xref: /freebsd/sys/netinet/tcp_input.c (revision 3110d4ebd6c0848cf5e25890d01791bb407e2a9b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2007-2008,2010
7  *	Swinburne University of Technology, Melbourne, Australia.
8  * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9  * Copyright (c) 2010 The FreeBSD Foundation
10  * Copyright (c) 2010-2011 Juniper Networks, Inc.
11  * All rights reserved.
12  *
13  * Portions of this software were developed at the Centre for Advanced Internet
14  * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15  * James Healy and David Hayes, made possible in part by a grant from the Cisco
16  * University Research Program Fund at Community Foundation Silicon Valley.
17  *
18  * Portions of this software were developed at the Centre for Advanced
19  * Internet Architectures, Swinburne University of Technology, Melbourne,
20  * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21  *
22  * Portions of this software were developed by Robert N. M. Watson under
23  * contract to Juniper Networks, Inc.
24  *
25  * Redistribution and use in source and binary forms, with or without
26  * modification, are permitted provided that the following conditions
27  * are met:
28  * 1. Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  * 2. Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in the
32  *    documentation and/or other materials provided with the distribution.
33  * 3. Neither the name of the University nor the names of its contributors
34  *    may be used to endorse or promote products derived from this software
35  *    without specific prior written permission.
36  *
37  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47  * SUCH DAMAGE.
48  *
49  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
50  */
51 
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
54 
55 #include "opt_inet.h"
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
59 
60 #include <sys/param.h>
61 #include <sys/arb.h>
62 #include <sys/kernel.h>
63 #ifdef TCP_HHOOK
64 #include <sys/hhook.h>
65 #endif
66 #include <sys/malloc.h>
67 #include <sys/mbuf.h>
68 #include <sys/proc.h>		/* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
71 #include <sys/sdt.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
79 
80 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
81 
82 #include <vm/uma.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #define TCPSTATES		/* for logging */
90 
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h>	/* required for icmp_var.h */
97 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
116 #ifdef TCPPCAP
117 #include <netinet/tcp_pcap.h>
118 #endif
119 #include <netinet/tcp_syncache.h>
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
123 #ifdef TCP_OFFLOAD
124 #include <netinet/tcp_offload.h>
125 #endif
126 
127 #include <netipsec/ipsec_support.h>
128 
129 #include <machine/in_cksum.h>
130 
131 #include <security/mac/mac_framework.h>
132 
133 const int tcprexmtthresh = 3;
134 
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137     &VNET_NAME(tcp_log_in_vain), 0,
138     "Log all incoming TCP segments to closed ports");
139 
140 VNET_DEFINE(int, blackhole) = 0;
141 #define	V_blackhole		VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143     &VNET_NAME(blackhole), 0,
144     "Do not send RST on segments to closed ports");
145 
146 VNET_DEFINE(int, tcp_delack_enabled) = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
148     &VNET_NAME(tcp_delack_enabled), 0,
149     "Delay ACK to try and piggyback it onto a data packet");
150 
151 VNET_DEFINE(int, drop_synfin) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153     &VNET_NAME(drop_synfin), 0,
154     "Drop TCP packets with SYN+FIN set");
155 
156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
158     &VNET_NAME(tcp_do_prr_conservative), 0,
159     "Do conservative Proportional Rate Reduction");
160 
161 VNET_DEFINE(int, tcp_do_prr) = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
163     &VNET_NAME(tcp_do_prr), 1,
164     "Enable Proportional Rate Reduction per RFC 6937");
165 
166 VNET_DEFINE(int, tcp_do_newcwv) = 0;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
168     &VNET_NAME(tcp_do_newcwv), 0,
169     "Enable New Congestion Window Validation per RFC7661");
170 
171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
173     &VNET_NAME(tcp_do_rfc6675_pipe), 0,
174     "Use calculated pipe/in-flight bytes per RFC 6675");
175 
176 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
178     &VNET_NAME(tcp_do_rfc3042), 0,
179     "Enable RFC 3042 (Limited Transmit)");
180 
181 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
183     &VNET_NAME(tcp_do_rfc3390), 0,
184     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
185 
186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
188     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
189     "Slow-start flight size (initial congestion window) in number of segments");
190 
191 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
193     &VNET_NAME(tcp_do_rfc3465), 0,
194     "Enable RFC 3465 (Appropriate Byte Counting)");
195 
196 VNET_DEFINE(int, tcp_abc_l_var) = 2;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
198     &VNET_NAME(tcp_abc_l_var), 2,
199     "Cap the max cwnd increment during slow-start to this number of segments");
200 
201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
202     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203     "TCP ECN");
204 
205 VNET_DEFINE(int, tcp_do_ecn) = 2;
206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
207     &VNET_NAME(tcp_do_ecn), 0,
208     "TCP ECN support");
209 
210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
212     &VNET_NAME(tcp_ecn_maxretries), 0,
213     "Max retries before giving up on ECN");
214 
215 VNET_DEFINE(int, tcp_insecure_syn) = 0;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
217     &VNET_NAME(tcp_insecure_syn), 0,
218     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
219 
220 VNET_DEFINE(int, tcp_insecure_rst) = 0;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_insecure_rst), 0,
223     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
224 
225 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
226 #define	V_tcp_recvspace	VNET(tcp_recvspace)
227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
228     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
229 
230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
232     &VNET_NAME(tcp_do_autorcvbuf), 0,
233     "Enable automatic receive buffer sizing");
234 
235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
237     &VNET_NAME(tcp_autorcvbuf_max), 0,
238     "Max size of automatic receive buffer");
239 
240 VNET_DEFINE(struct inpcbhead, tcb);
241 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
242 VNET_DEFINE(struct inpcbinfo, tcbinfo);
243 
244 /*
245  * TCP statistics are stored in an array of counter(9)s, which size matches
246  * size of struct tcpstat.  TCP running connection count is a regular array.
247  */
248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
250     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
253     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
254     "TCP connection counts by TCP state");
255 
256 static void
257 tcp_vnet_init(const void *unused)
258 {
259 
260 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
261 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
262 }
263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264     tcp_vnet_init, NULL);
265 
266 #ifdef VIMAGE
267 static void
268 tcp_vnet_uninit(const void *unused)
269 {
270 
271 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
272 	VNET_PCPUSTAT_FREE(tcpstat);
273 }
274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     tcp_vnet_uninit, NULL);
276 #endif /* VIMAGE */
277 
278 /*
279  * Kernel module interface for updating tcpstat.  The first argument is an index
280  * into tcpstat treated as an array.
281  */
282 void
283 kmod_tcpstat_add(int statnum, int val)
284 {
285 
286 	counter_u64_add(VNET(tcpstat)[statnum], val);
287 }
288 
289 #ifdef TCP_HHOOK
290 /*
291  * Wrapper for the TCP established input helper hook.
292  */
293 void
294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
295 {
296 	struct tcp_hhook_data hhook_data;
297 
298 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
299 		hhook_data.tp = tp;
300 		hhook_data.th = th;
301 		hhook_data.to = to;
302 
303 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 		    tp->osd);
305 	}
306 }
307 #endif
308 
309 /*
310  * CC wrapper hook functions
311  */
312 void
313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
314     uint16_t type)
315 {
316 #ifdef STATS
317 	int32_t gput;
318 #endif
319 
320 	INP_WLOCK_ASSERT(tp->t_inpcb);
321 
322 	tp->ccv->nsegs = nsegs;
323 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
324 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
325 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
326 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
327 		tp->ccv->flags |= CCF_CWND_LIMITED;
328 	else
329 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
330 
331 	if (type == CC_ACK) {
332 #ifdef STATS
333 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
334 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
335 		if (!IN_RECOVERY(tp->t_flags))
336 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
337 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
338 		if ((tp->t_flags & TF_GPUTINPROG) &&
339 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
340 			/*
341 			 * Compute goodput in bits per millisecond.
342 			 */
343 			gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
344 			    max(1, tcp_ts_getticks() - tp->gput_ts);
345 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
346 			    gput);
347 			/*
348 			 * XXXLAS: This is a temporary hack, and should be
349 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
350 			 * to deal with chained VOIs.
351 			 */
352 			if (tp->t_stats_gput_prev > 0)
353 				stats_voi_update_abs_s32(tp->t_stats,
354 				    VOI_TCP_GPUT_ND,
355 				    ((gput - tp->t_stats_gput_prev) * 100) /
356 				    tp->t_stats_gput_prev);
357 			tp->t_flags &= ~TF_GPUTINPROG;
358 			tp->t_stats_gput_prev = gput;
359 		}
360 #endif /* STATS */
361 		if (tp->snd_cwnd > tp->snd_ssthresh) {
362 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
363 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
364 				tp->t_bytes_acked -= tp->snd_cwnd;
365 				tp->ccv->flags |= CCF_ABC_SENTAWND;
366 			}
367 		} else {
368 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
369 				tp->t_bytes_acked = 0;
370 		}
371 	}
372 
373 	if (CC_ALGO(tp)->ack_received != NULL) {
374 		/* XXXLAS: Find a way to live without this */
375 		tp->ccv->curack = th->th_ack;
376 		CC_ALGO(tp)->ack_received(tp->ccv, type);
377 	}
378 #ifdef STATS
379 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
380 #endif
381 }
382 
383 void
384 cc_conn_init(struct tcpcb *tp)
385 {
386 	struct hc_metrics_lite metrics;
387 	struct inpcb *inp = tp->t_inpcb;
388 	u_int maxseg;
389 	int rtt;
390 
391 	INP_WLOCK_ASSERT(tp->t_inpcb);
392 
393 	tcp_hc_get(&inp->inp_inc, &metrics);
394 	maxseg = tcp_maxseg(tp);
395 
396 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
397 		tp->t_srtt = rtt;
398 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
399 		TCPSTAT_INC(tcps_usedrtt);
400 		if (metrics.rmx_rttvar) {
401 			tp->t_rttvar = metrics.rmx_rttvar;
402 			TCPSTAT_INC(tcps_usedrttvar);
403 		} else {
404 			/* default variation is +- 1 rtt */
405 			tp->t_rttvar =
406 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
407 		}
408 		TCPT_RANGESET(tp->t_rxtcur,
409 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
410 		    tp->t_rttmin, TCPTV_REXMTMAX);
411 	}
412 	if (metrics.rmx_ssthresh) {
413 		/*
414 		 * There's some sort of gateway or interface
415 		 * buffer limit on the path.  Use this to set
416 		 * the slow start threshold, but set the
417 		 * threshold to no less than 2*mss.
418 		 */
419 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
420 		TCPSTAT_INC(tcps_usedssthresh);
421 	}
422 
423 	/*
424 	 * Set the initial slow-start flight size.
425 	 *
426 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
427 	 * reduce the initial CWND to one segment as congestion is likely
428 	 * requiring us to be cautious.
429 	 */
430 	if (tp->snd_cwnd == 1)
431 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
432 	else
433 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
434 
435 	if (CC_ALGO(tp)->conn_init != NULL)
436 		CC_ALGO(tp)->conn_init(tp->ccv);
437 }
438 
439 void inline
440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
441 {
442 	INP_WLOCK_ASSERT(tp->t_inpcb);
443 
444 #ifdef STATS
445 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
446 #endif
447 
448 	switch(type) {
449 	case CC_NDUPACK:
450 		if (!IN_FASTRECOVERY(tp->t_flags)) {
451 			tp->snd_recover = tp->snd_max;
452 			if (tp->t_flags2 & TF2_ECN_PERMIT)
453 				tp->t_flags2 |= TF2_ECN_SND_CWR;
454 		}
455 		break;
456 	case CC_ECN:
457 		if (!IN_CONGRECOVERY(tp->t_flags) ||
458 		    /*
459 		     * Allow ECN reaction on ACK to CWR, if
460 		     * that data segment was also CE marked.
461 		     */
462 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
463 			EXIT_CONGRECOVERY(tp->t_flags);
464 			TCPSTAT_INC(tcps_ecn_rcwnd);
465 			tp->snd_recover = tp->snd_max + 1;
466 			if (tp->t_flags2 & TF2_ECN_PERMIT)
467 				tp->t_flags2 |= TF2_ECN_SND_CWR;
468 		}
469 		break;
470 	case CC_RTO:
471 		tp->t_dupacks = 0;
472 		tp->t_bytes_acked = 0;
473 		EXIT_RECOVERY(tp->t_flags);
474 		if (tp->t_flags2 & TF2_ECN_PERMIT)
475 			tp->t_flags2 |= TF2_ECN_SND_CWR;
476 		break;
477 	case CC_RTO_ERR:
478 		TCPSTAT_INC(tcps_sndrexmitbad);
479 		/* RTO was unnecessary, so reset everything. */
480 		tp->snd_cwnd = tp->snd_cwnd_prev;
481 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
482 		tp->snd_recover = tp->snd_recover_prev;
483 		if (tp->t_flags & TF_WASFRECOVERY)
484 			ENTER_FASTRECOVERY(tp->t_flags);
485 		if (tp->t_flags & TF_WASCRECOVERY)
486 			ENTER_CONGRECOVERY(tp->t_flags);
487 		tp->snd_nxt = tp->snd_max;
488 		tp->t_flags &= ~TF_PREVVALID;
489 		tp->t_badrxtwin = 0;
490 		break;
491 	}
492 
493 	if (CC_ALGO(tp)->cong_signal != NULL) {
494 		if (th != NULL)
495 			tp->ccv->curack = th->th_ack;
496 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 	}
498 }
499 
500 void inline
501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
502 {
503 	INP_WLOCK_ASSERT(tp->t_inpcb);
504 
505 	/* XXXLAS: KASSERT that we're in recovery? */
506 
507 	if (CC_ALGO(tp)->post_recovery != NULL) {
508 		tp->ccv->curack = th->th_ack;
509 		CC_ALGO(tp)->post_recovery(tp->ccv);
510 	}
511 	/* XXXLAS: EXIT_RECOVERY ? */
512 	tp->t_bytes_acked = 0;
513 	tp->sackhint.recover_fs = 0;
514 }
515 
516 /*
517  * Indicate whether this ack should be delayed.  We can delay the ack if
518  * following conditions are met:
519  *	- There is no delayed ack timer in progress.
520  *	- Our last ack wasn't a 0-sized window. We never want to delay
521  *	  the ack that opens up a 0-sized window.
522  *	- LRO wasn't used for this segment. We make sure by checking that the
523  *	  segment size is not larger than the MSS.
524  */
525 #define DELAY_ACK(tp, tlen)						\
526 	((!tcp_timer_active(tp, TT_DELACK) &&				\
527 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
528 	    (tlen <= tp->t_maxseg) &&					\
529 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
530 
531 void inline
532 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
533 {
534 	INP_WLOCK_ASSERT(tp->t_inpcb);
535 
536 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
537 		switch (iptos & IPTOS_ECN_MASK) {
538 		case IPTOS_ECN_CE:
539 			tp->ccv->flags |= CCF_IPHDR_CE;
540 			break;
541 		case IPTOS_ECN_ECT0:
542 			/* FALLTHROUGH */
543 		case IPTOS_ECN_ECT1:
544 			/* FALLTHROUGH */
545 		case IPTOS_ECN_NOTECT:
546 			tp->ccv->flags &= ~CCF_IPHDR_CE;
547 			break;
548 		}
549 
550 		if (th->th_flags & TH_CWR)
551 			tp->ccv->flags |= CCF_TCPHDR_CWR;
552 		else
553 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
554 
555 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
556 
557 		if (tp->ccv->flags & CCF_ACKNOW) {
558 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
559 			tp->t_flags |= TF_ACKNOW;
560 		}
561 	}
562 }
563 
564 /*
565  * TCP input handling is split into multiple parts:
566  *   tcp6_input is a thin wrapper around tcp_input for the extended
567  *	ip6_protox[] call format in ip6_input
568  *   tcp_input handles primary segment validation, inpcb lookup and
569  *	SYN processing on listen sockets
570  *   tcp_do_segment processes the ACK and text of the segment for
571  *	establishing, established and closing connections
572  */
573 #ifdef INET6
574 int
575 tcp6_input(struct mbuf **mp, int *offp, int proto)
576 {
577 	struct mbuf *m;
578 	struct in6_ifaddr *ia6;
579 	struct ip6_hdr *ip6;
580 
581 	m = *mp;
582 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
583 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
584 		if (m == NULL) {
585 			*mp = m;
586 			TCPSTAT_INC(tcps_rcvshort);
587 			return (IPPROTO_DONE);
588 		}
589 	}
590 
591 	/*
592 	 * draft-itojun-ipv6-tcp-to-anycast
593 	 * better place to put this in?
594 	 */
595 	ip6 = mtod(m, struct ip6_hdr *);
596 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
597 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
598 		ifa_free(&ia6->ia_ifa);
599 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
600 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
601 		*mp = NULL;
602 		return (IPPROTO_DONE);
603 	}
604 	if (ia6)
605 		ifa_free(&ia6->ia_ifa);
606 
607 	*mp = m;
608 	return (tcp_input(mp, offp, proto));
609 }
610 #endif /* INET6 */
611 
612 int
613 tcp_input(struct mbuf **mp, int *offp, int proto)
614 {
615 	struct mbuf *m = *mp;
616 	struct tcphdr *th = NULL;
617 	struct ip *ip = NULL;
618 	struct inpcb *inp = NULL;
619 	struct tcpcb *tp = NULL;
620 	struct socket *so = NULL;
621 	u_char *optp = NULL;
622 	int off0;
623 	int optlen = 0;
624 #ifdef INET
625 	int len;
626 	uint8_t ipttl;
627 #endif
628 	int tlen = 0, off;
629 	int drop_hdrlen;
630 	int thflags;
631 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
632 	uint8_t iptos;
633 	struct m_tag *fwd_tag = NULL;
634 #ifdef INET6
635 	struct ip6_hdr *ip6 = NULL;
636 	int isipv6;
637 #else
638 	const void *ip6 = NULL;
639 #endif /* INET6 */
640 	struct tcpopt to;		/* options in this segment */
641 	char *s = NULL;			/* address and port logging */
642 #ifdef TCPDEBUG
643 	/*
644 	 * The size of tcp_saveipgen must be the size of the max ip header,
645 	 * now IPv6.
646 	 */
647 	u_char tcp_saveipgen[IP6_HDR_LEN];
648 	struct tcphdr tcp_savetcp;
649 	short ostate = 0;
650 #endif
651 
652 	NET_EPOCH_ASSERT();
653 
654 #ifdef INET6
655 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
656 #endif
657 
658 	off0 = *offp;
659 	m = *mp;
660 	*mp = NULL;
661 	to.to_flags = 0;
662 	TCPSTAT_INC(tcps_rcvtotal);
663 
664 #ifdef INET6
665 	if (isipv6) {
666 		ip6 = mtod(m, struct ip6_hdr *);
667 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
668 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
669 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
670 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
671 				th->th_sum = m->m_pkthdr.csum_data;
672 			else
673 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
674 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
675 			th->th_sum ^= 0xffff;
676 		} else
677 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
678 		if (th->th_sum) {
679 			TCPSTAT_INC(tcps_rcvbadsum);
680 			goto drop;
681 		}
682 
683 		/*
684 		 * Be proactive about unspecified IPv6 address in source.
685 		 * As we use all-zero to indicate unbounded/unconnected pcb,
686 		 * unspecified IPv6 address can be used to confuse us.
687 		 *
688 		 * Note that packets with unspecified IPv6 destination is
689 		 * already dropped in ip6_input.
690 		 */
691 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
692 			/* XXX stat */
693 			goto drop;
694 		}
695 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
696 	}
697 #endif
698 #if defined(INET) && defined(INET6)
699 	else
700 #endif
701 #ifdef INET
702 	{
703 		/*
704 		 * Get IP and TCP header together in first mbuf.
705 		 * Note: IP leaves IP header in first mbuf.
706 		 */
707 		if (off0 > sizeof (struct ip)) {
708 			ip_stripoptions(m);
709 			off0 = sizeof(struct ip);
710 		}
711 		if (m->m_len < sizeof (struct tcpiphdr)) {
712 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
713 			    == NULL) {
714 				TCPSTAT_INC(tcps_rcvshort);
715 				return (IPPROTO_DONE);
716 			}
717 		}
718 		ip = mtod(m, struct ip *);
719 		th = (struct tcphdr *)((caddr_t)ip + off0);
720 		tlen = ntohs(ip->ip_len) - off0;
721 
722 		iptos = ip->ip_tos;
723 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
724 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
725 				th->th_sum = m->m_pkthdr.csum_data;
726 			else
727 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
728 				    ip->ip_dst.s_addr,
729 				    htonl(m->m_pkthdr.csum_data + tlen +
730 				    IPPROTO_TCP));
731 			th->th_sum ^= 0xffff;
732 		} else {
733 			struct ipovly *ipov = (struct ipovly *)ip;
734 
735 			/*
736 			 * Checksum extended TCP header and data.
737 			 */
738 			len = off0 + tlen;
739 			ipttl = ip->ip_ttl;
740 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
741 			ipov->ih_len = htons(tlen);
742 			th->th_sum = in_cksum(m, len);
743 			/* Reset length for SDT probes. */
744 			ip->ip_len = htons(len);
745 			/* Reset TOS bits */
746 			ip->ip_tos = iptos;
747 			/* Re-initialization for later version check */
748 			ip->ip_ttl = ipttl;
749 			ip->ip_v = IPVERSION;
750 			ip->ip_hl = off0 >> 2;
751 		}
752 
753 		if (th->th_sum) {
754 			TCPSTAT_INC(tcps_rcvbadsum);
755 			goto drop;
756 		}
757 	}
758 #endif /* INET */
759 
760 	/*
761 	 * Check that TCP offset makes sense,
762 	 * pull out TCP options and adjust length.		XXX
763 	 */
764 	off = th->th_off << 2;
765 	if (off < sizeof (struct tcphdr) || off > tlen) {
766 		TCPSTAT_INC(tcps_rcvbadoff);
767 		goto drop;
768 	}
769 	tlen -= off;	/* tlen is used instead of ti->ti_len */
770 	if (off > sizeof (struct tcphdr)) {
771 #ifdef INET6
772 		if (isipv6) {
773 			if (m->m_len < off0 + off) {
774 				m = m_pullup(m, off0 + off);
775 				if (m == NULL) {
776 					TCPSTAT_INC(tcps_rcvshort);
777 					return (IPPROTO_DONE);
778 				}
779 			}
780 			ip6 = mtod(m, struct ip6_hdr *);
781 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
782 		}
783 #endif
784 #if defined(INET) && defined(INET6)
785 		else
786 #endif
787 #ifdef INET
788 		{
789 			if (m->m_len < sizeof(struct ip) + off) {
790 				if ((m = m_pullup(m, sizeof (struct ip) + off))
791 				    == NULL) {
792 					TCPSTAT_INC(tcps_rcvshort);
793 					return (IPPROTO_DONE);
794 				}
795 				ip = mtod(m, struct ip *);
796 				th = (struct tcphdr *)((caddr_t)ip + off0);
797 			}
798 		}
799 #endif
800 		optlen = off - sizeof (struct tcphdr);
801 		optp = (u_char *)(th + 1);
802 	}
803 	thflags = th->th_flags;
804 
805 	/*
806 	 * Convert TCP protocol specific fields to host format.
807 	 */
808 	tcp_fields_to_host(th);
809 
810 	/*
811 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
812 	 */
813 	drop_hdrlen = off0 + off;
814 
815 	/*
816 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
817 	 */
818         if (
819 #ifdef INET6
820 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
821 #ifdef INET
822 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
823 #endif
824 #endif
825 #if defined(INET) && !defined(INET6)
826 	    (m->m_flags & M_IP_NEXTHOP)
827 #endif
828 	    )
829 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
830 
831 findpcb:
832 #ifdef INET6
833 	if (isipv6 && fwd_tag != NULL) {
834 		struct sockaddr_in6 *next_hop6;
835 
836 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
837 		/*
838 		 * Transparently forwarded. Pretend to be the destination.
839 		 * Already got one like this?
840 		 */
841 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
842 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
843 		    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
844 		if (!inp) {
845 			/*
846 			 * It's new.  Try to find the ambushing socket.
847 			 * Because we've rewritten the destination address,
848 			 * any hardware-generated hash is ignored.
849 			 */
850 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
851 			    th->th_sport, &next_hop6->sin6_addr,
852 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
853 			    th->th_dport, INPLOOKUP_WILDCARD |
854 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
855 		}
856 	} else if (isipv6) {
857 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
858 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
859 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
860 		    m->m_pkthdr.rcvif, m);
861 	}
862 #endif /* INET6 */
863 #if defined(INET6) && defined(INET)
864 	else
865 #endif
866 #ifdef INET
867 	if (fwd_tag != NULL) {
868 		struct sockaddr_in *next_hop;
869 
870 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
871 		/*
872 		 * Transparently forwarded. Pretend to be the destination.
873 		 * already got one like this?
874 		 */
875 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
876 		    ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
877 		    m->m_pkthdr.rcvif, m);
878 		if (!inp) {
879 			/*
880 			 * It's new.  Try to find the ambushing socket.
881 			 * Because we've rewritten the destination address,
882 			 * any hardware-generated hash is ignored.
883 			 */
884 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
885 			    th->th_sport, next_hop->sin_addr,
886 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
887 			    th->th_dport, INPLOOKUP_WILDCARD |
888 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
889 		}
890 	} else
891 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
892 		    th->th_sport, ip->ip_dst, th->th_dport,
893 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
894 		    m->m_pkthdr.rcvif, m);
895 #endif /* INET */
896 
897 	/*
898 	 * If the INPCB does not exist then all data in the incoming
899 	 * segment is discarded and an appropriate RST is sent back.
900 	 * XXX MRT Send RST using which routing table?
901 	 */
902 	if (inp == NULL) {
903 		/*
904 		 * Log communication attempts to ports that are not
905 		 * in use.
906 		 */
907 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
908 		    V_tcp_log_in_vain == 2) {
909 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
910 				log(LOG_INFO, "%s; %s: Connection attempt "
911 				    "to closed port\n", s, __func__);
912 		}
913 		/*
914 		 * When blackholing do not respond with a RST but
915 		 * completely ignore the segment and drop it.
916 		 */
917 		if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
918 		    V_blackhole == 2)
919 			goto dropunlock;
920 
921 		rstreason = BANDLIM_RST_CLOSEDPORT;
922 		goto dropwithreset;
923 	}
924 	INP_WLOCK_ASSERT(inp);
925 	/*
926 	 * While waiting for inp lock during the lookup, another thread
927 	 * can have dropped the inpcb, in which case we need to loop back
928 	 * and try to find a new inpcb to deliver to.
929 	 */
930 	if (inp->inp_flags & INP_DROPPED) {
931 		INP_WUNLOCK(inp);
932 		inp = NULL;
933 		goto findpcb;
934 	}
935 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
936 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
937 	    ((inp->inp_socket == NULL) ||
938 	    (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
939 		inp->inp_flowid = m->m_pkthdr.flowid;
940 		inp->inp_flowtype = M_HASHTYPE_GET(m);
941 	}
942 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
943 #ifdef INET6
944 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
945 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
946 		goto dropunlock;
947 	}
948 #ifdef INET
949 	else
950 #endif
951 #endif /* INET6 */
952 #ifdef INET
953 	if (IPSEC_ENABLED(ipv4) &&
954 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
955 		goto dropunlock;
956 	}
957 #endif /* INET */
958 #endif /* IPSEC */
959 
960 	/*
961 	 * Check the minimum TTL for socket.
962 	 */
963 	if (inp->inp_ip_minttl != 0) {
964 #ifdef INET6
965 		if (isipv6) {
966 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
967 				goto dropunlock;
968 		} else
969 #endif
970 		if (inp->inp_ip_minttl > ip->ip_ttl)
971 			goto dropunlock;
972 	}
973 
974 	/*
975 	 * A previous connection in TIMEWAIT state is supposed to catch stray
976 	 * or duplicate segments arriving late.  If this segment was a
977 	 * legitimate new connection attempt, the old INPCB gets removed and
978 	 * we can try again to find a listening socket.
979 	 *
980 	 * At this point, due to earlier optimism, we may hold only an inpcb
981 	 * lock, and not the inpcbinfo write lock.  If so, we need to try to
982 	 * acquire it, or if that fails, acquire a reference on the inpcb,
983 	 * drop all locks, acquire a global write lock, and then re-acquire
984 	 * the inpcb lock.  We may at that point discover that another thread
985 	 * has tried to free the inpcb, in which case we need to loop back
986 	 * and try to find a new inpcb to deliver to.
987 	 *
988 	 * XXXRW: It may be time to rethink timewait locking.
989 	 */
990 	if (inp->inp_flags & INP_TIMEWAIT) {
991 		tcp_dooptions(&to, optp, optlen,
992 		    (thflags & TH_SYN) ? TO_SYN : 0);
993 		/*
994 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
995 		 */
996 		if (tcp_twcheck(inp, &to, th, m, tlen))
997 			goto findpcb;
998 		return (IPPROTO_DONE);
999 	}
1000 	/*
1001 	 * The TCPCB may no longer exist if the connection is winding
1002 	 * down or it is in the CLOSED state.  Either way we drop the
1003 	 * segment and send an appropriate response.
1004 	 */
1005 	tp = intotcpcb(inp);
1006 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1007 		rstreason = BANDLIM_RST_CLOSEDPORT;
1008 		goto dropwithreset;
1009 	}
1010 
1011 #ifdef TCP_OFFLOAD
1012 	if (tp->t_flags & TF_TOE) {
1013 		tcp_offload_input(tp, m);
1014 		m = NULL;	/* consumed by the TOE driver */
1015 		goto dropunlock;
1016 	}
1017 #endif
1018 
1019 #ifdef MAC
1020 	INP_WLOCK_ASSERT(inp);
1021 	if (mac_inpcb_check_deliver(inp, m))
1022 		goto dropunlock;
1023 #endif
1024 	so = inp->inp_socket;
1025 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1026 #ifdef TCPDEBUG
1027 	if (so->so_options & SO_DEBUG) {
1028 		ostate = tp->t_state;
1029 #ifdef INET6
1030 		if (isipv6) {
1031 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1032 		} else
1033 #endif
1034 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1035 		tcp_savetcp = *th;
1036 	}
1037 #endif /* TCPDEBUG */
1038 	/*
1039 	 * When the socket is accepting connections (the INPCB is in LISTEN
1040 	 * state) we look into the SYN cache if this is a new connection
1041 	 * attempt or the completion of a previous one.
1042 	 */
1043 	KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1044 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1045 	if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1046 		struct in_conninfo inc;
1047 
1048 		bzero(&inc, sizeof(inc));
1049 #ifdef INET6
1050 		if (isipv6) {
1051 			inc.inc_flags |= INC_ISIPV6;
1052 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1053 				inc.inc_flags |= INC_IPV6MINMTU;
1054 			inc.inc6_faddr = ip6->ip6_src;
1055 			inc.inc6_laddr = ip6->ip6_dst;
1056 		} else
1057 #endif
1058 		{
1059 			inc.inc_faddr = ip->ip_src;
1060 			inc.inc_laddr = ip->ip_dst;
1061 		}
1062 		inc.inc_fport = th->th_sport;
1063 		inc.inc_lport = th->th_dport;
1064 		inc.inc_fibnum = so->so_fibnum;
1065 
1066 		/*
1067 		 * Check for an existing connection attempt in syncache if
1068 		 * the flag is only ACK.  A successful lookup creates a new
1069 		 * socket appended to the listen queue in SYN_RECEIVED state.
1070 		 */
1071 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1072 			/*
1073 			 * Parse the TCP options here because
1074 			 * syncookies need access to the reflected
1075 			 * timestamp.
1076 			 */
1077 			tcp_dooptions(&to, optp, optlen, 0);
1078 			/*
1079 			 * NB: syncache_expand() doesn't unlock
1080 			 * inp and tcpinfo locks.
1081 			 */
1082 			rstreason = syncache_expand(&inc, &to, th, &so, m);
1083 			if (rstreason < 0) {
1084 				/*
1085 				 * A failing TCP MD5 signature comparison
1086 				 * must result in the segment being dropped
1087 				 * and must not produce any response back
1088 				 * to the sender.
1089 				 */
1090 				goto dropunlock;
1091 			} else if (rstreason == 0) {
1092 				/*
1093 				 * No syncache entry or ACK was not
1094 				 * for our SYN/ACK.  Send a RST.
1095 				 * NB: syncache did its own logging
1096 				 * of the failure cause.
1097 				 */
1098 				rstreason = BANDLIM_RST_OPENPORT;
1099 				goto dropwithreset;
1100 			}
1101 tfo_socket_result:
1102 			if (so == NULL) {
1103 				/*
1104 				 * We completed the 3-way handshake
1105 				 * but could not allocate a socket
1106 				 * either due to memory shortage,
1107 				 * listen queue length limits or
1108 				 * global socket limits.  Send RST
1109 				 * or wait and have the remote end
1110 				 * retransmit the ACK for another
1111 				 * try.
1112 				 */
1113 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1114 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1115 					    "Socket allocation failed due to "
1116 					    "limits or memory shortage, %s\n",
1117 					    s, __func__,
1118 					    V_tcp_sc_rst_sock_fail ?
1119 					    "sending RST" : "try again");
1120 				if (V_tcp_sc_rst_sock_fail) {
1121 					rstreason = BANDLIM_UNLIMITED;
1122 					goto dropwithreset;
1123 				} else
1124 					goto dropunlock;
1125 			}
1126 			/*
1127 			 * Socket is created in state SYN_RECEIVED.
1128 			 * Unlock the listen socket, lock the newly
1129 			 * created socket and update the tp variable.
1130 			 */
1131 			INP_WUNLOCK(inp);	/* listen socket */
1132 			inp = sotoinpcb(so);
1133 			/*
1134 			 * New connection inpcb is already locked by
1135 			 * syncache_expand().
1136 			 */
1137 			INP_WLOCK_ASSERT(inp);
1138 			tp = intotcpcb(inp);
1139 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1140 			    ("%s: ", __func__));
1141 			/*
1142 			 * Process the segment and the data it
1143 			 * contains.  tcp_do_segment() consumes
1144 			 * the mbuf chain and unlocks the inpcb.
1145 			 */
1146 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1147 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1148 			    iptos);
1149 			return (IPPROTO_DONE);
1150 		}
1151 		/*
1152 		 * Segment flag validation for new connection attempts:
1153 		 *
1154 		 * Our (SYN|ACK) response was rejected.
1155 		 * Check with syncache and remove entry to prevent
1156 		 * retransmits.
1157 		 *
1158 		 * NB: syncache_chkrst does its own logging of failure
1159 		 * causes.
1160 		 */
1161 		if (thflags & TH_RST) {
1162 			syncache_chkrst(&inc, th, m);
1163 			goto dropunlock;
1164 		}
1165 		/*
1166 		 * We can't do anything without SYN.
1167 		 */
1168 		if ((thflags & TH_SYN) == 0) {
1169 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 				    "SYN is missing, segment ignored\n",
1172 				    s, __func__);
1173 			TCPSTAT_INC(tcps_badsyn);
1174 			goto dropunlock;
1175 		}
1176 		/*
1177 		 * (SYN|ACK) is bogus on a listen socket.
1178 		 */
1179 		if (thflags & TH_ACK) {
1180 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1181 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1182 				    "SYN|ACK invalid, segment rejected\n",
1183 				    s, __func__);
1184 			syncache_badack(&inc);	/* XXX: Not needed! */
1185 			TCPSTAT_INC(tcps_badsyn);
1186 			rstreason = BANDLIM_RST_OPENPORT;
1187 			goto dropwithreset;
1188 		}
1189 		/*
1190 		 * If the drop_synfin option is enabled, drop all
1191 		 * segments with both the SYN and FIN bits set.
1192 		 * This prevents e.g. nmap from identifying the
1193 		 * TCP/IP stack.
1194 		 * XXX: Poor reasoning.  nmap has other methods
1195 		 * and is constantly refining its stack detection
1196 		 * strategies.
1197 		 * XXX: This is a violation of the TCP specification
1198 		 * and was used by RFC1644.
1199 		 */
1200 		if ((thflags & TH_FIN) && V_drop_synfin) {
1201 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1202 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1203 				    "SYN|FIN segment ignored (based on "
1204 				    "sysctl setting)\n", s, __func__);
1205 			TCPSTAT_INC(tcps_badsyn);
1206 			goto dropunlock;
1207 		}
1208 		/*
1209 		 * Segment's flags are (SYN) or (SYN|FIN).
1210 		 *
1211 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1212 		 * as they do not affect the state of the TCP FSM.
1213 		 * The data pointed to by TH_URG and th_urp is ignored.
1214 		 */
1215 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1216 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1217 		KASSERT(thflags & (TH_SYN),
1218 		    ("%s: Listen socket: TH_SYN not set", __func__));
1219 #ifdef INET6
1220 		/*
1221 		 * If deprecated address is forbidden,
1222 		 * we do not accept SYN to deprecated interface
1223 		 * address to prevent any new inbound connection from
1224 		 * getting established.
1225 		 * When we do not accept SYN, we send a TCP RST,
1226 		 * with deprecated source address (instead of dropping
1227 		 * it).  We compromise it as it is much better for peer
1228 		 * to send a RST, and RST will be the final packet
1229 		 * for the exchange.
1230 		 *
1231 		 * If we do not forbid deprecated addresses, we accept
1232 		 * the SYN packet.  RFC2462 does not suggest dropping
1233 		 * SYN in this case.
1234 		 * If we decipher RFC2462 5.5.4, it says like this:
1235 		 * 1. use of deprecated addr with existing
1236 		 *    communication is okay - "SHOULD continue to be
1237 		 *    used"
1238 		 * 2. use of it with new communication:
1239 		 *   (2a) "SHOULD NOT be used if alternate address
1240 		 *        with sufficient scope is available"
1241 		 *   (2b) nothing mentioned otherwise.
1242 		 * Here we fall into (2b) case as we have no choice in
1243 		 * our source address selection - we must obey the peer.
1244 		 *
1245 		 * The wording in RFC2462 is confusing, and there are
1246 		 * multiple description text for deprecated address
1247 		 * handling - worse, they are not exactly the same.
1248 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1249 		 */
1250 		if (isipv6 && !V_ip6_use_deprecated) {
1251 			struct in6_ifaddr *ia6;
1252 
1253 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1254 			if (ia6 != NULL &&
1255 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1256 				ifa_free(&ia6->ia_ifa);
1257 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1258 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1259 					"Connection attempt to deprecated "
1260 					"IPv6 address rejected\n",
1261 					s, __func__);
1262 				rstreason = BANDLIM_RST_OPENPORT;
1263 				goto dropwithreset;
1264 			}
1265 			if (ia6)
1266 				ifa_free(&ia6->ia_ifa);
1267 		}
1268 #endif /* INET6 */
1269 		/*
1270 		 * Basic sanity checks on incoming SYN requests:
1271 		 *   Don't respond if the destination is a link layer
1272 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1273 		 *   If it is from this socket it must be forged.
1274 		 *   Don't respond if the source or destination is a
1275 		 *	global or subnet broad- or multicast address.
1276 		 *   Note that it is quite possible to receive unicast
1277 		 *	link-layer packets with a broadcast IP address. Use
1278 		 *	in_broadcast() to find them.
1279 		 */
1280 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1281 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1282 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1283 				"Connection attempt from broad- or multicast "
1284 				"link layer address ignored\n", s, __func__);
1285 			goto dropunlock;
1286 		}
1287 #ifdef INET6
1288 		if (isipv6) {
1289 			if (th->th_dport == th->th_sport &&
1290 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1291 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1292 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1293 					"Connection attempt to/from self "
1294 					"ignored\n", s, __func__);
1295 				goto dropunlock;
1296 			}
1297 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1298 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1299 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1300 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1301 					"Connection attempt from/to multicast "
1302 					"address ignored\n", s, __func__);
1303 				goto dropunlock;
1304 			}
1305 		}
1306 #endif
1307 #if defined(INET) && defined(INET6)
1308 		else
1309 #endif
1310 #ifdef INET
1311 		{
1312 			if (th->th_dport == th->th_sport &&
1313 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1314 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1315 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1316 					"Connection attempt from/to self "
1317 					"ignored\n", s, __func__);
1318 				goto dropunlock;
1319 			}
1320 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1321 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1322 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1323 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1324 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1325 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1326 					"Connection attempt from/to broad- "
1327 					"or multicast address ignored\n",
1328 					s, __func__);
1329 				goto dropunlock;
1330 			}
1331 		}
1332 #endif
1333 		/*
1334 		 * SYN appears to be valid.  Create compressed TCP state
1335 		 * for syncache.
1336 		 */
1337 #ifdef TCPDEBUG
1338 		if (so->so_options & SO_DEBUG)
1339 			tcp_trace(TA_INPUT, ostate, tp,
1340 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1341 #endif
1342 		TCP_PROBE3(debug__input, tp, th, m);
1343 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1344 		if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1345 			goto tfo_socket_result;
1346 
1347 		/*
1348 		 * Entry added to syncache and mbuf consumed.
1349 		 * Only the listen socket is unlocked by syncache_add().
1350 		 */
1351 		INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1352 		return (IPPROTO_DONE);
1353 	} else if (tp->t_state == TCPS_LISTEN) {
1354 		/*
1355 		 * When a listen socket is torn down the SO_ACCEPTCONN
1356 		 * flag is removed first while connections are drained
1357 		 * from the accept queue in a unlock/lock cycle of the
1358 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1359 		 * attempt go through unhandled.
1360 		 */
1361 		goto dropunlock;
1362 	}
1363 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1364 	if (tp->t_flags & TF_SIGNATURE) {
1365 		tcp_dooptions(&to, optp, optlen, thflags);
1366 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1367 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1368 			goto dropunlock;
1369 		}
1370 		if (!TCPMD5_ENABLED() ||
1371 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1372 			goto dropunlock;
1373 	}
1374 #endif
1375 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1376 
1377 	/*
1378 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1379 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1380 	 * the inpcb, and unlocks pcbinfo.
1381 	 */
1382 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1383 	return (IPPROTO_DONE);
1384 
1385 dropwithreset:
1386 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1387 
1388 	if (inp != NULL) {
1389 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1390 		INP_WUNLOCK(inp);
1391 	} else
1392 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1393 	m = NULL;	/* mbuf chain got consumed. */
1394 	goto drop;
1395 
1396 dropunlock:
1397 	if (m != NULL)
1398 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1399 
1400 	if (inp != NULL)
1401 		INP_WUNLOCK(inp);
1402 
1403 drop:
1404 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1405 	if (s != NULL)
1406 		free(s, M_TCPLOG);
1407 	if (m != NULL)
1408 		m_freem(m);
1409 	return (IPPROTO_DONE);
1410 }
1411 
1412 /*
1413  * Automatic sizing of receive socket buffer.  Often the send
1414  * buffer size is not optimally adjusted to the actual network
1415  * conditions at hand (delay bandwidth product).  Setting the
1416  * buffer size too small limits throughput on links with high
1417  * bandwidth and high delay (eg. trans-continental/oceanic links).
1418  *
1419  * On the receive side the socket buffer memory is only rarely
1420  * used to any significant extent.  This allows us to be much
1421  * more aggressive in scaling the receive socket buffer.  For
1422  * the case that the buffer space is actually used to a large
1423  * extent and we run out of kernel memory we can simply drop
1424  * the new segments; TCP on the sender will just retransmit it
1425  * later.  Setting the buffer size too big may only consume too
1426  * much kernel memory if the application doesn't read() from
1427  * the socket or packet loss or reordering makes use of the
1428  * reassembly queue.
1429  *
1430  * The criteria to step up the receive buffer one notch are:
1431  *  1. Application has not set receive buffer size with
1432  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1433  *  2. the number of bytes received during 1/2 of an sRTT
1434  *     is at least 3/8 of the current socket buffer size.
1435  *  3. receive buffer size has not hit maximal automatic size;
1436  *
1437  * If all of the criteria are met we increaset the socket buffer
1438  * by a 1/2 (bounded by the max). This allows us to keep ahead
1439  * of slow-start but also makes it so our peer never gets limited
1440  * by our rwnd which we then open up causing a burst.
1441  *
1442  * This algorithm does two steps per RTT at most and only if
1443  * we receive a bulk stream w/o packet losses or reorderings.
1444  * Shrinking the buffer during idle times is not necessary as
1445  * it doesn't consume any memory when idle.
1446  *
1447  * TODO: Only step up if the application is actually serving
1448  * the buffer to better manage the socket buffer resources.
1449  */
1450 int
1451 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1452     struct tcpcb *tp, int tlen)
1453 {
1454 	int newsize = 0;
1455 
1456 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1457 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1458 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1459 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1460 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1461 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1462 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1463 		}
1464 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1465 
1466 		/* Start over with next RTT. */
1467 		tp->rfbuf_ts = 0;
1468 		tp->rfbuf_cnt = 0;
1469 	} else {
1470 		tp->rfbuf_cnt += tlen;	/* add up */
1471 	}
1472 	return (newsize);
1473 }
1474 
1475 void
1476 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1477 {
1478 	/*
1479 	 * Since tp might be gone if the session entered
1480 	 * the TIME_WAIT state before coming here, we need
1481 	 * to check if the socket is still connected.
1482 	 */
1483 	if ((so->so_state & SS_ISCONNECTED) == 0)
1484 		return;
1485 	INP_LOCK_ASSERT(tp->t_inpcb);
1486 	if (tp->t_flags & TF_WAKESOR) {
1487 		tp->t_flags &= ~TF_WAKESOR;
1488 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1489 		sorwakeup(so);
1490 	}
1491 	if (tp->t_flags & TF_WAKESOW) {
1492 		tp->t_flags &= ~TF_WAKESOW;
1493 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1494 		sowwakeup(so);
1495 	}
1496 }
1497 
1498 void
1499 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1500     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1501 {
1502 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1503 	int rstreason, todrop, win, incforsyn = 0;
1504 	uint32_t tiwin;
1505 	uint16_t nsegs;
1506 	char *s;
1507 	struct in_conninfo *inc;
1508 	struct mbuf *mfree;
1509 	struct tcpopt to;
1510 	int tfo_syn;
1511 
1512 #ifdef TCPDEBUG
1513 	/*
1514 	 * The size of tcp_saveipgen must be the size of the max ip header,
1515 	 * now IPv6.
1516 	 */
1517 	u_char tcp_saveipgen[IP6_HDR_LEN];
1518 	struct tcphdr tcp_savetcp;
1519 	short ostate = 0;
1520 #endif
1521 	thflags = th->th_flags;
1522 	inc = &tp->t_inpcb->inp_inc;
1523 	tp->sackhint.last_sack_ack = 0;
1524 	sack_changed = 0;
1525 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1526 
1527 	NET_EPOCH_ASSERT();
1528 	INP_WLOCK_ASSERT(tp->t_inpcb);
1529 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1530 	    __func__));
1531 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1532 	    __func__));
1533 
1534 #ifdef TCPPCAP
1535 	/* Save segment, if requested. */
1536 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1537 #endif
1538 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1539 	    tlen, NULL, true);
1540 
1541 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1542 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1543 			log(LOG_DEBUG, "%s; %s: "
1544 			    "SYN|FIN segment ignored (based on "
1545 			    "sysctl setting)\n", s, __func__);
1546 			free(s, M_TCPLOG);
1547 		}
1548 		goto drop;
1549 	}
1550 
1551 	/*
1552 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1553 	 * check SEQ.ACK first.
1554 	 */
1555 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1556 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1557 		rstreason = BANDLIM_UNLIMITED;
1558 		goto dropwithreset;
1559 	}
1560 
1561 	/*
1562 	 * Segment received on connection.
1563 	 * Reset idle time and keep-alive timer.
1564 	 * XXX: This should be done after segment
1565 	 * validation to ignore broken/spoofed segs.
1566 	 */
1567 	tp->t_rcvtime = ticks;
1568 
1569 	/*
1570 	 * Scale up the window into a 32-bit value.
1571 	 * For the SYN_SENT state the scale is zero.
1572 	 */
1573 	tiwin = th->th_win << tp->snd_scale;
1574 #ifdef STATS
1575 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1576 #endif
1577 
1578 	/*
1579 	 * TCP ECN processing.
1580 	 */
1581 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1582 		if (thflags & TH_CWR) {
1583 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1584 			tp->t_flags |= TF_ACKNOW;
1585 		}
1586 		switch (iptos & IPTOS_ECN_MASK) {
1587 		case IPTOS_ECN_CE:
1588 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1589 			TCPSTAT_INC(tcps_ecn_ce);
1590 			break;
1591 		case IPTOS_ECN_ECT0:
1592 			TCPSTAT_INC(tcps_ecn_ect0);
1593 			break;
1594 		case IPTOS_ECN_ECT1:
1595 			TCPSTAT_INC(tcps_ecn_ect1);
1596 			break;
1597 		}
1598 
1599 		/* Process a packet differently from RFC3168. */
1600 		cc_ecnpkt_handler(tp, th, iptos);
1601 
1602 		/* Congestion experienced. */
1603 		if (thflags & TH_ECE) {
1604 			cc_cong_signal(tp, th, CC_ECN);
1605 		}
1606 	}
1607 
1608 	/*
1609 	 * Parse options on any incoming segment.
1610 	 */
1611 	tcp_dooptions(&to, (u_char *)(th + 1),
1612 	    (th->th_off << 2) - sizeof(struct tcphdr),
1613 	    (thflags & TH_SYN) ? TO_SYN : 0);
1614 
1615 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1616 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1617 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1618 		TCPSTAT_INC(tcps_sig_err_sigopt);
1619 		/* XXX: should drop? */
1620 	}
1621 #endif
1622 	/*
1623 	 * If echoed timestamp is later than the current time,
1624 	 * fall back to non RFC1323 RTT calculation.  Normalize
1625 	 * timestamp if syncookies were used when this connection
1626 	 * was established.
1627 	 */
1628 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1629 		to.to_tsecr -= tp->ts_offset;
1630 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1631 			to.to_tsecr = 0;
1632 		else if (tp->t_flags & TF_PREVVALID &&
1633 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1634 			cc_cong_signal(tp, th, CC_RTO_ERR);
1635 	}
1636 	/*
1637 	 * Process options only when we get SYN/ACK back. The SYN case
1638 	 * for incoming connections is handled in tcp_syncache.
1639 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1640 	 * or <SYN,ACK>) segment itself is never scaled.
1641 	 * XXX this is traditional behavior, may need to be cleaned up.
1642 	 */
1643 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1644 		/* Handle parallel SYN for ECN */
1645 		if (!(thflags & TH_ACK) &&
1646 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1647 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1648 			tp->t_flags2 |= TF2_ECN_PERMIT;
1649 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1650 			TCPSTAT_INC(tcps_ecn_shs);
1651 		}
1652 		if ((to.to_flags & TOF_SCALE) &&
1653 		    (tp->t_flags & TF_REQ_SCALE)) {
1654 			tp->t_flags |= TF_RCVD_SCALE;
1655 			tp->snd_scale = to.to_wscale;
1656 		} else
1657 			tp->t_flags &= ~TF_REQ_SCALE;
1658 		/*
1659 		 * Initial send window.  It will be updated with
1660 		 * the next incoming segment to the scaled value.
1661 		 */
1662 		tp->snd_wnd = th->th_win;
1663 		if ((to.to_flags & TOF_TS) &&
1664 		    (tp->t_flags & TF_REQ_TSTMP)) {
1665 			tp->t_flags |= TF_RCVD_TSTMP;
1666 			tp->ts_recent = to.to_tsval;
1667 			tp->ts_recent_age = tcp_ts_getticks();
1668 		} else
1669 			tp->t_flags &= ~TF_REQ_TSTMP;
1670 		if (to.to_flags & TOF_MSS)
1671 			tcp_mss(tp, to.to_mss);
1672 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1673 		    (to.to_flags & TOF_SACKPERM) == 0)
1674 			tp->t_flags &= ~TF_SACK_PERMIT;
1675 		if (IS_FASTOPEN(tp->t_flags)) {
1676 			if (to.to_flags & TOF_FASTOPEN) {
1677 				uint16_t mss;
1678 
1679 				if (to.to_flags & TOF_MSS)
1680 					mss = to.to_mss;
1681 				else
1682 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1683 						mss = TCP6_MSS;
1684 					else
1685 						mss = TCP_MSS;
1686 				tcp_fastopen_update_cache(tp, mss,
1687 				    to.to_tfo_len, to.to_tfo_cookie);
1688 			} else
1689 				tcp_fastopen_disable_path(tp);
1690 		}
1691 	}
1692 
1693 	/*
1694 	 * If timestamps were negotiated during SYN/ACK and a
1695 	 * segment without a timestamp is received, silently drop
1696 	 * the segment, unless it is a RST segment or missing timestamps are
1697 	 * tolerated.
1698 	 * See section 3.2 of RFC 7323.
1699 	 */
1700 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1701 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1702 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1703 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1704 				    "segment processed normally\n",
1705 				    s, __func__);
1706 				free(s, M_TCPLOG);
1707 			}
1708 		} else {
1709 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1710 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1711 				    "segment silently dropped\n", s, __func__);
1712 				free(s, M_TCPLOG);
1713 			}
1714 			goto drop;
1715 		}
1716 	}
1717 	/*
1718 	 * If timestamps were not negotiated during SYN/ACK and a
1719 	 * segment with a timestamp is received, ignore the
1720 	 * timestamp and process the packet normally.
1721 	 * See section 3.2 of RFC 7323.
1722 	 */
1723 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1724 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1725 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1726 			    "segment processed normally\n", s, __func__);
1727 			free(s, M_TCPLOG);
1728 		}
1729 	}
1730 
1731 	/*
1732 	 * Header prediction: check for the two common cases
1733 	 * of a uni-directional data xfer.  If the packet has
1734 	 * no control flags, is in-sequence, the window didn't
1735 	 * change and we're not retransmitting, it's a
1736 	 * candidate.  If the length is zero and the ack moved
1737 	 * forward, we're the sender side of the xfer.  Just
1738 	 * free the data acked & wake any higher level process
1739 	 * that was blocked waiting for space.  If the length
1740 	 * is non-zero and the ack didn't move, we're the
1741 	 * receiver side.  If we're getting packets in-order
1742 	 * (the reassembly queue is empty), add the data to
1743 	 * the socket buffer and note that we need a delayed ack.
1744 	 * Make sure that the hidden state-flags are also off.
1745 	 * Since we check for TCPS_ESTABLISHED first, it can only
1746 	 * be TH_NEEDSYN.
1747 	 */
1748 	if (tp->t_state == TCPS_ESTABLISHED &&
1749 	    th->th_seq == tp->rcv_nxt &&
1750 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1751 	    tp->snd_nxt == tp->snd_max &&
1752 	    tiwin && tiwin == tp->snd_wnd &&
1753 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1754 	    SEGQ_EMPTY(tp) &&
1755 	    ((to.to_flags & TOF_TS) == 0 ||
1756 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1757 		/*
1758 		 * If last ACK falls within this segment's sequence numbers,
1759 		 * record the timestamp.
1760 		 * NOTE that the test is modified according to the latest
1761 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1762 		 */
1763 		if ((to.to_flags & TOF_TS) != 0 &&
1764 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1765 			tp->ts_recent_age = tcp_ts_getticks();
1766 			tp->ts_recent = to.to_tsval;
1767 		}
1768 
1769 		if (tlen == 0) {
1770 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1771 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1772 			    !IN_RECOVERY(tp->t_flags) &&
1773 			    (to.to_flags & TOF_SACK) == 0 &&
1774 			    TAILQ_EMPTY(&tp->snd_holes)) {
1775 				/*
1776 				 * This is a pure ack for outstanding data.
1777 				 */
1778 				TCPSTAT_INC(tcps_predack);
1779 
1780 				/*
1781 				 * "bad retransmit" recovery without timestamps.
1782 				 */
1783 				if ((to.to_flags & TOF_TS) == 0 &&
1784 				    tp->t_rxtshift == 1 &&
1785 				    tp->t_flags & TF_PREVVALID &&
1786 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1787 					cc_cong_signal(tp, th, CC_RTO_ERR);
1788 				}
1789 
1790 				/*
1791 				 * Recalculate the transmit timer / rtt.
1792 				 *
1793 				 * Some boxes send broken timestamp replies
1794 				 * during the SYN+ACK phase, ignore
1795 				 * timestamps of 0 or we could calculate a
1796 				 * huge RTT and blow up the retransmit timer.
1797 				 */
1798 				if ((to.to_flags & TOF_TS) != 0 &&
1799 				    to.to_tsecr) {
1800 					uint32_t t;
1801 
1802 					t = tcp_ts_getticks() - to.to_tsecr;
1803 					if (!tp->t_rttlow || tp->t_rttlow > t)
1804 						tp->t_rttlow = t;
1805 					tcp_xmit_timer(tp,
1806 					    TCP_TS_TO_TICKS(t) + 1);
1807 				} else if (tp->t_rtttime &&
1808 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1809 					if (!tp->t_rttlow ||
1810 					    tp->t_rttlow > ticks - tp->t_rtttime)
1811 						tp->t_rttlow = ticks - tp->t_rtttime;
1812 					tcp_xmit_timer(tp,
1813 							ticks - tp->t_rtttime);
1814 				}
1815 				acked = BYTES_THIS_ACK(tp, th);
1816 
1817 #ifdef TCP_HHOOK
1818 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1819 				hhook_run_tcp_est_in(tp, th, &to);
1820 #endif
1821 
1822 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1823 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1824 				sbdrop(&so->so_snd, acked);
1825 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1826 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1827 					tp->snd_recover = th->th_ack - 1;
1828 
1829 				/*
1830 				 * Let the congestion control algorithm update
1831 				 * congestion control related information. This
1832 				 * typically means increasing the congestion
1833 				 * window.
1834 				 */
1835 				cc_ack_received(tp, th, nsegs, CC_ACK);
1836 
1837 				tp->snd_una = th->th_ack;
1838 				/*
1839 				 * Pull snd_wl2 up to prevent seq wrap relative
1840 				 * to th_ack.
1841 				 */
1842 				tp->snd_wl2 = th->th_ack;
1843 				tp->t_dupacks = 0;
1844 				m_freem(m);
1845 
1846 				/*
1847 				 * If all outstanding data are acked, stop
1848 				 * retransmit timer, otherwise restart timer
1849 				 * using current (possibly backed-off) value.
1850 				 * If process is waiting for space,
1851 				 * wakeup/selwakeup/signal.  If data
1852 				 * are ready to send, let tcp_output
1853 				 * decide between more output or persist.
1854 				 */
1855 #ifdef TCPDEBUG
1856 				if (so->so_options & SO_DEBUG)
1857 					tcp_trace(TA_INPUT, ostate, tp,
1858 					    (void *)tcp_saveipgen,
1859 					    &tcp_savetcp, 0);
1860 #endif
1861 				TCP_PROBE3(debug__input, tp, th, m);
1862 				if (tp->snd_una == tp->snd_max)
1863 					tcp_timer_activate(tp, TT_REXMT, 0);
1864 				else if (!tcp_timer_active(tp, TT_PERSIST))
1865 					tcp_timer_activate(tp, TT_REXMT,
1866 						      tp->t_rxtcur);
1867 				tp->t_flags |= TF_WAKESOW;
1868 				if (sbavail(&so->so_snd))
1869 					(void) tp->t_fb->tfb_tcp_output(tp);
1870 				goto check_delack;
1871 			}
1872 		} else if (th->th_ack == tp->snd_una &&
1873 		    tlen <= sbspace(&so->so_rcv)) {
1874 			int newsize = 0;	/* automatic sockbuf scaling */
1875 
1876 			/*
1877 			 * This is a pure, in-sequence data packet with
1878 			 * nothing on the reassembly queue and we have enough
1879 			 * buffer space to take it.
1880 			 */
1881 			/* Clean receiver SACK report if present */
1882 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1883 				tcp_clean_sackreport(tp);
1884 			TCPSTAT_INC(tcps_preddat);
1885 			tp->rcv_nxt += tlen;
1886 			if (tlen &&
1887 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1888 			    (tp->t_fbyte_in == 0)) {
1889 				tp->t_fbyte_in = ticks;
1890 				if (tp->t_fbyte_in == 0)
1891 					tp->t_fbyte_in = 1;
1892 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1893 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1894 			}
1895 			/*
1896 			 * Pull snd_wl1 up to prevent seq wrap relative to
1897 			 * th_seq.
1898 			 */
1899 			tp->snd_wl1 = th->th_seq;
1900 			/*
1901 			 * Pull rcv_up up to prevent seq wrap relative to
1902 			 * rcv_nxt.
1903 			 */
1904 			tp->rcv_up = tp->rcv_nxt;
1905 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1906 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1907 #ifdef TCPDEBUG
1908 			if (so->so_options & SO_DEBUG)
1909 				tcp_trace(TA_INPUT, ostate, tp,
1910 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1911 #endif
1912 			TCP_PROBE3(debug__input, tp, th, m);
1913 
1914 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1915 
1916 			/* Add data to socket buffer. */
1917 			SOCKBUF_LOCK(&so->so_rcv);
1918 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1919 				m_freem(m);
1920 			} else {
1921 				/*
1922 				 * Set new socket buffer size.
1923 				 * Give up when limit is reached.
1924 				 */
1925 				if (newsize)
1926 					if (!sbreserve_locked(&so->so_rcv,
1927 					    newsize, so, NULL))
1928 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1929 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1930 				sbappendstream_locked(&so->so_rcv, m, 0);
1931 			}
1932 			SOCKBUF_UNLOCK(&so->so_rcv);
1933 			tp->t_flags |= TF_WAKESOR;
1934 			if (DELAY_ACK(tp, tlen)) {
1935 				tp->t_flags |= TF_DELACK;
1936 			} else {
1937 				tp->t_flags |= TF_ACKNOW;
1938 				tp->t_fb->tfb_tcp_output(tp);
1939 			}
1940 			goto check_delack;
1941 		}
1942 	}
1943 
1944 	/*
1945 	 * Calculate amount of space in receive window,
1946 	 * and then do TCP input processing.
1947 	 * Receive window is amount of space in rcv queue,
1948 	 * but not less than advertised window.
1949 	 */
1950 	win = sbspace(&so->so_rcv);
1951 	if (win < 0)
1952 		win = 0;
1953 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1954 
1955 	switch (tp->t_state) {
1956 	/*
1957 	 * If the state is SYN_RECEIVED:
1958 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1959 	 */
1960 	case TCPS_SYN_RECEIVED:
1961 		if ((thflags & TH_ACK) &&
1962 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1963 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1964 				rstreason = BANDLIM_RST_OPENPORT;
1965 				goto dropwithreset;
1966 		}
1967 		if (IS_FASTOPEN(tp->t_flags)) {
1968 			/*
1969 			 * When a TFO connection is in SYN_RECEIVED, the
1970 			 * only valid packets are the initial SYN, a
1971 			 * retransmit/copy of the initial SYN (possibly with
1972 			 * a subset of the original data), a valid ACK, a
1973 			 * FIN, or a RST.
1974 			 */
1975 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1976 				rstreason = BANDLIM_RST_OPENPORT;
1977 				goto dropwithreset;
1978 			} else if (thflags & TH_SYN) {
1979 				/* non-initial SYN is ignored */
1980 				if ((tcp_timer_active(tp, TT_DELACK) ||
1981 				     tcp_timer_active(tp, TT_REXMT)))
1982 					goto drop;
1983 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1984 				goto drop;
1985 			}
1986 		}
1987 		break;
1988 
1989 	/*
1990 	 * If the state is SYN_SENT:
1991 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
1992 	 *	    been verified), then drop the connection.
1993 	 *	if seg contains a RST without an ACK, drop the seg.
1994 	 *	if seg does not contain SYN, then drop the seg.
1995 	 * Otherwise this is an acceptable SYN segment
1996 	 *	initialize tp->rcv_nxt and tp->irs
1997 	 *	if seg contains ack then advance tp->snd_una
1998 	 *	if seg contains an ECE and ECN support is enabled, the stream
1999 	 *	    is ECN capable.
2000 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2001 	 *	arrange for segment to be acked (eventually)
2002 	 *	continue processing rest of data/controls, beginning with URG
2003 	 */
2004 	case TCPS_SYN_SENT:
2005 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2006 			TCP_PROBE5(connect__refused, NULL, tp,
2007 			    m, tp, th);
2008 			tp = tcp_drop(tp, ECONNREFUSED);
2009 		}
2010 		if (thflags & TH_RST)
2011 			goto drop;
2012 		if (!(thflags & TH_SYN))
2013 			goto drop;
2014 
2015 		tp->irs = th->th_seq;
2016 		tcp_rcvseqinit(tp);
2017 		if (thflags & TH_ACK) {
2018 			int tfo_partial_ack = 0;
2019 
2020 			TCPSTAT_INC(tcps_connects);
2021 			soisconnected(so);
2022 #ifdef MAC
2023 			mac_socketpeer_set_from_mbuf(m, so);
2024 #endif
2025 			/* Do window scaling on this connection? */
2026 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2027 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2028 				tp->rcv_scale = tp->request_r_scale;
2029 			}
2030 			tp->rcv_adv += min(tp->rcv_wnd,
2031 			    TCP_MAXWIN << tp->rcv_scale);
2032 			tp->snd_una++;		/* SYN is acked */
2033 			/*
2034 			 * If not all the data that was sent in the TFO SYN
2035 			 * has been acked, resend the remainder right away.
2036 			 */
2037 			if (IS_FASTOPEN(tp->t_flags) &&
2038 			    (tp->snd_una != tp->snd_max)) {
2039 				tp->snd_nxt = th->th_ack;
2040 				tfo_partial_ack = 1;
2041 			}
2042 			/*
2043 			 * If there's data, delay ACK; if there's also a FIN
2044 			 * ACKNOW will be turned on later.
2045 			 */
2046 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2047 				tcp_timer_activate(tp, TT_DELACK,
2048 				    tcp_delacktime);
2049 			else
2050 				tp->t_flags |= TF_ACKNOW;
2051 
2052 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2053 			    (V_tcp_do_ecn == 1)) {
2054 				tp->t_flags2 |= TF2_ECN_PERMIT;
2055 				TCPSTAT_INC(tcps_ecn_shs);
2056 			}
2057 
2058 			/*
2059 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2060 			 * Transitions:
2061 			 *	SYN_SENT  --> ESTABLISHED
2062 			 *	SYN_SENT* --> FIN_WAIT_1
2063 			 */
2064 			tp->t_starttime = ticks;
2065 			if (tp->t_flags & TF_NEEDFIN) {
2066 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2067 				tp->t_flags &= ~TF_NEEDFIN;
2068 				thflags &= ~TH_SYN;
2069 			} else {
2070 				tcp_state_change(tp, TCPS_ESTABLISHED);
2071 				TCP_PROBE5(connect__established, NULL, tp,
2072 				    m, tp, th);
2073 				cc_conn_init(tp);
2074 				tcp_timer_activate(tp, TT_KEEP,
2075 				    TP_KEEPIDLE(tp));
2076 			}
2077 		} else {
2078 			/*
2079 			 * Received initial SYN in SYN-SENT[*] state =>
2080 			 * simultaneous open.
2081 			 * If it succeeds, connection is * half-synchronized.
2082 			 * Otherwise, do 3-way handshake:
2083 			 *        SYN-SENT -> SYN-RECEIVED
2084 			 *        SYN-SENT* -> SYN-RECEIVED*
2085 			 */
2086 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2087 			tcp_timer_activate(tp, TT_REXMT, 0);
2088 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2089 		}
2090 
2091 		INP_WLOCK_ASSERT(tp->t_inpcb);
2092 
2093 		/*
2094 		 * Advance th->th_seq to correspond to first data byte.
2095 		 * If data, trim to stay within window,
2096 		 * dropping FIN if necessary.
2097 		 */
2098 		th->th_seq++;
2099 		if (tlen > tp->rcv_wnd) {
2100 			todrop = tlen - tp->rcv_wnd;
2101 			m_adj(m, -todrop);
2102 			tlen = tp->rcv_wnd;
2103 			thflags &= ~TH_FIN;
2104 			TCPSTAT_INC(tcps_rcvpackafterwin);
2105 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2106 		}
2107 		tp->snd_wl1 = th->th_seq - 1;
2108 		tp->rcv_up = th->th_seq;
2109 		/*
2110 		 * Client side of transaction: already sent SYN and data.
2111 		 * If the remote host used T/TCP to validate the SYN,
2112 		 * our data will be ACK'd; if so, enter normal data segment
2113 		 * processing in the middle of step 5, ack processing.
2114 		 * Otherwise, goto step 6.
2115 		 */
2116 		if (thflags & TH_ACK)
2117 			goto process_ACK;
2118 
2119 		goto step6;
2120 
2121 	/*
2122 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2123 	 *      do normal processing.
2124 	 *
2125 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2126 	 */
2127 	case TCPS_LAST_ACK:
2128 	case TCPS_CLOSING:
2129 		break;  /* continue normal processing */
2130 	}
2131 
2132 	/*
2133 	 * States other than LISTEN or SYN_SENT.
2134 	 * First check the RST flag and sequence number since reset segments
2135 	 * are exempt from the timestamp and connection count tests.  This
2136 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2137 	 * below which allowed reset segments in half the sequence space
2138 	 * to fall though and be processed (which gives forged reset
2139 	 * segments with a random sequence number a 50 percent chance of
2140 	 * killing a connection).
2141 	 * Then check timestamp, if present.
2142 	 * Then check the connection count, if present.
2143 	 * Then check that at least some bytes of segment are within
2144 	 * receive window.  If segment begins before rcv_nxt,
2145 	 * drop leading data (and SYN); if nothing left, just ack.
2146 	 */
2147 	if (thflags & TH_RST) {
2148 		/*
2149 		 * RFC5961 Section 3.2
2150 		 *
2151 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2152 		 * - If RST is in window, we send challenge ACK.
2153 		 *
2154 		 * Note: to take into account delayed ACKs, we should
2155 		 *   test against last_ack_sent instead of rcv_nxt.
2156 		 * Note 2: we handle special case of closed window, not
2157 		 *   covered by the RFC.
2158 		 */
2159 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2160 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2161 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2162 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2163 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2164 			    __func__, th, tp));
2165 
2166 			if (V_tcp_insecure_rst ||
2167 			    tp->last_ack_sent == th->th_seq) {
2168 				TCPSTAT_INC(tcps_drops);
2169 				/* Drop the connection. */
2170 				switch (tp->t_state) {
2171 				case TCPS_SYN_RECEIVED:
2172 					so->so_error = ECONNREFUSED;
2173 					goto close;
2174 				case TCPS_ESTABLISHED:
2175 				case TCPS_FIN_WAIT_1:
2176 				case TCPS_FIN_WAIT_2:
2177 				case TCPS_CLOSE_WAIT:
2178 				case TCPS_CLOSING:
2179 				case TCPS_LAST_ACK:
2180 					so->so_error = ECONNRESET;
2181 				close:
2182 					/* FALLTHROUGH */
2183 				default:
2184 					tp = tcp_close(tp);
2185 				}
2186 			} else {
2187 				TCPSTAT_INC(tcps_badrst);
2188 				/* Send challenge ACK. */
2189 				tcp_respond(tp, mtod(m, void *), th, m,
2190 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2191 				tp->last_ack_sent = tp->rcv_nxt;
2192 				m = NULL;
2193 			}
2194 		}
2195 		goto drop;
2196 	}
2197 
2198 	/*
2199 	 * RFC5961 Section 4.2
2200 	 * Send challenge ACK for any SYN in synchronized state.
2201 	 */
2202 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2203 	    tp->t_state != TCPS_SYN_RECEIVED) {
2204 		TCPSTAT_INC(tcps_badsyn);
2205 		if (V_tcp_insecure_syn &&
2206 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2207 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2208 			tp = tcp_drop(tp, ECONNRESET);
2209 			rstreason = BANDLIM_UNLIMITED;
2210 		} else {
2211 			/* Send challenge ACK. */
2212 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2213 			    tp->snd_nxt, TH_ACK);
2214 			tp->last_ack_sent = tp->rcv_nxt;
2215 			m = NULL;
2216 		}
2217 		goto drop;
2218 	}
2219 
2220 	/*
2221 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2222 	 * and it's less than ts_recent, drop it.
2223 	 */
2224 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2225 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2226 		/* Check to see if ts_recent is over 24 days old.  */
2227 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2228 			/*
2229 			 * Invalidate ts_recent.  If this segment updates
2230 			 * ts_recent, the age will be reset later and ts_recent
2231 			 * will get a valid value.  If it does not, setting
2232 			 * ts_recent to zero will at least satisfy the
2233 			 * requirement that zero be placed in the timestamp
2234 			 * echo reply when ts_recent isn't valid.  The
2235 			 * age isn't reset until we get a valid ts_recent
2236 			 * because we don't want out-of-order segments to be
2237 			 * dropped when ts_recent is old.
2238 			 */
2239 			tp->ts_recent = 0;
2240 		} else {
2241 			TCPSTAT_INC(tcps_rcvduppack);
2242 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2243 			TCPSTAT_INC(tcps_pawsdrop);
2244 			if (tlen)
2245 				goto dropafterack;
2246 			goto drop;
2247 		}
2248 	}
2249 
2250 	/*
2251 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2252 	 * this connection before trimming the data to fit the receive
2253 	 * window.  Check the sequence number versus IRS since we know
2254 	 * the sequence numbers haven't wrapped.  This is a partial fix
2255 	 * for the "LAND" DoS attack.
2256 	 */
2257 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2258 		rstreason = BANDLIM_RST_OPENPORT;
2259 		goto dropwithreset;
2260 	}
2261 
2262 	todrop = tp->rcv_nxt - th->th_seq;
2263 	if (todrop > 0) {
2264 		if (thflags & TH_SYN) {
2265 			thflags &= ~TH_SYN;
2266 			th->th_seq++;
2267 			if (th->th_urp > 1)
2268 				th->th_urp--;
2269 			else
2270 				thflags &= ~TH_URG;
2271 			todrop--;
2272 		}
2273 		/*
2274 		 * Following if statement from Stevens, vol. 2, p. 960.
2275 		 */
2276 		if (todrop > tlen
2277 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2278 			/*
2279 			 * Any valid FIN must be to the left of the window.
2280 			 * At this point the FIN must be a duplicate or out
2281 			 * of sequence; drop it.
2282 			 */
2283 			thflags &= ~TH_FIN;
2284 
2285 			/*
2286 			 * Send an ACK to resynchronize and drop any data.
2287 			 * But keep on processing for RST or ACK.
2288 			 */
2289 			tp->t_flags |= TF_ACKNOW;
2290 			todrop = tlen;
2291 			TCPSTAT_INC(tcps_rcvduppack);
2292 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2293 		} else {
2294 			TCPSTAT_INC(tcps_rcvpartduppack);
2295 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2296 		}
2297 		/*
2298 		 * DSACK - add SACK block for dropped range
2299 		 */
2300 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2301 			tcp_update_sack_list(tp, th->th_seq,
2302 			    th->th_seq + todrop);
2303 			/*
2304 			 * ACK now, as the next in-sequence segment
2305 			 * will clear the DSACK block again
2306 			 */
2307 			tp->t_flags |= TF_ACKNOW;
2308 		}
2309 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2310 		th->th_seq += todrop;
2311 		tlen -= todrop;
2312 		if (th->th_urp > todrop)
2313 			th->th_urp -= todrop;
2314 		else {
2315 			thflags &= ~TH_URG;
2316 			th->th_urp = 0;
2317 		}
2318 	}
2319 
2320 	/*
2321 	 * If new data are received on a connection after the
2322 	 * user processes are gone, then RST the other end.
2323 	 */
2324 	if ((so->so_state & SS_NOFDREF) &&
2325 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2326 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2327 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2328 			    "after socket was closed, "
2329 			    "sending RST and removing tcpcb\n",
2330 			    s, __func__, tcpstates[tp->t_state], tlen);
2331 			free(s, M_TCPLOG);
2332 		}
2333 		tp = tcp_close(tp);
2334 		TCPSTAT_INC(tcps_rcvafterclose);
2335 		rstreason = BANDLIM_UNLIMITED;
2336 		goto dropwithreset;
2337 	}
2338 
2339 	/*
2340 	 * If segment ends after window, drop trailing data
2341 	 * (and PUSH and FIN); if nothing left, just ACK.
2342 	 */
2343 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2344 	if (todrop > 0) {
2345 		TCPSTAT_INC(tcps_rcvpackafterwin);
2346 		if (todrop >= tlen) {
2347 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2348 			/*
2349 			 * If window is closed can only take segments at
2350 			 * window edge, and have to drop data and PUSH from
2351 			 * incoming segments.  Continue processing, but
2352 			 * remember to ack.  Otherwise, drop segment
2353 			 * and ack.
2354 			 */
2355 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2356 				tp->t_flags |= TF_ACKNOW;
2357 				TCPSTAT_INC(tcps_rcvwinprobe);
2358 			} else
2359 				goto dropafterack;
2360 		} else
2361 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2362 		m_adj(m, -todrop);
2363 		tlen -= todrop;
2364 		thflags &= ~(TH_PUSH|TH_FIN);
2365 	}
2366 
2367 	/*
2368 	 * If last ACK falls within this segment's sequence numbers,
2369 	 * record its timestamp.
2370 	 * NOTE:
2371 	 * 1) That the test incorporates suggestions from the latest
2372 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2373 	 * 2) That updating only on newer timestamps interferes with
2374 	 *    our earlier PAWS tests, so this check should be solely
2375 	 *    predicated on the sequence space of this segment.
2376 	 * 3) That we modify the segment boundary check to be
2377 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2378 	 *    instead of RFC1323's
2379 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2380 	 *    This modified check allows us to overcome RFC1323's
2381 	 *    limitations as described in Stevens TCP/IP Illustrated
2382 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2383 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2384 	 */
2385 	if ((to.to_flags & TOF_TS) != 0 &&
2386 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2387 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2388 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2389 		tp->ts_recent_age = tcp_ts_getticks();
2390 		tp->ts_recent = to.to_tsval;
2391 	}
2392 
2393 	/*
2394 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2395 	 * flag is on (half-synchronized state), then queue data for
2396 	 * later processing; else drop segment and return.
2397 	 */
2398 	if ((thflags & TH_ACK) == 0) {
2399 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2400 		    (tp->t_flags & TF_NEEDSYN)) {
2401 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2402 			    IS_FASTOPEN(tp->t_flags)) {
2403 				tp->snd_wnd = tiwin;
2404 				cc_conn_init(tp);
2405 			}
2406 			goto step6;
2407 		} else if (tp->t_flags & TF_ACKNOW)
2408 			goto dropafterack;
2409 		else
2410 			goto drop;
2411 	}
2412 
2413 	/*
2414 	 * Ack processing.
2415 	 */
2416 	switch (tp->t_state) {
2417 	/*
2418 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2419 	 * ESTABLISHED state and continue processing.
2420 	 * The ACK was checked above.
2421 	 */
2422 	case TCPS_SYN_RECEIVED:
2423 
2424 		TCPSTAT_INC(tcps_connects);
2425 		soisconnected(so);
2426 		/* Do window scaling? */
2427 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2428 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2429 			tp->rcv_scale = tp->request_r_scale;
2430 		}
2431 		tp->snd_wnd = tiwin;
2432 		/*
2433 		 * Make transitions:
2434 		 *      SYN-RECEIVED  -> ESTABLISHED
2435 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2436 		 */
2437 		tp->t_starttime = ticks;
2438 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2439 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2440 			tp->t_tfo_pending = NULL;
2441 		}
2442 		if (tp->t_flags & TF_NEEDFIN) {
2443 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2444 			tp->t_flags &= ~TF_NEEDFIN;
2445 		} else {
2446 			tcp_state_change(tp, TCPS_ESTABLISHED);
2447 			TCP_PROBE5(accept__established, NULL, tp,
2448 			    m, tp, th);
2449 			/*
2450 			 * TFO connections call cc_conn_init() during SYN
2451 			 * processing.  Calling it again here for such
2452 			 * connections is not harmless as it would undo the
2453 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2454 			 * is retransmitted.
2455 			 */
2456 			if (!IS_FASTOPEN(tp->t_flags))
2457 				cc_conn_init(tp);
2458 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2459 		}
2460 		/*
2461 		 * Account for the ACK of our SYN prior to
2462 		 * regular ACK processing below, except for
2463 		 * simultaneous SYN, which is handled later.
2464 		 */
2465 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2466 			incforsyn = 1;
2467 		/*
2468 		 * If segment contains data or ACK, will call tcp_reass()
2469 		 * later; if not, do so now to pass queued data to user.
2470 		 */
2471 		if (tlen == 0 && (thflags & TH_FIN) == 0)
2472 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2473 			    (struct mbuf *)0);
2474 		tp->snd_wl1 = th->th_seq - 1;
2475 		/* FALLTHROUGH */
2476 
2477 	/*
2478 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2479 	 * ACKs.  If the ack is in the range
2480 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2481 	 * then advance tp->snd_una to th->th_ack and drop
2482 	 * data from the retransmission queue.  If this ACK reflects
2483 	 * more up to date window information we update our window information.
2484 	 */
2485 	case TCPS_ESTABLISHED:
2486 	case TCPS_FIN_WAIT_1:
2487 	case TCPS_FIN_WAIT_2:
2488 	case TCPS_CLOSE_WAIT:
2489 	case TCPS_CLOSING:
2490 	case TCPS_LAST_ACK:
2491 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2492 			TCPSTAT_INC(tcps_rcvacktoomuch);
2493 			goto dropafterack;
2494 		}
2495 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2496 		    ((to.to_flags & TOF_SACK) ||
2497 		     !TAILQ_EMPTY(&tp->snd_holes)))
2498 			sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2499 		else
2500 			/*
2501 			 * Reset the value so that previous (valid) value
2502 			 * from the last ack with SACK doesn't get used.
2503 			 */
2504 			tp->sackhint.sacked_bytes = 0;
2505 
2506 #ifdef TCP_HHOOK
2507 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2508 		hhook_run_tcp_est_in(tp, th, &to);
2509 #endif
2510 
2511 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2512 			u_int maxseg;
2513 
2514 			maxseg = tcp_maxseg(tp);
2515 			if (tlen == 0 &&
2516 			    (tiwin == tp->snd_wnd ||
2517 			    (tp->t_flags & TF_SACK_PERMIT))) {
2518 				/*
2519 				 * If this is the first time we've seen a
2520 				 * FIN from the remote, this is not a
2521 				 * duplicate and it needs to be processed
2522 				 * normally.  This happens during a
2523 				 * simultaneous close.
2524 				 */
2525 				if ((thflags & TH_FIN) &&
2526 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2527 					tp->t_dupacks = 0;
2528 					break;
2529 				}
2530 				TCPSTAT_INC(tcps_rcvdupack);
2531 				/*
2532 				 * If we have outstanding data (other than
2533 				 * a window probe), this is a completely
2534 				 * duplicate ack (ie, window info didn't
2535 				 * change and FIN isn't set),
2536 				 * the ack is the biggest we've
2537 				 * seen and we've seen exactly our rexmt
2538 				 * threshold of them, assume a packet
2539 				 * has been dropped and retransmit it.
2540 				 * Kludge snd_nxt & the congestion
2541 				 * window so we send only this one
2542 				 * packet.
2543 				 *
2544 				 * We know we're losing at the current
2545 				 * window size so do congestion avoidance
2546 				 * (set ssthresh to half the current window
2547 				 * and pull our congestion window back to
2548 				 * the new ssthresh).
2549 				 *
2550 				 * Dup acks mean that packets have left the
2551 				 * network (they're now cached at the receiver)
2552 				 * so bump cwnd by the amount in the receiver
2553 				 * to keep a constant cwnd packets in the
2554 				 * network.
2555 				 *
2556 				 * When using TCP ECN, notify the peer that
2557 				 * we reduced the cwnd.
2558 				 */
2559 				/*
2560 				 * Following 2 kinds of acks should not affect
2561 				 * dupack counting:
2562 				 * 1) Old acks
2563 				 * 2) Acks with SACK but without any new SACK
2564 				 * information in them. These could result from
2565 				 * any anomaly in the network like a switch
2566 				 * duplicating packets or a possible DoS attack.
2567 				 */
2568 				if (th->th_ack != tp->snd_una ||
2569 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2570 				    !sack_changed))
2571 					break;
2572 				else if (!tcp_timer_active(tp, TT_REXMT))
2573 					tp->t_dupacks = 0;
2574 				else if (++tp->t_dupacks > tcprexmtthresh ||
2575 				     IN_FASTRECOVERY(tp->t_flags)) {
2576 					cc_ack_received(tp, th, nsegs,
2577 					    CC_DUPACK);
2578 					if (V_tcp_do_prr &&
2579 					    IN_FASTRECOVERY(tp->t_flags) &&
2580 					    (tp->t_flags & TF_SACK_PERMIT)) {
2581 						long snd_cnt = 0, limit = 0;
2582 						long del_data = 0, pipe = 0;
2583 						/*
2584 						 * In a duplicate ACK del_data is only the
2585 						 * diff_in_sack. If no SACK is used del_data
2586 						 * will be 0. Pipe is the amount of data we
2587 						 * estimate to be in the network.
2588 						 */
2589 						del_data = tp->sackhint.delivered_data;
2590 						pipe = (tp->snd_nxt - tp->snd_fack) +
2591 							tp->sackhint.sack_bytes_rexmit;
2592 						tp->sackhint.prr_delivered += del_data;
2593 						if (pipe > tp->snd_ssthresh) {
2594 							if (tp->sackhint.recover_fs == 0)
2595 								tp->sackhint.recover_fs =
2596 								    max(1, tp->snd_nxt - tp->snd_una);
2597 							snd_cnt = (tp->sackhint.prr_delivered *
2598 							    tp->snd_ssthresh /
2599 							    tp->sackhint.recover_fs) +
2600 							    1 - tp->sackhint.sack_bytes_rexmit;
2601 						} else {
2602 							if (V_tcp_do_prr_conservative)
2603 								limit = tp->sackhint.prr_delivered -
2604 									tp->sackhint.sack_bytes_rexmit;
2605 							else
2606 								if ((tp->sackhint.prr_delivered -
2607 								    tp->sackhint.sack_bytes_rexmit) >
2608 								    del_data)
2609 									limit = tp->sackhint.prr_delivered -
2610 									    tp->sackhint.sack_bytes_rexmit +
2611 									    maxseg;
2612 								else
2613 									limit = del_data + maxseg;
2614 							if ((tp->snd_ssthresh - pipe) < limit)
2615 								snd_cnt = tp->snd_ssthresh - pipe;
2616 							else
2617 								snd_cnt = limit;
2618 						}
2619 						snd_cnt = max((snd_cnt / maxseg), 0);
2620 						/*
2621 						 * Send snd_cnt new data into the network in
2622 						 * response to this ACK. If there is a going
2623 						 * to be a SACK retransmission, adjust snd_cwnd
2624 						 * accordingly.
2625 						 */
2626 						tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
2627 						    tp->sackhint.sack_bytes_rexmit +
2628 						    (snd_cnt * maxseg);
2629 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2630 					    IN_FASTRECOVERY(tp->t_flags)) {
2631 						int awnd;
2632 
2633 						/*
2634 						 * Compute the amount of data in flight first.
2635 						 * We can inject new data into the pipe iff
2636 						 * we have less than 1/2 the original window's
2637 						 * worth of data in flight.
2638 						 */
2639 						if (V_tcp_do_rfc6675_pipe)
2640 							awnd = tcp_compute_pipe(tp);
2641 						else
2642 							awnd = (tp->snd_nxt - tp->snd_fack) +
2643 								tp->sackhint.sack_bytes_rexmit;
2644 
2645 						if (awnd < tp->snd_ssthresh) {
2646 							tp->snd_cwnd += maxseg;
2647 							if (tp->snd_cwnd > tp->snd_ssthresh)
2648 								tp->snd_cwnd = tp->snd_ssthresh;
2649 						}
2650 					} else
2651 						tp->snd_cwnd += maxseg;
2652 					(void) tp->t_fb->tfb_tcp_output(tp);
2653 					goto drop;
2654 				} else if (tp->t_dupacks == tcprexmtthresh) {
2655 					tcp_seq onxt = tp->snd_nxt;
2656 
2657 					/*
2658 					 * If we're doing sack, or prr, check
2659 					 * to see if we're already in sack
2660 					 * recovery. If we're not doing sack,
2661 					 * check to see if we're in newreno
2662 					 * recovery.
2663 					 */
2664 					if (V_tcp_do_prr ||
2665 					    (tp->t_flags & TF_SACK_PERMIT)) {
2666 						if (IN_FASTRECOVERY(tp->t_flags)) {
2667 							tp->t_dupacks = 0;
2668 							break;
2669 						}
2670 					} else {
2671 						if (SEQ_LEQ(th->th_ack,
2672 						    tp->snd_recover)) {
2673 							tp->t_dupacks = 0;
2674 							break;
2675 						}
2676 					}
2677 					/* Congestion signal before ack. */
2678 					cc_cong_signal(tp, th, CC_NDUPACK);
2679 					cc_ack_received(tp, th, nsegs,
2680 					    CC_DUPACK);
2681 					tcp_timer_activate(tp, TT_REXMT, 0);
2682 					tp->t_rtttime = 0;
2683 					if (V_tcp_do_prr) {
2684 						/*
2685 						 * snd_ssthresh is already updated by
2686 						 * cc_cong_signal.
2687 						 */
2688 						tp->sackhint.prr_delivered = 0;
2689 						tp->sackhint.sack_bytes_rexmit = 0;
2690 						tp->sackhint.recover_fs = max(1,
2691 						    tp->snd_nxt - tp->snd_una);
2692 					}
2693 					if (tp->t_flags & TF_SACK_PERMIT) {
2694 						TCPSTAT_INC(
2695 						    tcps_sack_recovery_episode);
2696 						tp->snd_recover = tp->snd_nxt;
2697 						tp->snd_cwnd = maxseg;
2698 						(void) tp->t_fb->tfb_tcp_output(tp);
2699 						goto drop;
2700 					}
2701 					tp->snd_nxt = th->th_ack;
2702 					tp->snd_cwnd = maxseg;
2703 					(void) tp->t_fb->tfb_tcp_output(tp);
2704 					KASSERT(tp->snd_limited <= 2,
2705 					    ("%s: tp->snd_limited too big",
2706 					    __func__));
2707 					tp->snd_cwnd = tp->snd_ssthresh +
2708 					     maxseg *
2709 					     (tp->t_dupacks - tp->snd_limited);
2710 					if (SEQ_GT(onxt, tp->snd_nxt))
2711 						tp->snd_nxt = onxt;
2712 					goto drop;
2713 				} else if (V_tcp_do_rfc3042) {
2714 					/*
2715 					 * Process first and second duplicate
2716 					 * ACKs. Each indicates a segment
2717 					 * leaving the network, creating room
2718 					 * for more. Make sure we can send a
2719 					 * packet on reception of each duplicate
2720 					 * ACK by increasing snd_cwnd by one
2721 					 * segment. Restore the original
2722 					 * snd_cwnd after packet transmission.
2723 					 */
2724 					cc_ack_received(tp, th, nsegs,
2725 					    CC_DUPACK);
2726 					uint32_t oldcwnd = tp->snd_cwnd;
2727 					tcp_seq oldsndmax = tp->snd_max;
2728 					u_int sent;
2729 					int avail;
2730 
2731 					KASSERT(tp->t_dupacks == 1 ||
2732 					    tp->t_dupacks == 2,
2733 					    ("%s: dupacks not 1 or 2",
2734 					    __func__));
2735 					if (tp->t_dupacks == 1)
2736 						tp->snd_limited = 0;
2737 					tp->snd_cwnd =
2738 					    (tp->snd_nxt - tp->snd_una) +
2739 					    (tp->t_dupacks - tp->snd_limited) *
2740 					    maxseg;
2741 					/*
2742 					 * Only call tcp_output when there
2743 					 * is new data available to be sent.
2744 					 * Otherwise we would send pure ACKs.
2745 					 */
2746 					SOCKBUF_LOCK(&so->so_snd);
2747 					avail = sbavail(&so->so_snd) -
2748 					    (tp->snd_nxt - tp->snd_una);
2749 					SOCKBUF_UNLOCK(&so->so_snd);
2750 					if (avail > 0)
2751 						(void) tp->t_fb->tfb_tcp_output(tp);
2752 					sent = tp->snd_max - oldsndmax;
2753 					if (sent > maxseg) {
2754 						KASSERT((tp->t_dupacks == 2 &&
2755 						    tp->snd_limited == 0) ||
2756 						   (sent == maxseg + 1 &&
2757 						    tp->t_flags & TF_SENTFIN),
2758 						    ("%s: sent too much",
2759 						    __func__));
2760 						tp->snd_limited = 2;
2761 					} else if (sent > 0)
2762 						++tp->snd_limited;
2763 					tp->snd_cwnd = oldcwnd;
2764 					goto drop;
2765 				}
2766 			}
2767 			break;
2768 		} else {
2769 			/*
2770 			 * This ack is advancing the left edge, reset the
2771 			 * counter.
2772 			 */
2773 			tp->t_dupacks = 0;
2774 			/*
2775 			 * If this ack also has new SACK info, increment the
2776 			 * counter as per rfc6675. The variable
2777 			 * sack_changed tracks all changes to the SACK
2778 			 * scoreboard, including when partial ACKs without
2779 			 * SACK options are received, and clear the scoreboard
2780 			 * from the left side. Such partial ACKs should not be
2781 			 * counted as dupacks here.
2782 			 */
2783 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2784 			    (to.to_flags & TOF_SACK) &&
2785 			    sack_changed)
2786 				tp->t_dupacks++;
2787 		}
2788 
2789 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2790 		    ("%s: th_ack <= snd_una", __func__));
2791 
2792 		/*
2793 		 * If the congestion window was inflated to account
2794 		 * for the other side's cached packets, retract it.
2795 		 */
2796 		if (IN_FASTRECOVERY(tp->t_flags)) {
2797 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2798 				if (tp->t_flags & TF_SACK_PERMIT)
2799 					if (V_tcp_do_prr)
2800 						tcp_prr_partialack(tp, th);
2801 					else
2802 						tcp_sack_partialack(tp, th);
2803 				else
2804 					tcp_newreno_partial_ack(tp, th);
2805 			} else
2806 				cc_post_recovery(tp, th);
2807 		}
2808 		/*
2809 		 * If we reach this point, ACK is not a duplicate,
2810 		 *     i.e., it ACKs something we sent.
2811 		 */
2812 		if (tp->t_flags & TF_NEEDSYN) {
2813 			/*
2814 			 * T/TCP: Connection was half-synchronized, and our
2815 			 * SYN has been ACK'd (so connection is now fully
2816 			 * synchronized).  Go to non-starred state,
2817 			 * increment snd_una for ACK of SYN, and check if
2818 			 * we can do window scaling.
2819 			 */
2820 			tp->t_flags &= ~TF_NEEDSYN;
2821 			tp->snd_una++;
2822 			/* Do window scaling? */
2823 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2824 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2825 				tp->rcv_scale = tp->request_r_scale;
2826 				/* Send window already scaled. */
2827 			}
2828 		}
2829 
2830 process_ACK:
2831 		INP_WLOCK_ASSERT(tp->t_inpcb);
2832 
2833 		/*
2834 		 * Adjust for the SYN bit in sequence space,
2835 		 * but don't account for it in cwnd calculations.
2836 		 * This is for the SYN_RECEIVED, non-simultaneous
2837 		 * SYN case. SYN_SENT and simultaneous SYN are
2838 		 * treated elsewhere.
2839 		 */
2840 		if (incforsyn)
2841 			tp->snd_una++;
2842 		acked = BYTES_THIS_ACK(tp, th);
2843 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2844 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2845 		    tp->snd_una, th->th_ack, tp, m));
2846 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2847 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2848 
2849 		/*
2850 		 * If we just performed our first retransmit, and the ACK
2851 		 * arrives within our recovery window, then it was a mistake
2852 		 * to do the retransmit in the first place.  Recover our
2853 		 * original cwnd and ssthresh, and proceed to transmit where
2854 		 * we left off.
2855 		 */
2856 		if (tp->t_rxtshift == 1 &&
2857 		    tp->t_flags & TF_PREVVALID &&
2858 		    tp->t_badrxtwin &&
2859 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2860 			cc_cong_signal(tp, th, CC_RTO_ERR);
2861 
2862 		/*
2863 		 * If we have a timestamp reply, update smoothed
2864 		 * round trip time.  If no timestamp is present but
2865 		 * transmit timer is running and timed sequence
2866 		 * number was acked, update smoothed round trip time.
2867 		 * Since we now have an rtt measurement, cancel the
2868 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2869 		 * Recompute the initial retransmit timer.
2870 		 *
2871 		 * Some boxes send broken timestamp replies
2872 		 * during the SYN+ACK phase, ignore
2873 		 * timestamps of 0 or we could calculate a
2874 		 * huge RTT and blow up the retransmit timer.
2875 		 */
2876 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2877 			uint32_t t;
2878 
2879 			t = tcp_ts_getticks() - to.to_tsecr;
2880 			if (!tp->t_rttlow || tp->t_rttlow > t)
2881 				tp->t_rttlow = t;
2882 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2883 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2884 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2885 				tp->t_rttlow = ticks - tp->t_rtttime;
2886 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2887 		}
2888 
2889 		/*
2890 		 * If all outstanding data is acked, stop retransmit
2891 		 * timer and remember to restart (more output or persist).
2892 		 * If there is more data to be acked, restart retransmit
2893 		 * timer, using current (possibly backed-off) value.
2894 		 */
2895 		if (th->th_ack == tp->snd_max) {
2896 			tcp_timer_activate(tp, TT_REXMT, 0);
2897 			needoutput = 1;
2898 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2899 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2900 
2901 		/*
2902 		 * If no data (only SYN) was ACK'd,
2903 		 *    skip rest of ACK processing.
2904 		 */
2905 		if (acked == 0)
2906 			goto step6;
2907 
2908 		/*
2909 		 * Let the congestion control algorithm update congestion
2910 		 * control related information. This typically means increasing
2911 		 * the congestion window.
2912 		 */
2913 		cc_ack_received(tp, th, nsegs, CC_ACK);
2914 
2915 		SOCKBUF_LOCK(&so->so_snd);
2916 		if (acked > sbavail(&so->so_snd)) {
2917 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2918 				tp->snd_wnd -= sbavail(&so->so_snd);
2919 			else
2920 				tp->snd_wnd = 0;
2921 			mfree = sbcut_locked(&so->so_snd,
2922 			    (int)sbavail(&so->so_snd));
2923 			ourfinisacked = 1;
2924 		} else {
2925 			mfree = sbcut_locked(&so->so_snd, acked);
2926 			if (tp->snd_wnd >= (uint32_t) acked)
2927 				tp->snd_wnd -= acked;
2928 			else
2929 				tp->snd_wnd = 0;
2930 			ourfinisacked = 0;
2931 		}
2932 		SOCKBUF_UNLOCK(&so->so_snd);
2933 		tp->t_flags |= TF_WAKESOW;
2934 		m_freem(mfree);
2935 		/* Detect una wraparound. */
2936 		if (!IN_RECOVERY(tp->t_flags) &&
2937 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2938 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2939 			tp->snd_recover = th->th_ack - 1;
2940 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2941 		if (IN_RECOVERY(tp->t_flags) &&
2942 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2943 			EXIT_RECOVERY(tp->t_flags);
2944 		}
2945 		tp->snd_una = th->th_ack;
2946 		if (tp->t_flags & TF_SACK_PERMIT) {
2947 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2948 				tp->snd_recover = tp->snd_una;
2949 		}
2950 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2951 			tp->snd_nxt = tp->snd_una;
2952 
2953 		switch (tp->t_state) {
2954 		/*
2955 		 * In FIN_WAIT_1 STATE in addition to the processing
2956 		 * for the ESTABLISHED state if our FIN is now acknowledged
2957 		 * then enter FIN_WAIT_2.
2958 		 */
2959 		case TCPS_FIN_WAIT_1:
2960 			if (ourfinisacked) {
2961 				/*
2962 				 * If we can't receive any more
2963 				 * data, then closing user can proceed.
2964 				 * Starting the timer is contrary to the
2965 				 * specification, but if we don't get a FIN
2966 				 * we'll hang forever.
2967 				 *
2968 				 * XXXjl:
2969 				 * we should release the tp also, and use a
2970 				 * compressed state.
2971 				 */
2972 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2973 					soisdisconnected(so);
2974 					tcp_timer_activate(tp, TT_2MSL,
2975 					    (tcp_fast_finwait2_recycle ?
2976 					    tcp_finwait2_timeout :
2977 					    TP_MAXIDLE(tp)));
2978 				}
2979 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
2980 			}
2981 			break;
2982 
2983 		/*
2984 		 * In CLOSING STATE in addition to the processing for
2985 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2986 		 * then enter the TIME-WAIT state, otherwise ignore
2987 		 * the segment.
2988 		 */
2989 		case TCPS_CLOSING:
2990 			if (ourfinisacked) {
2991 				tcp_twstart(tp);
2992 				m_freem(m);
2993 				return;
2994 			}
2995 			break;
2996 
2997 		/*
2998 		 * In LAST_ACK, we may still be waiting for data to drain
2999 		 * and/or to be acked, as well as for the ack of our FIN.
3000 		 * If our FIN is now acknowledged, delete the TCB,
3001 		 * enter the closed state and return.
3002 		 */
3003 		case TCPS_LAST_ACK:
3004 			if (ourfinisacked) {
3005 				tp = tcp_close(tp);
3006 				goto drop;
3007 			}
3008 			break;
3009 		}
3010 	}
3011 
3012 step6:
3013 	INP_WLOCK_ASSERT(tp->t_inpcb);
3014 
3015 	/*
3016 	 * Update window information.
3017 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3018 	 */
3019 	if ((thflags & TH_ACK) &&
3020 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3021 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3022 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3023 		/* keep track of pure window updates */
3024 		if (tlen == 0 &&
3025 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3026 			TCPSTAT_INC(tcps_rcvwinupd);
3027 		tp->snd_wnd = tiwin;
3028 		tp->snd_wl1 = th->th_seq;
3029 		tp->snd_wl2 = th->th_ack;
3030 		if (tp->snd_wnd > tp->max_sndwnd)
3031 			tp->max_sndwnd = tp->snd_wnd;
3032 		needoutput = 1;
3033 	}
3034 
3035 	/*
3036 	 * Process segments with URG.
3037 	 */
3038 	if ((thflags & TH_URG) && th->th_urp &&
3039 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3040 		/*
3041 		 * This is a kludge, but if we receive and accept
3042 		 * random urgent pointers, we'll crash in
3043 		 * soreceive.  It's hard to imagine someone
3044 		 * actually wanting to send this much urgent data.
3045 		 */
3046 		SOCKBUF_LOCK(&so->so_rcv);
3047 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3048 			th->th_urp = 0;			/* XXX */
3049 			thflags &= ~TH_URG;		/* XXX */
3050 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3051 			goto dodata;			/* XXX */
3052 		}
3053 		/*
3054 		 * If this segment advances the known urgent pointer,
3055 		 * then mark the data stream.  This should not happen
3056 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3057 		 * a FIN has been received from the remote side.
3058 		 * In these states we ignore the URG.
3059 		 *
3060 		 * According to RFC961 (Assigned Protocols),
3061 		 * the urgent pointer points to the last octet
3062 		 * of urgent data.  We continue, however,
3063 		 * to consider it to indicate the first octet
3064 		 * of data past the urgent section as the original
3065 		 * spec states (in one of two places).
3066 		 */
3067 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3068 			tp->rcv_up = th->th_seq + th->th_urp;
3069 			so->so_oobmark = sbavail(&so->so_rcv) +
3070 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3071 			if (so->so_oobmark == 0)
3072 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3073 			sohasoutofband(so);
3074 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3075 		}
3076 		SOCKBUF_UNLOCK(&so->so_rcv);
3077 		/*
3078 		 * Remove out of band data so doesn't get presented to user.
3079 		 * This can happen independent of advancing the URG pointer,
3080 		 * but if two URG's are pending at once, some out-of-band
3081 		 * data may creep in... ick.
3082 		 */
3083 		if (th->th_urp <= (uint32_t)tlen &&
3084 		    !(so->so_options & SO_OOBINLINE)) {
3085 			/* hdr drop is delayed */
3086 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3087 		}
3088 	} else {
3089 		/*
3090 		 * If no out of band data is expected,
3091 		 * pull receive urgent pointer along
3092 		 * with the receive window.
3093 		 */
3094 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3095 			tp->rcv_up = tp->rcv_nxt;
3096 	}
3097 dodata:							/* XXX */
3098 	INP_WLOCK_ASSERT(tp->t_inpcb);
3099 
3100 	/*
3101 	 * Process the segment text, merging it into the TCP sequencing queue,
3102 	 * and arranging for acknowledgment of receipt if necessary.
3103 	 * This process logically involves adjusting tp->rcv_wnd as data
3104 	 * is presented to the user (this happens in tcp_usrreq.c,
3105 	 * case PRU_RCVD).  If a FIN has already been received on this
3106 	 * connection then we just ignore the text.
3107 	 */
3108 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3109 		   IS_FASTOPEN(tp->t_flags));
3110 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3111 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3112 		tcp_seq save_start = th->th_seq;
3113 		tcp_seq save_rnxt  = tp->rcv_nxt;
3114 		int     save_tlen  = tlen;
3115 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3116 		/*
3117 		 * Insert segment which includes th into TCP reassembly queue
3118 		 * with control block tp.  Set thflags to whether reassembly now
3119 		 * includes a segment with FIN.  This handles the common case
3120 		 * inline (segment is the next to be received on an established
3121 		 * connection, and the queue is empty), avoiding linkage into
3122 		 * and removal from the queue and repetition of various
3123 		 * conversions.
3124 		 * Set DELACK for segments received in order, but ack
3125 		 * immediately when segments are out of order (so
3126 		 * fast retransmit can work).
3127 		 */
3128 		if (th->th_seq == tp->rcv_nxt &&
3129 		    SEGQ_EMPTY(tp) &&
3130 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3131 		     tfo_syn)) {
3132 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3133 				tp->t_flags |= TF_DELACK;
3134 			else
3135 				tp->t_flags |= TF_ACKNOW;
3136 			tp->rcv_nxt += tlen;
3137 			if (tlen &&
3138 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3139 			    (tp->t_fbyte_in == 0)) {
3140 				tp->t_fbyte_in = ticks;
3141 				if (tp->t_fbyte_in == 0)
3142 					tp->t_fbyte_in = 1;
3143 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3144 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3145 			}
3146 			thflags = th->th_flags & TH_FIN;
3147 			TCPSTAT_INC(tcps_rcvpack);
3148 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3149 			SOCKBUF_LOCK(&so->so_rcv);
3150 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3151 				m_freem(m);
3152 			else
3153 				sbappendstream_locked(&so->so_rcv, m, 0);
3154 			SOCKBUF_UNLOCK(&so->so_rcv);
3155 			tp->t_flags |= TF_WAKESOR;
3156 		} else {
3157 			/*
3158 			 * XXX: Due to the header drop above "th" is
3159 			 * theoretically invalid by now.  Fortunately
3160 			 * m_adj() doesn't actually frees any mbufs
3161 			 * when trimming from the head.
3162 			 */
3163 			tcp_seq temp = save_start;
3164 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3165 			tp->t_flags |= TF_ACKNOW;
3166 		}
3167 		if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3168 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3169 				/*
3170 				 * DSACK actually handled in the fastpath
3171 				 * above.
3172 				 */
3173 				tcp_update_sack_list(tp, save_start,
3174 				    save_start + save_tlen);
3175 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3176 				if ((tp->rcv_numsacks >= 1) &&
3177 				    (tp->sackblks[0].end == save_start)) {
3178 					/*
3179 					 * Partial overlap, recorded at todrop
3180 					 * above.
3181 					 */
3182 					tcp_update_sack_list(tp,
3183 					    tp->sackblks[0].start,
3184 					    tp->sackblks[0].end);
3185 				} else {
3186 					tcp_update_dsack_list(tp, save_start,
3187 					    save_start + save_tlen);
3188 				}
3189 			} else if (tlen >= save_tlen) {
3190 				/* Update of sackblks. */
3191 				tcp_update_dsack_list(tp, save_start,
3192 				    save_start + save_tlen);
3193 			} else if (tlen > 0) {
3194 				tcp_update_dsack_list(tp, save_start,
3195 				    save_start + tlen);
3196 			}
3197 		}
3198 #if 0
3199 		/*
3200 		 * Note the amount of data that peer has sent into
3201 		 * our window, in order to estimate the sender's
3202 		 * buffer size.
3203 		 * XXX: Unused.
3204 		 */
3205 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3206 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3207 		else
3208 			len = so->so_rcv.sb_hiwat;
3209 #endif
3210 	} else {
3211 		m_freem(m);
3212 		thflags &= ~TH_FIN;
3213 	}
3214 
3215 	/*
3216 	 * If FIN is received ACK the FIN and let the user know
3217 	 * that the connection is closing.
3218 	 */
3219 	if (thflags & TH_FIN) {
3220 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3221 			socantrcvmore(so);
3222 			/* The socket upcall is handled by socantrcvmore. */
3223 			tp->t_flags &= ~TF_WAKESOR;
3224 			/*
3225 			 * If connection is half-synchronized
3226 			 * (ie NEEDSYN flag on) then delay ACK,
3227 			 * so it may be piggybacked when SYN is sent.
3228 			 * Otherwise, since we received a FIN then no
3229 			 * more input can be expected, send ACK now.
3230 			 */
3231 			if (tp->t_flags & TF_NEEDSYN)
3232 				tp->t_flags |= TF_DELACK;
3233 			else
3234 				tp->t_flags |= TF_ACKNOW;
3235 			tp->rcv_nxt++;
3236 		}
3237 		switch (tp->t_state) {
3238 		/*
3239 		 * In SYN_RECEIVED and ESTABLISHED STATES
3240 		 * enter the CLOSE_WAIT state.
3241 		 */
3242 		case TCPS_SYN_RECEIVED:
3243 			tp->t_starttime = ticks;
3244 			/* FALLTHROUGH */
3245 		case TCPS_ESTABLISHED:
3246 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3247 			break;
3248 
3249 		/*
3250 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3251 		 * enter the CLOSING state.
3252 		 */
3253 		case TCPS_FIN_WAIT_1:
3254 			tcp_state_change(tp, TCPS_CLOSING);
3255 			break;
3256 
3257 		/*
3258 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3259 		 * starting the time-wait timer, turning off the other
3260 		 * standard timers.
3261 		 */
3262 		case TCPS_FIN_WAIT_2:
3263 			tcp_twstart(tp);
3264 			return;
3265 		}
3266 	}
3267 #ifdef TCPDEBUG
3268 	if (so->so_options & SO_DEBUG)
3269 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3270 			  &tcp_savetcp, 0);
3271 #endif
3272 	TCP_PROBE3(debug__input, tp, th, m);
3273 
3274 	/*
3275 	 * Return any desired output.
3276 	 */
3277 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3278 		(void) tp->t_fb->tfb_tcp_output(tp);
3279 
3280 check_delack:
3281 	INP_WLOCK_ASSERT(tp->t_inpcb);
3282 
3283 	if (tp->t_flags & TF_DELACK) {
3284 		tp->t_flags &= ~TF_DELACK;
3285 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3286 	}
3287 	tcp_handle_wakeup(tp, so);
3288 	INP_WUNLOCK(tp->t_inpcb);
3289 	return;
3290 
3291 dropafterack:
3292 	/*
3293 	 * Generate an ACK dropping incoming segment if it occupies
3294 	 * sequence space, where the ACK reflects our state.
3295 	 *
3296 	 * We can now skip the test for the RST flag since all
3297 	 * paths to this code happen after packets containing
3298 	 * RST have been dropped.
3299 	 *
3300 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3301 	 * segment we received passes the SYN-RECEIVED ACK test.
3302 	 * If it fails send a RST.  This breaks the loop in the
3303 	 * "LAND" DoS attack, and also prevents an ACK storm
3304 	 * between two listening ports that have been sent forged
3305 	 * SYN segments, each with the source address of the other.
3306 	 */
3307 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3308 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3309 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3310 		rstreason = BANDLIM_RST_OPENPORT;
3311 		goto dropwithreset;
3312 	}
3313 #ifdef TCPDEBUG
3314 	if (so->so_options & SO_DEBUG)
3315 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3316 			  &tcp_savetcp, 0);
3317 #endif
3318 	TCP_PROBE3(debug__input, tp, th, m);
3319 	tp->t_flags |= TF_ACKNOW;
3320 	(void) tp->t_fb->tfb_tcp_output(tp);
3321 	tcp_handle_wakeup(tp, so);
3322 	INP_WUNLOCK(tp->t_inpcb);
3323 	m_freem(m);
3324 	return;
3325 
3326 dropwithreset:
3327 	if (tp != NULL) {
3328 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3329 		tcp_handle_wakeup(tp, so);
3330 		INP_WUNLOCK(tp->t_inpcb);
3331 	} else
3332 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3333 	return;
3334 
3335 drop:
3336 	/*
3337 	 * Drop space held by incoming segment and return.
3338 	 */
3339 #ifdef TCPDEBUG
3340 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3341 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3342 			  &tcp_savetcp, 0);
3343 #endif
3344 	TCP_PROBE3(debug__input, tp, th, m);
3345 	if (tp != NULL) {
3346 		tcp_handle_wakeup(tp, so);
3347 		INP_WUNLOCK(tp->t_inpcb);
3348 	}
3349 	m_freem(m);
3350 }
3351 
3352 /*
3353  * Issue RST and make ACK acceptable to originator of segment.
3354  * The mbuf must still include the original packet header.
3355  * tp may be NULL.
3356  */
3357 void
3358 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3359     int tlen, int rstreason)
3360 {
3361 #ifdef INET
3362 	struct ip *ip;
3363 #endif
3364 #ifdef INET6
3365 	struct ip6_hdr *ip6;
3366 #endif
3367 
3368 	if (tp != NULL) {
3369 		INP_WLOCK_ASSERT(tp->t_inpcb);
3370 	}
3371 
3372 	/* Don't bother if destination was broadcast/multicast. */
3373 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3374 		goto drop;
3375 #ifdef INET6
3376 	if (mtod(m, struct ip *)->ip_v == 6) {
3377 		ip6 = mtod(m, struct ip6_hdr *);
3378 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3379 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3380 			goto drop;
3381 		/* IPv6 anycast check is done at tcp6_input() */
3382 	}
3383 #endif
3384 #if defined(INET) && defined(INET6)
3385 	else
3386 #endif
3387 #ifdef INET
3388 	{
3389 		ip = mtod(m, struct ip *);
3390 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3391 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3392 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3393 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3394 			goto drop;
3395 	}
3396 #endif
3397 
3398 	/* Perform bandwidth limiting. */
3399 	if (badport_bandlim(rstreason) < 0)
3400 		goto drop;
3401 
3402 	/* tcp_respond consumes the mbuf chain. */
3403 	if (th->th_flags & TH_ACK) {
3404 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3405 		    th->th_ack, TH_RST);
3406 	} else {
3407 		if (th->th_flags & TH_SYN)
3408 			tlen++;
3409 		if (th->th_flags & TH_FIN)
3410 			tlen++;
3411 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3412 		    (tcp_seq)0, TH_RST|TH_ACK);
3413 	}
3414 	return;
3415 drop:
3416 	m_freem(m);
3417 }
3418 
3419 /*
3420  * Parse TCP options and place in tcpopt.
3421  */
3422 void
3423 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3424 {
3425 	int opt, optlen;
3426 
3427 	to->to_flags = 0;
3428 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3429 		opt = cp[0];
3430 		if (opt == TCPOPT_EOL)
3431 			break;
3432 		if (opt == TCPOPT_NOP)
3433 			optlen = 1;
3434 		else {
3435 			if (cnt < 2)
3436 				break;
3437 			optlen = cp[1];
3438 			if (optlen < 2 || optlen > cnt)
3439 				break;
3440 		}
3441 		switch (opt) {
3442 		case TCPOPT_MAXSEG:
3443 			if (optlen != TCPOLEN_MAXSEG)
3444 				continue;
3445 			if (!(flags & TO_SYN))
3446 				continue;
3447 			to->to_flags |= TOF_MSS;
3448 			bcopy((char *)cp + 2,
3449 			    (char *)&to->to_mss, sizeof(to->to_mss));
3450 			to->to_mss = ntohs(to->to_mss);
3451 			break;
3452 		case TCPOPT_WINDOW:
3453 			if (optlen != TCPOLEN_WINDOW)
3454 				continue;
3455 			if (!(flags & TO_SYN))
3456 				continue;
3457 			to->to_flags |= TOF_SCALE;
3458 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3459 			break;
3460 		case TCPOPT_TIMESTAMP:
3461 			if (optlen != TCPOLEN_TIMESTAMP)
3462 				continue;
3463 			to->to_flags |= TOF_TS;
3464 			bcopy((char *)cp + 2,
3465 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3466 			to->to_tsval = ntohl(to->to_tsval);
3467 			bcopy((char *)cp + 6,
3468 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3469 			to->to_tsecr = ntohl(to->to_tsecr);
3470 			break;
3471 		case TCPOPT_SIGNATURE:
3472 			/*
3473 			 * In order to reply to a host which has set the
3474 			 * TCP_SIGNATURE option in its initial SYN, we have
3475 			 * to record the fact that the option was observed
3476 			 * here for the syncache code to perform the correct
3477 			 * response.
3478 			 */
3479 			if (optlen != TCPOLEN_SIGNATURE)
3480 				continue;
3481 			to->to_flags |= TOF_SIGNATURE;
3482 			to->to_signature = cp + 2;
3483 			break;
3484 		case TCPOPT_SACK_PERMITTED:
3485 			if (optlen != TCPOLEN_SACK_PERMITTED)
3486 				continue;
3487 			if (!(flags & TO_SYN))
3488 				continue;
3489 			if (!V_tcp_do_sack)
3490 				continue;
3491 			to->to_flags |= TOF_SACKPERM;
3492 			break;
3493 		case TCPOPT_SACK:
3494 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3495 				continue;
3496 			if (flags & TO_SYN)
3497 				continue;
3498 			to->to_flags |= TOF_SACK;
3499 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3500 			to->to_sacks = cp + 2;
3501 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3502 			break;
3503 		case TCPOPT_FAST_OPEN:
3504 			/*
3505 			 * Cookie length validation is performed by the
3506 			 * server side cookie checking code or the client
3507 			 * side cookie cache update code.
3508 			 */
3509 			if (!(flags & TO_SYN))
3510 				continue;
3511 			if (!V_tcp_fastopen_client_enable &&
3512 			    !V_tcp_fastopen_server_enable)
3513 				continue;
3514 			to->to_flags |= TOF_FASTOPEN;
3515 			to->to_tfo_len = optlen - 2;
3516 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3517 			break;
3518 		default:
3519 			continue;
3520 		}
3521 	}
3522 }
3523 
3524 /*
3525  * Pull out of band byte out of a segment so
3526  * it doesn't appear in the user's data queue.
3527  * It is still reflected in the segment length for
3528  * sequencing purposes.
3529  */
3530 void
3531 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3532     int off)
3533 {
3534 	int cnt = off + th->th_urp - 1;
3535 
3536 	while (cnt >= 0) {
3537 		if (m->m_len > cnt) {
3538 			char *cp = mtod(m, caddr_t) + cnt;
3539 			struct tcpcb *tp = sototcpcb(so);
3540 
3541 			INP_WLOCK_ASSERT(tp->t_inpcb);
3542 
3543 			tp->t_iobc = *cp;
3544 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3545 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3546 			m->m_len--;
3547 			if (m->m_flags & M_PKTHDR)
3548 				m->m_pkthdr.len--;
3549 			return;
3550 		}
3551 		cnt -= m->m_len;
3552 		m = m->m_next;
3553 		if (m == NULL)
3554 			break;
3555 	}
3556 	panic("tcp_pulloutofband");
3557 }
3558 
3559 /*
3560  * Collect new round-trip time estimate
3561  * and update averages and current timeout.
3562  */
3563 void
3564 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3565 {
3566 	int delta;
3567 
3568 	INP_WLOCK_ASSERT(tp->t_inpcb);
3569 
3570 	TCPSTAT_INC(tcps_rttupdated);
3571 	tp->t_rttupdated++;
3572 #ifdef STATS
3573 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3574 	    imax(0, rtt * 1000 / hz));
3575 #endif
3576 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3577 		/*
3578 		 * srtt is stored as fixed point with 5 bits after the
3579 		 * binary point (i.e., scaled by 8).  The following magic
3580 		 * is equivalent to the smoothing algorithm in rfc793 with
3581 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3582 		 * point).  Adjust rtt to origin 0.
3583 		 */
3584 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3585 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3586 
3587 		if ((tp->t_srtt += delta) <= 0)
3588 			tp->t_srtt = 1;
3589 
3590 		/*
3591 		 * We accumulate a smoothed rtt variance (actually, a
3592 		 * smoothed mean difference), then set the retransmit
3593 		 * timer to smoothed rtt + 4 times the smoothed variance.
3594 		 * rttvar is stored as fixed point with 4 bits after the
3595 		 * binary point (scaled by 16).  The following is
3596 		 * equivalent to rfc793 smoothing with an alpha of .75
3597 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3598 		 * rfc793's wired-in beta.
3599 		 */
3600 		if (delta < 0)
3601 			delta = -delta;
3602 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3603 		if ((tp->t_rttvar += delta) <= 0)
3604 			tp->t_rttvar = 1;
3605 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3606 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3607 	} else {
3608 		/*
3609 		 * No rtt measurement yet - use the unsmoothed rtt.
3610 		 * Set the variance to half the rtt (so our first
3611 		 * retransmit happens at 3*rtt).
3612 		 */
3613 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3614 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3615 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3616 	}
3617 	tp->t_rtttime = 0;
3618 	tp->t_rxtshift = 0;
3619 
3620 	/*
3621 	 * the retransmit should happen at rtt + 4 * rttvar.
3622 	 * Because of the way we do the smoothing, srtt and rttvar
3623 	 * will each average +1/2 tick of bias.  When we compute
3624 	 * the retransmit timer, we want 1/2 tick of rounding and
3625 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3626 	 * firing of the timer.  The bias will give us exactly the
3627 	 * 1.5 tick we need.  But, because the bias is
3628 	 * statistical, we have to test that we don't drop below
3629 	 * the minimum feasible timer (which is 2 ticks).
3630 	 */
3631 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3632 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3633 
3634 	/*
3635 	 * We received an ack for a packet that wasn't retransmitted;
3636 	 * it is probably safe to discard any error indications we've
3637 	 * received recently.  This isn't quite right, but close enough
3638 	 * for now (a route might have failed after we sent a segment,
3639 	 * and the return path might not be symmetrical).
3640 	 */
3641 	tp->t_softerror = 0;
3642 }
3643 
3644 /*
3645  * Determine a reasonable value for maxseg size.
3646  * If the route is known, check route for mtu.
3647  * If none, use an mss that can be handled on the outgoing interface
3648  * without forcing IP to fragment.  If no route is found, route has no mtu,
3649  * or the destination isn't local, use a default, hopefully conservative
3650  * size (usually 512 or the default IP max size, but no more than the mtu
3651  * of the interface), as we can't discover anything about intervening
3652  * gateways or networks.  We also initialize the congestion/slow start
3653  * window to be a single segment if the destination isn't local.
3654  * While looking at the routing entry, we also initialize other path-dependent
3655  * parameters from pre-set or cached values in the routing entry.
3656  *
3657  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3658  * IP options, e.g. IPSEC data, since length of this data may vary, and
3659  * thus it is calculated for every segment separately in tcp_output().
3660  *
3661  * NOTE that this routine is only called when we process an incoming
3662  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3663  * settings are handled in tcp_mssopt().
3664  */
3665 void
3666 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3667     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3668 {
3669 	int mss = 0;
3670 	uint32_t maxmtu = 0;
3671 	struct inpcb *inp = tp->t_inpcb;
3672 	struct hc_metrics_lite metrics;
3673 #ifdef INET6
3674 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3675 	size_t min_protoh = isipv6 ?
3676 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3677 			    sizeof (struct tcpiphdr);
3678 #else
3679 	const size_t min_protoh = sizeof(struct tcpiphdr);
3680 #endif
3681 
3682 	INP_WLOCK_ASSERT(tp->t_inpcb);
3683 
3684 	if (mtuoffer != -1) {
3685 		KASSERT(offer == -1, ("%s: conflict", __func__));
3686 		offer = mtuoffer - min_protoh;
3687 	}
3688 
3689 	/* Initialize. */
3690 #ifdef INET6
3691 	if (isipv6) {
3692 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3693 		tp->t_maxseg = V_tcp_v6mssdflt;
3694 	}
3695 #endif
3696 #if defined(INET) && defined(INET6)
3697 	else
3698 #endif
3699 #ifdef INET
3700 	{
3701 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3702 		tp->t_maxseg = V_tcp_mssdflt;
3703 	}
3704 #endif
3705 
3706 	/*
3707 	 * No route to sender, stay with default mss and return.
3708 	 */
3709 	if (maxmtu == 0) {
3710 		/*
3711 		 * In case we return early we need to initialize metrics
3712 		 * to a defined state as tcp_hc_get() would do for us
3713 		 * if there was no cache hit.
3714 		 */
3715 		if (metricptr != NULL)
3716 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3717 		return;
3718 	}
3719 
3720 	/* What have we got? */
3721 	switch (offer) {
3722 		case 0:
3723 			/*
3724 			 * Offer == 0 means that there was no MSS on the SYN
3725 			 * segment, in this case we use tcp_mssdflt as
3726 			 * already assigned to t_maxseg above.
3727 			 */
3728 			offer = tp->t_maxseg;
3729 			break;
3730 
3731 		case -1:
3732 			/*
3733 			 * Offer == -1 means that we didn't receive SYN yet.
3734 			 */
3735 			/* FALLTHROUGH */
3736 
3737 		default:
3738 			/*
3739 			 * Prevent DoS attack with too small MSS. Round up
3740 			 * to at least minmss.
3741 			 */
3742 			offer = max(offer, V_tcp_minmss);
3743 	}
3744 
3745 	/*
3746 	 * rmx information is now retrieved from tcp_hostcache.
3747 	 */
3748 	tcp_hc_get(&inp->inp_inc, &metrics);
3749 	if (metricptr != NULL)
3750 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3751 
3752 	/*
3753 	 * If there's a discovered mtu in tcp hostcache, use it.
3754 	 * Else, use the link mtu.
3755 	 */
3756 	if (metrics.rmx_mtu)
3757 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3758 	else {
3759 #ifdef INET6
3760 		if (isipv6) {
3761 			mss = maxmtu - min_protoh;
3762 			if (!V_path_mtu_discovery &&
3763 			    !in6_localaddr(&inp->in6p_faddr))
3764 				mss = min(mss, V_tcp_v6mssdflt);
3765 		}
3766 #endif
3767 #if defined(INET) && defined(INET6)
3768 		else
3769 #endif
3770 #ifdef INET
3771 		{
3772 			mss = maxmtu - min_protoh;
3773 			if (!V_path_mtu_discovery &&
3774 			    !in_localaddr(inp->inp_faddr))
3775 				mss = min(mss, V_tcp_mssdflt);
3776 		}
3777 #endif
3778 		/*
3779 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3780 		 * probably violates the TCP spec.
3781 		 * The problem is that, since we don't know the
3782 		 * other end's MSS, we are supposed to use a conservative
3783 		 * default.  But, if we do that, then MTU discovery will
3784 		 * never actually take place, because the conservative
3785 		 * default is much less than the MTUs typically seen
3786 		 * on the Internet today.  For the moment, we'll sweep
3787 		 * this under the carpet.
3788 		 *
3789 		 * The conservative default might not actually be a problem
3790 		 * if the only case this occurs is when sending an initial
3791 		 * SYN with options and data to a host we've never talked
3792 		 * to before.  Then, they will reply with an MSS value which
3793 		 * will get recorded and the new parameters should get
3794 		 * recomputed.  For Further Study.
3795 		 */
3796 	}
3797 	mss = min(mss, offer);
3798 
3799 	/*
3800 	 * Sanity check: make sure that maxseg will be large
3801 	 * enough to allow some data on segments even if the
3802 	 * all the option space is used (40bytes).  Otherwise
3803 	 * funny things may happen in tcp_output.
3804 	 *
3805 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3806 	 */
3807 	mss = max(mss, 64);
3808 
3809 	tp->t_maxseg = mss;
3810 }
3811 
3812 void
3813 tcp_mss(struct tcpcb *tp, int offer)
3814 {
3815 	int mss;
3816 	uint32_t bufsize;
3817 	struct inpcb *inp;
3818 	struct socket *so;
3819 	struct hc_metrics_lite metrics;
3820 	struct tcp_ifcap cap;
3821 
3822 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3823 
3824 	bzero(&cap, sizeof(cap));
3825 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3826 
3827 	mss = tp->t_maxseg;
3828 	inp = tp->t_inpcb;
3829 
3830 	/*
3831 	 * If there's a pipesize, change the socket buffer to that size,
3832 	 * don't change if sb_hiwat is different than default (then it
3833 	 * has been changed on purpose with setsockopt).
3834 	 * Make the socket buffers an integral number of mss units;
3835 	 * if the mss is larger than the socket buffer, decrease the mss.
3836 	 */
3837 	so = inp->inp_socket;
3838 	SOCKBUF_LOCK(&so->so_snd);
3839 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3840 		bufsize = metrics.rmx_sendpipe;
3841 	else
3842 		bufsize = so->so_snd.sb_hiwat;
3843 	if (bufsize < mss)
3844 		mss = bufsize;
3845 	else {
3846 		bufsize = roundup(bufsize, mss);
3847 		if (bufsize > sb_max)
3848 			bufsize = sb_max;
3849 		if (bufsize > so->so_snd.sb_hiwat)
3850 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3851 	}
3852 	SOCKBUF_UNLOCK(&so->so_snd);
3853 	/*
3854 	 * Sanity check: make sure that maxseg will be large
3855 	 * enough to allow some data on segments even if the
3856 	 * all the option space is used (40bytes).  Otherwise
3857 	 * funny things may happen in tcp_output.
3858 	 *
3859 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3860 	 */
3861 	tp->t_maxseg = max(mss, 64);
3862 
3863 	SOCKBUF_LOCK(&so->so_rcv);
3864 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3865 		bufsize = metrics.rmx_recvpipe;
3866 	else
3867 		bufsize = so->so_rcv.sb_hiwat;
3868 	if (bufsize > mss) {
3869 		bufsize = roundup(bufsize, mss);
3870 		if (bufsize > sb_max)
3871 			bufsize = sb_max;
3872 		if (bufsize > so->so_rcv.sb_hiwat)
3873 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3874 	}
3875 	SOCKBUF_UNLOCK(&so->so_rcv);
3876 
3877 	/* Check the interface for TSO capabilities. */
3878 	if (cap.ifcap & CSUM_TSO) {
3879 		tp->t_flags |= TF_TSO;
3880 		tp->t_tsomax = cap.tsomax;
3881 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3882 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3883 	}
3884 }
3885 
3886 /*
3887  * Determine the MSS option to send on an outgoing SYN.
3888  */
3889 int
3890 tcp_mssopt(struct in_conninfo *inc)
3891 {
3892 	int mss = 0;
3893 	uint32_t thcmtu = 0;
3894 	uint32_t maxmtu = 0;
3895 	size_t min_protoh;
3896 
3897 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3898 
3899 #ifdef INET6
3900 	if (inc->inc_flags & INC_ISIPV6) {
3901 		mss = V_tcp_v6mssdflt;
3902 		maxmtu = tcp_maxmtu6(inc, NULL);
3903 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3904 	}
3905 #endif
3906 #if defined(INET) && defined(INET6)
3907 	else
3908 #endif
3909 #ifdef INET
3910 	{
3911 		mss = V_tcp_mssdflt;
3912 		maxmtu = tcp_maxmtu(inc, NULL);
3913 		min_protoh = sizeof(struct tcpiphdr);
3914 	}
3915 #endif
3916 #if defined(INET6) || defined(INET)
3917 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3918 #endif
3919 
3920 	if (maxmtu && thcmtu)
3921 		mss = min(maxmtu, thcmtu) - min_protoh;
3922 	else if (maxmtu || thcmtu)
3923 		mss = max(maxmtu, thcmtu) - min_protoh;
3924 
3925 	return (mss);
3926 }
3927 
3928 void
3929 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3930 {
3931 	long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3932 	int maxseg = tcp_maxseg(tp);
3933 
3934 	INP_WLOCK_ASSERT(tp->t_inpcb);
3935 
3936 	tcp_timer_activate(tp, TT_REXMT, 0);
3937 	tp->t_rtttime = 0;
3938 	/*
3939 	 * Compute the amount of data that this ACK is indicating
3940 	 * (del_data) and an estimate of how many bytes are in the
3941 	 * network.
3942 	 */
3943 	if (SEQ_GEQ(th->th_ack, tp->snd_una))
3944 		del_data = BYTES_THIS_ACK(tp, th);
3945 	del_data += tp->sackhint.delivered_data;
3946 	pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3947 	tp->sackhint.prr_delivered += del_data;
3948 	/*
3949 	 * Proportional Rate Reduction
3950 	 */
3951 	if (pipe > tp->snd_ssthresh)
3952 		snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh / tp->sackhint.recover_fs) -
3953 		    tp->sackhint.sack_bytes_rexmit;
3954 	else {
3955 		if (V_tcp_do_prr_conservative)
3956 			limit = tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit;
3957 		else
3958 			if ((tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit) > del_data)
3959 				limit = tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit + maxseg;
3960 			else
3961 				limit = del_data + maxseg;
3962 		snd_cnt = min((tp->snd_ssthresh - pipe), limit);
3963 	}
3964 	snd_cnt = max((snd_cnt / maxseg), 0);
3965 	/*
3966 	 * Send snd_cnt new data into the network in response to this ack.
3967 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
3968 	 * accordingly.
3969 	 */
3970 	tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3971 		tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg);
3972 	tp->t_flags |= TF_ACKNOW;
3973 	(void) tcp_output(tp);
3974 }
3975 
3976 /*
3977  * On a partial ack arrives, force the retransmission of the
3978  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3979  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3980  * be started again.
3981  */
3982 void
3983 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3984 {
3985 	tcp_seq onxt = tp->snd_nxt;
3986 	uint32_t ocwnd = tp->snd_cwnd;
3987 	u_int maxseg = tcp_maxseg(tp);
3988 
3989 	INP_WLOCK_ASSERT(tp->t_inpcb);
3990 
3991 	tcp_timer_activate(tp, TT_REXMT, 0);
3992 	tp->t_rtttime = 0;
3993 	tp->snd_nxt = th->th_ack;
3994 	/*
3995 	 * Set snd_cwnd to one segment beyond acknowledged offset.
3996 	 * (tp->snd_una has not yet been updated when this function is called.)
3997 	 */
3998 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3999 	tp->t_flags |= TF_ACKNOW;
4000 	(void) tp->t_fb->tfb_tcp_output(tp);
4001 	tp->snd_cwnd = ocwnd;
4002 	if (SEQ_GT(onxt, tp->snd_nxt))
4003 		tp->snd_nxt = onxt;
4004 	/*
4005 	 * Partial window deflation.  Relies on fact that tp->snd_una
4006 	 * not updated yet.
4007 	 */
4008 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4009 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4010 	else
4011 		tp->snd_cwnd = 0;
4012 	tp->snd_cwnd += maxseg;
4013 }
4014 
4015 int
4016 tcp_compute_pipe(struct tcpcb *tp)
4017 {
4018 	return (tp->snd_max - tp->snd_una +
4019 		tp->sackhint.sack_bytes_rexmit -
4020 		tp->sackhint.sacked_bytes);
4021 }
4022 
4023 uint32_t
4024 tcp_compute_initwnd(uint32_t maxseg)
4025 {
4026 	/*
4027 	 * Calculate the Initial Window, also used as Restart Window
4028 	 *
4029 	 * RFC5681 Section 3.1 specifies the default conservative values.
4030 	 * RFC3390 specifies slightly more aggressive values.
4031 	 * RFC6928 increases it to ten segments.
4032 	 * Support for user specified value for initial flight size.
4033 	 */
4034 	if (V_tcp_initcwnd_segments)
4035 		return min(V_tcp_initcwnd_segments * maxseg,
4036 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4037 	else if (V_tcp_do_rfc3390)
4038 		return min(4 * maxseg, max(2 * maxseg, 4380));
4039 	else {
4040 		/* Per RFC5681 Section 3.1 */
4041 		if (maxseg > 2190)
4042 			return (2 * maxseg);
4043 		else if (maxseg > 1095)
4044 			return (3 * maxseg);
4045 		else
4046 			return (4 * maxseg);
4047 	}
4048 }
4049