rcvbyteafterwin += tlen;
1784 /*
1785 * If a new connection request is received
1786 * while in TIME_WAIT, drop the old connection
1787 * and start over if the sequence numbers
1788 * are above the previous ones.
1789 */
1790 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1791 if (thflags & TH_SYN &&
1792 tp->t_state == TCPS_TIME_WAIT &&
1793 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1794 tp = tcp_close(tp);
1795 goto findpcb;
1796 }
1797 /*
1798 * If window is closed can only take segments at
1799 * window edge, and have to drop data and PUSH from
1800 * incoming segments. Continue processing, but
1801 * remember to ack. Otherwise, drop segment
1802 * and ack.
1803 */
1804 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1805 tp->t_flags |= TF_ACKNOW;
1806 tcpstat.tcps_rcvwinprobe++;
1807 } else
1808 goto dropafterack;
1809 } else
1810 tcpstat.tcps_rcvbyteafterwin += todrop;
1811 m_adj(m, -todrop);
1812 tlen -= todrop;
1813 thflags &= ~(TH_PUSH|TH_FIN);
1814 }
1815
1816 /*
1817 * If last ACK falls within this segment's sequence numbers,
1818 * record its timestamp.
1819 * NOTE:
1820 * 1) That the test incorporates suggestions from the latest
1821 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1822 * 2) That updating only on newer timestamps interferes with
1823 * our earlier PAWS tests, so this check should be solely
1824 * predicated on the sequence space of this segment.
1825 * 3) That we modify the segment boundary check to be
1826 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1827 * instead of RFC1323's
1828 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1829 * This modified check allows us to overcome RFC1323's
1830 * limitations as described in Stevens TCP/IP Illustrated
1831 * Vol. 2 p.869. In such cases, we can still calculate the
1832 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1833 */
1834 if ((to.to_flags & TOF_TS) != 0 &&
1835 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1836 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1837 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1838 tp->ts_recent_age = ticks;
1839 tp->ts_recent = to.to_tsval;
1840 }
1841
1842 /*
1843 * If a SYN is in the window, then this is an
1844 * error and we send an RST and drop the connection.
1845 */
1846 if (thflags & TH_SYN) {
1847 tp = tcp_drop(tp, ECONNRESET);
1848 rstreason = BANDLIM_UNLIMITED;
1849 goto drop;
1850 }
1851
1852 /*
1853 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1854 * flag is on (half-synchronized state), then queue data for
1855 * later processing; else drop segment and return.
1856 */
1857 if ((thflags & TH_ACK) == 0) {
1858 if (tp->t_state == TCPS_SYN_RECEIVED ||
1859 (tp->t_flags & TF_NEEDSYN))
1860 goto step6;
1861 else
1862 goto drop;
1863 }
1864
1865 /*
1866 * Ack processing.
1867 */
1868 switch (tp->t_state) {
1869
1870 /*
1871 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1872 * ESTABLISHED state and continue processing.
1873 * The ACK was checked above.
1874 */
1875 case TCPS_SYN_RECEIVED:
1876
1877 tcpstat.tcps_connects++;
1878 soisconnected(so);
1879 /* Do window scaling? */
1880 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1881 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1882 tp->snd_scale = tp->requested_s_scale;
1883 tp->rcv_scale = tp->request_r_scale;
1884 }
1885 /*
1886 * Upon successful completion of 3-way handshake,
1887 * update cache.CC, pass any queued data to the user,
1888 * and advance state appropriately.
1889 */
1890 if (tcp_do_rfc1644) {
1891 tao.tao_cc = tp->cc_recv;
1892 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC,
1893 tp->cc_recv, 0);
1894 }
1895 /*
1896 * Make transitions:
1897 * SYN-RECEIVED -> ESTABLISHED
1898 * SYN-RECEIVED* -> FIN-WAIT-1
1899 */
1900 tp->t_starttime = ticks;
1901 if (tp->t_flags & TF_NEEDFIN) {
1902 tp->t_state = TCPS_FIN_WAIT_1;
1903 tp->t_flags &= ~TF_NEEDFIN;
1904 } else {
1905 tp->t_state = TCPS_ESTABLISHED;
1906 callout_reset(tp->tt_keep, tcp_keepidle,
1907 tcp_timer_keep, tp);
1908 }
1909 /*
1910 * If segment contains data or ACK, will call tcp_reass()
1911 * later; if not, do so now to pass queued data to user.
1912 */
1913 if (tlen == 0 && (thflags & TH_FIN) == 0)
1914 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1915 (struct mbuf *)0);
1916 tp->snd_wl1 = th->th_seq - 1;
1917 /* FALLTHROUGH */
1918
1919 /*
1920 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1921 * ACKs. If the ack is in the range
1922 * tp->snd_una < th->th_ack <= tp->snd_max
1923 * then advance tp->snd_una to th->th_ack and drop
1924 * data from the retransmission queue. If this ACK reflects
1925 * more up to date window information we update our window information.
1926 */
1927 case TCPS_ESTABLISHED:
1928 case TCPS_FIN_WAIT_1:
1929 case TCPS_FIN_WAIT_2:
1930 case TCPS_CLOSE_WAIT:
1931 case TCPS_CLOSING:
1932 case TCPS_LAST_ACK:
1933 case TCPS_TIME_WAIT:
1934 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1935 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1936 if (tlen == 0 && tiwin == tp->snd_wnd) {
1937 tcpstat.tcps_rcvdupack++;
1938 /*
1939 * If we have outstanding data (other than
1940 * a window probe), this is a completely
1941 * duplicate ack (ie, window info didn't
1942 * change), the ack is the biggest we've
1943 * seen and we've seen exactly our rexmt
1944 * threshhold of them, assume a packet
1945 * has been dropped and retransmit it.
1946 * Kludge snd_nxt & the congestion
1947 * window so we send only this one
1948 * packet.
1949 *
1950 * We know we're losing at the current
1951 * window size so do congestion avoidance
1952 * (set ssthresh to half the current window
1953 * and pull our congestion window back to
1954 * the new ssthresh).
1955 *
1956 * Dup acks mean that packets have left the
1957 * network (they're now cached at the receiver)
1958 * so bump cwnd by the amount in the receiver
1959 * to keep a constant cwnd packets in the
1960 * network.
1961 */
1962 if (!callout_active(tp->tt_rexmt) ||
1963 th->th_ack != tp->snd_una)
1964 tp->t_dupacks = 0;
1965 else if (++tp->t_dupacks > tcprexmtthresh ||
1966 ((tcp_do_newreno || tp->sack_enable) &&
1967 IN_FASTRECOVERY(tp))) {
1968 tp->snd_cwnd += tp->t_maxseg;
1969 (void) tcp_output(tp);
1970 goto drop;
1971 } else if (tp->t_dupacks == tcprexmtthresh) {
1972 tcp_seq onxt = tp->snd_nxt;
1973 u_int win;
1974
1975 /*
1976 * If we're doing sack, check to
1977 * see if we're already in sack
1978 * recovery. If we're not doing sack,
1979 * check to see if we're in newreno
1980 * recovery.
1981 */
1982 if (tp->sack_enable) {
1983 if (IN_FASTRECOVERY(tp)) {
1984 tp->t_dupacks = 0;
1985 break;
1986 }
1987 } else if (tcp_do_newreno) {
1988 if (SEQ_LEQ(th->th_ack,
1989 tp->snd_recover)) {
1990 tp->t_dupacks = 0;
1991 break;
1992 }
1993 }
1994 win = min(tp->snd_wnd, tp->snd_cwnd) /
1995 2 / tp->t_maxseg;
1996 if (win < 2)
1997 win = 2;
1998 tp->snd_ssthresh = win * tp->t_maxseg;
1999 ENTER_FASTRECOVERY(tp);
2000 tp->snd_recover = tp->snd_max;
2001 callout_stop(tp->tt_rexmt);
2002 tp->t_rtttime = 0;
2003 if (tp->sack_enable) {
2004 tcpstat.tcps_sack_recovery_episode++;
2005 tp->sack_newdata = tp->snd_nxt;
2006 tp->snd_cwnd = tp->t_maxseg;
2007 (void) tcp_output(tp);
2008 goto drop;
2009 }
2010 tp->snd_nxt = th->th_ack;
2011 tp->snd_cwnd = tp->t_maxseg;
2012 (void) tcp_output(tp);
2013 KASSERT(tp->snd_limited <= 2,
2014 ("tp->snd_limited too big"));
2015 tp->snd_cwnd = tp->snd_ssthresh +
2016 tp->t_maxseg *
2017 (tp->t_dupacks - tp->snd_limited);
2018 if (SEQ_GT(onxt, tp->snd_nxt))
2019 tp->snd_nxt = onxt;
2020 goto drop;
2021 } else if (tcp_do_rfc3042) {
2022 u_long oldcwnd = tp->snd_cwnd;
2023 tcp_seq oldsndmax = tp->snd_max;
2024 u_int sent;
2025
2026 KASSERT(tp->t_dupacks == 1 ||
2027 tp->t_dupacks == 2,
2028 ("dupacks not 1 or 2"));
2029 if (tp->t_dupacks == 1)
2030 tp->snd_limited = 0;
2031 tp->snd_cwnd =
2032 (tp->snd_nxt - tp->snd_una) +
2033 (tp->t_dupacks - tp->snd_limited) *
2034 tp->t_maxseg;
2035 (void) tcp_output(tp);
2036 sent = tp->snd_max - oldsndmax;
2037 if (sent > tp->t_maxseg) {
2038 KASSERT((tp->t_dupacks == 2 &&
2039 tp->snd_limited == 0) ||
2040 (sent == tp->t_maxseg + 1 &&
2041 tp->t_flags & TF_SENTFIN),
2042 ("sent too much"));
2043 tp->snd_limited = 2;
2044 } else if (sent > 0)
2045 ++tp->snd_limited;
2046 tp->snd_cwnd = oldcwnd;
2047 goto drop;
2048 }
2049 } else
2050 tp->t_dupacks = 0;
2051 break;
2052 }
2053
2054 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2055
2056 /*
2057 * If the congestion window was inflated to account
2058 * for the other side's cached packets, retract it.
2059 */
2060 if (tcp_do_newreno || tp->sack_enable) {
2061 if (IN_FASTRECOVERY(tp)) {
2062 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2063 if (tp->sack_enable)
2064 tcp_sack_partialack(tp, th);
2065 else
2066 tcp_newreno_partial_ack(tp, th);
2067 } else {
2068 /*
2069 * Out of fast recovery.
2070 * Window inflation should have left us
2071 * with approximately snd_ssthresh
2072 * outstanding data.
2073 * But in case we would be inclined to
2074 * send a burst, better to do it via
2075 * the slow start mechanism.
2076 */
2077 if (SEQ_GT(th->th_ack +
2078 tp->snd_ssthresh,
2079 tp->snd_max))
2080 tp->snd_cwnd = tp->snd_max -
2081 th->th_ack +
2082 tp->t_maxseg;
2083 else
2084 tp->snd_cwnd = tp->snd_ssthresh;
2085 }
2086 }
2087 } else {
2088 if (tp->t_dupacks >= tcprexmtthresh &&
2089 tp->snd_cwnd > tp->snd_ssthresh)
2090 tp->snd_cwnd = tp->snd_ssthresh;
2091 }
2092 tp->t_dupacks = 0;
2093 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2094 tcpstat.tcps_rcvacktoomuch++;
2095 goto dropafterack;
2096 }
2097 /*
2098 * If we reach this point, ACK is not a duplicate,
2099 * i.e., it ACKs something we sent.
2100 */
2101 if (tp->t_flags & TF_NEEDSYN) {
2102 /*
2103 * T/TCP: Connection was half-synchronized, and our
2104 * SYN has been ACK'd (so connection is now fully
2105 * synchronized). Go to non-starred state,
2106 * increment snd_una for ACK of SYN, and check if
2107 * we can do window scaling.
2108 */
2109 tp->t_flags &= ~TF_NEEDSYN;
2110 tp->snd_una++;
2111 /* Do window scaling? */
2112 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2113 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2114 tp->snd_scale = tp->requested_s_scale;
2115 tp->rcv_scale = tp->request_r_scale;
2116 }
2117 }
2118
2119 process_ACK:
2120 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2121 "locked"));
2122 INP_LOCK_ASSERT(inp);
2123
2124 acked = th->th_ack - tp->snd_una;
2125 tcpstat.tcps_rcvackpack++;
2126 tcpstat.tcps_rcvackbyte += acked;
2127
2128 /*
2129 * If we just performed our first retransmit, and the ACK
2130 * arrives within our recovery window, then it was a mistake
2131 * to do the retransmit in the first place. Recover our
2132 * original cwnd and ssthresh, and proceed to transmit where
2133 * we left off.
2134 */
2135 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2136 ++tcpstat.tcps_sndrexmitbad;
2137 tp->snd_cwnd = tp->snd_cwnd_prev;
2138 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2139 tp->snd_recover = tp->snd_recover_prev;
2140 if (tp->t_flags & TF_WASFRECOVERY)
2141 ENTER_FASTRECOVERY(tp);
2142 tp->snd_nxt = tp->snd_max;
2143 tp->t_badrxtwin = 0; /* XXX probably not required */
2144 }
2145
2146 /*
2147 * If we have a timestamp reply, update smoothed
2148 * round trip time. If no timestamp is present but
2149 * transmit timer is running and timed sequence
2150 * number was acked, update smoothed round trip time.
2151 * Since we now have an rtt measurement, cancel the
2152 * timer backoff (cf., Phil Karn's retransmit alg.).
2153 * Recompute the initial retransmit timer.
2154 *
2155 * Some boxes send broken timestamp replies
2156 * during the SYN+ACK phase, ignore
2157 * timestamps of 0 or we could calculate a
2158 * huge RTT and blow up the retransmit timer.
2159 */
2160 if ((to.to_flags & TOF_TS) != 0 &&
2161 to.to_tsecr) {
2162 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2163 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2164 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2165 }
2166 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2167
2168 /*
2169 * If all outstanding data is acked, stop retransmit
2170 * timer and remember to restart (more output or persist).
2171 * If there is more data to be acked, restart retransmit
2172 * timer, using current (possibly backed-off) value.
2173 */
2174 if (th->th_ack == tp->snd_max) {
2175 callout_stop(tp->tt_rexmt);
2176 needoutput = 1;
2177 } else if (!callout_active(tp->tt_persist))
2178 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2179 tcp_timer_rexmt, tp);
2180
2181 /*
2182 * If no data (only SYN) was ACK'd,
2183 * skip rest of ACK processing.
2184 */
2185 if (acked == 0)
2186 goto step6;
2187
2188 /*
2189 * When new data is acked, open the congestion window.
2190 * If the window gives us less than ssthresh packets
2191 * in flight, open exponentially (maxseg per packet).
2192 * Otherwise open linearly: maxseg per window
2193 * (maxseg^2 / cwnd per packet).
2194 */
2195 if ((!tcp_do_newreno && !tp->sack_enable) ||
2196 !IN_FASTRECOVERY(tp)) {
2197 register u_int cw = tp->snd_cwnd;
2198 register u_int incr = tp->t_maxseg;
2199 if (cw > tp->snd_ssthresh)
2200 incr = incr * incr / cw;
2201 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2202 }
2203 SOCKBUF_LOCK(&so->so_snd);
2204 if (acked > so->so_snd.sb_cc) {
2205 tp->snd_wnd -= so->so_snd.sb_cc;
2206 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2207 ourfinisacked = 1;
2208 } else {
2209 sbdrop_locked(&so->so_snd, acked);
2210 tp->snd_wnd -= acked;
2211 ourfinisacked = 0;
2212 }
2213 sowwakeup_locked(so);
2214 /* detect una wraparound */
2215 if ((tcp_do_newreno || tp->sack_enable) &&
2216 !IN_FASTRECOVERY(tp) &&
2217 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2218 SEQ_LEQ(th->th_ack, tp->snd_recover))
2219 tp->snd_recover = th->th_ack - 1;
2220 if ((tcp_do_newreno || tp->sack_enable) &&
2221 IN_FASTRECOVERY(tp) &&
2222 SEQ_GEQ(th->th_ack, tp->snd_recover))
2223 EXIT_FASTRECOVERY(tp);
2224 tp->snd_una = th->th_ack;
2225 if (tp->sack_enable) {
2226 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2227 tp->snd_recover = tp->snd_una;
2228 }
2229 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2230 tp->snd_nxt = tp->snd_una;
2231
2232 switch (tp->t_state) {
2233
2234 /*
2235 * In FIN_WAIT_1 STATE in addition to the processing
2236 * for the ESTABLISHED state if our FIN is now acknowledged
2237 * then enter FIN_WAIT_2.
2238 */
2239 case TCPS_FIN_WAIT_1:
2240 if (ourfinisacked) {
2241 /*
2242 * If we can't receive any more
2243 * data, then closing user can proceed.
2244 * Starting the timer is contrary to the
2245 * specification, but if we don't get a FIN
2246 * we'll hang forever.
2247 */
2248 /* XXXjl
2249 * we should release the tp also, and use a
2250 * compressed state.
2251 */
2252 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2253 soisdisconnected(so);
2254 callout_reset(tp->tt_2msl, tcp_maxidle,
2255 tcp_timer_2msl, tp);
2256 }
2257 tp->t_state = TCPS_FIN_WAIT_2;
2258 }
2259 break;
2260
2261 /*
2262 * In CLOSING STATE in addition to the processing for
2263 * the ESTABLISHED state if the ACK acknowledges our FIN
2264 * then enter the TIME-WAIT state, otherwise ignore
2265 * the segment.
2266 */
2267 case TCPS_CLOSING:
2268 if (ourfinisacked) {
2269 KASSERT(headlocked, ("tcp_input: process_ACK: "
2270 "head not locked"));
2271 tcp_twstart(tp);
2272 INP_INFO_WUNLOCK(&tcbinfo);
2273 m_freem(m);
2274 return;
2275 }
2276 break;
2277
2278 /*
2279 * In LAST_ACK, we may still be waiting for data to drain
2280 * and/or to be acked, as well as for the ack of our FIN.
2281 * If our FIN is now acknowledged, delete the TCB,
2282 * enter the closed state and return.
2283 */
2284 case TCPS_LAST_ACK:
2285 if (ourfinisacked) {
2286 tp = tcp_close(tp);
2287 goto drop;
2288 }
2289 break;
2290
2291 /*
2292 * In TIME_WAIT state the only thing that should arrive
2293 * is a retransmission of the remote FIN. Acknowledge
2294 * it and restart the finack timer.
2295 */
2296 case TCPS_TIME_WAIT:
2297 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2298 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2299 tcp_timer_2msl, tp);
2300 goto dropafterack;
2301 }
2302 }
2303
2304 step6:
2305 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2306 INP_LOCK_ASSERT(inp);
2307
2308 /*
2309 * Update window information.
2310 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2311 */
2312 if ((thflags & TH_ACK) &&
2313 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2314 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2315 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2316 /* keep track of pure window updates */
2317 if (tlen == 0 &&
2318 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2319 tcpstat.tcps_rcvwinupd++;
2320 tp->snd_wnd = tiwin;
2321 tp->snd_wl1 = th->th_seq;
2322 tp->snd_wl2 = th->th_ack;
2323 if (tp->snd_wnd > tp->max_sndwnd)
2324 tp->max_sndwnd = tp->snd_wnd;
2325 needoutput = 1;
2326 }
2327
2328 /*
2329 * Process segments with URG.
2330 */
2331 if ((thflags & TH_URG) && th->th_urp &&
2332 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2333 /*
2334 * This is a kludge, but if we receive and accept
2335 * random urgent pointers, we'll crash in
2336 * soreceive. It's hard to imagine someone
2337 * actually wanting to send this much urgent data.
2338 */
2339 SOCKBUF_LOCK(&so->so_rcv);
2340 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2341 th->th_urp = 0; /* XXX */
2342 thflags &= ~TH_URG; /* XXX */
2343 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2344 goto dodata; /* XXX */
2345 }
2346 /*
2347 * If this segment advances the known urgent pointer,
2348 * then mark the data stream. This should not happen
2349 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2350 * a FIN has been received from the remote side.
2351 * In these states we ignore the URG.
2352 *
2353 * According to RFC961 (Assigned Protocols),
2354 * the urgent pointer points to the last octet
2355 * of urgent data. We continue, however,
2356 * to consider it to indicate the first octet
2357 * of data past the urgent section as the original
2358 * spec states (in one of two places).
2359 */
2360 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2361 tp->rcv_up = th->th_seq + th->th_urp;
2362 so->so_oobmark = so->so_rcv.sb_cc +
2363 (tp->rcv_up - tp->rcv_nxt) - 1;
2364 if (so->so_oobmark == 0)
2365 so->so_rcv.sb_state |= SBS_RCVATMARK;
2366 sohasoutofband(so);
2367 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2368 }
2369 SOCKBUF_UNLOCK(&so->so_rcv);
2370 /*
2371 * Remove out of band data so doesn't get presented to user.
2372 * This can happen independent of advancing the URG pointer,
2373 * but if two URG's are pending at once, some out-of-band
2374 * data may creep in... ick.
2375 */
2376 if (th->th_urp <= (u_long)tlen &&
2377 !(so->so_options & SO_OOBINLINE)) {
2378 /* hdr drop is delayed */
2379 tcp_pulloutofband(so, th, m, drop_hdrlen);
2380 }
2381 } else {
2382 /*
2383 * If no out of band data is expected,
2384 * pull receive urgent pointer along
2385 * with the receive window.
2386 */
2387 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2388 tp->rcv_up = tp->rcv_nxt;
2389 }
2390 dodata: /* XXX */
2391 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2392 INP_LOCK_ASSERT(inp);
2393
2394 /*
2395 * Process the segment text, merging it into the TCP sequencing queue,
2396 * and arranging for acknowledgment of receipt if necessary.
2397 * This process logically involves adjusting tp->rcv_wnd as data
2398 * is presented to the user (this happens in tcp_usrreq.c,
2399 * case PRU_RCVD). If a FIN has already been received on this
2400 * connection then we just ignore the text.
2401 */
2402 if ((tlen || (thflags & TH_FIN)) &&
2403 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2404 m_adj(m, drop_hdrlen); /* delayed header drop */
2405 /*
2406 * Insert segment which includes th into TCP reassembly queue
2407 * with control block tp. Set thflags to whether reassembly now
2408 * includes a segment with FIN. This handles the common case
2409 * inline (segment is the next to be received on an established
2410 * connection, and the queue is empty), avoiding linkage into
2411 * and removal from the queue and repetition of various
2412 * conversions.
2413 * Set DELACK for segments received in order, but ack
2414 * immediately when segments are out of order (so
2415 * fast retransmit can work).
2416 */
2417 if (th->th_seq == tp->rcv_nxt &&
2418 LIST_EMPTY(&tp->t_segq) &&
2419 TCPS_HAVEESTABLISHED(tp->t_state)) {
2420 if (DELAY_ACK(tp))
2421 tp->t_flags |= TF_DELACK;
2422 else
2423 tp->t_flags |= TF_ACKNOW;
2424 tp->rcv_nxt += tlen;
2425 thflags = th->th_flags & TH_FIN;
2426 tcpstat.tcps_rcvpack++;
2427 tcpstat.tcps_rcvbyte += tlen;
2428 ND6_HINT(tp);
2429 SOCKBUF_LOCK(&so->so_rcv);
2430 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2431 m_freem(m);
2432 else
2433 sbappendstream_locked(&so->so_rcv, m);
2434 sorwakeup_locked(so);
2435 } else {
2436 thflags = tcp_reass(tp, th, &tlen, m);
2437 tp->t_flags |= TF_ACKNOW;
2438 }
2439 if (tp->sack_enable)
2440 tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen);
2441 /*
2442 * Note the amount of data that peer has sent into
2443 * our window, in order to estimate the sender's
2444 * buffer size.
2445 */
2446 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2447 } else {
2448 m_freem(m);
2449 thflags &= ~TH_FIN;
2450 }
2451
2452 /*
2453 * If FIN is received ACK the FIN and let the user know
2454 * that the connection is closing.
2455 */
2456 if (thflags & TH_FIN) {
2457 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2458 socantrcvmore(so);
2459 /*
2460 * If connection is half-synchronized
2461 * (ie NEEDSYN flag on) then delay ACK,
2462 * so it may be piggybacked when SYN is sent.
2463 * Otherwise, since we received a FIN then no
2464 * more input can be expected, send ACK now.
2465 */
2466 if (tp->t_flags & TF_NEEDSYN)
2467 tp->t_flags |= TF_DELACK;
2468 else
2469 tp->t_flags |= TF_ACKNOW;
2470 tp->rcv_nxt++;
2471 }
2472 switch (tp->t_state) {
2473
2474 /*
2475 * In SYN_RECEIVED and ESTABLISHED STATES
2476 * enter the CLOSE_WAIT state.
2477 */
2478 case TCPS_SYN_RECEIVED:
2479 tp->t_starttime = ticks;
2480 /*FALLTHROUGH*/
2481 case TCPS_ESTABLISHED:
2482 tp->t_state = TCPS_CLOSE_WAIT;
2483 break;
2484
2485 /*
2486 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2487 * enter the CLOSING state.
2488 */
2489 case TCPS_FIN_WAIT_1:
2490 tp->t_state = TCPS_CLOSING;
2491 break;
2492
2493 /*
2494 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2495 * starting the time-wait timer, turning off the other
2496 * standard timers.
2497 */
2498 case TCPS_FIN_WAIT_2:
2499 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2500 "TCP_FIN_WAIT_2: head not locked"));
2501 tcp_twstart(tp);
2502 INP_INFO_WUNLOCK(&tcbinfo);
2503 return;
2504
2505 /*
2506 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2507 */
2508 case TCPS_TIME_WAIT:
2509 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2510 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2511 tcp_timer_2msl, tp);
2512 break;
2513 }
2514 }
2515 INP_INFO_WUNLOCK(&tcbinfo);
2516 headlocked = 0;
2517 #ifdef TCPDEBUG
2518 if (so->so_options & SO_DEBUG)
2519 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2520 &tcp_savetcp, 0);
2521 #endif
2522
2523 /*
2524 * Return any desired output.
2525 */
2526 if (needoutput || (tp->t_flags & TF_ACKNOW))
2527 (void) tcp_output(tp);
2528
2529 check_delack:
2530 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2531 INP_LOCK_ASSERT(inp);
2532 if (tp->t_flags & TF_DELACK) {
2533 tp->t_flags &= ~TF_DELACK;
2534 callout_reset(tp->tt_delack, tcp_delacktime,
2535 tcp_timer_delack, tp);
2536 }
2537 INP_UNLOCK(inp);
2538 return;
2539
2540 dropafterack:
2541 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2542 /*
2543 * Generate an ACK dropping incoming segment if it occupies
2544 * sequence space, where the ACK reflects our state.
2545 *
2546 * We can now skip the test for the RST flag since all
2547 * paths to this code happen after packets containing
2548 * RST have been dropped.
2549 *
2550 * In the SYN-RECEIVED state, don't send an ACK unless the
2551 * segment we received passes the SYN-RECEIVED ACK test.
2552 * If it fails send a RST. This breaks the loop in the
2553 * "LAND" DoS attack, and also prevents an ACK storm
2554 * between two listening ports that have been sent forged
2555 * SYN segments, each with the source address of the other.
2556 */
2557 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2558 (SEQ_GT(tp->snd_una, th->th_ack) ||
2559 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2560 rstreason = BANDLIM_RST_OPENPORT;
2561 goto dropwithreset;
2562 }
2563 #ifdef TCPDEBUG
2564 if (so->so_options & SO_DEBUG)
2565 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2566 &tcp_savetcp, 0);
2567 #endif
2568 KASSERT(headlocked, ("headlocked should be 1"));
2569 INP_INFO_WUNLOCK(&tcbinfo);
2570 tp->t_flags |= TF_ACKNOW;
2571 (void) tcp_output(tp);
2572 INP_UNLOCK(inp);
2573 m_freem(m);
2574 return;
2575
2576 dropwithreset:
2577 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2578 /*
2579 * Generate a RST, dropping incoming segment.
2580 * Make ACK acceptable to originator of segment.
2581 * Don't bother to respond if destination was broadcast/multicast.
2582 */
2583 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2584 goto drop;
2585 if (isipv6) {
2586 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2587 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2588 goto drop;
2589 } else {
2590 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2591 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2592 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2593 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2594 goto drop;
2595 }
2596 /* IPv6 anycast check is done at tcp6_input() */
2597
2598 /*
2599 * Perform bandwidth limiting.
2600 */
2601 if (badport_bandlim(rstreason) < 0)
2602 goto drop;
2603
2604 #ifdef TCPDEBUG
2605 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2606 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2607 &tcp_savetcp, 0);
2608 #endif
2609
2610 if (thflags & TH_ACK)
2611 /* mtod() below is safe as long as hdr dropping is delayed */
2612 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2613 TH_RST);
2614 else {
2615 if (thflags & TH_SYN)
2616 tlen++;
2617 /* mtod() below is safe as long as hdr dropping is delayed */
2618 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2619 (tcp_seq)0, TH_RST|TH_ACK);
2620 }
2621
2622 if (tp)
2623 INP_UNLOCK(inp);
2624 if (headlocked)
2625 INP_INFO_WUNLOCK(&tcbinfo);
2626 return;
2627
2628 drop:
2629 /*
2630 * Drop space held by incoming segment and return.
2631 */
2632 #ifdef TCPDEBUG
2633 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2634 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2635 &tcp_savetcp, 0);
2636 #endif
2637 if (tp)
2638 INP_UNLOCK(inp);
2639 if (headlocked)
2640 INP_INFO_WUNLOCK(&tcbinfo);
2641 m_freem(m);
2642 return;
2643 }
2644
2645 /*
2646 * Parse TCP options and place in tcpopt.
2647 */
2648 static void
2649 tcp_dooptions(tp, to, cp, cnt, is_syn, th)
2650 struct tcpcb *tp;
2651 struct tcpopt *to;
2652 u_char *cp;
2653 int cnt;
2654 int is_syn;
2655 struct tcphdr *th;
2656 {
2657 int opt, optlen;
2658
2659 to->to_flags = 0;
2660 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2661 opt = cp[0];
2662 if (opt == TCPOPT_EOL)
2663 break;
2664 if (opt == TCPOPT_NOP)
2665 optlen = 1;
2666 else {
2667 if (cnt < 2)
2668 break;
2669 optlen = cp[1];
2670 if (optlen < 2 || optlen > cnt)
2671 break;
2672 }
2673 switch (opt) {
2674 case TCPOPT_MAXSEG:
2675 if (optlen != TCPOLEN_MAXSEG)
2676 continue;
2677 if (!is_syn)
2678 continue;
2679 to->to_flags |= TOF_MSS;
2680 bcopy((char *)cp + 2,
2681 (char *)&to->to_mss, sizeof(to->to_mss));
2682 to->to_mss = ntohs(to->to_mss);
2683 break;
2684 case TCPOPT_WINDOW:
2685 if (optlen != TCPOLEN_WINDOW)
2686 continue;
2687 if (! is_syn)
2688 continue;
2689 to->to_flags |= TOF_SCALE;
2690 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2691 break;
2692 case TCPOPT_TIMESTAMP:
2693 if (optlen != TCPOLEN_TIMESTAMP)
2694 continue;
2695 to->to_flags |= TOF_TS;
2696 bcopy((char *)cp + 2,
2697 (char *)&to->to_tsval, sizeof(to->to_tsval));
2698 to->to_tsval = ntohl(to->to_tsval);
2699 bcopy((char *)cp + 6,
2700 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2701 to->to_tsecr = ntohl(to->to_tsecr);
2702 /*
2703 * If echoed timestamp is later than the current time,
2704 * fall back to non RFC1323 RTT calculation.
2705 */
2706 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2707 to->to_tsecr = 0;
2708 break;
2709 case TCPOPT_CC:
2710 if (optlen != TCPOLEN_CC)
2711 continue;
2712 to->to_flags |= TOF_CC;
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_CCNEW:
2718 if (optlen != TCPOLEN_CC)
2719 continue;
2720 if (!is_syn)
2721 continue;
2722 to->to_flags |= TOF_CCNEW;
2723 bcopy((char *)cp + 2,
2724 (char *)&to->to_cc, sizeof(to->to_cc));
2725 to->to_cc = ntohl(to->to_cc);
2726 break;
2727 case TCPOPT_CCECHO:
2728 if (optlen != TCPOLEN_CC)
2729 continue;
2730 if (!is_syn)
2731 continue;
2732 to->to_flags |= TOF_CCECHO;
2733 bcopy((char *)cp + 2,
2734 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2735 to->to_ccecho = ntohl(to->to_ccecho);
2736 break;
2737 #ifdef TCP_SIGNATURE
2738 /*
2739 * XXX In order to reply to a host which has set the
2740 * TCP_SIGNATURE option in its initial SYN, we have to
2741 * record the fact that the option was observed here
2742 * for the syncache code to perform the correct response.
2743 */
2744 case TCPOPT_SIGNATURE:
2745 if (optlen != TCPOLEN_SIGNATURE)
2746 continue;
2747 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2748 break;
2749 #endif
2750 case TCPOPT_SACK_PERMITTED:
2751 if (!tcp_do_sack ||
2752 optlen != TCPOLEN_SACK_PERMITTED)
2753 continue;
2754 if (is_syn) {
2755 /* MUST only be set on SYN */
2756 to->to_flags |= TOF_SACK;
2757 }
2758 break;
2759
2760 case TCPOPT_SACK:
2761 if (!tp || tcp_sack_option(tp, th, cp, optlen))
2762 continue;
2763 break;
2764 default:
2765 continue;
2766 }
2767 }
2768 }
2769
2770 /*
2771 * Pull out of band byte out of a segment so
2772 * it doesn't appear in the user's data queue.
2773 * It is still reflected in the segment length for
2774 * sequencing purposes.
2775 */
2776 static void
2777 tcp_pulloutofband(so, th, m, off)
2778 struct socket *so;
2779 struct tcphdr *th;
2780 register struct mbuf *m;
2781 int off; /* delayed to be droped hdrlen */
2782 {
2783 int cnt = off + th->th_urp - 1;
2784
2785 while (cnt >= 0) {
2786 if (m->m_len > cnt) {
2787 char *cp = mtod(m, caddr_t) + cnt;
2788 struct tcpcb *tp = sototcpcb(so);
2789
2790 tp->t_iobc = *cp;
2791 tp->t_oobflags |= TCPOOB_HAVEDATA;
2792 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2793 m->m_len--;
2794 if (m->m_flags & M_PKTHDR)
2795 m->m_pkthdr.len--;
2796 return;
2797 }
2798 cnt -= m->m_len;
2799 m = m->m_next;
2800 if (m == 0)
2801 break;
2802 }
2803 panic("tcp_pulloutofband");
2804 }
2805
2806 /*
2807 * Collect new round-trip time estimate
2808 * and update averages and current timeout.
2809 */
2810 static void
2811 tcp_xmit_timer(tp, rtt)
2812 register struct tcpcb *tp;
2813 int rtt;
2814 {
2815 register int delta;
2816
2817 INP_LOCK_ASSERT(tp->t_inpcb);
2818
2819 tcpstat.tcps_rttupdated++;
2820 tp->t_rttupdated++;
2821 if (tp->t_srtt != 0) {
2822 /*
2823 * srtt is stored as fixed point with 5 bits after the
2824 * binary point (i.e., scaled by 8). The following magic
2825 * is equivalent to the smoothing algorithm in rfc793 with
2826 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2827 * point). Adjust rtt to origin 0.
2828 */
2829 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2830 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2831
2832 if ((tp->t_srtt += delta) <= 0)
2833 tp->t_srtt = 1;
2834
2835 /*
2836 * We accumulate a smoothed rtt variance (actually, a
2837 * smoothed mean difference), then set the retransmit
2838 * timer to smoothed rtt + 4 times the smoothed variance.
2839 * rttvar is stored as fixed point with 4 bits after the
2840 * binary point (scaled by 16). The following is
2841 * equivalent to rfc793 smoothing with an alpha of .75
2842 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2843 * rfc793's wired-in beta.
2844 */
2845 if (delta < 0)
2846 delta = -delta;
2847 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2848 if ((tp->t_rttvar += delta) <= 0)
2849 tp->t_rttvar = 1;
2850 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2851 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2852 } else {
2853 /*
2854 * No rtt measurement yet - use the unsmoothed rtt.
2855 * Set the variance to half the rtt (so our first
2856 * retransmit happens at 3*rtt).
2857 */
2858 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2859 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2860 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2861 }
2862 tp->t_rtttime = 0;
2863 tp->t_rxtshift = 0;
2864
2865 /*
2866 * the retransmit should happen at rtt + 4 * rttvar.
2867 * Because of the way we do the smoothing, srtt and rttvar
2868 * will each average +1/2 tick of bias. When we compute
2869 * the retransmit timer, we want 1/2 tick of rounding and
2870 * 1 extra tick because of +-1/2 tick uncertainty in the
2871 * firing of the timer. The bias will give us exactly the
2872 * 1.5 tick we need. But, because the bias is
2873 * statistical, we have to test that we don't drop below
2874 * the minimum feasible timer (which is 2 ticks).
2875 */
2876 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2877 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2878
2879 /*
2880 * We received an ack for a packet that wasn't retransmitted;
2881 * it is probably safe to discard any error indications we've
2882 * received recently. This isn't quite right, but close enough
2883 * for now (a route might have failed after we sent a segment,
2884 * and the return path might not be symmetrical).
2885 */
2886 tp->t_softerror = 0;
2887 }
2888
2889 /*
2890 * Determine a reasonable value for maxseg size.
2891 * If the route is known, check route for mtu.
2892 * If none, use an mss that can be handled on the outgoing
2893 * interface without forcing IP to fragment; if bigger than
2894 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2895 * to utilize large mbufs. If no route is found, route has no mtu,
2896 * or the destination isn't local, use a default, hopefully conservative
2897 * size (usually 512 or the default IP max size, but no more than the mtu
2898 * of the interface), as we can't discover anything about intervening
2899 * gateways or networks. We also initialize the congestion/slow start
2900 * window to be a single segment if the destination isn't local.
2901 * While looking at the routing entry, we also initialize other path-dependent
2902 * parameters from pre-set or cached values in the routing entry.
2903 *
2904 * Also take into account the space needed for options that we
2905 * send regularly. Make maxseg shorter by that amount to assure
2906 * that we can send maxseg amount of data even when the options
2907 * are present. Store the upper limit of the length of options plus
2908 * data in maxopd.
2909 *
2910 *
2911 * In case of T/TCP, we call this routine during implicit connection
2912 * setup as well (offer = -1), to initialize maxseg from the cached
2913 * MSS of our peer.
2914 *
2915 * NOTE that this routine is only called when we process an incoming
2916 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2917 */
2918 void
2919 tcp_mss(tp, offer)
2920 struct tcpcb *tp;
2921 int offer;
2922 {
2923 int rtt, mss;
2924 u_long bufsize;
2925 u_long maxmtu;
2926 struct inpcb *inp = tp->t_inpcb;
2927 struct socket *so;
2928 struct hc_metrics_lite metrics;
2929 struct rmxp_tao tao;
2930 int origoffer = offer;
2931 #ifdef INET6
2932 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2933 size_t min_protoh = isipv6 ?
2934 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2935 sizeof (struct tcpiphdr);
2936 #else
2937 const size_t min_protoh = sizeof(struct tcpiphdr);
2938 #endif
2939 bzero(&tao, sizeof(tao));
2940
2941 /* initialize */
2942 #ifdef INET6
2943 if (isipv6) {
2944 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2945 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2946 } else
2947 #endif
2948 {
2949 maxmtu = tcp_maxmtu(&inp->inp_inc);
2950 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2951 }
2952 so = inp->inp_socket;
2953
2954 /*
2955 * no route to sender, stay with default mss and return
2956 */
2957 if (maxmtu == 0)
2958 return;
2959
2960 /* what have we got? */
2961 switch (offer) {
2962 case 0:
2963 /*
2964 * Offer == 0 means that there was no MSS on the SYN
2965 * segment, in this case we use tcp_mssdflt.
2966 */
2967 offer =
2968 #ifdef INET6
2969 isipv6 ? tcp_v6mssdflt :
2970 #endif
2971 tcp_mssdflt;
2972 break;
2973
2974 case -1:
2975 /*
2976 * Offer == -1 means that we didn't receive SYN yet,
2977 * use cached value in that case;
2978 */
2979 if (tcp_do_rfc1644)
2980 tcp_hc_gettao(&inp->inp_inc, &tao);
2981 if (tao.tao_mssopt != 0)
2982 offer = tao.tao_mssopt;
2983 /* FALLTHROUGH */
2984
2985 default:
2986 /*
2987 * Prevent DoS attack with too small MSS. Round up
2988 * to at least minmss.
2989 */
2990 offer = max(offer, tcp_minmss);
2991 /*
2992 * Sanity check: make sure that maxopd will be large
2993 * enough to allow some data on segments even if the
2994 * all the option space is used (40bytes). Otherwise
2995 * funny things may happen in tcp_output.
2996 */
2997 offer = max(offer, 64);
2998 if (tcp_do_rfc1644)
2999 tcp_hc_updatetao(&inp->inp_inc,
3000 TCP_HC_TAO_MSSOPT, 0, offer);
3001 }
3002
3003 /*
3004 * rmx information is now retrieved from tcp_hostcache
3005 */
3006 tcp_hc_get(&inp->inp_inc, &metrics);
3007
3008 /*
3009 * if there's a discovered mtu int tcp hostcache, use it
3010 * else, use the link mtu.
3011 */
3012 if (metrics.rmx_mtu)
3013 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3014 else {
3015 #ifdef INET6
3016 if (isipv6) {
3017 mss = maxmtu - min_protoh;
3018 if (!path_mtu_discovery &&
3019 !in6_localaddr(&inp->in6p_faddr))
3020 mss = min(mss, tcp_v6mssdflt);
3021 } else
3022 #endif
3023 {
3024 mss = maxmtu - min_protoh;
3025 if (!path_mtu_discovery &&
3026 !in_localaddr(inp->inp_faddr))
3027 mss = min(mss, tcp_mssdflt);
3028 }
3029 }
3030 mss = min(mss, offer);
3031
3032 /*
3033 * maxopd stores the maximum length of data AND options
3034 * in a segment; maxseg is the amount of data in a normal
3035 * segment. We need to store this value (maxopd) apart
3036 * from maxseg, because now every segment carries options
3037 * and thus we normally have somewhat less data in segments.
3038 */
3039 tp->t_maxopd = mss;
3040
3041 /*
3042 * In case of T/TCP, origoffer==-1 indicates, that no segments
3043 * were received yet. In this case we just guess, otherwise
3044 * we do the same as before T/TCP.
3045 */
3046 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3047 (origoffer == -1 ||
3048 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3049 mss -= TCPOLEN_TSTAMP_APPA;
3050 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
3051 (origoffer == -1 ||
3052 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3053 mss -= TCPOLEN_CC_APPA;
3054 tp->t_maxseg = mss;
3055
3056 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3057 if (mss > MCLBYTES)
3058 mss &= ~(MCLBYTES-1);
3059 #else
3060 if (mss > MCLBYTES)
3061 mss = mss / MCLBYTES * MCLBYTES;
3062 #endif
3063 tp->t_maxseg = mss;
3064
3065 /*
3066 * If there's a pipesize, change the socket buffer to that size,
3067 * don't change if sb_hiwat is different than default (then it
3068 * has been changed on purpose with setsockopt).
3069 * Make the socket buffers an integral number of mss units;
3070 * if the mss is larger than the socket buffer, decrease the mss.
3071 */
3072 SOCKBUF_LOCK(&so->so_snd);
3073 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3074 bufsize = metrics.rmx_sendpipe;
3075 else
3076 bufsize = so->so_snd.sb_hiwat;
3077 if (bufsize < mss)
3078 mss = bufsize;
3079 else {
3080 bufsize = roundup(bufsize, mss);
3081 if (bufsize > sb_max)
3082 bufsize = sb_max;
3083 if (bufsize > so->so_snd.sb_hiwat)
3084 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3085 }
3086 SOCKBUF_UNLOCK(&so->so_snd);
3087 tp->t_maxseg = mss;
3088
3089 SOCKBUF_LOCK(&so->so_rcv);
3090 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3091 bufsize = metrics.rmx_recvpipe;
3092 else
3093 bufsize = so->so_rcv.sb_hiwat;
3094 if (bufsize > mss) {
3095 bufsize = roundup(bufsize, mss);
3096 if (bufsize > sb_max)
3097 bufsize = sb_max;
3098 if (bufsize > so->so_rcv.sb_hiwat)
3099 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3100 }
3101 SOCKBUF_UNLOCK(&so->so_rcv);
3102 /*
3103 * While we're here, check the others too
3104 */
3105 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3106 tp->t_srtt = rtt;
3107 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3108 tcpstat.tcps_usedrtt++;
3109 if (metrics.rmx_rttvar) {
3110 tp->t_rttvar = metrics.rmx_rttvar;
3111 tcpstat.tcps_usedrttvar++;
3112 } else {
3113 /* default variation is +- 1 rtt */
3114 tp->t_rttvar =
3115 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3116 }
3117 TCPT_RANGESET(tp->t_rxtcur,
3118 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3119 tp->t_rttmin, TCPTV_REXMTMAX);
3120 }
3121 if (metrics.rmx_ssthresh) {
3122 /*
3123 * There's some sort of gateway or interface
3124 * buffer limit on the path. Use this to set
3125 * the slow start threshhold, but set the
3126 * threshold to no less than 2*mss.
3127 */
3128 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3129 tcpstat.tcps_usedssthresh++;
3130 }
3131 if (metrics.rmx_bandwidth)
3132 tp->snd_bandwidth = metrics.rmx_bandwidth;
3133
3134 /*
3135 * Set the slow-start flight size depending on whether this
3136 * is a local network or not.
3137 *
3138 * Extend this so we cache the cwnd too and retrieve it here.
3139 * Make cwnd even bigger than RFC3390 suggests but only if we
3140 * have previous experience with the remote host. Be careful
3141 * not make cwnd bigger than remote receive window or our own
3142 * send socket buffer. Maybe put some additional upper bound
3143 * on the retrieved cwnd. Should do incremental updates to
3144 * hostcache when cwnd collapses so next connection doesn't
3145 * overloads the path again.
3146 *
3147 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3148 * We currently check only in syncache_socket for that.
3149 */
3150 #define TCP_METRICS_CWND
3151 #ifdef TCP_METRICS_CWND
3152 if (metrics.rmx_cwnd)
3153 tp->snd_cwnd = max(mss,
3154 min(metrics.rmx_cwnd / 2,
3155 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3156 else
3157 #endif
3158 if (tcp_do_rfc3390)
3159 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3160 #ifdef INET6
3161 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3162 (!isipv6 && in_localaddr(inp->inp_faddr)))
3163 #else
3164 else if (in_localaddr(inp->inp_faddr))
3165 #endif
3166 tp->snd_cwnd = mss * ss_fltsz_local;
3167 else
3168 tp->snd_cwnd = mss * ss_fltsz;
3169 }
3170
3171 /*
3172 * Determine the MSS option to send on an outgoing SYN.
3173 */
3174 int
3175 tcp_mssopt(inc)
3176 struct in_conninfo *inc;
3177 {
3178 int mss = 0;
3179 u_long maxmtu = 0;
3180 u_long thcmtu = 0;
3181 size_t min_protoh;
3182 #ifdef INET6
3183 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3184 #endif
3185
3186 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3187
3188 #ifdef INET6
3189 if (isipv6) {
3190 mss = tcp_v6mssdflt;
3191 maxmtu = tcp_maxmtu6(inc);
3192 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3193 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3194 } else
3195 #endif
3196 {
3197 mss = tcp_mssdflt;
3198 maxmtu = tcp_maxmtu(inc);
3199 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3200 min_protoh = sizeof(struct tcpiphdr);
3201 }
3202 if (maxmtu && thcmtu)
3203 mss = min(maxmtu, thcmtu) - min_protoh;
3204 else if (maxmtu || thcmtu)
3205 mss = max(maxmtu, thcmtu) - min_protoh;
3206
3207 return (mss);
3208 }
3209
3210
3211 /*
3212 * On a partial ack arrives, force the retransmission of the
3213 * next unacknowledged segment. Do not clear tp->t_dupacks.
3214 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3215 * be started again.
3216 */
3217 static void
3218 tcp_newreno_partial_ack(tp, th)
3219 struct tcpcb *tp;
3220 struct tcphdr *th;
3221 {
3222 tcp_seq onxt = tp->snd_nxt;
3223 u_long ocwnd = tp->snd_cwnd;
3224
3225 callout_stop(tp->tt_rexmt);
3226 tp->t_rtttime = 0;
3227 tp->snd_nxt = th->th_ack;
3228 /*
3229 * Set snd_cwnd to one segment beyond acknowledged offset.
3230 * (tp->snd_una has not yet been updated when this function is called.)
3231 */
3232 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3233 tp->t_flags |= TF_ACKNOW;
3234 (void) tcp_output(tp);
3235 tp->snd_cwnd = ocwnd;
3236 if (SEQ_GT(onxt, tp->snd_nxt))
3237 tp->snd_nxt = onxt;
3238 /*
3239 * Partial window deflation. Relies on fact that tp->snd_una
3240 * not updated yet.
3241 */
3242 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3243 }
3244
3245 /*
3246 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3247 * looking for a pcb in the listen state. Returns 0 otherwise.
3248 */
3249 static int
3250 tcp_timewait(tw, to, th, m, tlen)
3251 struct tcptw *tw;
3252 struct tcpopt *to;
3253 struct tcphdr *th;
3254 struct mbuf *m;
3255 int tlen;
3256 {
3257 int thflags;
3258 tcp_seq seq;
3259 #ifdef INET6
3260 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3261 #else
3262 const int isipv6 = 0;
3263 #endif
3264
3265 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */
3266 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3267 INP_LOCK_ASSERT(tw->tw_inpcb);
3268
3269 thflags = th->th_flags;
3270
3271 /*
3272 * NOTE: for FIN_WAIT_2 (to be added later),
3273 * must validate sequence number before accepting RST
3274 */
3275
3276 /*
3277 * If the segment contains RST:
3278 * Drop the segment - see Stevens, vol. 2, p. 964 and
3279 * RFC 1337.
3280 */
3281 if (thflags & TH_RST)
3282 goto drop;
3283
3284 /*
3285 * If segment contains a SYN and CC [not CC.NEW] option:
3286 * if connection duration > MSL, drop packet and send RST;
3287 *
3288 * if SEG.CC > CCrecv then is new SYN.
3289 * Complete close and delete TCPCB. Then reprocess
3290 * segment, hoping to find new TCPCB in LISTEN state;
3291 *
3292 * else must be old SYN; drop it.
3293 * else do normal processing.
3294 */
3295 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) {
3296 if ((ticks - tw->t_starttime) > tcp_msl)
3297 goto reset;
3298 if (CC_GT(to->to_cc, tw->cc_recv)) {
3299 (void) tcp_twclose(tw, 0);
3300 return (1);
3301 }
3302 goto drop;
3303 }
3304
3305 #if 0
3306 /* PAWS not needed at the moment */
3307 /*
3308 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3309 * and it's less than ts_recent, drop it.
3310 */
3311 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3312 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3313 if ((thflags & TH_ACK) == 0)
3314 goto drop;
3315 goto ack;
3316 }
3317 /*
3318 * ts_recent is never updated because we never accept new segments.
3319 */
3320 #endif
3321
3322 /*
3323 * If a new connection request is received
3324 * while in TIME_WAIT, drop the old connection
3325 * and start over if the sequence numbers
3326 * are above the previous ones.
3327 */
3328 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3329 (void) tcp_twclose(tw, 0);
3330 return (1);
3331 }
3332
3333 /*
3334 * Drop the the segment if it does not contain an ACK.
3335 */
3336 if ((thflags & TH_ACK) == 0)
3337 goto drop;
3338
3339 /*
3340 * Reset the 2MSL timer if this is a duplicate FIN.
3341 */
3342 if (thflags & TH_FIN) {
3343 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3344 if (seq + 1 == tw->rcv_nxt)
3345 tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3346 }
3347
3348 /*
3349 * Acknowledge the segment if it has data or is not a duplicate ACK.
3350 */
3351 if (thflags != TH_ACK || tlen != 0 ||
3352 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3353 tcp_twrespond(tw, TH_ACK);
3354 goto drop;
3355
3356 reset:
3357 /*
3358 * Generate a RST, dropping incoming segment.
3359 * Make ACK acceptable to originator of segment.
3360 * Don't bother to respond if destination was broadcast/multicast.
3361 */
3362 if (m->m_flags & (M_BCAST|M_MCAST))
3363 goto drop;
3364 if (isipv6) {
3365 struct ip6_hdr *ip6;
3366
3367 /* IPv6 anycast check is done at tcp6_input() */
3368 ip6 = mtod(m, struct ip6_hdr *);
3369 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3370 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3371 goto drop;
3372 } else {
3373 struct ip *ip;
3374
3375 ip = mtod(m, struct ip *);
3376 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3377 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3378 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3379 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3380 goto drop;
3381 }
3382 if (thflags & TH_ACK) {
3383 tcp_respond(NULL,
3384 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3385 } else {
3386 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3387 tcp_respond(NULL,
3388 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3389 }
3390 INP_UNLOCK(tw->tw_inpcb);
3391 return (0);
3392
3393 drop:
3394 INP_UNLOCK(tw->tw_inpcb);
3395 m_freem(m);
3396 return (0);
3397 }
Cache object: e5cde52d3f600a0a31d4936449e24f10
|