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: src/sys/netinet/tcp_input.c,v 1.252.2.21 2005/07/05 19:25:42 ps Exp $
31 */
32
33 #include "opt_ipfw.h" /* for ipfw_fwd */
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 #include "opt_tcp_sack.h"
41
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/mac.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h> /* for proc0 declaration */
48 #include <sys/protosw.h>
49 #include <sys/signalvar.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/systm.h>
55
56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
57
58 #include <vm/uma.h>
59
60 #include <net/if.h>
61 #include <net/route.h>
62
63 #include <netinet/in.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
70 #include <netinet/ip_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
83 #ifdef TCPDEBUG
84 #include <netinet/tcp_debug.h>
85 #endif /* TCPDEBUG */
86
87 #ifdef FAST_IPSEC
88 #include <netipsec/ipsec.h>
89 #include <netipsec/ipsec6.h>
90 #endif /*FAST_IPSEC*/
91
92 #ifdef IPSEC
93 #include <netinet6/ipsec.h>
94 #include <netinet6/ipsec6.h>
95 #include <netkey/key.h>
96 #endif /*IPSEC*/
97
98 #include <machine/in_cksum.h>
99
100 static const int tcprexmtthresh = 3;
101 tcp_cc tcp_ccgen;
102
103 struct tcpstat tcpstat;
104 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
105 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
106
107 static int log_in_vain = 0;
108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
109 &log_in_vain, 0, "Log all incoming TCP connections");
110
111 static int blackhole = 0;
112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
113 &blackhole, 0, "Do not send RST when dropping refused connections");
114
115 int tcp_delack_enabled = 1;
116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
117 &tcp_delack_enabled, 0,
118 "Delay ACK to try and piggyback it onto a data packet");
119
120 #ifdef TCP_DROP_SYNFIN
121 static int drop_synfin = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
123 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
124 #endif
125
126 static int tcp_do_rfc3042 = 1;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
128 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
129
130 static int tcp_do_rfc3390 = 1;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
132 &tcp_do_rfc3390, 0,
133 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
134
135 static int tcp_insecure_rst = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
137 &tcp_insecure_rst, 0,
138 "Follow the old (insecure) criteria for accepting RST packets.");
139
140 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
141 "TCP Segment Reassembly Queue");
142
143 static int tcp_reass_maxseg = 0;
144 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
145 &tcp_reass_maxseg, 0,
146 "Global maximum number of TCP Segments in Reassembly Queue");
147
148 int tcp_reass_qsize = 0;
149 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
150 &tcp_reass_qsize, 0,
151 "Global number of TCP Segments currently in Reassembly Queue");
152
153 static int tcp_reass_maxqlen = 48;
154 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
155 &tcp_reass_maxqlen, 0,
156 "Maximum number of TCP Segments per individual Reassembly Queue");
157
158 static int tcp_reass_overflows = 0;
159 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
160 &tcp_reass_overflows, 0,
161 "Global number of TCP Segment Reassembly Queue Overflows");
162
163 struct inpcbhead tcb;
164 #define tcb6 tcb /* for KAME src sync over BSD*'s */
165 struct inpcbinfo tcbinfo;
166 struct mtx *tcbinfo_mtx;
167
168 static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *,
169 int, int, struct tcphdr *);
170
171 static void tcp_pulloutofband(struct socket *,
172 struct tcphdr *, struct mbuf *, int);
173 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
174 struct mbuf *);
175 static void tcp_xmit_timer(struct tcpcb *, int);
176 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
177 static int tcp_timewait(struct tcptw *, struct tcpopt *,
178 struct tcphdr *, struct mbuf *, int);
179
180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
181 #ifdef INET6
182 #define ND6_HINT(tp) \
183 do { \
184 if ((tp) && (tp)->t_inpcb && \
185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
186 nd6_nud_hint(NULL, NULL, 0); \
187 } while (0)
188 #else
189 #define ND6_HINT(tp)
190 #endif
191
192 /*
193 * Indicate whether this ack should be delayed. We can delay the ack if
194 * - there is no delayed ack timer in progress and
195 * - our last ack wasn't a 0-sized window. We never want to delay
196 * the ack that opens up a 0-sized window and
197 * - delayed acks are enabled or
198 * - this is a half-synchronized T/TCP connection.
199 */
200 #define DELAY_ACK(tp) \
201 ((!callout_active(tp->tt_delack) && \
202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
204
205 /* Initialize TCP reassembly queue */
206 uma_zone_t tcp_reass_zone;
207 void
208 tcp_reass_init()
209 {
210 tcp_reass_maxseg = nmbclusters / 16;
211 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
212 &tcp_reass_maxseg);
213 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
215 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
216 }
217
218 static int
219 tcp_reass(tp, th, tlenp, m)
220 register struct tcpcb *tp;
221 register struct tcphdr *th;
222 int *tlenp;
223 struct mbuf *m;
224 {
225 struct tseg_qent *q;
226 struct tseg_qent *p = NULL;
227 struct tseg_qent *nq;
228 struct tseg_qent *te = NULL;
229 struct socket *so = tp->t_inpcb->inp_socket;
230 int flags;
231
232 INP_LOCK_ASSERT(tp->t_inpcb);
233
234 /*
235 * XXX: tcp_reass() is rather inefficient with its data structures
236 * and should be rewritten (see NetBSD for optimizations). While
237 * doing that it should move to its own file tcp_reass.c.
238 */
239
240 /*
241 * Call with th==NULL after become established to
242 * force pre-ESTABLISHED data up to user socket.
243 */
244 if (th == NULL)
245 goto present;
246
247 /*
248 * Limit the number of segments in the reassembly queue to prevent
249 * holding on to too many segments (and thus running out of mbufs).
250 * Make sure to let the missing segment through which caused this
251 * queue. Always keep one global queue entry spare to be able to
252 * process the missing segment.
253 */
254 if (th->th_seq != tp->rcv_nxt &&
255 (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
256 tp->t_segqlen >= tcp_reass_maxqlen)) {
257 tcp_reass_overflows++;
258 tcpstat.tcps_rcvmemdrop++;
259 m_freem(m);
260 *tlenp = 0;
261 return (0);
262 }
263
264 /*
265 * Allocate a new queue entry. If we can't, or hit the zone limit
266 * just drop the pkt.
267 */
268 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
269 if (te == NULL) {
270 tcpstat.tcps_rcvmemdrop++;
271 m_freem(m);
272 *tlenp = 0;
273 return (0);
274 }
275 tp->t_segqlen++;
276 tcp_reass_qsize++;
277
278 /*
279 * Find a segment which begins after this one does.
280 */
281 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
282 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
283 break;
284 p = q;
285 }
286
287 /*
288 * If there is a preceding segment, it may provide some of
289 * our data already. If so, drop the data from the incoming
290 * segment. If it provides all of our data, drop us.
291 */
292 if (p != NULL) {
293 register int i;
294 /* conversion to int (in i) handles seq wraparound */
295 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
296 if (i > 0) {
297 if (i >= *tlenp) {
298 tcpstat.tcps_rcvduppack++;
299 tcpstat.tcps_rcvdupbyte += *tlenp;
300 m_freem(m);
301 uma_zfree(tcp_reass_zone, te);
302 tp->t_segqlen--;
303 tcp_reass_qsize--;
304 /*
305 * Try to present any queued data
306 * at the left window edge to the user.
307 * This is needed after the 3-WHS
308 * completes.
309 */
310 goto present; /* ??? */
311 }
312 m_adj(m, i);
313 *tlenp -= i;
314 th->th_seq += i;
315 }
316 }
317 tcpstat.tcps_rcvoopack++;
318 tcpstat.tcps_rcvoobyte += *tlenp;
319
320 /*
321 * While we overlap succeeding segments trim them or,
322 * if they are completely covered, dequeue them.
323 */
324 while (q) {
325 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
326 if (i <= 0)
327 break;
328 if (i < q->tqe_len) {
329 q->tqe_th->th_seq += i;
330 q->tqe_len -= i;
331 m_adj(q->tqe_m, i);
332 break;
333 }
334
335 nq = LIST_NEXT(q, tqe_q);
336 LIST_REMOVE(q, tqe_q);
337 m_freem(q->tqe_m);
338 uma_zfree(tcp_reass_zone, q);
339 tp->t_segqlen--;
340 tcp_reass_qsize--;
341 q = nq;
342 }
343
344 /* Insert the new segment queue entry into place. */
345 te->tqe_m = m;
346 te->tqe_th = th;
347 te->tqe_len = *tlenp;
348
349 if (p == NULL) {
350 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
351 } else {
352 LIST_INSERT_AFTER(p, te, tqe_q);
353 }
354
355 present:
356 /*
357 * Present data to user, advancing rcv_nxt through
358 * completed sequence space.
359 */
360 if (!TCPS_HAVEESTABLISHED(tp->t_state))
361 return (0);
362 q = LIST_FIRST(&tp->t_segq);
363 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
364 return (0);
365 SOCKBUF_LOCK(&so->so_rcv);
366 do {
367 tp->rcv_nxt += q->tqe_len;
368 flags = q->tqe_th->th_flags & TH_FIN;
369 nq = LIST_NEXT(q, tqe_q);
370 LIST_REMOVE(q, tqe_q);
371 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
372 m_freem(q->tqe_m);
373 else
374 sbappendstream_locked(&so->so_rcv, q->tqe_m);
375 uma_zfree(tcp_reass_zone, q);
376 tp->t_segqlen--;
377 tcp_reass_qsize--;
378 q = nq;
379 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
380 ND6_HINT(tp);
381 sorwakeup_locked(so);
382 return (flags);
383 }
384
385 /*
386 * TCP input routine, follows pages 65-76 of the
387 * protocol specification dated September, 1981 very closely.
388 */
389 #ifdef INET6
390 int
391 tcp6_input(mp, offp, proto)
392 struct mbuf **mp;
393 int *offp, proto;
394 {
395 register struct mbuf *m = *mp;
396 struct in6_ifaddr *ia6;
397
398 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
399
400 /*
401 * draft-itojun-ipv6-tcp-to-anycast
402 * better place to put this in?
403 */
404 ia6 = ip6_getdstifaddr(m);
405 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
406 struct ip6_hdr *ip6;
407
408 ip6 = mtod(m, struct ip6_hdr *);
409 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
410 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
411 return IPPROTO_DONE;
412 }
413
414 tcp_input(m, *offp);
415 return IPPROTO_DONE;
416 }
417 #endif
418
419 void
420 tcp_input(m, off0)
421 register struct mbuf *m;
422 int off0;
423 {
424 register struct tcphdr *th;
425 register struct ip *ip = NULL;
426 register struct ipovly *ipov;
427 register struct inpcb *inp = NULL;
428 u_char *optp = NULL;
429 int optlen = 0;
430 int len, tlen, off;
431 int drop_hdrlen;
432 register struct tcpcb *tp = 0;
433 register int thflags;
434 struct socket *so = 0;
435 int todrop, acked, ourfinisacked, needoutput = 0;
436 u_long tiwin;
437 struct tcpopt to; /* options in this segment */
438 struct rmxp_tao tao; /* our TAO cache entry */
439 int headlocked = 0;
440 #ifdef IPFIREWALL_FORWARD
441 struct m_tag *fwd_tag;
442 #endif
443 int rstreason; /* For badport_bandlim accounting purposes */
444
445 struct ip6_hdr *ip6 = NULL;
446 #ifdef INET6
447 int isipv6;
448 #else
449 const int isipv6 = 0;
450 #endif
451
452 #ifdef TCPDEBUG
453 /*
454 * The size of tcp_saveipgen must be the size of the max ip header,
455 * now IPv6.
456 */
457 u_char tcp_saveipgen[40];
458 struct tcphdr tcp_savetcp;
459 short ostate = 0;
460 #endif
461
462 #ifdef INET6
463 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
464 #endif
465 bzero(&tao, sizeof(tao));
466 bzero((char *)&to, sizeof(to));
467
468 tcpstat.tcps_rcvtotal++;
469
470 if (isipv6) {
471 #ifdef INET6
472 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
473 ip6 = mtod(m, struct ip6_hdr *);
474 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
475 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
476 tcpstat.tcps_rcvbadsum++;
477 goto drop;
478 }
479 th = (struct tcphdr *)((caddr_t)ip6 + off0);
480
481 /*
482 * Be proactive about unspecified IPv6 address in source.
483 * As we use all-zero to indicate unbounded/unconnected pcb,
484 * unspecified IPv6 address can be used to confuse us.
485 *
486 * Note that packets with unspecified IPv6 destination is
487 * already dropped in ip6_input.
488 */
489 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
490 /* XXX stat */
491 goto drop;
492 }
493 #else
494 th = NULL; /* XXX: avoid compiler warning */
495 #endif
496 } else {
497 /*
498 * Get IP and TCP header together in first mbuf.
499 * Note: IP leaves IP header in first mbuf.
500 */
501 if (off0 > sizeof (struct ip)) {
502 ip_stripoptions(m, (struct mbuf *)0);
503 off0 = sizeof(struct ip);
504 }
505 if (m->m_len < sizeof (struct tcpiphdr)) {
506 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
507 tcpstat.tcps_rcvshort++;
508 return;
509 }
510 }
511 ip = mtod(m, struct ip *);
512 ipov = (struct ipovly *)ip;
513 th = (struct tcphdr *)((caddr_t)ip + off0);
514 tlen = ip->ip_len;
515
516 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
517 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
518 th->th_sum = m->m_pkthdr.csum_data;
519 else
520 th->th_sum = in_pseudo(ip->ip_src.s_addr,
521 ip->ip_dst.s_addr,
522 htonl(m->m_pkthdr.csum_data +
523 ip->ip_len +
524 IPPROTO_TCP));
525 th->th_sum ^= 0xffff;
526 #ifdef TCPDEBUG
527 ipov->ih_len = (u_short)tlen;
528 ipov->ih_len = htons(ipov->ih_len);
529 #endif
530 } else {
531 /*
532 * Checksum extended TCP header and data.
533 */
534 len = sizeof (struct ip) + tlen;
535 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
536 ipov->ih_len = (u_short)tlen;
537 ipov->ih_len = htons(ipov->ih_len);
538 th->th_sum = in_cksum(m, len);
539 }
540 if (th->th_sum) {
541 tcpstat.tcps_rcvbadsum++;
542 goto drop;
543 }
544 #ifdef INET6
545 /* Re-initialization for later version check */
546 ip->ip_v = IPVERSION;
547 #endif
548 }
549
550 /*
551 * Check that TCP offset makes sense,
552 * pull out TCP options and adjust length. XXX
553 */
554 off = th->th_off << 2;
555 if (off < sizeof (struct tcphdr) || off > tlen) {
556 tcpstat.tcps_rcvbadoff++;
557 goto drop;
558 }
559 tlen -= off; /* tlen is used instead of ti->ti_len */
560 if (off > sizeof (struct tcphdr)) {
561 if (isipv6) {
562 #ifdef INET6
563 IP6_EXTHDR_CHECK(m, off0, off, );
564 ip6 = mtod(m, struct ip6_hdr *);
565 th = (struct tcphdr *)((caddr_t)ip6 + off0);
566 #endif
567 } else {
568 if (m->m_len < sizeof(struct ip) + off) {
569 if ((m = m_pullup(m, sizeof (struct ip) + off))
570 == 0) {
571 tcpstat.tcps_rcvshort++;
572 return;
573 }
574 ip = mtod(m, struct ip *);
575 ipov = (struct ipovly *)ip;
576 th = (struct tcphdr *)((caddr_t)ip + off0);
577 }
578 }
579 optlen = off - sizeof (struct tcphdr);
580 optp = (u_char *)(th + 1);
581 }
582 thflags = th->th_flags;
583
584 #ifdef TCP_DROP_SYNFIN
585 /*
586 * If the drop_synfin option is enabled, drop all packets with
587 * both the SYN and FIN bits set. This prevents e.g. nmap from
588 * identifying the TCP/IP stack.
589 *
590 * This is a violation of the TCP specification.
591 */
592 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
593 goto drop;
594 #endif
595
596 /*
597 * Convert TCP protocol specific fields to host format.
598 */
599 th->th_seq = ntohl(th->th_seq);
600 th->th_ack = ntohl(th->th_ack);
601 th->th_win = ntohs(th->th_win);
602 th->th_urp = ntohs(th->th_urp);
603
604 /*
605 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
606 * until after ip6_savecontrol() is called and before other functions
607 * which don't want those proto headers.
608 * Because ip6_savecontrol() is going to parse the mbuf to
609 * search for data to be passed up to user-land, it wants mbuf
610 * parameters to be unchanged.
611 * XXX: the call of ip6_savecontrol() has been obsoleted based on
612 * latest version of the advanced API (20020110).
613 */
614 drop_hdrlen = off0 + off;
615
616 /*
617 * Locate pcb for segment.
618 */
619 INP_INFO_WLOCK(&tcbinfo);
620 headlocked = 1;
621 findpcb:
622 KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
623 #ifdef IPFIREWALL_FORWARD
624 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
625 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
626
627 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
628 struct sockaddr_in *next_hop;
629
630 next_hop = (struct sockaddr_in *)(fwd_tag+1);
631 /*
632 * Transparently forwarded. Pretend to be the destination.
633 * already got one like this?
634 */
635 inp = in_pcblookup_hash(&tcbinfo,
636 ip->ip_src, th->th_sport,
637 ip->ip_dst, th->th_dport,
638 0, m->m_pkthdr.rcvif);
639 if (!inp) {
640 /* It's new. Try to find the ambushing socket. */
641 inp = in_pcblookup_hash(&tcbinfo,
642 ip->ip_src, th->th_sport,
643 next_hop->sin_addr,
644 next_hop->sin_port ?
645 ntohs(next_hop->sin_port) :
646 th->th_dport,
647 1, m->m_pkthdr.rcvif);
648 }
649 /* Remove the tag from the packet. We don't need it anymore. */
650 m_tag_delete(m, fwd_tag);
651 } else {
652 #endif /* IPFIREWALL_FORWARD */
653 if (isipv6) {
654 #ifdef INET6
655 inp = in6_pcblookup_hash(&tcbinfo,
656 &ip6->ip6_src, th->th_sport,
657 &ip6->ip6_dst, th->th_dport,
658 1, m->m_pkthdr.rcvif);
659 #endif
660 } else
661 inp = in_pcblookup_hash(&tcbinfo,
662 ip->ip_src, th->th_sport,
663 ip->ip_dst, th->th_dport,
664 1, m->m_pkthdr.rcvif);
665 #ifdef IPFIREWALL_FORWARD
666 }
667 #endif /* IPFIREWALL_FORWARD */
668
669 #if defined(IPSEC) || defined(FAST_IPSEC)
670 #ifdef INET6
671 if (isipv6) {
672 if (inp != NULL && ipsec6_in_reject(m, inp)) {
673 #ifdef IPSEC
674 ipsec6stat.in_polvio++;
675 #endif
676 goto drop;
677 }
678 } else
679 #endif /* INET6 */
680 if (inp != NULL && ipsec4_in_reject(m, inp)) {
681 #ifdef IPSEC
682 ipsecstat.in_polvio++;
683 #endif
684 goto drop;
685 }
686 #endif /*IPSEC || FAST_IPSEC*/
687
688 /*
689 * If the state is CLOSED (i.e., TCB does not exist) then
690 * all data in the incoming segment is discarded.
691 * If the TCB exists but is in CLOSED state, it is embryonic,
692 * but should either do a listen or a connect soon.
693 */
694 if (inp == NULL) {
695 if (log_in_vain) {
696 #ifdef INET6
697 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
698 #else
699 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
700 #endif
701
702 if (isipv6) {
703 #ifdef INET6
704 strcpy(dbuf, "[");
705 strcpy(sbuf, "[");
706 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
707 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
708 strcat(dbuf, "]");
709 strcat(sbuf, "]");
710 #endif
711 } else {
712 strcpy(dbuf, inet_ntoa(ip->ip_dst));
713 strcpy(sbuf, inet_ntoa(ip->ip_src));
714 }
715 switch (log_in_vain) {
716 case 1:
717 if ((thflags & TH_SYN) == 0)
718 break;
719 /* FALLTHROUGH */
720 case 2:
721 log(LOG_INFO,
722 "Connection attempt to TCP %s:%d "
723 "from %s:%d flags:0x%02x\n",
724 dbuf, ntohs(th->th_dport), sbuf,
725 ntohs(th->th_sport), thflags);
726 break;
727 default:
728 break;
729 }
730 }
731 if (blackhole) {
732 switch (blackhole) {
733 case 1:
734 if (thflags & TH_SYN)
735 goto drop;
736 break;
737 case 2:
738 goto drop;
739 default:
740 goto drop;
741 }
742 }
743 rstreason = BANDLIM_RST_CLOSEDPORT;
744 goto dropwithreset;
745 }
746 INP_LOCK(inp);
747 if (inp->inp_vflag & INP_TIMEWAIT) {
748 /*
749 * The only option of relevance is TOF_CC, and only if
750 * present in a SYN segment. See tcp_timewait().
751 */
752 if (thflags & TH_SYN)
753 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th);
754 if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
755 &to, th, m, tlen))
756 goto findpcb;
757 /*
758 * tcp_timewait unlocks inp.
759 */
760 INP_INFO_WUNLOCK(&tcbinfo);
761 return;
762 }
763 tp = intotcpcb(inp);
764 if (tp == 0) {
765 INP_UNLOCK(inp);
766 rstreason = BANDLIM_RST_CLOSEDPORT;
767 goto dropwithreset;
768 }
769 if (tp->t_state == TCPS_CLOSED)
770 goto drop;
771
772 /* Unscale the window into a 32-bit value. */
773 if ((thflags & TH_SYN) == 0)
774 tiwin = th->th_win << tp->snd_scale;
775 else
776 tiwin = th->th_win;
777
778 #ifdef MAC
779 INP_LOCK_ASSERT(inp);
780 if (mac_check_inpcb_deliver(inp, m))
781 goto drop;
782 #endif
783 so = inp->inp_socket;
784 #ifdef TCPDEBUG
785 if (so->so_options & SO_DEBUG) {
786 ostate = tp->t_state;
787 if (isipv6)
788 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
789 else
790 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
791 tcp_savetcp = *th;
792 }
793 #endif
794 if (so->so_options & SO_ACCEPTCONN) {
795 struct in_conninfo inc;
796
797 #ifdef INET6
798 inc.inc_isipv6 = isipv6;
799 #endif
800 if (isipv6) {
801 inc.inc6_faddr = ip6->ip6_src;
802 inc.inc6_laddr = ip6->ip6_dst;
803 } else {
804 inc.inc_faddr = ip->ip_src;
805 inc.inc_laddr = ip->ip_dst;
806 }
807 inc.inc_fport = th->th_sport;
808 inc.inc_lport = th->th_dport;
809
810 /*
811 * If the state is LISTEN then ignore segment if it contains
812 * a RST. If the segment contains an ACK then it is bad and
813 * send a RST. If it does not contain a SYN then it is not
814 * interesting; drop it.
815 *
816 * If the state is SYN_RECEIVED (syncache) and seg contains
817 * an ACK, but not for our SYN/ACK, send a RST. If the seg
818 * contains a RST, check the sequence number to see if it
819 * is a valid reset segment.
820 */
821 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
822 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
823 if (!syncache_expand(&inc, th, &so, m)) {
824 /*
825 * No syncache entry, or ACK was not
826 * for our SYN/ACK. Send a RST.
827 */
828 tcpstat.tcps_badsyn++;
829 rstreason = BANDLIM_RST_OPENPORT;
830 goto dropwithreset;
831 }
832 if (so == NULL) {
833 /*
834 * Could not complete 3-way handshake,
835 * connection is being closed down, and
836 * syncache will free mbuf.
837 */
838 INP_UNLOCK(inp);
839 INP_INFO_WUNLOCK(&tcbinfo);
840 return;
841 }
842 /*
843 * Socket is created in state SYN_RECEIVED.
844 * Continue processing segment.
845 */
846 INP_UNLOCK(inp);
847 inp = sotoinpcb(so);
848 INP_LOCK(inp);
849 tp = intotcpcb(inp);
850 /*
851 * This is what would have happened in
852 * tcp_output() when the SYN,ACK was sent.
853 */
854 tp->snd_up = tp->snd_una;
855 tp->snd_max = tp->snd_nxt = tp->iss + 1;
856 tp->last_ack_sent = tp->rcv_nxt;
857 /*
858 * RFC1323: The window in SYN & SYN/ACK
859 * segments is never scaled.
860 */
861 tp->snd_wnd = tiwin; /* unscaled */
862 goto after_listen;
863 }
864 if (thflags & TH_RST) {
865 syncache_chkrst(&inc, th);
866 goto drop;
867 }
868 if (thflags & TH_ACK) {
869 syncache_badack(&inc);
870 tcpstat.tcps_badsyn++;
871 rstreason = BANDLIM_RST_OPENPORT;
872 goto dropwithreset;
873 }
874 goto drop;
875 }
876
877 /*
878 * Segment's flags are (SYN) or (SYN|FIN).
879 */
880 #ifdef INET6
881 /*
882 * If deprecated address is forbidden,
883 * we do not accept SYN to deprecated interface
884 * address to prevent any new inbound connection from
885 * getting established.
886 * When we do not accept SYN, we send a TCP RST,
887 * with deprecated source address (instead of dropping
888 * it). We compromise it as it is much better for peer
889 * to send a RST, and RST will be the final packet
890 * for the exchange.
891 *
892 * If we do not forbid deprecated addresses, we accept
893 * the SYN packet. RFC2462 does not suggest dropping
894 * SYN in this case.
895 * If we decipher RFC2462 5.5.4, it says like this:
896 * 1. use of deprecated addr with existing
897 * communication is okay - "SHOULD continue to be
898 * used"
899 * 2. use of it with new communication:
900 * (2a) "SHOULD NOT be used if alternate address
901 * with sufficient scope is available"
902 * (2b) nothing mentioned otherwise.
903 * Here we fall into (2b) case as we have no choice in
904 * our source address selection - we must obey the peer.
905 *
906 * The wording in RFC2462 is confusing, and there are
907 * multiple description text for deprecated address
908 * handling - worse, they are not exactly the same.
909 * I believe 5.5.4 is the best one, so we follow 5.5.4.
910 */
911 if (isipv6 && !ip6_use_deprecated) {
912 struct in6_ifaddr *ia6;
913
914 if ((ia6 = ip6_getdstifaddr(m)) &&
915 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
916 INP_UNLOCK(inp);
917 tp = NULL;
918 rstreason = BANDLIM_RST_OPENPORT;
919 goto dropwithreset;
920 }
921 }
922 #endif
923 /*
924 * If it is from this socket, drop it, it must be forged.
925 * Don't bother responding if the destination was a broadcast.
926 */
927 if (th->th_dport == th->th_sport) {
928 if (isipv6) {
929 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
930 &ip6->ip6_src))
931 goto drop;
932 } else {
933 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
934 goto drop;
935 }
936 }
937 /*
938 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
939 *
940 * Note that it is quite possible to receive unicast
941 * link-layer packets with a broadcast IP address. Use
942 * in_broadcast() to find them.
943 */
944 if (m->m_flags & (M_BCAST|M_MCAST))
945 goto drop;
946 if (isipv6) {
947 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
948 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
949 goto drop;
950 } else {
951 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
952 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
953 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
954 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
955 goto drop;
956 }
957 /*
958 * SYN appears to be valid; create compressed TCP state
959 * for syncache, or perform t/tcp connection.
960 */
961 if (so->so_qlen <= so->so_qlimit) {
962 #ifdef TCPDEBUG
963 if (so->so_options & SO_DEBUG)
964 tcp_trace(TA_INPUT, ostate, tp,
965 (void *)tcp_saveipgen, &tcp_savetcp, 0);
966 #endif
967 tcp_dooptions(tp, &to, optp, optlen, 1, th);
968 if (!syncache_add(&inc, &to, th, &so, m))
969 goto drop;
970 if (so == NULL) {
971 /*
972 * Entry added to syncache, mbuf used to
973 * send SYN,ACK packet.
974 */
975 KASSERT(headlocked, ("headlocked"));
976 INP_UNLOCK(inp);
977 INP_INFO_WUNLOCK(&tcbinfo);
978 return;
979 }
980 /*
981 * Segment passed TAO tests.
982 */
983 INP_UNLOCK(inp);
984 inp = sotoinpcb(so);
985 INP_LOCK(inp);
986 tp = intotcpcb(inp);
987 tp->snd_wnd = tiwin;
988 tp->t_starttime = ticks;
989 tp->t_state = TCPS_ESTABLISHED;
990
991 /*
992 * T/TCP logic:
993 * If there is a FIN or if there is data, then
994 * delay SYN,ACK(SYN) in the hope of piggy-backing
995 * it on a response segment. Otherwise must send
996 * ACK now in case the other side is slow starting.
997 */
998 if (thflags & TH_FIN || tlen != 0)
999 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1000 else
1001 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1002 tcpstat.tcps_connects++;
1003 soisconnected(so);
1004 goto trimthenstep6;
1005 }
1006 goto drop;
1007 }
1008 after_listen:
1009 KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1010 INP_LOCK_ASSERT(inp);
1011
1012 /* XXX temp debugging */
1013 /* should not happen - syncache should pick up these connections */
1014 if (tp->t_state == TCPS_LISTEN)
1015 panic("tcp_input: TCPS_LISTEN");
1016
1017 /*
1018 * This is the second part of the MSS DoS prevention code (after
1019 * minmss on the sending side) and it deals with too many too small
1020 * tcp packets in a too short timeframe (1 second).
1021 *
1022 * For every full second we count the number of received packets
1023 * and bytes. If we get a lot of packets per second for this connection
1024 * (tcp_minmssoverload) we take a closer look at it and compute the
1025 * average packet size for the past second. If that is less than
1026 * tcp_minmss we get too many packets with very small payload which
1027 * is not good and burdens our system (and every packet generates
1028 * a wakeup to the process connected to our socket). We can reasonable
1029 * expect this to be small packet DoS attack to exhaust our CPU
1030 * cycles.
1031 *
1032 * Care has to be taken for the minimum packet overload value. This
1033 * value defines the minimum number of packets per second before we
1034 * start to worry. This must not be too low to avoid killing for
1035 * example interactive connections with many small packets like
1036 * telnet or SSH.
1037 *
1038 * Setting either tcp_minmssoverload or tcp_minmss to "" disables
1039 * this check.
1040 *
1041 * Account for packet if payload packet, skip over ACK, etc.
1042 */
1043 if (tcp_minmss && tcp_minmssoverload &&
1044 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1045 if (tp->rcv_second > ticks) {
1046 tp->rcv_pps++;
1047 tp->rcv_byps += tlen + off;
1048 if (tp->rcv_pps > tcp_minmssoverload) {
1049 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1050 printf("too many small tcp packets from "
1051 "%s:%u, av. %lubyte/packet, "
1052 "dropping connection\n",
1053 #ifdef INET6
1054 isipv6 ?
1055 ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1056 #endif
1057 inet_ntoa(inp->inp_inc.inc_faddr),
1058 inp->inp_inc.inc_fport,
1059 tp->rcv_byps / tp->rcv_pps);
1060 KASSERT(headlocked, ("tcp_input: "
1061 "after_listen: tcp_drop: head "
1062 "not locked"));
1063 tp = tcp_drop(tp, ECONNRESET);
1064 tcpstat.tcps_minmssdrops++;
1065 goto drop;
1066 }
1067 }
1068 } else {
1069 tp->rcv_second = ticks + hz;
1070 tp->rcv_pps = 1;
1071 tp->rcv_byps = tlen + off;
1072 }
1073 }
1074
1075 /*
1076 * Segment received on connection.
1077 * Reset idle time and keep-alive timer.
1078 */
1079 tp->t_rcvtime = ticks;
1080 if (TCPS_HAVEESTABLISHED(tp->t_state))
1081 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1082
1083 /*
1084 * Process options only when we get SYN/ACK back. The SYN case
1085 * for incoming connections is handled in tcp_syncache.
1086 * XXX this is traditional behavior, may need to be cleaned up.
1087 */
1088 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th);
1089 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1090 if (to.to_flags & TOF_SCALE) {
1091 tp->t_flags |= TF_RCVD_SCALE;
1092 tp->requested_s_scale = to.to_requested_s_scale;
1093 }
1094 if (to.to_flags & TOF_TS) {
1095 tp->t_flags |= TF_RCVD_TSTMP;
1096 tp->ts_recent = to.to_tsval;
1097 tp->ts_recent_age = ticks;
1098 }
1099 if (to.to_flags & (TOF_CC|TOF_CCNEW))
1100 tp->t_flags |= TF_RCVD_CC;
1101 if (to.to_flags & TOF_MSS)
1102 tcp_mss(tp, to.to_mss);
1103 if (tp->sack_enable) {
1104 if (!(to.to_flags & TOF_SACK))
1105 tp->sack_enable = 0;
1106 else
1107 tp->t_flags |= TF_SACK_PERMIT;
1108 }
1109
1110 }
1111
1112 if (tp->sack_enable) {
1113 /* Delete stale (cumulatively acked) SACK holes */
1114 tcp_del_sackholes(tp, th);
1115 }
1116
1117 /*
1118 * Header prediction: check for the two common cases
1119 * of a uni-directional data xfer. If the packet has
1120 * no control flags, is in-sequence, the window didn't
1121 * change and we're not retransmitting, it's a
1122 * candidate. If the length is zero and the ack moved
1123 * forward, we're the sender side of the xfer. Just
1124 * free the data acked & wake any higher level process
1125 * that was blocked waiting for space. If the length
1126 * is non-zero and the ack didn't move, we're the
1127 * receiver side. If we're getting packets in-order
1128 * (the reassembly queue is empty), add the data to
1129 * the socket buffer and note that we need a delayed ack.
1130 * Make sure that the hidden state-flags are also off.
1131 * Since we check for TCPS_ESTABLISHED above, it can only
1132 * be TH_NEEDSYN.
1133 */
1134 if (tp->t_state == TCPS_ESTABLISHED &&
1135 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1136 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1137 ((to.to_flags & TOF_TS) == 0 ||
1138 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1139 /*
1140 * Using the CC option is compulsory if once started:
1141 * the segment is OK if no T/TCP was negotiated or
1142 * if the segment has a CC option equal to CCrecv
1143 */
1144 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1145 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1146 th->th_seq == tp->rcv_nxt &&
1147 tiwin && tiwin == tp->snd_wnd &&
1148 tp->snd_nxt == tp->snd_max) {
1149
1150 /*
1151 * If last ACK falls within this segment's sequence numbers,
1152 * record the timestamp.
1153 * NOTE that the test is modified according to the latest
1154 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1155 */
1156 if ((to.to_flags & TOF_TS) != 0 &&
1157 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1158 tp->ts_recent_age = ticks;
1159 tp->ts_recent = to.to_tsval;
1160 }
1161
1162 if (tlen == 0) {
1163 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1164 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1165 tp->snd_cwnd >= tp->snd_wnd &&
1166 ((!tcp_do_newreno && !tp->sack_enable &&
1167 tp->t_dupacks < tcprexmtthresh) ||
1168 ((tcp_do_newreno || tp->sack_enable) &&
1169 !IN_FASTRECOVERY(tp)))) {
1170 KASSERT(headlocked, ("headlocked"));
1171 INP_INFO_WUNLOCK(&tcbinfo);
1172 headlocked = 0;
1173 /*
1174 * this is a pure ack for outstanding data.
1175 */
1176 ++tcpstat.tcps_predack;
1177 /*
1178 * "bad retransmit" recovery
1179 */
1180 if (tp->t_rxtshift == 1 &&
1181 ticks < tp->t_badrxtwin) {
1182 ++tcpstat.tcps_sndrexmitbad;
1183 tp->snd_cwnd = tp->snd_cwnd_prev;
1184 tp->snd_ssthresh =
1185 tp->snd_ssthresh_prev;
1186 tp->snd_recover = tp->snd_recover_prev;
1187 if (tp->t_flags & TF_WASFRECOVERY)
1188 ENTER_FASTRECOVERY(tp);
1189 tp->snd_nxt = tp->snd_max;
1190 tp->t_badrxtwin = 0;
1191 }
1192
1193 /*
1194 * Recalculate the transmit timer / rtt.
1195 *
1196 * Some boxes send broken timestamp replies
1197 * during the SYN+ACK phase, ignore
1198 * timestamps of 0 or we could calculate a
1199 * huge RTT and blow up the retransmit timer.
1200 */
1201 if ((to.to_flags & TOF_TS) != 0 &&
1202 to.to_tsecr) {
1203 tcp_xmit_timer(tp,
1204 ticks - to.to_tsecr + 1);
1205 } else if (tp->t_rtttime &&
1206 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1207 tcp_xmit_timer(tp,
1208 ticks - tp->t_rtttime);
1209 }
1210 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1211 acked = th->th_ack - tp->snd_una;
1212 tcpstat.tcps_rcvackpack++;
1213 tcpstat.tcps_rcvackbyte += acked;
1214 sbdrop(&so->so_snd, acked);
1215 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1216 SEQ_LEQ(th->th_ack, tp->snd_recover))
1217 tp->snd_recover = th->th_ack - 1;
1218 tp->snd_una = th->th_ack;
1219 /*
1220 * pull snd_wl2 up to prevent seq wrap relative
1221 * to th_ack.
1222 */
1223 tp->snd_wl2 = th->th_ack;
1224 tp->t_dupacks = 0;
1225 m_freem(m);
1226 ND6_HINT(tp); /* some progress has been done */
1227
1228 /*
1229 * If all outstanding data are acked, stop
1230 * retransmit timer, otherwise restart timer
1231 * using current (possibly backed-off) value.
1232 * If process is waiting for space,
1233 * wakeup/selwakeup/signal. If data
1234 * are ready to send, let tcp_output
1235 * decide between more output or persist.
1236
1237 #ifdef TCPDEBUG
1238 if (so->so_options & SO_DEBUG)
1239 tcp_trace(TA_INPUT, ostate, tp,
1240 (void *)tcp_saveipgen,
1241 &tcp_savetcp, 0);
1242 #endif
1243 */
1244 if (tp->snd_una == tp->snd_max)
1245 callout_stop(tp->tt_rexmt);
1246 else if (!callout_active(tp->tt_persist))
1247 callout_reset(tp->tt_rexmt,
1248 tp->t_rxtcur,
1249 tcp_timer_rexmt, tp);
1250
1251 sowwakeup(so);
1252 if (so->so_snd.sb_cc)
1253 (void) tcp_output(tp);
1254 goto check_delack;
1255 }
1256 } else if (th->th_ack == tp->snd_una &&
1257 LIST_EMPTY(&tp->t_segq) &&
1258 tlen <= sbspace(&so->so_rcv)) {
1259 KASSERT(headlocked, ("headlocked"));
1260 INP_INFO_WUNLOCK(&tcbinfo);
1261 headlocked = 0;
1262 /*
1263 * this is a pure, in-sequence data packet
1264 * with nothing on the reassembly queue and
1265 * we have enough buffer space to take it.
1266 */
1267 /* Clean receiver SACK report if present */
1268 if (tp->sack_enable && tp->rcv_numsacks)
1269 tcp_clean_sackreport(tp);
1270 ++tcpstat.tcps_preddat;
1271 tp->rcv_nxt += tlen;
1272 /*
1273 * Pull snd_wl1 up to prevent seq wrap relative to
1274 * th_seq.
1275 */
1276 tp->snd_wl1 = th->th_seq;
1277 /*
1278 * Pull rcv_up up to prevent seq wrap relative to
1279 * rcv_nxt.
1280 */
1281 tp->rcv_up = tp->rcv_nxt;
1282 tcpstat.tcps_rcvpack++;
1283 tcpstat.tcps_rcvbyte += tlen;
1284 ND6_HINT(tp); /* some progress has been done */
1285 /*
1286 #ifdef TCPDEBUG
1287 if (so->so_options & SO_DEBUG)
1288 tcp_trace(TA_INPUT, ostate, tp,
1289 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1290 #endif
1291 * Add data to socket buffer.
1292 */
1293 SOCKBUF_LOCK(&so->so_rcv);
1294 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1295 m_freem(m);
1296 } else {
1297 m_adj(m, drop_hdrlen); /* delayed header drop */
1298 sbappendstream_locked(&so->so_rcv, m);
1299 }
1300 sorwakeup_locked(so);
1301 if (DELAY_ACK(tp)) {
1302 tp->t_flags |= TF_DELACK;
1303 } else {
1304 tp->t_flags |= TF_ACKNOW;
1305 tcp_output(tp);
1306 }
1307 goto check_delack;
1308 }
1309 }
1310
1311 /*
1312 * Calculate amount of space in receive window,
1313 * and then do TCP input processing.
1314 * Receive window is amount of space in rcv queue,
1315 * but not less than advertised window.
1316 */
1317 { int win;
1318
1319 win = sbspace(&so->so_rcv);
1320 if (win < 0)
1321 win = 0;
1322 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1323 }
1324
1325 switch (tp->t_state) {
1326
1327 /*
1328 * If the state is SYN_RECEIVED:
1329 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1330 */
1331 case TCPS_SYN_RECEIVED:
1332 if ((thflags & TH_ACK) &&
1333 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1334 SEQ_GT(th->th_ack, tp->snd_max))) {
1335 rstreason = BANDLIM_RST_OPENPORT;
1336 goto dropwithreset;
1337 }
1338 break;
1339
1340 /*
1341 * If the state is SYN_SENT:
1342 * if seg contains an ACK, but not for our SYN, drop the input.
1343 * if seg contains a RST, then drop the connection.
1344 * if seg does not contain SYN, then drop it.
1345 * Otherwise this is an acceptable SYN segment
1346 * initialize tp->rcv_nxt and tp->irs
1347 * if seg contains ack then advance tp->snd_una
1348 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1349 * arrange for segment to be acked (eventually)
1350 * continue processing rest of data/controls, beginning with URG
1351 */
1352 case TCPS_SYN_SENT:
1353 if (tcp_do_rfc1644)
1354 tcp_hc_gettao(&inp->inp_inc, &tao);
1355
1356 if ((thflags & TH_ACK) &&
1357 (SEQ_LEQ(th->th_ack, tp->iss) ||
1358 SEQ_GT(th->th_ack, tp->snd_max))) {
1359 /*
1360 * If we have a cached CCsent for the remote host,
1361 * hence we haven't just crashed and restarted,
1362 * do not send a RST. This may be a retransmission
1363 * from the other side after our earlier ACK was lost.
1364 * Our new SYN, when it arrives, will serve as the
1365 * needed ACK.
1366 */
1367 if (tao.tao_ccsent != 0)
1368 goto drop;
1369 else {
1370 rstreason = BANDLIM_UNLIMITED;
1371 goto dropwithreset;
1372 }
1373 }
1374 if (thflags & TH_RST) {
1375 if (thflags & TH_ACK) {
1376 KASSERT(headlocked, ("tcp_input: after_listen"
1377 ": tcp_drop.2: head not locked"));
1378 tp = tcp_drop(tp, ECONNREFUSED);
1379 }
1380 goto drop;
1381 }
1382 if ((thflags & TH_SYN) == 0)
1383 goto drop;
1384 tp->snd_wnd = th->th_win; /* initial send window */
1385 tp->cc_recv = to.to_cc; /* foreign CC */
1386
1387 tp->irs = th->th_seq;
1388 tcp_rcvseqinit(tp);
1389 if (thflags & TH_ACK) {
1390 /*
1391 * Our SYN was acked. If segment contains CC.ECHO
1392 * option, check it to make sure this segment really
1393 * matches our SYN. If not, just drop it as old
1394 * duplicate, but send an RST if we're still playing
1395 * by the old rules. If no CC.ECHO option, make sure
1396 * we don't get fooled into using T/TCP.
1397 */
1398 if (to.to_flags & TOF_CCECHO) {
1399 if (tp->cc_send != to.to_ccecho) {
1400 if (tao.tao_ccsent != 0)
1401 goto drop;
1402 else {
1403 rstreason = BANDLIM_UNLIMITED;
1404 goto dropwithreset;
1405 }
1406 }
1407 } else
1408 tp->t_flags &= ~TF_RCVD_CC;
1409 tcpstat.tcps_connects++;
1410 soisconnected(so);
1411 #ifdef MAC
1412 SOCK_LOCK(so);
1413 mac_set_socket_peer_from_mbuf(m, so);
1414 SOCK_UNLOCK(so);
1415 #endif
1416 /* Do window scaling on this connection? */
1417 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1418 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1419 tp->snd_scale = tp->requested_s_scale;
1420 tp->rcv_scale = tp->request_r_scale;
1421 }
1422 /* Segment is acceptable, update cache if undefined. */
1423 if (tao.tao_ccsent == 0 && tcp_do_rfc1644)
1424 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0);
1425
1426 tp->rcv_adv += tp->rcv_wnd;
1427 tp->snd_una++; /* SYN is acked */
1428 /*
1429 * If there's data, delay ACK; if there's also a FIN
1430 * ACKNOW will be turned on later.
1431 */
1432 if (DELAY_ACK(tp) && tlen != 0)
1433 callout_reset(tp->tt_delack, tcp_delacktime,
1434 tcp_timer_delack, tp);
1435 else
1436 tp->t_flags |= TF_ACKNOW;
1437 /*
1438 * Received <SYN,ACK> in SYN_SENT[*] state.
1439 * Transitions:
1440 * SYN_SENT --> ESTABLISHED
1441 * SYN_SENT* --> FIN_WAIT_1
1442 */
1443 tp->t_starttime = ticks;
1444 if (tp->t_flags & TF_NEEDFIN) {
1445 tp->t_state = TCPS_FIN_WAIT_1;
1446 tp->t_flags &= ~TF_NEEDFIN;
1447 thflags &= ~TH_SYN;
1448 } else {
1449 tp->t_state = TCPS_ESTABLISHED;
1450 callout_reset(tp->tt_keep, tcp_keepidle,
1451 tcp_timer_keep, tp);
1452 }
1453 } else {
1454 /*
1455 * Received initial SYN in SYN-SENT[*] state =>
1456 * simultaneous open. If segment contains CC option
1457 * and there is a cached CC, apply TAO test.
1458 * If it succeeds, connection is * half-synchronized.
1459 * Otherwise, do 3-way handshake:
1460 * SYN-SENT -> SYN-RECEIVED
1461 * SYN-SENT* -> SYN-RECEIVED*
1462 * If there was no CC option, clear cached CC value.
1463 */
1464 tp->t_flags |= TF_ACKNOW;
1465 callout_stop(tp->tt_rexmt);
1466 if (to.to_flags & TOF_CC) {
1467 if (tao.tao_cc != 0 &&
1468 CC_GT(to.to_cc, tao.tao_cc)) {
1469 /*
1470 * update cache and make transition:
1471 * SYN-SENT -> ESTABLISHED*
1472 * SYN-SENT* -> FIN-WAIT-1*
1473 */
1474 tao.tao_cc = to.to_cc;
1475 tcp_hc_updatetao(&inp->inp_inc,
1476 TCP_HC_TAO_CC, to.to_cc, 0);
1477 tp->t_starttime = ticks;
1478 if (tp->t_flags & TF_NEEDFIN) {
1479 tp->t_state = TCPS_FIN_WAIT_1;
1480 tp->t_flags &= ~TF_NEEDFIN;
1481 } else {
1482 tp->t_state = TCPS_ESTABLISHED;
1483 callout_reset(tp->tt_keep,
1484 tcp_keepidle,
1485 tcp_timer_keep,
1486 tp);
1487 }
1488 tp->t_flags |= TF_NEEDSYN;
1489 } else
1490 tp->t_state = TCPS_SYN_RECEIVED;
1491 } else {
1492 if (tcp_do_rfc1644) {
1493 /* CC.NEW or no option => invalidate cache */
1494 tao.tao_cc = 0;
1495 tcp_hc_updatetao(&inp->inp_inc,
1496 TCP_HC_TAO_CC, to.to_cc, 0);
1497 }
1498 tp->t_state = TCPS_SYN_RECEIVED;
1499 }
1500 }
1501
1502 trimthenstep6:
1503 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1504 "locked"));
1505 INP_LOCK_ASSERT(inp);
1506
1507 /*
1508 * Advance th->th_seq to correspond to first data byte.
1509 * If data, trim to stay within window,
1510 * dropping FIN if necessary.
1511 */
1512 th->th_seq++;
1513 if (tlen > tp->rcv_wnd) {
1514 todrop = tlen - tp->rcv_wnd;
1515 m_adj(m, -todrop);
1516 tlen = tp->rcv_wnd;
1517 thflags &= ~TH_FIN;
1518 tcpstat.tcps_rcvpackafterwin++;
1519 tcpstat.tcps_rcvbyteafterwin += todrop;
1520 }
1521 tp->snd_wl1 = th->th_seq - 1;
1522 tp->rcv_up = th->th_seq;
1523 /*
1524 * Client side of transaction: already sent SYN and data.
1525 * If the remote host used T/TCP to validate the SYN,
1526 * our data will be ACK'd; if so, enter normal data segment
1527 * processing in the middle of step 5, ack processing.
1528 * Otherwise, goto step 6.
1529 */
1530 if (thflags & TH_ACK)
1531 goto process_ACK;
1532
1533 goto step6;
1534
1535 /*
1536 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1537 * if segment contains a SYN and CC [not CC.NEW] option:
1538 * if state == TIME_WAIT and connection duration > MSL,
1539 * drop packet and send RST;
1540 *
1541 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1542 * ack the FIN (and data) in retransmission queue.
1543 * Complete close and delete TCPCB. Then reprocess
1544 * segment, hoping to find new TCPCB in LISTEN state;
1545 *
1546 * else must be old SYN; drop it.
1547 * else do normal processing.
1548 */
1549 case TCPS_LAST_ACK:
1550 case TCPS_CLOSING:
1551 case TCPS_TIME_WAIT:
1552 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1553 if ((thflags & TH_SYN) &&
1554 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1555 if (tp->t_state == TCPS_TIME_WAIT &&
1556 (ticks - tp->t_starttime) > tcp_msl) {
1557 rstreason = BANDLIM_UNLIMITED;
1558 goto dropwithreset;
1559 }
1560 if (CC_GT(to.to_cc, tp->cc_recv)) {
1561 tp = tcp_close(tp);
1562 goto findpcb;
1563 }
1564 else
1565 goto drop;
1566 }
1567 break; /* continue normal processing */
1568 }
1569
1570 /*
1571 * States other than LISTEN or SYN_SENT.
1572 * First check the RST flag and sequence number since reset segments
1573 * are exempt from the timestamp and connection count tests. This
1574 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1575 * below which allowed reset segments in half the sequence space
1576 * to fall though and be processed (which gives forged reset
1577 * segments with a random sequence number a 50 percent chance of
1578 * killing a connection).
1579 * Then check timestamp, if present.
1580 * Then check the connection count, if present.
1581 * Then check that at least some bytes of segment are within
1582 * receive window. If segment begins before rcv_nxt,
1583 * drop leading data (and SYN); if nothing left, just ack.
1584 *
1585 *
1586 * If the RST bit is set, check the sequence number to see
1587 * if this is a valid reset segment.
1588 * RFC 793 page 37:
1589 * In all states except SYN-SENT, all reset (RST) segments
1590 * are validated by checking their SEQ-fields. A reset is
1591 * valid if its sequence number is in the window.
1592 * Note: this does not take into account delayed ACKs, so
1593 * we should test against last_ack_sent instead of rcv_nxt.
1594 * The sequence number in the reset segment is normally an
1595 * echo of our outgoing acknowlegement numbers, but some hosts
1596 * send a reset with the sequence number at the rightmost edge
1597 * of our receive window, and we have to handle this case.
1598 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1599 * that brute force RST attacks are possible. To combat this,
1600 * we use a much stricter check while in the ESTABLISHED state,
1601 * only accepting RSTs where the sequence number is equal to
1602 * last_ack_sent. In all other states (the states in which a
1603 * RST is more likely), the more permissive check is used.
1604 * If we have multiple segments in flight, the intial reset
1605 * segment sequence numbers will be to the left of last_ack_sent,
1606 * but they will eventually catch up.
1607 * In any case, it never made sense to trim reset segments to
1608 * fit the receive window since RFC 1122 says:
1609 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1610 *
1611 * A TCP SHOULD allow a received RST segment to include data.
1612 *
1613 * DISCUSSION
1614 * It has been suggested that a RST segment could contain
1615 * ASCII text that encoded and explained the cause of the
1616 * RST. No standard has yet been established for such
1617 * data.
1618 *
1619 * If the reset segment passes the sequence number test examine
1620 * the state:
1621 * SYN_RECEIVED STATE:
1622 * If passive open, return to LISTEN state.
1623 * If active open, inform user that connection was refused.
1624 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1625 * Inform user that connection was reset, and close tcb.
1626 * CLOSING, LAST_ACK STATES:
1627 * Close the tcb.
1628 * TIME_WAIT STATE:
1629 * Drop the segment - see Stevens, vol. 2, p. 964 and
1630 * RFC 1337.
1631 */
1632 if (thflags & TH_RST) {
1633 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1634 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1635 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1636 switch (tp->t_state) {
1637
1638 case TCPS_SYN_RECEIVED:
1639 so->so_error = ECONNREFUSED;
1640 goto close;
1641
1642 case TCPS_ESTABLISHED:
1643 if (tp->last_ack_sent != th->th_seq &&
1644 tcp_insecure_rst == 0) {
1645 tcpstat.tcps_badrst++;
1646 goto drop;
1647 }
1648 case TCPS_FIN_WAIT_1:
1649 case TCPS_FIN_WAIT_2:
1650 case TCPS_CLOSE_WAIT:
1651 so->so_error = ECONNRESET;
1652 close:
1653 tp->t_state = TCPS_CLOSED;
1654 tcpstat.tcps_drops++;
1655 KASSERT(headlocked, ("tcp_input: "
1656 "trimthenstep6: tcp_close: head not "
1657 "locked"));
1658 tp = tcp_close(tp);
1659 break;
1660
1661 case TCPS_CLOSING:
1662 case TCPS_LAST_ACK:
1663 KASSERT(headlocked, ("trimthenstep6: "
1664 "tcp_close.2: head not locked"));
1665 tp = tcp_close(tp);
1666 break;
1667
1668 case TCPS_TIME_WAIT:
1669 KASSERT(tp->t_state != TCPS_TIME_WAIT,
1670 ("timewait"));
1671 break;
1672 }
1673 }
1674 goto drop;
1675 }
1676
1677 /*
1678 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1679 * and it's less than ts_recent, drop it.
1680 */
1681 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1682 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1683
1684 /* Check to see if ts_recent is over 24 days old. */
1685 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1686 /*
1687 * Invalidate ts_recent. If this segment updates
1688 * ts_recent, the age will be reset later and ts_recent
1689 * will get a valid value. If it does not, setting
1690 * ts_recent to zero will at least satisfy the
1691 * requirement that zero be placed in the timestamp
1692 * echo reply when ts_recent isn't valid. The
1693 * age isn't reset until we get a valid ts_recent
1694 * because we don't want out-of-order segments to be
1695 * dropped when ts_recent is old.
1696 */
1697 tp->ts_recent = 0;
1698 } else {
1699 tcpstat.tcps_rcvduppack++;
1700 tcpstat.tcps_rcvdupbyte += tlen;
1701 tcpstat.tcps_pawsdrop++;
1702 if (tlen)
1703 goto dropafterack;
1704 goto drop;
1705 }
1706 }
1707
1708 /*
1709 * T/TCP mechanism
1710 * If T/TCP was negotiated and the segment doesn't have CC,
1711 * or if its CC is wrong then drop the segment.
1712 * RST segments do not have to comply with this.
1713 */
1714 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1715 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1716 goto dropafterack;
1717
1718 /*
1719 * In the SYN-RECEIVED state, validate that the packet belongs to
1720 * this connection before trimming the data to fit the receive
1721 * window. Check the sequence number versus IRS since we know
1722 * the sequence numbers haven't wrapped. This is a partial fix
1723 * for the "LAND" DoS attack.
1724 */
1725 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1726 rstreason = BANDLIM_RST_OPENPORT;
1727 goto dropwithreset;
1728 }
1729
1730 todrop = tp->rcv_nxt - th->th_seq;
1731 if (todrop > 0) {
1732 if (thflags & TH_SYN) {
1733 thflags &= ~TH_SYN;
1734 th->th_seq++;
1735 if (th->th_urp > 1)
1736 th->th_urp--;
1737 else
1738 thflags &= ~TH_URG;
1739 todrop--;
1740 }
1741 /*
1742 * Following if statement from Stevens, vol. 2, p. 960.
1743 */
1744 if (todrop > tlen
1745 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1746 /*
1747 * Any valid FIN must be to the left of the window.
1748 * At this point the FIN must be a duplicate or out
1749 * of sequence; drop it.
1750 */
1751 thflags &= ~TH_FIN;
1752
1753 /*
1754 * Send an ACK to resynchronize and drop any data.
1755 * But keep on processing for RST or ACK.
1756 */
1757 tp->t_flags |= TF_ACKNOW;
1758 todrop = tlen;
1759 tcpstat.tcps_rcvduppack++;
1760 tcpstat.tcps_rcvdupbyte += todrop;
1761 } else {
1762 tcpstat.tcps_rcvpartduppack++;
1763 tcpstat.tcps_rcvpartdupbyte += todrop;
1764 }
1765 drop_hdrlen += todrop; /* drop from the top afterwards */
1766 th->th_seq += todrop;
1767 tlen -= todrop;
1768 if (th->th_urp > todrop)
1769 th->th_urp -= todrop;
1770 else {
1771 thflags &= ~TH_URG;
1772 th->th_urp = 0;
1773 }
1774 }
1775
1776 /*
1777 * If new data are received on a connection after the
1778 * user processes are gone, then RST the other end.
1779 */
1780 if ((so->so_state & SS_NOFDREF) &&
1781 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1782 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1783 "locked"));
1784 tp = tcp_close(tp);
1785 tcpstat.tcps_rcvafterclose++;
1786 rstreason = BANDLIM_UNLIMITED;
1787 goto dropwithreset;
1788 }
1789
1790 /*
1791 * If segment ends after window, drop trailing data
1792 * (and PUSH and FIN); if nothing left, just ACK.
1793 */
1794 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1795 if (todrop > 0) {
1796 tcpstat.tcps_rcvpackafterwin++;
1797 if (todrop >= tlen) {
1798 tcpstat.tcps_rcvbyteafterwin += tlen;
1799 /*
1800 * If a new connection request is received
1801 * while in TIME_WAIT, drop the old connection
1802 * and start over if the sequence numbers
1803 * are above the previous ones.
1804 */
1805 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1806 if (thflags & TH_SYN &&
1807 tp->t_state == TCPS_TIME_WAIT &&
1808 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1809 KASSERT(headlocked, ("trimthenstep6: "
1810 "tcp_close.4: head not locked"));
1811 tp = tcp_close(tp);
1812 goto findpcb;
1813 }
1814 /*
1815 * If window is closed can only take segments at
1816 * window edge, and have to drop data and PUSH from
1817 * incoming segments. Continue processing, but
1818 * remember to ack. Otherwise, drop segment
1819 * and ack.
1820 */
1821 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1822 tp->t_flags |= TF_ACKNOW;
1823 tcpstat.tcps_rcvwinprobe++;
1824 } else
1825 goto dropafterack;
1826 } else
1827 tcpstat.tcps_rcvbyteafterwin += todrop;
1828 m_adj(m, -todrop);
1829 tlen -= todrop;
1830 thflags &= ~(TH_PUSH|TH_FIN);
1831 }
1832
1833 /*
1834 * If last ACK falls within this segment's sequence numbers,
1835 * record its timestamp.
1836 * NOTE:
1837 * 1) That the test incorporates suggestions from the latest
1838 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1839 * 2) That updating only on newer timestamps interferes with
1840 * our earlier PAWS tests, so this check should be solely
1841 * predicated on the sequence space of this segment.
1842 * 3) That we modify the segment boundary check to be
1843 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1844 * instead of RFC1323's
1845 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1846 * This modified check allows us to overcome RFC1323's
1847 * limitations as described in Stevens TCP/IP Illustrated
1848 * Vol. 2 p.869. In such cases, we can still calculate the
1849 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1850 */
1851 if ((to.to_flags & TOF_TS) != 0 &&
1852 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1853 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1854 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1855 tp->ts_recent_age = ticks;
1856 tp->ts_recent = to.to_tsval;
1857 }
1858
1859 /*
1860 * If a SYN is in the window, then this is an
1861 * error and we send an RST and drop the connection.
1862 */
1863 if (thflags & TH_SYN) {
1864 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1865 "head not locked"));
1866 tp = tcp_drop(tp, ECONNRESET);
1867 rstreason = BANDLIM_UNLIMITED;
1868 goto drop;
1869 }
1870
1871 /*
1872 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1873 * flag is on (half-synchronized state), then queue data for
1874 * later processing; else drop segment and return.
1875 */
1876 if ((thflags & TH_ACK) == 0) {
1877 if (tp->t_state == TCPS_SYN_RECEIVED ||
1878 (tp->t_flags & TF_NEEDSYN))
1879 goto step6;
1880 else
1881 goto drop;
1882 }
1883
1884 /*
1885 * Ack processing.
1886 */
1887 switch (tp->t_state) {
1888
1889 /*
1890 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1891 * ESTABLISHED state and continue processing.
1892 * The ACK was checked above.
1893 */
1894 case TCPS_SYN_RECEIVED:
1895
1896 tcpstat.tcps_connects++;
1897 soisconnected(so);
1898 /* Do window scaling? */
1899 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1900 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1901 tp->snd_scale = tp->requested_s_scale;
1902 tp->rcv_scale = tp->request_r_scale;
1903 }
1904 /*
1905 * Upon successful completion of 3-way handshake,
1906 * update cache.CC, pass any queued data to the user,
1907 * and advance state appropriately.
1908 */
1909 if (tcp_do_rfc1644) {
1910 tao.tao_cc = tp->cc_recv;
1911 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC,
1912 tp->cc_recv, 0);
1913 }
1914 /*
1915 * Make transitions:
1916 * SYN-RECEIVED -> ESTABLISHED
1917 * SYN-RECEIVED* -> FIN-WAIT-1
1918 */
1919 tp->t_starttime = ticks;
1920 if (tp->t_flags & TF_NEEDFIN) {
1921 tp->t_state = TCPS_FIN_WAIT_1;
1922 tp->t_flags &= ~TF_NEEDFIN;
1923 } else {
1924 tp->t_state = TCPS_ESTABLISHED;
1925 callout_reset(tp->tt_keep, tcp_keepidle,
1926 tcp_timer_keep, tp);
1927 }
1928 /*
1929 * If segment contains data or ACK, will call tcp_reass()
1930 * later; if not, do so now to pass queued data to user.
1931 */
1932 if (tlen == 0 && (thflags & TH_FIN) == 0)
1933 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1934 (struct mbuf *)0);
1935 tp->snd_wl1 = th->th_seq - 1;
1936 /* FALLTHROUGH */
1937
1938 /*
1939 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1940 * ACKs. If the ack is in the range
1941 * tp->snd_una < th->th_ack <= tp->snd_max
1942 * then advance tp->snd_una to th->th_ack and drop
1943 * data from the retransmission queue. If this ACK reflects
1944 * more up to date window information we update our window information.
1945 */
1946 case TCPS_ESTABLISHED:
1947 case TCPS_FIN_WAIT_1:
1948 case TCPS_FIN_WAIT_2:
1949 case TCPS_CLOSE_WAIT:
1950 case TCPS_CLOSING:
1951 case TCPS_LAST_ACK:
1952 case TCPS_TIME_WAIT:
1953 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1954 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1955 if (tlen == 0 && tiwin == tp->snd_wnd) {
1956 tcpstat.tcps_rcvdupack++;
1957 /*
1958 * If we have outstanding data (other than
1959 * a window probe), this is a completely
1960 * duplicate ack (ie, window info didn't
1961 * change), the ack is the biggest we've
1962 * seen and we've seen exactly our rexmt
1963 * threshhold of them, assume a packet
1964 * has been dropped and retransmit it.
1965 * Kludge snd_nxt & the congestion
1966 * window so we send only this one
1967 * packet.
1968 *
1969 * We know we're losing at the current
1970 * window size so do congestion avoidance
1971 * (set ssthresh to half the current window
1972 * and pull our congestion window back to
1973 * the new ssthresh).
1974 *
1975 * Dup acks mean that packets have left the
1976 * network (they're now cached at the receiver)
1977 * so bump cwnd by the amount in the receiver
1978 * to keep a constant cwnd packets in the
1979 * network.
1980 */
1981 if (!callout_active(tp->tt_rexmt) ||
1982 th->th_ack != tp->snd_una)
1983 tp->t_dupacks = 0;
1984 else if (++tp->t_dupacks > tcprexmtthresh ||
1985 ((tcp_do_newreno || tp->sack_enable) &&
1986 IN_FASTRECOVERY(tp))) {
1987
1988 if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1989 int data_in_pipe;
1990 int sacked, lost_not_rexmitted;
1991
1992 /*
1993 * Compute the amount of data in flight first.
1994 * We can inject new data into the pipe iff
1995 * we have less than 1/2 the original window's
1996 * worth of data in flight.
1997 */
1998 sacked = tcp_sacked_bytes(tp, &lost_not_rexmitted);
1999 data_in_pipe = (tp->snd_nxt - tp->snd_una) -
2000 (sacked + lost_not_rexmitted);
2001 if (data_in_pipe < tp->snd_ssthresh) {
2002 tp->snd_cwnd += tp->t_maxseg;
2003 if (tp->snd_cwnd > tp->snd_ssthresh)
2004 tp->snd_cwnd = tp->snd_ssthresh;
2005 }
2006 } else
2007 tp->snd_cwnd += tp->t_maxseg;
2008 (void) tcp_output(tp);
2009 goto drop;
2010 } else if (tp->t_dupacks == tcprexmtthresh) {
2011 tcp_seq onxt = tp->snd_nxt;
2012 u_int win;
2013
2014 /*
2015 * If we're doing sack, check to
2016 * see if we're already in sack
2017 * recovery. If we're not doing sack,
2018 * check to see if we're in newreno
2019 * recovery.
2020 */
2021 if (tp->sack_enable) {
2022 if (IN_FASTRECOVERY(tp)) {
2023 tp->t_dupacks = 0;
2024 break;
2025 }
2026 } else if (tcp_do_newreno) {
2027 if (SEQ_LEQ(th->th_ack,
2028 tp->snd_recover)) {
2029 tp->t_dupacks = 0;
2030 break;
2031 }
2032 }
2033 win = min(tp->snd_wnd, tp->snd_cwnd) /
2034 2 / tp->t_maxseg;
2035 if (win < 2)
2036 win = 2;
2037 tp->snd_ssthresh = win * tp->t_maxseg;
2038 ENTER_FASTRECOVERY(tp);
2039 tp->snd_recover = tp->snd_max;
2040 callout_stop(tp->tt_rexmt);
2041 tp->t_rtttime = 0;
2042 if (tp->sack_enable) {
2043 tcpstat.tcps_sack_recovery_episode++;
2044 tp->sack_newdata = tp->snd_nxt;
2045 tp->snd_cwnd = tp->t_maxseg;
2046 (void) tcp_output(tp);
2047 goto drop;
2048 }
2049 tp->snd_nxt = th->th_ack;
2050 tp->snd_cwnd = tp->t_maxseg;
2051 (void) tcp_output(tp);
2052 KASSERT(tp->snd_limited <= 2,
2053 ("tp->snd_limited too big"));
2054 tp->snd_cwnd = tp->snd_ssthresh +
2055 tp->t_maxseg *
2056 (tp->t_dupacks - tp->snd_limited);
2057 if (SEQ_GT(onxt, tp->snd_nxt))
2058 tp->snd_nxt = onxt;
2059 goto drop;
2060 } else if (tcp_do_rfc3042) {
2061 u_long oldcwnd = tp->snd_cwnd;
2062 tcp_seq oldsndmax = tp->snd_max;
2063 u_int sent;
2064
2065 KASSERT(tp->t_dupacks == 1 ||
2066 tp->t_dupacks == 2,
2067 ("dupacks not 1 or 2"));
2068 if (tp->t_dupacks == 1)
2069 tp->snd_limited = 0;
2070 tp->snd_cwnd =
2071 (tp->snd_nxt - tp->snd_una) +
2072 (tp->t_dupacks - tp->snd_limited) *
2073 tp->t_maxseg;
2074 (void) tcp_output(tp);
2075 sent = tp->snd_max - oldsndmax;
2076 if (sent > tp->t_maxseg) {
2077 KASSERT((tp->t_dupacks == 2 &&
2078 tp->snd_limited == 0) ||
2079 (sent == tp->t_maxseg + 1 &&
2080 tp->t_flags & TF_SENTFIN),
2081 ("sent too much"));
2082 tp->snd_limited = 2;
2083 } else if (sent > 0)
2084 ++tp->snd_limited;
2085 tp->snd_cwnd = oldcwnd;
2086 goto drop;
2087 }
2088 } else
2089 tp->t_dupacks = 0;
2090 break;
2091 }
2092
2093 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2094
2095 /*
2096 * If the congestion window was inflated to account
2097 * for the other side's cached packets, retract it.
2098 */
2099 if (tcp_do_newreno || tp->sack_enable) {
2100 if (IN_FASTRECOVERY(tp)) {
2101 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2102 if (tp->sack_enable)
2103 tcp_sack_partialack(tp, th);
2104 else
2105 tcp_newreno_partial_ack(tp, th);
2106 } else {
2107 /*
2108 * Out of fast recovery.
2109 * Window inflation should have left us
2110 * with approximately snd_ssthresh
2111 * outstanding data.
2112 * But in case we would be inclined to
2113 * send a burst, better to do it via
2114 * the slow start mechanism.
2115 */
2116 if (SEQ_GT(th->th_ack +
2117 tp->snd_ssthresh,
2118 tp->snd_max))
2119 tp->snd_cwnd = tp->snd_max -
2120 th->th_ack +
2121 tp->t_maxseg;
2122 else
2123 tp->snd_cwnd = tp->snd_ssthresh;
2124 }
2125 }
2126 } else {
2127 if (tp->t_dupacks >= tcprexmtthresh &&
2128 tp->snd_cwnd > tp->snd_ssthresh)
2129 tp->snd_cwnd = tp->snd_ssthresh;
2130 }
2131 tp->t_dupacks = 0;
2132 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2133 tcpstat.tcps_rcvacktoomuch++;
2134 goto dropafterack;
2135 }
2136 /*
2137 * If we reach this point, ACK is not a duplicate,
2138 * i.e., it ACKs something we sent.
2139 */
2140 if (tp->t_flags & TF_NEEDSYN) {
2141 /*
2142 * T/TCP: Connection was half-synchronized, and our
2143 * SYN has been ACK'd (so connection is now fully
2144 * synchronized). Go to non-starred state,
2145 * increment snd_una for ACK of SYN, and check if
2146 * we can do window scaling.
2147 */
2148 tp->t_flags &= ~TF_NEEDSYN;
2149 tp->snd_una++;
2150 /* Do window scaling? */
2151 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2152 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2153 tp->snd_scale = tp->requested_s_scale;
2154 tp->rcv_scale = tp->request_r_scale;
2155 }
2156 }
2157
2158 process_ACK:
2159 KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2160 "locked"));
2161 INP_LOCK_ASSERT(inp);
2162
2163 acked = th->th_ack - tp->snd_una;
2164 tcpstat.tcps_rcvackpack++;
2165 tcpstat.tcps_rcvackbyte += acked;
2166
2167 /*
2168 * If we just performed our first retransmit, and the ACK
2169 * arrives within our recovery window, then it was a mistake
2170 * to do the retransmit in the first place. Recover our
2171 * original cwnd and ssthresh, and proceed to transmit where
2172 * we left off.
2173 */
2174 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2175 ++tcpstat.tcps_sndrexmitbad;
2176 tp->snd_cwnd = tp->snd_cwnd_prev;
2177 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2178 tp->snd_recover = tp->snd_recover_prev;
2179 if (tp->t_flags & TF_WASFRECOVERY)
2180 ENTER_FASTRECOVERY(tp);
2181 tp->snd_nxt = tp->snd_max;
2182 tp->t_badrxtwin = 0; /* XXX probably not required */
2183 }
2184
2185 /*
2186 * If we have a timestamp reply, update smoothed
2187 * round trip time. If no timestamp is present but
2188 * transmit timer is running and timed sequence
2189 * number was acked, update smoothed round trip time.
2190 * Since we now have an rtt measurement, cancel the
2191 * timer backoff (cf., Phil Karn's retransmit alg.).
2192 * Recompute the initial retransmit timer.
2193 *
2194 * Some boxes send broken timestamp replies
2195 * during the SYN+ACK phase, ignore
2196 * timestamps of 0 or we could calculate a
2197 * huge RTT and blow up the retransmit timer.
2198 */
2199 if ((to.to_flags & TOF_TS) != 0 &&
2200 to.to_tsecr) {
2201 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2202 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2203 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2204 }
2205 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2206
2207 /*
2208 * If all outstanding data is acked, stop retransmit
2209 * timer and remember to restart (more output or persist).
2210 * If there is more data to be acked, restart retransmit
2211 * timer, using current (possibly backed-off) value.
2212 */
2213 if (th->th_ack == tp->snd_max) {
2214 callout_stop(tp->tt_rexmt);
2215 needoutput = 1;
2216 } else if (!callout_active(tp->tt_persist))
2217 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2218 tcp_timer_rexmt, tp);
2219
2220 /*
2221 * If no data (only SYN) was ACK'd,
2222 * skip rest of ACK processing.
2223 */
2224 if (acked == 0)
2225 goto step6;
2226
2227 /*
2228 * When new data is acked, open the congestion window.
2229 * If the window gives us less than ssthresh packets
2230 * in flight, open exponentially (maxseg per packet).
2231 * Otherwise open linearly: maxseg per window
2232 * (maxseg^2 / cwnd per packet).
2233 */
2234 if ((!tcp_do_newreno && !tp->sack_enable) ||
2235 !IN_FASTRECOVERY(tp)) {
2236 register u_int cw = tp->snd_cwnd;
2237 register u_int incr = tp->t_maxseg;
2238 if (cw > tp->snd_ssthresh)
2239 incr = incr * incr / cw;
2240 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2241 }
2242 SOCKBUF_LOCK(&so->so_snd);
2243 if (acked > so->so_snd.sb_cc) {
2244 tp->snd_wnd -= so->so_snd.sb_cc;
2245 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2246 ourfinisacked = 1;
2247 } else {
2248 sbdrop_locked(&so->so_snd, acked);
2249 tp->snd_wnd -= acked;
2250 ourfinisacked = 0;
2251 }
2252 sowwakeup_locked(so);
2253 /* detect una wraparound */
2254 if ((tcp_do_newreno || tp->sack_enable) &&
2255 !IN_FASTRECOVERY(tp) &&
2256 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2257 SEQ_LEQ(th->th_ack, tp->snd_recover))
2258 tp->snd_recover = th->th_ack - 1;
2259 if ((tcp_do_newreno || tp->sack_enable) &&
2260 IN_FASTRECOVERY(tp) &&
2261 SEQ_GEQ(th->th_ack, tp->snd_recover))
2262 EXIT_FASTRECOVERY(tp);
2263 tp->snd_una = th->th_ack;
2264 if (tp->sack_enable) {
2265 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2266 tp->snd_recover = tp->snd_una;
2267 }
2268 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2269 tp->snd_nxt = tp->snd_una;
2270
2271 switch (tp->t_state) {
2272
2273 /*
2274 * In FIN_WAIT_1 STATE in addition to the processing
2275 * for the ESTABLISHED state if our FIN is now acknowledged
2276 * then enter FIN_WAIT_2.
2277 */
2278 case TCPS_FIN_WAIT_1:
2279 if (ourfinisacked) {
2280 /*
2281 * If we can't receive any more
2282 * data, then closing user can proceed.
2283 * Starting the timer is contrary to the
2284 * specification, but if we don't get a FIN
2285 * we'll hang forever.
2286 */
2287 /* XXXjl
2288 * we should release the tp also, and use a
2289 * compressed state.
2290 */
2291 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2292 soisdisconnected(so);
2293 callout_reset(tp->tt_2msl, tcp_maxidle,
2294 tcp_timer_2msl, tp);
2295 }
2296 tp->t_state = TCPS_FIN_WAIT_2;
2297 }
2298 break;
2299
2300 /*
2301 * In CLOSING STATE in addition to the processing for
2302 * the ESTABLISHED state if the ACK acknowledges our FIN
2303 * then enter the TIME-WAIT state, otherwise ignore
2304 * the segment.
2305 */
2306 case TCPS_CLOSING:
2307 if (ourfinisacked) {
2308 KASSERT(headlocked, ("tcp_input: process_ACK: "
2309 "head not locked"));
2310 tcp_twstart(tp);
2311 INP_INFO_WUNLOCK(&tcbinfo);
2312 m_freem(m);
2313 return;
2314 }
2315 break;
2316
2317 /*
2318 * In LAST_ACK, we may still be waiting for data to drain
2319 * and/or to be acked, as well as for the ack of our FIN.
2320 * If our FIN is now acknowledged, delete the TCB,
2321 * enter the closed state and return.
2322 */
2323 case TCPS_LAST_ACK:
2324 if (ourfinisacked) {
2325 KASSERT(headlocked, ("tcp_input: process_ACK:"
2326 " tcp_close: head not locked"));
2327 tp = tcp_close(tp);
2328 goto drop;
2329 }
2330 break;
2331
2332 /*
2333 * In TIME_WAIT state the only thing that should arrive
2334 * is a retransmission of the remote FIN. Acknowledge
2335 * it and restart the finack timer.
2336 */
2337 case TCPS_TIME_WAIT:
2338 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2339 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2340 tcp_timer_2msl, tp);
2341 goto dropafterack;
2342 }
2343 }
2344
2345 step6:
2346 KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2347 INP_LOCK_ASSERT(inp);
2348
2349 /*
2350 * Update window information.
2351 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2352 */
2353 if ((thflags & TH_ACK) &&
2354 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2355 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2356 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2357 /* keep track of pure window updates */
2358 if (tlen == 0 &&
2359 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2360 tcpstat.tcps_rcvwinupd++;
2361 tp->snd_wnd = tiwin;
2362 tp->snd_wl1 = th->th_seq;
2363 tp->snd_wl2 = th->th_ack;
2364 if (tp->snd_wnd > tp->max_sndwnd)
2365 tp->max_sndwnd = tp->snd_wnd;
2366 needoutput = 1;
2367 }
2368
2369 /*
2370 * Process segments with URG.
2371 */
2372 if ((thflags & TH_URG) && th->th_urp &&
2373 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2374 /*
2375 * This is a kludge, but if we receive and accept
2376 * random urgent pointers, we'll crash in
2377 * soreceive. It's hard to imagine someone
2378 * actually wanting to send this much urgent data.
2379 */
2380 SOCKBUF_LOCK(&so->so_rcv);
2381 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2382 th->th_urp = 0; /* XXX */
2383 thflags &= ~TH_URG; /* XXX */
2384 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2385 goto dodata; /* XXX */
2386 }
2387 /*
2388 * If this segment advances the known urgent pointer,
2389 * then mark the data stream. This should not happen
2390 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2391 * a FIN has been received from the remote side.
2392 * In these states we ignore the URG.
2393 *
2394 * According to RFC961 (Assigned Protocols),
2395 * the urgent pointer points to the last octet
2396 * of urgent data. We continue, however,
2397 * to consider it to indicate the first octet
2398 * of data past the urgent section as the original
2399 * spec states (in one of two places).
2400 */
2401 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2402 tp->rcv_up = th->th_seq + th->th_urp;
2403 so->so_oobmark = so->so_rcv.sb_cc +
2404 (tp->rcv_up - tp->rcv_nxt) - 1;
2405 if (so->so_oobmark == 0)
2406 so->so_rcv.sb_state |= SBS_RCVATMARK;
2407 sohasoutofband(so);
2408 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2409 }
2410 SOCKBUF_UNLOCK(&so->so_rcv);
2411 /*
2412 * Remove out of band data so doesn't get presented to user.
2413 * This can happen independent of advancing the URG pointer,
2414 * but if two URG's are pending at once, some out-of-band
2415 * data may creep in... ick.
2416 */
2417 if (th->th_urp <= (u_long)tlen &&
2418 !(so->so_options & SO_OOBINLINE)) {
2419 /* hdr drop is delayed */
2420 tcp_pulloutofband(so, th, m, drop_hdrlen);
2421 }
2422 } else {
2423 /*
2424 * If no out of band data is expected,
2425 * pull receive urgent pointer along
2426 * with the receive window.
2427 */
2428 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2429 tp->rcv_up = tp->rcv_nxt;
2430 }
2431 dodata: /* XXX */
2432 KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2433 INP_LOCK_ASSERT(inp);
2434
2435 /*
2436 * Process the segment text, merging it into the TCP sequencing queue,
2437 * and arranging for acknowledgment of receipt if necessary.
2438 * This process logically involves adjusting tp->rcv_wnd as data
2439 * is presented to the user (this happens in tcp_usrreq.c,
2440 * case PRU_RCVD). If a FIN has already been received on this
2441 * connection then we just ignore the text.
2442 */
2443 if ((tlen || (thflags & TH_FIN)) &&
2444 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2445 tcp_seq save_start = th->th_seq;
2446 tcp_seq save_end = th->th_seq + tlen;
2447 m_adj(m, drop_hdrlen); /* delayed header drop */
2448 /*
2449 * Insert segment which includes th into TCP reassembly queue
2450 * with control block tp. Set thflags to whether reassembly now
2451 * includes a segment with FIN. This handles the common case
2452 * inline (segment is the next to be received on an established
2453 * connection, and the queue is empty), avoiding linkage into
2454 * and removal from the queue and repetition of various
2455 * conversions.
2456 * Set DELACK for segments received in order, but ack
2457 * immediately when segments are out of order (so
2458 * fast retransmit can work).
2459 */
2460 if (th->th_seq == tp->rcv_nxt &&
2461 LIST_EMPTY(&tp->t_segq) &&
2462 TCPS_HAVEESTABLISHED(tp->t_state)) {
2463 if (DELAY_ACK(tp))
2464 tp->t_flags |= TF_DELACK;
2465 else
2466 tp->t_flags |= TF_ACKNOW;
2467 tp->rcv_nxt += tlen;
2468 thflags = th->th_flags & TH_FIN;
2469 tcpstat.tcps_rcvpack++;
2470 tcpstat.tcps_rcvbyte += tlen;
2471 ND6_HINT(tp);
2472 SOCKBUF_LOCK(&so->so_rcv);
2473 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2474 m_freem(m);
2475 else
2476 sbappendstream_locked(&so->so_rcv, m);
2477 sorwakeup_locked(so);
2478 } else {
2479 thflags = tcp_reass(tp, th, &tlen, m);
2480 tp->t_flags |= TF_ACKNOW;
2481 }
2482 if (tlen > 0 && tp->sack_enable)
2483 tcp_update_sack_list(tp, save_start, save_end);
2484 /*
2485 * Note the amount of data that peer has sent into
2486 * our window, in order to estimate the sender's
2487 * buffer size.
2488 */
2489 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2490 } else {
2491 m_freem(m);
2492 thflags &= ~TH_FIN;
2493 }
2494
2495 /*
2496 * If FIN is received ACK the FIN and let the user know
2497 * that the connection is closing.
2498 */
2499 if (thflags & TH_FIN) {
2500 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2501 socantrcvmore(so);
2502 /*
2503 * If connection is half-synchronized
2504 * (ie NEEDSYN flag on) then delay ACK,
2505 * so it may be piggybacked when SYN is sent.
2506 * Otherwise, since we received a FIN then no
2507 * more input can be expected, send ACK now.
2508 */
2509 if (tp->t_flags & TF_NEEDSYN)
2510 tp->t_flags |= TF_DELACK;
2511 else
2512 tp->t_flags |= TF_ACKNOW;
2513 tp->rcv_nxt++;
2514 }
2515 switch (tp->t_state) {
2516
2517 /*
2518 * In SYN_RECEIVED and ESTABLISHED STATES
2519 * enter the CLOSE_WAIT state.
2520 */
2521 case TCPS_SYN_RECEIVED:
2522 tp->t_starttime = ticks;
2523 /*FALLTHROUGH*/
2524 case TCPS_ESTABLISHED:
2525 tp->t_state = TCPS_CLOSE_WAIT;
2526 break;
2527
2528 /*
2529 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2530 * enter the CLOSING state.
2531 */
2532 case TCPS_FIN_WAIT_1:
2533 tp->t_state = TCPS_CLOSING;
2534 break;
2535
2536 /*
2537 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2538 * starting the time-wait timer, turning off the other
2539 * standard timers.
2540 */
2541 case TCPS_FIN_WAIT_2:
2542 KASSERT(headlocked == 1, ("tcp_input: dodata: "
2543 "TCP_FIN_WAIT_2: head not locked"));
2544 tcp_twstart(tp);
2545 INP_INFO_WUNLOCK(&tcbinfo);
2546 return;
2547
2548 /*
2549 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2550 */
2551 case TCPS_TIME_WAIT:
2552 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2553 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2554 tcp_timer_2msl, tp);
2555 break;
2556 }
2557 }
2558 INP_INFO_WUNLOCK(&tcbinfo);
2559 headlocked = 0;
2560 #ifdef TCPDEBUG
2561 if (so->so_options & SO_DEBUG)
2562 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2563 &tcp_savetcp, 0);
2564 #endif
2565
2566 /*
2567 * Return any desired output.
2568 */
2569 if (needoutput || (tp->t_flags & TF_ACKNOW))
2570 (void) tcp_output(tp);
2571
2572 check_delack:
2573 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2574 INP_LOCK_ASSERT(inp);
2575 if (tp->t_flags & TF_DELACK) {
2576 tp->t_flags &= ~TF_DELACK;
2577 callout_reset(tp->tt_delack, tcp_delacktime,
2578 tcp_timer_delack, tp);
2579 }
2580 INP_UNLOCK(inp);
2581 return;
2582
2583 dropafterack:
2584 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2585 /*
2586 * Generate an ACK dropping incoming segment if it occupies
2587 * sequence space, where the ACK reflects our state.
2588 *
2589 * We can now skip the test for the RST flag since all
2590 * paths to this code happen after packets containing
2591 * RST have been dropped.
2592 *
2593 * In the SYN-RECEIVED state, don't send an ACK unless the
2594 * segment we received passes the SYN-RECEIVED ACK test.
2595 * If it fails send a RST. This breaks the loop in the
2596 * "LAND" DoS attack, and also prevents an ACK storm
2597 * between two listening ports that have been sent forged
2598 * SYN segments, each with the source address of the other.
2599 */
2600 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2601 (SEQ_GT(tp->snd_una, th->th_ack) ||
2602 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2603 rstreason = BANDLIM_RST_OPENPORT;
2604 goto dropwithreset;
2605 }
2606 #ifdef TCPDEBUG
2607 if (so->so_options & SO_DEBUG)
2608 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2609 &tcp_savetcp, 0);
2610 #endif
2611 KASSERT(headlocked, ("headlocked should be 1"));
2612 INP_INFO_WUNLOCK(&tcbinfo);
2613 tp->t_flags |= TF_ACKNOW;
2614 (void) tcp_output(tp);
2615 INP_UNLOCK(inp);
2616 m_freem(m);
2617 return;
2618
2619 dropwithreset:
2620 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2621 /*
2622 * Generate a RST, dropping incoming segment.
2623 * Make ACK acceptable to originator of segment.
2624 * Don't bother to respond if destination was broadcast/multicast.
2625 */
2626 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2627 goto drop;
2628 if (isipv6) {
2629 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2630 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2631 goto drop;
2632 } else {
2633 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2634 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2635 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2636 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2637 goto drop;
2638 }
2639 /* IPv6 anycast check is done at tcp6_input() */
2640
2641 /*
2642 * Perform bandwidth limiting.
2643 */
2644 if (badport_bandlim(rstreason) < 0)
2645 goto drop;
2646
2647 #ifdef TCPDEBUG
2648 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2649 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2650 &tcp_savetcp, 0);
2651 #endif
2652
2653 if (thflags & TH_ACK)
2654 /* mtod() below is safe as long as hdr dropping is delayed */
2655 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2656 TH_RST);
2657 else {
2658 if (thflags & TH_SYN)
2659 tlen++;
2660 /* mtod() below is safe as long as hdr dropping is delayed */
2661 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2662 (tcp_seq)0, TH_RST|TH_ACK);
2663 }
2664
2665 if (tp)
2666 INP_UNLOCK(inp);
2667 if (headlocked)
2668 INP_INFO_WUNLOCK(&tcbinfo);
2669 return;
2670
2671 drop:
2672 /*
2673 * Drop space held by incoming segment and return.
2674 */
2675 #ifdef TCPDEBUG
2676 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2677 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2678 &tcp_savetcp, 0);
2679 #endif
2680 if (tp)
2681 INP_UNLOCK(inp);
2682 if (headlocked)
2683 INP_INFO_WUNLOCK(&tcbinfo);
2684 m_freem(m);
2685 return;
2686 }
2687
2688 /*
2689 * Parse TCP options and place in tcpopt.
2690 */
2691 static void
2692 tcp_dooptions(tp, to, cp, cnt, is_syn, th)
2693 struct tcpcb *tp;
2694 struct tcpopt *to;
2695 u_char *cp;
2696 int cnt;
2697 int is_syn;
2698 struct tcphdr *th;
2699 {
2700 int opt, optlen;
2701
2702 to->to_flags = 0;
2703 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2704 opt = cp[0];
2705 if (opt == TCPOPT_EOL)
2706 break;
2707 if (opt == TCPOPT_NOP)
2708 optlen = 1;
2709 else {
2710 if (cnt < 2)
2711 break;
2712 optlen = cp[1];
2713 if (optlen < 2 || optlen > cnt)
2714 break;
2715 }
2716 switch (opt) {
2717 case TCPOPT_MAXSEG:
2718 if (optlen != TCPOLEN_MAXSEG)
2719 continue;
2720 if (!is_syn)
2721 continue;
2722 to->to_flags |= TOF_MSS;
2723 bcopy((char *)cp + 2,
2724 (char *)&to->to_mss, sizeof(to->to_mss));
2725 to->to_mss = ntohs(to->to_mss);
2726 break;
2727 case TCPOPT_WINDOW:
2728 if (optlen != TCPOLEN_WINDOW)
2729 continue;
2730 if (! is_syn)
2731 continue;
2732 to->to_flags |= TOF_SCALE;
2733 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2734 break;
2735 case TCPOPT_TIMESTAMP:
2736 if (optlen != TCPOLEN_TIMESTAMP)
2737 continue;
2738 to->to_flags |= TOF_TS;
2739 bcopy((char *)cp + 2,
2740 (char *)&to->to_tsval, sizeof(to->to_tsval));
2741 to->to_tsval = ntohl(to->to_tsval);
2742 bcopy((char *)cp + 6,
2743 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2744 to->to_tsecr = ntohl(to->to_tsecr);
2745 /*
2746 * If echoed timestamp is later than the current time,
2747 * fall back to non RFC1323 RTT calculation.
2748 */
2749 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2750 to->to_tsecr = 0;
2751 break;
2752 case TCPOPT_CC:
2753 if (optlen != TCPOLEN_CC)
2754 continue;
2755 to->to_flags |= TOF_CC;
2756 bcopy((char *)cp + 2,
2757 (char *)&to->to_cc, sizeof(to->to_cc));
2758 to->to_cc = ntohl(to->to_cc);
2759 break;
2760 case TCPOPT_CCNEW:
2761 if (optlen != TCPOLEN_CC)
2762 continue;
2763 if (!is_syn)
2764 continue;
2765 to->to_flags |= TOF_CCNEW;
2766 bcopy((char *)cp + 2,
2767 (char *)&to->to_cc, sizeof(to->to_cc));
2768 to->to_cc = ntohl(to->to_cc);
2769 break;
2770 case TCPOPT_CCECHO:
2771 if (optlen != TCPOLEN_CC)
2772 continue;
2773 if (!is_syn)
2774 continue;
2775 to->to_flags |= TOF_CCECHO;
2776 bcopy((char *)cp + 2,
2777 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2778 to->to_ccecho = ntohl(to->to_ccecho);
2779 break;
2780 #ifdef TCP_SIGNATURE
2781 /*
2782 * XXX In order to reply to a host which has set the
2783 * TCP_SIGNATURE option in its initial SYN, we have to
2784 * record the fact that the option was observed here
2785 * for the syncache code to perform the correct response.
2786 */
2787 case TCPOPT_SIGNATURE:
2788 if (optlen != TCPOLEN_SIGNATURE)
2789 continue;
2790 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2791 break;
2792 #endif
2793 case TCPOPT_SACK_PERMITTED:
2794 if (!tcp_do_sack ||
2795 optlen != TCPOLEN_SACK_PERMITTED)
2796 continue;
2797 if (is_syn) {
2798 /* MUST only be set on SYN */
2799 to->to_flags |= TOF_SACK;
2800 }
2801 break;
2802
2803 case TCPOPT_SACK:
2804 if (!tp || tcp_sack_option(tp, th, cp, optlen))
2805 continue;
2806 break;
2807 default:
2808 continue;
2809 }
2810 }
2811 }
2812
2813 /*
2814 * Pull out of band byte out of a segment so
2815 * it doesn't appear in the user's data queue.
2816 * It is still reflected in the segment length for
2817 * sequencing purposes.
2818 */
2819 static void
2820 tcp_pulloutofband(so, th, m, off)
2821 struct socket *so;
2822 struct tcphdr *th;
2823 register struct mbuf *m;
2824 int off; /* delayed to be droped hdrlen */
2825 {
2826 int cnt = off + th->th_urp - 1;
2827
2828 while (cnt >= 0) {
2829 if (m->m_len > cnt) {
2830 char *cp = mtod(m, caddr_t) + cnt;
2831 struct tcpcb *tp = sototcpcb(so);
2832
2833 tp->t_iobc = *cp;
2834 tp->t_oobflags |= TCPOOB_HAVEDATA;
2835 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2836 m->m_len--;
2837 if (m->m_flags & M_PKTHDR)
2838 m->m_pkthdr.len--;
2839 return;
2840 }
2841 cnt -= m->m_len;
2842 m = m->m_next;
2843 if (m == 0)
2844 break;
2845 }
2846 panic("tcp_pulloutofband");
2847 }
2848
2849 /*
2850 * Collect new round-trip time estimate
2851 * and update averages and current timeout.
2852 */
2853 static void
2854 tcp_xmit_timer(tp, rtt)
2855 register struct tcpcb *tp;
2856 int rtt;
2857 {
2858 register int delta;
2859
2860 INP_LOCK_ASSERT(tp->t_inpcb);
2861
2862 tcpstat.tcps_rttupdated++;
2863 tp->t_rttupdated++;
2864 if (tp->t_srtt != 0) {
2865 /*
2866 * srtt is stored as fixed point with 5 bits after the
2867 * binary point (i.e., scaled by 8). The following magic
2868 * is equivalent to the smoothing algorithm in rfc793 with
2869 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2870 * point). Adjust rtt to origin 0.
2871 */
2872 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2873 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2874
2875 if ((tp->t_srtt += delta) <= 0)
2876 tp->t_srtt = 1;
2877
2878 /*
2879 * We accumulate a smoothed rtt variance (actually, a
2880 * smoothed mean difference), then set the retransmit
2881 * timer to smoothed rtt + 4 times the smoothed variance.
2882 * rttvar is stored as fixed point with 4 bits after the
2883 * binary point (scaled by 16). The following is
2884 * equivalent to rfc793 smoothing with an alpha of .75
2885 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2886 * rfc793's wired-in beta.
2887 */
2888 if (delta < 0)
2889 delta = -delta;
2890 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2891 if ((tp->t_rttvar += delta) <= 0)
2892 tp->t_rttvar = 1;
2893 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2894 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2895 } else {
2896 /*
2897 * No rtt measurement yet - use the unsmoothed rtt.
2898 * Set the variance to half the rtt (so our first
2899 * retransmit happens at 3*rtt).
2900 */
2901 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2902 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2903 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2904 }
2905 tp->t_rtttime = 0;
2906 tp->t_rxtshift = 0;
2907
2908 /*
2909 * the retransmit should happen at rtt + 4 * rttvar.
2910 * Because of the way we do the smoothing, srtt and rttvar
2911 * will each average +1/2 tick of bias. When we compute
2912 * the retransmit timer, we want 1/2 tick of rounding and
2913 * 1 extra tick because of +-1/2 tick uncertainty in the
2914 * firing of the timer. The bias will give us exactly the
2915 * 1.5 tick we need. But, because the bias is
2916 * statistical, we have to test that we don't drop below
2917 * the minimum feasible timer (which is 2 ticks).
2918 */
2919 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2920 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2921
2922 /*
2923 * We received an ack for a packet that wasn't retransmitted;
2924 * it is probably safe to discard any error indications we've
2925 * received recently. This isn't quite right, but close enough
2926 * for now (a route might have failed after we sent a segment,
2927 * and the return path might not be symmetrical).
2928 */
2929 tp->t_softerror = 0;
2930 }
2931
2932 /*
2933 * Determine a reasonable value for maxseg size.
2934 * If the route is known, check route for mtu.
2935 * If none, use an mss that can be handled on the outgoing
2936 * interface without forcing IP to fragment; if bigger than
2937 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2938 * to utilize large mbufs. If no route is found, route has no mtu,
2939 * or the destination isn't local, use a default, hopefully conservative
2940 * size (usually 512 or the default IP max size, but no more than the mtu
2941 * of the interface), as we can't discover anything about intervening
2942 * gateways or networks. We also initialize the congestion/slow start
2943 * window to be a single segment if the destination isn't local.
2944 * While looking at the routing entry, we also initialize other path-dependent
2945 * parameters from pre-set or cached values in the routing entry.
2946 *
2947 * Also take into account the space needed for options that we
2948 * send regularly. Make maxseg shorter by that amount to assure
2949 * that we can send maxseg amount of data even when the options
2950 * are present. Store the upper limit of the length of options plus
2951 * data in maxopd.
2952 *
2953 *
2954 * In case of T/TCP, we call this routine during implicit connection
2955 * setup as well (offer = -1), to initialize maxseg from the cached
2956 * MSS of our peer.
2957 *
2958 * NOTE that this routine is only called when we process an incoming
2959 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2960 */
2961 void
2962 tcp_mss(tp, offer)
2963 struct tcpcb *tp;
2964 int offer;
2965 {
2966 int rtt, mss;
2967 u_long bufsize;
2968 u_long maxmtu;
2969 struct inpcb *inp = tp->t_inpcb;
2970 struct socket *so;
2971 struct hc_metrics_lite metrics;
2972 struct rmxp_tao tao;
2973 int origoffer = offer;
2974 #ifdef INET6
2975 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2976 size_t min_protoh = isipv6 ?
2977 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2978 sizeof (struct tcpiphdr);
2979 #else
2980 const size_t min_protoh = sizeof(struct tcpiphdr);
2981 #endif
2982 bzero(&tao, sizeof(tao));
2983
2984 /* initialize */
2985 #ifdef INET6
2986 if (isipv6) {
2987 maxmtu = tcp_maxmtu6(&inp->inp_inc);
2988 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2989 } else
2990 #endif
2991 {
2992 maxmtu = tcp_maxmtu(&inp->inp_inc);
2993 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2994 }
2995 so = inp->inp_socket;
2996
2997 /*
2998 * no route to sender, stay with default mss and return
2999 */
3000 if (maxmtu == 0)
3001 return;
3002
3003 /* what have we got? */
3004 switch (offer) {
3005 case 0:
3006 /*
3007 * Offer == 0 means that there was no MSS on the SYN
3008 * segment, in this case we use tcp_mssdflt.
3009 */
3010 offer =
3011 #ifdef INET6
3012 isipv6 ? tcp_v6mssdflt :
3013 #endif
3014 tcp_mssdflt;
3015 break;
3016
3017 case -1:
3018 /*
3019 * Offer == -1 means that we didn't receive SYN yet,
3020 * use cached value in that case;
3021 */
3022 if (tcp_do_rfc1644)
3023 tcp_hc_gettao(&inp->inp_inc, &tao);
3024 if (tao.tao_mssopt != 0)
3025 offer = tao.tao_mssopt;
3026 /* FALLTHROUGH */
3027
3028 default:
3029 /*
3030 * Prevent DoS attack with too small MSS. Round up
3031 * to at least minmss.
3032 */
3033 offer = max(offer, tcp_minmss);
3034 /*
3035 * Sanity check: make sure that maxopd will be large
3036 * enough to allow some data on segments even if the
3037 * all the option space is used (40bytes). Otherwise
3038 * funny things may happen in tcp_output.
3039 */
3040 offer = max(offer, 64);
3041 if (tcp_do_rfc1644)
3042 tcp_hc_updatetao(&inp->inp_inc,
3043 TCP_HC_TAO_MSSOPT, 0, offer);
3044 }
3045
3046 /*
3047 * rmx information is now retrieved from tcp_hostcache
3048 */
3049 tcp_hc_get(&inp->inp_inc, &metrics);
3050
3051 /*
3052 * if there's a discovered mtu int tcp hostcache, use it
3053 * else, use the link mtu.
3054 */
3055 if (metrics.rmx_mtu)
3056 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3057 else {
3058 #ifdef INET6
3059 if (isipv6) {
3060 mss = maxmtu - min_protoh;
3061 if (!path_mtu_discovery &&
3062 !in6_localaddr(&inp->in6p_faddr))
3063 mss = min(mss, tcp_v6mssdflt);
3064 } else
3065 #endif
3066 {
3067 mss = maxmtu - min_protoh;
3068 if (!path_mtu_discovery &&
3069 !in_localaddr(inp->inp_faddr))
3070 mss = min(mss, tcp_mssdflt);
3071 }
3072 }
3073 mss = min(mss, offer);
3074
3075 /*
3076 * maxopd stores the maximum length of data AND options
3077 * in a segment; maxseg is the amount of data in a normal
3078 * segment. We need to store this value (maxopd) apart
3079 * from maxseg, because now every segment carries options
3080 * and thus we normally have somewhat less data in segments.
3081 */
3082 tp->t_maxopd = mss;
3083
3084 /*
3085 * In case of T/TCP, origoffer==-1 indicates, that no segments
3086 * were received yet. In this case we just guess, otherwise
3087 * we do the same as before T/TCP.
3088 */
3089 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3090 (origoffer == -1 ||
3091 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3092 mss -= TCPOLEN_TSTAMP_APPA;
3093 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
3094 (origoffer == -1 ||
3095 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3096 mss -= TCPOLEN_CC_APPA;
3097 tp->t_maxseg = mss;
3098
3099 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3100 if (mss > MCLBYTES)
3101 mss &= ~(MCLBYTES-1);
3102 #else
3103 if (mss > MCLBYTES)
3104 mss = mss / MCLBYTES * MCLBYTES;
3105 #endif
3106 tp->t_maxseg = mss;
3107
3108 /*
3109 * If there's a pipesize, change the socket buffer to that size,
3110 * don't change if sb_hiwat is different than default (then it
3111 * has been changed on purpose with setsockopt).
3112 * Make the socket buffers an integral number of mss units;
3113 * if the mss is larger than the socket buffer, decrease the mss.
3114 */
3115 SOCKBUF_LOCK(&so->so_snd);
3116 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3117 bufsize = metrics.rmx_sendpipe;
3118 else
3119 bufsize = so->so_snd.sb_hiwat;
3120 if (bufsize < mss)
3121 mss = bufsize;
3122 else {
3123 bufsize = roundup(bufsize, mss);
3124 if (bufsize > sb_max)
3125 bufsize = sb_max;
3126 if (bufsize > so->so_snd.sb_hiwat)
3127 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3128 }
3129 SOCKBUF_UNLOCK(&so->so_snd);
3130 tp->t_maxseg = mss;
3131
3132 SOCKBUF_LOCK(&so->so_rcv);
3133 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3134 bufsize = metrics.rmx_recvpipe;
3135 else
3136 bufsize = so->so_rcv.sb_hiwat;
3137 if (bufsize > mss) {
3138 bufsize = roundup(bufsize, mss);
3139 if (bufsize > sb_max)
3140 bufsize = sb_max;
3141 if (bufsize > so->so_rcv.sb_hiwat)
3142 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3143 }
3144 SOCKBUF_UNLOCK(&so->so_rcv);
3145 /*
3146 * While we're here, check the others too
3147 */
3148 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3149 tp->t_srtt = rtt;
3150 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3151 tcpstat.tcps_usedrtt++;
3152 if (metrics.rmx_rttvar) {
3153 tp->t_rttvar = metrics.rmx_rttvar;
3154 tcpstat.tcps_usedrttvar++;
3155 } else {
3156 /* default variation is +- 1 rtt */
3157 tp->t_rttvar =
3158 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3159 }
3160 TCPT_RANGESET(tp->t_rxtcur,
3161 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3162 tp->t_rttmin, TCPTV_REXMTMAX);
3163 }
3164 if (metrics.rmx_ssthresh) {
3165 /*
3166 * There's some sort of gateway or interface
3167 * buffer limit on the path. Use this to set
3168 * the slow start threshhold, but set the
3169 * threshold to no less than 2*mss.
3170 */
3171 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3172 tcpstat.tcps_usedssthresh++;
3173 }
3174 if (metrics.rmx_bandwidth)
3175 tp->snd_bandwidth = metrics.rmx_bandwidth;
3176
3177 /*
3178 * Set the slow-start flight size depending on whether this
3179 * is a local network or not.
3180 *
3181 * Extend this so we cache the cwnd too and retrieve it here.
3182 * Make cwnd even bigger than RFC3390 suggests but only if we
3183 * have previous experience with the remote host. Be careful
3184 * not make cwnd bigger than remote receive window or our own
3185 * send socket buffer. Maybe put some additional upper bound
3186 * on the retrieved cwnd. Should do incremental updates to
3187 * hostcache when cwnd collapses so next connection doesn't
3188 * overloads the path again.
3189 *
3190 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3191 * We currently check only in syncache_socket for that.
3192 */
3193 #define TCP_METRICS_CWND
3194 #ifdef TCP_METRICS_CWND
3195 if (metrics.rmx_cwnd)
3196 tp->snd_cwnd = max(mss,
3197 min(metrics.rmx_cwnd / 2,
3198 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3199 else
3200 #endif
3201 if (tcp_do_rfc3390)
3202 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3203 #ifdef INET6
3204 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3205 (!isipv6 && in_localaddr(inp->inp_faddr)))
3206 #else
3207 else if (in_localaddr(inp->inp_faddr))
3208 #endif
3209 tp->snd_cwnd = mss * ss_fltsz_local;
3210 else
3211 tp->snd_cwnd = mss * ss_fltsz;
3212 }
3213
3214 /*
3215 * Determine the MSS option to send on an outgoing SYN.
3216 */
3217 int
3218 tcp_mssopt(inc)
3219 struct in_conninfo *inc;
3220 {
3221 int mss = 0;
3222 u_long maxmtu = 0;
3223 u_long thcmtu = 0;
3224 size_t min_protoh;
3225 #ifdef INET6
3226 int isipv6 = inc->inc_isipv6 ? 1 : 0;
3227 #endif
3228
3229 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3230
3231 #ifdef INET6
3232 if (isipv6) {
3233 mss = tcp_v6mssdflt;
3234 maxmtu = tcp_maxmtu6(inc);
3235 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3236 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3237 } else
3238 #endif
3239 {
3240 mss = tcp_mssdflt;
3241 maxmtu = tcp_maxmtu(inc);
3242 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3243 min_protoh = sizeof(struct tcpiphdr);
3244 }
3245 if (maxmtu && thcmtu)
3246 mss = min(maxmtu, thcmtu) - min_protoh;
3247 else if (maxmtu || thcmtu)
3248 mss = max(maxmtu, thcmtu) - min_protoh;
3249
3250 return (mss);
3251 }
3252
3253
3254 /*
3255 * On a partial ack arrives, force the retransmission of the
3256 * next unacknowledged segment. Do not clear tp->t_dupacks.
3257 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3258 * be started again.
3259 */
3260 static void
3261 tcp_newreno_partial_ack(tp, th)
3262 struct tcpcb *tp;
3263 struct tcphdr *th;
3264 {
3265 tcp_seq onxt = tp->snd_nxt;
3266 u_long ocwnd = tp->snd_cwnd;
3267
3268 callout_stop(tp->tt_rexmt);
3269 tp->t_rtttime = 0;
3270 tp->snd_nxt = th->th_ack;
3271 /*
3272 * Set snd_cwnd to one segment beyond acknowledged offset.
3273 * (tp->snd_una has not yet been updated when this function is called.)
3274 */
3275 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3276 tp->t_flags |= TF_ACKNOW;
3277 (void) tcp_output(tp);
3278 tp->snd_cwnd = ocwnd;
3279 if (SEQ_GT(onxt, tp->snd_nxt))
3280 tp->snd_nxt = onxt;
3281 /*
3282 * Partial window deflation. Relies on fact that tp->snd_una
3283 * not updated yet.
3284 */
3285 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3286 }
3287
3288 /*
3289 * Returns 1 if the TIME_WAIT state was killed and we should start over,
3290 * looking for a pcb in the listen state. Returns 0 otherwise.
3291 */
3292 static int
3293 tcp_timewait(tw, to, th, m, tlen)
3294 struct tcptw *tw;
3295 struct tcpopt *to;
3296 struct tcphdr *th;
3297 struct mbuf *m;
3298 int tlen;
3299 {
3300 int thflags;
3301 tcp_seq seq;
3302 #ifdef INET6
3303 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3304 #else
3305 const int isipv6 = 0;
3306 #endif
3307
3308 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */
3309 INP_INFO_WLOCK_ASSERT(&tcbinfo);
3310 INP_LOCK_ASSERT(tw->tw_inpcb);
3311
3312 thflags = th->th_flags;
3313
3314 /*
3315 * NOTE: for FIN_WAIT_2 (to be added later),
3316 * must validate sequence number before accepting RST
3317 */
3318
3319 /*
3320 * If the segment contains RST:
3321 * Drop the segment - see Stevens, vol. 2, p. 964 and
3322 * RFC 1337.
3323 */
3324 if (thflags & TH_RST)
3325 goto drop;
3326
3327 /*
3328 * If segment contains a SYN and CC [not CC.NEW] option:
3329 * if connection duration > MSL, drop packet and send RST;
3330 *
3331 * if SEG.CC > CCrecv then is new SYN.
3332 * Complete close and delete TCPCB. Then reprocess
3333 * segment, hoping to find new TCPCB in LISTEN state;
3334 *
3335 * else must be old SYN; drop it.
3336 * else do normal processing.
3337 */
3338 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) {
3339 if ((ticks - tw->t_starttime) > tcp_msl)
3340 goto reset;
3341 if (CC_GT(to->to_cc, tw->cc_recv)) {
3342 (void) tcp_twclose(tw, 0);
3343 return (1);
3344 }
3345 goto drop;
3346 }
3347
3348 #if 0
3349 /* PAWS not needed at the moment */
3350 /*
3351 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3352 * and it's less than ts_recent, drop it.
3353 */
3354 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3355 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3356 if ((thflags & TH_ACK) == 0)
3357 goto drop;
3358 goto ack;
3359 }
3360 /*
3361 * ts_recent is never updated because we never accept new segments.
3362 */
3363 #endif
3364
3365 /*
3366 * If a new connection request is received
3367 * while in TIME_WAIT, drop the old connection
3368 * and start over if the sequence numbers
3369 * are above the previous ones.
3370 */
3371 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3372 (void) tcp_twclose(tw, 0);
3373 return (1);
3374 }
3375
3376 /*
3377 * Drop the the segment if it does not contain an ACK.
3378 */
3379 if ((thflags & TH_ACK) == 0)
3380 goto drop;
3381
3382 /*
3383 * Reset the 2MSL timer if this is a duplicate FIN.
3384 */
3385 if (thflags & TH_FIN) {
3386 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3387 if (seq + 1 == tw->rcv_nxt)
3388 tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3389 }
3390
3391 /*
3392 * Acknowledge the segment if it has data or is not a duplicate ACK.
3393 */
3394 if (thflags != TH_ACK || tlen != 0 ||
3395 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3396 tcp_twrespond(tw, TH_ACK);
3397 goto drop;
3398
3399 reset:
3400 /*
3401 * Generate a RST, dropping incoming segment.
3402 * Make ACK acceptable to originator of segment.
3403 * Don't bother to respond if destination was broadcast/multicast.
3404 */
3405 if (m->m_flags & (M_BCAST|M_MCAST))
3406 goto drop;
3407 if (isipv6) {
3408 struct ip6_hdr *ip6;
3409
3410 /* IPv6 anycast check is done at tcp6_input() */
3411 ip6 = mtod(m, struct ip6_hdr *);
3412 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3413 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3414 goto drop;
3415 } else {
3416 struct ip *ip;
3417
3418 ip = mtod(m, struct ip *);
3419 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3420 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3421 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3422 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3423 goto drop;
3424 }
3425 if (thflags & TH_ACK) {
3426 tcp_respond(NULL,
3427 mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3428 } else {
3429 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3430 tcp_respond(NULL,
3431 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3432 }
3433 INP_UNLOCK(tw->tw_inpcb);
3434 return (0);
3435
3436 drop:
3437 INP_UNLOCK(tw->tw_inpcb);
3438 m_freem(m);
3439 return (0);
3440 }
Cache object: 25a8db9b5ceacc63ef0b77310111785b
|