xref: /freebsd/sys/netinet/tcp_input.c (revision d485c77f203fb0f4cdc08dea5ff81631b51d8809)
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 }
514 
515 /*
516  * Indicate whether this ack should be delayed.  We can delay the ack if
517  * following conditions are met:
518  *	- There is no delayed ack timer in progress.
519  *	- Our last ack wasn't a 0-sized window. We never want to delay
520  *	  the ack that opens up a 0-sized window.
521  *	- LRO wasn't used for this segment. We make sure by checking that the
522  *	  segment size is not larger than the MSS.
523  */
524 #define DELAY_ACK(tp, tlen)						\
525 	((!tcp_timer_active(tp, TT_DELACK) &&				\
526 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
527 	    (tlen <= tp->t_maxseg) &&					\
528 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
529 
530 void inline
531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
532 {
533 	INP_WLOCK_ASSERT(tp->t_inpcb);
534 
535 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
536 		switch (iptos & IPTOS_ECN_MASK) {
537 		case IPTOS_ECN_CE:
538 			tp->ccv->flags |= CCF_IPHDR_CE;
539 			break;
540 		case IPTOS_ECN_ECT0:
541 			/* FALLTHROUGH */
542 		case IPTOS_ECN_ECT1:
543 			/* FALLTHROUGH */
544 		case IPTOS_ECN_NOTECT:
545 			tp->ccv->flags &= ~CCF_IPHDR_CE;
546 			break;
547 		}
548 
549 		if (th->th_flags & TH_CWR)
550 			tp->ccv->flags |= CCF_TCPHDR_CWR;
551 		else
552 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
553 
554 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
555 
556 		if (tp->ccv->flags & CCF_ACKNOW) {
557 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
558 			tp->t_flags |= TF_ACKNOW;
559 		}
560 	}
561 }
562 
563 /*
564  * TCP input handling is split into multiple parts:
565  *   tcp6_input is a thin wrapper around tcp_input for the extended
566  *	ip6_protox[] call format in ip6_input
567  *   tcp_input handles primary segment validation, inpcb lookup and
568  *	SYN processing on listen sockets
569  *   tcp_do_segment processes the ACK and text of the segment for
570  *	establishing, established and closing connections
571  */
572 #ifdef INET6
573 int
574 tcp6_input(struct mbuf **mp, int *offp, int proto)
575 {
576 	struct mbuf *m;
577 	struct in6_ifaddr *ia6;
578 	struct ip6_hdr *ip6;
579 
580 	m = *mp;
581 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
582 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
583 		if (m == NULL) {
584 			*mp = m;
585 			TCPSTAT_INC(tcps_rcvshort);
586 			return (IPPROTO_DONE);
587 		}
588 	}
589 
590 	/*
591 	 * draft-itojun-ipv6-tcp-to-anycast
592 	 * better place to put this in?
593 	 */
594 	ip6 = mtod(m, struct ip6_hdr *);
595 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
596 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
597 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
598 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
599 		*mp = NULL;
600 		return (IPPROTO_DONE);
601 	}
602 
603 	*mp = m;
604 	return (tcp_input(mp, offp, proto));
605 }
606 #endif /* INET6 */
607 
608 int
609 tcp_input(struct mbuf **mp, int *offp, int proto)
610 {
611 	struct mbuf *m = *mp;
612 	struct tcphdr *th = NULL;
613 	struct ip *ip = NULL;
614 	struct inpcb *inp = NULL;
615 	struct tcpcb *tp = NULL;
616 	struct socket *so = NULL;
617 	u_char *optp = NULL;
618 	int off0;
619 	int optlen = 0;
620 #ifdef INET
621 	int len;
622 	uint8_t ipttl;
623 #endif
624 	int tlen = 0, off;
625 	int drop_hdrlen;
626 	int thflags;
627 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
628 	uint8_t iptos;
629 	struct m_tag *fwd_tag = NULL;
630 #ifdef INET6
631 	struct ip6_hdr *ip6 = NULL;
632 	int isipv6;
633 #else
634 	const void *ip6 = NULL;
635 #endif /* INET6 */
636 	struct tcpopt to;		/* options in this segment */
637 	char *s = NULL;			/* address and port logging */
638 #ifdef TCPDEBUG
639 	/*
640 	 * The size of tcp_saveipgen must be the size of the max ip header,
641 	 * now IPv6.
642 	 */
643 	u_char tcp_saveipgen[IP6_HDR_LEN];
644 	struct tcphdr tcp_savetcp;
645 	short ostate = 0;
646 #endif
647 
648 	NET_EPOCH_ASSERT();
649 
650 #ifdef INET6
651 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
652 #endif
653 
654 	off0 = *offp;
655 	m = *mp;
656 	*mp = NULL;
657 	to.to_flags = 0;
658 	TCPSTAT_INC(tcps_rcvtotal);
659 
660 #ifdef INET6
661 	if (isipv6) {
662 		ip6 = mtod(m, struct ip6_hdr *);
663 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
664 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
665 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
666 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
667 				th->th_sum = m->m_pkthdr.csum_data;
668 			else
669 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
670 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
671 			th->th_sum ^= 0xffff;
672 		} else
673 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
674 		if (th->th_sum) {
675 			TCPSTAT_INC(tcps_rcvbadsum);
676 			goto drop;
677 		}
678 
679 		/*
680 		 * Be proactive about unspecified IPv6 address in source.
681 		 * As we use all-zero to indicate unbounded/unconnected pcb,
682 		 * unspecified IPv6 address can be used to confuse us.
683 		 *
684 		 * Note that packets with unspecified IPv6 destination is
685 		 * already dropped in ip6_input.
686 		 */
687 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
688 			/* XXX stat */
689 			goto drop;
690 		}
691 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
692 	}
693 #endif
694 #if defined(INET) && defined(INET6)
695 	else
696 #endif
697 #ifdef INET
698 	{
699 		/*
700 		 * Get IP and TCP header together in first mbuf.
701 		 * Note: IP leaves IP header in first mbuf.
702 		 */
703 		if (off0 > sizeof (struct ip)) {
704 			ip_stripoptions(m);
705 			off0 = sizeof(struct ip);
706 		}
707 		if (m->m_len < sizeof (struct tcpiphdr)) {
708 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
709 			    == NULL) {
710 				TCPSTAT_INC(tcps_rcvshort);
711 				return (IPPROTO_DONE);
712 			}
713 		}
714 		ip = mtod(m, struct ip *);
715 		th = (struct tcphdr *)((caddr_t)ip + off0);
716 		tlen = ntohs(ip->ip_len) - off0;
717 
718 		iptos = ip->ip_tos;
719 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
720 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
721 				th->th_sum = m->m_pkthdr.csum_data;
722 			else
723 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
724 				    ip->ip_dst.s_addr,
725 				    htonl(m->m_pkthdr.csum_data + tlen +
726 				    IPPROTO_TCP));
727 			th->th_sum ^= 0xffff;
728 		} else {
729 			struct ipovly *ipov = (struct ipovly *)ip;
730 
731 			/*
732 			 * Checksum extended TCP header and data.
733 			 */
734 			len = off0 + tlen;
735 			ipttl = ip->ip_ttl;
736 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
737 			ipov->ih_len = htons(tlen);
738 			th->th_sum = in_cksum(m, len);
739 			/* Reset length for SDT probes. */
740 			ip->ip_len = htons(len);
741 			/* Reset TOS bits */
742 			ip->ip_tos = iptos;
743 			/* Re-initialization for later version check */
744 			ip->ip_ttl = ipttl;
745 			ip->ip_v = IPVERSION;
746 			ip->ip_hl = off0 >> 2;
747 		}
748 
749 		if (th->th_sum) {
750 			TCPSTAT_INC(tcps_rcvbadsum);
751 			goto drop;
752 		}
753 	}
754 #endif /* INET */
755 
756 	/*
757 	 * Check that TCP offset makes sense,
758 	 * pull out TCP options and adjust length.		XXX
759 	 */
760 	off = th->th_off << 2;
761 	if (off < sizeof (struct tcphdr) || off > tlen) {
762 		TCPSTAT_INC(tcps_rcvbadoff);
763 		goto drop;
764 	}
765 	tlen -= off;	/* tlen is used instead of ti->ti_len */
766 	if (off > sizeof (struct tcphdr)) {
767 #ifdef INET6
768 		if (isipv6) {
769 			if (m->m_len < off0 + off) {
770 				m = m_pullup(m, off0 + off);
771 				if (m == NULL) {
772 					TCPSTAT_INC(tcps_rcvshort);
773 					return (IPPROTO_DONE);
774 				}
775 			}
776 			ip6 = mtod(m, struct ip6_hdr *);
777 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
778 		}
779 #endif
780 #if defined(INET) && defined(INET6)
781 		else
782 #endif
783 #ifdef INET
784 		{
785 			if (m->m_len < sizeof(struct ip) + off) {
786 				if ((m = m_pullup(m, sizeof (struct ip) + off))
787 				    == NULL) {
788 					TCPSTAT_INC(tcps_rcvshort);
789 					return (IPPROTO_DONE);
790 				}
791 				ip = mtod(m, struct ip *);
792 				th = (struct tcphdr *)((caddr_t)ip + off0);
793 			}
794 		}
795 #endif
796 		optlen = off - sizeof (struct tcphdr);
797 		optp = (u_char *)(th + 1);
798 	}
799 	thflags = th->th_flags;
800 
801 	/*
802 	 * Convert TCP protocol specific fields to host format.
803 	 */
804 	tcp_fields_to_host(th);
805 
806 	/*
807 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
808 	 */
809 	drop_hdrlen = off0 + off;
810 
811 	/*
812 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
813 	 */
814         if (
815 #ifdef INET6
816 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
817 #ifdef INET
818 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
819 #endif
820 #endif
821 #if defined(INET) && !defined(INET6)
822 	    (m->m_flags & M_IP_NEXTHOP)
823 #endif
824 	    )
825 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
826 
827 findpcb:
828 #ifdef INET6
829 	if (isipv6 && fwd_tag != NULL) {
830 		struct sockaddr_in6 *next_hop6;
831 
832 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
833 		/*
834 		 * Transparently forwarded. Pretend to be the destination.
835 		 * Already got one like this?
836 		 */
837 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
838 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
839 		    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
840 		if (!inp) {
841 			/*
842 			 * It's new.  Try to find the ambushing socket.
843 			 * Because we've rewritten the destination address,
844 			 * any hardware-generated hash is ignored.
845 			 */
846 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
847 			    th->th_sport, &next_hop6->sin6_addr,
848 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
849 			    th->th_dport, INPLOOKUP_WILDCARD |
850 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
851 		}
852 	} else if (isipv6) {
853 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
854 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
855 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
856 		    m->m_pkthdr.rcvif, m);
857 	}
858 #endif /* INET6 */
859 #if defined(INET6) && defined(INET)
860 	else
861 #endif
862 #ifdef INET
863 	if (fwd_tag != NULL) {
864 		struct sockaddr_in *next_hop;
865 
866 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
867 		/*
868 		 * Transparently forwarded. Pretend to be the destination.
869 		 * already got one like this?
870 		 */
871 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
872 		    ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
873 		    m->m_pkthdr.rcvif, m);
874 		if (!inp) {
875 			/*
876 			 * It's new.  Try to find the ambushing socket.
877 			 * Because we've rewritten the destination address,
878 			 * any hardware-generated hash is ignored.
879 			 */
880 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
881 			    th->th_sport, next_hop->sin_addr,
882 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
883 			    th->th_dport, INPLOOKUP_WILDCARD |
884 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
885 		}
886 	} else
887 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
888 		    th->th_sport, ip->ip_dst, th->th_dport,
889 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
890 		    m->m_pkthdr.rcvif, m);
891 #endif /* INET */
892 
893 	/*
894 	 * If the INPCB does not exist then all data in the incoming
895 	 * segment is discarded and an appropriate RST is sent back.
896 	 * XXX MRT Send RST using which routing table?
897 	 */
898 	if (inp == NULL) {
899 		/*
900 		 * Log communication attempts to ports that are not
901 		 * in use.
902 		 */
903 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
904 		    V_tcp_log_in_vain == 2) {
905 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
906 				log(LOG_INFO, "%s; %s: Connection attempt "
907 				    "to closed port\n", s, __func__);
908 		}
909 		/*
910 		 * When blackholing do not respond with a RST but
911 		 * completely ignore the segment and drop it.
912 		 */
913 		if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
914 		    V_blackhole == 2)
915 			goto dropunlock;
916 
917 		rstreason = BANDLIM_RST_CLOSEDPORT;
918 		goto dropwithreset;
919 	}
920 	INP_WLOCK_ASSERT(inp);
921 	/*
922 	 * While waiting for inp lock during the lookup, another thread
923 	 * can have dropped the inpcb, in which case we need to loop back
924 	 * and try to find a new inpcb to deliver to.
925 	 */
926 	if (inp->inp_flags & INP_DROPPED) {
927 		INP_WUNLOCK(inp);
928 		inp = NULL;
929 		goto findpcb;
930 	}
931 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
932 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
933 	    ((inp->inp_socket == NULL) ||
934 	    (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
935 		inp->inp_flowid = m->m_pkthdr.flowid;
936 		inp->inp_flowtype = M_HASHTYPE_GET(m);
937 	}
938 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
939 #ifdef INET6
940 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
941 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
942 		goto dropunlock;
943 	}
944 #ifdef INET
945 	else
946 #endif
947 #endif /* INET6 */
948 #ifdef INET
949 	if (IPSEC_ENABLED(ipv4) &&
950 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
951 		goto dropunlock;
952 	}
953 #endif /* INET */
954 #endif /* IPSEC */
955 
956 	/*
957 	 * Check the minimum TTL for socket.
958 	 */
959 	if (inp->inp_ip_minttl != 0) {
960 #ifdef INET6
961 		if (isipv6) {
962 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
963 				goto dropunlock;
964 		} else
965 #endif
966 		if (inp->inp_ip_minttl > ip->ip_ttl)
967 			goto dropunlock;
968 	}
969 
970 	/*
971 	 * A previous connection in TIMEWAIT state is supposed to catch stray
972 	 * or duplicate segments arriving late.  If this segment was a
973 	 * legitimate new connection attempt, the old INPCB gets removed and
974 	 * we can try again to find a listening socket.
975 	 *
976 	 * At this point, due to earlier optimism, we may hold only an inpcb
977 	 * lock, and not the inpcbinfo write lock.  If so, we need to try to
978 	 * acquire it, or if that fails, acquire a reference on the inpcb,
979 	 * drop all locks, acquire a global write lock, and then re-acquire
980 	 * the inpcb lock.  We may at that point discover that another thread
981 	 * has tried to free the inpcb, in which case we need to loop back
982 	 * and try to find a new inpcb to deliver to.
983 	 *
984 	 * XXXRW: It may be time to rethink timewait locking.
985 	 */
986 	if (inp->inp_flags & INP_TIMEWAIT) {
987 		tcp_dooptions(&to, optp, optlen,
988 		    (thflags & TH_SYN) ? TO_SYN : 0);
989 		/*
990 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
991 		 */
992 		if (tcp_twcheck(inp, &to, th, m, tlen))
993 			goto findpcb;
994 		return (IPPROTO_DONE);
995 	}
996 	/*
997 	 * The TCPCB may no longer exist if the connection is winding
998 	 * down or it is in the CLOSED state.  Either way we drop the
999 	 * segment and send an appropriate response.
1000 	 */
1001 	tp = intotcpcb(inp);
1002 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1003 		rstreason = BANDLIM_RST_CLOSEDPORT;
1004 		goto dropwithreset;
1005 	}
1006 
1007 #ifdef TCP_OFFLOAD
1008 	if (tp->t_flags & TF_TOE) {
1009 		tcp_offload_input(tp, m);
1010 		m = NULL;	/* consumed by the TOE driver */
1011 		goto dropunlock;
1012 	}
1013 #endif
1014 
1015 #ifdef MAC
1016 	INP_WLOCK_ASSERT(inp);
1017 	if (mac_inpcb_check_deliver(inp, m))
1018 		goto dropunlock;
1019 #endif
1020 	so = inp->inp_socket;
1021 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1022 #ifdef TCPDEBUG
1023 	if (so->so_options & SO_DEBUG) {
1024 		ostate = tp->t_state;
1025 #ifdef INET6
1026 		if (isipv6) {
1027 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1028 		} else
1029 #endif
1030 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1031 		tcp_savetcp = *th;
1032 	}
1033 #endif /* TCPDEBUG */
1034 	/*
1035 	 * When the socket is accepting connections (the INPCB is in LISTEN
1036 	 * state) we look into the SYN cache if this is a new connection
1037 	 * attempt or the completion of a previous one.
1038 	 */
1039 	KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1040 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1041 	if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1042 		struct in_conninfo inc;
1043 
1044 		bzero(&inc, sizeof(inc));
1045 #ifdef INET6
1046 		if (isipv6) {
1047 			inc.inc_flags |= INC_ISIPV6;
1048 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1049 				inc.inc_flags |= INC_IPV6MINMTU;
1050 			inc.inc6_faddr = ip6->ip6_src;
1051 			inc.inc6_laddr = ip6->ip6_dst;
1052 		} else
1053 #endif
1054 		{
1055 			inc.inc_faddr = ip->ip_src;
1056 			inc.inc_laddr = ip->ip_dst;
1057 		}
1058 		inc.inc_fport = th->th_sport;
1059 		inc.inc_lport = th->th_dport;
1060 		inc.inc_fibnum = so->so_fibnum;
1061 
1062 		/*
1063 		 * Check for an existing connection attempt in syncache if
1064 		 * the flag is only ACK.  A successful lookup creates a new
1065 		 * socket appended to the listen queue in SYN_RECEIVED state.
1066 		 */
1067 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1068 			/*
1069 			 * Parse the TCP options here because
1070 			 * syncookies need access to the reflected
1071 			 * timestamp.
1072 			 */
1073 			tcp_dooptions(&to, optp, optlen, 0);
1074 			/*
1075 			 * NB: syncache_expand() doesn't unlock
1076 			 * inp and tcpinfo locks.
1077 			 */
1078 			rstreason = syncache_expand(&inc, &to, th, &so, m);
1079 			if (rstreason < 0) {
1080 				/*
1081 				 * A failing TCP MD5 signature comparison
1082 				 * must result in the segment being dropped
1083 				 * and must not produce any response back
1084 				 * to the sender.
1085 				 */
1086 				goto dropunlock;
1087 			} else if (rstreason == 0) {
1088 				/*
1089 				 * No syncache entry or ACK was not
1090 				 * for our SYN/ACK.  Send a RST.
1091 				 * NB: syncache did its own logging
1092 				 * of the failure cause.
1093 				 */
1094 				rstreason = BANDLIM_RST_OPENPORT;
1095 				goto dropwithreset;
1096 			}
1097 tfo_socket_result:
1098 			if (so == NULL) {
1099 				/*
1100 				 * We completed the 3-way handshake
1101 				 * but could not allocate a socket
1102 				 * either due to memory shortage,
1103 				 * listen queue length limits or
1104 				 * global socket limits.  Send RST
1105 				 * or wait and have the remote end
1106 				 * retransmit the ACK for another
1107 				 * try.
1108 				 */
1109 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1110 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1111 					    "Socket allocation failed due to "
1112 					    "limits or memory shortage, %s\n",
1113 					    s, __func__,
1114 					    V_tcp_sc_rst_sock_fail ?
1115 					    "sending RST" : "try again");
1116 				if (V_tcp_sc_rst_sock_fail) {
1117 					rstreason = BANDLIM_UNLIMITED;
1118 					goto dropwithreset;
1119 				} else
1120 					goto dropunlock;
1121 			}
1122 			/*
1123 			 * Socket is created in state SYN_RECEIVED.
1124 			 * Unlock the listen socket, lock the newly
1125 			 * created socket and update the tp variable.
1126 			 */
1127 			INP_WUNLOCK(inp);	/* listen socket */
1128 			inp = sotoinpcb(so);
1129 			/*
1130 			 * New connection inpcb is already locked by
1131 			 * syncache_expand().
1132 			 */
1133 			INP_WLOCK_ASSERT(inp);
1134 			tp = intotcpcb(inp);
1135 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1136 			    ("%s: ", __func__));
1137 			/*
1138 			 * Process the segment and the data it
1139 			 * contains.  tcp_do_segment() consumes
1140 			 * the mbuf chain and unlocks the inpcb.
1141 			 */
1142 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1143 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1144 			    iptos);
1145 			return (IPPROTO_DONE);
1146 		}
1147 		/*
1148 		 * Segment flag validation for new connection attempts:
1149 		 *
1150 		 * Our (SYN|ACK) response was rejected.
1151 		 * Check with syncache and remove entry to prevent
1152 		 * retransmits.
1153 		 *
1154 		 * NB: syncache_chkrst does its own logging of failure
1155 		 * causes.
1156 		 */
1157 		if (thflags & TH_RST) {
1158 			syncache_chkrst(&inc, th, m);
1159 			goto dropunlock;
1160 		}
1161 		/*
1162 		 * We can't do anything without SYN.
1163 		 */
1164 		if ((thflags & TH_SYN) == 0) {
1165 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1166 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1167 				    "SYN is missing, segment ignored\n",
1168 				    s, __func__);
1169 			TCPSTAT_INC(tcps_badsyn);
1170 			goto dropunlock;
1171 		}
1172 		/*
1173 		 * (SYN|ACK) is bogus on a listen socket.
1174 		 */
1175 		if (thflags & TH_ACK) {
1176 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 				    "SYN|ACK invalid, segment rejected\n",
1179 				    s, __func__);
1180 			syncache_badack(&inc);	/* XXX: Not needed! */
1181 			TCPSTAT_INC(tcps_badsyn);
1182 			rstreason = BANDLIM_RST_OPENPORT;
1183 			goto dropwithreset;
1184 		}
1185 		/*
1186 		 * If the drop_synfin option is enabled, drop all
1187 		 * segments with both the SYN and FIN bits set.
1188 		 * This prevents e.g. nmap from identifying the
1189 		 * TCP/IP stack.
1190 		 * XXX: Poor reasoning.  nmap has other methods
1191 		 * and is constantly refining its stack detection
1192 		 * strategies.
1193 		 * XXX: This is a violation of the TCP specification
1194 		 * and was used by RFC1644.
1195 		 */
1196 		if ((thflags & TH_FIN) && V_drop_synfin) {
1197 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1198 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1199 				    "SYN|FIN segment ignored (based on "
1200 				    "sysctl setting)\n", s, __func__);
1201 			TCPSTAT_INC(tcps_badsyn);
1202 			goto dropunlock;
1203 		}
1204 		/*
1205 		 * Segment's flags are (SYN) or (SYN|FIN).
1206 		 *
1207 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1208 		 * as they do not affect the state of the TCP FSM.
1209 		 * The data pointed to by TH_URG and th_urp is ignored.
1210 		 */
1211 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1212 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1213 		KASSERT(thflags & (TH_SYN),
1214 		    ("%s: Listen socket: TH_SYN not set", __func__));
1215 #ifdef INET6
1216 		/*
1217 		 * If deprecated address is forbidden,
1218 		 * we do not accept SYN to deprecated interface
1219 		 * address to prevent any new inbound connection from
1220 		 * getting established.
1221 		 * When we do not accept SYN, we send a TCP RST,
1222 		 * with deprecated source address (instead of dropping
1223 		 * it).  We compromise it as it is much better for peer
1224 		 * to send a RST, and RST will be the final packet
1225 		 * for the exchange.
1226 		 *
1227 		 * If we do not forbid deprecated addresses, we accept
1228 		 * the SYN packet.  RFC2462 does not suggest dropping
1229 		 * SYN in this case.
1230 		 * If we decipher RFC2462 5.5.4, it says like this:
1231 		 * 1. use of deprecated addr with existing
1232 		 *    communication is okay - "SHOULD continue to be
1233 		 *    used"
1234 		 * 2. use of it with new communication:
1235 		 *   (2a) "SHOULD NOT be used if alternate address
1236 		 *        with sufficient scope is available"
1237 		 *   (2b) nothing mentioned otherwise.
1238 		 * Here we fall into (2b) case as we have no choice in
1239 		 * our source address selection - we must obey the peer.
1240 		 *
1241 		 * The wording in RFC2462 is confusing, and there are
1242 		 * multiple description text for deprecated address
1243 		 * handling - worse, they are not exactly the same.
1244 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1245 		 */
1246 		if (isipv6 && !V_ip6_use_deprecated) {
1247 			struct in6_ifaddr *ia6;
1248 
1249 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1250 			if (ia6 != NULL &&
1251 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1252 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1253 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1254 					"Connection attempt to deprecated "
1255 					"IPv6 address rejected\n",
1256 					s, __func__);
1257 				rstreason = BANDLIM_RST_OPENPORT;
1258 				goto dropwithreset;
1259 			}
1260 		}
1261 #endif /* INET6 */
1262 		/*
1263 		 * Basic sanity checks on incoming SYN requests:
1264 		 *   Don't respond if the destination is a link layer
1265 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1266 		 *   If it is from this socket it must be forged.
1267 		 *   Don't respond if the source or destination is a
1268 		 *	global or subnet broad- or multicast address.
1269 		 *   Note that it is quite possible to receive unicast
1270 		 *	link-layer packets with a broadcast IP address. Use
1271 		 *	in_broadcast() to find them.
1272 		 */
1273 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1274 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1275 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1276 				"Connection attempt from broad- or multicast "
1277 				"link layer address ignored\n", s, __func__);
1278 			goto dropunlock;
1279 		}
1280 #ifdef INET6
1281 		if (isipv6) {
1282 			if (th->th_dport == th->th_sport &&
1283 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1284 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1285 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1286 					"Connection attempt to/from self "
1287 					"ignored\n", s, __func__);
1288 				goto dropunlock;
1289 			}
1290 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1291 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1292 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1293 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1294 					"Connection attempt from/to multicast "
1295 					"address ignored\n", s, __func__);
1296 				goto dropunlock;
1297 			}
1298 		}
1299 #endif
1300 #if defined(INET) && defined(INET6)
1301 		else
1302 #endif
1303 #ifdef INET
1304 		{
1305 			if (th->th_dport == th->th_sport &&
1306 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1307 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1308 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1309 					"Connection attempt from/to self "
1310 					"ignored\n", s, __func__);
1311 				goto dropunlock;
1312 			}
1313 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1314 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1315 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1316 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1317 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1318 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1319 					"Connection attempt from/to broad- "
1320 					"or multicast address ignored\n",
1321 					s, __func__);
1322 				goto dropunlock;
1323 			}
1324 		}
1325 #endif
1326 		/*
1327 		 * SYN appears to be valid.  Create compressed TCP state
1328 		 * for syncache.
1329 		 */
1330 #ifdef TCPDEBUG
1331 		if (so->so_options & SO_DEBUG)
1332 			tcp_trace(TA_INPUT, ostate, tp,
1333 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1334 #endif
1335 		TCP_PROBE3(debug__input, tp, th, m);
1336 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1337 		if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1338 			goto tfo_socket_result;
1339 
1340 		/*
1341 		 * Entry added to syncache and mbuf consumed.
1342 		 * Only the listen socket is unlocked by syncache_add().
1343 		 */
1344 		INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1345 		return (IPPROTO_DONE);
1346 	} else if (tp->t_state == TCPS_LISTEN) {
1347 		/*
1348 		 * When a listen socket is torn down the SO_ACCEPTCONN
1349 		 * flag is removed first while connections are drained
1350 		 * from the accept queue in a unlock/lock cycle of the
1351 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1352 		 * attempt go through unhandled.
1353 		 */
1354 		goto dropunlock;
1355 	}
1356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1357 	if (tp->t_flags & TF_SIGNATURE) {
1358 		tcp_dooptions(&to, optp, optlen, thflags);
1359 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1360 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1361 			goto dropunlock;
1362 		}
1363 		if (!TCPMD5_ENABLED() ||
1364 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1365 			goto dropunlock;
1366 	}
1367 #endif
1368 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1369 
1370 	/*
1371 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1372 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1373 	 * the inpcb, and unlocks pcbinfo.
1374 	 */
1375 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1376 	return (IPPROTO_DONE);
1377 
1378 dropwithreset:
1379 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1380 
1381 	if (inp != NULL) {
1382 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1383 		INP_WUNLOCK(inp);
1384 	} else
1385 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1386 	m = NULL;	/* mbuf chain got consumed. */
1387 	goto drop;
1388 
1389 dropunlock:
1390 	if (m != NULL)
1391 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1392 
1393 	if (inp != NULL)
1394 		INP_WUNLOCK(inp);
1395 
1396 drop:
1397 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1398 	if (s != NULL)
1399 		free(s, M_TCPLOG);
1400 	if (m != NULL)
1401 		m_freem(m);
1402 	return (IPPROTO_DONE);
1403 }
1404 
1405 /*
1406  * Automatic sizing of receive socket buffer.  Often the send
1407  * buffer size is not optimally adjusted to the actual network
1408  * conditions at hand (delay bandwidth product).  Setting the
1409  * buffer size too small limits throughput on links with high
1410  * bandwidth and high delay (eg. trans-continental/oceanic links).
1411  *
1412  * On the receive side the socket buffer memory is only rarely
1413  * used to any significant extent.  This allows us to be much
1414  * more aggressive in scaling the receive socket buffer.  For
1415  * the case that the buffer space is actually used to a large
1416  * extent and we run out of kernel memory we can simply drop
1417  * the new segments; TCP on the sender will just retransmit it
1418  * later.  Setting the buffer size too big may only consume too
1419  * much kernel memory if the application doesn't read() from
1420  * the socket or packet loss or reordering makes use of the
1421  * reassembly queue.
1422  *
1423  * The criteria to step up the receive buffer one notch are:
1424  *  1. Application has not set receive buffer size with
1425  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1426  *  2. the number of bytes received during 1/2 of an sRTT
1427  *     is at least 3/8 of the current socket buffer size.
1428  *  3. receive buffer size has not hit maximal automatic size;
1429  *
1430  * If all of the criteria are met we increaset the socket buffer
1431  * by a 1/2 (bounded by the max). This allows us to keep ahead
1432  * of slow-start but also makes it so our peer never gets limited
1433  * by our rwnd which we then open up causing a burst.
1434  *
1435  * This algorithm does two steps per RTT at most and only if
1436  * we receive a bulk stream w/o packet losses or reorderings.
1437  * Shrinking the buffer during idle times is not necessary as
1438  * it doesn't consume any memory when idle.
1439  *
1440  * TODO: Only step up if the application is actually serving
1441  * the buffer to better manage the socket buffer resources.
1442  */
1443 int
1444 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1445     struct tcpcb *tp, int tlen)
1446 {
1447 	int newsize = 0;
1448 
1449 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1450 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1451 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1452 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1453 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1454 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1455 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1456 		}
1457 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1458 
1459 		/* Start over with next RTT. */
1460 		tp->rfbuf_ts = 0;
1461 		tp->rfbuf_cnt = 0;
1462 	} else {
1463 		tp->rfbuf_cnt += tlen;	/* add up */
1464 	}
1465 	return (newsize);
1466 }
1467 
1468 void
1469 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1470 {
1471 	/*
1472 	 * Since tp might be gone if the session entered
1473 	 * the TIME_WAIT state before coming here, we need
1474 	 * to check if the socket is still connected.
1475 	 */
1476 	if ((so->so_state & SS_ISCONNECTED) == 0)
1477 		return;
1478 	INP_LOCK_ASSERT(tp->t_inpcb);
1479 	if (tp->t_flags & TF_WAKESOR) {
1480 		tp->t_flags &= ~TF_WAKESOR;
1481 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1482 		sorwakeup(so);
1483 	}
1484 	if (tp->t_flags & TF_WAKESOW) {
1485 		tp->t_flags &= ~TF_WAKESOW;
1486 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1487 		sowwakeup(so);
1488 	}
1489 }
1490 
1491 void
1492 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1493     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1494 {
1495 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1496 	int rstreason, todrop, win, incforsyn = 0;
1497 	uint32_t tiwin;
1498 	uint16_t nsegs;
1499 	char *s;
1500 	struct in_conninfo *inc;
1501 	struct mbuf *mfree;
1502 	struct tcpopt to;
1503 	int tfo_syn;
1504 	u_int maxseg;
1505 
1506 #ifdef TCPDEBUG
1507 	/*
1508 	 * The size of tcp_saveipgen must be the size of the max ip header,
1509 	 * now IPv6.
1510 	 */
1511 	u_char tcp_saveipgen[IP6_HDR_LEN];
1512 	struct tcphdr tcp_savetcp;
1513 	short ostate = 0;
1514 #endif
1515 	thflags = th->th_flags;
1516 	inc = &tp->t_inpcb->inp_inc;
1517 	tp->sackhint.last_sack_ack = 0;
1518 	sack_changed = 0;
1519 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1520 
1521 	NET_EPOCH_ASSERT();
1522 	INP_WLOCK_ASSERT(tp->t_inpcb);
1523 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1524 	    __func__));
1525 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1526 	    __func__));
1527 
1528 #ifdef TCPPCAP
1529 	/* Save segment, if requested. */
1530 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1531 #endif
1532 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1533 	    tlen, NULL, true);
1534 
1535 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1536 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1537 			log(LOG_DEBUG, "%s; %s: "
1538 			    "SYN|FIN segment ignored (based on "
1539 			    "sysctl setting)\n", s, __func__);
1540 			free(s, M_TCPLOG);
1541 		}
1542 		goto drop;
1543 	}
1544 
1545 	/*
1546 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1547 	 * check SEQ.ACK first.
1548 	 */
1549 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1550 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1551 		rstreason = BANDLIM_UNLIMITED;
1552 		goto dropwithreset;
1553 	}
1554 
1555 	/*
1556 	 * Segment received on connection.
1557 	 * Reset idle time and keep-alive timer.
1558 	 * XXX: This should be done after segment
1559 	 * validation to ignore broken/spoofed segs.
1560 	 */
1561 	tp->t_rcvtime = ticks;
1562 
1563 	/*
1564 	 * Scale up the window into a 32-bit value.
1565 	 * For the SYN_SENT state the scale is zero.
1566 	 */
1567 	tiwin = th->th_win << tp->snd_scale;
1568 #ifdef STATS
1569 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1570 #endif
1571 
1572 	/*
1573 	 * TCP ECN processing.
1574 	 */
1575 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1576 		if (thflags & TH_CWR) {
1577 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1578 			tp->t_flags |= TF_ACKNOW;
1579 		}
1580 		switch (iptos & IPTOS_ECN_MASK) {
1581 		case IPTOS_ECN_CE:
1582 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1583 			TCPSTAT_INC(tcps_ecn_ce);
1584 			break;
1585 		case IPTOS_ECN_ECT0:
1586 			TCPSTAT_INC(tcps_ecn_ect0);
1587 			break;
1588 		case IPTOS_ECN_ECT1:
1589 			TCPSTAT_INC(tcps_ecn_ect1);
1590 			break;
1591 		}
1592 
1593 		/* Process a packet differently from RFC3168. */
1594 		cc_ecnpkt_handler(tp, th, iptos);
1595 
1596 		/* Congestion experienced. */
1597 		if (thflags & TH_ECE) {
1598 			cc_cong_signal(tp, th, CC_ECN);
1599 		}
1600 	}
1601 
1602 	/*
1603 	 * Parse options on any incoming segment.
1604 	 */
1605 	tcp_dooptions(&to, (u_char *)(th + 1),
1606 	    (th->th_off << 2) - sizeof(struct tcphdr),
1607 	    (thflags & TH_SYN) ? TO_SYN : 0);
1608 
1609 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1610 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1611 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1612 		TCPSTAT_INC(tcps_sig_err_sigopt);
1613 		/* XXX: should drop? */
1614 	}
1615 #endif
1616 	/*
1617 	 * If echoed timestamp is later than the current time,
1618 	 * fall back to non RFC1323 RTT calculation.  Normalize
1619 	 * timestamp if syncookies were used when this connection
1620 	 * was established.
1621 	 */
1622 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1623 		to.to_tsecr -= tp->ts_offset;
1624 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1625 			to.to_tsecr = 0;
1626 		else if (tp->t_flags & TF_PREVVALID &&
1627 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1628 			cc_cong_signal(tp, th, CC_RTO_ERR);
1629 	}
1630 	/*
1631 	 * Process options only when we get SYN/ACK back. The SYN case
1632 	 * for incoming connections is handled in tcp_syncache.
1633 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1634 	 * or <SYN,ACK>) segment itself is never scaled.
1635 	 * XXX this is traditional behavior, may need to be cleaned up.
1636 	 */
1637 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1638 		/* Handle parallel SYN for ECN */
1639 		if (!(thflags & TH_ACK) &&
1640 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1641 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1642 			tp->t_flags2 |= TF2_ECN_PERMIT;
1643 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1644 			TCPSTAT_INC(tcps_ecn_shs);
1645 		}
1646 		if ((to.to_flags & TOF_SCALE) &&
1647 		    (tp->t_flags & TF_REQ_SCALE)) {
1648 			tp->t_flags |= TF_RCVD_SCALE;
1649 			tp->snd_scale = to.to_wscale;
1650 		} else
1651 			tp->t_flags &= ~TF_REQ_SCALE;
1652 		/*
1653 		 * Initial send window.  It will be updated with
1654 		 * the next incoming segment to the scaled value.
1655 		 */
1656 		tp->snd_wnd = th->th_win;
1657 		if ((to.to_flags & TOF_TS) &&
1658 		    (tp->t_flags & TF_REQ_TSTMP)) {
1659 			tp->t_flags |= TF_RCVD_TSTMP;
1660 			tp->ts_recent = to.to_tsval;
1661 			tp->ts_recent_age = tcp_ts_getticks();
1662 		} else
1663 			tp->t_flags &= ~TF_REQ_TSTMP;
1664 		if (to.to_flags & TOF_MSS)
1665 			tcp_mss(tp, to.to_mss);
1666 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1667 		    (to.to_flags & TOF_SACKPERM) == 0)
1668 			tp->t_flags &= ~TF_SACK_PERMIT;
1669 		if (IS_FASTOPEN(tp->t_flags)) {
1670 			if (to.to_flags & TOF_FASTOPEN) {
1671 				uint16_t mss;
1672 
1673 				if (to.to_flags & TOF_MSS)
1674 					mss = to.to_mss;
1675 				else
1676 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1677 						mss = TCP6_MSS;
1678 					else
1679 						mss = TCP_MSS;
1680 				tcp_fastopen_update_cache(tp, mss,
1681 				    to.to_tfo_len, to.to_tfo_cookie);
1682 			} else
1683 				tcp_fastopen_disable_path(tp);
1684 		}
1685 	}
1686 
1687 	/*
1688 	 * If timestamps were negotiated during SYN/ACK and a
1689 	 * segment without a timestamp is received, silently drop
1690 	 * the segment, unless it is a RST segment or missing timestamps are
1691 	 * tolerated.
1692 	 * See section 3.2 of RFC 7323.
1693 	 */
1694 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1695 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1696 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1697 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1698 				    "segment processed normally\n",
1699 				    s, __func__);
1700 				free(s, M_TCPLOG);
1701 			}
1702 		} else {
1703 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1704 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1705 				    "segment silently dropped\n", s, __func__);
1706 				free(s, M_TCPLOG);
1707 			}
1708 			goto drop;
1709 		}
1710 	}
1711 	/*
1712 	 * If timestamps were not negotiated during SYN/ACK and a
1713 	 * segment with a timestamp is received, ignore the
1714 	 * timestamp and process the packet normally.
1715 	 * See section 3.2 of RFC 7323.
1716 	 */
1717 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1718 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1719 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1720 			    "segment processed normally\n", s, __func__);
1721 			free(s, M_TCPLOG);
1722 		}
1723 	}
1724 
1725 	/*
1726 	 * Header prediction: check for the two common cases
1727 	 * of a uni-directional data xfer.  If the packet has
1728 	 * no control flags, is in-sequence, the window didn't
1729 	 * change and we're not retransmitting, it's a
1730 	 * candidate.  If the length is zero and the ack moved
1731 	 * forward, we're the sender side of the xfer.  Just
1732 	 * free the data acked & wake any higher level process
1733 	 * that was blocked waiting for space.  If the length
1734 	 * is non-zero and the ack didn't move, we're the
1735 	 * receiver side.  If we're getting packets in-order
1736 	 * (the reassembly queue is empty), add the data to
1737 	 * the socket buffer and note that we need a delayed ack.
1738 	 * Make sure that the hidden state-flags are also off.
1739 	 * Since we check for TCPS_ESTABLISHED first, it can only
1740 	 * be TH_NEEDSYN.
1741 	 */
1742 	if (tp->t_state == TCPS_ESTABLISHED &&
1743 	    th->th_seq == tp->rcv_nxt &&
1744 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1745 	    tp->snd_nxt == tp->snd_max &&
1746 	    tiwin && tiwin == tp->snd_wnd &&
1747 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1748 	    SEGQ_EMPTY(tp) &&
1749 	    ((to.to_flags & TOF_TS) == 0 ||
1750 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1751 		/*
1752 		 * If last ACK falls within this segment's sequence numbers,
1753 		 * record the timestamp.
1754 		 * NOTE that the test is modified according to the latest
1755 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1756 		 */
1757 		if ((to.to_flags & TOF_TS) != 0 &&
1758 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1759 			tp->ts_recent_age = tcp_ts_getticks();
1760 			tp->ts_recent = to.to_tsval;
1761 		}
1762 
1763 		if (tlen == 0) {
1764 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1765 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1766 			    !IN_RECOVERY(tp->t_flags) &&
1767 			    (to.to_flags & TOF_SACK) == 0 &&
1768 			    TAILQ_EMPTY(&tp->snd_holes)) {
1769 				/*
1770 				 * This is a pure ack for outstanding data.
1771 				 */
1772 				TCPSTAT_INC(tcps_predack);
1773 
1774 				/*
1775 				 * "bad retransmit" recovery without timestamps.
1776 				 */
1777 				if ((to.to_flags & TOF_TS) == 0 &&
1778 				    tp->t_rxtshift == 1 &&
1779 				    tp->t_flags & TF_PREVVALID &&
1780 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1781 					cc_cong_signal(tp, th, CC_RTO_ERR);
1782 				}
1783 
1784 				/*
1785 				 * Recalculate the transmit timer / rtt.
1786 				 *
1787 				 * Some boxes send broken timestamp replies
1788 				 * during the SYN+ACK phase, ignore
1789 				 * timestamps of 0 or we could calculate a
1790 				 * huge RTT and blow up the retransmit timer.
1791 				 */
1792 				if ((to.to_flags & TOF_TS) != 0 &&
1793 				    to.to_tsecr) {
1794 					uint32_t t;
1795 
1796 					t = tcp_ts_getticks() - to.to_tsecr;
1797 					if (!tp->t_rttlow || tp->t_rttlow > t)
1798 						tp->t_rttlow = t;
1799 					tcp_xmit_timer(tp,
1800 					    TCP_TS_TO_TICKS(t) + 1);
1801 				} else if (tp->t_rtttime &&
1802 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1803 					if (!tp->t_rttlow ||
1804 					    tp->t_rttlow > ticks - tp->t_rtttime)
1805 						tp->t_rttlow = ticks - tp->t_rtttime;
1806 					tcp_xmit_timer(tp,
1807 							ticks - tp->t_rtttime);
1808 				}
1809 				acked = BYTES_THIS_ACK(tp, th);
1810 
1811 #ifdef TCP_HHOOK
1812 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1813 				hhook_run_tcp_est_in(tp, th, &to);
1814 #endif
1815 
1816 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1817 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1818 				sbdrop(&so->so_snd, acked);
1819 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1820 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1821 					tp->snd_recover = th->th_ack - 1;
1822 
1823 				/*
1824 				 * Let the congestion control algorithm update
1825 				 * congestion control related information. This
1826 				 * typically means increasing the congestion
1827 				 * window.
1828 				 */
1829 				cc_ack_received(tp, th, nsegs, CC_ACK);
1830 
1831 				tp->snd_una = th->th_ack;
1832 				/*
1833 				 * Pull snd_wl2 up to prevent seq wrap relative
1834 				 * to th_ack.
1835 				 */
1836 				tp->snd_wl2 = th->th_ack;
1837 				tp->t_dupacks = 0;
1838 				m_freem(m);
1839 
1840 				/*
1841 				 * If all outstanding data are acked, stop
1842 				 * retransmit timer, otherwise restart timer
1843 				 * using current (possibly backed-off) value.
1844 				 * If process is waiting for space,
1845 				 * wakeup/selwakeup/signal.  If data
1846 				 * are ready to send, let tcp_output
1847 				 * decide between more output or persist.
1848 				 */
1849 #ifdef TCPDEBUG
1850 				if (so->so_options & SO_DEBUG)
1851 					tcp_trace(TA_INPUT, ostate, tp,
1852 					    (void *)tcp_saveipgen,
1853 					    &tcp_savetcp, 0);
1854 #endif
1855 				TCP_PROBE3(debug__input, tp, th, m);
1856 				if (tp->snd_una == tp->snd_max)
1857 					tcp_timer_activate(tp, TT_REXMT, 0);
1858 				else if (!tcp_timer_active(tp, TT_PERSIST))
1859 					tcp_timer_activate(tp, TT_REXMT,
1860 						      tp->t_rxtcur);
1861 				tp->t_flags |= TF_WAKESOW;
1862 				if (sbavail(&so->so_snd))
1863 					(void) tp->t_fb->tfb_tcp_output(tp);
1864 				goto check_delack;
1865 			}
1866 		} else if (th->th_ack == tp->snd_una &&
1867 		    tlen <= sbspace(&so->so_rcv)) {
1868 			int newsize = 0;	/* automatic sockbuf scaling */
1869 
1870 			/*
1871 			 * This is a pure, in-sequence data packet with
1872 			 * nothing on the reassembly queue and we have enough
1873 			 * buffer space to take it.
1874 			 */
1875 			/* Clean receiver SACK report if present */
1876 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1877 				tcp_clean_sackreport(tp);
1878 			TCPSTAT_INC(tcps_preddat);
1879 			tp->rcv_nxt += tlen;
1880 			if (tlen &&
1881 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1882 			    (tp->t_fbyte_in == 0)) {
1883 				tp->t_fbyte_in = ticks;
1884 				if (tp->t_fbyte_in == 0)
1885 					tp->t_fbyte_in = 1;
1886 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1887 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1888 			}
1889 			/*
1890 			 * Pull snd_wl1 up to prevent seq wrap relative to
1891 			 * th_seq.
1892 			 */
1893 			tp->snd_wl1 = th->th_seq;
1894 			/*
1895 			 * Pull rcv_up up to prevent seq wrap relative to
1896 			 * rcv_nxt.
1897 			 */
1898 			tp->rcv_up = tp->rcv_nxt;
1899 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1900 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1901 #ifdef TCPDEBUG
1902 			if (so->so_options & SO_DEBUG)
1903 				tcp_trace(TA_INPUT, ostate, tp,
1904 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1905 #endif
1906 			TCP_PROBE3(debug__input, tp, th, m);
1907 
1908 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1909 
1910 			/* Add data to socket buffer. */
1911 			SOCKBUF_LOCK(&so->so_rcv);
1912 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1913 				m_freem(m);
1914 			} else {
1915 				/*
1916 				 * Set new socket buffer size.
1917 				 * Give up when limit is reached.
1918 				 */
1919 				if (newsize)
1920 					if (!sbreserve_locked(&so->so_rcv,
1921 					    newsize, so, NULL))
1922 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1923 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1924 				sbappendstream_locked(&so->so_rcv, m, 0);
1925 			}
1926 			SOCKBUF_UNLOCK(&so->so_rcv);
1927 			tp->t_flags |= TF_WAKESOR;
1928 			if (DELAY_ACK(tp, tlen)) {
1929 				tp->t_flags |= TF_DELACK;
1930 			} else {
1931 				tp->t_flags |= TF_ACKNOW;
1932 				tp->t_fb->tfb_tcp_output(tp);
1933 			}
1934 			goto check_delack;
1935 		}
1936 	}
1937 
1938 	/*
1939 	 * Calculate amount of space in receive window,
1940 	 * and then do TCP input processing.
1941 	 * Receive window is amount of space in rcv queue,
1942 	 * but not less than advertised window.
1943 	 */
1944 	win = sbspace(&so->so_rcv);
1945 	if (win < 0)
1946 		win = 0;
1947 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1948 
1949 	switch (tp->t_state) {
1950 	/*
1951 	 * If the state is SYN_RECEIVED:
1952 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1953 	 */
1954 	case TCPS_SYN_RECEIVED:
1955 		if ((thflags & TH_ACK) &&
1956 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1957 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1958 				rstreason = BANDLIM_RST_OPENPORT;
1959 				goto dropwithreset;
1960 		}
1961 		if (IS_FASTOPEN(tp->t_flags)) {
1962 			/*
1963 			 * When a TFO connection is in SYN_RECEIVED, the
1964 			 * only valid packets are the initial SYN, a
1965 			 * retransmit/copy of the initial SYN (possibly with
1966 			 * a subset of the original data), a valid ACK, a
1967 			 * FIN, or a RST.
1968 			 */
1969 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1970 				rstreason = BANDLIM_RST_OPENPORT;
1971 				goto dropwithreset;
1972 			} else if (thflags & TH_SYN) {
1973 				/* non-initial SYN is ignored */
1974 				if ((tcp_timer_active(tp, TT_DELACK) ||
1975 				     tcp_timer_active(tp, TT_REXMT)))
1976 					goto drop;
1977 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1978 				goto drop;
1979 			}
1980 		}
1981 		break;
1982 
1983 	/*
1984 	 * If the state is SYN_SENT:
1985 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
1986 	 *	    been verified), then drop the connection.
1987 	 *	if seg contains a RST without an ACK, drop the seg.
1988 	 *	if seg does not contain SYN, then drop the seg.
1989 	 * Otherwise this is an acceptable SYN segment
1990 	 *	initialize tp->rcv_nxt and tp->irs
1991 	 *	if seg contains ack then advance tp->snd_una
1992 	 *	if seg contains an ECE and ECN support is enabled, the stream
1993 	 *	    is ECN capable.
1994 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1995 	 *	arrange for segment to be acked (eventually)
1996 	 *	continue processing rest of data/controls, beginning with URG
1997 	 */
1998 	case TCPS_SYN_SENT:
1999 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2000 			TCP_PROBE5(connect__refused, NULL, tp,
2001 			    m, tp, th);
2002 			tp = tcp_drop(tp, ECONNREFUSED);
2003 		}
2004 		if (thflags & TH_RST)
2005 			goto drop;
2006 		if (!(thflags & TH_SYN))
2007 			goto drop;
2008 
2009 		tp->irs = th->th_seq;
2010 		tcp_rcvseqinit(tp);
2011 		if (thflags & TH_ACK) {
2012 			int tfo_partial_ack = 0;
2013 
2014 			TCPSTAT_INC(tcps_connects);
2015 			soisconnected(so);
2016 #ifdef MAC
2017 			mac_socketpeer_set_from_mbuf(m, so);
2018 #endif
2019 			/* Do window scaling on this connection? */
2020 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2021 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2022 				tp->rcv_scale = tp->request_r_scale;
2023 			}
2024 			tp->rcv_adv += min(tp->rcv_wnd,
2025 			    TCP_MAXWIN << tp->rcv_scale);
2026 			tp->snd_una++;		/* SYN is acked */
2027 			/*
2028 			 * If not all the data that was sent in the TFO SYN
2029 			 * has been acked, resend the remainder right away.
2030 			 */
2031 			if (IS_FASTOPEN(tp->t_flags) &&
2032 			    (tp->snd_una != tp->snd_max)) {
2033 				tp->snd_nxt = th->th_ack;
2034 				tfo_partial_ack = 1;
2035 			}
2036 			/*
2037 			 * If there's data, delay ACK; if there's also a FIN
2038 			 * ACKNOW will be turned on later.
2039 			 */
2040 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2041 				tcp_timer_activate(tp, TT_DELACK,
2042 				    tcp_delacktime);
2043 			else
2044 				tp->t_flags |= TF_ACKNOW;
2045 
2046 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2047 			    (V_tcp_do_ecn == 1)) {
2048 				tp->t_flags2 |= TF2_ECN_PERMIT;
2049 				TCPSTAT_INC(tcps_ecn_shs);
2050 			}
2051 
2052 			/*
2053 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2054 			 * Transitions:
2055 			 *	SYN_SENT  --> ESTABLISHED
2056 			 *	SYN_SENT* --> FIN_WAIT_1
2057 			 */
2058 			tp->t_starttime = ticks;
2059 			if (tp->t_flags & TF_NEEDFIN) {
2060 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2061 				tp->t_flags &= ~TF_NEEDFIN;
2062 				thflags &= ~TH_SYN;
2063 			} else {
2064 				tcp_state_change(tp, TCPS_ESTABLISHED);
2065 				TCP_PROBE5(connect__established, NULL, tp,
2066 				    m, tp, th);
2067 				cc_conn_init(tp);
2068 				tcp_timer_activate(tp, TT_KEEP,
2069 				    TP_KEEPIDLE(tp));
2070 			}
2071 		} else {
2072 			/*
2073 			 * Received initial SYN in SYN-SENT[*] state =>
2074 			 * simultaneous open.
2075 			 * If it succeeds, connection is * half-synchronized.
2076 			 * Otherwise, do 3-way handshake:
2077 			 *        SYN-SENT -> SYN-RECEIVED
2078 			 *        SYN-SENT* -> SYN-RECEIVED*
2079 			 */
2080 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2081 			tcp_timer_activate(tp, TT_REXMT, 0);
2082 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2083 		}
2084 
2085 		INP_WLOCK_ASSERT(tp->t_inpcb);
2086 
2087 		/*
2088 		 * Advance th->th_seq to correspond to first data byte.
2089 		 * If data, trim to stay within window,
2090 		 * dropping FIN if necessary.
2091 		 */
2092 		th->th_seq++;
2093 		if (tlen > tp->rcv_wnd) {
2094 			todrop = tlen - tp->rcv_wnd;
2095 			m_adj(m, -todrop);
2096 			tlen = tp->rcv_wnd;
2097 			thflags &= ~TH_FIN;
2098 			TCPSTAT_INC(tcps_rcvpackafterwin);
2099 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2100 		}
2101 		tp->snd_wl1 = th->th_seq - 1;
2102 		tp->rcv_up = th->th_seq;
2103 		/*
2104 		 * Client side of transaction: already sent SYN and data.
2105 		 * If the remote host used T/TCP to validate the SYN,
2106 		 * our data will be ACK'd; if so, enter normal data segment
2107 		 * processing in the middle of step 5, ack processing.
2108 		 * Otherwise, goto step 6.
2109 		 */
2110 		if (thflags & TH_ACK)
2111 			goto process_ACK;
2112 
2113 		goto step6;
2114 
2115 	/*
2116 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2117 	 *      do normal processing.
2118 	 *
2119 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2120 	 */
2121 	case TCPS_LAST_ACK:
2122 	case TCPS_CLOSING:
2123 		break;  /* continue normal processing */
2124 	}
2125 
2126 	/*
2127 	 * States other than LISTEN or SYN_SENT.
2128 	 * First check the RST flag and sequence number since reset segments
2129 	 * are exempt from the timestamp and connection count tests.  This
2130 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2131 	 * below which allowed reset segments in half the sequence space
2132 	 * to fall though and be processed (which gives forged reset
2133 	 * segments with a random sequence number a 50 percent chance of
2134 	 * killing a connection).
2135 	 * Then check timestamp, if present.
2136 	 * Then check the connection count, if present.
2137 	 * Then check that at least some bytes of segment are within
2138 	 * receive window.  If segment begins before rcv_nxt,
2139 	 * drop leading data (and SYN); if nothing left, just ack.
2140 	 */
2141 	if (thflags & TH_RST) {
2142 		/*
2143 		 * RFC5961 Section 3.2
2144 		 *
2145 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2146 		 * - If RST is in window, we send challenge ACK.
2147 		 *
2148 		 * Note: to take into account delayed ACKs, we should
2149 		 *   test against last_ack_sent instead of rcv_nxt.
2150 		 * Note 2: we handle special case of closed window, not
2151 		 *   covered by the RFC.
2152 		 */
2153 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2154 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2155 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2156 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2157 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2158 			    __func__, th, tp));
2159 
2160 			if (V_tcp_insecure_rst ||
2161 			    tp->last_ack_sent == th->th_seq) {
2162 				TCPSTAT_INC(tcps_drops);
2163 				/* Drop the connection. */
2164 				switch (tp->t_state) {
2165 				case TCPS_SYN_RECEIVED:
2166 					so->so_error = ECONNREFUSED;
2167 					goto close;
2168 				case TCPS_ESTABLISHED:
2169 				case TCPS_FIN_WAIT_1:
2170 				case TCPS_FIN_WAIT_2:
2171 				case TCPS_CLOSE_WAIT:
2172 				case TCPS_CLOSING:
2173 				case TCPS_LAST_ACK:
2174 					so->so_error = ECONNRESET;
2175 				close:
2176 					/* FALLTHROUGH */
2177 				default:
2178 					tp = tcp_close(tp);
2179 				}
2180 			} else {
2181 				TCPSTAT_INC(tcps_badrst);
2182 				/* Send challenge ACK. */
2183 				tcp_respond(tp, mtod(m, void *), th, m,
2184 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2185 				tp->last_ack_sent = tp->rcv_nxt;
2186 				m = NULL;
2187 			}
2188 		}
2189 		goto drop;
2190 	}
2191 
2192 	/*
2193 	 * RFC5961 Section 4.2
2194 	 * Send challenge ACK for any SYN in synchronized state.
2195 	 */
2196 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2197 	    tp->t_state != TCPS_SYN_RECEIVED) {
2198 		TCPSTAT_INC(tcps_badsyn);
2199 		if (V_tcp_insecure_syn &&
2200 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2201 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2202 			tp = tcp_drop(tp, ECONNRESET);
2203 			rstreason = BANDLIM_UNLIMITED;
2204 		} else {
2205 			/* Send challenge ACK. */
2206 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2207 			    tp->snd_nxt, TH_ACK);
2208 			tp->last_ack_sent = tp->rcv_nxt;
2209 			m = NULL;
2210 		}
2211 		goto drop;
2212 	}
2213 
2214 	/*
2215 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2216 	 * and it's less than ts_recent, drop it.
2217 	 */
2218 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2219 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2220 		/* Check to see if ts_recent is over 24 days old.  */
2221 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2222 			/*
2223 			 * Invalidate ts_recent.  If this segment updates
2224 			 * ts_recent, the age will be reset later and ts_recent
2225 			 * will get a valid value.  If it does not, setting
2226 			 * ts_recent to zero will at least satisfy the
2227 			 * requirement that zero be placed in the timestamp
2228 			 * echo reply when ts_recent isn't valid.  The
2229 			 * age isn't reset until we get a valid ts_recent
2230 			 * because we don't want out-of-order segments to be
2231 			 * dropped when ts_recent is old.
2232 			 */
2233 			tp->ts_recent = 0;
2234 		} else {
2235 			TCPSTAT_INC(tcps_rcvduppack);
2236 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2237 			TCPSTAT_INC(tcps_pawsdrop);
2238 			if (tlen)
2239 				goto dropafterack;
2240 			goto drop;
2241 		}
2242 	}
2243 
2244 	/*
2245 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2246 	 * this connection before trimming the data to fit the receive
2247 	 * window.  Check the sequence number versus IRS since we know
2248 	 * the sequence numbers haven't wrapped.  This is a partial fix
2249 	 * for the "LAND" DoS attack.
2250 	 */
2251 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2252 		rstreason = BANDLIM_RST_OPENPORT;
2253 		goto dropwithreset;
2254 	}
2255 
2256 	todrop = tp->rcv_nxt - th->th_seq;
2257 	if (todrop > 0) {
2258 		if (thflags & TH_SYN) {
2259 			thflags &= ~TH_SYN;
2260 			th->th_seq++;
2261 			if (th->th_urp > 1)
2262 				th->th_urp--;
2263 			else
2264 				thflags &= ~TH_URG;
2265 			todrop--;
2266 		}
2267 		/*
2268 		 * Following if statement from Stevens, vol. 2, p. 960.
2269 		 */
2270 		if (todrop > tlen
2271 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2272 			/*
2273 			 * Any valid FIN must be to the left of the window.
2274 			 * At this point the FIN must be a duplicate or out
2275 			 * of sequence; drop it.
2276 			 */
2277 			thflags &= ~TH_FIN;
2278 
2279 			/*
2280 			 * Send an ACK to resynchronize and drop any data.
2281 			 * But keep on processing for RST or ACK.
2282 			 */
2283 			tp->t_flags |= TF_ACKNOW;
2284 			todrop = tlen;
2285 			TCPSTAT_INC(tcps_rcvduppack);
2286 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2287 		} else {
2288 			TCPSTAT_INC(tcps_rcvpartduppack);
2289 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2290 		}
2291 		/*
2292 		 * DSACK - add SACK block for dropped range
2293 		 */
2294 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2295 			tcp_update_sack_list(tp, th->th_seq,
2296 			    th->th_seq + todrop);
2297 			/*
2298 			 * ACK now, as the next in-sequence segment
2299 			 * will clear the DSACK block again
2300 			 */
2301 			tp->t_flags |= TF_ACKNOW;
2302 		}
2303 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2304 		th->th_seq += todrop;
2305 		tlen -= todrop;
2306 		if (th->th_urp > todrop)
2307 			th->th_urp -= todrop;
2308 		else {
2309 			thflags &= ~TH_URG;
2310 			th->th_urp = 0;
2311 		}
2312 	}
2313 
2314 	/*
2315 	 * If new data are received on a connection after the
2316 	 * user processes are gone, then RST the other end.
2317 	 */
2318 	if ((so->so_state & SS_NOFDREF) &&
2319 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2320 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2321 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2322 			    "after socket was closed, "
2323 			    "sending RST and removing tcpcb\n",
2324 			    s, __func__, tcpstates[tp->t_state], tlen);
2325 			free(s, M_TCPLOG);
2326 		}
2327 		tp = tcp_close(tp);
2328 		TCPSTAT_INC(tcps_rcvafterclose);
2329 		rstreason = BANDLIM_UNLIMITED;
2330 		goto dropwithreset;
2331 	}
2332 
2333 	/*
2334 	 * If segment ends after window, drop trailing data
2335 	 * (and PUSH and FIN); if nothing left, just ACK.
2336 	 */
2337 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2338 	if (todrop > 0) {
2339 		TCPSTAT_INC(tcps_rcvpackafterwin);
2340 		if (todrop >= tlen) {
2341 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2342 			/*
2343 			 * If window is closed can only take segments at
2344 			 * window edge, and have to drop data and PUSH from
2345 			 * incoming segments.  Continue processing, but
2346 			 * remember to ack.  Otherwise, drop segment
2347 			 * and ack.
2348 			 */
2349 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2350 				tp->t_flags |= TF_ACKNOW;
2351 				TCPSTAT_INC(tcps_rcvwinprobe);
2352 			} else
2353 				goto dropafterack;
2354 		} else
2355 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2356 		m_adj(m, -todrop);
2357 		tlen -= todrop;
2358 		thflags &= ~(TH_PUSH|TH_FIN);
2359 	}
2360 
2361 	/*
2362 	 * If last ACK falls within this segment's sequence numbers,
2363 	 * record its timestamp.
2364 	 * NOTE:
2365 	 * 1) That the test incorporates suggestions from the latest
2366 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2367 	 * 2) That updating only on newer timestamps interferes with
2368 	 *    our earlier PAWS tests, so this check should be solely
2369 	 *    predicated on the sequence space of this segment.
2370 	 * 3) That we modify the segment boundary check to be
2371 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2372 	 *    instead of RFC1323's
2373 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2374 	 *    This modified check allows us to overcome RFC1323's
2375 	 *    limitations as described in Stevens TCP/IP Illustrated
2376 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2377 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2378 	 */
2379 	if ((to.to_flags & TOF_TS) != 0 &&
2380 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2381 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2382 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2383 		tp->ts_recent_age = tcp_ts_getticks();
2384 		tp->ts_recent = to.to_tsval;
2385 	}
2386 
2387 	/*
2388 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2389 	 * flag is on (half-synchronized state), then queue data for
2390 	 * later processing; else drop segment and return.
2391 	 */
2392 	if ((thflags & TH_ACK) == 0) {
2393 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2394 		    (tp->t_flags & TF_NEEDSYN)) {
2395 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2396 			    IS_FASTOPEN(tp->t_flags)) {
2397 				tp->snd_wnd = tiwin;
2398 				cc_conn_init(tp);
2399 			}
2400 			goto step6;
2401 		} else if (tp->t_flags & TF_ACKNOW)
2402 			goto dropafterack;
2403 		else
2404 			goto drop;
2405 	}
2406 
2407 	/*
2408 	 * Ack processing.
2409 	 */
2410 	switch (tp->t_state) {
2411 	/*
2412 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2413 	 * ESTABLISHED state and continue processing.
2414 	 * The ACK was checked above.
2415 	 */
2416 	case TCPS_SYN_RECEIVED:
2417 
2418 		TCPSTAT_INC(tcps_connects);
2419 		soisconnected(so);
2420 		/* Do window scaling? */
2421 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2422 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2423 			tp->rcv_scale = tp->request_r_scale;
2424 		}
2425 		tp->snd_wnd = tiwin;
2426 		/*
2427 		 * Make transitions:
2428 		 *      SYN-RECEIVED  -> ESTABLISHED
2429 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2430 		 */
2431 		tp->t_starttime = ticks;
2432 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2433 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2434 			tp->t_tfo_pending = NULL;
2435 		}
2436 		if (tp->t_flags & TF_NEEDFIN) {
2437 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2438 			tp->t_flags &= ~TF_NEEDFIN;
2439 		} else {
2440 			tcp_state_change(tp, TCPS_ESTABLISHED);
2441 			TCP_PROBE5(accept__established, NULL, tp,
2442 			    m, tp, th);
2443 			/*
2444 			 * TFO connections call cc_conn_init() during SYN
2445 			 * processing.  Calling it again here for such
2446 			 * connections is not harmless as it would undo the
2447 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2448 			 * is retransmitted.
2449 			 */
2450 			if (!IS_FASTOPEN(tp->t_flags))
2451 				cc_conn_init(tp);
2452 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2453 		}
2454 		/*
2455 		 * Account for the ACK of our SYN prior to
2456 		 * regular ACK processing below, except for
2457 		 * simultaneous SYN, which is handled later.
2458 		 */
2459 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2460 			incforsyn = 1;
2461 		/*
2462 		 * If segment contains data or ACK, will call tcp_reass()
2463 		 * later; if not, do so now to pass queued data to user.
2464 		 */
2465 		if (tlen == 0 && (thflags & TH_FIN) == 0)
2466 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2467 			    (struct mbuf *)0);
2468 		tp->snd_wl1 = th->th_seq - 1;
2469 		/* FALLTHROUGH */
2470 
2471 	/*
2472 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2473 	 * ACKs.  If the ack is in the range
2474 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2475 	 * then advance tp->snd_una to th->th_ack and drop
2476 	 * data from the retransmission queue.  If this ACK reflects
2477 	 * more up to date window information we update our window information.
2478 	 */
2479 	case TCPS_ESTABLISHED:
2480 	case TCPS_FIN_WAIT_1:
2481 	case TCPS_FIN_WAIT_2:
2482 	case TCPS_CLOSE_WAIT:
2483 	case TCPS_CLOSING:
2484 	case TCPS_LAST_ACK:
2485 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2486 			TCPSTAT_INC(tcps_rcvacktoomuch);
2487 			goto dropafterack;
2488 		}
2489 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2490 		    ((to.to_flags & TOF_SACK) ||
2491 		     !TAILQ_EMPTY(&tp->snd_holes)))
2492 			sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2493 		else
2494 			/*
2495 			 * Reset the value so that previous (valid) value
2496 			 * from the last ack with SACK doesn't get used.
2497 			 */
2498 			tp->sackhint.sacked_bytes = 0;
2499 
2500 #ifdef TCP_HHOOK
2501 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2502 		hhook_run_tcp_est_in(tp, th, &to);
2503 #endif
2504 
2505 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2506 			maxseg = tcp_maxseg(tp);
2507 			if (tlen == 0 &&
2508 			    (tiwin == tp->snd_wnd ||
2509 			    (tp->t_flags & TF_SACK_PERMIT))) {
2510 				/*
2511 				 * If this is the first time we've seen a
2512 				 * FIN from the remote, this is not a
2513 				 * duplicate and it needs to be processed
2514 				 * normally.  This happens during a
2515 				 * simultaneous close.
2516 				 */
2517 				if ((thflags & TH_FIN) &&
2518 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2519 					tp->t_dupacks = 0;
2520 					break;
2521 				}
2522 				TCPSTAT_INC(tcps_rcvdupack);
2523 				/*
2524 				 * If we have outstanding data (other than
2525 				 * a window probe), this is a completely
2526 				 * duplicate ack (ie, window info didn't
2527 				 * change and FIN isn't set),
2528 				 * the ack is the biggest we've
2529 				 * seen and we've seen exactly our rexmt
2530 				 * threshold of them, assume a packet
2531 				 * has been dropped and retransmit it.
2532 				 * Kludge snd_nxt & the congestion
2533 				 * window so we send only this one
2534 				 * packet.
2535 				 *
2536 				 * We know we're losing at the current
2537 				 * window size so do congestion avoidance
2538 				 * (set ssthresh to half the current window
2539 				 * and pull our congestion window back to
2540 				 * the new ssthresh).
2541 				 *
2542 				 * Dup acks mean that packets have left the
2543 				 * network (they're now cached at the receiver)
2544 				 * so bump cwnd by the amount in the receiver
2545 				 * to keep a constant cwnd packets in the
2546 				 * network.
2547 				 *
2548 				 * When using TCP ECN, notify the peer that
2549 				 * we reduced the cwnd.
2550 				 */
2551 				/*
2552 				 * Following 2 kinds of acks should not affect
2553 				 * dupack counting:
2554 				 * 1) Old acks
2555 				 * 2) Acks with SACK but without any new SACK
2556 				 * information in them. These could result from
2557 				 * any anomaly in the network like a switch
2558 				 * duplicating packets or a possible DoS attack.
2559 				 */
2560 				if (th->th_ack != tp->snd_una ||
2561 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2562 				    !sack_changed))
2563 					break;
2564 				else if (!tcp_timer_active(tp, TT_REXMT))
2565 					tp->t_dupacks = 0;
2566 				else if (++tp->t_dupacks > tcprexmtthresh ||
2567 				     IN_FASTRECOVERY(tp->t_flags)) {
2568 					cc_ack_received(tp, th, nsegs,
2569 					    CC_DUPACK);
2570 					if (V_tcp_do_prr &&
2571 					    IN_FASTRECOVERY(tp->t_flags) &&
2572 					    (tp->t_flags & TF_SACK_PERMIT)) {
2573 						long snd_cnt = 0, limit = 0;
2574 						long del_data = 0, pipe = 0;
2575 						/*
2576 						 * In a duplicate ACK del_data is only the
2577 						 * diff_in_sack. If no SACK is used del_data
2578 						 * will be 0. Pipe is the amount of data we
2579 						 * estimate to be in the network.
2580 						 */
2581 						del_data = tp->sackhint.delivered_data;
2582 						if (V_tcp_do_rfc6675_pipe)
2583 							pipe = tcp_compute_pipe(tp);
2584 						else
2585 							pipe = (tp->snd_nxt - tp->snd_fack) +
2586 								tp->sackhint.sack_bytes_rexmit;
2587 						tp->sackhint.prr_delivered += del_data;
2588 						if (pipe > tp->snd_ssthresh) {
2589 							if (tp->sackhint.recover_fs == 0)
2590 								tp->sackhint.recover_fs =
2591 								    max(1, tp->snd_nxt - tp->snd_una);
2592 							snd_cnt = (tp->sackhint.prr_delivered *
2593 							    tp->snd_ssthresh /
2594 							    tp->sackhint.recover_fs) +
2595 							    1 - tp->sackhint.sack_bytes_rexmit;
2596 						} else {
2597 							if (V_tcp_do_prr_conservative)
2598 								limit = tp->sackhint.prr_delivered -
2599 									tp->sackhint.sack_bytes_rexmit;
2600 							else
2601 								if ((tp->sackhint.prr_delivered -
2602 								    tp->sackhint.sack_bytes_rexmit) >
2603 								    del_data)
2604 									limit = tp->sackhint.prr_delivered -
2605 									    tp->sackhint.sack_bytes_rexmit +
2606 									    maxseg;
2607 								else
2608 									limit = del_data + maxseg;
2609 							if ((tp->snd_ssthresh - pipe) < limit)
2610 								snd_cnt = tp->snd_ssthresh - pipe;
2611 							else
2612 								snd_cnt = limit;
2613 						}
2614 						snd_cnt = max((snd_cnt / maxseg), 0);
2615 						/*
2616 						 * Send snd_cnt new data into the network in
2617 						 * response to this ACK. If there is a going
2618 						 * to be a SACK retransmission, adjust snd_cwnd
2619 						 * accordingly.
2620 						 */
2621 						tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
2622 						    tp->sackhint.sack_bytes_rexmit +
2623 						    (snd_cnt * maxseg);
2624 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2625 					    IN_FASTRECOVERY(tp->t_flags)) {
2626 						int awnd;
2627 
2628 						/*
2629 						 * Compute the amount of data in flight first.
2630 						 * We can inject new data into the pipe iff
2631 						 * we have less than 1/2 the original window's
2632 						 * worth of data in flight.
2633 						 */
2634 						if (V_tcp_do_rfc6675_pipe)
2635 							awnd = tcp_compute_pipe(tp);
2636 						else
2637 							awnd = (tp->snd_nxt - tp->snd_fack) +
2638 								tp->sackhint.sack_bytes_rexmit;
2639 
2640 						if (awnd < tp->snd_ssthresh) {
2641 							tp->snd_cwnd += maxseg;
2642 							if (tp->snd_cwnd > tp->snd_ssthresh)
2643 								tp->snd_cwnd = tp->snd_ssthresh;
2644 						}
2645 					} else
2646 						tp->snd_cwnd += maxseg;
2647 					(void) tp->t_fb->tfb_tcp_output(tp);
2648 					goto drop;
2649 				} else if (tp->t_dupacks == tcprexmtthresh ||
2650 					    (tp->t_flags & TF_SACK_PERMIT &&
2651 					     V_tcp_do_rfc6675_pipe &&
2652 					     tp->sackhint.sacked_bytes >
2653 					     (tcprexmtthresh - 1) * maxseg)) {
2654 enter_recovery:
2655 					/*
2656 					 * Above is the RFC6675 trigger condition of
2657 					 * more than (dupthresh-1)*maxseg sacked data.
2658 					 * If the count of holes in the
2659 					 * scoreboard is >= dupthresh, we could
2660 					 * also enter loss recovery, but don't
2661 					 * have that value readily available.
2662 					 */
2663 					tp->t_dupacks = tcprexmtthresh;
2664 					tcp_seq onxt = tp->snd_nxt;
2665 
2666 					/*
2667 					 * If we're doing sack, or prr, check
2668 					 * to see if we're already in sack
2669 					 * recovery. If we're not doing sack,
2670 					 * check to see if we're in newreno
2671 					 * recovery.
2672 					 */
2673 					if (V_tcp_do_prr ||
2674 					    (tp->t_flags & TF_SACK_PERMIT)) {
2675 						if (IN_FASTRECOVERY(tp->t_flags)) {
2676 							tp->t_dupacks = 0;
2677 							break;
2678 						}
2679 					} else {
2680 						if (SEQ_LEQ(th->th_ack,
2681 						    tp->snd_recover)) {
2682 							tp->t_dupacks = 0;
2683 							break;
2684 						}
2685 					}
2686 					/* Congestion signal before ack. */
2687 					cc_cong_signal(tp, th, CC_NDUPACK);
2688 					cc_ack_received(tp, th, nsegs,
2689 					    CC_DUPACK);
2690 					tcp_timer_activate(tp, TT_REXMT, 0);
2691 					tp->t_rtttime = 0;
2692 					if (V_tcp_do_prr) {
2693 						/*
2694 						 * snd_ssthresh is already updated by
2695 						 * cc_cong_signal.
2696 						 */
2697 						tp->sackhint.prr_delivered = 0;
2698 						tp->sackhint.sack_bytes_rexmit = 0;
2699 						tp->sackhint.recover_fs = max(1,
2700 						    tp->snd_nxt - tp->snd_una);
2701 					}
2702 					if (tp->t_flags & TF_SACK_PERMIT) {
2703 						TCPSTAT_INC(
2704 						    tcps_sack_recovery_episode);
2705 						tp->snd_recover = tp->snd_nxt;
2706 						tp->snd_cwnd = maxseg;
2707 						(void) tp->t_fb->tfb_tcp_output(tp);
2708 						if (SEQ_GT(th->th_ack, tp->snd_una))
2709 							goto resume_partialack;
2710 						goto drop;
2711 					}
2712 					tp->snd_nxt = th->th_ack;
2713 					tp->snd_cwnd = maxseg;
2714 					(void) tp->t_fb->tfb_tcp_output(tp);
2715 					KASSERT(tp->snd_limited <= 2,
2716 					    ("%s: tp->snd_limited too big",
2717 					    __func__));
2718 					tp->snd_cwnd = tp->snd_ssthresh +
2719 					     maxseg *
2720 					     (tp->t_dupacks - tp->snd_limited);
2721 					if (SEQ_GT(onxt, tp->snd_nxt))
2722 						tp->snd_nxt = onxt;
2723 					goto drop;
2724 				} else if (V_tcp_do_rfc3042) {
2725 					/*
2726 					 * Process first and second duplicate
2727 					 * ACKs. Each indicates a segment
2728 					 * leaving the network, creating room
2729 					 * for more. Make sure we can send a
2730 					 * packet on reception of each duplicate
2731 					 * ACK by increasing snd_cwnd by one
2732 					 * segment. Restore the original
2733 					 * snd_cwnd after packet transmission.
2734 					 */
2735 					cc_ack_received(tp, th, nsegs,
2736 					    CC_DUPACK);
2737 					uint32_t oldcwnd = tp->snd_cwnd;
2738 					tcp_seq oldsndmax = tp->snd_max;
2739 					u_int sent;
2740 					int avail;
2741 
2742 					KASSERT(tp->t_dupacks == 1 ||
2743 					    tp->t_dupacks == 2,
2744 					    ("%s: dupacks not 1 or 2",
2745 					    __func__));
2746 					if (tp->t_dupacks == 1)
2747 						tp->snd_limited = 0;
2748 					tp->snd_cwnd =
2749 					    (tp->snd_nxt - tp->snd_una) +
2750 					    (tp->t_dupacks - tp->snd_limited) *
2751 					    maxseg;
2752 					/*
2753 					 * Only call tcp_output when there
2754 					 * is new data available to be sent.
2755 					 * Otherwise we would send pure ACKs.
2756 					 */
2757 					SOCKBUF_LOCK(&so->so_snd);
2758 					avail = sbavail(&so->so_snd) -
2759 					    (tp->snd_nxt - tp->snd_una);
2760 					SOCKBUF_UNLOCK(&so->so_snd);
2761 					if (avail > 0)
2762 						(void) tp->t_fb->tfb_tcp_output(tp);
2763 					sent = tp->snd_max - oldsndmax;
2764 					if (sent > maxseg) {
2765 						KASSERT((tp->t_dupacks == 2 &&
2766 						    tp->snd_limited == 0) ||
2767 						   (sent == maxseg + 1 &&
2768 						    tp->t_flags & TF_SENTFIN),
2769 						    ("%s: sent too much",
2770 						    __func__));
2771 						tp->snd_limited = 2;
2772 					} else if (sent > 0)
2773 						++tp->snd_limited;
2774 					tp->snd_cwnd = oldcwnd;
2775 					goto drop;
2776 				}
2777 			}
2778 			break;
2779 		} else {
2780 			/*
2781 			 * This ack is advancing the left edge, reset the
2782 			 * counter.
2783 			 */
2784 			tp->t_dupacks = 0;
2785 			/*
2786 			 * If this ack also has new SACK info, increment the
2787 			 * counter as per rfc6675. The variable
2788 			 * sack_changed tracks all changes to the SACK
2789 			 * scoreboard, including when partial ACKs without
2790 			 * SACK options are received, and clear the scoreboard
2791 			 * from the left side. Such partial ACKs should not be
2792 			 * counted as dupacks here.
2793 			 */
2794 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2795 			    (to.to_flags & TOF_SACK) &&
2796 			    sack_changed) {
2797 				tp->t_dupacks++;
2798 				/* limit overhead by setting maxseg last */
2799 				if (!IN_FASTRECOVERY(tp->t_flags) &&
2800 				    (tp->sackhint.sacked_bytes >
2801 				    ((tcprexmtthresh - 1) *
2802 				    (maxseg = tcp_maxseg(tp))))) {
2803 					goto enter_recovery;
2804 				}
2805 			}
2806 		}
2807 
2808 resume_partialack:
2809 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2810 		    ("%s: th_ack <= snd_una", __func__));
2811 
2812 		/*
2813 		 * If the congestion window was inflated to account
2814 		 * for the other side's cached packets, retract it.
2815 		 */
2816 		if (IN_FASTRECOVERY(tp->t_flags)) {
2817 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2818 				if (tp->t_flags & TF_SACK_PERMIT)
2819 					if (V_tcp_do_prr && to.to_flags & TOF_SACK)
2820 						tcp_prr_partialack(tp, th);
2821 					else
2822 						tcp_sack_partialack(tp, th);
2823 				else
2824 					tcp_newreno_partial_ack(tp, th);
2825 			} else
2826 				cc_post_recovery(tp, th);
2827 		}
2828 		/*
2829 		 * If we reach this point, ACK is not a duplicate,
2830 		 *     i.e., it ACKs something we sent.
2831 		 */
2832 		if (tp->t_flags & TF_NEEDSYN) {
2833 			/*
2834 			 * T/TCP: Connection was half-synchronized, and our
2835 			 * SYN has been ACK'd (so connection is now fully
2836 			 * synchronized).  Go to non-starred state,
2837 			 * increment snd_una for ACK of SYN, and check if
2838 			 * we can do window scaling.
2839 			 */
2840 			tp->t_flags &= ~TF_NEEDSYN;
2841 			tp->snd_una++;
2842 			/* Do window scaling? */
2843 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2844 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2845 				tp->rcv_scale = tp->request_r_scale;
2846 				/* Send window already scaled. */
2847 			}
2848 		}
2849 
2850 process_ACK:
2851 		INP_WLOCK_ASSERT(tp->t_inpcb);
2852 
2853 		/*
2854 		 * Adjust for the SYN bit in sequence space,
2855 		 * but don't account for it in cwnd calculations.
2856 		 * This is for the SYN_RECEIVED, non-simultaneous
2857 		 * SYN case. SYN_SENT and simultaneous SYN are
2858 		 * treated elsewhere.
2859 		 */
2860 		if (incforsyn)
2861 			tp->snd_una++;
2862 		acked = BYTES_THIS_ACK(tp, th);
2863 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2864 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2865 		    tp->snd_una, th->th_ack, tp, m));
2866 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2867 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2868 
2869 		/*
2870 		 * If we just performed our first retransmit, and the ACK
2871 		 * arrives within our recovery window, then it was a mistake
2872 		 * to do the retransmit in the first place.  Recover our
2873 		 * original cwnd and ssthresh, and proceed to transmit where
2874 		 * we left off.
2875 		 */
2876 		if (tp->t_rxtshift == 1 &&
2877 		    tp->t_flags & TF_PREVVALID &&
2878 		    tp->t_badrxtwin &&
2879 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2880 			cc_cong_signal(tp, th, CC_RTO_ERR);
2881 
2882 		/*
2883 		 * If we have a timestamp reply, update smoothed
2884 		 * round trip time.  If no timestamp is present but
2885 		 * transmit timer is running and timed sequence
2886 		 * number was acked, update smoothed round trip time.
2887 		 * Since we now have an rtt measurement, cancel the
2888 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2889 		 * Recompute the initial retransmit timer.
2890 		 *
2891 		 * Some boxes send broken timestamp replies
2892 		 * during the SYN+ACK phase, ignore
2893 		 * timestamps of 0 or we could calculate a
2894 		 * huge RTT and blow up the retransmit timer.
2895 		 */
2896 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2897 			uint32_t t;
2898 
2899 			t = tcp_ts_getticks() - to.to_tsecr;
2900 			if (!tp->t_rttlow || tp->t_rttlow > t)
2901 				tp->t_rttlow = t;
2902 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2903 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2904 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2905 				tp->t_rttlow = ticks - tp->t_rtttime;
2906 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2907 		}
2908 
2909 		/*
2910 		 * If all outstanding data is acked, stop retransmit
2911 		 * timer and remember to restart (more output or persist).
2912 		 * If there is more data to be acked, restart retransmit
2913 		 * timer, using current (possibly backed-off) value.
2914 		 */
2915 		if (th->th_ack == tp->snd_max) {
2916 			tcp_timer_activate(tp, TT_REXMT, 0);
2917 			needoutput = 1;
2918 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2919 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2920 
2921 		/*
2922 		 * If no data (only SYN) was ACK'd,
2923 		 *    skip rest of ACK processing.
2924 		 */
2925 		if (acked == 0)
2926 			goto step6;
2927 
2928 		/*
2929 		 * Let the congestion control algorithm update congestion
2930 		 * control related information. This typically means increasing
2931 		 * the congestion window.
2932 		 */
2933 		cc_ack_received(tp, th, nsegs, CC_ACK);
2934 
2935 		SOCKBUF_LOCK(&so->so_snd);
2936 		if (acked > sbavail(&so->so_snd)) {
2937 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2938 				tp->snd_wnd -= sbavail(&so->so_snd);
2939 			else
2940 				tp->snd_wnd = 0;
2941 			mfree = sbcut_locked(&so->so_snd,
2942 			    (int)sbavail(&so->so_snd));
2943 			ourfinisacked = 1;
2944 		} else {
2945 			mfree = sbcut_locked(&so->so_snd, acked);
2946 			if (tp->snd_wnd >= (uint32_t) acked)
2947 				tp->snd_wnd -= acked;
2948 			else
2949 				tp->snd_wnd = 0;
2950 			ourfinisacked = 0;
2951 		}
2952 		SOCKBUF_UNLOCK(&so->so_snd);
2953 		tp->t_flags |= TF_WAKESOW;
2954 		m_freem(mfree);
2955 		/* Detect una wraparound. */
2956 		if (!IN_RECOVERY(tp->t_flags) &&
2957 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2958 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2959 			tp->snd_recover = th->th_ack - 1;
2960 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2961 		if (IN_RECOVERY(tp->t_flags) &&
2962 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2963 			EXIT_RECOVERY(tp->t_flags);
2964 		}
2965 		tp->snd_una = th->th_ack;
2966 		if (tp->t_flags & TF_SACK_PERMIT) {
2967 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2968 				tp->snd_recover = tp->snd_una;
2969 		}
2970 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2971 			tp->snd_nxt = tp->snd_una;
2972 
2973 		switch (tp->t_state) {
2974 		/*
2975 		 * In FIN_WAIT_1 STATE in addition to the processing
2976 		 * for the ESTABLISHED state if our FIN is now acknowledged
2977 		 * then enter FIN_WAIT_2.
2978 		 */
2979 		case TCPS_FIN_WAIT_1:
2980 			if (ourfinisacked) {
2981 				/*
2982 				 * If we can't receive any more
2983 				 * data, then closing user can proceed.
2984 				 * Starting the timer is contrary to the
2985 				 * specification, but if we don't get a FIN
2986 				 * we'll hang forever.
2987 				 *
2988 				 * XXXjl:
2989 				 * we should release the tp also, and use a
2990 				 * compressed state.
2991 				 */
2992 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2993 					soisdisconnected(so);
2994 					tcp_timer_activate(tp, TT_2MSL,
2995 					    (tcp_fast_finwait2_recycle ?
2996 					    tcp_finwait2_timeout :
2997 					    TP_MAXIDLE(tp)));
2998 				}
2999 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
3000 			}
3001 			break;
3002 
3003 		/*
3004 		 * In CLOSING STATE in addition to the processing for
3005 		 * the ESTABLISHED state if the ACK acknowledges our FIN
3006 		 * then enter the TIME-WAIT state, otherwise ignore
3007 		 * the segment.
3008 		 */
3009 		case TCPS_CLOSING:
3010 			if (ourfinisacked) {
3011 				tcp_twstart(tp);
3012 				m_freem(m);
3013 				return;
3014 			}
3015 			break;
3016 
3017 		/*
3018 		 * In LAST_ACK, we may still be waiting for data to drain
3019 		 * and/or to be acked, as well as for the ack of our FIN.
3020 		 * If our FIN is now acknowledged, delete the TCB,
3021 		 * enter the closed state and return.
3022 		 */
3023 		case TCPS_LAST_ACK:
3024 			if (ourfinisacked) {
3025 				tp = tcp_close(tp);
3026 				goto drop;
3027 			}
3028 			break;
3029 		}
3030 	}
3031 
3032 step6:
3033 	INP_WLOCK_ASSERT(tp->t_inpcb);
3034 
3035 	/*
3036 	 * Update window information.
3037 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3038 	 */
3039 	if ((thflags & TH_ACK) &&
3040 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3041 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3042 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3043 		/* keep track of pure window updates */
3044 		if (tlen == 0 &&
3045 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3046 			TCPSTAT_INC(tcps_rcvwinupd);
3047 		tp->snd_wnd = tiwin;
3048 		tp->snd_wl1 = th->th_seq;
3049 		tp->snd_wl2 = th->th_ack;
3050 		if (tp->snd_wnd > tp->max_sndwnd)
3051 			tp->max_sndwnd = tp->snd_wnd;
3052 		needoutput = 1;
3053 	}
3054 
3055 	/*
3056 	 * Process segments with URG.
3057 	 */
3058 	if ((thflags & TH_URG) && th->th_urp &&
3059 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3060 		/*
3061 		 * This is a kludge, but if we receive and accept
3062 		 * random urgent pointers, we'll crash in
3063 		 * soreceive.  It's hard to imagine someone
3064 		 * actually wanting to send this much urgent data.
3065 		 */
3066 		SOCKBUF_LOCK(&so->so_rcv);
3067 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3068 			th->th_urp = 0;			/* XXX */
3069 			thflags &= ~TH_URG;		/* XXX */
3070 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3071 			goto dodata;			/* XXX */
3072 		}
3073 		/*
3074 		 * If this segment advances the known urgent pointer,
3075 		 * then mark the data stream.  This should not happen
3076 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3077 		 * a FIN has been received from the remote side.
3078 		 * In these states we ignore the URG.
3079 		 *
3080 		 * According to RFC961 (Assigned Protocols),
3081 		 * the urgent pointer points to the last octet
3082 		 * of urgent data.  We continue, however,
3083 		 * to consider it to indicate the first octet
3084 		 * of data past the urgent section as the original
3085 		 * spec states (in one of two places).
3086 		 */
3087 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3088 			tp->rcv_up = th->th_seq + th->th_urp;
3089 			so->so_oobmark = sbavail(&so->so_rcv) +
3090 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3091 			if (so->so_oobmark == 0)
3092 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3093 			sohasoutofband(so);
3094 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3095 		}
3096 		SOCKBUF_UNLOCK(&so->so_rcv);
3097 		/*
3098 		 * Remove out of band data so doesn't get presented to user.
3099 		 * This can happen independent of advancing the URG pointer,
3100 		 * but if two URG's are pending at once, some out-of-band
3101 		 * data may creep in... ick.
3102 		 */
3103 		if (th->th_urp <= (uint32_t)tlen &&
3104 		    !(so->so_options & SO_OOBINLINE)) {
3105 			/* hdr drop is delayed */
3106 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3107 		}
3108 	} else {
3109 		/*
3110 		 * If no out of band data is expected,
3111 		 * pull receive urgent pointer along
3112 		 * with the receive window.
3113 		 */
3114 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3115 			tp->rcv_up = tp->rcv_nxt;
3116 	}
3117 dodata:							/* XXX */
3118 	INP_WLOCK_ASSERT(tp->t_inpcb);
3119 
3120 	/*
3121 	 * Process the segment text, merging it into the TCP sequencing queue,
3122 	 * and arranging for acknowledgment of receipt if necessary.
3123 	 * This process logically involves adjusting tp->rcv_wnd as data
3124 	 * is presented to the user (this happens in tcp_usrreq.c,
3125 	 * case PRU_RCVD).  If a FIN has already been received on this
3126 	 * connection then we just ignore the text.
3127 	 */
3128 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3129 		   IS_FASTOPEN(tp->t_flags));
3130 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3131 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3132 		tcp_seq save_start = th->th_seq;
3133 		tcp_seq save_rnxt  = tp->rcv_nxt;
3134 		int     save_tlen  = tlen;
3135 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3136 		/*
3137 		 * Insert segment which includes th into TCP reassembly queue
3138 		 * with control block tp.  Set thflags to whether reassembly now
3139 		 * includes a segment with FIN.  This handles the common case
3140 		 * inline (segment is the next to be received on an established
3141 		 * connection, and the queue is empty), avoiding linkage into
3142 		 * and removal from the queue and repetition of various
3143 		 * conversions.
3144 		 * Set DELACK for segments received in order, but ack
3145 		 * immediately when segments are out of order (so
3146 		 * fast retransmit can work).
3147 		 */
3148 		if (th->th_seq == tp->rcv_nxt &&
3149 		    SEGQ_EMPTY(tp) &&
3150 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3151 		     tfo_syn)) {
3152 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3153 				tp->t_flags |= TF_DELACK;
3154 			else
3155 				tp->t_flags |= TF_ACKNOW;
3156 			tp->rcv_nxt += tlen;
3157 			if (tlen &&
3158 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3159 			    (tp->t_fbyte_in == 0)) {
3160 				tp->t_fbyte_in = ticks;
3161 				if (tp->t_fbyte_in == 0)
3162 					tp->t_fbyte_in = 1;
3163 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3164 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3165 			}
3166 			thflags = th->th_flags & TH_FIN;
3167 			TCPSTAT_INC(tcps_rcvpack);
3168 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3169 			SOCKBUF_LOCK(&so->so_rcv);
3170 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3171 				m_freem(m);
3172 			else
3173 				sbappendstream_locked(&so->so_rcv, m, 0);
3174 			SOCKBUF_UNLOCK(&so->so_rcv);
3175 			tp->t_flags |= TF_WAKESOR;
3176 		} else {
3177 			/*
3178 			 * XXX: Due to the header drop above "th" is
3179 			 * theoretically invalid by now.  Fortunately
3180 			 * m_adj() doesn't actually frees any mbufs
3181 			 * when trimming from the head.
3182 			 */
3183 			tcp_seq temp = save_start;
3184 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3185 			tp->t_flags |= TF_ACKNOW;
3186 		}
3187 		if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3188 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3189 				/*
3190 				 * DSACK actually handled in the fastpath
3191 				 * above.
3192 				 */
3193 				tcp_update_sack_list(tp, save_start,
3194 				    save_start + save_tlen);
3195 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3196 				if ((tp->rcv_numsacks >= 1) &&
3197 				    (tp->sackblks[0].end == save_start)) {
3198 					/*
3199 					 * Partial overlap, recorded at todrop
3200 					 * above.
3201 					 */
3202 					tcp_update_sack_list(tp,
3203 					    tp->sackblks[0].start,
3204 					    tp->sackblks[0].end);
3205 				} else {
3206 					tcp_update_dsack_list(tp, save_start,
3207 					    save_start + save_tlen);
3208 				}
3209 			} else if (tlen >= save_tlen) {
3210 				/* Update of sackblks. */
3211 				tcp_update_dsack_list(tp, save_start,
3212 				    save_start + save_tlen);
3213 			} else if (tlen > 0) {
3214 				tcp_update_dsack_list(tp, save_start,
3215 				    save_start + tlen);
3216 			}
3217 		}
3218 #if 0
3219 		/*
3220 		 * Note the amount of data that peer has sent into
3221 		 * our window, in order to estimate the sender's
3222 		 * buffer size.
3223 		 * XXX: Unused.
3224 		 */
3225 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3226 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3227 		else
3228 			len = so->so_rcv.sb_hiwat;
3229 #endif
3230 	} else {
3231 		m_freem(m);
3232 		thflags &= ~TH_FIN;
3233 	}
3234 
3235 	/*
3236 	 * If FIN is received ACK the FIN and let the user know
3237 	 * that the connection is closing.
3238 	 */
3239 	if (thflags & TH_FIN) {
3240 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3241 			socantrcvmore(so);
3242 			/* The socket upcall is handled by socantrcvmore. */
3243 			tp->t_flags &= ~TF_WAKESOR;
3244 			/*
3245 			 * If connection is half-synchronized
3246 			 * (ie NEEDSYN flag on) then delay ACK,
3247 			 * so it may be piggybacked when SYN is sent.
3248 			 * Otherwise, since we received a FIN then no
3249 			 * more input can be expected, send ACK now.
3250 			 */
3251 			if (tp->t_flags & TF_NEEDSYN)
3252 				tp->t_flags |= TF_DELACK;
3253 			else
3254 				tp->t_flags |= TF_ACKNOW;
3255 			tp->rcv_nxt++;
3256 		}
3257 		switch (tp->t_state) {
3258 		/*
3259 		 * In SYN_RECEIVED and ESTABLISHED STATES
3260 		 * enter the CLOSE_WAIT state.
3261 		 */
3262 		case TCPS_SYN_RECEIVED:
3263 			tp->t_starttime = ticks;
3264 			/* FALLTHROUGH */
3265 		case TCPS_ESTABLISHED:
3266 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3267 			break;
3268 
3269 		/*
3270 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3271 		 * enter the CLOSING state.
3272 		 */
3273 		case TCPS_FIN_WAIT_1:
3274 			tcp_state_change(tp, TCPS_CLOSING);
3275 			break;
3276 
3277 		/*
3278 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3279 		 * starting the time-wait timer, turning off the other
3280 		 * standard timers.
3281 		 */
3282 		case TCPS_FIN_WAIT_2:
3283 			tcp_twstart(tp);
3284 			return;
3285 		}
3286 	}
3287 #ifdef TCPDEBUG
3288 	if (so->so_options & SO_DEBUG)
3289 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3290 			  &tcp_savetcp, 0);
3291 #endif
3292 	TCP_PROBE3(debug__input, tp, th, m);
3293 
3294 	/*
3295 	 * Return any desired output.
3296 	 */
3297 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3298 		(void) tp->t_fb->tfb_tcp_output(tp);
3299 
3300 check_delack:
3301 	INP_WLOCK_ASSERT(tp->t_inpcb);
3302 
3303 	if (tp->t_flags & TF_DELACK) {
3304 		tp->t_flags &= ~TF_DELACK;
3305 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3306 	}
3307 	tcp_handle_wakeup(tp, so);
3308 	INP_WUNLOCK(tp->t_inpcb);
3309 	return;
3310 
3311 dropafterack:
3312 	/*
3313 	 * Generate an ACK dropping incoming segment if it occupies
3314 	 * sequence space, where the ACK reflects our state.
3315 	 *
3316 	 * We can now skip the test for the RST flag since all
3317 	 * paths to this code happen after packets containing
3318 	 * RST have been dropped.
3319 	 *
3320 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3321 	 * segment we received passes the SYN-RECEIVED ACK test.
3322 	 * If it fails send a RST.  This breaks the loop in the
3323 	 * "LAND" DoS attack, and also prevents an ACK storm
3324 	 * between two listening ports that have been sent forged
3325 	 * SYN segments, each with the source address of the other.
3326 	 */
3327 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3328 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3329 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3330 		rstreason = BANDLIM_RST_OPENPORT;
3331 		goto dropwithreset;
3332 	}
3333 #ifdef TCPDEBUG
3334 	if (so->so_options & SO_DEBUG)
3335 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3336 			  &tcp_savetcp, 0);
3337 #endif
3338 	TCP_PROBE3(debug__input, tp, th, m);
3339 	tp->t_flags |= TF_ACKNOW;
3340 	(void) tp->t_fb->tfb_tcp_output(tp);
3341 	tcp_handle_wakeup(tp, so);
3342 	INP_WUNLOCK(tp->t_inpcb);
3343 	m_freem(m);
3344 	return;
3345 
3346 dropwithreset:
3347 	if (tp != NULL) {
3348 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3349 		tcp_handle_wakeup(tp, so);
3350 		INP_WUNLOCK(tp->t_inpcb);
3351 	} else
3352 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3353 	return;
3354 
3355 drop:
3356 	/*
3357 	 * Drop space held by incoming segment and return.
3358 	 */
3359 #ifdef TCPDEBUG
3360 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3361 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3362 			  &tcp_savetcp, 0);
3363 #endif
3364 	TCP_PROBE3(debug__input, tp, th, m);
3365 	if (tp != NULL) {
3366 		tcp_handle_wakeup(tp, so);
3367 		INP_WUNLOCK(tp->t_inpcb);
3368 	}
3369 	m_freem(m);
3370 }
3371 
3372 /*
3373  * Issue RST and make ACK acceptable to originator of segment.
3374  * The mbuf must still include the original packet header.
3375  * tp may be NULL.
3376  */
3377 void
3378 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3379     int tlen, int rstreason)
3380 {
3381 #ifdef INET
3382 	struct ip *ip;
3383 #endif
3384 #ifdef INET6
3385 	struct ip6_hdr *ip6;
3386 #endif
3387 
3388 	if (tp != NULL) {
3389 		INP_WLOCK_ASSERT(tp->t_inpcb);
3390 	}
3391 
3392 	/* Don't bother if destination was broadcast/multicast. */
3393 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3394 		goto drop;
3395 #ifdef INET6
3396 	if (mtod(m, struct ip *)->ip_v == 6) {
3397 		ip6 = mtod(m, struct ip6_hdr *);
3398 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3399 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3400 			goto drop;
3401 		/* IPv6 anycast check is done at tcp6_input() */
3402 	}
3403 #endif
3404 #if defined(INET) && defined(INET6)
3405 	else
3406 #endif
3407 #ifdef INET
3408 	{
3409 		ip = mtod(m, struct ip *);
3410 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3411 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3412 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3413 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3414 			goto drop;
3415 	}
3416 #endif
3417 
3418 	/* Perform bandwidth limiting. */
3419 	if (badport_bandlim(rstreason) < 0)
3420 		goto drop;
3421 
3422 	/* tcp_respond consumes the mbuf chain. */
3423 	if (th->th_flags & TH_ACK) {
3424 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3425 		    th->th_ack, TH_RST);
3426 	} else {
3427 		if (th->th_flags & TH_SYN)
3428 			tlen++;
3429 		if (th->th_flags & TH_FIN)
3430 			tlen++;
3431 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3432 		    (tcp_seq)0, TH_RST|TH_ACK);
3433 	}
3434 	return;
3435 drop:
3436 	m_freem(m);
3437 }
3438 
3439 /*
3440  * Parse TCP options and place in tcpopt.
3441  */
3442 void
3443 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3444 {
3445 	int opt, optlen;
3446 
3447 	to->to_flags = 0;
3448 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3449 		opt = cp[0];
3450 		if (opt == TCPOPT_EOL)
3451 			break;
3452 		if (opt == TCPOPT_NOP)
3453 			optlen = 1;
3454 		else {
3455 			if (cnt < 2)
3456 				break;
3457 			optlen = cp[1];
3458 			if (optlen < 2 || optlen > cnt)
3459 				break;
3460 		}
3461 		switch (opt) {
3462 		case TCPOPT_MAXSEG:
3463 			if (optlen != TCPOLEN_MAXSEG)
3464 				continue;
3465 			if (!(flags & TO_SYN))
3466 				continue;
3467 			to->to_flags |= TOF_MSS;
3468 			bcopy((char *)cp + 2,
3469 			    (char *)&to->to_mss, sizeof(to->to_mss));
3470 			to->to_mss = ntohs(to->to_mss);
3471 			break;
3472 		case TCPOPT_WINDOW:
3473 			if (optlen != TCPOLEN_WINDOW)
3474 				continue;
3475 			if (!(flags & TO_SYN))
3476 				continue;
3477 			to->to_flags |= TOF_SCALE;
3478 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3479 			break;
3480 		case TCPOPT_TIMESTAMP:
3481 			if (optlen != TCPOLEN_TIMESTAMP)
3482 				continue;
3483 			to->to_flags |= TOF_TS;
3484 			bcopy((char *)cp + 2,
3485 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3486 			to->to_tsval = ntohl(to->to_tsval);
3487 			bcopy((char *)cp + 6,
3488 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3489 			to->to_tsecr = ntohl(to->to_tsecr);
3490 			break;
3491 		case TCPOPT_SIGNATURE:
3492 			/*
3493 			 * In order to reply to a host which has set the
3494 			 * TCP_SIGNATURE option in its initial SYN, we have
3495 			 * to record the fact that the option was observed
3496 			 * here for the syncache code to perform the correct
3497 			 * response.
3498 			 */
3499 			if (optlen != TCPOLEN_SIGNATURE)
3500 				continue;
3501 			to->to_flags |= TOF_SIGNATURE;
3502 			to->to_signature = cp + 2;
3503 			break;
3504 		case TCPOPT_SACK_PERMITTED:
3505 			if (optlen != TCPOLEN_SACK_PERMITTED)
3506 				continue;
3507 			if (!(flags & TO_SYN))
3508 				continue;
3509 			if (!V_tcp_do_sack)
3510 				continue;
3511 			to->to_flags |= TOF_SACKPERM;
3512 			break;
3513 		case TCPOPT_SACK:
3514 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3515 				continue;
3516 			if (flags & TO_SYN)
3517 				continue;
3518 			to->to_flags |= TOF_SACK;
3519 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3520 			to->to_sacks = cp + 2;
3521 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3522 			break;
3523 		case TCPOPT_FAST_OPEN:
3524 			/*
3525 			 * Cookie length validation is performed by the
3526 			 * server side cookie checking code or the client
3527 			 * side cookie cache update code.
3528 			 */
3529 			if (!(flags & TO_SYN))
3530 				continue;
3531 			if (!V_tcp_fastopen_client_enable &&
3532 			    !V_tcp_fastopen_server_enable)
3533 				continue;
3534 			to->to_flags |= TOF_FASTOPEN;
3535 			to->to_tfo_len = optlen - 2;
3536 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3537 			break;
3538 		default:
3539 			continue;
3540 		}
3541 	}
3542 }
3543 
3544 /*
3545  * Pull out of band byte out of a segment so
3546  * it doesn't appear in the user's data queue.
3547  * It is still reflected in the segment length for
3548  * sequencing purposes.
3549  */
3550 void
3551 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3552     int off)
3553 {
3554 	int cnt = off + th->th_urp - 1;
3555 
3556 	while (cnt >= 0) {
3557 		if (m->m_len > cnt) {
3558 			char *cp = mtod(m, caddr_t) + cnt;
3559 			struct tcpcb *tp = sototcpcb(so);
3560 
3561 			INP_WLOCK_ASSERT(tp->t_inpcb);
3562 
3563 			tp->t_iobc = *cp;
3564 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3565 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3566 			m->m_len--;
3567 			if (m->m_flags & M_PKTHDR)
3568 				m->m_pkthdr.len--;
3569 			return;
3570 		}
3571 		cnt -= m->m_len;
3572 		m = m->m_next;
3573 		if (m == NULL)
3574 			break;
3575 	}
3576 	panic("tcp_pulloutofband");
3577 }
3578 
3579 /*
3580  * Collect new round-trip time estimate
3581  * and update averages and current timeout.
3582  */
3583 void
3584 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3585 {
3586 	int delta;
3587 
3588 	INP_WLOCK_ASSERT(tp->t_inpcb);
3589 
3590 	TCPSTAT_INC(tcps_rttupdated);
3591 	tp->t_rttupdated++;
3592 #ifdef STATS
3593 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3594 	    imax(0, rtt * 1000 / hz));
3595 #endif
3596 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3597 		/*
3598 		 * srtt is stored as fixed point with 5 bits after the
3599 		 * binary point (i.e., scaled by 8).  The following magic
3600 		 * is equivalent to the smoothing algorithm in rfc793 with
3601 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3602 		 * point).  Adjust rtt to origin 0.
3603 		 */
3604 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3605 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3606 
3607 		if ((tp->t_srtt += delta) <= 0)
3608 			tp->t_srtt = 1;
3609 
3610 		/*
3611 		 * We accumulate a smoothed rtt variance (actually, a
3612 		 * smoothed mean difference), then set the retransmit
3613 		 * timer to smoothed rtt + 4 times the smoothed variance.
3614 		 * rttvar is stored as fixed point with 4 bits after the
3615 		 * binary point (scaled by 16).  The following is
3616 		 * equivalent to rfc793 smoothing with an alpha of .75
3617 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3618 		 * rfc793's wired-in beta.
3619 		 */
3620 		if (delta < 0)
3621 			delta = -delta;
3622 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3623 		if ((tp->t_rttvar += delta) <= 0)
3624 			tp->t_rttvar = 1;
3625 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3626 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3627 	} else {
3628 		/*
3629 		 * No rtt measurement yet - use the unsmoothed rtt.
3630 		 * Set the variance to half the rtt (so our first
3631 		 * retransmit happens at 3*rtt).
3632 		 */
3633 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3634 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3635 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3636 	}
3637 	tp->t_rtttime = 0;
3638 	tp->t_rxtshift = 0;
3639 
3640 	/*
3641 	 * the retransmit should happen at rtt + 4 * rttvar.
3642 	 * Because of the way we do the smoothing, srtt and rttvar
3643 	 * will each average +1/2 tick of bias.  When we compute
3644 	 * the retransmit timer, we want 1/2 tick of rounding and
3645 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3646 	 * firing of the timer.  The bias will give us exactly the
3647 	 * 1.5 tick we need.  But, because the bias is
3648 	 * statistical, we have to test that we don't drop below
3649 	 * the minimum feasible timer (which is 2 ticks).
3650 	 */
3651 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3652 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3653 
3654 	/*
3655 	 * We received an ack for a packet that wasn't retransmitted;
3656 	 * it is probably safe to discard any error indications we've
3657 	 * received recently.  This isn't quite right, but close enough
3658 	 * for now (a route might have failed after we sent a segment,
3659 	 * and the return path might not be symmetrical).
3660 	 */
3661 	tp->t_softerror = 0;
3662 }
3663 
3664 /*
3665  * Determine a reasonable value for maxseg size.
3666  * If the route is known, check route for mtu.
3667  * If none, use an mss that can be handled on the outgoing interface
3668  * without forcing IP to fragment.  If no route is found, route has no mtu,
3669  * or the destination isn't local, use a default, hopefully conservative
3670  * size (usually 512 or the default IP max size, but no more than the mtu
3671  * of the interface), as we can't discover anything about intervening
3672  * gateways or networks.  We also initialize the congestion/slow start
3673  * window to be a single segment if the destination isn't local.
3674  * While looking at the routing entry, we also initialize other path-dependent
3675  * parameters from pre-set or cached values in the routing entry.
3676  *
3677  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3678  * IP options, e.g. IPSEC data, since length of this data may vary, and
3679  * thus it is calculated for every segment separately in tcp_output().
3680  *
3681  * NOTE that this routine is only called when we process an incoming
3682  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3683  * settings are handled in tcp_mssopt().
3684  */
3685 void
3686 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3687     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3688 {
3689 	int mss = 0;
3690 	uint32_t maxmtu = 0;
3691 	struct inpcb *inp = tp->t_inpcb;
3692 	struct hc_metrics_lite metrics;
3693 #ifdef INET6
3694 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3695 	size_t min_protoh = isipv6 ?
3696 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3697 			    sizeof (struct tcpiphdr);
3698 #else
3699 	const size_t min_protoh = sizeof(struct tcpiphdr);
3700 #endif
3701 
3702 	INP_WLOCK_ASSERT(tp->t_inpcb);
3703 
3704 	if (mtuoffer != -1) {
3705 		KASSERT(offer == -1, ("%s: conflict", __func__));
3706 		offer = mtuoffer - min_protoh;
3707 	}
3708 
3709 	/* Initialize. */
3710 #ifdef INET6
3711 	if (isipv6) {
3712 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3713 		tp->t_maxseg = V_tcp_v6mssdflt;
3714 	}
3715 #endif
3716 #if defined(INET) && defined(INET6)
3717 	else
3718 #endif
3719 #ifdef INET
3720 	{
3721 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3722 		tp->t_maxseg = V_tcp_mssdflt;
3723 	}
3724 #endif
3725 
3726 	/*
3727 	 * No route to sender, stay with default mss and return.
3728 	 */
3729 	if (maxmtu == 0) {
3730 		/*
3731 		 * In case we return early we need to initialize metrics
3732 		 * to a defined state as tcp_hc_get() would do for us
3733 		 * if there was no cache hit.
3734 		 */
3735 		if (metricptr != NULL)
3736 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3737 		return;
3738 	}
3739 
3740 	/* What have we got? */
3741 	switch (offer) {
3742 		case 0:
3743 			/*
3744 			 * Offer == 0 means that there was no MSS on the SYN
3745 			 * segment, in this case we use tcp_mssdflt as
3746 			 * already assigned to t_maxseg above.
3747 			 */
3748 			offer = tp->t_maxseg;
3749 			break;
3750 
3751 		case -1:
3752 			/*
3753 			 * Offer == -1 means that we didn't receive SYN yet.
3754 			 */
3755 			/* FALLTHROUGH */
3756 
3757 		default:
3758 			/*
3759 			 * Prevent DoS attack with too small MSS. Round up
3760 			 * to at least minmss.
3761 			 */
3762 			offer = max(offer, V_tcp_minmss);
3763 	}
3764 
3765 	/*
3766 	 * rmx information is now retrieved from tcp_hostcache.
3767 	 */
3768 	tcp_hc_get(&inp->inp_inc, &metrics);
3769 	if (metricptr != NULL)
3770 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3771 
3772 	/*
3773 	 * If there's a discovered mtu in tcp hostcache, use it.
3774 	 * Else, use the link mtu.
3775 	 */
3776 	if (metrics.rmx_mtu)
3777 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3778 	else {
3779 #ifdef INET6
3780 		if (isipv6) {
3781 			mss = maxmtu - min_protoh;
3782 			if (!V_path_mtu_discovery &&
3783 			    !in6_localaddr(&inp->in6p_faddr))
3784 				mss = min(mss, V_tcp_v6mssdflt);
3785 		}
3786 #endif
3787 #if defined(INET) && defined(INET6)
3788 		else
3789 #endif
3790 #ifdef INET
3791 		{
3792 			mss = maxmtu - min_protoh;
3793 			if (!V_path_mtu_discovery &&
3794 			    !in_localaddr(inp->inp_faddr))
3795 				mss = min(mss, V_tcp_mssdflt);
3796 		}
3797 #endif
3798 		/*
3799 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3800 		 * probably violates the TCP spec.
3801 		 * The problem is that, since we don't know the
3802 		 * other end's MSS, we are supposed to use a conservative
3803 		 * default.  But, if we do that, then MTU discovery will
3804 		 * never actually take place, because the conservative
3805 		 * default is much less than the MTUs typically seen
3806 		 * on the Internet today.  For the moment, we'll sweep
3807 		 * this under the carpet.
3808 		 *
3809 		 * The conservative default might not actually be a problem
3810 		 * if the only case this occurs is when sending an initial
3811 		 * SYN with options and data to a host we've never talked
3812 		 * to before.  Then, they will reply with an MSS value which
3813 		 * will get recorded and the new parameters should get
3814 		 * recomputed.  For Further Study.
3815 		 */
3816 	}
3817 	mss = min(mss, offer);
3818 
3819 	/*
3820 	 * Sanity check: make sure that maxseg will be large
3821 	 * enough to allow some data on segments even if the
3822 	 * all the option space is used (40bytes).  Otherwise
3823 	 * funny things may happen in tcp_output.
3824 	 *
3825 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3826 	 */
3827 	mss = max(mss, 64);
3828 
3829 	tp->t_maxseg = mss;
3830 }
3831 
3832 void
3833 tcp_mss(struct tcpcb *tp, int offer)
3834 {
3835 	int mss;
3836 	uint32_t bufsize;
3837 	struct inpcb *inp;
3838 	struct socket *so;
3839 	struct hc_metrics_lite metrics;
3840 	struct tcp_ifcap cap;
3841 
3842 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3843 
3844 	bzero(&cap, sizeof(cap));
3845 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3846 
3847 	mss = tp->t_maxseg;
3848 	inp = tp->t_inpcb;
3849 
3850 	/*
3851 	 * If there's a pipesize, change the socket buffer to that size,
3852 	 * don't change if sb_hiwat is different than default (then it
3853 	 * has been changed on purpose with setsockopt).
3854 	 * Make the socket buffers an integral number of mss units;
3855 	 * if the mss is larger than the socket buffer, decrease the mss.
3856 	 */
3857 	so = inp->inp_socket;
3858 	SOCKBUF_LOCK(&so->so_snd);
3859 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3860 		bufsize = metrics.rmx_sendpipe;
3861 	else
3862 		bufsize = so->so_snd.sb_hiwat;
3863 	if (bufsize < mss)
3864 		mss = bufsize;
3865 	else {
3866 		bufsize = roundup(bufsize, mss);
3867 		if (bufsize > sb_max)
3868 			bufsize = sb_max;
3869 		if (bufsize > so->so_snd.sb_hiwat)
3870 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3871 	}
3872 	SOCKBUF_UNLOCK(&so->so_snd);
3873 	/*
3874 	 * Sanity check: make sure that maxseg will be large
3875 	 * enough to allow some data on segments even if the
3876 	 * all the option space is used (40bytes).  Otherwise
3877 	 * funny things may happen in tcp_output.
3878 	 *
3879 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3880 	 */
3881 	tp->t_maxseg = max(mss, 64);
3882 
3883 	SOCKBUF_LOCK(&so->so_rcv);
3884 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3885 		bufsize = metrics.rmx_recvpipe;
3886 	else
3887 		bufsize = so->so_rcv.sb_hiwat;
3888 	if (bufsize > mss) {
3889 		bufsize = roundup(bufsize, mss);
3890 		if (bufsize > sb_max)
3891 			bufsize = sb_max;
3892 		if (bufsize > so->so_rcv.sb_hiwat)
3893 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3894 	}
3895 	SOCKBUF_UNLOCK(&so->so_rcv);
3896 
3897 	/* Check the interface for TSO capabilities. */
3898 	if (cap.ifcap & CSUM_TSO) {
3899 		tp->t_flags |= TF_TSO;
3900 		tp->t_tsomax = cap.tsomax;
3901 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3902 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3903 	}
3904 }
3905 
3906 /*
3907  * Determine the MSS option to send on an outgoing SYN.
3908  */
3909 int
3910 tcp_mssopt(struct in_conninfo *inc)
3911 {
3912 	int mss = 0;
3913 	uint32_t thcmtu = 0;
3914 	uint32_t maxmtu = 0;
3915 	size_t min_protoh;
3916 
3917 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3918 
3919 #ifdef INET6
3920 	if (inc->inc_flags & INC_ISIPV6) {
3921 		mss = V_tcp_v6mssdflt;
3922 		maxmtu = tcp_maxmtu6(inc, NULL);
3923 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3924 	}
3925 #endif
3926 #if defined(INET) && defined(INET6)
3927 	else
3928 #endif
3929 #ifdef INET
3930 	{
3931 		mss = V_tcp_mssdflt;
3932 		maxmtu = tcp_maxmtu(inc, NULL);
3933 		min_protoh = sizeof(struct tcpiphdr);
3934 	}
3935 #endif
3936 #if defined(INET6) || defined(INET)
3937 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3938 #endif
3939 
3940 	if (maxmtu && thcmtu)
3941 		mss = min(maxmtu, thcmtu) - min_protoh;
3942 	else if (maxmtu || thcmtu)
3943 		mss = max(maxmtu, thcmtu) - min_protoh;
3944 
3945 	return (mss);
3946 }
3947 
3948 void
3949 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3950 {
3951 	long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3952 	int maxseg = tcp_maxseg(tp);
3953 
3954 	INP_WLOCK_ASSERT(tp->t_inpcb);
3955 
3956 	tcp_timer_activate(tp, TT_REXMT, 0);
3957 	tp->t_rtttime = 0;
3958 	/*
3959 	 * Compute the amount of data that this ACK is indicating
3960 	 * (del_data) and an estimate of how many bytes are in the
3961 	 * network.
3962 	 */
3963 	if (SEQ_GEQ(th->th_ack, tp->snd_una))
3964 		del_data = BYTES_THIS_ACK(tp, th);
3965 	del_data += tp->sackhint.delivered_data;
3966 	if (V_tcp_do_rfc6675_pipe)
3967 		pipe = tcp_compute_pipe(tp);
3968 	else
3969 		pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3970 	tp->sackhint.prr_delivered += del_data;
3971 	/*
3972 	 * Proportional Rate Reduction
3973 	 */
3974 	if (pipe > tp->snd_ssthresh) {
3975 		if (tp->sackhint.recover_fs == 0)
3976 			tp->sackhint.recover_fs =
3977 			    max(1, tp->snd_nxt - tp->snd_una);
3978 		snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh /
3979 		    tp->sackhint.recover_fs) - tp->sackhint.sack_bytes_rexmit;
3980 	} else {
3981 		if (V_tcp_do_prr_conservative)
3982 			limit = tp->sackhint.prr_delivered -
3983 			    tp->sackhint.sack_bytes_rexmit;
3984 		else
3985 			if ((tp->sackhint.prr_delivered -
3986 			    tp->sackhint.sack_bytes_rexmit) > del_data)
3987 				limit = tp->sackhint.prr_delivered -
3988 				    tp->sackhint.sack_bytes_rexmit + maxseg;
3989 			else
3990 				limit = del_data + maxseg;
3991 		snd_cnt = min((tp->snd_ssthresh - pipe), limit);
3992 	}
3993 	snd_cnt = max((snd_cnt / maxseg), 0);
3994 	/*
3995 	 * Send snd_cnt new data into the network in response to this ack.
3996 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
3997 	 * accordingly.
3998 	 */
3999 	tp->snd_cwnd = max(maxseg, (int64_t)tp->snd_nxt - tp->snd_recover +
4000 		tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
4001 	tp->t_flags |= TF_ACKNOW;
4002 	(void) tcp_output(tp);
4003 }
4004 
4005 /*
4006  * On a partial ack arrives, force the retransmission of the
4007  * next unacknowledged segment.  Do not clear tp->t_dupacks.
4008  * By setting snd_nxt to ti_ack, this forces retransmission timer to
4009  * be started again.
4010  */
4011 void
4012 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4013 {
4014 	tcp_seq onxt = tp->snd_nxt;
4015 	uint32_t ocwnd = tp->snd_cwnd;
4016 	u_int maxseg = tcp_maxseg(tp);
4017 
4018 	INP_WLOCK_ASSERT(tp->t_inpcb);
4019 
4020 	tcp_timer_activate(tp, TT_REXMT, 0);
4021 	tp->t_rtttime = 0;
4022 	tp->snd_nxt = th->th_ack;
4023 	/*
4024 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4025 	 * (tp->snd_una has not yet been updated when this function is called.)
4026 	 */
4027 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4028 	tp->t_flags |= TF_ACKNOW;
4029 	(void) tp->t_fb->tfb_tcp_output(tp);
4030 	tp->snd_cwnd = ocwnd;
4031 	if (SEQ_GT(onxt, tp->snd_nxt))
4032 		tp->snd_nxt = onxt;
4033 	/*
4034 	 * Partial window deflation.  Relies on fact that tp->snd_una
4035 	 * not updated yet.
4036 	 */
4037 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4038 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4039 	else
4040 		tp->snd_cwnd = 0;
4041 	tp->snd_cwnd += maxseg;
4042 }
4043 
4044 int
4045 tcp_compute_pipe(struct tcpcb *tp)
4046 {
4047 	return (tp->snd_max - tp->snd_una +
4048 		tp->sackhint.sack_bytes_rexmit -
4049 		tp->sackhint.sacked_bytes);
4050 }
4051 
4052 uint32_t
4053 tcp_compute_initwnd(uint32_t maxseg)
4054 {
4055 	/*
4056 	 * Calculate the Initial Window, also used as Restart Window
4057 	 *
4058 	 * RFC5681 Section 3.1 specifies the default conservative values.
4059 	 * RFC3390 specifies slightly more aggressive values.
4060 	 * RFC6928 increases it to ten segments.
4061 	 * Support for user specified value for initial flight size.
4062 	 */
4063 	if (V_tcp_initcwnd_segments)
4064 		return min(V_tcp_initcwnd_segments * maxseg,
4065 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4066 	else if (V_tcp_do_rfc3390)
4067 		return min(4 * maxseg, max(2 * maxseg, 4380));
4068 	else {
4069 		/* Per RFC5681 Section 3.1 */
4070 		if (maxseg > 2190)
4071 			return (2 * maxseg);
4072 		else if (maxseg > 1095)
4073 			return (3 * maxseg);
4074 		else
4075 			return (4 * maxseg);
4076 	}
4077 }
4078