xref: /freebsd/sys/netinet/tcp_input.c (revision 6be3386466ab79a84b48429ae66244f21526d3df)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2007-2008,2010
7  *	Swinburne University of Technology, Melbourne, Australia.
8  * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9  * Copyright (c) 2010 The FreeBSD Foundation
10  * Copyright (c) 2010-2011 Juniper Networks, Inc.
11  * All rights reserved.
12  *
13  * Portions of this software were developed at the Centre for Advanced Internet
14  * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15  * James Healy and David Hayes, made possible in part by a grant from the Cisco
16  * University Research Program Fund at Community Foundation Silicon Valley.
17  *
18  * Portions of this software were developed at the Centre for Advanced
19  * Internet Architectures, Swinburne University of Technology, Melbourne,
20  * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21  *
22  * Portions of this software were developed by Robert N. M. Watson under
23  * contract to Juniper Networks, Inc.
24  *
25  * Redistribution and use in source and binary forms, with or without
26  * modification, are permitted provided that the following conditions
27  * are met:
28  * 1. Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  * 2. Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in the
32  *    documentation and/or other materials provided with the distribution.
33  * 3. Neither the name of the University nor the names of its contributors
34  *    may be used to endorse or promote products derived from this software
35  *    without specific prior written permission.
36  *
37  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47  * SUCH DAMAGE.
48  *
49  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
50  */
51 
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
54 
55 #include "opt_inet.h"
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
59 
60 #include <sys/param.h>
61 #include <sys/arb.h>
62 #include <sys/kernel.h>
63 #ifdef TCP_HHOOK
64 #include <sys/hhook.h>
65 #endif
66 #include <sys/malloc.h>
67 #include <sys/mbuf.h>
68 #include <sys/proc.h>		/* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
71 #include <sys/sdt.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
79 
80 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
81 
82 #include <vm/uma.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #define TCPSTATES		/* for logging */
90 
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h>	/* required for icmp_var.h */
97 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
116 #ifdef TCPPCAP
117 #include <netinet/tcp_pcap.h>
118 #endif
119 #include <netinet/tcp_syncache.h>
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
123 #ifdef TCP_OFFLOAD
124 #include <netinet/tcp_offload.h>
125 #endif
126 
127 #include <netipsec/ipsec_support.h>
128 
129 #include <machine/in_cksum.h>
130 
131 #include <security/mac/mac_framework.h>
132 
133 const int tcprexmtthresh = 3;
134 
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137     &VNET_NAME(tcp_log_in_vain), 0,
138     "Log all incoming TCP segments to closed ports");
139 
140 VNET_DEFINE(int, blackhole) = 0;
141 #define	V_blackhole		VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143     &VNET_NAME(blackhole), 0,
144     "Do not send RST on segments to closed ports");
145 
146 VNET_DEFINE(int, tcp_delack_enabled) = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
148     &VNET_NAME(tcp_delack_enabled), 0,
149     "Delay ACK to try and piggyback it onto a data packet");
150 
151 VNET_DEFINE(int, drop_synfin) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153     &VNET_NAME(drop_synfin), 0,
154     "Drop TCP packets with SYN+FIN set");
155 
156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
158     &VNET_NAME(tcp_do_prr_conservative), 0,
159     "Do conservative Proportional Rate Reduction");
160 
161 VNET_DEFINE(int, tcp_do_prr) = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
163     &VNET_NAME(tcp_do_prr), 1,
164     "Enable Proportional Rate Reduction per RFC 6937");
165 
166 VNET_DEFINE(int, tcp_do_newcwv) = 0;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
168     &VNET_NAME(tcp_do_newcwv), 0,
169     "Enable New Congestion Window Validation per RFC7661");
170 
171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
173     &VNET_NAME(tcp_do_rfc6675_pipe), 0,
174     "Use calculated pipe/in-flight bytes per RFC 6675");
175 
176 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
178     &VNET_NAME(tcp_do_rfc3042), 0,
179     "Enable RFC 3042 (Limited Transmit)");
180 
181 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
183     &VNET_NAME(tcp_do_rfc3390), 0,
184     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
185 
186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
188     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
189     "Slow-start flight size (initial congestion window) in number of segments");
190 
191 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
193     &VNET_NAME(tcp_do_rfc3465), 0,
194     "Enable RFC 3465 (Appropriate Byte Counting)");
195 
196 VNET_DEFINE(int, tcp_abc_l_var) = 2;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
198     &VNET_NAME(tcp_abc_l_var), 2,
199     "Cap the max cwnd increment during slow-start to this number of segments");
200 
201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
202     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203     "TCP ECN");
204 
205 VNET_DEFINE(int, tcp_do_ecn) = 2;
206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
207     &VNET_NAME(tcp_do_ecn), 0,
208     "TCP ECN support");
209 
210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
212     &VNET_NAME(tcp_ecn_maxretries), 0,
213     "Max retries before giving up on ECN");
214 
215 VNET_DEFINE(int, tcp_insecure_syn) = 0;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
217     &VNET_NAME(tcp_insecure_syn), 0,
218     "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
219 
220 VNET_DEFINE(int, tcp_insecure_rst) = 0;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_insecure_rst), 0,
223     "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
224 
225 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
226 #define	V_tcp_recvspace	VNET(tcp_recvspace)
227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
228     &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
229 
230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
232     &VNET_NAME(tcp_do_autorcvbuf), 0,
233     "Enable automatic receive buffer sizing");
234 
235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
237     &VNET_NAME(tcp_autorcvbuf_max), 0,
238     "Max size of automatic receive buffer");
239 
240 VNET_DEFINE(struct inpcbhead, tcb);
241 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
242 VNET_DEFINE(struct inpcbinfo, tcbinfo);
243 
244 /*
245  * TCP statistics are stored in an array of counter(9)s, which size matches
246  * size of struct tcpstat.  TCP running connection count is a regular array.
247  */
248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
250     tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
253     CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
254     "TCP connection counts by TCP state");
255 
256 static void
257 tcp_vnet_init(const void *unused)
258 {
259 
260 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
261 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
262 }
263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264     tcp_vnet_init, NULL);
265 
266 #ifdef VIMAGE
267 static void
268 tcp_vnet_uninit(const void *unused)
269 {
270 
271 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
272 	VNET_PCPUSTAT_FREE(tcpstat);
273 }
274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275     tcp_vnet_uninit, NULL);
276 #endif /* VIMAGE */
277 
278 /*
279  * Kernel module interface for updating tcpstat.  The first argument is an index
280  * into tcpstat treated as an array.
281  */
282 void
283 kmod_tcpstat_add(int statnum, int val)
284 {
285 
286 	counter_u64_add(VNET(tcpstat)[statnum], val);
287 }
288 
289 #ifdef TCP_HHOOK
290 /*
291  * Wrapper for the TCP established input helper hook.
292  */
293 void
294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
295 {
296 	struct tcp_hhook_data hhook_data;
297 
298 	if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
299 		hhook_data.tp = tp;
300 		hhook_data.th = th;
301 		hhook_data.to = to;
302 
303 		hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 		    tp->osd);
305 	}
306 }
307 #endif
308 
309 /*
310  * CC wrapper hook functions
311  */
312 void
313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
314     uint16_t type)
315 {
316 #ifdef STATS
317 	int32_t gput;
318 #endif
319 
320 	INP_WLOCK_ASSERT(tp->t_inpcb);
321 
322 	tp->ccv->nsegs = nsegs;
323 	tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
324 	if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
325 	    (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
326 	     (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
327 		tp->ccv->flags |= CCF_CWND_LIMITED;
328 	else
329 		tp->ccv->flags &= ~CCF_CWND_LIMITED;
330 
331 	if (type == CC_ACK) {
332 #ifdef STATS
333 		stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
334 		    ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
335 		if (!IN_RECOVERY(tp->t_flags))
336 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
337 			   tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
338 		if ((tp->t_flags & TF_GPUTINPROG) &&
339 		    SEQ_GEQ(th->th_ack, tp->gput_ack)) {
340 			/*
341 			 * Compute goodput in bits per millisecond.
342 			 */
343 			gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
344 			    max(1, tcp_ts_getticks() - tp->gput_ts);
345 			stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
346 			    gput);
347 			/*
348 			 * XXXLAS: This is a temporary hack, and should be
349 			 * chained off VOI_TCP_GPUT when stats(9) grows an API
350 			 * to deal with chained VOIs.
351 			 */
352 			if (tp->t_stats_gput_prev > 0)
353 				stats_voi_update_abs_s32(tp->t_stats,
354 				    VOI_TCP_GPUT_ND,
355 				    ((gput - tp->t_stats_gput_prev) * 100) /
356 				    tp->t_stats_gput_prev);
357 			tp->t_flags &= ~TF_GPUTINPROG;
358 			tp->t_stats_gput_prev = gput;
359 		}
360 #endif /* STATS */
361 		if (tp->snd_cwnd > tp->snd_ssthresh) {
362 			tp->t_bytes_acked += tp->ccv->bytes_this_ack;
363 			if (tp->t_bytes_acked >= tp->snd_cwnd) {
364 				tp->t_bytes_acked -= tp->snd_cwnd;
365 				tp->ccv->flags |= CCF_ABC_SENTAWND;
366 			}
367 		} else {
368 				tp->ccv->flags &= ~CCF_ABC_SENTAWND;
369 				tp->t_bytes_acked = 0;
370 		}
371 	}
372 
373 	if (CC_ALGO(tp)->ack_received != NULL) {
374 		/* XXXLAS: Find a way to live without this */
375 		tp->ccv->curack = th->th_ack;
376 		CC_ALGO(tp)->ack_received(tp->ccv, type);
377 	}
378 #ifdef STATS
379 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
380 #endif
381 }
382 
383 void
384 cc_conn_init(struct tcpcb *tp)
385 {
386 	struct hc_metrics_lite metrics;
387 	struct inpcb *inp = tp->t_inpcb;
388 	u_int maxseg;
389 	int rtt;
390 
391 	INP_WLOCK_ASSERT(tp->t_inpcb);
392 
393 	tcp_hc_get(&inp->inp_inc, &metrics);
394 	maxseg = tcp_maxseg(tp);
395 
396 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
397 		tp->t_srtt = rtt;
398 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
399 		TCPSTAT_INC(tcps_usedrtt);
400 		if (metrics.rmx_rttvar) {
401 			tp->t_rttvar = metrics.rmx_rttvar;
402 			TCPSTAT_INC(tcps_usedrttvar);
403 		} else {
404 			/* default variation is +- 1 rtt */
405 			tp->t_rttvar =
406 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
407 		}
408 		TCPT_RANGESET(tp->t_rxtcur,
409 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
410 		    tp->t_rttmin, TCPTV_REXMTMAX);
411 	}
412 	if (metrics.rmx_ssthresh) {
413 		/*
414 		 * There's some sort of gateway or interface
415 		 * buffer limit on the path.  Use this to set
416 		 * the slow start threshold, but set the
417 		 * threshold to no less than 2*mss.
418 		 */
419 		tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
420 		TCPSTAT_INC(tcps_usedssthresh);
421 	}
422 
423 	/*
424 	 * Set the initial slow-start flight size.
425 	 *
426 	 * If a SYN or SYN/ACK was lost and retransmitted, we have to
427 	 * reduce the initial CWND to one segment as congestion is likely
428 	 * requiring us to be cautious.
429 	 */
430 	if (tp->snd_cwnd == 1)
431 		tp->snd_cwnd = maxseg;		/* SYN(-ACK) lost */
432 	else
433 		tp->snd_cwnd = tcp_compute_initwnd(maxseg);
434 
435 	if (CC_ALGO(tp)->conn_init != NULL)
436 		CC_ALGO(tp)->conn_init(tp->ccv);
437 }
438 
439 void inline
440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
441 {
442 	INP_WLOCK_ASSERT(tp->t_inpcb);
443 
444 #ifdef STATS
445 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
446 #endif
447 
448 	switch(type) {
449 	case CC_NDUPACK:
450 		if (!IN_FASTRECOVERY(tp->t_flags)) {
451 			tp->snd_recover = tp->snd_max;
452 			if (tp->t_flags2 & TF2_ECN_PERMIT)
453 				tp->t_flags2 |= TF2_ECN_SND_CWR;
454 		}
455 		break;
456 	case CC_ECN:
457 		if (!IN_CONGRECOVERY(tp->t_flags) ||
458 		    /*
459 		     * Allow ECN reaction on ACK to CWR, if
460 		     * that data segment was also CE marked.
461 		     */
462 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
463 			EXIT_CONGRECOVERY(tp->t_flags);
464 			TCPSTAT_INC(tcps_ecn_rcwnd);
465 			tp->snd_recover = tp->snd_max + 1;
466 			if (tp->t_flags2 & TF2_ECN_PERMIT)
467 				tp->t_flags2 |= TF2_ECN_SND_CWR;
468 		}
469 		break;
470 	case CC_RTO:
471 		tp->t_dupacks = 0;
472 		tp->t_bytes_acked = 0;
473 		EXIT_RECOVERY(tp->t_flags);
474 		if (tp->t_flags2 & TF2_ECN_PERMIT)
475 			tp->t_flags2 |= TF2_ECN_SND_CWR;
476 		break;
477 	case CC_RTO_ERR:
478 		TCPSTAT_INC(tcps_sndrexmitbad);
479 		/* RTO was unnecessary, so reset everything. */
480 		tp->snd_cwnd = tp->snd_cwnd_prev;
481 		tp->snd_ssthresh = tp->snd_ssthresh_prev;
482 		tp->snd_recover = tp->snd_recover_prev;
483 		if (tp->t_flags & TF_WASFRECOVERY)
484 			ENTER_FASTRECOVERY(tp->t_flags);
485 		if (tp->t_flags & TF_WASCRECOVERY)
486 			ENTER_CONGRECOVERY(tp->t_flags);
487 		tp->snd_nxt = tp->snd_max;
488 		tp->t_flags &= ~TF_PREVVALID;
489 		tp->t_badrxtwin = 0;
490 		break;
491 	}
492 
493 	if (CC_ALGO(tp)->cong_signal != NULL) {
494 		if (th != NULL)
495 			tp->ccv->curack = th->th_ack;
496 		CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 	}
498 }
499 
500 void inline
501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
502 {
503 	INP_WLOCK_ASSERT(tp->t_inpcb);
504 
505 	/* XXXLAS: KASSERT that we're in recovery? */
506 
507 	if (CC_ALGO(tp)->post_recovery != NULL) {
508 		tp->ccv->curack = th->th_ack;
509 		CC_ALGO(tp)->post_recovery(tp->ccv);
510 	}
511 	/* XXXLAS: EXIT_RECOVERY ? */
512 	tp->t_bytes_acked = 0;
513 	tp->sackhint.prr_out = 0;
514 }
515 
516 /*
517  * Indicate whether this ack should be delayed.  We can delay the ack if
518  * following conditions are met:
519  *	- There is no delayed ack timer in progress.
520  *	- Our last ack wasn't a 0-sized window. We never want to delay
521  *	  the ack that opens up a 0-sized window.
522  *	- LRO wasn't used for this segment. We make sure by checking that the
523  *	  segment size is not larger than the MSS.
524  */
525 #define DELAY_ACK(tp, tlen)						\
526 	((!tcp_timer_active(tp, TT_DELACK) &&				\
527 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
528 	    (tlen <= tp->t_maxseg) &&					\
529 	    (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
530 
531 void inline
532 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
533 {
534 	INP_WLOCK_ASSERT(tp->t_inpcb);
535 
536 	if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
537 		switch (iptos & IPTOS_ECN_MASK) {
538 		case IPTOS_ECN_CE:
539 			tp->ccv->flags |= CCF_IPHDR_CE;
540 			break;
541 		case IPTOS_ECN_ECT0:
542 			/* FALLTHROUGH */
543 		case IPTOS_ECN_ECT1:
544 			/* FALLTHROUGH */
545 		case IPTOS_ECN_NOTECT:
546 			tp->ccv->flags &= ~CCF_IPHDR_CE;
547 			break;
548 		}
549 
550 		if (th->th_flags & TH_CWR)
551 			tp->ccv->flags |= CCF_TCPHDR_CWR;
552 		else
553 			tp->ccv->flags &= ~CCF_TCPHDR_CWR;
554 
555 		CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
556 
557 		if (tp->ccv->flags & CCF_ACKNOW) {
558 			tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
559 			tp->t_flags |= TF_ACKNOW;
560 		}
561 	}
562 }
563 
564 /*
565  * TCP input handling is split into multiple parts:
566  *   tcp6_input is a thin wrapper around tcp_input for the extended
567  *	ip6_protox[] call format in ip6_input
568  *   tcp_input handles primary segment validation, inpcb lookup and
569  *	SYN processing on listen sockets
570  *   tcp_do_segment processes the ACK and text of the segment for
571  *	establishing, established and closing connections
572  */
573 #ifdef INET6
574 int
575 tcp6_input(struct mbuf **mp, int *offp, int proto)
576 {
577 	struct mbuf *m;
578 	struct in6_ifaddr *ia6;
579 	struct ip6_hdr *ip6;
580 
581 	m = *mp;
582 	if (m->m_len < *offp + sizeof(struct tcphdr)) {
583 		m = m_pullup(m, *offp + sizeof(struct tcphdr));
584 		if (m == NULL) {
585 			*mp = m;
586 			TCPSTAT_INC(tcps_rcvshort);
587 			return (IPPROTO_DONE);
588 		}
589 	}
590 
591 	/*
592 	 * draft-itojun-ipv6-tcp-to-anycast
593 	 * better place to put this in?
594 	 */
595 	ip6 = mtod(m, struct ip6_hdr *);
596 	ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
597 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
598 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
599 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
600 		*mp = NULL;
601 		return (IPPROTO_DONE);
602 	}
603 
604 	*mp = m;
605 	return (tcp_input(mp, offp, proto));
606 }
607 #endif /* INET6 */
608 
609 int
610 tcp_input(struct mbuf **mp, int *offp, int proto)
611 {
612 	struct mbuf *m = *mp;
613 	struct tcphdr *th = NULL;
614 	struct ip *ip = NULL;
615 	struct inpcb *inp = NULL;
616 	struct tcpcb *tp = NULL;
617 	struct socket *so = NULL;
618 	u_char *optp = NULL;
619 	int off0;
620 	int optlen = 0;
621 #ifdef INET
622 	int len;
623 	uint8_t ipttl;
624 #endif
625 	int tlen = 0, off;
626 	int drop_hdrlen;
627 	int thflags;
628 	int rstreason = 0;	/* For badport_bandlim accounting purposes */
629 	uint8_t iptos;
630 	struct m_tag *fwd_tag = NULL;
631 #ifdef INET6
632 	struct ip6_hdr *ip6 = NULL;
633 	int isipv6;
634 #else
635 	const void *ip6 = NULL;
636 #endif /* INET6 */
637 	struct tcpopt to;		/* options in this segment */
638 	char *s = NULL;			/* address and port logging */
639 #ifdef TCPDEBUG
640 	/*
641 	 * The size of tcp_saveipgen must be the size of the max ip header,
642 	 * now IPv6.
643 	 */
644 	u_char tcp_saveipgen[IP6_HDR_LEN];
645 	struct tcphdr tcp_savetcp;
646 	short ostate = 0;
647 #endif
648 
649 	NET_EPOCH_ASSERT();
650 
651 #ifdef INET6
652 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
653 #endif
654 
655 	off0 = *offp;
656 	m = *mp;
657 	*mp = NULL;
658 	to.to_flags = 0;
659 	TCPSTAT_INC(tcps_rcvtotal);
660 
661 #ifdef INET6
662 	if (isipv6) {
663 		ip6 = mtod(m, struct ip6_hdr *);
664 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
665 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
666 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
667 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
668 				th->th_sum = m->m_pkthdr.csum_data;
669 			else
670 				th->th_sum = in6_cksum_pseudo(ip6, tlen,
671 				    IPPROTO_TCP, m->m_pkthdr.csum_data);
672 			th->th_sum ^= 0xffff;
673 		} else
674 			th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
675 		if (th->th_sum) {
676 			TCPSTAT_INC(tcps_rcvbadsum);
677 			goto drop;
678 		}
679 
680 		/*
681 		 * Be proactive about unspecified IPv6 address in source.
682 		 * As we use all-zero to indicate unbounded/unconnected pcb,
683 		 * unspecified IPv6 address can be used to confuse us.
684 		 *
685 		 * Note that packets with unspecified IPv6 destination is
686 		 * already dropped in ip6_input.
687 		 */
688 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
689 			/* XXX stat */
690 			goto drop;
691 		}
692 		iptos = IPV6_TRAFFIC_CLASS(ip6);
693 	}
694 #endif
695 #if defined(INET) && defined(INET6)
696 	else
697 #endif
698 #ifdef INET
699 	{
700 		/*
701 		 * Get IP and TCP header together in first mbuf.
702 		 * Note: IP leaves IP header in first mbuf.
703 		 */
704 		if (off0 > sizeof (struct ip)) {
705 			ip_stripoptions(m);
706 			off0 = sizeof(struct ip);
707 		}
708 		if (m->m_len < sizeof (struct tcpiphdr)) {
709 			if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
710 			    == NULL) {
711 				TCPSTAT_INC(tcps_rcvshort);
712 				return (IPPROTO_DONE);
713 			}
714 		}
715 		ip = mtod(m, struct ip *);
716 		th = (struct tcphdr *)((caddr_t)ip + off0);
717 		tlen = ntohs(ip->ip_len) - off0;
718 
719 		iptos = ip->ip_tos;
720 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
721 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
722 				th->th_sum = m->m_pkthdr.csum_data;
723 			else
724 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
725 				    ip->ip_dst.s_addr,
726 				    htonl(m->m_pkthdr.csum_data + tlen +
727 				    IPPROTO_TCP));
728 			th->th_sum ^= 0xffff;
729 		} else {
730 			struct ipovly *ipov = (struct ipovly *)ip;
731 
732 			/*
733 			 * Checksum extended TCP header and data.
734 			 */
735 			len = off0 + tlen;
736 			ipttl = ip->ip_ttl;
737 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
738 			ipov->ih_len = htons(tlen);
739 			th->th_sum = in_cksum(m, len);
740 			/* Reset length for SDT probes. */
741 			ip->ip_len = htons(len);
742 			/* Reset TOS bits */
743 			ip->ip_tos = iptos;
744 			/* Re-initialization for later version check */
745 			ip->ip_ttl = ipttl;
746 			ip->ip_v = IPVERSION;
747 			ip->ip_hl = off0 >> 2;
748 		}
749 
750 		if (th->th_sum) {
751 			TCPSTAT_INC(tcps_rcvbadsum);
752 			goto drop;
753 		}
754 	}
755 #endif /* INET */
756 
757 	/*
758 	 * Check that TCP offset makes sense,
759 	 * pull out TCP options and adjust length.		XXX
760 	 */
761 	off = th->th_off << 2;
762 	if (off < sizeof (struct tcphdr) || off > tlen) {
763 		TCPSTAT_INC(tcps_rcvbadoff);
764 		goto drop;
765 	}
766 	tlen -= off;	/* tlen is used instead of ti->ti_len */
767 	if (off > sizeof (struct tcphdr)) {
768 #ifdef INET6
769 		if (isipv6) {
770 			if (m->m_len < off0 + off) {
771 				m = m_pullup(m, off0 + off);
772 				if (m == NULL) {
773 					TCPSTAT_INC(tcps_rcvshort);
774 					return (IPPROTO_DONE);
775 				}
776 			}
777 			ip6 = mtod(m, struct ip6_hdr *);
778 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
779 		}
780 #endif
781 #if defined(INET) && defined(INET6)
782 		else
783 #endif
784 #ifdef INET
785 		{
786 			if (m->m_len < sizeof(struct ip) + off) {
787 				if ((m = m_pullup(m, sizeof (struct ip) + off))
788 				    == NULL) {
789 					TCPSTAT_INC(tcps_rcvshort);
790 					return (IPPROTO_DONE);
791 				}
792 				ip = mtod(m, struct ip *);
793 				th = (struct tcphdr *)((caddr_t)ip + off0);
794 			}
795 		}
796 #endif
797 		optlen = off - sizeof (struct tcphdr);
798 		optp = (u_char *)(th + 1);
799 	}
800 	thflags = th->th_flags;
801 
802 	/*
803 	 * Convert TCP protocol specific fields to host format.
804 	 */
805 	tcp_fields_to_host(th);
806 
807 	/*
808 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
809 	 */
810 	drop_hdrlen = off0 + off;
811 
812 	/*
813 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
814 	 */
815         if (
816 #ifdef INET6
817 	    (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
818 #ifdef INET
819 	    || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
820 #endif
821 #endif
822 #if defined(INET) && !defined(INET6)
823 	    (m->m_flags & M_IP_NEXTHOP)
824 #endif
825 	    )
826 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
827 
828 findpcb:
829 #ifdef INET6
830 	if (isipv6 && fwd_tag != NULL) {
831 		struct sockaddr_in6 *next_hop6;
832 
833 		next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
834 		/*
835 		 * Transparently forwarded. Pretend to be the destination.
836 		 * Already got one like this?
837 		 */
838 		inp = in6_pcblookup_mbuf(&V_tcbinfo,
839 		    &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
840 		    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
841 		if (!inp) {
842 			/*
843 			 * It's new.  Try to find the ambushing socket.
844 			 * Because we've rewritten the destination address,
845 			 * any hardware-generated hash is ignored.
846 			 */
847 			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
848 			    th->th_sport, &next_hop6->sin6_addr,
849 			    next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
850 			    th->th_dport, INPLOOKUP_WILDCARD |
851 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
852 		}
853 	} else if (isipv6) {
854 		inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
855 		    th->th_sport, &ip6->ip6_dst, th->th_dport,
856 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
857 		    m->m_pkthdr.rcvif, m);
858 	}
859 #endif /* INET6 */
860 #if defined(INET6) && defined(INET)
861 	else
862 #endif
863 #ifdef INET
864 	if (fwd_tag != NULL) {
865 		struct sockaddr_in *next_hop;
866 
867 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
868 		/*
869 		 * Transparently forwarded. Pretend to be the destination.
870 		 * already got one like this?
871 		 */
872 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
873 		    ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
874 		    m->m_pkthdr.rcvif, m);
875 		if (!inp) {
876 			/*
877 			 * It's new.  Try to find the ambushing socket.
878 			 * Because we've rewritten the destination address,
879 			 * any hardware-generated hash is ignored.
880 			 */
881 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
882 			    th->th_sport, next_hop->sin_addr,
883 			    next_hop->sin_port ? ntohs(next_hop->sin_port) :
884 			    th->th_dport, INPLOOKUP_WILDCARD |
885 			    INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
886 		}
887 	} else
888 		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
889 		    th->th_sport, ip->ip_dst, th->th_dport,
890 		    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
891 		    m->m_pkthdr.rcvif, m);
892 #endif /* INET */
893 
894 	/*
895 	 * If the INPCB does not exist then all data in the incoming
896 	 * segment is discarded and an appropriate RST is sent back.
897 	 * XXX MRT Send RST using which routing table?
898 	 */
899 	if (inp == NULL) {
900 		/*
901 		 * Log communication attempts to ports that are not
902 		 * in use.
903 		 */
904 		if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
905 		    V_tcp_log_in_vain == 2) {
906 			if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
907 				log(LOG_INFO, "%s; %s: Connection attempt "
908 				    "to closed port\n", s, __func__);
909 		}
910 		/*
911 		 * When blackholing do not respond with a RST but
912 		 * completely ignore the segment and drop it.
913 		 */
914 		if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
915 		    V_blackhole == 2)
916 			goto dropunlock;
917 
918 		rstreason = BANDLIM_RST_CLOSEDPORT;
919 		goto dropwithreset;
920 	}
921 	INP_WLOCK_ASSERT(inp);
922 	/*
923 	 * While waiting for inp lock during the lookup, another thread
924 	 * can have dropped the inpcb, in which case we need to loop back
925 	 * and try to find a new inpcb to deliver to.
926 	 */
927 	if (inp->inp_flags & INP_DROPPED) {
928 		INP_WUNLOCK(inp);
929 		inp = NULL;
930 		goto findpcb;
931 	}
932 	if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
933 	    (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
934 	    ((inp->inp_socket == NULL) ||
935 	    (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
936 		inp->inp_flowid = m->m_pkthdr.flowid;
937 		inp->inp_flowtype = M_HASHTYPE_GET(m);
938 	}
939 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
940 #ifdef INET6
941 	if (isipv6 && IPSEC_ENABLED(ipv6) &&
942 	    IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
943 		goto dropunlock;
944 	}
945 #ifdef INET
946 	else
947 #endif
948 #endif /* INET6 */
949 #ifdef INET
950 	if (IPSEC_ENABLED(ipv4) &&
951 	    IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
952 		goto dropunlock;
953 	}
954 #endif /* INET */
955 #endif /* IPSEC */
956 
957 	/*
958 	 * Check the minimum TTL for socket.
959 	 */
960 	if (inp->inp_ip_minttl != 0) {
961 #ifdef INET6
962 		if (isipv6) {
963 			if (inp->inp_ip_minttl > ip6->ip6_hlim)
964 				goto dropunlock;
965 		} else
966 #endif
967 		if (inp->inp_ip_minttl > ip->ip_ttl)
968 			goto dropunlock;
969 	}
970 
971 	/*
972 	 * A previous connection in TIMEWAIT state is supposed to catch stray
973 	 * or duplicate segments arriving late.  If this segment was a
974 	 * legitimate new connection attempt, the old INPCB gets removed and
975 	 * we can try again to find a listening socket.
976 	 *
977 	 * At this point, due to earlier optimism, we may hold only an inpcb
978 	 * lock, and not the inpcbinfo write lock.  If so, we need to try to
979 	 * acquire it, or if that fails, acquire a reference on the inpcb,
980 	 * drop all locks, acquire a global write lock, and then re-acquire
981 	 * the inpcb lock.  We may at that point discover that another thread
982 	 * has tried to free the inpcb, in which case we need to loop back
983 	 * and try to find a new inpcb to deliver to.
984 	 *
985 	 * XXXRW: It may be time to rethink timewait locking.
986 	 */
987 	if (inp->inp_flags & INP_TIMEWAIT) {
988 		tcp_dooptions(&to, optp, optlen,
989 		    (thflags & TH_SYN) ? TO_SYN : 0);
990 		/*
991 		 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
992 		 */
993 		if (tcp_twcheck(inp, &to, th, m, tlen))
994 			goto findpcb;
995 		return (IPPROTO_DONE);
996 	}
997 	/*
998 	 * The TCPCB may no longer exist if the connection is winding
999 	 * down or it is in the CLOSED state.  Either way we drop the
1000 	 * segment and send an appropriate response.
1001 	 */
1002 	tp = intotcpcb(inp);
1003 	if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1004 		rstreason = BANDLIM_RST_CLOSEDPORT;
1005 		goto dropwithreset;
1006 	}
1007 
1008 #ifdef TCP_OFFLOAD
1009 	if (tp->t_flags & TF_TOE) {
1010 		tcp_offload_input(tp, m);
1011 		m = NULL;	/* consumed by the TOE driver */
1012 		goto dropunlock;
1013 	}
1014 #endif
1015 
1016 #ifdef MAC
1017 	INP_WLOCK_ASSERT(inp);
1018 	if (mac_inpcb_check_deliver(inp, m))
1019 		goto dropunlock;
1020 #endif
1021 	so = inp->inp_socket;
1022 	KASSERT(so != NULL, ("%s: so == NULL", __func__));
1023 #ifdef TCPDEBUG
1024 	if (so->so_options & SO_DEBUG) {
1025 		ostate = tp->t_state;
1026 #ifdef INET6
1027 		if (isipv6) {
1028 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1029 		} else
1030 #endif
1031 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1032 		tcp_savetcp = *th;
1033 	}
1034 #endif /* TCPDEBUG */
1035 	/*
1036 	 * When the socket is accepting connections (the INPCB is in LISTEN
1037 	 * state) we look into the SYN cache if this is a new connection
1038 	 * attempt or the completion of a previous one.
1039 	 */
1040 	KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1041 	    ("%s: so accepting but tp %p not listening", __func__, tp));
1042 	if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1043 		struct in_conninfo inc;
1044 
1045 		bzero(&inc, sizeof(inc));
1046 #ifdef INET6
1047 		if (isipv6) {
1048 			inc.inc_flags |= INC_ISIPV6;
1049 			if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1050 				inc.inc_flags |= INC_IPV6MINMTU;
1051 			inc.inc6_faddr = ip6->ip6_src;
1052 			inc.inc6_laddr = ip6->ip6_dst;
1053 		} else
1054 #endif
1055 		{
1056 			inc.inc_faddr = ip->ip_src;
1057 			inc.inc_laddr = ip->ip_dst;
1058 		}
1059 		inc.inc_fport = th->th_sport;
1060 		inc.inc_lport = th->th_dport;
1061 		inc.inc_fibnum = so->so_fibnum;
1062 
1063 		/*
1064 		 * Check for an existing connection attempt in syncache if
1065 		 * the flag is only ACK.  A successful lookup creates a new
1066 		 * socket appended to the listen queue in SYN_RECEIVED state.
1067 		 */
1068 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1069 			/*
1070 			 * Parse the TCP options here because
1071 			 * syncookies need access to the reflected
1072 			 * timestamp.
1073 			 */
1074 			tcp_dooptions(&to, optp, optlen, 0);
1075 			/*
1076 			 * NB: syncache_expand() doesn't unlock
1077 			 * inp and tcpinfo locks.
1078 			 */
1079 			rstreason = syncache_expand(&inc, &to, th, &so, m);
1080 			if (rstreason < 0) {
1081 				/*
1082 				 * A failing TCP MD5 signature comparison
1083 				 * must result in the segment being dropped
1084 				 * and must not produce any response back
1085 				 * to the sender.
1086 				 */
1087 				goto dropunlock;
1088 			} else if (rstreason == 0) {
1089 				/*
1090 				 * No syncache entry or ACK was not
1091 				 * for our SYN/ACK.  Send a RST.
1092 				 * NB: syncache did its own logging
1093 				 * of the failure cause.
1094 				 */
1095 				rstreason = BANDLIM_RST_OPENPORT;
1096 				goto dropwithreset;
1097 			}
1098 tfo_socket_result:
1099 			if (so == NULL) {
1100 				/*
1101 				 * We completed the 3-way handshake
1102 				 * but could not allocate a socket
1103 				 * either due to memory shortage,
1104 				 * listen queue length limits or
1105 				 * global socket limits.  Send RST
1106 				 * or wait and have the remote end
1107 				 * retransmit the ACK for another
1108 				 * try.
1109 				 */
1110 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1111 					log(LOG_DEBUG, "%s; %s: Listen socket: "
1112 					    "Socket allocation failed due to "
1113 					    "limits or memory shortage, %s\n",
1114 					    s, __func__,
1115 					    V_tcp_sc_rst_sock_fail ?
1116 					    "sending RST" : "try again");
1117 				if (V_tcp_sc_rst_sock_fail) {
1118 					rstreason = BANDLIM_UNLIMITED;
1119 					goto dropwithreset;
1120 				} else
1121 					goto dropunlock;
1122 			}
1123 			/*
1124 			 * Socket is created in state SYN_RECEIVED.
1125 			 * Unlock the listen socket, lock the newly
1126 			 * created socket and update the tp variable.
1127 			 */
1128 			INP_WUNLOCK(inp);	/* listen socket */
1129 			inp = sotoinpcb(so);
1130 			/*
1131 			 * New connection inpcb is already locked by
1132 			 * syncache_expand().
1133 			 */
1134 			INP_WLOCK_ASSERT(inp);
1135 			tp = intotcpcb(inp);
1136 			KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1137 			    ("%s: ", __func__));
1138 			/*
1139 			 * Process the segment and the data it
1140 			 * contains.  tcp_do_segment() consumes
1141 			 * the mbuf chain and unlocks the inpcb.
1142 			 */
1143 			TCP_PROBE5(receive, NULL, tp, m, tp, th);
1144 			tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1145 			    iptos);
1146 			return (IPPROTO_DONE);
1147 		}
1148 		/*
1149 		 * Segment flag validation for new connection attempts:
1150 		 *
1151 		 * Our (SYN|ACK) response was rejected.
1152 		 * Check with syncache and remove entry to prevent
1153 		 * retransmits.
1154 		 *
1155 		 * NB: syncache_chkrst does its own logging of failure
1156 		 * causes.
1157 		 */
1158 		if (thflags & TH_RST) {
1159 			syncache_chkrst(&inc, th, m);
1160 			goto dropunlock;
1161 		}
1162 		/*
1163 		 * We can't do anything without SYN.
1164 		 */
1165 		if ((thflags & TH_SYN) == 0) {
1166 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1167 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1168 				    "SYN is missing, segment ignored\n",
1169 				    s, __func__);
1170 			TCPSTAT_INC(tcps_badsyn);
1171 			goto dropunlock;
1172 		}
1173 		/*
1174 		 * (SYN|ACK) is bogus on a listen socket.
1175 		 */
1176 		if (thflags & TH_ACK) {
1177 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1178 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1179 				    "SYN|ACK invalid, segment rejected\n",
1180 				    s, __func__);
1181 			syncache_badack(&inc);	/* XXX: Not needed! */
1182 			TCPSTAT_INC(tcps_badsyn);
1183 			rstreason = BANDLIM_RST_OPENPORT;
1184 			goto dropwithreset;
1185 		}
1186 		/*
1187 		 * If the drop_synfin option is enabled, drop all
1188 		 * segments with both the SYN and FIN bits set.
1189 		 * This prevents e.g. nmap from identifying the
1190 		 * TCP/IP stack.
1191 		 * XXX: Poor reasoning.  nmap has other methods
1192 		 * and is constantly refining its stack detection
1193 		 * strategies.
1194 		 * XXX: This is a violation of the TCP specification
1195 		 * and was used by RFC1644.
1196 		 */
1197 		if ((thflags & TH_FIN) && V_drop_synfin) {
1198 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1199 				log(LOG_DEBUG, "%s; %s: Listen socket: "
1200 				    "SYN|FIN segment ignored (based on "
1201 				    "sysctl setting)\n", s, __func__);
1202 			TCPSTAT_INC(tcps_badsyn);
1203 			goto dropunlock;
1204 		}
1205 		/*
1206 		 * Segment's flags are (SYN) or (SYN|FIN).
1207 		 *
1208 		 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1209 		 * as they do not affect the state of the TCP FSM.
1210 		 * The data pointed to by TH_URG and th_urp is ignored.
1211 		 */
1212 		KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1213 		    ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1214 		KASSERT(thflags & (TH_SYN),
1215 		    ("%s: Listen socket: TH_SYN not set", __func__));
1216 #ifdef INET6
1217 		/*
1218 		 * If deprecated address is forbidden,
1219 		 * we do not accept SYN to deprecated interface
1220 		 * address to prevent any new inbound connection from
1221 		 * getting established.
1222 		 * When we do not accept SYN, we send a TCP RST,
1223 		 * with deprecated source address (instead of dropping
1224 		 * it).  We compromise it as it is much better for peer
1225 		 * to send a RST, and RST will be the final packet
1226 		 * for the exchange.
1227 		 *
1228 		 * If we do not forbid deprecated addresses, we accept
1229 		 * the SYN packet.  RFC2462 does not suggest dropping
1230 		 * SYN in this case.
1231 		 * If we decipher RFC2462 5.5.4, it says like this:
1232 		 * 1. use of deprecated addr with existing
1233 		 *    communication is okay - "SHOULD continue to be
1234 		 *    used"
1235 		 * 2. use of it with new communication:
1236 		 *   (2a) "SHOULD NOT be used if alternate address
1237 		 *        with sufficient scope is available"
1238 		 *   (2b) nothing mentioned otherwise.
1239 		 * Here we fall into (2b) case as we have no choice in
1240 		 * our source address selection - we must obey the peer.
1241 		 *
1242 		 * The wording in RFC2462 is confusing, and there are
1243 		 * multiple description text for deprecated address
1244 		 * handling - worse, they are not exactly the same.
1245 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1246 		 */
1247 		if (isipv6 && !V_ip6_use_deprecated) {
1248 			struct in6_ifaddr *ia6;
1249 
1250 			ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1251 			if (ia6 != NULL &&
1252 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1253 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1254 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1255 					"Connection attempt to deprecated "
1256 					"IPv6 address rejected\n",
1257 					s, __func__);
1258 				rstreason = BANDLIM_RST_OPENPORT;
1259 				goto dropwithreset;
1260 			}
1261 		}
1262 #endif /* INET6 */
1263 		/*
1264 		 * Basic sanity checks on incoming SYN requests:
1265 		 *   Don't respond if the destination is a link layer
1266 		 *	broadcast according to RFC1122 4.2.3.10, p. 104.
1267 		 *   If it is from this socket it must be forged.
1268 		 *   Don't respond if the source or destination is a
1269 		 *	global or subnet broad- or multicast address.
1270 		 *   Note that it is quite possible to receive unicast
1271 		 *	link-layer packets with a broadcast IP address. Use
1272 		 *	in_broadcast() to find them.
1273 		 */
1274 		if (m->m_flags & (M_BCAST|M_MCAST)) {
1275 			if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1276 			    log(LOG_DEBUG, "%s; %s: Listen socket: "
1277 				"Connection attempt from broad- or multicast "
1278 				"link layer address ignored\n", s, __func__);
1279 			goto dropunlock;
1280 		}
1281 #ifdef INET6
1282 		if (isipv6) {
1283 			if (th->th_dport == th->th_sport &&
1284 			    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1285 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1286 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1287 					"Connection attempt to/from self "
1288 					"ignored\n", s, __func__);
1289 				goto dropunlock;
1290 			}
1291 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1292 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1293 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 					"Connection attempt from/to multicast "
1296 					"address ignored\n", s, __func__);
1297 				goto dropunlock;
1298 			}
1299 		}
1300 #endif
1301 #if defined(INET) && defined(INET6)
1302 		else
1303 #endif
1304 #ifdef INET
1305 		{
1306 			if (th->th_dport == th->th_sport &&
1307 			    ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1308 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 					"Connection attempt from/to self "
1311 					"ignored\n", s, __func__);
1312 				goto dropunlock;
1313 			}
1314 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1315 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1316 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1317 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1318 				if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1319 				    log(LOG_DEBUG, "%s; %s: Listen socket: "
1320 					"Connection attempt from/to broad- "
1321 					"or multicast address ignored\n",
1322 					s, __func__);
1323 				goto dropunlock;
1324 			}
1325 		}
1326 #endif
1327 		/*
1328 		 * SYN appears to be valid.  Create compressed TCP state
1329 		 * for syncache.
1330 		 */
1331 #ifdef TCPDEBUG
1332 		if (so->so_options & SO_DEBUG)
1333 			tcp_trace(TA_INPUT, ostate, tp,
1334 			    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1335 #endif
1336 		TCP_PROBE3(debug__input, tp, th, m);
1337 		tcp_dooptions(&to, optp, optlen, TO_SYN);
1338 		if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1339 			goto tfo_socket_result;
1340 
1341 		/*
1342 		 * Entry added to syncache and mbuf consumed.
1343 		 * Only the listen socket is unlocked by syncache_add().
1344 		 */
1345 		INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1346 		return (IPPROTO_DONE);
1347 	} else if (tp->t_state == TCPS_LISTEN) {
1348 		/*
1349 		 * When a listen socket is torn down the SO_ACCEPTCONN
1350 		 * flag is removed first while connections are drained
1351 		 * from the accept queue in a unlock/lock cycle of the
1352 		 * ACCEPT_LOCK, opening a race condition allowing a SYN
1353 		 * attempt go through unhandled.
1354 		 */
1355 		goto dropunlock;
1356 	}
1357 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1358 	if (tp->t_flags & TF_SIGNATURE) {
1359 		tcp_dooptions(&to, optp, optlen, thflags);
1360 		if ((to.to_flags & TOF_SIGNATURE) == 0) {
1361 			TCPSTAT_INC(tcps_sig_err_nosigopt);
1362 			goto dropunlock;
1363 		}
1364 		if (!TCPMD5_ENABLED() ||
1365 		    TCPMD5_INPUT(m, th, to.to_signature) != 0)
1366 			goto dropunlock;
1367 	}
1368 #endif
1369 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1370 
1371 	/*
1372 	 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1373 	 * state.  tcp_do_segment() always consumes the mbuf chain, unlocks
1374 	 * the inpcb, and unlocks pcbinfo.
1375 	 */
1376 	tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1377 	return (IPPROTO_DONE);
1378 
1379 dropwithreset:
1380 	TCP_PROBE5(receive, NULL, tp, m, tp, th);
1381 
1382 	if (inp != NULL) {
1383 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
1384 		INP_WUNLOCK(inp);
1385 	} else
1386 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1387 	m = NULL;	/* mbuf chain got consumed. */
1388 	goto drop;
1389 
1390 dropunlock:
1391 	if (m != NULL)
1392 		TCP_PROBE5(receive, NULL, tp, m, tp, th);
1393 
1394 	if (inp != NULL)
1395 		INP_WUNLOCK(inp);
1396 
1397 drop:
1398 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1399 	if (s != NULL)
1400 		free(s, M_TCPLOG);
1401 	if (m != NULL)
1402 		m_freem(m);
1403 	return (IPPROTO_DONE);
1404 }
1405 
1406 /*
1407  * Automatic sizing of receive socket buffer.  Often the send
1408  * buffer size is not optimally adjusted to the actual network
1409  * conditions at hand (delay bandwidth product).  Setting the
1410  * buffer size too small limits throughput on links with high
1411  * bandwidth and high delay (eg. trans-continental/oceanic links).
1412  *
1413  * On the receive side the socket buffer memory is only rarely
1414  * used to any significant extent.  This allows us to be much
1415  * more aggressive in scaling the receive socket buffer.  For
1416  * the case that the buffer space is actually used to a large
1417  * extent and we run out of kernel memory we can simply drop
1418  * the new segments; TCP on the sender will just retransmit it
1419  * later.  Setting the buffer size too big may only consume too
1420  * much kernel memory if the application doesn't read() from
1421  * the socket or packet loss or reordering makes use of the
1422  * reassembly queue.
1423  *
1424  * The criteria to step up the receive buffer one notch are:
1425  *  1. Application has not set receive buffer size with
1426  *     SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1427  *  2. the number of bytes received during 1/2 of an sRTT
1428  *     is at least 3/8 of the current socket buffer size.
1429  *  3. receive buffer size has not hit maximal automatic size;
1430  *
1431  * If all of the criteria are met we increaset the socket buffer
1432  * by a 1/2 (bounded by the max). This allows us to keep ahead
1433  * of slow-start but also makes it so our peer never gets limited
1434  * by our rwnd which we then open up causing a burst.
1435  *
1436  * This algorithm does two steps per RTT at most and only if
1437  * we receive a bulk stream w/o packet losses or reorderings.
1438  * Shrinking the buffer during idle times is not necessary as
1439  * it doesn't consume any memory when idle.
1440  *
1441  * TODO: Only step up if the application is actually serving
1442  * the buffer to better manage the socket buffer resources.
1443  */
1444 int
1445 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1446     struct tcpcb *tp, int tlen)
1447 {
1448 	int newsize = 0;
1449 
1450 	if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1451 	    tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1452 	    TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1453 	    ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1454 		if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1455 		    so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1456 			newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1457 		}
1458 		TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1459 
1460 		/* Start over with next RTT. */
1461 		tp->rfbuf_ts = 0;
1462 		tp->rfbuf_cnt = 0;
1463 	} else {
1464 		tp->rfbuf_cnt += tlen;	/* add up */
1465 	}
1466 	return (newsize);
1467 }
1468 
1469 void
1470 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1471 {
1472 	/*
1473 	 * Since tp might be gone if the session entered
1474 	 * the TIME_WAIT state before coming here, we need
1475 	 * to check if the socket is still connected.
1476 	 */
1477 	if ((so->so_state & SS_ISCONNECTED) == 0)
1478 		return;
1479 	INP_LOCK_ASSERT(tp->t_inpcb);
1480 	if (tp->t_flags & TF_WAKESOR) {
1481 		tp->t_flags &= ~TF_WAKESOR;
1482 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1483 		sorwakeup(so);
1484 	}
1485 	if (tp->t_flags & TF_WAKESOW) {
1486 		tp->t_flags &= ~TF_WAKESOW;
1487 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1488 		sowwakeup(so);
1489 	}
1490 }
1491 
1492 void
1493 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1494     struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1495 {
1496 	int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1497 	int rstreason, todrop, win, incforsyn = 0;
1498 	uint32_t tiwin;
1499 	uint16_t nsegs;
1500 	char *s;
1501 	struct in_conninfo *inc;
1502 	struct mbuf *mfree;
1503 	struct tcpopt to;
1504 	int tfo_syn;
1505 	u_int maxseg;
1506 
1507 #ifdef TCPDEBUG
1508 	/*
1509 	 * The size of tcp_saveipgen must be the size of the max ip header,
1510 	 * now IPv6.
1511 	 */
1512 	u_char tcp_saveipgen[IP6_HDR_LEN];
1513 	struct tcphdr tcp_savetcp;
1514 	short ostate = 0;
1515 #endif
1516 	thflags = th->th_flags;
1517 	inc = &tp->t_inpcb->inp_inc;
1518 	tp->sackhint.last_sack_ack = 0;
1519 	sack_changed = 0;
1520 	nsegs = max(1, m->m_pkthdr.lro_nsegs);
1521 
1522 	NET_EPOCH_ASSERT();
1523 	INP_WLOCK_ASSERT(tp->t_inpcb);
1524 	KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1525 	    __func__));
1526 	KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1527 	    __func__));
1528 
1529 #ifdef TCPPCAP
1530 	/* Save segment, if requested. */
1531 	tcp_pcap_add(th, m, &(tp->t_inpkts));
1532 #endif
1533 	TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1534 	    tlen, NULL, true);
1535 
1536 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1537 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1538 			log(LOG_DEBUG, "%s; %s: "
1539 			    "SYN|FIN segment ignored (based on "
1540 			    "sysctl setting)\n", s, __func__);
1541 			free(s, M_TCPLOG);
1542 		}
1543 		goto drop;
1544 	}
1545 
1546 	/*
1547 	 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1548 	 * check SEQ.ACK first.
1549 	 */
1550 	if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1551 	    (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1552 		rstreason = BANDLIM_UNLIMITED;
1553 		goto dropwithreset;
1554 	}
1555 
1556 	/*
1557 	 * Segment received on connection.
1558 	 * Reset idle time and keep-alive timer.
1559 	 * XXX: This should be done after segment
1560 	 * validation to ignore broken/spoofed segs.
1561 	 */
1562 	tp->t_rcvtime = ticks;
1563 
1564 	/*
1565 	 * Scale up the window into a 32-bit value.
1566 	 * For the SYN_SENT state the scale is zero.
1567 	 */
1568 	tiwin = th->th_win << tp->snd_scale;
1569 #ifdef STATS
1570 	stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1571 #endif
1572 
1573 	/*
1574 	 * TCP ECN processing.
1575 	 */
1576 	if (tp->t_flags2 & TF2_ECN_PERMIT) {
1577 		if (thflags & TH_CWR) {
1578 			tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1579 			tp->t_flags |= TF_ACKNOW;
1580 		}
1581 		switch (iptos & IPTOS_ECN_MASK) {
1582 		case IPTOS_ECN_CE:
1583 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1584 			TCPSTAT_INC(tcps_ecn_ce);
1585 			break;
1586 		case IPTOS_ECN_ECT0:
1587 			TCPSTAT_INC(tcps_ecn_ect0);
1588 			break;
1589 		case IPTOS_ECN_ECT1:
1590 			TCPSTAT_INC(tcps_ecn_ect1);
1591 			break;
1592 		}
1593 
1594 		/* Process a packet differently from RFC3168. */
1595 		cc_ecnpkt_handler(tp, th, iptos);
1596 
1597 		/* Congestion experienced. */
1598 		if (thflags & TH_ECE) {
1599 			cc_cong_signal(tp, th, CC_ECN);
1600 		}
1601 	}
1602 
1603 	/*
1604 	 * Parse options on any incoming segment.
1605 	 */
1606 	tcp_dooptions(&to, (u_char *)(th + 1),
1607 	    (th->th_off << 2) - sizeof(struct tcphdr),
1608 	    (thflags & TH_SYN) ? TO_SYN : 0);
1609 
1610 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1611 	if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1612 	    (to.to_flags & TOF_SIGNATURE) == 0) {
1613 		TCPSTAT_INC(tcps_sig_err_sigopt);
1614 		/* XXX: should drop? */
1615 	}
1616 #endif
1617 	/*
1618 	 * If echoed timestamp is later than the current time,
1619 	 * fall back to non RFC1323 RTT calculation.  Normalize
1620 	 * timestamp if syncookies were used when this connection
1621 	 * was established.
1622 	 */
1623 	if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1624 		to.to_tsecr -= tp->ts_offset;
1625 		if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1626 			to.to_tsecr = 0;
1627 		else if (tp->t_flags & TF_PREVVALID &&
1628 			 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1629 			cc_cong_signal(tp, th, CC_RTO_ERR);
1630 	}
1631 	/*
1632 	 * Process options only when we get SYN/ACK back. The SYN case
1633 	 * for incoming connections is handled in tcp_syncache.
1634 	 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1635 	 * or <SYN,ACK>) segment itself is never scaled.
1636 	 * XXX this is traditional behavior, may need to be cleaned up.
1637 	 */
1638 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1639 		/* Handle parallel SYN for ECN */
1640 		if (!(thflags & TH_ACK) &&
1641 		    ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1642 		    ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1643 			tp->t_flags2 |= TF2_ECN_PERMIT;
1644 			tp->t_flags2 |= TF2_ECN_SND_ECE;
1645 			TCPSTAT_INC(tcps_ecn_shs);
1646 		}
1647 		if ((to.to_flags & TOF_SCALE) &&
1648 		    (tp->t_flags & TF_REQ_SCALE) &&
1649 		    !(tp->t_flags & TF_NOOPT)) {
1650 			tp->t_flags |= TF_RCVD_SCALE;
1651 			tp->snd_scale = to.to_wscale;
1652 		} else
1653 			tp->t_flags &= ~TF_REQ_SCALE;
1654 		/*
1655 		 * Initial send window.  It will be updated with
1656 		 * the next incoming segment to the scaled value.
1657 		 */
1658 		tp->snd_wnd = th->th_win;
1659 		if ((to.to_flags & TOF_TS) &&
1660 		    (tp->t_flags & TF_REQ_TSTMP) &&
1661 		    !(tp->t_flags & TF_NOOPT)) {
1662 			tp->t_flags |= TF_RCVD_TSTMP;
1663 			tp->ts_recent = to.to_tsval;
1664 			tp->ts_recent_age = tcp_ts_getticks();
1665 		} else
1666 			tp->t_flags &= ~TF_REQ_TSTMP;
1667 		if (to.to_flags & TOF_MSS)
1668 			tcp_mss(tp, to.to_mss);
1669 		if ((tp->t_flags & TF_SACK_PERMIT) &&
1670 		    (!(to.to_flags & TOF_SACKPERM) ||
1671 		    (tp->t_flags & TF_NOOPT)))
1672 			tp->t_flags &= ~TF_SACK_PERMIT;
1673 		if (IS_FASTOPEN(tp->t_flags)) {
1674 			if ((to.to_flags & TOF_FASTOPEN) &&
1675 			    !(tp->t_flags & TF_NOOPT)) {
1676 				uint16_t mss;
1677 
1678 				if (to.to_flags & TOF_MSS)
1679 					mss = to.to_mss;
1680 				else
1681 					if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1682 						mss = TCP6_MSS;
1683 					else
1684 						mss = TCP_MSS;
1685 				tcp_fastopen_update_cache(tp, mss,
1686 				    to.to_tfo_len, to.to_tfo_cookie);
1687 			} else
1688 				tcp_fastopen_disable_path(tp);
1689 		}
1690 	}
1691 
1692 	/*
1693 	 * If timestamps were negotiated during SYN/ACK and a
1694 	 * segment without a timestamp is received, silently drop
1695 	 * the segment, unless it is a RST segment or missing timestamps are
1696 	 * tolerated.
1697 	 * See section 3.2 of RFC 7323.
1698 	 */
1699 	if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1700 		if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1701 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1702 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1703 				    "segment processed normally\n",
1704 				    s, __func__);
1705 				free(s, M_TCPLOG);
1706 			}
1707 		} else {
1708 			if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1709 				log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1710 				    "segment silently dropped\n", s, __func__);
1711 				free(s, M_TCPLOG);
1712 			}
1713 			goto drop;
1714 		}
1715 	}
1716 	/*
1717 	 * If timestamps were not negotiated during SYN/ACK and a
1718 	 * segment with a timestamp is received, ignore the
1719 	 * timestamp and process the packet normally.
1720 	 * See section 3.2 of RFC 7323.
1721 	 */
1722 	if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1723 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1724 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1725 			    "segment processed normally\n", s, __func__);
1726 			free(s, M_TCPLOG);
1727 		}
1728 	}
1729 
1730 	/*
1731 	 * Header prediction: check for the two common cases
1732 	 * of a uni-directional data xfer.  If the packet has
1733 	 * no control flags, is in-sequence, the window didn't
1734 	 * change and we're not retransmitting, it's a
1735 	 * candidate.  If the length is zero and the ack moved
1736 	 * forward, we're the sender side of the xfer.  Just
1737 	 * free the data acked & wake any higher level process
1738 	 * that was blocked waiting for space.  If the length
1739 	 * is non-zero and the ack didn't move, we're the
1740 	 * receiver side.  If we're getting packets in-order
1741 	 * (the reassembly queue is empty), add the data to
1742 	 * the socket buffer and note that we need a delayed ack.
1743 	 * Make sure that the hidden state-flags are also off.
1744 	 * Since we check for TCPS_ESTABLISHED first, it can only
1745 	 * be TH_NEEDSYN.
1746 	 */
1747 	if (tp->t_state == TCPS_ESTABLISHED &&
1748 	    th->th_seq == tp->rcv_nxt &&
1749 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1750 	    tp->snd_nxt == tp->snd_max &&
1751 	    tiwin && tiwin == tp->snd_wnd &&
1752 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1753 	    SEGQ_EMPTY(tp) &&
1754 	    ((to.to_flags & TOF_TS) == 0 ||
1755 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1756 		/*
1757 		 * If last ACK falls within this segment's sequence numbers,
1758 		 * record the timestamp.
1759 		 * NOTE that the test is modified according to the latest
1760 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1761 		 */
1762 		if ((to.to_flags & TOF_TS) != 0 &&
1763 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1764 			tp->ts_recent_age = tcp_ts_getticks();
1765 			tp->ts_recent = to.to_tsval;
1766 		}
1767 
1768 		if (tlen == 0) {
1769 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1770 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1771 			    !IN_RECOVERY(tp->t_flags) &&
1772 			    (to.to_flags & TOF_SACK) == 0 &&
1773 			    TAILQ_EMPTY(&tp->snd_holes)) {
1774 				/*
1775 				 * This is a pure ack for outstanding data.
1776 				 */
1777 				TCPSTAT_INC(tcps_predack);
1778 
1779 				/*
1780 				 * "bad retransmit" recovery without timestamps.
1781 				 */
1782 				if ((to.to_flags & TOF_TS) == 0 &&
1783 				    tp->t_rxtshift == 1 &&
1784 				    tp->t_flags & TF_PREVVALID &&
1785 				    (int)(ticks - tp->t_badrxtwin) < 0) {
1786 					cc_cong_signal(tp, th, CC_RTO_ERR);
1787 				}
1788 
1789 				/*
1790 				 * Recalculate the transmit timer / rtt.
1791 				 *
1792 				 * Some boxes send broken timestamp replies
1793 				 * during the SYN+ACK phase, ignore
1794 				 * timestamps of 0 or we could calculate a
1795 				 * huge RTT and blow up the retransmit timer.
1796 				 */
1797 				if ((to.to_flags & TOF_TS) != 0 &&
1798 				    to.to_tsecr) {
1799 					uint32_t t;
1800 
1801 					t = tcp_ts_getticks() - to.to_tsecr;
1802 					if (!tp->t_rttlow || tp->t_rttlow > t)
1803 						tp->t_rttlow = t;
1804 					tcp_xmit_timer(tp,
1805 					    TCP_TS_TO_TICKS(t) + 1);
1806 				} else if (tp->t_rtttime &&
1807 				    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1808 					if (!tp->t_rttlow ||
1809 					    tp->t_rttlow > ticks - tp->t_rtttime)
1810 						tp->t_rttlow = ticks - tp->t_rtttime;
1811 					tcp_xmit_timer(tp,
1812 							ticks - tp->t_rtttime);
1813 				}
1814 				acked = BYTES_THIS_ACK(tp, th);
1815 
1816 #ifdef TCP_HHOOK
1817 				/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1818 				hhook_run_tcp_est_in(tp, th, &to);
1819 #endif
1820 
1821 				TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1822 				TCPSTAT_ADD(tcps_rcvackbyte, acked);
1823 				sbdrop(&so->so_snd, acked);
1824 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1825 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1826 					tp->snd_recover = th->th_ack - 1;
1827 
1828 				/*
1829 				 * Let the congestion control algorithm update
1830 				 * congestion control related information. This
1831 				 * typically means increasing the congestion
1832 				 * window.
1833 				 */
1834 				cc_ack_received(tp, th, nsegs, CC_ACK);
1835 
1836 				tp->snd_una = th->th_ack;
1837 				/*
1838 				 * Pull snd_wl2 up to prevent seq wrap relative
1839 				 * to th_ack.
1840 				 */
1841 				tp->snd_wl2 = th->th_ack;
1842 				tp->t_dupacks = 0;
1843 				m_freem(m);
1844 
1845 				/*
1846 				 * If all outstanding data are acked, stop
1847 				 * retransmit timer, otherwise restart timer
1848 				 * using current (possibly backed-off) value.
1849 				 * If process is waiting for space,
1850 				 * wakeup/selwakeup/signal.  If data
1851 				 * are ready to send, let tcp_output
1852 				 * decide between more output or persist.
1853 				 */
1854 #ifdef TCPDEBUG
1855 				if (so->so_options & SO_DEBUG)
1856 					tcp_trace(TA_INPUT, ostate, tp,
1857 					    (void *)tcp_saveipgen,
1858 					    &tcp_savetcp, 0);
1859 #endif
1860 				TCP_PROBE3(debug__input, tp, th, m);
1861 				if (tp->snd_una == tp->snd_max)
1862 					tcp_timer_activate(tp, TT_REXMT, 0);
1863 				else if (!tcp_timer_active(tp, TT_PERSIST))
1864 					tcp_timer_activate(tp, TT_REXMT,
1865 						      tp->t_rxtcur);
1866 				tp->t_flags |= TF_WAKESOW;
1867 				if (sbavail(&so->so_snd))
1868 					(void) tp->t_fb->tfb_tcp_output(tp);
1869 				goto check_delack;
1870 			}
1871 		} else if (th->th_ack == tp->snd_una &&
1872 		    tlen <= sbspace(&so->so_rcv)) {
1873 			int newsize = 0;	/* automatic sockbuf scaling */
1874 
1875 			/*
1876 			 * This is a pure, in-sequence data packet with
1877 			 * nothing on the reassembly queue and we have enough
1878 			 * buffer space to take it.
1879 			 */
1880 			/* Clean receiver SACK report if present */
1881 			if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1882 				tcp_clean_sackreport(tp);
1883 			TCPSTAT_INC(tcps_preddat);
1884 			tp->rcv_nxt += tlen;
1885 			if (tlen &&
1886 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1887 			    (tp->t_fbyte_in == 0)) {
1888 				tp->t_fbyte_in = ticks;
1889 				if (tp->t_fbyte_in == 0)
1890 					tp->t_fbyte_in = 1;
1891 				if (tp->t_fbyte_out && tp->t_fbyte_in)
1892 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1893 			}
1894 			/*
1895 			 * Pull snd_wl1 up to prevent seq wrap relative to
1896 			 * th_seq.
1897 			 */
1898 			tp->snd_wl1 = th->th_seq;
1899 			/*
1900 			 * Pull rcv_up up to prevent seq wrap relative to
1901 			 * rcv_nxt.
1902 			 */
1903 			tp->rcv_up = tp->rcv_nxt;
1904 			TCPSTAT_ADD(tcps_rcvpack, nsegs);
1905 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
1906 #ifdef TCPDEBUG
1907 			if (so->so_options & SO_DEBUG)
1908 				tcp_trace(TA_INPUT, ostate, tp,
1909 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1910 #endif
1911 			TCP_PROBE3(debug__input, tp, th, m);
1912 
1913 			newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1914 
1915 			/* Add data to socket buffer. */
1916 			SOCKBUF_LOCK(&so->so_rcv);
1917 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1918 				m_freem(m);
1919 			} else {
1920 				/*
1921 				 * Set new socket buffer size.
1922 				 * Give up when limit is reached.
1923 				 */
1924 				if (newsize)
1925 					if (!sbreserve_locked(&so->so_rcv,
1926 					    newsize, so, NULL))
1927 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1928 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1929 				sbappendstream_locked(&so->so_rcv, m, 0);
1930 			}
1931 			SOCKBUF_UNLOCK(&so->so_rcv);
1932 			tp->t_flags |= TF_WAKESOR;
1933 			if (DELAY_ACK(tp, tlen)) {
1934 				tp->t_flags |= TF_DELACK;
1935 			} else {
1936 				tp->t_flags |= TF_ACKNOW;
1937 				tp->t_fb->tfb_tcp_output(tp);
1938 			}
1939 			goto check_delack;
1940 		}
1941 	}
1942 
1943 	/*
1944 	 * Calculate amount of space in receive window,
1945 	 * and then do TCP input processing.
1946 	 * Receive window is amount of space in rcv queue,
1947 	 * but not less than advertised window.
1948 	 */
1949 	win = sbspace(&so->so_rcv);
1950 	if (win < 0)
1951 		win = 0;
1952 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1953 
1954 	switch (tp->t_state) {
1955 	/*
1956 	 * If the state is SYN_RECEIVED:
1957 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1958 	 */
1959 	case TCPS_SYN_RECEIVED:
1960 		if ((thflags & TH_ACK) &&
1961 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1962 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1963 				rstreason = BANDLIM_RST_OPENPORT;
1964 				goto dropwithreset;
1965 		}
1966 		if (IS_FASTOPEN(tp->t_flags)) {
1967 			/*
1968 			 * When a TFO connection is in SYN_RECEIVED, the
1969 			 * only valid packets are the initial SYN, a
1970 			 * retransmit/copy of the initial SYN (possibly with
1971 			 * a subset of the original data), a valid ACK, a
1972 			 * FIN, or a RST.
1973 			 */
1974 			if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1975 				rstreason = BANDLIM_RST_OPENPORT;
1976 				goto dropwithreset;
1977 			} else if (thflags & TH_SYN) {
1978 				/* non-initial SYN is ignored */
1979 				if ((tcp_timer_active(tp, TT_DELACK) ||
1980 				     tcp_timer_active(tp, TT_REXMT)))
1981 					goto drop;
1982 			} else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1983 				goto drop;
1984 			}
1985 		}
1986 		break;
1987 
1988 	/*
1989 	 * If the state is SYN_SENT:
1990 	 *	if seg contains a RST with valid ACK (SEQ.ACK has already
1991 	 *	    been verified), then drop the connection.
1992 	 *	if seg contains a RST without an ACK, drop the seg.
1993 	 *	if seg does not contain SYN, then drop the seg.
1994 	 * Otherwise this is an acceptable SYN segment
1995 	 *	initialize tp->rcv_nxt and tp->irs
1996 	 *	if seg contains ack then advance tp->snd_una
1997 	 *	if seg contains an ECE and ECN support is enabled, the stream
1998 	 *	    is ECN capable.
1999 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2000 	 *	arrange for segment to be acked (eventually)
2001 	 *	continue processing rest of data/controls, beginning with URG
2002 	 */
2003 	case TCPS_SYN_SENT:
2004 		if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2005 			TCP_PROBE5(connect__refused, NULL, tp,
2006 			    m, tp, th);
2007 			tp = tcp_drop(tp, ECONNREFUSED);
2008 		}
2009 		if (thflags & TH_RST)
2010 			goto drop;
2011 		if (!(thflags & TH_SYN))
2012 			goto drop;
2013 
2014 		tp->irs = th->th_seq;
2015 		tcp_rcvseqinit(tp);
2016 		if (thflags & TH_ACK) {
2017 			int tfo_partial_ack = 0;
2018 
2019 			TCPSTAT_INC(tcps_connects);
2020 			soisconnected(so);
2021 #ifdef MAC
2022 			mac_socketpeer_set_from_mbuf(m, so);
2023 #endif
2024 			/* Do window scaling on this connection? */
2025 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2026 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2027 				tp->rcv_scale = tp->request_r_scale;
2028 			}
2029 			tp->rcv_adv += min(tp->rcv_wnd,
2030 			    TCP_MAXWIN << tp->rcv_scale);
2031 			tp->snd_una++;		/* SYN is acked */
2032 			/*
2033 			 * If not all the data that was sent in the TFO SYN
2034 			 * has been acked, resend the remainder right away.
2035 			 */
2036 			if (IS_FASTOPEN(tp->t_flags) &&
2037 			    (tp->snd_una != tp->snd_max)) {
2038 				tp->snd_nxt = th->th_ack;
2039 				tfo_partial_ack = 1;
2040 			}
2041 			/*
2042 			 * If there's data, delay ACK; if there's also a FIN
2043 			 * ACKNOW will be turned on later.
2044 			 */
2045 			if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2046 				tcp_timer_activate(tp, TT_DELACK,
2047 				    tcp_delacktime);
2048 			else
2049 				tp->t_flags |= TF_ACKNOW;
2050 
2051 			if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2052 			    (V_tcp_do_ecn == 1)) {
2053 				tp->t_flags2 |= TF2_ECN_PERMIT;
2054 				TCPSTAT_INC(tcps_ecn_shs);
2055 			}
2056 
2057 			/*
2058 			 * Received <SYN,ACK> in SYN_SENT[*] state.
2059 			 * Transitions:
2060 			 *	SYN_SENT  --> ESTABLISHED
2061 			 *	SYN_SENT* --> FIN_WAIT_1
2062 			 */
2063 			tp->t_starttime = ticks;
2064 			if (tp->t_flags & TF_NEEDFIN) {
2065 				tcp_state_change(tp, TCPS_FIN_WAIT_1);
2066 				tp->t_flags &= ~TF_NEEDFIN;
2067 				thflags &= ~TH_SYN;
2068 			} else {
2069 				tcp_state_change(tp, TCPS_ESTABLISHED);
2070 				TCP_PROBE5(connect__established, NULL, tp,
2071 				    m, tp, th);
2072 				cc_conn_init(tp);
2073 				tcp_timer_activate(tp, TT_KEEP,
2074 				    TP_KEEPIDLE(tp));
2075 			}
2076 		} else {
2077 			/*
2078 			 * Received initial SYN in SYN-SENT[*] state =>
2079 			 * simultaneous open.
2080 			 * If it succeeds, connection is * half-synchronized.
2081 			 * Otherwise, do 3-way handshake:
2082 			 *        SYN-SENT -> SYN-RECEIVED
2083 			 *        SYN-SENT* -> SYN-RECEIVED*
2084 			 */
2085 			tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2086 			tcp_timer_activate(tp, TT_REXMT, 0);
2087 			tcp_state_change(tp, TCPS_SYN_RECEIVED);
2088 		}
2089 
2090 		INP_WLOCK_ASSERT(tp->t_inpcb);
2091 
2092 		/*
2093 		 * Advance th->th_seq to correspond to first data byte.
2094 		 * If data, trim to stay within window,
2095 		 * dropping FIN if necessary.
2096 		 */
2097 		th->th_seq++;
2098 		if (tlen > tp->rcv_wnd) {
2099 			todrop = tlen - tp->rcv_wnd;
2100 			m_adj(m, -todrop);
2101 			tlen = tp->rcv_wnd;
2102 			thflags &= ~TH_FIN;
2103 			TCPSTAT_INC(tcps_rcvpackafterwin);
2104 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2105 		}
2106 		tp->snd_wl1 = th->th_seq - 1;
2107 		tp->rcv_up = th->th_seq;
2108 		/*
2109 		 * Client side of transaction: already sent SYN and data.
2110 		 * If the remote host used T/TCP to validate the SYN,
2111 		 * our data will be ACK'd; if so, enter normal data segment
2112 		 * processing in the middle of step 5, ack processing.
2113 		 * Otherwise, goto step 6.
2114 		 */
2115 		if (thflags & TH_ACK)
2116 			goto process_ACK;
2117 
2118 		goto step6;
2119 
2120 	/*
2121 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2122 	 *      do normal processing.
2123 	 *
2124 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
2125 	 */
2126 	case TCPS_LAST_ACK:
2127 	case TCPS_CLOSING:
2128 		break;  /* continue normal processing */
2129 	}
2130 
2131 	/*
2132 	 * States other than LISTEN or SYN_SENT.
2133 	 * First check the RST flag and sequence number since reset segments
2134 	 * are exempt from the timestamp and connection count tests.  This
2135 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2136 	 * below which allowed reset segments in half the sequence space
2137 	 * to fall though and be processed (which gives forged reset
2138 	 * segments with a random sequence number a 50 percent chance of
2139 	 * killing a connection).
2140 	 * Then check timestamp, if present.
2141 	 * Then check the connection count, if present.
2142 	 * Then check that at least some bytes of segment are within
2143 	 * receive window.  If segment begins before rcv_nxt,
2144 	 * drop leading data (and SYN); if nothing left, just ack.
2145 	 */
2146 	if (thflags & TH_RST) {
2147 		/*
2148 		 * RFC5961 Section 3.2
2149 		 *
2150 		 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2151 		 * - If RST is in window, we send challenge ACK.
2152 		 *
2153 		 * Note: to take into account delayed ACKs, we should
2154 		 *   test against last_ack_sent instead of rcv_nxt.
2155 		 * Note 2: we handle special case of closed window, not
2156 		 *   covered by the RFC.
2157 		 */
2158 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2159 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2160 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2161 			KASSERT(tp->t_state != TCPS_SYN_SENT,
2162 			    ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2163 			    __func__, th, tp));
2164 
2165 			if (V_tcp_insecure_rst ||
2166 			    tp->last_ack_sent == th->th_seq) {
2167 				TCPSTAT_INC(tcps_drops);
2168 				/* Drop the connection. */
2169 				switch (tp->t_state) {
2170 				case TCPS_SYN_RECEIVED:
2171 					so->so_error = ECONNREFUSED;
2172 					goto close;
2173 				case TCPS_ESTABLISHED:
2174 				case TCPS_FIN_WAIT_1:
2175 				case TCPS_FIN_WAIT_2:
2176 				case TCPS_CLOSE_WAIT:
2177 				case TCPS_CLOSING:
2178 				case TCPS_LAST_ACK:
2179 					so->so_error = ECONNRESET;
2180 				close:
2181 					/* FALLTHROUGH */
2182 				default:
2183 					tp = tcp_close(tp);
2184 				}
2185 			} else {
2186 				TCPSTAT_INC(tcps_badrst);
2187 				/* Send challenge ACK. */
2188 				tcp_respond(tp, mtod(m, void *), th, m,
2189 				    tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2190 				tp->last_ack_sent = tp->rcv_nxt;
2191 				m = NULL;
2192 			}
2193 		}
2194 		goto drop;
2195 	}
2196 
2197 	/*
2198 	 * RFC5961 Section 4.2
2199 	 * Send challenge ACK for any SYN in synchronized state.
2200 	 */
2201 	if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2202 	    tp->t_state != TCPS_SYN_RECEIVED) {
2203 		TCPSTAT_INC(tcps_badsyn);
2204 		if (V_tcp_insecure_syn &&
2205 		    SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2206 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2207 			tp = tcp_drop(tp, ECONNRESET);
2208 			rstreason = BANDLIM_UNLIMITED;
2209 		} else {
2210 			/* Send challenge ACK. */
2211 			tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2212 			    tp->snd_nxt, TH_ACK);
2213 			tp->last_ack_sent = tp->rcv_nxt;
2214 			m = NULL;
2215 		}
2216 		goto drop;
2217 	}
2218 
2219 	/*
2220 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2221 	 * and it's less than ts_recent, drop it.
2222 	 */
2223 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2224 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2225 		/* Check to see if ts_recent is over 24 days old.  */
2226 		if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2227 			/*
2228 			 * Invalidate ts_recent.  If this segment updates
2229 			 * ts_recent, the age will be reset later and ts_recent
2230 			 * will get a valid value.  If it does not, setting
2231 			 * ts_recent to zero will at least satisfy the
2232 			 * requirement that zero be placed in the timestamp
2233 			 * echo reply when ts_recent isn't valid.  The
2234 			 * age isn't reset until we get a valid ts_recent
2235 			 * because we don't want out-of-order segments to be
2236 			 * dropped when ts_recent is old.
2237 			 */
2238 			tp->ts_recent = 0;
2239 		} else {
2240 			TCPSTAT_INC(tcps_rcvduppack);
2241 			TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2242 			TCPSTAT_INC(tcps_pawsdrop);
2243 			if (tlen)
2244 				goto dropafterack;
2245 			goto drop;
2246 		}
2247 	}
2248 
2249 	/*
2250 	 * In the SYN-RECEIVED state, validate that the packet belongs to
2251 	 * this connection before trimming the data to fit the receive
2252 	 * window.  Check the sequence number versus IRS since we know
2253 	 * the sequence numbers haven't wrapped.  This is a partial fix
2254 	 * for the "LAND" DoS attack.
2255 	 */
2256 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2257 		rstreason = BANDLIM_RST_OPENPORT;
2258 		goto dropwithreset;
2259 	}
2260 
2261 	todrop = tp->rcv_nxt - th->th_seq;
2262 	if (todrop > 0) {
2263 		if (thflags & TH_SYN) {
2264 			thflags &= ~TH_SYN;
2265 			th->th_seq++;
2266 			if (th->th_urp > 1)
2267 				th->th_urp--;
2268 			else
2269 				thflags &= ~TH_URG;
2270 			todrop--;
2271 		}
2272 		/*
2273 		 * Following if statement from Stevens, vol. 2, p. 960.
2274 		 */
2275 		if (todrop > tlen
2276 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2277 			/*
2278 			 * Any valid FIN must be to the left of the window.
2279 			 * At this point the FIN must be a duplicate or out
2280 			 * of sequence; drop it.
2281 			 */
2282 			thflags &= ~TH_FIN;
2283 
2284 			/*
2285 			 * Send an ACK to resynchronize and drop any data.
2286 			 * But keep on processing for RST or ACK.
2287 			 */
2288 			tp->t_flags |= TF_ACKNOW;
2289 			todrop = tlen;
2290 			TCPSTAT_INC(tcps_rcvduppack);
2291 			TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2292 		} else {
2293 			TCPSTAT_INC(tcps_rcvpartduppack);
2294 			TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2295 		}
2296 		/*
2297 		 * DSACK - add SACK block for dropped range
2298 		 */
2299 		if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2300 			tcp_update_sack_list(tp, th->th_seq,
2301 			    th->th_seq + todrop);
2302 			/*
2303 			 * ACK now, as the next in-sequence segment
2304 			 * will clear the DSACK block again
2305 			 */
2306 			tp->t_flags |= TF_ACKNOW;
2307 		}
2308 		drop_hdrlen += todrop;	/* drop from the top afterwards */
2309 		th->th_seq += todrop;
2310 		tlen -= todrop;
2311 		if (th->th_urp > todrop)
2312 			th->th_urp -= todrop;
2313 		else {
2314 			thflags &= ~TH_URG;
2315 			th->th_urp = 0;
2316 		}
2317 	}
2318 
2319 	/*
2320 	 * If new data are received on a connection after the
2321 	 * user processes are gone, then RST the other end.
2322 	 */
2323 	if ((so->so_state & SS_NOFDREF) &&
2324 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2325 		if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2326 			log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2327 			    "after socket was closed, "
2328 			    "sending RST and removing tcpcb\n",
2329 			    s, __func__, tcpstates[tp->t_state], tlen);
2330 			free(s, M_TCPLOG);
2331 		}
2332 		tp = tcp_close(tp);
2333 		TCPSTAT_INC(tcps_rcvafterclose);
2334 		rstreason = BANDLIM_UNLIMITED;
2335 		goto dropwithreset;
2336 	}
2337 
2338 	/*
2339 	 * If segment ends after window, drop trailing data
2340 	 * (and PUSH and FIN); if nothing left, just ACK.
2341 	 */
2342 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2343 	if (todrop > 0) {
2344 		TCPSTAT_INC(tcps_rcvpackafterwin);
2345 		if (todrop >= tlen) {
2346 			TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2347 			/*
2348 			 * If window is closed can only take segments at
2349 			 * window edge, and have to drop data and PUSH from
2350 			 * incoming segments.  Continue processing, but
2351 			 * remember to ack.  Otherwise, drop segment
2352 			 * and ack.
2353 			 */
2354 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2355 				tp->t_flags |= TF_ACKNOW;
2356 				TCPSTAT_INC(tcps_rcvwinprobe);
2357 			} else
2358 				goto dropafterack;
2359 		} else
2360 			TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2361 		m_adj(m, -todrop);
2362 		tlen -= todrop;
2363 		thflags &= ~(TH_PUSH|TH_FIN);
2364 	}
2365 
2366 	/*
2367 	 * If last ACK falls within this segment's sequence numbers,
2368 	 * record its timestamp.
2369 	 * NOTE:
2370 	 * 1) That the test incorporates suggestions from the latest
2371 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2372 	 * 2) That updating only on newer timestamps interferes with
2373 	 *    our earlier PAWS tests, so this check should be solely
2374 	 *    predicated on the sequence space of this segment.
2375 	 * 3) That we modify the segment boundary check to be
2376 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2377 	 *    instead of RFC1323's
2378 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2379 	 *    This modified check allows us to overcome RFC1323's
2380 	 *    limitations as described in Stevens TCP/IP Illustrated
2381 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2382 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2383 	 */
2384 	if ((to.to_flags & TOF_TS) != 0 &&
2385 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2386 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2387 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
2388 		tp->ts_recent_age = tcp_ts_getticks();
2389 		tp->ts_recent = to.to_tsval;
2390 	}
2391 
2392 	/*
2393 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
2394 	 * flag is on (half-synchronized state), then queue data for
2395 	 * later processing; else drop segment and return.
2396 	 */
2397 	if ((thflags & TH_ACK) == 0) {
2398 		if (tp->t_state == TCPS_SYN_RECEIVED ||
2399 		    (tp->t_flags & TF_NEEDSYN)) {
2400 			if (tp->t_state == TCPS_SYN_RECEIVED &&
2401 			    IS_FASTOPEN(tp->t_flags)) {
2402 				tp->snd_wnd = tiwin;
2403 				cc_conn_init(tp);
2404 			}
2405 			goto step6;
2406 		} else if (tp->t_flags & TF_ACKNOW)
2407 			goto dropafterack;
2408 		else
2409 			goto drop;
2410 	}
2411 
2412 	/*
2413 	 * Ack processing.
2414 	 */
2415 	switch (tp->t_state) {
2416 	/*
2417 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2418 	 * ESTABLISHED state and continue processing.
2419 	 * The ACK was checked above.
2420 	 */
2421 	case TCPS_SYN_RECEIVED:
2422 
2423 		TCPSTAT_INC(tcps_connects);
2424 		soisconnected(so);
2425 		/* Do window scaling? */
2426 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2427 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2428 			tp->rcv_scale = tp->request_r_scale;
2429 		}
2430 		tp->snd_wnd = tiwin;
2431 		/*
2432 		 * Make transitions:
2433 		 *      SYN-RECEIVED  -> ESTABLISHED
2434 		 *      SYN-RECEIVED* -> FIN-WAIT-1
2435 		 */
2436 		tp->t_starttime = ticks;
2437 		if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2438 			tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2439 			tp->t_tfo_pending = NULL;
2440 		}
2441 		if (tp->t_flags & TF_NEEDFIN) {
2442 			tcp_state_change(tp, TCPS_FIN_WAIT_1);
2443 			tp->t_flags &= ~TF_NEEDFIN;
2444 		} else {
2445 			tcp_state_change(tp, TCPS_ESTABLISHED);
2446 			TCP_PROBE5(accept__established, NULL, tp,
2447 			    m, tp, th);
2448 			/*
2449 			 * TFO connections call cc_conn_init() during SYN
2450 			 * processing.  Calling it again here for such
2451 			 * connections is not harmless as it would undo the
2452 			 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2453 			 * is retransmitted.
2454 			 */
2455 			if (!IS_FASTOPEN(tp->t_flags))
2456 				cc_conn_init(tp);
2457 			tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2458 		}
2459 		/*
2460 		 * Account for the ACK of our SYN prior to
2461 		 * regular ACK processing below, except for
2462 		 * simultaneous SYN, which is handled later.
2463 		 */
2464 		if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2465 			incforsyn = 1;
2466 		/*
2467 		 * If segment contains data or ACK, will call tcp_reass()
2468 		 * later; if not, do so now to pass queued data to user.
2469 		 */
2470 		if (tlen == 0 && (thflags & TH_FIN) == 0)
2471 			(void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2472 			    (struct mbuf *)0);
2473 		tp->snd_wl1 = th->th_seq - 1;
2474 		/* FALLTHROUGH */
2475 
2476 	/*
2477 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2478 	 * ACKs.  If the ack is in the range
2479 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2480 	 * then advance tp->snd_una to th->th_ack and drop
2481 	 * data from the retransmission queue.  If this ACK reflects
2482 	 * more up to date window information we update our window information.
2483 	 */
2484 	case TCPS_ESTABLISHED:
2485 	case TCPS_FIN_WAIT_1:
2486 	case TCPS_FIN_WAIT_2:
2487 	case TCPS_CLOSE_WAIT:
2488 	case TCPS_CLOSING:
2489 	case TCPS_LAST_ACK:
2490 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2491 			TCPSTAT_INC(tcps_rcvacktoomuch);
2492 			goto dropafterack;
2493 		}
2494 		if ((tp->t_flags & TF_SACK_PERMIT) &&
2495 		    ((to.to_flags & TOF_SACK) ||
2496 		     !TAILQ_EMPTY(&tp->snd_holes)))
2497 			sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2498 		else
2499 			/*
2500 			 * Reset the value so that previous (valid) value
2501 			 * from the last ack with SACK doesn't get used.
2502 			 */
2503 			tp->sackhint.sacked_bytes = 0;
2504 
2505 #ifdef TCP_HHOOK
2506 		/* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2507 		hhook_run_tcp_est_in(tp, th, &to);
2508 #endif
2509 
2510 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2511 			maxseg = tcp_maxseg(tp);
2512 			if (tlen == 0 &&
2513 			    (tiwin == tp->snd_wnd ||
2514 			    (tp->t_flags & TF_SACK_PERMIT))) {
2515 				/*
2516 				 * If this is the first time we've seen a
2517 				 * FIN from the remote, this is not a
2518 				 * duplicate and it needs to be processed
2519 				 * normally.  This happens during a
2520 				 * simultaneous close.
2521 				 */
2522 				if ((thflags & TH_FIN) &&
2523 				    (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2524 					tp->t_dupacks = 0;
2525 					break;
2526 				}
2527 				TCPSTAT_INC(tcps_rcvdupack);
2528 				/*
2529 				 * If we have outstanding data (other than
2530 				 * a window probe), this is a completely
2531 				 * duplicate ack (ie, window info didn't
2532 				 * change and FIN isn't set),
2533 				 * the ack is the biggest we've
2534 				 * seen and we've seen exactly our rexmt
2535 				 * threshold of them, assume a packet
2536 				 * has been dropped and retransmit it.
2537 				 * Kludge snd_nxt & the congestion
2538 				 * window so we send only this one
2539 				 * packet.
2540 				 *
2541 				 * We know we're losing at the current
2542 				 * window size so do congestion avoidance
2543 				 * (set ssthresh to half the current window
2544 				 * and pull our congestion window back to
2545 				 * the new ssthresh).
2546 				 *
2547 				 * Dup acks mean that packets have left the
2548 				 * network (they're now cached at the receiver)
2549 				 * so bump cwnd by the amount in the receiver
2550 				 * to keep a constant cwnd packets in the
2551 				 * network.
2552 				 *
2553 				 * When using TCP ECN, notify the peer that
2554 				 * we reduced the cwnd.
2555 				 */
2556 				/*
2557 				 * Following 2 kinds of acks should not affect
2558 				 * dupack counting:
2559 				 * 1) Old acks
2560 				 * 2) Acks with SACK but without any new SACK
2561 				 * information in them. These could result from
2562 				 * any anomaly in the network like a switch
2563 				 * duplicating packets or a possible DoS attack.
2564 				 */
2565 				if (th->th_ack != tp->snd_una ||
2566 				    ((tp->t_flags & TF_SACK_PERMIT) &&
2567 				    !sack_changed))
2568 					break;
2569 				else if (!tcp_timer_active(tp, TT_REXMT))
2570 					tp->t_dupacks = 0;
2571 				else if (++tp->t_dupacks > tcprexmtthresh ||
2572 				     IN_FASTRECOVERY(tp->t_flags)) {
2573 					cc_ack_received(tp, th, nsegs,
2574 					    CC_DUPACK);
2575 					if (V_tcp_do_prr &&
2576 					    IN_FASTRECOVERY(tp->t_flags) &&
2577 					    (tp->t_flags & TF_SACK_PERMIT)) {
2578 						int snd_cnt = 0, limit = 0;
2579 						int del_data = 0, pipe = 0;
2580 						/*
2581 						 * In a duplicate ACK del_data is only the
2582 						 * diff_in_sack. If no SACK is used del_data
2583 						 * will be 0. Pipe is the amount of data we
2584 						 * estimate to be in the network.
2585 						 */
2586 						del_data = tp->sackhint.delivered_data;
2587 						if (V_tcp_do_rfc6675_pipe)
2588 							pipe = tcp_compute_pipe(tp);
2589 						else
2590 							pipe = (tp->snd_nxt - tp->snd_fack) +
2591 								tp->sackhint.sack_bytes_rexmit;
2592 						tp->sackhint.prr_delivered += del_data;
2593 						if (pipe >= tp->snd_ssthresh) {
2594 							if (tp->sackhint.recover_fs == 0)
2595 								tp->sackhint.recover_fs =
2596 								    imax(1, tp->snd_nxt - tp->snd_una);
2597 							snd_cnt = howmany((long)tp->sackhint.prr_delivered *
2598 							    tp->snd_ssthresh, tp->sackhint.recover_fs) -
2599 							    tp->sackhint.prr_out;
2600 						} else {
2601 							if (V_tcp_do_prr_conservative)
2602 								limit = tp->sackhint.prr_delivered -
2603 									tp->sackhint.prr_out;
2604 							else
2605 								limit = imax(tp->sackhint.prr_delivered -
2606 									    tp->sackhint.prr_out,
2607 									    del_data) + maxseg;
2608 							snd_cnt = imin(tp->snd_ssthresh - pipe, limit);
2609 						}
2610 						snd_cnt = imax(snd_cnt, 0) / maxseg;
2611 						/*
2612 						 * Send snd_cnt new data into the network in
2613 						 * response to this ACK. If there is a going
2614 						 * to be a SACK retransmission, adjust snd_cwnd
2615 						 * accordingly.
2616 						 */
2617 						tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
2618 						    tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
2619 					} else if ((tp->t_flags & TF_SACK_PERMIT) &&
2620 					    IN_FASTRECOVERY(tp->t_flags)) {
2621 						int awnd;
2622 
2623 						/*
2624 						 * Compute the amount of data in flight first.
2625 						 * We can inject new data into the pipe iff
2626 						 * we have less than 1/2 the original window's
2627 						 * worth of data in flight.
2628 						 */
2629 						if (V_tcp_do_rfc6675_pipe)
2630 							awnd = tcp_compute_pipe(tp);
2631 						else
2632 							awnd = (tp->snd_nxt - tp->snd_fack) +
2633 								tp->sackhint.sack_bytes_rexmit;
2634 
2635 						if (awnd < tp->snd_ssthresh) {
2636 							tp->snd_cwnd += maxseg;
2637 							if (tp->snd_cwnd > tp->snd_ssthresh)
2638 								tp->snd_cwnd = tp->snd_ssthresh;
2639 						}
2640 					} else
2641 						tp->snd_cwnd += maxseg;
2642 					(void) tp->t_fb->tfb_tcp_output(tp);
2643 					goto drop;
2644 				} else if (tp->t_dupacks == tcprexmtthresh ||
2645 					    (tp->t_flags & TF_SACK_PERMIT &&
2646 					     V_tcp_do_rfc6675_pipe &&
2647 					     tp->sackhint.sacked_bytes >
2648 					     (tcprexmtthresh - 1) * maxseg)) {
2649 enter_recovery:
2650 					/*
2651 					 * Above is the RFC6675 trigger condition of
2652 					 * more than (dupthresh-1)*maxseg sacked data.
2653 					 * If the count of holes in the
2654 					 * scoreboard is >= dupthresh, we could
2655 					 * also enter loss recovery, but don't
2656 					 * have that value readily available.
2657 					 */
2658 					tp->t_dupacks = tcprexmtthresh;
2659 					tcp_seq onxt = tp->snd_nxt;
2660 
2661 					/*
2662 					 * If we're doing sack, or prr, check
2663 					 * to see if we're already in sack
2664 					 * recovery. If we're not doing sack,
2665 					 * check to see if we're in newreno
2666 					 * recovery.
2667 					 */
2668 					if (V_tcp_do_prr ||
2669 					    (tp->t_flags & TF_SACK_PERMIT)) {
2670 						if (IN_FASTRECOVERY(tp->t_flags)) {
2671 							tp->t_dupacks = 0;
2672 							break;
2673 						}
2674 					} else {
2675 						if (SEQ_LEQ(th->th_ack,
2676 						    tp->snd_recover)) {
2677 							tp->t_dupacks = 0;
2678 							break;
2679 						}
2680 					}
2681 					/* Congestion signal before ack. */
2682 					cc_cong_signal(tp, th, CC_NDUPACK);
2683 					cc_ack_received(tp, th, nsegs,
2684 					    CC_DUPACK);
2685 					tcp_timer_activate(tp, TT_REXMT, 0);
2686 					tp->t_rtttime = 0;
2687 					if (V_tcp_do_prr) {
2688 						/*
2689 						 * snd_ssthresh is already updated by
2690 						 * cc_cong_signal.
2691 						 */
2692 						tp->sackhint.prr_delivered =
2693 						    tp->sackhint.sacked_bytes;
2694 						tp->sackhint.recover_fs = max(1,
2695 						    tp->snd_nxt - tp->snd_una);
2696 					}
2697 					if (tp->t_flags & TF_SACK_PERMIT) {
2698 						TCPSTAT_INC(
2699 						    tcps_sack_recovery_episode);
2700 						tp->snd_recover = tp->snd_nxt;
2701 						tp->snd_cwnd = maxseg;
2702 						(void) tp->t_fb->tfb_tcp_output(tp);
2703 						if (SEQ_GT(th->th_ack, tp->snd_una))
2704 							goto resume_partialack;
2705 						goto drop;
2706 					}
2707 					tp->snd_nxt = th->th_ack;
2708 					tp->snd_cwnd = maxseg;
2709 					(void) tp->t_fb->tfb_tcp_output(tp);
2710 					KASSERT(tp->snd_limited <= 2,
2711 					    ("%s: tp->snd_limited too big",
2712 					    __func__));
2713 					tp->snd_cwnd = tp->snd_ssthresh +
2714 					     maxseg *
2715 					     (tp->t_dupacks - tp->snd_limited);
2716 					if (SEQ_GT(onxt, tp->snd_nxt))
2717 						tp->snd_nxt = onxt;
2718 					goto drop;
2719 				} else if (V_tcp_do_rfc3042) {
2720 					/*
2721 					 * Process first and second duplicate
2722 					 * ACKs. Each indicates a segment
2723 					 * leaving the network, creating room
2724 					 * for more. Make sure we can send a
2725 					 * packet on reception of each duplicate
2726 					 * ACK by increasing snd_cwnd by one
2727 					 * segment. Restore the original
2728 					 * snd_cwnd after packet transmission.
2729 					 */
2730 					cc_ack_received(tp, th, nsegs,
2731 					    CC_DUPACK);
2732 					uint32_t oldcwnd = tp->snd_cwnd;
2733 					tcp_seq oldsndmax = tp->snd_max;
2734 					u_int sent;
2735 					int avail;
2736 
2737 					KASSERT(tp->t_dupacks == 1 ||
2738 					    tp->t_dupacks == 2,
2739 					    ("%s: dupacks not 1 or 2",
2740 					    __func__));
2741 					if (tp->t_dupacks == 1)
2742 						tp->snd_limited = 0;
2743 					tp->snd_cwnd =
2744 					    (tp->snd_nxt - tp->snd_una) +
2745 					    (tp->t_dupacks - tp->snd_limited) *
2746 					    maxseg;
2747 					/*
2748 					 * Only call tcp_output when there
2749 					 * is new data available to be sent.
2750 					 * Otherwise we would send pure ACKs.
2751 					 */
2752 					SOCKBUF_LOCK(&so->so_snd);
2753 					avail = sbavail(&so->so_snd) -
2754 					    (tp->snd_nxt - tp->snd_una);
2755 					SOCKBUF_UNLOCK(&so->so_snd);
2756 					if (avail > 0)
2757 						(void) tp->t_fb->tfb_tcp_output(tp);
2758 					sent = tp->snd_max - oldsndmax;
2759 					if (sent > maxseg) {
2760 						KASSERT((tp->t_dupacks == 2 &&
2761 						    tp->snd_limited == 0) ||
2762 						   (sent == maxseg + 1 &&
2763 						    tp->t_flags & TF_SENTFIN),
2764 						    ("%s: sent too much",
2765 						    __func__));
2766 						tp->snd_limited = 2;
2767 					} else if (sent > 0)
2768 						++tp->snd_limited;
2769 					tp->snd_cwnd = oldcwnd;
2770 					goto drop;
2771 				}
2772 			}
2773 			break;
2774 		} else {
2775 			/*
2776 			 * This ack is advancing the left edge, reset the
2777 			 * counter.
2778 			 */
2779 			tp->t_dupacks = 0;
2780 			/*
2781 			 * If this ack also has new SACK info, increment the
2782 			 * counter as per rfc6675. The variable
2783 			 * sack_changed tracks all changes to the SACK
2784 			 * scoreboard, including when partial ACKs without
2785 			 * SACK options are received, and clear the scoreboard
2786 			 * from the left side. Such partial ACKs should not be
2787 			 * counted as dupacks here.
2788 			 */
2789 			if ((tp->t_flags & TF_SACK_PERMIT) &&
2790 			    (to.to_flags & TOF_SACK) &&
2791 			    sack_changed) {
2792 				tp->t_dupacks++;
2793 				/* limit overhead by setting maxseg last */
2794 				if (!IN_FASTRECOVERY(tp->t_flags) &&
2795 				    (tp->sackhint.sacked_bytes >
2796 				    ((tcprexmtthresh - 1) *
2797 				    (maxseg = tcp_maxseg(tp))))) {
2798 					goto enter_recovery;
2799 				}
2800 			}
2801 		}
2802 
2803 resume_partialack:
2804 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2805 		    ("%s: th_ack <= snd_una", __func__));
2806 
2807 		/*
2808 		 * If the congestion window was inflated to account
2809 		 * for the other side's cached packets, retract it.
2810 		 */
2811 		if (IN_FASTRECOVERY(tp->t_flags)) {
2812 			if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2813 				if (tp->t_flags & TF_SACK_PERMIT)
2814 					if (V_tcp_do_prr && to.to_flags & TOF_SACK)
2815 						tcp_prr_partialack(tp, th);
2816 					else
2817 						tcp_sack_partialack(tp, th);
2818 				else
2819 					tcp_newreno_partial_ack(tp, th);
2820 			} else
2821 				cc_post_recovery(tp, th);
2822 		}
2823 		/*
2824 		 * If we reach this point, ACK is not a duplicate,
2825 		 *     i.e., it ACKs something we sent.
2826 		 */
2827 		if (tp->t_flags & TF_NEEDSYN) {
2828 			/*
2829 			 * T/TCP: Connection was half-synchronized, and our
2830 			 * SYN has been ACK'd (so connection is now fully
2831 			 * synchronized).  Go to non-starred state,
2832 			 * increment snd_una for ACK of SYN, and check if
2833 			 * we can do window scaling.
2834 			 */
2835 			tp->t_flags &= ~TF_NEEDSYN;
2836 			tp->snd_una++;
2837 			/* Do window scaling? */
2838 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2839 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2840 				tp->rcv_scale = tp->request_r_scale;
2841 				/* Send window already scaled. */
2842 			}
2843 		}
2844 
2845 process_ACK:
2846 		INP_WLOCK_ASSERT(tp->t_inpcb);
2847 
2848 		/*
2849 		 * Adjust for the SYN bit in sequence space,
2850 		 * but don't account for it in cwnd calculations.
2851 		 * This is for the SYN_RECEIVED, non-simultaneous
2852 		 * SYN case. SYN_SENT and simultaneous SYN are
2853 		 * treated elsewhere.
2854 		 */
2855 		if (incforsyn)
2856 			tp->snd_una++;
2857 		acked = BYTES_THIS_ACK(tp, th);
2858 		KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2859 		    "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2860 		    tp->snd_una, th->th_ack, tp, m));
2861 		TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2862 		TCPSTAT_ADD(tcps_rcvackbyte, acked);
2863 
2864 		/*
2865 		 * If we just performed our first retransmit, and the ACK
2866 		 * arrives within our recovery window, then it was a mistake
2867 		 * to do the retransmit in the first place.  Recover our
2868 		 * original cwnd and ssthresh, and proceed to transmit where
2869 		 * we left off.
2870 		 */
2871 		if (tp->t_rxtshift == 1 &&
2872 		    tp->t_flags & TF_PREVVALID &&
2873 		    tp->t_badrxtwin &&
2874 		    SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2875 			cc_cong_signal(tp, th, CC_RTO_ERR);
2876 
2877 		/*
2878 		 * If we have a timestamp reply, update smoothed
2879 		 * round trip time.  If no timestamp is present but
2880 		 * transmit timer is running and timed sequence
2881 		 * number was acked, update smoothed round trip time.
2882 		 * Since we now have an rtt measurement, cancel the
2883 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2884 		 * Recompute the initial retransmit timer.
2885 		 *
2886 		 * Some boxes send broken timestamp replies
2887 		 * during the SYN+ACK phase, ignore
2888 		 * timestamps of 0 or we could calculate a
2889 		 * huge RTT and blow up the retransmit timer.
2890 		 */
2891 		if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2892 			uint32_t t;
2893 
2894 			t = tcp_ts_getticks() - to.to_tsecr;
2895 			if (!tp->t_rttlow || tp->t_rttlow > t)
2896 				tp->t_rttlow = t;
2897 			tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2898 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2899 			if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2900 				tp->t_rttlow = ticks - tp->t_rtttime;
2901 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2902 		}
2903 
2904 		/*
2905 		 * If all outstanding data is acked, stop retransmit
2906 		 * timer and remember to restart (more output or persist).
2907 		 * If there is more data to be acked, restart retransmit
2908 		 * timer, using current (possibly backed-off) value.
2909 		 */
2910 		if (th->th_ack == tp->snd_max) {
2911 			tcp_timer_activate(tp, TT_REXMT, 0);
2912 			needoutput = 1;
2913 		} else if (!tcp_timer_active(tp, TT_PERSIST))
2914 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2915 
2916 		/*
2917 		 * If no data (only SYN) was ACK'd,
2918 		 *    skip rest of ACK processing.
2919 		 */
2920 		if (acked == 0)
2921 			goto step6;
2922 
2923 		/*
2924 		 * Let the congestion control algorithm update congestion
2925 		 * control related information. This typically means increasing
2926 		 * the congestion window.
2927 		 */
2928 		cc_ack_received(tp, th, nsegs, CC_ACK);
2929 
2930 		SOCKBUF_LOCK(&so->so_snd);
2931 		if (acked > sbavail(&so->so_snd)) {
2932 			if (tp->snd_wnd >= sbavail(&so->so_snd))
2933 				tp->snd_wnd -= sbavail(&so->so_snd);
2934 			else
2935 				tp->snd_wnd = 0;
2936 			mfree = sbcut_locked(&so->so_snd,
2937 			    (int)sbavail(&so->so_snd));
2938 			ourfinisacked = 1;
2939 		} else {
2940 			mfree = sbcut_locked(&so->so_snd, acked);
2941 			if (tp->snd_wnd >= (uint32_t) acked)
2942 				tp->snd_wnd -= acked;
2943 			else
2944 				tp->snd_wnd = 0;
2945 			ourfinisacked = 0;
2946 		}
2947 		SOCKBUF_UNLOCK(&so->so_snd);
2948 		tp->t_flags |= TF_WAKESOW;
2949 		m_freem(mfree);
2950 		/* Detect una wraparound. */
2951 		if (!IN_RECOVERY(tp->t_flags) &&
2952 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2953 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2954 			tp->snd_recover = th->th_ack - 1;
2955 		/* XXXLAS: Can this be moved up into cc_post_recovery? */
2956 		if (IN_RECOVERY(tp->t_flags) &&
2957 		    SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2958 			EXIT_RECOVERY(tp->t_flags);
2959 		}
2960 		tp->snd_una = th->th_ack;
2961 		if (tp->t_flags & TF_SACK_PERMIT) {
2962 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2963 				tp->snd_recover = tp->snd_una;
2964 		}
2965 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2966 			tp->snd_nxt = tp->snd_una;
2967 
2968 		switch (tp->t_state) {
2969 		/*
2970 		 * In FIN_WAIT_1 STATE in addition to the processing
2971 		 * for the ESTABLISHED state if our FIN is now acknowledged
2972 		 * then enter FIN_WAIT_2.
2973 		 */
2974 		case TCPS_FIN_WAIT_1:
2975 			if (ourfinisacked) {
2976 				/*
2977 				 * If we can't receive any more
2978 				 * data, then closing user can proceed.
2979 				 * Starting the timer is contrary to the
2980 				 * specification, but if we don't get a FIN
2981 				 * we'll hang forever.
2982 				 *
2983 				 * XXXjl:
2984 				 * we should release the tp also, and use a
2985 				 * compressed state.
2986 				 */
2987 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2988 					soisdisconnected(so);
2989 					tcp_timer_activate(tp, TT_2MSL,
2990 					    (tcp_fast_finwait2_recycle ?
2991 					    tcp_finwait2_timeout :
2992 					    TP_MAXIDLE(tp)));
2993 				}
2994 				tcp_state_change(tp, TCPS_FIN_WAIT_2);
2995 			}
2996 			break;
2997 
2998 		/*
2999 		 * In CLOSING STATE in addition to the processing for
3000 		 * the ESTABLISHED state if the ACK acknowledges our FIN
3001 		 * then enter the TIME-WAIT state, otherwise ignore
3002 		 * the segment.
3003 		 */
3004 		case TCPS_CLOSING:
3005 			if (ourfinisacked) {
3006 				tcp_twstart(tp);
3007 				m_freem(m);
3008 				return;
3009 			}
3010 			break;
3011 
3012 		/*
3013 		 * In LAST_ACK, we may still be waiting for data to drain
3014 		 * and/or to be acked, as well as for the ack of our FIN.
3015 		 * If our FIN is now acknowledged, delete the TCB,
3016 		 * enter the closed state and return.
3017 		 */
3018 		case TCPS_LAST_ACK:
3019 			if (ourfinisacked) {
3020 				tp = tcp_close(tp);
3021 				goto drop;
3022 			}
3023 			break;
3024 		}
3025 	}
3026 
3027 step6:
3028 	INP_WLOCK_ASSERT(tp->t_inpcb);
3029 
3030 	/*
3031 	 * Update window information.
3032 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3033 	 */
3034 	if ((thflags & TH_ACK) &&
3035 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3036 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3037 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3038 		/* keep track of pure window updates */
3039 		if (tlen == 0 &&
3040 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3041 			TCPSTAT_INC(tcps_rcvwinupd);
3042 		tp->snd_wnd = tiwin;
3043 		tp->snd_wl1 = th->th_seq;
3044 		tp->snd_wl2 = th->th_ack;
3045 		if (tp->snd_wnd > tp->max_sndwnd)
3046 			tp->max_sndwnd = tp->snd_wnd;
3047 		needoutput = 1;
3048 	}
3049 
3050 	/*
3051 	 * Process segments with URG.
3052 	 */
3053 	if ((thflags & TH_URG) && th->th_urp &&
3054 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3055 		/*
3056 		 * This is a kludge, but if we receive and accept
3057 		 * random urgent pointers, we'll crash in
3058 		 * soreceive.  It's hard to imagine someone
3059 		 * actually wanting to send this much urgent data.
3060 		 */
3061 		SOCKBUF_LOCK(&so->so_rcv);
3062 		if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3063 			th->th_urp = 0;			/* XXX */
3064 			thflags &= ~TH_URG;		/* XXX */
3065 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
3066 			goto dodata;			/* XXX */
3067 		}
3068 		/*
3069 		 * If this segment advances the known urgent pointer,
3070 		 * then mark the data stream.  This should not happen
3071 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3072 		 * a FIN has been received from the remote side.
3073 		 * In these states we ignore the URG.
3074 		 *
3075 		 * According to RFC961 (Assigned Protocols),
3076 		 * the urgent pointer points to the last octet
3077 		 * of urgent data.  We continue, however,
3078 		 * to consider it to indicate the first octet
3079 		 * of data past the urgent section as the original
3080 		 * spec states (in one of two places).
3081 		 */
3082 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3083 			tp->rcv_up = th->th_seq + th->th_urp;
3084 			so->so_oobmark = sbavail(&so->so_rcv) +
3085 			    (tp->rcv_up - tp->rcv_nxt) - 1;
3086 			if (so->so_oobmark == 0)
3087 				so->so_rcv.sb_state |= SBS_RCVATMARK;
3088 			sohasoutofband(so);
3089 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3090 		}
3091 		SOCKBUF_UNLOCK(&so->so_rcv);
3092 		/*
3093 		 * Remove out of band data so doesn't get presented to user.
3094 		 * This can happen independent of advancing the URG pointer,
3095 		 * but if two URG's are pending at once, some out-of-band
3096 		 * data may creep in... ick.
3097 		 */
3098 		if (th->th_urp <= (uint32_t)tlen &&
3099 		    !(so->so_options & SO_OOBINLINE)) {
3100 			/* hdr drop is delayed */
3101 			tcp_pulloutofband(so, th, m, drop_hdrlen);
3102 		}
3103 	} else {
3104 		/*
3105 		 * If no out of band data is expected,
3106 		 * pull receive urgent pointer along
3107 		 * with the receive window.
3108 		 */
3109 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3110 			tp->rcv_up = tp->rcv_nxt;
3111 	}
3112 dodata:							/* XXX */
3113 	INP_WLOCK_ASSERT(tp->t_inpcb);
3114 
3115 	/*
3116 	 * Process the segment text, merging it into the TCP sequencing queue,
3117 	 * and arranging for acknowledgment of receipt if necessary.
3118 	 * This process logically involves adjusting tp->rcv_wnd as data
3119 	 * is presented to the user (this happens in tcp_usrreq.c,
3120 	 * case PRU_RCVD).  If a FIN has already been received on this
3121 	 * connection then we just ignore the text.
3122 	 */
3123 	tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3124 		   IS_FASTOPEN(tp->t_flags));
3125 	if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3126 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3127 		tcp_seq save_start = th->th_seq;
3128 		tcp_seq save_rnxt  = tp->rcv_nxt;
3129 		int     save_tlen  = tlen;
3130 		m_adj(m, drop_hdrlen);	/* delayed header drop */
3131 		/*
3132 		 * Insert segment which includes th into TCP reassembly queue
3133 		 * with control block tp.  Set thflags to whether reassembly now
3134 		 * includes a segment with FIN.  This handles the common case
3135 		 * inline (segment is the next to be received on an established
3136 		 * connection, and the queue is empty), avoiding linkage into
3137 		 * and removal from the queue and repetition of various
3138 		 * conversions.
3139 		 * Set DELACK for segments received in order, but ack
3140 		 * immediately when segments are out of order (so
3141 		 * fast retransmit can work).
3142 		 */
3143 		if (th->th_seq == tp->rcv_nxt &&
3144 		    SEGQ_EMPTY(tp) &&
3145 		    (TCPS_HAVEESTABLISHED(tp->t_state) ||
3146 		     tfo_syn)) {
3147 			if (DELAY_ACK(tp, tlen) || tfo_syn)
3148 				tp->t_flags |= TF_DELACK;
3149 			else
3150 				tp->t_flags |= TF_ACKNOW;
3151 			tp->rcv_nxt += tlen;
3152 			if (tlen &&
3153 			    ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3154 			    (tp->t_fbyte_in == 0)) {
3155 				tp->t_fbyte_in = ticks;
3156 				if (tp->t_fbyte_in == 0)
3157 					tp->t_fbyte_in = 1;
3158 				if (tp->t_fbyte_out && tp->t_fbyte_in)
3159 					tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3160 			}
3161 			thflags = th->th_flags & TH_FIN;
3162 			TCPSTAT_INC(tcps_rcvpack);
3163 			TCPSTAT_ADD(tcps_rcvbyte, tlen);
3164 			SOCKBUF_LOCK(&so->so_rcv);
3165 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3166 				m_freem(m);
3167 			else
3168 				sbappendstream_locked(&so->so_rcv, m, 0);
3169 			SOCKBUF_UNLOCK(&so->so_rcv);
3170 			tp->t_flags |= TF_WAKESOR;
3171 		} else {
3172 			/*
3173 			 * XXX: Due to the header drop above "th" is
3174 			 * theoretically invalid by now.  Fortunately
3175 			 * m_adj() doesn't actually frees any mbufs
3176 			 * when trimming from the head.
3177 			 */
3178 			tcp_seq temp = save_start;
3179 			thflags = tcp_reass(tp, th, &temp, &tlen, m);
3180 			tp->t_flags |= TF_ACKNOW;
3181 		}
3182 		if ((tp->t_flags & TF_SACK_PERMIT) &&
3183 		    (save_tlen > 0) &&
3184 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
3185 			if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3186 				/*
3187 				 * DSACK actually handled in the fastpath
3188 				 * above.
3189 				 */
3190 				tcp_update_sack_list(tp, save_start,
3191 				    save_start + save_tlen);
3192 			} else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3193 				if ((tp->rcv_numsacks >= 1) &&
3194 				    (tp->sackblks[0].end == save_start)) {
3195 					/*
3196 					 * Partial overlap, recorded at todrop
3197 					 * above.
3198 					 */
3199 					tcp_update_sack_list(tp,
3200 					    tp->sackblks[0].start,
3201 					    tp->sackblks[0].end);
3202 				} else {
3203 					tcp_update_dsack_list(tp, save_start,
3204 					    save_start + save_tlen);
3205 				}
3206 			} else if (tlen >= save_tlen) {
3207 				/* Update of sackblks. */
3208 				tcp_update_dsack_list(tp, save_start,
3209 				    save_start + save_tlen);
3210 			} else if (tlen > 0) {
3211 				tcp_update_dsack_list(tp, save_start,
3212 				    save_start + tlen);
3213 			}
3214 		}
3215 #if 0
3216 		/*
3217 		 * Note the amount of data that peer has sent into
3218 		 * our window, in order to estimate the sender's
3219 		 * buffer size.
3220 		 * XXX: Unused.
3221 		 */
3222 		if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3223 			len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3224 		else
3225 			len = so->so_rcv.sb_hiwat;
3226 #endif
3227 	} else {
3228 		m_freem(m);
3229 		thflags &= ~TH_FIN;
3230 	}
3231 
3232 	/*
3233 	 * If FIN is received ACK the FIN and let the user know
3234 	 * that the connection is closing.
3235 	 */
3236 	if (thflags & TH_FIN) {
3237 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3238 			socantrcvmore(so);
3239 			/* The socket upcall is handled by socantrcvmore. */
3240 			tp->t_flags &= ~TF_WAKESOR;
3241 			/*
3242 			 * If connection is half-synchronized
3243 			 * (ie NEEDSYN flag on) then delay ACK,
3244 			 * so it may be piggybacked when SYN is sent.
3245 			 * Otherwise, since we received a FIN then no
3246 			 * more input can be expected, send ACK now.
3247 			 */
3248 			if (tp->t_flags & TF_NEEDSYN)
3249 				tp->t_flags |= TF_DELACK;
3250 			else
3251 				tp->t_flags |= TF_ACKNOW;
3252 			tp->rcv_nxt++;
3253 		}
3254 		switch (tp->t_state) {
3255 		/*
3256 		 * In SYN_RECEIVED and ESTABLISHED STATES
3257 		 * enter the CLOSE_WAIT state.
3258 		 */
3259 		case TCPS_SYN_RECEIVED:
3260 			tp->t_starttime = ticks;
3261 			/* FALLTHROUGH */
3262 		case TCPS_ESTABLISHED:
3263 			tcp_state_change(tp, TCPS_CLOSE_WAIT);
3264 			break;
3265 
3266 		/*
3267 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3268 		 * enter the CLOSING state.
3269 		 */
3270 		case TCPS_FIN_WAIT_1:
3271 			tcp_state_change(tp, TCPS_CLOSING);
3272 			break;
3273 
3274 		/*
3275 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3276 		 * starting the time-wait timer, turning off the other
3277 		 * standard timers.
3278 		 */
3279 		case TCPS_FIN_WAIT_2:
3280 			tcp_twstart(tp);
3281 			return;
3282 		}
3283 	}
3284 #ifdef TCPDEBUG
3285 	if (so->so_options & SO_DEBUG)
3286 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3287 			  &tcp_savetcp, 0);
3288 #endif
3289 	TCP_PROBE3(debug__input, tp, th, m);
3290 
3291 	/*
3292 	 * Return any desired output.
3293 	 */
3294 	if (needoutput || (tp->t_flags & TF_ACKNOW))
3295 		(void) tp->t_fb->tfb_tcp_output(tp);
3296 
3297 check_delack:
3298 	INP_WLOCK_ASSERT(tp->t_inpcb);
3299 
3300 	if (tp->t_flags & TF_DELACK) {
3301 		tp->t_flags &= ~TF_DELACK;
3302 		tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3303 	}
3304 	tcp_handle_wakeup(tp, so);
3305 	INP_WUNLOCK(tp->t_inpcb);
3306 	return;
3307 
3308 dropafterack:
3309 	/*
3310 	 * Generate an ACK dropping incoming segment if it occupies
3311 	 * sequence space, where the ACK reflects our state.
3312 	 *
3313 	 * We can now skip the test for the RST flag since all
3314 	 * paths to this code happen after packets containing
3315 	 * RST have been dropped.
3316 	 *
3317 	 * In the SYN-RECEIVED state, don't send an ACK unless the
3318 	 * segment we received passes the SYN-RECEIVED ACK test.
3319 	 * If it fails send a RST.  This breaks the loop in the
3320 	 * "LAND" DoS attack, and also prevents an ACK storm
3321 	 * between two listening ports that have been sent forged
3322 	 * SYN segments, each with the source address of the other.
3323 	 */
3324 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3325 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
3326 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
3327 		rstreason = BANDLIM_RST_OPENPORT;
3328 		goto dropwithreset;
3329 	}
3330 #ifdef TCPDEBUG
3331 	if (so->so_options & SO_DEBUG)
3332 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3333 			  &tcp_savetcp, 0);
3334 #endif
3335 	TCP_PROBE3(debug__input, tp, th, m);
3336 	tp->t_flags |= TF_ACKNOW;
3337 	(void) tp->t_fb->tfb_tcp_output(tp);
3338 	tcp_handle_wakeup(tp, so);
3339 	INP_WUNLOCK(tp->t_inpcb);
3340 	m_freem(m);
3341 	return;
3342 
3343 dropwithreset:
3344 	if (tp != NULL) {
3345 		tcp_dropwithreset(m, th, tp, tlen, rstreason);
3346 		tcp_handle_wakeup(tp, so);
3347 		INP_WUNLOCK(tp->t_inpcb);
3348 	} else
3349 		tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3350 	return;
3351 
3352 drop:
3353 	/*
3354 	 * Drop space held by incoming segment and return.
3355 	 */
3356 #ifdef TCPDEBUG
3357 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3358 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3359 			  &tcp_savetcp, 0);
3360 #endif
3361 	TCP_PROBE3(debug__input, tp, th, m);
3362 	if (tp != NULL) {
3363 		tcp_handle_wakeup(tp, so);
3364 		INP_WUNLOCK(tp->t_inpcb);
3365 	}
3366 	m_freem(m);
3367 }
3368 
3369 /*
3370  * Issue RST and make ACK acceptable to originator of segment.
3371  * The mbuf must still include the original packet header.
3372  * tp may be NULL.
3373  */
3374 void
3375 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3376     int tlen, int rstreason)
3377 {
3378 #ifdef INET
3379 	struct ip *ip;
3380 #endif
3381 #ifdef INET6
3382 	struct ip6_hdr *ip6;
3383 #endif
3384 
3385 	if (tp != NULL) {
3386 		INP_WLOCK_ASSERT(tp->t_inpcb);
3387 	}
3388 
3389 	/* Don't bother if destination was broadcast/multicast. */
3390 	if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3391 		goto drop;
3392 #ifdef INET6
3393 	if (mtod(m, struct ip *)->ip_v == 6) {
3394 		ip6 = mtod(m, struct ip6_hdr *);
3395 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3396 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3397 			goto drop;
3398 		/* IPv6 anycast check is done at tcp6_input() */
3399 	}
3400 #endif
3401 #if defined(INET) && defined(INET6)
3402 	else
3403 #endif
3404 #ifdef INET
3405 	{
3406 		ip = mtod(m, struct ip *);
3407 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3408 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3409 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3410 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3411 			goto drop;
3412 	}
3413 #endif
3414 
3415 	/* Perform bandwidth limiting. */
3416 	if (badport_bandlim(rstreason) < 0)
3417 		goto drop;
3418 
3419 	/* tcp_respond consumes the mbuf chain. */
3420 	if (th->th_flags & TH_ACK) {
3421 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3422 		    th->th_ack, TH_RST);
3423 	} else {
3424 		if (th->th_flags & TH_SYN)
3425 			tlen++;
3426 		if (th->th_flags & TH_FIN)
3427 			tlen++;
3428 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3429 		    (tcp_seq)0, TH_RST|TH_ACK);
3430 	}
3431 	return;
3432 drop:
3433 	m_freem(m);
3434 }
3435 
3436 /*
3437  * Parse TCP options and place in tcpopt.
3438  */
3439 void
3440 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3441 {
3442 	int opt, optlen;
3443 
3444 	to->to_flags = 0;
3445 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
3446 		opt = cp[0];
3447 		if (opt == TCPOPT_EOL)
3448 			break;
3449 		if (opt == TCPOPT_NOP)
3450 			optlen = 1;
3451 		else {
3452 			if (cnt < 2)
3453 				break;
3454 			optlen = cp[1];
3455 			if (optlen < 2 || optlen > cnt)
3456 				break;
3457 		}
3458 		switch (opt) {
3459 		case TCPOPT_MAXSEG:
3460 			if (optlen != TCPOLEN_MAXSEG)
3461 				continue;
3462 			if (!(flags & TO_SYN))
3463 				continue;
3464 			to->to_flags |= TOF_MSS;
3465 			bcopy((char *)cp + 2,
3466 			    (char *)&to->to_mss, sizeof(to->to_mss));
3467 			to->to_mss = ntohs(to->to_mss);
3468 			break;
3469 		case TCPOPT_WINDOW:
3470 			if (optlen != TCPOLEN_WINDOW)
3471 				continue;
3472 			if (!(flags & TO_SYN))
3473 				continue;
3474 			to->to_flags |= TOF_SCALE;
3475 			to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3476 			break;
3477 		case TCPOPT_TIMESTAMP:
3478 			if (optlen != TCPOLEN_TIMESTAMP)
3479 				continue;
3480 			to->to_flags |= TOF_TS;
3481 			bcopy((char *)cp + 2,
3482 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
3483 			to->to_tsval = ntohl(to->to_tsval);
3484 			bcopy((char *)cp + 6,
3485 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3486 			to->to_tsecr = ntohl(to->to_tsecr);
3487 			break;
3488 		case TCPOPT_SIGNATURE:
3489 			/*
3490 			 * In order to reply to a host which has set the
3491 			 * TCP_SIGNATURE option in its initial SYN, we have
3492 			 * to record the fact that the option was observed
3493 			 * here for the syncache code to perform the correct
3494 			 * response.
3495 			 */
3496 			if (optlen != TCPOLEN_SIGNATURE)
3497 				continue;
3498 			to->to_flags |= TOF_SIGNATURE;
3499 			to->to_signature = cp + 2;
3500 			break;
3501 		case TCPOPT_SACK_PERMITTED:
3502 			if (optlen != TCPOLEN_SACK_PERMITTED)
3503 				continue;
3504 			if (!(flags & TO_SYN))
3505 				continue;
3506 			if (!V_tcp_do_sack)
3507 				continue;
3508 			to->to_flags |= TOF_SACKPERM;
3509 			break;
3510 		case TCPOPT_SACK:
3511 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3512 				continue;
3513 			if (flags & TO_SYN)
3514 				continue;
3515 			to->to_flags |= TOF_SACK;
3516 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3517 			to->to_sacks = cp + 2;
3518 			TCPSTAT_INC(tcps_sack_rcv_blocks);
3519 			break;
3520 		case TCPOPT_FAST_OPEN:
3521 			/*
3522 			 * Cookie length validation is performed by the
3523 			 * server side cookie checking code or the client
3524 			 * side cookie cache update code.
3525 			 */
3526 			if (!(flags & TO_SYN))
3527 				continue;
3528 			if (!V_tcp_fastopen_client_enable &&
3529 			    !V_tcp_fastopen_server_enable)
3530 				continue;
3531 			to->to_flags |= TOF_FASTOPEN;
3532 			to->to_tfo_len = optlen - 2;
3533 			to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3534 			break;
3535 		default:
3536 			continue;
3537 		}
3538 	}
3539 }
3540 
3541 /*
3542  * Pull out of band byte out of a segment so
3543  * it doesn't appear in the user's data queue.
3544  * It is still reflected in the segment length for
3545  * sequencing purposes.
3546  */
3547 void
3548 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3549     int off)
3550 {
3551 	int cnt = off + th->th_urp - 1;
3552 
3553 	while (cnt >= 0) {
3554 		if (m->m_len > cnt) {
3555 			char *cp = mtod(m, caddr_t) + cnt;
3556 			struct tcpcb *tp = sototcpcb(so);
3557 
3558 			INP_WLOCK_ASSERT(tp->t_inpcb);
3559 
3560 			tp->t_iobc = *cp;
3561 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3562 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3563 			m->m_len--;
3564 			if (m->m_flags & M_PKTHDR)
3565 				m->m_pkthdr.len--;
3566 			return;
3567 		}
3568 		cnt -= m->m_len;
3569 		m = m->m_next;
3570 		if (m == NULL)
3571 			break;
3572 	}
3573 	panic("tcp_pulloutofband");
3574 }
3575 
3576 /*
3577  * Collect new round-trip time estimate
3578  * and update averages and current timeout.
3579  */
3580 void
3581 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3582 {
3583 	int delta;
3584 
3585 	INP_WLOCK_ASSERT(tp->t_inpcb);
3586 
3587 	TCPSTAT_INC(tcps_rttupdated);
3588 	tp->t_rttupdated++;
3589 #ifdef STATS
3590 	stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3591 	    imax(0, rtt * 1000 / hz));
3592 #endif
3593 	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3594 		/*
3595 		 * srtt is stored as fixed point with 5 bits after the
3596 		 * binary point (i.e., scaled by 8).  The following magic
3597 		 * is equivalent to the smoothing algorithm in rfc793 with
3598 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3599 		 * point).  Adjust rtt to origin 0.
3600 		 */
3601 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3602 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3603 
3604 		if ((tp->t_srtt += delta) <= 0)
3605 			tp->t_srtt = 1;
3606 
3607 		/*
3608 		 * We accumulate a smoothed rtt variance (actually, a
3609 		 * smoothed mean difference), then set the retransmit
3610 		 * timer to smoothed rtt + 4 times the smoothed variance.
3611 		 * rttvar is stored as fixed point with 4 bits after the
3612 		 * binary point (scaled by 16).  The following is
3613 		 * equivalent to rfc793 smoothing with an alpha of .75
3614 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3615 		 * rfc793's wired-in beta.
3616 		 */
3617 		if (delta < 0)
3618 			delta = -delta;
3619 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3620 		if ((tp->t_rttvar += delta) <= 0)
3621 			tp->t_rttvar = 1;
3622 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3623 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3624 	} else {
3625 		/*
3626 		 * No rtt measurement yet - use the unsmoothed rtt.
3627 		 * Set the variance to half the rtt (so our first
3628 		 * retransmit happens at 3*rtt).
3629 		 */
3630 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
3631 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3632 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3633 	}
3634 	tp->t_rtttime = 0;
3635 	tp->t_rxtshift = 0;
3636 
3637 	/*
3638 	 * the retransmit should happen at rtt + 4 * rttvar.
3639 	 * Because of the way we do the smoothing, srtt and rttvar
3640 	 * will each average +1/2 tick of bias.  When we compute
3641 	 * the retransmit timer, we want 1/2 tick of rounding and
3642 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3643 	 * firing of the timer.  The bias will give us exactly the
3644 	 * 1.5 tick we need.  But, because the bias is
3645 	 * statistical, we have to test that we don't drop below
3646 	 * the minimum feasible timer (which is 2 ticks).
3647 	 */
3648 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3649 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3650 
3651 	/*
3652 	 * We received an ack for a packet that wasn't retransmitted;
3653 	 * it is probably safe to discard any error indications we've
3654 	 * received recently.  This isn't quite right, but close enough
3655 	 * for now (a route might have failed after we sent a segment,
3656 	 * and the return path might not be symmetrical).
3657 	 */
3658 	tp->t_softerror = 0;
3659 }
3660 
3661 /*
3662  * Determine a reasonable value for maxseg size.
3663  * If the route is known, check route for mtu.
3664  * If none, use an mss that can be handled on the outgoing interface
3665  * without forcing IP to fragment.  If no route is found, route has no mtu,
3666  * or the destination isn't local, use a default, hopefully conservative
3667  * size (usually 512 or the default IP max size, but no more than the mtu
3668  * of the interface), as we can't discover anything about intervening
3669  * gateways or networks.  We also initialize the congestion/slow start
3670  * window to be a single segment if the destination isn't local.
3671  * While looking at the routing entry, we also initialize other path-dependent
3672  * parameters from pre-set or cached values in the routing entry.
3673  *
3674  * NOTE that resulting t_maxseg doesn't include space for TCP options or
3675  * IP options, e.g. IPSEC data, since length of this data may vary, and
3676  * thus it is calculated for every segment separately in tcp_output().
3677  *
3678  * NOTE that this routine is only called when we process an incoming
3679  * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3680  * settings are handled in tcp_mssopt().
3681  */
3682 void
3683 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3684     struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3685 {
3686 	int mss = 0;
3687 	uint32_t maxmtu = 0;
3688 	struct inpcb *inp = tp->t_inpcb;
3689 	struct hc_metrics_lite metrics;
3690 #ifdef INET6
3691 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3692 	size_t min_protoh = isipv6 ?
3693 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3694 			    sizeof (struct tcpiphdr);
3695 #else
3696 	const size_t min_protoh = sizeof(struct tcpiphdr);
3697 #endif
3698 
3699 	INP_WLOCK_ASSERT(tp->t_inpcb);
3700 
3701 	if (mtuoffer != -1) {
3702 		KASSERT(offer == -1, ("%s: conflict", __func__));
3703 		offer = mtuoffer - min_protoh;
3704 	}
3705 
3706 	/* Initialize. */
3707 #ifdef INET6
3708 	if (isipv6) {
3709 		maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3710 		tp->t_maxseg = V_tcp_v6mssdflt;
3711 	}
3712 #endif
3713 #if defined(INET) && defined(INET6)
3714 	else
3715 #endif
3716 #ifdef INET
3717 	{
3718 		maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3719 		tp->t_maxseg = V_tcp_mssdflt;
3720 	}
3721 #endif
3722 
3723 	/*
3724 	 * No route to sender, stay with default mss and return.
3725 	 */
3726 	if (maxmtu == 0) {
3727 		/*
3728 		 * In case we return early we need to initialize metrics
3729 		 * to a defined state as tcp_hc_get() would do for us
3730 		 * if there was no cache hit.
3731 		 */
3732 		if (metricptr != NULL)
3733 			bzero(metricptr, sizeof(struct hc_metrics_lite));
3734 		return;
3735 	}
3736 
3737 	/* What have we got? */
3738 	switch (offer) {
3739 		case 0:
3740 			/*
3741 			 * Offer == 0 means that there was no MSS on the SYN
3742 			 * segment, in this case we use tcp_mssdflt as
3743 			 * already assigned to t_maxseg above.
3744 			 */
3745 			offer = tp->t_maxseg;
3746 			break;
3747 
3748 		case -1:
3749 			/*
3750 			 * Offer == -1 means that we didn't receive SYN yet.
3751 			 */
3752 			/* FALLTHROUGH */
3753 
3754 		default:
3755 			/*
3756 			 * Prevent DoS attack with too small MSS. Round up
3757 			 * to at least minmss.
3758 			 */
3759 			offer = max(offer, V_tcp_minmss);
3760 	}
3761 
3762 	/*
3763 	 * rmx information is now retrieved from tcp_hostcache.
3764 	 */
3765 	tcp_hc_get(&inp->inp_inc, &metrics);
3766 	if (metricptr != NULL)
3767 		bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3768 
3769 	/*
3770 	 * If there's a discovered mtu in tcp hostcache, use it.
3771 	 * Else, use the link mtu.
3772 	 */
3773 	if (metrics.rmx_mtu)
3774 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3775 	else {
3776 #ifdef INET6
3777 		if (isipv6) {
3778 			mss = maxmtu - min_protoh;
3779 			if (!V_path_mtu_discovery &&
3780 			    !in6_localaddr(&inp->in6p_faddr))
3781 				mss = min(mss, V_tcp_v6mssdflt);
3782 		}
3783 #endif
3784 #if defined(INET) && defined(INET6)
3785 		else
3786 #endif
3787 #ifdef INET
3788 		{
3789 			mss = maxmtu - min_protoh;
3790 			if (!V_path_mtu_discovery &&
3791 			    !in_localaddr(inp->inp_faddr))
3792 				mss = min(mss, V_tcp_mssdflt);
3793 		}
3794 #endif
3795 		/*
3796 		 * XXX - The above conditional (mss = maxmtu - min_protoh)
3797 		 * probably violates the TCP spec.
3798 		 * The problem is that, since we don't know the
3799 		 * other end's MSS, we are supposed to use a conservative
3800 		 * default.  But, if we do that, then MTU discovery will
3801 		 * never actually take place, because the conservative
3802 		 * default is much less than the MTUs typically seen
3803 		 * on the Internet today.  For the moment, we'll sweep
3804 		 * this under the carpet.
3805 		 *
3806 		 * The conservative default might not actually be a problem
3807 		 * if the only case this occurs is when sending an initial
3808 		 * SYN with options and data to a host we've never talked
3809 		 * to before.  Then, they will reply with an MSS value which
3810 		 * will get recorded and the new parameters should get
3811 		 * recomputed.  For Further Study.
3812 		 */
3813 	}
3814 	mss = min(mss, offer);
3815 
3816 	/*
3817 	 * Sanity check: make sure that maxseg will be large
3818 	 * enough to allow some data on segments even if the
3819 	 * all the option space is used (40bytes).  Otherwise
3820 	 * funny things may happen in tcp_output.
3821 	 *
3822 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3823 	 */
3824 	mss = max(mss, 64);
3825 
3826 	tp->t_maxseg = mss;
3827 }
3828 
3829 void
3830 tcp_mss(struct tcpcb *tp, int offer)
3831 {
3832 	int mss;
3833 	uint32_t bufsize;
3834 	struct inpcb *inp;
3835 	struct socket *so;
3836 	struct hc_metrics_lite metrics;
3837 	struct tcp_ifcap cap;
3838 
3839 	KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3840 
3841 	bzero(&cap, sizeof(cap));
3842 	tcp_mss_update(tp, offer, -1, &metrics, &cap);
3843 
3844 	mss = tp->t_maxseg;
3845 	inp = tp->t_inpcb;
3846 
3847 	/*
3848 	 * If there's a pipesize, change the socket buffer to that size,
3849 	 * don't change if sb_hiwat is different than default (then it
3850 	 * has been changed on purpose with setsockopt).
3851 	 * Make the socket buffers an integral number of mss units;
3852 	 * if the mss is larger than the socket buffer, decrease the mss.
3853 	 */
3854 	so = inp->inp_socket;
3855 	SOCKBUF_LOCK(&so->so_snd);
3856 	if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3857 		bufsize = metrics.rmx_sendpipe;
3858 	else
3859 		bufsize = so->so_snd.sb_hiwat;
3860 	if (bufsize < mss)
3861 		mss = bufsize;
3862 	else {
3863 		bufsize = roundup(bufsize, mss);
3864 		if (bufsize > sb_max)
3865 			bufsize = sb_max;
3866 		if (bufsize > so->so_snd.sb_hiwat)
3867 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3868 	}
3869 	SOCKBUF_UNLOCK(&so->so_snd);
3870 	/*
3871 	 * Sanity check: make sure that maxseg will be large
3872 	 * enough to allow some data on segments even if the
3873 	 * all the option space is used (40bytes).  Otherwise
3874 	 * funny things may happen in tcp_output.
3875 	 *
3876 	 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3877 	 */
3878 	tp->t_maxseg = max(mss, 64);
3879 
3880 	SOCKBUF_LOCK(&so->so_rcv);
3881 	if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3882 		bufsize = metrics.rmx_recvpipe;
3883 	else
3884 		bufsize = so->so_rcv.sb_hiwat;
3885 	if (bufsize > mss) {
3886 		bufsize = roundup(bufsize, mss);
3887 		if (bufsize > sb_max)
3888 			bufsize = sb_max;
3889 		if (bufsize > so->so_rcv.sb_hiwat)
3890 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3891 	}
3892 	SOCKBUF_UNLOCK(&so->so_rcv);
3893 
3894 	/* Check the interface for TSO capabilities. */
3895 	if (cap.ifcap & CSUM_TSO) {
3896 		tp->t_flags |= TF_TSO;
3897 		tp->t_tsomax = cap.tsomax;
3898 		tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3899 		tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3900 	}
3901 }
3902 
3903 /*
3904  * Determine the MSS option to send on an outgoing SYN.
3905  */
3906 int
3907 tcp_mssopt(struct in_conninfo *inc)
3908 {
3909 	int mss = 0;
3910 	uint32_t thcmtu = 0;
3911 	uint32_t maxmtu = 0;
3912 	size_t min_protoh;
3913 
3914 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3915 
3916 #ifdef INET6
3917 	if (inc->inc_flags & INC_ISIPV6) {
3918 		mss = V_tcp_v6mssdflt;
3919 		maxmtu = tcp_maxmtu6(inc, NULL);
3920 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3921 	}
3922 #endif
3923 #if defined(INET) && defined(INET6)
3924 	else
3925 #endif
3926 #ifdef INET
3927 	{
3928 		mss = V_tcp_mssdflt;
3929 		maxmtu = tcp_maxmtu(inc, NULL);
3930 		min_protoh = sizeof(struct tcpiphdr);
3931 	}
3932 #endif
3933 #if defined(INET6) || defined(INET)
3934 	thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3935 #endif
3936 
3937 	if (maxmtu && thcmtu)
3938 		mss = min(maxmtu, thcmtu) - min_protoh;
3939 	else if (maxmtu || thcmtu)
3940 		mss = max(maxmtu, thcmtu) - min_protoh;
3941 
3942 	return (mss);
3943 }
3944 
3945 void
3946 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3947 {
3948 	int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3949 	int maxseg = tcp_maxseg(tp);
3950 
3951 	INP_WLOCK_ASSERT(tp->t_inpcb);
3952 
3953 	tcp_timer_activate(tp, TT_REXMT, 0);
3954 	tp->t_rtttime = 0;
3955 	/*
3956 	 * Compute the amount of data that this ACK is indicating
3957 	 * (del_data) and an estimate of how many bytes are in the
3958 	 * network.
3959 	 */
3960 	del_data = tp->sackhint.delivered_data;
3961 	if (V_tcp_do_rfc6675_pipe)
3962 		pipe = tcp_compute_pipe(tp);
3963 	else
3964 		pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3965 	tp->sackhint.prr_delivered += del_data;
3966 	/*
3967 	 * Proportional Rate Reduction
3968 	 */
3969 	if (pipe >= tp->snd_ssthresh) {
3970 		if (tp->sackhint.recover_fs == 0)
3971 			tp->sackhint.recover_fs =
3972 			    imax(1, tp->snd_nxt - tp->snd_una);
3973 		snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3974 			    tp->snd_ssthresh, tp->sackhint.recover_fs) -
3975 			    tp->sackhint.prr_out;
3976 	} else {
3977 		if (V_tcp_do_prr_conservative)
3978 			limit = tp->sackhint.prr_delivered -
3979 			    tp->sackhint.prr_out;
3980 		else
3981 			limit = imax(tp->sackhint.prr_delivered -
3982 				    tp->sackhint.prr_out, del_data) +
3983 				    maxseg;
3984 		snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3985 	}
3986 	snd_cnt = imax(snd_cnt, 0) / maxseg;
3987 	/*
3988 	 * Send snd_cnt new data into the network in response to this ack.
3989 	 * If there is going to be a SACK retransmission, adjust snd_cwnd
3990 	 * accordingly.
3991 	 */
3992 	tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
3993 		tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
3994 	tp->t_flags |= TF_ACKNOW;
3995 	(void) tcp_output(tp);
3996 }
3997 
3998 /*
3999  * On a partial ack arrives, force the retransmission of the
4000  * next unacknowledged segment.  Do not clear tp->t_dupacks.
4001  * By setting snd_nxt to ti_ack, this forces retransmission timer to
4002  * be started again.
4003  */
4004 void
4005 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4006 {
4007 	tcp_seq onxt = tp->snd_nxt;
4008 	uint32_t ocwnd = tp->snd_cwnd;
4009 	u_int maxseg = tcp_maxseg(tp);
4010 
4011 	INP_WLOCK_ASSERT(tp->t_inpcb);
4012 
4013 	tcp_timer_activate(tp, TT_REXMT, 0);
4014 	tp->t_rtttime = 0;
4015 	tp->snd_nxt = th->th_ack;
4016 	/*
4017 	 * Set snd_cwnd to one segment beyond acknowledged offset.
4018 	 * (tp->snd_una has not yet been updated when this function is called.)
4019 	 */
4020 	tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4021 	tp->t_flags |= TF_ACKNOW;
4022 	(void) tp->t_fb->tfb_tcp_output(tp);
4023 	tp->snd_cwnd = ocwnd;
4024 	if (SEQ_GT(onxt, tp->snd_nxt))
4025 		tp->snd_nxt = onxt;
4026 	/*
4027 	 * Partial window deflation.  Relies on fact that tp->snd_una
4028 	 * not updated yet.
4029 	 */
4030 	if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4031 		tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4032 	else
4033 		tp->snd_cwnd = 0;
4034 	tp->snd_cwnd += maxseg;
4035 }
4036 
4037 int
4038 tcp_compute_pipe(struct tcpcb *tp)
4039 {
4040 	return (tp->snd_max - tp->snd_una +
4041 		tp->sackhint.sack_bytes_rexmit -
4042 		tp->sackhint.sacked_bytes);
4043 }
4044 
4045 uint32_t
4046 tcp_compute_initwnd(uint32_t maxseg)
4047 {
4048 	/*
4049 	 * Calculate the Initial Window, also used as Restart Window
4050 	 *
4051 	 * RFC5681 Section 3.1 specifies the default conservative values.
4052 	 * RFC3390 specifies slightly more aggressive values.
4053 	 * RFC6928 increases it to ten segments.
4054 	 * Support for user specified value for initial flight size.
4055 	 */
4056 	if (V_tcp_initcwnd_segments)
4057 		return min(V_tcp_initcwnd_segments * maxseg,
4058 		    max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4059 	else if (V_tcp_do_rfc3390)
4060 		return min(4 * maxseg, max(2 * maxseg, 4380));
4061 	else {
4062 		/* Per RFC5681 Section 3.1 */
4063 		if (maxseg > 2190)
4064 			return (2 * maxseg);
4065 		else if (maxseg > 1095)
4066 			return (3 * maxseg);
4067 		else
4068 			return (4 * maxseg);
4069 	}
4070 }
4071