Undo part of the last commit. OBJFORMAT_PATH controls how the cross
[dragonfly.git] / sys / netinet / tcp_input.c
CommitLineData
984263bc 1/*
8819433a 2 * Copyright (c) 2002-2004 Jeffrey Hsu. All rights reserved.
984263bc
MD
3 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
4 * The Regents of the University of California. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by the University of
17 * California, Berkeley and its contributors.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
35 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
83be63fe 36 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.20 2004/03/09 21:21:54 hsu Exp $
984263bc
MD
37 */
38
39#include "opt_ipfw.h" /* for ipfw_fwd */
40#include "opt_inet6.h"
41#include "opt_ipsec.h"
42#include "opt_tcpdebug.h"
43#include "opt_tcp_input.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/kernel.h>
48#include <sys/sysctl.h>
49#include <sys/malloc.h>
50#include <sys/mbuf.h>
51#include <sys/proc.h> /* for proc0 declaration */
52#include <sys/protosw.h>
53#include <sys/socket.h>
54#include <sys/socketvar.h>
55#include <sys/syslog.h>
3f9db7f8 56#include <sys/in_cksum.h>
984263bc
MD
57
58#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
59
60#include <net/if.h>
61#include <net/route.h>
62
63#include <netinet/in.h>
64#include <netinet/in_systm.h>
65#include <netinet/ip.h>
66#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
67#include <netinet/in_var.h>
68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
69#include <netinet/in_pcb.h>
70#include <netinet/ip_var.h>
71#include <netinet/ip6.h>
72#include <netinet/icmp6.h>
73#include <netinet6/nd6.h>
74#include <netinet6/ip6_var.h>
75#include <netinet6/in6_pcb.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
86u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
87struct tcphdr tcp_savetcp;
88#endif /* TCPDEBUG */
89
90#ifdef FAST_IPSEC
91#include <netipsec/ipsec.h>
92#include <netipsec/ipsec6.h>
93#endif
94
95#ifdef IPSEC
96#include <netinet6/ipsec.h>
97#include <netinet6/ipsec6.h>
d2438d69 98#include <netproto/key/key.h>
984263bc
MD
99#endif /*IPSEC*/
100
984263bc
MD
101MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
102
103static const int tcprexmtthresh = 3;
104tcp_cc tcp_ccgen;
105
106struct tcpstat tcpstat;
107SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
108 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
109
110static int log_in_vain = 0;
111SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &log_in_vain, 0, "Log all incoming TCP connections");
113
114static int blackhole = 0;
115SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
116 &blackhole, 0, "Do not send RST when dropping refused connections");
117
118int tcp_delack_enabled = 1;
119SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
120 &tcp_delack_enabled, 0,
121 "Delay ACK to try and piggyback it onto a data packet");
122
123#ifdef TCP_DROP_SYNFIN
124static int drop_synfin = 0;
125SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
126 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
127#endif
128
33abdd1c
MD
129static int tcp_do_limitedtransmit = 1;
130SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
131 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
132
8819433a
JH
133static int tcp_do_early_retransmit = 0;
134SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
135 &tcp_do_early_retransmit, 0, "Early retransmit");
136
4b52d5ee
JH
137static int tcp_do_rfc3390 = 1;
138SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
139 &tcp_do_rfc3390, 0,
140 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
141
efd4b327
JH
142static int tcp_do_eifel_detect = 1;
143SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
144 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
145
3edf7c37
RG
146SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
147 "TCP Segment Reassembly Queue");
148
149int tcp_reass_maxseg = 0;
150SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
151 &tcp_reass_maxseg, 0,
152 "Global maximum number of TCP Segments in Reassembly Queue");
153
154int tcp_reass_qsize = 0;
155SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
156 &tcp_reass_qsize, 0,
157 "Global number of TCP Segments currently in Reassembly Queue");
158
159static int tcp_reass_overflows = 0;
160SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
161 &tcp_reass_overflows, 0,
162 "Global number of TCP Segment Reassembly Queue Overflows");
163
d371a63a 164struct inpcbinfo tcbinfo[MAXCPU];
984263bc
MD
165
166static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
167static void tcp_pulloutofband(struct socket *,
168 struct tcphdr *, struct mbuf *, int);
169static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
170 struct mbuf *);
171static void tcp_xmit_timer(struct tcpcb *, int);
172static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
173
174/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
175#ifdef INET6
176#define ND6_HINT(tp) \
177do { \
178 if ((tp) && (tp)->t_inpcb && \
179 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
180 (tp)->t_inpcb->in6p_route.ro_rt) \
181 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
182} while (0)
183#else
184#define ND6_HINT(tp)
185#endif
186
187/*
188 * Indicate whether this ack should be delayed. We can delay the ack if
189 * - delayed acks are enabled and
190 * - there is no delayed ack timer in progress and
191 * - our last ack wasn't a 0-sized window. We never want to delay
192 * the ack that opens up a 0-sized window.
193 */
194#define DELAY_ACK(tp) \
195 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
196 (tp->t_flags & TF_RXWIN0SENT) == 0)
197
198static int
199tcp_reass(tp, th, tlenp, m)
2256ba69
RG
200 struct tcpcb *tp;
201 struct tcphdr *th;
984263bc
MD
202 int *tlenp;
203 struct mbuf *m;
204{
205 struct tseg_qent *q;
206 struct tseg_qent *p = NULL;
207 struct tseg_qent *nq;
208 struct tseg_qent *te;
209 struct socket *so = tp->t_inpcb->inp_socket;
210 int flags;
211
212 /*
213 * Call with th==0 after become established to
214 * force pre-ESTABLISHED data up to user socket.
215 */
216 if (th == 0)
217 goto present;
218
3edf7c37
RG
219 /*
220 * Limit the number of segments in the reassembly queue to prevent
221 * holding on to too many segments (and thus running out of mbufs).
222 * Make sure to let the missing segment through which caused this
223 * queue. Always keep one global queue entry spare to be able to
224 * process the missing segment.
225 */
226 if (th->th_seq != tp->rcv_nxt &&
227 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
228 tcp_reass_overflows++;
229 tcpstat.tcps_rcvmemdrop++;
230 m_freem(m);
231 return (0);
232 }
233
984263bc
MD
234 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
235 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
236 M_NOWAIT);
237 if (te == NULL) {
238 tcpstat.tcps_rcvmemdrop++;
239 m_freem(m);
240 return (0);
241 }
3edf7c37 242 tcp_reass_qsize++;
984263bc
MD
243
244 /*
245 * Find a segment which begins after this one does.
246 */
247 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
248 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
249 break;
250 p = q;
251 }
252
253 /*
254 * If there is a preceding segment, it may provide some of
255 * our data already. If so, drop the data from the incoming
256 * segment. If it provides all of our data, drop us.
257 */
258 if (p != NULL) {
2256ba69 259 int i;
984263bc
MD
260 /* conversion to int (in i) handles seq wraparound */
261 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
262 if (i > 0) {
263 if (i >= *tlenp) {
264 tcpstat.tcps_rcvduppack++;
265 tcpstat.tcps_rcvdupbyte += *tlenp;
266 m_freem(m);
267 free(te, M_TSEGQ);
3edf7c37 268 tcp_reass_qsize--;
984263bc
MD
269 /*
270 * Try to present any queued data
271 * at the left window edge to the user.
272 * This is needed after the 3-WHS
273 * completes.
274 */
275 goto present; /* ??? */
276 }
277 m_adj(m, i);
278 *tlenp -= i;
279 th->th_seq += i;
280 }
281 }
282 tcpstat.tcps_rcvoopack++;
283 tcpstat.tcps_rcvoobyte += *tlenp;
284
285 /*
286 * While we overlap succeeding segments trim them or,
287 * if they are completely covered, dequeue them.
288 */
289 while (q) {
2256ba69 290 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
984263bc
MD
291 if (i <= 0)
292 break;
293 if (i < q->tqe_len) {
294 q->tqe_th->th_seq += i;
295 q->tqe_len -= i;
296 m_adj(q->tqe_m, i);
297 break;
298 }
299
300 nq = LIST_NEXT(q, tqe_q);
301 LIST_REMOVE(q, tqe_q);
302 m_freem(q->tqe_m);
303 free(q, M_TSEGQ);
3edf7c37 304 tcp_reass_qsize--;
984263bc
MD
305 q = nq;
306 }
307
308 /* Insert the new segment queue entry into place. */
309 te->tqe_m = m;
310 te->tqe_th = th;
311 te->tqe_len = *tlenp;
312
313 if (p == NULL) {
314 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
315 } else {
316 LIST_INSERT_AFTER(p, te, tqe_q);
317 }
318
319present:
320 /*
321 * Present data to user, advancing rcv_nxt through
322 * completed sequence space.
323 */
324 if (!TCPS_HAVEESTABLISHED(tp->t_state))
325 return (0);
326 q = LIST_FIRST(&tp->t_segq);
327 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
328 return (0);
329 do {
330 tp->rcv_nxt += q->tqe_len;
331 flags = q->tqe_th->th_flags & TH_FIN;
332 nq = LIST_NEXT(q, tqe_q);
333 LIST_REMOVE(q, tqe_q);
334 if (so->so_state & SS_CANTRCVMORE)
335 m_freem(q->tqe_m);
336 else
337 sbappend(&so->so_rcv, q->tqe_m);
338 free(q, M_TSEGQ);
3edf7c37 339 tcp_reass_qsize--;
984263bc
MD
340 q = nq;
341 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
342 ND6_HINT(tp);
343 sorwakeup(so);
344 return (flags);
345}
346
347/*
348 * TCP input routine, follows pages 65-76 of the
349 * protocol specification dated September, 1981 very closely.
350 */
351#ifdef INET6
352int
353tcp6_input(mp, offp, proto)
354 struct mbuf **mp;
355 int *offp, proto;
356{
2256ba69 357 struct mbuf *m = *mp;
984263bc
MD
358 struct in6_ifaddr *ia6;
359
360 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
361
362 /*
363 * draft-itojun-ipv6-tcp-to-anycast
364 * better place to put this in?
365 */
366 ia6 = ip6_getdstifaddr(m);
367 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
368 struct ip6_hdr *ip6;
369
370 ip6 = mtod(m, struct ip6_hdr *);
371 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
372 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
373 return IPPROTO_DONE;
374 }
375
376 tcp_input(m, *offp, proto);
377 return IPPROTO_DONE;
378}
379#endif
380
381void
382tcp_input(m, off0, proto)
2256ba69 383 struct mbuf *m;
984263bc
MD
384 int off0, proto;
385{
2256ba69
RG
386 struct tcphdr *th;
387 struct ip *ip = NULL;
388 struct ipovly *ipov;
389 struct inpcb *inp = NULL;
984263bc
MD
390 u_char *optp = NULL;
391 int optlen = 0;
392 int len, tlen, off;
393 int drop_hdrlen;
2256ba69
RG
394 struct tcpcb *tp = NULL;
395 int thflags;
984263bc
MD
396 struct socket *so = 0;
397 int todrop, acked, ourfinisacked, needoutput = 0;
398 u_long tiwin;
399 struct tcpopt to; /* options in this segment */
400 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
401 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
402 struct sockaddr_in *next_hop = NULL;
403 int rstreason; /* For badport_bandlim accounting purposes */
d371a63a 404 int cpu;
984263bc
MD
405 struct ip6_hdr *ip6 = NULL;
406#ifdef INET6
407 int isipv6;
408#else
409 const int isipv6 = 0;
410#endif
411#ifdef TCPDEBUG
412 short ostate = 0;
413#endif
414
415 /* Grab info from MT_TAG mbufs prepended to the chain. */
416 for (;m && m->m_type == MT_TAG; m = m->m_next) {
417 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
418 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
419 }
420#ifdef INET6
421 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
422#endif
423 bzero((char *)&to, sizeof(to));
424
425 tcpstat.tcps_rcvtotal++;
426
427 if (isipv6) {
428 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
429 ip6 = mtod(m, struct ip6_hdr *);
430 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
431 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
432 tcpstat.tcps_rcvbadsum++;
433 goto drop;
434 }
435 th = (struct tcphdr *)((caddr_t)ip6 + off0);
436
437 /*
438 * Be proactive about unspecified IPv6 address in source.
439 * As we use all-zero to indicate unbounded/unconnected pcb,
440 * unspecified IPv6 address can be used to confuse us.
441 *
442 * Note that packets with unspecified IPv6 destination is
443 * already dropped in ip6_input.
444 */
445 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
446 /* XXX stat */
447 goto drop;
448 }
449 } else {
450 /*
451 * Get IP and TCP header together in first mbuf.
452 * Note: IP leaves IP header in first mbuf.
453 */
454 if (off0 > sizeof(struct ip)) {
bddf0751 455 ip_stripoptions(m);
984263bc
MD
456 off0 = sizeof(struct ip);
457 }
458 if (m->m_len < sizeof(struct tcpiphdr)) {
459 if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) {
460 tcpstat.tcps_rcvshort++;
461 return;
462 }
463 }
464 ip = mtod(m, struct ip *);
465 ipov = (struct ipovly *)ip;
466 th = (struct tcphdr *)((caddr_t)ip + off0);
467 tlen = ip->ip_len;
468
469 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
470 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
471 th->th_sum = m->m_pkthdr.csum_data;
472 else
473 th->th_sum = in_pseudo(ip->ip_src.s_addr,
474 ip->ip_dst.s_addr,
475 htonl(m->m_pkthdr.csum_data +
476 ip->ip_len +
477 IPPROTO_TCP));
478 th->th_sum ^= 0xffff;
479 } else {
480 /*
481 * Checksum extended TCP header and data.
482 */
483 len = sizeof(struct ip) + tlen;
484 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
485 ipov->ih_len = (u_short)tlen;
486 ipov->ih_len = htons(ipov->ih_len);
487 th->th_sum = in_cksum(m, len);
488 }
489 if (th->th_sum) {
490 tcpstat.tcps_rcvbadsum++;
491 goto drop;
492 }
493#ifdef INET6
494 /* Re-initialization for later version check */
495 ip->ip_v = IPVERSION;
496#endif
497 }
498
499 /*
500 * Check that TCP offset makes sense,
501 * pull out TCP options and adjust length. XXX
502 */
503 off = th->th_off << 2;
504 if (off < sizeof(struct tcphdr) || off > tlen) {
505 tcpstat.tcps_rcvbadoff++;
506 goto drop;
507 }
508 tlen -= off; /* tlen is used instead of ti->ti_len */
509 if (off > sizeof(struct tcphdr)) {
510 if (isipv6) {
511 IP6_EXTHDR_CHECK(m, off0, off, );
512 ip6 = mtod(m, struct ip6_hdr *);
513 th = (struct tcphdr *)((caddr_t)ip6 + off0);
514 } else {
515 if (m->m_len < sizeof(struct ip) + off) {
516 if ((m = m_pullup(m, sizeof(struct ip) + off))
517 == 0) {
518 tcpstat.tcps_rcvshort++;
519 return;
520 }
521 ip = mtod(m, struct ip *);
522 ipov = (struct ipovly *)ip;
523 th = (struct tcphdr *)((caddr_t)ip + off0);
524 }
525 }
526 optlen = off - sizeof(struct tcphdr);
527 optp = (u_char *)(th + 1);
528 }
529 thflags = th->th_flags;
530
531#ifdef TCP_DROP_SYNFIN
532 /*
533 * If the drop_synfin option is enabled, drop all packets with
534 * both the SYN and FIN bits set. This prevents e.g. nmap from
535 * identifying the TCP/IP stack.
536 *
537 * This is a violation of the TCP specification.
538 */
539 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
540 goto drop;
541#endif
542
543 /*
544 * Convert TCP protocol specific fields to host format.
545 */
546 th->th_seq = ntohl(th->th_seq);
547 th->th_ack = ntohl(th->th_ack);
548 th->th_win = ntohs(th->th_win);
549 th->th_urp = ntohs(th->th_urp);
550
551 /*
552 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
553 * until after ip6_savecontrol() is called and before other functions
554 * which don't want those proto headers.
555 * Because ip6_savecontrol() is going to parse the mbuf to
556 * search for data to be passed up to user-land, it wants mbuf
557 * parameters to be unchanged.
558 * XXX: the call of ip6_savecontrol() has been obsoleted based on
559 * latest version of the advanced API (20020110).
560 */
561 drop_hdrlen = off0 + off;
562
563 /*
564 * Locate pcb for segment.
565 */
566findpcb:
567 /* IPFIREWALL_FORWARD section */
568 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */
569 /*
570 * Transparently forwarded. Pretend to be the destination.
571 * already got one like this?
572 */
d371a63a
JH
573 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
574 ip->ip_src, th->th_sport,
984263bc
MD
575 ip->ip_dst, th->th_dport,
576 0, m->m_pkthdr.rcvif);
577 if (!inp) {
83be63fe
JH
578 /*
579 * It's new. Try to find the ambushing socket.
580 */
581
582 /*
583 * The rest of the ipfw code stores the port in
584 * host order. XXX
585 * (The IP address is still in network order.)
586 */
587 in_port_t dport = next_hop->sin_port ?
588 htons(next_hop->sin_port) :
589 th->th_dport;
590
d371a63a 591 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
83be63fe 592 next_hop->sin_addr.s_addr, dport);
d371a63a 593 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc 594 ip->ip_src, th->th_sport,
83be63fe 595 next_hop->sin_addr, dport,
984263bc
MD
596 1, m->m_pkthdr.rcvif);
597 }
598 } else {
599 if (isipv6)
d371a63a 600 inp = in6_pcblookup_hash(&tcbinfo[0],
984263bc
MD
601 &ip6->ip6_src, th->th_sport,
602 &ip6->ip6_dst, th->th_dport,
603 1, m->m_pkthdr.rcvif);
604 else
d371a63a 605 inp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
984263bc
MD
606 ip->ip_src, th->th_sport,
607 ip->ip_dst, th->th_dport,
608 1, m->m_pkthdr.rcvif);
609 }
610
611#ifdef IPSEC
612 if (isipv6) {
613 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
614 ipsec6stat.in_polvio++;
615 goto drop;
616 }
617 } else {
618 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
619 ipsecstat.in_polvio++;
620 goto drop;
621 }
622 }
623#endif
624#ifdef FAST_IPSEC
625 if (isipv6) {
626 if (inp != NULL && ipsec6_in_reject(m, inp)) {
627 goto drop;
628 }
629 } else {
630 if (inp != NULL && ipsec4_in_reject(m, inp)) {
631 goto drop;
632 }
633 }
634#endif
635
636 /*
637 * If the state is CLOSED (i.e., TCB does not exist) then
638 * all data in the incoming segment is discarded.
639 * If the TCB exists but is in CLOSED state, it is embryonic,
640 * but should either do a listen or a connect soon.
641 */
642 if (inp == NULL) {
643 if (log_in_vain) {
644#ifdef INET6
645 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
646#else
647 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
648#endif
649 if (isipv6) {
650 strcpy(dbuf, "[");
651 strcpy(sbuf, "[");
652 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
653 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
654 strcat(dbuf, "]");
655 strcat(sbuf, "]");
656 } else {
657 strcpy(dbuf, inet_ntoa(ip->ip_dst));
658 strcpy(sbuf, inet_ntoa(ip->ip_src));
659 }
660 switch (log_in_vain) {
661 case 1:
662 if ((thflags & TH_SYN) == 0)
663 break;
664 case 2:
665 log(LOG_INFO,
666 "Connection attempt to TCP %s:%d "
667 "from %s:%d flags:0x%02x\n",
668 dbuf, ntohs(th->th_dport), sbuf,
669 ntohs(th->th_sport), thflags);
670 break;
671 default:
672 break;
673 }
674 }
675 if (blackhole) {
676 switch (blackhole) {
677 case 1:
678 if (thflags & TH_SYN)
679 goto drop;
680 break;
681 case 2:
682 goto drop;
683 default:
684 goto drop;
685 }
686 }
687 rstreason = BANDLIM_RST_CLOSEDPORT;
688 goto dropwithreset;
689 }
690 tp = intotcpcb(inp);
691 if (tp == NULL) {
692 rstreason = BANDLIM_RST_CLOSEDPORT;
693 goto dropwithreset;
694 }
695 if (tp->t_state == TCPS_CLOSED)
696 goto drop;
697
698 /* Unscale the window into a 32-bit value. */
699 if ((thflags & TH_SYN) == 0)
700 tiwin = th->th_win << tp->snd_scale;
701 else
702 tiwin = th->th_win;
703
704 so = inp->inp_socket;
47654766 705
984263bc 706#ifdef TCPDEBUG
47654766
JH
707 if (so->so_options & SO_DEBUG) {
708 ostate = tp->t_state;
709 if (isipv6)
710 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
711 else
712 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
713 tcp_savetcp = *th;
714 }
984263bc 715#endif
47654766
JH
716
717 if (so->so_options & SO_ACCEPTCONN) {
718 struct in_conninfo inc;
719
984263bc
MD
720#ifdef INET6
721 inc.inc_isipv6 = isipv6;
722#endif
723 if (isipv6) {
724 inc.inc6_faddr = ip6->ip6_src;
725 inc.inc6_laddr = ip6->ip6_dst;
726 inc.inc6_route.ro_rt = NULL; /* XXX */
727 } else {
728 inc.inc_faddr = ip->ip_src;
729 inc.inc_laddr = ip->ip_dst;
730 inc.inc_route.ro_rt = NULL; /* XXX */
731 }
732 inc.inc_fport = th->th_sport;
733 inc.inc_lport = th->th_dport;
734
735 /*
736 * If the state is LISTEN then ignore segment if it contains
737 * a RST. If the segment contains an ACK then it is bad and
738 * send a RST. If it does not contain a SYN then it is not
739 * interesting; drop it.
740 *
741 * If the state is SYN_RECEIVED (syncache) and seg contains
742 * an ACK, but not for our SYN/ACK, send a RST. If the seg
743 * contains a RST, check the sequence number to see if it
744 * is a valid reset segment.
745 */
746 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
747 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
748 if (!syncache_expand(&inc, th, &so, m)) {
749 /*
750 * No syncache entry, or ACK was not
751 * for our SYN/ACK. Send a RST.
752 */
753 tcpstat.tcps_badsyn++;
754 rstreason = BANDLIM_RST_OPENPORT;
755 goto dropwithreset;
756 }
757 if (so == NULL)
758 /*
759 * Could not complete 3-way handshake,
760 * connection is being closed down, and
761 * syncache will free mbuf.
762 */
763 return;
764 /*
765 * Socket is created in state SYN_RECEIVED.
766 * Continue processing segment.
767 */
768 inp = sotoinpcb(so);
769 tp = intotcpcb(inp);
770 /*
771 * This is what would have happened in
772 * tcp_output() when the SYN,ACK was sent.
773 */
774 tp->snd_up = tp->snd_una;
775 tp->snd_max = tp->snd_nxt = tp->iss + 1;
776 tp->last_ack_sent = tp->rcv_nxt;
777/*
778 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
779 * until the _second_ ACK is received:
780 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
781 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
782 * move to ESTAB, set snd_wnd to tiwin.
783 */
784 tp->snd_wnd = tiwin; /* unscaled */
785 goto after_listen;
786 }
787 if (thflags & TH_RST) {
788 syncache_chkrst(&inc, th);
789 goto drop;
790 }
791 if (thflags & TH_ACK) {
792 syncache_badack(&inc);
793 tcpstat.tcps_badsyn++;
794 rstreason = BANDLIM_RST_OPENPORT;
795 goto dropwithreset;
796 }
797 goto drop;
798 }
799
800 /*
801 * Segment's flags are (SYN) or (SYN|FIN).
802 */
803#ifdef INET6
804 /*
805 * If deprecated address is forbidden,
806 * we do not accept SYN to deprecated interface
807 * address to prevent any new inbound connection from
808 * getting established.
809 * When we do not accept SYN, we send a TCP RST,
810 * with deprecated source address (instead of dropping
811 * it). We compromise it as it is much better for peer
812 * to send a RST, and RST will be the final packet
813 * for the exchange.
814 *
815 * If we do not forbid deprecated addresses, we accept
816 * the SYN packet. RFC2462 does not suggest dropping
817 * SYN in this case.
818 * If we decipher RFC2462 5.5.4, it says like this:
819 * 1. use of deprecated addr with existing
820 * communication is okay - "SHOULD continue to be
821 * used"
822 * 2. use of it with new communication:
823 * (2a) "SHOULD NOT be used if alternate address
824 * with sufficient scope is available"
825 * (2b) nothing mentioned otherwise.
826 * Here we fall into (2b) case as we have no choice in
827 * our source address selection - we must obey the peer.
828 *
829 * The wording in RFC2462 is confusing, and there are
830 * multiple description text for deprecated address
831 * handling - worse, they are not exactly the same.
832 * I believe 5.5.4 is the best one, so we follow 5.5.4.
833 */
834 if (isipv6 && !ip6_use_deprecated) {
835 struct in6_ifaddr *ia6;
836
837 if ((ia6 = ip6_getdstifaddr(m)) &&
838 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
839 tp = NULL;
840 rstreason = BANDLIM_RST_OPENPORT;
841 goto dropwithreset;
842 }
843 }
844#endif
845 /*
846 * If it is from this socket, drop it, it must be forged.
847 * Don't bother responding if the destination was a broadcast.
848 */
849 if (th->th_dport == th->th_sport) {
850 if (isipv6) {
851 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
852 &ip6->ip6_src))
853 goto drop;
854 } else {
855 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
856 goto drop;
857 }
858 }
859 /*
860 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
861 *
862 * Note that it is quite possible to receive unicast
863 * link-layer packets with a broadcast IP address. Use
864 * in_broadcast() to find them.
865 */
866 if (m->m_flags & (M_BCAST|M_MCAST))
867 goto drop;
868 if (isipv6) {
869 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
870 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
871 goto drop;
872 } else {
873 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
874 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
875 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
876 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
877 goto drop;
878 }
879 /*
880 * SYN appears to be valid; create compressed TCP state
881 * for syncache, or perform t/tcp connection.
882 */
883 if (so->so_qlen <= so->so_qlimit) {
884 tcp_dooptions(&to, optp, optlen, 1);
885 if (!syncache_add(&inc, &to, th, &so, m))
886 goto drop;
887 if (so == NULL)
888 /*
889 * Entry added to syncache, mbuf used to
890 * send SYN,ACK packet.
891 */
892 return;
893 /*
894 * Segment passed TAO tests.
895 */
896 inp = sotoinpcb(so);
897 tp = intotcpcb(inp);
898 tp->snd_wnd = tiwin;
899 tp->t_starttime = ticks;
900 tp->t_state = TCPS_ESTABLISHED;
901
902 /*
903 * If there is a FIN, or if there is data and the
904 * connection is local, then delay SYN,ACK(SYN) in
905 * the hope of piggy-backing it on a response
906 * segment. Otherwise must send ACK now in case
907 * the other side is slow starting.
908 */
909 if (DELAY_ACK(tp) &&
910 ((thflags & TH_FIN) ||
911 (tlen != 0 &&
912 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
913 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
914 callout_reset(tp->tt_delack, tcp_delacktime,
915 tcp_timer_delack, tp);
916 tp->t_flags |= TF_NEEDSYN;
917 } else
918 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
919
920 tcpstat.tcps_connects++;
921 soisconnected(so);
922 goto trimthenstep6;
923 }
924 goto drop;
925 }
926after_listen:
927
928/* XXX temp debugging */
929 /* should not happen - syncache should pick up these connections */
930 if (tp->t_state == TCPS_LISTEN)
931 panic("tcp_input: TCPS_LISTEN");
932
933 /*
934 * Segment received on connection.
935 * Reset idle time and keep-alive timer.
936 */
937 tp->t_rcvtime = ticks;
938 if (TCPS_HAVEESTABLISHED(tp->t_state))
939 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
940
941 /*
942 * Process options.
943 * XXX this is tradtitional behavior, may need to be cleaned up.
944 */
945 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
946 if (thflags & TH_SYN) {
947 if (to.to_flags & TOF_SCALE) {
948 tp->t_flags |= TF_RCVD_SCALE;
949 tp->requested_s_scale = to.to_requested_s_scale;
950 }
951 if (to.to_flags & TOF_TS) {
952 tp->t_flags |= TF_RCVD_TSTMP;
953 tp->ts_recent = to.to_tsval;
954 tp->ts_recent_age = ticks;
955 }
956 if (to.to_flags & (TOF_CC|TOF_CCNEW))
957 tp->t_flags |= TF_RCVD_CC;
958 if (to.to_flags & TOF_MSS)
959 tcp_mss(tp, to.to_mss);
960 }
961
962 /*
963 * Header prediction: check for the two common cases
964 * of a uni-directional data xfer. If the packet has
965 * no control flags, is in-sequence, the window didn't
966 * change and we're not retransmitting, it's a
967 * candidate. If the length is zero and the ack moved
968 * forward, we're the sender side of the xfer. Just
969 * free the data acked & wake any higher level process
970 * that was blocked waiting for space. If the length
971 * is non-zero and the ack didn't move, we're the
972 * receiver side. If we're getting packets in-order
973 * (the reassembly queue is empty), add the data to
974 * the socket buffer and note that we need a delayed ack.
975 * Make sure that the hidden state-flags are also off.
976 * Since we check for TCPS_ESTABLISHED above, it can only
977 * be TH_NEEDSYN.
978 */
979 if (tp->t_state == TCPS_ESTABLISHED &&
980 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
981 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
982 ((to.to_flags & TOF_TS) == 0 ||
983 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
984 /*
985 * Using the CC option is compulsory if once started:
986 * the segment is OK if no T/TCP was negotiated or
987 * if the segment has a CC option equal to CCrecv
988 */
989 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
990 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
991 th->th_seq == tp->rcv_nxt &&
992 tiwin && tiwin == tp->snd_wnd &&
993 tp->snd_nxt == tp->snd_max) {
994
995 /*
996 * If last ACK falls within this segment's sequence numbers,
997 * record the timestamp.
998 * NOTE that the test is modified according to the latest
999 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1000 */
1001 if ((to.to_flags & TOF_TS) != 0 &&
1002 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1003 tp->ts_recent_age = ticks;
1004 tp->ts_recent = to.to_tsval;
1005 }
1006
1007 if (tlen == 0) {
1008 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1009 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1010 tp->snd_cwnd >= tp->snd_wnd &&
1011 ((!tcp_do_newreno &&
9845754e
MD
1012 tp->t_dupacks < tcprexmtthresh) ||
1013 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) {
984263bc
MD
1014 /*
1015 * this is a pure ack for outstanding data.
1016 */
1017 ++tcpstat.tcps_predack;
1018 /*
1019 * "bad retransmit" recovery
bfdb979e
JH
1020 *
1021 * If Eifel detection applies, then
1022 * it is deterministic, so use it
1023 * unconditionally over the old heuristic.
1024 * Otherwise, fall back to the old heuristic.
984263bc 1025 */
bfdb979e
JH
1026 if (tcp_do_eifel_detect &&
1027 (to.to_flags & TOF_TS) && to.to_tsecr &&
1028 (tp->t_flags & TF_FIRSTACCACK)) {
1029 /* Eifel detection applicable. */
1030 if (to.to_tsecr < tp->t_rexmtTS) {
1031 tcp_revert_congestion_state(tp);
1032 ++tcpstat.tcps_eifeldetected;
1033 }
1034 } else if (tp->t_rxtshift == 1 &&
1035 ticks < tp->t_badrxtwin) {
1036 tcp_revert_congestion_state(tp);
1037 ++tcpstat.tcps_rttdetected;
984263bc 1038 }
8819433a
JH
1039 tp->t_flags &= ~(TF_FIRSTACCACK |
1040 TF_FASTREXMT | TF_EARLYREXMT);
984263bc
MD
1041 /*
1042 * Recalculate the retransmit timer / rtt.
1043 *
1044 * Some machines (certain windows boxes)
1045 * send broken timestamp replies during the
1046 * SYN+ACK phase, ignore timestamps of 0.
1047 */
1048 if ((to.to_flags & TOF_TS) != 0 &&
1049 to.to_tsecr) {
1050 tcp_xmit_timer(tp,
1051 ticks - to.to_tsecr + 1);
1052 } else if (tp->t_rtttime &&
1053 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1054 tcp_xmit_timer(tp,
1055 ticks - tp->t_rtttime);
1056 }
1057 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1058 acked = th->th_ack - tp->snd_una;
1059 tcpstat.tcps_rcvackpack++;
1060 tcpstat.tcps_rcvackbyte += acked;
1061 sbdrop(&so->so_snd, acked);
9845754e
MD
1062 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1063 SEQ_LEQ(th->th_ack, tp->snd_recover))
1064 tp->snd_recover = th->th_ack - 1;
1065 tp->snd_una = th->th_ack;
984263bc
MD
1066 tp->t_dupacks = 0;
1067 m_freem(m);
1068 ND6_HINT(tp); /* some progress has been done */
1069
1070 /*
1071 * If all outstanding data are acked, stop
1072 * retransmit timer, otherwise restart timer
1073 * using current (possibly backed-off) value.
1074 * If process is waiting for space,
1075 * wakeup/selwakeup/signal. If data
1076 * are ready to send, let tcp_output
1077 * decide between more output or persist.
1078 */
1079 if (tp->snd_una == tp->snd_max)
1080 callout_stop(tp->tt_rexmt);
1081 else if (!callout_active(tp->tt_persist))
1082 callout_reset(tp->tt_rexmt,
1083 tp->t_rxtcur,
1084 tcp_timer_rexmt, tp);
1085
1086 sowwakeup(so);
1087 if (so->so_snd.sb_cc)
1088 (void) tcp_output(tp);
1089 return;
1090 }
1091 } else if (th->th_ack == tp->snd_una &&
1092 LIST_EMPTY(&tp->t_segq) &&
1093 tlen <= sbspace(&so->so_rcv)) {
1094 /*
1095 * this is a pure, in-sequence data packet
1096 * with nothing on the reassembly queue and
1097 * we have enough buffer space to take it.
1098 */
1099 ++tcpstat.tcps_preddat;
1100 tp->rcv_nxt += tlen;
1101 tcpstat.tcps_rcvpack++;
1102 tcpstat.tcps_rcvbyte += tlen;
1103 ND6_HINT(tp); /* some progress has been done */
1104 /*
1105 * Add data to socket buffer.
1106 */
1107 if (so->so_state & SS_CANTRCVMORE) {
1108 m_freem(m);
1109 } else {
1110 m_adj(m, drop_hdrlen); /* delayed header drop */
1111 sbappend(&so->so_rcv, m);
1112 }
1113 sorwakeup(so);
1114 if (DELAY_ACK(tp)) {
1115 callout_reset(tp->tt_delack, tcp_delacktime,
1116 tcp_timer_delack, tp);
1117 } else {
1118 tp->t_flags |= TF_ACKNOW;
1119 tcp_output(tp);
1120 }
1121 return;
1122 }
1123 }
1124
1125 /*
1126 * Calculate amount of space in receive window,
1127 * and then do TCP input processing.
1128 * Receive window is amount of space in rcv queue,
1129 * but not less than advertised window.
1130 */
1131 { int win;
1132
1133 win = sbspace(&so->so_rcv);
1134 if (win < 0)
1135 win = 0;
1136 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1137 }
1138
1139 switch (tp->t_state) {
1140
1141 /*
1142 * If the state is SYN_RECEIVED:
1143 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1144 */
1145 case TCPS_SYN_RECEIVED:
1146 if ((thflags & TH_ACK) &&
1147 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1148 SEQ_GT(th->th_ack, tp->snd_max))) {
1149 rstreason = BANDLIM_RST_OPENPORT;
1150 goto dropwithreset;
1151 }
1152 break;
1153
1154 /*
1155 * If the state is SYN_SENT:
1156 * if seg contains an ACK, but not for our SYN, drop the input.
1157 * if seg contains a RST, then drop the connection.
1158 * if seg does not contain SYN, then drop it.
1159 * Otherwise this is an acceptable SYN segment
1160 * initialize tp->rcv_nxt and tp->irs
1161 * if seg contains ack then advance tp->snd_una
1162 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1163 * arrange for segment to be acked (eventually)
1164 * continue processing rest of data/controls, beginning with URG
1165 */
1166 case TCPS_SYN_SENT:
1167 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1168 taop = &tao_noncached;
1169 bzero(taop, sizeof(*taop));
1170 }
1171
1172 if ((thflags & TH_ACK) &&
1173 (SEQ_LEQ(th->th_ack, tp->iss) ||
1174 SEQ_GT(th->th_ack, tp->snd_max))) {
1175 /*
1176 * If we have a cached CCsent for the remote host,
1177 * hence we haven't just crashed and restarted,
1178 * do not send a RST. This may be a retransmission
1179 * from the other side after our earlier ACK was lost.
1180 * Our new SYN, when it arrives, will serve as the
1181 * needed ACK.
1182 */
1183 if (taop->tao_ccsent != 0)
1184 goto drop;
1185 else {
1186 rstreason = BANDLIM_UNLIMITED;
1187 goto dropwithreset;
1188 }
1189 }
1190 if (thflags & TH_RST) {
1191 if (thflags & TH_ACK)
1192 tp = tcp_drop(tp, ECONNREFUSED);
1193 goto drop;
1194 }
1195 if ((thflags & TH_SYN) == 0)
1196 goto drop;
1197 tp->snd_wnd = th->th_win; /* initial send window */
1198 tp->cc_recv = to.to_cc; /* foreign CC */
1199
1200 tp->irs = th->th_seq;
1201 tcp_rcvseqinit(tp);
1202 if (thflags & TH_ACK) {
1203 /*
1204 * Our SYN was acked. If segment contains CC.ECHO
1205 * option, check it to make sure this segment really
1206 * matches our SYN. If not, just drop it as old
1207 * duplicate, but send an RST if we're still playing
1208 * by the old rules. If no CC.ECHO option, make sure
1209 * we don't get fooled into using T/TCP.
1210 */
1211 if (to.to_flags & TOF_CCECHO) {
1212 if (tp->cc_send != to.to_ccecho) {
1213 if (taop->tao_ccsent != 0)
1214 goto drop;
1215 else {
1216 rstreason = BANDLIM_UNLIMITED;
1217 goto dropwithreset;
1218 }
1219 }
1220 } else
1221 tp->t_flags &= ~TF_RCVD_CC;
1222 tcpstat.tcps_connects++;
1223 soisconnected(so);
1224 /* Do window scaling on this connection? */
1225 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1226 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1227 tp->snd_scale = tp->requested_s_scale;
1228 tp->rcv_scale = tp->request_r_scale;
1229 }
1230 /* Segment is acceptable, update cache if undefined. */
1231 if (taop->tao_ccsent == 0)
1232 taop->tao_ccsent = to.to_ccecho;
1233
1234 tp->rcv_adv += tp->rcv_wnd;
1235 tp->snd_una++; /* SYN is acked */
1236 /*
1237 * If there's data, delay ACK; if there's also a FIN
1238 * ACKNOW will be turned on later.
1239 */
1240 if (DELAY_ACK(tp) && tlen != 0)
1241 callout_reset(tp->tt_delack, tcp_delacktime,
1242 tcp_timer_delack, tp);
1243 else
1244 tp->t_flags |= TF_ACKNOW;
1245 /*
1246 * Received <SYN,ACK> in SYN_SENT[*] state.
1247 * Transitions:
1248 * SYN_SENT --> ESTABLISHED
1249 * SYN_SENT* --> FIN_WAIT_1
1250 */
1251 tp->t_starttime = ticks;
1252 if (tp->t_flags & TF_NEEDFIN) {
1253 tp->t_state = TCPS_FIN_WAIT_1;
1254 tp->t_flags &= ~TF_NEEDFIN;
1255 thflags &= ~TH_SYN;
1256 } else {
1257 tp->t_state = TCPS_ESTABLISHED;
1258 callout_reset(tp->tt_keep, tcp_keepidle,
1259 tcp_timer_keep, tp);
1260 }
1261 } else {
1262 /*
1263 * Received initial SYN in SYN-SENT[*] state =>
1264 * simultaneous open. If segment contains CC option
1265 * and there is a cached CC, apply TAO test.
1266 * If it succeeds, connection is * half-synchronized.
1267 * Otherwise, do 3-way handshake:
1268 * SYN-SENT -> SYN-RECEIVED
1269 * SYN-SENT* -> SYN-RECEIVED*
1270 * If there was no CC option, clear cached CC value.
1271 */
1272 tp->t_flags |= TF_ACKNOW;
1273 callout_stop(tp->tt_rexmt);
1274 if (to.to_flags & TOF_CC) {
1275 if (taop->tao_cc != 0 &&
1276 CC_GT(to.to_cc, taop->tao_cc)) {
1277 /*
1278 * update cache and make transition:
1279 * SYN-SENT -> ESTABLISHED*
1280 * SYN-SENT* -> FIN-WAIT-1*
1281 */
1282 taop->tao_cc = to.to_cc;
1283 tp->t_starttime = ticks;
1284 if (tp->t_flags & TF_NEEDFIN) {
1285 tp->t_state = TCPS_FIN_WAIT_1;
1286 tp->t_flags &= ~TF_NEEDFIN;
1287 } else {
1288 tp->t_state = TCPS_ESTABLISHED;
1289 callout_reset(tp->tt_keep,
1290 tcp_keepidle,
1291 tcp_timer_keep,
1292 tp);
1293 }
1294 tp->t_flags |= TF_NEEDSYN;
1295 } else
1296 tp->t_state = TCPS_SYN_RECEIVED;
1297 } else {
1298 /* CC.NEW or no option => invalidate cache */
1299 taop->tao_cc = 0;
1300 tp->t_state = TCPS_SYN_RECEIVED;
1301 }
1302 }
1303
1304trimthenstep6:
1305 /*
1306 * Advance th->th_seq to correspond to first data byte.
1307 * If data, trim to stay within window,
1308 * dropping FIN if necessary.
1309 */
1310 th->th_seq++;
1311 if (tlen > tp->rcv_wnd) {
1312 todrop = tlen - tp->rcv_wnd;
1313 m_adj(m, -todrop);
1314 tlen = tp->rcv_wnd;
1315 thflags &= ~TH_FIN;
1316 tcpstat.tcps_rcvpackafterwin++;
1317 tcpstat.tcps_rcvbyteafterwin += todrop;
1318 }
1319 tp->snd_wl1 = th->th_seq - 1;
1320 tp->rcv_up = th->th_seq;
1321 /*
1322 * Client side of transaction: already sent SYN and data.
1323 * If the remote host used T/TCP to validate the SYN,
1324 * our data will be ACK'd; if so, enter normal data segment
1325 * processing in the middle of step 5, ack processing.
1326 * Otherwise, goto step 6.
1327 */
1328 if (thflags & TH_ACK)
1329 goto process_ACK;
1330
1331 goto step6;
1332
1333 /*
1334 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1335 * if segment contains a SYN and CC [not CC.NEW] option:
1336 * if state == TIME_WAIT and connection duration > MSL,
1337 * drop packet and send RST;
1338 *
1339 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1340 * ack the FIN (and data) in retransmission queue.
1341 * Complete close and delete TCPCB. Then reprocess
1342 * segment, hoping to find new TCPCB in LISTEN state;
1343 *
1344 * else must be old SYN; drop it.
1345 * else do normal processing.
1346 */
1347 case TCPS_LAST_ACK:
1348 case TCPS_CLOSING:
1349 case TCPS_TIME_WAIT:
1350 if ((thflags & TH_SYN) &&
1351 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1352 if (tp->t_state == TCPS_TIME_WAIT &&
1353 (ticks - tp->t_starttime) > tcp_msl) {
1354 rstreason = BANDLIM_UNLIMITED;
1355 goto dropwithreset;
1356 }
1357 if (CC_GT(to.to_cc, tp->cc_recv)) {
1358 tp = tcp_close(tp);
1359 goto findpcb;
1360 }
1361 else
1362 goto drop;
1363 }
1364 break; /* continue normal processing */
1365 }
1366
1367 /*
1368 * States other than LISTEN or SYN_SENT.
1369 * First check the RST flag and sequence number since reset segments
1370 * are exempt from the timestamp and connection count tests. This
1371 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1372 * below which allowed reset segments in half the sequence space
1373 * to fall though and be processed (which gives forged reset
1374 * segments with a random sequence number a 50 percent chance of
1375 * killing a connection).
1376 * Then check timestamp, if present.
1377 * Then check the connection count, if present.
1378 * Then check that at least some bytes of segment are within
1379 * receive window. If segment begins before rcv_nxt,
1380 * drop leading data (and SYN); if nothing left, just ack.
1381 *
1382 *
1383 * If the RST bit is set, check the sequence number to see
1384 * if this is a valid reset segment.
1385 * RFC 793 page 37:
1386 * In all states except SYN-SENT, all reset (RST) segments
1387 * are validated by checking their SEQ-fields. A reset is
1388 * valid if its sequence number is in the window.
1389 * Note: this does not take into account delayed ACKs, so
1390 * we should test against last_ack_sent instead of rcv_nxt.
1391 * The sequence number in the reset segment is normally an
1392 * echo of our outgoing acknowlegement numbers, but some hosts
1393 * send a reset with the sequence number at the rightmost edge
1394 * of our receive window, and we have to handle this case.
1395 * If we have multiple segments in flight, the intial reset
1396 * segment sequence numbers will be to the left of last_ack_sent,
1397 * but they will eventually catch up.
1398 * In any case, it never made sense to trim reset segments to
1399 * fit the receive window since RFC 1122 says:
1400 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1401 *
1402 * A TCP SHOULD allow a received RST segment to include data.
1403 *
1404 * DISCUSSION
1405 * It has been suggested that a RST segment could contain
1406 * ASCII text that encoded and explained the cause of the
1407 * RST. No standard has yet been established for such
1408 * data.
1409 *
1410 * If the reset segment passes the sequence number test examine
1411 * the state:
1412 * SYN_RECEIVED STATE:
1413 * If passive open, return to LISTEN state.
1414 * If active open, inform user that connection was refused.
1415 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1416 * Inform user that connection was reset, and close tcb.
1417 * CLOSING, LAST_ACK STATES:
1418 * Close the tcb.
1419 * TIME_WAIT STATE:
1420 * Drop the segment - see Stevens, vol. 2, p. 964 and
1421 * RFC 1337.
1422 */
1423 if (thflags & TH_RST) {
1424 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1425 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1426 switch (tp->t_state) {
1427
1428 case TCPS_SYN_RECEIVED:
1429 so->so_error = ECONNREFUSED;
1430 goto close;
1431
1432 case TCPS_ESTABLISHED:
1433 case TCPS_FIN_WAIT_1:
1434 case TCPS_FIN_WAIT_2:
1435 case TCPS_CLOSE_WAIT:
1436 so->so_error = ECONNRESET;
1437 close:
1438 tp->t_state = TCPS_CLOSED;
1439 tcpstat.tcps_drops++;
1440 tp = tcp_close(tp);
1441 break;
1442
1443 case TCPS_CLOSING:
1444 case TCPS_LAST_ACK:
1445 tp = tcp_close(tp);
1446 break;
1447
1448 case TCPS_TIME_WAIT:
1449 break;
1450 }
1451 }
1452 goto drop;
1453 }
1454
1455 /*
1456 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1457 * and it's less than ts_recent, drop it.
1458 */
1459 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1460 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1461
1462 /* Check to see if ts_recent is over 24 days old. */
1463 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1464 /*
1465 * Invalidate ts_recent. If this segment updates
1466 * ts_recent, the age will be reset later and ts_recent
1467 * will get a valid value. If it does not, setting
1468 * ts_recent to zero will at least satisfy the
1469 * requirement that zero be placed in the timestamp
1470 * echo reply when ts_recent isn't valid. The
1471 * age isn't reset until we get a valid ts_recent
1472 * because we don't want out-of-order segments to be
1473 * dropped when ts_recent is old.
1474 */
1475 tp->ts_recent = 0;
1476 } else {
1477 tcpstat.tcps_rcvduppack++;
1478 tcpstat.tcps_rcvdupbyte += tlen;
1479 tcpstat.tcps_pawsdrop++;
1480 if (tlen)
1481 goto dropafterack;
1482 goto drop;
1483 }
1484 }
1485
1486 /*
1487 * T/TCP mechanism
1488 * If T/TCP was negotiated and the segment doesn't have CC,
1489 * or if its CC is wrong then drop the segment.
1490 * RST segments do not have to comply with this.
1491 */
1492 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1493 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1494 goto dropafterack;
1495
1496 /*
1497 * In the SYN-RECEIVED state, validate that the packet belongs to
1498 * this connection before trimming the data to fit the receive
1499 * window. Check the sequence number versus IRS since we know
1500 * the sequence numbers haven't wrapped. This is a partial fix
1501 * for the "LAND" DoS attack.
1502 */
1503 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1504 rstreason = BANDLIM_RST_OPENPORT;
1505 goto dropwithreset;
1506 }
1507
1508 todrop = tp->rcv_nxt - th->th_seq;
1509 if (todrop > 0) {
1510 if (thflags & TH_SYN) {
1511 thflags &= ~TH_SYN;
1512 th->th_seq++;
1513 if (th->th_urp > 1)
1514 th->th_urp--;
1515 else
1516 thflags &= ~TH_URG;
1517 todrop--;
1518 }
1519 /*
1520 * Following if statement from Stevens, vol. 2, p. 960.
1521 */
1522 if (todrop > tlen
1523 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1524 /*
1525 * Any valid FIN must be to the left of the window.
1526 * At this point the FIN must be a duplicate or out
1527 * of sequence; drop it.
1528 */
1529 thflags &= ~TH_FIN;
1530
1531 /*
1532 * Send an ACK to resynchronize and drop any data.
1533 * But keep on processing for RST or ACK.
1534 */
1535 tp->t_flags |= TF_ACKNOW;
1536 todrop = tlen;
1537 tcpstat.tcps_rcvduppack++;
1538 tcpstat.tcps_rcvdupbyte += todrop;
1539 } else {
1540 tcpstat.tcps_rcvpartduppack++;
1541 tcpstat.tcps_rcvpartdupbyte += todrop;
1542 }
1543 drop_hdrlen += todrop; /* drop from the top afterwards */
1544 th->th_seq += todrop;
1545 tlen -= todrop;
1546 if (th->th_urp > todrop)
1547 th->th_urp -= todrop;
1548 else {
1549 thflags &= ~TH_URG;
1550 th->th_urp = 0;
1551 }
1552 }
1553
1554 /*
1555 * If new data are received on a connection after the
1556 * user processes are gone, then RST the other end.
1557 */
1558 if ((so->so_state & SS_NOFDREF) &&
1559 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1560 tp = tcp_close(tp);
1561 tcpstat.tcps_rcvafterclose++;
1562 rstreason = BANDLIM_UNLIMITED;
1563 goto dropwithreset;
1564 }
1565
1566 /*
1567 * If segment ends after window, drop trailing data
1568 * (and PUSH and FIN); if nothing left, just ACK.
1569 */
1570 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1571 if (todrop > 0) {
1572 tcpstat.tcps_rcvpackafterwin++;
1573 if (todrop >= tlen) {
1574 tcpstat.tcps_rcvbyteafterwin += tlen;
1575 /*
1576 * If a new connection request is received
1577 * while in TIME_WAIT, drop the old connection
1578 * and start over if the sequence numbers
1579 * are above the previous ones.
1580 */
1581 if (thflags & TH_SYN &&
1582 tp->t_state == TCPS_TIME_WAIT &&
1583 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1584 tp = tcp_close(tp);
1585 goto findpcb;
1586 }
1587 /*
1588 * If window is closed can only take segments at
1589 * window edge, and have to drop data and PUSH from
1590 * incoming segments. Continue processing, but
1591 * remember to ack. Otherwise, drop segment
1592 * and ack.
1593 */
1594 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1595 tp->t_flags |= TF_ACKNOW;
1596 tcpstat.tcps_rcvwinprobe++;
1597 } else
1598 goto dropafterack;
1599 } else
1600 tcpstat.tcps_rcvbyteafterwin += todrop;
1601 m_adj(m, -todrop);
1602 tlen -= todrop;
1603 thflags &= ~(TH_PUSH|TH_FIN);
1604 }
1605
1606 /*
1607 * If last ACK falls within this segment's sequence numbers,
1608 * record its timestamp.
1609 * NOTE that the test is modified according to the latest
1610 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1611 */
1612 if ((to.to_flags & TOF_TS) != 0 &&
1613 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1614 tp->ts_recent_age = ticks;
1615 tp->ts_recent = to.to_tsval;
1616 }
1617
1618 /*
1619 * If a SYN is in the window, then this is an
1620 * error and we send an RST and drop the connection.
1621 */
1622 if (thflags & TH_SYN) {
1623 tp = tcp_drop(tp, ECONNRESET);
1624 rstreason = BANDLIM_UNLIMITED;
1625 goto dropwithreset;
1626 }
1627
1628 /*
1629 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1630 * flag is on (half-synchronized state), then queue data for
1631 * later processing; else drop segment and return.
1632 */
1633 if ((thflags & TH_ACK) == 0) {
1634 if (tp->t_state == TCPS_SYN_RECEIVED ||
1635 (tp->t_flags & TF_NEEDSYN))
1636 goto step6;
1637 else
1638 goto drop;
1639 }
1640
1641 /*
1642 * Ack processing.
1643 */
1644 switch (tp->t_state) {
1645
1646 /*
1647 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1648 * ESTABLISHED state and continue processing.
1649 * The ACK was checked above.
1650 */
1651 case TCPS_SYN_RECEIVED:
1652
1653 tcpstat.tcps_connects++;
1654 soisconnected(so);
1655 /* Do window scaling? */
1656 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1657 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1658 tp->snd_scale = tp->requested_s_scale;
1659 tp->rcv_scale = tp->request_r_scale;
1660 }
1661 /*
1662 * Upon successful completion of 3-way handshake,
1663 * update cache.CC if it was undefined, pass any queued
1664 * data to the user, and advance state appropriately.
1665 */
1666 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1667 taop->tao_cc == 0)
1668 taop->tao_cc = tp->cc_recv;
1669
1670 /*
1671 * Make transitions:
1672 * SYN-RECEIVED -> ESTABLISHED
1673 * SYN-RECEIVED* -> FIN-WAIT-1
1674 */
1675 tp->t_starttime = ticks;
1676 if (tp->t_flags & TF_NEEDFIN) {
1677 tp->t_state = TCPS_FIN_WAIT_1;
1678 tp->t_flags &= ~TF_NEEDFIN;
1679 } else {
1680 tp->t_state = TCPS_ESTABLISHED;
1681 callout_reset(tp->tt_keep, tcp_keepidle,
1682 tcp_timer_keep, tp);
1683 }
1684 /*
1685 * If segment contains data or ACK, will call tcp_reass()
1686 * later; if not, do so now to pass queued data to user.
1687 */
1688 if (tlen == 0 && (thflags & TH_FIN) == 0)
1689 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1690 (struct mbuf *)0);
1691 tp->snd_wl1 = th->th_seq - 1;
1692 /* fall into ... */
1693
1694 /*
1695 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1696 * ACKs. If the ack is in the range
1697 * tp->snd_una < th->th_ack <= tp->snd_max
1698 * then advance tp->snd_una to th->th_ack and drop
1699 * data from the retransmission queue. If this ACK reflects
1700 * more up to date window information we update our window information.
1701 */
1702 case TCPS_ESTABLISHED:
1703 case TCPS_FIN_WAIT_1:
1704 case TCPS_FIN_WAIT_2:
1705 case TCPS_CLOSE_WAIT:
1706 case TCPS_CLOSING:
1707 case TCPS_LAST_ACK:
1708 case TCPS_TIME_WAIT:
1709
1710 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1711 if (tlen == 0 && tiwin == tp->snd_wnd) {
1712 tcpstat.tcps_rcvdupack++;
1713 /*
1714 * If we have outstanding data (other than
1715 * a window probe), this is a completely
1716 * duplicate ack (ie, window info didn't
1717 * change), the ack is the biggest we've
1718 * seen and we've seen exactly our rexmt
1719 * threshhold of them, assume a packet
1720 * has been dropped and retransmit it.
1721 * Kludge snd_nxt & the congestion
1722 * window so we send only this one
1723 * packet.
1724 *
1725 * We know we're losing at the current
1726 * window size so do congestion avoidance
1727 * (set ssthresh to half the current window
1728 * and pull our congestion window back to
1729 * the new ssthresh).
1730 *
1731 * Dup acks mean that packets have left the
1732 * network (they're now cached at the receiver)
1733 * so bump cwnd by the amount in the receiver
1734 * to keep a constant cwnd packets in the
1735 * network.
1736 */
1737 if (!callout_active(tp->tt_rexmt) ||
1738 th->th_ack != tp->snd_una)
1739 tp->t_dupacks = 0;
1740 else if (++tp->t_dupacks > tcprexmtthresh ||
1741 (tcp_do_newreno &&
9845754e 1742 IN_FASTRECOVERY(tp))) {
984263bc
MD
1743 tp->snd_cwnd += tp->t_maxseg;
1744 (void) tcp_output(tp);
1745 goto drop;
1746 } else if (tp->t_dupacks == tcprexmtthresh) {
8819433a 1747 tcp_seq onxt;
984263bc 1748 u_int win;
8819433a 1749
984263bc 1750 if (tcp_do_newreno &&
9845754e
MD
1751 SEQ_LEQ(th->th_ack,
1752 tp->snd_recover)) {
984263bc
MD
1753 tp->t_dupacks = 0;
1754 break;
1755 }
8819433a 1756fastretransmit:
efd4b327
JH
1757 if (tcp_do_eifel_detect &&
1758 (tp->t_flags & TF_RCVD_TSTMP)) {
1759 tcp_save_congestion_state(tp);
1760 tp->t_flags |= TF_FASTREXMT;
1761 }
984263bc
MD
1762 win = min(tp->snd_wnd, tp->snd_cwnd) /
1763 2 / tp->t_maxseg;
1764 if (win < 2)
1765 win = 2;
1766 tp->snd_ssthresh = win * tp->t_maxseg;
9845754e 1767 ENTER_FASTRECOVERY(tp);
984263bc
MD
1768 tp->snd_recover = tp->snd_max;
1769 callout_stop(tp->tt_rexmt);
1770 tp->t_rtttime = 0;
8819433a 1771 onxt = tp->snd_nxt;
984263bc
MD
1772 tp->snd_nxt = th->th_ack;
1773 tp->snd_cwnd = tp->t_maxseg;
1774 (void) tcp_output(tp);
8819433a 1775 ++tcpstat.tcps_sndfastrexmit;
33abdd1c
MD
1776 KASSERT(tp->snd_limited <= 2,
1777 ("tp->snd_limited too big"));
984263bc 1778 tp->snd_cwnd = tp->snd_ssthresh +
33abdd1c
MD
1779 (tp->t_maxseg *
1780 (tp->t_dupacks - tp->snd_limited));
984263bc
MD
1781 if (SEQ_GT(onxt, tp->snd_nxt))
1782 tp->snd_nxt = onxt;
1783 goto drop;
33abdd1c
MD
1784 } else if (tcp_do_limitedtransmit) {
1785 u_long oldcwnd = tp->snd_cwnd;
1786 tcp_seq oldsndmax = tp->snd_max;
8819433a
JH
1787 /* outstanding data */
1788 uint32_t ownd =
1789 tp->snd_max - tp->snd_una;
33abdd1c 1790 u_int sent;
ca94b4a6 1791
8819433a
JH
1792#define iceildiv(n, d) (((n)+(d)-1) / (d))
1793
33abdd1c
MD
1794 KASSERT(tp->t_dupacks == 1 ||
1795 tp->t_dupacks == 2,
1796 ("dupacks not 1 or 2"));
52c1865b 1797 if (tp->t_dupacks == 1)
33abdd1c 1798 tp->snd_limited = 0;
8819433a 1799 tp->snd_cwnd = ownd +
52c1865b
JH
1800 (tp->t_dupacks - tp->snd_limited) *
1801 tp->t_maxseg;
33abdd1c 1802 (void) tcp_output(tp);
8819433a 1803 tp->snd_cwnd = oldcwnd;
33abdd1c
MD
1804 sent = tp->snd_max - oldsndmax;
1805 if (sent > tp->t_maxseg) {
ca94b4a6
JH
1806 KASSERT((tp->t_dupacks == 2 &&
1807 tp->snd_limited == 0) ||
1808 (sent == tp->t_maxseg + 1 &&
1809 tp->t_flags & TF_SENTFIN),
33abdd1c 1810 ("sent too much"));
8819433a
JH
1811 KASSERT(sent <=
1812 tp->t_maxseg * 2,
1813 ("sent too many segments"));
33abdd1c 1814 tp->snd_limited = 2;
8819433a
JH
1815 tcpstat.tcps_sndlimited += 2;
1816 } else if (sent > 0) {
33abdd1c 1817 ++tp->snd_limited;
8819433a
JH
1818 ++tcpstat.tcps_sndlimited;
1819 } else if (tcp_do_early_retransmit &&
1820 (tcp_do_eifel_detect &&
1821 (tp->t_flags & TF_RCVD_TSTMP)) &&
1822 tcp_do_newreno &&
1823 tp->t_dupacks + 1 >=
1824 iceildiv(ownd, tp->t_maxseg)) {
1825 ++tcpstat.tcps_sndearlyrexmit;
1826 tp->t_flags |= TF_EARLYREXMT;
1827 goto fastretransmit;
1828 }
33abdd1c 1829 goto drop;
984263bc
MD
1830 }
1831 } else
1832 tp->t_dupacks = 0;
1833 break;
1834 }
1835
1836 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1837
1838 /*
1839 * If the congestion window was inflated to account
1840 * for the other side's cached packets, retract it.
1841 */
1842 if (tcp_do_newreno) {
9845754e 1843 if (IN_FASTRECOVERY(tp)) {
984263bc
MD
1844 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1845 tcp_newreno_partial_ack(tp, th);
1846 } else {
1847 /*
1848 * Window inflation should have left us
1849 * with approximately snd_ssthresh
1850 * outstanding data.
1851 * But in case we would be inclined to
1852 * send a burst, better to do it via
1853 * the slow start mechanism.
1854 */
1855 if (SEQ_GT(th->th_ack +
1856 tp->snd_ssthresh,
1857 tp->snd_max))
1858 tp->snd_cwnd = tp->snd_max -
1859 th->th_ack +
1860 tp->t_maxseg;
1861 else
1862 tp->snd_cwnd = tp->snd_ssthresh;
1863 }
1864 }
1865 } else {
1866 if (tp->t_dupacks >= tcprexmtthresh &&
1867 tp->snd_cwnd > tp->snd_ssthresh)
1868 tp->snd_cwnd = tp->snd_ssthresh;
1869 }
1870 tp->t_dupacks = 0;
1871 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1872 tcpstat.tcps_rcvacktoomuch++;
1873 goto dropafterack;
1874 }
1875 /*
1876 * If we reach this point, ACK is not a duplicate,
1877 * i.e., it ACKs something we sent.
1878 */
1879 if (tp->t_flags & TF_NEEDSYN) {
1880 /*
1881 * T/TCP: Connection was half-synchronized, and our
1882 * SYN has been ACK'd (so connection is now fully
1883 * synchronized). Go to non-starred state,
1884 * increment snd_una for ACK of SYN, and check if
1885 * we can do window scaling.
1886 */
1887 tp->t_flags &= ~TF_NEEDSYN;
1888 tp->snd_una++;
1889 /* Do window scaling? */
1890 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1891 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1892 tp->snd_scale = tp->requested_s_scale;
1893 tp->rcv_scale = tp->request_r_scale;
1894 }
1895 }
1896
1897process_ACK:
1898 acked = th->th_ack - tp->snd_una;
1899 tcpstat.tcps_rcvackpack++;
1900 tcpstat.tcps_rcvackbyte += acked;
1901
1902 /*
1903 * If we just performed our first retransmit, and the ACK
1904 * arrives within our recovery window, then it was a mistake
1905 * to do the retransmit in the first place. Recover our
1906 * original cwnd and ssthresh, and proceed to transmit where
1907 * we left off.
1908 */
bfdb979e
JH
1909 if (tcp_do_eifel_detect && acked &&
1910 (to.to_flags & TOF_TS) && to.to_tsecr &&
1911 (tp->t_flags & TF_FIRSTACCACK)) {
1912 /* Eifel detection applicable. */
1913 if (to.to_tsecr < tp->t_rexmtTS) {
bfdb979e 1914 ++tcpstat.tcps_eifeldetected;
8819433a
JH
1915 tcp_revert_congestion_state(tp);
1916 if (tp->t_rxtshift == 1 &&
1917 ticks >= tp->t_badrxtwin)
1918 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
1919 }
1920 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1921 tcp_revert_congestion_state(tp);
1922 ++tcpstat.tcps_rttdetected;
984263bc
MD
1923 }
1924
1925 /*
1926 * If we have a timestamp reply, update smoothed
1927 * round trip time. If no timestamp is present but
1928 * transmit timer is running and timed sequence
1929 * number was acked, update smoothed round trip time.
1930 * Since we now have an rtt measurement, cancel the
1931 * timer backoff (cf., Phil Karn's retransmit alg.).
1932 * Recompute the initial retransmit timer.
1933 *
1934 * Some machines (certain windows boxes) send broken
1935 * timestamp replies during the SYN+ACK phase, ignore
1936 * timestamps of 0.
1937 */
1938 if ((to.to_flags & TOF_TS) != 0 &&
1939 to.to_tsecr) {
1940 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1941 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1942 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1943 }
1944 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1945
1946 /*
1947 * If all outstanding data is acked, stop retransmit
1948 * timer and remember to restart (more output or persist).
1949 * If there is more data to be acked, restart retransmit
1950 * timer, using current (possibly backed-off) value.
1951 */
1952 if (th->th_ack == tp->snd_max) {
1953 callout_stop(tp->tt_rexmt);
1954 needoutput = 1;
1955 } else if (!callout_active(tp->tt_persist))
1956 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1957 tcp_timer_rexmt, tp);
1958
1959 /*
1960 * If no data (only SYN) was ACK'd,
1961 * skip rest of ACK processing.
1962 */
1963 if (acked == 0)
1964 goto step6;
1965
efd4b327 1966 /* Stop looking for an acceptable ACK since one was received. */
8819433a 1967 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
efd4b327 1968
984263bc
MD
1969 /*
1970 * When new data is acked, open the congestion window.
1971 * If the window gives us less than ssthresh packets
1972 * in flight, open exponentially (maxseg per packet).
1973 * Otherwise open linearly: maxseg per window
1974 * (maxseg^2 / cwnd per packet).
1975 */
9845754e 1976 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) {
2256ba69
RG
1977 u_int cw = tp->snd_cwnd;
1978 u_int incr = tp->t_maxseg;
984263bc
MD
1979 if (cw > tp->snd_ssthresh)
1980 incr = incr * incr / cw;
1981 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
1982 }
1983 if (acked > so->so_snd.sb_cc) {
1984 tp->snd_wnd -= so->so_snd.sb_cc;
1985 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1986 ourfinisacked = 1;
1987 } else {
1988 sbdrop(&so->so_snd, acked);
1989 tp->snd_wnd -= acked;
1990 ourfinisacked = 0;
1991 }
1992 sowwakeup(so);
1993 /* detect una wraparound */
9845754e
MD
1994 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) &&
1995 SEQ_GT(tp->snd_una, tp->snd_recover) &&
1996 SEQ_LEQ(th->th_ack, tp->snd_recover))
1997 tp->snd_recover = th->th_ack - 1;
1998 if (tcp_do_newreno && IN_FASTRECOVERY(tp) &&
1999 SEQ_GEQ(th->th_ack, tp->snd_recover))
2000 EXIT_FASTRECOVERY(tp);
984263bc
MD
2001 tp->snd_una = th->th_ack;
2002 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2003 tp->snd_nxt = tp->snd_una;
2004
2005 switch (tp->t_state) {
2006
2007 /*
2008 * In FIN_WAIT_1 STATE in addition to the processing
2009 * for the ESTABLISHED state if our FIN is now acknowledged
2010 * then enter FIN_WAIT_2.
2011 */
2012 case TCPS_FIN_WAIT_1:
2013 if (ourfinisacked) {
2014 /*
2015 * If we can't receive any more
2016 * data, then closing user can proceed.
2017 * Starting the timer is contrary to the
2018 * specification, but if we don't get a FIN
2019 * we'll hang forever.
2020 */
2021 if (so->so_state & SS_CANTRCVMORE) {
2022 soisdisconnected(so);
2023 callout_reset(tp->tt_2msl, tcp_maxidle,
2024 tcp_timer_2msl, tp);
2025 }
2026 tp->t_state = TCPS_FIN_WAIT_2;
2027 }
2028 break;
2029
2030 /*
2031 * In CLOSING STATE in addition to the processing for
2032 * the ESTABLISHED state if the ACK acknowledges our FIN
2033 * then enter the TIME-WAIT state, otherwise ignore
2034 * the segment.
2035 */
2036 case TCPS_CLOSING:
2037 if (ourfinisacked) {
2038 tp->t_state = TCPS_TIME_WAIT;
2039 tcp_canceltimers(tp);
2040 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2041 if (tp->cc_recv != 0 &&
2042 (ticks - tp->t_starttime) < tcp_msl)
2043 callout_reset(tp->tt_2msl,
2044 tp->t_rxtcur *
2045 TCPTV_TWTRUNC,
2046 tcp_timer_2msl, tp);
2047 else
2048 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2049 tcp_timer_2msl, tp);
2050 soisdisconnected(so);
2051 }
2052 break;
2053
2054 /*
2055 * In LAST_ACK, we may still be waiting for data to drain
2056 * and/or to be acked, as well as for the ack of our FIN.
2057 * If our FIN is now acknowledged, delete the TCB,
2058 * enter the closed state and return.
2059 */
2060 case TCPS_LAST_ACK:
2061 if (ourfinisacked) {
2062 tp = tcp_close(tp);
2063 goto drop;
2064 }
2065 break;
2066
2067 /*
2068 * In TIME_WAIT state the only thing that should arrive
2069 * is a retransmission of the remote FIN. Acknowledge
2070 * it and restart the finack timer.
2071 */
2072 case TCPS_TIME_WAIT:
2073 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2074 tcp_timer_2msl, tp);
2075 goto dropafterack;
2076 }
2077 }
2078
2079step6:
2080 /*
2081 * Update window information.
2082 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2083 */
2084 if ((thflags & TH_ACK) &&
2085 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2086 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2087 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2088 /* keep track of pure window updates */
2089 if (tlen == 0 &&
2090 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2091 tcpstat.tcps_rcvwinupd++;
2092 tp->snd_wnd = tiwin;
2093 tp->snd_wl1 = th->th_seq;
2094 tp->snd_wl2 = th->th_ack;
2095 if (tp->snd_wnd > tp->max_sndwnd)
2096 tp->max_sndwnd = tp->snd_wnd;
2097 needoutput = 1;
2098 }
2099
2100 /*
2101 * Process segments with URG.
2102 */
2103 if ((thflags & TH_URG) && th->th_urp &&
2104 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2105 /*
2106 * This is a kludge, but if we receive and accept
2107 * random urgent pointers, we'll crash in
2108 * soreceive. It's hard to imagine someone
2109 * actually wanting to send this much urgent data.
2110 */
2111 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2112 th->th_urp = 0; /* XXX */
2113 thflags &= ~TH_URG; /* XXX */
2114 goto dodata; /* XXX */
2115 }
2116 /*
2117 * If this segment advances the known urgent pointer,
2118 * then mark the data stream. This should not happen
2119 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2120 * a FIN has been received from the remote side.
2121 * In these states we ignore the URG.
2122 *
2123 * According to RFC961 (Assigned Protocols),
2124 * the urgent pointer points to the last octet
2125 * of urgent data. We continue, however,
2126 * to consider it to indicate the first octet
2127 * of data past the urgent section as the original
2128 * spec states (in one of two places).
2129 */
2130 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2131 tp->rcv_up = th->th_seq + th->th_urp;
2132 so->so_oobmark = so->so_rcv.sb_cc +
2133 (tp->rcv_up - tp->rcv_nxt) - 1;
2134 if (so->so_oobmark == 0)
2135 so->so_state |= SS_RCVATMARK;
2136 sohasoutofband(so);
2137 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2138 }
2139 /*
2140 * Remove out of band data so doesn't get presented to user.
2141 * This can happen independent of advancing the URG pointer,
2142 * but if two URG's are pending at once, some out-of-band
2143 * data may creep in... ick.
2144 */
2145 if (th->th_urp <= (u_long)tlen
2146#ifdef SO_OOBINLINE
2147 && (so->so_options & SO_OOBINLINE) == 0
2148#endif
2149 )
2150 tcp_pulloutofband(so, th, m,
2151 drop_hdrlen); /* hdr drop is delayed */
2152 } else {
2153 /*
2154 * If no out of band data is expected,
2155 * pull receive urgent pointer along
2156 * with the receive window.
2157 */
2158 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2159 tp->rcv_up = tp->rcv_nxt;
2160 }
2161dodata: /* XXX */
2162
2163 /*
2164 * Process the segment text, merging it into the TCP sequencing queue,
2165 * and arranging for acknowledgment of receipt if necessary.
2166 * This process logically involves adjusting tp->rcv_wnd as data
2167 * is presented to the user (this happens in tcp_usrreq.c,
2168 * case PRU_RCVD). If a FIN has already been received on this
2169 * connection then we just ignore the text.
2170 */
2171 if ((tlen || (thflags & TH_FIN)) &&
2172 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2173 m_adj(m, drop_hdrlen); /* delayed header drop */
2174 /*
2175 * Insert segment which includes th into TCP reassembly queue
2176 * with control block tp. Set thflags to whether reassembly now
2177 * includes a segment with FIN. This handles the common case
2178 * inline (segment is the next to be received on an established
2179 * connection, and the queue is empty), avoiding linkage into
2180 * and removal from the queue and repetition of various
2181 * conversions.
2182 * Set DELACK for segments received in order, but ack
2183 * immediately when segments are out of order (so
2184 * fast retransmit can work).
2185 */
2186 if (th->th_seq == tp->rcv_nxt &&
2187 LIST_EMPTY(&tp->t_segq) &&
2188 TCPS_HAVEESTABLISHED(tp->t_state)) {
2189 if (DELAY_ACK(tp))
2190 callout_reset(tp->tt_delack, tcp_delacktime,
2191 tcp_timer_delack, tp);
2192 else
2193 tp->t_flags |= TF_ACKNOW;
2194 tp->rcv_nxt += tlen;
2195 thflags = th->th_flags & TH_FIN;
2196 tcpstat.tcps_rcvpack++;
2197 tcpstat.tcps_rcvbyte += tlen;
2198 ND6_HINT(tp);
2199 if (so->so_state & SS_CANTRCVMORE)
2200 m_freem(m);
2201 else
2202 sbappend(&so->so_rcv, m);
2203 sorwakeup(so);
2204 } else {
2205 thflags = tcp_reass(tp, th, &tlen, m);
2206 tp->t_flags |= TF_ACKNOW;
2207 }
2208
2209 /*
2210 * Note the amount of data that peer has sent into
2211 * our window, in order to estimate the sender's
2212 * buffer size.
2213 */
2214 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2215 } else {
2216 m_freem(m);
2217 thflags &= ~TH_FIN;
2218 }
2219
2220 /*
2221 * If FIN is received ACK the FIN and let the user know
2222 * that the connection is closing.
2223 */
2224 if (thflags & TH_FIN) {
2225 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2226 socantrcvmore(so);
2227 /*
2228 * If connection is half-synchronized
2229 * (ie NEEDSYN flag on) then delay ACK,
2230 * so it may be piggybacked when SYN is sent.
2231 * Otherwise, since we received a FIN then no
2232 * more input can be expected, send ACK now.
2233 */
2234 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2235 callout_reset(tp->tt_delack, tcp_delacktime,
2236 tcp_timer_delack, tp);
2237 else
2238 tp->t_flags |= TF_ACKNOW;
2239 tp->rcv_nxt++;
2240 }
2241 switch (tp->t_state) {
2242
2243 /*
2244 * In SYN_RECEIVED and ESTABLISHED STATES
2245 * enter the CLOSE_WAIT state.
2246 */
2247 case TCPS_SYN_RECEIVED:
2248 tp->t_starttime = ticks;
2249 /*FALLTHROUGH*/
2250 case TCPS_ESTABLISHED:
2251 tp->t_state = TCPS_CLOSE_WAIT;
2252 break;
2253
2254 /*
2255 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2256 * enter the CLOSING state.
2257 */
2258 case TCPS_FIN_WAIT_1:
2259 tp->t_state = TCPS_CLOSING;
2260 break;
2261
2262 /*
2263 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2264 * starting the time-wait timer, turning off the other
2265 * standard timers.
2266 */
2267 case TCPS_FIN_WAIT_2:
2268 tp->t_state = TCPS_TIME_WAIT;
2269 tcp_canceltimers(tp);
2270 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2271 if (tp->cc_recv != 0 &&
2272 (ticks - tp->t_starttime) < tcp_msl) {
2273 callout_reset(tp->tt_2msl,
2274 tp->t_rxtcur * TCPTV_TWTRUNC,
2275 tcp_timer_2msl, tp);
2276 /* For transaction client, force ACK now. */
2277 tp->t_flags |= TF_ACKNOW;
2278 }
2279 else
2280 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2281 tcp_timer_2msl, tp);
2282 soisdisconnected(so);
2283 break;
2284
2285 /*
2286 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2287 */
2288 case TCPS_TIME_WAIT:
2289 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2290 tcp_timer_2msl, tp);
2291 break;
2292 }
2293 }
2294#ifdef TCPDEBUG
2295 if (so->so_options & SO_DEBUG)
2296 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2297 &tcp_savetcp, 0);
2298#endif
2299
2300 /*
2301 * Return any desired output.
2302 */
2303 if (needoutput || (tp->t_flags & TF_ACKNOW))
2304 (void) tcp_output(tp);
2305 return;
2306
2307dropafterack:
2308 /*
2309 * Generate an ACK dropping incoming segment if it occupies
2310 * sequence space, where the ACK reflects our state.
2311 *
2312 * We can now skip the test for the RST flag since all
2313 * paths to this code happen after packets containing
2314 * RST have been dropped.
2315 *
2316 * In the SYN-RECEIVED state, don't send an ACK unless the
2317 * segment we received passes the SYN-RECEIVED ACK test.
2318 * If it fails send a RST. This breaks the loop in the
2319 * "LAND" DoS attack, and also prevents an ACK storm
2320 * between two listening ports that have been sent forged
2321 * SYN segments, each with the source address of the other.
2322 */
2323 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2324 (SEQ_GT(tp->snd_una, th->th_ack) ||
2325 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2326 rstreason = BANDLIM_RST_OPENPORT;
2327 goto dropwithreset;
2328 }
2329#ifdef TCPDEBUG
2330 if (so->so_options & SO_DEBUG)
2331 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2332 &tcp_savetcp, 0);
2333#endif
2334 m_freem(m);
2335 tp->t_flags |= TF_ACKNOW;
2336 (void) tcp_output(tp);
2337 return;
2338
2339dropwithreset:
2340 /*
2341 * Generate a RST, dropping incoming segment.
2342 * Make ACK acceptable to originator of segment.
2343 * Don't bother to respond if destination was broadcast/multicast.
2344 */
2345 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2346 goto drop;
2347 if (isipv6) {
2348 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2349 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2350 goto drop;
2351 } else {
2352 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2353 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2354 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2355 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2356 goto drop;
2357 }
2358 /* IPv6 anycast check is done at tcp6_input() */
2359
2360 /*
2361 * Perform bandwidth limiting.
2362 */
2363#ifdef ICMP_BANDLIM
2364 if (badport_bandlim(rstreason) < 0)
2365 goto drop;
2366#endif
2367
2368#ifdef TCPDEBUG
2369 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2370 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2371 &tcp_savetcp, 0);
2372#endif
2373 if (thflags & TH_ACK)
2374 /* mtod() below is safe as long as hdr dropping is delayed */
2375 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2376 TH_RST);
2377 else {
2378 if (thflags & TH_SYN)
2379 tlen++;
2380 /* mtod() below is safe as long as hdr dropping is delayed */
2381 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2382 (tcp_seq)0, TH_RST|TH_ACK);
2383 }
2384 return;
2385
2386drop:
2387 /*
2388 * Drop space held by incoming segment and return.
2389 */
2390#ifdef TCPDEBUG
2391 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2392 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2393 &tcp_savetcp, 0);
2394#endif
2395 m_freem(m);
2396 return;
2397}
2398
2399/*
2400 * Parse TCP options and place in tcpopt.
2401 */
2402static void
2403tcp_dooptions(to, cp, cnt, is_syn)
2404 struct tcpopt *to;
2405 u_char *cp;
2406 int cnt;
2407{
2408 int opt, optlen;
2409
2410 to->to_flags = 0;
2411 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2412 opt = cp[0];
2413 if (opt == TCPOPT_EOL)
2414 break;
2415 if (opt == TCPOPT_NOP)
2416 optlen = 1;
2417 else {
2418 if (cnt < 2)
2419 break;
2420 optlen = cp[1];
2421 if (optlen < 2 || optlen > cnt)
2422 break;
2423 }
2424 switch (opt) {
2425 case TCPOPT_MAXSEG:
2426 if (optlen != TCPOLEN_MAXSEG)
2427 continue;
2428 if (!is_syn)
2429 continue;
2430 to->to_flags |= TOF_MSS;
2431 bcopy((char *)cp + 2,
2432 (char *)&to->to_mss, sizeof(to->to_mss));
2433 to->to_mss = ntohs(to->to_mss);
2434 break;
2435 case TCPOPT_WINDOW:
2436 if (optlen != TCPOLEN_WINDOW)
2437 continue;
2438 if (! is_syn)
2439 continue;
2440 to->to_flags |= TOF_SCALE;
2441 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2442 break;
2443 case TCPOPT_TIMESTAMP:
2444 if (optlen != TCPOLEN_TIMESTAMP)
2445 continue;
2446 to->to_flags |= TOF_TS;
2447 bcopy((char *)cp + 2,
2448 (char *)&to->to_tsval, sizeof(to->to_tsval));
2449 to->to_tsval = ntohl(to->to_tsval);
2450 bcopy((char *)cp + 6,
2451 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2452 to->to_tsecr = ntohl(to->to_tsecr);
2453 break;
2454 case TCPOPT_CC:
2455 if (optlen != TCPOLEN_CC)
2456 continue;
2457 to->to_flags |= TOF_CC;
2458 bcopy((char *)cp + 2,
2459 (char *)&to->to_cc, sizeof(to->to_cc));
2460 to->to_cc = ntohl(to->to_cc);
2461 break;
2462 case TCPOPT_CCNEW:
2463 if (optlen != TCPOLEN_CC)
2464 continue;
2465 if (!is_syn)
2466 continue;
2467 to->to_flags |= TOF_CCNEW;
2468 bcopy((char *)cp + 2,
2469 (char *)&to->to_cc, sizeof(to->to_cc));
2470 to->to_cc = ntohl(to->to_cc);
2471 break;
2472 case TCPOPT_CCECHO:
2473 if (optlen != TCPOLEN_CC)
2474 continue;
2475 if (!is_syn)
2476 continue;
2477 to->to_flags |= TOF_CCECHO;
2478 bcopy((char *)cp + 2,
2479 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2480 to->to_ccecho = ntohl(to->to_ccecho);
2481 break;
2482 default:
2483 continue;
2484 }
2485 }
2486}
2487
2488/*
2489 * Pull out of band byte out of a segment so
2490 * it doesn't appear in the user's data queue.
2491 * It is still reflected in the segment length for
2492 * sequencing purposes.
2493 */
2494static void
2495tcp_pulloutofband(so, th, m, off)
2496 struct socket *so;
2497 struct tcphdr *th;
2256ba69 2498 struct mbuf *m;
984263bc
MD
2499 int off; /* delayed to be droped hdrlen */
2500{
2501 int cnt = off + th->th_urp - 1;
2502
2503 while (cnt >= 0) {
2504 if (m->m_len > cnt) {
2505 char *cp = mtod(m, caddr_t) + cnt;
2506 struct tcpcb *tp = sototcpcb(so);
2507
2508 tp->t_iobc = *cp;
2509 tp->t_oobflags |= TCPOOB_HAVEDATA;
2510 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2511 m->m_len--;
2512 if (m->m_flags & M_PKTHDR)
2513 m->m_pkthdr.len--;
2514 return;
2515 }
2516 cnt -= m->m_len;
2517 m = m->m_next;
2518 if (m == 0)
2519 break;
2520 }
2521 panic("tcp_pulloutofband");
2522}
2523
2524/*
2525 * Collect new round-trip time estimate
2526 * and update averages and current timeout.
2527 */
2528static void
2529tcp_xmit_timer(tp, rtt)
2256ba69 2530 struct tcpcb *tp;
984263bc
MD
2531 int rtt;
2532{
2256ba69 2533 int delta;
984263bc
MD
2534
2535 tcpstat.tcps_rttupdated++;
2536 tp->t_rttupdated++;
2537 if (tp->t_srtt != 0) {
2538 /*
2539 * srtt is stored as fixed point with 5 bits after the
2540 * binary point (i.e., scaled by 8). The following magic
2541 * is equivalent to the smoothing algorithm in rfc793 with
2542 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2543 * point). Adjust rtt to origin 0.
2544 */
2545 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2546 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2547
2548 if ((tp->t_srtt += delta) <= 0)
2549 tp->t_srtt = 1;
2550
2551 /*
2552 * We accumulate a smoothed rtt variance (actually, a
2553 * smoothed mean difference), then set the retransmit
2554 * timer to smoothed rtt + 4 times the smoothed variance.
2555 * rttvar is stored as fixed point with 4 bits after the
2556 * binary point (scaled by 16). The following is
2557 * equivalent to rfc793 smoothing with an alpha of .75
2558 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2559 * rfc793's wired-in beta.
2560 */
2561 if (delta < 0)
2562 delta = -delta;
2563 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2564 if ((tp->t_rttvar += delta) <= 0)
2565 tp->t_rttvar = 1;
2566 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2567 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2568 } else {
2569 /*
2570 * No rtt measurement yet - use the unsmoothed rtt.
2571 * Set the variance to half the rtt (so our first
2572 * retransmit happens at 3*rtt).
2573 */
2574 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2575 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2576 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2577 }
2578 tp->t_rtttime = 0;
2579 tp->t_rxtshift = 0;
2580
2581 /*
2582 * the retransmit should happen at rtt + 4 * rttvar.
2583 * Because of the way we do the smoothing, srtt and rttvar
2584 * will each average +1/2 tick of bias. When we compute
2585 * the retransmit timer, we want 1/2 tick of rounding and
2586 * 1 extra tick because of +-1/2 tick uncertainty in the
2587 * firing of the timer. The bias will give us exactly the
2588 * 1.5 tick we need. But, because the bias is
2589 * statistical, we have to test that we don't drop below
2590 * the minimum feasible timer (which is 2 ticks).
2591 */
2592 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2593 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2594
2595 /*
2596 * We received an ack for a packet that wasn't retransmitted;
2597 * it is probably safe to discard any error indications we've
2598 * received recently. This isn't quite right, but close enough
2599 * for now (a route might have failed after we sent a segment,
2600 * and the return path might not be symmetrical).
2601 */
2602 tp->t_softerror = 0;
2603}
2604
2605/*
2606 * Determine a reasonable value for maxseg size.
2607 * If the route is known, check route for mtu.
2608 * If none, use an mss that can be handled on the outgoing
2609 * interface without forcing IP to fragment; if bigger than
2610 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2611 * to utilize large mbufs. If no route is found, route has no mtu,
2612 * or the destination isn't local, use a default, hopefully conservative
2613 * size (usually 512 or the default IP max size, but no more than the mtu
2614 * of the interface), as we can't discover anything about intervening
2615 * gateways or networks. We also initialize the congestion/slow start
2616 * window to be a single segment if the destination isn't local.
2617 * While looking at the routing entry, we also initialize other path-dependent
2618 * parameters from pre-set or cached values in the routing entry.
2619 *
2620 * Also take into account the space needed for options that we
2621 * send regularly. Make maxseg shorter by that amount to assure
2622 * that we can send maxseg amount of data even when the options
2623 * are present. Store the upper limit of the length of options plus
2624 * data in maxopd.
2625 *
2626 * NOTE that this routine is only called when we process an incoming
2627 * segment, for outgoing segments only tcp_mssopt is called.
2628 *
2629 * In case of T/TCP, we call this routine during implicit connection
2630 * setup as well (offer = -1), to initialize maxseg from the cached
2631 * MSS of our peer.
2632 */
2633void
2634tcp_mss(tp, offer)
2635 struct tcpcb *tp;
2636 int offer;
2637{
2256ba69 2638 struct rtentry *rt;
984263bc 2639 struct ifnet *ifp;
2256ba69 2640 int rtt, mss;
984263bc
MD
2641 u_long bufsize;
2642 struct inpcb *inp = tp->t_inpcb;
2643 struct socket *so;
2644 struct rmxp_tao *taop;
2645 int origoffer = offer;
2646#ifdef INET6
2647 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2648 size_t min_protoh = isipv6 ?
2649 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2650 sizeof(struct tcpiphdr);
2651#else
2652 const int isipv6 = 0;
2653 const size_t min_protoh = sizeof(struct tcpiphdr);
2654#endif
2655
2656 if (isipv6)
2657 rt = tcp_rtlookup6(&inp->inp_inc);
2658 else
2659 rt = tcp_rtlookup(&inp->inp_inc);
2660 if (rt == NULL) {
2661 tp->t_maxopd = tp->t_maxseg =
2662 isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2663 return;
2664 }
2665 ifp = rt->rt_ifp;
2666 so = inp->inp_socket;
2667
2668 taop = rmx_taop(rt->rt_rmx);
2669 /*
2670 * Offer == -1 means that we didn't receive SYN yet,
2671 * use cached value in that case;
2672 */
2673 if (offer == -1)
2674 offer = taop->tao_mssopt;
2675 /*
2676 * Offer == 0 means that there was no MSS on the SYN segment,
2677 * in this case we use tcp_mssdflt.
2678 */
2679 if (offer == 0)
2680 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
2681 else
2682 /*
2683 * Sanity check: make sure that maxopd will be large
2684 * enough to allow some data on segments even is the
2685 * all the option space is used (40bytes). Otherwise
2686 * funny things may happen in tcp_output.
2687 */
2688 offer = max(offer, 64);
2689 taop->tao_mssopt = offer;
2690
2691 /*
2692 * While we're here, check if there's an initial rtt
2693 * or rttvar. Convert from the route-table units
2694 * to scaled multiples of the slow timeout timer.
2695 */
2696 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2697 /*
2698 * XXX the lock bit for RTT indicates that the value
2699 * is also a minimum value; this is subject to time.
2700 */
2701 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2702 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2703 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2704 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2705 tcpstat.tcps_usedrtt++;
2706 if (rt->rt_rmx.rmx_rttvar) {
2707 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2708 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2709 tcpstat.tcps_usedrttvar++;
2710 } else {
2711 /* default variation is +- 1 rtt */
2712 tp->t_rttvar =
2713 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2714 }
2715 TCPT_RANGESET(tp->t_rxtcur,
2716 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2717 tp->t_rttmin, TCPTV_REXMTMAX);
2718 }
2719 /*
2720 * if there's an mtu associated with the route, use it
2721 * else, use the link mtu.
2722 */
2723 if (rt->rt_rmx.rmx_mtu)
2724 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2725 else {
2726 if (isipv6) {
2727 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2728 min_protoh;
2729 if (!in6_localaddr(&inp->in6p_faddr))
2730 mss = min(mss, tcp_v6mssdflt);
2731 } else {
2732 mss = ifp->if_mtu - min_protoh;
2733 if (!in_localaddr(inp->inp_faddr))
2734 mss = min(mss, tcp_mssdflt);
2735 }
2736 }
2737 mss = min(mss, offer);
2738 /*
2739 * maxopd stores the maximum length of data AND options
2740 * in a segment; maxseg is the amount of data in a normal
2741 * segment. We need to store this value (maxopd) apart
2742 * from maxseg, because now every segment carries options
2743 * and thus we normally have somewhat less data in segments.
2744 */
2745 tp->t_maxopd = mss;
2746
2747 /*
2748 * In case of T/TCP, origoffer==-1 indicates, that no segments
2749 * were received yet. In this case we just guess, otherwise
2750 * we do the same as before T/TCP.
2751 */
2752 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2753 (origoffer == -1 ||
2754 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2755 mss -= TCPOLEN_TSTAMP_APPA;
2756 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2757 (origoffer == -1 ||
2758 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2759 mss -= TCPOLEN_CC_APPA;
2760
2761#if (MCLBYTES & (MCLBYTES - 1)) == 0
2762 if (mss > MCLBYTES)
2763 mss &= ~(MCLBYTES-1);
2764#else
2765 if (mss > MCLBYTES)
2766 mss = mss / MCLBYTES * MCLBYTES;
2767#endif
2768 /*
2769 * If there's a pipesize, change the socket buffer
2770 * to that size. Make the socket buffers an integral
2771 * number of mss units; if the mss is larger than
2772 * the socket buffer, decrease the mss.
2773 */
2774#ifdef RTV_SPIPE
2775 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2776#endif
2777 bufsize = so->so_snd.sb_hiwat;
2778 if (bufsize < mss)
2779 mss = bufsize;
2780 else {
2781 bufsize = roundup(bufsize, mss);
2782 if (bufsize > sb_max)
2783 bufsize = sb_max;
2784 if (bufsize > so->so_snd.sb_hiwat)
2785 (void)sbreserve(&so->so_snd, bufsize, so, NULL);
2786 }
2787 tp->t_maxseg = mss;
2788
2789#ifdef RTV_RPIPE
2790 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2791#endif
2792 bufsize = so->so_rcv.sb_hiwat;
2793 if (bufsize > mss) {
2794 bufsize = roundup(bufsize, mss);
2795 if (bufsize > sb_max)
2796 bufsize = sb_max;
2797 if (bufsize > so->so_rcv.sb_hiwat)
2798 (void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2799 }
2800
2801 /*
2802 * Set the slow-start flight size depending on whether this
2803 * is a local network or not.
2804 */
4b52d5ee
JH
2805 if (tcp_do_rfc3390)
2806 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2807 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2808 (!isipv6 && in_localaddr(inp->inp_faddr)))
984263bc 2809 tp->snd_cwnd = mss * ss_fltsz_local;
4b52d5ee 2810 else
984263bc
MD
2811 tp->snd_cwnd = mss * ss_fltsz;
2812
2813 if (rt->rt_rmx.rmx_ssthresh) {
2814 /*
2815 * There's some sort of gateway or interface
2816 * buffer limit on the path. Use this to set
2817 * the slow start threshhold, but set the
2818 * threshold to no less than 2*mss.
2819 */
2820 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2821 tcpstat.tcps_usedssthresh++;
2822 }
2823}
2824
2825/*
2826 * Determine the MSS option to send on an outgoing SYN.
2827 */
2828int
2829tcp_mssopt(tp)
2830 struct tcpcb *tp;
2831{
2832 struct rtentry *rt;
2833#ifdef INET6
2834 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2835 int min_protoh = isipv6 ?
2836 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2837 sizeof(struct tcpiphdr);
2838#else
2839 const int isipv6 = 0;
2840 const size_t min_protoh = sizeof(struct tcpiphdr);
2841#endif
2842
2843 if (isipv6)
2844 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
2845 else
2846 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
2847 if (rt == NULL)
2848 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2849
2850 return (rt->rt_ifp->if_mtu - min_protoh);
2851}
2852
2853
2854/*
2855 * When a partial ack arrives, force the retransmission of the
2856 * next unacknowledged segment. Do not clear tp->t_dupacks.
2857 * By setting snd_nxt to ti_ack, this forces retransmission timer to
2858 * be started again.
2859 */
2860static void
2861tcp_newreno_partial_ack(tp, th)
2862 struct tcpcb *tp;
2863 struct tcphdr *th;
2864{
2865 tcp_seq onxt = tp->snd_nxt;
2866 u_long ocwnd = tp->snd_cwnd;
2867
2868 callout_stop(tp->tt_rexmt);
2869 tp->t_rtttime = 0;
2870 tp->snd_nxt = th->th_ack;
2871 /*
2872 * Set snd_cwnd to one segment beyond acknowledged offset
2873 * (tp->snd_una has not yet been updated when this function is called.)
2874 */
2875 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
2876 tp->t_flags |= TF_ACKNOW;
2877 (void) tcp_output(tp);
2878 tp->snd_cwnd = ocwnd;
2879 if (SEQ_GT(onxt, tp->snd_nxt))
2880 tp->snd_nxt = onxt;
2881 /*
2882 * Partial window deflation. Relies on fact that tp->snd_una
2883 * not updated yet.
2884 */
2885 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
2886}