xref: /dflybsd-src/sys/netinet/tcp_input.c (revision ed20d0e3546bb6e350a4091da731a060b0050541)
1 /*
2  * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2002, 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
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.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 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 
70 #include "opt_inet.h"
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>
87 #include <sys/in_cksum.h>
88 
89 #include <sys/socketvar2.h>
90 
91 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
92 #include <machine/stdarg.h>
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>
100 #include <netinet/ip_icmp.h>	/* for ICMP_BANDLIM */
101 #include <netinet/in_var.h>
102 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM */
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>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcpip.h>
118 
119 #ifdef TCPDEBUG
120 #include <netinet/tcp_debug.h>
121 
122 u_char tcp_saveipgen[40];    /* the size must be of max ip header, now IPv6 */
123 struct tcphdr tcp_savetcp;
124 #endif
125 
126 #ifdef FAST_IPSEC
127 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
129 #endif
130 
131 #ifdef IPSEC
132 #include <netinet6/ipsec.h>
133 #include <netinet6/ipsec6.h>
134 #include <netproto/key/key.h>
135 #endif
136 
137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
138 
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141     &log_in_vain, 0, "Log all incoming TCP connections");
142 
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145     &blackhole, 0, "Do not send RST when dropping refused connections");
146 
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149     &tcp_delack_enabled, 0,
150     "Delay ACK to try and piggyback it onto a data packet");
151 
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
156 #endif
157 
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160     &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
161 
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164     &tcp_do_early_retransmit, 0, "Early retransmit");
165 
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168     &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
169 
170 static int tcp_do_eifel_detect = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172     &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
173 
174 static int tcp_do_abc = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
176     &tcp_do_abc, 0,
177     "TCP Appropriate Byte Counting (RFC 3465)");
178 
179 /*
180  * Define as tunable for easy testing with SACK on and off.
181  * Warning:  do not change setting in the middle of an existing active TCP flow,
182  *   else strange things might happen to that flow.
183  */
184 int tcp_do_sack = 1;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
186     &tcp_do_sack, 0, "Enable SACK Algorithms");
187 
188 int tcp_do_smartsack = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
190     &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
191 
192 int tcp_do_rescuesack = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
194     &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
195 
196 int tcp_aggressive_rescuesack = 1;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
198     &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
199 
200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
201     "TCP Segment Reassembly Queue");
202 
203 int tcp_reass_maxseg = 0;
204 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
205     &tcp_reass_maxseg, 0,
206     "Global maximum number of TCP Segments in Reassembly Queue");
207 
208 int tcp_reass_qsize = 0;
209 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
210     &tcp_reass_qsize, 0,
211     "Global number of TCP Segments currently in Reassembly Queue");
212 
213 static int tcp_reass_overflows = 0;
214 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
215     &tcp_reass_overflows, 0,
216     "Global number of TCP Segment Reassembly Queue Overflows");
217 
218 int tcp_do_autorcvbuf = 1;
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
220     &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
221 
222 int tcp_autorcvbuf_inc = 16*1024;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
224     &tcp_autorcvbuf_inc, 0,
225     "Incrementor step size of automatic receive buffer");
226 
227 int tcp_autorcvbuf_max = 2*1024*1024;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
229     &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
230 
231 int tcp_sosend_agglim = 2;
232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
233     &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
234 
235 int tcp_sosend_async = 1;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
237     &tcp_sosend_async, 0, "TCP asynchronized pru_send");
238 
239 static void	 tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
240 static void	 tcp_pulloutofband(struct socket *,
241 		     struct tcphdr *, struct mbuf *, int);
242 static int	 tcp_reass(struct tcpcb *, struct tcphdr *, int *,
243 		     struct mbuf *);
244 static void	 tcp_xmit_timer(struct tcpcb *, int);
245 static void	 tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
246 static void	 tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
247 static int	 tcp_rmx_msl(const struct tcpcb *);
248 static void	 tcp_established(struct tcpcb *);
249 
250 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
251 #ifdef INET6
252 #define ND6_HINT(tp) \
253 do { \
254 	if ((tp) && (tp)->t_inpcb && \
255 	    ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
256 	    (tp)->t_inpcb->in6p_route.ro_rt) \
257 		nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
258 } while (0)
259 #else
260 #define ND6_HINT(tp)
261 #endif
262 
263 /*
264  * Indicate whether this ack should be delayed.  We can delay the ack if
265  *	- delayed acks are enabled and
266  *	- there is no delayed ack timer in progress and
267  *	- our last ack wasn't a 0-sized window.  We never want to delay
268  *	  the ack that opens up a 0-sized window.
269  */
270 #define DELAY_ACK(tp) \
271 	(tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
272 	!(tp->t_flags & TF_RXWIN0SENT))
273 
274 #define acceptable_window_update(tp, th, tiwin)				\
275     (SEQ_LT(tp->snd_wl1, th->th_seq) ||					\
276      (tp->snd_wl1 == th->th_seq &&					\
277       (SEQ_LT(tp->snd_wl2, th->th_ack) ||				\
278        (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
279 
280 static int
281 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
282 {
283 	struct tseg_qent *q;
284 	struct tseg_qent *p = NULL;
285 	struct tseg_qent *te;
286 	struct socket *so = tp->t_inpcb->inp_socket;
287 	int flags;
288 
289 	/*
290 	 * Call with th == NULL after become established to
291 	 * force pre-ESTABLISHED data up to user socket.
292 	 */
293 	if (th == NULL)
294 		goto present;
295 
296 	/*
297 	 * Limit the number of segments in the reassembly queue to prevent
298 	 * holding on to too many segments (and thus running out of mbufs).
299 	 * Make sure to let the missing segment through which caused this
300 	 * queue.  Always keep one global queue entry spare to be able to
301 	 * process the missing segment.
302 	 */
303 	if (th->th_seq != tp->rcv_nxt &&
304 	    tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
305 		tcp_reass_overflows++;
306 		tcpstat.tcps_rcvmemdrop++;
307 		m_freem(m);
308 		/* no SACK block to report */
309 		tp->reportblk.rblk_start = tp->reportblk.rblk_end;
310 		return (0);
311 	}
312 
313 	/* Allocate a new queue entry. */
314 	te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
315 	if (te == NULL) {
316 		tcpstat.tcps_rcvmemdrop++;
317 		m_freem(m);
318 		/* no SACK block to report */
319 		tp->reportblk.rblk_start = tp->reportblk.rblk_end;
320 		return (0);
321 	}
322 	atomic_add_int(&tcp_reass_qsize, 1);
323 
324 	/*
325 	 * Find a segment which begins after this one does.
326 	 */
327 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
328 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
329 			break;
330 		p = q;
331 	}
332 
333 	/*
334 	 * If there is a preceding segment, it may provide some of
335 	 * our data already.  If so, drop the data from the incoming
336 	 * segment.  If it provides all of our data, drop us.
337 	 */
338 	if (p != NULL) {
339 		tcp_seq_diff_t i;
340 
341 		/* conversion to int (in i) handles seq wraparound */
342 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
343 		if (i > 0) {		/* overlaps preceding segment */
344 			tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
345 			/* enclosing block starts w/ preceding segment */
346 			tp->encloseblk.rblk_start = p->tqe_th->th_seq;
347 			if (i >= *tlenp) {
348 				/* preceding encloses incoming segment */
349 				tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
350 				    p->tqe_th->th_seq + p->tqe_len,
351 				    p->tqe_th->th_flags);
352 				tcpstat.tcps_rcvduppack++;
353 				tcpstat.tcps_rcvdupbyte += *tlenp;
354 				m_freem(m);
355 				kfree(te, M_TSEGQ);
356 				atomic_add_int(&tcp_reass_qsize, -1);
357 				/*
358 				 * Try to present any queued data
359 				 * at the left window edge to the user.
360 				 * This is needed after the 3-WHS
361 				 * completes.
362 				 */
363 				goto present;	/* ??? */
364 			}
365 			m_adj(m, i);
366 			*tlenp -= i;
367 			th->th_seq += i;
368 			/* incoming segment end is enclosing block end */
369 			tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
370 			    th->th_seq + *tlenp, th->th_flags);
371 			/* trim end of reported D-SACK block */
372 			tp->reportblk.rblk_end = th->th_seq;
373 		}
374 	}
375 	tcpstat.tcps_rcvoopack++;
376 	tcpstat.tcps_rcvoobyte += *tlenp;
377 
378 	/*
379 	 * While we overlap succeeding segments trim them or,
380 	 * if they are completely covered, dequeue them.
381 	 */
382 	while (q) {
383 		tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
384 		tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
385 		tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
386 		struct tseg_qent *nq;
387 
388 		if (i <= 0)
389 			break;
390 		if (!(tp->t_flags & TF_DUPSEG)) {    /* first time through */
391 			tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
392 			tp->encloseblk = tp->reportblk;
393 			/* report trailing duplicate D-SACK segment */
394 			tp->reportblk.rblk_start = q->tqe_th->th_seq;
395 		}
396 		if ((tp->t_flags & TF_ENCLOSESEG) &&
397 		    SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
398 			/* extend enclosing block if one exists */
399 			tp->encloseblk.rblk_end = qend_sack;
400 		}
401 		if (i < q->tqe_len) {
402 			q->tqe_th->th_seq += i;
403 			q->tqe_len -= i;
404 			m_adj(q->tqe_m, i);
405 			break;
406 		}
407 
408 		nq = LIST_NEXT(q, tqe_q);
409 		LIST_REMOVE(q, tqe_q);
410 		m_freem(q->tqe_m);
411 		kfree(q, M_TSEGQ);
412 		atomic_add_int(&tcp_reass_qsize, -1);
413 		q = nq;
414 	}
415 
416 	/* Insert the new segment queue entry into place. */
417 	te->tqe_m = m;
418 	te->tqe_th = th;
419 	te->tqe_len = *tlenp;
420 
421 	/* check if can coalesce with following segment */
422 	if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
423 		tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
424 		tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
425 
426 		te->tqe_len += q->tqe_len;
427 		if (q->tqe_th->th_flags & TH_FIN)
428 			te->tqe_th->th_flags |= TH_FIN;
429 		m_cat(te->tqe_m, q->tqe_m);
430 		tp->encloseblk.rblk_end = tend_sack;
431 		/*
432 		 * When not reporting a duplicate segment, use
433 		 * the larger enclosing block as the SACK block.
434 		 */
435 		if (!(tp->t_flags & TF_DUPSEG))
436 			tp->reportblk.rblk_end = tend_sack;
437 		LIST_REMOVE(q, tqe_q);
438 		kfree(q, M_TSEGQ);
439 		atomic_add_int(&tcp_reass_qsize, -1);
440 	}
441 
442 	if (p == NULL) {
443 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
444 	} else {
445 		/* check if can coalesce with preceding segment */
446 		if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
447 			p->tqe_len += te->tqe_len;
448 			m_cat(p->tqe_m, te->tqe_m);
449 			tp->encloseblk.rblk_start = p->tqe_th->th_seq;
450 			/*
451 			 * When not reporting a duplicate segment, use
452 			 * the larger enclosing block as the SACK block.
453 			 */
454 			if (!(tp->t_flags & TF_DUPSEG))
455 				tp->reportblk.rblk_start = p->tqe_th->th_seq;
456 			kfree(te, M_TSEGQ);
457 			atomic_add_int(&tcp_reass_qsize, -1);
458 		} else {
459 			LIST_INSERT_AFTER(p, te, tqe_q);
460 		}
461 	}
462 
463 present:
464 	/*
465 	 * Present data to user, advancing rcv_nxt through
466 	 * completed sequence space.
467 	 */
468 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
469 		return (0);
470 	q = LIST_FIRST(&tp->t_segq);
471 	if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
472 		return (0);
473 	tp->rcv_nxt += q->tqe_len;
474 	if (!(tp->t_flags & TF_DUPSEG))	{
475 		/* no SACK block to report since ACK advanced */
476 		tp->reportblk.rblk_start = tp->reportblk.rblk_end;
477 	}
478 	/* no enclosing block to report since ACK advanced */
479 	tp->t_flags &= ~TF_ENCLOSESEG;
480 	flags = q->tqe_th->th_flags & TH_FIN;
481 	LIST_REMOVE(q, tqe_q);
482 	KASSERT(LIST_EMPTY(&tp->t_segq) ||
483 		LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
484 		("segment not coalesced"));
485 	if (so->so_state & SS_CANTRCVMORE) {
486 		m_freem(q->tqe_m);
487 	} else {
488 		lwkt_gettoken(&so->so_rcv.ssb_token);
489 		ssb_appendstream(&so->so_rcv, q->tqe_m);
490 		lwkt_reltoken(&so->so_rcv.ssb_token);
491 	}
492 	kfree(q, M_TSEGQ);
493 	atomic_add_int(&tcp_reass_qsize, -1);
494 	ND6_HINT(tp);
495 	sorwakeup(so);
496 	return (flags);
497 }
498 
499 /*
500  * TCP input routine, follows pages 65-76 of the
501  * protocol specification dated September, 1981 very closely.
502  */
503 #ifdef INET6
504 int
505 tcp6_input(struct mbuf **mp, int *offp, int proto)
506 {
507 	struct mbuf *m = *mp;
508 	struct in6_ifaddr *ia6;
509 
510 	IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
511 
512 	/*
513 	 * draft-itojun-ipv6-tcp-to-anycast
514 	 * better place to put this in?
515 	 */
516 	ia6 = ip6_getdstifaddr(m);
517 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
518 		struct ip6_hdr *ip6;
519 
520 		ip6 = mtod(m, struct ip6_hdr *);
521 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
522 			    offsetof(struct ip6_hdr, ip6_dst));
523 		return (IPPROTO_DONE);
524 	}
525 
526 	tcp_input(mp, offp, proto);
527 	return (IPPROTO_DONE);
528 }
529 #endif
530 
531 int
532 tcp_input(struct mbuf **mp, int *offp, int proto)
533 {
534 	int off0;
535 	struct tcphdr *th;
536 	struct ip *ip = NULL;
537 	struct ipovly *ipov;
538 	struct inpcb *inp = NULL;
539 	u_char *optp = NULL;
540 	int optlen = 0;
541 	int tlen, off;
542 	int len = 0;
543 	int drop_hdrlen;
544 	struct tcpcb *tp = NULL;
545 	int thflags;
546 	struct socket *so = NULL;
547 	int todrop, acked;
548 	boolean_t ourfinisacked, needoutput = FALSE;
549 	u_long tiwin;
550 	int recvwin;
551 	struct tcpopt to;		/* options in this segment */
552 	struct sockaddr_in *next_hop = NULL;
553 	int rstreason; /* For badport_bandlim accounting purposes */
554 	int cpu;
555 	struct ip6_hdr *ip6 = NULL;
556 	struct mbuf *m;
557 #ifdef INET6
558 	boolean_t isipv6;
559 #else
560 	const boolean_t isipv6 = FALSE;
561 #endif
562 #ifdef TCPDEBUG
563 	short ostate = 0;
564 #endif
565 
566 	off0 = *offp;
567 	m = *mp;
568 	*mp = NULL;
569 
570 	tcpstat.tcps_rcvtotal++;
571 
572 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
573 		struct m_tag *mtag;
574 
575 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
576 		KKASSERT(mtag != NULL);
577 		next_hop = m_tag_data(mtag);
578 	}
579 
580 #ifdef INET6
581 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
582 #endif
583 
584 	if (isipv6) {
585 		/* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
586 		ip6 = mtod(m, struct ip6_hdr *);
587 		tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
588 		if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
589 			tcpstat.tcps_rcvbadsum++;
590 			goto drop;
591 		}
592 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
593 
594 		/*
595 		 * Be proactive about unspecified IPv6 address in source.
596 		 * As we use all-zero to indicate unbounded/unconnected pcb,
597 		 * unspecified IPv6 address can be used to confuse us.
598 		 *
599 		 * Note that packets with unspecified IPv6 destination is
600 		 * already dropped in ip6_input.
601 		 */
602 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
603 			/* XXX stat */
604 			goto drop;
605 		}
606 	} else {
607 		/*
608 		 * Get IP and TCP header together in first mbuf.
609 		 * Note: IP leaves IP header in first mbuf.
610 		 */
611 		if (off0 > sizeof(struct ip)) {
612 			ip_stripoptions(m);
613 			off0 = sizeof(struct ip);
614 		}
615 		/* already checked and pulled up in ip_demux() */
616 		KASSERT(m->m_len >= sizeof(struct tcpiphdr),
617 		    ("TCP header not in one mbuf: m->m_len %d", m->m_len));
618 		ip = mtod(m, struct ip *);
619 		ipov = (struct ipovly *)ip;
620 		th = (struct tcphdr *)((caddr_t)ip + off0);
621 		tlen = ip->ip_len;
622 
623 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
624 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
625 				th->th_sum = m->m_pkthdr.csum_data;
626 			else
627 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
628 						ip->ip_dst.s_addr,
629 						htonl(m->m_pkthdr.csum_data +
630 							ip->ip_len +
631 							IPPROTO_TCP));
632 			th->th_sum ^= 0xffff;
633 		} else {
634 			/*
635 			 * Checksum extended TCP header and data.
636 			 */
637 			len = sizeof(struct ip) + tlen;
638 			bzero(ipov->ih_x1, sizeof ipov->ih_x1);
639 			ipov->ih_len = (u_short)tlen;
640 			ipov->ih_len = htons(ipov->ih_len);
641 			th->th_sum = in_cksum(m, len);
642 		}
643 		if (th->th_sum) {
644 			tcpstat.tcps_rcvbadsum++;
645 			goto drop;
646 		}
647 #ifdef INET6
648 		/* Re-initialization for later version check */
649 		ip->ip_v = IPVERSION;
650 #endif
651 	}
652 
653 	/*
654 	 * Check that TCP offset makes sense,
655 	 * pull out TCP options and adjust length.		XXX
656 	 */
657 	off = th->th_off << 2;
658 	/* already checked and pulled up in ip_demux() */
659 	KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
660 	    ("bad TCP data offset %d (tlen %d)", off, tlen));
661 	tlen -= off;	/* tlen is used instead of ti->ti_len */
662 	if (off > sizeof(struct tcphdr)) {
663 		if (isipv6) {
664 			IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
665 			ip6 = mtod(m, struct ip6_hdr *);
666 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
667 		} else {
668 			/* already pulled up in ip_demux() */
669 			KASSERT(m->m_len >= sizeof(struct ip) + off,
670 			    ("TCP header and options not in one mbuf: "
671 			     "m_len %d, off %d", m->m_len, off));
672 		}
673 		optlen = off - sizeof(struct tcphdr);
674 		optp = (u_char *)(th + 1);
675 	}
676 	thflags = th->th_flags;
677 
678 #ifdef TCP_DROP_SYNFIN
679 	/*
680 	 * If the drop_synfin option is enabled, drop all packets with
681 	 * both the SYN and FIN bits set. This prevents e.g. nmap from
682 	 * identifying the TCP/IP stack.
683 	 *
684 	 * This is a violation of the TCP specification.
685 	 */
686 	if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
687 		goto drop;
688 #endif
689 
690 	/*
691 	 * Convert TCP protocol specific fields to host format.
692 	 */
693 	th->th_seq = ntohl(th->th_seq);
694 	th->th_ack = ntohl(th->th_ack);
695 	th->th_win = ntohs(th->th_win);
696 	th->th_urp = ntohs(th->th_urp);
697 
698 	/*
699 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
700 	 * until after ip6_savecontrol() is called and before other functions
701 	 * which don't want those proto headers.
702 	 * Because ip6_savecontrol() is going to parse the mbuf to
703 	 * search for data to be passed up to user-land, it wants mbuf
704 	 * parameters to be unchanged.
705 	 * XXX: the call of ip6_savecontrol() has been obsoleted based on
706 	 * latest version of the advanced API (20020110).
707 	 */
708 	drop_hdrlen = off0 + off;
709 
710 	/*
711 	 * Locate pcb for segment.
712 	 */
713 findpcb:
714 	/* IPFIREWALL_FORWARD section */
715 	if (next_hop != NULL && !isipv6) {  /* IPv6 support is not there yet */
716 		/*
717 		 * Transparently forwarded. Pretend to be the destination.
718 		 * already got one like this?
719 		 */
720 		cpu = mycpu->gd_cpuid;
721 		inp = in_pcblookup_hash(&tcbinfo[cpu],
722 					ip->ip_src, th->th_sport,
723 					ip->ip_dst, th->th_dport,
724 					0, m->m_pkthdr.rcvif);
725 		if (!inp) {
726 			/*
727 			 * It's new.  Try to find the ambushing socket.
728 			 */
729 
730 			/*
731 			 * The rest of the ipfw code stores the port in
732 			 * host order.  XXX
733 			 * (The IP address is still in network order.)
734 			 */
735 			in_port_t dport = next_hop->sin_port ?
736 						htons(next_hop->sin_port) :
737 						th->th_dport;
738 
739 			cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
740 					  next_hop->sin_addr.s_addr, dport);
741 			inp = in_pcblookup_hash(&tcbinfo[cpu],
742 						ip->ip_src, th->th_sport,
743 						next_hop->sin_addr, dport,
744 						1, m->m_pkthdr.rcvif);
745 		}
746 	} else {
747 		if (isipv6) {
748 			inp = in6_pcblookup_hash(&tcbinfo[0],
749 						 &ip6->ip6_src, th->th_sport,
750 						 &ip6->ip6_dst, th->th_dport,
751 						 1, m->m_pkthdr.rcvif);
752 		} else {
753 			cpu = mycpu->gd_cpuid;
754 			inp = in_pcblookup_hash(&tcbinfo[cpu],
755 						ip->ip_src, th->th_sport,
756 						ip->ip_dst, th->th_dport,
757 						1, m->m_pkthdr.rcvif);
758 		}
759 	}
760 
761 	/*
762 	 * If the state is CLOSED (i.e., TCB does not exist) then
763 	 * all data in the incoming segment is discarded.
764 	 * If the TCB exists but is in CLOSED state, it is embryonic,
765 	 * but should either do a listen or a connect soon.
766 	 */
767 	if (inp == NULL) {
768 		if (log_in_vain) {
769 #ifdef INET6
770 			char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
771 #else
772 			char dbuf[sizeof "aaa.bbb.ccc.ddd"];
773 			char sbuf[sizeof "aaa.bbb.ccc.ddd"];
774 #endif
775 			if (isipv6) {
776 				strcpy(dbuf, "[");
777 				strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
778 				strcat(dbuf, "]");
779 				strcpy(sbuf, "[");
780 				strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
781 				strcat(sbuf, "]");
782 			} else {
783 				strcpy(dbuf, inet_ntoa(ip->ip_dst));
784 				strcpy(sbuf, inet_ntoa(ip->ip_src));
785 			}
786 			switch (log_in_vain) {
787 			case 1:
788 				if (!(thflags & TH_SYN))
789 					break;
790 			case 2:
791 				log(LOG_INFO,
792 				    "Connection attempt to TCP %s:%d "
793 				    "from %s:%d flags:0x%02x\n",
794 				    dbuf, ntohs(th->th_dport), sbuf,
795 				    ntohs(th->th_sport), thflags);
796 				break;
797 			default:
798 				break;
799 			}
800 		}
801 		if (blackhole) {
802 			switch (blackhole) {
803 			case 1:
804 				if (thflags & TH_SYN)
805 					goto drop;
806 				break;
807 			case 2:
808 				goto drop;
809 			default:
810 				goto drop;
811 			}
812 		}
813 		rstreason = BANDLIM_RST_CLOSEDPORT;
814 		goto dropwithreset;
815 	}
816 
817 #ifdef IPSEC
818 	if (isipv6) {
819 		if (ipsec6_in_reject_so(m, inp->inp_socket)) {
820 			ipsec6stat.in_polvio++;
821 			goto drop;
822 		}
823 	} else {
824 		if (ipsec4_in_reject_so(m, inp->inp_socket)) {
825 			ipsecstat.in_polvio++;
826 			goto drop;
827 		}
828 	}
829 #endif
830 #ifdef FAST_IPSEC
831 	if (isipv6) {
832 		if (ipsec6_in_reject(m, inp))
833 			goto drop;
834 	} else {
835 		if (ipsec4_in_reject(m, inp))
836 			goto drop;
837 	}
838 #endif
839 	/* Check the minimum TTL for socket. */
840 #ifdef INET6
841 	if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
842 		goto drop;
843 #endif
844 
845 	tp = intotcpcb(inp);
846 	if (tp == NULL) {
847 		rstreason = BANDLIM_RST_CLOSEDPORT;
848 		goto dropwithreset;
849 	}
850 	if (tp->t_state <= TCPS_CLOSED)
851 		goto drop;
852 
853 	so = inp->inp_socket;
854 
855 #ifdef TCPDEBUG
856 	if (so->so_options & SO_DEBUG) {
857 		ostate = tp->t_state;
858 		if (isipv6)
859 			bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
860 		else
861 			bcopy(ip, tcp_saveipgen, sizeof(*ip));
862 		tcp_savetcp = *th;
863 	}
864 #endif
865 
866 	bzero(&to, sizeof to);
867 
868 	if (so->so_options & SO_ACCEPTCONN) {
869 		struct in_conninfo inc;
870 
871 #ifdef INET6
872 		inc.inc_isipv6 = (isipv6 == TRUE);
873 #endif
874 		if (isipv6) {
875 			inc.inc6_faddr = ip6->ip6_src;
876 			inc.inc6_laddr = ip6->ip6_dst;
877 			inc.inc6_route.ro_rt = NULL;		/* XXX */
878 		} else {
879 			inc.inc_faddr = ip->ip_src;
880 			inc.inc_laddr = ip->ip_dst;
881 			inc.inc_route.ro_rt = NULL;		/* XXX */
882 		}
883 		inc.inc_fport = th->th_sport;
884 		inc.inc_lport = th->th_dport;
885 
886 		/*
887 		 * If the state is LISTEN then ignore segment if it contains
888 		 * a RST.  If the segment contains an ACK then it is bad and
889 		 * send a RST.  If it does not contain a SYN then it is not
890 		 * interesting; drop it.
891 		 *
892 		 * If the state is SYN_RECEIVED (syncache) and seg contains
893 		 * an ACK, but not for our SYN/ACK, send a RST.  If the seg
894 		 * contains a RST, check the sequence number to see if it
895 		 * is a valid reset segment.
896 		 */
897 		if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
898 			if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
899 				if (!syncache_expand(&inc, th, &so, m)) {
900 					/*
901 					 * No syncache entry, or ACK was not
902 					 * for our SYN/ACK.  Send a RST.
903 					 */
904 					tcpstat.tcps_badsyn++;
905 					rstreason = BANDLIM_RST_OPENPORT;
906 					goto dropwithreset;
907 				}
908 
909 				/*
910 				 * Could not complete 3-way handshake,
911 				 * connection is being closed down, and
912 				 * syncache will free mbuf.
913 				 */
914 				if (so == NULL)
915 					return(IPPROTO_DONE);
916 
917 				/*
918 				 * We must be in the correct protocol thread
919 				 * for this connection.
920 				 */
921 				KKASSERT(so->so_port == &curthread->td_msgport);
922 
923 				/*
924 				 * Socket is created in state SYN_RECEIVED.
925 				 * Continue processing segment.
926 				 */
927 				inp = so->so_pcb;
928 				tp = intotcpcb(inp);
929 				/*
930 				 * This is what would have happened in
931 				 * tcp_output() when the SYN,ACK was sent.
932 				 */
933 				tp->snd_up = tp->snd_una;
934 				tp->snd_max = tp->snd_nxt = tp->iss + 1;
935 				tp->last_ack_sent = tp->rcv_nxt;
936 
937 				goto after_listen;
938 			}
939 			if (thflags & TH_RST) {
940 				syncache_chkrst(&inc, th);
941 				goto drop;
942 			}
943 			if (thflags & TH_ACK) {
944 				syncache_badack(&inc);
945 				tcpstat.tcps_badsyn++;
946 				rstreason = BANDLIM_RST_OPENPORT;
947 				goto dropwithreset;
948 			}
949 			goto drop;
950 		}
951 
952 		/*
953 		 * Segment's flags are (SYN) or (SYN | FIN).
954 		 */
955 #ifdef INET6
956 		/*
957 		 * If deprecated address is forbidden,
958 		 * we do not accept SYN to deprecated interface
959 		 * address to prevent any new inbound connection from
960 		 * getting established.
961 		 * When we do not accept SYN, we send a TCP RST,
962 		 * with deprecated source address (instead of dropping
963 		 * it).  We compromise it as it is much better for peer
964 		 * to send a RST, and RST will be the final packet
965 		 * for the exchange.
966 		 *
967 		 * If we do not forbid deprecated addresses, we accept
968 		 * the SYN packet.  RFC2462 does not suggest dropping
969 		 * SYN in this case.
970 		 * If we decipher RFC2462 5.5.4, it says like this:
971 		 * 1. use of deprecated addr with existing
972 		 *    communication is okay - "SHOULD continue to be
973 		 *    used"
974 		 * 2. use of it with new communication:
975 		 *   (2a) "SHOULD NOT be used if alternate address
976 		 *	  with sufficient scope is available"
977 		 *   (2b) nothing mentioned otherwise.
978 		 * Here we fall into (2b) case as we have no choice in
979 		 * our source address selection - we must obey the peer.
980 		 *
981 		 * The wording in RFC2462 is confusing, and there are
982 		 * multiple description text for deprecated address
983 		 * handling - worse, they are not exactly the same.
984 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
985 		 */
986 		if (isipv6 && !ip6_use_deprecated) {
987 			struct in6_ifaddr *ia6;
988 
989 			if ((ia6 = ip6_getdstifaddr(m)) &&
990 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
991 				tp = NULL;
992 				rstreason = BANDLIM_RST_OPENPORT;
993 				goto dropwithreset;
994 			}
995 		}
996 #endif
997 		/*
998 		 * If it is from this socket, drop it, it must be forged.
999 		 * Don't bother responding if the destination was a broadcast.
1000 		 */
1001 		if (th->th_dport == th->th_sport) {
1002 			if (isipv6) {
1003 				if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1004 						       &ip6->ip6_src))
1005 					goto drop;
1006 			} else {
1007 				if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1008 					goto drop;
1009 			}
1010 		}
1011 		/*
1012 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1013 		 *
1014 		 * Note that it is quite possible to receive unicast
1015 		 * link-layer packets with a broadcast IP address. Use
1016 		 * in_broadcast() to find them.
1017 		 */
1018 		if (m->m_flags & (M_BCAST | M_MCAST))
1019 			goto drop;
1020 		if (isipv6) {
1021 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1022 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1023 				goto drop;
1024 		} else {
1025 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1026 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1027 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1028 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1029 				goto drop;
1030 		}
1031 		/*
1032 		 * SYN appears to be valid; create compressed TCP state
1033 		 * for syncache, or perform t/tcp connection.
1034 		 */
1035 		if (so->so_qlen <= so->so_qlimit) {
1036 			tcp_dooptions(&to, optp, optlen, TRUE);
1037 			if (!syncache_add(&inc, &to, th, so, m))
1038 				goto drop;
1039 
1040 			/*
1041 			 * Entry added to syncache, mbuf used to
1042 			 * send SYN,ACK packet.
1043 			 */
1044 			return(IPPROTO_DONE);
1045 		}
1046 		goto drop;
1047 	}
1048 
1049 after_listen:
1050 	/*
1051 	 * Should not happen - syncache should pick up these connections.
1052 	 *
1053 	 * Once we are past handling listen sockets we must be in the
1054 	 * correct protocol processing thread.
1055 	 */
1056 	KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1057 	KKASSERT(so->so_port == &curthread->td_msgport);
1058 
1059 	/* Unscale the window into a 32-bit value. */
1060 	if (!(thflags & TH_SYN))
1061 		tiwin = th->th_win << tp->snd_scale;
1062 	else
1063 		tiwin = th->th_win;
1064 
1065 	/*
1066 	 * This is the second part of the MSS DoS prevention code (after
1067 	 * minmss on the sending side) and it deals with too many too small
1068 	 * tcp packets in a too short timeframe (1 second).
1069 	 *
1070 	 * XXX Removed.  This code was crap.  It does not scale to network
1071 	 *     speed, and default values break NFS.  Gone.
1072 	 */
1073 	/* REMOVED */
1074 
1075 	/*
1076 	 * Segment received on connection.
1077 	 *
1078 	 * Reset idle time and keep-alive timer.  Don't waste time if less
1079 	 * then a second has elapsed.
1080 	 */
1081 	if ((int)(ticks - tp->t_rcvtime) > hz)
1082 		tcp_timer_keep_activity(tp, thflags);
1083 
1084 	/*
1085 	 * Process options.
1086 	 * XXX this is tradtitional behavior, may need to be cleaned up.
1087 	 */
1088 	tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1089 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1090 		if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1091 			tp->t_flags |= TF_RCVD_SCALE;
1092 			tp->snd_scale = to.to_requested_s_scale;
1093 		}
1094 
1095 		/*
1096 		 * Initial send window; will be updated upon next ACK
1097 		 */
1098 		tp->snd_wnd = th->th_win;
1099 
1100 		if (to.to_flags & TOF_TS) {
1101 			tp->t_flags |= TF_RCVD_TSTMP;
1102 			tp->ts_recent = to.to_tsval;
1103 			tp->ts_recent_age = ticks;
1104 		}
1105 		if (!(to.to_flags & TOF_MSS))
1106 			to.to_mss = 0;
1107 		tcp_mss(tp, to.to_mss);
1108 		/*
1109 		 * Only set the TF_SACK_PERMITTED per-connection flag
1110 		 * if we got a SACK_PERMITTED option from the other side
1111 		 * and the global tcp_do_sack variable is true.
1112 		 */
1113 		if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1114 			tp->t_flags |= TF_SACK_PERMITTED;
1115 	}
1116 
1117 	/*
1118 	 * Header prediction: check for the two common cases
1119 	 * of a uni-directional data xfer.  If the packet has
1120 	 * no control flags, is in-sequence, the window didn't
1121 	 * change and we're not retransmitting, it's a
1122 	 * candidate.  If the length is zero and the ack moved
1123 	 * forward, we're the sender side of the xfer.  Just
1124 	 * free the data acked & wake any higher level process
1125 	 * that was blocked waiting for space.  If the length
1126 	 * is non-zero and the ack didn't move, we're the
1127 	 * receiver side.  If we're getting packets in-order
1128 	 * (the reassembly queue is empty), add the data to
1129 	 * the socket buffer and note that we need a delayed ack.
1130 	 * Make sure that the hidden state-flags are also off.
1131 	 * Since we check for TCPS_ESTABLISHED above, it can only
1132 	 * be TH_NEEDSYN.
1133 	 */
1134 	if (tp->t_state == TCPS_ESTABLISHED &&
1135 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1136 	    !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1137 	    (!(to.to_flags & TOF_TS) ||
1138 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1139 	    th->th_seq == tp->rcv_nxt &&
1140 	    tp->snd_nxt == tp->snd_max) {
1141 
1142 		/*
1143 		 * If last ACK falls within this segment's sequence numbers,
1144 		 * record the timestamp.
1145 		 * NOTE that the test is modified according to the latest
1146 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1147 		 */
1148 		if ((to.to_flags & TOF_TS) &&
1149 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1150 			tp->ts_recent_age = ticks;
1151 			tp->ts_recent = to.to_tsval;
1152 		}
1153 
1154 		if (tlen == 0) {
1155 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1156 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1157 			    tp->snd_cwnd >= tp->snd_wnd &&
1158 			    !IN_FASTRECOVERY(tp)) {
1159 				/*
1160 				 * This is a pure ack for outstanding data.
1161 				 */
1162 				++tcpstat.tcps_predack;
1163 				/*
1164 				 * "bad retransmit" recovery
1165 				 *
1166 				 * If Eifel detection applies, then
1167 				 * it is deterministic, so use it
1168 				 * unconditionally over the old heuristic.
1169 				 * Otherwise, fall back to the old heuristic.
1170 				 */
1171 				if (tcp_do_eifel_detect &&
1172 				    (to.to_flags & TOF_TS) && to.to_tsecr &&
1173 				    (tp->t_flags & TF_FIRSTACCACK)) {
1174 					/* Eifel detection applicable. */
1175 					if (to.to_tsecr < tp->t_rexmtTS) {
1176 						tcp_revert_congestion_state(tp);
1177 						++tcpstat.tcps_eifeldetected;
1178 						if (tp->t_rxtshift != 1 ||
1179 						    ticks >= tp->t_badrxtwin)
1180 							++tcpstat.tcps_rttcantdetect;
1181 					}
1182 				} else if (tp->t_rxtshift == 1 &&
1183 					   ticks < tp->t_badrxtwin) {
1184 					tcp_revert_congestion_state(tp);
1185 					++tcpstat.tcps_rttdetected;
1186 				}
1187 				tp->t_flags &= ~(TF_FIRSTACCACK |
1188 						 TF_FASTREXMT | TF_EARLYREXMT);
1189 				/*
1190 				 * Recalculate the retransmit timer / rtt.
1191 				 *
1192 				 * Some machines (certain windows boxes)
1193 				 * send broken timestamp replies during the
1194 				 * SYN+ACK phase, ignore timestamps of 0.
1195 				 */
1196 				if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1197 					tcp_xmit_timer(tp,
1198 						       ticks - to.to_tsecr + 1);
1199 				} else if (tp->t_rtttime &&
1200 					   SEQ_GT(th->th_ack, tp->t_rtseq)) {
1201 					tcp_xmit_timer(tp,
1202 						       ticks - tp->t_rtttime);
1203 				}
1204 				tcp_xmit_bandwidth_limit(tp, th->th_ack);
1205 				acked = th->th_ack - tp->snd_una;
1206 				tcpstat.tcps_rcvackpack++;
1207 				tcpstat.tcps_rcvackbyte += acked;
1208 				sbdrop(&so->so_snd.sb, acked);
1209 				tp->snd_recover = th->th_ack - 1;
1210 				tp->snd_una = th->th_ack;
1211 				tp->t_dupacks = 0;
1212 				/*
1213 				 * Update window information.
1214 				 */
1215 				if (tiwin != tp->snd_wnd &&
1216 				    acceptable_window_update(tp, th, tiwin)) {
1217 					/* keep track of pure window updates */
1218 					if (tp->snd_wl2 == th->th_ack &&
1219 					    tiwin > tp->snd_wnd)
1220 						tcpstat.tcps_rcvwinupd++;
1221 					tp->snd_wnd = tiwin;
1222 					tp->snd_wl1 = th->th_seq;
1223 					tp->snd_wl2 = th->th_ack;
1224 					if (tp->snd_wnd > tp->max_sndwnd)
1225 						tp->max_sndwnd = tp->snd_wnd;
1226 				}
1227 				m_freem(m);
1228 				ND6_HINT(tp); /* some progress has been done */
1229 				/*
1230 				 * If all outstanding data are acked, stop
1231 				 * retransmit timer, otherwise restart timer
1232 				 * using current (possibly backed-off) value.
1233 				 * If process is waiting for space,
1234 				 * wakeup/selwakeup/signal.  If data
1235 				 * are ready to send, let tcp_output
1236 				 * decide between more output or persist.
1237 				 */
1238 				if (tp->snd_una == tp->snd_max) {
1239 					tcp_callout_stop(tp, tp->tt_rexmt);
1240 				} else if (!tcp_callout_active(tp,
1241 					    tp->tt_persist)) {
1242 					tcp_callout_reset(tp, tp->tt_rexmt,
1243 					    tp->t_rxtcur, tcp_timer_rexmt);
1244 				}
1245 				sowwakeup(so);
1246 				if (so->so_snd.ssb_cc > 0)
1247 					tcp_output(tp);
1248 				return(IPPROTO_DONE);
1249 			}
1250 		} else if (tiwin == tp->snd_wnd &&
1251 		    th->th_ack == tp->snd_una &&
1252 		    LIST_EMPTY(&tp->t_segq) &&
1253 		    tlen <= ssb_space(&so->so_rcv)) {
1254 			u_long newsize = 0;	/* automatic sockbuf scaling */
1255 			/*
1256 			 * This is a pure, in-sequence data packet
1257 			 * with nothing on the reassembly queue and
1258 			 * we have enough buffer space to take it.
1259 			 */
1260 			++tcpstat.tcps_preddat;
1261 			tp->rcv_nxt += tlen;
1262 			tcpstat.tcps_rcvpack++;
1263 			tcpstat.tcps_rcvbyte += tlen;
1264 			ND6_HINT(tp);	/* some progress has been done */
1265 		/*
1266 		 * Automatic sizing of receive socket buffer.  Often the send
1267 		 * buffer size is not optimally adjusted to the actual network
1268 		 * conditions at hand (delay bandwidth product).  Setting the
1269 		 * buffer size too small limits throughput on links with high
1270 		 * bandwidth and high delay (eg. trans-continental/oceanic links).
1271 		 *
1272 		 * On the receive side the socket buffer memory is only rarely
1273 		 * used to any significant extent.  This allows us to be much
1274 		 * more aggressive in scaling the receive socket buffer.  For
1275 		 * the case that the buffer space is actually used to a large
1276 		 * extent and we run out of kernel memory we can simply drop
1277 		 * the new segments; TCP on the sender will just retransmit it
1278 		 * later.  Setting the buffer size too big may only consume too
1279 		 * much kernel memory if the application doesn't read() from
1280 		 * the socket or packet loss or reordering makes use of the
1281 		 * reassembly queue.
1282 		 *
1283 		 * The criteria to step up the receive buffer one notch are:
1284 		 *  1. the number of bytes received during the time it takes
1285 		 *     one timestamp to be reflected back to us (the RTT);
1286 		 *  2. received bytes per RTT is within seven eighth of the
1287 		 *     current socket buffer size;
1288 		 *  3. receive buffer size has not hit maximal automatic size;
1289 		 *
1290 		 * This algorithm does one step per RTT at most and only if
1291 		 * we receive a bulk stream w/o packet losses or reorderings.
1292 		 * Shrinking the buffer during idle times is not necessary as
1293 		 * it doesn't consume any memory when idle.
1294 		 *
1295 		 * TODO: Only step up if the application is actually serving
1296 		 * the buffer to better manage the socket buffer resources.
1297 		 */
1298 			if (tcp_do_autorcvbuf &&
1299 			    to.to_tsecr &&
1300 			    (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1301 				if (to.to_tsecr > tp->rfbuf_ts &&
1302 				    to.to_tsecr - tp->rfbuf_ts < hz) {
1303 					if (tp->rfbuf_cnt >
1304 					    (so->so_rcv.ssb_hiwat / 8 * 7) &&
1305 					    so->so_rcv.ssb_hiwat <
1306 					    tcp_autorcvbuf_max) {
1307 						newsize =
1308 						    ulmin(so->so_rcv.ssb_hiwat +
1309 							  tcp_autorcvbuf_inc,
1310 							  tcp_autorcvbuf_max);
1311 					}
1312 					/* Start over with next RTT. */
1313 					tp->rfbuf_ts = 0;
1314 					tp->rfbuf_cnt = 0;
1315 				} else
1316 					tp->rfbuf_cnt += tlen;	/* add up */
1317 			}
1318 			/*
1319 			 * Add data to socket buffer.
1320 			 */
1321 			if (so->so_state & SS_CANTRCVMORE) {
1322 				m_freem(m);
1323 			} else {
1324 				/*
1325 				 * Set new socket buffer size, give up when
1326 				 * limit is reached.
1327 				 *
1328 				 * Adjusting the size can mess up ACK
1329 				 * sequencing when pure window updates are
1330 				 * being avoided (which is the default),
1331 				 * so force an ack.
1332 				 */
1333 				lwkt_gettoken(&so->so_rcv.ssb_token);
1334 				if (newsize) {
1335 					tp->t_flags |= TF_RXRESIZED;
1336 					if (!ssb_reserve(&so->so_rcv, newsize,
1337 							 so, NULL)) {
1338 						atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1339 					}
1340 					if (newsize >=
1341 					    (TCP_MAXWIN << tp->rcv_scale)) {
1342 						atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1343 					}
1344 				}
1345 				m_adj(m, drop_hdrlen); /* delayed header drop */
1346 				ssb_appendstream(&so->so_rcv, m);
1347 				lwkt_reltoken(&so->so_rcv.ssb_token);
1348 			}
1349 			sorwakeup(so);
1350 			/*
1351 			 * This code is responsible for most of the ACKs
1352 			 * the TCP stack sends back after receiving a data
1353 			 * packet.  Note that the DELAY_ACK check fails if
1354 			 * the delack timer is already running, which results
1355 			 * in an ack being sent every other packet (which is
1356 			 * what we want).
1357 			 *
1358 			 * We then further aggregate acks by not actually
1359 			 * sending one until the protocol thread has completed
1360 			 * processing the current backlog of packets.  This
1361 			 * does not delay the ack any further, but allows us
1362 			 * to take advantage of the packet aggregation that
1363 			 * high speed NICs do (usually blocks of 8-10 packets)
1364 			 * to send a single ack rather then four or five acks,
1365 			 * greatly reducing the ack rate, the return channel
1366 			 * bandwidth, and the protocol overhead on both ends.
1367 			 *
1368 			 * Since this also has the effect of slowing down
1369 			 * the exponential slow-start ramp-up, systems with
1370 			 * very large bandwidth-delay products might want
1371 			 * to turn the feature off.
1372 			 */
1373 			if (DELAY_ACK(tp)) {
1374 				tcp_callout_reset(tp, tp->tt_delack,
1375 				    tcp_delacktime, tcp_timer_delack);
1376 			} else if (tcp_aggregate_acks) {
1377 				tp->t_flags |= TF_ACKNOW;
1378 				if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1379 					tp->t_flags |= TF_ONOUTPUTQ;
1380 					tp->tt_cpu = mycpu->gd_cpuid;
1381 					TAILQ_INSERT_TAIL(
1382 					    &tcpcbackq[tp->tt_cpu],
1383 					    tp, t_outputq);
1384 				}
1385 			} else {
1386 				tp->t_flags |= TF_ACKNOW;
1387 				tcp_output(tp);
1388 			}
1389 			return(IPPROTO_DONE);
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * Calculate amount of space in receive window,
1395 	 * and then do TCP input processing.
1396 	 * Receive window is amount of space in rcv queue,
1397 	 * but not less than advertised window.
1398 	 */
1399 	recvwin = ssb_space(&so->so_rcv);
1400 	if (recvwin < 0)
1401 		recvwin = 0;
1402 	tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1403 
1404 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1405 	tp->rfbuf_ts = 0;
1406 	tp->rfbuf_cnt = 0;
1407 
1408 	switch (tp->t_state) {
1409 	/*
1410 	 * If the state is SYN_RECEIVED:
1411 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1412 	 */
1413 	case TCPS_SYN_RECEIVED:
1414 		if ((thflags & TH_ACK) &&
1415 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1416 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1417 				rstreason = BANDLIM_RST_OPENPORT;
1418 				goto dropwithreset;
1419 		}
1420 		break;
1421 
1422 	/*
1423 	 * If the state is SYN_SENT:
1424 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1425 	 *	if seg contains a RST, then drop the connection.
1426 	 *	if seg does not contain SYN, then drop it.
1427 	 * Otherwise this is an acceptable SYN segment
1428 	 *	initialize tp->rcv_nxt and tp->irs
1429 	 *	if seg contains ack then advance tp->snd_una
1430 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1431 	 *	arrange for segment to be acked (eventually)
1432 	 *	continue processing rest of data/controls, beginning with URG
1433 	 */
1434 	case TCPS_SYN_SENT:
1435 		if ((thflags & TH_ACK) &&
1436 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1437 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1438 			rstreason = BANDLIM_UNLIMITED;
1439 			goto dropwithreset;
1440 		}
1441 		if (thflags & TH_RST) {
1442 			if (thflags & TH_ACK)
1443 				tp = tcp_drop(tp, ECONNREFUSED);
1444 			goto drop;
1445 		}
1446 		if (!(thflags & TH_SYN))
1447 			goto drop;
1448 
1449 		tp->irs = th->th_seq;
1450 		tcp_rcvseqinit(tp);
1451 		if (thflags & TH_ACK) {
1452 			/* Our SYN was acked. */
1453 			tcpstat.tcps_connects++;
1454 			soisconnected(so);
1455 			/* Do window scaling on this connection? */
1456 			if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1457 			    (TF_RCVD_SCALE | TF_REQ_SCALE))
1458 				tp->rcv_scale = tp->request_r_scale;
1459 			tp->rcv_adv += tp->rcv_wnd;
1460 			tp->snd_una++;		/* SYN is acked */
1461 			tcp_callout_stop(tp, tp->tt_rexmt);
1462 			/*
1463 			 * If there's data, delay ACK; if there's also a FIN
1464 			 * ACKNOW will be turned on later.
1465 			 */
1466 			if (DELAY_ACK(tp) && tlen != 0) {
1467 				tcp_callout_reset(tp, tp->tt_delack,
1468 				    tcp_delacktime, tcp_timer_delack);
1469 			} else {
1470 				tp->t_flags |= TF_ACKNOW;
1471 			}
1472 			/*
1473 			 * Received <SYN,ACK> in SYN_SENT[*] state.
1474 			 * Transitions:
1475 			 *	SYN_SENT  --> ESTABLISHED
1476 			 *	SYN_SENT* --> FIN_WAIT_1
1477 			 */
1478 			tp->t_starttime = ticks;
1479 			if (tp->t_flags & TF_NEEDFIN) {
1480 				tp->t_state = TCPS_FIN_WAIT_1;
1481 				tp->t_flags &= ~TF_NEEDFIN;
1482 				thflags &= ~TH_SYN;
1483 			} else {
1484 				tcp_established(tp);
1485 			}
1486 		} else {
1487 			/*
1488 			 * Received initial SYN in SYN-SENT[*] state =>
1489 			 * simultaneous open.
1490 			 * Do 3-way handshake:
1491 			 *	  SYN-SENT -> SYN-RECEIVED
1492 			 *	  SYN-SENT* -> SYN-RECEIVED*
1493 			 */
1494 			tp->t_flags |= TF_ACKNOW;
1495 			tcp_callout_stop(tp, tp->tt_rexmt);
1496 			tp->t_state = TCPS_SYN_RECEIVED;
1497 		}
1498 
1499 		/*
1500 		 * Advance th->th_seq to correspond to first data byte.
1501 		 * If data, trim to stay within window,
1502 		 * dropping FIN if necessary.
1503 		 */
1504 		th->th_seq++;
1505 		if (tlen > tp->rcv_wnd) {
1506 			todrop = tlen - tp->rcv_wnd;
1507 			m_adj(m, -todrop);
1508 			tlen = tp->rcv_wnd;
1509 			thflags &= ~TH_FIN;
1510 			tcpstat.tcps_rcvpackafterwin++;
1511 			tcpstat.tcps_rcvbyteafterwin += todrop;
1512 		}
1513 		tp->snd_wl1 = th->th_seq - 1;
1514 		tp->rcv_up = th->th_seq;
1515 		/*
1516 		 * Client side of transaction: already sent SYN and data.
1517 		 * If the remote host used T/TCP to validate the SYN,
1518 		 * our data will be ACK'd; if so, enter normal data segment
1519 		 * processing in the middle of step 5, ack processing.
1520 		 * Otherwise, goto step 6.
1521 		 */
1522 		if (thflags & TH_ACK)
1523 			goto process_ACK;
1524 
1525 		goto step6;
1526 
1527 	/*
1528 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1529 	 *	do normal processing (we no longer bother with T/TCP).
1530 	 */
1531 	case TCPS_LAST_ACK:
1532 	case TCPS_CLOSING:
1533 	case TCPS_TIME_WAIT:
1534 		break;  /* continue normal processing */
1535 	}
1536 
1537 	/*
1538 	 * States other than LISTEN or SYN_SENT.
1539 	 * First check the RST flag and sequence number since reset segments
1540 	 * are exempt from the timestamp and connection count tests.  This
1541 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1542 	 * below which allowed reset segments in half the sequence space
1543 	 * to fall though and be processed (which gives forged reset
1544 	 * segments with a random sequence number a 50 percent chance of
1545 	 * killing a connection).
1546 	 * Then check timestamp, if present.
1547 	 * Then check the connection count, if present.
1548 	 * Then check that at least some bytes of segment are within
1549 	 * receive window.  If segment begins before rcv_nxt,
1550 	 * drop leading data (and SYN); if nothing left, just ack.
1551 	 *
1552 	 *
1553 	 * If the RST bit is set, check the sequence number to see
1554 	 * if this is a valid reset segment.
1555 	 * RFC 793 page 37:
1556 	 *   In all states except SYN-SENT, all reset (RST) segments
1557 	 *   are validated by checking their SEQ-fields.  A reset is
1558 	 *   valid if its sequence number is in the window.
1559 	 * Note: this does not take into account delayed ACKs, so
1560 	 *   we should test against last_ack_sent instead of rcv_nxt.
1561 	 *   The sequence number in the reset segment is normally an
1562 	 *   echo of our outgoing acknowledgement numbers, but some hosts
1563 	 *   send a reset with the sequence number at the rightmost edge
1564 	 *   of our receive window, and we have to handle this case.
1565 	 * If we have multiple segments in flight, the intial reset
1566 	 * segment sequence numbers will be to the left of last_ack_sent,
1567 	 * but they will eventually catch up.
1568 	 * In any case, it never made sense to trim reset segments to
1569 	 * fit the receive window since RFC 1122 says:
1570 	 *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1571 	 *
1572 	 *    A TCP SHOULD allow a received RST segment to include data.
1573 	 *
1574 	 *    DISCUSSION
1575 	 *	   It has been suggested that a RST segment could contain
1576 	 *	   ASCII text that encoded and explained the cause of the
1577 	 *	   RST.  No standard has yet been established for such
1578 	 *	   data.
1579 	 *
1580 	 * If the reset segment passes the sequence number test examine
1581 	 * the state:
1582 	 *    SYN_RECEIVED STATE:
1583 	 *	If passive open, return to LISTEN state.
1584 	 *	If active open, inform user that connection was refused.
1585 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1586 	 *	Inform user that connection was reset, and close tcb.
1587 	 *    CLOSING, LAST_ACK STATES:
1588 	 *	Close the tcb.
1589 	 *    TIME_WAIT STATE:
1590 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
1591 	 *	RFC 1337.
1592 	 */
1593 	if (thflags & TH_RST) {
1594 		if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1595 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1596 			switch (tp->t_state) {
1597 
1598 			case TCPS_SYN_RECEIVED:
1599 				so->so_error = ECONNREFUSED;
1600 				goto close;
1601 
1602 			case TCPS_ESTABLISHED:
1603 			case TCPS_FIN_WAIT_1:
1604 			case TCPS_FIN_WAIT_2:
1605 			case TCPS_CLOSE_WAIT:
1606 				so->so_error = ECONNRESET;
1607 			close:
1608 				tp->t_state = TCPS_CLOSED;
1609 				tcpstat.tcps_drops++;
1610 				tp = tcp_close(tp);
1611 				break;
1612 
1613 			case TCPS_CLOSING:
1614 			case TCPS_LAST_ACK:
1615 				tp = tcp_close(tp);
1616 				break;
1617 
1618 			case TCPS_TIME_WAIT:
1619 				break;
1620 			}
1621 		}
1622 		goto drop;
1623 	}
1624 
1625 	/*
1626 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1627 	 * and it's less than ts_recent, drop it.
1628 	 */
1629 	if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1630 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1631 
1632 		/* Check to see if ts_recent is over 24 days old.  */
1633 		if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1634 			/*
1635 			 * Invalidate ts_recent.  If this segment updates
1636 			 * ts_recent, the age will be reset later and ts_recent
1637 			 * will get a valid value.  If it does not, setting
1638 			 * ts_recent to zero will at least satisfy the
1639 			 * requirement that zero be placed in the timestamp
1640 			 * echo reply when ts_recent isn't valid.  The
1641 			 * age isn't reset until we get a valid ts_recent
1642 			 * because we don't want out-of-order segments to be
1643 			 * dropped when ts_recent is old.
1644 			 */
1645 			tp->ts_recent = 0;
1646 		} else {
1647 			tcpstat.tcps_rcvduppack++;
1648 			tcpstat.tcps_rcvdupbyte += tlen;
1649 			tcpstat.tcps_pawsdrop++;
1650 			if (tlen)
1651 				goto dropafterack;
1652 			goto drop;
1653 		}
1654 	}
1655 
1656 	/*
1657 	 * In the SYN-RECEIVED state, validate that the packet belongs to
1658 	 * this connection before trimming the data to fit the receive
1659 	 * window.  Check the sequence number versus IRS since we know
1660 	 * the sequence numbers haven't wrapped.  This is a partial fix
1661 	 * for the "LAND" DoS attack.
1662 	 */
1663 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1664 		rstreason = BANDLIM_RST_OPENPORT;
1665 		goto dropwithreset;
1666 	}
1667 
1668 	todrop = tp->rcv_nxt - th->th_seq;
1669 	if (todrop > 0) {
1670 		if (TCP_DO_SACK(tp)) {
1671 			/* Report duplicate segment at head of packet. */
1672 			tp->reportblk.rblk_start = th->th_seq;
1673 			tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1674 			    th->th_seq + tlen, thflags);
1675 			if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1676 				tp->reportblk.rblk_end = tp->rcv_nxt;
1677 			tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1678 		}
1679 		if (thflags & TH_SYN) {
1680 			thflags &= ~TH_SYN;
1681 			th->th_seq++;
1682 			if (th->th_urp > 1)
1683 				th->th_urp--;
1684 			else
1685 				thflags &= ~TH_URG;
1686 			todrop--;
1687 		}
1688 		/*
1689 		 * Following if statement from Stevens, vol. 2, p. 960.
1690 		 */
1691 		if (todrop > tlen ||
1692 		    (todrop == tlen && !(thflags & TH_FIN))) {
1693 			/*
1694 			 * Any valid FIN must be to the left of the window.
1695 			 * At this point the FIN must be a duplicate or out
1696 			 * of sequence; drop it.
1697 			 */
1698 			thflags &= ~TH_FIN;
1699 
1700 			/*
1701 			 * Send an ACK to resynchronize and drop any data.
1702 			 * But keep on processing for RST or ACK.
1703 			 */
1704 			tp->t_flags |= TF_ACKNOW;
1705 			todrop = tlen;
1706 			tcpstat.tcps_rcvduppack++;
1707 			tcpstat.tcps_rcvdupbyte += todrop;
1708 		} else {
1709 			tcpstat.tcps_rcvpartduppack++;
1710 			tcpstat.tcps_rcvpartdupbyte += todrop;
1711 		}
1712 		drop_hdrlen += todrop;	/* drop from the top afterwards */
1713 		th->th_seq += todrop;
1714 		tlen -= todrop;
1715 		if (th->th_urp > todrop)
1716 			th->th_urp -= todrop;
1717 		else {
1718 			thflags &= ~TH_URG;
1719 			th->th_urp = 0;
1720 		}
1721 	}
1722 
1723 	/*
1724 	 * If new data are received on a connection after the
1725 	 * user processes are gone, then RST the other end.
1726 	 */
1727 	if ((so->so_state & SS_NOFDREF) &&
1728 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1729 		tp = tcp_close(tp);
1730 		tcpstat.tcps_rcvafterclose++;
1731 		rstreason = BANDLIM_UNLIMITED;
1732 		goto dropwithreset;
1733 	}
1734 
1735 	/*
1736 	 * If segment ends after window, drop trailing data
1737 	 * (and PUSH and FIN); if nothing left, just ACK.
1738 	 */
1739 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1740 	if (todrop > 0) {
1741 		tcpstat.tcps_rcvpackafterwin++;
1742 		if (todrop >= tlen) {
1743 			tcpstat.tcps_rcvbyteafterwin += tlen;
1744 			/*
1745 			 * If a new connection request is received
1746 			 * while in TIME_WAIT, drop the old connection
1747 			 * and start over if the sequence numbers
1748 			 * are above the previous ones.
1749 			 */
1750 			if (thflags & TH_SYN &&
1751 			    tp->t_state == TCPS_TIME_WAIT &&
1752 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1753 				tp = tcp_close(tp);
1754 				goto findpcb;
1755 			}
1756 			/*
1757 			 * If window is closed can only take segments at
1758 			 * window edge, and have to drop data and PUSH from
1759 			 * incoming segments.  Continue processing, but
1760 			 * remember to ack.  Otherwise, drop segment
1761 			 * and ack.
1762 			 */
1763 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1764 				tp->t_flags |= TF_ACKNOW;
1765 				tcpstat.tcps_rcvwinprobe++;
1766 			} else
1767 				goto dropafterack;
1768 		} else
1769 			tcpstat.tcps_rcvbyteafterwin += todrop;
1770 		m_adj(m, -todrop);
1771 		tlen -= todrop;
1772 		thflags &= ~(TH_PUSH | TH_FIN);
1773 	}
1774 
1775 	/*
1776 	 * If last ACK falls within this segment's sequence numbers,
1777 	 * record its timestamp.
1778 	 * NOTE:
1779 	 * 1) That the test incorporates suggestions from the latest
1780 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1781 	 * 2) That updating only on newer timestamps interferes with
1782 	 *    our earlier PAWS tests, so this check should be solely
1783 	 *    predicated on the sequence space of this segment.
1784 	 * 3) That we modify the segment boundary check to be
1785 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1786 	 *    instead of RFC1323's
1787 	 *        Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1788 	 *    This modified check allows us to overcome RFC1323's
1789 	 *    limitations as described in Stevens TCP/IP Illustrated
1790 	 *    Vol. 2 p.869. In such cases, we can still calculate the
1791 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1792 	 */
1793 	if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1794 	    SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1795 					+ ((thflags & TH_SYN) != 0)
1796 					+ ((thflags & TH_FIN) != 0)))) {
1797 		tp->ts_recent_age = ticks;
1798 		tp->ts_recent = to.to_tsval;
1799 	}
1800 
1801 	/*
1802 	 * If a SYN is in the window, then this is an
1803 	 * error and we send an RST and drop the connection.
1804 	 */
1805 	if (thflags & TH_SYN) {
1806 		tp = tcp_drop(tp, ECONNRESET);
1807 		rstreason = BANDLIM_UNLIMITED;
1808 		goto dropwithreset;
1809 	}
1810 
1811 	/*
1812 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1813 	 * flag is on (half-synchronized state), then queue data for
1814 	 * later processing; else drop segment and return.
1815 	 */
1816 	if (!(thflags & TH_ACK)) {
1817 		if (tp->t_state == TCPS_SYN_RECEIVED ||
1818 		    (tp->t_flags & TF_NEEDSYN))
1819 			goto step6;
1820 		else
1821 			goto drop;
1822 	}
1823 
1824 	/*
1825 	 * Ack processing.
1826 	 */
1827 	switch (tp->t_state) {
1828 	/*
1829 	 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1830 	 * ESTABLISHED state and continue processing.
1831 	 * The ACK was checked above.
1832 	 */
1833 	case TCPS_SYN_RECEIVED:
1834 
1835 		tcpstat.tcps_connects++;
1836 		soisconnected(so);
1837 		/* Do window scaling? */
1838 		if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1839 		    (TF_RCVD_SCALE | TF_REQ_SCALE))
1840 			tp->rcv_scale = tp->request_r_scale;
1841 		/*
1842 		 * Make transitions:
1843 		 *      SYN-RECEIVED  -> ESTABLISHED
1844 		 *      SYN-RECEIVED* -> FIN-WAIT-1
1845 		 */
1846 		tp->t_starttime = ticks;
1847 		if (tp->t_flags & TF_NEEDFIN) {
1848 			tp->t_state = TCPS_FIN_WAIT_1;
1849 			tp->t_flags &= ~TF_NEEDFIN;
1850 		} else {
1851 			tcp_established(tp);
1852 		}
1853 		/*
1854 		 * If segment contains data or ACK, will call tcp_reass()
1855 		 * later; if not, do so now to pass queued data to user.
1856 		 */
1857 		if (tlen == 0 && !(thflags & TH_FIN))
1858 			tcp_reass(tp, NULL, NULL, NULL);
1859 		/* fall into ... */
1860 
1861 	/*
1862 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1863 	 * ACKs.  If the ack is in the range
1864 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1865 	 * then advance tp->snd_una to th->th_ack and drop
1866 	 * data from the retransmission queue.  If this ACK reflects
1867 	 * more up to date window information we update our window information.
1868 	 */
1869 	case TCPS_ESTABLISHED:
1870 	case TCPS_FIN_WAIT_1:
1871 	case TCPS_FIN_WAIT_2:
1872 	case TCPS_CLOSE_WAIT:
1873 	case TCPS_CLOSING:
1874 	case TCPS_LAST_ACK:
1875 	case TCPS_TIME_WAIT:
1876 
1877 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1878 			if (TCP_DO_SACK(tp))
1879 				tcp_sack_update_scoreboard(tp, &to);
1880 			if (tlen != 0 || tiwin != tp->snd_wnd) {
1881 				tp->t_dupacks = 0;
1882 				break;
1883 			}
1884 			tcpstat.tcps_rcvdupack++;
1885 			if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1886 			    th->th_ack != tp->snd_una) {
1887 				tp->t_dupacks = 0;
1888 				break;
1889 			}
1890 			/*
1891 			 * We have outstanding data (other than
1892 			 * a window probe), this is a completely
1893 			 * duplicate ack (ie, window info didn't
1894 			 * change), the ack is the biggest we've
1895 			 * seen and we've seen exactly our rexmt
1896 			 * threshhold of them, so assume a packet
1897 			 * has been dropped and retransmit it.
1898 			 * Kludge snd_nxt & the congestion
1899 			 * window so we send only this one
1900 			 * packet.
1901 			 */
1902 			if (IN_FASTRECOVERY(tp)) {
1903 				if (TCP_DO_SACK(tp)) {
1904 					/* No artifical cwnd inflation. */
1905 					tcp_sack_rexmt(tp, th);
1906 				} else {
1907 					/*
1908 					 * Dup acks mean that packets
1909 					 * have left the network
1910 					 * (they're now cached at the
1911 					 * receiver) so bump cwnd by
1912 					 * the amount in the receiver
1913 					 * to keep a constant cwnd
1914 					 * packets in the network.
1915 					 */
1916 					tp->snd_cwnd += tp->t_maxseg;
1917 					tcp_output(tp);
1918 				}
1919 			} else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1920 				tp->t_dupacks = 0;
1921 				break;
1922 			} else if (++tp->t_dupacks == tcprexmtthresh) {
1923 				tcp_seq old_snd_nxt;
1924 				u_int win;
1925 
1926 fastretransmit:
1927 				if (tcp_do_eifel_detect &&
1928 				    (tp->t_flags & TF_RCVD_TSTMP)) {
1929 					tcp_save_congestion_state(tp);
1930 					tp->t_flags |= TF_FASTREXMT;
1931 				}
1932 				/*
1933 				 * We know we're losing at the current
1934 				 * window size, so do congestion avoidance:
1935 				 * set ssthresh to half the current window
1936 				 * and pull our congestion window back to the
1937 				 * new ssthresh.
1938 				 */
1939 				win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1940 				    tp->t_maxseg;
1941 				if (win < 2)
1942 					win = 2;
1943 				tp->snd_ssthresh = win * tp->t_maxseg;
1944 				ENTER_FASTRECOVERY(tp);
1945 				tp->snd_recover = tp->snd_max;
1946 				tcp_callout_stop(tp, tp->tt_rexmt);
1947 				tp->t_rtttime = 0;
1948 				old_snd_nxt = tp->snd_nxt;
1949 				tp->snd_nxt = th->th_ack;
1950 				tp->snd_cwnd = tp->t_maxseg;
1951 				tcp_output(tp);
1952 				++tcpstat.tcps_sndfastrexmit;
1953 				tp->snd_cwnd = tp->snd_ssthresh;
1954 				tp->rexmt_high = tp->snd_nxt;
1955 				tp->t_flags &= ~TF_SACKRESCUED;
1956 				if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1957 					tp->snd_nxt = old_snd_nxt;
1958 				KASSERT(tp->snd_limited <= 2,
1959 				    ("tp->snd_limited too big"));
1960 				if (TCP_DO_SACK(tp))
1961 					tcp_sack_rexmt(tp, th);
1962 				else
1963 					tp->snd_cwnd += tp->t_maxseg *
1964 					    (tp->t_dupacks - tp->snd_limited);
1965 			} else if (tcp_do_limitedtransmit) {
1966 				u_long oldcwnd = tp->snd_cwnd;
1967 				tcp_seq oldsndmax = tp->snd_max;
1968 				tcp_seq oldsndnxt = tp->snd_nxt;
1969 				/* outstanding data */
1970 				uint32_t ownd = tp->snd_max - tp->snd_una;
1971 				u_int sent;
1972 
1973 #define	iceildiv(n, d)		(((n)+(d)-1) / (d))
1974 
1975 				KASSERT(tp->t_dupacks == 1 ||
1976 					tp->t_dupacks == 2,
1977 				    ("dupacks not 1 or 2"));
1978 				if (tp->t_dupacks == 1)
1979 					tp->snd_limited = 0;
1980 				tp->snd_nxt = tp->snd_max;
1981 				tp->snd_cwnd = ownd +
1982 				    (tp->t_dupacks - tp->snd_limited) *
1983 				    tp->t_maxseg;
1984 				tcp_output(tp);
1985 
1986 				if (SEQ_LT(oldsndnxt, oldsndmax)) {
1987 					KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
1988 					    ("snd_una moved in other threads"));
1989 					tp->snd_nxt = oldsndnxt;
1990 				}
1991 				tp->snd_cwnd = oldcwnd;
1992 				sent = tp->snd_max - oldsndmax;
1993 				if (sent > tp->t_maxseg) {
1994 					KASSERT((tp->t_dupacks == 2 &&
1995 						 tp->snd_limited == 0) ||
1996 						(sent == tp->t_maxseg + 1 &&
1997 						 tp->t_flags & TF_SENTFIN),
1998 					    ("sent too much"));
1999 					KASSERT(sent <= tp->t_maxseg * 2,
2000 					    ("sent too many segments"));
2001 					tp->snd_limited = 2;
2002 					tcpstat.tcps_sndlimited += 2;
2003 				} else if (sent > 0) {
2004 					++tp->snd_limited;
2005 					++tcpstat.tcps_sndlimited;
2006 				} else if (tcp_do_early_retransmit &&
2007 				    (tcp_do_eifel_detect &&
2008 				     (tp->t_flags & TF_RCVD_TSTMP)) &&
2009 				    ownd < 4 * tp->t_maxseg &&
2010 				    tp->t_dupacks + 1 >=
2011 				      iceildiv(ownd, tp->t_maxseg) &&
2012 				    (!TCP_DO_SACK(tp) ||
2013 				     ownd <= tp->t_maxseg ||
2014 				     tcp_sack_has_sacked(&tp->scb,
2015 							ownd - tp->t_maxseg))) {
2016 					++tcpstat.tcps_sndearlyrexmit;
2017 					tp->t_flags |= TF_EARLYREXMT;
2018 					goto fastretransmit;
2019 				}
2020 			}
2021 			goto drop;
2022 		}
2023 
2024 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2025 		tp->t_dupacks = 0;
2026 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2027 			/*
2028 			 * Detected optimistic ACK attack.
2029 			 * Force slow-start to de-synchronize attack.
2030 			 */
2031 			tp->snd_cwnd = tp->t_maxseg;
2032 			tp->snd_wacked = 0;
2033 
2034 			tcpstat.tcps_rcvacktoomuch++;
2035 			goto dropafterack;
2036 		}
2037 		/*
2038 		 * If we reach this point, ACK is not a duplicate,
2039 		 *     i.e., it ACKs something we sent.
2040 		 */
2041 		if (tp->t_flags & TF_NEEDSYN) {
2042 			/*
2043 			 * T/TCP: Connection was half-synchronized, and our
2044 			 * SYN has been ACK'd (so connection is now fully
2045 			 * synchronized).  Go to non-starred state,
2046 			 * increment snd_una for ACK of SYN, and check if
2047 			 * we can do window scaling.
2048 			 */
2049 			tp->t_flags &= ~TF_NEEDSYN;
2050 			tp->snd_una++;
2051 			/* Do window scaling? */
2052 			if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2053 			    (TF_RCVD_SCALE | TF_REQ_SCALE))
2054 				tp->rcv_scale = tp->request_r_scale;
2055 		}
2056 
2057 process_ACK:
2058 		acked = th->th_ack - tp->snd_una;
2059 		tcpstat.tcps_rcvackpack++;
2060 		tcpstat.tcps_rcvackbyte += acked;
2061 
2062 		if (tcp_do_eifel_detect && acked > 0 &&
2063 		    (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2064 		    (tp->t_flags & TF_FIRSTACCACK)) {
2065 			/* Eifel detection applicable. */
2066 			if (to.to_tsecr < tp->t_rexmtTS) {
2067 				++tcpstat.tcps_eifeldetected;
2068 				tcp_revert_congestion_state(tp);
2069 				if (tp->t_rxtshift != 1 ||
2070 				    ticks >= tp->t_badrxtwin)
2071 					++tcpstat.tcps_rttcantdetect;
2072 			}
2073 		} else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2074 			/*
2075 			 * If we just performed our first retransmit,
2076 			 * and the ACK arrives within our recovery window,
2077 			 * then it was a mistake to do the retransmit
2078 			 * in the first place.  Recover our original cwnd
2079 			 * and ssthresh, and proceed to transmit where we
2080 			 * left off.
2081 			 */
2082 			tcp_revert_congestion_state(tp);
2083 			++tcpstat.tcps_rttdetected;
2084 		}
2085 
2086 		/*
2087 		 * If we have a timestamp reply, update smoothed
2088 		 * round trip time.  If no timestamp is present but
2089 		 * transmit timer is running and timed sequence
2090 		 * number was acked, update smoothed round trip time.
2091 		 * Since we now have an rtt measurement, cancel the
2092 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2093 		 * Recompute the initial retransmit timer.
2094 		 *
2095 		 * Some machines (certain windows boxes) send broken
2096 		 * timestamp replies during the SYN+ACK phase, ignore
2097 		 * timestamps of 0.
2098 		 */
2099 		if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2100 			tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2101 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2102 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2103 		tcp_xmit_bandwidth_limit(tp, th->th_ack);
2104 
2105 		/*
2106 		 * If no data (only SYN) was ACK'd,
2107 		 *    skip rest of ACK processing.
2108 		 */
2109 		if (acked == 0)
2110 			goto step6;
2111 
2112 		/* Stop looking for an acceptable ACK since one was received. */
2113 		tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2114 
2115 		if (acked > so->so_snd.ssb_cc) {
2116 			tp->snd_wnd -= so->so_snd.ssb_cc;
2117 			sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2118 			ourfinisacked = TRUE;
2119 		} else {
2120 			sbdrop(&so->so_snd.sb, acked);
2121 			tp->snd_wnd -= acked;
2122 			ourfinisacked = FALSE;
2123 		}
2124 		sowwakeup(so);
2125 
2126 		/*
2127 		 * Update window information.
2128 		 */
2129 		if (acceptable_window_update(tp, th, tiwin)) {
2130 			/* keep track of pure window updates */
2131 			if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2132 			    tiwin > tp->snd_wnd)
2133 				tcpstat.tcps_rcvwinupd++;
2134 			tp->snd_wnd = tiwin;
2135 			tp->snd_wl1 = th->th_seq;
2136 			tp->snd_wl2 = th->th_ack;
2137 			if (tp->snd_wnd > tp->max_sndwnd)
2138 				tp->max_sndwnd = tp->snd_wnd;
2139 			needoutput = TRUE;
2140 		}
2141 
2142 		tp->snd_una = th->th_ack;
2143 		if (TCP_DO_SACK(tp))
2144 			tcp_sack_update_scoreboard(tp, &to);
2145 		if (IN_FASTRECOVERY(tp)) {
2146 			if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2147 				EXIT_FASTRECOVERY(tp);
2148 				needoutput = TRUE;
2149 				/*
2150 				 * If the congestion window was inflated
2151 				 * to account for the other side's
2152 				 * cached packets, retract it.
2153 				 */
2154 				if (!TCP_DO_SACK(tp))
2155 					tp->snd_cwnd = tp->snd_ssthresh;
2156 
2157 				/*
2158 				 * Window inflation should have left us
2159 				 * with approximately snd_ssthresh outstanding
2160 				 * data.  But, in case we would be inclined
2161 				 * to send a burst, better do it using
2162 				 * slow start.
2163 				 */
2164 				if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2165 					   tp->snd_max + 2 * tp->t_maxseg))
2166 					tp->snd_cwnd =
2167 					    (tp->snd_max - tp->snd_una) +
2168 					    2 * tp->t_maxseg;
2169 
2170 				tp->snd_wacked = 0;
2171 			} else {
2172 				if (TCP_DO_SACK(tp)) {
2173 					tp->snd_max_rexmt = tp->snd_max;
2174 					tcp_sack_rexmt(tp, th);
2175 				} else {
2176 					tcp_newreno_partial_ack(tp, th, acked);
2177 				}
2178 				needoutput = FALSE;
2179 			}
2180 		} else {
2181 			/*
2182 			 * Open the congestion window.  When in slow-start,
2183 			 * open exponentially: maxseg per packet.  Otherwise,
2184 			 * open linearly: maxseg per window.
2185 			 */
2186 			if (tp->snd_cwnd <= tp->snd_ssthresh) {
2187 				u_int abc_sslimit =
2188 				    (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2189 				     tp->t_maxseg : 2 * tp->t_maxseg);
2190 
2191 				/* slow-start */
2192 				tp->snd_cwnd += tcp_do_abc ?
2193 				    min(acked, abc_sslimit) : tp->t_maxseg;
2194 			} else {
2195 				/* linear increase */
2196 				tp->snd_wacked += tcp_do_abc ? acked :
2197 				    tp->t_maxseg;
2198 				if (tp->snd_wacked >= tp->snd_cwnd) {
2199 					tp->snd_wacked -= tp->snd_cwnd;
2200 					tp->snd_cwnd += tp->t_maxseg;
2201 				}
2202 			}
2203 			tp->snd_cwnd = min(tp->snd_cwnd,
2204 					   TCP_MAXWIN << tp->snd_scale);
2205 			tp->snd_recover = th->th_ack - 1;
2206 		}
2207 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2208 			tp->snd_nxt = tp->snd_una;
2209 
2210 		/*
2211 		 * If all outstanding data is acked, stop retransmit
2212 		 * timer and remember to restart (more output or persist).
2213 		 * If there is more data to be acked, restart retransmit
2214 		 * timer, using current (possibly backed-off) value.
2215 		 */
2216 		if (th->th_ack == tp->snd_max) {
2217 			tcp_callout_stop(tp, tp->tt_rexmt);
2218 			needoutput = TRUE;
2219 		} else if (!tcp_callout_active(tp, tp->tt_persist)) {
2220 			tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2221 			    tcp_timer_rexmt);
2222 		}
2223 
2224 		switch (tp->t_state) {
2225 		/*
2226 		 * In FIN_WAIT_1 STATE in addition to the processing
2227 		 * for the ESTABLISHED state if our FIN is now acknowledged
2228 		 * then enter FIN_WAIT_2.
2229 		 */
2230 		case TCPS_FIN_WAIT_1:
2231 			if (ourfinisacked) {
2232 				/*
2233 				 * If we can't receive any more
2234 				 * data, then closing user can proceed.
2235 				 * Starting the timer is contrary to the
2236 				 * specification, but if we don't get a FIN
2237 				 * we'll hang forever.
2238 				 */
2239 				if (so->so_state & SS_CANTRCVMORE) {
2240 					soisdisconnected(so);
2241 					tcp_callout_reset(tp, tp->tt_2msl,
2242 					    tp->t_maxidle, tcp_timer_2msl);
2243 				}
2244 				tp->t_state = TCPS_FIN_WAIT_2;
2245 			}
2246 			break;
2247 
2248 		/*
2249 		 * In CLOSING STATE in addition to the processing for
2250 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2251 		 * then enter the TIME-WAIT state, otherwise ignore
2252 		 * the segment.
2253 		 */
2254 		case TCPS_CLOSING:
2255 			if (ourfinisacked) {
2256 				tp->t_state = TCPS_TIME_WAIT;
2257 				tcp_canceltimers(tp);
2258 				tcp_callout_reset(tp, tp->tt_2msl,
2259 					    2 * tcp_rmx_msl(tp),
2260 					    tcp_timer_2msl);
2261 				soisdisconnected(so);
2262 			}
2263 			break;
2264 
2265 		/*
2266 		 * In LAST_ACK, we may still be waiting for data to drain
2267 		 * and/or to be acked, as well as for the ack of our FIN.
2268 		 * If our FIN is now acknowledged, delete the TCB,
2269 		 * enter the closed state and return.
2270 		 */
2271 		case TCPS_LAST_ACK:
2272 			if (ourfinisacked) {
2273 				tp = tcp_close(tp);
2274 				goto drop;
2275 			}
2276 			break;
2277 
2278 		/*
2279 		 * In TIME_WAIT state the only thing that should arrive
2280 		 * is a retransmission of the remote FIN.  Acknowledge
2281 		 * it and restart the finack timer.
2282 		 */
2283 		case TCPS_TIME_WAIT:
2284 			tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2285 			    tcp_timer_2msl);
2286 			goto dropafterack;
2287 		}
2288 	}
2289 
2290 step6:
2291 	/*
2292 	 * Update window information.
2293 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2294 	 */
2295 	if ((thflags & TH_ACK) &&
2296 	    acceptable_window_update(tp, th, tiwin)) {
2297 		/* keep track of pure window updates */
2298 		if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2299 		    tiwin > tp->snd_wnd)
2300 			tcpstat.tcps_rcvwinupd++;
2301 		tp->snd_wnd = tiwin;
2302 		tp->snd_wl1 = th->th_seq;
2303 		tp->snd_wl2 = th->th_ack;
2304 		if (tp->snd_wnd > tp->max_sndwnd)
2305 			tp->max_sndwnd = tp->snd_wnd;
2306 		needoutput = TRUE;
2307 	}
2308 
2309 	/*
2310 	 * Process segments with URG.
2311 	 */
2312 	if ((thflags & TH_URG) && th->th_urp &&
2313 	    !TCPS_HAVERCVDFIN(tp->t_state)) {
2314 		/*
2315 		 * This is a kludge, but if we receive and accept
2316 		 * random urgent pointers, we'll crash in
2317 		 * soreceive.  It's hard to imagine someone
2318 		 * actually wanting to send this much urgent data.
2319 		 */
2320 		if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2321 			th->th_urp = 0;			/* XXX */
2322 			thflags &= ~TH_URG;		/* XXX */
2323 			goto dodata;			/* XXX */
2324 		}
2325 		/*
2326 		 * If this segment advances the known urgent pointer,
2327 		 * then mark the data stream.  This should not happen
2328 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2329 		 * a FIN has been received from the remote side.
2330 		 * In these states we ignore the URG.
2331 		 *
2332 		 * According to RFC961 (Assigned Protocols),
2333 		 * the urgent pointer points to the last octet
2334 		 * of urgent data.  We continue, however,
2335 		 * to consider it to indicate the first octet
2336 		 * of data past the urgent section as the original
2337 		 * spec states (in one of two places).
2338 		 */
2339 		if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2340 			tp->rcv_up = th->th_seq + th->th_urp;
2341 			so->so_oobmark = so->so_rcv.ssb_cc +
2342 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2343 			if (so->so_oobmark == 0)
2344 				sosetstate(so, SS_RCVATMARK);
2345 			sohasoutofband(so);
2346 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2347 		}
2348 		/*
2349 		 * Remove out of band data so doesn't get presented to user.
2350 		 * This can happen independent of advancing the URG pointer,
2351 		 * but if two URG's are pending at once, some out-of-band
2352 		 * data may creep in... ick.
2353 		 */
2354 		if (th->th_urp <= (u_long)tlen &&
2355 		    !(so->so_options & SO_OOBINLINE)) {
2356 			/* hdr drop is delayed */
2357 			tcp_pulloutofband(so, th, m, drop_hdrlen);
2358 		}
2359 	} else {
2360 		/*
2361 		 * If no out of band data is expected,
2362 		 * pull receive urgent pointer along
2363 		 * with the receive window.
2364 		 */
2365 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2366 			tp->rcv_up = tp->rcv_nxt;
2367 	}
2368 
2369 dodata:							/* XXX */
2370 	/*
2371 	 * Process the segment text, merging it into the TCP sequencing queue,
2372 	 * and arranging for acknowledgment of receipt if necessary.
2373 	 * This process logically involves adjusting tp->rcv_wnd as data
2374 	 * is presented to the user (this happens in tcp_usrreq.c,
2375 	 * case PRU_RCVD).  If a FIN has already been received on this
2376 	 * connection then we just ignore the text.
2377 	 */
2378 	if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2379 		m_adj(m, drop_hdrlen);	/* delayed header drop */
2380 		/*
2381 		 * Insert segment which includes th into TCP reassembly queue
2382 		 * with control block tp.  Set thflags to whether reassembly now
2383 		 * includes a segment with FIN.  This handles the common case
2384 		 * inline (segment is the next to be received on an established
2385 		 * connection, and the queue is empty), avoiding linkage into
2386 		 * and removal from the queue and repetition of various
2387 		 * conversions.
2388 		 * Set DELACK for segments received in order, but ack
2389 		 * immediately when segments are out of order (so
2390 		 * fast retransmit can work).
2391 		 */
2392 		if (th->th_seq == tp->rcv_nxt &&
2393 		    LIST_EMPTY(&tp->t_segq) &&
2394 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
2395 			if (DELAY_ACK(tp)) {
2396 				tcp_callout_reset(tp, tp->tt_delack,
2397 				    tcp_delacktime, tcp_timer_delack);
2398 			} else {
2399 				tp->t_flags |= TF_ACKNOW;
2400 			}
2401 			tp->rcv_nxt += tlen;
2402 			thflags = th->th_flags & TH_FIN;
2403 			tcpstat.tcps_rcvpack++;
2404 			tcpstat.tcps_rcvbyte += tlen;
2405 			ND6_HINT(tp);
2406 			if (so->so_state & SS_CANTRCVMORE) {
2407 				m_freem(m);
2408 			} else {
2409 				lwkt_gettoken(&so->so_rcv.ssb_token);
2410 				ssb_appendstream(&so->so_rcv, m);
2411 				lwkt_reltoken(&so->so_rcv.ssb_token);
2412 			}
2413 			sorwakeup(so);
2414 		} else {
2415 			if (!(tp->t_flags & TF_DUPSEG)) {
2416 				/* Initialize SACK report block. */
2417 				tp->reportblk.rblk_start = th->th_seq;
2418 				tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2419 				    th->th_seq + tlen, thflags);
2420 			}
2421 			thflags = tcp_reass(tp, th, &tlen, m);
2422 			tp->t_flags |= TF_ACKNOW;
2423 		}
2424 
2425 		/*
2426 		 * Note the amount of data that peer has sent into
2427 		 * our window, in order to estimate the sender's
2428 		 * buffer size.
2429 		 */
2430 		len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2431 	} else {
2432 		m_freem(m);
2433 		thflags &= ~TH_FIN;
2434 	}
2435 
2436 	/*
2437 	 * If FIN is received ACK the FIN and let the user know
2438 	 * that the connection is closing.
2439 	 */
2440 	if (thflags & TH_FIN) {
2441 		if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2442 			socantrcvmore(so);
2443 			/*
2444 			 * If connection is half-synchronized
2445 			 * (ie NEEDSYN flag on) then delay ACK,
2446 			 * so it may be piggybacked when SYN is sent.
2447 			 * Otherwise, since we received a FIN then no
2448 			 * more input can be expected, send ACK now.
2449 			 */
2450 			if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2451 				tcp_callout_reset(tp, tp->tt_delack,
2452 				    tcp_delacktime, tcp_timer_delack);
2453 			} else {
2454 				tp->t_flags |= TF_ACKNOW;
2455 			}
2456 			tp->rcv_nxt++;
2457 		}
2458 
2459 		switch (tp->t_state) {
2460 		/*
2461 		 * In SYN_RECEIVED and ESTABLISHED STATES
2462 		 * enter the CLOSE_WAIT state.
2463 		 */
2464 		case TCPS_SYN_RECEIVED:
2465 			tp->t_starttime = ticks;
2466 			/*FALLTHROUGH*/
2467 		case TCPS_ESTABLISHED:
2468 			tp->t_state = TCPS_CLOSE_WAIT;
2469 			break;
2470 
2471 		/*
2472 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2473 		 * enter the CLOSING state.
2474 		 */
2475 		case TCPS_FIN_WAIT_1:
2476 			tp->t_state = TCPS_CLOSING;
2477 			break;
2478 
2479 		/*
2480 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2481 		 * starting the time-wait timer, turning off the other
2482 		 * standard timers.
2483 		 */
2484 		case TCPS_FIN_WAIT_2:
2485 			tp->t_state = TCPS_TIME_WAIT;
2486 			tcp_canceltimers(tp);
2487 			tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2488 				    tcp_timer_2msl);
2489 			soisdisconnected(so);
2490 			break;
2491 
2492 		/*
2493 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2494 		 */
2495 		case TCPS_TIME_WAIT:
2496 			tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2497 			    tcp_timer_2msl);
2498 			break;
2499 		}
2500 	}
2501 
2502 #ifdef TCPDEBUG
2503 	if (so->so_options & SO_DEBUG)
2504 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2505 #endif
2506 
2507 	/*
2508 	 * Return any desired output.
2509 	 */
2510 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2511 		tcp_output(tp);
2512 	tcp_sack_report_cleanup(tp);
2513 	return(IPPROTO_DONE);
2514 
2515 dropafterack:
2516 	/*
2517 	 * Generate an ACK dropping incoming segment if it occupies
2518 	 * sequence space, where the ACK reflects our state.
2519 	 *
2520 	 * We can now skip the test for the RST flag since all
2521 	 * paths to this code happen after packets containing
2522 	 * RST have been dropped.
2523 	 *
2524 	 * In the SYN-RECEIVED state, don't send an ACK unless the
2525 	 * segment we received passes the SYN-RECEIVED ACK test.
2526 	 * If it fails send a RST.  This breaks the loop in the
2527 	 * "LAND" DoS attack, and also prevents an ACK storm
2528 	 * between two listening ports that have been sent forged
2529 	 * SYN segments, each with the source address of the other.
2530 	 */
2531 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2532 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
2533 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
2534 		rstreason = BANDLIM_RST_OPENPORT;
2535 		goto dropwithreset;
2536 	}
2537 #ifdef TCPDEBUG
2538 	if (so->so_options & SO_DEBUG)
2539 		tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2540 #endif
2541 	m_freem(m);
2542 	tp->t_flags |= TF_ACKNOW;
2543 	tcp_output(tp);
2544 	tcp_sack_report_cleanup(tp);
2545 	return(IPPROTO_DONE);
2546 
2547 dropwithreset:
2548 	/*
2549 	 * Generate a RST, dropping incoming segment.
2550 	 * Make ACK acceptable to originator of segment.
2551 	 * Don't bother to respond if destination was broadcast/multicast.
2552 	 */
2553 	if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2554 		goto drop;
2555 	if (isipv6) {
2556 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2557 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2558 			goto drop;
2559 	} else {
2560 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2561 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2562 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2563 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2564 			goto drop;
2565 	}
2566 	/* IPv6 anycast check is done at tcp6_input() */
2567 
2568 	/*
2569 	 * Perform bandwidth limiting.
2570 	 */
2571 #ifdef ICMP_BANDLIM
2572 	if (badport_bandlim(rstreason) < 0)
2573 		goto drop;
2574 #endif
2575 
2576 #ifdef TCPDEBUG
2577 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2578 		tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2579 #endif
2580 	if (thflags & TH_ACK)
2581 		/* mtod() below is safe as long as hdr dropping is delayed */
2582 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2583 			    TH_RST);
2584 	else {
2585 		if (thflags & TH_SYN)
2586 			tlen++;
2587 		/* mtod() below is safe as long as hdr dropping is delayed */
2588 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2589 			    (tcp_seq)0, TH_RST | TH_ACK);
2590 	}
2591 	if (tp != NULL)
2592 		tcp_sack_report_cleanup(tp);
2593 	return(IPPROTO_DONE);
2594 
2595 drop:
2596 	/*
2597 	 * Drop space held by incoming segment and return.
2598 	 */
2599 #ifdef TCPDEBUG
2600 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2601 		tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2602 #endif
2603 	m_freem(m);
2604 	if (tp != NULL)
2605 		tcp_sack_report_cleanup(tp);
2606 	return(IPPROTO_DONE);
2607 }
2608 
2609 /*
2610  * Parse TCP options and place in tcpopt.
2611  */
2612 static void
2613 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2614 {
2615 	int opt, optlen, i;
2616 
2617 	to->to_flags = 0;
2618 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2619 		opt = cp[0];
2620 		if (opt == TCPOPT_EOL)
2621 			break;
2622 		if (opt == TCPOPT_NOP)
2623 			optlen = 1;
2624 		else {
2625 			if (cnt < 2)
2626 				break;
2627 			optlen = cp[1];
2628 			if (optlen < 2 || optlen > cnt)
2629 				break;
2630 		}
2631 		switch (opt) {
2632 		case TCPOPT_MAXSEG:
2633 			if (optlen != TCPOLEN_MAXSEG)
2634 				continue;
2635 			if (!is_syn)
2636 				continue;
2637 			to->to_flags |= TOF_MSS;
2638 			bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2639 			to->to_mss = ntohs(to->to_mss);
2640 			break;
2641 		case TCPOPT_WINDOW:
2642 			if (optlen != TCPOLEN_WINDOW)
2643 				continue;
2644 			if (!is_syn)
2645 				continue;
2646 			to->to_flags |= TOF_SCALE;
2647 			to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2648 			break;
2649 		case TCPOPT_TIMESTAMP:
2650 			if (optlen != TCPOLEN_TIMESTAMP)
2651 				continue;
2652 			to->to_flags |= TOF_TS;
2653 			bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2654 			to->to_tsval = ntohl(to->to_tsval);
2655 			bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2656 			to->to_tsecr = ntohl(to->to_tsecr);
2657 			/*
2658 			 * If echoed timestamp is later than the current time,
2659 			 * fall back to non RFC1323 RTT calculation.
2660 			 */
2661 			if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2662 				to->to_tsecr = 0;
2663 			break;
2664 		case TCPOPT_SACK_PERMITTED:
2665 			if (optlen != TCPOLEN_SACK_PERMITTED)
2666 				continue;
2667 			if (!is_syn)
2668 				continue;
2669 			to->to_flags |= TOF_SACK_PERMITTED;
2670 			break;
2671 		case TCPOPT_SACK:
2672 			if ((optlen - 2) & 0x07)	/* not multiple of 8 */
2673 				continue;
2674 			to->to_nsackblocks = (optlen - 2) / 8;
2675 			to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2676 			to->to_flags |= TOF_SACK;
2677 			for (i = 0; i < to->to_nsackblocks; i++) {
2678 				struct raw_sackblock *r = &to->to_sackblocks[i];
2679 
2680 				r->rblk_start = ntohl(r->rblk_start);
2681 				r->rblk_end = ntohl(r->rblk_end);
2682 
2683 				if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2684 					/*
2685 					 * Invalid SACK block; discard all
2686 					 * SACK blocks
2687 					 */
2688 					tcpstat.tcps_rcvbadsackopt++;
2689 					to->to_nsackblocks = 0;
2690 					to->to_sackblocks = NULL;
2691 					to->to_flags &= ~TOF_SACK;
2692 					break;
2693 				}
2694 			}
2695 			break;
2696 #ifdef TCP_SIGNATURE
2697 		/*
2698 		 * XXX In order to reply to a host which has set the
2699 		 * TCP_SIGNATURE option in its initial SYN, we have to
2700 		 * record the fact that the option was observed here
2701 		 * for the syncache code to perform the correct response.
2702 		 */
2703 		case TCPOPT_SIGNATURE:
2704 			if (optlen != TCPOLEN_SIGNATURE)
2705 				continue;
2706 			to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2707 			break;
2708 #endif /* TCP_SIGNATURE */
2709 		default:
2710 			continue;
2711 		}
2712 	}
2713 }
2714 
2715 /*
2716  * Pull out of band byte out of a segment so
2717  * it doesn't appear in the user's data queue.
2718  * It is still reflected in the segment length for
2719  * sequencing purposes.
2720  * "off" is the delayed to be dropped hdrlen.
2721  */
2722 static void
2723 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2724 {
2725 	int cnt = off + th->th_urp - 1;
2726 
2727 	while (cnt >= 0) {
2728 		if (m->m_len > cnt) {
2729 			char *cp = mtod(m, caddr_t) + cnt;
2730 			struct tcpcb *tp = sototcpcb(so);
2731 
2732 			tp->t_iobc = *cp;
2733 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2734 			bcopy(cp + 1, cp, m->m_len - cnt - 1);
2735 			m->m_len--;
2736 			if (m->m_flags & M_PKTHDR)
2737 				m->m_pkthdr.len--;
2738 			return;
2739 		}
2740 		cnt -= m->m_len;
2741 		m = m->m_next;
2742 		if (m == NULL)
2743 			break;
2744 	}
2745 	panic("tcp_pulloutofband");
2746 }
2747 
2748 /*
2749  * Collect new round-trip time estimate
2750  * and update averages and current timeout.
2751  */
2752 static void
2753 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2754 {
2755 	int delta;
2756 
2757 	tcpstat.tcps_rttupdated++;
2758 	tp->t_rttupdated++;
2759 	if (tp->t_srtt != 0) {
2760 		/*
2761 		 * srtt is stored as fixed point with 5 bits after the
2762 		 * binary point (i.e., scaled by 8).  The following magic
2763 		 * is equivalent to the smoothing algorithm in rfc793 with
2764 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2765 		 * point).  Adjust rtt to origin 0.
2766 		 */
2767 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2768 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2769 
2770 		if ((tp->t_srtt += delta) <= 0)
2771 			tp->t_srtt = 1;
2772 
2773 		/*
2774 		 * We accumulate a smoothed rtt variance (actually, a
2775 		 * smoothed mean difference), then set the retransmit
2776 		 * timer to smoothed rtt + 4 times the smoothed variance.
2777 		 * rttvar is stored as fixed point with 4 bits after the
2778 		 * binary point (scaled by 16).  The following is
2779 		 * equivalent to rfc793 smoothing with an alpha of .75
2780 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2781 		 * rfc793's wired-in beta.
2782 		 */
2783 		if (delta < 0)
2784 			delta = -delta;
2785 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2786 		if ((tp->t_rttvar += delta) <= 0)
2787 			tp->t_rttvar = 1;
2788 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2789 			tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2790 	} else {
2791 		/*
2792 		 * No rtt measurement yet - use the unsmoothed rtt.
2793 		 * Set the variance to half the rtt (so our first
2794 		 * retransmit happens at 3*rtt).
2795 		 */
2796 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
2797 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2798 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2799 	}
2800 	tp->t_rtttime = 0;
2801 	tp->t_rxtshift = 0;
2802 
2803 	/*
2804 	 * the retransmit should happen at rtt + 4 * rttvar.
2805 	 * Because of the way we do the smoothing, srtt and rttvar
2806 	 * will each average +1/2 tick of bias.  When we compute
2807 	 * the retransmit timer, we want 1/2 tick of rounding and
2808 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2809 	 * firing of the timer.  The bias will give us exactly the
2810 	 * 1.5 tick we need.  But, because the bias is
2811 	 * statistical, we have to test that we don't drop below
2812 	 * the minimum feasible timer (which is 2 ticks).
2813 	 */
2814 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2815 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2816 
2817 	/*
2818 	 * We received an ack for a packet that wasn't retransmitted;
2819 	 * it is probably safe to discard any error indications we've
2820 	 * received recently.  This isn't quite right, but close enough
2821 	 * for now (a route might have failed after we sent a segment,
2822 	 * and the return path might not be symmetrical).
2823 	 */
2824 	tp->t_softerror = 0;
2825 }
2826 
2827 /*
2828  * Determine a reasonable value for maxseg size.
2829  * If the route is known, check route for mtu.
2830  * If none, use an mss that can be handled on the outgoing
2831  * interface without forcing IP to fragment; if bigger than
2832  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2833  * to utilize large mbufs.  If no route is found, route has no mtu,
2834  * or the destination isn't local, use a default, hopefully conservative
2835  * size (usually 512 or the default IP max size, but no more than the mtu
2836  * of the interface), as we can't discover anything about intervening
2837  * gateways or networks.  We also initialize the congestion/slow start
2838  * window to be a single segment if the destination isn't local.
2839  * While looking at the routing entry, we also initialize other path-dependent
2840  * parameters from pre-set or cached values in the routing entry.
2841  *
2842  * Also take into account the space needed for options that we
2843  * send regularly.  Make maxseg shorter by that amount to assure
2844  * that we can send maxseg amount of data even when the options
2845  * are present.  Store the upper limit of the length of options plus
2846  * data in maxopd.
2847  *
2848  * NOTE that this routine is only called when we process an incoming
2849  * segment, for outgoing segments only tcp_mssopt is called.
2850  */
2851 void
2852 tcp_mss(struct tcpcb *tp, int offer)
2853 {
2854 	struct rtentry *rt;
2855 	struct ifnet *ifp;
2856 	int rtt, mss;
2857 	u_long bufsize;
2858 	struct inpcb *inp = tp->t_inpcb;
2859 	struct socket *so;
2860 #ifdef INET6
2861 	boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2862 	size_t min_protoh = isipv6 ?
2863 			    sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2864 			    sizeof(struct tcpiphdr);
2865 #else
2866 	const boolean_t isipv6 = FALSE;
2867 	const size_t min_protoh = sizeof(struct tcpiphdr);
2868 #endif
2869 
2870 	if (isipv6)
2871 		rt = tcp_rtlookup6(&inp->inp_inc);
2872 	else
2873 		rt = tcp_rtlookup(&inp->inp_inc);
2874 	if (rt == NULL) {
2875 		tp->t_maxopd = tp->t_maxseg =
2876 		    (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2877 		return;
2878 	}
2879 	ifp = rt->rt_ifp;
2880 	so = inp->inp_socket;
2881 
2882 	/*
2883 	 * Offer == 0 means that there was no MSS on the SYN segment,
2884 	 * in this case we use either the interface mtu or tcp_mssdflt.
2885 	 *
2886 	 * An offer which is too large will be cut down later.
2887 	 */
2888 	if (offer == 0) {
2889 		if (isipv6) {
2890 			if (in6_localaddr(&inp->in6p_faddr)) {
2891 				offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2892 					min_protoh;
2893 			} else {
2894 				offer = tcp_v6mssdflt;
2895 			}
2896 		} else {
2897 			if (in_localaddr(inp->inp_faddr))
2898 				offer = ifp->if_mtu - min_protoh;
2899 			else
2900 				offer = tcp_mssdflt;
2901 		}
2902 	}
2903 
2904 	/*
2905 	 * Prevent DoS attack with too small MSS. Round up
2906 	 * to at least minmss.
2907 	 *
2908 	 * Sanity check: make sure that maxopd will be large
2909 	 * enough to allow some data on segments even is the
2910 	 * all the option space is used (40bytes).  Otherwise
2911 	 * funny things may happen in tcp_output.
2912 	 */
2913 	offer = max(offer, tcp_minmss);
2914 	offer = max(offer, 64);
2915 
2916 	rt->rt_rmx.rmx_mssopt = offer;
2917 
2918 	/*
2919 	 * While we're here, check if there's an initial rtt
2920 	 * or rttvar.  Convert from the route-table units
2921 	 * to scaled multiples of the slow timeout timer.
2922 	 */
2923 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2924 		/*
2925 		 * XXX the lock bit for RTT indicates that the value
2926 		 * is also a minimum value; this is subject to time.
2927 		 */
2928 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2929 			tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2930 		tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2931 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2932 		tcpstat.tcps_usedrtt++;
2933 		if (rt->rt_rmx.rmx_rttvar) {
2934 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2935 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2936 			tcpstat.tcps_usedrttvar++;
2937 		} else {
2938 			/* default variation is +- 1 rtt */
2939 			tp->t_rttvar =
2940 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2941 		}
2942 		TCPT_RANGESET(tp->t_rxtcur,
2943 			      ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2944 			      tp->t_rttmin, TCPTV_REXMTMAX);
2945 	}
2946 
2947 	/*
2948 	 * if there's an mtu associated with the route, use it
2949 	 * else, use the link mtu.  Take the smaller of mss or offer
2950 	 * as our final mss.
2951 	 */
2952 	if (rt->rt_rmx.rmx_mtu) {
2953 		mss = rt->rt_rmx.rmx_mtu - min_protoh;
2954 	} else {
2955 		if (isipv6)
2956 			mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2957 		else
2958 			mss = ifp->if_mtu - min_protoh;
2959 	}
2960 	mss = min(mss, offer);
2961 
2962 	/*
2963 	 * maxopd stores the maximum length of data AND options
2964 	 * in a segment; maxseg is the amount of data in a normal
2965 	 * segment.  We need to store this value (maxopd) apart
2966 	 * from maxseg, because now every segment carries options
2967 	 * and thus we normally have somewhat less data in segments.
2968 	 */
2969 	tp->t_maxopd = mss;
2970 
2971 	if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2972 	    ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2973 		mss -= TCPOLEN_TSTAMP_APPA;
2974 
2975 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
2976 		if (mss > MCLBYTES)
2977 			mss &= ~(MCLBYTES-1);
2978 #else
2979 		if (mss > MCLBYTES)
2980 			mss = mss / MCLBYTES * MCLBYTES;
2981 #endif
2982 	/*
2983 	 * If there's a pipesize, change the socket buffer
2984 	 * to that size.  Make the socket buffers an integral
2985 	 * number of mss units; if the mss is larger than
2986 	 * the socket buffer, decrease the mss.
2987 	 */
2988 #ifdef RTV_SPIPE
2989 	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2990 #endif
2991 		bufsize = so->so_snd.ssb_hiwat;
2992 	if (bufsize < mss)
2993 		mss = bufsize;
2994 	else {
2995 		bufsize = roundup(bufsize, mss);
2996 		if (bufsize > sb_max)
2997 			bufsize = sb_max;
2998 		if (bufsize > so->so_snd.ssb_hiwat)
2999 			ssb_reserve(&so->so_snd, bufsize, so, NULL);
3000 	}
3001 	tp->t_maxseg = mss;
3002 
3003 #ifdef RTV_RPIPE
3004 	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3005 #endif
3006 		bufsize = so->so_rcv.ssb_hiwat;
3007 	if (bufsize > mss) {
3008 		bufsize = roundup(bufsize, mss);
3009 		if (bufsize > sb_max)
3010 			bufsize = sb_max;
3011 		if (bufsize > so->so_rcv.ssb_hiwat) {
3012 			lwkt_gettoken(&so->so_rcv.ssb_token);
3013 			ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3014 			lwkt_reltoken(&so->so_rcv.ssb_token);
3015 		}
3016 	}
3017 
3018 	/*
3019 	 * Set the slow-start flight size
3020 	 *
3021 	 * NOTE: t_maxseg must have been configured!
3022 	 */
3023 	tp->snd_cwnd = tcp_initial_window(tp);
3024 
3025 	if (rt->rt_rmx.rmx_ssthresh) {
3026 		/*
3027 		 * There's some sort of gateway or interface
3028 		 * buffer limit on the path.  Use this to set
3029 		 * the slow start threshhold, but set the
3030 		 * threshold to no less than 2*mss.
3031 		 */
3032 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3033 		tcpstat.tcps_usedssthresh++;
3034 	}
3035 }
3036 
3037 /*
3038  * Determine the MSS option to send on an outgoing SYN.
3039  */
3040 int
3041 tcp_mssopt(struct tcpcb *tp)
3042 {
3043 	struct rtentry *rt;
3044 #ifdef INET6
3045 	boolean_t isipv6 =
3046 	    ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3047 	int min_protoh = isipv6 ?
3048 			     sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3049 			     sizeof(struct tcpiphdr);
3050 #else
3051 	const boolean_t isipv6 = FALSE;
3052 	const size_t min_protoh = sizeof(struct tcpiphdr);
3053 #endif
3054 
3055 	if (isipv6)
3056 		rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3057 	else
3058 		rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3059 	if (rt == NULL)
3060 		return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3061 
3062 	return (rt->rt_ifp->if_mtu - min_protoh);
3063 }
3064 
3065 /*
3066  * When a partial ack arrives, force the retransmission of the
3067  * next unacknowledged segment.  Do not exit Fast Recovery.
3068  *
3069  * Implement the Slow-but-Steady variant of NewReno by restarting the
3070  * the retransmission timer.  Turn it off here so it can be restarted
3071  * later in tcp_output().
3072  */
3073 static void
3074 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3075 {
3076 	tcp_seq old_snd_nxt = tp->snd_nxt;
3077 	u_long ocwnd = tp->snd_cwnd;
3078 
3079 	tcp_callout_stop(tp, tp->tt_rexmt);
3080 	tp->t_rtttime = 0;
3081 	tp->snd_nxt = th->th_ack;
3082 	/* Set snd_cwnd to one segment beyond acknowledged offset. */
3083 	tp->snd_cwnd = tp->t_maxseg;
3084 	tp->t_flags |= TF_ACKNOW;
3085 	tcp_output(tp);
3086 	if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3087 		tp->snd_nxt = old_snd_nxt;
3088 	/* partial window deflation */
3089 	if (ocwnd > acked)
3090 		tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3091 	else
3092 		tp->snd_cwnd = tp->t_maxseg;
3093 }
3094 
3095 /*
3096  * In contrast to the Slow-but-Steady NewReno variant,
3097  * we do not reset the retransmission timer for SACK retransmissions,
3098  * except when retransmitting snd_una.
3099  */
3100 static void
3101 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3102 {
3103 	tcp_seq old_snd_nxt = tp->snd_nxt;
3104 	u_long ocwnd = tp->snd_cwnd;
3105 	uint32_t pipe;
3106 	int nseg = 0;		/* consecutive new segments */
3107 	int nseg_rexmt = 0;	/* retransmitted segments */
3108 #define MAXBURST 4		/* limit burst of new packets on partial ack */
3109 
3110 	tp->t_rtttime = 0;
3111 	pipe = tcp_sack_compute_pipe(tp);
3112 	while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3113 	    (!tcp_do_smartsack || nseg < MAXBURST)) {
3114 		tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3115 		uint32_t sent, seglen;
3116 		boolean_t rescue;
3117 		int error;
3118 
3119 		old_rexmt_high = tp->rexmt_high;
3120 	    	if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3121 			tp->rexmt_high = old_rexmt_high;
3122 			break;
3123 		}
3124 
3125 		/*
3126 		 * If the next tranmission is a rescue retranmission,
3127 		 * we check whether we have already sent some data
3128 		 * (either new segments or retransmitted segments)
3129 		 * into the the network or not.  Since the idea of rescue
3130 		 * retransmission is to sustain ACK clock, as long as
3131 		 * some segments are in the network, ACK clock will be
3132 		 * kept ticking.
3133 		 */
3134 		if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3135 			tp->rexmt_high = old_rexmt_high;
3136 			break;
3137 		}
3138 
3139 		if (nextrexmt == tp->snd_max)
3140 			++nseg;
3141 		else
3142 			++nseg_rexmt;
3143 		tp->snd_nxt = nextrexmt;
3144 		tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3145 		old_snd_max = tp->snd_max;
3146 		if (nextrexmt == tp->snd_una)
3147 			tcp_callout_stop(tp, tp->tt_rexmt);
3148 		error = tcp_output(tp);
3149 		if (error != 0) {
3150 			tp->rexmt_high = old_rexmt_high;
3151 			break;
3152 		}
3153 		sent = tp->snd_nxt - nextrexmt;
3154 		if (sent <= 0) {
3155 			tp->rexmt_high = old_rexmt_high;
3156 			break;
3157 		}
3158 		pipe += sent;
3159 		tcpstat.tcps_sndsackpack++;
3160 		tcpstat.tcps_sndsackbyte += sent;
3161 
3162 		if (rescue) {
3163 			tcpstat.tcps_sackrescue++;
3164 			tp->rexmt_rescue = tp->snd_nxt;
3165 			tp->t_flags |= TF_SACKRESCUED;
3166 			break;
3167 		}
3168 		if (SEQ_LT(nextrexmt, old_snd_max) &&
3169 		    SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3170 			tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3171 			if ((tp->t_flags & TF_SACKRESCUED) &&
3172 			    SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3173 				/* Drag RescueRxt along with HighRxt */
3174 				tp->rexmt_rescue = tp->rexmt_high;
3175 			}
3176 		}
3177 	}
3178 	if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3179 		tp->snd_nxt = old_snd_nxt;
3180 	tp->snd_cwnd = ocwnd;
3181 }
3182 
3183 /*
3184  * Reset idle time and keep-alive timer, typically called when a valid
3185  * tcp packet is received but may also be called when FASTKEEP is set
3186  * to prevent the previous long-timeout from calculating to a drop.
3187  *
3188  * Only update t_rcvtime for non-SYN packets.
3189  *
3190  * Handle the case where one side thinks the connection is established
3191  * but the other side has, say, rebooted without cleaning out the
3192  * connection.   The SYNs could be construed as an attack and wind
3193  * up ignored, but in case it isn't an attack we can validate the
3194  * connection by forcing a keepalive.
3195  */
3196 void
3197 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3198 {
3199 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3200 		if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3201 			tp->t_flags |= TF_KEEPALIVE;
3202 			tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3203 					  tcp_timer_keep);
3204 		} else {
3205 			tp->t_rcvtime = ticks;
3206 			tp->t_flags &= ~TF_KEEPALIVE;
3207 			tcp_callout_reset(tp, tp->tt_keep,
3208 					  tp->t_keepidle,
3209 					  tcp_timer_keep);
3210 		}
3211 	}
3212 }
3213 
3214 static int
3215 tcp_rmx_msl(const struct tcpcb *tp)
3216 {
3217 	struct rtentry *rt;
3218 	struct inpcb *inp = tp->t_inpcb;
3219 	int msl;
3220 #ifdef INET6
3221 	boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3222 #else
3223 	const boolean_t isipv6 = FALSE;
3224 #endif
3225 
3226 	if (isipv6)
3227 		rt = tcp_rtlookup6(&inp->inp_inc);
3228 	else
3229 		rt = tcp_rtlookup(&inp->inp_inc);
3230 	if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3231 		return tcp_msl;
3232 
3233 	msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3234 	if (msl == 0)
3235 		msl = 1;
3236 
3237 	return msl;
3238 }
3239 
3240 static void
3241 tcp_established(struct tcpcb *tp)
3242 {
3243 	tp->t_state = TCPS_ESTABLISHED;
3244 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3245 
3246 	if (tp->t_rxtsyn > 0) {
3247 		/*
3248 		 * RFC6298:
3249 		 * "If the timer expires awaiting the ACK of a SYN segment
3250 		 *  and the TCP implementation is using an RTO less than 3
3251 		 *  seconds, the RTO MUST be re-initialized to 3 seconds
3252 		 *  when data transmission begins"
3253 		 */
3254 		if (tp->t_rxtcur < TCPTV_RTOBASE3)
3255 			tp->t_rxtcur = TCPTV_RTOBASE3;
3256 	}
3257 }
3258