Get rid of dead non-DragonFly code.
[dragonfly.git] / sys / netinet / tcp_input.c
CommitLineData
984263bc 1/*
66d6c637
JH
2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
95b22adf 4 *
66d6c637
JH
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
95b22adf 7 *
66d6c637
JH
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
95b22adf 19 *
66d6c637
JH
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34/*
35 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
36 *
37 * License terms: all terms for the DragonFly license above plus the following:
38 *
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
41 *
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
44 *
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
48 * apply.
49 */
50
51/*
984263bc
MD
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
53 * The Regents of the University of California. All rights reserved.
54 *
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
57 * are met:
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
70 *
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
81 * SUCH DAMAGE.
82 *
83 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
84 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
1141eb20 85 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.47 2004/12/29 01:19:53 hsu Exp $
984263bc
MD
86 */
87
88#include "opt_ipfw.h" /* for ipfw_fwd */
89#include "opt_inet6.h"
90#include "opt_ipsec.h"
91#include "opt_tcpdebug.h"
92#include "opt_tcp_input.h"
93
94#include <sys/param.h>
95#include <sys/systm.h>
96#include <sys/kernel.h>
97#include <sys/sysctl.h>
98#include <sys/malloc.h>
99#include <sys/mbuf.h>
100#include <sys/proc.h> /* for proc0 declaration */
101#include <sys/protosw.h>
102#include <sys/socket.h>
103#include <sys/socketvar.h>
104#include <sys/syslog.h>
3f9db7f8 105#include <sys/in_cksum.h>
984263bc
MD
106
107#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
a00138cb 108#include <machine/stdarg.h>
984263bc
MD
109
110#include <net/if.h>
111#include <net/route.h>
112
113#include <netinet/in.h>
114#include <netinet/in_systm.h>
115#include <netinet/ip.h>
95b22adf 116#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
984263bc 117#include <netinet/in_var.h>
95b22adf 118#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
984263bc
MD
119#include <netinet/in_pcb.h>
120#include <netinet/ip_var.h>
121#include <netinet/ip6.h>
122#include <netinet/icmp6.h>
123#include <netinet6/nd6.h>
124#include <netinet6/ip6_var.h>
125#include <netinet6/in6_pcb.h>
126#include <netinet/tcp.h>
127#include <netinet/tcp_fsm.h>
128#include <netinet/tcp_seq.h>
129#include <netinet/tcp_timer.h>
130#include <netinet/tcp_var.h>
131#include <netinet6/tcp6_var.h>
132#include <netinet/tcpip.h>
95b22adf 133
984263bc
MD
134#ifdef TCPDEBUG
135#include <netinet/tcp_debug.h>
136
95b22adf 137u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
984263bc 138struct tcphdr tcp_savetcp;
95b22adf 139#endif
984263bc
MD
140
141#ifdef FAST_IPSEC
bf844ffa
JH
142#include <netproto/ipsec/ipsec.h>
143#include <netproto/ipsec/ipsec6.h>
984263bc
MD
144#endif
145
146#ifdef IPSEC
147#include <netinet6/ipsec.h>
148#include <netinet6/ipsec6.h>
d2438d69 149#include <netproto/key/key.h>
95b22adf 150#endif
984263bc 151
984263bc
MD
152MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
153
984263bc 154tcp_cc tcp_ccgen;
984263bc 155static int log_in_vain = 0;
d24ce1dc 156SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
984263bc
MD
157 &log_in_vain, 0, "Log all incoming TCP connections");
158
159static int blackhole = 0;
160SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
161 &blackhole, 0, "Do not send RST when dropping refused connections");
162
163int tcp_delack_enabled = 1;
d24ce1dc
JH
164SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
165 &tcp_delack_enabled, 0,
984263bc
MD
166 "Delay ACK to try and piggyback it onto a data packet");
167
168#ifdef TCP_DROP_SYNFIN
169static int drop_synfin = 0;
170SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
171 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
172#endif
173
33abdd1c
MD
174static int tcp_do_limitedtransmit = 1;
175SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
176 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
177
91489f6b 178static int tcp_do_early_retransmit = 1;
8819433a
JH
179SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
180 &tcp_do_early_retransmit, 0, "Early retransmit");
181
4b52d5ee
JH
182static int tcp_do_rfc3390 = 1;
183SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
184 &tcp_do_rfc3390, 0,
185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
186
efd4b327
JH
187static int tcp_do_eifel_detect = 1;
188SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
189 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
190
91489f6b
JH
191/*
192 * Define as tunable for easy testing with SACK on and off.
193 * Warning: do not change setting in the middle of an existing active TCP flow,
194 * else strange things might happen to that flow.
195 */
6e5bbdda 196int tcp_do_sack = 1;
91489f6b
JH
197SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
198 &tcp_do_sack, 0, "Enable SACK Algorithms");
199
200int tcp_do_smartsack = 1;
201SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
202 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
203
3edf7c37
RG
204SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
205 "TCP Segment Reassembly Queue");
206
207int tcp_reass_maxseg = 0;
208SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
209 &tcp_reass_maxseg, 0,
210 "Global maximum number of TCP Segments in Reassembly Queue");
211
212int tcp_reass_qsize = 0;
213SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
214 &tcp_reass_qsize, 0,
215 "Global number of TCP Segments currently in Reassembly Queue");
216
217static int tcp_reass_overflows = 0;
218SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
219 &tcp_reass_overflows, 0,
220 "Global number of TCP Segment Reassembly Queue Overflows");
221
95b22adf 222static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
984263bc
MD
223static void tcp_pulloutofband(struct socket *,
224 struct tcphdr *, struct mbuf *, int);
225static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
226 struct mbuf *);
227static void tcp_xmit_timer(struct tcpcb *, int);
91489f6b
JH
228static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
229static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
984263bc
MD
230
231/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
232#ifdef INET6
233#define ND6_HINT(tp) \
234do { \
235 if ((tp) && (tp)->t_inpcb && \
61896e3c 236 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
984263bc
MD
237 (tp)->t_inpcb->in6p_route.ro_rt) \
238 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
239} while (0)
240#else
241#define ND6_HINT(tp)
242#endif
243
244/*
245 * Indicate whether this ack should be delayed. We can delay the ack if
246 * - delayed acks are enabled and
247 * - there is no delayed ack timer in progress and
248 * - our last ack wasn't a 0-sized window. We never want to delay
249 * the ack that opens up a 0-sized window.
250 */
251#define DELAY_ACK(tp) \
252 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
61896e3c 253 !(tp->t_flags & TF_RXWIN0SENT))
984263bc 254
df9d7670
JH
255#define acceptable_window_update(tp, th, tiwin) \
256 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
257 (tp->snd_wl1 == th->th_seq && \
258 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
259 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
260
984263bc 261static int
95b22adf 262tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
984263bc
MD
263{
264 struct tseg_qent *q;
265 struct tseg_qent *p = NULL;
984263bc
MD
266 struct tseg_qent *te;
267 struct socket *so = tp->t_inpcb->inp_socket;
268 int flags;
269
270 /*
61896e3c 271 * Call with th == NULL after become established to
984263bc
MD
272 * force pre-ESTABLISHED data up to user socket.
273 */
61896e3c 274 if (th == NULL)
984263bc
MD
275 goto present;
276
3edf7c37
RG
277 /*
278 * Limit the number of segments in the reassembly queue to prevent
279 * holding on to too many segments (and thus running out of mbufs).
280 * Make sure to let the missing segment through which caused this
281 * queue. Always keep one global queue entry spare to be able to
282 * process the missing segment.
283 */
284 if (th->th_seq != tp->rcv_nxt &&
285 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
286 tcp_reass_overflows++;
287 tcpstat.tcps_rcvmemdrop++;
288 m_freem(m);
91489f6b
JH
289 /* no SACK block to report */
290 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
3edf7c37
RG
291 return (0);
292 }
293
ba4e3dbe 294 /* Allocate a new queue entry. */
984263bc 295 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
ba4e3dbe 296 M_INTWAIT | M_NULLOK);
984263bc
MD
297 if (te == NULL) {
298 tcpstat.tcps_rcvmemdrop++;
299 m_freem(m);
91489f6b
JH
300 /* no SACK block to report */
301 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
984263bc
MD
302 return (0);
303 }
3edf7c37 304 tcp_reass_qsize++;
984263bc
MD
305
306 /*
307 * Find a segment which begins after this one does.
308 */
309 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
310 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
311 break;
312 p = q;
313 }
314
315 /*
316 * If there is a preceding segment, it may provide some of
317 * our data already. If so, drop the data from the incoming
318 * segment. If it provides all of our data, drop us.
319 */
320 if (p != NULL) {
2256ba69 321 int i;
91489f6b 322
984263bc
MD
323 /* conversion to int (in i) handles seq wraparound */
324 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
91489f6b
JH
325 if (i > 0) { /* overlaps preceding segment */
326 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
327 /* enclosing block starts w/ preceding segment */
328 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
984263bc 329 if (i >= *tlenp) {
91489f6b
JH
330 /* preceding encloses incoming segment */
331 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
332 p->tqe_len;
984263bc
MD
333 tcpstat.tcps_rcvduppack++;
334 tcpstat.tcps_rcvdupbyte += *tlenp;
335 m_freem(m);
336 free(te, M_TSEGQ);
3edf7c37 337 tcp_reass_qsize--;
984263bc
MD
338 /*
339 * Try to present any queued data
340 * at the left window edge to the user.
341 * This is needed after the 3-WHS
342 * completes.
343 */
344 goto present; /* ??? */
345 }
346 m_adj(m, i);
347 *tlenp -= i;
348 th->th_seq += i;
91489f6b
JH
349 /* incoming segment end is enclosing block end */
350 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
351 ((th->th_flags & TH_FIN) != 0);
352 /* trim end of reported D-SACK block */
353 tp->reportblk.rblk_end = th->th_seq;
984263bc
MD
354 }
355 }
356 tcpstat.tcps_rcvoopack++;
357 tcpstat.tcps_rcvoobyte += *tlenp;
358
359 /*
360 * While we overlap succeeding segments trim them or,
361 * if they are completely covered, dequeue them.
362 */
363 while (q) {
2256ba69 364 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
91489f6b
JH
365 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
366 struct tseg_qent *nq;
367
984263bc
MD
368 if (i <= 0)
369 break;
91489f6b
JH
370 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
371 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
372 tp->encloseblk = tp->reportblk;
373 /* report trailing duplicate D-SACK segment */
374 tp->reportblk.rblk_start = q->tqe_th->th_seq;
375 }
376 if ((tp->t_flags & TF_ENCLOSESEG) &&
377 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
378 /* extend enclosing block if one exists */
379 tp->encloseblk.rblk_end = qend;
380 }
984263bc
MD
381 if (i < q->tqe_len) {
382 q->tqe_th->th_seq += i;
383 q->tqe_len -= i;
384 m_adj(q->tqe_m, i);
385 break;
386 }
387
388 nq = LIST_NEXT(q, tqe_q);
389 LIST_REMOVE(q, tqe_q);
390 m_freem(q->tqe_m);
391 free(q, M_TSEGQ);
3edf7c37 392 tcp_reass_qsize--;
984263bc
MD
393 q = nq;
394 }
395
396 /* Insert the new segment queue entry into place. */
397 te->tqe_m = m;
398 te->tqe_th = th;
399 te->tqe_len = *tlenp;
400
91489f6b
JH
401 /* check if can coalesce with following segment */
402 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
403 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
404
405 te->tqe_len += q->tqe_len;
a174690a 406 if (q->tqe_th->th_flags & TH_FIN)
f16f8cc3 407 te->tqe_th->th_flags |= TH_FIN;
91489f6b
JH
408 m_cat(te->tqe_m, q->tqe_m);
409 tp->encloseblk.rblk_end = tend;
410 /*
411 * When not reporting a duplicate segment, use
412 * the larger enclosing block as the SACK block.
413 */
414 if (!(tp->t_flags & TF_DUPSEG))
415 tp->reportblk.rblk_end = tend;
416 LIST_REMOVE(q, tqe_q);
417 free(q, M_TSEGQ);
418 tcp_reass_qsize--;
419 }
420
984263bc
MD
421 if (p == NULL) {
422 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
423 } else {
91489f6b
JH
424 /* check if can coalesce with preceding segment */
425 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
a174690a
JH
426 p->tqe_len += te->tqe_len;
427 m_cat(p->tqe_m, te->tqe_m);
91489f6b
JH
428 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
429 /*
430 * When not reporting a duplicate segment, use
431 * the larger enclosing block as the SACK block.
432 */
433 if (!(tp->t_flags & TF_DUPSEG))
434 tp->reportblk.rblk_start = p->tqe_th->th_seq;
a174690a
JH
435 free(te, M_TSEGQ);
436 tcp_reass_qsize--;
91489f6b
JH
437 } else
438 LIST_INSERT_AFTER(p, te, tqe_q);
984263bc
MD
439 }
440
441present:
442 /*
443 * Present data to user, advancing rcv_nxt through
444 * completed sequence space.
445 */
446 if (!TCPS_HAVEESTABLISHED(tp->t_state))
447 return (0);
448 q = LIST_FIRST(&tp->t_segq);
61896e3c 449 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
984263bc 450 return (0);
91489f6b
JH
451 tp->rcv_nxt += q->tqe_len;
452 if (!(tp->t_flags & TF_DUPSEG)) {
453 /* no SACK block to report since ACK advanced */
454 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
455 }
456 /* no enclosing block to report since ACK advanced */
457 tp->t_flags &= ~TF_ENCLOSESEG;
458 flags = q->tqe_th->th_flags & TH_FIN;
459 LIST_REMOVE(q, tqe_q);
460 KASSERT(LIST_EMPTY(&tp->t_segq) ||
461 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
462 ("segment not coalesced"));
463 if (so->so_state & SS_CANTRCVMORE)
464 m_freem(q->tqe_m);
465 else
507106e0 466 sbappendstream(&so->so_rcv, q->tqe_m);
91489f6b
JH
467 free(q, M_TSEGQ);
468 tcp_reass_qsize--;
984263bc
MD
469 ND6_HINT(tp);
470 sorwakeup(so);
471 return (flags);
472}
473
474/*
475 * TCP input routine, follows pages 65-76 of the
476 * protocol specification dated September, 1981 very closely.
477 */
478#ifdef INET6
479int
95b22adf 480tcp6_input(struct mbuf **mp, int *offp, int proto)
984263bc 481{
2256ba69 482 struct mbuf *m = *mp;
984263bc
MD
483 struct in6_ifaddr *ia6;
484
485 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
486
487 /*
488 * draft-itojun-ipv6-tcp-to-anycast
489 * better place to put this in?
490 */
491 ia6 = ip6_getdstifaddr(m);
d24ce1dc 492 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
984263bc
MD
493 struct ip6_hdr *ip6;
494
495 ip6 = mtod(m, struct ip6_hdr *);
496 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
f23061d4 497 offsetof(struct ip6_hdr, ip6_dst));
d24ce1dc 498 return (IPPROTO_DONE);
984263bc
MD
499 }
500
501 tcp_input(m, *offp, proto);
d24ce1dc 502 return (IPPROTO_DONE);
984263bc
MD
503}
504#endif
505
506void
a00138cb 507tcp_input(struct mbuf *m, ...)
984263bc 508{
a00138cb
JS
509 __va_list ap;
510 int off0, proto;
2256ba69
RG
511 struct tcphdr *th;
512 struct ip *ip = NULL;
513 struct ipovly *ipov;
514 struct inpcb *inp = NULL;
984263bc
MD
515 u_char *optp = NULL;
516 int optlen = 0;
517 int len, tlen, off;
518 int drop_hdrlen;
2256ba69
RG
519 struct tcpcb *tp = NULL;
520 int thflags;
984263bc 521 struct socket *so = 0;
61896e3c
JH
522 int todrop, acked;
523 boolean_t ourfinisacked, needoutput = FALSE;
984263bc 524 u_long tiwin;
61896e3c 525 int recvwin;
984263bc
MD
526 struct tcpopt to; /* options in this segment */
527 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
528 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
529 struct sockaddr_in *next_hop = NULL;
530 int rstreason; /* For badport_bandlim accounting purposes */
d371a63a 531 int cpu;
984263bc
MD
532 struct ip6_hdr *ip6 = NULL;
533#ifdef INET6
d24ce1dc 534 boolean_t isipv6;
984263bc 535#else
d24ce1dc 536 const boolean_t isipv6 = FALSE;
984263bc
MD
537#endif
538#ifdef TCPDEBUG
539 short ostate = 0;
540#endif
541
a00138cb
JS
542 __va_start(ap, m);
543 off0 = __va_arg(ap, int);
544 proto = __va_arg(ap, int);
545 __va_end(ap);
546
d24ce1dc
JH
547 tcpstat.tcps_rcvtotal++;
548
549 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
550 while (m->m_type == MT_TAG) {
984263bc
MD
551 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
552 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
d24ce1dc 553 m = m->m_next;
984263bc 554 }
d24ce1dc 555
984263bc 556#ifdef INET6
d24ce1dc 557 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
984263bc 558#endif
984263bc
MD
559
560 if (isipv6) {
561 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
562 ip6 = mtod(m, struct ip6_hdr *);
563 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
564 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
565 tcpstat.tcps_rcvbadsum++;
566 goto drop;
567 }
568 th = (struct tcphdr *)((caddr_t)ip6 + off0);
569
570 /*
571 * Be proactive about unspecified IPv6 address in source.
572 * As we use all-zero to indicate unbounded/unconnected pcb,
573 * unspecified IPv6 address can be used to confuse us.
574 *
575 * Note that packets with unspecified IPv6 destination is
576 * already dropped in ip6_input.
577 */
578 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
579 /* XXX stat */
580 goto drop;
581 }
582 } else {
583 /*
584 * Get IP and TCP header together in first mbuf.
585 * Note: IP leaves IP header in first mbuf.
586 */
587 if (off0 > sizeof(struct ip)) {
bddf0751 588 ip_stripoptions(m);
984263bc
MD
589 off0 = sizeof(struct ip);
590 }
55d829f8
JH
591 /* already checked and pulled up in ip_demux() */
592 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
61896e3c 593 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
984263bc
MD
594 ip = mtod(m, struct ip *);
595 ipov = (struct ipovly *)ip;
596 th = (struct tcphdr *)((caddr_t)ip + off0);
597 tlen = ip->ip_len;
598
599 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
600 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
601 th->th_sum = m->m_pkthdr.csum_data;
602 else
603 th->th_sum = in_pseudo(ip->ip_src.s_addr,
604 ip->ip_dst.s_addr,
605 htonl(m->m_pkthdr.csum_data +
606 ip->ip_len +
607 IPPROTO_TCP));
608 th->th_sum ^= 0xffff;
609 } else {
610 /*
611 * Checksum extended TCP header and data.
612 */
613 len = sizeof(struct ip) + tlen;
614 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
615 ipov->ih_len = (u_short)tlen;
616 ipov->ih_len = htons(ipov->ih_len);
617 th->th_sum = in_cksum(m, len);
618 }
619 if (th->th_sum) {
620 tcpstat.tcps_rcvbadsum++;
621 goto drop;
622 }
623#ifdef INET6
624 /* Re-initialization for later version check */
625 ip->ip_v = IPVERSION;
626#endif
627 }
628
629 /*
630 * Check that TCP offset makes sense,
631 * pull out TCP options and adjust length. XXX
632 */
633 off = th->th_off << 2;
55d829f8
JH
634 /* already checked and pulled up in ip_demux() */
635 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
61896e3c 636 ("bad TCP data offset %d (tlen %d)", off, tlen));
984263bc
MD
637 tlen -= off; /* tlen is used instead of ti->ti_len */
638 if (off > sizeof(struct tcphdr)) {
639 if (isipv6) {
640 IP6_EXTHDR_CHECK(m, off0, off, );
641 ip6 = mtod(m, struct ip6_hdr *);
642 th = (struct tcphdr *)((caddr_t)ip6 + off0);
643 } else {
55d829f8
JH
644 /* already pulled up in ip_demux() */
645 KASSERT(m->m_len >= sizeof(struct ip) + off,
61896e3c
JH
646 ("TCP header and options not in one mbuf: "
647 "m_len %d, off %d", m->m_len, off));
984263bc
MD
648 }
649 optlen = off - sizeof(struct tcphdr);
650 optp = (u_char *)(th + 1);
651 }
652 thflags = th->th_flags;
653
654#ifdef TCP_DROP_SYNFIN
655 /*
656 * If the drop_synfin option is enabled, drop all packets with
657 * both the SYN and FIN bits set. This prevents e.g. nmap from
658 * identifying the TCP/IP stack.
659 *
660 * This is a violation of the TCP specification.
661 */
61896e3c 662 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
984263bc
MD
663 goto drop;
664#endif
665
666 /*
667 * Convert TCP protocol specific fields to host format.
668 */
669 th->th_seq = ntohl(th->th_seq);
670 th->th_ack = ntohl(th->th_ack);
671 th->th_win = ntohs(th->th_win);
672 th->th_urp = ntohs(th->th_urp);
673
674 /*
d24ce1dc 675 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
984263bc
MD
676 * until after ip6_savecontrol() is called and before other functions
677 * which don't want those proto headers.
678 * Because ip6_savecontrol() is going to parse the mbuf to
679 * search for data to be passed up to user-land, it wants mbuf
680 * parameters to be unchanged.
681 * XXX: the call of ip6_savecontrol() has been obsoleted based on
682 * latest version of the advanced API (20020110).
683 */
684 drop_hdrlen = off0 + off;
685
686 /*
687 * Locate pcb for segment.
688 */
689findpcb:
690 /* IPFIREWALL_FORWARD section */
d24ce1dc 691 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
984263bc
MD
692 /*
693 * Transparently forwarded. Pretend to be the destination.
d24ce1dc 694 * already got one like this?
984263bc 695 */
6ca1a1cd 696 cpu = mycpu->gd_cpuid;
6ca1a1cd 697 inp = in_pcblookup_hash(&tcbinfo[cpu],
d371a63a 698 ip->ip_src, th->th_sport,
984263bc
MD
699 ip->ip_dst, th->th_dport,
700 0, m->m_pkthdr.rcvif);
701 if (!inp) {
83be63fe
JH
702 /*
703 * It's new. Try to find the ambushing socket.
704 */
705
706 /*
707 * The rest of the ipfw code stores the port in
708 * host order. XXX
709 * (The IP address is still in network order.)
710 */
711 in_port_t dport = next_hop->sin_port ?
712 htons(next_hop->sin_port) :
713 th->th_dport;
714
d371a63a 715 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
83be63fe 716 next_hop->sin_addr.s_addr, dport);
d371a63a 717 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc 718 ip->ip_src, th->th_sport,
83be63fe 719 next_hop->sin_addr, dport,
984263bc
MD
720 1, m->m_pkthdr.rcvif);
721 }
722 } else {
6ca1a1cd 723 if (isipv6) {
d371a63a 724 inp = in6_pcblookup_hash(&tcbinfo[0],
984263bc
MD
725 &ip6->ip6_src, th->th_sport,
726 &ip6->ip6_dst, th->th_dport,
727 1, m->m_pkthdr.rcvif);
6ca1a1cd 728 } else {
6ca1a1cd 729 cpu = mycpu->gd_cpuid;
6ca1a1cd 730 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc
MD
731 ip->ip_src, th->th_sport,
732 ip->ip_dst, th->th_dport,
733 1, m->m_pkthdr.rcvif);
6ca1a1cd 734 }
984263bc 735 }
984263bc
MD
736
737 /*
738 * If the state is CLOSED (i.e., TCB does not exist) then
739 * all data in the incoming segment is discarded.
740 * If the TCB exists but is in CLOSED state, it is embryonic,
741 * but should either do a listen or a connect soon.
742 */
743 if (inp == NULL) {
744 if (log_in_vain) {
745#ifdef INET6
746 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
747#else
1141eb20
JH
748 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
749 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
984263bc
MD
750#endif
751 if (isipv6) {
752 strcpy(dbuf, "[");
984263bc 753 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
984263bc 754 strcat(dbuf, "]");
95b22adf
JH
755 strcpy(sbuf, "[");
756 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
984263bc
MD
757 strcat(sbuf, "]");
758 } else {
759 strcpy(dbuf, inet_ntoa(ip->ip_dst));
760 strcpy(sbuf, inet_ntoa(ip->ip_src));
761 }
762 switch (log_in_vain) {
763 case 1:
61896e3c 764 if (!(thflags & TH_SYN))
984263bc
MD
765 break;
766 case 2:
767 log(LOG_INFO,
768 "Connection attempt to TCP %s:%d "
769 "from %s:%d flags:0x%02x\n",
770 dbuf, ntohs(th->th_dport), sbuf,
771 ntohs(th->th_sport), thflags);
772 break;
773 default:
774 break;
775 }
776 }
d24ce1dc 777 if (blackhole) {
984263bc
MD
778 switch (blackhole) {
779 case 1:
780 if (thflags & TH_SYN)
781 goto drop;
782 break;
783 case 2:
784 goto drop;
785 default:
786 goto drop;
787 }
788 }
789 rstreason = BANDLIM_RST_CLOSEDPORT;
790 goto dropwithreset;
791 }
61896e3c
JH
792
793#ifdef IPSEC
794 if (isipv6) {
795 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
796 ipsec6stat.in_polvio++;
797 goto drop;
798 }
799 } else {
800 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
801 ipsecstat.in_polvio++;
802 goto drop;
803 }
804 }
805#endif
806#ifdef FAST_IPSEC
807 if (isipv6) {
808 if (ipsec6_in_reject(m, inp))
809 goto drop;
810 } else {
811 if (ipsec4_in_reject(m, inp))
812 goto drop;
813 }
814#endif
815
984263bc
MD
816 tp = intotcpcb(inp);
817 if (tp == NULL) {
818 rstreason = BANDLIM_RST_CLOSEDPORT;
819 goto dropwithreset;
820 }
eb594563 821 if (tp->t_state <= TCPS_CLOSED)
984263bc
MD
822 goto drop;
823
824 /* Unscale the window into a 32-bit value. */
95b22adf 825 if (!(thflags & TH_SYN))
984263bc
MD
826 tiwin = th->th_win << tp->snd_scale;
827 else
828 tiwin = th->th_win;
829
830 so = inp->inp_socket;
47654766 831
984263bc 832#ifdef TCPDEBUG
47654766
JH
833 if (so->so_options & SO_DEBUG) {
834 ostate = tp->t_state;
835 if (isipv6)
95b22adf 836 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
47654766 837 else
95b22adf 838 bcopy(ip, tcp_saveipgen, sizeof(*ip));
47654766
JH
839 tcp_savetcp = *th;
840 }
984263bc 841#endif
47654766 842
95b22adf 843 bzero(&to, sizeof(to));
d24ce1dc 844
47654766
JH
845 if (so->so_options & SO_ACCEPTCONN) {
846 struct in_conninfo inc;
847
984263bc 848#ifdef INET6
d24ce1dc 849 inc.inc_isipv6 = (isipv6 == TRUE);
984263bc
MD
850#endif
851 if (isipv6) {
852 inc.inc6_faddr = ip6->ip6_src;
853 inc.inc6_laddr = ip6->ip6_dst;
854 inc.inc6_route.ro_rt = NULL; /* XXX */
855 } else {
856 inc.inc_faddr = ip->ip_src;
857 inc.inc_laddr = ip->ip_dst;
858 inc.inc_route.ro_rt = NULL; /* XXX */
859 }
860 inc.inc_fport = th->th_sport;
861 inc.inc_lport = th->th_dport;
862
61896e3c
JH
863 /*
864 * If the state is LISTEN then ignore segment if it contains
984263bc
MD
865 * a RST. If the segment contains an ACK then it is bad and
866 * send a RST. If it does not contain a SYN then it is not
867 * interesting; drop it.
868 *
869 * If the state is SYN_RECEIVED (syncache) and seg contains
870 * an ACK, but not for our SYN/ACK, send a RST. If the seg
871 * contains a RST, check the sequence number to see if it
872 * is a valid reset segment.
873 */
61896e3c
JH
874 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
875 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
984263bc
MD
876 if (!syncache_expand(&inc, th, &so, m)) {
877 /*
878 * No syncache entry, or ACK was not
879 * for our SYN/ACK. Send a RST.
880 */
881 tcpstat.tcps_badsyn++;
882 rstreason = BANDLIM_RST_OPENPORT;
883 goto dropwithreset;
884 }
885 if (so == NULL)
886 /*
887 * Could not complete 3-way handshake,
888 * connection is being closed down, and
889 * syncache will free mbuf.
890 */
891 return;
892 /*
893 * Socket is created in state SYN_RECEIVED.
894 * Continue processing segment.
895 */
896 inp = sotoinpcb(so);
897 tp = intotcpcb(inp);
898 /*
899 * This is what would have happened in
900 * tcp_output() when the SYN,ACK was sent.
901 */
902 tp->snd_up = tp->snd_una;
903 tp->snd_max = tp->snd_nxt = tp->iss + 1;
904 tp->last_ack_sent = tp->rcv_nxt;
905/*
906 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
907 * until the _second_ ACK is received:
908 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
909 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
61896e3c 910 * move to ESTAB, set snd_wnd to tiwin.
d24ce1dc 911 */
984263bc
MD
912 tp->snd_wnd = tiwin; /* unscaled */
913 goto after_listen;
914 }
915 if (thflags & TH_RST) {
916 syncache_chkrst(&inc, th);
917 goto drop;
918 }
919 if (thflags & TH_ACK) {
920 syncache_badack(&inc);
921 tcpstat.tcps_badsyn++;
922 rstreason = BANDLIM_RST_OPENPORT;
923 goto dropwithreset;
924 }
925 goto drop;
926 }
927
928 /*
61896e3c 929 * Segment's flags are (SYN) or (SYN | FIN).
984263bc
MD
930 */
931#ifdef INET6
932 /*
933 * If deprecated address is forbidden,
934 * we do not accept SYN to deprecated interface
935 * address to prevent any new inbound connection from
936 * getting established.
937 * When we do not accept SYN, we send a TCP RST,
938 * with deprecated source address (instead of dropping
939 * it). We compromise it as it is much better for peer
940 * to send a RST, and RST will be the final packet
941 * for the exchange.
942 *
943 * If we do not forbid deprecated addresses, we accept
944 * the SYN packet. RFC2462 does not suggest dropping
945 * SYN in this case.
946 * If we decipher RFC2462 5.5.4, it says like this:
947 * 1. use of deprecated addr with existing
948 * communication is okay - "SHOULD continue to be
949 * used"
950 * 2. use of it with new communication:
951 * (2a) "SHOULD NOT be used if alternate address
61896e3c 952 * with sufficient scope is available"
984263bc
MD
953 * (2b) nothing mentioned otherwise.
954 * Here we fall into (2b) case as we have no choice in
955 * our source address selection - we must obey the peer.
956 *
957 * The wording in RFC2462 is confusing, and there are
958 * multiple description text for deprecated address
959 * handling - worse, they are not exactly the same.
960 * I believe 5.5.4 is the best one, so we follow 5.5.4.
961 */
962 if (isipv6 && !ip6_use_deprecated) {
963 struct in6_ifaddr *ia6;
964
965 if ((ia6 = ip6_getdstifaddr(m)) &&
966 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
967 tp = NULL;
968 rstreason = BANDLIM_RST_OPENPORT;
969 goto dropwithreset;
970 }
971 }
972#endif
973 /*
974 * If it is from this socket, drop it, it must be forged.
975 * Don't bother responding if the destination was a broadcast.
976 */
977 if (th->th_dport == th->th_sport) {
978 if (isipv6) {
979 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
980 &ip6->ip6_src))
981 goto drop;
982 } else {
983 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
984 goto drop;
985 }
986 }
987 /*
988 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
989 *
990 * Note that it is quite possible to receive unicast
991 * link-layer packets with a broadcast IP address. Use
992 * in_broadcast() to find them.
993 */
61896e3c 994 if (m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
995 goto drop;
996 if (isipv6) {
997 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
998 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
999 goto drop;
1000 } else {
1001 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1002 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1003 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1004 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1005 goto drop;
1006 }
1007 /*
1008 * SYN appears to be valid; create compressed TCP state
1009 * for syncache, or perform t/tcp connection.
1010 */
1011 if (so->so_qlen <= so->so_qlimit) {
95b22adf 1012 tcp_dooptions(&to, optp, optlen, TRUE);
984263bc
MD
1013 if (!syncache_add(&inc, &to, th, &so, m))
1014 goto drop;
1015 if (so == NULL)
1016 /*
1017 * Entry added to syncache, mbuf used to
1018 * send SYN,ACK packet.
1019 */
1020 return;
1021 /*
1022 * Segment passed TAO tests.
1023 */
1024 inp = sotoinpcb(so);
1025 tp = intotcpcb(inp);
1026 tp->snd_wnd = tiwin;
1027 tp->t_starttime = ticks;
1028 tp->t_state = TCPS_ESTABLISHED;
1029
1030 /*
1031 * If there is a FIN, or if there is data and the
1032 * connection is local, then delay SYN,ACK(SYN) in
1033 * the hope of piggy-backing it on a response
1034 * segment. Otherwise must send ACK now in case
1035 * the other side is slow starting.
1036 */
1037 if (DELAY_ACK(tp) &&
1038 ((thflags & TH_FIN) ||
1039 (tlen != 0 &&
1040 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1041 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
d24ce1dc
JH
1042 callout_reset(tp->tt_delack, tcp_delacktime,
1043 tcp_timer_delack, tp);
984263bc 1044 tp->t_flags |= TF_NEEDSYN;
d24ce1dc 1045 } else
984263bc
MD
1046 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1047
1048 tcpstat.tcps_connects++;
1049 soisconnected(so);
1050 goto trimthenstep6;
1051 }
1052 goto drop;
1053 }
1054after_listen:
1055
984263bc 1056 /* should not happen - syncache should pick up these connections */
61896e3c 1057 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
984263bc
MD
1058
1059 /*
1060 * Segment received on connection.
1061 * Reset idle time and keep-alive timer.
1062 */
1063 tp->t_rcvtime = ticks;
1064 if (TCPS_HAVEESTABLISHED(tp->t_state))
1065 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1066
1067 /*
1068 * Process options.
1069 * XXX this is tradtitional behavior, may need to be cleaned up.
1070 */
95b22adf 1071 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
984263bc
MD
1072 if (thflags & TH_SYN) {
1073 if (to.to_flags & TOF_SCALE) {
1074 tp->t_flags |= TF_RCVD_SCALE;
1075 tp->requested_s_scale = to.to_requested_s_scale;
1076 }
1077 if (to.to_flags & TOF_TS) {
1078 tp->t_flags |= TF_RCVD_TSTMP;
1079 tp->ts_recent = to.to_tsval;
1080 tp->ts_recent_age = ticks;
1081 }
61896e3c 1082 if (to.to_flags & (TOF_CC | TOF_CCNEW))
984263bc
MD
1083 tp->t_flags |= TF_RCVD_CC;
1084 if (to.to_flags & TOF_MSS)
1085 tcp_mss(tp, to.to_mss);
91489f6b
JH
1086 /*
1087 * Only set the TF_SACK_PERMITTED per-connection flag
1088 * if we got a SACK_PERMITTED option from the other side
1089 * and the global tcp_do_sack variable is true.
1090 */
1091 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1092 tp->t_flags |= TF_SACK_PERMITTED;
984263bc
MD
1093 }
1094
1095 /*
1096 * Header prediction: check for the two common cases
1097 * of a uni-directional data xfer. If the packet has
1098 * no control flags, is in-sequence, the window didn't
1099 * change and we're not retransmitting, it's a
1100 * candidate. If the length is zero and the ack moved
1101 * forward, we're the sender side of the xfer. Just
1102 * free the data acked & wake any higher level process
1103 * that was blocked waiting for space. If the length
1104 * is non-zero and the ack didn't move, we're the
1105 * receiver side. If we're getting packets in-order
1106 * (the reassembly queue is empty), add the data to
1107 * the socket buffer and note that we need a delayed ack.
1108 * Make sure that the hidden state-flags are also off.
1109 * Since we check for TCPS_ESTABLISHED above, it can only
1110 * be TH_NEEDSYN.
1111 */
1112 if (tp->t_state == TCPS_ESTABLISHED &&
1113 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
61896e3c
JH
1114 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1115 (!(to.to_flags & TOF_TS) ||
984263bc
MD
1116 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1117 /*
1118 * Using the CC option is compulsory if once started:
1119 * the segment is OK if no T/TCP was negotiated or
1120 * if the segment has a CC option equal to CCrecv
1121 */
1122 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
61896e3c 1123 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
984263bc 1124 th->th_seq == tp->rcv_nxt &&
984263bc
MD
1125 tp->snd_nxt == tp->snd_max) {
1126
1127 /*
1128 * If last ACK falls within this segment's sequence numbers,
1129 * record the timestamp.
1130 * NOTE that the test is modified according to the latest
1131 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1132 */
61896e3c 1133 if ((to.to_flags & TOF_TS) &&
984263bc
MD
1134 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1135 tp->ts_recent_age = ticks;
1136 tp->ts_recent = to.to_tsval;
1137 }
1138
1139 if (tlen == 0) {
1140 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1141 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1142 tp->snd_cwnd >= tp->snd_wnd &&
95b22adf 1143 !IN_FASTRECOVERY(tp)) {
984263bc
MD
1144 /*
1145 * this is a pure ack for outstanding data.
1146 */
1147 ++tcpstat.tcps_predack;
1148 /*
1149 * "bad retransmit" recovery
bfdb979e
JH
1150 *
1151 * If Eifel detection applies, then
1152 * it is deterministic, so use it
1153 * unconditionally over the old heuristic.
1154 * Otherwise, fall back to the old heuristic.
984263bc 1155 */
bfdb979e
JH
1156 if (tcp_do_eifel_detect &&
1157 (to.to_flags & TOF_TS) && to.to_tsecr &&
1158 (tp->t_flags & TF_FIRSTACCACK)) {
1159 /* Eifel detection applicable. */
1160 if (to.to_tsecr < tp->t_rexmtTS) {
1161 tcp_revert_congestion_state(tp);
1162 ++tcpstat.tcps_eifeldetected;
1163 }
1164 } else if (tp->t_rxtshift == 1 &&
1165 ticks < tp->t_badrxtwin) {
1166 tcp_revert_congestion_state(tp);
1167 ++tcpstat.tcps_rttdetected;
984263bc 1168 }
8819433a
JH
1169 tp->t_flags &= ~(TF_FIRSTACCACK |
1170 TF_FASTREXMT | TF_EARLYREXMT);
984263bc
MD
1171 /*
1172 * Recalculate the retransmit timer / rtt.
1173 *
d24ce1dc 1174 * Some machines (certain windows boxes)
984263bc
MD
1175 * send broken timestamp replies during the
1176 * SYN+ACK phase, ignore timestamps of 0.
1177 */
95b22adf 1178 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
984263bc 1179 tcp_xmit_timer(tp,
95b22adf 1180 ticks - to.to_tsecr + 1);
984263bc 1181 } else if (tp->t_rtttime &&
95b22adf 1182 SEQ_GT(th->th_ack, tp->t_rtseq)) {
984263bc
MD
1183 tcp_xmit_timer(tp,
1184 ticks - tp->t_rtttime);
1185 }
1186 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1187 acked = th->th_ack - tp->snd_una;
1188 tcpstat.tcps_rcvackpack++;
1189 tcpstat.tcps_rcvackbyte += acked;
1190 sbdrop(&so->so_snd, acked);
cfb3f3f4 1191 tp->snd_recover = th->th_ack - 1;
9845754e 1192 tp->snd_una = th->th_ack;
984263bc 1193 tp->t_dupacks = 0;
df9d7670
JH
1194 /*
1195 * Update window information.
1196 */
1197 if (tiwin != tp->snd_wnd &&
1198 acceptable_window_update(tp, th, tiwin)) {
1199 /* keep track of pure window updates */
1200 if (tp->snd_wl2 == th->th_ack &&
1201 tiwin > tp->snd_wnd)
1202 tcpstat.tcps_rcvwinupd++;
1203 tp->snd_wnd = tiwin;
1204 tp->snd_wl1 = th->th_seq;
1205 tp->snd_wl2 = th->th_ack;
1206 if (tp->snd_wnd > tp->max_sndwnd)
1207 tp->max_sndwnd = tp->snd_wnd;
1208 }
984263bc
MD
1209 m_freem(m);
1210 ND6_HINT(tp); /* some progress has been done */
984263bc
MD
1211 /*
1212 * If all outstanding data are acked, stop
1213 * retransmit timer, otherwise restart timer
1214 * using current (possibly backed-off) value.
1215 * If process is waiting for space,
1216 * wakeup/selwakeup/signal. If data
1217 * are ready to send, let tcp_output
1218 * decide between more output or persist.
1219 */
1220 if (tp->snd_una == tp->snd_max)
1221 callout_stop(tp->tt_rexmt);
1222 else if (!callout_active(tp->tt_persist))
d24ce1dc 1223 callout_reset(tp->tt_rexmt,
984263bc
MD
1224 tp->t_rxtcur,
1225 tcp_timer_rexmt, tp);
1226
1227 sowwakeup(so);
1228 if (so->so_snd.sb_cc)
f23061d4 1229 tcp_output(tp);
984263bc
MD
1230 return;
1231 }
df9d7670
JH
1232 } else if (tiwin == tp->snd_wnd &&
1233 th->th_ack == tp->snd_una &&
984263bc
MD
1234 LIST_EMPTY(&tp->t_segq) &&
1235 tlen <= sbspace(&so->so_rcv)) {
1236 /*
1237 * this is a pure, in-sequence data packet
1238 * with nothing on the reassembly queue and
1239 * we have enough buffer space to take it.
1240 */
1241 ++tcpstat.tcps_preddat;
1242 tp->rcv_nxt += tlen;
1243 tcpstat.tcps_rcvpack++;
1244 tcpstat.tcps_rcvbyte += tlen;
1245 ND6_HINT(tp); /* some progress has been done */
1246 /*
1247 * Add data to socket buffer.
1248 */
1249 if (so->so_state & SS_CANTRCVMORE) {
1250 m_freem(m);
1251 } else {
09cf73e0 1252 m_adj(m, drop_hdrlen); /* delayed header drop */
507106e0 1253 sbappendstream(&so->so_rcv, m);
984263bc
MD
1254 }
1255 sorwakeup(so);
2b1ce38a
MD
1256
1257 /*
1258 * This code is responsible for most of the ACKs
1259 * the TCP stack sends back after receiving a data
1260 * packet. Note that the DELAY_ACK check fails if
1261 * the delack timer is already running, which results
1262 * in an ack being sent every other packet (which is
1263 * what we want).
1264 */
984263bc 1265 if (DELAY_ACK(tp)) {
61896e3c
JH
1266 callout_reset(tp->tt_delack, tcp_delacktime,
1267 tcp_timer_delack, tp);
984263bc
MD
1268 } else {
1269 tp->t_flags |= TF_ACKNOW;
61896e3c 1270 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
2b1ce38a
MD
1271 tp->t_flags |= TF_ONOUTPUTQ;
1272 tp->tt_cpu = mycpu->gd_cpuid;
1273 TAILQ_INSERT_TAIL(
1274 &tcpcbackq[tp->tt_cpu],
1275 tp, t_outputq);
1276 }
984263bc
MD
1277 }
1278 return;
1279 }
1280 }
1281
1282 /*
1283 * Calculate amount of space in receive window,
1284 * and then do TCP input processing.
1285 * Receive window is amount of space in rcv queue,
1286 * but not less than advertised window.
1287 */
61896e3c
JH
1288 recvwin = sbspace(&so->so_rcv);
1289 if (recvwin < 0)
1290 recvwin = 0;
1291 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
984263bc
MD
1292
1293 switch (tp->t_state) {
984263bc
MD
1294 /*
1295 * If the state is SYN_RECEIVED:
1296 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1297 */
1298 case TCPS_SYN_RECEIVED:
1299 if ((thflags & TH_ACK) &&
1300 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1301 SEQ_GT(th->th_ack, tp->snd_max))) {
1302 rstreason = BANDLIM_RST_OPENPORT;
1303 goto dropwithreset;
1304 }
1305 break;
1306
1307 /*
1308 * If the state is SYN_SENT:
1309 * if seg contains an ACK, but not for our SYN, drop the input.
1310 * if seg contains a RST, then drop the connection.
1311 * if seg does not contain SYN, then drop it.
1312 * Otherwise this is an acceptable SYN segment
1313 * initialize tp->rcv_nxt and tp->irs
1314 * if seg contains ack then advance tp->snd_una
1315 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1316 * arrange for segment to be acked (eventually)
1317 * continue processing rest of data/controls, beginning with URG
1318 */
1319 case TCPS_SYN_SENT:
1320 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1321 taop = &tao_noncached;
1322 bzero(taop, sizeof(*taop));
1323 }
1324
1325 if ((thflags & TH_ACK) &&
1326 (SEQ_LEQ(th->th_ack, tp->iss) ||
1327 SEQ_GT(th->th_ack, tp->snd_max))) {
1328 /*
1329 * If we have a cached CCsent for the remote host,
1330 * hence we haven't just crashed and restarted,
1331 * do not send a RST. This may be a retransmission
1332 * from the other side after our earlier ACK was lost.
1333 * Our new SYN, when it arrives, will serve as the
1334 * needed ACK.
1335 */
1336 if (taop->tao_ccsent != 0)
1337 goto drop;
1338 else {
1339 rstreason = BANDLIM_UNLIMITED;
1340 goto dropwithreset;
1341 }
1342 }
1343 if (thflags & TH_RST) {
1344 if (thflags & TH_ACK)
1345 tp = tcp_drop(tp, ECONNREFUSED);
1346 goto drop;
1347 }
61896e3c 1348 if (!(thflags & TH_SYN))
984263bc
MD
1349 goto drop;
1350 tp->snd_wnd = th->th_win; /* initial send window */
1351 tp->cc_recv = to.to_cc; /* foreign CC */
1352
1353 tp->irs = th->th_seq;
1354 tcp_rcvseqinit(tp);
1355 if (thflags & TH_ACK) {
1356 /*
1357 * Our SYN was acked. If segment contains CC.ECHO
1358 * option, check it to make sure this segment really
1359 * matches our SYN. If not, just drop it as old
1360 * duplicate, but send an RST if we're still playing
1361 * by the old rules. If no CC.ECHO option, make sure
1362 * we don't get fooled into using T/TCP.
1363 */
1364 if (to.to_flags & TOF_CCECHO) {
1365 if (tp->cc_send != to.to_ccecho) {
1366 if (taop->tao_ccsent != 0)
1367 goto drop;
1368 else {
1369 rstreason = BANDLIM_UNLIMITED;
1370 goto dropwithreset;
1371 }
1372 }
1373 } else
1374 tp->t_flags &= ~TF_RCVD_CC;
1375 tcpstat.tcps_connects++;
1376 soisconnected(so);
1377 /* Do window scaling on this connection? */
61896e3c 1378 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
b5572302 1379 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
1380 tp->snd_scale = tp->requested_s_scale;
1381 tp->rcv_scale = tp->request_r_scale;
1382 }
1383 /* Segment is acceptable, update cache if undefined. */
1384 if (taop->tao_ccsent == 0)
1385 taop->tao_ccsent = to.to_ccecho;
1386
1387 tp->rcv_adv += tp->rcv_wnd;
1388 tp->snd_una++; /* SYN is acked */
91489f6b 1389 callout_stop(tp->tt_rexmt);
984263bc
MD
1390 /*
1391 * If there's data, delay ACK; if there's also a FIN
1392 * ACKNOW will be turned on later.
1393 */
1394 if (DELAY_ACK(tp) && tlen != 0)
61896e3c
JH
1395 callout_reset(tp->tt_delack, tcp_delacktime,
1396 tcp_timer_delack, tp);
984263bc
MD
1397 else
1398 tp->t_flags |= TF_ACKNOW;
1399 /*
1400 * Received <SYN,ACK> in SYN_SENT[*] state.
1401 * Transitions:
1402 * SYN_SENT --> ESTABLISHED
1403 * SYN_SENT* --> FIN_WAIT_1
1404 */
1405 tp->t_starttime = ticks;
1406 if (tp->t_flags & TF_NEEDFIN) {
1407 tp->t_state = TCPS_FIN_WAIT_1;
1408 tp->t_flags &= ~TF_NEEDFIN;
1409 thflags &= ~TH_SYN;
1410 } else {
1411 tp->t_state = TCPS_ESTABLISHED;
1412 callout_reset(tp->tt_keep, tcp_keepidle,
1413 tcp_timer_keep, tp);
1414 }
1415 } else {
1416 /*
95b22adf
JH
1417 * Received initial SYN in SYN-SENT[*] state =>
1418 * simultaneous open. If segment contains CC option
1419 * and there is a cached CC, apply TAO test.
1420 * If it succeeds, connection is * half-synchronized.
1421 * Otherwise, do 3-way handshake:
61896e3c
JH
1422 * SYN-SENT -> SYN-RECEIVED
1423 * SYN-SENT* -> SYN-RECEIVED*
95b22adf
JH
1424 * If there was no CC option, clear cached CC value.
1425 */
984263bc
MD
1426 tp->t_flags |= TF_ACKNOW;
1427 callout_stop(tp->tt_rexmt);
1428 if (to.to_flags & TOF_CC) {
1429 if (taop->tao_cc != 0 &&
1430 CC_GT(to.to_cc, taop->tao_cc)) {
1431 /*
1432 * update cache and make transition:
61896e3c
JH
1433 * SYN-SENT -> ESTABLISHED*
1434 * SYN-SENT* -> FIN-WAIT-1*
984263bc
MD
1435 */
1436 taop->tao_cc = to.to_cc;
1437 tp->t_starttime = ticks;
1438 if (tp->t_flags & TF_NEEDFIN) {
1439 tp->t_state = TCPS_FIN_WAIT_1;
1440 tp->t_flags &= ~TF_NEEDFIN;
1441 } else {
1442 tp->t_state = TCPS_ESTABLISHED;
1443 callout_reset(tp->tt_keep,
1444 tcp_keepidle,
1445 tcp_timer_keep,
1446 tp);
1447 }
1448 tp->t_flags |= TF_NEEDSYN;
1449 } else
1450 tp->t_state = TCPS_SYN_RECEIVED;
1451 } else {
1452 /* CC.NEW or no option => invalidate cache */
1453 taop->tao_cc = 0;
1454 tp->t_state = TCPS_SYN_RECEIVED;
1455 }
1456 }
1457
1458trimthenstep6:
1459 /*
1460 * Advance th->th_seq to correspond to first data byte.
1461 * If data, trim to stay within window,
1462 * dropping FIN if necessary.
1463 */
1464 th->th_seq++;
1465 if (tlen > tp->rcv_wnd) {
1466 todrop = tlen - tp->rcv_wnd;
1467 m_adj(m, -todrop);
1468 tlen = tp->rcv_wnd;
1469 thflags &= ~TH_FIN;
1470 tcpstat.tcps_rcvpackafterwin++;
1471 tcpstat.tcps_rcvbyteafterwin += todrop;
1472 }
1473 tp->snd_wl1 = th->th_seq - 1;
1474 tp->rcv_up = th->th_seq;
1475 /*
1476 * Client side of transaction: already sent SYN and data.
1477 * If the remote host used T/TCP to validate the SYN,
1478 * our data will be ACK'd; if so, enter normal data segment
1479 * processing in the middle of step 5, ack processing.
1480 * Otherwise, goto step 6.
1481 */
95b22adf 1482 if (thflags & TH_ACK)
984263bc
MD
1483 goto process_ACK;
1484
1485 goto step6;
1486
1487 /*
1488 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1489 * if segment contains a SYN and CC [not CC.NEW] option:
61896e3c
JH
1490 * if state == TIME_WAIT and connection duration > MSL,
1491 * drop packet and send RST;
984263bc
MD
1492 *
1493 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1494 * ack the FIN (and data) in retransmission queue.
61896e3c
JH
1495 * Complete close and delete TCPCB. Then reprocess
1496 * segment, hoping to find new TCPCB in LISTEN state;
984263bc
MD
1497 *
1498 * else must be old SYN; drop it.
61896e3c 1499 * else do normal processing.
984263bc
MD
1500 */
1501 case TCPS_LAST_ACK:
1502 case TCPS_CLOSING:
1503 case TCPS_TIME_WAIT:
1504 if ((thflags & TH_SYN) &&
1505 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1506 if (tp->t_state == TCPS_TIME_WAIT &&
1507 (ticks - tp->t_starttime) > tcp_msl) {
1508 rstreason = BANDLIM_UNLIMITED;
1509 goto dropwithreset;
1510 }
1511 if (CC_GT(to.to_cc, tp->cc_recv)) {
1512 tp = tcp_close(tp);
1513 goto findpcb;
1514 }
1515 else
1516 goto drop;
1517 }
95b22adf 1518 break; /* continue normal processing */
984263bc
MD
1519 }
1520
1521 /*
1522 * States other than LISTEN or SYN_SENT.
1523 * First check the RST flag and sequence number since reset segments
1524 * are exempt from the timestamp and connection count tests. This
1525 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1526 * below which allowed reset segments in half the sequence space
1527 * to fall though and be processed (which gives forged reset
1528 * segments with a random sequence number a 50 percent chance of
1529 * killing a connection).
1530 * Then check timestamp, if present.
1531 * Then check the connection count, if present.
1532 * Then check that at least some bytes of segment are within
1533 * receive window. If segment begins before rcv_nxt,
1534 * drop leading data (and SYN); if nothing left, just ack.
1535 *
1536 *
1537 * If the RST bit is set, check the sequence number to see
1538 * if this is a valid reset segment.
1539 * RFC 793 page 37:
1540 * In all states except SYN-SENT, all reset (RST) segments
1541 * are validated by checking their SEQ-fields. A reset is
1542 * valid if its sequence number is in the window.
1543 * Note: this does not take into account delayed ACKs, so
1544 * we should test against last_ack_sent instead of rcv_nxt.
1545 * The sequence number in the reset segment is normally an
91489f6b 1546 * echo of our outgoing acknowledgement numbers, but some hosts
984263bc
MD
1547 * send a reset with the sequence number at the rightmost edge
1548 * of our receive window, and we have to handle this case.
1549 * If we have multiple segments in flight, the intial reset
1550 * segment sequence numbers will be to the left of last_ack_sent,
1551 * but they will eventually catch up.
1552 * In any case, it never made sense to trim reset segments to
1553 * fit the receive window since RFC 1122 says:
1554 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1555 *
1556 * A TCP SHOULD allow a received RST segment to include data.
1557 *
1558 * DISCUSSION
61896e3c
JH
1559 * It has been suggested that a RST segment could contain
1560 * ASCII text that encoded and explained the cause of the
91489f6b 1561 * RST. No standard has yet been established for such
61896e3c 1562 * data.
984263bc
MD
1563 *
1564 * If the reset segment passes the sequence number test examine
1565 * the state:
1566 * SYN_RECEIVED STATE:
1567 * If passive open, return to LISTEN state.
1568 * If active open, inform user that connection was refused.
1569 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1570 * Inform user that connection was reset, and close tcb.
1571 * CLOSING, LAST_ACK STATES:
1572 * Close the tcb.
1573 * TIME_WAIT STATE:
1574 * Drop the segment - see Stevens, vol. 2, p. 964 and
61896e3c 1575 * RFC 1337.
984263bc
MD
1576 */
1577 if (thflags & TH_RST) {
1578 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
d4dbb5be 1579 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
984263bc
MD
1580 switch (tp->t_state) {
1581
1582 case TCPS_SYN_RECEIVED:
1583 so->so_error = ECONNREFUSED;
1584 goto close;
1585
1586 case TCPS_ESTABLISHED:
1587 case TCPS_FIN_WAIT_1:
1588 case TCPS_FIN_WAIT_2:
1589 case TCPS_CLOSE_WAIT:
1590 so->so_error = ECONNRESET;
1591 close:
1592 tp->t_state = TCPS_CLOSED;
1593 tcpstat.tcps_drops++;
1594 tp = tcp_close(tp);
1595 break;
1596
1597 case TCPS_CLOSING:
1598 case TCPS_LAST_ACK:
1599 tp = tcp_close(tp);
1600 break;
1601
1602 case TCPS_TIME_WAIT:
1603 break;
1604 }
1605 }
1606 goto drop;
1607 }
1608
1609 /*
1610 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1611 * and it's less than ts_recent, drop it.
1612 */
61896e3c 1613 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
984263bc
MD
1614 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1615
1616 /* Check to see if ts_recent is over 24 days old. */
1617 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1618 /*
1619 * Invalidate ts_recent. If this segment updates
1620 * ts_recent, the age will be reset later and ts_recent
1621 * will get a valid value. If it does not, setting
1622 * ts_recent to zero will at least satisfy the
1623 * requirement that zero be placed in the timestamp
1624 * echo reply when ts_recent isn't valid. The
1625 * age isn't reset until we get a valid ts_recent
1626 * because we don't want out-of-order segments to be
1627 * dropped when ts_recent is old.
1628 */
1629 tp->ts_recent = 0;
1630 } else {
1631 tcpstat.tcps_rcvduppack++;
1632 tcpstat.tcps_rcvdupbyte += tlen;
1633 tcpstat.tcps_pawsdrop++;
1634 if (tlen)
1635 goto dropafterack;
1636 goto drop;
1637 }
1638 }
1639
1640 /*
1641 * T/TCP mechanism
1642 * If T/TCP was negotiated and the segment doesn't have CC,
1643 * or if its CC is wrong then drop the segment.
1644 * RST segments do not have to comply with this.
1645 */
1646 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
61896e3c 1647 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
95b22adf 1648 goto dropafterack;
984263bc
MD
1649
1650 /*
1651 * In the SYN-RECEIVED state, validate that the packet belongs to
1652 * this connection before trimming the data to fit the receive
1653 * window. Check the sequence number versus IRS since we know
1654 * the sequence numbers haven't wrapped. This is a partial fix
1655 * for the "LAND" DoS attack.
1656 */
1657 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1658 rstreason = BANDLIM_RST_OPENPORT;
1659 goto dropwithreset;
1660 }
1661
1662 todrop = tp->rcv_nxt - th->th_seq;
1663 if (todrop > 0) {
91489f6b
JH
1664 if (TCP_DO_SACK(tp)) {
1665 /* Report duplicate segment at head of packet. */
1666 tp->reportblk.rblk_start = th->th_seq;
1667 tp->reportblk.rblk_end = th->th_seq + tlen;
1668 if (thflags & TH_FIN)
1669 ++tp->reportblk.rblk_end;
1670 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1671 tp->reportblk.rblk_end = tp->rcv_nxt;
1672 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1673 }
984263bc
MD
1674 if (thflags & TH_SYN) {
1675 thflags &= ~TH_SYN;
1676 th->th_seq++;
1677 if (th->th_urp > 1)
1678 th->th_urp--;
1679 else
1680 thflags &= ~TH_URG;
1681 todrop--;
1682 }
1683 /*
1684 * Following if statement from Stevens, vol. 2, p. 960.
1685 */
61896e3c
JH
1686 if (todrop > tlen ||
1687 (todrop == tlen && !(thflags & TH_FIN))) {
984263bc
MD
1688 /*
1689 * Any valid FIN must be to the left of the window.
1690 * At this point the FIN must be a duplicate or out
1691 * of sequence; drop it.
1692 */
1693 thflags &= ~TH_FIN;
1694
1695 /*
1696 * Send an ACK to resynchronize and drop any data.
1697 * But keep on processing for RST or ACK.
1698 */
1699 tp->t_flags |= TF_ACKNOW;
1700 todrop = tlen;
1701 tcpstat.tcps_rcvduppack++;
1702 tcpstat.tcps_rcvdupbyte += todrop;
1703 } else {
1704 tcpstat.tcps_rcvpartduppack++;
1705 tcpstat.tcps_rcvpartdupbyte += todrop;
1706 }
1707 drop_hdrlen += todrop; /* drop from the top afterwards */
1708 th->th_seq += todrop;
1709 tlen -= todrop;
1710 if (th->th_urp > todrop)
1711 th->th_urp -= todrop;
1712 else {
1713 thflags &= ~TH_URG;
1714 th->th_urp = 0;
1715 }
1716 }
1717
1718 /*
1719 * If new data are received on a connection after the
1720 * user processes are gone, then RST the other end.
1721 */
1722 if ((so->so_state & SS_NOFDREF) &&
1723 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1724 tp = tcp_close(tp);
1725 tcpstat.tcps_rcvafterclose++;
1726 rstreason = BANDLIM_UNLIMITED;
1727 goto dropwithreset;
1728 }
1729
1730 /*
1731 * If segment ends after window, drop trailing data
1732 * (and PUSH and FIN); if nothing left, just ACK.
1733 */
61896e3c 1734 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
984263bc
MD
1735 if (todrop > 0) {
1736 tcpstat.tcps_rcvpackafterwin++;
1737 if (todrop >= tlen) {
1738 tcpstat.tcps_rcvbyteafterwin += tlen;
1739 /*
1740 * If a new connection request is received
1741 * while in TIME_WAIT, drop the old connection
1742 * and start over if the sequence numbers
1743 * are above the previous ones.
1744 */
1745 if (thflags & TH_SYN &&
1746 tp->t_state == TCPS_TIME_WAIT &&
1747 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1748 tp = tcp_close(tp);
1749 goto findpcb;
1750 }
1751 /*
1752 * If window is closed can only take segments at
1753 * window edge, and have to drop data and PUSH from
1754 * incoming segments. Continue processing, but
1755 * remember to ack. Otherwise, drop segment
1756 * and ack.
1757 */
1758 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1759 tp->t_flags |= TF_ACKNOW;
1760 tcpstat.tcps_rcvwinprobe++;
1761 } else
1762 goto dropafterack;
1763 } else
1764 tcpstat.tcps_rcvbyteafterwin += todrop;
1765 m_adj(m, -todrop);
1766 tlen -= todrop;
61896e3c 1767 thflags &= ~(TH_PUSH | TH_FIN);
984263bc
MD
1768 }
1769
1770 /*
1771 * If last ACK falls within this segment's sequence numbers,
1772 * record its timestamp.
1773 * NOTE that the test is modified according to the latest
1774 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1775 */
61896e3c 1776 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
984263bc
MD
1777 tp->ts_recent_age = ticks;
1778 tp->ts_recent = to.to_tsval;
1779 }
1780
1781 /*
1782 * If a SYN is in the window, then this is an
1783 * error and we send an RST and drop the connection.
1784 */
1785 if (thflags & TH_SYN) {
1786 tp = tcp_drop(tp, ECONNRESET);
1787 rstreason = BANDLIM_UNLIMITED;
1788 goto dropwithreset;
1789 }
1790
1791 /*
1792 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1793 * flag is on (half-synchronized state), then queue data for
1794 * later processing; else drop segment and return.
1795 */
61896e3c 1796 if (!(thflags & TH_ACK)) {
984263bc
MD
1797 if (tp->t_state == TCPS_SYN_RECEIVED ||
1798 (tp->t_flags & TF_NEEDSYN))
1799 goto step6;
1800 else
1801 goto drop;
1802 }
1803
1804 /*
1805 * Ack processing.
1806 */
1807 switch (tp->t_state) {
984263bc 1808 /*
91489f6b 1809 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
984263bc
MD
1810 * ESTABLISHED state and continue processing.
1811 * The ACK was checked above.
1812 */
1813 case TCPS_SYN_RECEIVED:
1814
1815 tcpstat.tcps_connects++;
1816 soisconnected(so);
1817 /* Do window scaling? */
61896e3c 1818 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
b5572302 1819 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
1820 tp->snd_scale = tp->requested_s_scale;
1821 tp->rcv_scale = tp->request_r_scale;
1822 }
1823 /*
1824 * Upon successful completion of 3-way handshake,
1825 * update cache.CC if it was undefined, pass any queued
1826 * data to the user, and advance state appropriately.
1827 */
1828 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1829 taop->tao_cc == 0)
1830 taop->tao_cc = tp->cc_recv;
1831
1832 /*
1833 * Make transitions:
1834 * SYN-RECEIVED -> ESTABLISHED
1835 * SYN-RECEIVED* -> FIN-WAIT-1
1836 */
1837 tp->t_starttime = ticks;
1838 if (tp->t_flags & TF_NEEDFIN) {
1839 tp->t_state = TCPS_FIN_WAIT_1;
1840 tp->t_flags &= ~TF_NEEDFIN;
1841 } else {
1842 tp->t_state = TCPS_ESTABLISHED;
d24ce1dc 1843 callout_reset(tp->tt_keep, tcp_keepidle,
984263bc
MD
1844 tcp_timer_keep, tp);
1845 }
1846 /*
1847 * If segment contains data or ACK, will call tcp_reass()
1848 * later; if not, do so now to pass queued data to user.
1849 */
61896e3c 1850 if (tlen == 0 && !(thflags & TH_FIN))
f23061d4 1851 tcp_reass(tp, NULL, NULL, NULL);
984263bc
MD
1852 /* fall into ... */
1853
1854 /*
1855 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1856 * ACKs. If the ack is in the range
1857 * tp->snd_una < th->th_ack <= tp->snd_max
1858 * then advance tp->snd_una to th->th_ack and drop
1859 * data from the retransmission queue. If this ACK reflects
1860 * more up to date window information we update our window information.
1861 */
1862 case TCPS_ESTABLISHED:
1863 case TCPS_FIN_WAIT_1:
1864 case TCPS_FIN_WAIT_2:
1865 case TCPS_CLOSE_WAIT:
1866 case TCPS_CLOSING:
1867 case TCPS_LAST_ACK:
1868 case TCPS_TIME_WAIT:
1869
1870 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
91489f6b
JH
1871 if (TCP_DO_SACK(tp))
1872 tcp_sack_update_scoreboard(tp, &to);
95b22adf
JH
1873 if (tlen != 0 || tiwin != tp->snd_wnd) {
1874 tp->t_dupacks = 0;
1875 break;
1876 }
1877 tcpstat.tcps_rcvdupack++;
91489f6b
JH
1878 if (!callout_active(tp->tt_rexmt) ||
1879 th->th_ack != tp->snd_una) {
1880 tp->t_dupacks = 0;
1881 break;
1882 }
95b22adf 1883 /*
91489f6b 1884 * We have outstanding data (other than
95b22adf
JH
1885 * a window probe), this is a completely
1886 * duplicate ack (ie, window info didn't
1887 * change), the ack is the biggest we've
1888 * seen and we've seen exactly our rexmt
91489f6b 1889 * threshhold of them, so assume a packet
95b22adf
JH
1890 * has been dropped and retransmit it.
1891 * Kludge snd_nxt & the congestion
1892 * window so we send only this one
1893 * packet.
95b22adf 1894 */
91489f6b
JH
1895 if (IN_FASTRECOVERY(tp)) {
1896 if (TCP_DO_SACK(tp)) {
1897 /* No artifical cwnd inflation. */
1898 tcp_sack_rexmt(tp, th);
1899 } else {
1900 /*
1901 * Dup acks mean that packets
1902 * have left the network
1903 * (they're now cached at the
1904 * receiver) so bump cwnd by
1905 * the amount in the receiver
1906 * to keep a constant cwnd
1907 * packets in the network.
1908 */
1909 tp->snd_cwnd += tp->t_maxseg;
f23061d4 1910 tcp_output(tp);
91489f6b
JH
1911 }
1912 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
95b22adf
JH
1913 tp->t_dupacks = 0;
1914 break;
95b22adf 1915 } else if (++tp->t_dupacks == tcprexmtthresh) {
91489f6b 1916 tcp_seq old_snd_nxt;
95b22adf
JH
1917 u_int win;
1918
8819433a 1919fastretransmit:
95b22adf
JH
1920 if (tcp_do_eifel_detect &&
1921 (tp->t_flags & TF_RCVD_TSTMP)) {
1922 tcp_save_congestion_state(tp);
1923 tp->t_flags |= TF_FASTREXMT;
1924 }
91489f6b
JH
1925 /*
1926 * We know we're losing at the current
1927 * window size, so do congestion avoidance:
1928 * set ssthresh to half the current window
1929 * and pull our congestion window back to the
1930 * new ssthresh.
1931 */
95b22adf
JH
1932 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1933 tp->t_maxseg;
1934 if (win < 2)
1935 win = 2;
1936 tp->snd_ssthresh = win * tp->t_maxseg;
1937 ENTER_FASTRECOVERY(tp);
1938 tp->snd_recover = tp->snd_max;
1939 callout_stop(tp->tt_rexmt);
1940 tp->t_rtttime = 0;
91489f6b 1941 old_snd_nxt = tp->snd_nxt;
95b22adf
JH
1942 tp->snd_nxt = th->th_ack;
1943 tp->snd_cwnd = tp->t_maxseg;
f23061d4 1944 tcp_output(tp);
95b22adf 1945 ++tcpstat.tcps_sndfastrexmit;
91489f6b
JH
1946 tp->snd_cwnd = tp->snd_ssthresh;
1947 tp->rexmt_high = tp->snd_nxt;
1948 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1949 tp->snd_nxt = old_snd_nxt;
95b22adf
JH
1950 KASSERT(tp->snd_limited <= 2,
1951 ("tp->snd_limited too big"));
91489f6b
JH
1952 if (TCP_DO_SACK(tp))
1953 tcp_sack_rexmt(tp, th);
1954 else
1955 tp->snd_cwnd += tp->t_maxseg *
1956 (tp->t_dupacks - tp->snd_limited);
95b22adf
JH
1957 } else if (tcp_do_limitedtransmit) {
1958 u_long oldcwnd = tp->snd_cwnd;
1959 tcp_seq oldsndmax = tp->snd_max;
1960 /* outstanding data */
1961 uint32_t ownd = tp->snd_max - tp->snd_una;
1962 u_int sent;
ca94b4a6 1963
8819433a
JH
1964#define iceildiv(n, d) (((n)+(d)-1) / (d))
1965
95b22adf
JH
1966 KASSERT(tp->t_dupacks == 1 ||
1967 tp->t_dupacks == 2,
1968 ("dupacks not 1 or 2"));
1969 if (tp->t_dupacks == 1)
1970 tp->snd_limited = 0;
1971 tp->snd_cwnd = ownd +
1972 (tp->t_dupacks - tp->snd_limited) *
1973 tp->t_maxseg;
f23061d4 1974 tcp_output(tp);
95b22adf
JH
1975 tp->snd_cwnd = oldcwnd;
1976 sent = tp->snd_max - oldsndmax;
1977 if (sent > tp->t_maxseg) {
1978 KASSERT((tp->t_dupacks == 2 &&
1979 tp->snd_limited == 0) ||
1980 (sent == tp->t_maxseg + 1 &&
1981 tp->t_flags & TF_SENTFIN),
1982 ("sent too much"));
1983 KASSERT(sent <= tp->t_maxseg * 2,
1984 ("sent too many segments"));
1985 tp->snd_limited = 2;
1986 tcpstat.tcps_sndlimited += 2;
1987 } else if (sent > 0) {
1988 ++tp->snd_limited;
1989 ++tcpstat.tcps_sndlimited;
1990 } else if (tcp_do_early_retransmit &&
1991 (tcp_do_eifel_detect &&
1992 (tp->t_flags & TF_RCVD_TSTMP)) &&
91489f6b 1993 ownd < 4 * tp->t_maxseg &&
95b22adf 1994 tp->t_dupacks + 1 >=
91489f6b
JH
1995 iceildiv(ownd, tp->t_maxseg) &&
1996 (!TCP_DO_SACK(tp) ||
f23061d4 1997 ownd <= tp->t_maxseg ||
91489f6b 1998 tcp_sack_has_sacked(&tp->scb,
f23061d4 1999 ownd - tp->t_maxseg))) {
95b22adf
JH
2000 ++tcpstat.tcps_sndearlyrexmit;
2001 tp->t_flags |= TF_EARLYREXMT;
2002 goto fastretransmit;
984263bc 2003 }
95b22adf
JH
2004 }
2005 goto drop;
984263bc
MD
2006 }
2007
2008 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
984263bc
MD
2009 tp->t_dupacks = 0;
2010 if (SEQ_GT(th->th_ack, tp->snd_max)) {
5a274421
JH
2011 /*
2012 * Detected optimistic ACK attack.
2013 * Force slow-start to de-synchronize attack.
2014 */
2015 tp->snd_cwnd = tp->t_maxseg;
2016
984263bc
MD
2017 tcpstat.tcps_rcvacktoomuch++;
2018 goto dropafterack;
2019 }
2020 /*
2021 * If we reach this point, ACK is not a duplicate,
2022 * i.e., it ACKs something we sent.
2023 */
2024 if (tp->t_flags & TF_NEEDSYN) {
2025 /*
2026 * T/TCP: Connection was half-synchronized, and our
2027 * SYN has been ACK'd (so connection is now fully
2028 * synchronized). Go to non-starred state,
2029 * increment snd_una for ACK of SYN, and check if
2030 * we can do window scaling.
2031 */
2032 tp->t_flags &= ~TF_NEEDSYN;
2033 tp->snd_una++;
2034 /* Do window scaling? */
61896e3c
JH
2035 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2036 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
2037 tp->snd_scale = tp->requested_s_scale;
2038 tp->rcv_scale = tp->request_r_scale;
2039 }
2040 }
2041
2042process_ACK:
2043 acked = th->th_ack - tp->snd_una;
2044 tcpstat.tcps_rcvackpack++;
2045 tcpstat.tcps_rcvackbyte += acked;
2046
b5572302 2047 if (tcp_do_eifel_detect && acked > 0 &&
95b22adf 2048 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
bfdb979e
JH
2049 (tp->t_flags & TF_FIRSTACCACK)) {
2050 /* Eifel detection applicable. */
2051 if (to.to_tsecr < tp->t_rexmtTS) {
bfdb979e 2052 ++tcpstat.tcps_eifeldetected;
8819433a
JH
2053 tcp_revert_congestion_state(tp);
2054 if (tp->t_rxtshift == 1 &&
2055 ticks >= tp->t_badrxtwin)
2056 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
2057 }
2058 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
b5572302
JH
2059 /*
2060 * If we just performed our first retransmit,
2061 * and the ACK arrives within our recovery window,
2062 * then it was a mistake to do the retransmit
91489f6b 2063 * in the first place. Recover our original cwnd
b5572302
JH
2064 * and ssthresh, and proceed to transmit where we
2065 * left off.
2066 */
bfdb979e
JH
2067 tcp_revert_congestion_state(tp);
2068 ++tcpstat.tcps_rttdetected;
984263bc
MD
2069 }
2070
2071 /*
2072 * If we have a timestamp reply, update smoothed
2073 * round trip time. If no timestamp is present but
2074 * transmit timer is running and timed sequence
2075 * number was acked, update smoothed round trip time.
2076 * Since we now have an rtt measurement, cancel the
2077 * timer backoff (cf., Phil Karn's retransmit alg.).
2078 * Recompute the initial retransmit timer.
2079 *
2080 * Some machines (certain windows boxes) send broken
d24ce1dc 2081 * timestamp replies during the SYN+ACK phase, ignore
984263bc
MD
2082 * timestamps of 0.
2083 */
95b22adf 2084 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
984263bc 2085 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
95b22adf 2086 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
984263bc 2087 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
984263bc
MD
2088 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2089
984263bc
MD
2090 /*
2091 * If no data (only SYN) was ACK'd,
2092 * skip rest of ACK processing.
2093 */
2094 if (acked == 0)
2095 goto step6;
2096
efd4b327 2097 /* Stop looking for an acceptable ACK since one was received. */
8819433a 2098 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
efd4b327 2099
984263bc
MD
2100 if (acked > so->so_snd.sb_cc) {
2101 tp->snd_wnd -= so->so_snd.sb_cc;
2102 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
61896e3c 2103 ourfinisacked = TRUE;
984263bc
MD
2104 } else {
2105 sbdrop(&so->so_snd, acked);
2106 tp->snd_wnd -= acked;
61896e3c 2107 ourfinisacked = FALSE;
984263bc
MD
2108 }
2109 sowwakeup(so);
91489f6b
JH
2110
2111 /*
2112 * Update window information.
2113 * Don't look at window if no ACK:
2114 * TAC's send garbage on first SYN.
2115 */
2116 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2117 (tp->snd_wl1 == th->th_seq &&
2118 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2119 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2120 /* keep track of pure window updates */
2121 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2122 tiwin > tp->snd_wnd)
2123 tcpstat.tcps_rcvwinupd++;
2124 tp->snd_wnd = tiwin;
2125 tp->snd_wl1 = th->th_seq;
2126 tp->snd_wl2 = th->th_ack;
2127 if (tp->snd_wnd > tp->max_sndwnd)
2128 tp->max_sndwnd = tp->snd_wnd;
2129 needoutput = TRUE;
2130 }
2131
2132 tp->snd_una = th->th_ack;
2133 if (TCP_DO_SACK(tp))
2134 tcp_sack_update_scoreboard(tp, &to);
95b22adf 2135 if (IN_FASTRECOVERY(tp)) {
91489f6b 2136 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
95b22adf 2137 EXIT_FASTRECOVERY(tp);
91489f6b
JH
2138 needoutput = TRUE;
2139 /*
2140 * If the congestion window was inflated
2141 * to account for the other side's
2142 * cached packets, retract it.
2143 *
2144 * Window inflation should have left us
2145 * with approximately snd_ssthresh outstanding
2146 * data. But, in case we would be inclined
2147 * to send a burst, better do it using
2148 * slow start.
2149 */
2150 if (!TCP_DO_SACK(tp))
2151 tp->snd_cwnd = tp->snd_ssthresh;
2152
2153 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2154 tp->snd_max + 2 * tp->t_maxseg))
2155 tp->snd_cwnd =
2156 (tp->snd_max - tp->snd_una) +
2157 2 * tp->t_maxseg;
2158 } else {
2159 if (TCP_DO_SACK(tp)) {
2160 tp->snd_max_rexmt = tp->snd_max;
2161 tcp_sack_rexmt(tp, th);
2162 } else {
2163 tcp_newreno_partial_ack(tp, th, acked);
2164 }
2165 needoutput = FALSE;
2166 }
95b22adf 2167 } else {
91489f6b
JH
2168 /*
2169 * When new data is acked, open the congestion window.
2170 * If the window gives us less than ssthresh packets
2171 * in flight, open exponentially (maxseg per packet).
2172 * Otherwise open linearly: maxseg per window
2173 * (maxseg^2 / cwnd per packet).
2174 */
2175 u_int cw = tp->snd_cwnd;
2176 u_int incr;
2177
2178 if (cw > tp->snd_ssthresh)
2179 incr = tp->t_maxseg * tp->t_maxseg / cw;
2180 else
2181 incr = tp->t_maxseg;
2182 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
95b22adf 2183 tp->snd_recover = th->th_ack - 1;
cfb3f3f4 2184 }
984263bc
MD
2185 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2186 tp->snd_nxt = tp->snd_una;
2187
91489f6b
JH
2188 /*
2189 * If all outstanding data is acked, stop retransmit
2190 * timer and remember to restart (more output or persist).
2191 * If there is more data to be acked, restart retransmit
2192 * timer, using current (possibly backed-off) value.
2193 */
2194 if (th->th_ack == tp->snd_max) {
2195 callout_stop(tp->tt_rexmt);
2196 needoutput = TRUE;
2197 } else if (!callout_active(tp->tt_persist))
2198 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2199 tcp_timer_rexmt, tp);
2200
984263bc 2201 switch (tp->t_state) {
984263bc
MD
2202 /*
2203 * In FIN_WAIT_1 STATE in addition to the processing
2204 * for the ESTABLISHED state if our FIN is now acknowledged
2205 * then enter FIN_WAIT_2.
2206 */
2207 case TCPS_FIN_WAIT_1:
2208 if (ourfinisacked) {
2209 /*
2210 * If we can't receive any more
2211 * data, then closing user can proceed.
2212 * Starting the timer is contrary to the
2213 * specification, but if we don't get a FIN
2214 * we'll hang forever.
2215 */
2216 if (so->so_state & SS_CANTRCVMORE) {
2217 soisdisconnected(so);
2218 callout_reset(tp->tt_2msl, tcp_maxidle,
2219 tcp_timer_2msl, tp);
2220 }
2221 tp->t_state = TCPS_FIN_WAIT_2;
2222 }
2223 break;
2224
95b22adf 2225 /*
984263bc
MD
2226 * In CLOSING STATE in addition to the processing for
2227 * the ESTABLISHED state if the ACK acknowledges our FIN
2228 * then enter the TIME-WAIT state, otherwise ignore
2229 * the segment.
2230 */
2231 case TCPS_CLOSING:
2232 if (ourfinisacked) {
2233 tp->t_state = TCPS_TIME_WAIT;
2234 tcp_canceltimers(tp);
2235 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2236 if (tp->cc_recv != 0 &&
2237 (ticks - tp->t_starttime) < tcp_msl)
2238 callout_reset(tp->tt_2msl,
b5572302
JH
2239 tp->t_rxtcur * TCPTV_TWTRUNC,
2240 tcp_timer_2msl, tp);
984263bc
MD
2241 else
2242 callout_reset(tp->tt_2msl, 2 * tcp_msl,
b5572302 2243 tcp_timer_2msl, tp);
984263bc
MD
2244 soisdisconnected(so);
2245 }
2246 break;
2247
2248 /*
2249 * In LAST_ACK, we may still be waiting for data to drain
2250 * and/or to be acked, as well as for the ack of our FIN.
2251 * If our FIN is now acknowledged, delete the TCB,
2252 * enter the closed state and return.
2253 */
2254 case TCPS_LAST_ACK:
2255 if (ourfinisacked) {
2256 tp = tcp_close(tp);
2257 goto drop;
2258 }
2259 break;
2260
2261 /*
2262 * In TIME_WAIT state the only thing that should arrive
2263 * is a retransmission of the remote FIN. Acknowledge
2264 * it and restart the finack timer.
2265 */
2266 case TCPS_TIME_WAIT:
2267 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2268 tcp_timer_2msl, tp);
2269 goto dropafterack;
2270 }
2271 }
2272
2273step6:
2274 /*
2275 * Update window information.
2276 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2277 */
2278 if ((thflags & TH_ACK) &&
df9d7670 2279 acceptable_window_update(tp, th, tiwin)) {
984263bc 2280 /* keep track of pure window updates */
b5572302
JH
2281 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2282 tiwin > tp->snd_wnd)
984263bc
MD
2283 tcpstat.tcps_rcvwinupd++;
2284 tp->snd_wnd = tiwin;
2285 tp->snd_wl1 = th->th_seq;
2286 tp->snd_wl2 = th->th_ack;
2287 if (tp->snd_wnd > tp->max_sndwnd)
2288 tp->max_sndwnd = tp->snd_wnd;
61896e3c 2289 needoutput = TRUE;
984263bc
MD
2290 }
2291
2292 /*
2293 * Process segments with URG.
2294 */
2295 if ((thflags & TH_URG) && th->th_urp &&
95b22adf 2296 !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2297 /*
2298 * This is a kludge, but if we receive and accept
2299 * random urgent pointers, we'll crash in
2300 * soreceive. It's hard to imagine someone
2301 * actually wanting to send this much urgent data.
2302 */
2303 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2304 th->th_urp = 0; /* XXX */
2305 thflags &= ~TH_URG; /* XXX */
2306 goto dodata; /* XXX */
2307 }
2308 /*
2309 * If this segment advances the known urgent pointer,
2310 * then mark the data stream. This should not happen
2311 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2312 * a FIN has been received from the remote side.
2313 * In these states we ignore the URG.
2314 *
2315 * According to RFC961 (Assigned Protocols),
2316 * the urgent pointer points to the last octet
2317 * of urgent data. We continue, however,
2318 * to consider it to indicate the first octet
2319 * of data past the urgent section as the original
2320 * spec states (in one of two places).
2321 */
61896e3c 2322 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
984263bc
MD
2323 tp->rcv_up = th->th_seq + th->th_urp;
2324 so->so_oobmark = so->so_rcv.sb_cc +
2325 (tp->rcv_up - tp->rcv_nxt) - 1;
2326 if (so->so_oobmark == 0)
2327 so->so_state |= SS_RCVATMARK;
2328 sohasoutofband(so);
2329 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2330 }
2331 /*
2332 * Remove out of band data so doesn't get presented to user.
2333 * This can happen independent of advancing the URG pointer,
2334 * but if two URG's are pending at once, some out-of-band
2335 * data may creep in... ick.
2336 */
61896e3c
JH
2337 if (th->th_urp <= (u_long)tlen &&
2338 !(so->so_options & SO_OOBINLINE)) {
2339 /* hdr drop is delayed */
2340 tcp_pulloutofband(so, th, m, drop_hdrlen);
2341 }
984263bc
MD
2342 } else {
2343 /*
2344 * If no out of band data is expected,
2345 * pull receive urgent pointer along
2346 * with the receive window.
2347 */
2348 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2349 tp->rcv_up = tp->rcv_nxt;
2350 }
984263bc 2351
61896e3c 2352dodata: /* XXX */
984263bc
MD
2353 /*
2354 * Process the segment text, merging it into the TCP sequencing queue,
2355 * and arranging for acknowledgment of receipt if necessary.
2356 * This process logically involves adjusting tp->rcv_wnd as data
2357 * is presented to the user (this happens in tcp_usrreq.c,
2358 * case PRU_RCVD). If a FIN has already been received on this
2359 * connection then we just ignore the text.
2360 */
95b22adf 2361 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2362 m_adj(m, drop_hdrlen); /* delayed header drop */
2363 /*
2364 * Insert segment which includes th into TCP reassembly queue
2365 * with control block tp. Set thflags to whether reassembly now
2366 * includes a segment with FIN. This handles the common case
2367 * inline (segment is the next to be received on an established
2368 * connection, and the queue is empty), avoiding linkage into
2369 * and removal from the queue and repetition of various
2370 * conversions.
2371 * Set DELACK for segments received in order, but ack
2372 * immediately when segments are out of order (so
2373 * fast retransmit can work).
2374 */
2375 if (th->th_seq == tp->rcv_nxt &&
2376 LIST_EMPTY(&tp->t_segq) &&
2377 TCPS_HAVEESTABLISHED(tp->t_state)) {
2378 if (DELAY_ACK(tp))
2379 callout_reset(tp->tt_delack, tcp_delacktime,
2380 tcp_timer_delack, tp);
2381 else
2382 tp->t_flags |= TF_ACKNOW;
2383 tp->rcv_nxt += tlen;
2384 thflags = th->th_flags & TH_FIN;
2385 tcpstat.tcps_rcvpack++;
2386 tcpstat.tcps_rcvbyte += tlen;
2387 ND6_HINT(tp);
2388 if (so->so_state & SS_CANTRCVMORE)
2389 m_freem(m);
2390 else
507106e0 2391 sbappendstream(&so->so_rcv, m);
984263bc
MD
2392 sorwakeup(so);
2393 } else {
91489f6b
JH
2394 if (!(tp->t_flags & TF_DUPSEG)) {
2395 /* Initialize SACK report block. */
2396 tp->reportblk.rblk_start = th->th_seq;
2397 tp->reportblk.rblk_end = th->th_seq + tlen +
2398 ((thflags & TH_FIN) != 0);
2399 }
984263bc
MD
2400 thflags = tcp_reass(tp, th, &tlen, m);
2401 tp->t_flags |= TF_ACKNOW;
2402 }
2403
2404 /*
2405 * Note the amount of data that peer has sent into
2406 * our window, in order to estimate the sender's
2407 * buffer size.
2408 */
2409 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2410 } else {
2411 m_freem(m);
2412 thflags &= ~TH_FIN;
2413 }
2414
2415 /*
2416 * If FIN is received ACK the FIN and let the user know
2417 * that the connection is closing.
2418 */
2419 if (thflags & TH_FIN) {
95b22adf 2420 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2421 socantrcvmore(so);
2422 /*
2423 * If connection is half-synchronized
2424 * (ie NEEDSYN flag on) then delay ACK,
2425 * so it may be piggybacked when SYN is sent.
2426 * Otherwise, since we received a FIN then no
2427 * more input can be expected, send ACK now.
2428 */
2429 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
61896e3c
JH
2430 callout_reset(tp->tt_delack, tcp_delacktime,
2431 tcp_timer_delack, tp);
984263bc
MD
2432 else
2433 tp->t_flags |= TF_ACKNOW;
2434 tp->rcv_nxt++;
2435 }
984263bc 2436
61896e3c 2437 switch (tp->t_state) {
95b22adf 2438 /*
984263bc
MD
2439 * In SYN_RECEIVED and ESTABLISHED STATES
2440 * enter the CLOSE_WAIT state.
2441 */
2442 case TCPS_SYN_RECEIVED:
2443 tp->t_starttime = ticks;
2444 /*FALLTHROUGH*/
2445 case TCPS_ESTABLISHED:
2446 tp->t_state = TCPS_CLOSE_WAIT;
2447 break;
2448
95b22adf 2449 /*
984263bc
MD
2450 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2451 * enter the CLOSING state.
2452 */
2453 case TCPS_FIN_WAIT_1:
2454 tp->t_state = TCPS_CLOSING;
2455 break;
2456
95b22adf 2457 /*
984263bc
MD
2458 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2459 * starting the time-wait timer, turning off the other
2460 * standard timers.
2461 */
2462 case TCPS_FIN_WAIT_2:
2463 tp->t_state = TCPS_TIME_WAIT;
2464 tcp_canceltimers(tp);
2465 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2466 if (tp->cc_recv != 0 &&
2467 (ticks - tp->t_starttime) < tcp_msl) {
2468 callout_reset(tp->tt_2msl,
2469 tp->t_rxtcur * TCPTV_TWTRUNC,
2470 tcp_timer_2msl, tp);
2471 /* For transaction client, force ACK now. */
2472 tp->t_flags |= TF_ACKNOW;
2473 }
2474 else
2475 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2476 tcp_timer_2msl, tp);
2477 soisdisconnected(so);
2478 break;
2479
2480 /*
2481 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2482 */
2483 case TCPS_TIME_WAIT:
2484 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2485 tcp_timer_2msl, tp);
2486 break;
2487 }
2488 }
61896e3c 2489
984263bc
MD
2490#ifdef TCPDEBUG
2491 if (so->so_options & SO_DEBUG)
f23061d4 2492 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2493#endif
2494
2495 /*
2496 * Return any desired output.
2497 */
2498 if (needoutput || (tp->t_flags & TF_ACKNOW))
f23061d4 2499 tcp_output(tp);
984263bc
MD
2500 return;
2501
2502dropafterack:
2503 /*
2504 * Generate an ACK dropping incoming segment if it occupies
2505 * sequence space, where the ACK reflects our state.
2506 *
2507 * We can now skip the test for the RST flag since all
2508 * paths to this code happen after packets containing
2509 * RST have been dropped.
2510 *
2511 * In the SYN-RECEIVED state, don't send an ACK unless the
2512 * segment we received passes the SYN-RECEIVED ACK test.
2513 * If it fails send a RST. This breaks the loop in the
2514 * "LAND" DoS attack, and also prevents an ACK storm
2515 * between two listening ports that have been sent forged
2516 * SYN segments, each with the source address of the other.
2517 */
2518 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2519 (SEQ_GT(tp->snd_una, th->th_ack) ||
2520 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2521 rstreason = BANDLIM_RST_OPENPORT;
2522 goto dropwithreset;
2523 }
2524#ifdef TCPDEBUG
2525 if (so->so_options & SO_DEBUG)
f23061d4 2526 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2527#endif
2528 m_freem(m);
2529 tp->t_flags |= TF_ACKNOW;
f23061d4 2530 tcp_output(tp);
984263bc
MD
2531 return;
2532
2533dropwithreset:
2534 /*
2535 * Generate a RST, dropping incoming segment.
2536 * Make ACK acceptable to originator of segment.
2537 * Don't bother to respond if destination was broadcast/multicast.
2538 */
61896e3c 2539 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
2540 goto drop;
2541 if (isipv6) {
2542 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2543 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2544 goto drop;
2545 } else {
2546 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2547 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
95b22adf
JH
2548 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2549 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
984263bc
MD
2550 goto drop;
2551 }
2552 /* IPv6 anycast check is done at tcp6_input() */
2553
2554 /*
2555 * Perform bandwidth limiting.
2556 */
2557#ifdef ICMP_BANDLIM
2558 if (badport_bandlim(rstreason) < 0)
2559 goto drop;
2560#endif
2561
2562#ifdef TCPDEBUG
2563 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2564 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2565#endif
2566 if (thflags & TH_ACK)
2567 /* mtod() below is safe as long as hdr dropping is delayed */
2568 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2569 TH_RST);
2570 else {
2571 if (thflags & TH_SYN)
2572 tlen++;
2573 /* mtod() below is safe as long as hdr dropping is delayed */
61896e3c
JH
2574 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2575 (tcp_seq)0, TH_RST | TH_ACK);
984263bc
MD
2576 }
2577 return;
2578
2579drop:
2580 /*
2581 * Drop space held by incoming segment and return.
2582 */
2583#ifdef TCPDEBUG
2584 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2585 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2586#endif
2587 m_freem(m);
2588 return;
2589}
2590
2591/*
2592 * Parse TCP options and place in tcpopt.
2593 */
2594static void
95b22adf 2595tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
984263bc 2596{
91489f6b 2597 int opt, optlen, i;
984263bc
MD
2598
2599 to->to_flags = 0;
2600 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2601 opt = cp[0];
2602 if (opt == TCPOPT_EOL)
2603 break;
2604 if (opt == TCPOPT_NOP)
2605 optlen = 1;
2606 else {
2607 if (cnt < 2)
2608 break;
2609 optlen = cp[1];
2610 if (optlen < 2 || optlen > cnt)
2611 break;
2612 }
2613 switch (opt) {
2614 case TCPOPT_MAXSEG:
2615 if (optlen != TCPOLEN_MAXSEG)
2616 continue;
2617 if (!is_syn)
2618 continue;
2619 to->to_flags |= TOF_MSS;
95b22adf 2620 bcopy(cp + 2, &to->to_mss, sizeof(to->to_mss));
984263bc
MD
2621 to->to_mss = ntohs(to->to_mss);
2622 break;
2623 case TCPOPT_WINDOW:
2624 if (optlen != TCPOLEN_WINDOW)
2625 continue;
95b22adf 2626 if (!is_syn)
984263bc
MD
2627 continue;
2628 to->to_flags |= TOF_SCALE;
2629 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2630 break;
2631 case TCPOPT_TIMESTAMP:
2632 if (optlen != TCPOLEN_TIMESTAMP)
2633 continue;
2634 to->to_flags |= TOF_TS;
95b22adf 2635 bcopy(cp + 2, &to->to_tsval, sizeof(to->to_tsval));
984263bc 2636 to->to_tsval = ntohl(to->to_tsval);
95b22adf 2637 bcopy(cp + 6, &to->to_tsecr, sizeof(to->to_tsecr));
984263bc
MD
2638 to->to_tsecr = ntohl(to->to_tsecr);
2639 break;
2640 case TCPOPT_CC:
2641 if (optlen != TCPOLEN_CC)
2642 continue;
2643 to->to_flags |= TOF_CC;
95b22adf 2644 bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc));
984263bc
MD
2645 to->to_cc = ntohl(to->to_cc);
2646 break;
2647 case TCPOPT_CCNEW:
2648 if (optlen != TCPOLEN_CC)
2649 continue;
2650 if (!is_syn)
2651 continue;
2652 to->to_flags |= TOF_CCNEW;
95b22adf 2653 bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc));
984263bc
MD
2654 to->to_cc = ntohl(to->to_cc);
2655 break;
2656 case TCPOPT_CCECHO:
2657 if (optlen != TCPOLEN_CC)
2658 continue;
2659 if (!is_syn)
2660 continue;
2661 to->to_flags |= TOF_CCECHO;
95b22adf 2662 bcopy(cp + 2, &to->to_ccecho, sizeof(to->to_ccecho));
984263bc
MD
2663 to->to_ccecho = ntohl(to->to_ccecho);
2664 break;
91489f6b
JH
2665 case TCPOPT_SACK_PERMITTED:
2666 if (optlen != TCPOLEN_SACK_PERMITTED)
2667 continue;
2668 if (!is_syn)
2669 continue;
2670 to->to_flags |= TOF_SACK_PERMITTED;
2671 break;
2672 case TCPOPT_SACK:
2673 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2674 continue;
2675 to->to_nsackblocks = (optlen - 2) / 8;
2676 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2677 to->to_flags |= TOF_SACK;
2678 for (i = 0; i < to->to_nsackblocks; i++) {
2679 struct raw_sackblock *r = &to->to_sackblocks[i];
2680
2681 r->rblk_start = ntohl(r->rblk_start);
2682 r->rblk_end = ntohl(r->rblk_end);
2683 }
2684 break;
984263bc
MD
2685 default:
2686 continue;
2687 }
2688 }
2689}
2690
2691/*
2692 * Pull out of band byte out of a segment so
2693 * it doesn't appear in the user's data queue.
2694 * It is still reflected in the segment length for
2695 * sequencing purposes.
95b22adf 2696 * "off" is the delayed to be dropped hdrlen.
984263bc
MD
2697 */
2698static void
95b22adf 2699tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
984263bc
MD
2700{
2701 int cnt = off + th->th_urp - 1;
2702
2703 while (cnt >= 0) {
2704 if (m->m_len > cnt) {
2705 char *cp = mtod(m, caddr_t) + cnt;
2706 struct tcpcb *tp = sototcpcb(so);
2707
2708 tp->t_iobc = *cp;
2709 tp->t_oobflags |= TCPOOB_HAVEDATA;
95b22adf 2710 bcopy(cp + 1, cp, m->m_len - cnt - 1);
984263bc
MD
2711 m->m_len--;
2712 if (m->m_flags & M_PKTHDR)
2713 m->m_pkthdr.len--;
2714 return;
2715 }
2716 cnt -= m->m_len;
2717 m = m->m_next;
2718 if (m == 0)
2719 break;
2720 }
2721 panic("tcp_pulloutofband");
2722}
2723
2724/*
2725 * Collect new round-trip time estimate
2726 * and update averages and current timeout.
2727 */
2728static void
95b22adf 2729tcp_xmit_timer(struct tcpcb *tp, int rtt)
984263bc 2730{
2256ba69 2731 int delta;
984263bc
MD
2732
2733 tcpstat.tcps_rttupdated++;
2734 tp->t_rttupdated++;
2735 if (tp->t_srtt != 0) {
2736 /*
2737 * srtt is stored as fixed point with 5 bits after the
2738 * binary point (i.e., scaled by 8). The following magic
2739 * is equivalent to the smoothing algorithm in rfc793 with
2740 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2741 * point). Adjust rtt to origin 0.
2742 */
2743 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2744 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2745
2746 if ((tp->t_srtt += delta) <= 0)
2747 tp->t_srtt = 1;
2748
2749 /*
2750 * We accumulate a smoothed rtt variance (actually, a
2751 * smoothed mean difference), then set the retransmit
2752 * timer to smoothed rtt + 4 times the smoothed variance.
2753 * rttvar is stored as fixed point with 4 bits after the
2754 * binary point (scaled by 16). The following is
2755 * equivalent to rfc793 smoothing with an alpha of .75
2756 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2757 * rfc793's wired-in beta.
2758 */
2759 if (delta < 0)
2760 delta = -delta;
2761 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2762 if ((tp->t_rttvar += delta) <= 0)
2763 tp->t_rttvar = 1;
2764 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2765 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2766 } else {
2767 /*
2768 * No rtt measurement yet - use the unsmoothed rtt.
2769 * Set the variance to half the rtt (so our first
2770 * retransmit happens at 3*rtt).
2771 */
2772 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2773 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2774 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2775 }
2776 tp->t_rtttime = 0;
2777 tp->t_rxtshift = 0;
2778
2779 /*
2780 * the retransmit should happen at rtt + 4 * rttvar.
2781 * Because of the way we do the smoothing, srtt and rttvar
2782 * will each average +1/2 tick of bias. When we compute
2783 * the retransmit timer, we want 1/2 tick of rounding and
2784 * 1 extra tick because of +-1/2 tick uncertainty in the
2785 * firing of the timer. The bias will give us exactly the
2786 * 1.5 tick we need. But, because the bias is
2787 * statistical, we have to test that we don't drop below
2788 * the minimum feasible timer (which is 2 ticks).
2789 */
2790 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2791 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2792
2793 /*
2794 * We received an ack for a packet that wasn't retransmitted;
2795 * it is probably safe to discard any error indications we've
2796 * received recently. This isn't quite right, but close enough
2797 * for now (a route might have failed after we sent a segment,
2798 * and the return path might not be symmetrical).
2799 */
2800 tp->t_softerror = 0;
2801}
2802
2803/*
2804 * Determine a reasonable value for maxseg size.
2805 * If the route is known, check route for mtu.
2806 * If none, use an mss that can be handled on the outgoing
2807 * interface without forcing IP to fragment; if bigger than
2808 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2809 * to utilize large mbufs. If no route is found, route has no mtu,
2810 * or the destination isn't local, use a default, hopefully conservative
2811 * size (usually 512 or the default IP max size, but no more than the mtu
2812 * of the interface), as we can't discover anything about intervening
2813 * gateways or networks. We also initialize the congestion/slow start
2814 * window to be a single segment if the destination isn't local.
2815 * While looking at the routing entry, we also initialize other path-dependent
2816 * parameters from pre-set or cached values in the routing entry.
2817 *
2818 * Also take into account the space needed for options that we
2819 * send regularly. Make maxseg shorter by that amount to assure
2820 * that we can send maxseg amount of data even when the options
2821 * are present. Store the upper limit of the length of options plus
2822 * data in maxopd.
2823 *
2824 * NOTE that this routine is only called when we process an incoming
2825 * segment, for outgoing segments only tcp_mssopt is called.
2826 *
2827 * In case of T/TCP, we call this routine during implicit connection
2828 * setup as well (offer = -1), to initialize maxseg from the cached
2829 * MSS of our peer.
2830 */
2831void
95b22adf 2832tcp_mss(struct tcpcb *tp, int offer)
984263bc 2833{
2256ba69 2834 struct rtentry *rt;
984263bc 2835 struct ifnet *ifp;
2256ba69 2836 int rtt, mss;
984263bc
MD
2837 u_long bufsize;
2838 struct inpcb *inp = tp->t_inpcb;
2839 struct socket *so;
2840 struct rmxp_tao *taop;
2841 int origoffer = offer;
2842#ifdef INET6
d24ce1dc 2843 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
2844 size_t min_protoh = isipv6 ?
2845 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2846 sizeof(struct tcpiphdr);
2847#else
d24ce1dc 2848 const boolean_t isipv6 = FALSE;
984263bc
MD
2849 const size_t min_protoh = sizeof(struct tcpiphdr);
2850#endif
2851
2852 if (isipv6)
2853 rt = tcp_rtlookup6(&inp->inp_inc);
2854 else
2855 rt = tcp_rtlookup(&inp->inp_inc);
2856 if (rt == NULL) {
2857 tp->t_maxopd = tp->t_maxseg =
d24ce1dc 2858 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
984263bc
MD
2859 return;
2860 }
2861 ifp = rt->rt_ifp;
2862 so = inp->inp_socket;
2863
2864 taop = rmx_taop(rt->rt_rmx);
2865 /*
2866 * Offer == -1 means that we didn't receive SYN yet,
2867 * use cached value in that case;
2868 */
2869 if (offer == -1)
2870 offer = taop->tao_mssopt;
2871 /*
2872 * Offer == 0 means that there was no MSS on the SYN segment,
2873 * in this case we use tcp_mssdflt.
2874 */
2875 if (offer == 0)
d24ce1dc 2876 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
984263bc
MD
2877 else
2878 /*
2879 * Sanity check: make sure that maxopd will be large
2880 * enough to allow some data on segments even is the
2881 * all the option space is used (40bytes). Otherwise
2882 * funny things may happen in tcp_output.
2883 */
2884 offer = max(offer, 64);
2885 taop->tao_mssopt = offer;
2886
2887 /*
2888 * While we're here, check if there's an initial rtt
2889 * or rttvar. Convert from the route-table units
2890 * to scaled multiples of the slow timeout timer.
2891 */
2892 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2893 /*
2894 * XXX the lock bit for RTT indicates that the value
2895 * is also a minimum value; this is subject to time.
2896 */
2897 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2898 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2899 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2900 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2901 tcpstat.tcps_usedrtt++;
2902 if (rt->rt_rmx.rmx_rttvar) {
2903 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2904 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2905 tcpstat.tcps_usedrttvar++;
2906 } else {
2907 /* default variation is +- 1 rtt */
2908 tp->t_rttvar =
2909 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2910 }
2911 TCPT_RANGESET(tp->t_rxtcur,
2912 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2913 tp->t_rttmin, TCPTV_REXMTMAX);
2914 }
2915 /*
2916 * if there's an mtu associated with the route, use it
2917 * else, use the link mtu.
2918 */
2919 if (rt->rt_rmx.rmx_mtu)
2920 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2921 else {
2922 if (isipv6) {
2923 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu -
2924 min_protoh;
2925 if (!in6_localaddr(&inp->in6p_faddr))
2926 mss = min(mss, tcp_v6mssdflt);
2927 } else {
2928 mss = ifp->if_mtu - min_protoh;
2929 if (!in_localaddr(inp->inp_faddr))
2930 mss = min(mss, tcp_mssdflt);
2931 }
2932 }
2933 mss = min(mss, offer);
2934 /*
2935 * maxopd stores the maximum length of data AND options
2936 * in a segment; maxseg is the amount of data in a normal
2937 * segment. We need to store this value (maxopd) apart
2938 * from maxseg, because now every segment carries options
2939 * and thus we normally have somewhat less data in segments.
2940 */
2941 tp->t_maxopd = mss;
2942
2943 /*
2944 * In case of T/TCP, origoffer==-1 indicates, that no segments
2945 * were received yet. In this case we just guess, otherwise
2946 * we do the same as before T/TCP.
2947 */
61896e3c 2948 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
984263bc
MD
2949 (origoffer == -1 ||
2950 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2951 mss -= TCPOLEN_TSTAMP_APPA;
61896e3c 2952 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
984263bc
MD
2953 (origoffer == -1 ||
2954 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2955 mss -= TCPOLEN_CC_APPA;
2956
2957#if (MCLBYTES & (MCLBYTES - 1)) == 0
2958 if (mss > MCLBYTES)
2959 mss &= ~(MCLBYTES-1);
2960#else
2961 if (mss > MCLBYTES)
2962 mss = mss / MCLBYTES * MCLBYTES;
2963#endif
2964 /*
2965 * If there's a pipesize, change the socket buffer
2966 * to that size. Make the socket buffers an integral
2967 * number of mss units; if the mss is larger than
2968 * the socket buffer, decrease the mss.
2969 */
2970#ifdef RTV_SPIPE
2971 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2972#endif
2973 bufsize = so->so_snd.sb_hiwat;
2974 if (bufsize < mss)
2975 mss = bufsize;
2976 else {
2977 bufsize = roundup(bufsize, mss);
2978 if (bufsize > sb_max)
2979 bufsize = sb_max;
2980 if (bufsize > so->so_snd.sb_hiwat)
f23061d4 2981 sbreserve(&so->so_snd, bufsize, so, NULL);
984263bc
MD
2982 }
2983 tp->t_maxseg = mss;
2984
2985#ifdef RTV_RPIPE
2986 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2987#endif
2988 bufsize = so->so_rcv.sb_hiwat;
2989 if (bufsize > mss) {
2990 bufsize = roundup(bufsize, mss);
2991 if (bufsize > sb_max)
2992 bufsize = sb_max;
2993 if (bufsize > so->so_rcv.sb_hiwat)
f23061d4 2994 sbreserve(&so->so_rcv, bufsize, so, NULL);
984263bc
MD
2995 }
2996
2997 /*
2998 * Set the slow-start flight size depending on whether this
2999 * is a local network or not.
3000 */
4b52d5ee
JH
3001 if (tcp_do_rfc3390)
3002 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3003 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3004 (!isipv6 && in_localaddr(inp->inp_faddr)))
984263bc 3005 tp->snd_cwnd = mss * ss_fltsz_local;
4b52d5ee 3006 else
984263bc
MD
3007 tp->snd_cwnd = mss * ss_fltsz;
3008
3009 if (rt->rt_rmx.rmx_ssthresh) {
3010 /*
3011 * There's some sort of gateway or interface
3012 * buffer limit on the path. Use this to set
3013 * the slow start threshhold, but set the
3014 * threshold to no less than 2*mss.
3015 */
3016 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3017 tcpstat.tcps_usedssthresh++;
3018 }
3019}
3020
3021/*
3022 * Determine the MSS option to send on an outgoing SYN.
3023 */
3024int
95b22adf 3025tcp_mssopt(struct tcpcb *tp)
984263bc
MD
3026{
3027 struct rtentry *rt;
3028#ifdef INET6
d24ce1dc
JH
3029 boolean_t isipv6 =
3030 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
3031 int min_protoh = isipv6 ?
3032 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3033 sizeof(struct tcpiphdr);
3034#else
d24ce1dc 3035 const boolean_t isipv6 = FALSE;
984263bc
MD
3036 const size_t min_protoh = sizeof(struct tcpiphdr);
3037#endif
3038
3039 if (isipv6)
3040 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3041 else
3042 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3043 if (rt == NULL)
3044 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3045
3046 return (rt->rt_ifp->if_mtu - min_protoh);
3047}
3048
984263bc
MD
3049/*
3050 * When a partial ack arrives, force the retransmission of the
91489f6b
JH
3051 * next unacknowledged segment. Do not exit Fast Recovery.
3052 *
3053 * Implement the Slow-but-Steady variant of NewReno by restarting the
3054 * the retransmission timer. Turn it off here so it can be restarted
3055 * later in tcp_output().
984263bc
MD
3056 */
3057static void
91489f6b 3058tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
984263bc 3059{
91489f6b
JH
3060 tcp_seq old_snd_nxt = tp->snd_nxt;
3061 u_long ocwnd = tp->snd_cwnd;
984263bc
MD
3062
3063 callout_stop(tp->tt_rexmt);
3064 tp->t_rtttime = 0;
3065 tp->snd_nxt = th->th_ack;
91489f6b
JH
3066 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3067 tp->snd_cwnd = tp->t_maxseg;
984263bc 3068 tp->t_flags |= TF_ACKNOW;
f23061d4 3069 tcp_output(tp);
91489f6b
JH
3070 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3071 tp->snd_nxt = old_snd_nxt;
3072 /* partial window deflation */
3073 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3074}
3075
3076/*
3077 * In contrast to the Slow-but-Steady NewReno variant,
3078 * we do not reset the retransmission timer for SACK retransmissions,
3079 * except when retransmitting snd_una.
3080 */
3081static void
3082tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3083{
3084 uint32_t pipe, seglen;
3085 tcp_seq nextrexmt;
3086 boolean_t lostdup;
3087 tcp_seq old_snd_nxt = tp->snd_nxt;
3088 u_long ocwnd = tp->snd_cwnd;
3089 int nseg = 0; /* consecutive new segments */
3090#define MAXBURST 4 /* limit burst of new packets on partial ack */
3091
3092 tp->t_rtttime = 0;
3093 pipe = tcp_sack_compute_pipe(tp);
3094 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3095 (!tcp_do_smartsack || nseg < MAXBURST) &&
3096 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3097 uint32_t sent;
3098 tcp_seq old_snd_max;
3099 int error;
3100
3101 if (nextrexmt == tp->snd_max) ++nseg;
3102 tp->snd_nxt = nextrexmt;
3103 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3104 old_snd_max = tp->snd_max;
3105 if (nextrexmt == tp->snd_una)
3106 callout_stop(tp->tt_rexmt);
3107 error = tcp_output(tp);
3108 if (error != 0)
3109 break;
3110 sent = tp->snd_nxt - nextrexmt;
3111 if (sent <= 0)
3112 break;
3113 if (!lostdup)
3114 pipe += sent;
3115 tcpstat.tcps_sndsackpack++;
3116 tcpstat.tcps_sndsackbyte += sent;
3117 if (SEQ_LT(nextrexmt, old_snd_max) &&
3118 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3119 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3120 }
3121 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3122 tp->snd_nxt = old_snd_nxt;
984263bc 3123 tp->snd_cwnd = ocwnd;
984263bc 3124}