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/6.1/sys/netinet/tcp_input.c 156186 2006-03-01 21:13:29Z andre $
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
102 struct tcpstat tcpstat;
103 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
105
106 static int log_in_vain = 0;
107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
108 &log_in_vain, 0, "Log all incoming TCP connections");
109
110 static int blackhole = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
112 &blackhole, 0, "Do not send RST when dropping refused connections");
113
114 int tcp_delack_enabled = 1;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
116 &tcp_delack_enabled, 0,
117 "Delay ACK to try and piggyback it onto a data packet");
118
119 #ifdef TCP_DROP_SYNFIN
120 static int drop_synfin = 0;
121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
123 #endif
124
125 static int tcp_do_rfc3042 = 1;
126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
128
129 static int tcp_do_rfc3390 = 1;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
131 &tcp_do_rfc3390, 0,
132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
133
134 static int tcp_insecure_rst = 0;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
136 &tcp_insecure_rst, 0,
137 "Follow the old (insecure) criteria for accepting RST packets.");
138
139 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
140 "TCP Segment Reassembly Queue");
141
142 static int tcp_reass_maxseg = 0;
143 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
144 &tcp_reass_maxseg, 0,
145 "Global maximum number of TCP Segments in Reassembly Queue");
146
147 int tcp_reass_qsize = 0;
148 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
149 &tcp_reass_qsize, 0,
150 "Global number of TCP Segments currently in Reassembly Queue");
151
152 static int tcp_reass_maxqlen = 48;
153 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
154 &tcp_reass_maxqlen, 0,
155 "Maximum number of TCP Segments per individual Reassembly Queue");
156
157 static int tcp_reass_overflows = 0;
158 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
159 &tcp_reass_overflows, 0,
160 "Global number of TCP Segment Reassembly Queue Overflows");
161
162 struct inpcbhead tcb;
163 #define tcb6 tcb /* for KAME src sync over BSD*'s */
164 struct inpcbinfo tcbinfo;
165 struct mtx *tcbinfo_mtx;
166
167 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
168
169 static void tcp_pulloutofband(struct socket *,
170 struct tcphdr *, struct mbuf *, int);
171 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
172 struct mbuf *);
173 static void tcp_xmit_timer(struct tcpcb *, int);
174 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
175 static int tcp_timewait(struct tcptw *, struct tcpopt *,
176 struct tcphdr *, struct mbuf *, int);
177
178 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
179 #ifdef INET6
180 #define ND6_HINT(tp) \
181 do { \
182 if ((tp) && (tp)->t_inpcb && \
183 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
184 nd6_nud_hint(NULL, NULL, 0); \
185 } while (0)
186 #else
187 #define ND6_HINT(tp)
188 #endif
189
190 /*
191 * Indicate whether this ack should be delayed. We can delay the ack if
192 * - there is no delayed ack timer in progress and
193 * - our last ack wasn't a 0-sized window. We never want to delay
194 * the ack that opens up a 0-sized window and
195 * - delayed acks are enabled or
196 * - this is a half-synchronized T/TCP connection.
197 */
198 #define DELAY_ACK(tp) \
199 ((!callout_active(tp->tt_delack) && \
200 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
201 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
202
203 /* Initialize TCP reassembly queue */
204 uma_zone_t tcp_reass_zone;
205 void
206 tcp_reass_init()
207 {
208 tcp_reass_maxseg = nmbclusters / 16;
209 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
210 &tcp_reass_maxseg);
211 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
213 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
214 }
215
216 static int
217 tcp_reass(tp, th, tlenp, m)
218 register struct tcpcb *tp;
219 register struct tcphdr *th;
220 int *tlenp;
221 struct mbuf *m;
222 {
223 struct tseg_qent *q;
224 struct tseg_qent *p = NULL;
225 struct tseg_qent *nq;
226 struct tseg_qent *te = NULL;
227 struct socket *so = tp->t_inpcb->inp_socket;
228 int flags;
229
230 INP_LOCK_ASSERT(tp->t_inpcb);
231
232 /*
233 * XXX: tcp_reass() is rather inefficient with its data structures
234 * and should be rewritten (see NetBSD for optimizations). While
235 * doing that it should move to its own file tcp_reass.c.
236 */
237
238 /*
239 * Call with th==NULL after become established to
240 * force pre-ESTABLISHED data up to user socket.
241 */
242 if (th == NULL)
243 goto present;
244
245 /*
246 * Limit the number of segments in the reassembly queue to prevent
247 * holding on to too many segments (and thus running out of mbufs).
248 * Make sure to let the missing segment through which caused this
249 * queue. Always keep one global queue entry spare to be able to
250 * process the missing segment.
251 */
252 if (th->th_seq != tp->rcv_nxt &&
253 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
254 tp->t_segqlen >= tcp_reass_maxqlen)) {
255 tcp_reass_overflows++;
256 tcpstat.tcps_rcvmemdrop++;
257 m_freem(m);
258 *tlenp = 0;
259 return (0);
260 }
261
262 /*
263 * Allocate a new queue entry. If we can't, or hit the zone limit
264 * just drop the pkt.
265 */
266 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
267 if (te == NULL) {
268 tcpstat.tcps_rcvmemdrop++;
269 m_freem(m);
270 *tlenp = 0;
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 int headlocked = 0;
437 #ifdef IPFIREWALL_FORWARD
438 struct m_tag *fwd_tag;
439 #endif
440 int rstreason; /* For badport_bandlim accounting purposes */
441
442 struct ip6_hdr *ip6 = NULL;
443 #ifdef INET6
444 int isipv6;
445 #else
446 const int isipv6 = 0;
447 #endif
448
449 #ifdef TCPDEBUG
450 /*
451 * The size of tcp_saveipgen must be the size of the max ip header,
452 * now IPv6.
453 */
454 u_char tcp_saveipgen[40];
455 struct tcphdr tcp_savetcp;
456 short ostate = 0;
457 #endif
458
459 #ifdef INET6
460 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
461 #endif
462 bzero((char *)&to, sizeof(to));
463
464 tcpstat.tcps_rcvtotal++;
465
466 if (isipv6) {
467 #ifdef INET6
468 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
469 ip6 = mtod(m, struct ip6_hdr *);
470 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
471 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
472 tcpstat.tcps_rcvbadsum++;
473 goto drop;
474 }
475 th = (struct tcphdr *)((caddr_t)ip6 + off0);
476
477 /*
478 * Be proactive about unspecified IPv6 address in source.
479 * As we use all-zero to indicate unbounded/unconnected pcb,
480 * unspecified IPv6 address can be used to confuse us.
481 *
482 * Note that packets with unspecified IPv6 destination is
483 * already dropped in ip6_input.
484 */
485 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
486 /* XXX stat */
487 goto drop;
488 }
489 #else
490 th = NULL; /* XXX: avoid compiler warning */
491 #endif
492 } else {
493 /*
494 * Get IP and TCP header together in first mbuf.
495 * Note: IP leaves IP header in first mbuf.
496 */
497 if (off0 > sizeof (struct ip)) {
498 ip_stripoptions(m, (struct mbuf *)0);
499 off0 = sizeof(struct ip);
500 }
501 if (m->m_len < sizeof (struct tcpiphdr)) {
502 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
503 tcpstat.tcps_rcvshort++;
504 return;
505 }
506 }
507 ip = mtod(m, struct ip *);
508 ipov = (struct ipovly *)ip;
509 th = (struct tcphdr *)((caddr_t)ip + off0);
510 tlen = ip->ip_len;
511
512 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
513 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
514 th->th_sum = m->m_pkthdr.csum_data;
515 else
516 th->th_sum = in_pseudo(ip->ip_src.s_addr,
517 ip->ip_dst.s_addr,
518 htonl(m->m_pkthdr.csum_data +
519 ip->ip_len +
520 IPPROTO_TCP));
521 th->th_sum ^= 0xffff;
522 #ifdef TCPDEBUG
523 ipov->ih_len = (u_short)tlen;
524 ipov->ih_len = htons(ipov->ih_len);
525 #endif
526 } else {
527 /*
528 * Checksum extended TCP header and data.
529 */
530 len = sizeof (struct ip) + tlen;
531 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
532 ipov->ih_len = (u_short)tlen;
533 ipov->ih_len = htons(ipov->ih_len);
534 th->th_sum = in_cksum(m, len);
535 }
536 if (th->th_sum) {
537 tcpstat.tcps_rcvbadsum++;
538 goto drop;
539 }
540 #ifdef INET6
541 /* Re-initialization for later version check */
542 ip->ip_v = IPVERSION;
543 #endif
544 }
545
546 /*
547 * Check that TCP offset makes sense,
548 * pull out TCP options and adjust length. XXX
549 */
550 off = th->th_off << 2;
551 if (off < sizeof (struct tcphdr) || off > tlen) {
552 tcpstat.tcps_rcvbadoff++;
553 goto drop;
554 }
555 tlen -= off; /* tlen is used instead of ti->ti_len */
556 if (off > sizeof (struct tcphdr)) {
557 if (isipv6) {
558 #ifdef INET6
559 IP6_EXTHDR_CHECK(m, off0, off, );
560 ip6 = mtod(m, struct ip6_hdr *);
561 th = (struct tcphdr *)((caddr_t)ip6 + off0);
562 #endif
563 } else {
564 if (m->m_len < sizeof(struct ip) + off) {
565 if ((m = m_pullup(m, sizeof (struct ip) + off))
566 == 0) {
567 tcpstat.tcps_rcvshort++;
568 return;
569 }
570 ip = mtod(m, struct ip *);
571 ipov = (struct ipovly *)ip;
572 th = (struct tcphdr *)((caddr_t)ip + off0);
573 }
574 }
575 optlen = off - sizeof (struct tcphdr);
576 optp = (u_char *)(th + 1);
577 }
578 thflags = th->th_flags;
579
580 #ifdef TCP_DROP_SYNFIN
581 /*
582 * If the drop_synfin option is enabled, drop all packets with
583 * both the SYN and FIN bits set. This prevents e.g. nmap from
584 * identifying the TCP/IP stack.
585 *
586 * This is a violation of the TCP specification.
587 */
588 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
589 goto drop;
590 #endif
591
592 /*
593 * Convert TCP protocol specific fields to host format.
594 */
595 th->th_seq = ntohl(th->th_seq);
596 th->th_ack = ntohl(th->th_ack);
597 th->th_win = ntohs(th->th_win);
598 th->th_urp = ntohs(th->th_urp);
599
600 /*
601 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
602 * until after ip6_savecontrol() is called and before other functions
603 * which don't want those proto headers.
604 * Because ip6_savecontrol() is going to parse the mbuf to
605 * search for data to be passed up to user-land, it wants mbuf
606 * parameters to be unchanged.
607 * XXX: the call of ip6_savecontrol() has been obsoleted based on
608 * latest version of the advanced API (20020110).
609 */
610 drop_hdrlen = off0 + off;
611
612 /*
613 * Locate pcb for segment.
614 */
615 INP_INFO_WLOCK(&tcbinfo);
616 headlocked = 1;
617 findpcb:
618 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
619 #ifdef IPFIREWALL_FORWARD
620 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
621 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
622
623 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
624 struct sockaddr_in *next_hop;
625
626 next_hop = (struct sockaddr_in *)(fwd_tag+1);
627 /*
628 * Transparently forwarded. Pretend to be the destination.
629 * already got one like this?
630 */
631 inp = in_pcblookup_hash(&tcbinfo,
632 ip->ip_src, th->th_sport,
633 ip->ip_dst, th->th_dport,
634 0, m->m_pkthdr.rcvif);
635 if (!inp) {
636 /* It's new. Try to find the ambushing socket. */
637 inp = in_pcblookup_hash(&tcbinfo,
638 ip->ip_src, th->th_sport,
639 next_hop->sin_addr,
640 next_hop->sin_port ?
641 ntohs(next_hop->sin_port) :
642 th->th_dport,
643 1, m->m_pkthdr.rcvif);
644 }
645 /* Remove the tag from the packet. We don't need it anymore. */
646 m_tag_delete(m, fwd_tag);
647 } else {
648 #endif /* IPFIREWALL_FORWARD */
649 if (isipv6) {
650 #ifdef INET6
651 inp = in6_pcblookup_hash(&tcbinfo,
652 &ip6->ip6_src, th->th_sport,
653 &ip6->ip6_dst, th->th_dport,
654 1, m->m_pkthdr.rcvif);
655 #endif
656 } else
657 inp = in_pcblookup_hash(&tcbinfo,
658 ip->ip_src, th->th_sport,
659 ip->ip_dst, th->th_dport,
660 1, m->m_pkthdr.rcvif);
661 #ifdef IPFIREWALL_FORWARD
662 }
663 #endif /* IPFIREWALL_FORWARD */
664
665 #if defined(IPSEC) || defined(FAST_IPSEC)
666 #ifdef INET6
667 if (isipv6) {
668 if (inp != NULL && ipsec6_in_reject(m, inp)) {
669 #ifdef IPSEC
670 ipsec6stat.in_polvio++;
671 #endif
672 goto drop;
673 }
674 } else
675 #endif /* INET6 */
676 if (inp != NULL && ipsec4_in_reject(m, inp)) {
677 #ifdef IPSEC
678 ipsecstat.in_polvio++;
679 #endif
680 goto drop;
681 }
682 #endif /*IPSEC || FAST_IPSEC*/
683
684 /*
685 * If the state is CLOSED (i.e., TCB does not exist) then
686 * all data in the incoming segment is discarded.
687 * If the TCB exists but is in CLOSED state, it is embryonic,
688 * but should either do a listen or a connect soon.
689 */
690 if (inp == NULL) {
691 if (log_in_vain) {
692 #ifdef INET6
693 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
694 #else
695 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
696 #endif
697
698 if (isipv6) {
699 #ifdef INET6
700 strcpy(dbuf, "[");
701 strcpy(sbuf, "[");
702 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
703 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
704 strcat(dbuf, "]");
705 strcat(sbuf, "]");
706 #endif
707 } else {
708 strcpy(dbuf, inet_ntoa(ip->ip_dst));
709 strcpy(sbuf, inet_ntoa(ip->ip_src));
710 }
711 switch (log_in_vain) {
712 case 1:
713 if ((thflags & TH_SYN) == 0)
714 break;
715 /* FALLTHROUGH */
716 case 2:
717 log(LOG_INFO,
718 "Connection attempt to TCP %s:%d "
719 "from %s:%d flags:0x%02x\n",
720 dbuf, ntohs(th->th_dport), sbuf,
721 ntohs(th->th_sport), thflags);
722 break;
723 default:
724 break;
725 }
726 }
727 if (blackhole) {
728 switch (blackhole) {
729 case 1:
730 if (thflags & TH_SYN)
731 goto drop;
732 break;
733 case 2:
734 goto drop;
735 default:
736 goto drop;
737 }
738 }
739 rstreason = BANDLIM_RST_CLOSEDPORT;
740 goto dropwithreset;
741 }
742 INP_LOCK(inp);
743
744 /* Check the minimum TTL for socket. */
745 if (inp->inp_ip_minttl != 0) {
746 #ifdef INET6
747 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
748 goto drop;
749 else
750 #endif
751 if (inp->inp_ip_minttl > ip->ip_ttl)
752 goto drop;
753 }
754
755 if (inp->inp_vflag & INP_TIMEWAIT) {
756 /*
757 * The only option of relevance is TOF_CC, and only if
758 * present in a SYN segment. See tcp_timewait().
759 */
760 if (thflags & TH_SYN)
761 tcp_dooptions(&to, optp, optlen, 1);
762 if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
763 &to, th, m, tlen))
764 goto findpcb;
765 /*
766 * tcp_timewait unlocks inp.
767 */
768 INP_INFO_WUNLOCK(&tcbinfo);
769 return;
770 }
771 tp = intotcpcb(inp);
772 if (tp == 0) {
773 INP_UNLOCK(inp);
774 rstreason = BANDLIM_RST_CLOSEDPORT;
775 goto dropwithreset;
776 }
777 if (tp->t_state == TCPS_CLOSED)
778 goto drop;
779
780 /* Unscale the window into a 32-bit value. */
781 if ((thflags & TH_SYN) == 0)
782 tiwin = th->th_win << tp->snd_scale;
783 else
784 tiwin = th->th_win;
785
786 #ifdef MAC
787 INP_LOCK_ASSERT(inp);
788 if (mac_check_inpcb_deliver(inp, m))
789 goto drop;
790 #endif
791 so = inp->inp_socket;
792 #ifdef TCPDEBUG
793 if (so->so_options & SO_DEBUG) {
794 ostate = tp->t_state;
795 if (isipv6)
796 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
797 else
798 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
799 tcp_savetcp = *th;
800 }
801 #endif
802 if (so->so_options & SO_ACCEPTCONN) {
803 struct in_conninfo inc;
804
805 #ifdef INET6
806 inc.inc_isipv6 = isipv6;
807 #endif
808 if (isipv6) {
809 inc.inc6_faddr = ip6->ip6_src;
810 inc.inc6_laddr = ip6->ip6_dst;
811 } else {
812 inc.inc_faddr = ip->ip_src;
813 inc.inc_laddr = ip->ip_dst;
814 }
815 inc.inc_fport = th->th_sport;
816 inc.inc_lport = th->th_dport;
817
818 /*
819 * If the state is LISTEN then ignore segment if it contains
820 * a RST. If the segment contains an ACK then it is bad and
821 * send a RST. If it does not contain a SYN then it is not
822 * interesting; drop it.
823 *
824 * If the state is SYN_RECEIVED (syncache) and seg contains
825 * an ACK, but not for our SYN/ACK, send a RST. If the seg
826 * contains a RST, check the sequence number to see if it
827 * is a valid reset segment.
828 */
829 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
830 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
831 if (!syncache_expand(&inc, th, &so, m)) {
832 /*
833 * No syncache entry, or ACK was not
834 * for our SYN/ACK. Send a RST.
835 */
836 tcpstat.tcps_badsyn++;
837 rstreason = BANDLIM_RST_OPENPORT;
838 goto dropwithreset;
839 }
840 if (so == NULL) {
841 /*
842 * Could not complete 3-way handshake,
843 * connection is being closed down, and
844 * syncache will free mbuf.
845 */
846 INP_UNLOCK(inp);
847 INP_INFO_WUNLOCK(&tcbinfo);
848 return;
849 }
850 /*
851 * Socket is created in state SYN_RECEIVED.
852 * Continue processing segment.
853 */
854 INP_UNLOCK(inp);
855 inp = sotoinpcb(so);
856 INP_LOCK(inp);
857 tp = intotcpcb(inp);
858 /*
859 * This is what would have happened in
860 * tcp_output() when the SYN,ACK was sent.
861 */
862 tp->snd_up = tp->snd_una;
863 tp->snd_max = tp->snd_nxt = tp->iss + 1;
864 tp->last_ack_sent = tp->rcv_nxt;
865 /*
866 * RFC1323: The window in SYN & SYN/ACK
867 * segments is never scaled.
868 */
869 tp->snd_wnd = tiwin; /* unscaled */
870 goto after_listen;
871 }
872 if (thflags & TH_RST) {
873 syncache_chkrst(&inc, th);
874 goto drop;
875 }
876 if (thflags & TH_ACK) {
877 syncache_badack(&inc);
878 tcpstat.tcps_badsyn++;
879 rstreason = BANDLIM_RST_OPENPORT;
880 goto dropwithreset;
881 }
882 goto drop;
883 }
884
885 /*
886 * Segment's flags are (SYN) or (SYN|FIN).
887 */
888 #ifdef INET6
889 /*
890 * If deprecated address is forbidden,
891 * we do not accept SYN to deprecated interface
892 * address to prevent any new inbound connection from
893 * getting established.
894 * When we do not accept SYN, we send a TCP RST,
895 * with deprecated source address (instead of dropping
896 * it). We compromise it as it is much better for peer
897 * to send a RST, and RST will be the final packet
898 * for the exchange.
899 *
900 * If we do not forbid deprecated addresses, we accept
901 * the SYN packet. RFC2462 does not suggest dropping
902 * SYN in this case.
903 * If we decipher RFC2462 5.5.4, it says like this:
904 * 1. use of deprecated addr with existing
905 * communication is okay - "SHOULD continue to be
906 * used"
907 * 2. use of it with new communication:
908 * (2a) "SHOULD NOT be used if alternate address
909 * with sufficient scope is available"
910 * (2b) nothing mentioned otherwise.
911 * Here we fall into (2b) case as we have no choice in
912 * our source address selection - we must obey the peer.
913 *
914 * The wording in RFC2462 is confusing, and there are
915 * multiple description text for deprecated address
916 * handling - worse, they are not exactly the same.
917 * I believe 5.5.4 is the best one, so we follow 5.5.4.
918 */
919 if (isipv6 && !ip6_use_deprecated) {
920 struct in6_ifaddr *ia6;
921
922 if ((ia6 = ip6_getdstifaddr(m)) &&
923 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
924 INP_UNLOCK(inp);
925 tp = NULL;
926 rstreason = BANDLIM_RST_OPENPORT;
927 goto dropwithreset;
928 }
929 }
930 #endif
931 /*
932 * If it is from this socket, drop it, it must be forged.
933 * Don't bother responding if the destination was a broadcast.
934 */
935 if (th->th_dport == th->th_sport) {
936 if (isipv6) {
937 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
938 &ip6->ip6_src))
939 goto drop;
940 } else {
941 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
942 goto drop;
943 }
944 }
945 /*
946 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
947 *
948 * Note that it is quite possible to receive unicast
949 * link-layer packets with a broadcast IP address. Use
950 * in_broadcast() to find them.
951 */
952 if (m->m_flags & (M_BCAST|M_MCAST))
953 goto drop;
954 if (isipv6) {
955 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
956 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
957 goto drop;
958 } else {
959 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
960 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
961 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
962 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
963 goto drop;
964 }
965 /*
966 * SYN appears to be valid; create compressed TCP state
967 * for syncache, or perform t/tcp connection.
968 */
969 if (so->so_qlen <= so->so_qlimit) {
970 #ifdef TCPDEBUG
971 if (so->so_options & SO_DEBUG)
972 tcp_trace(TA_INPUT, ostate, tp,
973 (void *)tcp_saveipgen, &tcp_savetcp, 0);
974 #endif
975 tcp_dooptions(&to, optp, optlen, 1);
976 if (!syncache_add(&inc, &to, th, &so, m))
977 goto drop;
978 if (so == NULL) {
979 /*
980 * Entry added to syncache, mbuf used to
981 * send SYN,ACK packet.
982 */
983 KASSERT(headlocked, ("headlocked"));
984 INP_UNLOCK(inp);
985 INP_INFO_WUNLOCK(&tcbinfo);
986 return;
987 }
988 /*
989 * Segment passed TAO tests.
990 */
991 INP_UNLOCK(inp);
992 inp = sotoinpcb(so);
993 INP_LOCK(inp);
994 tp = intotcpcb(inp);
995 tp->snd_wnd = tiwin;
996 tp->t_starttime = ticks;
997 tp->t_state = TCPS_ESTABLISHED;
998
999 /*
1000 * T/TCP logic:
1001 * If there is a FIN or if there is data, then
1002 * delay SYN,ACK(SYN) in the hope of piggy-backing
1003 * it on a response segment. Otherwise must send
1004 * ACK now in case the other side is slow starting.
1005 */
1006 if (thflags & TH_FIN || tlen != 0)
1007 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1008 else
1009 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1010 tcpstat.tcps_connects++;
1011 soisconnected(so);
1012 goto trimthenstep6;
1013 }
1014 goto drop;
1015 }
1016 after_listen:
1017 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1018 INP_LOCK_ASSERT(inp);
1019
1020 /* Syncache takes care of sockets in the listen state. */
1021 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1022
1023 /*
1024 * This is the second part of the MSS DoS prevention code (after
1025 * minmss on the sending side) and it deals with too many too small
1026 * tcp packets in a too short timeframe (1 second).
1027 *
1028 * For every full second we count the number of received packets
1029 * and bytes. If we get a lot of packets per second for this connection
1030 * (tcp_minmssoverload) we take a closer look at it and compute the
1031 * average packet size for the past second. If that is less than
1032 * tcp_minmss we get too many packets with very small payload which
1033 * is not good and burdens our system (and every packet generates
1034 * a wakeup to the process connected to our socket). We can reasonable
1035 * expect this to be small packet DoS attack to exhaust our CPU
1036 * cycles.
1037 *
1038 * Care has to be taken for the minimum packet overload value. This
1039 * value defines the minimum number of packets per second before we
1040 * start to worry. This must not be too low to avoid killing for
1041 * example interactive connections with many small packets like
1042 * telnet or SSH.
1043 *
1044 * Setting either tcp_minmssoverload or tcp_minmss to "" disables
1045 * this check.
1046 *
1047 * Account for packet if payload packet, skip over ACK, etc.
1048 */
1049 if (tcp_minmss && tcp_minmssoverload &&
1050 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1051 if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1052 tp->rcv_pps++;
1053 tp->rcv_byps += tlen + off;
1054 if (tp->rcv_pps > tcp_minmssoverload) {
1055 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1056 printf("too many small tcp packets from "
1057 "%s:%u, av. %lubyte/packet, "
1058 "dropping connection\n",
1059 #ifdef INET6
1060 isipv6 ?
1061 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1062 #endif
1063 inet_ntoa(inp->inp_inc.inc_faddr),
1064 inp->inp_inc.inc_fport,
1065 tp->rcv_byps / tp->rcv_pps);
1066 KASSERT(headlocked, ("tcp_input: "
1067 "after_listen: tcp_drop: head "
1068 "not locked"));
1069 tp = tcp_drop(tp, ECONNRESET);
1070 tcpstat.tcps_minmssdrops++;
1071 goto drop;
1072 }
1073 }
1074 } else {
1075 tp->rcv_second = ticks + hz;
1076 tp->rcv_pps = 1;
1077 tp->rcv_byps = tlen + off;
1078 }
1079 }
1080
1081 /*
1082 * Segment received on connection.
1083 * Reset idle time and keep-alive timer.
1084 */
1085 tp->t_rcvtime = ticks;
1086 if (TCPS_HAVEESTABLISHED(tp->t_state))
1087 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1088
1089 /*
1090 * Process options only when we get SYN/ACK back. The SYN case
1091 * for incoming connections is handled in tcp_syncache.
1092 * XXX this is traditional behavior, may need to be cleaned up.
1093 */
1094 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
1095 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1096 if (to.to_flags & TOF_SCALE) {
1097 tp->t_flags |= TF_RCVD_SCALE;
1098 tp->requested_s_scale = to.to_requested_s_scale;
1099 }
1100 if (to.to_flags & TOF_TS) {
1101 tp->t_flags |= TF_RCVD_TSTMP;
1102 tp->ts_recent = to.to_tsval;
1103 tp->ts_recent_age = ticks;
1104 }
1105 if (to.to_flags & TOF_MSS)
1106 tcp_mss(tp, to.to_mss);
1107 if (tp->sack_enable) {
1108 if (!(to.to_flags & TOF_SACK))
1109 tp->sack_enable = 0;
1110 else
1111 tp->t_flags |= TF_SACK_PERMIT;
1112 }
1113
1114 }
1115
1116 /*
1117 * Header prediction: check for the two common cases
1118 * of a uni-directional data xfer. If the packet has
1119 * no control flags, is in-sequence, the window didn't
1120 * change and we're not retransmitting, it's a
1121 * candidate. If the length is zero and the ack moved
1122 * forward, we're the sender side of the xfer. Just
1123 * free the data acked & wake any higher level process
1124 * that was blocked waiting for space. If the length
1125 * is non-zero and the ack didn't move, we're the
1126 * receiver side. If we're getting packets in-order
1127 * (the reassembly queue is empty), add the data to
1128 * the socket buffer and note that we need a delayed ack.
1129 * Make sure that the hidden state-flags are also off.
1130 * Since we check for TCPS_ESTABLISHED above, it can only
1131 * be TH_NEEDSYN.
1132 */
1133 if (tp->t_state == TCPS_ESTABLISHED &&
1134 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1135 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1136 ((to.to_flags & TOF_TS) == 0 ||
1137 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1138 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1139 tp->snd_nxt == tp->snd_max) {
1140
1141 /*
1142 * If last ACK falls within this segment's sequence numbers,
1143 * record the timestamp.
1144 * NOTE that the test is modified according to the latest
1145 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1146 */
1147 if ((to.to_flags & TOF_TS) != 0 &&
1148 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1149 tp->ts_recent_age = ticks;
1150 tp->ts_recent = to.to_tsval;
1151 }
1152
1153 if (tlen == 0) {
1154 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1155 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1156 tp->snd_cwnd >= tp->snd_wnd &&
1157 ((!tcp_do_newreno && !tp->sack_enable &&
1158 tp->t_dupacks < tcprexmtthresh) ||
1159 ((tcp_do_newreno || tp->sack_enable) &&
1160 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1161 TAILQ_EMPTY(&tp->snd_holes)))) {
1162 KASSERT(headlocked, ("headlocked"));
1163 INP_INFO_WUNLOCK(&tcbinfo);
1164 headlocked = 0;
1165 /*
1166 * this is a pure ack for outstanding data.
1167 */
1168 ++tcpstat.tcps_predack;
1169 /*
1170 * "bad retransmit" recovery
1171 */
1172 if (tp->t_rxtshift == 1 &&
1173 ticks < tp->t_badrxtwin) {
1174 ++tcpstat.tcps_sndrexmitbad;
1175 tp->snd_cwnd = tp->snd_cwnd_prev;
1176 tp->snd_ssthresh =
1177 tp->snd_ssthresh_prev;
1178 tp->snd_recover = tp->snd_recover_prev;
1179 if (tp->t_flags & TF_WASFRECOVERY)
1180 ENTER_FASTRECOVERY(tp);
1181 tp->snd_nxt = tp->snd_max;
1182 tp->t_badrxtwin = 0;
1183 }
1184
1185 /*
1186 * Recalculate the transmit timer / rtt.
1187 *
1188 * Some boxes send broken timestamp replies
1189 * during the SYN+ACK phase, ignore
1190 * timestamps of 0 or we could calculate a
1191 * huge RTT and blow up the retransmit timer.
1192 */
1193 if ((to.to_flags & TOF_TS) != 0 &&
1194 to.to_tsecr) {
1195 if (!tp->t_rttlow ||
1196 tp->t_rttlow > ticks - to.to_tsecr)
1197 tp->t_rttlow = ticks - 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 if (!tp->t_rttlow ||
1203 tp->t_rttlow > ticks - tp->t_rtttime)
1204 tp->t_rttlow = ticks - tp->t_rtttime;
1205 tcp_xmit_timer(tp,
1206 ticks - tp->t_rtttime);
1207 }
1208 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1209 acked = th->th_ack - tp->snd_una;
1210 tcpstat.tcps_rcvackpack++;
1211 tcpstat.tcps_rcvackbyte += acked;
1212 sbdrop(&so->so_snd, acked);
1213 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1214 SEQ_LEQ(th->th_ack, tp->snd_recover))
1215 tp->snd_recover = th->th_ack - 1;
1216 tp->snd_una = th->th_ack;
1217 /*
1218 * pull snd_wl2 up to prevent seq wrap relative
1219 * to th_ack.
1220 */
1221 tp->snd_wl2 = th->th_ack;
1222 tp->t_dupacks = 0;
1223 m_freem(m);
1224 ND6_HINT(tp); /* some progress has been done */
1225
1226 /*
1227 * If all outstanding data are acked, stop
1228 * retransmit timer, otherwise restart timer
1229 * using current (possibly backed-off) value.
1230 * If process is waiting for space,
1231 * wakeup/selwakeup/signal. If data
1232 * are ready to send, let tcp_output
1233 * decide between more output or persist.
1234
1235 #ifdef TCPDEBUG
1236 if (so->so_options & SO_DEBUG)
1237 tcp_trace(TA_INPUT, ostate, tp,
1238 (void *)tcp_saveipgen,
1239 &tcp_savetcp, 0);
1240 #endif
1241 */
1242 if (tp->snd_una == tp->snd_max)
1243 callout_stop(tp->tt_rexmt);
1244 else if (!callout_active(tp->tt_persist))
1245 callout_reset(tp->tt_rexmt,
1246 tp->t_rxtcur,
1247 tcp_timer_rexmt, tp);
1248
1249 sowwakeup(so);
1250 if (so->so_snd.sb_cc)
1251 (void) tcp_output(tp);
1252 goto check_delack;
1253 }
1254 } else if (th->th_ack == tp->snd_una &&
1255 LIST_EMPTY(&tp->t_segq) &&
1256 tlen <= sbspace(&so->so_rcv)) {
1257 KASSERT(headlocked, ("headlocked"));
1258 INP_INFO_WUNLOCK(&tcbinfo);
1259 headlocked = 0;
1260 /*
1261 * this is a pure, in-sequence data packet
1262 * with nothing on the reassembly queue and
1263 * we have enough buffer space to take it.
1264 */
1265 /* Clean receiver SACK report if present */
1266 if (tp->sack_enable && tp->rcv_numsacks)
1267 tcp_clean_sackreport(tp);
1268 ++tcpstat.tcps_preddat;
1269 tp->rcv_nxt += tlen;
1270 /*
1271 * Pull snd_wl1 up to prevent seq wrap relative to
1272 * th_seq.
1273 */
1274 tp->snd_wl1 = th->th_seq;
1275 /*
1276 * Pull rcv_up up to prevent seq wrap relative to
1277 * rcv_nxt.
1278 */
1279 tp->rcv_up = tp->rcv_nxt;
1280 tcpstat.tcps_rcvpack++;
1281 tcpstat.tcps_rcvbyte += tlen;
1282 ND6_HINT(tp); /* some progress has been done */
1283 /*
1284 #ifdef TCPDEBUG
1285 if (so->so_options & SO_DEBUG)
1286 tcp_trace(TA_INPUT, ostate, tp,
1287 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1288 #endif
1289 * Add data to socket buffer.
1290 */
1291 SOCKBUF_LOCK(&so->so_rcv);
1292 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1293 m_freem(m);
1294 } else {
1295 m_adj(m, drop_hdrlen); /* delayed header drop */
1296 sbappendstream_locked(&so->so_rcv, m);
1297 }
1298 sorwakeup_locked(so);
1299 if (DELAY_ACK(tp)) {
1300 tp->t_flags |= TF_DELACK;
1301 } else {
1302 tp->t_flags |= TF_ACKNOW;
1303 tcp_output(tp);
1304 }
1305 goto check_delack;
1306 }
1307 }
1308
1309 /*
1310 * Calculate amount of space in receive window,
1311 * and then do TCP input processing.
1312 * Receive window is amount of space in rcv queue,
1313 * but not less than advertised window.
1314 */
1315 { int win;
1316
1317 win = sbspace(&so->so_rcv);
1318 if (win < 0)
1319 win = 0;
1320 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1321 }
1322
1323 switch (tp->t_state) {
1324
1325 /*
1326 * If the state is SYN_RECEIVED:
1327 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1328 */
1329 case TCPS_SYN_RECEIVED:
1330 if ((thflags & TH_ACK) &&
1331 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1332 SEQ_GT(th->th_ack, tp->snd_max))) {
1333 rstreason = BANDLIM_RST_OPENPORT;
1334 goto dropwithreset;
1335 }
1336 break;
1337
1338 /*
1339 * If the state is SYN_SENT:
1340 * if seg contains an ACK, but not for our SYN, drop the input.
1341 * if seg contains a RST, then drop the connection.
1342 * if seg does not contain SYN, then drop it.
1343 * Otherwise this is an acceptable SYN segment
1344 * initialize tp->rcv_nxt and tp->irs
1345 * if seg contains ack then advance tp->snd_una
1346 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1347 * arrange for segment to be acked (eventually)
1348 * continue processing rest of data/controls, beginning with URG
1349 */
1350 case TCPS_SYN_SENT:
1351 if ((thflags & TH_ACK) &&
1352 (SEQ_LEQ(th->th_ack, tp->iss) ||
1353 SEQ_GT(th->th_ack, tp->snd_max))) {
1354 rstreason = BANDLIM_UNLIMITED;
1355 goto dropwithreset;
1356 }
1357 if (thflags & TH_RST) {
1358 if (thflags & TH_ACK) {
1359 KASSERT(headlocked, ("tcp_input: after_listen"
1360 ": tcp_drop.2: head not locked"));
1361 tp = tcp_drop(tp, ECONNREFUSED);
1362 }
1363 goto drop;
1364 }
1365 if ((thflags & TH_SYN) == 0)
1366 goto drop;
1367 tp->snd_wnd = th->th_win; /* initial send window */
1368
1369 tp->irs = th->th_seq;
1370 tcp_rcvseqinit(tp);
1371 if (thflags & TH_ACK) {
1372 tcpstat.tcps_connects++;
1373 soisconnected(so);
1374 #ifdef MAC
1375 SOCK_LOCK(so);
1376 mac_set_socket_peer_from_mbuf(m, so);
1377 SOCK_UNLOCK(so);
1378 #endif
1379 /* Do window scaling on this connection? */
1380 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1381 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1382 tp->snd_scale = tp->requested_s_scale;
1383 tp->rcv_scale = tp->request_r_scale;
1384 }
1385 tp->rcv_adv += tp->rcv_wnd;
1386 tp->snd_una++; /* SYN is acked */
1387 /*
1388 * If there's data, delay ACK; if there's also a FIN
1389 * ACKNOW will be turned on later.
1390 */
1391 if (DELAY_ACK(tp) && tlen != 0)
1392 callout_reset(tp->tt_delack, tcp_delacktime,
1393 tcp_timer_delack, tp);
1394 else
1395 tp->t_flags |= TF_ACKNOW;
1396 /*
1397 * Received <SYN,ACK> in SYN_SENT[*] state.
1398 * Transitions:
1399 * SYN_SENT --> ESTABLISHED
1400 * SYN_SENT* --> FIN_WAIT_1
1401 */
1402 tp->t_starttime = ticks;
1403 if (tp->t_flags & TF_NEEDFIN) {
1404 tp->t_state = TCPS_FIN_WAIT_1;
1405 tp->t_flags &= ~TF_NEEDFIN;
1406 thflags &= ~TH_SYN;
1407 } else {
1408 tp->t_state = TCPS_ESTABLISHED;
1409 callout_reset(tp->tt_keep, tcp_keepidle,
1410 tcp_timer_keep, tp);
1411 }
1412 } else {
1413 /*
1414 * Received initial SYN in SYN-SENT[*] state =>
1415 * simultaneous open. If segment contains CC option
1416 * and there is a cached CC, apply TAO test.
1417 * If it succeeds, connection is * half-synchronized.
1418 * Otherwise, do 3-way handshake:
1419 * SYN-SENT -> SYN-RECEIVED
1420 * SYN-SENT* -> SYN-RECEIVED*
1421 * If there was no CC option, clear cached CC value.
1422 */
1423 tp->t_flags |= TF_ACKNOW;
1424 callout_stop(tp->tt_rexmt);
1425 tp->t_state = TCPS_SYN_RECEIVED;
1426 }
1427
1428 trimthenstep6:
1429 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1430 "locked"));
1431 INP_LOCK_ASSERT(inp);
1432
1433 /*
1434 * Advance th->th_seq to correspond to first data byte.
1435 * If data, trim to stay within window,
1436 * dropping FIN if necessary.
1437 */
1438 th->th_seq++;
1439 if (tlen > tp->rcv_wnd) {
1440 todrop = tlen - tp->rcv_wnd;
1441 m_adj(m, -todrop);
1442 tlen = tp->rcv_wnd;
1443 thflags &= ~TH_FIN;
1444 tcpstat.tcps_rcvpackafterwin++;
1445 tcpstat.tcps_rcvbyteafterwin += todrop;
1446 }
1447 tp->snd_wl1 = th->th_seq - 1;
1448 tp->rcv_up = th->th_seq;
1449 /*
1450 * Client side of transaction: already sent SYN and data.
1451 * If the remote host used T/TCP to validate the SYN,
1452 * our data will be ACK'd; if so, enter normal data segment
1453 * processing in the middle of step 5, ack processing.
1454 * Otherwise, goto step 6.
1455 */
1456 if (thflags & TH_ACK)
1457 goto process_ACK;
1458
1459 goto step6;
1460
1461 /*
1462 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1463 * do normal processing.
1464 *
1465 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1466 */
1467 case TCPS_LAST_ACK:
1468 case TCPS_CLOSING:
1469 case TCPS_TIME_WAIT:
1470 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1471 break; /* continue normal processing */
1472 }
1473
1474 /*
1475 * States other than LISTEN or SYN_SENT.
1476 * First check the RST flag and sequence number since reset segments
1477 * are exempt from the timestamp and connection count tests. This
1478 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1479 * below which allowed reset segments in half the sequence space
1480 * to fall though and be processed (which gives forged reset
1481 * segments with a random sequence number a 50 percent chance of
1482 * killing a connection).
1483 * Then check timestamp, if present.
1484 * Then check the connection count, if present.
1485 * Then check that at least some bytes of segment are within
1486 * receive window. If segment begins before rcv_nxt,
1487 * drop leading data (and SYN); if nothing left, just ack.
1488 *
1489 *
1490 * If the RST bit is set, check the sequence number to see
1491 * if this is a valid reset segment.
1492 * RFC 793 page 37:
1493 * In all states except SYN-SENT, all reset (RST) segments
1494 * are validated by checking their SEQ-fields. A reset is
1495 * valid if its sequence number is in the window.
1496 * Note: this does not take into account delayed ACKs, so
1497 * we should test against last_ack_sent instead of rcv_nxt.
1498 * The sequence number in the reset segment is normally an
1499 * echo of our outgoing acknowlegement numbers, but some hosts
1500 * send a reset with the sequence number at the rightmost edge
1501 * of our receive window, and we have to handle this case.
1502 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1503 * that brute force RST attacks are possible. To combat this,
1504 * we use a much stricter check while in the ESTABLISHED state,
1505 * only accepting RSTs where the sequence number is equal to
1506 * last_ack_sent. In all other states (the states in which a
1507 * RST is more likely), the more permissive check is used.
1508 * If we have multiple segments in flight, the intial reset
1509 * segment sequence numbers will be to the left of last_ack_sent,
1510 * but they will eventually catch up.
1511 * In any case, it never made sense to trim reset segments to
1512 * fit the receive window since RFC 1122 says:
1513 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1514 *
1515 * A TCP SHOULD allow a received RST segment to include data.
1516 *
1517 * DISCUSSION
1518 * It has been suggested that a RST segment could contain
1519 * ASCII text that encoded and explained the cause of the
1520 * RST. No standard has yet been established for such
1521 * data.
1522 *
1523 * If the reset segment passes the sequence number test examine
1524 * the state:
1525 * SYN_RECEIVED STATE:
1526 * If passive open, return to LISTEN state.
1527 * If active open, inform user that connection was refused.
1528 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1529 * Inform user that connection was reset, and close tcb.
1530 * CLOSING, LAST_ACK STATES:
1531 * Close the tcb.
1532 * TIME_WAIT STATE:
1533 * Drop the segment - see Stevens, vol. 2, p. 964 and
1534 * RFC 1337.
1535 */
1536 if (thflags & TH_RST) {
1537 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1538 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1539 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1540 switch (tp->t_state) {
1541
1542 case TCPS_SYN_RECEIVED:
1543 so->so_error = ECONNREFUSED;
1544 goto close;
1545
1546 case TCPS_ESTABLISHED:
1547 if (tp->last_ack_sent != th->th_seq &&
1548 tcp_insecure_rst == 0) {
1549 tcpstat.tcps_badrst++;
1550 goto drop;
1551 }
1552 case TCPS_FIN_WAIT_1:
1553 case TCPS_FIN_WAIT_2:
1554 case TCPS_CLOSE_WAIT:
1555 so->so_error = ECONNRESET;
1556 close:
1557 tp->t_state = TCPS_CLOSED;
1558 tcpstat.tcps_drops++;
1559 KASSERT(headlocked, ("tcp_input: "
1560 "trimthenstep6: tcp_close: head not "
1561 "locked"));
1562 tp = tcp_close(tp);
1563 break;
1564
1565 case TCPS_CLOSING:
1566 case TCPS_LAST_ACK:
1567 KASSERT(headlocked, ("trimthenstep6: "
1568 "tcp_close.2: head not locked"));
1569 tp = tcp_close(tp);
1570 break;
1571
1572 case TCPS_TIME_WAIT:
1573 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1574 ("timewait"));
1575 break;
1576 }
1577 }
1578 goto drop;
1579 }
1580
1581 /*
1582 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1583 * and it's less than ts_recent, drop it.
1584 */
1585 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1586 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1587
1588 /* Check to see if ts_recent is over 24 days old. */
1589 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1590 /*
1591 * Invalidate ts_recent. If this segment updates
1592 * ts_recent, the age will be reset later and ts_recent
1593 * will get a valid value. If it does not, setting
1594 * ts_recent to zero will at least satisfy the
1595 * requirement that zero be placed in the timestamp
1596 * echo reply when ts_recent isn't valid. The
1597 * age isn't reset until we get a valid ts_recent
1598 * because we don't want out-of-order segments to be
1599 * dropped when ts_recent is old.
1600 */
1601 tp->ts_recent = 0;
1602 } else {
1603 tcpstat.tcps_rcvduppack++;
1604 tcpstat.tcps_rcvdupbyte += tlen;
1605 tcpstat.tcps_pawsdrop++;
1606 if (tlen)
1607 goto dropafterack;
1608 goto drop;
1609 }
1610 }
1611
1612 /*
1613 * In the SYN-RECEIVED state, validate that the packet belongs to
1614 * this connection before trimming the data to fit the receive
1615 * window. Check the sequence number versus IRS since we know
1616 * the sequence numbers haven't wrapped. This is a partial fix
1617 * for the "LAND" DoS attack.
1618 */
1619 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1620 rstreason = BANDLIM_RST_OPENPORT;
1621 goto dropwithreset;
1622 }
1623
1624 todrop = tp->rcv_nxt - th->th_seq;
1625 if (todrop > 0) {
1626 if (thflags & TH_SYN) {
1627 thflags &= ~TH_SYN;
1628 th->th_seq++;
1629 if (th->th_urp > 1)
1630 th->th_urp--;
1631 else
1632 thflags &= ~TH_URG;
1633 todrop--;
1634 }
1635 /*
1636 * Following if statement from Stevens, vol. 2, p. 960.
1637 */
1638 if (todrop > tlen
1639 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1640 /*
1641 * Any valid FIN must be to the left of the window.
1642 * At this point the FIN must be a duplicate or out
1643 * of sequence; drop it.
1644 */
1645 thflags &= ~TH_FIN;
1646
1647 /*
1648 * Send an ACK to resynchronize and drop any data.
1649 * But keep on processing for RST or ACK.
1650 */
1651 tp->t_flags |= TF_ACKNOW;
1652 todrop = tlen;
1653 tcpstat.tcps_rcvduppack++;
1654 tcpstat.tcps_rcvdupbyte += todrop;
1655 } else {
1656 tcpstat.tcps_rcvpartduppack++;
1657 tcpstat.tcps_rcvpartdupbyte += todrop;
1658 }
1659 drop_hdrlen += todrop; /* drop from the top afterwards */
1660 th->th_seq += todrop;
1661 tlen -= todrop;
1662 if (th->th_urp > todrop)
1663 th->th_urp -= todrop;
1664 else {
1665 thflags &= ~TH_URG;
1666 th->th_urp = 0;
1667 }
1668 }
1669
1670 /*
1671 * If new data are received on a connection after the
1672 * user processes are gone, then RST the other end.
1673 */
1674 if ((so->so_state & SS_NOFDREF) &&
1675 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1676 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1677 "locked"));
1678 tp = tcp_close(tp);
1679 tcpstat.tcps_rcvafterclose++;
1680 rstreason = BANDLIM_UNLIMITED;
1681 goto dropwithreset;
1682 }
1683
1684 /*
1685 * If segment ends after window, drop trailing data
1686 * (and PUSH and FIN); if nothing left, just ACK.
1687 */
1688 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1689 if (todrop > 0) {
1690 tcpstat.tcps_rcvpackafterwin++;
1691 if (todrop >= tlen) {
1692 tcpstat.tcps_rcvbyteafterwin += tlen;
1693 /*
1694 * If a new connection request is received
1695 * while in TIME_WAIT, drop the old connection
1696 * and start over if the sequence numbers
1697 * are above the previous ones.
1698 */
1699 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1700 if (thflags & TH_SYN &&
1701 tp->t_state == TCPS_TIME_WAIT &&
1702 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1703 KASSERT(headlocked, ("trimthenstep6: "
1704 "tcp_close.4: head not locked"));
1705 tp = tcp_close(tp);
1706 goto findpcb;
1707 }
1708 /*
1709 * If window is closed can only take segments at
1710 * window edge, and have to drop data and PUSH from
1711 * incoming segments. Continue processing, but
1712 * remember to ack. Otherwise, drop segment
1713 * and ack.
1714 */
1715 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1716 tp->t_flags |= TF_ACKNOW;
1717 tcpstat.tcps_rcvwinprobe++;
1718 } else
1719 goto dropafterack;
1720 } else
1721 tcpstat.tcps_rcvbyteafterwin += todrop;
1722 m_adj(m, -todrop);
1723 tlen -= todrop;
1724 thflags &= ~(TH_PUSH|TH_FIN);
1725 }
1726
1727 /*
1728 * If last ACK falls within this segment's sequence numbers,
1729 * record its timestamp.
1730 * NOTE:
1731 * 1) That the test incorporates suggestions from the latest
1732 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1733 * 2) That updating only on newer timestamps interferes with
1734 * our earlier PAWS tests, so this check should be solely
1735 * predicated on the sequence space of this segment.
1736 * 3) That we modify the segment boundary check to be
1737 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1738 * instead of RFC1323's
1739 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1740 * This modified check allows us to overcome RFC1323's
1741 * limitations as described in Stevens TCP/IP Illustrated
1742 * Vol. 2 p.869. In such cases, we can still calculate the
1743 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1744 */
1745 if ((to.to_flags & TOF_TS) != 0 &&
1746 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1747 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1748 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1749 tp->ts_recent_age = ticks;
1750 tp->ts_recent = to.to_tsval;
1751 }
1752
1753 /*
1754 * If a SYN is in the window, then this is an
1755 * error and we send an RST and drop the connection.
1756 */
1757 if (thflags & TH_SYN) {
1758 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1759 "head not locked"));
1760 tp = tcp_drop(tp, ECONNRESET);
1761 rstreason = BANDLIM_UNLIMITED;
1762 goto drop;
1763 }
1764
1765 /*
1766 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1767 * flag is on (half-synchronized state), then queue data for
1768 * later processing; else drop segment and return.
1769 */
1770 if ((thflags & TH_ACK) == 0) {
1771 if (tp->t_state == TCPS_SYN_RECEIVED ||
1772 (tp->t_flags & TF_NEEDSYN))
1773 goto step6;
1774 else
1775 goto drop;
1776 }
1777
1778 /*
1779 * Ack processing.
1780 */
1781 switch (tp->t_state) {
1782
1783 /*
1784 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1785 * ESTABLISHED state and continue processing.
1786 * The ACK was checked above.
1787 */
1788 case TCPS_SYN_RECEIVED:
1789
1790 tcpstat.tcps_connects++;
1791 soisconnected(so);
1792 /* Do window scaling? */
1793 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1794 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1795 tp->snd_scale = tp->requested_s_scale;
1796 tp->rcv_scale = tp->request_r_scale;
1797 }
1798 /*
1799 * Make transitions:
1800 * SYN-RECEIVED -> ESTABLISHED
1801 * SYN-RECEIVED* -> FIN-WAIT-1
1802 */
1803 tp->t_starttime = ticks;
1804 if (tp->t_flags & TF_NEEDFIN) {
1805 tp->t_state = TCPS_FIN_WAIT_1;
1806 tp->t_flags &= ~TF_NEEDFIN;
1807 } else {
1808 tp->t_state = TCPS_ESTABLISHED;
1809 callout_reset(tp->tt_keep, tcp_keepidle,
1810 tcp_timer_keep, tp);
1811 }
1812 /*
1813 * If segment contains data or ACK, will call tcp_reass()
1814 * later; if not, do so now to pass queued data to user.
1815 */
1816 if (tlen == 0 && (thflags & TH_FIN) == 0)
1817 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1818 (struct mbuf *)0);
1819 tp->snd_wl1 = th->th_seq - 1;
1820 /* FALLTHROUGH */
1821
1822 /*
1823 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1824 * ACKs. If the ack is in the range
1825 * tp->snd_una < th->th_ack <= tp->snd_max
1826 * then advance tp->snd_una to th->th_ack and drop
1827 * data from the retransmission queue. If this ACK reflects
1828 * more up to date window information we update our window information.
1829 */
1830 case TCPS_ESTABLISHED:
1831 case TCPS_FIN_WAIT_1:
1832 case TCPS_FIN_WAIT_2:
1833 case TCPS_CLOSE_WAIT:
1834 case TCPS_CLOSING:
1835 case TCPS_LAST_ACK:
1836 case TCPS_TIME_WAIT:
1837 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1838 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1839 tcpstat.tcps_rcvacktoomuch++;
1840 goto dropafterack;
1841 }
1842 if (tp->sack_enable &&
1843 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1844 tcp_sack_doack(tp, &to, th->th_ack);
1845 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1846 if (tlen == 0 && tiwin == tp->snd_wnd) {
1847 tcpstat.tcps_rcvdupack++;
1848 /*
1849 * If we have outstanding data (other than
1850 * a window probe), this is a completely
1851 * duplicate ack (ie, window info didn't
1852 * change), the ack is the biggest we've
1853 * seen and we've seen exactly our rexmt
1854 * threshhold of them, assume a packet
1855 * has been dropped and retransmit it.
1856 * Kludge snd_nxt & the congestion
1857 * window so we send only this one
1858 * packet.
1859 *
1860 * We know we're losing at the current
1861 * window size so do congestion avoidance
1862 * (set ssthresh to half the current window
1863 * and pull our congestion window back to
1864 * the new ssthresh).
1865 *
1866 * Dup acks mean that packets have left the
1867 * network (they're now cached at the receiver)
1868 * so bump cwnd by the amount in the receiver
1869 * to keep a constant cwnd packets in the
1870 * network.
1871 */
1872 if (!callout_active(tp->tt_rexmt) ||
1873 th->th_ack != tp->snd_una)
1874 tp->t_dupacks = 0;
1875 else if (++tp->t_dupacks > tcprexmtthresh ||
1876 ((tcp_do_newreno || tp->sack_enable) &&
1877 IN_FASTRECOVERY(tp))) {
1878 if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1879 int awnd;
1880
1881 /*
1882 * Compute the amount of data in flight first.
1883 * We can inject new data into the pipe iff
1884 * we have less than 1/2 the original window's
1885 * worth of data in flight.
1886 */
1887 awnd = (tp->snd_nxt - tp->snd_fack) +
1888 tp->sackhint.sack_bytes_rexmit;
1889 if (awnd < tp->snd_ssthresh) {
1890 tp->snd_cwnd += tp->t_maxseg;
1891 if (tp->snd_cwnd > tp->snd_ssthresh)
1892 tp->snd_cwnd = tp->snd_ssthresh;
1893 }
1894 } else
1895 tp->snd_cwnd += tp->t_maxseg;
1896 (void) tcp_output(tp);
1897 goto drop;
1898 } else if (tp->t_dupacks == tcprexmtthresh) {
1899 tcp_seq onxt = tp->snd_nxt;
1900 u_int win;
1901
1902 /*
1903 * If we're doing sack, check to
1904 * see if we're already in sack
1905 * recovery. If we're not doing sack,
1906 * check to see if we're in newreno
1907 * recovery.
1908 */
1909 if (tp->sack_enable) {
1910 if (IN_FASTRECOVERY(tp)) {
1911 tp->t_dupacks = 0;
1912 break;
1913 }
1914 } else if (tcp_do_newreno) {
1915 if (SEQ_LEQ(th->th_ack,
1916 tp->snd_recover)) {
1917 tp->t_dupacks = 0;
1918 break;
1919 }
1920 }
1921 win = min(tp->snd_wnd, tp->snd_cwnd) /
1922 2 / tp->t_maxseg;
1923 if (win < 2)
1924 win = 2;
1925 tp->snd_ssthresh = win * tp->t_maxseg;
1926 ENTER_FASTRECOVERY(tp);
1927 tp->snd_recover = tp->snd_max;
1928 callout_stop(tp->tt_rexmt);
1929 tp->t_rtttime = 0;
1930 if (tp->sack_enable) {
1931 tcpstat.tcps_sack_recovery_episode++;
1932 tp->sack_newdata = tp->snd_nxt;
1933 tp->snd_cwnd = tp->t_maxseg;
1934 (void) tcp_output(tp);
1935 goto drop;
1936 }
1937 tp->snd_nxt = th->th_ack;
1938 tp->snd_cwnd = tp->t_maxseg;
1939 (void) tcp_output(tp);
1940 KASSERT(tp->snd_limited <= 2,
1941 ("tp->snd_limited too big"));
1942 tp->snd_cwnd = tp->snd_ssthresh +
1943 tp->t_maxseg *
1944 (tp->t_dupacks - tp->snd_limited);
1945 if (SEQ_GT(onxt, tp->snd_nxt))
1946 tp->snd_nxt = onxt;
1947 goto drop;
1948 } else if (tcp_do_rfc3042) {
1949 u_long oldcwnd = tp->snd_cwnd;
1950 tcp_seq oldsndmax = tp->snd_max;
1951 u_int sent;
1952
1953 KASSERT(tp->t_dupacks == 1 ||
1954 tp->t_dupacks == 2,
1955 ("dupacks not 1 or 2"));
1956 if (tp->t_dupacks == 1)
1957 tp->snd_limited = 0;
1958 tp->snd_cwnd =
1959 (tp->snd_nxt - tp->snd_una) +
1960 (tp->t_dupacks - tp->snd_limited) *
1961 tp->t_maxseg;
1962 (void) tcp_output(tp);
1963 sent = tp->snd_max - oldsndmax;
1964 if (sent > tp->t_maxseg) {
1965 KASSERT((tp->t_dupacks == 2 &&
1966 tp->snd_limited == 0) ||
1967 (sent == tp->t_maxseg + 1 &&
1968 tp->t_flags & TF_SENTFIN),
1969 ("sent too much"));
1970 tp->snd_limited = 2;
1971 } else if (sent > 0)
1972 ++tp->snd_limited;
1973 tp->snd_cwnd = oldcwnd;
1974 goto drop;
1975 }
1976 } else
1977 tp->t_dupacks = 0;
1978 break;
1979 }
1980
1981 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1982
1983 /*
1984 * If the congestion window was inflated to account
1985 * for the other side's cached packets, retract it.
1986 */
1987 if (tcp_do_newreno || tp->sack_enable) {
1988 if (IN_FASTRECOVERY(tp)) {
1989 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1990 if (tp->sack_enable)
1991 tcp_sack_partialack(tp, th);
1992 else
1993 tcp_newreno_partial_ack(tp, th);
1994 } else {
1995 /*
1996 * Out of fast recovery.
1997 * Window inflation should have left us
1998 * with approximately snd_ssthresh
1999 * outstanding data.
2000 * But in case we would be inclined to
2001 * send a burst, better to do it via
2002 * the slow start mechanism.
2003 */
2004 if (SEQ_GT(th->th_ack +
2005 tp->snd_ssthresh,
2006 tp->snd_max))
2007 tp->snd_cwnd = tp->snd_max -
2008 th->th_ack +
2009 tp->t_maxseg;
2010 else
2011 tp->snd_cwnd = tp->snd_ssthresh;
2012 }
2013 }
2014 } else {
2015 if (tp->t_dupacks >= tcprexmtthresh &&
2016 tp->snd_cwnd > tp->snd_ssthresh)
2017 tp->snd_cwnd = tp->snd_ssthresh;
2018 }
2019 tp->t_dupacks = 0;
2020 /*
2021 * If we reach this point, ACK is not a duplicate,
2022 * i.e., it ACKs something we sent.
2023 */
2024 if (tp->t_flags & TF_NEEDSYN) {
2025 /*
2026 * T/TCP: Connection was half-synchronized, and our
2027 * SYN has been ACK'd (so connection is now fully
2028 * synchronized). Go to non-starred state,
2029 * increment snd_una for ACK of SYN, and check if
2030 * we can do window scaling.
2031 */
2032 tp->t_flags &= ~TF_NEEDSYN;
2033 tp->snd_una++;
2034 /* Do window scaling? */
2035 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2036 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2037 tp->snd_scale = tp->requested_s_scale;
2038 tp->rcv_scale = tp->request_r_scale;
2039 }
2040 }
2041
2042 process_ACK:
2043 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2044 "locked"));
2045 INP_LOCK_ASSERT(inp);
2046
2047 acked = th->th_ack - tp->snd_una;
2048 tcpstat.tcps_rcvackpack++;
2049 tcpstat.tcps_rcvackbyte += acked;
2050
2051 /*
2052 * If we just performed our first retransmit, and the ACK
2053 * arrives within our recovery window, then it was a mistake
2054 * to do the retransmit in the first place. Recover our
2055 * original cwnd and ssthresh, and proceed to transmit where
2056 * we left off.
2057 */
2058 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2059 ++tcpstat.tcps_sndrexmitbad;
2060 tp->snd_cwnd = tp->snd_cwnd_prev;
2061 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2062 tp->snd_recover = tp->snd_recover_prev;
2063 if (tp->t_flags & TF_WASFRECOVERY)
2064 ENTER_FASTRECOVERY(tp);
2065 tp->snd_nxt = tp->snd_max;
2066 tp->t_badrxtwin = 0; /* XXX probably not required */
2067 }
2068
2069 /*
2070 * If we have a timestamp reply, update smoothed
2071 * round trip time. If no timestamp is present but
2072 * transmit timer is running and timed sequence
2073 * number was acked, update smoothed round trip time.
2074 * Since we now have an rtt measurement, cancel the
2075 * timer backoff (cf., Phil Karn's retransmit alg.).
2076 * Recompute the initial retransmit timer.
2077 *
2078 * Some boxes send broken timestamp replies
2079 * during the SYN+ACK phase, ignore
2080 * timestamps of 0 or we could calculate a
2081 * huge RTT and blow up the retransmit timer.
2082 */
2083 if ((to.to_flags & TOF_TS) != 0 &&
2084 to.to_tsecr) {
2085 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2086 tp->t_rttlow = ticks - to.to_tsecr;
2087 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2088 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2089 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2090 tp->t_rttlow = ticks - tp->t_rtttime;
2091 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2092 }
2093 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2094
2095 /*
2096 * If all outstanding data is acked, stop retransmit
2097 * timer and remember to restart (more output or persist).
2098 * If there is more data to be acked, restart retransmit
2099 * timer, using current (possibly backed-off) value.
2100 */
2101 if (th->th_ack == tp->snd_max) {
2102 callout_stop(tp->tt_rexmt);
2103 needoutput = 1;
2104 } else if (!callout_active(tp->tt_persist))
2105 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2106 tcp_timer_rexmt, tp);
2107
2108 /*
2109 * If no data (only SYN) was ACK'd,
2110 * skip rest of ACK processing.
2111 */
2112 if (acked == 0)
2113 goto step6;
2114
2115 /*
2116 * When new data is acked, open the congestion window.
2117 * If the window gives us less than ssthresh packets
2118 * in flight, open exponentially (maxseg per packet).
2119 * Otherwise open linearly: maxseg per window
2120 * (maxseg^2 / cwnd per packet).
2121 */
2122 if ((!tcp_do_newreno && !tp->sack_enable) ||
2123 !IN_FASTRECOVERY(tp)) {
2124 register u_int cw = tp->snd_cwnd;
2125 register u_int incr = tp->t_maxseg;
2126 if (cw > tp->snd_ssthresh)
2127 incr = incr * incr / cw;
2128 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2129 }
2130 SOCKBUF_LOCK(&so->so_snd);
2131 if (acked > so->so_snd.sb_cc) {
2132 tp->snd_wnd -= so->so_snd.sb_cc;
2133 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2134 ourfinisacked = 1;
2135 } else {
2136 sbdrop_locked(&so->so_snd, acked);
2137 tp->snd_wnd -= acked;
2138 ourfinisacked = 0;
2139 }
2140 sowwakeup_locked(so);
2141 /* detect una wraparound */
2142 if ((tcp_do_newreno || tp->sack_enable) &&
2143 !IN_FASTRECOVERY(tp) &&
2144 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2145 SEQ_LEQ(th->th_ack, tp->snd_recover))
2146 tp->snd_recover = th->th_ack - 1;
2147 if ((tcp_do_newreno || tp->sack_enable) &&
2148 IN_FASTRECOVERY(tp) &&
2149 SEQ_GEQ(th->th_ack, tp->snd_recover))
2150 EXIT_FASTRECOVERY(tp);
2151 tp->snd_una = th->th_ack;
2152 if (tp->sack_enable) {
2153 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2154 tp->snd_recover = tp->snd_una;
2155 }
2156 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2157 tp->snd_nxt = tp->snd_una;
2158
2159 switch (tp->t_state) {
2160
2161 /*
2162 * In FIN_WAIT_1 STATE in addition to the processing
2163 * for the ESTABLISHED state if our FIN is now acknowledged
2164 * then enter FIN_WAIT_2.
2165 */
2166 case TCPS_FIN_WAIT_1:
2167 if (ourfinisacked) {
2168 /*
2169 * If we can't receive any more
2170 * data, then closing user can proceed.
2171 * Starting the timer is contrary to the
2172 * specification, but if we don't get a FIN
2173 * we'll hang forever.
2174 */
2175 /* XXXjl
2176 * we should release the tp also, and use a
2177 * compressed state.
2178 */
2179 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2180 soisdisconnected(so);
2181 callout_reset(tp->tt_2msl, tcp_maxidle,
2182 tcp_timer_2msl, tp);
2183 }
2184 tp->t_state = TCPS_FIN_WAIT_2;
2185 }
2186 break;
2187
2188 /*
2189 * In CLOSING STATE in addition to the processing for
2190 * the ESTABLISHED state if the ACK acknowledges our FIN
2191 * then enter the TIME-WAIT state, otherwise ignore
2192 * the segment.
2193 */
2194 case TCPS_CLOSING:
2195 if (ourfinisacked) {
2196 KASSERT(headlocked, ("tcp_input: process_ACK: "
2197 "head not locked"));
2198 tcp_twstart(tp);
2199 INP_INFO_WUNLOCK(&tcbinfo);
2200 m_freem(m);
2201 return;
2202 }
2203 break;
2204
2205 /*
2206 * In LAST_ACK, we may still be waiting for data to drain
2207 * and/or to be acked, as well as for the ack of our FIN.
2208 * If our FIN is now acknowledged, delete the TCB,
2209 * enter the closed state and return.
2210 */
2211 case TCPS_LAST_ACK:
2212 if (ourfinisacked) {
2213 KASSERT(headlocked, ("tcp_input: process_ACK:"
2214 " tcp_close: head not locked"));
2215 tp = tcp_close(tp);
2216 goto drop;
2217 }
2218 break;
2219
2220 /*
2221 * In TIME_WAIT state the only thing that should arrive
2222 * is a retransmission of the remote FIN. Acknowledge
2223 * it and restart the finack timer.
2224 */
2225 case TCPS_TIME_WAIT:
2226 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2227 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2228 tcp_timer_2msl, tp);
2229 goto dropafterack;
2230 }
2231 }
2232
2233 step6:
2234 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2235 INP_LOCK_ASSERT(inp);
2236
2237 /*
2238 * Update window information.
2239 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2240 */
2241 if ((thflags & TH_ACK) &&
2242 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2243 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2244 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2245 /* keep track of pure window updates */
2246 if (tlen == 0 &&
2247 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2248 tcpstat.tcps_rcvwinupd++;
2249 tp->snd_wnd = tiwin;
2250 tp->snd_wl1 = th->th_seq;
2251 tp->snd_wl2 = th->th_ack;
2252 if (tp->snd_wnd > tp->max_sndwnd)
2253 tp->max_sndwnd = tp->snd_wnd;
2254 needoutput = 1;
2255 }
2256
2257 /*
2258 * Process segments with URG.
2259 */
2260 if ((thflags & TH_URG) && th->th_urp &&
2261 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2262 /*
2263 * This is a kludge, but if we receive and accept
2264 * random urgent pointers, we'll crash in
2265 * soreceive. It's hard to imagine someone
2266 * actually wanting to send this much urgent data.
2267 */
2268 SOCKBUF_LOCK(&so->so_rcv);
2269 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2270 th->th_urp = 0; /* XXX */
2271 thflags &= ~TH_URG; /* XXX */
2272 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2273 goto dodata; /* XXX */
2274 }
2275 /*
2276 * If this segment advances the known urgent pointer,
2277 * then mark the data stream. This should not happen
2278 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2279 * a FIN has been received from the remote side.
2280 * In these states we ignore the URG.
2281 *
2282 * According to RFC961 (Assigned Protocols),
2283 * the urgent pointer points to the last octet
2284 * of urgent data. We continue, however,
2285 * to consider it to indicate the first octet
2286 * of data past the urgent section as the original
2287 * spec states (in one of two places).
2288 */
2289 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2290 tp->rcv_up = th->th_seq + th->th_urp;
2291 so->so_oobmark = so->so_rcv.sb_cc +
2292 (tp->rcv_up - tp->rcv_nxt) - 1;
2293 if (so->so_oobmark == 0)
2294 so->so_rcv.sb_state |= SBS_RCVATMARK;
2295 sohasoutofband(so);
2296 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2297 }
2298 SOCKBUF_UNLOCK(&so->so_rcv);
2299 /*
2300 * Remove out of band data so doesn't get presented to user.
2301 * This can happen independent of advancing the URG pointer,
2302 * but if two URG's are pending at once, some out-of-band
2303 * data may creep in... ick.
2304 */
2305 if (th->th_urp <= (u_long)tlen &&
2306 !(so->so_options & SO_OOBINLINE)) {
2307 /* hdr drop is delayed */
2308 tcp_pulloutofband(so, th, m, drop_hdrlen);
2309 }
2310 } else {
2311 /*
2312 * If no out of band data is expected,
2313 * pull receive urgent pointer along
2314 * with the receive window.
2315 */
2316 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2317 tp->rcv_up = tp->rcv_nxt;
2318 }
2319 dodata: /* XXX */
2320 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2321 INP_LOCK_ASSERT(inp);
2322
2323 /*
2324 * Process the segment text, merging it into the TCP sequencing queue,
2325 * and arranging for acknowledgment of receipt if necessary.
2326 * This process logically involves adjusting tp->rcv_wnd as data
2327 * is presented to the user (this happens in tcp_usrreq.c,
2328 * case PRU_RCVD). If a FIN has already been received on this
2329 * connection then we just ignore the text.
2330 */
2331 if ((tlen || (thflags & TH_FIN)) &&
2332 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2333 tcp_seq save_start = th->th_seq;
2334 tcp_seq save_end = th->th_seq + tlen;
2335 m_adj(m, drop_hdrlen); /* delayed header drop */
2336 /*
2337 * Insert segment which includes th into TCP reassembly queue
2338 * with control block tp. Set thflags to whether reassembly now
2339 * includes a segment with FIN. This handles the common case
2340 * inline (segment is the next to be received on an established
2341 * connection, and the queue is empty), avoiding linkage into
2342 * and removal from the queue and repetition of various
2343 * conversions.
2344 * Set DELACK for segments received in order, but ack
2345 * immediately when segments are out of order (so
2346 * fast retransmit can work).
2347 */
2348 if (th->th_seq == tp->rcv_nxt &&
2349 LIST_EMPTY(&tp->t_segq) &&
2350 TCPS_HAVEESTABLISHED(tp->t_state)) {
2351 if (DELAY_ACK(tp))
2352 tp->t_flags |= TF_DELACK;
2353 else
2354 tp->t_flags |= TF_ACKNOW;
2355 tp->rcv_nxt += tlen;
2356 thflags = th->th_flags & TH_FIN;
2357 tcpstat.tcps_rcvpack++;
2358 tcpstat.tcps_rcvbyte += tlen;
2359 ND6_HINT(tp);
2360 SOCKBUF_LOCK(&so->so_rcv);
2361 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2362 m_freem(m);
2363 else
2364 sbappendstream_locked(&so->so_rcv, m);
2365 sorwakeup_locked(so);
2366 } else {
2367 thflags = tcp_reass(tp, th, &tlen, m);
2368 tp->t_flags |= TF_ACKNOW;
2369 }
2370 if (tlen > 0 && tp->sack_enable)
2371 tcp_update_sack_list(tp, save_start, save_end);
2372 /*
2373 * Note the amount of data that peer has sent into
2374 * our window, in order to estimate the sender's
2375 * buffer size.
2376 */
2377 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2378 } else {
2379 m_freem(m);
2380 thflags &= ~TH_FIN;
2381 }
2382
2383 /*
2384 * If FIN is received ACK the FIN and let the user know
2385 * that the connection is closing.
2386 */
2387 if (thflags & TH_FIN) {
2388 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2389 socantrcvmore(so);
2390 /*
2391 * If connection is half-synchronized
2392 * (ie NEEDSYN flag on) then delay ACK,
2393 * so it may be piggybacked when SYN is sent.
2394 * Otherwise, since we received a FIN then no
2395 * more input can be expected, send ACK now.
2396 */
2397 if (tp->t_flags & TF_NEEDSYN)
2398 tp->t_flags |= TF_DELACK;
2399 else
2400 tp->t_flags |= TF_ACKNOW;
2401 tp->rcv_nxt++;
2402 }
2403 switch (tp->t_state) {
2404
2405 /*
2406 * In SYN_RECEIVED and ESTABLISHED STATES
2407 * enter the CLOSE_WAIT state.
2408 */
2409 case TCPS_SYN_RECEIVED:
2410 tp->t_starttime = ticks;
2411 /*FALLTHROUGH*/
2412 case TCPS_ESTABLISHED:
2413 tp->t_state = TCPS_CLOSE_WAIT;
2414 break;
2415
2416 /*
2417 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2418 * enter the CLOSING state.
2419 */
2420 case TCPS_FIN_WAIT_1:
2421 tp->t_state = TCPS_CLOSING;
2422 break;
2423
2424 /*
2425 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2426 * starting the time-wait timer, turning off the other
2427 * standard timers.
2428 */
2429 case TCPS_FIN_WAIT_2:
2430 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2431 "TCP_FIN_WAIT_2: head not locked"));
2432 tcp_twstart(tp);
2433 INP_INFO_WUNLOCK(&tcbinfo);
2434 return;
2435
2436 /*
2437 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2438 */
2439 case TCPS_TIME_WAIT:
2440 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2441 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2442 tcp_timer_2msl, tp);
2443 break;
2444 }
2445 }
2446 INP_INFO_WUNLOCK(&tcbinfo);
2447 headlocked = 0;
2448 #ifdef TCPDEBUG
2449 if (so->so_options & SO_DEBUG)
2450 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2451 &tcp_savetcp, 0);
2452 #endif
2453
2454 /*
2455 * Return any desired output.
2456 */
2457 if (needoutput || (tp->t_flags & TF_ACKNOW))
2458 (void) tcp_output(tp);
2459
2460 check_delack:
2461 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2462 INP_LOCK_ASSERT(inp);
2463 if (tp->t_flags & TF_DELACK) {
2464 tp->t_flags &= ~TF_DELACK;
2465 callout_reset(tp->tt_delack, tcp_delacktime,
2466 tcp_timer_delack, tp);
2467 }
2468 INP_UNLOCK(inp);
2469 return;
2470
2471 dropafterack:
2472 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2473 /*
2474 * Generate an ACK dropping incoming segment if it occupies
2475 * sequence space, where the ACK reflects our state.
2476 *
2477 * We can now skip the test for the RST flag since all
2478 * paths to this code happen after packets containing
2479 * RST have been dropped.
2480 *
2481 * In the SYN-RECEIVED state, don't send an ACK unless the
2482 * segment we received passes the SYN-RECEIVED ACK test.
2483 * If it fails send a RST. This breaks the loop in the
2484 * "LAND" DoS attack, and also prevents an ACK storm
2485 * between two listening ports that have been sent forged
2486 * SYN segments, each with the source address of the other.
2487 */
2488 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2489 (SEQ_GT(tp->snd_una, th->th_ack) ||
2490 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2491 rstreason = BANDLIM_RST_OPENPORT;
2492 goto dropwithreset;
2493 }
2494 #ifdef TCPDEBUG
2495 if (so->so_options & SO_DEBUG)
2496 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2497 &tcp_savetcp, 0);
2498 #endif
2499 KASSERT(headlocked, ("headlocked should be 1"));
2500 INP_INFO_WUNLOCK(&tcbinfo);
2501 tp->t_flags |= TF_ACKNOW;
2502 (void) tcp_output(tp);
2503 INP_UNLOCK(inp);
2504 m_freem(m);
2505 return;
2506
2507 dropwithreset:
2508 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2509 /*
2510 * Generate a RST, dropping incoming segment.
2511 * Make ACK acceptable to originator of segment.
2512 * Don't bother to respond if destination was broadcast/multicast.
2513 */
2514 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2515 goto drop;
2516 if (isipv6) {
2517 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2518 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2519 goto drop;
2520 } else {
2521 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2522 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2523 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2524 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2525 goto drop;
2526 }
2527 /* IPv6 anycast check is done at tcp6_input() */
2528
2529 /*
2530 * Perform bandwidth limiting.
2531 */
2532 if (badport_bandlim(rstreason) < 0)
2533 goto drop;
2534
2535 #ifdef TCPDEBUG
2536 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2537 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2538 &tcp_savetcp, 0);
2539 #endif
2540
2541 if (thflags & TH_ACK)
2542 /* mtod() below is safe as long as hdr dropping is delayed */
2543 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2544 TH_RST);
2545 else {
2546 if (thflags & TH_SYN)
2547 tlen++;
2548 /* mtod() below is safe as long as hdr dropping is delayed */
2549 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2550 (tcp_seq)0, TH_RST|TH_ACK);
2551 }
2552
2553 if (tp)
2554 INP_UNLOCK(inp);
2555 if (headlocked)
2556 INP_INFO_WUNLOCK(&tcbinfo);
2557 return;
2558
2559 drop:
2560 /*
2561 * Drop space held by incoming segment and return.
2562 */
2563 #ifdef TCPDEBUG
2564 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2565 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2566 &tcp_savetcp, 0);
2567 #endif
2568 if (tp)
2569 INP_UNLOCK(inp);
2570 if (headlocked)
2571 INP_INFO_WUNLOCK(&tcbinfo);
2572 m_freem(m);
2573 return;
2574 }
2575
2576 /*
2577 * Parse TCP options and place in tcpopt.
2578 */
2579 static void
2580 tcp_dooptions(to, cp, cnt, is_syn)
2581 struct tcpopt *to;
2582 u_char *cp;
2583 int cnt;
2584 int is_syn;
2585 {
2586 int opt, optlen;
2587
2588 to->to_flags = 0;
2589 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2590 opt = cp[0];
2591 if (opt == TCPOPT_EOL)
2592 break;
2593 if (opt == TCPOPT_NOP)
2594 optlen = 1;
2595 else {
2596 if (cnt < 2)
2597 break;
2598 optlen = cp[1];
2599 if (optlen < 2 || optlen > cnt)
2600 break;
2601 }
2602 switch (opt) {
2603 case TCPOPT_MAXSEG:
2604 if (optlen != TCPOLEN_MAXSEG)
2605 continue;
2606 if (!is_syn)
2607 continue;
2608 to->to_flags |= TOF_MSS;
2609 bcopy((char *)cp + 2,
2610 (char *)&to->to_mss, sizeof(to->to_mss));
2611 to->to_mss = ntohs(to->to_mss);
2612 break;
2613 case TCPOPT_WINDOW:
2614 if (optlen != TCPOLEN_WINDOW)
2615 continue;
2616 if (! is_syn)
2617 continue;
2618 to->to_flags |= TOF_SCALE;
2619 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2620 break;
2621 case TCPOPT_TIMESTAMP:
2622 if (optlen != TCPOLEN_TIMESTAMP)
2623 continue;
2624 to->to_flags |= TOF_TS;
2625 bcopy((char *)cp + 2,
2626 (char *)&to->to_tsval, sizeof(to->to_tsval));
2627 to->to_tsval = ntohl(to->to_tsval);
2628 bcopy((char *)cp + 6,
2629 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2630 to->to_tsecr = ntohl(to->to_tsecr);
2631 /*
2632 * If echoed timestamp is later than the current time,
2633 * fall back to non RFC1323 RTT calculation.
2634 */
2635 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2636 to->to_tsecr = 0;
2637 break;
2638 #ifdef TCP_SIGNATURE
2639 /*
2640 * XXX In order to reply to a host which has set the
2641 * TCP_SIGNATURE option in its initial SYN, we have to
2642 * record the fact that the option was observed here
2643 * for the syncache code to perform the correct response.
2644 */
2645 case TCPOPT_SIGNATURE:
2646 if (optlen != TCPOLEN_SIGNATURE)
2647 continue;
2648 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2649 break;
2650 #endif
2651 case TCPOPT_SACK_PERMITTED:
2652 if (!tcp_do_sack ||
2653 optlen != TCPOLEN_SACK_PERMITTED)
2654 continue;
2655 if (is_syn) {
2656 /* MUST only be set on SYN */
2657 to->to_flags |= TOF_SACK;
2658 }
2659 break;
2660 case TCPOPT_SACK:
2661 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2662 continue;
2663 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2664 to->to_sacks = cp + 2;
2665 tcpstat.tcps_sack_rcv_blocks++;
2666 break;
2667 default:
2668 continue;
2669 }
2670 }
2671 }
2672
2673 /*
2674 * Pull out of band byte out of a segment so
2675 * it doesn't appear in the user's data queue.
2676 * It is still reflected in the segment length for
2677 * sequencing purposes.
2678 */
2679 static void
2680 tcp_pulloutofband(so, th, m, off)
2681 struct socket *so;
2682 struct tcphdr *th;
2683 register struct mbuf *m;
2684 int off; /* delayed to be droped hdrlen */
2685 {
2686 int cnt = off + th->th_urp - 1;
2687
2688 while (cnt >= 0) {
2689 if (m->m_len > cnt) {
2690 char *cp = mtod(m, caddr_t) + cnt;
2691 struct tcpcb *tp = sototcpcb(so);
2692
2693 tp->t_iobc = *cp;
2694 tp->t_oobflags |= TCPOOB_HAVEDATA;
2695 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2696 m->m_len--;
2697 if (m->m_flags & M_PKTHDR)
2698 m->m_pkthdr.len--;
2699 return;
2700 }
2701 cnt -= m->m_len;
2702 m = m->m_next;
2703 if (m == 0)
2704 break;
2705 }
2706 panic("tcp_pulloutofband");
2707 }
2708
2709 /*
2710 * Collect new round-trip time estimate
2711 * and update averages and current timeout.
2712 */
2713 static void
2714 tcp_xmit_timer(tp, rtt)
2715 register struct tcpcb *tp;
2716 int rtt;
2717 {
2718 register int delta;
2719
2720 INP_LOCK_ASSERT(tp->t_inpcb);
2721
2722 tcpstat.tcps_rttupdated++;
2723 tp->t_rttupdated++;
2724 if (tp->t_srtt != 0) {
2725 /*
2726 * srtt is stored as fixed point with 5 bits after the
2727 * binary point (i.e., scaled by 8). The following magic
2728 * is equivalent to the smoothing algorithm in rfc793 with
2729 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2730 * point). Adjust rtt to origin 0.
2731 */
2732 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2733 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2734
2735 if ((tp->t_srtt += delta) <= 0)
2736 tp->t_srtt = 1;
2737
2738 /*
2739 * We accumulate a smoothed rtt variance (actually, a
2740 * smoothed mean difference), then set the retransmit
2741 * timer to smoothed rtt + 4 times the smoothed variance.
2742 * rttvar is stored as fixed point with 4 bits after the
2743 * binary point (scaled by 16). The following is
2744 * equivalent to rfc793 smoothing with an alpha of .75
2745 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2746 * rfc793's wired-in beta.
2747 */
2748 if (delta < 0)
2749 delta = -delta;
2750 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2751 if ((tp->t_rttvar += delta) <= 0)
2752 tp->t_rttvar = 1;
2753 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2754 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2755 } else {
2756 /*
2757 * No rtt measurement yet - use the unsmoothed rtt.
2758 * Set the variance to half the rtt (so our first
2759 * retransmit happens at 3*rtt).
2760 */
2761 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2762 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2763 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2764 }
2765 tp->t_rtttime = 0;
2766 tp->t_rxtshift = 0;
2767
2768 /*
2769 * the retransmit should happen at rtt + 4 * rttvar.
2770 * Because of the way we do the smoothing, srtt and rttvar
2771 * will each average +1/2 tick of bias. When we compute
2772 * the retransmit timer, we want 1/2 tick of rounding and
2773 * 1 extra tick because of +-1/2 tick uncertainty in the
2774 * firing of the timer. The bias will give us exactly the
2775 * 1.5 tick we need. But, because the bias is
2776 * statistical, we have to test that we don't drop below
2777 * the minimum feasible timer (which is 2 ticks).
2778 */
2779 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2780 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2781
2782 /*
2783 * We received an ack for a packet that wasn't retransmitted;
2784 * it is probably safe to discard any error indications we've
2785 * received recently. This isn't quite right, but close enough
2786 * for now (a route might have failed after we sent a segment,
2787 * and the return path might not be symmetrical).
2788 */
2789 tp->t_softerror = 0;
2790 }
2791
2792 /*
2793 * Determine a reasonable value for maxseg size.
2794 * If the route is known, check route for mtu.
2795 * If none, use an mss that can be handled on the outgoing
2796 * interface without forcing IP to fragment; if bigger than
2797 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2798 * to utilize large mbufs. If no route is found, route has no mtu,
2799 * or the destination isn't local, use a default, hopefully conservative
2800 * size (usually 512 or the default IP max size, but no more than the mtu
2801 * of the interface), as we can't discover anything about intervening
2802 * gateways or networks. We also initialize the congestion/slow start
2803 * window to be a single segment if the destination isn't local.
2804 * While looking at the routing entry, we also initialize other path-dependent
2805 * parameters from pre-set or cached values in the routing entry.
2806 *
2807 * Also take into account the space needed for options that we
2808 * send regularly. Make maxseg shorter by that amount to assure
2809 * that we can send maxseg amount of data even when the options
2810 * are present. Store the upper limit of the length of options plus
2811 * data in maxopd.
2812 *
2813 *
2814 * In case of T/TCP, we call this routine during implicit connection
2815 * setup as well (offer = -1), to initialize maxseg from the cached
2816 * MSS of our peer.
2817 *
2818 * NOTE that this routine is only called when we process an incoming
2819 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2820 */
2821 void
2822 tcp_mss(tp, offer)
2823 struct tcpcb *tp;
2824 int offer;
2825 {
2826 int rtt, mss;
2827 u_long bufsize;
2828 u_long maxmtu;
2829 struct inpcb *inp = tp->t_inpcb;
2830 struct socket *so;
2831 struct hc_metrics_lite metrics;
2832 int origoffer = offer;
2833 #ifdef INET6
2834 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2835 size_t min_protoh = isipv6 ?
2836 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2837 sizeof (struct tcpiphdr);
2838 #else
2839 const size_t min_protoh = sizeof(struct tcpiphdr);
2840 #endif
2841
2842 /* initialize */
2843 #ifdef INET6
2844 if (isipv6) {
2845 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2846 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2847 } else
2848 #endif
2849 {
2850 maxmtu = tcp_maxmtu(&inp->inp_inc);
2851 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2852 }
2853 so = inp->inp_socket;
2854
2855 /*
2856 * no route to sender, stay with default mss and return
2857 */
2858 if (maxmtu == 0)
2859 return;
2860
2861 /* what have we got? */
2862 switch (offer) {
2863 case 0:
2864 /*
2865 * Offer == 0 means that there was no MSS on the SYN
2866 * segment, in this case we use tcp_mssdflt.
2867 */
2868 offer =
2869 #ifdef INET6
2870 isipv6 ? tcp_v6mssdflt :
2871 #endif
2872 tcp_mssdflt;
2873 break;
2874
2875 case -1:
2876 /*
2877 * Offer == -1 means that we didn't receive SYN yet.
2878 */
2879 /* FALLTHROUGH */
2880
2881 default:
2882 /*
2883 * Prevent DoS attack with too small MSS. Round up
2884 * to at least minmss.
2885 */
2886 offer = max(offer, tcp_minmss);
2887 /*
2888 * Sanity check: make sure that maxopd will be large
2889 * enough to allow some data on segments even if the
2890 * all the option space is used (40bytes). Otherwise
2891 * funny things may happen in tcp_output.
2892 */
2893 offer = max(offer, 64);
2894 }
2895
2896 /*
2897 * rmx information is now retrieved from tcp_hostcache
2898 */
2899 tcp_hc_get(&inp->inp_inc, &metrics);
2900
2901 /*
2902 * if there's a discovered mtu int tcp hostcache, use it
2903 * else, use the link mtu.
2904 */
2905 if (metrics.rmx_mtu)
2906 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2907 else {
2908 #ifdef INET6
2909 if (isipv6) {
2910 mss = maxmtu - min_protoh;
2911 if (!path_mtu_discovery &&
2912 !in6_localaddr(&inp->in6p_faddr))
2913 mss = min(mss, tcp_v6mssdflt);
2914 } else
2915 #endif
2916 {
2917 mss = maxmtu - min_protoh;
2918 if (!path_mtu_discovery &&
2919 !in_localaddr(inp->inp_faddr))
2920 mss = min(mss, tcp_mssdflt);
2921 }
2922 }
2923 mss = min(mss, offer);
2924
2925 /*
2926 * maxopd stores the maximum length of data AND options
2927 * in a segment; maxseg is the amount of data in a normal
2928 * segment. We need to store this value (maxopd) apart
2929 * from maxseg, because now every segment carries options
2930 * and thus we normally have somewhat less data in segments.
2931 */
2932 tp->t_maxopd = mss;
2933
2934 /*
2935 * origoffer==-1 indicates, that no segments were received yet.
2936 * In this case we just guess.
2937 */
2938 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2939 (origoffer == -1 ||
2940 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2941 mss -= TCPOLEN_TSTAMP_APPA;
2942 tp->t_maxseg = mss;
2943
2944 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2945 if (mss > MCLBYTES)
2946 mss &= ~(MCLBYTES-1);
2947 #else
2948 if (mss > MCLBYTES)
2949 mss = mss / MCLBYTES * MCLBYTES;
2950 #endif
2951 tp->t_maxseg = mss;
2952
2953 /*
2954 * If there's a pipesize, change the socket buffer to that size,
2955 * don't change if sb_hiwat is different than default (then it
2956 * has been changed on purpose with setsockopt).
2957 * Make the socket buffers an integral number of mss units;
2958 * if the mss is larger than the socket buffer, decrease the mss.
2959 */
2960 SOCKBUF_LOCK(&so->so_snd);
2961 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2962 bufsize = metrics.rmx_sendpipe;
2963 else
2964 bufsize = so->so_snd.sb_hiwat;
2965 if (bufsize < mss)
2966 mss = bufsize;
2967 else {
2968 bufsize = roundup(bufsize, mss);
2969 if (bufsize > sb_max)
2970 bufsize = sb_max;
2971 if (bufsize > so->so_snd.sb_hiwat)
2972 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
2973 }
2974 SOCKBUF_UNLOCK(&so->so_snd);
2975 tp->t_maxseg = mss;
2976
2977 SOCKBUF_LOCK(&so->so_rcv);
2978 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2979 bufsize = metrics.rmx_recvpipe;
2980 else
2981 bufsize = so->so_rcv.sb_hiwat;
2982 if (bufsize > mss) {
2983 bufsize = roundup(bufsize, mss);
2984 if (bufsize > sb_max)
2985 bufsize = sb_max;
2986 if (bufsize > so->so_rcv.sb_hiwat)
2987 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
2988 }
2989 SOCKBUF_UNLOCK(&so->so_rcv);
2990 /*
2991 * While we're here, check the others too
2992 */
2993 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2994 tp->t_srtt = rtt;
2995 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2996 tcpstat.tcps_usedrtt++;
2997 if (metrics.rmx_rttvar) {
2998 tp->t_rttvar = metrics.rmx_rttvar;
2999 tcpstat.tcps_usedrttvar++;
3000 } else {
3001 /* default variation is +- 1 rtt */
3002 tp->t_rttvar =
3003 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3004 }
3005 TCPT_RANGESET(tp->t_rxtcur,
3006 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3007 tp->t_rttmin, TCPTV_REXMTMAX);
3008 }
3009 if (metrics.rmx_ssthresh) {
3010 /*
3011 * There's some sort of gateway or interface
3012 * buffer limit on the path. Use this to set
3013 * the slow start threshhold, but set the
3014 * threshold to no less than 2*mss.
3015 */
3016 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3017 tcpstat.tcps_usedssthresh++;
3018 }
3019 if (metrics.rmx_bandwidth)
3020 tp->snd_bandwidth = metrics.rmx_bandwidth;
3021
3022 /*
3023 * Set the slow-start flight size depending on whether this
3024 * is a local network or not.
3025 *
3026 * Extend this so we cache the cwnd too and retrieve it here.
3027 * Make cwnd even bigger than RFC3390 suggests but only if we
3028 * have previous experience with the remote host. Be careful
3029 * not make cwnd bigger than remote receive window or our own
3030 * send socket buffer. Maybe put some additional upper bound
3031 * on the retrieved cwnd. Should do incremental updates to
3032 * hostcache when cwnd collapses so next connection doesn't
3033 * overloads the path again.
3034 *
3035 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3036 * We currently check only in syncache_socket for that.
3037 */
3038 #define TCP_METRICS_CWND
3039 #ifdef TCP_METRICS_CWND
3040 if (metrics.rmx_cwnd)
3041 tp->snd_cwnd = max(mss,
3042 min(metrics.rmx_cwnd / 2,
3043 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3044 else
3045 #endif
3046 if (tcp_do_rfc3390)
3047 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3048 #ifdef INET6
3049 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3050 (!isipv6 && in_localaddr(inp->inp_faddr)))
3051 #else
3052 else if (in_localaddr(inp->inp_faddr))
3053 #endif
3054 tp->snd_cwnd = mss * ss_fltsz_local;
3055 else
3056 tp->snd_cwnd = mss * ss_fltsz;
3057 }
3058
3059 /*
3060 * Determine the MSS option to send on an outgoing SYN.
3061 */
3062 int
3063 tcp_mssopt(inc)
3064 struct in_conninfo *inc;
3065 {
3066 int mss = 0;
3067 u_long maxmtu = 0;
3068 u_long thcmtu = 0;
3069 size_t min_protoh;
3070 #ifdef INET6
3071 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3072 #endif
3073
3074 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3075
3076 #ifdef INET6
3077 if (isipv6) {
3078 mss = tcp_v6mssdflt;
3079 maxmtu = tcp_maxmtu6(inc);
3080 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3081 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3082 } else
3083 #endif
3084 {
3085 mss = tcp_mssdflt;
3086 maxmtu = tcp_maxmtu(inc);
3087 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3088 min_protoh = sizeof(struct tcpiphdr);
3089 }
3090 if (maxmtu && thcmtu)
3091 mss = min(maxmtu, thcmtu) - min_protoh;
3092 else if (maxmtu || thcmtu)
3093 mss = max(maxmtu, thcmtu) - min_protoh;
3094
3095 return (mss);
3096 }
3097
3098
3099 /*
3100 * On a partial ack arrives, force the retransmission of the
3101 * next unacknowledged segment. Do not clear tp->t_dupacks.
3102 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3103 * be started again.
3104 */
3105 static void
3106 tcp_newreno_partial_ack(tp, th)
3107 struct tcpcb *tp;
3108 struct tcphdr *th;
3109 {
3110 tcp_seq onxt = tp->snd_nxt;
3111 u_long ocwnd = tp->snd_cwnd;
3112
3113 callout_stop(tp->tt_rexmt);
3114 tp->t_rtttime = 0;
3115 tp->snd_nxt = th->th_ack;
3116 /*
3117 * Set snd_cwnd to one segment beyond acknowledged offset.
3118 * (tp->snd_una has not yet been updated when this function is called.)
3119 */
3120 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3121 tp->t_flags |= TF_ACKNOW;
3122 (void) tcp_output(tp);
3123 tp->snd_cwnd = ocwnd;
3124 if (SEQ_GT(onxt, tp->snd_nxt))
3125 tp->snd_nxt = onxt;
3126 /*
3127 * Partial window deflation. Relies on fact that tp->snd_una
3128 * not updated yet.
3129 */
3130 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3131 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3132 else
3133 tp->snd_cwnd = 0;
3134 tp->snd_cwnd += tp->t_maxseg;
3135 }
3136
3137 /*
3138 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3139 * looking for a pcb in the listen state. Returns 0 otherwise.
3140 */
3141 static int
3142 tcp_timewait(tw, to, th, m, tlen)
3143 struct tcptw *tw;
3144 struct tcpopt *to;
3145 struct tcphdr *th;
3146 struct mbuf *m;
3147 int tlen;
3148 {
3149 int thflags;
3150 tcp_seq seq;
3151 #ifdef INET6
3152 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3153 #else
3154 const int isipv6 = 0;
3155 #endif
3156
3157 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */
3158 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3159 INP_LOCK_ASSERT(tw->tw_inpcb);
3160
3161 thflags = th->th_flags;
3162
3163 /*
3164 * NOTE: for FIN_WAIT_2 (to be added later),
3165 * must validate sequence number before accepting RST
3166 */
3167
3168 /*
3169 * If the segment contains RST:
3170 * Drop the segment - see Stevens, vol. 2, p. 964 and
3171 * RFC 1337.
3172 */
3173 if (thflags & TH_RST)
3174 goto drop;
3175
3176 #if 0
3177 /* PAWS not needed at the moment */
3178 /*
3179 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3180 * and it's less than ts_recent, drop it.
3181 */
3182 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3183 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3184 if ((thflags & TH_ACK) == 0)
3185 goto drop;
3186 goto ack;
3187 }
3188 /*
3189 * ts_recent is never updated because we never accept new segments.
3190 */
3191 #endif
3192
3193 /*
3194 * If a new connection request is received
3195 * while in TIME_WAIT, drop the old connection
3196 * and start over if the sequence numbers
3197 * are above the previous ones.
3198 */
3199 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3200 (void) tcp_twclose(tw, 0);
3201 return (1);
3202 }
3203
3204 /*
3205 * Drop the the segment if it does not contain an ACK.
3206 */
3207 if ((thflags & TH_ACK) == 0)
3208 goto drop;
3209
3210 /*
3211 * Reset the 2MSL timer if this is a duplicate FIN.
3212 */
3213 if (thflags & TH_FIN) {
3214 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3215 if (seq + 1 == tw->rcv_nxt)
3216 tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3217 }
3218
3219 /*
3220 * Acknowledge the segment if it has data or is not a duplicate ACK.
3221 */
3222 if (thflags != TH_ACK || tlen != 0 ||
3223 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3224 tcp_twrespond(tw, TH_ACK);
3225 goto drop;
3226
3227 /*
3228 * Generate a RST, dropping incoming segment.
3229 * Make ACK acceptable to originator of segment.
3230 * Don't bother to respond if destination was broadcast/multicast.
3231 */
3232 if (m->m_flags & (M_BCAST|M_MCAST))
3233 goto drop;
3234 if (isipv6) {
3235 struct ip6_hdr *ip6;
3236
3237 /* IPv6 anycast check is done at tcp6_input() */
3238 ip6 = mtod(m, struct ip6_hdr *);
3239 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3240 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3241 goto drop;
3242 } else {
3243 struct ip *ip;
3244
3245 ip = mtod(m, struct ip *);
3246 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3247 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3248 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3249 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3250 goto drop;
3251 }
3252 if (thflags & TH_ACK) {
3253 tcp_respond(NULL,
3254 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3255 } else {
3256 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3257 tcp_respond(NULL,
3258 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3259 }
3260 INP_UNLOCK(tw->tw_inpcb);
3261 return (0);
3262
3263 drop:
3264 INP_UNLOCK(tw->tw_inpcb);
3265 m_freem(m);
3266 return (0);
3267 }
Cache object: cbedebaad36bc6676f7f286831b53954
|