tcp_input: Partly revert 98cb2337e1000153d84b64a5e9aabbecda8e8dbf
[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/*
984263bc
MD
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68 */
69
b1992928 70#include "opt_inet.h"
984263bc
MD
71#include "opt_inet6.h"
72#include "opt_ipsec.h"
73#include "opt_tcpdebug.h"
74#include "opt_tcp_input.h"
75
76#include <sys/param.h>
77#include <sys/systm.h>
78#include <sys/kernel.h>
79#include <sys/sysctl.h>
80#include <sys/malloc.h>
81#include <sys/mbuf.h>
82#include <sys/proc.h> /* for proc0 declaration */
83#include <sys/protosw.h>
84#include <sys/socket.h>
85#include <sys/socketvar.h>
86#include <sys/syslog.h>
3f9db7f8 87#include <sys/in_cksum.h>
984263bc 88
6cef7136
MD
89#include <sys/socketvar2.h>
90
984263bc 91#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
a00138cb 92#include <machine/stdarg.h>
984263bc
MD
93
94#include <net/if.h>
95#include <net/route.h>
96
97#include <netinet/in.h>
98#include <netinet/in_systm.h>
99#include <netinet/ip.h>
95b22adf 100#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
984263bc 101#include <netinet/in_var.h>
95b22adf 102#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
984263bc
MD
103#include <netinet/in_pcb.h>
104#include <netinet/ip_var.h>
105#include <netinet/ip6.h>
106#include <netinet/icmp6.h>
107#include <netinet6/nd6.h>
108#include <netinet6/ip6_var.h>
109#include <netinet6/in6_pcb.h>
110#include <netinet/tcp.h>
111#include <netinet/tcp_fsm.h>
112#include <netinet/tcp_seq.h>
113#include <netinet/tcp_timer.h>
a48c5dd5 114#include <netinet/tcp_timer2.h>
984263bc
MD
115#include <netinet/tcp_var.h>
116#include <netinet6/tcp6_var.h>
117#include <netinet/tcpip.h>
95b22adf 118
984263bc
MD
119#ifdef TCPDEBUG
120#include <netinet/tcp_debug.h>
121
95b22adf 122u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
984263bc 123struct tcphdr tcp_savetcp;
95b22adf 124#endif
984263bc
MD
125
126#ifdef FAST_IPSEC
bf844ffa
JH
127#include <netproto/ipsec/ipsec.h>
128#include <netproto/ipsec/ipsec6.h>
984263bc
MD
129#endif
130
131#ifdef IPSEC
132#include <netinet6/ipsec.h>
133#include <netinet6/ipsec6.h>
d2438d69 134#include <netproto/key/key.h>
95b22adf 135#endif
984263bc 136
984263bc
MD
137MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
138
984263bc 139static int log_in_vain = 0;
d24ce1dc 140SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
984263bc
MD
141 &log_in_vain, 0, "Log all incoming TCP connections");
142
143static int blackhole = 0;
144SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145 &blackhole, 0, "Do not send RST when dropping refused connections");
146
147int tcp_delack_enabled = 1;
d24ce1dc
JH
148SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149 &tcp_delack_enabled, 0,
984263bc
MD
150 "Delay ACK to try and piggyback it onto a data packet");
151
152#ifdef TCP_DROP_SYNFIN
153static int drop_synfin = 0;
154SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
156#endif
157
33abdd1c
MD
158static int tcp_do_limitedtransmit = 1;
159SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
161
91489f6b 162static int tcp_do_early_retransmit = 1;
8819433a
JH
163SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164 &tcp_do_early_retransmit, 0, "Early retransmit");
165
8acdb67c 166int tcp_aggregate_acks = 1;
72b37eeb
MD
167SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
169
8d573eef 170int tcp_do_rfc3390 = 1;
4b52d5ee
JH
171SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
172 &tcp_do_rfc3390, 0,
173 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
174
efd4b327
JH
175static int tcp_do_eifel_detect = 1;
176SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
177 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
178
8acdb67c
JH
179static int tcp_do_abc = 1;
180SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
181 &tcp_do_abc, 0,
182 "TCP Appropriate Byte Counting (RFC 3465)");
183
91489f6b
JH
184/*
185 * Define as tunable for easy testing with SACK on and off.
186 * Warning: do not change setting in the middle of an existing active TCP flow,
187 * else strange things might happen to that flow.
188 */
6e5bbdda 189int tcp_do_sack = 1;
91489f6b
JH
190SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
191 &tcp_do_sack, 0, "Enable SACK Algorithms");
192
193int tcp_do_smartsack = 1;
194SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
195 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
196
3edf7c37
RG
197SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
198 "TCP Segment Reassembly Queue");
199
200int tcp_reass_maxseg = 0;
201SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
202 &tcp_reass_maxseg, 0,
203 "Global maximum number of TCP Segments in Reassembly Queue");
204
205int tcp_reass_qsize = 0;
206SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
207 &tcp_reass_qsize, 0,
208 "Global number of TCP Segments currently in Reassembly Queue");
209
210static int tcp_reass_overflows = 0;
211SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
212 &tcp_reass_overflows, 0,
213 "Global number of TCP Segment Reassembly Queue Overflows");
214
5b0b9fa5
PA
215int tcp_do_autorcvbuf = 1;
216SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
217 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
218
219int tcp_autorcvbuf_inc = 16*1024;
220SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
221 &tcp_autorcvbuf_inc, 0,
222 "Incrementor step size of automatic receive buffer");
223
46e92930 224int tcp_autorcvbuf_max = 2*1024*1024;
5b0b9fa5
PA
225SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
226 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
227
c16aca65
SZ
228int tcp_sosnd_agglim = 2;
229SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_agglim, CTLFLAG_RW,
230 &tcp_sosnd_agglim, 0, "TCP sosend mbuf aggregation limit");
5b0b9fa5 231
f786a6ac 232int tcp_sosnd_async = 1;
f2a3782e
SZ
233SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_async, CTLFLAG_RW,
234 &tcp_sosnd_async, 0, "TCP asynchronized pru_send");
235
95b22adf 236static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
984263bc
MD
237static void tcp_pulloutofband(struct socket *,
238 struct tcphdr *, struct mbuf *, int);
239static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
240 struct mbuf *);
241static void tcp_xmit_timer(struct tcpcb *, int);
91489f6b
JH
242static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
243static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
01d3427a 244static int tcp_rmx_msl(const struct tcpcb *);
8651f7f8 245static void tcp_established(struct tcpcb *);
984263bc
MD
246
247/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
248#ifdef INET6
249#define ND6_HINT(tp) \
250do { \
251 if ((tp) && (tp)->t_inpcb && \
61896e3c 252 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
984263bc
MD
253 (tp)->t_inpcb->in6p_route.ro_rt) \
254 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
255} while (0)
256#else
257#define ND6_HINT(tp)
258#endif
259
260/*
261 * Indicate whether this ack should be delayed. We can delay the ack if
262 * - delayed acks are enabled and
263 * - there is no delayed ack timer in progress and
264 * - our last ack wasn't a 0-sized window. We never want to delay
265 * the ack that opens up a 0-sized window.
266 */
267#define DELAY_ACK(tp) \
a48c5dd5 268 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
61896e3c 269 !(tp->t_flags & TF_RXWIN0SENT))
984263bc 270
df9d7670
JH
271#define acceptable_window_update(tp, th, tiwin) \
272 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
273 (tp->snd_wl1 == th->th_seq && \
274 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
275 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
276
984263bc 277static int
95b22adf 278tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
984263bc
MD
279{
280 struct tseg_qent *q;
281 struct tseg_qent *p = NULL;
984263bc
MD
282 struct tseg_qent *te;
283 struct socket *so = tp->t_inpcb->inp_socket;
284 int flags;
285
286 /*
61896e3c 287 * Call with th == NULL after become established to
984263bc
MD
288 * force pre-ESTABLISHED data up to user socket.
289 */
61896e3c 290 if (th == NULL)
984263bc
MD
291 goto present;
292
3edf7c37
RG
293 /*
294 * Limit the number of segments in the reassembly queue to prevent
295 * holding on to too many segments (and thus running out of mbufs).
296 * Make sure to let the missing segment through which caused this
297 * queue. Always keep one global queue entry spare to be able to
298 * process the missing segment.
299 */
300 if (th->th_seq != tp->rcv_nxt &&
301 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
302 tcp_reass_overflows++;
303 tcpstat.tcps_rcvmemdrop++;
304 m_freem(m);
91489f6b
JH
305 /* no SACK block to report */
306 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
3edf7c37
RG
307 return (0);
308 }
309
ba4e3dbe 310 /* Allocate a new queue entry. */
884717e1 311 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
984263bc
MD
312 if (te == NULL) {
313 tcpstat.tcps_rcvmemdrop++;
314 m_freem(m);
91489f6b
JH
315 /* no SACK block to report */
316 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
984263bc
MD
317 return (0);
318 }
2d23a8be 319 atomic_add_int(&tcp_reass_qsize, 1);
984263bc
MD
320
321 /*
322 * Find a segment which begins after this one does.
323 */
324 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
325 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
326 break;
327 p = q;
328 }
329
330 /*
331 * If there is a preceding segment, it may provide some of
332 * our data already. If so, drop the data from the incoming
333 * segment. If it provides all of our data, drop us.
334 */
335 if (p != NULL) {
76cf5e41 336 tcp_seq_diff_t i;
91489f6b 337
984263bc
MD
338 /* conversion to int (in i) handles seq wraparound */
339 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
91489f6b
JH
340 if (i > 0) { /* overlaps preceding segment */
341 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
342 /* enclosing block starts w/ preceding segment */
343 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
984263bc 344 if (i >= *tlenp) {
91489f6b 345 /* preceding encloses incoming segment */
3a5d999b
SZ
346 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
347 p->tqe_th->th_seq + p->tqe_len,
348 p->tqe_th->th_flags);
984263bc
MD
349 tcpstat.tcps_rcvduppack++;
350 tcpstat.tcps_rcvdupbyte += *tlenp;
351 m_freem(m);
efda3bd0 352 kfree(te, M_TSEGQ);
2d23a8be 353 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
354 /*
355 * Try to present any queued data
356 * at the left window edge to the user.
357 * This is needed after the 3-WHS
358 * completes.
359 */
360 goto present; /* ??? */
361 }
362 m_adj(m, i);
363 *tlenp -= i;
364 th->th_seq += i;
91489f6b 365 /* incoming segment end is enclosing block end */
3a5d999b
SZ
366 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
367 th->th_seq + *tlenp, th->th_flags);
91489f6b
JH
368 /* trim end of reported D-SACK block */
369 tp->reportblk.rblk_end = th->th_seq;
984263bc
MD
370 }
371 }
372 tcpstat.tcps_rcvoopack++;
373 tcpstat.tcps_rcvoobyte += *tlenp;
374
375 /*
376 * While we overlap succeeding segments trim them or,
377 * if they are completely covered, dequeue them.
378 */
379 while (q) {
76cf5e41 380 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
91489f6b 381 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
3a5d999b 382 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
91489f6b
JH
383 struct tseg_qent *nq;
384
984263bc
MD
385 if (i <= 0)
386 break;
91489f6b
JH
387 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
388 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
389 tp->encloseblk = tp->reportblk;
390 /* report trailing duplicate D-SACK segment */
391 tp->reportblk.rblk_start = q->tqe_th->th_seq;
392 }
393 if ((tp->t_flags & TF_ENCLOSESEG) &&
3a5d999b 394 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
91489f6b 395 /* extend enclosing block if one exists */
3a5d999b 396 tp->encloseblk.rblk_end = qend_sack;
91489f6b 397 }
984263bc
MD
398 if (i < q->tqe_len) {
399 q->tqe_th->th_seq += i;
400 q->tqe_len -= i;
401 m_adj(q->tqe_m, i);
402 break;
403 }
404
405 nq = LIST_NEXT(q, tqe_q);
406 LIST_REMOVE(q, tqe_q);
407 m_freem(q->tqe_m);
efda3bd0 408 kfree(q, M_TSEGQ);
2d23a8be 409 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
410 q = nq;
411 }
412
413 /* Insert the new segment queue entry into place. */
414 te->tqe_m = m;
415 te->tqe_th = th;
416 te->tqe_len = *tlenp;
417
91489f6b
JH
418 /* check if can coalesce with following segment */
419 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
420 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
3a5d999b 421 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
91489f6b
JH
422
423 te->tqe_len += q->tqe_len;
a174690a 424 if (q->tqe_th->th_flags & TH_FIN)
f16f8cc3 425 te->tqe_th->th_flags |= TH_FIN;
91489f6b 426 m_cat(te->tqe_m, q->tqe_m);
3a5d999b 427 tp->encloseblk.rblk_end = tend_sack;
91489f6b
JH
428 /*
429 * When not reporting a duplicate segment, use
430 * the larger enclosing block as the SACK block.
431 */
432 if (!(tp->t_flags & TF_DUPSEG))
3a5d999b 433 tp->reportblk.rblk_end = tend_sack;
91489f6b 434 LIST_REMOVE(q, tqe_q);
efda3bd0 435 kfree(q, M_TSEGQ);
2d23a8be 436 atomic_add_int(&tcp_reass_qsize, -1);
91489f6b
JH
437 }
438
984263bc
MD
439 if (p == NULL) {
440 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
441 } else {
91489f6b
JH
442 /* check if can coalesce with preceding segment */
443 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
a174690a
JH
444 p->tqe_len += te->tqe_len;
445 m_cat(p->tqe_m, te->tqe_m);
91489f6b
JH
446 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
447 /*
448 * When not reporting a duplicate segment, use
449 * the larger enclosing block as the SACK block.
450 */
451 if (!(tp->t_flags & TF_DUPSEG))
452 tp->reportblk.rblk_start = p->tqe_th->th_seq;
efda3bd0 453 kfree(te, M_TSEGQ);
2d23a8be
MD
454 atomic_add_int(&tcp_reass_qsize, -1);
455 } else {
91489f6b 456 LIST_INSERT_AFTER(p, te, tqe_q);
2d23a8be 457 }
984263bc
MD
458 }
459
460present:
461 /*
462 * Present data to user, advancing rcv_nxt through
463 * completed sequence space.
464 */
465 if (!TCPS_HAVEESTABLISHED(tp->t_state))
466 return (0);
467 q = LIST_FIRST(&tp->t_segq);
61896e3c 468 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
984263bc 469 return (0);
91489f6b
JH
470 tp->rcv_nxt += q->tqe_len;
471 if (!(tp->t_flags & TF_DUPSEG)) {
472 /* no SACK block to report since ACK advanced */
473 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
474 }
475 /* no enclosing block to report since ACK advanced */
476 tp->t_flags &= ~TF_ENCLOSESEG;
477 flags = q->tqe_th->th_flags & TH_FIN;
478 LIST_REMOVE(q, tqe_q);
479 KASSERT(LIST_EMPTY(&tp->t_segq) ||
480 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
481 ("segment not coalesced"));
6cef7136 482 if (so->so_state & SS_CANTRCVMORE) {
91489f6b 483 m_freem(q->tqe_m);
6cef7136
MD
484 } else {
485 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 486 ssb_appendstream(&so->so_rcv, q->tqe_m);
6cef7136
MD
487 lwkt_reltoken(&so->so_rcv.ssb_token);
488 }
efda3bd0 489 kfree(q, M_TSEGQ);
2d23a8be 490 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
491 ND6_HINT(tp);
492 sorwakeup(so);
493 return (flags);
494}
495
496/*
497 * TCP input routine, follows pages 65-76 of the
498 * protocol specification dated September, 1981 very closely.
499 */
500#ifdef INET6
501int
95b22adf 502tcp6_input(struct mbuf **mp, int *offp, int proto)
984263bc 503{
2256ba69 504 struct mbuf *m = *mp;
984263bc
MD
505 struct in6_ifaddr *ia6;
506
507 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
508
509 /*
510 * draft-itojun-ipv6-tcp-to-anycast
511 * better place to put this in?
512 */
513 ia6 = ip6_getdstifaddr(m);
d24ce1dc 514 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
984263bc
MD
515 struct ip6_hdr *ip6;
516
517 ip6 = mtod(m, struct ip6_hdr *);
518 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
f23061d4 519 offsetof(struct ip6_hdr, ip6_dst));
d24ce1dc 520 return (IPPROTO_DONE);
984263bc
MD
521 }
522
002c1265 523 tcp_input(mp, offp, proto);
d24ce1dc 524 return (IPPROTO_DONE);
984263bc
MD
525}
526#endif
527
002c1265
MD
528int
529tcp_input(struct mbuf **mp, int *offp, int proto)
984263bc 530{
002c1265 531 int off0;
2256ba69
RG
532 struct tcphdr *th;
533 struct ip *ip = NULL;
534 struct ipovly *ipov;
535 struct inpcb *inp = NULL;
984263bc
MD
536 u_char *optp = NULL;
537 int optlen = 0;
b1992928
MD
538 int tlen, off;
539 int len = 0;
984263bc 540 int drop_hdrlen;
2256ba69
RG
541 struct tcpcb *tp = NULL;
542 int thflags;
4090d6ff 543 struct socket *so = NULL;
61896e3c
JH
544 int todrop, acked;
545 boolean_t ourfinisacked, needoutput = FALSE;
984263bc 546 u_long tiwin;
61896e3c 547 int recvwin;
984263bc 548 struct tcpopt to; /* options in this segment */
984263bc
MD
549 struct sockaddr_in *next_hop = NULL;
550 int rstreason; /* For badport_bandlim accounting purposes */
d371a63a 551 int cpu;
984263bc 552 struct ip6_hdr *ip6 = NULL;
002c1265 553 struct mbuf *m;
984263bc 554#ifdef INET6
d24ce1dc 555 boolean_t isipv6;
984263bc 556#else
d24ce1dc 557 const boolean_t isipv6 = FALSE;
984263bc
MD
558#endif
559#ifdef TCPDEBUG
560 short ostate = 0;
561#endif
562
002c1265
MD
563 off0 = *offp;
564 m = *mp;
565 *mp = NULL;
a00138cb 566
d24ce1dc
JH
567 tcpstat.tcps_rcvtotal++;
568
5de23090
SZ
569 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
570 struct m_tag *mtag;
571
572 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
573 KKASSERT(mtag != NULL);
574 next_hop = m_tag_data(mtag);
984263bc 575 }
d24ce1dc 576
984263bc 577#ifdef INET6
d24ce1dc 578 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
984263bc 579#endif
984263bc
MD
580
581 if (isipv6) {
582 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
583 ip6 = mtod(m, struct ip6_hdr *);
407e896e 584 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
984263bc
MD
585 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
586 tcpstat.tcps_rcvbadsum++;
587 goto drop;
588 }
589 th = (struct tcphdr *)((caddr_t)ip6 + off0);
590
591 /*
592 * Be proactive about unspecified IPv6 address in source.
593 * As we use all-zero to indicate unbounded/unconnected pcb,
594 * unspecified IPv6 address can be used to confuse us.
595 *
596 * Note that packets with unspecified IPv6 destination is
597 * already dropped in ip6_input.
598 */
599 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
600 /* XXX stat */
601 goto drop;
602 }
603 } else {
604 /*
605 * Get IP and TCP header together in first mbuf.
606 * Note: IP leaves IP header in first mbuf.
607 */
608 if (off0 > sizeof(struct ip)) {
bddf0751 609 ip_stripoptions(m);
984263bc
MD
610 off0 = sizeof(struct ip);
611 }
55d829f8
JH
612 /* already checked and pulled up in ip_demux() */
613 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
61896e3c 614 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
984263bc
MD
615 ip = mtod(m, struct ip *);
616 ipov = (struct ipovly *)ip;
617 th = (struct tcphdr *)((caddr_t)ip + off0);
618 tlen = ip->ip_len;
619
620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
622 th->th_sum = m->m_pkthdr.csum_data;
623 else
624 th->th_sum = in_pseudo(ip->ip_src.s_addr,
625 ip->ip_dst.s_addr,
626 htonl(m->m_pkthdr.csum_data +
627 ip->ip_len +
628 IPPROTO_TCP));
629 th->th_sum ^= 0xffff;
630 } else {
631 /*
632 * Checksum extended TCP header and data.
633 */
634 len = sizeof(struct ip) + tlen;
407e896e 635 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
984263bc
MD
636 ipov->ih_len = (u_short)tlen;
637 ipov->ih_len = htons(ipov->ih_len);
638 th->th_sum = in_cksum(m, len);
639 }
640 if (th->th_sum) {
641 tcpstat.tcps_rcvbadsum++;
642 goto drop;
643 }
644#ifdef INET6
645 /* Re-initialization for later version check */
646 ip->ip_v = IPVERSION;
647#endif
648 }
649
650 /*
651 * Check that TCP offset makes sense,
652 * pull out TCP options and adjust length. XXX
653 */
654 off = th->th_off << 2;
55d829f8
JH
655 /* already checked and pulled up in ip_demux() */
656 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
61896e3c 657 ("bad TCP data offset %d (tlen %d)", off, tlen));
984263bc
MD
658 tlen -= off; /* tlen is used instead of ti->ti_len */
659 if (off > sizeof(struct tcphdr)) {
660 if (isipv6) {
002c1265 661 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
984263bc
MD
662 ip6 = mtod(m, struct ip6_hdr *);
663 th = (struct tcphdr *)((caddr_t)ip6 + off0);
664 } else {
55d829f8
JH
665 /* already pulled up in ip_demux() */
666 KASSERT(m->m_len >= sizeof(struct ip) + off,
61896e3c
JH
667 ("TCP header and options not in one mbuf: "
668 "m_len %d, off %d", m->m_len, off));
984263bc
MD
669 }
670 optlen = off - sizeof(struct tcphdr);
671 optp = (u_char *)(th + 1);
672 }
673 thflags = th->th_flags;
674
675#ifdef TCP_DROP_SYNFIN
676 /*
677 * If the drop_synfin option is enabled, drop all packets with
678 * both the SYN and FIN bits set. This prevents e.g. nmap from
679 * identifying the TCP/IP stack.
680 *
681 * This is a violation of the TCP specification.
682 */
61896e3c 683 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
984263bc
MD
684 goto drop;
685#endif
686
687 /*
688 * Convert TCP protocol specific fields to host format.
689 */
690 th->th_seq = ntohl(th->th_seq);
691 th->th_ack = ntohl(th->th_ack);
692 th->th_win = ntohs(th->th_win);
693 th->th_urp = ntohs(th->th_urp);
694
695 /*
d24ce1dc 696 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
984263bc
MD
697 * until after ip6_savecontrol() is called and before other functions
698 * which don't want those proto headers.
699 * Because ip6_savecontrol() is going to parse the mbuf to
700 * search for data to be passed up to user-land, it wants mbuf
701 * parameters to be unchanged.
702 * XXX: the call of ip6_savecontrol() has been obsoleted based on
703 * latest version of the advanced API (20020110).
704 */
705 drop_hdrlen = off0 + off;
706
707 /*
708 * Locate pcb for segment.
709 */
710findpcb:
711 /* IPFIREWALL_FORWARD section */
d24ce1dc 712 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
984263bc
MD
713 /*
714 * Transparently forwarded. Pretend to be the destination.
d24ce1dc 715 * already got one like this?
984263bc 716 */
6ca1a1cd 717 cpu = mycpu->gd_cpuid;
6ca1a1cd 718 inp = in_pcblookup_hash(&tcbinfo[cpu],
d371a63a 719 ip->ip_src, th->th_sport,
984263bc
MD
720 ip->ip_dst, th->th_dport,
721 0, m->m_pkthdr.rcvif);
722 if (!inp) {
83be63fe
JH
723 /*
724 * It's new. Try to find the ambushing socket.
725 */
726
727 /*
728 * The rest of the ipfw code stores the port in
729 * host order. XXX
730 * (The IP address is still in network order.)
731 */
732 in_port_t dport = next_hop->sin_port ?
733 htons(next_hop->sin_port) :
734 th->th_dport;
735
d371a63a 736 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
83be63fe 737 next_hop->sin_addr.s_addr, dport);
d371a63a 738 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc 739 ip->ip_src, th->th_sport,
83be63fe 740 next_hop->sin_addr, dport,
984263bc
MD
741 1, m->m_pkthdr.rcvif);
742 }
743 } else {
6ca1a1cd 744 if (isipv6) {
d371a63a 745 inp = in6_pcblookup_hash(&tcbinfo[0],
984263bc
MD
746 &ip6->ip6_src, th->th_sport,
747 &ip6->ip6_dst, th->th_dport,
748 1, m->m_pkthdr.rcvif);
6ca1a1cd 749 } else {
6ca1a1cd 750 cpu = mycpu->gd_cpuid;
6ca1a1cd 751 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc
MD
752 ip->ip_src, th->th_sport,
753 ip->ip_dst, th->th_dport,
754 1, m->m_pkthdr.rcvif);
6ca1a1cd 755 }
984263bc 756 }
984263bc
MD
757
758 /*
759 * If the state is CLOSED (i.e., TCB does not exist) then
760 * all data in the incoming segment is discarded.
761 * If the TCB exists but is in CLOSED state, it is embryonic,
762 * but should either do a listen or a connect soon.
763 */
764 if (inp == NULL) {
765 if (log_in_vain) {
766#ifdef INET6
767 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
768#else
1141eb20
JH
769 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
770 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
984263bc
MD
771#endif
772 if (isipv6) {
773 strcpy(dbuf, "[");
984263bc 774 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
984263bc 775 strcat(dbuf, "]");
95b22adf
JH
776 strcpy(sbuf, "[");
777 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
984263bc
MD
778 strcat(sbuf, "]");
779 } else {
780 strcpy(dbuf, inet_ntoa(ip->ip_dst));
781 strcpy(sbuf, inet_ntoa(ip->ip_src));
782 }
783 switch (log_in_vain) {
784 case 1:
61896e3c 785 if (!(thflags & TH_SYN))
984263bc
MD
786 break;
787 case 2:
788 log(LOG_INFO,
789 "Connection attempt to TCP %s:%d "
790 "from %s:%d flags:0x%02x\n",
791 dbuf, ntohs(th->th_dport), sbuf,
792 ntohs(th->th_sport), thflags);
793 break;
794 default:
795 break;
796 }
797 }
d24ce1dc 798 if (blackhole) {
984263bc
MD
799 switch (blackhole) {
800 case 1:
801 if (thflags & TH_SYN)
802 goto drop;
803 break;
804 case 2:
805 goto drop;
806 default:
807 goto drop;
808 }
809 }
810 rstreason = BANDLIM_RST_CLOSEDPORT;
811 goto dropwithreset;
812 }
61896e3c
JH
813
814#ifdef IPSEC
815 if (isipv6) {
816 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
817 ipsec6stat.in_polvio++;
818 goto drop;
819 }
820 } else {
821 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
822 ipsecstat.in_polvio++;
823 goto drop;
824 }
825 }
826#endif
827#ifdef FAST_IPSEC
828 if (isipv6) {
829 if (ipsec6_in_reject(m, inp))
830 goto drop;
831 } else {
832 if (ipsec4_in_reject(m, inp))
833 goto drop;
834 }
835#endif
95926362
MD
836 /* Check the minimum TTL for socket. */
837#ifdef INET6
838 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
839 goto drop;
840#endif
61896e3c 841
984263bc
MD
842 tp = intotcpcb(inp);
843 if (tp == NULL) {
844 rstreason = BANDLIM_RST_CLOSEDPORT;
845 goto dropwithreset;
846 }
eb594563 847 if (tp->t_state <= TCPS_CLOSED)
984263bc
MD
848 goto drop;
849
850 /* Unscale the window into a 32-bit value. */
95b22adf 851 if (!(thflags & TH_SYN))
984263bc
MD
852 tiwin = th->th_win << tp->snd_scale;
853 else
854 tiwin = th->th_win;
855
856 so = inp->inp_socket;
47654766 857
984263bc 858#ifdef TCPDEBUG
47654766
JH
859 if (so->so_options & SO_DEBUG) {
860 ostate = tp->t_state;
861 if (isipv6)
95b22adf 862 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
47654766 863 else
95b22adf 864 bcopy(ip, tcp_saveipgen, sizeof(*ip));
47654766
JH
865 tcp_savetcp = *th;
866 }
984263bc 867#endif
47654766 868
407e896e 869 bzero(&to, sizeof to);
d24ce1dc 870
47654766
JH
871 if (so->so_options & SO_ACCEPTCONN) {
872 struct in_conninfo inc;
873
984263bc 874#ifdef INET6
d24ce1dc 875 inc.inc_isipv6 = (isipv6 == TRUE);
984263bc
MD
876#endif
877 if (isipv6) {
878 inc.inc6_faddr = ip6->ip6_src;
879 inc.inc6_laddr = ip6->ip6_dst;
880 inc.inc6_route.ro_rt = NULL; /* XXX */
881 } else {
882 inc.inc_faddr = ip->ip_src;
883 inc.inc_laddr = ip->ip_dst;
884 inc.inc_route.ro_rt = NULL; /* XXX */
885 }
886 inc.inc_fport = th->th_sport;
887 inc.inc_lport = th->th_dport;
888
61896e3c
JH
889 /*
890 * If the state is LISTEN then ignore segment if it contains
984263bc
MD
891 * a RST. If the segment contains an ACK then it is bad and
892 * send a RST. If it does not contain a SYN then it is not
893 * interesting; drop it.
894 *
895 * If the state is SYN_RECEIVED (syncache) and seg contains
896 * an ACK, but not for our SYN/ACK, send a RST. If the seg
897 * contains a RST, check the sequence number to see if it
898 * is a valid reset segment.
899 */
61896e3c
JH
900 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
901 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
984263bc
MD
902 if (!syncache_expand(&inc, th, &so, m)) {
903 /*
904 * No syncache entry, or ACK was not
905 * for our SYN/ACK. Send a RST.
906 */
907 tcpstat.tcps_badsyn++;
908 rstreason = BANDLIM_RST_OPENPORT;
909 goto dropwithreset;
910 }
2d23a8be
MD
911
912 /*
913 * Could not complete 3-way handshake,
914 * connection is being closed down, and
915 * syncache will free mbuf.
916 */
984263bc 917 if (so == NULL)
002c1265 918 return(IPPROTO_DONE);
2d23a8be
MD
919
920 /*
921 * We must be in the correct protocol thread
922 * for this connection.
923 */
924 KKASSERT(so->so_port == &curthread->td_msgport);
925
984263bc
MD
926 /*
927 * Socket is created in state SYN_RECEIVED.
928 * Continue processing segment.
929 */
ed894f8c 930 inp = so->so_pcb;
984263bc
MD
931 tp = intotcpcb(inp);
932 /*
933 * This is what would have happened in
934 * tcp_output() when the SYN,ACK was sent.
935 */
936 tp->snd_up = tp->snd_una;
937 tp->snd_max = tp->snd_nxt = tp->iss + 1;
938 tp->last_ack_sent = tp->rcv_nxt;
939/*
940 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
941 * until the _second_ ACK is received:
942 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
943 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
61896e3c 944 * move to ESTAB, set snd_wnd to tiwin.
d24ce1dc 945 */
984263bc
MD
946 tp->snd_wnd = tiwin; /* unscaled */
947 goto after_listen;
948 }
949 if (thflags & TH_RST) {
950 syncache_chkrst(&inc, th);
951 goto drop;
952 }
953 if (thflags & TH_ACK) {
954 syncache_badack(&inc);
955 tcpstat.tcps_badsyn++;
956 rstreason = BANDLIM_RST_OPENPORT;
957 goto dropwithreset;
958 }
959 goto drop;
960 }
961
962 /*
61896e3c 963 * Segment's flags are (SYN) or (SYN | FIN).
984263bc
MD
964 */
965#ifdef INET6
966 /*
967 * If deprecated address is forbidden,
968 * we do not accept SYN to deprecated interface
969 * address to prevent any new inbound connection from
970 * getting established.
971 * When we do not accept SYN, we send a TCP RST,
972 * with deprecated source address (instead of dropping
973 * it). We compromise it as it is much better for peer
974 * to send a RST, and RST will be the final packet
975 * for the exchange.
976 *
977 * If we do not forbid deprecated addresses, we accept
978 * the SYN packet. RFC2462 does not suggest dropping
979 * SYN in this case.
980 * If we decipher RFC2462 5.5.4, it says like this:
981 * 1. use of deprecated addr with existing
982 * communication is okay - "SHOULD continue to be
983 * used"
984 * 2. use of it with new communication:
985 * (2a) "SHOULD NOT be used if alternate address
61896e3c 986 * with sufficient scope is available"
984263bc
MD
987 * (2b) nothing mentioned otherwise.
988 * Here we fall into (2b) case as we have no choice in
989 * our source address selection - we must obey the peer.
990 *
991 * The wording in RFC2462 is confusing, and there are
992 * multiple description text for deprecated address
993 * handling - worse, they are not exactly the same.
994 * I believe 5.5.4 is the best one, so we follow 5.5.4.
995 */
996 if (isipv6 && !ip6_use_deprecated) {
997 struct in6_ifaddr *ia6;
998
999 if ((ia6 = ip6_getdstifaddr(m)) &&
1000 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1001 tp = NULL;
1002 rstreason = BANDLIM_RST_OPENPORT;
1003 goto dropwithreset;
1004 }
1005 }
1006#endif
1007 /*
1008 * If it is from this socket, drop it, it must be forged.
1009 * Don't bother responding if the destination was a broadcast.
1010 */
1011 if (th->th_dport == th->th_sport) {
1012 if (isipv6) {
1013 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1014 &ip6->ip6_src))
1015 goto drop;
1016 } else {
1017 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1018 goto drop;
1019 }
1020 }
1021 /*
1022 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1023 *
1024 * Note that it is quite possible to receive unicast
1025 * link-layer packets with a broadcast IP address. Use
1026 * in_broadcast() to find them.
1027 */
61896e3c 1028 if (m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
1029 goto drop;
1030 if (isipv6) {
1031 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1032 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1033 goto drop;
1034 } else {
1035 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1036 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1037 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1038 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1039 goto drop;
1040 }
1041 /*
1042 * SYN appears to be valid; create compressed TCP state
1043 * for syncache, or perform t/tcp connection.
1044 */
1045 if (so->so_qlen <= so->so_qlimit) {
95b22adf 1046 tcp_dooptions(&to, optp, optlen, TRUE);
b09567cc 1047 if (!syncache_add(&inc, &to, th, so, m))
984263bc 1048 goto drop;
2d23a8be
MD
1049
1050 /*
1051 * Entry added to syncache, mbuf used to
1052 * send SYN,ACK packet.
1053 */
b09567cc 1054 return(IPPROTO_DONE);
984263bc
MD
1055 }
1056 goto drop;
1057 }
984263bc 1058
2d23a8be
MD
1059after_listen:
1060 /*
1061 * Should not happen - syncache should pick up these connections.
1062 *
1063 * Once we are past handling listen sockets we must be in the
1064 * correct protocol processing thread.
1065 */
61896e3c 1066 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
2d23a8be 1067 KKASSERT(so->so_port == &curthread->td_msgport);
984263bc
MD
1068
1069 /*
5b0b9fa5
PA
1070 * This is the second part of the MSS DoS prevention code (after
1071 * minmss on the sending side) and it deals with too many too small
1072 * tcp packets in a too short timeframe (1 second).
1073 *
153d34ac
MD
1074 * XXX Removed. This code was crap. It does not scale to network
1075 * speed, and default values break NFS. Gone.
5b0b9fa5 1076 */
153d34ac 1077 /* REMOVED */
5b0b9fa5
PA
1078
1079 /*
984263bc 1080 * Segment received on connection.
efca2b8e
MD
1081 *
1082 * Reset idle time and keep-alive timer. Don't waste time if less
0ecd93f9 1083 * then a second has elapsed.
984263bc 1084 */
0ecd93f9
MD
1085 if ((int)(ticks - tp->t_rcvtime) > hz)
1086 tcp_timer_keep_activity(tp, thflags);
984263bc
MD
1087
1088 /*
1089 * Process options.
1090 * XXX this is tradtitional behavior, may need to be cleaned up.
1091 */
95b22adf 1092 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
ad0af98b 1093 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
984263bc
MD
1094 if (to.to_flags & TOF_SCALE) {
1095 tp->t_flags |= TF_RCVD_SCALE;
1096 tp->requested_s_scale = to.to_requested_s_scale;
1097 }
1098 if (to.to_flags & TOF_TS) {
1099 tp->t_flags |= TF_RCVD_TSTMP;
1100 tp->ts_recent = to.to_tsval;
1101 tp->ts_recent_age = ticks;
1102 }
f34fd0f2
SZ
1103 if (!(to.to_flags & TOF_MSS))
1104 to.to_mss = 0;
1105 tcp_mss(tp, to.to_mss);
91489f6b
JH
1106 /*
1107 * Only set the TF_SACK_PERMITTED per-connection flag
1108 * if we got a SACK_PERMITTED option from the other side
1109 * and the global tcp_do_sack variable is true.
1110 */
1111 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1112 tp->t_flags |= TF_SACK_PERMITTED;
984263bc
MD
1113 }
1114
1115 /*
1116 * Header prediction: check for the two common cases
1117 * of a uni-directional data xfer. If the packet has
1118 * no control flags, is in-sequence, the window didn't
1119 * change and we're not retransmitting, it's a
1120 * candidate. If the length is zero and the ack moved
1121 * forward, we're the sender side of the xfer. Just
1122 * free the data acked & wake any higher level process
1123 * that was blocked waiting for space. If the length
1124 * is non-zero and the ack didn't move, we're the
1125 * receiver side. If we're getting packets in-order
1126 * (the reassembly queue is empty), add the data to
1127 * the socket buffer and note that we need a delayed ack.
1128 * Make sure that the hidden state-flags are also off.
1129 * Since we check for TCPS_ESTABLISHED above, it can only
1130 * be TH_NEEDSYN.
1131 */
1132 if (tp->t_state == TCPS_ESTABLISHED &&
1133 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
61896e3c
JH
1134 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1135 (!(to.to_flags & TOF_TS) ||
984263bc 1136 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
984263bc 1137 th->th_seq == tp->rcv_nxt &&
984263bc
MD
1138 tp->snd_nxt == tp->snd_max) {
1139
1140 /*
1141 * If last ACK falls within this segment's sequence numbers,
1142 * record the timestamp.
1143 * NOTE that the test is modified according to the latest
1144 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1145 */
61896e3c 1146 if ((to.to_flags & TOF_TS) &&
984263bc
MD
1147 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1148 tp->ts_recent_age = ticks;
1149 tp->ts_recent = to.to_tsval;
1150 }
1151
1152 if (tlen == 0) {
1153 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1154 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1155 tp->snd_cwnd >= tp->snd_wnd &&
95b22adf 1156 !IN_FASTRECOVERY(tp)) {
984263bc 1157 /*
be23faf1 1158 * This is a pure ack for outstanding data.
984263bc
MD
1159 */
1160 ++tcpstat.tcps_predack;
1161 /*
1162 * "bad retransmit" recovery
bfdb979e
JH
1163 *
1164 * If Eifel detection applies, then
1165 * it is deterministic, so use it
1166 * unconditionally over the old heuristic.
1167 * Otherwise, fall back to the old heuristic.
984263bc 1168 */
bfdb979e
JH
1169 if (tcp_do_eifel_detect &&
1170 (to.to_flags & TOF_TS) && to.to_tsecr &&
1171 (tp->t_flags & TF_FIRSTACCACK)) {
1172 /* Eifel detection applicable. */
1173 if (to.to_tsecr < tp->t_rexmtTS) {
1174 tcp_revert_congestion_state(tp);
1175 ++tcpstat.tcps_eifeldetected;
928c3291
SZ
1176 if (tp->t_rxtshift != 1 ||
1177 ticks >= tp->t_badrxtwin)
1178 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
1179 }
1180 } else if (tp->t_rxtshift == 1 &&
1181 ticks < tp->t_badrxtwin) {
1182 tcp_revert_congestion_state(tp);
1183 ++tcpstat.tcps_rttdetected;
984263bc 1184 }
8819433a
JH
1185 tp->t_flags &= ~(TF_FIRSTACCACK |
1186 TF_FASTREXMT | TF_EARLYREXMT);
984263bc
MD
1187 /*
1188 * Recalculate the retransmit timer / rtt.
1189 *
d24ce1dc 1190 * Some machines (certain windows boxes)
984263bc
MD
1191 * send broken timestamp replies during the
1192 * SYN+ACK phase, ignore timestamps of 0.
1193 */
95b22adf 1194 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
984263bc 1195 tcp_xmit_timer(tp,
95b22adf 1196 ticks - to.to_tsecr + 1);
984263bc 1197 } else if (tp->t_rtttime &&
95b22adf 1198 SEQ_GT(th->th_ack, tp->t_rtseq)) {
984263bc
MD
1199 tcp_xmit_timer(tp,
1200 ticks - tp->t_rtttime);
1201 }
1202 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1203 acked = th->th_ack - tp->snd_una;
1204 tcpstat.tcps_rcvackpack++;
1205 tcpstat.tcps_rcvackbyte += acked;
6d49aa6f 1206 sbdrop(&so->so_snd.sb, acked);
cfb3f3f4 1207 tp->snd_recover = th->th_ack - 1;
9845754e 1208 tp->snd_una = th->th_ack;
984263bc 1209 tp->t_dupacks = 0;
df9d7670
JH
1210 /*
1211 * Update window information.
1212 */
1213 if (tiwin != tp->snd_wnd &&
1214 acceptable_window_update(tp, th, tiwin)) {
1215 /* keep track of pure window updates */
1216 if (tp->snd_wl2 == th->th_ack &&
1217 tiwin > tp->snd_wnd)
1218 tcpstat.tcps_rcvwinupd++;
1219 tp->snd_wnd = tiwin;
1220 tp->snd_wl1 = th->th_seq;
1221 tp->snd_wl2 = th->th_ack;
1222 if (tp->snd_wnd > tp->max_sndwnd)
1223 tp->max_sndwnd = tp->snd_wnd;
1224 }
984263bc
MD
1225 m_freem(m);
1226 ND6_HINT(tp); /* some progress has been done */
984263bc
MD
1227 /*
1228 * If all outstanding data are acked, stop
1229 * retransmit timer, otherwise restart timer
1230 * using current (possibly backed-off) value.
1231 * If process is waiting for space,
1232 * wakeup/selwakeup/signal. If data
1233 * are ready to send, let tcp_output
1234 * decide between more output or persist.
1235 */
a48c5dd5
SZ
1236 if (tp->snd_una == tp->snd_max) {
1237 tcp_callout_stop(tp, tp->tt_rexmt);
1238 } else if (!tcp_callout_active(tp,
1239 tp->tt_persist)) {
1240 tcp_callout_reset(tp, tp->tt_rexmt,
1241 tp->t_rxtcur, tcp_timer_rexmt);
1242 }
984263bc 1243 sowwakeup(so);
6d49aa6f 1244 if (so->so_snd.ssb_cc > 0)
f23061d4 1245 tcp_output(tp);
002c1265 1246 return(IPPROTO_DONE);
984263bc 1247 }
df9d7670
JH
1248 } else if (tiwin == tp->snd_wnd &&
1249 th->th_ack == tp->snd_una &&
984263bc 1250 LIST_EMPTY(&tp->t_segq) &&
6d49aa6f 1251 tlen <= ssb_space(&so->so_rcv)) {
46e92930 1252 u_long newsize = 0; /* automatic sockbuf scaling */
984263bc 1253 /*
be23faf1 1254 * This is a pure, in-sequence data packet
984263bc
MD
1255 * with nothing on the reassembly queue and
1256 * we have enough buffer space to take it.
1257 */
1258 ++tcpstat.tcps_preddat;
1259 tp->rcv_nxt += tlen;
1260 tcpstat.tcps_rcvpack++;
1261 tcpstat.tcps_rcvbyte += tlen;
1262 ND6_HINT(tp); /* some progress has been done */
5b0b9fa5
PA
1263 /*
1264 * Automatic sizing of receive socket buffer. Often the send
1265 * buffer size is not optimally adjusted to the actual network
1266 * conditions at hand (delay bandwidth product). Setting the
1267 * buffer size too small limits throughput on links with high
1268 * bandwidth and high delay (eg. trans-continental/oceanic links).
1269 *
1270 * On the receive side the socket buffer memory is only rarely
1271 * used to any significant extent. This allows us to be much
1272 * more aggressive in scaling the receive socket buffer. For
1273 * the case that the buffer space is actually used to a large
1274 * extent and we run out of kernel memory we can simply drop
1275 * the new segments; TCP on the sender will just retransmit it
1276 * later. Setting the buffer size too big may only consume too
1277 * much kernel memory if the application doesn't read() from
1278 * the socket or packet loss or reordering makes use of the
1279 * reassembly queue.
1280 *
1281 * The criteria to step up the receive buffer one notch are:
1282 * 1. the number of bytes received during the time it takes
1283 * one timestamp to be reflected back to us (the RTT);
1284 * 2. received bytes per RTT is within seven eighth of the
1285 * current socket buffer size;
1286 * 3. receive buffer size has not hit maximal automatic size;
1287 *
1288 * This algorithm does one step per RTT at most and only if
1289 * we receive a bulk stream w/o packet losses or reorderings.
1290 * Shrinking the buffer during idle times is not necessary as
1291 * it doesn't consume any memory when idle.
1292 *
1293 * TODO: Only step up if the application is actually serving
1294 * the buffer to better manage the socket buffer resources.
1295 */
1296 if (tcp_do_autorcvbuf &&
1297 to.to_tsecr &&
1298 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1299 if (to.to_tsecr > tp->rfbuf_ts &&
1300 to.to_tsecr - tp->rfbuf_ts < hz) {
1301 if (tp->rfbuf_cnt >
1302 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1303 so->so_rcv.ssb_hiwat <
1304 tcp_autorcvbuf_max) {
1305 newsize =
46e92930
MD
1306 ulmin(so->so_rcv.ssb_hiwat +
1307 tcp_autorcvbuf_inc,
1308 tcp_autorcvbuf_max);
5b0b9fa5
PA
1309 }
1310 /* Start over with next RTT. */
1311 tp->rfbuf_ts = 0;
1312 tp->rfbuf_cnt = 0;
1313 } else
1314 tp->rfbuf_cnt += tlen; /* add up */
1315 }
984263bc
MD
1316 /*
1317 * Add data to socket buffer.
1318 */
1319 if (so->so_state & SS_CANTRCVMORE) {
1320 m_freem(m);
1321 } else {
5b0b9fa5 1322 /*
46e92930
MD
1323 * Set new socket buffer size, give up when
1324 * limit is reached.
1325 *
1326 * Adjusting the size can mess up ACK
1327 * sequencing when pure window updates are
1328 * being avoided (which is the default),
1329 * so force an ack.
5b0b9fa5 1330 */
6cef7136 1331 lwkt_gettoken(&so->so_rcv.ssb_token);
46e92930
MD
1332 if (newsize) {
1333 tp->t_flags |= TF_RXRESIZED;
5b0b9fa5 1334 if (!ssb_reserve(&so->so_rcv, newsize,
46e92930 1335 so, NULL)) {
14343ad3 1336 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
46e92930
MD
1337 }
1338 if (newsize >=
1339 (TCP_MAXWIN << tp->rcv_scale)) {
14343ad3 1340 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
46e92930
MD
1341 }
1342 }
09cf73e0 1343 m_adj(m, drop_hdrlen); /* delayed header drop */
6d49aa6f 1344 ssb_appendstream(&so->so_rcv, m);
6cef7136 1345 lwkt_reltoken(&so->so_rcv.ssb_token);
984263bc
MD
1346 }
1347 sorwakeup(so);
2b1ce38a
MD
1348 /*
1349 * This code is responsible for most of the ACKs
1350 * the TCP stack sends back after receiving a data
1351 * packet. Note that the DELAY_ACK check fails if
1352 * the delack timer is already running, which results
1353 * in an ack being sent every other packet (which is
1354 * what we want).
72b37eeb
MD
1355 *
1356 * We then further aggregate acks by not actually
1357 * sending one until the protocol thread has completed
1358 * processing the current backlog of packets. This
1359 * does not delay the ack any further, but allows us
1360 * to take advantage of the packet aggregation that
1361 * high speed NICs do (usually blocks of 8-10 packets)
1362 * to send a single ack rather then four or five acks,
1363 * greatly reducing the ack rate, the return channel
1364 * bandwidth, and the protocol overhead on both ends.
1365 *
1366 * Since this also has the effect of slowing down
1367 * the exponential slow-start ramp-up, systems with
1368 * very large bandwidth-delay products might want
1369 * to turn the feature off.
2b1ce38a 1370 */
984263bc 1371 if (DELAY_ACK(tp)) {
a48c5dd5
SZ
1372 tcp_callout_reset(tp, tp->tt_delack,
1373 tcp_delacktime, tcp_timer_delack);
72b37eeb 1374 } else if (tcp_aggregate_acks) {
984263bc 1375 tp->t_flags |= TF_ACKNOW;
61896e3c 1376 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
2b1ce38a
MD
1377 tp->t_flags |= TF_ONOUTPUTQ;
1378 tp->tt_cpu = mycpu->gd_cpuid;
1379 TAILQ_INSERT_TAIL(
1380 &tcpcbackq[tp->tt_cpu],
1381 tp, t_outputq);
1382 }
72b37eeb
MD
1383 } else {
1384 tp->t_flags |= TF_ACKNOW;
1385 tcp_output(tp);
984263bc 1386 }
002c1265 1387 return(IPPROTO_DONE);
984263bc
MD
1388 }
1389 }
1390
1391 /*
1392 * Calculate amount of space in receive window,
1393 * and then do TCP input processing.
1394 * Receive window is amount of space in rcv queue,
1395 * but not less than advertised window.
1396 */
6d49aa6f 1397 recvwin = ssb_space(&so->so_rcv);
61896e3c
JH
1398 if (recvwin < 0)
1399 recvwin = 0;
1400 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
984263bc 1401
5b0b9fa5
PA
1402 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1403 tp->rfbuf_ts = 0;
1404 tp->rfbuf_cnt = 0;
1405
984263bc 1406 switch (tp->t_state) {
984263bc
MD
1407 /*
1408 * If the state is SYN_RECEIVED:
1409 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1410 */
1411 case TCPS_SYN_RECEIVED:
1412 if ((thflags & TH_ACK) &&
1413 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1414 SEQ_GT(th->th_ack, tp->snd_max))) {
1415 rstreason = BANDLIM_RST_OPENPORT;
1416 goto dropwithreset;
1417 }
1418 break;
1419
1420 /*
1421 * If the state is SYN_SENT:
1422 * if seg contains an ACK, but not for our SYN, drop the input.
1423 * if seg contains a RST, then drop the connection.
1424 * if seg does not contain SYN, then drop it.
1425 * Otherwise this is an acceptable SYN segment
1426 * initialize tp->rcv_nxt and tp->irs
1427 * if seg contains ack then advance tp->snd_una
1428 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1429 * arrange for segment to be acked (eventually)
1430 * continue processing rest of data/controls, beginning with URG
1431 */
1432 case TCPS_SYN_SENT:
984263bc
MD
1433 if ((thflags & TH_ACK) &&
1434 (SEQ_LEQ(th->th_ack, tp->iss) ||
1435 SEQ_GT(th->th_ack, tp->snd_max))) {
27b8aee3
AE
1436 rstreason = BANDLIM_UNLIMITED;
1437 goto dropwithreset;
984263bc
MD
1438 }
1439 if (thflags & TH_RST) {
1440 if (thflags & TH_ACK)
1441 tp = tcp_drop(tp, ECONNREFUSED);
1442 goto drop;
1443 }
61896e3c 1444 if (!(thflags & TH_SYN))
984263bc
MD
1445 goto drop;
1446 tp->snd_wnd = th->th_win; /* initial send window */
984263bc
MD
1447
1448 tp->irs = th->th_seq;
1449 tcp_rcvseqinit(tp);
1450 if (thflags & TH_ACK) {
27b8aee3 1451 /* Our SYN was acked. */
984263bc
MD
1452 tcpstat.tcps_connects++;
1453 soisconnected(so);
1454 /* Do window scaling on this connection? */
61896e3c 1455 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
b5572302 1456 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
1457 tp->snd_scale = tp->requested_s_scale;
1458 tp->rcv_scale = tp->request_r_scale;
1459 }
984263bc
MD
1460 tp->rcv_adv += tp->rcv_wnd;
1461 tp->snd_una++; /* SYN is acked */
a48c5dd5 1462 tcp_callout_stop(tp, tp->tt_rexmt);
984263bc
MD
1463 /*
1464 * If there's data, delay ACK; if there's also a FIN
1465 * ACKNOW will be turned on later.
1466 */
a48c5dd5
SZ
1467 if (DELAY_ACK(tp) && tlen != 0) {
1468 tcp_callout_reset(tp, tp->tt_delack,
1469 tcp_delacktime, tcp_timer_delack);
1470 } else {
984263bc 1471 tp->t_flags |= TF_ACKNOW;
a48c5dd5 1472 }
984263bc
MD
1473 /*
1474 * Received <SYN,ACK> in SYN_SENT[*] state.
1475 * Transitions:
1476 * SYN_SENT --> ESTABLISHED
1477 * SYN_SENT* --> FIN_WAIT_1
1478 */
1479 tp->t_starttime = ticks;
1480 if (tp->t_flags & TF_NEEDFIN) {
1481 tp->t_state = TCPS_FIN_WAIT_1;
1482 tp->t_flags &= ~TF_NEEDFIN;
1483 thflags &= ~TH_SYN;
1484 } else {
8651f7f8 1485 tcp_established(tp);
984263bc
MD
1486 }
1487 } else {
1488 /*
95b22adf 1489 * Received initial SYN in SYN-SENT[*] state =>
27b8aee3
AE
1490 * simultaneous open.
1491 * Do 3-way handshake:
61896e3c
JH
1492 * SYN-SENT -> SYN-RECEIVED
1493 * SYN-SENT* -> SYN-RECEIVED*
95b22adf 1494 */
984263bc 1495 tp->t_flags |= TF_ACKNOW;
a48c5dd5 1496 tcp_callout_stop(tp, tp->tt_rexmt);
27b8aee3 1497 tp->t_state = TCPS_SYN_RECEIVED;
984263bc
MD
1498 }
1499
984263bc
MD
1500 /*
1501 * Advance th->th_seq to correspond to first data byte.
1502 * If data, trim to stay within window,
1503 * dropping FIN if necessary.
1504 */
1505 th->th_seq++;
1506 if (tlen > tp->rcv_wnd) {
1507 todrop = tlen - tp->rcv_wnd;
1508 m_adj(m, -todrop);
1509 tlen = tp->rcv_wnd;
1510 thflags &= ~TH_FIN;
1511 tcpstat.tcps_rcvpackafterwin++;
1512 tcpstat.tcps_rcvbyteafterwin += todrop;
1513 }
1514 tp->snd_wl1 = th->th_seq - 1;
1515 tp->rcv_up = th->th_seq;
1516 /*
1517 * Client side of transaction: already sent SYN and data.
1518 * If the remote host used T/TCP to validate the SYN,
1519 * our data will be ACK'd; if so, enter normal data segment
1520 * processing in the middle of step 5, ack processing.
1521 * Otherwise, goto step 6.
1522 */
95b22adf 1523 if (thflags & TH_ACK)
984263bc
MD
1524 goto process_ACK;
1525
1526 goto step6;
1527
1528 /*
1529 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
27b8aee3 1530 * do normal processing (we no longer bother with T/TCP).
984263bc
MD
1531 */
1532 case TCPS_LAST_ACK:
1533 case TCPS_CLOSING:
1534 case TCPS_TIME_WAIT:
95b22adf 1535 break; /* continue normal processing */
984263bc
MD
1536 }
1537
1538 /*
1539 * States other than LISTEN or SYN_SENT.
1540 * First check the RST flag and sequence number since reset segments
1541 * are exempt from the timestamp and connection count tests. This
1542 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1543 * below which allowed reset segments in half the sequence space
1544 * to fall though and be processed (which gives forged reset
1545 * segments with a random sequence number a 50 percent chance of
1546 * killing a connection).
1547 * Then check timestamp, if present.
1548 * Then check the connection count, if present.
1549 * Then check that at least some bytes of segment are within
1550 * receive window. If segment begins before rcv_nxt,
1551 * drop leading data (and SYN); if nothing left, just ack.
1552 *
1553 *
1554 * If the RST bit is set, check the sequence number to see
1555 * if this is a valid reset segment.
1556 * RFC 793 page 37:
1557 * In all states except SYN-SENT, all reset (RST) segments
1558 * are validated by checking their SEQ-fields. A reset is
1559 * valid if its sequence number is in the window.
1560 * Note: this does not take into account delayed ACKs, so
1561 * we should test against last_ack_sent instead of rcv_nxt.
1562 * The sequence number in the reset segment is normally an
91489f6b 1563 * echo of our outgoing acknowledgement numbers, but some hosts
984263bc
MD
1564 * send a reset with the sequence number at the rightmost edge
1565 * of our receive window, and we have to handle this case.
1566 * If we have multiple segments in flight, the intial reset
1567 * segment sequence numbers will be to the left of last_ack_sent,
1568 * but they will eventually catch up.
1569 * In any case, it never made sense to trim reset segments to
1570 * fit the receive window since RFC 1122 says:
1571 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1572 *
1573 * A TCP SHOULD allow a received RST segment to include data.
1574 *
1575 * DISCUSSION
61896e3c
JH
1576 * It has been suggested that a RST segment could contain
1577 * ASCII text that encoded and explained the cause of the
91489f6b 1578 * RST. No standard has yet been established for such
61896e3c 1579 * data.
984263bc
MD
1580 *
1581 * If the reset segment passes the sequence number test examine
1582 * the state:
1583 * SYN_RECEIVED STATE:
1584 * If passive open, return to LISTEN state.
1585 * If active open, inform user that connection was refused.
1586 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1587 * Inform user that connection was reset, and close tcb.
1588 * CLOSING, LAST_ACK STATES:
1589 * Close the tcb.
1590 * TIME_WAIT STATE:
1591 * Drop the segment - see Stevens, vol. 2, p. 964 and
61896e3c 1592 * RFC 1337.
984263bc
MD
1593 */
1594 if (thflags & TH_RST) {
1595 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
d4dbb5be 1596 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
984263bc
MD
1597 switch (tp->t_state) {
1598
1599 case TCPS_SYN_RECEIVED:
1600 so->so_error = ECONNREFUSED;
1601 goto close;
1602
1603 case TCPS_ESTABLISHED:
1604 case TCPS_FIN_WAIT_1:
1605 case TCPS_FIN_WAIT_2:
1606 case TCPS_CLOSE_WAIT:
1607 so->so_error = ECONNRESET;
1608 close:
1609 tp->t_state = TCPS_CLOSED;
1610 tcpstat.tcps_drops++;
1611 tp = tcp_close(tp);
1612 break;
1613
1614 case TCPS_CLOSING:
1615 case TCPS_LAST_ACK:
1616 tp = tcp_close(tp);
1617 break;
1618
1619 case TCPS_TIME_WAIT:
1620 break;
1621 }
1622 }
1623 goto drop;
1624 }
1625
1626 /*
1627 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1628 * and it's less than ts_recent, drop it.
1629 */
61896e3c 1630 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
984263bc
MD
1631 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1632
1633 /* Check to see if ts_recent is over 24 days old. */
1634 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1635 /*
1636 * Invalidate ts_recent. If this segment updates
1637 * ts_recent, the age will be reset later and ts_recent
1638 * will get a valid value. If it does not, setting
1639 * ts_recent to zero will at least satisfy the
1640 * requirement that zero be placed in the timestamp
1641 * echo reply when ts_recent isn't valid. The
1642 * age isn't reset until we get a valid ts_recent
1643 * because we don't want out-of-order segments to be
1644 * dropped when ts_recent is old.
1645 */
1646 tp->ts_recent = 0;
1647 } else {
1648 tcpstat.tcps_rcvduppack++;
1649 tcpstat.tcps_rcvdupbyte += tlen;
1650 tcpstat.tcps_pawsdrop++;
1651 if (tlen)
1652 goto dropafterack;
1653 goto drop;
1654 }
1655 }
1656
1657 /*
984263bc
MD
1658 * In the SYN-RECEIVED state, validate that the packet belongs to
1659 * this connection before trimming the data to fit the receive
1660 * window. Check the sequence number versus IRS since we know
1661 * the sequence numbers haven't wrapped. This is a partial fix
1662 * for the "LAND" DoS attack.
1663 */
1664 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1665 rstreason = BANDLIM_RST_OPENPORT;
1666 goto dropwithreset;
1667 }
1668
1669 todrop = tp->rcv_nxt - th->th_seq;
1670 if (todrop > 0) {
91489f6b
JH
1671 if (TCP_DO_SACK(tp)) {
1672 /* Report duplicate segment at head of packet. */
1673 tp->reportblk.rblk_start = th->th_seq;
3a5d999b
SZ
1674 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1675 th->th_seq + tlen, thflags);
91489f6b
JH
1676 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1677 tp->reportblk.rblk_end = tp->rcv_nxt;
1678 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1679 }
984263bc
MD
1680 if (thflags & TH_SYN) {
1681 thflags &= ~TH_SYN;
1682 th->th_seq++;
1683 if (th->th_urp > 1)
1684 th->th_urp--;
1685 else
1686 thflags &= ~TH_URG;
1687 todrop--;
1688 }
1689 /*
1690 * Following if statement from Stevens, vol. 2, p. 960.
1691 */
61896e3c
JH
1692 if (todrop > tlen ||
1693 (todrop == tlen && !(thflags & TH_FIN))) {
984263bc
MD
1694 /*
1695 * Any valid FIN must be to the left of the window.
1696 * At this point the FIN must be a duplicate or out
1697 * of sequence; drop it.
1698 */
1699 thflags &= ~TH_FIN;
1700
1701 /*
1702 * Send an ACK to resynchronize and drop any data.
1703 * But keep on processing for RST or ACK.
1704 */
1705 tp->t_flags |= TF_ACKNOW;
1706 todrop = tlen;
1707 tcpstat.tcps_rcvduppack++;
1708 tcpstat.tcps_rcvdupbyte += todrop;
1709 } else {
1710 tcpstat.tcps_rcvpartduppack++;
1711 tcpstat.tcps_rcvpartdupbyte += todrop;
1712 }
1713 drop_hdrlen += todrop; /* drop from the top afterwards */
1714 th->th_seq += todrop;
1715 tlen -= todrop;
1716 if (th->th_urp > todrop)
1717 th->th_urp -= todrop;
1718 else {
1719 thflags &= ~TH_URG;
1720 th->th_urp = 0;
1721 }
1722 }
1723
1724 /*
1725 * If new data are received on a connection after the
1726 * user processes are gone, then RST the other end.
1727 */
1728 if ((so->so_state & SS_NOFDREF) &&
1729 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1730 tp = tcp_close(tp);
1731 tcpstat.tcps_rcvafterclose++;
1732 rstreason = BANDLIM_UNLIMITED;
1733 goto dropwithreset;
1734 }
1735
1736 /*
1737 * If segment ends after window, drop trailing data
1738 * (and PUSH and FIN); if nothing left, just ACK.
1739 */
61896e3c 1740 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
984263bc
MD
1741 if (todrop > 0) {
1742 tcpstat.tcps_rcvpackafterwin++;
1743 if (todrop >= tlen) {
1744 tcpstat.tcps_rcvbyteafterwin += tlen;
1745 /*
1746 * If a new connection request is received
1747 * while in TIME_WAIT, drop the old connection
1748 * and start over if the sequence numbers
1749 * are above the previous ones.
1750 */
1751 if (thflags & TH_SYN &&
1752 tp->t_state == TCPS_TIME_WAIT &&
1753 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1754 tp = tcp_close(tp);
1755 goto findpcb;
1756 }
1757 /*
1758 * If window is closed can only take segments at
1759 * window edge, and have to drop data and PUSH from
1760 * incoming segments. Continue processing, but
1761 * remember to ack. Otherwise, drop segment
1762 * and ack.
1763 */
1764 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1765 tp->t_flags |= TF_ACKNOW;
1766 tcpstat.tcps_rcvwinprobe++;
1767 } else
1768 goto dropafterack;
1769 } else
1770 tcpstat.tcps_rcvbyteafterwin += todrop;
1771 m_adj(m, -todrop);
1772 tlen -= todrop;
61896e3c 1773 thflags &= ~(TH_PUSH | TH_FIN);
984263bc
MD
1774 }
1775
1776 /*
1777 * If last ACK falls within this segment's sequence numbers,
1778 * record its timestamp.
ad0af98b
ND
1779 * NOTE:
1780 * 1) That the test incorporates suggestions from the latest
1781 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1782 * 2) That updating only on newer timestamps interferes with
1783 * our earlier PAWS tests, so this check should be solely
1784 * predicated on the sequence space of this segment.
1785 * 3) That we modify the segment boundary check to be
1786 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1787 * instead of RFC1323's
1788 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1789 * This modified check allows us to overcome RFC1323's
1790 * limitations as described in Stevens TCP/IP Illustrated
1791 * Vol. 2 p.869. In such cases, we can still calculate the
1792 * RTT correctly when RCV.NXT == Last.ACK.Sent.
984263bc 1793 */
ad0af98b
ND
1794 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1795 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1796 + ((thflags & TH_SYN) != 0)
1797 + ((thflags & TH_FIN) != 0)))) {
984263bc
MD
1798 tp->ts_recent_age = ticks;
1799 tp->ts_recent = to.to_tsval;
1800 }
1801
1802 /*
1803 * If a SYN is in the window, then this is an
1804 * error and we send an RST and drop the connection.
1805 */
1806 if (thflags & TH_SYN) {
1807 tp = tcp_drop(tp, ECONNRESET);
1808 rstreason = BANDLIM_UNLIMITED;
1809 goto dropwithreset;
1810 }
1811
1812 /*
1813 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1814 * flag is on (half-synchronized state), then queue data for
1815 * later processing; else drop segment and return.
1816 */
61896e3c 1817 if (!(thflags & TH_ACK)) {
984263bc
MD
1818 if (tp->t_state == TCPS_SYN_RECEIVED ||
1819 (tp->t_flags & TF_NEEDSYN))
1820 goto step6;
1821 else
1822 goto drop;
1823 }
1824
1825 /*
1826 * Ack processing.
1827 */
1828 switch (tp->t_state) {
984263bc 1829 /*
91489f6b 1830 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
984263bc
MD
1831 * ESTABLISHED state and continue processing.
1832 * The ACK was checked above.
1833 */
1834 case TCPS_SYN_RECEIVED:
1835
1836 tcpstat.tcps_connects++;
1837 soisconnected(so);
1838 /* Do window scaling? */
61896e3c 1839 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
b5572302 1840 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
1841 tp->snd_scale = tp->requested_s_scale;
1842 tp->rcv_scale = tp->request_r_scale;
1843 }
1844 /*
984263bc
MD
1845 * Make transitions:
1846 * SYN-RECEIVED -> ESTABLISHED
1847 * SYN-RECEIVED* -> FIN-WAIT-1
1848 */
1849 tp->t_starttime = ticks;
1850 if (tp->t_flags & TF_NEEDFIN) {
1851 tp->t_state = TCPS_FIN_WAIT_1;
1852 tp->t_flags &= ~TF_NEEDFIN;
1853 } else {
8651f7f8 1854 tcp_established(tp);
984263bc
MD
1855 }
1856 /*
1857 * If segment contains data or ACK, will call tcp_reass()
1858 * later; if not, do so now to pass queued data to user.
1859 */
61896e3c 1860 if (tlen == 0 && !(thflags & TH_FIN))
f23061d4 1861 tcp_reass(tp, NULL, NULL, NULL);
984263bc
MD
1862 /* fall into ... */
1863
1864 /*
1865 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1866 * ACKs. If the ack is in the range
1867 * tp->snd_una < th->th_ack <= tp->snd_max
1868 * then advance tp->snd_una to th->th_ack and drop
1869 * data from the retransmission queue. If this ACK reflects
1870 * more up to date window information we update our window information.
1871 */
1872 case TCPS_ESTABLISHED:
1873 case TCPS_FIN_WAIT_1:
1874 case TCPS_FIN_WAIT_2:
1875 case TCPS_CLOSE_WAIT:
1876 case TCPS_CLOSING:
1877 case TCPS_LAST_ACK:
1878 case TCPS_TIME_WAIT:
1879
1880 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
91489f6b
JH
1881 if (TCP_DO_SACK(tp))
1882 tcp_sack_update_scoreboard(tp, &to);
95b22adf
JH
1883 if (tlen != 0 || tiwin != tp->snd_wnd) {
1884 tp->t_dupacks = 0;
1885 break;
1886 }
1887 tcpstat.tcps_rcvdupack++;
a48c5dd5 1888 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
91489f6b
JH
1889 th->th_ack != tp->snd_una) {
1890 tp->t_dupacks = 0;
1891 break;
1892 }
95b22adf 1893 /*
91489f6b 1894 * We have outstanding data (other than
95b22adf
JH
1895 * a window probe), this is a completely
1896 * duplicate ack (ie, window info didn't
1897 * change), the ack is the biggest we've
1898 * seen and we've seen exactly our rexmt
91489f6b 1899 * threshhold of them, so assume a packet
95b22adf
JH
1900 * has been dropped and retransmit it.
1901 * Kludge snd_nxt & the congestion
1902 * window so we send only this one
1903 * packet.
95b22adf 1904 */
91489f6b
JH
1905 if (IN_FASTRECOVERY(tp)) {
1906 if (TCP_DO_SACK(tp)) {
1907 /* No artifical cwnd inflation. */
1908 tcp_sack_rexmt(tp, th);
1909 } else {
1910 /*
1911 * Dup acks mean that packets
1912 * have left the network
1913 * (they're now cached at the
1914 * receiver) so bump cwnd by
1915 * the amount in the receiver
1916 * to keep a constant cwnd
1917 * packets in the network.
1918 */
1919 tp->snd_cwnd += tp->t_maxseg;
f23061d4 1920 tcp_output(tp);
91489f6b
JH
1921 }
1922 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
95b22adf
JH
1923 tp->t_dupacks = 0;
1924 break;
95b22adf 1925 } else if (++tp->t_dupacks == tcprexmtthresh) {
91489f6b 1926 tcp_seq old_snd_nxt;
95b22adf
JH
1927 u_int win;
1928
8819433a 1929fastretransmit:
95b22adf
JH
1930 if (tcp_do_eifel_detect &&
1931 (tp->t_flags & TF_RCVD_TSTMP)) {
1932 tcp_save_congestion_state(tp);
1933 tp->t_flags |= TF_FASTREXMT;
1934 }
91489f6b
JH
1935 /*
1936 * We know we're losing at the current
1937 * window size, so do congestion avoidance:
1938 * set ssthresh to half the current window
1939 * and pull our congestion window back to the
1940 * new ssthresh.
1941 */
95b22adf
JH
1942 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1943 tp->t_maxseg;
1944 if (win < 2)
1945 win = 2;
1946 tp->snd_ssthresh = win * tp->t_maxseg;
1947 ENTER_FASTRECOVERY(tp);
1948 tp->snd_recover = tp->snd_max;
a48c5dd5 1949 tcp_callout_stop(tp, tp->tt_rexmt);
95b22adf 1950 tp->t_rtttime = 0;
91489f6b 1951 old_snd_nxt = tp->snd_nxt;
95b22adf
JH
1952 tp->snd_nxt = th->th_ack;
1953 tp->snd_cwnd = tp->t_maxseg;
f23061d4 1954 tcp_output(tp);
95b22adf 1955 ++tcpstat.tcps_sndfastrexmit;
91489f6b
JH
1956 tp->snd_cwnd = tp->snd_ssthresh;
1957 tp->rexmt_high = tp->snd_nxt;
1958 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1959 tp->snd_nxt = old_snd_nxt;
95b22adf
JH
1960 KASSERT(tp->snd_limited <= 2,
1961 ("tp->snd_limited too big"));
91489f6b
JH
1962 if (TCP_DO_SACK(tp))
1963 tcp_sack_rexmt(tp, th);
1964 else
1965 tp->snd_cwnd += tp->t_maxseg *
1966 (tp->t_dupacks - tp->snd_limited);
95b22adf
JH
1967 } else if (tcp_do_limitedtransmit) {
1968 u_long oldcwnd = tp->snd_cwnd;
1969 tcp_seq oldsndmax = tp->snd_max;
9e707f76 1970 tcp_seq oldsndnxt = tp->snd_nxt;
95b22adf
JH
1971 /* outstanding data */
1972 uint32_t ownd = tp->snd_max - tp->snd_una;
1973 u_int sent;
ca94b4a6 1974
8819433a
JH
1975#define iceildiv(n, d) (((n)+(d)-1) / (d))
1976
95b22adf
JH
1977 KASSERT(tp->t_dupacks == 1 ||
1978 tp->t_dupacks == 2,
1979 ("dupacks not 1 or 2"));
1980 if (tp->t_dupacks == 1)
1981 tp->snd_limited = 0;
9e707f76 1982 tp->snd_nxt = tp->snd_max;
95b22adf
JH
1983 tp->snd_cwnd = ownd +
1984 (tp->t_dupacks - tp->snd_limited) *
1985 tp->t_maxseg;
f23061d4 1986 tcp_output(tp);
98cb2337 1987
98cb2337 1988 if (SEQ_LT(oldsndnxt, oldsndmax)) {
31ea9b1e
SZ
1989 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
1990 ("snd_una moved in other threads"));
6d08465d 1991 tp->snd_nxt = oldsndnxt;
98cb2337 1992 }
95b22adf
JH
1993 tp->snd_cwnd = oldcwnd;
1994 sent = tp->snd_max - oldsndmax;
1995 if (sent > tp->t_maxseg) {
1996 KASSERT((tp->t_dupacks == 2 &&
1997 tp->snd_limited == 0) ||
1998 (sent == tp->t_maxseg + 1 &&
1999 tp->t_flags & TF_SENTFIN),
2000 ("sent too much"));
2001 KASSERT(sent <= tp->t_maxseg * 2,
2002 ("sent too many segments"));
2003 tp->snd_limited = 2;
2004 tcpstat.tcps_sndlimited += 2;
2005 } else if (sent > 0) {
2006 ++tp->snd_limited;
2007 ++tcpstat.tcps_sndlimited;
2008 } else if (tcp_do_early_retransmit &&
2009 (tcp_do_eifel_detect &&
2010 (tp->t_flags & TF_RCVD_TSTMP)) &&
91489f6b 2011 ownd < 4 * tp->t_maxseg &&
95b22adf 2012 tp->t_dupacks + 1 >=
91489f6b
JH
2013 iceildiv(ownd, tp->t_maxseg) &&
2014 (!TCP_DO_SACK(tp) ||
f23061d4 2015 ownd <= tp->t_maxseg ||
91489f6b 2016 tcp_sack_has_sacked(&tp->scb,
f23061d4 2017 ownd - tp->t_maxseg))) {
95b22adf
JH
2018 ++tcpstat.tcps_sndearlyrexmit;
2019 tp->t_flags |= TF_EARLYREXMT;
2020 goto fastretransmit;
984263bc 2021 }
95b22adf
JH
2022 }
2023 goto drop;
984263bc
MD
2024 }
2025
2026 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
984263bc
MD
2027 tp->t_dupacks = 0;
2028 if (SEQ_GT(th->th_ack, tp->snd_max)) {
5a274421
JH
2029 /*
2030 * Detected optimistic ACK attack.
2031 * Force slow-start to de-synchronize attack.
2032 */
2033 tp->snd_cwnd = tp->t_maxseg;
8acdb67c 2034 tp->snd_wacked = 0;
5a274421 2035
984263bc
MD
2036 tcpstat.tcps_rcvacktoomuch++;
2037 goto dropafterack;
2038 }
2039 /*
2040 * If we reach this point, ACK is not a duplicate,
2041 * i.e., it ACKs something we sent.
2042 */
2043 if (tp->t_flags & TF_NEEDSYN) {
2044 /*
2045 * T/TCP: Connection was half-synchronized, and our
2046 * SYN has been ACK'd (so connection is now fully
2047 * synchronized). Go to non-starred state,
2048 * increment snd_una for ACK of SYN, and check if
2049 * we can do window scaling.
2050 */
2051 tp->t_flags &= ~TF_NEEDSYN;
2052 tp->snd_una++;
2053 /* Do window scaling? */
61896e3c
JH
2054 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2055 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
984263bc
MD
2056 tp->snd_scale = tp->requested_s_scale;
2057 tp->rcv_scale = tp->request_r_scale;
2058 }
2059 }
2060
2061process_ACK:
2062 acked = th->th_ack - tp->snd_una;
2063 tcpstat.tcps_rcvackpack++;
2064 tcpstat.tcps_rcvackbyte += acked;
2065
b5572302 2066 if (tcp_do_eifel_detect && acked > 0 &&
95b22adf 2067 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
bfdb979e
JH
2068 (tp->t_flags & TF_FIRSTACCACK)) {
2069 /* Eifel detection applicable. */
2070 if (to.to_tsecr < tp->t_rexmtTS) {
bfdb979e 2071 ++tcpstat.tcps_eifeldetected;
8819433a 2072 tcp_revert_congestion_state(tp);
928c3291 2073 if (tp->t_rxtshift != 1 ||
8819433a
JH
2074 ticks >= tp->t_badrxtwin)
2075 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
2076 }
2077 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
b5572302
JH
2078 /*
2079 * If we just performed our first retransmit,
2080 * and the ACK arrives within our recovery window,
2081 * then it was a mistake to do the retransmit
91489f6b 2082 * in the first place. Recover our original cwnd
b5572302
JH
2083 * and ssthresh, and proceed to transmit where we
2084 * left off.
2085 */
bfdb979e
JH
2086 tcp_revert_congestion_state(tp);
2087 ++tcpstat.tcps_rttdetected;
984263bc
MD
2088 }
2089
2090 /*
2091 * If we have a timestamp reply, update smoothed
2092 * round trip time. If no timestamp is present but
2093 * transmit timer is running and timed sequence
2094 * number was acked, update smoothed round trip time.
2095 * Since we now have an rtt measurement, cancel the
2096 * timer backoff (cf., Phil Karn's retransmit alg.).
2097 * Recompute the initial retransmit timer.
2098 *
2099 * Some machines (certain windows boxes) send broken
d24ce1dc 2100 * timestamp replies during the SYN+ACK phase, ignore
984263bc
MD
2101 * timestamps of 0.
2102 */
95b22adf 2103 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
984263bc 2104 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
95b22adf 2105 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
984263bc 2106 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
984263bc
MD
2107 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2108
2109 /*
984263bc
MD
2110 * If no data (only SYN) was ACK'd,
2111 * skip rest of ACK processing.
2112 */
2113 if (acked == 0)
2114 goto step6;
2115
efd4b327 2116 /* Stop looking for an acceptable ACK since one was received. */
8819433a 2117 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
efd4b327 2118
6d49aa6f
MD
2119 if (acked > so->so_snd.ssb_cc) {
2120 tp->snd_wnd -= so->so_snd.ssb_cc;
2121 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
61896e3c 2122 ourfinisacked = TRUE;
984263bc 2123 } else {
6d49aa6f 2124 sbdrop(&so->so_snd.sb, acked);
984263bc 2125 tp->snd_wnd -= acked;
61896e3c 2126 ourfinisacked = FALSE;
984263bc
MD
2127 }
2128 sowwakeup(so);
91489f6b
JH
2129
2130 /*
2131 * Update window information.
91489f6b 2132 */
e126661b 2133 if (acceptable_window_update(tp, th, tiwin)) {
91489f6b
JH
2134 /* keep track of pure window updates */
2135 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2136 tiwin > tp->snd_wnd)
2137 tcpstat.tcps_rcvwinupd++;
2138 tp->snd_wnd = tiwin;
2139 tp->snd_wl1 = th->th_seq;
2140 tp->snd_wl2 = th->th_ack;
2141 if (tp->snd_wnd > tp->max_sndwnd)
2142 tp->max_sndwnd = tp->snd_wnd;
2143 needoutput = TRUE;
2144 }
2145
2146 tp->snd_una = th->th_ack;
2147 if (TCP_DO_SACK(tp))
2148 tcp_sack_update_scoreboard(tp, &to);
95b22adf 2149 if (IN_FASTRECOVERY(tp)) {
91489f6b 2150 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
95b22adf 2151 EXIT_FASTRECOVERY(tp);
91489f6b
JH
2152 needoutput = TRUE;
2153 /*
2154 * If the congestion window was inflated
2155 * to account for the other side's
2156 * cached packets, retract it.
8acdb67c
JH
2157 */
2158 if (!TCP_DO_SACK(tp))
2159 tp->snd_cwnd = tp->snd_ssthresh;
2160
2161 /*
91489f6b
JH
2162 * Window inflation should have left us
2163 * with approximately snd_ssthresh outstanding
2164 * data. But, in case we would be inclined
2165 * to send a burst, better do it using
2166 * slow start.
2167 */
91489f6b
JH
2168 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2169 tp->snd_max + 2 * tp->t_maxseg))
2170 tp->snd_cwnd =
2171 (tp->snd_max - tp->snd_una) +
2172 2 * tp->t_maxseg;
8acdb67c
JH
2173
2174 tp->snd_wacked = 0;
91489f6b
JH
2175 } else {
2176 if (TCP_DO_SACK(tp)) {
2177 tp->snd_max_rexmt = tp->snd_max;
2178 tcp_sack_rexmt(tp, th);
2179 } else {
2180 tcp_newreno_partial_ack(tp, th, acked);
2181 }
2182 needoutput = FALSE;
2183 }
95b22adf 2184 } else {
91489f6b 2185 /*
8acdb67c
JH
2186 * Open the congestion window. When in slow-start,
2187 * open exponentially: maxseg per packet. Otherwise,
2188 * open linearly: maxseg per window.
91489f6b 2189 */
8acdb67c
JH
2190 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2191 u_int abc_sslimit =
2192 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2193 tp->t_maxseg : 2 * tp->t_maxseg);
2194
2195 /* slow-start */
2196 tp->snd_cwnd += tcp_do_abc ?
2197 min(acked, abc_sslimit) : tp->t_maxseg;
2198 } else {
2199 /* linear increase */
2200 tp->snd_wacked += tcp_do_abc ? acked :
2201 tp->t_maxseg;
2202 if (tp->snd_wacked >= tp->snd_cwnd) {
2203 tp->snd_wacked -= tp->snd_cwnd;
2204 tp->snd_cwnd += tp->t_maxseg;
2205 }
2206 }
2207 tp->snd_cwnd = min(tp->snd_cwnd,
2208 TCP_MAXWIN << tp->snd_scale);
95b22adf 2209 tp->snd_recover = th->th_ack - 1;
cfb3f3f4 2210 }
984263bc
MD
2211 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2212 tp->snd_nxt = tp->snd_una;
2213
91489f6b
JH
2214 /*
2215 * If all outstanding data is acked, stop retransmit
2216 * timer and remember to restart (more output or persist).
2217 * If there is more data to be acked, restart retransmit
2218 * timer, using current (possibly backed-off) value.
2219 */
2220 if (th->th_ack == tp->snd_max) {
a48c5dd5 2221 tcp_callout_stop(tp, tp->tt_rexmt);
91489f6b 2222 needoutput = TRUE;
a48c5dd5
SZ
2223 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2224 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2225 tcp_timer_rexmt);
2226 }
91489f6b 2227
984263bc 2228 switch (tp->t_state) {
984263bc
MD
2229 /*
2230 * In FIN_WAIT_1 STATE in addition to the processing
2231 * for the ESTABLISHED state if our FIN is now acknowledged
2232 * then enter FIN_WAIT_2.
2233 */
2234 case TCPS_FIN_WAIT_1:
2235 if (ourfinisacked) {
2236 /*
2237 * If we can't receive any more
2238 * data, then closing user can proceed.
2239 * Starting the timer is contrary to the
2240 * specification, but if we don't get a FIN
2241 * we'll hang forever.
2242 */
2243 if (so->so_state & SS_CANTRCVMORE) {
2244 soisdisconnected(so);
a48c5dd5 2245 tcp_callout_reset(tp, tp->tt_2msl,
5d61ded3 2246 tp->t_maxidle, tcp_timer_2msl);
984263bc
MD
2247 }
2248 tp->t_state = TCPS_FIN_WAIT_2;
2249 }
2250 break;
2251
95b22adf 2252 /*
984263bc
MD
2253 * In CLOSING STATE in addition to the processing for
2254 * the ESTABLISHED state if the ACK acknowledges our FIN
2255 * then enter the TIME-WAIT state, otherwise ignore
2256 * the segment.
2257 */
2258 case TCPS_CLOSING:
2259 if (ourfinisacked) {
2260 tp->t_state = TCPS_TIME_WAIT;
2261 tcp_canceltimers(tp);
27b8aee3 2262 tcp_callout_reset(tp, tp->tt_2msl,
01d3427a
SZ
2263 2 * tcp_rmx_msl(tp),
2264 tcp_timer_2msl);
984263bc
MD
2265 soisdisconnected(so);
2266 }
2267 break;
2268
2269 /*
2270 * In LAST_ACK, we may still be waiting for data to drain
2271 * and/or to be acked, as well as for the ack of our FIN.
2272 * If our FIN is now acknowledged, delete the TCB,
2273 * enter the closed state and return.
2274 */
2275 case TCPS_LAST_ACK:
2276 if (ourfinisacked) {
2277 tp = tcp_close(tp);
2278 goto drop;
2279 }
2280 break;
2281
2282 /*
2283 * In TIME_WAIT state the only thing that should arrive
2284 * is a retransmission of the remote FIN. Acknowledge
2285 * it and restart the finack timer.
2286 */
2287 case TCPS_TIME_WAIT:
01d3427a 2288 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2289 tcp_timer_2msl);
984263bc
MD
2290 goto dropafterack;
2291 }
2292 }
2293
2294step6:
2295 /*
2296 * Update window information.
2297 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2298 */
2299 if ((thflags & TH_ACK) &&
df9d7670 2300 acceptable_window_update(tp, th, tiwin)) {
984263bc 2301 /* keep track of pure window updates */
b5572302
JH
2302 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2303 tiwin > tp->snd_wnd)
984263bc
MD
2304 tcpstat.tcps_rcvwinupd++;
2305 tp->snd_wnd = tiwin;
2306 tp->snd_wl1 = th->th_seq;
2307 tp->snd_wl2 = th->th_ack;
2308 if (tp->snd_wnd > tp->max_sndwnd)
2309 tp->max_sndwnd = tp->snd_wnd;
61896e3c 2310 needoutput = TRUE;
984263bc
MD
2311 }
2312
2313 /*
2314 * Process segments with URG.
2315 */
2316 if ((thflags & TH_URG) && th->th_urp &&
95b22adf 2317 !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2318 /*
2319 * This is a kludge, but if we receive and accept
2320 * random urgent pointers, we'll crash in
2321 * soreceive. It's hard to imagine someone
2322 * actually wanting to send this much urgent data.
2323 */
6d49aa6f 2324 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
984263bc
MD
2325 th->th_urp = 0; /* XXX */
2326 thflags &= ~TH_URG; /* XXX */
2327 goto dodata; /* XXX */
2328 }
2329 /*
2330 * If this segment advances the known urgent pointer,
2331 * then mark the data stream. This should not happen
2332 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2333 * a FIN has been received from the remote side.
2334 * In these states we ignore the URG.
2335 *
2336 * According to RFC961 (Assigned Protocols),
2337 * the urgent pointer points to the last octet
2338 * of urgent data. We continue, however,
2339 * to consider it to indicate the first octet
2340 * of data past the urgent section as the original
2341 * spec states (in one of two places).
2342 */
61896e3c 2343 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
984263bc 2344 tp->rcv_up = th->th_seq + th->th_urp;
6d49aa6f 2345 so->so_oobmark = so->so_rcv.ssb_cc +
984263bc
MD
2346 (tp->rcv_up - tp->rcv_nxt) - 1;
2347 if (so->so_oobmark == 0)
6cef7136 2348 sosetstate(so, SS_RCVATMARK);
984263bc
MD
2349 sohasoutofband(so);
2350 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2351 }
2352 /*
2353 * Remove out of band data so doesn't get presented to user.
2354 * This can happen independent of advancing the URG pointer,
2355 * but if two URG's are pending at once, some out-of-band
2356 * data may creep in... ick.
2357 */
61896e3c
JH
2358 if (th->th_urp <= (u_long)tlen &&
2359 !(so->so_options & SO_OOBINLINE)) {
2360 /* hdr drop is delayed */
2361 tcp_pulloutofband(so, th, m, drop_hdrlen);
2362 }
984263bc
MD
2363 } else {
2364 /*
2365 * If no out of band data is expected,
2366 * pull receive urgent pointer along
2367 * with the receive window.
2368 */
2369 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2370 tp->rcv_up = tp->rcv_nxt;
2371 }
984263bc 2372
61896e3c 2373dodata: /* XXX */
984263bc
MD
2374 /*
2375 * Process the segment text, merging it into the TCP sequencing queue,
2376 * and arranging for acknowledgment of receipt if necessary.
2377 * This process logically involves adjusting tp->rcv_wnd as data
2378 * is presented to the user (this happens in tcp_usrreq.c,
2379 * case PRU_RCVD). If a FIN has already been received on this
2380 * connection then we just ignore the text.
2381 */
95b22adf 2382 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2383 m_adj(m, drop_hdrlen); /* delayed header drop */
2384 /*
2385 * Insert segment which includes th into TCP reassembly queue
2386 * with control block tp. Set thflags to whether reassembly now
2387 * includes a segment with FIN. This handles the common case
2388 * inline (segment is the next to be received on an established
2389 * connection, and the queue is empty), avoiding linkage into
2390 * and removal from the queue and repetition of various
2391 * conversions.
2392 * Set DELACK for segments received in order, but ack
2393 * immediately when segments are out of order (so
2394 * fast retransmit can work).
2395 */
2396 if (th->th_seq == tp->rcv_nxt &&
2397 LIST_EMPTY(&tp->t_segq) &&
2398 TCPS_HAVEESTABLISHED(tp->t_state)) {
a48c5dd5
SZ
2399 if (DELAY_ACK(tp)) {
2400 tcp_callout_reset(tp, tp->tt_delack,
2401 tcp_delacktime, tcp_timer_delack);
2402 } else {
984263bc 2403 tp->t_flags |= TF_ACKNOW;
a48c5dd5 2404 }
984263bc
MD
2405 tp->rcv_nxt += tlen;
2406 thflags = th->th_flags & TH_FIN;
2407 tcpstat.tcps_rcvpack++;
2408 tcpstat.tcps_rcvbyte += tlen;
2409 ND6_HINT(tp);
6cef7136 2410 if (so->so_state & SS_CANTRCVMORE) {
984263bc 2411 m_freem(m);
6cef7136
MD
2412 } else {
2413 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 2414 ssb_appendstream(&so->so_rcv, m);
6cef7136
MD
2415 lwkt_reltoken(&so->so_rcv.ssb_token);
2416 }
984263bc
MD
2417 sorwakeup(so);
2418 } else {
91489f6b
JH
2419 if (!(tp->t_flags & TF_DUPSEG)) {
2420 /* Initialize SACK report block. */
2421 tp->reportblk.rblk_start = th->th_seq;
3a5d999b
SZ
2422 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2423 th->th_seq + tlen, thflags);
91489f6b 2424 }
984263bc
MD
2425 thflags = tcp_reass(tp, th, &tlen, m);
2426 tp->t_flags |= TF_ACKNOW;
2427 }
2428
2429 /*
2430 * Note the amount of data that peer has sent into
2431 * our window, in order to estimate the sender's
2432 * buffer size.
2433 */
6d49aa6f 2434 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
984263bc
MD
2435 } else {
2436 m_freem(m);
2437 thflags &= ~TH_FIN;
2438 }
2439
2440 /*
2441 * If FIN is received ACK the FIN and let the user know
2442 * that the connection is closing.
2443 */
2444 if (thflags & TH_FIN) {
95b22adf 2445 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2446 socantrcvmore(so);
2447 /*
2448 * If connection is half-synchronized
2449 * (ie NEEDSYN flag on) then delay ACK,
2450 * so it may be piggybacked when SYN is sent.
2451 * Otherwise, since we received a FIN then no
2452 * more input can be expected, send ACK now.
2453 */
a48c5dd5
SZ
2454 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2455 tcp_callout_reset(tp, tp->tt_delack,
2456 tcp_delacktime, tcp_timer_delack);
2457 } else {
984263bc 2458 tp->t_flags |= TF_ACKNOW;
a48c5dd5 2459 }
984263bc
MD
2460 tp->rcv_nxt++;
2461 }
984263bc 2462
61896e3c 2463 switch (tp->t_state) {
95b22adf 2464 /*
984263bc
MD
2465 * In SYN_RECEIVED and ESTABLISHED STATES
2466 * enter the CLOSE_WAIT state.
2467 */
2468 case TCPS_SYN_RECEIVED:
2469 tp->t_starttime = ticks;
2470 /*FALLTHROUGH*/
2471 case TCPS_ESTABLISHED:
2472 tp->t_state = TCPS_CLOSE_WAIT;
2473 break;
2474
95b22adf 2475 /*
984263bc
MD
2476 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2477 * enter the CLOSING state.
2478 */
2479 case TCPS_FIN_WAIT_1:
2480 tp->t_state = TCPS_CLOSING;
2481 break;
2482
95b22adf 2483 /*
984263bc
MD
2484 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2485 * starting the time-wait timer, turning off the other
2486 * standard timers.
2487 */
2488 case TCPS_FIN_WAIT_2:
2489 tp->t_state = TCPS_TIME_WAIT;
2490 tcp_canceltimers(tp);
01d3427a 2491 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2492 tcp_timer_2msl);
984263bc
MD
2493 soisdisconnected(so);
2494 break;
2495
2496 /*
2497 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2498 */
2499 case TCPS_TIME_WAIT:
01d3427a 2500 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2501 tcp_timer_2msl);
984263bc
MD
2502 break;
2503 }
2504 }
61896e3c 2505
984263bc
MD
2506#ifdef TCPDEBUG
2507 if (so->so_options & SO_DEBUG)
f23061d4 2508 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2509#endif
2510
2511 /*
2512 * Return any desired output.
2513 */
2514 if (needoutput || (tp->t_flags & TF_ACKNOW))
f23061d4 2515 tcp_output(tp);
d58ca578 2516 tcp_sack_report_cleanup(tp);
002c1265 2517 return(IPPROTO_DONE);
984263bc
MD
2518
2519dropafterack:
2520 /*
2521 * Generate an ACK dropping incoming segment if it occupies
2522 * sequence space, where the ACK reflects our state.
2523 *
2524 * We can now skip the test for the RST flag since all
2525 * paths to this code happen after packets containing
2526 * RST have been dropped.
2527 *
2528 * In the SYN-RECEIVED state, don't send an ACK unless the
2529 * segment we received passes the SYN-RECEIVED ACK test.
2530 * If it fails send a RST. This breaks the loop in the
2531 * "LAND" DoS attack, and also prevents an ACK storm
2532 * between two listening ports that have been sent forged
2533 * SYN segments, each with the source address of the other.
2534 */
2535 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2536 (SEQ_GT(tp->snd_una, th->th_ack) ||
2537 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2538 rstreason = BANDLIM_RST_OPENPORT;
2539 goto dropwithreset;
2540 }
2541#ifdef TCPDEBUG
2542 if (so->so_options & SO_DEBUG)
f23061d4 2543 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2544#endif
2545 m_freem(m);
2546 tp->t_flags |= TF_ACKNOW;
f23061d4 2547 tcp_output(tp);
d58ca578 2548 tcp_sack_report_cleanup(tp);
002c1265 2549 return(IPPROTO_DONE);
984263bc
MD
2550
2551dropwithreset:
2552 /*
2553 * Generate a RST, dropping incoming segment.
2554 * Make ACK acceptable to originator of segment.
2555 * Don't bother to respond if destination was broadcast/multicast.
2556 */
61896e3c 2557 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
2558 goto drop;
2559 if (isipv6) {
2560 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2561 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2562 goto drop;
2563 } else {
2564 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2565 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
95b22adf
JH
2566 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2567 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
984263bc
MD
2568 goto drop;
2569 }
2570 /* IPv6 anycast check is done at tcp6_input() */
2571
2572 /*
2573 * Perform bandwidth limiting.
2574 */
2575#ifdef ICMP_BANDLIM
2576 if (badport_bandlim(rstreason) < 0)
2577 goto drop;
2578#endif
2579
2580#ifdef TCPDEBUG
2581 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2582 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2583#endif
2584 if (thflags & TH_ACK)
2585 /* mtod() below is safe as long as hdr dropping is delayed */
2586 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2587 TH_RST);
2588 else {
2589 if (thflags & TH_SYN)
2590 tlen++;
2591 /* mtod() below is safe as long as hdr dropping is delayed */
61896e3c
JH
2592 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2593 (tcp_seq)0, TH_RST | TH_ACK);
984263bc 2594 }
d58ca578
SZ
2595 if (tp != NULL)
2596 tcp_sack_report_cleanup(tp);
002c1265 2597 return(IPPROTO_DONE);
984263bc
MD
2598
2599drop:
2600 /*
2601 * Drop space held by incoming segment and return.
2602 */
2603#ifdef TCPDEBUG
2604 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2605 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2606#endif
2607 m_freem(m);
d58ca578
SZ
2608 if (tp != NULL)
2609 tcp_sack_report_cleanup(tp);
002c1265 2610 return(IPPROTO_DONE);
984263bc
MD
2611}
2612
2613/*
2614 * Parse TCP options and place in tcpopt.
2615 */
2616static void
95b22adf 2617tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
984263bc 2618{
91489f6b 2619 int opt, optlen, i;
984263bc
MD
2620
2621 to->to_flags = 0;
2622 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2623 opt = cp[0];
2624 if (opt == TCPOPT_EOL)
2625 break;
2626 if (opt == TCPOPT_NOP)
2627 optlen = 1;
2628 else {
2629 if (cnt < 2)
2630 break;
2631 optlen = cp[1];
2632 if (optlen < 2 || optlen > cnt)
2633 break;
2634 }
2635 switch (opt) {
2636 case TCPOPT_MAXSEG:
2637 if (optlen != TCPOLEN_MAXSEG)
2638 continue;
2639 if (!is_syn)
2640 continue;
2641 to->to_flags |= TOF_MSS;
407e896e 2642 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
984263bc
MD
2643 to->to_mss = ntohs(to->to_mss);
2644 break;
2645 case TCPOPT_WINDOW:
2646 if (optlen != TCPOLEN_WINDOW)
2647 continue;
95b22adf 2648 if (!is_syn)
984263bc
MD
2649 continue;
2650 to->to_flags |= TOF_SCALE;
2651 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2652 break;
2653 case TCPOPT_TIMESTAMP:
2654 if (optlen != TCPOLEN_TIMESTAMP)
2655 continue;
2656 to->to_flags |= TOF_TS;
407e896e 2657 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
984263bc 2658 to->to_tsval = ntohl(to->to_tsval);
407e896e 2659 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
984263bc 2660 to->to_tsecr = ntohl(to->to_tsecr);
ad0af98b
ND
2661 /*
2662 * If echoed timestamp is later than the current time,
2663 * fall back to non RFC1323 RTT calculation.
2664 */
2665 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2666 to->to_tsecr = 0;
984263bc 2667 break;
91489f6b
JH
2668 case TCPOPT_SACK_PERMITTED:
2669 if (optlen != TCPOLEN_SACK_PERMITTED)
2670 continue;
2671 if (!is_syn)
2672 continue;
2673 to->to_flags |= TOF_SACK_PERMITTED;
2674 break;
2675 case TCPOPT_SACK:
2676 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2677 continue;
2678 to->to_nsackblocks = (optlen - 2) / 8;
2679 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2680 to->to_flags |= TOF_SACK;
2681 for (i = 0; i < to->to_nsackblocks; i++) {
2682 struct raw_sackblock *r = &to->to_sackblocks[i];
2683
2684 r->rblk_start = ntohl(r->rblk_start);
2685 r->rblk_end = ntohl(r->rblk_end);
865b0477
SZ
2686
2687 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2688 /*
2689 * Invalid SACK block; discard all
2690 * SACK blocks
2691 */
02cc2f35 2692 tcpstat.tcps_rcvbadsackopt++;
865b0477
SZ
2693 to->to_nsackblocks = 0;
2694 to->to_sackblocks = NULL;
2695 to->to_flags &= ~TOF_SACK;
2696 break;
2697 }
91489f6b
JH
2698 }
2699 break;
b1992928
MD
2700#ifdef TCP_SIGNATURE
2701 /*
2702 * XXX In order to reply to a host which has set the
2703 * TCP_SIGNATURE option in its initial SYN, we have to
2704 * record the fact that the option was observed here
2705 * for the syncache code to perform the correct response.
2706 */
2707 case TCPOPT_SIGNATURE:
2708 if (optlen != TCPOLEN_SIGNATURE)
2709 continue;
2710 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2711 break;
2712#endif /* TCP_SIGNATURE */
984263bc
MD
2713 default:
2714 continue;
2715 }
2716 }
2717}
2718
2719/*
2720 * Pull out of band byte out of a segment so
2721 * it doesn't appear in the user's data queue.
2722 * It is still reflected in the segment length for
2723 * sequencing purposes.
95b22adf 2724 * "off" is the delayed to be dropped hdrlen.
984263bc
MD
2725 */
2726static void
95b22adf 2727tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
984263bc
MD
2728{
2729 int cnt = off + th->th_urp - 1;
2730
2731 while (cnt >= 0) {
2732 if (m->m_len > cnt) {
2733 char *cp = mtod(m, caddr_t) + cnt;
2734 struct tcpcb *tp = sototcpcb(so);
2735
2736 tp->t_iobc = *cp;
2737 tp->t_oobflags |= TCPOOB_HAVEDATA;
95b22adf 2738 bcopy(cp + 1, cp, m->m_len - cnt - 1);
984263bc
MD
2739 m->m_len--;
2740 if (m->m_flags & M_PKTHDR)
2741 m->m_pkthdr.len--;
2742 return;
2743 }
2744 cnt -= m->m_len;
2745 m = m->m_next;
4090d6ff 2746 if (m == NULL)
984263bc
MD
2747 break;
2748 }
2749 panic("tcp_pulloutofband");
2750}
2751
2752/*
2753 * Collect new round-trip time estimate
2754 * and update averages and current timeout.
2755 */
2756static void
95b22adf 2757tcp_xmit_timer(struct tcpcb *tp, int rtt)
984263bc 2758{
2256ba69 2759 int delta;
984263bc
MD
2760
2761 tcpstat.tcps_rttupdated++;
2762 tp->t_rttupdated++;
2763 if (tp->t_srtt != 0) {
2764 /*
2765 * srtt is stored as fixed point with 5 bits after the
2766 * binary point (i.e., scaled by 8). The following magic
2767 * is equivalent to the smoothing algorithm in rfc793 with
2768 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2769 * point). Adjust rtt to origin 0.
2770 */
2771 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2772 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2773
2774 if ((tp->t_srtt += delta) <= 0)
2775 tp->t_srtt = 1;
2776
2777 /*
2778 * We accumulate a smoothed rtt variance (actually, a
2779 * smoothed mean difference), then set the retransmit
2780 * timer to smoothed rtt + 4 times the smoothed variance.
2781 * rttvar is stored as fixed point with 4 bits after the
2782 * binary point (scaled by 16). The following is
2783 * equivalent to rfc793 smoothing with an alpha of .75
2784 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2785 * rfc793's wired-in beta.
2786 */
2787 if (delta < 0)
2788 delta = -delta;
2789 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2790 if ((tp->t_rttvar += delta) <= 0)
2791 tp->t_rttvar = 1;
2792 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2793 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2794 } else {
2795 /*
2796 * No rtt measurement yet - use the unsmoothed rtt.
2797 * Set the variance to half the rtt (so our first
2798 * retransmit happens at 3*rtt).
2799 */
2800 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2801 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2802 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2803 }
2804 tp->t_rtttime = 0;
2805 tp->t_rxtshift = 0;
2806
2807 /*
2808 * the retransmit should happen at rtt + 4 * rttvar.
2809 * Because of the way we do the smoothing, srtt and rttvar
2810 * will each average +1/2 tick of bias. When we compute
2811 * the retransmit timer, we want 1/2 tick of rounding and
2812 * 1 extra tick because of +-1/2 tick uncertainty in the
2813 * firing of the timer. The bias will give us exactly the
2814 * 1.5 tick we need. But, because the bias is
2815 * statistical, we have to test that we don't drop below
2816 * the minimum feasible timer (which is 2 ticks).
2817 */
2818 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2819 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2820
2821 /*
2822 * We received an ack for a packet that wasn't retransmitted;
2823 * it is probably safe to discard any error indications we've
2824 * received recently. This isn't quite right, but close enough
2825 * for now (a route might have failed after we sent a segment,
2826 * and the return path might not be symmetrical).
2827 */
2828 tp->t_softerror = 0;
2829}
2830
2831/*
2832 * Determine a reasonable value for maxseg size.
2833 * If the route is known, check route for mtu.
2834 * If none, use an mss that can be handled on the outgoing
2835 * interface without forcing IP to fragment; if bigger than
2836 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2837 * to utilize large mbufs. If no route is found, route has no mtu,
2838 * or the destination isn't local, use a default, hopefully conservative
2839 * size (usually 512 or the default IP max size, but no more than the mtu
2840 * of the interface), as we can't discover anything about intervening
2841 * gateways or networks. We also initialize the congestion/slow start
2842 * window to be a single segment if the destination isn't local.
2843 * While looking at the routing entry, we also initialize other path-dependent
2844 * parameters from pre-set or cached values in the routing entry.
2845 *
2846 * Also take into account the space needed for options that we
2847 * send regularly. Make maxseg shorter by that amount to assure
2848 * that we can send maxseg amount of data even when the options
2849 * are present. Store the upper limit of the length of options plus
2850 * data in maxopd.
2851 *
2852 * NOTE that this routine is only called when we process an incoming
2853 * segment, for outgoing segments only tcp_mssopt is called.
984263bc
MD
2854 */
2855void
95b22adf 2856tcp_mss(struct tcpcb *tp, int offer)
984263bc 2857{
2256ba69 2858 struct rtentry *rt;
984263bc 2859 struct ifnet *ifp;
2256ba69 2860 int rtt, mss;
984263bc
MD
2861 u_long bufsize;
2862 struct inpcb *inp = tp->t_inpcb;
2863 struct socket *so;
984263bc 2864#ifdef INET6
d24ce1dc 2865 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
2866 size_t min_protoh = isipv6 ?
2867 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2868 sizeof(struct tcpiphdr);
2869#else
d24ce1dc 2870 const boolean_t isipv6 = FALSE;
984263bc
MD
2871 const size_t min_protoh = sizeof(struct tcpiphdr);
2872#endif
2873
2874 if (isipv6)
2875 rt = tcp_rtlookup6(&inp->inp_inc);
2876 else
2877 rt = tcp_rtlookup(&inp->inp_inc);
2878 if (rt == NULL) {
2879 tp->t_maxopd = tp->t_maxseg =
d24ce1dc 2880 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
984263bc
MD
2881 return;
2882 }
2883 ifp = rt->rt_ifp;
2884 so = inp->inp_socket;
2885
984263bc
MD
2886 /*
2887 * Offer == 0 means that there was no MSS on the SYN segment,
b235ad6d
MD
2888 * in this case we use either the interface mtu or tcp_mssdflt.
2889 *
2890 * An offer which is too large will be cut down later.
984263bc 2891 */
5b0b9fa5 2892 if (offer == 0) {
b235ad6d
MD
2893 if (isipv6) {
2894 if (in6_localaddr(&inp->in6p_faddr)) {
2895 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2896 min_protoh;
2897 } else {
2898 offer = tcp_v6mssdflt;
2899 }
2900 } else {
2901 if (in_localaddr(inp->inp_faddr))
2902 offer = ifp->if_mtu - min_protoh;
2903 else
2904 offer = tcp_mssdflt;
2905 }
5b0b9fa5 2906 }
b235ad6d
MD
2907
2908 /*
2909 * Prevent DoS attack with too small MSS. Round up
2910 * to at least minmss.
2911 *
2912 * Sanity check: make sure that maxopd will be large
2913 * enough to allow some data on segments even is the
2914 * all the option space is used (40bytes). Otherwise
2915 * funny things may happen in tcp_output.
2916 */
2917 offer = max(offer, tcp_minmss);
2918 offer = max(offer, 64);
2919
27b8aee3 2920 rt->rt_rmx.rmx_mssopt = offer;
984263bc
MD
2921
2922 /*
2923 * While we're here, check if there's an initial rtt
2924 * or rttvar. Convert from the route-table units
2925 * to scaled multiples of the slow timeout timer.
2926 */
2927 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2928 /*
2929 * XXX the lock bit for RTT indicates that the value
2930 * is also a minimum value; this is subject to time.
2931 */
2932 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2933 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2934 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2935 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2936 tcpstat.tcps_usedrtt++;
2937 if (rt->rt_rmx.rmx_rttvar) {
2938 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2939 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2940 tcpstat.tcps_usedrttvar++;
2941 } else {
2942 /* default variation is +- 1 rtt */
2943 tp->t_rttvar =
2944 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2945 }
2946 TCPT_RANGESET(tp->t_rxtcur,
2947 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2948 tp->t_rttmin, TCPTV_REXMTMAX);
2949 }
b235ad6d 2950
984263bc
MD
2951 /*
2952 * if there's an mtu associated with the route, use it
b235ad6d
MD
2953 * else, use the link mtu. Take the smaller of mss or offer
2954 * as our final mss.
984263bc 2955 */
b235ad6d 2956 if (rt->rt_rmx.rmx_mtu) {
984263bc 2957 mss = rt->rt_rmx.rmx_mtu - min_protoh;
b235ad6d
MD
2958 } else {
2959 if (isipv6)
698ac46c 2960 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
b235ad6d 2961 else
984263bc 2962 mss = ifp->if_mtu - min_protoh;
984263bc
MD
2963 }
2964 mss = min(mss, offer);
b235ad6d 2965
984263bc
MD
2966 /*
2967 * maxopd stores the maximum length of data AND options
2968 * in a segment; maxseg is the amount of data in a normal
2969 * segment. We need to store this value (maxopd) apart
2970 * from maxseg, because now every segment carries options
2971 * and thus we normally have somewhat less data in segments.
2972 */
2973 tp->t_maxopd = mss;
2974
61896e3c 2975 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
27b8aee3 2976 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
984263bc 2977 mss -= TCPOLEN_TSTAMP_APPA;
984263bc
MD
2978
2979#if (MCLBYTES & (MCLBYTES - 1)) == 0
2980 if (mss > MCLBYTES)
2981 mss &= ~(MCLBYTES-1);
2982#else
2983 if (mss > MCLBYTES)
2984 mss = mss / MCLBYTES * MCLBYTES;
2985#endif
2986 /*
2987 * If there's a pipesize, change the socket buffer
2988 * to that size. Make the socket buffers an integral
2989 * number of mss units; if the mss is larger than
2990 * the socket buffer, decrease the mss.
2991 */
2992#ifdef RTV_SPIPE
2993 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2994#endif
6d49aa6f 2995 bufsize = so->so_snd.ssb_hiwat;
984263bc
MD
2996 if (bufsize < mss)
2997 mss = bufsize;
2998 else {
2999 bufsize = roundup(bufsize, mss);
3000 if (bufsize > sb_max)
3001 bufsize = sb_max;
6d49aa6f
MD
3002 if (bufsize > so->so_snd.ssb_hiwat)
3003 ssb_reserve(&so->so_snd, bufsize, so, NULL);
984263bc
MD
3004 }
3005 tp->t_maxseg = mss;
3006
3007#ifdef RTV_RPIPE
3008 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3009#endif
6d49aa6f 3010 bufsize = so->so_rcv.ssb_hiwat;
984263bc
MD
3011 if (bufsize > mss) {
3012 bufsize = roundup(bufsize, mss);
3013 if (bufsize > sb_max)
3014 bufsize = sb_max;
6cef7136
MD
3015 if (bufsize > so->so_rcv.ssb_hiwat) {
3016 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 3017 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
6cef7136
MD
3018 lwkt_reltoken(&so->so_rcv.ssb_token);
3019 }
984263bc
MD
3020 }
3021
3022 /*
3023 * Set the slow-start flight size depending on whether this
3024 * is a local network or not.
3025 */
4b52d5ee
JH
3026 if (tcp_do_rfc3390)
3027 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
4b52d5ee 3028 else
cd149066 3029 tp->snd_cwnd = mss;
984263bc
MD
3030
3031 if (rt->rt_rmx.rmx_ssthresh) {
3032 /*
3033 * There's some sort of gateway or interface
3034 * buffer limit on the path. Use this to set
3035 * the slow start threshhold, but set the
3036 * threshold to no less than 2*mss.
3037 */
3038 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3039 tcpstat.tcps_usedssthresh++;
3040 }
3041}
3042
3043/*
3044 * Determine the MSS option to send on an outgoing SYN.
3045 */
3046int
95b22adf 3047tcp_mssopt(struct tcpcb *tp)
984263bc
MD
3048{
3049 struct rtentry *rt;
3050#ifdef INET6
d24ce1dc
JH
3051 boolean_t isipv6 =
3052 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
3053 int min_protoh = isipv6 ?
3054 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3055 sizeof(struct tcpiphdr);
3056#else
d24ce1dc 3057 const boolean_t isipv6 = FALSE;
984263bc
MD
3058 const size_t min_protoh = sizeof(struct tcpiphdr);
3059#endif
3060
3061 if (isipv6)
3062 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3063 else
3064 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3065 if (rt == NULL)
3066 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3067
3068 return (rt->rt_ifp->if_mtu - min_protoh);
3069}
3070
984263bc
MD
3071/*
3072 * When a partial ack arrives, force the retransmission of the
91489f6b
JH
3073 * next unacknowledged segment. Do not exit Fast Recovery.
3074 *
3075 * Implement the Slow-but-Steady variant of NewReno by restarting the
3076 * the retransmission timer. Turn it off here so it can be restarted
3077 * later in tcp_output().
984263bc
MD
3078 */
3079static void
91489f6b 3080tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
984263bc 3081{
91489f6b
JH
3082 tcp_seq old_snd_nxt = tp->snd_nxt;
3083 u_long ocwnd = tp->snd_cwnd;
984263bc 3084
a48c5dd5 3085 tcp_callout_stop(tp, tp->tt_rexmt);
984263bc
MD
3086 tp->t_rtttime = 0;
3087 tp->snd_nxt = th->th_ack;
91489f6b
JH
3088 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3089 tp->snd_cwnd = tp->t_maxseg;
984263bc 3090 tp->t_flags |= TF_ACKNOW;
f23061d4 3091 tcp_output(tp);
91489f6b
JH
3092 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3093 tp->snd_nxt = old_snd_nxt;
3094 /* partial window deflation */
68947d12
ND
3095 if (ocwnd > acked)
3096 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3097 else
3098 tp->snd_cwnd = tp->t_maxseg;
91489f6b
JH
3099}
3100
3101/*
3102 * In contrast to the Slow-but-Steady NewReno variant,
3103 * we do not reset the retransmission timer for SACK retransmissions,
3104 * except when retransmitting snd_una.
3105 */
3106static void
3107tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3108{
3109 uint32_t pipe, seglen;
3110 tcp_seq nextrexmt;
3111 boolean_t lostdup;
3112 tcp_seq old_snd_nxt = tp->snd_nxt;
3113 u_long ocwnd = tp->snd_cwnd;
3114 int nseg = 0; /* consecutive new segments */
3115#define MAXBURST 4 /* limit burst of new packets on partial ack */
3116
3117 tp->t_rtttime = 0;
3118 pipe = tcp_sack_compute_pipe(tp);
3119 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3120 (!tcp_do_smartsack || nseg < MAXBURST) &&
3121 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3122 uint32_t sent;
3123 tcp_seq old_snd_max;
3124 int error;
3125
98cb2337
MD
3126 if (nextrexmt == tp->snd_max)
3127 ++nseg;
91489f6b
JH
3128 tp->snd_nxt = nextrexmt;
3129 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3130 old_snd_max = tp->snd_max;
3131 if (nextrexmt == tp->snd_una)
a48c5dd5 3132 tcp_callout_stop(tp, tp->tt_rexmt);
91489f6b
JH
3133 error = tcp_output(tp);
3134 if (error != 0)
3135 break;
3136 sent = tp->snd_nxt - nextrexmt;
3137 if (sent <= 0)
3138 break;
3139 if (!lostdup)
3140 pipe += sent;
3141 tcpstat.tcps_sndsackpack++;
3142 tcpstat.tcps_sndsackbyte += sent;
3143 if (SEQ_LT(nextrexmt, old_snd_max) &&
3144 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3145 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3146 }
3147 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3148 tp->snd_nxt = old_snd_nxt;
984263bc 3149 tp->snd_cwnd = ocwnd;
984263bc 3150}
0ecd93f9
MD
3151
3152/*
3153 * Reset idle time and keep-alive timer, typically called when a valid
3154 * tcp packet is received but may also be called when FASTKEEP is set
3155 * to prevent the previous long-timeout from calculating to a drop.
3156 *
3157 * Only update t_rcvtime for non-SYN packets.
3158 *
3159 * Handle the case where one side thinks the connection is established
3160 * but the other side has, say, rebooted without cleaning out the
3161 * connection. The SYNs could be construed as an attack and wind
3162 * up ignored, but in case it isn't an attack we can validate the
3163 * connection by forcing a keepalive.
3164 */
3165void
3166tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3167{
3168 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3169 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3170 tp->t_flags |= TF_KEEPALIVE;
3171 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3172 tcp_timer_keep);
3173 } else {
3174 tp->t_rcvtime = ticks;
3175 tp->t_flags &= ~TF_KEEPALIVE;
3176 tcp_callout_reset(tp, tp->tt_keep,
3177 tcp_getkeepidle(tp),
3178 tcp_timer_keep);
3179 }
3180 }
3181}
01d3427a
SZ
3182
3183static int
3184tcp_rmx_msl(const struct tcpcb *tp)
3185{
3186 struct rtentry *rt;
3187 struct inpcb *inp = tp->t_inpcb;
9855a4ef 3188 int msl;
01d3427a
SZ
3189#ifdef INET6
3190 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3191#else
3192 const boolean_t isipv6 = FALSE;
3193#endif
3194
3195 if (isipv6)
3196 rt = tcp_rtlookup6(&inp->inp_inc);
3197 else
3198 rt = tcp_rtlookup(&inp->inp_inc);
3199 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3200 return tcp_msl;
3201
9855a4ef
SZ
3202 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3203 if (msl == 0)
3204 msl = 1;
3205
3206 return msl;
01d3427a 3207}
8651f7f8
SZ
3208
3209static void
3210tcp_established(struct tcpcb *tp)
3211{
3212 tp->t_state = TCPS_ESTABLISHED;
3213 tcp_callout_reset(tp, tp->tt_keep, tcp_getkeepidle(tp), tcp_timer_keep);
be34e534
SZ
3214
3215 if (tp->t_flags & TF_SYN_WASLOST) {
3216 /*
3217 * RFC3390:
3218 * "If the SYN or SYN/ACK is lost, the initial window used by
3219 * a sender after a correctly transmitted SYN MUST be one
3220 * segment consisting of MSS bytes."
3221 */
3222 tp->snd_cwnd = tp->t_maxseg;
48a4676d
SZ
3223
3224 /*
3225 * RFC6298:
3226 * "If the timer expires awaiting the ACK of a SYN segment
3227 * and the TCP implementation is using an RTO less than 3
3228 * seconds, the RTO MUST be re-initialized to 3 seconds
3229 * when data transmission begins"
3230 */
3231 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3232 tp->t_rxtcur = TCPTV_RTOBASE3;
be34e534 3233 }
8651f7f8 3234}