tcp: Even for PAWS tolerance, no segments should follow segment with FIN
[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
66d6c637 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
efd4b327
JH
170static int tcp_do_eifel_detect = 1;
171SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
173
8acdb67c
JH
174static int tcp_do_abc = 1;
175SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
176 &tcp_do_abc, 0,
177 "TCP Appropriate Byte Counting (RFC 3465)");
178
9de1f696
SZ
179/*
180 * The following value actually takes range [25ms, 250ms],
181 * given that most modern systems use 1ms ~ 10ms as the unit
182 * of timestamp option.
183 */
184static u_int tcp_paws_tolerance = 25;
185SYSCTL_UINT(_net_inet_tcp, OID_AUTO, paws_tolerance, CTLFLAG_RW,
186 &tcp_paws_tolerance, 0, "RFC1323 PAWS tolerance");
187
91489f6b
JH
188/*
189 * Define as tunable for easy testing with SACK on and off.
190 * Warning: do not change setting in the middle of an existing active TCP flow,
191 * else strange things might happen to that flow.
192 */
6e5bbdda 193int tcp_do_sack = 1;
91489f6b
JH
194SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
195 &tcp_do_sack, 0, "Enable SACK Algorithms");
196
197int tcp_do_smartsack = 1;
198SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
199 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
200
1bdfd728
SZ
201int tcp_do_rescuesack = 1;
202SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
203 &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
204
c01f968f 205int tcp_aggressive_rescuesack = 0;
a098966f
SZ
206SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
207 &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
208
ccb518ea
SZ
209static int tcp_force_sackrxt = 1;
210SYSCTL_INT(_net_inet_tcp, OID_AUTO, force_sackrxt, CTLFLAG_RW,
211 &tcp_force_sackrxt, 0, "Allowed forced SACK retransmit burst");
212
338fe0b0 213int tcp_do_rfc3517bis = 1;
ffe35e17
SZ
214SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis, CTLFLAG_RW,
215 &tcp_do_rfc3517bis, 0, "Enable RFC3517 update");
216
5fd89c20
SZ
217int tcp_rfc3517bis_rxt = 0;
218SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis_rxt, CTLFLAG_RW,
219 &tcp_rfc3517bis_rxt, 0, "Enable RFC3517 retransmit update");
220
3edf7c37
RG
221SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
222 "TCP Segment Reassembly Queue");
223
224int tcp_reass_maxseg = 0;
225SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
226 &tcp_reass_maxseg, 0,
227 "Global maximum number of TCP Segments in Reassembly Queue");
228
229int tcp_reass_qsize = 0;
230SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
231 &tcp_reass_qsize, 0,
232 "Global number of TCP Segments currently in Reassembly Queue");
233
234static int tcp_reass_overflows = 0;
235SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
236 &tcp_reass_overflows, 0,
237 "Global number of TCP Segment Reassembly Queue Overflows");
238
5b0b9fa5
PA
239int tcp_do_autorcvbuf = 1;
240SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
241 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
242
243int tcp_autorcvbuf_inc = 16*1024;
244SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
245 &tcp_autorcvbuf_inc, 0,
246 "Incrementor step size of automatic receive buffer");
247
46e92930 248int tcp_autorcvbuf_max = 2*1024*1024;
5b0b9fa5
PA
249SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
250 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
251
0df7608b
SZ
252int tcp_sosend_agglim = 2;
253SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
254 &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
5b0b9fa5 255
0df7608b
SZ
256int tcp_sosend_async = 1;
257SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
258 &tcp_sosend_async, 0, "TCP asynchronized pru_send");
f2a3782e 259
12e60b57
SZ
260static int tcp_ignore_redun_dsack = 1;
261SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW,
262 &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK");
263
6c1bbf57
SZ
264static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t,
265 tcp_seq);
984263bc
MD
266static void tcp_pulloutofband(struct socket *,
267 struct tcphdr *, struct mbuf *, int);
268static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
269 struct mbuf *);
073ec6c4 270static void tcp_xmit_timer(struct tcpcb *, int, tcp_seq);
91489f6b 271static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
ccb518ea 272static void tcp_sack_rexmt(struct tcpcb *, boolean_t);
ffe35e17 273static boolean_t tcp_sack_limitedxmit(struct tcpcb *);
01d3427a 274static int tcp_rmx_msl(const struct tcpcb *);
8651f7f8 275static void tcp_established(struct tcpcb *);
5c99b248 276static boolean_t tcp_recv_dupack(struct tcpcb *, tcp_seq,
27f4bf33 277 const struct tcpopt *);
984263bc
MD
278
279/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
280#ifdef INET6
281#define ND6_HINT(tp) \
282do { \
283 if ((tp) && (tp)->t_inpcb && \
61896e3c 284 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
984263bc
MD
285 (tp)->t_inpcb->in6p_route.ro_rt) \
286 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
287} while (0)
288#else
289#define ND6_HINT(tp)
290#endif
291
292/*
293 * Indicate whether this ack should be delayed. We can delay the ack if
294 * - delayed acks are enabled and
295 * - there is no delayed ack timer in progress and
296 * - our last ack wasn't a 0-sized window. We never want to delay
297 * the ack that opens up a 0-sized window.
298 */
299#define DELAY_ACK(tp) \
a48c5dd5 300 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
61896e3c 301 !(tp->t_flags & TF_RXWIN0SENT))
984263bc 302
df9d7670
JH
303#define acceptable_window_update(tp, th, tiwin) \
304 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
305 (tp->snd_wl1 == th->th_seq && \
306 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
307 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
308
ffe35e17
SZ
309#define iceildiv(n, d) (((n)+(d)-1) / (d))
310#define need_early_retransmit(tp, ownd) \
311 (tcp_do_early_retransmit && \
312 (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \
4c0f4a00 313 ownd < ((tp->t_rxtthresh + 1) * tp->t_maxseg) && \
ffe35e17
SZ
314 tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \
315 (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \
316 tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg)))
317
9de1f696
SZ
318/*
319 * Returns TRUE, if this segment can be merged with the last
320 * pending segment in the reassemble queue and this segment
321 * does not overlap with the pending segment immediately
322 * preceeding the last pending segment.
323 */
324static __inline boolean_t
325tcp_paws_canreasslast(const struct tcpcb *tp, const struct tcphdr *th, int tlen)
326{
327 const struct tseg_qent *last, *prev;
328
329 last = TAILQ_LAST(&tp->t_segq, tsegqe_head);
330 if (last == NULL)
331 return FALSE;
332
333 /* This segment comes immediately after the last pending segment */
04e6f08e
SZ
334 if (last->tqe_th->th_seq + last->tqe_len == th->th_seq) {
335 if (last->tqe_th->th_flags & TH_FIN) {
336 /* No segments should follow segment w/ FIN */
337 return FALSE;
338 }
9de1f696 339 return TRUE;
04e6f08e 340 }
9de1f696
SZ
341
342 if (th->th_seq + tlen != last->tqe_th->th_seq)
343 return FALSE;
344 /* This segment comes immediately before the last pending segment */
345
346 prev = TAILQ_PREV(last, tsegqe_head, tqe_q);
347 if (prev == NULL) {
348 /*
349 * No pending preceeding segment, we assume this segment
350 * could be reassembled.
351 */
352 return TRUE;
353 }
354
355 /* This segment does not overlap with the preceeding segment */
356 if (SEQ_GEQ(th->th_seq, prev->tqe_th->th_seq + prev->tqe_len))
357 return TRUE;
358
359 return FALSE;
360}
361
e2289e66
SZ
362static __inline void
363tcp_ncr_update_rxtthresh(struct tcpcb *tp)
364{
365 int old_rxtthresh = tp->t_rxtthresh;
366 uint32_t ownd = tp->snd_max - tp->snd_una;
367
368 tp->t_rxtthresh = max(3, ((ownd / tp->t_maxseg) >> 1));
369 if (tp->t_rxtthresh != old_rxtthresh) {
370 tcp_sack_update_lostseq(&tp->scb, tp->snd_una,
371 tp->t_maxseg, tp->t_rxtthresh);
372 }
373}
374
984263bc 375static int
95b22adf 376tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
984263bc
MD
377{
378 struct tseg_qent *q;
379 struct tseg_qent *p = NULL;
984263bc
MD
380 struct tseg_qent *te;
381 struct socket *so = tp->t_inpcb->inp_socket;
382 int flags;
383
384 /*
61896e3c 385 * Call with th == NULL after become established to
984263bc
MD
386 * force pre-ESTABLISHED data up to user socket.
387 */
61896e3c 388 if (th == NULL)
984263bc
MD
389 goto present;
390
3edf7c37
RG
391 /*
392 * Limit the number of segments in the reassembly queue to prevent
393 * holding on to too many segments (and thus running out of mbufs).
394 * Make sure to let the missing segment through which caused this
395 * queue. Always keep one global queue entry spare to be able to
396 * process the missing segment.
397 */
398 if (th->th_seq != tp->rcv_nxt &&
399 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
400 tcp_reass_overflows++;
401 tcpstat.tcps_rcvmemdrop++;
402 m_freem(m);
91489f6b
JH
403 /* no SACK block to report */
404 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
3edf7c37
RG
405 return (0);
406 }
407
ba4e3dbe 408 /* Allocate a new queue entry. */
884717e1 409 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
984263bc
MD
410 if (te == NULL) {
411 tcpstat.tcps_rcvmemdrop++;
412 m_freem(m);
91489f6b
JH
413 /* no SACK block to report */
414 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
984263bc
MD
415 return (0);
416 }
2d23a8be 417 atomic_add_int(&tcp_reass_qsize, 1);
984263bc
MD
418
419 /*
420 * Find a segment which begins after this one does.
421 */
0f9e45de 422 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) {
984263bc
MD
423 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
424 break;
425 p = q;
426 }
427
428 /*
429 * If there is a preceding segment, it may provide some of
430 * our data already. If so, drop the data from the incoming
431 * segment. If it provides all of our data, drop us.
432 */
433 if (p != NULL) {
76cf5e41 434 tcp_seq_diff_t i;
91489f6b 435
984263bc
MD
436 /* conversion to int (in i) handles seq wraparound */
437 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
91489f6b 438 if (i > 0) { /* overlaps preceding segment */
c7e6499a
SZ
439 tp->sack_flags |=
440 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
91489f6b
JH
441 /* enclosing block starts w/ preceding segment */
442 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
984263bc 443 if (i >= *tlenp) {
53432bee
SZ
444 if (th->th_flags & TH_FIN)
445 p->tqe_th->th_flags |= TH_FIN;
446
91489f6b 447 /* preceding encloses incoming segment */
3a5d999b
SZ
448 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
449 p->tqe_th->th_seq + p->tqe_len,
450 p->tqe_th->th_flags);
984263bc
MD
451 tcpstat.tcps_rcvduppack++;
452 tcpstat.tcps_rcvdupbyte += *tlenp;
453 m_freem(m);
efda3bd0 454 kfree(te, M_TSEGQ);
2d23a8be 455 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
456 /*
457 * Try to present any queued data
458 * at the left window edge to the user.
459 * This is needed after the 3-WHS
460 * completes.
461 */
462 goto present; /* ??? */
463 }
464 m_adj(m, i);
465 *tlenp -= i;
466 th->th_seq += i;
91489f6b 467 /* incoming segment end is enclosing block end */
3a5d999b
SZ
468 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
469 th->th_seq + *tlenp, th->th_flags);
91489f6b
JH
470 /* trim end of reported D-SACK block */
471 tp->reportblk.rblk_end = th->th_seq;
984263bc
MD
472 }
473 }
474 tcpstat.tcps_rcvoopack++;
475 tcpstat.tcps_rcvoobyte += *tlenp;
476
477 /*
478 * While we overlap succeeding segments trim them or,
479 * if they are completely covered, dequeue them.
480 */
481 while (q) {
76cf5e41 482 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
91489f6b 483 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
3a5d999b 484 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
91489f6b
JH
485 struct tseg_qent *nq;
486
984263bc
MD
487 if (i <= 0)
488 break;
c7e6499a
SZ
489 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
490 /* first time through */
491 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
91489f6b
JH
492 tp->encloseblk = tp->reportblk;
493 /* report trailing duplicate D-SACK segment */
494 tp->reportblk.rblk_start = q->tqe_th->th_seq;
495 }
c7e6499a 496 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) &&
3a5d999b 497 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
91489f6b 498 /* extend enclosing block if one exists */
3a5d999b 499 tp->encloseblk.rblk_end = qend_sack;
91489f6b 500 }
984263bc
MD
501 if (i < q->tqe_len) {
502 q->tqe_th->th_seq += i;
503 q->tqe_len -= i;
504 m_adj(q->tqe_m, i);
505 break;
506 }
507
53432bee
SZ
508 if (q->tqe_th->th_flags & TH_FIN)
509 th->th_flags |= TH_FIN;
510
0f9e45de
SZ
511 nq = TAILQ_NEXT(q, tqe_q);
512 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
984263bc 513 m_freem(q->tqe_m);
efda3bd0 514 kfree(q, M_TSEGQ);
2d23a8be 515 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
516 q = nq;
517 }
518
519 /* Insert the new segment queue entry into place. */
520 te->tqe_m = m;
521 te->tqe_th = th;
522 te->tqe_len = *tlenp;
523
91489f6b
JH
524 /* check if can coalesce with following segment */
525 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
53432bee 526 tcp_seq tend_sack;
91489f6b
JH
527
528 te->tqe_len += q->tqe_len;
a174690a 529 if (q->tqe_th->th_flags & TH_FIN)
f16f8cc3 530 te->tqe_th->th_flags |= TH_FIN;
53432bee
SZ
531 tend_sack = TCP_SACK_BLKEND(te->tqe_th->th_seq + te->tqe_len,
532 te->tqe_th->th_flags);
533
91489f6b 534 m_cat(te->tqe_m, q->tqe_m);
3a5d999b 535 tp->encloseblk.rblk_end = tend_sack;
91489f6b
JH
536 /*
537 * When not reporting a duplicate segment, use
538 * the larger enclosing block as the SACK block.
539 */
c7e6499a 540 if (!(tp->sack_flags & TSACK_F_DUPSEG))
3a5d999b 541 tp->reportblk.rblk_end = tend_sack;
0f9e45de 542 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
efda3bd0 543 kfree(q, M_TSEGQ);
2d23a8be 544 atomic_add_int(&tcp_reass_qsize, -1);
91489f6b
JH
545 }
546
984263bc 547 if (p == NULL) {
0f9e45de 548 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q);
984263bc 549 } else {
91489f6b
JH
550 /* check if can coalesce with preceding segment */
551 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
53432bee
SZ
552 if (te->tqe_th->th_flags & TH_FIN)
553 p->tqe_th->th_flags |= TH_FIN;
a174690a
JH
554 p->tqe_len += te->tqe_len;
555 m_cat(p->tqe_m, te->tqe_m);
91489f6b
JH
556 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
557 /*
558 * When not reporting a duplicate segment, use
559 * the larger enclosing block as the SACK block.
560 */
c7e6499a 561 if (!(tp->sack_flags & TSACK_F_DUPSEG))
91489f6b 562 tp->reportblk.rblk_start = p->tqe_th->th_seq;
efda3bd0 563 kfree(te, M_TSEGQ);
2d23a8be
MD
564 atomic_add_int(&tcp_reass_qsize, -1);
565 } else {
0f9e45de 566 TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q);
2d23a8be 567 }
984263bc
MD
568 }
569
570present:
571 /*
572 * Present data to user, advancing rcv_nxt through
573 * completed sequence space.
574 */
575 if (!TCPS_HAVEESTABLISHED(tp->t_state))
576 return (0);
0f9e45de 577 q = TAILQ_FIRST(&tp->t_segq);
61896e3c 578 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
984263bc 579 return (0);
91489f6b 580 tp->rcv_nxt += q->tqe_len;
c7e6499a 581 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
91489f6b
JH
582 /* no SACK block to report since ACK advanced */
583 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
584 }
585 /* no enclosing block to report since ACK advanced */
c7e6499a 586 tp->sack_flags &= ~TSACK_F_ENCLOSESEG;
91489f6b 587 flags = q->tqe_th->th_flags & TH_FIN;
0f9e45de
SZ
588 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
589 KASSERT(TAILQ_EMPTY(&tp->t_segq) ||
590 TAILQ_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
91489f6b 591 ("segment not coalesced"));
6cef7136 592 if (so->so_state & SS_CANTRCVMORE) {
91489f6b 593 m_freem(q->tqe_m);
6cef7136
MD
594 } else {
595 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 596 ssb_appendstream(&so->so_rcv, q->tqe_m);
6cef7136
MD
597 lwkt_reltoken(&so->so_rcv.ssb_token);
598 }
efda3bd0 599 kfree(q, M_TSEGQ);
2d23a8be 600 atomic_add_int(&tcp_reass_qsize, -1);
984263bc
MD
601 ND6_HINT(tp);
602 sorwakeup(so);
603 return (flags);
604}
605
606/*
607 * TCP input routine, follows pages 65-76 of the
608 * protocol specification dated September, 1981 very closely.
609 */
610#ifdef INET6
611int
95b22adf 612tcp6_input(struct mbuf **mp, int *offp, int proto)
984263bc 613{
2256ba69 614 struct mbuf *m = *mp;
984263bc
MD
615 struct in6_ifaddr *ia6;
616
617 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
618
619 /*
620 * draft-itojun-ipv6-tcp-to-anycast
621 * better place to put this in?
622 */
623 ia6 = ip6_getdstifaddr(m);
d24ce1dc 624 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
984263bc
MD
625 struct ip6_hdr *ip6;
626
627 ip6 = mtod(m, struct ip6_hdr *);
628 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
f23061d4 629 offsetof(struct ip6_hdr, ip6_dst));
d24ce1dc 630 return (IPPROTO_DONE);
984263bc
MD
631 }
632
002c1265 633 tcp_input(mp, offp, proto);
d24ce1dc 634 return (IPPROTO_DONE);
984263bc
MD
635}
636#endif
637
002c1265
MD
638int
639tcp_input(struct mbuf **mp, int *offp, int proto)
984263bc 640{
002c1265 641 int off0;
2256ba69
RG
642 struct tcphdr *th;
643 struct ip *ip = NULL;
644 struct ipovly *ipov;
645 struct inpcb *inp = NULL;
984263bc
MD
646 u_char *optp = NULL;
647 int optlen = 0;
b1992928
MD
648 int tlen, off;
649 int len = 0;
984263bc 650 int drop_hdrlen;
2256ba69
RG
651 struct tcpcb *tp = NULL;
652 int thflags;
4090d6ff 653 struct socket *so = NULL;
61896e3c 654 int todrop, acked;
27f4bf33
SZ
655 boolean_t ourfinisacked, needoutput = FALSE, delayed_dupack = FALSE;
656 tcp_seq th_dupack = 0; /* XXX gcc warning */
984263bc 657 u_long tiwin;
61896e3c 658 int recvwin;
984263bc 659 struct tcpopt to; /* options in this segment */
984263bc
MD
660 struct sockaddr_in *next_hop = NULL;
661 int rstreason; /* For badport_bandlim accounting purposes */
d371a63a 662 int cpu;
984263bc 663 struct ip6_hdr *ip6 = NULL;
002c1265 664 struct mbuf *m;
984263bc 665#ifdef INET6
d24ce1dc 666 boolean_t isipv6;
984263bc 667#else
d24ce1dc 668 const boolean_t isipv6 = FALSE;
984263bc
MD
669#endif
670#ifdef TCPDEBUG
671 short ostate = 0;
672#endif
673
002c1265
MD
674 off0 = *offp;
675 m = *mp;
676 *mp = NULL;
a00138cb 677
d24ce1dc
JH
678 tcpstat.tcps_rcvtotal++;
679
5de23090
SZ
680 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
681 struct m_tag *mtag;
682
683 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
684 KKASSERT(mtag != NULL);
685 next_hop = m_tag_data(mtag);
984263bc 686 }
d24ce1dc 687
984263bc 688#ifdef INET6
d24ce1dc 689 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
984263bc 690#endif
984263bc
MD
691
692 if (isipv6) {
693 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
694 ip6 = mtod(m, struct ip6_hdr *);
407e896e 695 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
984263bc
MD
696 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
697 tcpstat.tcps_rcvbadsum++;
698 goto drop;
699 }
700 th = (struct tcphdr *)((caddr_t)ip6 + off0);
701
702 /*
703 * Be proactive about unspecified IPv6 address in source.
704 * As we use all-zero to indicate unbounded/unconnected pcb,
705 * unspecified IPv6 address can be used to confuse us.
706 *
707 * Note that packets with unspecified IPv6 destination is
708 * already dropped in ip6_input.
709 */
710 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
711 /* XXX stat */
712 goto drop;
713 }
714 } else {
715 /*
716 * Get IP and TCP header together in first mbuf.
717 * Note: IP leaves IP header in first mbuf.
718 */
719 if (off0 > sizeof(struct ip)) {
bddf0751 720 ip_stripoptions(m);
984263bc
MD
721 off0 = sizeof(struct ip);
722 }
55d829f8
JH
723 /* already checked and pulled up in ip_demux() */
724 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
61896e3c 725 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
984263bc
MD
726 ip = mtod(m, struct ip *);
727 ipov = (struct ipovly *)ip;
728 th = (struct tcphdr *)((caddr_t)ip + off0);
729 tlen = ip->ip_len;
730
731 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
732 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
733 th->th_sum = m->m_pkthdr.csum_data;
734 else
735 th->th_sum = in_pseudo(ip->ip_src.s_addr,
736 ip->ip_dst.s_addr,
737 htonl(m->m_pkthdr.csum_data +
738 ip->ip_len +
739 IPPROTO_TCP));
740 th->th_sum ^= 0xffff;
741 } else {
742 /*
743 * Checksum extended TCP header and data.
744 */
745 len = sizeof(struct ip) + tlen;
407e896e 746 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
984263bc
MD
747 ipov->ih_len = (u_short)tlen;
748 ipov->ih_len = htons(ipov->ih_len);
749 th->th_sum = in_cksum(m, len);
750 }
751 if (th->th_sum) {
752 tcpstat.tcps_rcvbadsum++;
753 goto drop;
754 }
755#ifdef INET6
756 /* Re-initialization for later version check */
757 ip->ip_v = IPVERSION;
758#endif
759 }
760
761 /*
762 * Check that TCP offset makes sense,
763 * pull out TCP options and adjust length. XXX
764 */
765 off = th->th_off << 2;
55d829f8
JH
766 /* already checked and pulled up in ip_demux() */
767 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
61896e3c 768 ("bad TCP data offset %d (tlen %d)", off, tlen));
984263bc
MD
769 tlen -= off; /* tlen is used instead of ti->ti_len */
770 if (off > sizeof(struct tcphdr)) {
771 if (isipv6) {
002c1265 772 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
984263bc
MD
773 ip6 = mtod(m, struct ip6_hdr *);
774 th = (struct tcphdr *)((caddr_t)ip6 + off0);
775 } else {
55d829f8
JH
776 /* already pulled up in ip_demux() */
777 KASSERT(m->m_len >= sizeof(struct ip) + off,
61896e3c
JH
778 ("TCP header and options not in one mbuf: "
779 "m_len %d, off %d", m->m_len, off));
984263bc
MD
780 }
781 optlen = off - sizeof(struct tcphdr);
782 optp = (u_char *)(th + 1);
783 }
784 thflags = th->th_flags;
785
786#ifdef TCP_DROP_SYNFIN
787 /*
788 * If the drop_synfin option is enabled, drop all packets with
789 * both the SYN and FIN bits set. This prevents e.g. nmap from
790 * identifying the TCP/IP stack.
791 *
792 * This is a violation of the TCP specification.
793 */
61896e3c 794 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
984263bc
MD
795 goto drop;
796#endif
797
798 /*
799 * Convert TCP protocol specific fields to host format.
800 */
801 th->th_seq = ntohl(th->th_seq);
802 th->th_ack = ntohl(th->th_ack);
803 th->th_win = ntohs(th->th_win);
804 th->th_urp = ntohs(th->th_urp);
805
806 /*
d24ce1dc 807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
984263bc
MD
808 * until after ip6_savecontrol() is called and before other functions
809 * which don't want those proto headers.
810 * Because ip6_savecontrol() is going to parse the mbuf to
811 * search for data to be passed up to user-land, it wants mbuf
812 * parameters to be unchanged.
813 * XXX: the call of ip6_savecontrol() has been obsoleted based on
814 * latest version of the advanced API (20020110).
815 */
816 drop_hdrlen = off0 + off;
817
818 /*
819 * Locate pcb for segment.
820 */
821findpcb:
822 /* IPFIREWALL_FORWARD section */
d24ce1dc 823 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
984263bc
MD
824 /*
825 * Transparently forwarded. Pretend to be the destination.
d24ce1dc 826 * already got one like this?
984263bc 827 */
6ca1a1cd 828 cpu = mycpu->gd_cpuid;
6ca1a1cd 829 inp = in_pcblookup_hash(&tcbinfo[cpu],
d371a63a 830 ip->ip_src, th->th_sport,
984263bc
MD
831 ip->ip_dst, th->th_dport,
832 0, m->m_pkthdr.rcvif);
833 if (!inp) {
83be63fe
JH
834 /*
835 * It's new. Try to find the ambushing socket.
836 */
837
838 /*
839 * The rest of the ipfw code stores the port in
840 * host order. XXX
841 * (The IP address is still in network order.)
842 */
843 in_port_t dport = next_hop->sin_port ?
844 htons(next_hop->sin_port) :
845 th->th_dport;
846
d371a63a 847 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
83be63fe 848 next_hop->sin_addr.s_addr, dport);
d371a63a 849 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc 850 ip->ip_src, th->th_sport,
83be63fe 851 next_hop->sin_addr, dport,
984263bc
MD
852 1, m->m_pkthdr.rcvif);
853 }
854 } else {
6ca1a1cd 855 if (isipv6) {
d371a63a 856 inp = in6_pcblookup_hash(&tcbinfo[0],
984263bc
MD
857 &ip6->ip6_src, th->th_sport,
858 &ip6->ip6_dst, th->th_dport,
859 1, m->m_pkthdr.rcvif);
6ca1a1cd 860 } else {
6ca1a1cd 861 cpu = mycpu->gd_cpuid;
6ca1a1cd 862 inp = in_pcblookup_hash(&tcbinfo[cpu],
984263bc
MD
863 ip->ip_src, th->th_sport,
864 ip->ip_dst, th->th_dport,
865 1, m->m_pkthdr.rcvif);
6ca1a1cd 866 }
984263bc 867 }
984263bc
MD
868
869 /*
870 * If the state is CLOSED (i.e., TCB does not exist) then
871 * all data in the incoming segment is discarded.
872 * If the TCB exists but is in CLOSED state, it is embryonic,
873 * but should either do a listen or a connect soon.
874 */
875 if (inp == NULL) {
876 if (log_in_vain) {
877#ifdef INET6
878 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
879#else
1141eb20
JH
880 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
881 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
984263bc
MD
882#endif
883 if (isipv6) {
884 strcpy(dbuf, "[");
984263bc 885 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
984263bc 886 strcat(dbuf, "]");
95b22adf
JH
887 strcpy(sbuf, "[");
888 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
984263bc
MD
889 strcat(sbuf, "]");
890 } else {
891 strcpy(dbuf, inet_ntoa(ip->ip_dst));
892 strcpy(sbuf, inet_ntoa(ip->ip_src));
893 }
894 switch (log_in_vain) {
895 case 1:
61896e3c 896 if (!(thflags & TH_SYN))
984263bc
MD
897 break;
898 case 2:
899 log(LOG_INFO,
900 "Connection attempt to TCP %s:%d "
901 "from %s:%d flags:0x%02x\n",
902 dbuf, ntohs(th->th_dport), sbuf,
903 ntohs(th->th_sport), thflags);
904 break;
905 default:
906 break;
907 }
908 }
d24ce1dc 909 if (blackhole) {
984263bc
MD
910 switch (blackhole) {
911 case 1:
912 if (thflags & TH_SYN)
913 goto drop;
914 break;
915 case 2:
916 goto drop;
917 default:
918 goto drop;
919 }
920 }
921 rstreason = BANDLIM_RST_CLOSEDPORT;
922 goto dropwithreset;
923 }
61896e3c
JH
924
925#ifdef IPSEC
926 if (isipv6) {
927 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
928 ipsec6stat.in_polvio++;
929 goto drop;
930 }
931 } else {
932 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
933 ipsecstat.in_polvio++;
934 goto drop;
935 }
936 }
937#endif
938#ifdef FAST_IPSEC
939 if (isipv6) {
940 if (ipsec6_in_reject(m, inp))
941 goto drop;
942 } else {
943 if (ipsec4_in_reject(m, inp))
944 goto drop;
945 }
946#endif
95926362
MD
947 /* Check the minimum TTL for socket. */
948#ifdef INET6
949 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
950 goto drop;
951#endif
61896e3c 952
984263bc
MD
953 tp = intotcpcb(inp);
954 if (tp == NULL) {
955 rstreason = BANDLIM_RST_CLOSEDPORT;
956 goto dropwithreset;
957 }
eb594563 958 if (tp->t_state <= TCPS_CLOSED)
984263bc
MD
959 goto drop;
960
984263bc 961 so = inp->inp_socket;
47654766 962
984263bc 963#ifdef TCPDEBUG
47654766
JH
964 if (so->so_options & SO_DEBUG) {
965 ostate = tp->t_state;
966 if (isipv6)
95b22adf 967 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
47654766 968 else
95b22adf 969 bcopy(ip, tcp_saveipgen, sizeof(*ip));
47654766
JH
970 tcp_savetcp = *th;
971 }
984263bc 972#endif
47654766 973
407e896e 974 bzero(&to, sizeof to);
d24ce1dc 975
47654766
JH
976 if (so->so_options & SO_ACCEPTCONN) {
977 struct in_conninfo inc;
978
984263bc 979#ifdef INET6
d24ce1dc 980 inc.inc_isipv6 = (isipv6 == TRUE);
984263bc
MD
981#endif
982 if (isipv6) {
983 inc.inc6_faddr = ip6->ip6_src;
984 inc.inc6_laddr = ip6->ip6_dst;
985 inc.inc6_route.ro_rt = NULL; /* XXX */
986 } else {
987 inc.inc_faddr = ip->ip_src;
988 inc.inc_laddr = ip->ip_dst;
989 inc.inc_route.ro_rt = NULL; /* XXX */
990 }
991 inc.inc_fport = th->th_sport;
992 inc.inc_lport = th->th_dport;
993
61896e3c
JH
994 /*
995 * If the state is LISTEN then ignore segment if it contains
984263bc
MD
996 * a RST. If the segment contains an ACK then it is bad and
997 * send a RST. If it does not contain a SYN then it is not
998 * interesting; drop it.
999 *
1000 * If the state is SYN_RECEIVED (syncache) and seg contains
1001 * an ACK, but not for our SYN/ACK, send a RST. If the seg
1002 * contains a RST, check the sequence number to see if it
1003 * is a valid reset segment.
1004 */
61896e3c
JH
1005 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
1006 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
984263bc
MD
1007 if (!syncache_expand(&inc, th, &so, m)) {
1008 /*
1009 * No syncache entry, or ACK was not
1010 * for our SYN/ACK. Send a RST.
1011 */
1012 tcpstat.tcps_badsyn++;
1013 rstreason = BANDLIM_RST_OPENPORT;
1014 goto dropwithreset;
1015 }
2d23a8be
MD
1016
1017 /*
1018 * Could not complete 3-way handshake,
1019 * connection is being closed down, and
1020 * syncache will free mbuf.
1021 */
984263bc 1022 if (so == NULL)
002c1265 1023 return(IPPROTO_DONE);
2d23a8be
MD
1024
1025 /*
1026 * We must be in the correct protocol thread
1027 * for this connection.
1028 */
1029 KKASSERT(so->so_port == &curthread->td_msgport);
1030
984263bc
MD
1031 /*
1032 * Socket is created in state SYN_RECEIVED.
1033 * Continue processing segment.
1034 */
ed894f8c 1035 inp = so->so_pcb;
984263bc
MD
1036 tp = intotcpcb(inp);
1037 /*
1038 * This is what would have happened in
1039 * tcp_output() when the SYN,ACK was sent.
1040 */
1041 tp->snd_up = tp->snd_una;
1042 tp->snd_max = tp->snd_nxt = tp->iss + 1;
1043 tp->last_ack_sent = tp->rcv_nxt;
df1d2774 1044
984263bc
MD
1045 goto after_listen;
1046 }
1047 if (thflags & TH_RST) {
1048 syncache_chkrst(&inc, th);
1049 goto drop;
1050 }
1051 if (thflags & TH_ACK) {
1052 syncache_badack(&inc);
1053 tcpstat.tcps_badsyn++;
1054 rstreason = BANDLIM_RST_OPENPORT;
1055 goto dropwithreset;
1056 }
1057 goto drop;
1058 }
1059
1060 /*
61896e3c 1061 * Segment's flags are (SYN) or (SYN | FIN).
984263bc
MD
1062 */
1063#ifdef INET6
1064 /*
1065 * If deprecated address is forbidden,
1066 * we do not accept SYN to deprecated interface
1067 * address to prevent any new inbound connection from
1068 * getting established.
1069 * When we do not accept SYN, we send a TCP RST,
1070 * with deprecated source address (instead of dropping
1071 * it). We compromise it as it is much better for peer
1072 * to send a RST, and RST will be the final packet
1073 * for the exchange.
1074 *
1075 * If we do not forbid deprecated addresses, we accept
1076 * the SYN packet. RFC2462 does not suggest dropping
1077 * SYN in this case.
1078 * If we decipher RFC2462 5.5.4, it says like this:
1079 * 1. use of deprecated addr with existing
1080 * communication is okay - "SHOULD continue to be
1081 * used"
1082 * 2. use of it with new communication:
1083 * (2a) "SHOULD NOT be used if alternate address
61896e3c 1084 * with sufficient scope is available"
984263bc
MD
1085 * (2b) nothing mentioned otherwise.
1086 * Here we fall into (2b) case as we have no choice in
1087 * our source address selection - we must obey the peer.
1088 *
1089 * The wording in RFC2462 is confusing, and there are
1090 * multiple description text for deprecated address
1091 * handling - worse, they are not exactly the same.
1092 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1093 */
1094 if (isipv6 && !ip6_use_deprecated) {
1095 struct in6_ifaddr *ia6;
1096
1097 if ((ia6 = ip6_getdstifaddr(m)) &&
1098 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1099 tp = NULL;
1100 rstreason = BANDLIM_RST_OPENPORT;
1101 goto dropwithreset;
1102 }
1103 }
1104#endif
1105 /*
1106 * If it is from this socket, drop it, it must be forged.
1107 * Don't bother responding if the destination was a broadcast.
1108 */
1109 if (th->th_dport == th->th_sport) {
1110 if (isipv6) {
1111 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1112 &ip6->ip6_src))
1113 goto drop;
1114 } else {
1115 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1116 goto drop;
1117 }
1118 }
1119 /*
1120 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1121 *
1122 * Note that it is quite possible to receive unicast
1123 * link-layer packets with a broadcast IP address. Use
1124 * in_broadcast() to find them.
1125 */
61896e3c 1126 if (m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
1127 goto drop;
1128 if (isipv6) {
1129 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1130 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1131 goto drop;
1132 } else {
1133 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1134 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1135 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1136 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1137 goto drop;
1138 }
1139 /*
1140 * SYN appears to be valid; create compressed TCP state
1141 * for syncache, or perform t/tcp connection.
1142 */
1143 if (so->so_qlen <= so->so_qlimit) {
6c1bbf57 1144 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
b09567cc 1145 if (!syncache_add(&inc, &to, th, so, m))
984263bc 1146 goto drop;
2d23a8be
MD
1147
1148 /*
1149 * Entry added to syncache, mbuf used to
1150 * send SYN,ACK packet.
1151 */
b09567cc 1152 return(IPPROTO_DONE);
984263bc
MD
1153 }
1154 goto drop;
1155 }
984263bc 1156
2d23a8be
MD
1157after_listen:
1158 /*
1159 * Should not happen - syncache should pick up these connections.
1160 *
1161 * Once we are past handling listen sockets we must be in the
1162 * correct protocol processing thread.
1163 */
61896e3c 1164 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
2d23a8be 1165 KKASSERT(so->so_port == &curthread->td_msgport);
984263bc 1166
df1d2774
SZ
1167 /* Unscale the window into a 32-bit value. */
1168 if (!(thflags & TH_SYN))
1169 tiwin = th->th_win << tp->snd_scale;
1170 else
1171 tiwin = th->th_win;
1172
5b0b9fa5
PA
1173 /*
1174 * This is the second part of the MSS DoS prevention code (after
1175 * minmss on the sending side) and it deals with too many too small
1176 * tcp packets in a too short timeframe (1 second).
1177 *
153d34ac
MD
1178 * XXX Removed. This code was crap. It does not scale to network
1179 * speed, and default values break NFS. Gone.
5b0b9fa5 1180 */
153d34ac 1181 /* REMOVED */
5b0b9fa5 1182
984263bc
MD
1183 /*
1184 * Segment received on connection.
efca2b8e
MD
1185 *
1186 * Reset idle time and keep-alive timer. Don't waste time if less
0ecd93f9 1187 * then a second has elapsed.
984263bc 1188 */
0ecd93f9
MD
1189 if ((int)(ticks - tp->t_rcvtime) > hz)
1190 tcp_timer_keep_activity(tp, thflags);
984263bc
MD
1191
1192 /*
1193 * Process options.
1194 * XXX this is tradtitional behavior, may need to be cleaned up.
1195 */
6c1bbf57 1196 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
ad0af98b 1197 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
df1d2774 1198 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
984263bc 1199 tp->t_flags |= TF_RCVD_SCALE;
df1d2774 1200 tp->snd_scale = to.to_requested_s_scale;
984263bc 1201 }
df1d2774
SZ
1202
1203 /*
1204 * Initial send window; will be updated upon next ACK
1205 */
1206 tp->snd_wnd = th->th_win;
1207
984263bc
MD
1208 if (to.to_flags & TOF_TS) {
1209 tp->t_flags |= TF_RCVD_TSTMP;
1210 tp->ts_recent = to.to_tsval;
1211 tp->ts_recent_age = ticks;
1212 }
f34fd0f2
SZ
1213 if (!(to.to_flags & TOF_MSS))
1214 to.to_mss = 0;
1215 tcp_mss(tp, to.to_mss);
91489f6b
JH
1216 /*
1217 * Only set the TF_SACK_PERMITTED per-connection flag
1218 * if we got a SACK_PERMITTED option from the other side
1219 * and the global tcp_do_sack variable is true.
1220 */
1221 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1222 tp->t_flags |= TF_SACK_PERMITTED;
984263bc
MD
1223 }
1224
1225 /*
1226 * Header prediction: check for the two common cases
1227 * of a uni-directional data xfer. If the packet has
1228 * no control flags, is in-sequence, the window didn't
1229 * change and we're not retransmitting, it's a
1230 * candidate. If the length is zero and the ack moved
1231 * forward, we're the sender side of the xfer. Just
1232 * free the data acked & wake any higher level process
1233 * that was blocked waiting for space. If the length
1234 * is non-zero and the ack didn't move, we're the
1235 * receiver side. If we're getting packets in-order
1236 * (the reassembly queue is empty), add the data to
1237 * the socket buffer and note that we need a delayed ack.
1238 * Make sure that the hidden state-flags are also off.
1239 * Since we check for TCPS_ESTABLISHED above, it can only
1240 * be TH_NEEDSYN.
1241 */
1242 if (tp->t_state == TCPS_ESTABLISHED &&
1243 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
61896e3c
JH
1244 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1245 (!(to.to_flags & TOF_TS) ||
984263bc 1246 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
984263bc 1247 th->th_seq == tp->rcv_nxt &&
984263bc
MD
1248 tp->snd_nxt == tp->snd_max) {
1249
1250 /*
1251 * If last ACK falls within this segment's sequence numbers,
1252 * record the timestamp.
1253 * NOTE that the test is modified according to the latest
1254 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1255 */
61896e3c 1256 if ((to.to_flags & TOF_TS) &&
984263bc
MD
1257 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1258 tp->ts_recent_age = ticks;
1259 tp->ts_recent = to.to_tsval;
1260 }
1261
1262 if (tlen == 0) {
1263 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1264 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1265 tp->snd_cwnd >= tp->snd_wnd &&
95b22adf 1266 !IN_FASTRECOVERY(tp)) {
984263bc 1267 /*
be23faf1 1268 * This is a pure ack for outstanding data.
984263bc
MD
1269 */
1270 ++tcpstat.tcps_predack;
1271 /*
1272 * "bad retransmit" recovery
bfdb979e
JH
1273 *
1274 * If Eifel detection applies, then
1275 * it is deterministic, so use it
1276 * unconditionally over the old heuristic.
1277 * Otherwise, fall back to the old heuristic.
984263bc 1278 */
bfdb979e
JH
1279 if (tcp_do_eifel_detect &&
1280 (to.to_flags & TOF_TS) && to.to_tsecr &&
1c8b7a61 1281 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
bfdb979e
JH
1282 /* Eifel detection applicable. */
1283 if (to.to_tsecr < tp->t_rexmtTS) {
1284 tcp_revert_congestion_state(tp);
1285 ++tcpstat.tcps_eifeldetected;
928c3291
SZ
1286 if (tp->t_rxtshift != 1 ||
1287 ticks >= tp->t_badrxtwin)
1288 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
1289 }
1290 } else if (tp->t_rxtshift == 1 &&
1291 ticks < tp->t_badrxtwin) {
1292 tcp_revert_congestion_state(tp);
1293 ++tcpstat.tcps_rttdetected;
984263bc 1294 }
1c8b7a61
SZ
1295 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
1296 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
984263bc
MD
1297 /*
1298 * Recalculate the retransmit timer / rtt.
1299 *
d24ce1dc 1300 * Some machines (certain windows boxes)
984263bc
MD
1301 * send broken timestamp replies during the
1302 * SYN+ACK phase, ignore timestamps of 0.
1303 */
95b22adf 1304 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
984263bc 1305 tcp_xmit_timer(tp,
073ec6c4
SZ
1306 ticks - to.to_tsecr + 1,
1307 th->th_ack);
984263bc 1308 } else if (tp->t_rtttime &&
95b22adf 1309 SEQ_GT(th->th_ack, tp->t_rtseq)) {
984263bc 1310 tcp_xmit_timer(tp,
073ec6c4
SZ
1311 ticks - tp->t_rtttime,
1312 th->th_ack);
984263bc
MD
1313 }
1314 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1315 acked = th->th_ack - tp->snd_una;
1316 tcpstat.tcps_rcvackpack++;
1317 tcpstat.tcps_rcvackbyte += acked;
6d49aa6f 1318 sbdrop(&so->so_snd.sb, acked);
cfb3f3f4 1319 tp->snd_recover = th->th_ack - 1;
9845754e 1320 tp->snd_una = th->th_ack;
984263bc 1321 tp->t_dupacks = 0;
df9d7670
JH
1322 /*
1323 * Update window information.
1324 */
1325 if (tiwin != tp->snd_wnd &&
1326 acceptable_window_update(tp, th, tiwin)) {
1327 /* keep track of pure window updates */
1328 if (tp->snd_wl2 == th->th_ack &&
1329 tiwin > tp->snd_wnd)
1330 tcpstat.tcps_rcvwinupd++;
1331 tp->snd_wnd = tiwin;
1332 tp->snd_wl1 = th->th_seq;
1333 tp->snd_wl2 = th->th_ack;
1334 if (tp->snd_wnd > tp->max_sndwnd)
1335 tp->max_sndwnd = tp->snd_wnd;
1336 }
984263bc
MD
1337 m_freem(m);
1338 ND6_HINT(tp); /* some progress has been done */
984263bc
MD
1339 /*
1340 * If all outstanding data are acked, stop
1341 * retransmit timer, otherwise restart timer
1342 * using current (possibly backed-off) value.
1343 * If process is waiting for space,
1344 * wakeup/selwakeup/signal. If data
1345 * are ready to send, let tcp_output
1346 * decide between more output or persist.
1347 */
a48c5dd5
SZ
1348 if (tp->snd_una == tp->snd_max) {
1349 tcp_callout_stop(tp, tp->tt_rexmt);
1350 } else if (!tcp_callout_active(tp,
1351 tp->tt_persist)) {
1352 tcp_callout_reset(tp, tp->tt_rexmt,
1353 tp->t_rxtcur, tcp_timer_rexmt);
1354 }
984263bc 1355 sowwakeup(so);
6d49aa6f 1356 if (so->so_snd.ssb_cc > 0)
f23061d4 1357 tcp_output(tp);
002c1265 1358 return(IPPROTO_DONE);
984263bc 1359 }
df9d7670
JH
1360 } else if (tiwin == tp->snd_wnd &&
1361 th->th_ack == tp->snd_una &&
0f9e45de 1362 TAILQ_EMPTY(&tp->t_segq) &&
6d49aa6f 1363 tlen <= ssb_space(&so->so_rcv)) {
46e92930 1364 u_long newsize = 0; /* automatic sockbuf scaling */
984263bc 1365 /*
be23faf1 1366 * This is a pure, in-sequence data packet
984263bc
MD
1367 * with nothing on the reassembly queue and
1368 * we have enough buffer space to take it.
1369 */
1370 ++tcpstat.tcps_preddat;
1371 tp->rcv_nxt += tlen;
1372 tcpstat.tcps_rcvpack++;
1373 tcpstat.tcps_rcvbyte += tlen;
1374 ND6_HINT(tp); /* some progress has been done */
5b0b9fa5
PA
1375 /*
1376 * Automatic sizing of receive socket buffer. Often the send
1377 * buffer size is not optimally adjusted to the actual network
1378 * conditions at hand (delay bandwidth product). Setting the
1379 * buffer size too small limits throughput on links with high
1380 * bandwidth and high delay (eg. trans-continental/oceanic links).
1381 *
1382 * On the receive side the socket buffer memory is only rarely
1383 * used to any significant extent. This allows us to be much
1384 * more aggressive in scaling the receive socket buffer. For
1385 * the case that the buffer space is actually used to a large
1386 * extent and we run out of kernel memory we can simply drop
1387 * the new segments; TCP on the sender will just retransmit it
1388 * later. Setting the buffer size too big may only consume too
1389 * much kernel memory if the application doesn't read() from
1390 * the socket or packet loss or reordering makes use of the
1391 * reassembly queue.
1392 *
1393 * The criteria to step up the receive buffer one notch are:
1394 * 1. the number of bytes received during the time it takes
1395 * one timestamp to be reflected back to us (the RTT);
1396 * 2. received bytes per RTT is within seven eighth of the
1397 * current socket buffer size;
1398 * 3. receive buffer size has not hit maximal automatic size;
1399 *
1400 * This algorithm does one step per RTT at most and only if
1401 * we receive a bulk stream w/o packet losses or reorderings.
1402 * Shrinking the buffer during idle times is not necessary as
1403 * it doesn't consume any memory when idle.
1404 *
1405 * TODO: Only step up if the application is actually serving
1406 * the buffer to better manage the socket buffer resources.
1407 */
1408 if (tcp_do_autorcvbuf &&
1409 to.to_tsecr &&
1410 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1411 if (to.to_tsecr > tp->rfbuf_ts &&
1412 to.to_tsecr - tp->rfbuf_ts < hz) {
1413 if (tp->rfbuf_cnt >
1414 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1415 so->so_rcv.ssb_hiwat <
1416 tcp_autorcvbuf_max) {
1417 newsize =
46e92930
MD
1418 ulmin(so->so_rcv.ssb_hiwat +
1419 tcp_autorcvbuf_inc,
1420 tcp_autorcvbuf_max);
5b0b9fa5
PA
1421 }
1422 /* Start over with next RTT. */
1423 tp->rfbuf_ts = 0;
1424 tp->rfbuf_cnt = 0;
1425 } else
1426 tp->rfbuf_cnt += tlen; /* add up */
1427 }
984263bc
MD
1428 /*
1429 * Add data to socket buffer.
1430 */
1431 if (so->so_state & SS_CANTRCVMORE) {
1432 m_freem(m);
1433 } else {
5b0b9fa5 1434 /*
46e92930
MD
1435 * Set new socket buffer size, give up when
1436 * limit is reached.
1437 *
1438 * Adjusting the size can mess up ACK
1439 * sequencing when pure window updates are
1440 * being avoided (which is the default),
1441 * so force an ack.
5b0b9fa5 1442 */
6cef7136 1443 lwkt_gettoken(&so->so_rcv.ssb_token);
46e92930
MD
1444 if (newsize) {
1445 tp->t_flags |= TF_RXRESIZED;
5b0b9fa5 1446 if (!ssb_reserve(&so->so_rcv, newsize,
46e92930 1447 so, NULL)) {
14343ad3 1448 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
46e92930
MD
1449 }
1450 if (newsize >=
1451 (TCP_MAXWIN << tp->rcv_scale)) {
14343ad3 1452 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
46e92930
MD
1453 }
1454 }
09cf73e0 1455 m_adj(m, drop_hdrlen); /* delayed header drop */
6d49aa6f 1456 ssb_appendstream(&so->so_rcv, m);
6cef7136 1457 lwkt_reltoken(&so->so_rcv.ssb_token);
984263bc
MD
1458 }
1459 sorwakeup(so);
2b1ce38a
MD
1460 /*
1461 * This code is responsible for most of the ACKs
1462 * the TCP stack sends back after receiving a data
1463 * packet. Note that the DELAY_ACK check fails if
1464 * the delack timer is already running, which results
1465 * in an ack being sent every other packet (which is
1466 * what we want).
72b37eeb
MD
1467 *
1468 * We then further aggregate acks by not actually
1469 * sending one until the protocol thread has completed
1470 * processing the current backlog of packets. This
1471 * does not delay the ack any further, but allows us
1472 * to take advantage of the packet aggregation that
1473 * high speed NICs do (usually blocks of 8-10 packets)
1474 * to send a single ack rather then four or five acks,
1475 * greatly reducing the ack rate, the return channel
1476 * bandwidth, and the protocol overhead on both ends.
1477 *
1478 * Since this also has the effect of slowing down
1479 * the exponential slow-start ramp-up, systems with
1480 * very large bandwidth-delay products might want
1481 * to turn the feature off.
2b1ce38a 1482 */
984263bc 1483 if (DELAY_ACK(tp)) {
a48c5dd5
SZ
1484 tcp_callout_reset(tp, tp->tt_delack,
1485 tcp_delacktime, tcp_timer_delack);
72b37eeb 1486 } else if (tcp_aggregate_acks) {
984263bc 1487 tp->t_flags |= TF_ACKNOW;
61896e3c 1488 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
2b1ce38a
MD
1489 tp->t_flags |= TF_ONOUTPUTQ;
1490 tp->tt_cpu = mycpu->gd_cpuid;
1491 TAILQ_INSERT_TAIL(
1492 &tcpcbackq[tp->tt_cpu],
1493 tp, t_outputq);
1494 }
72b37eeb
MD
1495 } else {
1496 tp->t_flags |= TF_ACKNOW;
1497 tcp_output(tp);
984263bc 1498 }
002c1265 1499 return(IPPROTO_DONE);
984263bc
MD
1500 }
1501 }
1502
1503 /*
1504 * Calculate amount of space in receive window,
1505 * and then do TCP input processing.
1506 * Receive window is amount of space in rcv queue,
1507 * but not less than advertised window.
1508 */
6d49aa6f 1509 recvwin = ssb_space(&so->so_rcv);
61896e3c
JH
1510 if (recvwin < 0)
1511 recvwin = 0;
1512 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
984263bc 1513
5b0b9fa5
PA
1514 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1515 tp->rfbuf_ts = 0;
1516 tp->rfbuf_cnt = 0;
1517
984263bc 1518 switch (tp->t_state) {
984263bc
MD
1519 /*
1520 * If the state is SYN_RECEIVED:
1521 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1522 */
1523 case TCPS_SYN_RECEIVED:
1524 if ((thflags & TH_ACK) &&
1525 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1526 SEQ_GT(th->th_ack, tp->snd_max))) {
1527 rstreason = BANDLIM_RST_OPENPORT;
1528 goto dropwithreset;
1529 }
1530 break;
1531
1532 /*
1533 * If the state is SYN_SENT:
1534 * if seg contains an ACK, but not for our SYN, drop the input.
1535 * if seg contains a RST, then drop the connection.
1536 * if seg does not contain SYN, then drop it.
1537 * Otherwise this is an acceptable SYN segment
1538 * initialize tp->rcv_nxt and tp->irs
1539 * if seg contains ack then advance tp->snd_una
1540 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1541 * arrange for segment to be acked (eventually)
1542 * continue processing rest of data/controls, beginning with URG
1543 */
1544 case TCPS_SYN_SENT:
984263bc
MD
1545 if ((thflags & TH_ACK) &&
1546 (SEQ_LEQ(th->th_ack, tp->iss) ||
1547 SEQ_GT(th->th_ack, tp->snd_max))) {
27b8aee3
AE
1548 rstreason = BANDLIM_UNLIMITED;
1549 goto dropwithreset;
984263bc
MD
1550 }
1551 if (thflags & TH_RST) {
1552 if (thflags & TH_ACK)
1553 tp = tcp_drop(tp, ECONNREFUSED);
1554 goto drop;
1555 }
61896e3c 1556 if (!(thflags & TH_SYN))
984263bc 1557 goto drop;
984263bc
MD
1558
1559 tp->irs = th->th_seq;
1560 tcp_rcvseqinit(tp);
1561 if (thflags & TH_ACK) {
27b8aee3 1562 /* Our SYN was acked. */
984263bc
MD
1563 tcpstat.tcps_connects++;
1564 soisconnected(so);
1565 /* Do window scaling on this connection? */
61896e3c 1566 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
df1d2774 1567 (TF_RCVD_SCALE | TF_REQ_SCALE))
984263bc 1568 tp->rcv_scale = tp->request_r_scale;
984263bc
MD
1569 tp->rcv_adv += tp->rcv_wnd;
1570 tp->snd_una++; /* SYN is acked */
a48c5dd5 1571 tcp_callout_stop(tp, tp->tt_rexmt);
984263bc
MD
1572 /*
1573 * If there's data, delay ACK; if there's also a FIN
1574 * ACKNOW will be turned on later.
1575 */
a48c5dd5
SZ
1576 if (DELAY_ACK(tp) && tlen != 0) {
1577 tcp_callout_reset(tp, tp->tt_delack,
1578 tcp_delacktime, tcp_timer_delack);
1579 } else {
984263bc 1580 tp->t_flags |= TF_ACKNOW;
a48c5dd5 1581 }
984263bc
MD
1582 /*
1583 * Received <SYN,ACK> in SYN_SENT[*] state.
1584 * Transitions:
1585 * SYN_SENT --> ESTABLISHED
1586 * SYN_SENT* --> FIN_WAIT_1
1587 */
1588 tp->t_starttime = ticks;
1589 if (tp->t_flags & TF_NEEDFIN) {
1590 tp->t_state = TCPS_FIN_WAIT_1;
1591 tp->t_flags &= ~TF_NEEDFIN;
1592 thflags &= ~TH_SYN;
1593 } else {
8651f7f8 1594 tcp_established(tp);
984263bc
MD
1595 }
1596 } else {
1597 /*
95b22adf 1598 * Received initial SYN in SYN-SENT[*] state =>
27b8aee3
AE
1599 * simultaneous open.
1600 * Do 3-way handshake:
61896e3c
JH
1601 * SYN-SENT -> SYN-RECEIVED
1602 * SYN-SENT* -> SYN-RECEIVED*
95b22adf 1603 */
984263bc 1604 tp->t_flags |= TF_ACKNOW;
a48c5dd5 1605 tcp_callout_stop(tp, tp->tt_rexmt);
27b8aee3 1606 tp->t_state = TCPS_SYN_RECEIVED;
984263bc
MD
1607 }
1608
984263bc
MD
1609 /*
1610 * Advance th->th_seq to correspond to first data byte.
1611 * If data, trim to stay within window,
1612 * dropping FIN if necessary.
1613 */
1614 th->th_seq++;
1615 if (tlen > tp->rcv_wnd) {
1616 todrop = tlen - tp->rcv_wnd;
1617 m_adj(m, -todrop);
1618 tlen = tp->rcv_wnd;
1619 thflags &= ~TH_FIN;
1620 tcpstat.tcps_rcvpackafterwin++;
1621 tcpstat.tcps_rcvbyteafterwin += todrop;
1622 }
1623 tp->snd_wl1 = th->th_seq - 1;
1624 tp->rcv_up = th->th_seq;
1625 /*
1626 * Client side of transaction: already sent SYN and data.
1627 * If the remote host used T/TCP to validate the SYN,
1628 * our data will be ACK'd; if so, enter normal data segment
1629 * processing in the middle of step 5, ack processing.
1630 * Otherwise, goto step 6.
1631 */
95b22adf 1632 if (thflags & TH_ACK)
984263bc
MD
1633 goto process_ACK;
1634
1635 goto step6;
1636
1637 /*
1638 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
27b8aee3 1639 * do normal processing (we no longer bother with T/TCP).
984263bc
MD
1640 */
1641 case TCPS_LAST_ACK:
1642 case TCPS_CLOSING:
1643 case TCPS_TIME_WAIT:
95b22adf 1644 break; /* continue normal processing */
984263bc
MD
1645 }
1646
1647 /*
1648 * States other than LISTEN or SYN_SENT.
1649 * First check the RST flag and sequence number since reset segments
1650 * are exempt from the timestamp and connection count tests. This
1651 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1652 * below which allowed reset segments in half the sequence space
1653 * to fall though and be processed (which gives forged reset
1654 * segments with a random sequence number a 50 percent chance of
1655 * killing a connection).
1656 * Then check timestamp, if present.
1657 * Then check the connection count, if present.
1658 * Then check that at least some bytes of segment are within
1659 * receive window. If segment begins before rcv_nxt,
1660 * drop leading data (and SYN); if nothing left, just ack.
1661 *
1662 *
1663 * If the RST bit is set, check the sequence number to see
1664 * if this is a valid reset segment.
1665 * RFC 793 page 37:
1666 * In all states except SYN-SENT, all reset (RST) segments
1667 * are validated by checking their SEQ-fields. A reset is
1668 * valid if its sequence number is in the window.
1669 * Note: this does not take into account delayed ACKs, so
1670 * we should test against last_ack_sent instead of rcv_nxt.
1671 * The sequence number in the reset segment is normally an
91489f6b 1672 * echo of our outgoing acknowledgement numbers, but some hosts
984263bc
MD
1673 * send a reset with the sequence number at the rightmost edge
1674 * of our receive window, and we have to handle this case.
1675 * If we have multiple segments in flight, the intial reset
1676 * segment sequence numbers will be to the left of last_ack_sent,
1677 * but they will eventually catch up.
1678 * In any case, it never made sense to trim reset segments to
1679 * fit the receive window since RFC 1122 says:
1680 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1681 *
1682 * A TCP SHOULD allow a received RST segment to include data.
1683 *
1684 * DISCUSSION
61896e3c
JH
1685 * It has been suggested that a RST segment could contain
1686 * ASCII text that encoded and explained the cause of the
91489f6b 1687 * RST. No standard has yet been established for such
61896e3c 1688 * data.
984263bc
MD
1689 *
1690 * If the reset segment passes the sequence number test examine
1691 * the state:
1692 * SYN_RECEIVED STATE:
1693 * If passive open, return to LISTEN state.
1694 * If active open, inform user that connection was refused.
1695 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1696 * Inform user that connection was reset, and close tcb.
1697 * CLOSING, LAST_ACK STATES:
1698 * Close the tcb.
1699 * TIME_WAIT STATE:
1700 * Drop the segment - see Stevens, vol. 2, p. 964 and
61896e3c 1701 * RFC 1337.
984263bc
MD
1702 */
1703 if (thflags & TH_RST) {
1704 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
d4dbb5be 1705 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
984263bc
MD
1706 switch (tp->t_state) {
1707
1708 case TCPS_SYN_RECEIVED:
1709 so->so_error = ECONNREFUSED;
1710 goto close;
1711
1712 case TCPS_ESTABLISHED:
1713 case TCPS_FIN_WAIT_1:
1714 case TCPS_FIN_WAIT_2:
1715 case TCPS_CLOSE_WAIT:
1716 so->so_error = ECONNRESET;
1717 close:
1718 tp->t_state = TCPS_CLOSED;
1719 tcpstat.tcps_drops++;
1720 tp = tcp_close(tp);
1721 break;
1722
1723 case TCPS_CLOSING:
1724 case TCPS_LAST_ACK:
1725 tp = tcp_close(tp);
1726 break;
1727
1728 case TCPS_TIME_WAIT:
1729 break;
1730 }
1731 }
1732 goto drop;
1733 }
1734
1735 /*
1736 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1737 * and it's less than ts_recent, drop it.
1738 */
61896e3c 1739 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
984263bc 1740 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
984263bc
MD
1741 /* Check to see if ts_recent is over 24 days old. */
1742 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1743 /*
1744 * Invalidate ts_recent. If this segment updates
1745 * ts_recent, the age will be reset later and ts_recent
1746 * will get a valid value. If it does not, setting
1747 * ts_recent to zero will at least satisfy the
1748 * requirement that zero be placed in the timestamp
1749 * echo reply when ts_recent isn't valid. The
1750 * age isn't reset until we get a valid ts_recent
1751 * because we don't want out-of-order segments to be
1752 * dropped when ts_recent is old.
1753 */
1754 tp->ts_recent = 0;
9de1f696
SZ
1755 } else if (tcp_paws_tolerance && tlen != 0 &&
1756 tp->t_state == TCPS_ESTABLISHED &&
1757 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK&&
1758 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1759 th->th_ack == tp->snd_una &&
1760 tiwin == tp->snd_wnd &&
1761 TSTMP_GEQ(to.to_tsval + tcp_paws_tolerance, tp->ts_recent)&&
1762 (th->th_seq == tp->rcv_nxt ||
1763 (SEQ_GT(th->th_seq, tp->rcv_nxt) &&
1764 tcp_paws_canreasslast(tp, th, tlen)))) {
1765 /*
1766 * This tends to prevent valid new segments from being
1767 * dropped by the reordered segments sent by the fast
1768 * retransmission algorithm on the sending side, i.e.
1769 * the fast retransmitted segment w/ larger timestamp
1770 * arrives earlier than the previously sent new segments
1771 * w/ smaller timestamp.
1772 *
1773 * If following conditions are met, the segment is
1774 * accepted:
1775 * - The segment contains data
1776 * - The connection is established
1777 * - The header does not contain important flags
1778 * - SYN or FIN is not needed
1779 * - It does not acknowledge new data
1780 * - Receive window is not changed
1781 * - The timestamp is within "acceptable" range
1782 * - The new segment is what we are expecting or
1783 * the new segment could be merged w/ the last
1784 * pending segment on the reassemble queue
1785 */
1786 tcpstat.tcps_pawsaccept++;
1787 tcpstat.tcps_pawsdrop++;
984263bc
MD
1788 } else {
1789 tcpstat.tcps_rcvduppack++;
1790 tcpstat.tcps_rcvdupbyte += tlen;
1791 tcpstat.tcps_pawsdrop++;
1792 if (tlen)
1793 goto dropafterack;
1794 goto drop;
1795 }
1796 }
1797
984263bc
MD
1798 /*
1799 * In the SYN-RECEIVED state, validate that the packet belongs to
1800 * this connection before trimming the data to fit the receive
1801 * window. Check the sequence number versus IRS since we know
1802 * the sequence numbers haven't wrapped. This is a partial fix
1803 * for the "LAND" DoS attack.
1804 */
1805 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1806 rstreason = BANDLIM_RST_OPENPORT;
1807 goto dropwithreset;
1808 }
1809
1810 todrop = tp->rcv_nxt - th->th_seq;
1811 if (todrop > 0) {
91489f6b
JH
1812 if (TCP_DO_SACK(tp)) {
1813 /* Report duplicate segment at head of packet. */
1814 tp->reportblk.rblk_start = th->th_seq;
3a5d999b
SZ
1815 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1816 th->th_seq + tlen, thflags);
91489f6b
JH
1817 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1818 tp->reportblk.rblk_end = tp->rcv_nxt;
c7e6499a
SZ
1819 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT);
1820 tp->t_flags |= TF_ACKNOW;
91489f6b 1821 }
984263bc
MD
1822 if (thflags & TH_SYN) {
1823 thflags &= ~TH_SYN;
1824 th->th_seq++;
1825 if (th->th_urp > 1)
1826 th->th_urp--;
1827 else
1828 thflags &= ~TH_URG;
1829 todrop--;
1830 }
1831 /*
1832 * Following if statement from Stevens, vol. 2, p. 960.
1833 */
61896e3c
JH
1834 if (todrop > tlen ||
1835 (todrop == tlen && !(thflags & TH_FIN))) {
984263bc
MD
1836 /*
1837 * Any valid FIN must be to the left of the window.
1838 * At this point the FIN must be a duplicate or out
1839 * of sequence; drop it.
1840 */
1841 thflags &= ~TH_FIN;
1842
1843 /*
1844 * Send an ACK to resynchronize and drop any data.
1845 * But keep on processing for RST or ACK.
1846 */
1847 tp->t_flags |= TF_ACKNOW;
1848 todrop = tlen;
1849 tcpstat.tcps_rcvduppack++;
1850 tcpstat.tcps_rcvdupbyte += todrop;
1851 } else {
1852 tcpstat.tcps_rcvpartduppack++;
1853 tcpstat.tcps_rcvpartdupbyte += todrop;
1854 }
1855 drop_hdrlen += todrop; /* drop from the top afterwards */
1856 th->th_seq += todrop;
1857 tlen -= todrop;
1858 if (th->th_urp > todrop)
1859 th->th_urp -= todrop;
1860 else {
1861 thflags &= ~TH_URG;
1862 th->th_urp = 0;
1863 }
1864 }
1865
1866 /*
1867 * If new data are received on a connection after the
1868 * user processes are gone, then RST the other end.
1869 */
1870 if ((so->so_state & SS_NOFDREF) &&
1871 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1872 tp = tcp_close(tp);
1873 tcpstat.tcps_rcvafterclose++;
1874 rstreason = BANDLIM_UNLIMITED;
1875 goto dropwithreset;
1876 }
1877
1878 /*
1879 * If segment ends after window, drop trailing data
1880 * (and PUSH and FIN); if nothing left, just ACK.
1881 */
61896e3c 1882 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
984263bc
MD
1883 if (todrop > 0) {
1884 tcpstat.tcps_rcvpackafterwin++;
1885 if (todrop >= tlen) {
1886 tcpstat.tcps_rcvbyteafterwin += tlen;
1887 /*
1888 * If a new connection request is received
1889 * while in TIME_WAIT, drop the old connection
1890 * and start over if the sequence numbers
1891 * are above the previous ones.
1892 */
1893 if (thflags & TH_SYN &&
1894 tp->t_state == TCPS_TIME_WAIT &&
1895 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1896 tp = tcp_close(tp);
1897 goto findpcb;
1898 }
1899 /*
1900 * If window is closed can only take segments at
1901 * window edge, and have to drop data and PUSH from
1902 * incoming segments. Continue processing, but
1903 * remember to ack. Otherwise, drop segment
1904 * and ack.
1905 */
1906 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1907 tp->t_flags |= TF_ACKNOW;
1908 tcpstat.tcps_rcvwinprobe++;
1909 } else
1910 goto dropafterack;
1911 } else
1912 tcpstat.tcps_rcvbyteafterwin += todrop;
1913 m_adj(m, -todrop);
1914 tlen -= todrop;
61896e3c 1915 thflags &= ~(TH_PUSH | TH_FIN);
984263bc
MD
1916 }
1917
1918 /*
1919 * If last ACK falls within this segment's sequence numbers,
1920 * record its timestamp.
ad0af98b
ND
1921 * NOTE:
1922 * 1) That the test incorporates suggestions from the latest
1923 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1924 * 2) That updating only on newer timestamps interferes with
1925 * our earlier PAWS tests, so this check should be solely
1926 * predicated on the sequence space of this segment.
1927 * 3) That we modify the segment boundary check to be
1928 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1929 * instead of RFC1323's
1930 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1931 * This modified check allows us to overcome RFC1323's
1932 * limitations as described in Stevens TCP/IP Illustrated
1933 * Vol. 2 p.869. In such cases, we can still calculate the
1934 * RTT correctly when RCV.NXT == Last.ACK.Sent.
984263bc 1935 */
ad0af98b
ND
1936 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1937 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1938 + ((thflags & TH_SYN) != 0)
1939 + ((thflags & TH_FIN) != 0)))) {
984263bc
MD
1940 tp->ts_recent_age = ticks;
1941 tp->ts_recent = to.to_tsval;
1942 }
1943
1944 /*
1945 * If a SYN is in the window, then this is an
1946 * error and we send an RST and drop the connection.
1947 */
1948 if (thflags & TH_SYN) {
1949 tp = tcp_drop(tp, ECONNRESET);
1950 rstreason = BANDLIM_UNLIMITED;
1951 goto dropwithreset;
1952 }
1953
1954 /*
1955 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1956 * flag is on (half-synchronized state), then queue data for
1957 * later processing; else drop segment and return.
1958 */
61896e3c 1959 if (!(thflags & TH_ACK)) {
984263bc
MD
1960 if (tp->t_state == TCPS_SYN_RECEIVED ||
1961 (tp->t_flags & TF_NEEDSYN))
1962 goto step6;
1963 else
1964 goto drop;
1965 }
1966
1967 /*
1968 * Ack processing.
1969 */
1970 switch (tp->t_state) {
984263bc 1971 /*
91489f6b 1972 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
984263bc
MD
1973 * ESTABLISHED state and continue processing.
1974 * The ACK was checked above.
1975 */
1976 case TCPS_SYN_RECEIVED:
1977
1978 tcpstat.tcps_connects++;
1979 soisconnected(so);
1980 /* Do window scaling? */
61896e3c 1981 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
df1d2774 1982 (TF_RCVD_SCALE | TF_REQ_SCALE))
984263bc 1983 tp->rcv_scale = tp->request_r_scale;
984263bc
MD
1984 /*
1985 * Make transitions:
1986 * SYN-RECEIVED -> ESTABLISHED
1987 * SYN-RECEIVED* -> FIN-WAIT-1
1988 */
1989 tp->t_starttime = ticks;
1990 if (tp->t_flags & TF_NEEDFIN) {
1991 tp->t_state = TCPS_FIN_WAIT_1;
1992 tp->t_flags &= ~TF_NEEDFIN;
1993 } else {
8651f7f8 1994 tcp_established(tp);
984263bc
MD
1995 }
1996 /*
1997 * If segment contains data or ACK, will call tcp_reass()
1998 * later; if not, do so now to pass queued data to user.
1999 */
61896e3c 2000 if (tlen == 0 && !(thflags & TH_FIN))
f23061d4 2001 tcp_reass(tp, NULL, NULL, NULL);
984263bc
MD
2002 /* fall into ... */
2003
2004 /*
2005 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2006 * ACKs. If the ack is in the range
2007 * tp->snd_una < th->th_ack <= tp->snd_max
2008 * then advance tp->snd_una to th->th_ack and drop
2009 * data from the retransmission queue. If this ACK reflects
2010 * more up to date window information we update our window information.
2011 */
2012 case TCPS_ESTABLISHED:
2013 case TCPS_FIN_WAIT_1:
2014 case TCPS_FIN_WAIT_2:
2015 case TCPS_CLOSE_WAIT:
2016 case TCPS_CLOSING:
2017 case TCPS_LAST_ACK:
2018 case TCPS_TIME_WAIT:
2019
2020 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
91489f6b
JH
2021 if (TCP_DO_SACK(tp))
2022 tcp_sack_update_scoreboard(tp, &to);
a48c5dd5 2023 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
91489f6b 2024 th->th_ack != tp->snd_una) {
52f9ffcf
SZ
2025 if (tlen == 0 && tiwin == tp->snd_wnd)
2026 tcpstat.tcps_rcvdupack++;
91489f6b
JH
2027 tp->t_dupacks = 0;
2028 break;
2029 }
ffe35e17
SZ
2030 if (tlen != 0 || tiwin != tp->snd_wnd) {
2031 if (!tcp_do_rfc3517bis ||
2032 !TCP_DO_SACK(tp) ||
2033 (to.to_flags &
2034 (TOF_SACK | TOF_SACK_REDUNDANT))
2035 != TOF_SACK) {
2036 tp->t_dupacks = 0;
91489f6b 2037 } else {
27f4bf33
SZ
2038 delayed_dupack = TRUE;
2039 th_dupack = th->th_ack;
91489f6b 2040 }
95b22adf 2041 break;
95b22adf 2042 }
5c99b248 2043 if (tcp_recv_dupack(tp, th->th_ack, &to))
ffe35e17 2044 goto drop;
27f4bf33
SZ
2045 else
2046 break;
984263bc
MD
2047 }
2048
2049 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
984263bc
MD
2050 tp->t_dupacks = 0;
2051 if (SEQ_GT(th->th_ack, tp->snd_max)) {
5a274421
JH
2052 /*
2053 * Detected optimistic ACK attack.
2054 * Force slow-start to de-synchronize attack.
2055 */
2056 tp->snd_cwnd = tp->t_maxseg;
8acdb67c 2057 tp->snd_wacked = 0;
5a274421 2058
984263bc
MD
2059 tcpstat.tcps_rcvacktoomuch++;
2060 goto dropafterack;
2061 }
2062 /*
2063 * If we reach this point, ACK is not a duplicate,
2064 * i.e., it ACKs something we sent.
2065 */
2066 if (tp->t_flags & TF_NEEDSYN) {
2067 /*
2068 * T/TCP: Connection was half-synchronized, and our
2069 * SYN has been ACK'd (so connection is now fully
2070 * synchronized). Go to non-starred state,
2071 * increment snd_una for ACK of SYN, and check if
2072 * we can do window scaling.
2073 */
2074 tp->t_flags &= ~TF_NEEDSYN;
2075 tp->snd_una++;
2076 /* Do window scaling? */
61896e3c 2077 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
df1d2774 2078 (TF_RCVD_SCALE | TF_REQ_SCALE))
984263bc 2079 tp->rcv_scale = tp->request_r_scale;
984263bc
MD
2080 }
2081
2082process_ACK:
2083 acked = th->th_ack - tp->snd_una;
2084 tcpstat.tcps_rcvackpack++;
2085 tcpstat.tcps_rcvackbyte += acked;
2086
b5572302 2087 if (tcp_do_eifel_detect && acked > 0 &&
95b22adf 2088 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
1c8b7a61 2089 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
bfdb979e
JH
2090 /* Eifel detection applicable. */
2091 if (to.to_tsecr < tp->t_rexmtTS) {
bfdb979e 2092 ++tcpstat.tcps_eifeldetected;
8819433a 2093 tcp_revert_congestion_state(tp);
928c3291 2094 if (tp->t_rxtshift != 1 ||
8819433a
JH
2095 ticks >= tp->t_badrxtwin)
2096 ++tcpstat.tcps_rttcantdetect;
bfdb979e
JH
2097 }
2098 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
b5572302
JH
2099 /*
2100 * If we just performed our first retransmit,
2101 * and the ACK arrives within our recovery window,
2102 * then it was a mistake to do the retransmit
91489f6b 2103 * in the first place. Recover our original cwnd
b5572302
JH
2104 * and ssthresh, and proceed to transmit where we
2105 * left off.
2106 */
bfdb979e
JH
2107 tcp_revert_congestion_state(tp);
2108 ++tcpstat.tcps_rttdetected;
984263bc
MD
2109 }
2110
2111 /*
2112 * If we have a timestamp reply, update smoothed
2113 * round trip time. If no timestamp is present but
2114 * transmit timer is running and timed sequence
2115 * number was acked, update smoothed round trip time.
2116 * Since we now have an rtt measurement, cancel the
2117 * timer backoff (cf., Phil Karn's retransmit alg.).
2118 * Recompute the initial retransmit timer.
2119 *
2120 * Some machines (certain windows boxes) send broken
d24ce1dc 2121 * timestamp replies during the SYN+ACK phase, ignore
984263bc
MD
2122 * timestamps of 0.
2123 */
95b22adf 2124 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
073ec6c4 2125 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack);
95b22adf 2126 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
073ec6c4 2127 tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack);
984263bc
MD
2128 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2129
984263bc
MD
2130 /*
2131 * If no data (only SYN) was ACK'd,
2132 * skip rest of ACK processing.
2133 */
2134 if (acked == 0)
2135 goto step6;
2136
efd4b327 2137 /* Stop looking for an acceptable ACK since one was received. */
1c8b7a61
SZ
2138 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
2139 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
efd4b327 2140
6d49aa6f
MD
2141 if (acked > so->so_snd.ssb_cc) {
2142 tp->snd_wnd -= so->so_snd.ssb_cc;
2143 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
61896e3c 2144 ourfinisacked = TRUE;
984263bc 2145 } else {
6d49aa6f 2146 sbdrop(&so->so_snd.sb, acked);
984263bc 2147 tp->snd_wnd -= acked;
61896e3c 2148 ourfinisacked = FALSE;
984263bc
MD
2149 }
2150 sowwakeup(so);
91489f6b
JH
2151
2152 /*
2153 * Update window information.
91489f6b 2154 */
e126661b 2155 if (acceptable_window_update(tp, th, tiwin)) {
91489f6b
JH
2156 /* keep track of pure window updates */
2157 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2158 tiwin > tp->snd_wnd)
2159 tcpstat.tcps_rcvwinupd++;
2160 tp->snd_wnd = tiwin;
2161 tp->snd_wl1 = th->th_seq;
2162 tp->snd_wl2 = th->th_ack;
2163 if (tp->snd_wnd > tp->max_sndwnd)
2164 tp->max_sndwnd = tp->snd_wnd;
2165 needoutput = TRUE;
2166 }
2167
2168 tp->snd_una = th->th_ack;
2169 if (TCP_DO_SACK(tp))
2170 tcp_sack_update_scoreboard(tp, &to);
95b22adf 2171 if (IN_FASTRECOVERY(tp)) {
91489f6b 2172 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
95b22adf 2173 EXIT_FASTRECOVERY(tp);
91489f6b
JH
2174 needoutput = TRUE;
2175 /*
2176 * If the congestion window was inflated
2177 * to account for the other side's
2178 * cached packets, retract it.
8acdb67c
JH
2179 */
2180 if (!TCP_DO_SACK(tp))
2181 tp->snd_cwnd = tp->snd_ssthresh;
2182
2183 /*
91489f6b
JH
2184 * Window inflation should have left us
2185 * with approximately snd_ssthresh outstanding
2186 * data. But, in case we would be inclined
2187 * to send a burst, better do it using
2188 * slow start.
2189 */
91489f6b
JH
2190 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2191 tp->snd_max + 2 * tp->t_maxseg))
2192 tp->snd_cwnd =
2193 (tp->snd_max - tp->snd_una) +
2194 2 * tp->t_maxseg;
8acdb67c
JH
2195
2196 tp->snd_wacked = 0;
91489f6b
JH
2197 } else {
2198 if (TCP_DO_SACK(tp)) {
2199 tp->snd_max_rexmt = tp->snd_max;
ccb518ea
SZ
2200 tcp_sack_rexmt(tp,
2201 tp->snd_una == tp->rexmt_high);
91489f6b
JH
2202 } else {
2203 tcp_newreno_partial_ack(tp, th, acked);
2204 }
2205 needoutput = FALSE;
2206 }
95b22adf 2207 } else {
91489f6b 2208 /*
8acdb67c
JH
2209 * Open the congestion window. When in slow-start,
2210 * open exponentially: maxseg per packet. Otherwise,
2211 * open linearly: maxseg per window.
91489f6b 2212 */
8acdb67c
JH
2213 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2214 u_int abc_sslimit =
2215 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2216 tp->t_maxseg : 2 * tp->t_maxseg);
2217
2218 /* slow-start */
2219 tp->snd_cwnd += tcp_do_abc ?
2220 min(acked, abc_sslimit) : tp->t_maxseg;
2221 } else {
2222 /* linear increase */
2223 tp->snd_wacked += tcp_do_abc ? acked :
2224 tp->t_maxseg;
2225 if (tp->snd_wacked >= tp->snd_cwnd) {
2226 tp->snd_wacked -= tp->snd_cwnd;
2227 tp->snd_cwnd += tp->t_maxseg;
2228 }
2229 }
2230 tp->snd_cwnd = min(tp->snd_cwnd,
2231 TCP_MAXWIN << tp->snd_scale);
95b22adf 2232 tp->snd_recover = th->th_ack - 1;
cfb3f3f4 2233 }
984263bc
MD
2234 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2235 tp->snd_nxt = tp->snd_una;
2236
91489f6b
JH
2237 /*
2238 * If all outstanding data is acked, stop retransmit
2239 * timer and remember to restart (more output or persist).
2240 * If there is more data to be acked, restart retransmit
2241 * timer, using current (possibly backed-off) value.
2242 */
2243 if (th->th_ack == tp->snd_max) {
a48c5dd5 2244 tcp_callout_stop(tp, tp->tt_rexmt);
91489f6b 2245 needoutput = TRUE;
a48c5dd5
SZ
2246 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2247 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2248 tcp_timer_rexmt);
2249 }
91489f6b 2250
984263bc 2251 switch (tp->t_state) {
984263bc
MD
2252 /*
2253 * In FIN_WAIT_1 STATE in addition to the processing
2254 * for the ESTABLISHED state if our FIN is now acknowledged
2255 * then enter FIN_WAIT_2.
2256 */
2257 case TCPS_FIN_WAIT_1:
2258 if (ourfinisacked) {
2259 /*
2260 * If we can't receive any more
2261 * data, then closing user can proceed.
2262 * Starting the timer is contrary to the
2263 * specification, but if we don't get a FIN
2264 * we'll hang forever.
2265 */
2266 if (so->so_state & SS_CANTRCVMORE) {
2267 soisdisconnected(so);
a48c5dd5 2268 tcp_callout_reset(tp, tp->tt_2msl,
5d61ded3 2269 tp->t_maxidle, tcp_timer_2msl);
984263bc
MD
2270 }
2271 tp->t_state = TCPS_FIN_WAIT_2;
2272 }
2273 break;
2274
95b22adf 2275 /*
984263bc
MD
2276 * In CLOSING STATE in addition to the processing for
2277 * the ESTABLISHED state if the ACK acknowledges our FIN
2278 * then enter the TIME-WAIT state, otherwise ignore
2279 * the segment.
2280 */
2281 case TCPS_CLOSING:
2282 if (ourfinisacked) {
2283 tp->t_state = TCPS_TIME_WAIT;
2284 tcp_canceltimers(tp);
27b8aee3 2285 tcp_callout_reset(tp, tp->tt_2msl,
01d3427a
SZ
2286 2 * tcp_rmx_msl(tp),
2287 tcp_timer_2msl);
984263bc
MD
2288 soisdisconnected(so);
2289 }
2290 break;
2291
2292 /*
2293 * In LAST_ACK, we may still be waiting for data to drain
2294 * and/or to be acked, as well as for the ack of our FIN.
2295 * If our FIN is now acknowledged, delete the TCB,
2296 * enter the closed state and return.
2297 */
2298 case TCPS_LAST_ACK:
2299 if (ourfinisacked) {
2300 tp = tcp_close(tp);
2301 goto drop;
2302 }
2303 break;
2304
2305 /*
2306 * In TIME_WAIT state the only thing that should arrive
2307 * is a retransmission of the remote FIN. Acknowledge
2308 * it and restart the finack timer.
2309 */
2310 case TCPS_TIME_WAIT:
01d3427a 2311 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2312 tcp_timer_2msl);
984263bc
MD
2313 goto dropafterack;
2314 }
2315 }
2316
2317step6:
2318 /*
2319 * Update window information.
2320 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2321 */
2322 if ((thflags & TH_ACK) &&
df9d7670 2323 acceptable_window_update(tp, th, tiwin)) {
984263bc 2324 /* keep track of pure window updates */
b5572302
JH
2325 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2326 tiwin > tp->snd_wnd)
984263bc
MD
2327 tcpstat.tcps_rcvwinupd++;
2328 tp->snd_wnd = tiwin;
2329 tp->snd_wl1 = th->th_seq;
2330 tp->snd_wl2 = th->th_ack;
2331 if (tp->snd_wnd > tp->max_sndwnd)
2332 tp->max_sndwnd = tp->snd_wnd;
61896e3c 2333 needoutput = TRUE;
984263bc
MD
2334 }
2335
2336 /*
2337 * Process segments with URG.
2338 */
2339 if ((thflags & TH_URG) && th->th_urp &&
95b22adf 2340 !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2341 /*
2342 * This is a kludge, but if we receive and accept
2343 * random urgent pointers, we'll crash in
2344 * soreceive. It's hard to imagine someone
2345 * actually wanting to send this much urgent data.
2346 */
6d49aa6f 2347 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
984263bc
MD
2348 th->th_urp = 0; /* XXX */
2349 thflags &= ~TH_URG; /* XXX */
2350 goto dodata; /* XXX */
2351 }
2352 /*
2353 * If this segment advances the known urgent pointer,
2354 * then mark the data stream. This should not happen
2355 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2356 * a FIN has been received from the remote side.
2357 * In these states we ignore the URG.
2358 *
2359 * According to RFC961 (Assigned Protocols),
2360 * the urgent pointer points to the last octet
2361 * of urgent data. We continue, however,
2362 * to consider it to indicate the first octet
2363 * of data past the urgent section as the original
2364 * spec states (in one of two places).
2365 */
61896e3c 2366 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
984263bc 2367 tp->rcv_up = th->th_seq + th->th_urp;
6d49aa6f 2368 so->so_oobmark = so->so_rcv.ssb_cc +
984263bc
MD
2369 (tp->rcv_up - tp->rcv_nxt) - 1;
2370 if (so->so_oobmark == 0)
6cef7136 2371 sosetstate(so, SS_RCVATMARK);
984263bc
MD
2372 sohasoutofband(so);
2373 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2374 }
2375 /*
2376 * Remove out of band data so doesn't get presented to user.
2377 * This can happen independent of advancing the URG pointer,
2378 * but if two URG's are pending at once, some out-of-band
2379 * data may creep in... ick.
2380 */
61896e3c
JH
2381 if (th->th_urp <= (u_long)tlen &&
2382 !(so->so_options & SO_OOBINLINE)) {
2383 /* hdr drop is delayed */
2384 tcp_pulloutofband(so, th, m, drop_hdrlen);
2385 }
984263bc
MD
2386 } else {
2387 /*
2388 * If no out of band data is expected,
2389 * pull receive urgent pointer along
2390 * with the receive window.
2391 */
2392 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2393 tp->rcv_up = tp->rcv_nxt;
2394 }
984263bc 2395
61896e3c 2396dodata: /* XXX */
984263bc
MD
2397 /*
2398 * Process the segment text, merging it into the TCP sequencing queue,
2399 * and arranging for acknowledgment of receipt if necessary.
2400 * This process logically involves adjusting tp->rcv_wnd as data
2401 * is presented to the user (this happens in tcp_usrreq.c,
2402 * case PRU_RCVD). If a FIN has already been received on this
2403 * connection then we just ignore the text.
2404 */
95b22adf 2405 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2406 m_adj(m, drop_hdrlen); /* delayed header drop */
2407 /*
2408 * Insert segment which includes th into TCP reassembly queue
2409 * with control block tp. Set thflags to whether reassembly now
2410 * includes a segment with FIN. This handles the common case
2411 * inline (segment is the next to be received on an established
2412 * connection, and the queue is empty), avoiding linkage into
2413 * and removal from the queue and repetition of various
2414 * conversions.
2415 * Set DELACK for segments received in order, but ack
2416 * immediately when segments are out of order (so
2417 * fast retransmit can work).
2418 */
2419 if (th->th_seq == tp->rcv_nxt &&
0f9e45de 2420 TAILQ_EMPTY(&tp->t_segq) &&
984263bc 2421 TCPS_HAVEESTABLISHED(tp->t_state)) {
a48c5dd5
SZ
2422 if (DELAY_ACK(tp)) {
2423 tcp_callout_reset(tp, tp->tt_delack,
2424 tcp_delacktime, tcp_timer_delack);
2425 } else {
984263bc 2426 tp->t_flags |= TF_ACKNOW;
a48c5dd5 2427 }
984263bc
MD
2428 tp->rcv_nxt += tlen;
2429 thflags = th->th_flags & TH_FIN;
2430 tcpstat.tcps_rcvpack++;
2431 tcpstat.tcps_rcvbyte += tlen;
2432 ND6_HINT(tp);
6cef7136 2433 if (so->so_state & SS_CANTRCVMORE) {
984263bc 2434 m_freem(m);
6cef7136
MD
2435 } else {
2436 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 2437 ssb_appendstream(&so->so_rcv, m);
6cef7136
MD
2438 lwkt_reltoken(&so->so_rcv.ssb_token);
2439 }
984263bc
MD
2440 sorwakeup(so);
2441 } else {
c7e6499a 2442 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
91489f6b
JH
2443 /* Initialize SACK report block. */
2444 tp->reportblk.rblk_start = th->th_seq;
3a5d999b
SZ
2445 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2446 th->th_seq + tlen, thflags);
91489f6b 2447 }
984263bc
MD
2448 thflags = tcp_reass(tp, th, &tlen, m);
2449 tp->t_flags |= TF_ACKNOW;
2450 }
2451
2452 /*
2453 * Note the amount of data that peer has sent into
2454 * our window, in order to estimate the sender's
2455 * buffer size.
2456 */
6d49aa6f 2457 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
984263bc
MD
2458 } else {
2459 m_freem(m);
2460 thflags &= ~TH_FIN;
2461 }
2462
2463 /*
2464 * If FIN is received ACK the FIN and let the user know
2465 * that the connection is closing.
2466 */
2467 if (thflags & TH_FIN) {
95b22adf 2468 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
984263bc
MD
2469 socantrcvmore(so);
2470 /*
2471 * If connection is half-synchronized
2472 * (ie NEEDSYN flag on) then delay ACK,
2473 * so it may be piggybacked when SYN is sent.
2474 * Otherwise, since we received a FIN then no
2475 * more input can be expected, send ACK now.
2476 */
a48c5dd5
SZ
2477 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2478 tcp_callout_reset(tp, tp->tt_delack,
2479 tcp_delacktime, tcp_timer_delack);
2480 } else {
984263bc 2481 tp->t_flags |= TF_ACKNOW;
a48c5dd5 2482 }
984263bc
MD
2483 tp->rcv_nxt++;
2484 }
984263bc 2485
61896e3c 2486 switch (tp->t_state) {
95b22adf 2487 /*
984263bc
MD
2488 * In SYN_RECEIVED and ESTABLISHED STATES
2489 * enter the CLOSE_WAIT state.
2490 */
2491 case TCPS_SYN_RECEIVED:
2492 tp->t_starttime = ticks;
2493 /*FALLTHROUGH*/
2494 case TCPS_ESTABLISHED:
2495 tp->t_state = TCPS_CLOSE_WAIT;
2496 break;
2497
95b22adf 2498 /*
984263bc
MD
2499 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2500 * enter the CLOSING state.
2501 */
2502 case TCPS_FIN_WAIT_1:
2503 tp->t_state = TCPS_CLOSING;
2504 break;
2505
95b22adf 2506 /*
984263bc
MD
2507 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2508 * starting the time-wait timer, turning off the other
2509 * standard timers.
2510 */
2511 case TCPS_FIN_WAIT_2:
2512 tp->t_state = TCPS_TIME_WAIT;
2513 tcp_canceltimers(tp);
01d3427a 2514 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2515 tcp_timer_2msl);
984263bc
MD
2516 soisdisconnected(so);
2517 break;
2518
2519 /*
2520 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2521 */
2522 case TCPS_TIME_WAIT:
01d3427a 2523 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
a48c5dd5 2524 tcp_timer_2msl);
984263bc
MD
2525 break;
2526 }
2527 }
61896e3c 2528
984263bc
MD
2529#ifdef TCPDEBUG
2530 if (so->so_options & SO_DEBUG)
f23061d4 2531 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2532#endif
2533
27f4bf33
SZ
2534 /*
2535 * Delayed duplicated ACK processing
2536 */
5c99b248 2537 if (delayed_dupack && tcp_recv_dupack(tp, th_dupack, &to))
27f4bf33
SZ
2538 needoutput = FALSE;
2539
984263bc
MD
2540 /*
2541 * Return any desired output.
2542 */
2543 if (needoutput || (tp->t_flags & TF_ACKNOW))
f23061d4 2544 tcp_output(tp);
d58ca578 2545 tcp_sack_report_cleanup(tp);
002c1265 2546 return(IPPROTO_DONE);
984263bc
MD
2547
2548dropafterack:
2549 /*
2550 * Generate an ACK dropping incoming segment if it occupies
2551 * sequence space, where the ACK reflects our state.
2552 *
2553 * We can now skip the test for the RST flag since all
2554 * paths to this code happen after packets containing
2555 * RST have been dropped.
2556 *
2557 * In the SYN-RECEIVED state, don't send an ACK unless the
2558 * segment we received passes the SYN-RECEIVED ACK test.
2559 * If it fails send a RST. This breaks the loop in the
2560 * "LAND" DoS attack, and also prevents an ACK storm
2561 * between two listening ports that have been sent forged
2562 * SYN segments, each with the source address of the other.
2563 */
2564 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2565 (SEQ_GT(tp->snd_una, th->th_ack) ||
2566 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2567 rstreason = BANDLIM_RST_OPENPORT;
2568 goto dropwithreset;
2569 }
2570#ifdef TCPDEBUG
2571 if (so->so_options & SO_DEBUG)
f23061d4 2572 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2573#endif
2574 m_freem(m);
2575 tp->t_flags |= TF_ACKNOW;
f23061d4 2576 tcp_output(tp);
d58ca578 2577 tcp_sack_report_cleanup(tp);
002c1265 2578 return(IPPROTO_DONE);
984263bc
MD
2579
2580dropwithreset:
2581 /*
2582 * Generate a RST, dropping incoming segment.
2583 * Make ACK acceptable to originator of segment.
2584 * Don't bother to respond if destination was broadcast/multicast.
2585 */
61896e3c 2586 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
984263bc
MD
2587 goto drop;
2588 if (isipv6) {
2589 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2590 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2591 goto drop;
2592 } else {
2593 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2594 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
95b22adf
JH
2595 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2596 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
984263bc
MD
2597 goto drop;
2598 }
2599 /* IPv6 anycast check is done at tcp6_input() */
2600
2601 /*
2602 * Perform bandwidth limiting.
2603 */
2604#ifdef ICMP_BANDLIM
2605 if (badport_bandlim(rstreason) < 0)
2606 goto drop;
2607#endif
2608
2609#ifdef TCPDEBUG
2610 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2611 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2612#endif
2613 if (thflags & TH_ACK)
2614 /* mtod() below is safe as long as hdr dropping is delayed */
2615 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2616 TH_RST);
2617 else {
2618 if (thflags & TH_SYN)
2619 tlen++;
2620 /* mtod() below is safe as long as hdr dropping is delayed */
61896e3c
JH
2621 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2622 (tcp_seq)0, TH_RST | TH_ACK);
984263bc 2623 }
d58ca578
SZ
2624 if (tp != NULL)
2625 tcp_sack_report_cleanup(tp);
002c1265 2626 return(IPPROTO_DONE);
984263bc
MD
2627
2628drop:
2629 /*
2630 * Drop space held by incoming segment and return.
2631 */
2632#ifdef TCPDEBUG
2633 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
f23061d4 2634 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
984263bc
MD
2635#endif
2636 m_freem(m);
d58ca578
SZ
2637 if (tp != NULL)
2638 tcp_sack_report_cleanup(tp);
002c1265 2639 return(IPPROTO_DONE);
984263bc
MD
2640}
2641
2642/*
2643 * Parse TCP options and place in tcpopt.
2644 */
2645static void
6c1bbf57
SZ
2646tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2647 tcp_seq ack)
984263bc 2648{
91489f6b 2649 int opt, optlen, i;
984263bc
MD
2650
2651 to->to_flags = 0;
2652 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2653 opt = cp[0];
2654 if (opt == TCPOPT_EOL)
2655 break;
2656 if (opt == TCPOPT_NOP)
2657 optlen = 1;
2658 else {
2659 if (cnt < 2)
2660 break;
2661 optlen = cp[1];
2662 if (optlen < 2 || optlen > cnt)
2663 break;
2664 }
2665 switch (opt) {
2666 case TCPOPT_MAXSEG:
2667 if (optlen != TCPOLEN_MAXSEG)
2668 continue;
2669 if (!is_syn)
2670 continue;
2671 to->to_flags |= TOF_MSS;
407e896e 2672 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
984263bc
MD
2673 to->to_mss = ntohs(to->to_mss);
2674 break;
2675 case TCPOPT_WINDOW:
2676 if (optlen != TCPOLEN_WINDOW)
2677 continue;
95b22adf 2678 if (!is_syn)
984263bc
MD
2679 continue;
2680 to->to_flags |= TOF_SCALE;
2681 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2682 break;
2683 case TCPOPT_TIMESTAMP:
2684 if (optlen != TCPOLEN_TIMESTAMP)
2685 continue;
2686 to->to_flags |= TOF_TS;
407e896e 2687 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
984263bc 2688 to->to_tsval = ntohl(to->to_tsval);
407e896e 2689 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
984263bc 2690 to->to_tsecr = ntohl(to->to_tsecr);
ad0af98b
ND
2691 /*
2692 * If echoed timestamp is later than the current time,
2693 * fall back to non RFC1323 RTT calculation.
2694 */
2695 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2696 to->to_tsecr = 0;
984263bc 2697 break;
91489f6b
JH
2698 case TCPOPT_SACK_PERMITTED:
2699 if (optlen != TCPOLEN_SACK_PERMITTED)
2700 continue;
2701 if (!is_syn)
2702 continue;
2703 to->to_flags |= TOF_SACK_PERMITTED;
2704 break;
2705 case TCPOPT_SACK:
2706 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2707 continue;
2708 to->to_nsackblocks = (optlen - 2) / 8;
2709 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2710 to->to_flags |= TOF_SACK;
2711 for (i = 0; i < to->to_nsackblocks; i++) {
2712 struct raw_sackblock *r = &to->to_sackblocks[i];
2713
2714 r->rblk_start = ntohl(r->rblk_start);
2715 r->rblk_end = ntohl(r->rblk_end);
865b0477
SZ
2716
2717 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2718 /*
2719 * Invalid SACK block; discard all
2720 * SACK blocks
2721 */
02cc2f35 2722 tcpstat.tcps_rcvbadsackopt++;
865b0477
SZ
2723 to->to_nsackblocks = 0;
2724 to->to_sackblocks = NULL;
2725 to->to_flags &= ~TOF_SACK;
2726 break;
2727 }
91489f6b 2728 }
6c1bbf57
SZ
2729 if ((to->to_flags & TOF_SACK) &&
2730 tcp_sack_ndsack_blocks(to->to_sackblocks,
2731 to->to_nsackblocks, ack))
2732 to->to_flags |= TOF_DSACK;
91489f6b 2733 break;
b1992928
MD
2734#ifdef TCP_SIGNATURE
2735 /*
2736 * XXX In order to reply to a host which has set the
2737 * TCP_SIGNATURE option in its initial SYN, we have to
2738 * record the fact that the option was observed here
2739 * for the syncache code to perform the correct response.
2740 */
2741 case TCPOPT_SIGNATURE:
2742 if (optlen != TCPOLEN_SIGNATURE)
2743 continue;
2744 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2745 break;
2746#endif /* TCP_SIGNATURE */
984263bc
MD
2747 default:
2748 continue;
2749 }
2750 }
2751}
2752
2753/*
2754 * Pull out of band byte out of a segment so
2755 * it doesn't appear in the user's data queue.
2756 * It is still reflected in the segment length for
2757 * sequencing purposes.
95b22adf 2758 * "off" is the delayed to be dropped hdrlen.
984263bc
MD
2759 */
2760static void
95b22adf 2761tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
984263bc
MD
2762{
2763 int cnt = off + th->th_urp - 1;
2764
2765 while (cnt >= 0) {
2766 if (m->m_len > cnt) {
2767 char *cp = mtod(m, caddr_t) + cnt;
2768 struct tcpcb *tp = sototcpcb(so);
2769
2770 tp->t_iobc = *cp;
2771 tp->t_oobflags |= TCPOOB_HAVEDATA;
95b22adf 2772 bcopy(cp + 1, cp, m->m_len - cnt - 1);
984263bc
MD
2773 m->m_len--;
2774 if (m->m_flags & M_PKTHDR)
2775 m->m_pkthdr.len--;
2776 return;
2777 }
2778 cnt -= m->m_len;
2779 m = m->m_next;
4090d6ff 2780 if (m == NULL)
984263bc
MD
2781 break;
2782 }
2783 panic("tcp_pulloutofband");
2784}
2785
2786/*
2787 * Collect new round-trip time estimate
2788 * and update averages and current timeout.
2789 */
2790static void
073ec6c4 2791tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
984263bc 2792{
073ec6c4 2793 int rebaserto = 0;
984263bc
MD
2794
2795 tcpstat.tcps_rttupdated++;
2796 tp->t_rttupdated++;
1c8b7a61
SZ
2797 if ((tp->rxt_flags & TRXT_F_REBASERTO) &&
2798 SEQ_GT(ack, tp->snd_max_prev)) {
073ec6c4
SZ
2799#ifdef DEBUG_EIFEL_RESPONSE
2800 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2801 tp->t_srtt_prev, tp->t_rttvar_prev,
2802 tp->t_srtt, tp->t_rttvar);
2803#endif
2804
2805 tcpstat.tcps_eifelresponse++;
2806 rebaserto = 1;
1c8b7a61 2807 tp->rxt_flags &= ~TRXT_F_REBASERTO;
073ec6c4
SZ
2808 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2809 tp->t_rttvar = max(tp->t_rttvar_prev,
2810 (rtt << (TCP_RTTVAR_SHIFT - 1)));
2811 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2812 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2813
2814#ifdef DEBUG_EIFEL_RESPONSE
2815 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2816#endif
2817 } else if (tp->t_srtt != 0) {
2818 int delta;
2819
984263bc
MD
2820 /*
2821 * srtt is stored as fixed point with 5 bits after the
2822 * binary point (i.e., scaled by 8). The following magic
2823 * is equivalent to the smoothing algorithm in rfc793 with
2824 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2825 * point). Adjust rtt to origin 0.
2826 */
2827 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2828 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2829
2830 if ((tp->t_srtt += delta) <= 0)
2831 tp->t_srtt = 1;
2832
2833 /*
2834 * We accumulate a smoothed rtt variance (actually, a
2835 * smoothed mean difference), then set the retransmit
2836 * timer to smoothed rtt + 4 times the smoothed variance.
2837 * rttvar is stored as fixed point with 4 bits after the
2838 * binary point (scaled by 16). The following is
2839 * equivalent to rfc793 smoothing with an alpha of .75
2840 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2841 * rfc793's wired-in beta.
2842 */
2843 if (delta < 0)
2844 delta = -delta;
2845 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2846 if ((tp->t_rttvar += delta) <= 0)
2847 tp->t_rttvar = 1;
2848 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2849 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2850 } else {
2851 /*
2852 * No rtt measurement yet - use the unsmoothed rtt.
2853 * Set the variance to half the rtt (so our first
2854 * retransmit happens at 3*rtt).
2855 */
2856 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2857 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2858 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2859 }
2860 tp->t_rtttime = 0;
2861 tp->t_rxtshift = 0;
2862
073ec6c4
SZ
2863#ifdef DEBUG_EIFEL_RESPONSE
2864 if (rebaserto) {
2865 kprintf("| rxtcur prev %d, old %d, ",
2866 tp->t_rxtcur_prev, tp->t_rxtcur);
2867 }
2868#endif
2869
984263bc
MD
2870 /*
2871 * the retransmit should happen at rtt + 4 * rttvar.
2872 * Because of the way we do the smoothing, srtt and rttvar
2873 * will each average +1/2 tick of bias. When we compute
2874 * the retransmit timer, we want 1/2 tick of rounding and
2875 * 1 extra tick because of +-1/2 tick uncertainty in the
2876 * firing of the timer. The bias will give us exactly the
2877 * 1.5 tick we need. But, because the bias is
2878 * statistical, we have to test that we don't drop below
2879 * the minimum feasible timer (which is 2 ticks).
2880 */
2881 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2882 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2883
073ec6c4
SZ
2884 if (rebaserto) {
2885 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2886 /*
2887 * RFC4015 requires that the new RTO is at least
2888 * 2*G (tcp_eifel_rtoinc) greater then the RTO
2889 * (t_rxtcur_prev) when the spurious retransmit
2890 * timeout happens.
2891 *
2892 * The above condition could be true, if the SRTT
2893 * and RTTVAR used to calculate t_rxtcur_prev
2894 * resulted in a value less than t_rttmin. So
2895 * simply increasing SRTT by tcp_eifel_rtoinc when
e1dd9e15 2896 * preparing for the Eifel response could not ensure
073ec6c4
SZ
2897 * that the new RTO will be tcp_eifel_rtoinc greater
2898 * t_rxtcur_prev.
2899 */
2900 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2901 }
2902#ifdef DEBUG_EIFEL_RESPONSE
2903 kprintf("new %d\n", tp->t_rxtcur);
2904#endif
2905 }
2906
984263bc
MD
2907 /*
2908 * We received an ack for a packet that wasn't retransmitted;
2909 * it is probably safe to discard any error indications we've
2910 * received recently. This isn't quite right, but close enough
2911 * for now (a route might have failed after we sent a segment,
2912 * and the return path might not be symmetrical).
2913 */
2914 tp->t_softerror = 0;
2915}
2916
2917/*
2918 * Determine a reasonable value for maxseg size.
2919 * If the route is known, check route for mtu.
2920 * If none, use an mss that can be handled on the outgoing
2921 * interface without forcing IP to fragment; if bigger than
2922 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2923 * to utilize large mbufs. If no route is found, route has no mtu,
2924 * or the destination isn't local, use a default, hopefully conservative
2925 * size (usually 512 or the default IP max size, but no more than the mtu
2926 * of the interface), as we can't discover anything about intervening
2927 * gateways or networks. We also initialize the congestion/slow start
2928 * window to be a single segment if the destination isn't local.
2929 * While looking at the routing entry, we also initialize other path-dependent
2930 * parameters from pre-set or cached values in the routing entry.
2931 *
2932 * Also take into account the space needed for options that we
2933 * send regularly. Make maxseg shorter by that amount to assure
2934 * that we can send maxseg amount of data even when the options
2935 * are present. Store the upper limit of the length of options plus
2936 * data in maxopd.
2937 *
2938 * NOTE that this routine is only called when we process an incoming
2939 * segment, for outgoing segments only tcp_mssopt is called.
984263bc
MD
2940 */
2941void
95b22adf 2942tcp_mss(struct tcpcb *tp, int offer)
984263bc 2943{
2256ba69 2944 struct rtentry *rt;
984263bc 2945 struct ifnet *ifp;
2256ba69 2946 int rtt, mss;
984263bc
MD
2947 u_long bufsize;
2948 struct inpcb *inp = tp->t_inpcb;
2949 struct socket *so;
984263bc 2950#ifdef INET6
d24ce1dc 2951 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
2952 size_t min_protoh = isipv6 ?
2953 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2954 sizeof(struct tcpiphdr);
2955#else
d24ce1dc 2956 const boolean_t isipv6 = FALSE;
984263bc
MD
2957 const size_t min_protoh = sizeof(struct tcpiphdr);
2958#endif
2959
2960 if (isipv6)
2961 rt = tcp_rtlookup6(&inp->inp_inc);
2962 else
2963 rt = tcp_rtlookup(&inp->inp_inc);
2964 if (rt == NULL) {
2965 tp->t_maxopd = tp->t_maxseg =
d24ce1dc 2966 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
984263bc
MD
2967 return;
2968 }
2969 ifp = rt->rt_ifp;
2970 so = inp->inp_socket;
2971
984263bc
MD
2972 /*
2973 * Offer == 0 means that there was no MSS on the SYN segment,
b235ad6d
MD
2974 * in this case we use either the interface mtu or tcp_mssdflt.
2975 *
2976 * An offer which is too large will be cut down later.
984263bc 2977 */
5b0b9fa5 2978 if (offer == 0) {
b235ad6d
MD
2979 if (isipv6) {
2980 if (in6_localaddr(&inp->in6p_faddr)) {
2981 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2982 min_protoh;
2983 } else {
2984 offer = tcp_v6mssdflt;
2985 }
2986 } else {
2987 if (in_localaddr(inp->inp_faddr))
2988 offer = ifp->if_mtu - min_protoh;
2989 else
2990 offer = tcp_mssdflt;
2991 }
5b0b9fa5 2992 }
b235ad6d
MD
2993
2994 /*
2995 * Prevent DoS attack with too small MSS. Round up
2996 * to at least minmss.
2997 *
2998 * Sanity check: make sure that maxopd will be large
2999 * enough to allow some data on segments even is the
3000 * all the option space is used (40bytes). Otherwise
3001 * funny things may happen in tcp_output.
3002 */
3003 offer = max(offer, tcp_minmss);
3004 offer = max(offer, 64);
3005
27b8aee3 3006 rt->rt_rmx.rmx_mssopt = offer;
984263bc
MD
3007
3008 /*
3009 * While we're here, check if there's an initial rtt
3010 * or rttvar. Convert from the route-table units
3011 * to scaled multiples of the slow timeout timer.
3012 */
3013 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3014 /*
3015 * XXX the lock bit for RTT indicates that the value
3016 * is also a minimum value; this is subject to time.
3017 */
3018 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3019 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3020 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3021 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3022 tcpstat.tcps_usedrtt++;
3023 if (rt->rt_rmx.rmx_rttvar) {
3024 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3025 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3026 tcpstat.tcps_usedrttvar++;
3027 } else {
3028 /* default variation is +- 1 rtt */
3029 tp->t_rttvar =
3030 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3031 }
3032 TCPT_RANGESET(tp->t_rxtcur,
3033 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3034 tp->t_rttmin, TCPTV_REXMTMAX);
3035 }
b235ad6d 3036
984263bc
MD
3037 /*
3038 * if there's an mtu associated with the route, use it
b235ad6d
MD
3039 * else, use the link mtu. Take the smaller of mss or offer
3040 * as our final mss.
984263bc 3041 */
b235ad6d 3042 if (rt->rt_rmx.rmx_mtu) {
984263bc 3043 mss = rt->rt_rmx.rmx_mtu - min_protoh;
b235ad6d
MD
3044 } else {
3045 if (isipv6)
698ac46c 3046 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
b235ad6d 3047 else
984263bc 3048 mss = ifp->if_mtu - min_protoh;
984263bc
MD
3049 }
3050 mss = min(mss, offer);
b235ad6d 3051
984263bc
MD
3052 /*
3053 * maxopd stores the maximum length of data AND options
3054 * in a segment; maxseg is the amount of data in a normal
3055 * segment. We need to store this value (maxopd) apart
3056 * from maxseg, because now every segment carries options
3057 * and thus we normally have somewhat less data in segments.
3058 */
3059 tp->t_maxopd = mss;
3060
61896e3c 3061 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
27b8aee3 3062 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
984263bc 3063 mss -= TCPOLEN_TSTAMP_APPA;
984263bc
MD
3064
3065#if (MCLBYTES & (MCLBYTES - 1)) == 0
f99dae5e
SW
3066 if (mss > MCLBYTES)
3067 mss &= ~(MCLBYTES-1);
984263bc 3068#else
f99dae5e
SW
3069 if (mss > MCLBYTES)
3070 mss = mss / MCLBYTES * MCLBYTES;
984263bc
MD
3071#endif
3072 /*
3073 * If there's a pipesize, change the socket buffer
3074 * to that size. Make the socket buffers an integral
3075 * number of mss units; if the mss is larger than
3076 * the socket buffer, decrease the mss.
3077 */
3078#ifdef RTV_SPIPE
3079 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3080#endif
6d49aa6f 3081 bufsize = so->so_snd.ssb_hiwat;
984263bc
MD
3082 if (bufsize < mss)
3083 mss = bufsize;
3084 else {
3085 bufsize = roundup(bufsize, mss);
3086 if (bufsize > sb_max)
3087 bufsize = sb_max;
6d49aa6f
MD
3088 if (bufsize > so->so_snd.ssb_hiwat)
3089 ssb_reserve(&so->so_snd, bufsize, so, NULL);
984263bc
MD
3090 }
3091 tp->t_maxseg = mss;
3092
3093#ifdef RTV_RPIPE
3094 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3095#endif
6d49aa6f 3096 bufsize = so->so_rcv.ssb_hiwat;
984263bc
MD
3097 if (bufsize > mss) {
3098 bufsize = roundup(bufsize, mss);
3099 if (bufsize > sb_max)
3100 bufsize = sb_max;
6cef7136
MD
3101 if (bufsize > so->so_rcv.ssb_hiwat) {
3102 lwkt_gettoken(&so->so_rcv.ssb_token);
6d49aa6f 3103 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
6cef7136
MD
3104 lwkt_reltoken(&so->so_rcv.ssb_token);
3105 }
984263bc
MD
3106 }
3107
3108 /*
697020f5
SZ
3109 * Set the slow-start flight size
3110 *
3111 * NOTE: t_maxseg must have been configured!
984263bc 3112 */
697020f5 3113 tp->snd_cwnd = tcp_initial_window(tp);
984263bc
MD
3114
3115 if (rt->rt_rmx.rmx_ssthresh) {
3116 /*
3117 * There's some sort of gateway or interface
3118 * buffer limit on the path. Use this to set
3119 * the slow start threshhold, but set the
3120 * threshold to no less than 2*mss.
3121 */
3122 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3123 tcpstat.tcps_usedssthresh++;
3124 }
3125}
3126
3127/*
3128 * Determine the MSS option to send on an outgoing SYN.
3129 */
3130int
95b22adf 3131tcp_mssopt(struct tcpcb *tp)
984263bc
MD
3132{
3133 struct rtentry *rt;
3134#ifdef INET6
d24ce1dc
JH
3135 boolean_t isipv6 =
3136 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
984263bc
MD
3137 int min_protoh = isipv6 ?
3138 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3139 sizeof(struct tcpiphdr);
3140#else
d24ce1dc 3141 const boolean_t isipv6 = FALSE;
984263bc
MD
3142 const size_t min_protoh = sizeof(struct tcpiphdr);
3143#endif
3144
3145 if (isipv6)
3146 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3147 else
3148 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3149 if (rt == NULL)
3150 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3151
3152 return (rt->rt_ifp->if_mtu - min_protoh);
3153}
3154
984263bc
MD
3155/*
3156 * When a partial ack arrives, force the retransmission of the
91489f6b
JH
3157 * next unacknowledged segment. Do not exit Fast Recovery.
3158 *
3159 * Implement the Slow-but-Steady variant of NewReno by restarting the
3160 * the retransmission timer. Turn it off here so it can be restarted
3161 * later in tcp_output().
984263bc
MD
3162 */
3163static void
91489f6b 3164tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
984263bc 3165{
91489f6b
JH
3166 tcp_seq old_snd_nxt = tp->snd_nxt;
3167 u_long ocwnd = tp->snd_cwnd;
984263bc 3168
a48c5dd5 3169 tcp_callout_stop(tp, tp->tt_rexmt);
984263bc
MD
3170 tp->t_rtttime = 0;
3171 tp->snd_nxt = th->th_ack;
91489f6b
JH
3172 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3173 tp->snd_cwnd = tp->t_maxseg;
984263bc 3174 tp->t_flags |= TF_ACKNOW;
f23061d4 3175 tcp_output(tp);
91489f6b
JH
3176 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3177 tp->snd_nxt = old_snd_nxt;
3178 /* partial window deflation */
68947d12
ND
3179 if (ocwnd > acked)
3180 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3181 else
3182 tp->snd_cwnd = tp->t_maxseg;
91489f6b
JH
3183}
3184
3185/*
3186 * In contrast to the Slow-but-Steady NewReno variant,
3187 * we do not reset the retransmission timer for SACK retransmissions,
3188 * except when retransmitting snd_una.
3189 */
3190static void
ccb518ea 3191tcp_sack_rexmt(struct tcpcb *tp, boolean_t force)
91489f6b 3192{
91489f6b
JH
3193 tcp_seq old_snd_nxt = tp->snd_nxt;
3194 u_long ocwnd = tp->snd_cwnd;
1bdfd728 3195 uint32_t pipe;
91489f6b 3196 int nseg = 0; /* consecutive new segments */
a098966f 3197 int nseg_rexmt = 0; /* retransmitted segments */
ccb518ea 3198 int maxrexmt = 0;
91489f6b
JH
3199#define MAXBURST 4 /* limit burst of new packets on partial ack */
3200
ccb518ea
SZ
3201 if (force) {
3202 uint32_t unsacked = tcp_sack_first_unsacked_len(tp);
3203
3204 /*
3205 * Try to fill the first hole in the receiver's
3206 * reassemble queue.
3207 */
3208 maxrexmt = howmany(unsacked, tp->t_maxseg);
3209 if (maxrexmt > tcp_force_sackrxt)
3210 maxrexmt = tcp_force_sackrxt;
3211 }
3212
91489f6b
JH
3213 tp->t_rtttime = 0;
3214 pipe = tcp_sack_compute_pipe(tp);
ccb518ea
SZ
3215 while (((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg
3216 || (force && nseg_rexmt < maxrexmt && nseg == 0)) &&
1bdfd728
SZ
3217 (!tcp_do_smartsack || nseg < MAXBURST)) {
3218 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3219 uint32_t sent, seglen;
3220 boolean_t rescue;
91489f6b
JH
3221 int error;
3222
1bdfd728 3223 old_rexmt_high = tp->rexmt_high;
64debfd6 3224 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
1bdfd728
SZ
3225 tp->rexmt_high = old_rexmt_high;
3226 break;
3227 }
3228
a098966f
SZ
3229 /*
3230 * If the next tranmission is a rescue retranmission,
3231 * we check whether we have already sent some data
3232 * (either new segments or retransmitted segments)
3233 * into the the network or not. Since the idea of rescue
3234 * retransmission is to sustain ACK clock, as long as
3235 * some segments are in the network, ACK clock will be
3236 * kept ticking.
3237 */
3238 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3239 tp->rexmt_high = old_rexmt_high;
3240 break;
3241 }
3242
98cb2337
MD
3243 if (nextrexmt == tp->snd_max)
3244 ++nseg;
a098966f
SZ
3245 else
3246 ++nseg_rexmt;
91489f6b
JH
3247 tp->snd_nxt = nextrexmt;
3248 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3249 old_snd_max = tp->snd_max;
3250 if (nextrexmt == tp->snd_una)
a48c5dd5 3251 tcp_callout_stop(tp, tp->tt_rexmt);
91489f6b 3252 error = tcp_output(tp);
1bdfd728
SZ
3253 if (error != 0) {
3254 tp->rexmt_high = old_rexmt_high;
91489f6b 3255 break;
1bdfd728 3256 }
91489f6b 3257 sent = tp->snd_nxt - nextrexmt;
1bdfd728
SZ
3258 if (sent <= 0) {
3259 tp->rexmt_high = old_rexmt_high;
91489f6b 3260 break;
1bdfd728
SZ
3261 }
3262 pipe += sent;
91489f6b
JH
3263 tcpstat.tcps_sndsackpack++;
3264 tcpstat.tcps_sndsackbyte += sent;
1bdfd728
SZ
3265
3266 if (rescue) {
a098966f 3267 tcpstat.tcps_sackrescue++;
1bdfd728 3268 tp->rexmt_rescue = tp->snd_nxt;
c7e6499a 3269 tp->sack_flags |= TSACK_F_SACKRESCUED;
1bdfd728
SZ
3270 break;
3271 }
91489f6b 3272 if (SEQ_LT(nextrexmt, old_snd_max) &&
a098966f 3273 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
91489f6b 3274 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
21f337ca 3275 if (tcp_aggressive_rescuesack &&
c7e6499a 3276 (tp->sack_flags & TSACK_F_SACKRESCUED) &&
a098966f
SZ
3277 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3278 /* Drag RescueRxt along with HighRxt */
3279 tp->rexmt_rescue = tp->rexmt_high;
3280 }
3281 }
91489f6b
JH
3282 }
3283 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3284 tp->snd_nxt = old_snd_nxt;
984263bc 3285 tp->snd_cwnd = ocwnd;
984263bc 3286}
0ecd93f9 3287
e2289e66
SZ
3288/*
3289 * Return TRUE, if some new segments are sent
3290 */
ffe35e17
SZ
3291static boolean_t
3292tcp_sack_limitedxmit(struct tcpcb *tp)
3293{
3294 tcp_seq oldsndnxt = tp->snd_nxt;
3295 tcp_seq oldsndmax = tp->snd_max;
3296 u_long ocwnd = tp->snd_cwnd;
db421eef 3297 uint32_t pipe, sent;
ffe35e17 3298 boolean_t ret = FALSE;
db421eef
SZ
3299 tcp_seq_diff_t cwnd_left;
3300 tcp_seq next;
ffe35e17
SZ
3301
3302 tp->rexmt_high = tp->snd_una - 1;
3303 pipe = tcp_sack_compute_pipe(tp);
db421eef
SZ
3304 cwnd_left = (tcp_seq_diff_t)(ocwnd - pipe);
3305 if (cwnd_left < (tcp_seq_diff_t)tp->t_maxseg)
3306 return FALSE;
ffe35e17 3307
db421eef
SZ
3308 next = tp->snd_nxt = tp->snd_max;
3309 tp->snd_cwnd = tp->snd_nxt - tp->snd_una +
3310 rounddown(cwnd_left, tp->t_maxseg);
ffe35e17 3311
db421eef 3312 tcp_output(tp);
ffe35e17 3313
db421eef
SZ
3314 sent = tp->snd_nxt - next;
3315 if (sent > 0) {
3316 tcpstat.tcps_sndlimited += howmany(sent, tp->t_maxseg);
ffe35e17
SZ
3317 ret = TRUE;
3318 }
3319
3320 if (SEQ_LT(oldsndnxt, oldsndmax)) {
3321 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3322 ("snd_una moved in other threads"));
3323 tp->snd_nxt = oldsndnxt;
3324 }
3325 tp->snd_cwnd = ocwnd;
3326
e2289e66
SZ
3327 if (ret && TCP_DO_NCR(tp))
3328 tcp_ncr_update_rxtthresh(tp);
3329
ffe35e17
SZ
3330 return ret;
3331}
3332
0ecd93f9
MD
3333/*
3334 * Reset idle time and keep-alive timer, typically called when a valid
3335 * tcp packet is received but may also be called when FASTKEEP is set
3336 * to prevent the previous long-timeout from calculating to a drop.
3337 *
3338 * Only update t_rcvtime for non-SYN packets.
3339 *
3340 * Handle the case where one side thinks the connection is established
3341 * but the other side has, say, rebooted without cleaning out the
3342 * connection. The SYNs could be construed as an attack and wind
3343 * up ignored, but in case it isn't an attack we can validate the
3344 * connection by forcing a keepalive.
3345 */
3346void
3347tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3348{
3349 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3350 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3351 tp->t_flags |= TF_KEEPALIVE;
3352 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3353 tcp_timer_keep);
3354 } else {
3355 tp->t_rcvtime = ticks;
3356 tp->t_flags &= ~TF_KEEPALIVE;
3357 tcp_callout_reset(tp, tp->tt_keep,
fb8d5c6d 3358 tp->t_keepidle,
0ecd93f9
MD
3359 tcp_timer_keep);
3360 }
3361 }
3362}
01d3427a
SZ
3363
3364static int
3365tcp_rmx_msl(const struct tcpcb *tp)
3366{
3367 struct rtentry *rt;
3368 struct inpcb *inp = tp->t_inpcb;
9855a4ef 3369 int msl;
01d3427a
SZ
3370#ifdef INET6
3371 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3372#else
3373 const boolean_t isipv6 = FALSE;
3374#endif
3375
3376 if (isipv6)
3377 rt = tcp_rtlookup6(&inp->inp_inc);
3378 else
3379 rt = tcp_rtlookup(&inp->inp_inc);
3380 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3381 return tcp_msl;
3382
9855a4ef
SZ
3383 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3384 if (msl == 0)
3385 msl = 1;
3386
3387 return msl;
01d3427a 3388}
8651f7f8
SZ
3389
3390static void
3391tcp_established(struct tcpcb *tp)
3392{
3393 tp->t_state = TCPS_ESTABLISHED;
fb8d5c6d 3394 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
be34e534 3395
d5082e3d 3396 if (tp->t_rxtsyn > 0) {
48a4676d
SZ
3397 /*
3398 * RFC6298:
3399 * "If the timer expires awaiting the ACK of a SYN segment
3400 * and the TCP implementation is using an RTO less than 3
3401 * seconds, the RTO MUST be re-initialized to 3 seconds
3402 * when data transmission begins"
3403 */
3404 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3405 tp->t_rxtcur = TCPTV_RTOBASE3;
be34e534 3406 }
8651f7f8 3407}
27f4bf33
SZ
3408
3409/*
3410 * Returns TRUE, if the ACK should be dropped
3411 */
3412static boolean_t
5c99b248 3413tcp_recv_dupack(struct tcpcb *tp, tcp_seq th_ack, const struct tcpopt *to)
27f4bf33 3414{
ba0d6f99
SZ
3415 boolean_t fast_sack_rexmt = TRUE;
3416
27f4bf33
SZ
3417 tcpstat.tcps_rcvdupack++;
3418
3419 /*
3420 * We have outstanding data (other than a window probe),
3421 * this is a completely duplicate ack (ie, window info
3422 * didn't change), the ack is the biggest we've seen and
3423 * we've seen exactly our rexmt threshhold of them, so
3424 * assume a packet has been dropped and retransmit it.
3425 * Kludge snd_nxt & the congestion window so we send only
3426 * this one packet.
3427 */
3428 if (IN_FASTRECOVERY(tp)) {
3429 if (TCP_DO_SACK(tp)) {
ccb518ea
SZ
3430 boolean_t force = FALSE;
3431
3432 if (tp->snd_una == tp->rexmt_high &&
3433 (to->to_flags & (TOF_SACK | TOF_SACK_REDUNDANT)) ==
3434 TOF_SACK) {
3435 /*
3436 * New segments got SACKed and
3437 * no retransmit yet.
3438 */
3439 force = TRUE;
3440 }
3441
27f4bf33 3442 /* No artifical cwnd inflation. */
ccb518ea 3443 tcp_sack_rexmt(tp, force);
27f4bf33
SZ
3444 } else {
3445 /*
3446 * Dup acks mean that packets have left
3447 * the network (they're now cached at the
3448 * receiver) so bump cwnd by the amount in
3449 * the receiver to keep a constant cwnd
3450 * packets in the network.
3451 */
3452 tp->snd_cwnd += tp->t_maxseg;
3453 tcp_output(tp);
3454 }
e2289e66 3455 return TRUE;
27f4bf33
SZ
3456 } else if (SEQ_LT(th_ack, tp->snd_recover)) {
3457 tp->t_dupacks = 0;
3458 return FALSE;
3459 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
3460 (to->to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
3461 (TOF_DSACK | TOF_SACK_REDUNDANT)) {
3462 /*
3463 * If the ACK carries DSACK and other SACK blocks
3464 * carry information that we have already known,
3465 * don't count this ACK as duplicate ACK. This
3466 * prevents spurious early retransmit and fast
3467 * retransmit. This also meets the requirement of
3468 * RFC3042 that new segments should not be sent if
3469 * t